Fixes#3898
The entire `BinaryCaches` row used to get replaced after it became
stale according to the `timestamp` column. In a concurrent scenario,
this leads to foreign key conflicts as different instances of the
in-process `state.caches` cache now differ, with the consequence that
the older process still tries to use the `id` number of the old record.
Furthermore, this phenomenon appears to have caused the cache for
actual narinfos to be erased about every week, while the default
ttl for narinfos was supposed to be 30 days.
This is slightly more accurate considering that an outdated record
may exist in the persistent cache. Possibly-outdated records are
quite relevant as they may be foreign keys to more recent information
that we want to keep, but we will not return them here.
In unprivileged podman containers, /proc is not fully visible (there
are other filesystems mounted on subdirectories of /proc). Therefore
we can't mount a new /proc in the sandbox that matches the PID
namespace of the sandbox. So this commit automatically disables
sandboxing if /proc is not fully visible.
This didn't work because sandboxing doesn't work in Docker. However,
the sandboxing check is done lazily - after clone(CLONE_NEWNS) fails,
we retry with sandboxing disabled. But at that point, we've already
done UID allocation under the assumption that user namespaces are
enabled.
So let's get rid of the "goto fallback" logic and just detect early
whether user / mount namespaces are enabled.
This commit also gets rid of a compatibility hack for some ancient
Linux kernels (<2.13).
Previously we would completely refetch the submodules from the
network, even though the repo might already have them. Now we copy the
.git/modules directory from the repo as an optimisation. This speeds
up evaluating
builtins.fetchTree { type = "git"; url = "/path/to/blender"; submodules = true; }
(where /path/to/blender already has the needed submodules) from 121s
to 57s.
This is still pretty inefficient and a hack, but a better solution is
best done on the lazy-trees branch.
This change also help in the case where the repo already has the
submodules but the origin is unfetchable for whatever reason
(e.g. there have been cases where Nix in a GitHub action doesn't have
the right authentication set up).
We cannot use 'actualUrl', because for file:// repos that's not the
original URL that the repo was fetched from. This is a problem since
submodules may be relative to the original URL.
Fixes e.g.
nix eval --impure --json --expr 'builtins.fetchTree { type = "git"; url = "/path/to/blender"; submodules = true; }'
where /path/to/blender is a clone of
https://github.com/blender/blender.git (which has several relative
submodules like '../blender-addons.git').
Previously, build-remote would show a warning if all build slots were
taken, even if they would open up later. This caused a lot of spam in
the logs. Disable this warning when maxJobs > 0.
See #6263
Per the old FIXME, this flag was on too many commands, and mostly
ignored. Now it is just on the commands where it actually has an effect.
Per https://github.com/NixOS/nix/issues/7261, I would still like to get
rid of it entirely, but that is a separate project. This change should
be good with or without doing that.
`nix app` had something called `InstallableDerivedPath` which is
actually the same thing. We go with the later's name because it has
become more correct.
I originally did this change (more hurriedly) as part of #6225 --- a
mini store-only Nix and a full Nix need to share this code. In the first
RFC meeting for https://github.com/NixOS/rfcs/pull/134 we discussed how
some splitting of the massive `installables.cc` could begin prior, as
that is a good thing anyways. (@edolstra's words, not mine!) This would
be one such PR.
tl;dr: With this 1 line change I was able to get a speedup of 1.5x on 1Gbit/s
wan connections by enabling zstd compression in nginx.
Also nix already supported all common compression format for http
transfer, webservers usually only enable them if they are advertised
through the Accept-Encoding header.
This pull requests makes nix advertises content compression support for
zstd, br, gzip and deflate.
It's particular useful to add transparent compression for binary caches
that serve packages from the host nix store in particular nix-serve,
nix-serve-ng and harmonia.
I tried so far gzip, brotli and zstd, whereas only zstd was able to bring
me performance improvements for 1Gbit/s WAN connections.
