Calling both mono.Now() and time.Now() is slow and
leads to unnecessary precision errors.
Instead, directly compute mono.Time relative to baseMono and baseWall.
This is the opposite calculation as mono.Time.WallTime.
Updates tailscale/corp#8427
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
There are two race conditions in output handling.
The first race condition is due to a misuse of exec.Cmd.StdoutPipe.
The documentation explicitly forbids concurrent use of StdoutPipe
with exec.Cmd.Wait (see golang/go#60908) because Wait will
close both sides of the pipe once the process ends without
any guarantees that all data has been read from the pipe.
To fix this, we allocate the os.Pipes ourselves and
manage cleanup ourselves when the process has ended.
The second race condition is because sshSession.run waits
upon exec.Cmd to finish and then immediately proceeds to call ss.Exit,
which will close all output streams going to the SSH client.
This may interrupt any asynchronous io.Copy still copying data.
To fix this, we close the write-side of the os.Pipes after
the process has finished (and before calling ss.Exit) and
synchronously wait for the io.Copy routines to finish.
Fixes#7601
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Co-authored-by: Maisem Ali <maisem@tailscale.com>
The retry logic was pathological in the following ways:
* If we restarted the logging service, any pending uploads
would be placed in a retry-loop where it depended on backoff.Backoff,
which was too aggresive. It would retry failures within milliseconds,
taking at least 10 retries to hit a delay of 1 second.
* In the event where a logstream was rate limited,
the aggressive retry logic would severely exacerbate the problem
since each retry would also log an error message.
It is by chance that the rate of log error spam
does not happen to exceed the rate limit itself.
We modify the retry logic in the following ways:
* We now respect the "Retry-After" header sent by the logging service.
* Lacking a "Retry-After" header, we retry after a hard-coded period of
30 to 60 seconds. This avoids the thundering-herd effect when all nodes
try reconnecting to the logging service at the same time after a restart.
* We do not treat a status 400 as having been uploaded.
This is simply not the behavior of the logging service.
Updates #tailscale/corp#11213
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Exclude traffic with 100.100.100.100 (for IPv4) and
with fd7a:115c:a1e0::53 (for IPv6) since this traffic with the
Tailscale service running locally on the node.
This traffic never left the node.
It also happens to be a high volume amount of traffic since
DNS requests occur over UDP with each request coming from a
unique port, thus resulting in many discrete traffic flows.
Fixestailscale/corp#10554
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The addition of WaitGroup.Go in the standard library has been
repeatedly proposed and rejected.
See golang/go#18022, golang/go#23538, and golang/go#39863
In summary, the argument for WaitGroup.Go is that it avoids bugs like:
go func() {
wg.Add(1)
defer wg.Done()
...
}()
where the increment happens after execution (not before)
and also (to a lesser degree) because:
wg.Go(func() {
...
})
is shorter and more readble.
The argument against WaitGroup.Go is that the provided function
takes no arguments and so inputs and outputs must closed over
by the provided function. The most common race bug for goroutines
is that the caller forgot to capture the loop iteration variable,
so this pattern may make it easier to be accidentally racy.
However, that is changing with golang/go#57969.
In my experience the probability of race bugs due to the former
still outwighs the latter, but I have no concrete evidence to prove it.
The existence of errgroup.Group.Go and frequent utility of the method
at least proves that this is a workable pattern and
the possibility of accidental races do not appear to
manifest as frequently as feared.
A reason *not* to use errgroup.Group everywhere is that there are many
situations where it doesn't make sense for the goroutine to return an error
since the error is handled in a different mechanism
(e.g., logged and ignored, formatted and printed to the frontend, etc.).
While you can use errgroup.Group by always returning nil,
the fact that you *can* return nil makes it easy to accidentally return
an error when nothing is checking the return of group.Wait.
This is not a hypothetical problem, but something that has bitten us
in usages that was only using errgroup.Group without intending to use
the error reporting part of it.
Thus, add a (yet another) variant of WaitGroup here that
is identical to sync.WaitGroup, but with an extra method.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Add Value, which measures the rate at which an event occurs,
exponentially weighted towards recent activity.
