Define an auto cleanup handler for use with repo locking. This is based
on the existing auto transaction cleanup. A wrapper for
ostree_repo_lock_push() is added with it. The intended usage is like so:
g_autoptr(OstreeRepoAutoLock) lock = NULL;
lock = ostree_repo_auto_lock_push (repo, lock_type, cancellable, error);
if (!lock)
return FALSE;
The functions and type are marked to be skipped by introspection since I
can't see them being usable from bindings.
Closes: #1343
Approved by: cgwalters
Currently ostree has no method of guarding against concurrent pruning.
When there are multiple repo writers, it's possible to have a pull or
commit race against a prune and end up with missing objects.
This adds a file based repo locking mechanism. The intention is to take
a shared lock when writing objects and an exclusive lock when deleting
them. In order to make use of the locking throughout the library in a
fine grained fashion, the lock acts recursively with a stack of lock
states. If the lock becomes exclusive, it will stay in that state until
the stack is unwound past the initial exclusive push. The file locking
is similar to GLnxLockFile in that it uses open file descriptor locks
but falls back to flock when needed.
The lock also attempts to be thread safe by storing the lock state in
thread local storage with GPrivate. This means that each thread will
have an independent lock for each repository it opens. There are some
drawbacks to that, but it seemed impossible to manage the lock state
coherently in the face of multithreaded access.
The API is a push/pop interface in accordance with the recursive nature
of the locking. The push interface uses an enum that's translated to
LOCK_SH or LOCK_EX as needed. Both interfaces use an internal timeout
field to decide whether to manage the lock in a blocking or non-blocking
fashion. The intention is to allow ostree applications as well as
administrators to control this timeout. For now, the default is a 30
second timeout.
Note that the timeout is handled synchronously in thread since the lock
is maintained in thread local storage. I.e., the thread that acquires
the lock needs to be the same thread that runs the operation. There may
be a way to offer an asynchronous version, but it's not clear exactly
how that would work since it would likely involve a separate thread that
invokes a callback when the locking operation completes.
https://bugzilla.gnome.org/show_bug.cgi?id=759442Closes: #1343
Approved by: cgwalters
For rpm-ostree I'd like to do importing in parallel with threads; the code is
*almost* ready for that except today it calls
`ostree_repo_transaction_set_ref()`.
Looking at the code, there's really a "transaction" struct here,
not just stats. Let's lift that struct out, and move the refs
into it under the existing lock.
Clarify the documentation around multithreading for various functions.
Closes: #1358
Approved by: jlebon
This lowers into the commit core what the static delta code
was doing, and improves the API.
The bigger picture issue is that for writing large files, our current "pull" API
where the caller provides a `GInputStream` is very awkward in some scenarios.
For example, we have a whole "libarchive input stream" that is a ~200 line
GObject that boils down to wrapping `archive_read_data()`.
This came more to a head when I was working on rpm-ostree jigdo since I had to
copy that object.
One step we can take after this is to further split `write_content_object()`
into a "write symlink or archive object" versus "write bare content object"
(it already has a mess of conditionals) and teach the latter case to call
this.
The eventual goal here is to make this API public.
Closes: #1355
Approved by: jlebon
The way `_ostree_repo_open_content_bare()` did both looking for the object and
possibly creating a new fd was just weird and inconsistent with e.g. the pull
code where we always call `has_object()` first.
Just call `has_object()` in the delta paths that used this too, making the
implementation right now a thin wrapper around
`glnx_open_tmpfile_linkable_at()`, but this is prep for a later patch which does
more.
Closes: #1283
Approved by: jlebon
It's no longer called directly by the pull code, so make it static.
The goal here is to have the pull and local-fs commit paths use higher level
more efficient APIs, and eventually make those APIs public.
Closes: #1257
Approved by: jlebon
We added a `.dir-locals.el` in commit: 9a77017d87
There's no need to have it per-file, with that people might think
to add other editors, which is the wrong direction.