The following nginx configuration was used in combination with the
[zstd module](https://github.com/tokers/zstd-nginx-module) and
[harmonia](https://github.com/nix-community/harmonia/)
```nix
{
services.nginx.virtualHosts."cache.yourhost.com" = {
locations."/".extraConfig = ''
proxy_pass http://127.0.0.1:5000;
proxy_set_header Host $host;
proxy_redirect http:// https://;
proxy_http_version 1.1;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
zstd on;
zstd_types application/x-nix-archive;
'';
};
}
```
For testing I unpacked a linux kernel tarball to the nix store using
this command `nix-prefetch-url --unpack https://cdn.kernel.org/pub/linux/kernel/v6.x/linux-6.1.8.tar.gz`.
Before:
```console
$ nix build && rm -rf /tmp/hello && time ./result/bin/nix copy --no-check-sigs --from https://cache.thalheim.io --to 'file:///tmp/hello?compression=none' '/nix/store/j42mahch5f0jvfmayhzwbb88sw36fvah-linux-6.1.8.tar.gz'
warning: Git tree '/scratch/joerg/nix' is dirty
real 0m18,375s
user 0m2,889s
sys 0m1,558s
```
After:
```console
$ nix build && rm -rf /tmp/hello && time ./result/bin/nix copy --no-check-sigs --from https://cache.thalheim.io --to 'file:///tmp/hello?compression=none' '/nix/store/j42mahch5f0jvfmayhzwb
b88sw36fvah-linux-6.1.8.tar.gz'
real 0m11,884s
user 0m4,130s
sys 0m1,439s
```
Signed-off-by: Jörg Thalheim <joerg@thalheim.io>
Update src/libstore/filetransfer.cc
Co-authored-by: Théophane Hufschmitt <7226587+thufschmitt@users.noreply.github.com>
These settings are not needed for libstore at all, they are just used by
the nix daemon *command* for authorization on unix domain sockets. My
moving them to a new configuration struct just in that file, we avoid
them leaking anywhere else.
Also, it is good to break up the mammoth `Settings` struct in general.
Issue #5638 tracks this.
The message is not changed because I do not want to regress in
convenience to the user. Just saying "this connection is not trusted"
doesn't tell them out to fix the issue. The ideal thing to do would be
to somehow parameterize `processCommand` on how the error should be
displayed, so different sorts of connections can display different
information to the user based on how authentication is performed for the
connection in question. This, however, is a good bit more work, so it is
left for the future.
This came up with me thinking about the tcp:// store (#5265). The larger
project is not TCP *per se*, but the idea that it should be possible for
something else to manage access control to services like the Nix Daemon,
and those services simply trust or trust the incoming connection as they
are told. This is a more capability-oriented way of thinking about trust
than "every server implements its own auth separately" as we are used to today.
Its very great that libstore itself already implements just this model,
and so via this refactor I basically want to "enshrine" that so it
continues to be the case.
With the switch to C++20, the rules became more strict, and we can no
longer initialize base classes. Make them comments instead.
(BTW
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p2287r1.html
this offers some new syntax for this use-case. Hopefully this will be
adopted and we can eventually use it.)
I don't think the `narHash` is in need of documentation more than the
other undocumented fields, but regardless this change has nothing to do
with that field and so we should leave the comment as is.
`&` without space before is far more common on this codebase than I
thought, so it is not worth changing just this one file. Maybe we will
adopt a formatter someday but until then this is fine.
For frameworks it's important that structures are as lazy as possible
to prevent infinite recursions, performance issues and errors that
aren't related to the thing to evaluate. As a consequence, they have
to emit more attributes than strictly (sic) necessary.
However, these attributes with empty values are not useful to the user
so we omit them.
The references set seems to have been unused since `LegacySSHStore`
references were first created in
caa5793b4a.
The method decls never were upstream, and accidentally added by me in
062533f7cd (probably due to `git rerere`).
Sorry!
This reduces the diff from #3746.
No other getDefaultFlakeAttrPaths implementation has this trailing dot,
and the dot can show up in error messages like:
error: flake '...' does not provide attribute 'packages.x86_64-linux.', ...
