The locking protocol in nat.Mapping was racy:
* Mapping.addressMap RLock'd, but then returned a map shared between
caller and Mapping, so the lock didn't do anything.
* Operations inside Service.{verifyExistingMappings,acquireNewMappings}
would lock the map for every update, but that means callers to
Mapping.ExternalAddresses can be looping over the map while the
Service methods are concurrently modifying it. When the Go runtime
detects that happening, it panics.
* Mapping.expires was read and updated without locking.
The Service methods now lock the map once and release the lock only when
done.
Also, subscribers no longer get the added and removed addresses, because
none of them were using the information. This was changed for a previous
attempt to retain the fine-grained locking and not reverted because it
simplifies the code.
Accept a subcommand as an alternative to the --generate option. It
accepts a custom config directory through either the --home or
--config options, using the default location if neither is given.
Add the options --gui-user and --gui-password to "generate", but not
the "serve --generate" option form. If either is given, an existing
config will not abort the command, but rather load, modify and save it
with the new credentials. The password can be read from standard
input by passing only a single dash as argument.
Config modification is skipped if the value matches what's already in
the config.
* cmd/syncthing: Utilize lib/locations package in generate().
Instead of manually joining paths with "magic" file names, get them
from the centralized locations helper lib.
* cmd/syncthing: Simplify logging for --generate option.
Visible change: No more timestamp prefixes.
What hash is used to store the password should ideally be an
implementation detail, so that every user of the GUIConfiguration
object automatically agrees on how to handle it. That is currently
distribututed over the confighandler.go and api_auth.go files, plus
tests.
Add the SetHasedPassword() / CompareHashedPassword() API to keep the
hashing method encapsulated. Add a separate test for it and adjust
other users and tests. Remove all deprecated imports of the bcrypt
package.
LoadOrGenerateCertificate() takes two file path arguments, but then
uses the locations package to determine the actual path. Fix that
with a minimally invasive change, by using the arguments instead.
Factor out GenerateCertificate().
The only caller of this function is cmd/syncthing, which passes the
same values, so this is technically a no-op.
* lib/tlsutil: Make storing generated certificate optional. Avoid
temporary cert and key files in tests, keep cert in memory.
Consistently use double dashes and fix typos -conf, -data-dir and
-verify.
Applies also to tests running the syncthing binary for consistency.
* Fix mismatched option name --conf in cli subcommand.
According to the source code comments, the cli option flags should
mirror those from the serve subcommand where applicable. That one is
actually called --config though.
* cli: Fix help text option placeholders.
The urfave/cli package uses the Value field of StringFlag to provide a
default value, not to name the placeholder. That is instead done with
backticks around some part of the Usage field.
* cli: Add missing --data flag in subcommand help text.
The urfave/cli based option parsing uses a fake flags collection to
generate help texts matching the used global options. But the --data
option was omitted from it, although it is definitely required when
using --config as well. Note that it cannot just be ignored, as some
debug stuff actually uses the DB:
syncthing cli --data=/bar --config=/foo debug index dump
By truncating time.Time to an int64 nanosecond count, we lose the
ability to precisely order timestamps before 1678 or after 2262, but we
gain (linux/amd64, Go 1.17.1):
name old time/op new time/op delta
JobQueuePushPopDone10k-8 2.85ms ± 5% 2.29ms ± 2% -19.80% (p=0.000 n=20+18)
JobQueueBump-8 34.0µs ± 1% 29.8µs ± 1% -12.35% (p=0.000 n=19+19)
name old alloc/op new alloc/op delta
JobQueuePushPopDone10k-8 2.56MB ± 0% 1.76MB ± 0% -31.31% (p=0.000 n=18+13)
name old allocs/op new allocs/op delta
JobQueuePushPopDone10k-8 23.0 ± 0% 23.0 ± 0% ~ (all equal)
Results for BenchmarkJobQueueBump are with the fixed version, which no
longer depends on b.N for the amount of work performed. rand.Rand.Intn
is cheap at ~10ns per iteration.
Like Windows and Mac, Android is also an interactive operating system.
On top of that, it usually runs on much slower hardware than the other
two. Because of that, it makes sense to limit the number of hashes used
by default there too.
Signed-off-by: Tomasz Wilczyński <twilczynski@naver.com>
Registry.Get used a full sort to get the minimum of a list, and the sort
was broken because util.AddressUnspecifiedLess assumed it could find out
whether an address is IPv4 or IPv6 from its Network method. However,
net.(TCP|UDP)Addr.Network always returns "tcp"/"udp".
The current detection is flawed, because it looks for a few specific
file systems like "msdos" or "fat" to set the mtime window, while in
reality Android seems to report names like "fuseblk", which can stand
for fat, ext4, or even f2fs.