Closes: #1206
Approved by: jlebon
Conceptually `ostree-repo-pull.c` should be be written using
just public APIs; we theoretically support building without HTTP
for people who just want to use the object store portion and
do their own fetching.
We have some nontrivial behaviors in the pull layer though; one
of those is the "bareuseronly" verification. Make a new internal
API that accepts flags, move it into `commit.c`. This
is prep for further work in changing object import to support
reflinks.
Closes: #1193
Approved by: jlebon
Update libglnx, which is mostly port the repo stagedir code
to the new tmpdir API. This turned out to require some
libglnx changes to support de-allocating the tmpdir ref while
still maintaining the on-disk dir.
Update submodule: libglnx
Closes: #1172
Approved by: jlebon
Doing this in prep for libglnx tmpdir porting, but I think we should also do
this because the partial fetch code IMO was never fully baked; among other
things it was never integrated into the scheme we came up with for "boot id
sync" that we use for complete/staged objects.
There's a lot of complexity here that while we have some coverage for, I think
we need to refocus on the core functionality. The libcurl backend doesn't have
an equivalent to this today.
In particular for small objects, this is simply overly complex. The downside is
clearly for large objects like FAH's 61MB initramfs; not being able to resume
fetches of those is unfortunate.
In practice though, I think most people should be using deltas, and we need to
make sure deltas work for large objects anyways.
Further ultimately the peer-to-peer work should help a lot for people
with truly unreliable connections.
Closes: #1176
Approved by: jlebon
This option allows a repo to explicitly opt out of adding new remotes in
a remotes configuration directory. This currently defaults to true for
system repos and false for non-system repos to maintain legacy behavior
that non-system repos don't add remotes in a configuration directory.
That would be problematic for flatpak, which specifies a remotes config
dir but adds remotes in ways that are incompatible with it.
So, what this really does is allow system repos to control whether they
want to add remotes in the config dir or not. That's important if your
flatpak repo is the system repo like at Endless.
Closes: #1134Closes: #1155
Approved by: cgwalters
Steal some code from flatpak for this, which allows porting a few more things to
new style. I started on a public API version of this but was trying to roll some
other things into it and it snowballed. Let's do this version since it's easy
for now.
While here I changed things so that `generate_deployment_refs()` now just uses
`_set_ref_immediate()` rather than requring a txn.
Also, AFAICS there was no test coverage of `generate_deployment_refs()`; I tried
commenting it out and at least `admin-test.sh` passed. Add some coverage of this
- I verified that with this commenting out bits of that function cause the test
to fail.
Closes: #1132
Approved by: jlebon
When working with collections it can be useful to see remote refs rather
than just local and mirrored ones. This commit changes the "ostree refs
-c" output to include remote refs, and includes remote refs with
collection IDs in summary file generation as well. The former behavior
is consistent with how "ostree refs" works, and the latter behavior is
useful in facilitating P2P updates even when mirrors haven't been
configured.
To accomplish this, OstreeRepoListRefsExtFlags was extended with an
EXCLUDE_REMOTES flag. This was done rather than an INCLUDE_REMOTES flag
so that existing calls to ostree_repo_list_refs_ext continue to have the
same behavior. This flag was added to ostree_repo_list_collection_refs
(which is an experimental API break).
Also, add unit tests for the "refs -c" and summary file behavior, and
update relevant tests.
Closes: #1069
Approved by: cgwalters
This essentially completes our fd-relative conversion.
While here, I cleaned up the semantics of `ostree_repo_create()` and
`ostree_repo_create_at()` to be more atomic - basically various scripts were
testing for the `objects` subdirectory, so let's formalize that.
Closes: #820
Approved by: jlebon
There are multiple use cases where we'd like to alias refs.