It is guaranteed to occupy O(1) memory, operate in O(1) runtime,
and is safe for concurrent use.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The time.Parse function has been optimized to the point
where it is faster than our custom implementation.
See upstream changes in:
* https://go.dev/cl/429862
* https://go.dev/cl/425197
* https://go.dev/cl/425116
Performance:
BenchmarkGoParse3339/Z 38.75 ns/op 0 B/op 0 allocs/op
BenchmarkGoParse3339/TZ 54.02 ns/op 0 B/op 0 allocs/op
BenchmarkParse3339/Z 40.17 ns/op 0 B/op 0 allocs/op
BenchmarkParse3339/TZ 87.06 ns/op 0 B/op 0 allocs/op
We can see that the stdlib implementation is now faster.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Share the same underlying implementation for both PrivateID and PublicID.
For the shared methods, declare them in the same order.
Only keep documentation on methods without obvious meaning.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Ever since the introduction of the "must" package,
most MustXXX functions are no longer necessary.
Remove this as it is no longer depended upon
from within this repository and by the internal private repository.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The log ID types were moved to a separate package so that
code that only depend on log ID types do not need to link
in the logic for the logtail client itself.
Not all code need the logtail client.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The Tailscale logging service has a hard limit on the maximum
log message size that can be accepted.
We want to ensure that netlog messages never exceed
this limit otherwise a client cannot transmit logs.
Move the goroutine for periodically dumping netlog messages
from wgengine/netlog to net/connstats.
This allows net/connstats to manage when it dumps messages,
either based on time or by size.
Updates tailscale/corp#8427
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Consider the following pattern:
err1 := foo()
err2 := bar()
err3 := baz()
return multierr.New(err1, err2, err3)
If err1, err2, and err3 are all nil, then multierr.New should not allocate.
Thus, modify the logic of New to count the number of distinct error values
and allocate the exactly needed slice. This also speeds up non-empty error
situation since repeatedly growing with append is slow.
Performance:
name old time/op new time/op delta
Empty-24 41.8ns ± 2% 6.4ns ± 1% -84.73% (p=0.000 n=10+10)
NonEmpty-24 120ns ± 3% 69ns ± 1% -42.01% (p=0.000 n=9+10)
name old alloc/op new alloc/op delta
Empty-24 64.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
NonEmpty-24 168B ± 0% 88B ± 0% -47.62% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
Empty-24 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
NonEmpty-24 3.00 ± 0% 2.00 ± 0% -33.33% (p=0.000 n=10+10)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Use multierr.Range to iterate through an error tree
instead of multiple invocations of errors.As.
This scales better as we add more Go error types to the switch.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Errors in Go are no longer viewed as a linear chain, but a tree.
See golang/go#53435.
Add a Range function that iterates through an error
in a pre-order, depth-first order.
This matches the iteration order of errors.As in Go 1.20.
This adds the logic (but currently commented out) for having
Error implement the multi-error version of Unwrap in Go 1.20.
It is commented out currently since it causes "go vet"
to complain about having the "wrong" signature.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Previously, tstun.Wrapper and magicsock.Conn managed their
own statistics data structure and relied on an external call to
Extract to extract (and reset) the statistics.
This makes it difficult to ensure a maximum size on the statistics
as the caller has no introspection into whether the number
of unique connections is getting too large.
Invert the control flow such that a *connstats.Statistics
is registered with tstun.Wrapper and magicsock.Conn.
Methods on non-nil *connstats.Statistics are called for every packet.
This allows the implementation of connstats.Statistics (in the future)
to better control when it needs to flush to ensure
bounds on maximum sizes.
The value registered into tstun.Wrapper and magicsock.Conn could
be an interface, but that has two performance detriments:
1. Method calls on interface values are more expensive since
they must go through a virtual method dispatch.
2. The implementation would need a sync.Mutex to protect the
statistics value instead of using an atomic.Pointer.
Given that methods on constats.Statistics are called for every packet,
we want reduce the CPU cost on this hot path.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Many packages reference the logtail ID types,
but unfortunately pull in the transitive dependencies of logtail.
Fix this problem by putting the log ID types in its own package
with minimal dependencies.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The cutset provided to strings.TrimRight was missing the digit '6',
making it such that we couldn't parse something like "365d".
