vendor: bump runc to 1.0.1

The commands used were (roughly):

	hack/pin-dependency.sh github.com/opencontainers/runc v1.0.1
	hack/lint-dependencies.sh
	# Follow its recommendations.
	hack/pin-dependency.sh github.com/cilium/ebpf v0.6.2
	hack/pin-dependency.sh github.com/opencontainers/selinux v1.8.2
	hack/pin-dependency.sh github.com/sirupsen/logrus v1.8.1
	# Recheck.
	hack/lint-dependencies.sh
	GO111MODULE=on go mod edit -dropreplace github.com/willf/bitset
	hack/update-vendor.sh
	# Recheck.
	hack/lint-dependencies.sh
	hack/update-internal-modules.sh
	# Recheck.
	hack/lint-dependencies.sh

Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

LICENSES/vendor/github.com/bits-and-blooms/bitset/LICENSE

@@ -1,4 +1,4 @@
-= vendor/github.com/willf/bitset licensed under: =
+= vendor/github.com/bits-and-blooms/bitset licensed under: =
 
 Copyright (c) 2014 Will Fitzgerald. All rights reserved.
 
@@ -28,4 +28,4 @@ THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
-= vendor/github.com/willf/bitset/LICENSE 0d2b5d66dabaff0df8e1ffe191a21bd1
+= vendor/github.com/bits-and-blooms/bitset/LICENSE 0d2b5d66dabaff0df8e1ffe191a21bd1

go.mod

@@ -66,8 +66,8 @@ require (
 	github.com/onsi/ginkgo v1.14.0
 	github.com/onsi/gomega v1.10.1
 	github.com/opencontainers/go-digest v1.0.0
-	github.com/opencontainers/runc v1.0.0-rc95
+	github.com/opencontainers/runc v1.0.1
-	github.com/opencontainers/selinux v1.8.0
+	github.com/opencontainers/selinux v1.8.2
 	github.com/pkg/errors v0.9.1
 	github.com/pmezard/go-difflib v1.0.0
 	github.com/prometheus/client_golang v1.11.0
@@ -181,6 +181,7 @@ replace (
 	github.com/benbjohnson/clock => github.com/benbjohnson/clock v1.0.3
 	github.com/beorn7/perks => github.com/beorn7/perks v1.0.1
 	github.com/bgentry/speakeasy => github.com/bgentry/speakeasy v0.1.0
+	github.com/bits-and-blooms/bitset => github.com/bits-and-blooms/bitset v1.2.0
 	github.com/bketelsen/crypt => github.com/bketelsen/crypt v0.0.3-0.20200106085610-5cbc8cc4026c
 	github.com/blang/semver => github.com/blang/semver v3.5.1+incompatible
 	github.com/boltdb/bolt => github.com/boltdb/bolt v1.3.1
@@ -192,7 +193,7 @@ replace (
 	github.com/chzyer/logex => github.com/chzyer/logex v1.1.10
 	github.com/chzyer/readline => github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e
 	github.com/chzyer/test => github.com/chzyer/test v0.0.0-20180213035817-a1ea475d72b1
-	github.com/cilium/ebpf => github.com/cilium/ebpf v0.5.0
+	github.com/cilium/ebpf => github.com/cilium/ebpf v0.6.2
 	github.com/clusterhq/flocker-go => github.com/clusterhq/flocker-go v0.0.0-20160920122132-2b8b7259d313
 	github.com/cncf/udpa/go => github.com/cncf/udpa/go v0.0.0-20201120205902-5459f2c99403
 	github.com/cockroachdb/datadriven => github.com/cockroachdb/datadriven v0.0.0-20200714090401-bf6692d28da5
@@ -368,9 +369,9 @@ replace (
 	github.com/onsi/gomega => github.com/onsi/gomega v1.10.1
 	github.com/opencontainers/go-digest => github.com/opencontainers/go-digest v1.0.0
 	github.com/opencontainers/image-spec => github.com/opencontainers/image-spec v1.0.1
-	github.com/opencontainers/runc => github.com/opencontainers/runc v1.0.0-rc95
+	github.com/opencontainers/runc => github.com/opencontainers/runc v1.0.1
 	github.com/opencontainers/runtime-spec => github.com/opencontainers/runtime-spec v1.0.3-0.20210326190908-1c3f411f0417
-	github.com/opencontainers/selinux => github.com/opencontainers/selinux v1.8.0
+	github.com/opencontainers/selinux => github.com/opencontainers/selinux v1.8.2
 	github.com/opentracing/opentracing-go => github.com/opentracing/opentracing-go v1.1.0
 	github.com/pascaldekloe/goe => github.com/pascaldekloe/goe v0.0.0-20180627143212-57f6aae5913c
 	github.com/pelletier/go-toml => github.com/pelletier/go-toml v1.2.0
@@ -396,7 +397,7 @@ replace (
 	github.com/seccomp/libseccomp-golang => github.com/seccomp/libseccomp-golang v0.9.1
 	github.com/sergi/go-diff => github.com/sergi/go-diff v1.1.0
 	github.com/shurcooL/sanitized_anchor_name => github.com/shurcooL/sanitized_anchor_name v1.0.0
-	github.com/sirupsen/logrus => github.com/sirupsen/logrus v1.7.0
+	github.com/sirupsen/logrus => github.com/sirupsen/logrus v1.8.1
 	github.com/smartystreets/assertions => github.com/smartystreets/assertions v1.1.0
 	github.com/smartystreets/goconvey => github.com/smartystreets/goconvey v1.6.4
 	github.com/soheilhy/cmux => github.com/soheilhy/cmux v0.1.5
@@ -418,7 +419,6 @@ replace (
 	github.com/vishvananda/netlink => github.com/vishvananda/netlink v1.1.0
 	github.com/vishvananda/netns => github.com/vishvananda/netns v0.0.0-20200728191858-db3c7e526aae
 	github.com/vmware/govmomi => github.com/vmware/govmomi v0.20.3
-	github.com/willf/bitset => github.com/willf/bitset v1.1.11
 	github.com/xiang90/probing => github.com/xiang90/probing v0.0.0-20190116061207-43a291ad63a2
 	github.com/xlab/treeprint => github.com/xlab/treeprint v0.0.0-20181112141820-a009c3971eca
 	github.com/yuin/goldmark => github.com/yuin/goldmark v1.3.5

