The folder marker conversion forgot to hide the .stfolder. This adds
that, for those who have not yet been converted.
Also adds Hide() calls to the folder start, to mend historical
unhidedness. (I'm sure this will upset someone who is manually managing
their .stignores in the other direction...)
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4384
When STHASHING is set, don't benchmark as it's already decided. If weak
hashing isn't set to "auto", don't benchmark that either.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4349
These functions were very naive and slow. We haven't done much about
them because they pretty much don't matter at all for Syncthing
performance. They are however called very often in the discovery server
and these optimizations have a huge effect on the CPU load on the
public discovery servers.
The code isn't exactly obvious, but we have good test coverage on all
these functions.
benchmark old ns/op new ns/op delta
BenchmarkLuhnify-8 12458 1045 -91.61%
BenchmarkUnluhnify-8 12598 1074 -91.47%
BenchmarkChunkify-8 10792 104 -99.04%
benchmark old allocs new allocs delta
BenchmarkLuhnify-8 18 1 -94.44%
BenchmarkUnluhnify-8 18 1 -94.44%
BenchmarkChunkify-8 44 2 -95.45%
benchmark old bytes new bytes delta
BenchmarkLuhnify-8 1278 64 -94.99%
BenchmarkUnluhnify-8 1278 64 -94.99%
BenchmarkChunkify-8 42552 128 -99.70%
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4346
Currently all errors during pulling and the first of these errors again on
finishing are logged to info. Besides that the errors logged when finishing
are stored in f.errors. This PR moves all logging during pulling to the debug
channel (they might still be relevant in some obscure debugging case) and
uses the stored errors to log the main error per fail when all pulling
iterations are done and failed.
Additional instead of trying 11 times it now only tries 3 times.
This is the first part of what is discussed here:
https://forum.syncthing.net/t/reduce-verboseness-of-puller/10261
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4338
This updates kcp and uses our own fork which:
1. Keys sessions not just by remote address, but by remote address +
conversation id 2. Allows not to close connections that were passed directly
to the library. 3. Resets cache key if the session gets terminated.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4339
LGTM: calmh
Prior to this, the following is possible:
- Create a symlink "foo -> /somewhere", it gets synced
- Delete "foo", it gets versioned
- Create "foo/bar", it gets synced
- Delete "foo/bar", it gets versioned in "/somewhere/bar"
With this change, versioners should never version symlinks.
Otherwise all the lines from includes will be shown in the web UI instead of
just the #include ... line. This problem was introduced in #3996.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4248
LGTM: calmh
This removes the special handling of minor versions as major when the
actual major is zero, and adds the special case that upgrades from 0.x
to 1.x are considered minor. 0.x to 2.x or 1.x to 2.x etc are still
considered major.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4226
This solves the erratic test failures on model.TestIgnores by ensuring
that the ignore patterns are reloaded even in the face of unchanged
timestamps.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4208
The folder already knew how to stop properly, but the fs.Walk() didn't
and can potentially take a very long time. This adds context support to
Walk and the underlying scanning stuff, and passes in an appropriate
context from above. The stop channel in model.folder is replaced with a
context for this purpose.
To test I added an infiniteFS that represents a large amount of data
(not actually infinite, but close) and verify that walking it is
properly stopped. For that to be implemented smoothly I moved out the
Walk function to it's own type, as typically the implementer of a new
filesystem type might not need or want to reimplement Walk.
It's somewhat tricky to test that this actually works properly on the
actual sendReceiveFolder and so on, as those are started from inside the
model and the filesystem isn't easily pluggable etc. Instead I've tested
that part manually by adding a huge folder and verifying that pause,
resume and reconfig do the right things by looking at debug output.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4117
So, when first implementing the database layer I added panics on every
unexpected error condition mostly to be sure to flush out bugs and
inconsistencies. Then it became sort of standard, and we don't seem to
have many bugs here any more so the panics are usually caused by things
like checksum errors on read. But it's not an optimal user experience to
crash all the time.
Here I've weeded out most of the panics, while retaining a few "can't
happen" ones like errors on marshalling and write that we really can't
recover from.
