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
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
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
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
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
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
This mode is similar to bare-user, but does not store the permission,
ownership (uid/gid) and xattrs in an xattr on the file objects in the
repo. Additionally it stores symlinks as symlinks rather than as
regular files+xattrs, like the bare mode. The later is needed because
we can't store the is-symlink in the xattr.
This means that some metadata is lost, such as the uid. When reading a
repo like this we always report uid, gid as 0, and no xattrs, so
unless this is true in the commit the resulting repository will
not fsck correctly.
However, it the main usecase of the repository is to check out with
--user-mode, then no information is lost, and the repository can
work on filesystems without xattrs (such as tmpfs).
Closes: #750
Approved by: cgwalters
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
Commit 1d4f1b8878 started using hardlinks
checkouts of symlinks. However, symlinks are not stored as symlink in the
repo for bare-user repos, so this breaks user-mode checkouts of such repos.
We fix this by checking for !is_symlink in the bare-user case.
This fixes:
https://github.com/ostreedev/ostree/issues/537Closes: #538
Approved by: giuseppe
I was noticing a recent performance issue with checkouts
which seemed to be mostly us going back to doing a `fsync()` on
directories.
Regardless, while looking at that, I saw we were spending time
creating new symlinks. Even though symlinks are small, it's still
better to hardlink them.
Going way back in time, the reason we weren't doing this is
because we were hitting `EMFILE` on ext4, but that was for
gnome-continuous which creates *many* build roots. Even
there though, they're just a cache, and we handle `EMFILE`.
For ostree-for-host-system, we don't expect to have many roots (just 3
at most transiently), so hardlinking symlinks does make sense.
Closes: #521
Approved by: jlebon
My previous change in https://github.com/ostreedev/ostree/pull/425
actually broke things so we basically used the repository defaults =(
This is a subtle mess since we're only trying to flip things off
for the *new* API.
Clean this up so that the "default repo inheritance" lives only in one
place - in the compat layer for the old checkout API. The new
checkout API defaults to off period, so the repository state is
irrelevant.
Closes: #520
Approved by: jlebon
Since we're adding a new API, we have the opportunity to fix
the defaults. We expect clients to do a `syncfs()` or equivalent
on their own now, since it's way more efficient.
Flip the checkout fsync default to off.
Closes: #425
Approved by: giuseppe
Provide a gobject introspection safe version for
`ostree_repo_checkout_tree_at'.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Closes: #417
Approved by: cgwalters
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
I've seen a few people hit this and wonder why checkouts are slow/take
space. Really, ensuring this happens is the *point* of OSTree.
Physical copies should be a last resort fallback for very unusual
situations (one of those is rpm-ostree checking out the db since
librpm doesn't know how to read from libostree).
Even I hit the fact that `/var` is a mountpoint disallowing hardlinks
with `/ostree` once and was confused. =)
Add this to the rofiles-fuse test case because it creates a mount
point.
Closes: #368
Approved by: jlebon
Import `gs_file_enumerator_iterate()` for the next six months or
so...after RHEL 7.3 is released I'm strongly considering hard
requiring 2.46 or so.
Likely at some point we should figure out how to share more "glib
backport" code with NetworkManager at least.
Closes: #341
Approved by: jlebon
Just noticed this while reading some code, we didn't have many manual
`out: close()` bits left, this pushes us over the edge to autocleanup
almost everywhere.
Closes: #332
Approved by: jlebon
1 is a better choice than 0 because some programs use 0
as a special value; for example, GNU Tar warns of an
"implausibly old timestamp" with 0.
Closes: #330
Approved by: cgwalters
⚠️ There is a notable spiked pit trap here around
`posix_fallocate()` and `errno`. This has bit other projects,
see e.g.
7bb87460e6
Otherwise the port was straightforward.
This is a better followup to dc9239dd7b
since I wanted to do fsync-less checkouts in rpm-ostree too, and
replicating the "turn off fsync temporarily" was in retrospect just a
hack.