go.sum

@@ -67,6 +67,8 @@ github.com/benbjohnson/clock v1.0.3/go.mod h1:bGMdMPoPVvcYyt1gHDf4J2KE153Yf9BuiU
 github.com/beorn7/perks v1.0.1 h1:VlbKKnNfV8bJzeqoa4cOKqO6bYr3WgKZxO8Z16+hsOM=
 github.com/beorn7/perks v1.0.1/go.mod h1:G2ZrVWU2WbWT9wwq4/hrbKbnv/1ERSJQ0ibhJ6rlkpw=
 github.com/bgentry/speakeasy v0.1.0/go.mod h1:+zsyZBPWlz7T6j88CTgSN5bM796AkVf0kBD4zp0CCIs=
+github.com/bits-and-blooms/bitset v1.2.0 h1:Kn4yilvwNtMACtf1eYDlG8H77R07mZSPbMjLyS07ChA=
+github.com/bits-and-blooms/bitset v1.2.0/go.mod h1:gIdJ4wp64HaoK2YrL1Q5/N7Y16edYb8uY+O0FJTyyDA=
 github.com/bketelsen/crypt v0.0.3-0.20200106085610-5cbc8cc4026c/go.mod h1:MKsuJmJgSg28kpZDP6UIiPt0e0Oz0kqKNGyRaWEPv84=
 github.com/blang/semver v3.5.1+incompatible h1:cQNTCjp13qL8KC3Nbxr/y2Bqb63oX6wdnnjpJbkM4JQ=
 github.com/blang/semver v3.5.1+incompatible/go.mod h1:kRBLl5iJ+tD4TcOOxsy/0fnwebNt5EWlYSAyrTnjyyk=
@@ -84,8 +86,8 @@ github.com/checkpoint-restore/go-criu/v5 v5.0.0/go.mod h1:cfwC0EG7HMUenopBsUf9d8
 github.com/chzyer/logex v1.1.10/go.mod h1:+Ywpsq7O8HXn0nuIou7OrIPyXbp3wmkHB+jjWRnGsAI=
 github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e/go.mod h1:nSuG5e5PlCu98SY8svDHJxuZscDgtXS6KTTbou5AhLI=
 github.com/chzyer/test v0.0.0-20180213035817-a1ea475d72b1/go.mod h1:Q3SI9o4m/ZMnBNeIyt5eFwwo7qiLfzFZmjNmxjkiQlU=
-github.com/cilium/ebpf v0.5.0 h1:E1KshmrMEtkMP2UjlWzfmUV1owWY+BnbL5FxxuatnrU=
+github.com/cilium/ebpf v0.6.2 h1:iHsfF/t4aW4heW2YKfeHrVPGdtYTL4C4KocpM8KTSnI=
-github.com/cilium/ebpf v0.5.0/go.mod h1:4tRaxcgiL706VnOzHOdBlY8IEAIdxINsQBcU4xJJXRs=
+github.com/cilium/ebpf v0.6.2/go.mod h1:4tRaxcgiL706VnOzHOdBlY8IEAIdxINsQBcU4xJJXRs=
 github.com/clusterhq/flocker-go v0.0.0-20160920122132-2b8b7259d313 h1:eIHD9GNM3Hp7kcRW5mvcz7WTR3ETeoYYKwpgA04kaXE=
 github.com/clusterhq/flocker-go v0.0.0-20160920122132-2b8b7259d313/go.mod h1:P1wt9Z3DP8O6W3rvwCt0REIlshg1InHImaLW0t3ObY0=
 github.com/cncf/udpa/go v0.0.0-20201120205902-5459f2c99403/go.mod h1:WmhPx2Nbnhtbo57+VJT5O0JRkEi1Wbu0z5j0R8u5Hbk=
@@ -371,12 +373,12 @@ github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8
 github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/opencontainers/image-spec v1.0.1 h1:JMemWkRwHx4Zj+fVxWoMCFm/8sYGGrUVojFA6h/TRcI=
 github.com/opencontainers/image-spec v1.0.1/go.mod h1:BtxoFyWECRxE4U/7sNtV5W15zMzWCbyJoFRP3s7yZA0=
-github.com/opencontainers/runc v1.0.0-rc95 h1:RMuWVfY3E1ILlVsC3RhIq38n4sJtlOFwU9gfFZSqrd0=
+github.com/opencontainers/runc v1.0.1 h1:G18PGckGdAm3yVQRWDVQ1rLSLntiniKJ0cNRT2Tm5gs=
-github.com/opencontainers/runc v1.0.0-rc95/go.mod h1:z+bZxa/+Tz/FmYVWkhUajJdzFeOqjc5vrqskhVyHGUM=
+github.com/opencontainers/runc v1.0.1/go.mod h1:aTaHFFwQXuA71CiyxOdFFIorAoemI04suvGRQFzWTD0=
 github.com/opencontainers/runtime-spec v1.0.3-0.20210326190908-1c3f411f0417 h1:3snG66yBm59tKhhSPQrQ/0bCrv1LQbKt40LnUPiUxdc=
 github.com/opencontainers/runtime-spec v1.0.3-0.20210326190908-1c3f411f0417/go.mod h1:jwyrGlmzljRJv/Fgzds9SsS/C5hL+LL3ko9hs6T5lQ0=
-github.com/opencontainers/selinux v1.8.0 h1:+77ba4ar4jsCbL1GLbFL8fFM57w6suPfSS9PDLDY7KM=
+github.com/opencontainers/selinux v1.8.2 h1:c4ca10UMgRcvZ6h0K4HtS15UaVSBEaE+iln2LVpAuGc=
-github.com/opencontainers/selinux v1.8.0/go.mod h1:RScLhm78qiWa2gbVCcGkC7tCGdgk3ogry1nUQF8Evvo=
+github.com/opencontainers/selinux v1.8.2/go.mod h1:MUIHuUEvKB1wtJjQdOyYRgOnLD2xAPP8dBsCoU0KuF8=
 github.com/opentracing/opentracing-go v1.1.0/go.mod h1:UkNAQd3GIcIGf0SeVgPpRdFStlNbqXla1AfSYxPUl2o=
 github.com/pascaldekloe/goe v0.0.0-20180627143212-57f6aae5913c/go.mod h1:lzWF7FIEvWOWxwDKqyGYQf6ZUaNfKdP144TG7ZOy1lc=
 github.com/pelletier/go-toml v1.2.0/go.mod h1:5z9KED0ma1S8pY6P1sdut58dfprrGBbd/94hg7ilaic=
@@ -418,8 +420,8 @@ github.com/sergi/go-diff v1.1.0 h1:we8PVUC3FE2uYfodKH/nBHMSetSfHDR6scGdBi+erh0=
 github.com/sergi/go-diff v1.1.0/go.mod h1:STckp+ISIX8hZLjrqAeVduY0gWCT9IjLuqbuNXdaHfM=
 github.com/shurcooL/sanitized_anchor_name v1.0.0 h1:PdmoCO6wvbs+7yrJyMORt4/BmY5IYyJwS/kOiWx8mHo=
 github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeVRXNmyL/1OwPU0+IJeTBvfc=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
-github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v1.1.0/go.mod h1:tcbTF8ujkAEcZ8TElKY+i30BzYlVhC/LOxJk7iOWnoo=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.5 h1:jjzc5WVemNEDTLwv9tlmemhC73tI08BNOIGwBOo10Js=
@@ -454,8 +456,6 @@ github.com/vishvananda/netns v0.0.0-20200728191858-db3c7e526aae h1:4hwBBUfQCFe3C
 github.com/vishvananda/netns v0.0.0-20200728191858-db3c7e526aae/go.mod h1:DD4vA1DwXk04H54A1oHXtwZmA0grkVMdPxx/VGLCah0=
 github.com/vmware/govmomi v0.20.3 h1:gpw/0Ku+6RgF3jsi7fnCLmlcikBHfKBCUcu1qgc16OU=
 github.com/vmware/govmomi v0.20.3/go.mod h1:URlwyTFZX72RmxtxuaFL2Uj3fD1JTvZdx59bHWk6aFU=
-github.com/willf/bitset v1.1.11 h1:N7Z7E9UvjW+sGsEl7k/SJrvY2reP1A07MrGuCjIOjRE=
-github.com/willf/bitset v1.1.11/go.mod h1:83CECat5yLh5zVOf4P1ErAgKA5UDvKtgyUABdr3+MjI=
 github.com/xiang90/probing v0.0.0-20190116061207-43a291ad63a2 h1:eY9dn8+vbi4tKz5Qo6v2eYzo7kUS51QINcR5jNpbZS8=
 github.com/xiang90/probing v0.0.0-20190116061207-43a291ad63a2/go.mod h1:UETIi67q53MR2AWcXfiuqkDkRtnGDLqkBTpCHuJHxtU=
 github.com/xlab/treeprint v0.0.0-20181112141820-a009c3971eca h1:1CFlNzQhALwjS9mBAUkycX616GzgsuYUOCHA5+HSlXI=

staging/src/k8s.io/apiextensions-apiserver/go.sum

@@ -382,8 +382,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/apiserver/go.mod

@@ -22,6 +22,7 @@ require (
 	github.com/grpc-ecosystem/go-grpc-prometheus v1.2.0
 	github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822
 	github.com/pquerna/cachecontrol v0.0.0-20171018203845-0dec1b30a021 // indirect
+	github.com/sirupsen/logrus v1.8.1 // indirect
 	github.com/spf13/pflag v1.0.5
 	github.com/stretchr/testify v1.7.0
 	go.etcd.io/etcd/api/v3 v3.5.0

staging/src/k8s.io/apiserver/go.sum

@@ -376,8 +376,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/cloud-provider/go.sum

@@ -372,8 +372,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/controller-manager/go.sum

@@ -480,8 +480,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/kube-aggregator/go.sum

@@ -373,8 +373,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/kube-controller-manager/go.sum

@@ -324,6 +324,7 @@ github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPx
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/legacy-cloud-providers/go.sum

@@ -380,6 +380,7 @@ github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPx
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/pod-security-admission/go.sum

@@ -369,8 +369,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

staging/src/k8s.io/sample-apiserver/go.sum

@@ -370,8 +370,9 @@ github.com/shurcooL/sanitized_anchor_name v1.0.0/go.mod h1:1NzhyTcUVG4SuEtjjoZeV
 github.com/sirupsen/logrus v1.2.0/go.mod h1:LxeOpSwHxABJmUn/MG1IvRgCAasNZTLOkJPxbbu5VWo=
 github.com/sirupsen/logrus v1.4.2/go.mod h1:tLMulIdttU9McNUspp0xgXVQah82FyeX6MwdIuYE2rE=
 github.com/sirupsen/logrus v1.6.0/go.mod h1:7uNnSEd1DgxDLC74fIahvMZmmYsHGZGEOFrfsX/uA88=
-github.com/sirupsen/logrus v1.7.0 h1:ShrD1U9pZB12TX0cVy0DtePoCH97K8EtX+mg7ZARUtM=
 github.com/sirupsen/logrus v1.7.0/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
+github.com/sirupsen/logrus v1.8.1 h1:dJKuHgqk1NNQlqoA6BTlM1Wf9DOH3NBjQyu0h9+AZZE=
+github.com/sirupsen/logrus v1.8.1/go.mod h1:yWOB1SBYBC5VeMP7gHvWumXLIWorT60ONWic61uBYv0=
 github.com/smartystreets/assertions v0.0.0-20180927180507-b2de0cb4f26d/go.mod h1:OnSkiWE9lh6wB0YB77sQom3nweQdgAjqCqsofrRNTgc=
 github.com/smartystreets/goconvey v1.6.4/go.mod h1:syvi0/a8iFYH4r/RixwvyeAJjdLS9QV7WQ/tjFTllLA=
 github.com/soheilhy/cmux v0.1.4/go.mod h1:IM3LyeVVIOuxMH7sFAkER9+bJ4dT7Ms6E4xg4kGIyLM=