At the moment, we set the mtime window only for a few known names used
for the fat filesystem. With this change, we take a safer approach of
always setting the time window unless we explicitly detect file systems
like ext2/ext3/ex4, which are known not to experience issues with moving
timestamps on Android.
Signed-off-by: Tomasz Wilczyński <twilczynski@naver.com>
Before this patch, IPv4-compatible addresses (::ffff:aaa.bbb.ccc.ddd)
may be used if a quic6://some.domain:port is specified and both IPv4 and
IPv6 addresses exist for that domain name.
This changes the GC mechanism so that the first pass (which reads all
FileInfos to populate bloom filters with block & version hashes) can
happen concurrently with normal database operations.
The big gcMut still exists, and we grab it temporarily to block all
other modifications while we set up the bloom filters. We then release
the lock and let other things happen, with those other things also
updating the bloom filters as required. Once the first phase is done we
again grab the gcMut, knowing that we are the sole modifier of the
database, and do the cleanup.
I also removed the final compaction step.
* Trigger connection loop on config device addition (fixes#7600)
* Also check for device address equality
* Move EqualStrings from api_test to utils, and use in connections/service.go
* Make sure CommitConfiguration cannot block due on the deviceAddressesChanged channel
* Update lib/connections/service.go
Co-authored-by: Jakob Borg <jakob@kastelo.net>
This makes us use TLS 1.3+ on sync connections by default. A new option
`insecureAllowOldTLSVersions` exists to allow communication with TLS
1.2-only clients (roughly Syncthing 1.2.2 and older). Even with that
option set you get a slightly simplified setup, with the cipher suite
order fixed instead of auto detected.
No longer hide the web UI controls for the new untrusted/encrypted
device feature. Testing hasn't been very widespread, but there has been
some and quite a few bugs have been caught and fixed. I believe its time
to not hide it anymore, and cautiously recommend usage. E.g. mention
that the feature hasn't been widely used yet and anyone using it is an
early adopter, but drop the bit about not using it with production data.
We can maybe stress the need for backups in general and especially
using this.
There was a logic mistake, so the limit in question wasn't used. On my
macOS this doesn't seem to matter, the hard limit returned is 2^63-1 and
setting the soft limit to that works. However I'm assuming that's not
the case for older macOSes since it was so nicely documented, so we
should still have this working. (10240 FDs should be enough for
anybody.)
This is a mostly pointless change to make security scanners and static
analysis tools happy, as they all hate seeing md5. None of our md5 uses
were security relevant, but still. Only visible effect of this change is
that our temp file names for very long file names become slightly longer
than they were previously...
This adds a couple of dummy asset files protected by the "noassets"
build tag. The purpose is that it should be possible for, for example,
CI tools and static analysis things to compile and analyze the source
tree without our custom asset generation step. Also makes `go test -tags
noassets ./...` work without building assets first.
This truncates times meant for API consumption to second precision,
where fractions won't typically matter or add any value. Exception to
this is timestamps on logs and events, and of course I'm not touching
things like file metadata.
I'm not 100% certain this is an exhaustive change, but it's the things I
found by grepping and following the breadcrumbs from lib/api...
I also considered general-but-ugly solutions, like having the API
serializer itself do reflection magic or even regexps on returned
objects, but decided against it because aurgh...
This loosens the ‘is this localhost?’ check to include *.localhost host
names.
This allows for clearer (hence better) names to be used in browsers,
e.g. when accessing a remote syncthing instance ‘foo’ using a ssh port
forward, one can use foo.localhost to remind oneself which one is which.
💡 Without these changes, Syncthing shows a ‘Host check error’ when
pointing a browser at http://foo.localhost/, and with these changes, the
interface loads as usual.
The .localhost top level domain is a reserved top-level domain (RFC 2606):
> The ".localhost" TLD has traditionally been statically defined in
> host DNS implementations as having an A record pointing to the
> loop back IP address and is reserved for such use. Any other use
> would conflict with widely deployed code which assumes this use.
> – https://tools.ietf.org/html/rfc2606
As Wikipedia puts it:
> This allows the use of these names for either documentation purposes
or in local testing scenarios. – https://en.wikipedia.org/wiki/.localhost
On Linux systems, systemd-resolved resolves *.localhost, on purpose:
https://www.freedesktop.org/software/systemd/man/systemd-resolved.service.html
See also #4815, #4816.
An untrusted device will receive padded info for small blocks, and hence
sometimes request a larger block than actually exists on disk.
Previously we let this pass because we didn't have a hash to compare to
in that case and we ignored the EOF error based on that.
Now the untrusted device does pass an encrypted hash that we decrypt and
verify. This means we can't check for len(hash)==0 any more, but on the
other hand we do have a valid hash we can apply to the data we actually
read. If it matches then we don't need to worry about the read
supposedly being a bit short.
