avoid breaking intrinsics when obfuscating names
We obfuscate import paths as well as their declared names.
The compiler treats some packages and APIs in special ways,
and the way it detects those is by looking at import paths and names.
In the past, we have avoided obfuscating some names like embed.FS or
reflect.Value.MethodByName for this reason. Otherwise,
go:embed or the linker's deadcode elimination might be broken.
This matching by path and name also happens with compiler intrinsics.
Intrinsics allow the compiler to rewrite some standard library calls
with small and efficient assembly, depending on the target GOARCH.
For example, math/bits.TrailingZeros32 gets replaced with ssa.OpCtz32,
which on amd64 may result in using the TZCNTL instruction.
We never noticed that we were breaking many of these intrinsics.
The intrinsics for funcs declared in the runtime and its dependencies
still worked properly, as we do not obfuscate those packages yet.
However, for other packages like math/bits and sync/atomic,
the intrinsics were being entirely disabled due to obfuscated names.
Skipping intrinsics is particularly bad for performance,
and it also leads to slightly larger binaries:
│ old │ new │
│ bin-B │ bin-B vs base │
Build-16 5.450Mi ± ∞ ¹ 5.333Mi ± ∞ ¹ -2.15% (p=0.029 n=4)
Finally, the main reason we noticed that intrinsics were broken
is that apparently GOARCH=mips fails to link without them,
as some symbols end up being not defined at all.
This patch fixes builds for the MIPS family of architectures.
Rather than building and linking all of std for every GOARCH,
test that intrinsics work by asking the compiler to print which
intrinsics are being applied, and checking that math/bits gets them.
This fix is relatively unfortunate, as it means we stop obfuscating
about 120 function names and a handful of package paths.
However, fixing builds and intrinsics is much more important.
We can figure out better ways to deal with intrinsics in the future.
Fixes #646.
1 year ago
|
|
|
# A fairly average Go build, importing some std libraries.
|
|
|
|
|
# We always build for a foreign GOOS.
|
|
|
|
|
# GOOS=windows, unless the host is also windows; then linux.
|
|
|
|
|
# GOARCH=arm, unless the host is also arm; then amd64.
|
|
|
|
|
# Windows and ARM are both interesting,
|
|
|
|
|
# and it helps with coverage as we mainly test on linux/amd64.
|
avoid breaking intrinsics when obfuscating names
We obfuscate import paths as well as their declared names.
The compiler treats some packages and APIs in special ways,
and the way it detects those is by looking at import paths and names.
In the past, we have avoided obfuscating some names like embed.FS or
reflect.Value.MethodByName for this reason. Otherwise,
go:embed or the linker's deadcode elimination might be broken.
This matching by path and name also happens with compiler intrinsics.
Intrinsics allow the compiler to rewrite some standard library calls
with small and efficient assembly, depending on the target GOARCH.
For example, math/bits.TrailingZeros32 gets replaced with ssa.OpCtz32,
which on amd64 may result in using the TZCNTL instruction.
We never noticed that we were breaking many of these intrinsics.
The intrinsics for funcs declared in the runtime and its dependencies
still worked properly, as we do not obfuscate those packages yet.
However, for other packages like math/bits and sync/atomic,
the intrinsics were being entirely disabled due to obfuscated names.
Skipping intrinsics is particularly bad for performance,
and it also leads to slightly larger binaries:
│ old │ new │
│ bin-B │ bin-B vs base │
Build-16 5.450Mi ± ∞ ¹ 5.333Mi ± ∞ ¹ -2.15% (p=0.029 n=4)
Finally, the main reason we noticed that intrinsics were broken
is that apparently GOARCH=mips fails to link without them,
as some symbols end up being not defined at all.
This patch fixes builds for the MIPS family of architectures.
Rather than building and linking all of std for every GOARCH,
test that intrinsics work by asking the compiler to print which
intrinsics are being applied, and checking that math/bits gets them.
This fix is relatively unfortunate, as it means we stop obfuscating
about 120 function names and a handful of package paths.
However, fixing builds and intrinsics is much more important.