vendor/github.com/bits-and-blooms/bitset/README.md

@@ -2,10 +2,9 @@
 
 *Go language library to map between non-negative integers and boolean values*
 
-[![Test](https://github.com/willf/bitset/workflows/Test/badge.svg)](https://github.com/willf/bitset/actions?query=workflow%3ATest)
+[![Test](https://github.com/bits-and-blooms/bitset/workflows/Test/badge.svg)](https://github.com/willf/bitset/actions?query=workflow%3ATest)
-[![Master Coverage Status](https://coveralls.io/repos/willf/bitset/badge.svg?branch=master&service=github)](https://coveralls.io/github/willf/bitset?branch=master)
 [![Go Report Card](https://goreportcard.com/badge/github.com/willf/bitset)](https://goreportcard.com/report/github.com/willf/bitset)
-[![PkgGoDev](https://pkg.go.dev/badge/github.com/willf/bitset?tab=doc)](https://pkg.go.dev/github.com/willf/bitset?tab=doc)
+[![PkgGoDev](https://pkg.go.dev/badge/github.com/bits-and-blooms/bitset?tab=doc)](https://pkg.go.dev/github.com/bits-and-blooms/bitset?tab=doc)
 
 
 ## Description
@@ -30,7 +29,7 @@ import (
 	"fmt"
 	"math/rand"
 
-	"github.com/willf/bitset"
+	"github.com/bits-and-blooms/bitset"
 )
 
 func main() {
@@ -63,7 +62,7 @@ func main() {
 
 As an alternative to BitSets, one should check out the 'big' package, which provides a (less set-theoretical) view of bitsets.
 
-Package documentation is at: https://pkg.go.dev/github.com/willf/bitset?tab=doc
+Package documentation is at: https://pkg.go.dev/github.com/bits-and-blooms/bitset?tab=doc
 
 ## Memory Usage
 
@@ -78,7 +77,7 @@ It is possible that a later version will match the `math/bits` return signature
 ## Installation
 
 ```bash
-go get github.com/willf/bitset
+go get github.com/bits-and-blooms/bitset
 ```
 
 ## Contributing

vendor/github.com/bits-and-blooms/bitset/bitset.go

@@ -209,6 +209,27 @@ func (b *BitSet) Flip(i uint) *BitSet {
 	return b
 }
 
+// FlipRange bit in [start, end).
+// If end>= Cap(), this function will panic.
+// Warning: using a very large value for 'end'
+// may lead to a memory shortage and a panic: the caller is responsible
+// for providing sensible parameters in line with their memory capacity.
+func (b *BitSet) FlipRange(start, end uint) *BitSet {
+	if start >= end {
+		return b
+	}
+
+	b.extendSetMaybe(end - 1)
+	var startWord uint = start >> log2WordSize
+	var endWord uint = end >> log2WordSize
+	b.set[startWord] ^= ^(^uint64(0) << (start & (wordSize - 1)))
+	for i := startWord; i < endWord; i++ {
+		b.set[i] = ^b.set[i]
+	}
+	b.set[endWord] ^= ^uint64(0) >> (-end & (wordSize - 1))
+	return b
+}
+
 // Shrink shrinks BitSet so that the provided value is the last possible
 // set value. It clears all bits > the provided index and reduces the size
 // and length of the set.
@@ -519,7 +540,7 @@ func (b *BitSet) Copy(c *BitSet) (count uint) {
 }
 
 // Count (number of set bits).
-// Also known as "popcount" or "popularity count".
+// Also known as "popcount" or "population count".
 func (b *BitSet) Count() uint {
 	if b != nil && b.set != nil {
 		return uint(popcntSlice(b.set))

vendor/github.com/bits-and-blooms/bitset/go.mod

@@ -0,0 +1,3 @@
+module github.com/bits-and-blooms/bitset
+
+go 1.14

vendor/github.com/cilium/ebpf/CONTRIBUTING.md

@@ -18,6 +18,23 @@ reason about the proposed changes.
 ## Running the tests
 
 Many of the tests require privileges to set resource limits and load eBPF code.
-The easiest way to obtain these is to run the tests with `sudo`:
+The easiest way to obtain these is to run the tests with `sudo`.
+
+To test the current package with your local kernel you can simply run:
+```
+go test -exec sudo  ./...
+```
+
+To test the current package with a different kernel version you can use the [run-tests.sh](run-tests.sh) script.
+It requires [virtme](https://github.com/amluto/virtme) and qemu to be installed.
+
+Examples:
+
+```bash
+# Run all tests on a 5.4 kernel
+./run-tests.sh 5.4
+
+# Run a subset of tests:
+./run-tests.sh 5.4 go test ./link
+```
 
-    sudo go test ./...

vendor/github.com/cilium/ebpf/Makefile

@@ -1,7 +1,7 @@
 # The development version of clang is distributed as the 'clang' binary,
 # while stable/released versions have a version number attached.
 # Pin the default clang to a stable version.
-CLANG ?= clang-11
+CLANG ?= clang-12
 CFLAGS := -target bpf -O2 -g -Wall -Werror $(CFLAGS)
 
 # Obtain an absolute path to the directory of the Makefile.
@@ -17,7 +17,7 @@ VERSION := $(shell cat ${REPODIR}/testdata/docker/VERSION)
 TARGETS := \
 	testdata/loader-clang-7 \
 	testdata/loader-clang-9 \
-	testdata/loader-clang-11 \
+	testdata/loader-$(CLANG) \
 	testdata/invalid_map \
 	testdata/raw_tracepoint \
 	testdata/invalid_map_static \
@@ -33,6 +33,7 @@ TARGETS := \
 docker-all:
 	docker run --rm --user "${UIDGID}" \
 		-v "${REPODIR}":/ebpf -w /ebpf --env MAKEFLAGS \
+		--env CFLAGS="-fdebug-prefix-map=/ebpf=." \
 		"${IMAGE}:${VERSION}" \
 		make all
 
@@ -47,6 +48,8 @@ clean:
 	-$(RM) internal/btf/testdata/*.elf
 
 all: $(addsuffix -el.elf,$(TARGETS)) $(addsuffix -eb.elf,$(TARGETS))
+	ln -srf testdata/loader-$(CLANG)-el.elf testdata/loader-el.elf
+	ln -srf testdata/loader-$(CLANG)-eb.elf testdata/loader-eb.elf
 
 testdata/loader-%-el.elf: testdata/loader.c
 	$* $(CFLAGS) -mlittle-endian -c $< -o $@

vendor/github.com/cilium/ebpf/asm/func.go

@@ -132,6 +132,58 @@ const (
 	FnSkStorageDelete
 	FnSendSignal
 	FnTcpGenSyncookie
+	FnSkbOutput
+	FnProbeReadUser
+	FnProbeReadKernel
+	FnProbeReadUserStr
+	FnProbeReadKernelStr
+	FnTcpSendAck
+	FnSendSignalThread
+	FnJiffies64
+	FnReadBranchRecords
+	FnGetNsCurrentPidTgid
+	FnXdpOutput
+	FnGetNetnsCookie
+	FnGetCurrentAncestorCgroupId
+	FnSkAssign
+	FnKtimeGetBootNs
+	FnSeqPrintf
+	FnSeqWrite
+	FnSkCgroupId
+	FnSkAncestorCgroupId
+	FnRingbufOutput
+	FnRingbufReserve
+	FnRingbufSubmit
+	FnRingbufDiscard
+	FnRingbufQuery
+	FnCsumLevel
+	FnSkcToTcp6Sock
+	FnSkcToTcpSock
+	FnSkcToTcpTimewaitSock
+	FnSkcToTcpRequestSock
+	FnSkcToUdp6Sock
+	FnGetTaskStack
+	FnLoadHdrOpt
+	FnStoreHdrOpt
+	FnReserveHdrOpt
+	FnInodeStorageGet
+	FnInodeStorageDelete
+	FnDPath
+	FnCopyFromUser
+	FnSnprintfBtf
+	FnSeqPrintfBtf
+	FnSkbCgroupClassid
+	FnRedirectNeigh
+	FnPerCpuPtr
+	FnThisCpuPtr
+	FnRedirectPeer
+	FnTaskStorageGet
+	FnTaskStorageDelete
+	FnGetCurrentTaskBtf
+	FnBprmOptsSet
+	FnKtimeGetCoarseNs
+	FnImaInodeHash
+	FnSockFromFile
 )
 
 // Call emits a function call.