We can simply add a boolean to the checkout options.
https://github.com/GNOME/ostree/pull/172
A fast way to generate new OSTree content using an existing
tree is to checkout (as hard links), add/replace files, then
call `ostree_repo_scan_hardlinks()`, then commit.
But `ostree_repo_scan_hardlinks()` scans the entire repo, which
can be slow if you have a lot of content.
All we really need is a mapping of (device,inode) -> checksum
just for the objects we checked out, then use that mapping
for commits.
This patch adds API so that callers can create a mapping via
`ostree_repo_devino_cache_new()`, then pass it to
`ostree_repo_checkout_tree_at()` which will populate it, and then
`ostree_repo_write_directory_to_mtree()` can consume it.
I plan to use this in rpm-ostree for package layering work.
Notes:
- The old `ostree_repo_scan_hardlinks()` API still works.
- I tweaked the cache to be a set with the checksum colocated with
the key, to avoid a separate malloc block per entry.
https://github.com/GNOME/ostree/pull/167
rpm-ostree currently uses ostree_repo_checkout_tree(), which as a side
effect will use the uncompressed objects cache by default. This is
rather annoying if you're using rpm-ostree on a server-side
repository, because if you then rsync the repo, you'll be syncing out
the uncompressed objects unless you exclude them.
We added the ability to disable the uncompressed cache in the
repository config to fix this, but it's better to allow application
control over this. The uncompressed cache will in some future version
become opt in as well.
This new API further:
- Drops the `GFile` usage in favor of `openat` APIs
- Improves ergonomics by avoiding callers having to query the source
`GFileInfo` (and carry around a copy of `OSTREE_GIO_FAST_QUERYINFO`)
- Has a more extensible options structure
Per the comment, I rather crudely have the `ostree checkout` builtin
call both APIs to ensure some testing coverage.
However, I'd like to in the future have easier-to-set-up testing code
that calls `libtest.sh` to set up dummy data.
Checkout was one of the first complex code paths I tried to convert to
*at(). I ended up keeping both, because I hit the "xattrs for a
symlink" problem. Later, Florian gave me a workaround, and we started
using it here, but the GFile * parameters weren't deleted. They're
not used, so do so now.
We already set all file mtimes to 0 so that they are constant
over all checkouts, and can be made constant with a known value from
the system where the ostree was created.
However, this was not happening for directories. Zero their mtimes too.
This is important for shipping a fontconfig cache in the ostree;
the fontconfig cache files embed a directory mtime.
Some use cases for checkouts don't need to fsync during checkout.
Installer programs for example will just do a global fsync at the end.
In the future, the default "ostree admin" core could also be
rearchitected to only do a transaction commit right before reboot, and
do the fsync then.
https://bugzilla.gnome.org/show_bug.cgi?id=742482
This format is pretty much the same as the "bare" format, except the
file ownership and xattrs is not stored in the actual filesystem object, but
rather on the side in a user xattr. This means two things:
1) An unprivileged user can store such a repo independent of the types
of files in it or their xattrs. And you can later (as root)
reconstruct the real filesystem tree with ownership. Although you
can't do that using hardlink-sharing. This also means ostree
fsck does a full verification.
2) Such a repository can be checked out with user-mode (checkout -U)
as an unprivileged user using hardlinks for space sharing.
Additionally, symlinks are stored as regular files (with the content
being the symlink target) because user xattrs are not supported on
symlinks. We know at checkout time if the file is a symlink because
the original st_mode is stored in the xattr metadata.
https://bugzilla.gnome.org/show_bug.cgi?id=741125
Applying xattrs on a symlink during checkout failed since
it was setting the xattrs on the final filename, not the
temporary name.
This made the "checkout union 1" test in test-basic.sh
fail.
https://bugzilla.gnome.org/show_bug.cgi?id=741125
When commiting a symlink we do store the uid/gid of the actual
symlink (i.e. not target). However, this was not restored
on non-user-mode checkout as it should.
