This is what caused the merge of
https://github.com/projectatomic/rpm-ostree/pull/652
to blow up, since https://github.com/ostreedev/ostree/pull/848
landed right before we tried to merge it.
When I was writing that PR I remember having an uncertain feeling
since we were doing a `mkdirat` above, but at the time I thought
we'd have test suite coverage...turns out we didn't.
For backwards compatibility, we need to continue to do a `mkdirat` here of the
parent. However...I can't think of a reason anyone would *want* that behavior.
Hence, let's add a special trick - if the destination name is `.`, we skip
`mkdirat()`. That way rpm-ostree for example can open a dfd for `/etc` and avoid
the `mkdir`.
Fold the subpath tests into `test-basic.sh` since it's not worth a separate
file. Add a test case for checking out a file.
Closes: #854
Approved by: jlebon
As $DEITY intended.
I was reading the `prepare-root.c` code and the indentation damage was
distracting. Squash tabs that have leaked into various places in the code. I
didn't yet touch the `src/libostree` bits as that has higher potential for
conflict.
Closes: #852
Approved by: jlebon
Looking at `perf record ostree checkout`, some things stand out; e.g.:
```
+ 27.63% 0.07% ostree libgio-2.0.so.0.5000.3 [.] g_file_enumerator_iterate
+ 22.74% 0.28% ostree libostree-1.so.1.0.0 [.] ostree_repo_file_tree_query_child
+ 13.74% 0.08% ostree libostree-1.so.1.0.0 [.] ot_variant_bsearch_str
```
The GIO abstractions are already fairly heavyweight, and `OstreeRepoFile` mallocs
a lot too.
Make things more efficient here by dropping the GIO bits for reading ostree data -
we just read from the variants directly and iterate over them. The end result
here is that according to perf we go from ~40% of our time in the kernel to
~70%, and things like `g_file_enumerator_iterate()` drop entirely out of the
hot set.
Closes: #848
Approved by: jlebon
Since we now have a cleaner separation of "toplevel checkout prep"
versus "recursive checkout", handle the special case of checking out
a single file at first rather than later.
Prep for future work in optimizing this function more.
Closes: #848
Approved by: jlebon
Only non-mechanical bit here was creating a local autoptr for a bit
where we'd previously done an unref for a struct member.
Closes: #847
Approved by: jlebon
Rather than `g_output_stream_splice()`, where the input is a regular
file.
See https://github.com/GNOME/libglnx/pull/44 for some more information.
I didn't try to measure the performance difference, but seeing the
read()/write() to/from userspace mixed in with the pointless `poll()` annoyed me
when reading strace.
As a bonus, we will again start using reflinks (if available) for `/etc`,
which is a regression from the https://github.com/ostreedev/ostree/pull/797
changes (which before used `glnx_file_copy_at()`).
Also, for the first time we'll use reflinks when doing commits from file-backed
content. This happens in `rpm-ostree compose tree` today for example.
Update submodule: libglnx
Closes: #817
Approved by: jlebon
Continuing to chip away at this. Using `g_file_enumerator_iterate()`
here helps notably.
I started on the much bigger `ostree_diff_dirs_with_options()` but
it's a lot messier - for later.
Closes: #844
Approved by: jlebon
I think tests/ should be just that, ci/ is separate. Also rename
the files to include "travis" since that's what we use them
for right now.
Closes: #843
Approved by: jlebon
We need our `make install` to override the ostree RPM, so do it all in one txn.
This sort of thing is where a more rigorous model like rdgo/gcontinuous use
becomes better, but we'll hack it with shell for now.
Closes: #824
Approved by: jlebon
These at the moment aren't in a container, and may need space. In the future
overlay2 will help here, we can more easily extend the rootfs.
Closes: #840
Approved by: jlebon
`test-pull-many.sh` is was just too slow to be a unit test. Generating
a bunch of files via shell is slow, the delta generation is slow, etc.
Every developer doesn't need to run it every time.
Somewhat address this by converting it into our installed test framework, which
moves it out of the developer fast paths. Another advantage to this is
that we can simply reuse the FAH tree content rather than synthesizing
new bits each time.
Closes: #840
Approved by: jlebon
The major reason to do this is that running tests *both* installed
and uninstalled in our CI is a mostly pointless waste of time.
Particularly given we have a few expensive tests.
We *do* have tests that only run uninstalled (since they require
the source code) like `test-symbols.sh`.
Hence, add `--enable-installed-tests=exclusive` to mean *only* do installed for
most tests.
We'll still have uninstalled coverage via the Travis/Debian configs, and we
could perhaps do another build with a subset of uninstalled tests, but I'm not
really concerned about it.