vendor/github.com/cilium/ebpf/asm/func_string.go

@@ -119,11 +119,63 @@ func _() {
 	_ = x[FnSkStorageDelete-108]
 	_ = x[FnSendSignal-109]
 	_ = x[FnTcpGenSyncookie-110]
+	_ = x[FnSkbOutput-111]
+	_ = x[FnProbeReadUser-112]
+	_ = x[FnProbeReadKernel-113]
+	_ = x[FnProbeReadUserStr-114]
+	_ = x[FnProbeReadKernelStr-115]
+	_ = x[FnTcpSendAck-116]
+	_ = x[FnSendSignalThread-117]
+	_ = x[FnJiffies64-118]
+	_ = x[FnReadBranchRecords-119]
+	_ = x[FnGetNsCurrentPidTgid-120]
+	_ = x[FnXdpOutput-121]
+	_ = x[FnGetNetnsCookie-122]
+	_ = x[FnGetCurrentAncestorCgroupId-123]
+	_ = x[FnSkAssign-124]
+	_ = x[FnKtimeGetBootNs-125]
+	_ = x[FnSeqPrintf-126]
+	_ = x[FnSeqWrite-127]
+	_ = x[FnSkCgroupId-128]
+	_ = x[FnSkAncestorCgroupId-129]
+	_ = x[FnRingbufOutput-130]
+	_ = x[FnRingbufReserve-131]
+	_ = x[FnRingbufSubmit-132]
+	_ = x[FnRingbufDiscard-133]
+	_ = x[FnRingbufQuery-134]
+	_ = x[FnCsumLevel-135]
+	_ = x[FnSkcToTcp6Sock-136]
+	_ = x[FnSkcToTcpSock-137]
+	_ = x[FnSkcToTcpTimewaitSock-138]
+	_ = x[FnSkcToTcpRequestSock-139]
+	_ = x[FnSkcToUdp6Sock-140]
+	_ = x[FnGetTaskStack-141]
+	_ = x[FnLoadHdrOpt-142]
+	_ = x[FnStoreHdrOpt-143]
+	_ = x[FnReserveHdrOpt-144]
+	_ = x[FnInodeStorageGet-145]
+	_ = x[FnInodeStorageDelete-146]
+	_ = x[FnDPath-147]
+	_ = x[FnCopyFromUser-148]
+	_ = x[FnSnprintfBtf-149]
+	_ = x[FnSeqPrintfBtf-150]
+	_ = x[FnSkbCgroupClassid-151]
+	_ = x[FnRedirectNeigh-152]
+	_ = x[FnPerCpuPtr-153]
+	_ = x[FnThisCpuPtr-154]
+	_ = x[FnRedirectPeer-155]
+	_ = x[FnTaskStorageGet-156]
+	_ = x[FnTaskStorageDelete-157]
+	_ = x[FnGetCurrentTaskBtf-158]
+	_ = x[FnBprmOptsSet-159]
+	_ = x[FnKtimeGetCoarseNs-160]
+	_ = x[FnImaInodeHash-161]
+	_ = x[FnSockFromFile-162]
 }
 
-const _BuiltinFunc_name = "FnUnspecFnMapLookupElemFnMapUpdateElemFnMapDeleteElemFnProbeReadFnKtimeGetNsFnTracePrintkFnGetPrandomU32FnGetSmpProcessorIdFnSkbStoreBytesFnL3CsumReplaceFnL4CsumReplaceFnTailCallFnCloneRedirectFnGetCurrentPidTgidFnGetCurrentUidGidFnGetCurrentCommFnGetCgroupClassidFnSkbVlanPushFnSkbVlanPopFnSkbGetTunnelKeyFnSkbSetTunnelKeyFnPerfEventReadFnRedirectFnGetRouteRealmFnPerfEventOutputFnSkbLoadBytesFnGetStackidFnCsumDiffFnSkbGetTunnelOptFnSkbSetTunnelOptFnSkbChangeProtoFnSkbChangeTypeFnSkbUnderCgroupFnGetHashRecalcFnGetCurrentTaskFnProbeWriteUserFnCurrentTaskUnderCgroupFnSkbChangeTailFnSkbPullDataFnCsumUpdateFnSetHashInvalidFnGetNumaNodeIdFnSkbChangeHeadFnXdpAdjustHeadFnProbeReadStrFnGetSocketCookieFnGetSocketUidFnSetHashFnSetsockoptFnSkbAdjustRoomFnRedirectMapFnSkRedirectMapFnSockMapUpdateFnXdpAdjustMetaFnPerfEventReadValueFnPerfProgReadValueFnGetsockoptFnOverrideReturnFnSockOpsCbFlagsSetFnMsgRedirectMapFnMsgApplyBytesFnMsgCorkBytesFnMsgPullDataFnBindFnXdpAdjustTailFnSkbGetXfrmStateFnGetStackFnSkbLoadBytesRelativeFnFibLookupFnSockHashUpdateFnMsgRedirectHashFnSkRedirectHashFnLwtPushEncapFnLwtSeg6StoreBytesFnLwtSeg6AdjustSrhFnLwtSeg6ActionFnRcRepeatFnRcKeydownFnSkbCgroupIdFnGetCurrentCgroupIdFnGetLocalStorageFnSkSelectReuseportFnSkbAncestorCgroupIdFnSkLookupTcpFnSkLookupUdpFnSkReleaseFnMapPushElemFnMapPopElemFnMapPeekElemFnMsgPushDataFnMsgPopDataFnRcPointerRelFnSpinLockFnSpinUnlockFnSkFullsockFnTcpSockFnSkbEcnSetCeFnGetListenerSockFnSkcLookupTcpFnTcpCheckSyncookieFnSysctlGetNameFnSysctlGetCurrentValueFnSysctlGetNewValueFnSysctlSetNewValueFnStrtolFnStrtoulFnSkStorageGetFnSkStorageDeleteFnSendSignalFnTcpGenSyncookie"
+const _BuiltinFunc_name = "FnUnspecFnMapLookupElemFnMapUpdateElemFnMapDeleteElemFnProbeReadFnKtimeGetNsFnTracePrintkFnGetPrandomU32FnGetSmpProcessorIdFnSkbStoreBytesFnL3CsumReplaceFnL4CsumReplaceFnTailCallFnCloneRedirectFnGetCurrentPidTgidFnGetCurrentUidGidFnGetCurrentCommFnGetCgroupClassidFnSkbVlanPushFnSkbVlanPopFnSkbGetTunnelKeyFnSkbSetTunnelKeyFnPerfEventReadFnRedirectFnGetRouteRealmFnPerfEventOutputFnSkbLoadBytesFnGetStackidFnCsumDiffFnSkbGetTunnelOptFnSkbSetTunnelOptFnSkbChangeProtoFnSkbChangeTypeFnSkbUnderCgroupFnGetHashRecalcFnGetCurrentTaskFnProbeWriteUserFnCurrentTaskUnderCgroupFnSkbChangeTailFnSkbPullDataFnCsumUpdateFnSetHashInvalidFnGetNumaNodeIdFnSkbChangeHeadFnXdpAdjustHeadFnProbeReadStrFnGetSocketCookieFnGetSocketUidFnSetHashFnSetsockoptFnSkbAdjustRoomFnRedirectMapFnSkRedirectMapFnSockMapUpdateFnXdpAdjustMetaFnPerfEventReadValueFnPerfProgReadValueFnGetsockoptFnOverrideReturnFnSockOpsCbFlagsSetFnMsgRedirectMapFnMsgApplyBytesFnMsgCorkBytesFnMsgPullDataFnBindFnXdpAdjustTailFnSkbGetXfrmStateFnGetStackFnSkbLoadBytesRelativeFnFibLookupFnSockHashUpdateFnMsgRedirectHashFnSkRedirectHashFnLwtPushEncapFnLwtSeg6StoreBytesFnLwtSeg6AdjustSrhFnLwtSeg6ActionFnRcRepeatFnRcKeydownFnSkbCgroupIdFnGetCurrentCgroupIdFnGetLocalStorageFnSkSelectReuseportFnSkbAncestorCgroupIdFnSkLookupTcpFnSkLookupUdpFnSkReleaseFnMapPushElemFnMapPopElemFnMapPeekElemFnMsgPushDataFnMsgPopDataFnRcPointerRelFnSpinLockFnSpinUnlockFnSkFullsockFnTcpSockFnSkbEcnSetCeFnGetListenerSockFnSkcLookupTcpFnTcpCheckSyncookieFnSysctlGetNameFnSysctlGetCurrentValueFnSysctlGetNewValueFnSysctlSetNewValueFnStrtolFnStrtoulFnSkStorageGetFnSkStorageDeleteFnSendSignalFnTcpGenSyncookieFnSkbOutputFnProbeReadUserFnProbeReadKernelFnProbeReadUserStrFnProbeReadKernelStrFnTcpSendAckFnSendSignalThreadFnJiffies64FnReadBranchRecordsFnGetNsCurrentPidTgidFnXdpOutputFnGetNetnsCookieFnGetCurrentAncestorCgroupIdFnSkAssignFnKtimeGetBootNsFnSeqPrintfFnSeqWriteFnSkCgroupIdFnSkAncestorCgroupIdFnRingbufOutputFnRingbufReserveFnRingbufSubmitFnRingbufDiscardFnRingbufQueryFnCsumLevelFnSkcToTcp6SockFnSkcToTcpSockFnSkcToTcpTimewaitSockFnSkcToTcpRequestSockFnSkcToUdp6SockFnGetTaskStackFnLoadHdrOptFnStoreHdrOptFnReserveHdrOptFnInodeStorageGetFnInodeStorageDeleteFnDPathFnCopyFromUserFnSnprintfBtfFnSeqPrintfBtfFnSkbCgroupClassidFnRedirectNeighFnPerCpuPtrFnThisCpuPtrFnRedirectPeerFnTaskStorageGetFnTaskStorageDeleteFnGetCurrentTaskBtfFnBprmOptsSetFnKtimeGetCoarseNsFnImaInodeHashFnSockFromFile"
 