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This function is no longer necessary as you can trivially rewrite:
logtail.MustParsePublicID(...)
with:
must.Get(logtail.ParsePublicID(...))
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Map is a concurrent safe map that is a trivial wrapper
over a Go map and a sync.RWMutex.
It is optimized for use-cases where the entries change often,
which is the opposite use-case of what sync.Map is optimized for.
The API is patterned off of sync.Map, but made generic.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
There is a finite limit to the maximum message size that logtail can upload.
We need to make sure network logging messages remain under this size.
These constants allow us to compute the maximum number of ConnectionCounts
we can buffer before we must flush.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Exit node traffic is aggregated to protect the privacy
of those using an exit node. However, it is reasonable to
at least log which nodes are making most use of an exit node.
For a node using an exit node,
the source will be the taiscale IP address of itself,
while the destination will be zeroed out.
For a node that serves as an exit node,
the source will be zeroed out,
while the destination will be tailscale IP address
of the node that initiated the exit traffic.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
In the future this will cause a node to be unable to join the tailnet
if network logging is enabled.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
If the network logging configruation changes (and nothing else)
we will tear down the network logger and start it back up.
However, doing so will lose the router configuration state.
Manually reconfigure it with the routing state.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Make netlogfmt useful regardless of the exact schema of the input.
If a JSON object looks like a network log message,
then unmarshal it as one and then print it.
This allows netlogfmt to support both a stream of JSON objects
directly serialized from netlogtype.Message, or the schema
returned by the /api/v2/tailnet/{{tailnet}}/network-logs API endpoint.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This is a temporary hack to prevent logtail getting stuck
uploading the same excessive message over and over.
A better solution will be discussed and implemented.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
There is utility in logging traffic statistics that occurs at the physical layer.
That is, in order to send packets virtually to a particular tailscale IP address,
what physical endpoints did we need to communicate with?
This functionality logs IP addresses identical to
what had always been logged in magicsock prior to #5823,
so there is no increase in PII being logged.
ExtractStatistics returns a mapping of connections to counts.
The source is always a Tailscale IP address (without port),
while the destination is some endpoint reachable on WAN or LAN.
As a special case, traffic routed through DERP will use 127.3.3.40
as the destination address with the port being the DERP region.
This entire feature is only enabled if data-plane audit logging
is enabled on the tailnet (by default it is disabled).
Example of type of information logged:
------------------------------------ Tx[P/s] Tx[B/s] Rx[P/s] Rx[B/s]
PhysicalTraffic: 25.80 3.39Ki 38.80 5.57Ki
100.1.2.3 -> 143.11.22.33:41641 15.40 2.00Ki 23.20 3.37Ki
100.4.5.6 -> 192.168.0.100:41641 10.20 1.38Ki 15.60 2.20Ki
100.7.8.9 -> 127.3.3.40:2 0.20 6.40 0.00 0.00
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The wireguard-go code unfortunately calls this unconditionally
even when verbose logging is disabled.
Partial revert of #5911.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This package parses a JSON stream of netlog.Message from os.Stdin
and pretty prints the contents as a stream of tables.
It supports reverse lookup of tailscale IP addresses if given
an API key and the tailnet that these traffic logs belong to.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This field seems seldom used and the documentation is wrong.
It is simpler to just derive its original value dynamically
when endpoint.DstToString is called.
This method is potentially used by wireguard-go,
but not in any code path is performance sensitive.
All calls to it use it in conjunction with fmt.Printf,
which is going to be slow anyways since it uses Go reflection.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This sets up Logger to handle statistics at the magicsock layer,
where we can correlate traffic between a particular tailscale IP address
and any number of physical endpoints used to contact the node
that hosts that tailscale address.
We also export Message and TupleCounts to better document the JSON format
that is being sent to the logging infrastructure.
This commit does NOT yet enable the actual logging of magicsock statistics.
That will be a future commit.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
If the wgcfg.Config is specified with network logging arguments,
then Userspace.Reconfig starts up an asynchronous network logger,
which is shutdown either upon Userspace.Close or when Userspace.Reconfig
is called again without network logging or route arguments.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The Logger type managers a logtail.Logger for extracting
statistics from a tstun.Wrapper.