- Clarify doc comments, Installables::getCursors returns non-empty
vector
- Use vector::at in Installable::getCursor instead of checking for empty
vector and throwing an exception with error message.
Handle the case where none of getActualAttrPaths() actually exists,
in which case instead of returning an empty vector.
This fixes the case where the user misspells the attribute name in nix
search. Instead of getting no search results, now it shows an error with
suggestions.
Also remove InstallableFlake::getCursor() override since it's now
equivalent to the base class version.
Avoid needless work and throwing away invariants.
These conversions date back to when `StorePath` was in Rust and there
were issues with it missing utility methods.
Previously, getDefaultNixPath was called too early: at initialisation
time, before CLI and config have been processed, when `restrictEval` and
`pureEval` both have their default value `false`. Call it when
initialising the EvalState instead, and use `setDefault`.
clangStdenv compiles with a single warning:
```
warning: destructor called on non-final 'nix::PosAdapter' that has virtual functions but non-virtual destructor [-Wdelete-non-abstract-non-virtual-dtor]
```
This fixes the warning by making the destructor of PosAdapter virtual,
deffering to the correct destructor from the concrete child classes.
This has no impact in the end, as none of these classes have specific
destructors.
Technicaly, it may be faster not to have this indirection, but as per
the warning, there is only one place where we have to delete abstract
PosAdapter values.
Not worth bikesheding I guess.
Allows checking directory entry type of a single file/directory.
This was added to optimize the use of `builtins.readDir` on some
filesystems and operating systems which cannot detect this information
using POSIX's `readdir`.
Previously `builtins.readDir` would eagerly use system calls to lookup
these filetypes using other interfaces; this change makes these
operations lazy in the attribute values for each file with application
of `builtins.readFileType`.
We had some local variables left over from the older (more
complicated) implementation of this function. They should all be unused,
but one wasn't by mistake.
Delete them all, and replace the one that was still in use as intended.
It's used as the “system” profile in a bunch of places, so better not
touch it. Besides, it doesn't hurt to keep it since it's owned by root
any way, so it doesn't have the `chown` problem that the user profiles
had and that led to wanting to move them on the client-side.
Rather than using `/nix/var/nix/{profiles,gcroots}/per-user/`, put the user
profiles and gcroots under `$XDG_DATA_DIR/nix/{profiles,gcroots}`.
This means that the daemon no longer needs to manage these paths itself
(they are fully handled client-side). In particular, it doesn’t have to
`chown` them anymore (removing one need for root).
This does change the layout of the gc-roots created by nix-env, and is
likely to break some stuff, so I’m not sure how to properly handle that.
Originally there was no `path-info.*`, then there was `path-info.hh`,
then there was `path-info.cc`, but only for new things. Moving this
stuff over makes everything consistent.
Instead of needing to run `nix show-config --json | jq -r
'."warn-dirty".value'` to view the value of `warn-dirty`, you can now
run `nix show-config warn-dirty`.
This should be a non-empty set, and so we don't want people doing this
by accident. We remove the zero-0 constructor with a little inheritance
trickery.
`DerivedPath::Built` and `DerivationGoal` were previously using a
regular set with the convention that the empty set means all outputs.
But it is easy to forget about this rule when processing those sets.
Using `OutputSpec` forces us to get it right.
It appears that on current macOS versions, our use of poll() to detect
client disconnects no longer works. As a workaround, poll() for
POLLRDNORM, since this *will* wake up when the client has
disconnected. The downside is that it also wakes up when input is
available. So just sleep for a bit in that case. This means that on
macOS, a client disconnect may take up to a second to be detected,
but that's better than not being detected at all.
Fixes#7584.
This way the links are clearly within the manual (ie not absolute paths),
while allowing snippets to reference the documentation root reliably,
regardless of at which base url they're included.
Prior to this change, we had a bunch of ad-hoc string manipulation code
scattered around. This made it hard to figure out what data model for
string contexts is.