-var _BuiltinFunc_index = [...]uint16{0, 8, 23, 38, 53, 64, 76, 89, 104, 123, 138, 153, 168, 178, 193, 212, 230, 246, 264, 277, 289, 306, 323, 338, 348, 363, 380, 394, 406, 416, 433, 450, 466, 481, 497, 512, 528, 544, 568, 583, 596, 608, 624, 639, 654, 669, 683, 700, 714, 723, 735, 750, 763, 778, 793, 808, 828, 847, 859, 875, 894, 910, 925, 939, 952, 958, 973, 990, 1000, 1022, 1033, 1049, 1066, 1082, 1096, 1115, 1133, 1148, 1158, 1169, 1182, 1202, 1219, 1238, 1259, 1272, 1285, 1296, 1309, 1321, 1334, 1347, 1359, 1373, 1383, 1395, 1407, 1416, 1429, 1446, 1460, 1479, 1494, 1517, 1536, 1555, 1563, 1572, 1586, 1603, 1615, 1632}
+var _BuiltinFunc_index = [...]uint16{0, 8, 23, 38, 53, 64, 76, 89, 104, 123, 138, 153, 168, 178, 193, 212, 230, 246, 264, 277, 289, 306, 323, 338, 348, 363, 380, 394, 406, 416, 433, 450, 466, 481, 497, 512, 528, 544, 568, 583, 596, 608, 624, 639, 654, 669, 683, 700, 714, 723, 735, 750, 763, 778, 793, 808, 828, 847, 859, 875, 894, 910, 925, 939, 952, 958, 973, 990, 1000, 1022, 1033, 1049, 1066, 1082, 1096, 1115, 1133, 1148, 1158, 1169, 1182, 1202, 1219, 1238, 1259, 1272, 1285, 1296, 1309, 1321, 1334, 1347, 1359, 1373, 1383, 1395, 1407, 1416, 1429, 1446, 1460, 1479, 1494, 1517, 1536, 1555, 1563, 1572, 1586, 1603, 1615, 1632, 1643, 1658, 1675, 1693, 1713, 1725, 1743, 1754, 1773, 1794, 1805, 1821, 1849, 1859, 1875, 1886, 1896, 1908, 1928, 1943, 1959, 1974, 1990, 2004, 2015, 2030, 2044, 2066, 2087, 2102, 2116, 2128, 2141, 2156, 2173, 2193, 2200, 2214, 2227, 2241, 2259, 2274, 2285, 2297, 2311, 2327, 2346, 2365, 2378, 2396, 2410, 2424}
 
 func (i BuiltinFunc) String() string {
 	if i < 0 || i >= BuiltinFunc(len(_BuiltinFunc_index)-1) {

vendor/github.com/cilium/ebpf/asm/instruction.go

@@ -57,7 +57,7 @@ func (ins *Instruction) Unmarshal(r io.Reader, bo binary.ByteOrder) (uint64, err
 		return 0, fmt.Errorf("can't unmarshal registers: %s", err)
 	}
 
-	if !bi.OpCode.isDWordLoad() {
+	if !bi.OpCode.IsDWordLoad() {
 		return InstructionSize, nil
 	}
 
@@ -80,7 +80,7 @@ func (ins Instruction) Marshal(w io.Writer, bo binary.ByteOrder) (uint64, error)
 		return 0, errors.New("invalid opcode")
 	}
 
-	isDWordLoad := ins.OpCode.isDWordLoad()
+	isDWordLoad := ins.OpCode.IsDWordLoad()
 
 	cons := int32(ins.Constant)
 	if isDWordLoad {
@@ -123,7 +123,7 @@ func (ins Instruction) Marshal(w io.Writer, bo binary.ByteOrder) (uint64, error)
 //
 // Returns an error if the instruction doesn't load a map.
 func (ins *Instruction) RewriteMapPtr(fd int) error {
-	if !ins.OpCode.isDWordLoad() {
+	if !ins.OpCode.IsDWordLoad() {
 		return fmt.Errorf("%s is not a 64 bit load", ins.OpCode)
 	}
 
@@ -138,15 +138,19 @@ func (ins *Instruction) RewriteMapPtr(fd int) error {
 	return nil
 }
 
-func (ins *Instruction) mapPtr() uint32 {
+// MapPtr returns the map fd for this instruction.
-	return uint32(uint64(ins.Constant) & math.MaxUint32)
+//
+// The result is undefined if the instruction is not a load from a map,
+// see IsLoadFromMap.
+func (ins *Instruction) MapPtr() int {
+	return int(int32(uint64(ins.Constant) & math.MaxUint32))
 }
 
 // RewriteMapOffset changes the offset of a direct load from a map.
 //
 // Returns an error if the instruction is not a direct load.
 func (ins *Instruction) RewriteMapOffset(offset uint32) error {
-	if !ins.OpCode.isDWordLoad() {
+	if !ins.OpCode.IsDWordLoad() {
 		return fmt.Errorf("%s is not a 64 bit load", ins.OpCode)
 	}
 
@@ -163,10 +167,10 @@ func (ins *Instruction) mapOffset() uint32 {
 	return uint32(uint64(ins.Constant) >> 32)
 }
 
-// isLoadFromMap returns true if the instruction loads from a map.
+// IsLoadFromMap returns true if the instruction loads from a map.
 //
 // This covers both loading the map pointer and direct map value loads.
-func (ins *Instruction) isLoadFromMap() bool {
+func (ins *Instruction) IsLoadFromMap() bool {
 	return ins.OpCode == LoadImmOp(DWord) && (ins.Src == PseudoMapFD || ins.Src == PseudoMapValue)
 }
 
@@ -177,6 +181,12 @@ func (ins *Instruction) IsFunctionCall() bool {
 	return ins.OpCode.JumpOp() == Call && ins.Src == PseudoCall
 }
 
+// IsConstantLoad returns true if the instruction loads a constant of the
+// given size.
+func (ins *Instruction) IsConstantLoad(size Size) bool {
+	return ins.OpCode == LoadImmOp(size) && ins.Src == R0 && ins.Offset == 0
+}
+
 // Format implements fmt.Formatter.
 func (ins Instruction) Format(f fmt.State, c rune) {
 	if c != 'v' {
@@ -197,8 +207,8 @@ func (ins Instruction) Format(f fmt.State, c rune) {
 		return
 	}
 
-	if ins.isLoadFromMap() {
+	if ins.IsLoadFromMap() {
-		fd := int32(ins.mapPtr())
+		fd := ins.MapPtr()
 		switch ins.Src {
 		case PseudoMapFD:
 			fmt.Fprintf(f, "LoadMapPtr dst: %s fd: %d", ins.Dst, fd)
@@ -403,7 +413,7 @@ func (insns Instructions) Marshal(w io.Writer, bo binary.ByteOrder) error {
 func (insns Instructions) Tag(bo binary.ByteOrder) (string, error) {
 	h := sha1.New()
 	for i, ins := range insns {
-		if ins.isLoadFromMap() {
+		if ins.IsLoadFromMap() {
 			ins.Constant = 0
 		}
 		_, err := ins.Marshal(h, bo)

vendor/github.com/cilium/ebpf/asm/load_store.go

@@ -111,7 +111,7 @@ func LoadMapPtr(dst Register, fd int) Instruction {
 		OpCode:   LoadImmOp(DWord),
 		Dst:      dst,
 		Src:      PseudoMapFD,
-		Constant: int64(fd),
+		Constant: int64(uint32(fd)),
 	}
 }
 

vendor/github.com/cilium/ebpf/asm/opcode.go

@@ -69,13 +69,13 @@ const InvalidOpCode OpCode = 0xff
 // rawInstructions returns the number of BPF instructions required
 // to encode this opcode.
 func (op OpCode) rawInstructions() int {
-	if op.isDWordLoad() {
+	if op.IsDWordLoad() {
 		return 2
 	}
 	return 1
 }
 