So long as Shutdown is called, it ensures that logtail
and statistic gathering resources are properly cleared up.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The node and domain audit log IDs are provided in the map response,
but are ultimately going to be used in wgengine since
that's the layer that manages the tstun.Wrapper.
Do the plumbing work to get this field passed down the stack.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Rename StatisticsEnable as SetStatisticsEnabled to be consistent
with other similarly named methods.
Rename StatisticsExtract as ExtractStatistics to follow
the convention where methods start with a verb.
It was originally named with Statistics as a prefix so that
statistics related methods would sort well in godoc,
but that property no longer holds.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Upstream optimizations to the Go time package will make
unmarshaling of time.Time 3-6x faster. See:
* https://go.dev/cl/425116
* https://go.dev/cl/425197
* https://go.dev/cl/429862
The last optimization avoids a []byte -> string allocation
if the timestamp string less than than 32B.
Unfortunately, the presence of a timezone breaks that optimization.
Drop recording of timezone as this is non-essential information.
Most of the performance gains is upon unmarshal,
but there is also a slight performance benefit to
not marshaling the timezone as well.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The copy ID operates similar to a CC in email where
a message is sent to both the primary ID and also the copy ID.
A given log message is uploaded once, but the log server
records it twice for each ID.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
If Wrapper.StatisticsEnable is enabled,
then per-connection counters are maintained.
If enabled, Wrapper.StatisticsExtract must be periodically called
otherwise there is unbounded memory growth.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
High-level API:
type Statistics struct { ... }
type Counts struct { TxPackets, TxBytes, RxPackets, RxBytes uint64 }
func (*Statistics) UpdateTx([]byte)
func (*Statistics) UpdateRx([]byte)
func (*Statistics) Extract() map[flowtrack.Tuple]Counts
The API accepts a []byte instead of a packet.Parsed so that a future
implementation can directly hash the address and port bytes,
which are contiguous in most IP packets.
This will be useful for a custom concurrent-safe hashmap implementation.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Callers of LogHost often jump through hoops to undo the
loss of information dropped by LogHost (e.g., the HTTP scheme).
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The version.CmdName implementation is buggy such that it does not correctly
identify the binary name if it embeds other go binaries.
For now, add a NewWithConfigPath API that allows the caller to explicitly
specify this information.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
It is unclear whether the lack of checking nil-ness of slices
was an oversight or a deliberate feature.
Lacking a comment, the assumption is that this was an oversight.
Also, expand the logic to perform cycle detection for recursive slices.
We do this on a per-element basis since a slice is semantically
equivalent to a list of pointers.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Add a new lookupTypeHasher function that is just a cached front-end
around the makeTypeHasher function.
We do not need to worry about the recursive type cycle issue that
made getTypeInfo more complicated since makeTypeHasher
is not directly recursive. All calls to itself happen lazily
through a sync.Once upon first use.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The entry logic of Hash has extra complexity to make sure
we always have an addressable value on hand.
If not, we heap allocate the input.
For this reason we document that there are performance benefits
to always providing a pointer.
Rather than documenting this, just enforce it through generics.
Also, delete the unused HasherForType function.
It's an interesting use of generics, but not well tested.
We can resurrect it from code history if there's a need for it.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This helps pprof better identify which Go kinds take the most time
since the kind is always in the function name.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This helps pprof better identify which Go kinds take the most time
since the kind is always in the function name.
There is a minor adjustment where we hash the length of the map
to be more on the cautious side.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Rather than having two copies []fieldInfo,
just maintain one and perform merging in the same pass.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This helps pprof better identify which Go kinds take the most time
since the kind is always in the function name.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Use of reflect.Value.SetXXX panics if the provided argument was
obtained from an unexported struct field.
Instead, pass an unsafe.Pointer around and convert to a
reflect.Value when necessary (i.e., for maps and interfaces).
Converting from unsafe.Pointer to reflect.Value guarantees that
none of the read-only bits will be populated.
When running in race mode, we attach type information to the pointer
so that we can type check every pointer operation.
This also type-checks that direct memory hashing is within
the valid range of a struct value.
We add test cases that previously caused deephash to panic,
but now pass.