I'd like to do a renewed push for the InstalledTests model since
I feel it's just fundamentally better. (`g-d-t-r` kind of sucks,
but then so does the automake runner).
Also while we're here - fix the CI to use the correct context,
which started this mess.
Closes: #837
Approved by: dbnicholson
If we're freeing the segment, it's basically always better to use
`autoptr()`. Fewer lines, more reliable, etc.
Noticed an instance of this in the pull code while reviewing a different PR,
decided to do a grep for it and fix it tree wide.
Closes: #836
Approved by: pwithnall
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
This was part of the philosophy behind https://wiki.gnome.org/Initiatives/GnomeGoals/InstalledTests -
libraries like libostree don't need to replicate everything in unit tests, we
can use the tests from our dependencies directly too.
We'll also get API break coverage testing too.
Closes: #818
Approved by: jlebon
I happened to glance at the top of my most recent patch and
noticed that I used an `throw_errno()` function in a non-errno place.
I scanned the patch for other instances of this but didn't find one.
Closes: #811
Approved by: jlebon
I was planning to change some of the object loading code in the
future, so here's some porting.
Note that I rewrote `_ostree_repo_has_loose_object()` since it
used an error return across multiple functions.
Honestly I'm not sure about this `TEMP_FAILURE_RETRY()` business...
in reality we're going to end up with a ton of code linked in
process that doesn't do it. Unix sucks =( But I'm keeping
what was there out of consistency.
Closes: #809
Approved by: jlebon
This did a `closedir` in the `goto out` section before, but it
turns out more nicely if we follow the usual pattern of doing
the `open(O_DIRECTORY)` in the callee function and handle `ENOENT`
there.
Closes: #809
Approved by: jlebon
This is a variant of the efforts in https://github.com/ostreedev/ostree/pull/741
Working on `rpm-ostree livefs`, I realized though I needed to just
check out *new* files directly into the live `/etc` (and possibly
delete obsolete files).
The way the current `/etc` merge works is fundamentally different from
that. So my plan currently is to probably do something like:
- Compute diff
- Check out each *new* file individually (as a copy)
- Optionally delete obsolete files
Also, a few other things become more important - in the current deploy code, we
copy all of the files, then relabel them. But we shouldn't expose to *live*
systems the race conditions of doing that, plus we should only relabel files we
checked out.
By converting the deploy's /etc code to use this, we fix the same TODO item
there around atomically having the label set up as we create files. And further,
if we kill the `/var` relabeling which I think is unnecessary since Anaconda
does it, we could delete large chunks of code there.
In the implementation, there are two types of things: regular files, and
symlinks. For regular files, in the `O_TMPFILE` case, we have the ability to
do *everything* atomically (including SELinux labeling) before linking it into
place. So let's just use that. For symlinks, we use `setfscreatecon()`.
Closes: #797
Approved by: jlebon
Our container-driven tests can't e.g. test SELinux sanely, and
have to support being run as root *and* non-root too.
Use redhat-ci to provision a VM and run tests directly there. These are
installed tests too.
Closes: https://github.com/ostreedev/ostree/issues/806Closes: #807
Approved by: jlebon
This could be shared more easily with e.g. rpm-ostree, but what I'm currently
working on is installed, privileged (potentially destructive, i.e. VM) tests
that will source this separately from the current `libtest.sh`. That does work
installed, but in practice is oriented around unit (uninstalled, unprivileged)
tests.
Closes: #807
Approved by: jlebon
We really have an astonishing variety of similar functions which write files and
symlinks. I was working on a different PR and the duplication between the
union-mode and add-mode/none-mode checkout functions bothered me.
I realized that the "handle EEXIST" tri-state maps directly to the
`GLnxLinkTmpfileReplaceMode`, so deduping things makes even more sense.
Closes: #801
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
I never really liked the term "osname". I feel "stateroot" is a *lot* clearer,
since the osname/stateroot mostly just holds `/var`. Further it avoids the `os`
prefix which is already overloaded.
Some of the existing docs already talked about "operating system state", which
further reinforces this.
There's *lot* more things than this which reference the term "osname", but I
don't want to change *everything* yet in this patch in case we decide to do
something different - this just gets the highlights.
Closes: #794
Approved by: jlebon
This is intended to be used for copying `/usr/etc` → `/etc` for
deployments.
A TODO here is to use `glnx_file_copy_at()` if the repo mode allows
it - then we'd use reflinks if available.