-func (op OpCode) isDWordLoad() bool {
+func (op OpCode) IsDWordLoad() bool {
 	return op == LoadImmOp(DWord)
 }
 

vendor/github.com/cilium/ebpf/collection.go

@@ -3,6 +3,7 @@ package ebpf
 import (
 	"errors"
 	"fmt"
+	"io"
 	"math"
 	"reflect"
 	"strings"
@@ -89,8 +90,8 @@ func (cs *CollectionSpec) RewriteMaps(maps map[string]*Map) error {
 //
 // The constant must be defined like so in the C program:
 //
-//    static volatile const type foobar;
+//    volatile const type foobar;
-//    static volatile const type foobar = default;
+//    volatile const type foobar = default;
 //
 // Replacement values must be of the same length as the C sizeof(type).
 // If necessary, they are marshalled according to the same rules as
@@ -269,11 +270,21 @@ func NewCollectionWithOptions(spec *CollectionSpec, opts CollectionOptions) (*Co
 	}, nil
 }
 
-type btfHandleCache map[*btf.Spec]*btf.Handle
+type handleCache struct {
+	btfHandles map[*btf.Spec]*btf.Handle
+	btfSpecs   map[io.ReaderAt]*btf.Spec
+}
 
-func (btfs btfHandleCache) load(spec *btf.Spec) (*btf.Handle, error) {
+func newHandleCache() *handleCache {
-	if btfs[spec] != nil {
+	return &handleCache{
-		return btfs[spec], nil
+		btfHandles: make(map[*btf.Spec]*btf.Handle),
+		btfSpecs:   make(map[io.ReaderAt]*btf.Spec),
+	}
+}
+
+func (hc handleCache) btfHandle(spec *btf.Spec) (*btf.Handle, error) {
+	if hc.btfHandles[spec] != nil {
+		return hc.btfHandles[spec], nil
 	}
 
 	handle, err := btf.NewHandle(spec)
@@ -281,14 +292,30 @@ func (btfs btfHandleCache) load(spec *btf.Spec) (*btf.Handle, error) {
 		return nil, err
 	}
 
-	btfs[spec] = handle
+	hc.btfHandles[spec] = handle
 	return handle, nil
 }
 
-func (btfs btfHandleCache) close() {
+func (hc handleCache) btfSpec(rd io.ReaderAt) (*btf.Spec, error) {
-	for _, handle := range btfs {
+	if hc.btfSpecs[rd] != nil {
+		return hc.btfSpecs[rd], nil
+	}
+
+	spec, err := btf.LoadSpecFromReader(rd)
+	if err != nil {
+		return nil, err
+	}
+
+	hc.btfSpecs[rd] = spec
+	return spec, nil
+}
+
+func (hc handleCache) close() {
+	for _, handle := range hc.btfHandles {
 		handle.Close()
 	}
+	hc.btfHandles = nil
+	hc.btfSpecs = nil
 }
 
 func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
@@ -300,12 +327,12 @@ func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
 	var (
 		maps             = make(map[string]*Map)
 		progs            = make(map[string]*Program)
-		btfs             = make(btfHandleCache)
+		handles          = newHandleCache()
 		skipMapsAndProgs = false
 	)
 
 	cleanup = func() {
-		btfs.close()
+		handles.close()
 
 		if skipMapsAndProgs {
 			return
@@ -335,7 +362,7 @@ func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
 			return nil, fmt.Errorf("missing map %s", mapName)
 		}
 
-		m, err := newMapWithOptions(mapSpec, opts.Maps, btfs)
+		m, err := newMapWithOptions(mapSpec, opts.Maps, handles)
 		if err != nil {
 			return nil, fmt.Errorf("map %s: %w", mapName, err)
 		}
@@ -360,7 +387,7 @@ func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
 		for i := range progSpec.Instructions {
 			ins := &progSpec.Instructions[i]
 
-			if ins.OpCode != asm.LoadImmOp(asm.DWord) || ins.Reference == "" {
+			if !ins.IsLoadFromMap() || ins.Reference == "" {
 				continue
 			}
 
@@ -372,7 +399,7 @@ func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
 
 			m, err := loadMap(ins.Reference)
 			if err != nil {
-				return nil, fmt.Errorf("program %s: %s", progName, err)
+				return nil, fmt.Errorf("program %s: %w", progName, err)
 			}
 
 			fd := m.FD()
@@ -384,7 +411,7 @@ func lazyLoadCollection(coll *CollectionSpec, opts *CollectionOptions) (
 			}
 		}
 
-		prog, err := newProgramWithOptions(progSpec, opts.Programs, btfs)
+		prog, err := newProgramWithOptions(progSpec, opts.Programs, handles)
 		if err != nil {
 			return nil, fmt.Errorf("program %s: %w", progName, err)
 		}
@@ -534,7 +561,7 @@ func assignValues(to interface{}, valueOf func(reflect.Type, string) (reflect.Va
 			}
 
 			if err != nil {
-				return fmt.Errorf("field %s: %s", field.Name, err)
+				return fmt.Errorf("field %s: %w", field.Name, err)
 			}
 		}
 

vendor/github.com/cilium/ebpf/elf_reader.go

@@ -96,7 +96,7 @@ func LoadCollectionSpecFromReader(rd io.ReaderAt) (*CollectionSpec, error) {
 	}
 
 	btfSpec, err := btf.LoadSpecFromReader(rd)
-	if err != nil {
+	if err != nil && !errors.Is(err, btf.ErrNotFound) {
 		return nil, fmt.Errorf("load BTF: %w", err)
 	}
 
@@ -159,7 +159,7 @@ func LoadCollectionSpecFromReader(rd io.ReaderAt) (*CollectionSpec, error) {
 			}
 
 			if target.Flags&elf.SHF_STRINGS > 0 {
-				return nil, fmt.Errorf("section %q: string %q is not stack allocated: %w", section.Name, rel.Name, ErrNotSupported)
+				return nil, fmt.Errorf("section %q: string is not stack allocated: %w", section.Name, ErrNotSupported)
 			}
 
 			target.references++
@@ -374,17 +374,25 @@ func (ec *elfCode) relocateInstruction(ins *asm.Instruction, rel elf.Symbol) err
 		}
 
 	case dataSection:
+		var offset uint32
 		switch typ {
 		case elf.STT_SECTION:
 			if bind != elf.STB_LOCAL {
 				return fmt.Errorf("direct load: %s: unsupported relocation %s", name, bind)
 			}
 
+			// This is really a reference to a static symbol, which clang doesn't
+			// emit a symbol table entry for. Instead it encodes the offset in
+			// the instruction itself.
+			offset = uint32(uint64(ins.Constant))
+
 		case elf.STT_OBJECT:
 			if bind != elf.STB_GLOBAL {
 				return fmt.Errorf("direct load: %s: unsupported relocation %s", name, bind)
 			}
 
+			offset = uint32(rel.Value)
+
 		default:
 			return fmt.Errorf("incorrect relocation type %v for direct map load", typ)
 		}
@@ -394,10 +402,8 @@ func (ec *elfCode) relocateInstruction(ins *asm.Instruction, rel elf.Symbol) err
 		// it's not clear how to encode that into Instruction.
 		name = target.Name
 
-		// For some reason, clang encodes the offset of the symbol its
+		// The kernel expects the offset in the second basic BPF instruction.
-		// section in the first basic BPF instruction, while the kernel
+		ins.Constant = int64(uint64(offset) << 32)
-		// expects it in the second one.
-		ins.Constant <<= 32
 		ins.Src = asm.PseudoMapValue
 
 		// Mark the instruction as needing an update when creating the
@@ -491,33 +497,38 @@ func (ec *elfCode) loadMaps(maps map[string]*MapSpec) error {
 				return fmt.Errorf("section %s: missing symbol for map at offset %d", sec.Name, offset)
 			}
 
-			if maps[mapSym.Name] != nil {
+			mapName := mapSym.Name
+			if maps[mapName] != nil {
 				return fmt.Errorf("section %v: map %v already exists", sec.Name, mapSym)
 			}
 
 			lr := io.LimitReader(r, int64(size))
 
 			spec := MapSpec{
-				Name: SanitizeName(mapSym.Name, -1),
+				Name: SanitizeName(mapName, -1),
 			}
 			switch {
 			case binary.Read(lr, ec.ByteOrder, &spec.Type) != nil:
-				return fmt.Errorf("map %v: missing type", mapSym)
+				return fmt.Errorf("map %s: missing type", mapName)
 			case binary.Read(lr, ec.ByteOrder, &spec.KeySize) != nil:
-				return fmt.Errorf("map %v: missing key size", mapSym)
+				return fmt.Errorf("map %s: missing key size", mapName)
 			case binary.Read(lr, ec.ByteOrder, &spec.ValueSize) != nil:
-				return fmt.Errorf("map %v: missing value size", mapSym)
+				return fmt.Errorf("map %s: missing value size", mapName)
 			case binary.Read(lr, ec.ByteOrder, &spec.MaxEntries) != nil:
-				return fmt.Errorf("map %v: missing max entries", mapSym)
+				return fmt.Errorf("map %s: missing max entries", mapName)
 			case binary.Read(lr, ec.ByteOrder, &spec.Flags) != nil:
-				return fmt.Errorf("map %v: missing flags", mapSym)
+				return fmt.Errorf("map %s: missing flags", mapName)
 			}
 
 			if _, err := io.Copy(internal.DiscardZeroes{}, lr); err != nil {
-				return fmt.Errorf("map %v: unknown and non-zero fields in definition", mapSym)
+				return fmt.Errorf("map %s: unknown and non-zero fields in definition", mapName)
 			}
 