We can figure out better ways to deal with intrinsics in the future.
Fixes #646.
1 year ago
|
|
|
#
|
|
|
|
|
# We also ensure that intrinsics work as expected.
|
|
|
|
|
# The compiler replaces calls to some functions with intrinsics in its ssa stage,
|
|
|
|
|
# and it recognizes which functions via the package path and func name.
|
|
|
|
|
# If we obfuscate those package paths without adjusting the compiler,
|
|
|
|
|
# intrinsics aren't applied, causing performance loss or build errors.
|
|
|
|
|
# We use the math/bits package, as its Len64 intrinsic is present in both arm
|
|
|
|
|
# and arm64, and it is not part of the runtime nor its dependencies.
|
|
|
|
|
[!windows] env GOOS=windows
|
|
|
|
|
[windows] env GOOS=linux
|
|
|
|
|
[!arm] env GOARCH=arm
|
|
|
|
|
[arm] env GOARCH=arm64
|
|
|
|
|
exec garble build -gcflags=math/bits=-d=ssa/intrinsics/debug=1
|
avoid breaking intrinsics when obfuscating names
We obfuscate import paths as well as their declared names.
The compiler treats some packages and APIs in special ways,
and the way it detects those is by looking at import paths and names.
In the past, we have avoided obfuscating some names like embed.FS or
reflect.Value.MethodByName for this reason. Otherwise,
go:embed or the linker's deadcode elimination might be broken.
This matching by path and name also happens with compiler intrinsics.
Intrinsics allow the compiler to rewrite some standard library calls
with small and efficient assembly, depending on the target GOARCH.
For example, math/bits.TrailingZeros32 gets replaced with ssa.OpCtz32,
which on amd64 may result in using the TZCNTL instruction.
We never noticed that we were breaking many of these intrinsics.
The intrinsics for funcs declared in the runtime and its dependencies
still worked properly, as we do not obfuscate those packages yet.
However, for other packages like math/bits and sync/atomic,
the intrinsics were being entirely disabled due to obfuscated names.
Skipping intrinsics is particularly bad for performance,
and it also leads to slightly larger binaries:
│ old │ new │
│ bin-B │ bin-B vs base │
Build-16 5.450Mi ± ∞ ¹ 5.333Mi ± ∞ ¹ -2.15% (p=0.029 n=4)
Finally, the main reason we noticed that intrinsics were broken
is that apparently GOARCH=mips fails to link without them,
as some symbols end up being not defined at all.
This patch fixes builds for the MIPS family of architectures.
Rather than building and linking all of std for every GOARCH,
test that intrinsics work by asking the compiler to print which
intrinsics are being applied, and checking that math/bits gets them.
This fix is relatively unfortunate, as it means we stop obfuscating
about 120 function names and a handful of package paths.
However, fixing builds and intrinsics is much more important.
We can figure out better ways to deal with intrinsics in the future.
Fixes #646.
1 year ago
|
|
|
stderr 'intrinsic substitution for Len64.*BitLen64'
|
wrap types.Importer to canonicalize import paths
The docs for go/importer.ForCompiler say:
The lookup function is called each time the resulting importer
needs to resolve an import path. In this mode the importer can
only be invoked with canonical import paths (not relative or
absolute ones); it is assumed that the translation to canonical
import paths is being done by the client of the importer.
We use a lookup func for two reasons: first, to support modules, and
second, to be able to use our information from "go list -json -export".
However, go/types does not canonicalize import paths before calling
ImportFrom. This is somewhat understandable; it doesn't know whether an
importer was created with a lookup func, and ImportFrom only requires
the input path to be canonicalized in that scenario. When the lookup
func is nil, the importer canonicalizes by itself via go/build.Import.