Performance:
name old time/op new time/op delta
Hash 14.1µs ± 1% 14.1µs ± 1% ~ (p=0.590 n=10+9)
HashPacketFilter 2.53µs ± 2% 2.44µs ± 1% -3.79% (p=0.000 n=9+10)
TailcfgNode 1.45µs ± 1% 1.43µs ± 0% -1.36% (p=0.000 n=9+9)
HashArray 318ns ± 2% 318ns ± 2% ~ (p=0.541 n=10+10)
HashMapAcyclic 32.9µs ± 1% 31.6µs ± 1% -4.16% (p=0.000 n=10+9)
There is a slight performance gain due to the use of unsafe.Pointer
over reflect.Value methods. Also, passing an unsafe.Pointer (1 word)
on the stack is cheaper than passing a reflect.Value (3 words).
Performance gains are diminishing since SHA-256 hashing now dominates the runtime.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
When built with "deephash_debug", print the set of HashXXX methods.
Example usage:
$ go test -run=GetTypeHasher/string_slice -tags=deephash_debug
U64(2)+U64(3)+S("foo")+U64(3)+S("bar")+FIN
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Rather than separate functions to hash each kind,
just rely on the fact that these are direct memory hashable,
thus simplifying the code.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Every implementation of typeHasherFunc always returns true,
which implies that the slow path is no longer executed.
Delete it.
h.hashValueWithType(v, ti, ...) is deleted as it is equivalent to:
ti.hasher()(h, v)
h.hashValue(v, ...) is deleted as it is equivalent to:
ti := getTypeInfo(v.Type())
ti.hasher()(h, v)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Add support for maps and interfaces to the fast path.
Add cycle-detection to the pointer handling logic.
This logic is mostly copied from the slow path.
A future commit will delete the slow path once
the fast path never falls back to the slow path.
Performance:
name old time/op new time/op delta
Hash-24 18.5µs ± 1% 14.9µs ± 2% -19.52% (p=0.000 n=10+10)
HashPacketFilter-24 2.54µs ± 1% 2.60µs ± 1% +2.19% (p=0.000 n=10+10)
HashMapAcyclic-24 31.6µs ± 1% 30.5µs ± 1% -3.42% (p=0.000 n=9+8)
TailcfgNode-24 1.44µs ± 2% 1.43µs ± 1% ~ (p=0.171 n=10+10)
HashArray-24 324ns ± 1% 324ns ± 2% ~ (p=0.425 n=9+9)
The additional cycle detection logic doesn't incur much slow down
since it only activates if a type is recursive, which does not apply
for any of the types that we care about.
There is a notable performance boost since we switch from the fath path
to the slow path less often. Most notably, a struct with a field that
could not be handled by the fast path would previously cause
the entire struct to go through the slow path.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
There are 5 types that we care about that implement AppendTo:
key.DiscoPublic
key.NodePublic
netip.Prefix
netipx.IPRange
netip.Addr
The key types are thin wrappers around [32]byte and are memory hashable.
The netip.Prefix and netipx.IPRange types are thin wrappers over netip.Addr
and are hashable by default if netip.Addr is hashable.
The netip.Addr type is the only one with a complex structure where
the default behavior of deephash does not hash it correctly due to the presence
of the intern.Value type.
Drop support for AppendTo and instead add specialized hashing for netip.Addr
that would be semantically equivalent to == on the netip.Addr values.
The AppendTo support was already broken prior to this change.
It was fully removed (intentionally or not) in #4870.
It was partially restored in #4858 for the fast path,
but still broken in the slow path.
Just drop support for it altogether.
This does mean we lack any ability for types to self-hash themselves.
In the future we can add support for types that implement:
interface { DeepHash() Sum }
Test and fuzz cases were added for the relevant types that
used to rely on the AppendTo method.
FuzzAddr has been executed on 1 billion samples without issues.
Signed-off-by: Joe Tsai joetsai@digital-static.net
Rename Hash as Block512 to indicate that this is a general-purpose
hash.Hash for any algorithm that operates on 512-bit block sizes.
While we rename the package as hashx in this commit,
a subsequent commit will move the sha256x package to hashx.
This is done separately to avoid confusing git.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Also, rename canMemHash to typeIsMemHashable to be consistent.