First, having a "stable" alias which gets swapped across major
versions: https://pagure.io/atomic-wg/issue/228
Another case is when a ref is obsoleted;
<https://pagure.io/atomic-wg/issue/303>
This second one could be done with endoflife rebase, but I think
this case is better on the server side, as we might later change
our minds and do actual releases there.
I initially just added some test cases for symlinks in the `refs/heads` dir to
ensure this actually works (and it did), but I think it's worth having APIs.
Closes: #1033
Approved by: jlebon
These were previously private, but since we expect people to use them, let's add
`#define`s like we did for some of the other commit metadata.
Closes: #1028
Approved by: jlebon
(Note this PR was reverted in <https://github.com/ostreedev/ostree/pull/902>;
this version should be better)
Using `${sysroot}` to mean the physical storage root: We don't want to write to
`${sysroot}/etc/ostree/remotes.d`, since nothing will read it, and really
`${sysroot}` should just have `/ostree` (ideally). Today the Anaconda rpmostree
code ends up writing there. Fix this by adding a notion of "physical" sysroot.
We determine whether the path is physical by checking for `/sysroot`, which
exists in deployment roots (and there shouldn't be a `${sysroot}/sysroot`).
In order to unit test this, I added a `--sysroot` argument to `remote add`.
However, doing this better would require reworking the command line parsing for
the `remote` argument to support specifying `--repo` or `--sysroot`, and I
didn't quite want to do that yet in this patch.
This second iteration of this patch fixes the bug we hit the first time;
embarassingly enough I broke `ostree remote list` finding system remotes.
The fix is to have `ostree_repo_open()` figure out whether it's the same
as `/ostree/repo` for now.
Down the line...we might consider having the `ostree remote` command line itself
instatiate an `OstreeSysroot` by default, but this maximizes compatibility; we
just have to pay a small cost that `ostree` usage outside of that case like
`ostree static-delta` in a releng Jenkins job or whatever will do this `stat()`
too.
Closes: https://github.com/ostreedev/ostree/issues/892Closes: #1008
Approved by: mbarnes
The collection and ref bindings are stored in the commit metadata
under ostree.collection-binding and ostree.ref-binding,
respectively. They will be used to verify if the commit really comes
from the collection and ref we wanted to pull from.
Closes: #972
Approved by: cgwalters
And in general, if for some reason we can't write `user.` xattrs, provide an
error immediately rather than doing it during a later pull. This way the failure
cause is a lot more obvious.
Related: https://github.com/ostreedev/ostree/issues/991Closes: #993
Approved by: jlebon
For ostree-as-host, we're the superuser, so we'll blow past
any reserved free space by default. While deltas have size
metadata, if one happens to do a loose fetch, we can fill
up the disk.
Another case is flatpak: the system helper has similar concerns
here as ostree-as-host, and for `flatpak --user`, we also
want to be nice and avoid filling up the user's quota.
Closes: https://github.com/ostreedev/ostree/issues/962Closes: #987
Approved by: jlebon
There's lots of mechanically replacing `OtTmpFile` with `GLnxTmpfile`;
the biggest changes are in the commit path. Symlink commits are now
very clearly separated from regular files. Symlinks are `OtCleanupUnlinkat`,
and regular files are `GLnxTmpfile`.
The commit codepath separates those as `_ostree_repo_commit_path_final()` and
`_ostree_repo_commit_tmpf_final()`. A nice aspect of all of this is that they
both *consume* the temporary on success. This avoids an extra spurious
`unlink()` call.
One of the biggest bits of code motion is in `commit_loose_regfile_object()`,
which no longer needs to care about symlinks. For the most parth though it's
just removing conditionals.
Update submodule: libglnx
Closes: #958
Approved by: jlebon
The `OstreeRepoContentBareCommit` struct was basically an `OtTmpFile`, so let's
make it one. I moved the "convert to `GOutputStream`" logic into the callers,
since that bit can't fail; it makes the implementation much simpler since we can
just return the result of `ot_open_tmpfile_linkable_at()`.
