This prevents combining untrusted with introducer and auto-accept, and
also verifies that folders shared with untrusted devices have passwords
at config loading time.
Co-authored-by: Simon Frei <freisim93@gmail.com>
We usually want to ensure that our own device is present. However if the
given device ID is the empty ID, we shouldn't do that. This is a
legimate (though way too non-obvious) use-case when opening the config
without knowing/caring about the device ID.
* Platform data (ownership, xattrs, etc.) is now set correctly for newly-received folders, even if the received folder has the NoPermissions flag.
* Call setPlatformData on receivers that have ignorePerms set to true.
This fixes various test issues with Go 1.20.
- Most tests rewritten to use fakefs where possible
- Some tests that were already skipped, or dubious (invasive,
unmaintainable, unclear what they even tested) have been removed
- Some actual code rewritten to better support testing in fakefs
Co-authored-by: Eric P <eric@kastelo.net>
In the sequence of loading ignores, the error File Does Not Exist is not being considered a fatal error, since the .stignore file is allowed to not exist. However, included ignore files also tossed that same error in case those do not exist while in those cases it's considered an error and it should lead to the folder stopping. Changing the error when opening an included ignore file to something other than the regular does fix this issue, as in it now works again as described in the Documentation.
This makes the various protocol priorities configurable among the other
options. With this, it's possible to prefer QUIC over TCP for WAN
connections, for example. Both sides need to be similarly configured for
this to work properly.
The default priority order remains the same as previously (TCP, QUIC,
Relay, with LAN better than WAN).
To make this happen I made each dialer & listener more priority aware,
and moved the check for whether a connection is LAN or not into the
dialer / listener -- this is the new "lanChecker" type that's passed
around.
In the original fix in #8563 I simply forgot this. Which meant #8556
wasn't actually fixed, as the trialer size would have been 0 (default),
and thus we would have still sent the inflated size to encrypted peers.
lib/model: Fix file size inconsisency due to enc. trailer
Fixes a regression due to PR #8563, while arguable the bug was actually
introduced in a much older PR #7155, but didn't have any bad effects so
far:
We account for the encryption trailer in the db updater routine,
calculating the file-info size there. However there's no guarantee that
the file-info at this point is still the exact same as when it was
written. It was before, but isn't anymore since introducing the new
EncryptedTrailerSize field.
Fix: Adjust the size in the info at the same place where the trailer is
written, i.e. we definitely have the actual size on disk.
The layout of the request differs based on whether it comes from an
untrusted device or a trusted device with encrypted enabled. Handle
both.
Closes#8819.
Allow the watcher delay to take fractional values, effectively allowing
for much shorter delays. The minimum value is limited at 0.01, which
effectively translates to 10ms. This is required in order to guarantee
that there is still enough time to aggregate multiple single change
events.
Signed-off-by: Tomasz Wilczyński <twilczynski@naver.com>
This adds a cache to the expensive key generation operations. It's fixes
size LRU/MRU stuff to keep memory usage bounded under absurd conditions.
Also closes#8600.
This adds the BlocksHash field from the FileInfo to our API output. It
can be useful for debugging, or for external tools. I'm intentionally
leaving it as an opaque base64 string because no meaning should be
derived from it: it's just a string.
This makes sure the service manager doesn't interpret timeout errors, or any other error, as a signal to stop the service instead of restarting it.
I added it directly to our service utility function, as it may help catch other instances of the same problem... We would typically want timeouts etc to be a retryable error, unless it is the top level context that has timed out and we check for that specifically.
This adds a word to the version string when running containerized. The
purpose is mostly to facilitate troubleshooting via screenshot by
"leaking" this rather important aspect of the setup. Additionally, the
version row gets "no-overflow-ellipsis" treatment so that the whole
thing is actually visible in the GUI and the (now useless) tooltip is
removed. In production releases this won't make a difference as the
whole thing will typically fit, but in odd setups it provides more info
up front.
* lib/connections: Cache isLAN decision for later external access.
The check whether a remote device's address is on a local network
currently happens when handling the Hello message, to configure the
limiters. Save the result to the ConnectionInfo and pass it out as
part of the model's ConnectionInfo struct in ConnectionStats().