-			maps[mapSym.Name] = &spec
+			if err := spec.clampPerfEventArraySize(); err != nil {
+				return fmt.Errorf("map %s: %w", mapName, err)
+			}
+
+			maps[mapName] = &spec
 		}
 	}
 
@@ -565,6 +576,10 @@ func (ec *elfCode) loadBTFMaps(maps map[string]*MapSpec) error {
 				return fmt.Errorf("map %v: %w", name, err)
 			}
 
+			if err := mapSpec.clampPerfEventArraySize(); err != nil {
+				return fmt.Errorf("map %v: %w", name, err)
+			}
+
 			maps[name] = mapSpec
 		}
 	}
@@ -847,6 +862,8 @@ func getProgType(sectionName string) (ProgramType, AttachType, uint32, string) {
 		"uretprobe/":            {Kprobe, AttachNone, 0},
 		"tracepoint/":           {TracePoint, AttachNone, 0},
 		"raw_tracepoint/":       {RawTracepoint, AttachNone, 0},
+		"raw_tp/":               {RawTracepoint, AttachNone, 0},
+		"tp_btf/":               {Tracing, AttachTraceRawTp, 0},
 		"xdp":                   {XDP, AttachNone, 0},
 		"perf_event":            {PerfEvent, AttachNone, 0},
 		"lwt_in":                {LWTIn, AttachNone, 0},
@@ -860,6 +877,9 @@ func getProgType(sectionName string) (ProgramType, AttachType, uint32, string) {
 		"lirc_mode2":            {LircMode2, AttachLircMode2, 0},
 		"flow_dissector":        {FlowDissector, AttachFlowDissector, 0},
 		"iter/":                 {Tracing, AttachTraceIter, 0},
+		"fentry/":               {Tracing, AttachTraceFEntry, 0},
+		"fmod_ret/":             {Tracing, AttachModifyReturn, 0},
+		"fexit/":                {Tracing, AttachTraceFExit, 0},
 		"fentry.s/":             {Tracing, AttachTraceFEntry, unix.BPF_F_SLEEPABLE},
 		"fmod_ret.s/":           {Tracing, AttachModifyReturn, unix.BPF_F_SLEEPABLE},
 		"fexit.s/":              {Tracing, AttachTraceFExit, unix.BPF_F_SLEEPABLE},

vendor/github.com/cilium/ebpf/internal/btf/btf.go

@@ -35,7 +35,7 @@ type Spec struct {
 	namedTypes map[string][]namedType
 	funcInfos  map[string]extInfo
 	lineInfos  map[string]extInfo
-	coreRelos  map[string]bpfCoreRelos
+	coreRelos  map[string]coreRelos
 	byteOrder  binary.ByteOrder
 }
 
@@ -53,7 +53,7 @@ type btfHeader struct {
 
 // LoadSpecFromReader reads BTF sections from an ELF.
 //
-// Returns a nil Spec and no error if no BTF was present.
+// Returns ErrNotFound if the reader contains no BTF.
 func LoadSpecFromReader(rd io.ReaderAt) (*Spec, error) {
 	file, err := internal.NewSafeELFFile(rd)
 	if err != nil {
@@ -67,7 +67,7 @@ func LoadSpecFromReader(rd io.ReaderAt) (*Spec, error) {
 	}
 
 	if btfSection == nil {
-		return nil, nil
+		return nil, fmt.Errorf("btf: %w", ErrNotFound)
 	}
 
 	symbols, err := file.Symbols()
@@ -377,7 +377,7 @@ func (s *Spec) marshal(opts marshalOpts) ([]byte, error) {
 	for _, raw := range s.rawTypes {
 		switch {
 		case opts.StripFuncLinkage && raw.Kind() == kindFunc:
-			raw.SetLinkage(linkageStatic)
+			raw.SetLinkage(StaticFunc)
 		}
 
 		if err := raw.Marshal(&buf, opts.ByteOrder); err != nil {
@@ -438,13 +438,13 @@ func (s *Spec) Program(name string, length uint64) (*Program, error) {
 
 	funcInfos, funcOK := s.funcInfos[name]
 	lineInfos, lineOK := s.lineInfos[name]
-	coreRelos, coreOK := s.coreRelos[name]
+	relos, coreOK := s.coreRelos[name]
 
 	if !funcOK && !lineOK && !coreOK {
 		return nil, fmt.Errorf("no extended BTF info for section %s", name)
 	}
 
-	return &Program{s, length, funcInfos, lineInfos, coreRelos}, nil
+	return &Program{s, length, funcInfos, lineInfos, relos}, nil
 }
 
 // Datasec returns the BTF required to create maps which represent data sections.
@@ -491,7 +491,8 @@ func (s *Spec) FindType(name string, typ Type) error {
 		return fmt.Errorf("type %s: %w", name, ErrNotFound)
 	}
 
-	value := reflect.Indirect(reflect.ValueOf(copyType(candidate)))
+	cpy, _ := copyType(candidate, nil)
+	value := reflect.Indirect(reflect.ValueOf(cpy))
 	reflect.Indirect(reflect.ValueOf(typ)).Set(value)
 	return nil
 }
@@ -606,7 +607,7 @@ type Program struct {
 	spec                 *Spec
 	length               uint64
 	funcInfos, lineInfos extInfo
-	coreRelos            bpfCoreRelos
+	coreRelos            coreRelos
 }
 
 // ProgramSpec returns the Spec needed for loading function and line infos into the kernel.
@@ -665,16 +666,23 @@ func ProgramLineInfos(s *Program) (recordSize uint32, bytes []byte, err error) {
 	return s.lineInfos.recordSize, bytes, nil
 }
 
-// ProgramRelocations returns the CO-RE relocations required to adjust the
+// ProgramFixups returns the changes required to adjust the program to the target.
-// program to the target.
 //
 // This is a free function instead of a method to hide it from users
 // of package ebpf.
-func ProgramRelocations(s *Program, target *Spec) (map[uint64]Relocation, error) {
+func ProgramFixups(s *Program, target *Spec) (COREFixups, error) {
 	if len(s.coreRelos) == 0 {
 		return nil, nil
 	}
 
+	if target == nil {
+		var err error
+		target, err = LoadKernelSpec()
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	return coreRelocate(s.spec, target, s.coreRelos)
 }
 
@@ -771,7 +779,7 @@ var haveFuncLinkage = internal.FeatureTest("BTF func linkage", "5.6", func() err
 	types.Func.SetKind(kindFunc)
 	types.Func.SizeType = 1 // aka FuncProto
 	types.Func.NameOff = 1
-	types.Func.SetLinkage(linkageGlobal)
+	types.Func.SetLinkage(GlobalFunc)
 
 	btf := marshalBTF(&types, strings, internal.NativeEndian)
 

vendor/github.com/cilium/ebpf/internal/btf/btf_types.go

@@ -6,6 +6,8 @@ import (
 	"io"
 )
 
+//go:generate stringer -linecomment -output=btf_types_string.go -type=FuncLinkage,VarLinkage
+
 // btfKind describes a Type.
 type btfKind uint8
 
@@ -31,14 +33,23 @@ const (
 	kindDatasec
 )
 
-// btfFuncLinkage describes BTF function linkage metadata.
+// FuncLinkage describes BTF function linkage metadata.
-type btfFuncLinkage uint8
+type FuncLinkage int
 
 // Equivalent of enum btf_func_linkage.
 const (
-	linkageStatic btfFuncLinkage = iota
+	StaticFunc FuncLinkage = iota // static
-	linkageGlobal
+	GlobalFunc                    // global
-	// linkageExtern // Currently unused in libbpf.
+	ExternFunc                    // extern
+)
+
+// VarLinkage describes BTF variable linkage metadata.
+type VarLinkage int
+
+const (
+	StaticVar VarLinkage = iota // static
+	GlobalVar                   // global
+	ExternVar                   // extern
 )
 
 const (
@@ -144,11 +155,11 @@ func (bt *btfType) KindFlag() bool {
 	return bt.info(btfTypeKindFlagMask, btfTypeKindFlagShift) == 1
 }
 