There are zero changes to the semantics.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Any type that is memory hashable must not be recursive since
there are definitely no pointers involved to make a cycle.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Put the t.Size() == 0 check first since this is applicable in all cases.
Drop the last struct field conditional since this is covered by the
sumFieldSize check at the end.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Hashing []any is slow since hashing of interfaces is slow.
Hashing of interfaces is slow since we pessimistically assume
that cycles can occur through them and start cycle tracking.
Drop the variadic signature of Update and fix callers to pass in
an anonymous struct so that we are hashing concrete types
near the root of the value tree.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Formatting a time.Time as RFC3339 is slow.
See https://go.dev/issue/54093
Now that we have efficient hashing of fixed-width integers,
just hash the time.Time as a binary value.
Performance:
Hash-24 19.0µs ± 1% 18.6µs ± 1% -2.03% (p=0.000 n=10+9)
TailcfgNode-24 1.79µs ± 1% 1.40µs ± 1% -21.74% (p=0.000 n=10+9)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Switch deephash to use sha256x.Hash.
We add sha256x.HashString to efficiently hash a string.
It uses unsafe under the hood to convert a string to a []byte.
We also modify sha256x.Hash to export the underlying hash.Hash
for testing purposes so that we can intercept all hash.Hash calls.
Performance:
name old time/op new time/op delta
Hash-24 19.8µs ± 1% 19.2µs ± 1% -3.01% (p=0.000 n=10+10)
HashPacketFilter-24 2.61µs ± 0% 2.53µs ± 1% -3.01% (p=0.000 n=8+10)
HashMapAcyclic-24 31.3µs ± 1% 29.8µs ± 0% -4.80% (p=0.000 n=10+9)
TailcfgNode-24 1.83µs ± 1% 1.82µs ± 2% ~ (p=0.305 n=10+10)
HashArray-24 344ns ± 2% 323ns ± 1% -6.02% (p=0.000 n=9+10)
The performance gains is not as dramatic as sha256x over sha256 due to:
1. most of the hashing already occurring through the direct memory hashing logic, and
2. what does not go through direct memory hashing is slowed down by reflect.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
In Go 1.19, the reflect.Value.MapRange method uses "function outlining"
so that the allocation of reflect.MapIter is inlinable by the caller.
If the iterator doesn't escape the caller, it can be stack allocated.
See https://go.dev/cl/400675
Performance:
name old time/op new time/op delta
HashMapAcyclic-24 31.9µs ± 2% 32.1µs ± 1% ~ (p=0.075 n=10+10)
name old alloc/op new alloc/op delta
HashMapAcyclic-24 0.00B 0.00B ~ (all equal)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The hash.Hash provided by sha256.New is much more efficient
if we always provide it with data a multiple of the block size.
This avoids double-copying of data into the internal block
of sha256.digest.x. Effectively, we are managing a block ourselves
to ensure we only ever call hash.Hash.Write with full blocks.
Performance:
name old time/op new time/op delta
Hash 33.5µs ± 1% 20.6µs ± 1% -38.40% (p=0.000 n=10+9)
The logic has gone through CPU-hours of fuzzing.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The logic of deephash is both simpler and easier to reason about
if values are always addressable.
In Go, the composite kinds are slices, arrays, maps, structs,
interfaces, pointers, channels, and functions,
where we define "composite" as a Go value that encapsulates
some other Go value (e.g., a map is a collection of key-value entries).
In the cases of pointers and slices, the sub-values are always addressable.
In the cases of arrays and structs, the sub-values are always addressable
if and only if the parent value is addressable.
In the case of maps and interfaces, the sub-values are never addressable.
To make them addressable, we need to copy them onto the heap.
For the purposes of deephash, we do not care about channels and functions.
For all non-composite kinds (e.g., strings and ints), they are only addressable
if obtained from one of the composite kinds that produce addressable values
(i.e., pointers, slices, addressable arrays, and addressable structs).
A non-addressible, non-composite kind can be made addressable by
allocating it on the heap, obtaining a pointer to it, and dereferencing it.
Thus, if we can ensure that values are addressable at the entry points,
and shallow copy sub-values whenever we encounter an interface or map,
then we can ensure that all values are always addressable and
assume such property throughout all the logic.