For the rest, I'm mostly treating any read error as "entry didn't
exist". This should mean we'll rescan the file and correct the info (if
scanning) or treat it as a new file and do conflict handling (when
pulling). In some cases things like our global stats may be slightly
incorrect until a restart, if a database entry goes suddenly missing
during runtime.
All in all, I think this makes us a bit more robust and friendly without
introducing too many risks for the user. If the database is truly toast,
probably many other things on the system will be toast as well...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4118
Harmonize how we use batches in the model, using ProtoSize() to judge
the actual weight of the entire batch instead of estimating. Use smaller
batches in the block map - I think we might have though that batch.Len()
in the leveldb was the batch size in bytes, but it's actually number of
operations.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4114
The mechanism to disallow manual scans before the initial scan completed
(#3996) , had the side effect, that if the initial scan failed, no further
scans are allowed. So this marks the initial scan as finished regardless of
whether it succeeded or not.
There was also redundant code in rofolder and a pointless check for folder
health in scanSubsIfHealthy (happens in internalScanFolderSubdirs as well).
This also moves logging from folder.go to ro/rw-folder.go to include the
information about whether it is send-only or send-receive
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4104
This adds a parameter "events" to the /rest/events endpoint. It should
be a comma separated list of the events the consumer is interested in.
When not given it defaults to the current set of events, so it's
backwards compatible.
The API service then manages subscriptions, creating them as required
for each requested event mask. Old subscriptions are not "garbage
collected" - it's assumed that in normal usage the set of event
subscriptions will be small enough. Possibly lower than before, as we
will not set up the disk event subscription unless it's actually used.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4092
This deprecates the current minDiskFreePct setting and introduces
minDiskFree. The latter is, in it's serialized form, a string with a
unit. We accept percentages ("2.35%") and absolute values ("250 k", "12.5
Gi"). Common suffixes are understood. The config editor lets the user
enter the string, and validates it.
We still default to "1 %", but the user can change that to an absolute
value at will.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4087
LGTM: AudriusButkevicius, imsodin
This adds a new config AllowedNetworks per device, which when set should
contain a list of network prefixes (192.168.0.0/126 etc) that are
allowed for the given device. The connection service will not attempt
connections to addresses outside of the given networks and incoming
connections will be rejected as well.
I've added the config to the normal device editor and shown it (when
set) in the device summary on the main screen.
There's a unit test for the IsAllowedNetwork method, I've done some
manual sanity testing on top of that.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4073
One more step on the path of the great refactoring. Touches rwfolder a
little bit since it uses the Lstat from fs as well, but mostly this is
just on the scanner as rwfolder is scheduled for a later refactor.
There are a couple of usages of fs.DefaultFilesystem that will in the
end become a filesystem injected from the top, but that comes later.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4070
LGTM: AudriusButkevicius, imsodin
Adds a unit test to ensure we don't scan symlinks on Windows. For the
rwfolder, trusts that the logic in the invalid check is correct and that
the check is actually called from the need loop.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4042
Basically, if we don't care about the sync status of the file we should
not tag someone else out of sync because they don't have the latest
version. This solves *my* "Syncing - 100%" scenario at least.
The reason this happens seems to be like this, in my situation. I have
three devices, connected in a "line": A-B-C. A is a Mac and litters
.DS_Store files everywhere. I've ignored these, but some escaped into
the folders before I did so. I've also ignored them on B and C but at
different stages. B was flagging C as out of sync, because at the point
the ignores were introduced C had a lower version of .DS_Store than A.
Now none of them are sending updates about it any more since it's
ignored...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3981
Other routines use atomics, hence even if we are under a lock, we should
too.
We might atomically store with
Not sure how it happens, but it's between lines
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3974
After this change,
- Symlinks on Windows are always unsupported. Sorry.
- Symlinks are always enabled on other platforms. They are just a small
file like anything else. There is no need to special case them. If you
don't want to sync some symlinks, ignore them.