Now, we still store string contexts most of the time as encoded strings
--- I was wary of the performance implications of changing that --- but
whenever we parse them we do so only through the
`NixStringContextElem::parse` method, which handles all cases. This
creates a data type that is very similar to `DerivedPath` but:
- Represents the funky `=<drvpath>` case as properly distinct from the
others.
- Only encodes a single output, no wildcards and no set, for the
"built" case.
(I would like to deprecate `=<path>`, after which we are in spitting
distance of `DerivedPath` and could maybe get away with fewer types, but
that is another topic for another day.)
macOS doesn't have user namespacing, so the gid of the builder needs
to be nixbld. The logic got "has sandboxing enabled" confused with
"has user namespaces".
Fixes#7529.
This basically reverts 6e5165b773.
It fixes errors like
sandbox-exec: <internal init prelude>:292:47: unable to open sandbox-minimal.sb: not found
when trying to run a development Nix installed in a user's home
directory.
Also, we're trying to minimize the number of installed files
to make it possible to deploy Nix as a single statically-linked
binary.
Adds a new boolean structured attribute
`outputChecks.<output>.unsafeDiscardReferences` which disables scanning
an output for runtime references.
__structuredAttrs = true;
outputChecks.out.unsafeDiscardReferences = true;
This is useful when creating filesystem images containing their own embedded Nix
store: they are self-contained blobs of data with no runtime dependencies.
Setting this attribute requires the experimental feature
`discard-references` to be enabled.
Previously addTempRoot() acquired the LocalStore state lock and waited
for the garbage collector to reply. If the garbage collector is in the
same process (as it the case with auto-GC), this would deadlock as
soon as the garbage collector thread needs the LocalStore state lock.
So now addTempRoot() uses separate Syncs for the state that it
needs. As long at the auto-GC thread doesn't call addTempRoot() (which
it shouldn't), it shouldn't deadlock.
Fixes#3224.
This also moves the file handle into its own Sync object so we're not
holding the _state while acquiring the file lock. There was no real
deadlock risk here since locking a newly created file cannot block,
but it's still a bit nicer.
This has the same goal as b13fd4c58e81b2b2b0d72caa5ce80de861622610,but
achieves it in a different way in order to not break
`nix why-depends --derivation`.
In principle, this should avoid deadlocks where two instances of Nix are
holding a shared lock on big-lock and are both waiting to get an
exclusive lock.
However, it seems like `flock(2)` is supposed to do this automatically,
so it's not clear whether this is actually where the problem comes from.
This makes 'nix develop' set the Linux personality in the same way
that the actual build does, allowing a command like 'nix develop
nix#devShells.i686-linux.default' on x86_64-linux to work correctly.
Without this, the error is lost, and it makes for a hard to debug
situation. Also remove some of the busyness inside the sqlite_open_v2
args.
The errcode returned is not the extended one. The only way to make open
return an extended code, would be to add SQLITE_OPEN_EXRESCODE to the
flags. In the future it might be worth making this change,
which would also simplify the existing SQLiteError code.
This makes 'nix build' work on paths (which will be copied to the
store) and store paths (returned as is). E.g. the following flake
output attributes can be built using 'nix build .#foo':
foo = ./src;
foo = self.outPath;
foo = builtins.fetchTarball { ... };
foo = (builtins.fetchTree { .. }).outPath;
foo = builtins.fetchTree { .. } + "/README.md";
foo = builtins.storePath /nix/store/...;
Note that this is potentially risky, e.g.
foo = /.;
will cause Nix to try to copy the entire file system to the store.
What doesn't work yet:
foo = self;
foo = builtins.fetchTree { .. };
because we don't handle attrsets with an outPath attribute in it yet,
and
foo = builtins.storePath /nix/store/.../README.md;
since result symlinks have to point to a store path currently (rather
than a file inside a store path).
Fixes#7417.
They did not include the detailed error message, losing essential
information for troubleshooting.