Prep for `GLnxTmpfile` porting.
Closes: #957
Approved by: jlebon
These are tuples of (collection ID, ref name) which are a globally-unique
form of local ref. They use OstreeCollectionRef as an identifier, and hence
need to be accessed using new API, as the existing API uses string
identifiers and sometimes accepts refspecs. Remote names are not
supported as part an OstreeCollectionRef.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #924
Approved by: cgwalters
Add {get,set}_collection_id() methods to OstreeRepo and some documentation
about the concept of a collection ID which globally identifies an
upstream repository. See the documentation for more details.
This will be used in future commits. For now, the new API is marked as
experimental (--enable-experimental-api).
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #924
Approved by: cgwalters
This is followon work from previous cleanups. Basically
`stat_bare_content_object()` was the `fstatat()` logic
and `ostree_repo_read_bare_fd()` was the `openat()` implementation;
they duplicated some bits to find the object in staging, recurse
into parent etc.
Further, I wanted an internal-only version of this API which didn't allocate
`GFileInfo`/`GInputStream` but used a plain `fd` and `struct stat` to avoid
mallocs.
The end version here I think looks a lot nicer, since we deduplicate the various
`open()` calls in the different cases for example.
Closes: #952
Approved by: jlebon
Before this, if one had repos of matching mode but different owners,
which could happen if one e.g. makes a `bare` non-root repo in
`/ostree/deploy/$stateroot/var/tmp`, every time we tried to call `linkat()`
we'd get `EPERM` and fall back to a copy.
Fix this by saving the repo owner uid, and avoid trying to call `linkat()` if we
know it's going to fail. Of course most commonly in this scenario we'll
immediately fail trying to `chown` the files to `0`, but this is prep for a
future patch to improve `bare-user` → `bare-user-only` imports where we'll be a
bit more sophisticated.
Closes: #922
Approved by: alexlarsson
This reverts commit 1eff3e8343. There
are a few issues with it. It's not a critical thing for now, so
let's ugly up the git history and revisit when we have time to
debug it and add more tests.
Besides the below issue, I noticed that the simple `ostree remote add`
now writes to `/ostree/repo/config` because we *aren't* using the
`--sysroot` argument.
Closes: https://github.com/ostreedev/ostree/issues/901Closes: #902
Approved by: mike-nguyen
Using `${sysroot}` to mean the physical storage root: We don't want to write to
`${sysroot}/etc/ostree/remotes.d`, since nothing will read it, and really
`${sysroot}` should just have `/ostree` (ideally). Today the Anaconda rpmostree
code ends up writing there. Fix this by adding a notion of "physical" sysroot.
We determine whether the path is physical by checking for `/sysroot`, which
exists in deployment roots (and there shouldn't be a `${sysroot}/sysroot`).
In order to unit test this, I added a `--sysroot` argument to `remote add`.
However, doing this better would require reworking the command line parsing for
the `remote` argument to support specifying `--repo` or `--sysroot`, and I
didn't quite want to do that yet in this patch.
Closes: https://github.com/ostreedev/ostree/issues/892Closes: #896
Approved by: jlebon
This is prep for introducing a fd-relative `ostree_repo_new_at()`.
Previously, `ostree_repo_is_system()` compared `GFile` paths, but
there's a much simpler check we can do first - if this repository
was created via `OstreeSysroot`, it must be a system repo.