Performance:
name old time/op new time/op delta
Hash-24 21.5µs ± 1% 19.7µs ± 1% -8.29% (p=0.000 n=9+9)
HashPacketFilter-24 2.61µs ± 1% 2.62µs ± 0% +0.29% (p=0.037 n=10+9)
HashMapAcyclic-24 30.8µs ± 1% 30.9µs ± 1% ~ (p=0.400 n=9+10)
TailcfgNode-24 1.84µs ± 1% 1.84µs ± 2% ~ (p=0.928 n=10+10)
HashArray-24 324ns ± 2% 332ns ± 2% +2.45% (p=0.000 n=10+10)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The Do function assists in calling functions that must succeed.
It only interacts well with functions that return (T, err).
Signatures with more return arguments are not supported.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
It should be safe to initialize multiple Server instances
without any resource leaks what-so-ever.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The Header field allows the server to specify specific headers to set.
Example use case: server returns 429 with the "Retry-After" header set.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
If the field is the zero value, then avoid serializing the field.
This reduces verbosity in server logs.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
If ConfigFromFile cannot find the configuration file,
we must not initialize it with NewConfig.
Instead, we need it to fail validation so that it eventually writes
a newly constructed configuration file.
Otherwise, new tailscale instances will never be able store a persistent
log config and start with a new config file upon every bootup.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Use the "tailscaled" prefix instead of "tsnet" for state file names:
1. It is consistent with the pre-existing {{Dir}}/tailscaled.state file.
2. It makes the file layout of `tsnet` and `tailscaled` identical,
so that they are compatible with each other.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
It is not idiomatic for Go code to panic for situations that
can be normal. For example, if a server receives PrivateID
from a client, it is normal for the server to call
PrivateID.PublicID to validate that the PublicID matches.
However, doing so would panic prior to this change.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The hujson package transition to just being a pure AST
parser and formatter for HuJSON and not an unmarshaler.
Thus, parse HuJSON as such, convert it to JSON,
and then use the standard JSON unmarshaler.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
A large influx of new connections can bring down DERP
since it spins off a new goroutine for each connection,
where each routine may do significant amount of work
(e.g., allocating memory and crunching numbers for TLS crypto).
The momentary spike can cause the process to OOM.
This commit sets the groundwork for limiting connections,
but leaves the limit at infinite by default.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The new /keys endpoint allows you to list API and machine auth keys.
You can also create machine auth key.
It currently does not support creating another API key.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The "go generate" command blindly looks for "//go:generate" anywhere
in the file regardless of whether it is truly a comment.
Prevent this false positive in cloner.go by mangling the string
to look less like "//go:generate".
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Emit a go:generate pragma with the full set of flags passed to cloner.
This allows the user to simply run "go generate" at the location
of the generate file to reproduce the file.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Unfortunately this test fails on certain architectures.
The problem comes down to inconsistencies in the Go escape analysis
where specific variables are marked as escaping on certain architectures.
The variables escaping to the heap are unfortunately in crypto/sha256,
which makes it impossible to fixthis locally in deephash.
For now, fix the test by compensating for the allocations that
occur from calling sha256.digest.Sum.
See golang/go#48055
Fixes#2727
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
This test is highly dependent on the accuracy of OS timers.
Reduce the number of failures by decreasing the required
accuracy from 0.999 to 0.995.
Also, switch from repeated time.Sleep to using a time.Ticker
for improved accuracy.
Updates #2727
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The index for every struct field or slice element and
the number of fields for the struct is unncessary.
The hashing of Go values is unambiguous because every type (except maps)
encodes in a parsable manner. So long as we know the type information,
we could theoretically decode every value (except for maps).
At a high level:
* numbers are encoded as fixed-width records according to precision.
* strings (and AppendTo output) are encoded with a fixed-width length,
followed by the contents of the buffer.
* slices are prefixed by a fixed-width length, followed by the encoding
of each value. So long as we know the type of each element, we could
theoretically decode each element.
* arrays are encoded just like slices, but elide the length
since it is determined from the Go type.
* maps are encoded first with a byte indicating whether it is a cycle.
If a cycle, it is followed by a fixed-width index for the pointer,
otherwise followed by the SHA-256 hash of its contents. The encoding of maps
is not decodeable, but a SHA-256 hash is sufficient to avoid ambiguities.