Example message:
warning: creating statement 'insert or rplace into NARs(cache, hashPart, namePart, url, compression, fileHash, fileSize, narHash, narSize, refs, deriver, sigs, ca, timestamp, present) values (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 1)': at offset 10: SQL logic error, near "rplace": syntax error (in '/tmp/nix-shell.grQ6f7/nix-test/tests/binary-cache/test-home/.cache/nix/binary-cache-v6.sqlite')
It's not the best example; more important information will be in
the message for e.g. a constraint violation.
I don't see why this specific error is printed as a warning, but
that's for another commit.
Unsetting `build-users-group` (without `auto-allocate-uids` enabled)
gives the following error:
```
src/libstore/lock.cc:25: static std::unique_ptr<nix::UserLock> nix::SimpleUserLock::acquire(): Assertion `settings.buildUsersGroup != ""' failed.
```
Fix the logic in `useBuildUsers` and document the default value
for `build-users-group`.
This makes the position object used in exceptions abstract, with a
method getSource() to get the source code of the file in which the
error originated. This is needed for lazy trees because source files
don't necessarily exist in the filesystem, and we don't want to make
libutil depend on the InputAccessor type in libfetcher.
Make everything be in the form "while ..." (most things were already),
and in particular *don't* use other propositions that must go after or
before specific "while ..." clauses to make sense.
When debugging nix expressions the outermost trace tends to be more useful
than the innermost. It is therefore printed last to save developers from
scrolling.
We used to set enforceDeterminism to true in the settings (by default)
and thus did send a non-zero value over the wire. The value should
probably be ignored as it should only matter if nrRounds is non-zero
as well.
Having the old code here where the value is expected to be zero only
works with the same version of Nix where we are sending zero. We
should always test this against older Nix versions being client or
server as otherwise upgrade in larger networks might be a pain.
Fixes 8e0946e8df
Fix#6209
When trying to run `nix log <installable>`, try first to resolve the derivation pointed to
by `<installable>` as it is the resolved one that holds the build log.
This has a couple of shortcomings:
1. It’s expensive as it requires re-reading the derivation
2. It’s brittle because if the derivation doesn’t exist anymore or can’t
be resolved (which is the case if any one of its build inputs is missing),
then we can’t access the log anymore
However, I don’t think we can do better (at least not right now).
The alternatives I see are:
1. Copy the build log for the un-resolved derivation. But that means a
lot of duplication
2. Store the results of the resolving in the db. Which might be the best
long-term solution, but leads to a whole new class of potential
issues.
These only functioned if a very narrow combination of conditions held:
- The result path does not yet exist (--check did not result in
repeated builds), AND
- The result path is not available from any configured substituters, AND
- No remote builders that can build the path are available.
If any of these do not hold, a derivation would be built 0 or 1 times
regardless of the repeat option. Thus, remove it to avoid confusion.
The old way was not correct.
Here is an example:
```
$ nix-instantiate --eval --expr 'let x = a: throw "asdf"; in x 1' --show-trace
error: asdf
… while evaluating 'x'
at «string»:1:9:
1| let x = a: throw "asdf"; in x 1
| ^
… from call site
at «string»:1:29:
1| let x = a: throw "asdf"; in x 1
| ^
```
and yet also:
```
$ nix-instantiate --eval --expr 'let x = a: throw "asdf"; in x' --show-trace
<LAMBDA>
```
Here is the thing: in both cases we are evaluating `x`!
Nix is a higher-order languages, and functions are a sort of value. When
we write `x = a: ...`, `a: ...` is the expression that `x` is being
defined to be, and that is already a value. Therefore, we should *never*
get an trace that says "while evaluating `x`", because evaluating `a:
...` is *trival* and nothing happens during it!
What is actually happening here is we are applying `x` and evaluating
its *body* with arguments substituted for parameters. I think the
simplest way to say is just "while *calling* `x`", and so that is what I
changed it to.
We need to close the GC server socket before shutting down the active
GC client connections, otherwise a client may (re)connect and get
ECONNRESET. But also handle ECONNRESET for resilience.