Closes: #804
Approved by: jlebon
There aren't many users of `g_file_enumerator_iterate()` left - those
remaining are usually good candidates for porting. There's some more
porting to do in this file; a mix of trivial and harder. This
one is a good candidate for an individual commit.
Closes: #803
Approved by: jlebon
I plan to make some future changes here, and wanted to do this
first.
Random side note; how about converting the do/while loops for `EINTR` to
`TEMP_FAILURE_RETRY()`? We're very inconsistent about that...
Closes: #792
Approved by: jlebon
`perf record ostree checkout ...` for a bare-user repo was telling
me we were spending a good 13% of our time in the depchain of `ot_lgexattrat()`.
The problem here is that traversing the `/proc` path turns out to be
somewhat expensive - there are LSM (SELinux) checks, etc.
For regular files, opening and just getting the xattr, then closing is still
quite cheap. For symlinks, we'll always need to open anyways.
This appears to shave about ~0.1 seconds off of a checkout of
`fedora-atomic/25/x86_64/docker-host` on my workstation.
Oh, and this was the last user of `ot_lgexattrat()` so we can kill it, which is
nice - the xattr code should really live in libglnx.
Closes: #796
Approved by: jlebon
This is only about 40%, and mostly simpler functions. It's
nice to switch to `g_autoptr(GMatchInfo)` instead of our inline version.
I decided to add more usage of `ot_transfer_out_value()`, though it'd
be nice to try to have a copy of that in libglnx (or possibly glib).
Closes: #791
Approved by: jlebon
This commit won't actually *be* 2017.5 since due to the way our infrastructure
works, we still want to increment git master to 2017.5.
See https://github.com/ostreedev/ostree/pull/800Closes: #800
Approved by: jlebon
Logic error introduced after refactoring; we hoisted the
`is_bare_user_symlink` variable to the top, but its computation
below. But the `is_bare` symlink depended on it.
Closes: https://github.com/ostreedev/ostree/issues/798Closes: #799
Approved by: jlebon
I think the majority of OSTree usage calls pull with refs, not
explicit commits. We even added special "override syntax" with
`@` (e.g. `ostree pull foo@ab12c34`) as a hybrid.
However, some users may want to still pull explicit commits
for whatever reason. The old static delta logic looked at
the previous commit of the ref. However, in https://github.com/ostreedev/ostree/pull/710
we enhanced the logic to look at all local commits.
It's now a lot more natural to teach the delta logic
to support revisions, e.g. `ostree pull someorigin ab12c34`.
This also fixes the problem that before, `--require-static-deltas`
was completely ignored when processing revisions.
This is a nontrivial refactoring of the logic, but the end
result feels a lot more readable to me.
Closes: https://github.com/ostreedev/ostree/issues/783Closes: #787
Approved by: cgwalters
Previously we'd assert and dump core if one used `checkout -H` without
`-U` on a bare-user repo, because we'd hit the bare-user symlink case.
Rework the code to handle this, and add tests. I hit this when I was going to
suggest to someone to use `-H` to ensure they were getting hardlinks.
Closes: #779
Approved by: jlebon
Testing a fetch of `fedora-atomic/.../docker-host` from
an nginx instance over `https://127.0.0.1` using Fedora 25
versions. Average over 3 runs:
Before: ~24.6 seconds
After: ~19 seconds
Speedup: ~30%
Closes: https://github.com/ostreedev/ostree/issues/778Closes: #780
Approved by: jlebon
In particular the 26-variable monster 👹 in `install_deployment_kernel()` is
slain🗡. I didn't touch every function here, trying to keep things gradual.
Closes: #781
Approved by: jlebon
It was reported that in the range request handling, we called `remove_pending()`
twice (once in processing it, and once potentially in the local_error cleanup),
and this could be viewed as a use-after-free. However, right now the range
cleanup and `local_error` being set are mututally exclusive.
Further, the task object already holds a strong reference, so I observed the
refcount was 2. For both of these reasons, there is no use-after-free in
practice.
Reported-By: "Siddharth Sharma" <siddharth@redhat.com>
Closes: #774
Approved by: jlebon
I was planning to change one here, decided to do a conversion
of some of the simpler functions in this file to keep up momentum.
Closes: #776
Approved by: jlebon
I was working on `rpm-ostree livefs` which does some ostree-based
filesystem diffs, and noticed that we were ending up with `/proc`
not being labeled in our base trees.
Reading the selinux-policy source, indeed we have:
```
/proc -d <<none>>
/proc/.* <<none>>
```
This dates pretty far back. We really don't want unlabeled
content in ostree. In this case it's mostly OK since the kernel
will assign a label, but again *everything* should be labeled via
OSTree so that it's all consistent, which will fix `ostree diff`.