- The protocol doesn't differentiate between different "types" of
symlinks. If that distinction ever does become relevant the individual
devices can figure it out by looking at the destination when they
create the link.
It's backwards compatible in that all the old symlink types are still
understood to be symlinks, and the new SYMLINK type is equivalent to the
old SYMLINK_UNKNOWN which was always a valid way to do it.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3962
LGTM: AudriusButkevicius
Instead of just immediately dropping the event if the subscription isn't
ready to receive it, give it 15 ms to catch up. The value 15 ms is
grabbed out of thin air - it just seems reasonable to me.
The timer juggling makes the event send pretty much exactly twice as
slow as it was before, but we're still under a microsecond. I think it's
negligible compared to whatever event that just happened that we're
interested in logging (usually a file operation of some kind).
benchmark old ns/op new ns/op delta
BenchmarkBufferedSub-8 475 950 +100.00%
benchmark old allocs new allocs delta
BenchmarkBufferedSub-8 4 4 +0.00%
benchmark old bytes new bytes delta
BenchmarkBufferedSub-8 104 117 +12.50%
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3960
Instead of
[I6KAH] 19:05:56 INFO: Single thread hash performance is 359 MB/s using minio/sha256-simd (354 MB/s using crypto/sha256).
it now says
[I6KAH] 19:06:16 INFO: Single thread SHA256 performance is 359 MB/s using minio/sha256-simd (354 MB/s using crypto/sha256).
[I6KAH] 19:06:17 INFO: Actual hashing performance is 299.01 MB/s
which is more informative. This is also the number it reports in usage
reporting.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3918
Can't do what I did, as the rolling function is not the same as the
non-rolling one. Instead this uses an improved version of the rolling
adler32 to accomplish the same thing. (PR filed on upstream, so should
be able to use that directly in the future.)
The rolling version of adler32 is just a wrapper around the standard
hash/adler32 when used in a non-rolling fashion, but it's inefficient as
it allocates a new hash instance for every Write(). This uses the
default version instead in the block hasher, and adds a test to verify
the result is the same as they were before. It reduces allocations by
88% and increases speed about 5%.
benchmark old ns/op new ns/op delta
BenchmarkHashFile-8 64434698 61303647 -4.86%
benchmark old MB/s new MB/s speedup
BenchmarkHashFile-8 276.65 290.78 1.05x
benchmark old allocs new allocs delta
BenchmarkHashFile-8 1238 150 -87.88%
benchmark old bytes new bytes delta
BenchmarkHashFile-8 17877363 49292 -99.72%
Syncthing adds some hidden files when a folder is added, but there is currently
no equivalent cleanup procedure. This change is conservative as not to
accidentally cause data loss.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3874
On Windows we would descend into SYMLINKD type links when we scanned
them successfully, as we would return nil from the walk function and the
filepath.Walk iterator apparently thought it OK to descend into the
symlinked directory.
With this change we always return filepath.SkipDir no matter what.
Tested on Windows 10 as admin, does what it should.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3875
Also tweaks the proto definitions:
- [packed=false] on the block_indexes field to retain compat with
v0.14.16 and earlier.
- Uses the vendored protobuf package in include paths.
And, "build.go setup" will install the vendored protoc-gen-gogofast.
This should ensure that a proto rebuild isn't so dependent on whatever
version of the compiler and package the developer has installed...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3864
The protobuf encoder now produces packed arrays for things like []int32,
which is actually correct according to the proto3 spec. However
Syncthing v0.14.16 and earlier doesn't support this. This reverts the
encoding change, but keeps the updated decoder so that we are both more
compatible with other proto3 implementations and can move to the updated
encoder in the future.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3856
Since we anyway need the folderConfig for this I'm skipping the copying
of all it's attributes that rwfolder did and just keeping the original
around instead.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3825
This adds support for AES_256_GCM_SHA384 (in there since Go 1.5, a bit
of a shame we missed it) and ChaCha20-Poly1305 (if built with Go 1.8;
ignored on older Gos).