* interfaces are encoded first with a byte indicating whether it is nil.
If not nil, it is followed by a fixed-width index for the type,
and then the encoding for the underlying value. Having the type be encoded
first ensures that the value could theoretically be decoded next.
* pointers are encoded first with a byte indicating whether it is
1) nil, 2) a cycle, or 3) newly seen. If a cycle, it is followed by
a fixed-width index for the pointer. If newly seen, it is followed by
the encoding for the pointed-at value.
Removing unnecessary details speeds up hashing:
name old time/op new time/op delta
Hash-8 76.0µs ± 1% 55.8µs ± 2% -26.62% (p=0.000 n=10+10)
HashMapAcyclic-8 61.9µs ± 0% 62.0µs ± 0% ~ (p=0.666 n=9+9)
TailcfgNode-8 10.2µs ± 1% 7.5µs ± 1% -26.90% (p=0.000 n=10+9)
HashArray-8 1.07µs ± 1% 0.70µs ± 1% -34.67% (p=0.000 n=10+9)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Instead of hashing the humanly formatted forms of a number,
hash the native machine bits of the integers themselves.
There is a small performance gain for this:
name old time/op new time/op delta
Hash-8 75.7µs ± 1% 76.0µs ± 2% ~ (p=0.315 n=10+9)
HashMapAcyclic-8 63.1µs ± 3% 61.3µs ± 1% -2.77% (p=0.000 n=10+10)
TailcfgNode-8 10.3µs ± 1% 10.2µs ± 1% -1.48% (p=0.000 n=10+10)
HashArray-8 1.07µs ± 1% 1.05µs ± 1% -1.79% (p=0.000 n=10+10)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The swapping of bufio.Writer between hasher and mapHasher is subtle.
Just embed a hasher in mapHasher to avoid complexity here.
No notable change in performance:
name old time/op new time/op delta
Hash-8 76.7µs ± 1% 77.0µs ± 1% ~ (p=0.182 n=9+10)
HashMapAcyclic-8 62.4µs ± 1% 62.5µs ± 1% ~ (p=0.315 n=10+9)
TailcfgNode-8 10.3µs ± 1% 10.3µs ± 1% -0.62% (p=0.004 n=10+9)
HashArray-8 1.07µs ± 1% 1.06µs ± 1% -0.98% (p=0.001 n=8+9)
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
The previous algorithm used a map of all visited pointers.
The strength of this approach is that it quickly prunes any nodes
that we have ever visited before. The detriment of the approach
is that pruning is heavily dependent on the order that pointers
were visited. This is especially relevant for hashing a map
where map entries are visited in a non-deterministic manner,
which would cause the map hash to be non-deterministic
(which defeats the point of a hash).
This new algorithm uses a stack of all visited pointers,
similar to how github.com/google/go-cmp performs cycle detection.
When we visit a pointer, we push it onto the stack, and when
we leave a pointer, we pop it from the stack.
Before visiting a pointer, we first check whether the pointer exists
anywhere in the stack. If yes, then we prune the node.
The detriment of this approach is that we may hash a node more often
than before since we do not prune as aggressively.
The set of visited pointers up until any node is only the
path of nodes up to that node and not any other pointers
that may have been visited elsewhere. This provides us
deterministic hashing regardless of visit order.
We can now delete hashMapFallback and associated complexity,
which only exists because the previous approach was non-deterministic
in the presence of cycles.
This fixes a failure of the old algorithm where obviously different
values are treated as equal because the pruning was too aggresive.
See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534
The new algorithm is slightly slower since it prunes less aggresively:
name old time/op new time/op delta
Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19)
HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19)
TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17)
HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19)
However, a slower but more correct algorithm seems
more favorable than a faster but incorrect algorithm.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
A Go interface may hold any number of different concrete types.
Just because two underlying values hash to the same thing
does not mean the two values are identical if they have different
concrete types. As such, include the type in the hash.
Seed the hash upon first use with the current time.
This ensures that the stability of the hash is bounded within
the lifetime of one program execution.
Hopefully, this prevents future bugs where someone assumes that
this hash is stable.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Joe TsaiandGitHub