Closes: #886
Approved by: jlebon
Make it an internal, not static, API; like _ostree_repo_add_remote(). It
will be used in many the same situations.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #875
Approved by: cgwalters
Return whether the remote already existed. This is an internal API, so
it’s not an API break. The return value will be useful in upcoming
commits for working out whether to later remove a remote again.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #875
Approved by: cgwalters
Previously it was static to ostree-repo.c. Make it usable throughout
libostree so it can be used by an upcoming commit, but also expose the
typedef and reference counting functions so that opaque OstreeRemote
pointers can be used by user code, in anticipation of exposing more of
its API publicly in future.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #832
Approved by: cgwalters
• Commit timestamps, so it’s easy to work out whether a given commit is
newer than the one we have locally
• Summary file timestamp, so it’s easy to work out whether the summary
file is more up to date than another summary file
• Summary file expiry time, so clients can work out when they should
expect the summary file to next be updated, and hence can query for
it at roughly the right time
The expiry time requires input from the user, so is currently never set
automatically. Programs using libostree can set it if they wish.
Signed-off-by: Philip Withnall <withnall@endlessm.com>
Closes: #826
Approved by: cgwalters
The keyring isn't large, so let's just fall back to copying it
rather than requiring `renameat()`.
Prep for `ostree_repo_open_at()`.
Closes: #821
Approved by: jlebon
I was reading a strace the other day and noticed we were loading the same
`.dirmeta` object many times. Unlike the other object types, `.dirmeta` objects
don't accumulate much over time; there are only so many directory metadata types.
(Without SELinux involved it'd probably be 5-6 I'd guess offhand).
For `fedora-atomic/25/x86_64/docker-host` there are currently 34 `.dirmeta` in
the tree.
But how many times during a checkout did we load those 34 dirmeta objects?
With a quick strace:
```
$ strace -s 2048 -f -o strace.log ostree --repo=repo-build checkout -U fedora-atomic/25/x86_64/docker-host host-test-checkout
$ grep dirmeta strace.log | wc -l
7165
```
After, as you'd expect, we just loaded `34` from disk. We do
6 system calls (`openat+fstat+fstat+read+read+close`) per dirmeta,
so we dropped a total of 42780 system calls - which is about 20% of the total
system calls made.
`perf record` tells me that we're spending ~40 of our time in the kernel during
a checkout, so reducing syscall traffic helps. Though most of that appears to be
in the VFS and XFS layers for `linkat` (which isn't surprising).
So how much did perf improve? Well, on my workstation, I get a lot of
fluctuation in timing, sometimes by 30%, so this was well within the noise. But
it's well worth speeding up checkout, and I think this optimization will shine
more as we improve performance elsewhere.
Closes: #795
Approved by: jlebon
These are leftovers from the packfile code and should have been
deleted in commit: 2a0601efc7
I noticed this now since I wanted to add a new type of caching.
Closes: #795
Approved by: jlebon
There are a lot of things suboptimal about this approach, but
on the other hand we need to get our CI back up and running.
The basic approach is to - in the test suite, detect if we're on overlayfs. If
so, set a flag in the repo, which gets picked up by a few strategic places in
the core to turn on "ignore xattrs".
I also had to add a variant of this for the sysroot work.
The core problem here is while overlayfs will let us read and
see the SELinux labels, it won't let us write them.
Down the line, we should improve this so that we can selectively ignore e.g.
`security.*` attributes but not `user.*` say.
Closes: https://github.com/ostreedev/ostree/issues/758Closes: #759
Approved by: jlebon
When fetching over a fast enough connection, we can be receiving files
faster than we write them. This can then lead to EMFILE when we have
enough files open. This was made very easy to notice with the upcoming
libcurl backend, which makes use of pipelining.
Closes: #675
Approved by: cgwalters
Working on the libcurl backend, I didn't want to reimplement another queue. I
think the queue logic is really better done at the high level, since the fetcher
knows how we want to prioritize metadata over content, etc.
Adding another queue here is duplication, but things will look nicer when we can
actually delete the libsoup one in the next commit.
Closes: #654
Approved by: jlebon
The gzip default is 6. When I was writing this code, I chose 9 under
the assumption that for long-term archival, the extra compression was
worth it.