Fixes random failures like
GC socket disconnected
connecting to '/tmp/nix-shell.y07M0H/nix-test/default/var/nix/gc-socket/socket'
sending GC root '/tmp/nix-shell.y07M0H/nix-test/default/store/kb5yzija0f1x5xkqkgclrdzldxj6nnc6-non-blocking'
reading GC root from client: error: unexpected EOF reading a line
1 store paths deleted, 0.00 MiB freed
error: reading from file: Connection reset by peer
in gc-non-blocking.sh.
It calls strlen() on the input (rather than simply copying at most
`size` bytes), which can fail if the input is not zero-terminated and
is inefficient in any case.
Fixes#7347.
why-depends assumed that we knew the output path of the second argument.
For CA derivations, we might not know until it's built. One way to solve
this would be to build the second installable to get the output path.
In this case we don't need to, though. If the first installable (A)
depends on the second (B), then getting the store path of A will
necessitate having the store path B. The contrapositive is, if the store
path of B is not known (i.e. it's a CA derivation which hasn't been
built), then A does not depend on B.
We shouldn't skip this if the supplementary group list is empty,
because then the sandbox won't drop the supplementary groups of the
parent (like "root").
The new experimental feature 'cgroups' enables the use of cgroups for
all builds. This allows better containment and enables setting
resource limits and getting some build stats.
It occurred when a output of the dependency was already available,
so it didn't need rebuilding and didn't get added to the
inputDrvOutputs.
This process-related info wasn't suitable for the purpose of finding
the actual input paths for the builder. It is better to do this in
absolute terms by querying the store.
This change is needed to support aws-sdk-cpp 1.10 and newer.
I opted not to make this dependent on the sdk version because
the crt dependency has been in the interface of the older
sdk as well, and it was only coincidence that libstore didn't
make use of any privately defined symbols directly.
When calling `builtins.readFile` on a store path, the references of that
path are currently added to the resulting string's context.
This change makes those references the *possible* context of the string,
but filters them to keep only the references whose hash actually appears
in the string, similarly to what is done for determining the runtime
references of a path.
Cgroups are now only used for derivations that require the uid-range
range feature. This allows auto UID allocation even on systems that
don't have cgroups (like macOS).
Also, make things work on modern systems that use cgroups v2 (where
there is a single hierarchy and no "systemd" controller).
after discussing this with multiple people, I'm convinced that "build
task" is more precise: a derivation is not an action, but inert until it
is built. also it's easier to pronounce.
proposal: use "build task" for the generic concept "description of how
to derive new files from the contents of existing files". then it will
be easier to distinguish what we mean by "derivation" (a specific data
structure and Nix language value type) and "store derivation" (a
serialisation of a derivation into a file in the Nix store).
readline is not re-entrant, so entering the debugger from the
completioncallback results in an eventual segfault.
The workaround is to temporarily disable the debugger when searching
for possible completions.
Call it as `['nix', '__build-remote', ... ]` rather than the previous
`["__build-remote", "nix __build-remote", ... ]` which seemed to have
been most likely unintended
The description of the --profile option talks about the "update" operation.
This is probably meant for operations such as "nix profile install", but the
same option is reused in other subcommands, which do not update the profile,
such as "nix profile {list,history,diff-closures}".
We update the description to make sense in both contexts.
Currently, Nix passes `-a` when it runs commands on a remote machine via
SSH, which disables agent forwarding. This causes issues when the
`ForwardAgent` option is set in SSH config files, as the command line
operation always overrides those.
In particular, this causes issues if the command being run is `sudo`
and the remote machine is configured with the equivalent of NixOS's
`security.pam.enableSSHAgentAuth` option. Not allowing SSH agent
forwarding can cause authentication to fail unexpectedly.
This can currently be worked around by setting `NIX_SSHOPTS="-A"`, but
we should defer to the options in the SSH config files to be least
surprising for users.
Robert Hensing
Yorick
Jörg Thalheim
Linus Heckemann
Patrick Jackson