This commit fixes that, and additionally it ensures xattrs
on symlinks are not set in user-mode checkout.
https://bugzilla.gnome.org/show_bug.cgi?id=741125
fixes a coredump when using a command like:
$ ostree --repo=repo checkout -U --subpath=/usr/lib/passwd \
fedora-atomic/rawhide/x86_64/docker-host usrlib-new
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
While we did support disabling the uncompressed-objects-cache
per-repository:
1) We didn't actually respect that operation when doing
CHECKOUT_MODE_USER on archive-z2 repositories
2) It'd be better to automatically detect we can't write to the
repo and disable the uncompressed cache then.
Finally, fsync to ensure all entries are on disk, unless disabled.
We support disabling this for cases like server-side buildroot
construction where we don't need to be robust against power loss
The previous commit here changed things so that we do mkdir(x, 0700),
then fchmod later only if we created the directory.
However the logic was incorrect; we still need to chmod even in
MODE_USER if we created the directory.
This is a bit more efficient in that we're not walking full paths, and
it helps avoid security/reliability issues if an attacker (or just a
misbehaving process) has the ability to mutate paths in the middle.
This large patch moves the core xattr logic down into libgsystem,
which allows the gs_shutil_cp_a() API to copy them. In turn, this
allows us to just use that API instead of rolling our own recursive
copy here.
As noted in the new comment though, one case that we are explicitly
regressing is where the new /etc removes a parent directory that's
needed by a modified file. This seems unlikely for most vendors now,
but let's do that as a separate bug.
https://bugzilla.gnome.org/show_bug.cgi?id=711058
For the cases where we can't hardlink, use at-relative walking of the
path where possible. We still don't have lsetxattrat, so we also need
to deal with pathnames, but that is now only for symlinks.
Again, the advantages of this are a lot less malloc() of pathnames in
ostree, and much less time spent traversing paths inside the kernel.
https://bugzilla.gnome.org/show_bug.cgi?id=707733
Nothing external uses it. We keep ostree_get_xattrs_for_file() public
because it's convenient for external consumers to get xattrs in
exactly the format we desire.
https://bugzilla.gnome.org/show_bug.cgi?id=707733
Clean up how we deal with the uncompressed object cache; we now use
openat()/linkat() and such just like we do for the main objects/.
Use linkat() between the objects and the destination, if possible.
https://bugzilla.gnome.org/show_bug.cgi?id=707733
This avoids repeatedly traversing the target pathname, and is just
more efficient.
Part of a prelude to using fd-relative API for the source object path
too.
The way we recurse into subdirectories in parallel makes it far too
easy to hit up against the arbitrary Linux fd limit of 1024.
Since the fix here is about dropping parallelism, let's just go all
the way for now and make a plain old synchronous API =(
This does simplify both internal callers which wanted a sync API
anyways.
https://bugzilla.gnome.org/show_bug.cgi?id=706380
We'll always have "bare" mode for keeping files-as-hardlinks as root.
But "archive" was my second attempt at a format for non-root file
storage, used by the gnome-ostree buildsystem which runs as non-root.
It was really handy to have a "tar" like mode where I can create
tarballs as a user, that contain files owned by root for example.
The "archive" mode stored content files as two pieces in the
filesystem; ".file" contained metadata, and ".filecontent" was the
actual content, uncompressed. The nice thing about this was that to
check out a tree as non-root, you could just hardlink into the repo.
However, archive was fairly bad for serving via HTTP; it required
*two* HTTP requests per content object, greatly magnifing the already
inefficient fetch process. So "archive-z2" was introduced.
To allow gnome-ostree to still check out trees as a user, the
"uncompressed-object-cache" was introduced, and that's how things have
been working for a while.
So we should just be able to kill this code. Specifically note just
how much better the stage_object() function became.
https://bugzilla.gnome.org/show_bug.cgi?id=706057
Colin Walters
Alexander Larsson
Christian Hergert
Giuseppe Scrivano
Mathnerd314
Matthew Barnes
Daniel Svensson
Daniel Drake
Vadim Rutkovsky
Tobias Hunger