Notably, `/proc` is the *only* file path that isn't covered when composing a
Fedora Atomic Host. So I added a hack here to hardcode it (although I'm a bit
uncertain about whether it should really be `proc_t` on disk before systemd
mounts or not).
Out of conservatism, I made this a flag, so if we hit issues down the line, we
could easily change rpm-ostree to stumble on as it did before.
Closes: #768
Approved by: jlebon
Being bitten by lack of PR testing here. There are two bugs:
- First and foremost, I forgot that GObject will call the property setters with
the defaults. This meant we were getting both path="/var/tmp/blah" and fd=-1,
and we were accepting -1 as a fd, which then got converted into AT_FDCWD
which was wrong.
- Since these properties are construct only and mutually exclusive, don't
try to handle one resetting the other. Assert that exactly one of them is set.
Closes: #769
Approved by: jlebon
This is inspired by the [Coccinelle](http://coccinelle.lip6.fr/) usage
in systemd. I also took it a bit further and added infrastructure
to have spatches which should never apply. This acts as a blacklist.
The reason to do the latter is that coccinelle is *way* more powerful than the
regular expresssions we have in `make syntax-check`.
I started with blacklisting `g_error_free()` directly. The reason that's bad is
it leaves a dangling pointer.
Closes: #754
Approved by: jlebon
I happened to read this file and realized there's a lot of cruft left over from
the time when I liked `GFile` and `malloc()`ing like 50 times just to make a
pathname string. Delete it.
Closes: #767
Approved by: jlebon
Actually trying to use this in rpm-ostree, it kept returning successfully when I
didn't expect it to... The first conditional was always succeeding even when I
was asking for a newer minor.
Closes: #766
Approved by: jlebon
I'd like to do this in rpm-ostree at least. Originally I was looking at porting
to `G_DECLARE_FINAL_TYPE` but eh, this is easier for now and won't bump our GLib
dependency which might matter for our embedded users.
For now I just did a few replacements in the `remote` command line. A full port
can come as we do other code cleanups.
This will actually break the flatpak build right now, but
that's easy to fix. And we concluded in e.g.
https://bugs.freedesktop.org/show_bug.cgi?id=95065#c5
it's a bug for downstream projects to do that.
Closes: #756
Approved by: jlebon
I didn't touch everything since at least `commit_loose_object_trusted`
does this:
```
out:
if (G_UNLIKELY (error && *error))
g_prefix_error (error, "Writing object %s.%s: ", checksum, ostree_object_type_to_string (objtype));
```
Which...it'd be interesting to make into an autocleanup. But for now just
keeping up with converting things bit by bit.
Closes: #761
Approved by: jlebon
In [this commit](https://github.com/ostreedev/ostree/commit/6ce80f9685e6273dbcb4731d6650a10976ea220a)
for some reason I added a `sepolicy` member to the sysroot. I
have no idea why I did that, and it's conceptually wrong
since the policy is specific to a *deployment*.
This bit me when I was working on [a pull request](https://github.com/ostreedev/ostree/pull/763)
elsewhere, since at that point it was `NULL`.
We already pass around the sepolicy in the deployment code, so just stop caching
it.
Closes: #764
Approved by: jlebon
(Also rename the other CI contexts to be more consistent)
We pass this right now. I just noticed an instance of this in bwrap, and I think
we should be trying to match the RPM build baseline.
Closes: #765
Approved by: jlebon
I was playing around in a FAH vagrant box, and hit:
```
Receiving delta parts: 3/4 453.2 kB/s 1.8 MB/145.8 MB
error: opcode set-read-source: No such file object b6e54ba3471b9c116ce6b9bfbf9e55fec60d35cfdb9ae5ae1ee219af02a591b7
```
This is because this host version doesn't yet have
https://github.com/ostreedev/ostree/pull/710
which incidentally fixed this for the case where the OS vendor is using
summary files.
Some organizations may not be using summary files - at least we still try to
support that case. So let's copy the logic very recently added in that commit to
handle the legacy case too.
No new tests since this is a nice-to-have - we really do
expect people to be using summary files now.
Closes: #739
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
More sophisticated users of libostree like rpm-ostree need control over things
like the system repository. Previously we introduced a "no cleanup" flag to
`ostree_sysroot_simple_write_deployment()`, but that's a high level API that
does filtering on its own.
Since rpm-ostree needs more control, let's expose the bare essentials of the
"sysroot commit" operation with an extensible options structure, where one of
the options is whether or not to do post-transaction repository operations.