* gui: Use provided connection attribute to distinguish LAN / WAN.
Replace the dumb IP address check which didn't catch common cases and
actually could contradict what the backend decided. That could have
been confusing if the GUI says WAN, but the limiter is not actually
applied because the backend thinks it's a LAN.
Add strings for QUIC and relay connections to also differentiate
between LAN and WAN.
* gui: Redefine reception level icons for all connection types.
Move the mapping to the JS code, as it is much easier to handle
multiple switch cases by fall-through there.
QUIC is regarded no less than TCP anymore. LAN and WAN make the
difference between levels 4 / 3 and 2 / 1:
{TCP,QUIC} LAN --> {TCP,QUIC} WAN --> Relay LAN --> Relay WAN -->
Disconnected.
Previous debug input didn't really give enough info to show what was
happening, while it also printed full block lists which are enormously
verbose. Now it consistently prints 1. what it sees on disk, 2. what it
got from CurrentFile (without blocks), 3. the action taken on that file.
There are some situations where an upgrade wouldn't be supported, even though the noUpgrade bool isn't set. So when handling the errors that are caused by this, when attempting an upgrade, it shouldn't lead to some sort of offline-message/restart/warning/etc...
I added some checks on specific errors related to this and return a 501 (Not Implemented) response instead, in case of an "UpgradeUnsupported"-error. Additionally, on the GUI-side, the 501-response is now not to be considered an error to act upon.
* implement authentication via token for relaysrv
Make replaysrv check for a token before allowing clients to
join. The token can be set via the replay-uri.
* fix formatting
* key composite literal
* do not error out if auth material is provided but not needed
* remove unused method receiver
* clean up unused parameter in functions
* cleaner token handling, disable joining the pool if token is set.
* Keep backwards compatibility with older clients.
In prior versions of the protocol JoinRelayRequest did not have a
token field. Trying to unmarshal such a request will result in
an error. Return an empty JoinRelayRequest, that is a request
without token, instead.
Co-authored-by: entity0xfe <entity0xfe@my.domain>
The restore function of Trash Can ran a rename at the end regardless of whether there was anything to rename. In this case, when the file-to-be-restored did not exist in the destination folder, this resulted in an error. I added a simple check, keeping track of whether the file existed prior to restoring it in the destination folder and depending on this value it will now return nil after the restoration to prevent the renaming function to kick off. Added a test for this specific edge-case as well.
This adds support for syncing extended attributes on supported
filesystem on Linux, macOS, FreeBSD and NetBSD. Windows is currently
excluded because the APIs seem onerous and annoying and frankly the uses
cases seem few and far between. On Unixes this also covers ACLs as those
are stored as extended attributes.
Similar to ownership syncing this will optional & opt-in, which two
settings controlling the main behavior: one to "sync" xattrs (read &
write) and another one to "scan" xattrs (only read them so other devices
can "sync" them, but not apply any locally).
Co-authored-by: Tomasz Wilczyński <twilczynski@naver.com>
The cleaning logic in util.go was used by Simple and Trashcan but only
really suited Trashcan since it works based on mtimes which Simple does
not use. The cleaning logic in util.go was moved to trashcan.go.
Staggered and Simple seemed to be able to benefit from the same base so
util.go now has the base for those two with an added parameter which
takes a function so it can still handle versioner-specific logic to
decide which files to clean up. Simple now also correctly cleans files
based on their time-stamp in the title together with a specific maximum
amount to keep. The Archive function in Simple.go was changed to get rid
of duplicated code.
Additionally the trashcan testcase which was used by Trashcan as well as
Simple was moved from versioner_test.go to trashcan_test.go to keep it
clean, there was no need to keep it in a separate test file
This replaces old style errors.Wrap with modern fmt.Errorf and removes
the (direct) dependency on github.com/pkg/errors. A couple of cases are
adjusted by hand as previously errors.Wrap(nil, ...) would return nil,
which is not what fmt.Errorf does.
Locally paused folders will fail on checkFolderRunningLocked() and
therefore abort the loop. Avoid this by skipping paused folders
directly.