Before this change, the added crossbuild test would fail:
> garble build net/http
[stderr]
# vendor/golang.org/x/crypto/chacha20
typecheck error: /usr/lib/go/src/vendor/golang.org/x/crypto/chacha20/chacha_generic.go:10:2: could not import crypto/cipher (can't find import: "crypto/cipher")
# vendor/golang.org/x/text/secure/bidirule
typecheck error: /usr/lib/go/src/vendor/golang.org/x/text/secure/bidirule/bidirule.go:12:2: could not import errors (can't find import: "errors")
# vendor/golang.org/x/crypto/cryptobyte
typecheck error: /usr/lib/go/src/vendor/golang.org/x/crypto/cryptobyte/asn1.go:8:16: could not import encoding/asn1 (can't find import: "encoding/asn1")
# vendor/golang.org/x/text/unicode/norm
typecheck error: /usr/lib/go/src/vendor/golang.org/x/text/unicode/norm/composition.go:7:8: could not import unicode/utf8 (can't find import: "unicode/utf8")
This is because we'd fall back to importer.Default, which only knows how
to find packages in $GOROOT/pkg. Those are missing for cross-builds,
unsurprisingly, as those built archives end up in the build cache.
After this change, we properly support importing std-vendored packages,
so we can get rid of the importer.Default workaround. And, by extension,
cross-builds now work as well.
Note that, in the added test script, the full build of the binary fails,
as there seems to be some sort of linker problem:
> garble build
[stderr]
# test/main
d9rqJyxo.uoqIiDs5: relocation target runtime.os9A16A3 not defined
We leave that as a TODO for now, as this change is subtle enough as it
is.
3 years ago
|
|
|
-- go.mod --
|
|
|
|
|
module test/main
|
|
|
|
|
|
|
|
|
|
go 1.20
|
wrap types.Importer to canonicalize import paths
The docs for go/importer.ForCompiler say:
The lookup function is called each time the resulting importer
needs to resolve an import path. In this mode the importer can
only be invoked with canonical import paths (not relative or
absolute ones); it is assumed that the translation to canonical
import paths is being done by the client of the importer.
We use a lookup func for two reasons: first, to support modules, and
second, to be able to use our information from "go list -json -export".
However, go/types does not canonicalize import paths before calling
ImportFrom. This is somewhat understandable; it doesn't know whether an
importer was created with a lookup func, and ImportFrom only requires
the input path to be canonicalized in that scenario. When the lookup
func is nil, the importer canonicalizes by itself via go/build.Import.
Before this change, the added crossbuild test would fail:
> garble build net/http
[stderr]
# vendor/golang.org/x/crypto/chacha20
typecheck error: /usr/lib/go/src/vendor/golang.org/x/crypto/chacha20/chacha_generic.go:10:2: could not import crypto/cipher (can't find import: "crypto/cipher")
# vendor/golang.org/x/text/secure/bidirule
typecheck error: /usr/lib/go/src/vendor/golang.org/x/text/secure/bidirule/bidirule.go:12:2: could not import errors (can't find import: "errors")
# vendor/golang.org/x/crypto/cryptobyte
typecheck error: /usr/lib/go/src/vendor/golang.org/x/crypto/cryptobyte/asn1.go:8:16: could not import encoding/asn1 (can't find import: "encoding/asn1")
# vendor/golang.org/x/text/unicode/norm
typecheck error: /usr/lib/go/src/vendor/golang.org/x/text/unicode/norm/composition.go:7:8: could not import unicode/utf8 (can't find import: "unicode/utf8")
This is because we'd fall back to importer.Default, which only knows how
to find packages in $GOROOT/pkg. Those are missing for cross-builds,
unsurprisingly, as those built archives end up in the build cache.
After this change, we properly support importing std-vendored packages,
so we can get rid of the importer.Default workaround. And, by extension,
cross-builds now work as well.
Note that, in the added test script, the full build of the binary fails,
as there seems to be some sort of linker problem:
> garble build
[stderr]
# test/main
d9rqJyxo.uoqIiDs5: relocation target runtime.os9A16A3 not defined
We leave that as a TODO for now, as this change is subtle enough as it
is.
3 years ago
|
|
|
-- main.go --
|
|
|
|
|
package main
|
|
|
|
|
|
|
|
|
|
import "net/http"
|
|
|
|
|
|
|
|
|
|
func main() {
|
|
|
|
|
http.ListenAndServe("", nil)
|
|
|
|
|
}
|