Turns out level 9 is really, really not worth it. Here's run at level 9
compressing the current Fedora Atomic Host into archive:
```
ostree --repo=repo pull-local repo-build fedora-atomic/25/x86_64/docker-host
real 2m38.115s
user 2m31.210s
sys 0m3.114s
617M repo
```
And here's the new default level of 6:
```
ostree --repo=repo pull-local repo-build fedora-atomic/25/x86_64/docker-host
real 0m53.712s
user 0m43.727s
sys 0m3.601s
619M repo
619M total
```
As you can see, we run almost *three times* faster, and we take up *less
than one percent* more space.
Conclusion: Using level 9 is dumb. And here's a run at compression level 1:
```
ostree --repo=repo pull-local repo-build fedora-atomic/25/x86_64/docker-host
real 0m24.073s
user 0m17.574s
sys 0m2.636s
643M repo
643M total
```
I would argue actually many people would prefer even this for "devel" repos.
For production repos, you want static deltas anyways. (However, perhaps
we should support a model where generating a delta involves re-compressing
fallback objects with a bit stronger compression level).
Anyways, let's make everyone's life better and switch the default to 6.
Closes: #671
Approved by: jlebon
I was having this thought today about making more of the OS readonly,
and ultimately if we got to the point where all ostree operations are
through the repo and sysroot dfds, we could have rpm-ostree be the
only process holding those fds open, and have a read-only bind mount
on top.
Anyways, we're not there, likely won't be soon, but this gets us
closer to being fully fd relative.
Closes: #628
Approved by: jlebon
This way e.g. flatpak can detect which timestamp it should use.
See `flatpak/common/flatpak-utils.c:flatpak_zero_mtime`.
Closes: #501
Approved by: cgwalters
Quoting Dan Nicholson in
<https://github.com/ostreedev/ostree/pull/330#issuecomment-245499099>
mtime of 0 has been the semantics of ostree deployments from basically
the beginning of the project. We (and others, see
flatpak/flatpak@b5204c9) rely on that fact when generating trees.
In particular, this affects caches that use the mtime of the
associated file or directory to determine if the cache is valid. By
arbitrarily changing the mtime of the files to something else, all
the caches we setup in the build are now invalidated. Preseeding
caches is really important to the user experience as it avoids
having the user wait while they're regenerated on first run.
Now, we could change our build infrastructure to preset all the
mtimes to 1 to match this change, but what does that do for our
existing users who are on an ostree that deploys with mtimes of 0?
We could just revert this change at Endless (and the associated one
in Flatpak), and that would be fine for our users. However, if we
point non-Endless users to our apps, they'll have the great
experience of waiting 10 seconds the first time they launch it while
the fontconfig cache is rebuilt unnecessarily.
Closes: #495
Approved by: jlebon
I forgot to actually remove `config_file` in the previous
commit, the txn lock hasn't been used in a long time, and
for the uncompressed cache, everything uses the fd already.
Closes: #433
Approved by: giuseppe
In general this is even cleaner now, though it was better after I
extracted a helper function for the "write tempfile with contents"
bits that were shared between metadata and regular file codepaths.
Closes: #369
Approved by: jlebon
When reworking the ostree core [to use O_TMPFILE](https://github.com/ostreedev/ostree/pull/369),
I hit an issue in the way the untrusted delta codepath ends up trying
to re-open the file to checksum it. That's not possible with
`O_TMPFILE` since the fd (which we opened `O_WRONLY`) is the only
accessible reference to the content.
Fix this by changing the delta processing code to update a checksum as
we're doing writes, which is also faster, and ends up simplifying the
code as well.
What would be an even larger simplification here is if we e.g. used a
separate thread calling `write_object()` or something like that; the
main issue I see there is somehow bridging the fact that function
wants a `GInputStream*` but the delta code is generating stream of
writes.
Closes: #392
Approved by: jlebon
Dan Nicholson
Colin Walters
Matthew Leeds
Krzesimir Nowak
Philip Withnall
Jonathan Lebon