Co-authored-by: Jakob Borg <jakob@kastelo.net>
* lib/locations: Fix enum values camelCase.
* lib/locations: Remove unused FailuresFile.
* cmd/syncthing: Turn around role of locations storage.
Previously the locations package was used to provide default paths,
possibly with an overridden home directory. Extra paths supplied on
the command line were handled and passed around in the options object.
To make the changed paths available to any other interested package,
override the location setting from the option if supplied, instead of
vice versa when not supplied. Adapt code using this to read from the
locations package instead of passing through the options object.
* lib/locations: Refactor showPaths to locations package.
Generate a reusable string in locations.PrettyPrintPaths().
Enumerating all possible locations in different packages is error
prone, so add a new public function to generate the listing as a
string in the locations package. Adapt cmd/syncthing --paths to use
that instead of its own console output.
* lib/locations: Include CSRF token in pretty printed paths.
* lib/api: New endpoint /rest/system/paths.
The paths should be available for troubleshooting from a running
instance. Using the --paths CLI option is not easy in some
environments, so expose the locations mapping to a JSON endpoint.
Add utility function ListExpandedPaths() that also filters out any
entries which still contain variable placeholders.
* gui: List runtime paths in separate log viewer tab.
* Wrap paths.
* lib/syncthing: Utilize locations.Get() instead of passing an arg.
* Include base directories, move label to table caption.
* gui: Switch to hard-coded paths instead of iterating over all.
* gui: Break aboutModalView into tabs.
Use tabs to separate authors from included third-party software.
* gui: Move paths from log viewer to about modal.
* lib/locations: Adjust pretty print output order to match GUI.
* gui, authors: Remove additional bot names and fix indent.
The indentation changed because of the tabbed about dialog, fix the
authors script to respect that.
Skip Syncthing*Automation in authors list as well.
* Update AUTHORS list to remove bot names.
* Revert AUTHORS email order change.
* Do not emphasize DB and log file locations.
* Review line wrapping.
* review part 1: strings.Builder, naming
* Rename and extend locations.Set() with error handling.
Remodel the Override() function along the existing SetBaseDir() and
rename it to simply Set(). Make sure to use absolute paths when given
log file or GUI assets override options. Add proper error reporting
if that goes wrong.
* Remove obsolete comment about empty logfile option.
* Don't filter out unexpanded baseDir placeholders, only ${timestamp}.
* Restore behavior regarding special "-" logfile argument.
If the option is given, but with empty value, assume the no log
file (same as "-"). Don't try to convert the special value to an
absolute path though and document this fact in a comment for the Set()
function.
* Use template to check for location key validity.
* Don't filter out timestamp placeholders.
* lib/api: Remove paths from /rest/system/status.
* lib/ur: Properly initialize map in failure data (fixes#8479)
Co-authored-by: Jakob Borg <jakob@kastelo.net>
Also adds idle time and keepalive parameters because how this is
configured has changed in the new package version. The values are those
that seems like might already be default, if keep-alives were enabled,
which is not obvious from the doc comments.
Also, Go 1.19 gofmt reformatting of comments.
all: Add package runtimeos for runtime.GOOS comparisons
I grew tired of hand written string comparisons. This adds generated
constants for the GOOS values, and predefined Is$OS constants that can
be iffed on. In a couple of places I rewrote trivial switch:es to if:s,
and added Illumos where we checked for Solaris (because they are
effectively the same, and if we're going to target one of them that
would be Illumos...).
This adds support for syncing ownership on Unixes and on Windows. The
scanner always picks up ownership information, but it is not applied
unless the new folder option "Sync Ownership" is set.
Ownership data is stored in a new FileInfo field called "platform data". This
is intended to hold further platform-specific data in the future
(specifically, extended attributes), which is why the whole design is a
bit overkill for just ownership.
* lib/model: Override scan config for auto-accepted encrypted folders.
Encrypted folders should not have the fs watcher enabled and rarely
benefit from a scheduled rescan. The GUI adjusts the suggested
settings (watcher disabled, one day rescan interval) when accepting a
receive-encrypted folder. Mirror that behavior to the auto-accept
case where the GUI is not involved.
Versioning also does not work well for encrypted folders, same
treatment.