Closes: #745
Approved by: jlebon
I happened to be reading this one recently for a reason I forget,
and it's a relatively easy conversion.
Also one not conflicting with any outstanding patches.
Closes: #752
Approved by: jlebon
Use `g_auto()` more sanely with a struct implmenting the "is initialized"
pattern. This is way less ugly for callers, and fixes bugs like
us calling `setfscreatecon()` even if an error occurred beforehand.
Also fold in the logic for "NULL or not loaded" sepolicy into the setup rather
than requiring callers to inline it.
Prep for more users of this function.
Closes: #746
Approved by: jlebon
I'm porting other code away from `GFile`, and while we don't use this
internally, it will let us do so at a later date. I'm averse to changing the
code right now as we don't have good CI coverage of this.
Closes: #746
Approved by: jlebon
This actually worked before because `pkg-config --cflags openssl`
is empty, and the linker was satisfying `-lssl -lcrypto` from our
indirect dependencies.
Also, in fact we *currently* just want `pkg-config libcrypto` i.e.
`-lcrypto`, since we aren't actually using openssl to speak TLS.
This doesn't actually matter on Fedora at least since they're both in the
`openssl-libs` package, but they are separate for a reason.
Closes: #749
Approved by: jlebon
For future work I'm going to tweak how we handle cleanup, and
the private cleanup flags didn't really end up being used - we
only specify "prune repo or not". So fold that into a boolean for now.
The sysroot deploy logic then has a single "do_postclean" boolean, which is all
I want to expose as public API.
Closes: #744
Approved by: jlebon
Add an OpenSSL backend to the checksum input stream, which is where we do a lot
of checksumming (object commit, static deltas).
The raw OpenSSL performance is
[approximately double](https://gist.github.com/cgwalters/169349fd1c06fd4fb4d3a7ce33303222) on
my laptop; not only does OpenSSL have e.g. hand-tuned x86_64 assembly, the
current implementation uses the
[Intel SHA extensions](https://en.wikipedia.org/wiki/Intel_SHA_extensions).
Another reason to do this is I was idly thinking about adding
[Curve25519](https://en.wikipedia.org/wiki/Curve25519) signatures (like e.g.
Alpine does) instead of/in addition to GPG. The rationale for that is
that GPG is pretty heavyweight, both in code footprint and the simple
fact that EC keys are way smaller.
I didn't benchmark ostree with this; we have bigger performance problems
really like the fact we just malloc way too much. But, it's a step
in the right direction I think in combination with the libcurl work
where we're linking to openssl anyways.
Closes: #738
Approved by: jlebon
The current `OstreeChecksumInputStream` is public due to a historical
mistake. I'd like to add an OpenSSL checksum backend, but that's
harder without breaking this API.
Let's ignore it and create a new private version, so it's easier to do the
GLib/OpenSSL abstraction in one place.
Closes: #738
Approved by: jlebon
We should fix this more reliably by defining a separate variable for
introspection (and gtk-doc) sources at some point, but this does the trick for
now.
Avoids a spam of warnings, and it's definitely not public API.
Closes: #734
Approved by: jlebon
I just happened to be reading this code; still testing
the waters for the new code style. If we decide to do more
of this it's likely worth trying out a [Coccinelle](http://coccinelle.lip6.fr/)
patch or something.
Closes: #733
Approved by: jlebon
[Previously](https://github.com/ostreedev/ostree/pull/728) we added compile-time
checking for versions, but there are use cases for runtime checking as well,
because in a number of API calls we use `GVariant` as an API extension
mechanism.
Closes: #735
Approved by: jlebon
This goes farther along the path of deprecating it; consumers should
have been ported at this point. Though probably a lot of people
may still use `rpm-ostree-toolbox`.
Closes: #723
Approved by: jlebon
It's been almost a month, I think the current git is working well and
not too risky. We have some new API additions which I think the
various consumers of them are going to want.
Closes: #726
Approved by: jlebon
At least in all Linux kernels up to today, one can never `link()` across
devices, so we might as well verify that up front. This will help for a future
patch to add a new type of union-add checkout, since Linux checks for `EEXIST`
before `EXDEV`.
Closes: #714
Approved by: jlebon
I've seen code in a few places that I think on balance is definitely better this
way. Some of our functions have huge variable declaration sections.
This change includes one small example where we could start using declarations
after statements.
A concern I had was - how does this interact with `__attribute__((cleanup))` and
early returns? I tested it, and AFAICS the behavior is what you'd expect - the
cleanup function isn't called if its variable isn't reachable.
Closes: #718
Approved by: jlebon
Colin Walters