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3822
The test for the error string is fragile, and the error string changed
in Go 1.8 so the relevant part is no longer a prefix. This covers it
with a test though, so it should be fine in the future as well.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3818
Instead, trust (and test) that the temp file has appropriate permissions
from the start. The only place where this changes our behavior is for
ignores which go from 0644 to 0600. I'm OK with that.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3756
This changes the "seen" map that we're anyway keeping around to track
the modtimes of loaded files instead. When doing a Load() we check that
1) the file we are loading is in the modtime set, and 2) that none of
the files in the modtime set have changed modtimes. If that's the case
we do a quick return without parsing anything or clearing the cache.
This required adding two one seconds sleeps in the tests to make sure
the modtimes were updated when we expect cache reloads, because I'm on a
crappy filesystem with one second timestamp granularity. That also
proves it works...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3754
Fsyncing the file has a small performance penalty and seems unnecessary. The
file will be fsynced anyway, when the changes are commited to the database.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3749
This makes the device ID a real type that can be used in the protobuf
schema. That avoids the juggling back and forth from []byte in a bunch
of places and simplifies the code.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3695
This adds autodetection of the fastest hashing library on startup, thus
handling the performance regression. It also adds an environment
variable to control the selection, STHASHING=standard (Go standard
library version, avoids SIGILL crash when the minio library has bugs on
odd CPUs), STHASHING=minio (to force using the minio version) or unset
for the default autodetection.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3617
When the GUI/API is bound to localhost, we enforce that the Host header
looks like localhost. This can be disabled by setting
insecureSkipHostCheck in the GUI config.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3558
When files that were previously marked as deleted became ignored, we
used to do nothing at all. This changes that behavior to set the Invalid
bit (that we should rename to Ignored). This then becomes an update to
other devices that they should not trust our knowledge about the file in
question.
Read this diff without whitespace...
Tested by
- creating a bunch of files on s1
- letting them sync to s2
- shutting down s2
- deleting the files on s1 and rescanning
- adding the files to .stignore on s1 and rescanning
- starting up s2 and letting it sync
- observing the files are not deleted on s2, and it considers itself up
to date.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3557
We used to consider deleted files & directories 128 bytes large. After
the delta indexes change a bug slipped in where deleted files would be
weighted according to their old non-deleted size. Both ways are
incorrect (but the latest change made it worse), as if there are more
files deleted than remaining data in the repo the needSize can be
greater than the globalSize, resulting in a negative completion
percentage.
This change makes it so that deleted items are zero bytes large, which
makes more sense. Instead we expose the number of files that we need to
delete as a separate field in the Completion() result, and hack the
percentage down to 95% complete if it was 100% complete but we need to
delete files. This latter part is sort of ugly, but necessary to give
the user some sort of feedback.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3556
We previously set the mtime on the temp file, and then renamed it to the
real path. Unfortunately that means we'd save the real timestamp under
the under the temp name ".syncthing.foo.tmp" when the actual file that
we will look up on the next scan is "foo". This moves the Chtimes later,
ensuring that it gets recorded correctly under the right name.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3519
These are no longer required with Go 1.7. Change made by removing the
functions, doing a global s/osutil.Remove/os.Remove/.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3514
So there were some issues here. The main problem was that
model.Close(deviceID) was overloaded to mean "the connection was closed
by the protocol layer" and "i want to close this connection". That meant
it could get called twice - once *to* close the connection and then once
more when the connection *was* closed.
After this refactor there is instead a Closed(conn) method that is the
callback. I didn't need to change the parameter in the end, but I think
it's clearer what it means when it takes the connection that was closed
instead of a device ID. To close a connection, the new close(deviceID)
method is used instead, which only closes the underlying connection and
leaves the cleanup to the Closed() callback.
I also changed how we do connection switching. Instead of the connection
service calling close and then adding the connection, it just adds the
new connection. The model knows that it already has a connection and
makes sure to close and clean out that one before adding the new
connection.
To make sure to sequence this properly I added a new map of channels
that get created on connection add and closed by Closed(), so that
AddConnection() can do the close and wait for the cleanup to happen
before proceeding.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3490
Furthermore:
1. Cleans configs received, migrates them as we receive them.
2. Clears indexes of devices we no longer share the folder with
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3478
It seems that it would be impossible to drop down to relay after establishing a direct connection
Also, we should not drop the existing connection until after we've passed the validation steps,
and it seems it's being dropped in two places unnecesserily at the moment.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3480
This adds a new nanoseconds field to the FileInfo, populates it during
scans and sets the non-truncated time in Chtimes calls.
The actual file modification time is defined as modified_s seconds +
modified_ns nanoseconds. It's expected that the modified_ns field is <=
1e9 (that is, all whole seconds should go in the modified_s field) but
not really enforced. Given that it's an int32 the timestamp can be
adjusted += ~2.9 seconds by the modified_ns field...
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3431
This adds a config to enable debug functions on the API server, which is
by default disabled. When enabled, the /rest/debug things become
available and become available without requiring a CSRF token (although
authentication is required if configured).
We also add a new endpoint /rest/debug/cpuprof?duration=15s (with the
duration being configurable, defaulting to 30s). This runs a CPU profile
for the duration and returns it as a file. It sets headers so that a
browser will save the file with an informative name.
The same is done for heap profiles, /rest/debug/heapprof, which does not
take any parameters.
The purpose of this is that any user can enable debugging under
advanced, then point their browser to the endpoint above and get a file
that contains a CPU or heap profile we can use, with the filename
telling us what version and architecture the profile is from.
On the command line, this becomes
curl -O -J http://localhost:8082/rest/debug/cpuprof?duration=5s
curl: Saved to filename
'syncthing-cpu-darwin-amd64-v0.14.3+4-g935bcc0-110307.pprof'
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3467
Otherwise if the file grows during scanning the block list will be out
of sync with the stated size and things get confused. We could fixup the
size afterwards based on the block list, but then we might see other
inconsistencies as the mtime should have changed to reflect the new size
etc. Better stick to the original state and let the next scan pick up
the change.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3442
We could have a file to sync with permissions rw------- but we'd create
the temp file with rw-rw-rw- minus umask, usually rw-r--r--. This
potentially exposes private data while the file is being synced.
Similarly, when ignorePerms was set and we were reusing a temp files we
would set the permissions to rw-r--r-- explicitly, potentially
overriding a strict umask that would otherwise have had the file be
rw-------.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3437
The previous commit loosened the locking around database updates.
Apparently that was not fine - what happens is that parallell updates
to the same file for different devices stomp on each others updates to
the global index, leaving it missing one of the two devices.
This lets us add message types in the future, for authentication or
other purposes, without completely breaking old clients. I see this as
similar behavior to adding fields to messages - newer clients must
simple be aware that older ones may ignore the message and act
accordingly.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3390
This slightly changes the interface used for committing configuration
changes. The two parts are now:
- VerifyConfiguration, which runs synchronously and locked, and can
abort the config change. These callbacks shouldn't *do* anything
apart from looking at the config changes and saying yes or no. No
change from previously.
- CommitConfiguration, which runs asynchronously (one goroutine per
call) *after* replacing the config and releasing any locks. Returning
false from these methods sets the "requires restart" flag, which now
lives in the config.Wrapper.
This should be deadlock free as the CommitConfiguration calls can take
as long as they like and can wait for locks to be released when they
need to tweak things. I think this should be safe compared to before as
the CommitConfiguration calls were always made from a random background
goroutine (typically one from the HTTP server), so it was always
concurrent with everything else anyway.
Hence the CommitResponse type is gone, instead you get an error back on
verification failure only, and need to explicitly check
w.RequiresRestart() afterwards if you care.
As an added bonus this fixes a bug where we would reset the "requires
restart" indicator if a config that did not require restart was saved,
even if we already were in the requires-restart state.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3386
A random "instance ID" is generated on each start of the local discovery
service. The instance ID is included in the announcement. When we see a
new instance ID we treat is a new device and respond with an
announcement of our own. Hence devices get to know each other quickly on
restart.
GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/3385