-func (bt *btfType) Linkage() btfFuncLinkage {
+func (bt *btfType) Linkage() FuncLinkage {
-	return btfFuncLinkage(bt.info(btfTypeVlenMask, btfTypeVlenShift))
+	return FuncLinkage(bt.info(btfTypeVlenMask, btfTypeVlenShift))
 }
 
-func (bt *btfType) SetLinkage(linkage btfFuncLinkage) {
+func (bt *btfType) SetLinkage(linkage FuncLinkage) {
 	bt.setInfo(uint32(linkage), btfTypeVlenMask, btfTypeVlenShift)
 }
 

vendor/github.com/cilium/ebpf/internal/btf/btf_types_string.go

@@ -0,0 +1,44 @@
+// Code generated by "stringer -linecomment -output=btf_types_string.go -type=FuncLinkage,VarLinkage"; DO NOT EDIT.
+
+package btf
+
+import "strconv"
+
+func _() {
+	// An "invalid array index" compiler error signifies that the constant values have changed.
+	// Re-run the stringer command to generate them again.
+	var x [1]struct{}
+	_ = x[StaticFunc-0]
+	_ = x[GlobalFunc-1]
+	_ = x[ExternFunc-2]
+}
+
+const _FuncLinkage_name = "staticglobalextern"
+
+var _FuncLinkage_index = [...]uint8{0, 6, 12, 18}
+
+func (i FuncLinkage) String() string {
+	if i < 0 || i >= FuncLinkage(len(_FuncLinkage_index)-1) {
+		return "FuncLinkage(" + strconv.FormatInt(int64(i), 10) + ")"
+	}
+	return _FuncLinkage_name[_FuncLinkage_index[i]:_FuncLinkage_index[i+1]]
+}
+func _() {
+	// An "invalid array index" compiler error signifies that the constant values have changed.
+	// Re-run the stringer command to generate them again.
+	var x [1]struct{}
+	_ = x[StaticVar-0]
+	_ = x[GlobalVar-1]
+	_ = x[ExternVar-2]
+}
+
+const _VarLinkage_name = "staticglobalextern"
+
+var _VarLinkage_index = [...]uint8{0, 6, 12, 18}
+
+func (i VarLinkage) String() string {
+	if i < 0 || i >= VarLinkage(len(_VarLinkage_index)-1) {
+		return "VarLinkage(" + strconv.FormatInt(int64(i), 10) + ")"
+	}
+	return _VarLinkage_name[_VarLinkage_index[i]:_VarLinkage_index[i+1]]
+}

vendor/github.com/cilium/ebpf/internal/btf/core.go

@@ -3,29 +3,146 @@ package btf
 import (
 	"errors"
 	"fmt"
+	"math"
 	"reflect"
+	"sort"
 	"strconv"
 	"strings"
+
+	"github.com/cilium/ebpf/asm"
 )
 
 // Code in this file is derived from libbpf, which is available under a BSD
 // 2-Clause license.
 
-// Relocation describes a CO-RE relocation.
+// COREFixup is the result of computing a CO-RE relocation for a target.
-type Relocation struct {
+type COREFixup struct {
-	Current uint32
+	Kind   COREKind
-	New     uint32
+	Local  uint32
+	Target uint32
+	Poison bool
 }
 
-func (r Relocation) equal(other Relocation) bool {
+func (f COREFixup) equal(other COREFixup) bool {
-	return r.Current == other.Current && r.New == other.New
+	return f.Local == other.Local && f.Target == other.Target
 }
 
-// coreReloKind is the type of CO-RE relocation
+func (f COREFixup) String() string {
-type coreReloKind uint32
+	if f.Poison {
+		return fmt.Sprintf("%s=poison", f.Kind)
+	}
+	return fmt.Sprintf("%s=%d->%d", f.Kind, f.Local, f.Target)
+}
+
+func (f COREFixup) apply(ins *asm.Instruction) error {
+	if f.Poison {
+		return errors.New("can't poison individual instruction")
+	}
+
+	switch class := ins.OpCode.Class(); class {
+	case asm.LdXClass, asm.StClass, asm.StXClass:
+		if want := int16(f.Local); want != ins.Offset {
+			return fmt.Errorf("invalid offset %d, expected %d", ins.Offset, want)
+		}
+
+		if f.Target > math.MaxInt16 {
+			return fmt.Errorf("offset %d exceeds MaxInt16", f.Target)
+		}
+
+		ins.Offset = int16(f.Target)
+
+	case asm.LdClass:
+		if !ins.IsConstantLoad(asm.DWord) {
+			return fmt.Errorf("not a dword-sized immediate load")
+		}
+
+		if want := int64(f.Local); want != ins.Constant {
+			return fmt.Errorf("invalid immediate %d, expected %d", ins.Constant, want)
+		}
+
+		ins.Constant = int64(f.Target)
+
+	case asm.ALUClass:
+		if ins.OpCode.ALUOp() == asm.Swap {
+			return fmt.Errorf("relocation against swap")
+		}
+
+		fallthrough
+
+	case asm.ALU64Class:
+		if src := ins.OpCode.Source(); src != asm.ImmSource {
+			return fmt.Errorf("invalid source %s", src)
+		}
+
+		if want := int64(f.Local); want != ins.Constant {
+			return fmt.Errorf("invalid immediate %d, expected %d", ins.Constant, want)
+		}
+
+		if f.Target > math.MaxInt32 {
+			return fmt.Errorf("immediate %d exceeds MaxInt32", f.Target)
+		}
+
+		ins.Constant = int64(f.Target)
+
+	default:
+		return fmt.Errorf("invalid class %s", class)
+	}
+
+	return nil
+}
+
+func (f COREFixup) isNonExistant() bool {
+	return f.Kind.checksForExistence() && f.Target == 0
+}
+
+type COREFixups map[uint64]COREFixup
+
+// Apply a set of CO-RE relocations to a BPF program.
+func (fs COREFixups) Apply(insns asm.Instructions) (asm.Instructions, error) {
+	if len(fs) == 0 {
+		cpy := make(asm.Instructions, len(insns))
+		copy(cpy, insns)
+		return insns, nil
+	}
+
+	cpy := make(asm.Instructions, 0, len(insns))
+	iter := insns.Iterate()
+	for iter.Next() {
+		fixup, ok := fs[iter.Offset.Bytes()]
+		if !ok {
+			cpy = append(cpy, *iter.Ins)
+			continue
+		}
+
+		ins := *iter.Ins
+		if fixup.Poison {
+			const badRelo = asm.BuiltinFunc(0xbad2310)
+
+			cpy = append(cpy, badRelo.Call())
+			if ins.OpCode.IsDWordLoad() {
+				// 64 bit constant loads occupy two raw bpf instructions, so
+				// we need to add another instruction as padding.
+				cpy = append(cpy, badRelo.Call())
+			}
+
+			continue
+		}
+
+		if err := fixup.apply(&ins); err != nil {
+			return nil, fmt.Errorf("instruction %d, offset %d: %s: %w", iter.Index, iter.Offset.Bytes(), fixup.Kind, err)
+		}
+
+		cpy = append(cpy, ins)
+	}
+
+	return cpy, nil
+}
+
+// COREKind is the type of CO-RE relocation
+type COREKind uint32
 
 const (
-	reloFieldByteOffset coreReloKind = iota /* field byte offset */
+	reloFieldByteOffset COREKind = iota /* field byte offset */
 	reloFieldByteSize                   /* field size in bytes */
 	reloFieldExists                     /* field existence in target kernel */
 	reloFieldSigned                     /* field signedness (0 - unsigned, 1 - signed) */
@@ -39,7 +156,7 @@ const (
 	reloEnumvalValue                    /* enum value integer value */
 )
 
-func (k coreReloKind) String() string {
+func (k COREKind) String() string {
 	switch k {
 	case reloFieldByteOffset:
 		return "byte_off"
@@ -70,103 +187,249 @@ func (k coreReloKind) String() string {
 	}
 }
 
-func coreRelocate(local, target *Spec, coreRelos bpfCoreRelos) (map[uint64]Relocation, error) {
+func (k COREKind) checksForExistence() bool {
-	if target == nil {
+	return k == reloEnumvalExists || k == reloTypeExists || k == reloFieldExists
-		var err error
+}
-		target, err = loadKernelSpec()
-		if err != nil {
-			return nil, err
-		}
-	}
 
+func coreRelocate(local, target *Spec, relos coreRelos) (COREFixups, error) {
 	if local.byteOrder != target.byteOrder {
 		return nil, fmt.Errorf("can't relocate %s against %s", local.byteOrder, target.byteOrder)
 	}
 
-	relocations := make(map[uint64]Relocation, len(coreRelos))
+	var ids []TypeID
-	for _, relo := range coreRelos {
+	relosByID := make(map[TypeID]coreRelos)
-		accessorStr, err := local.strings.Lookup(relo.AccessStrOff)
+	result := make(COREFixups, len(relos))
+	for _, relo := range relos {
+		if relo.kind == reloTypeIDLocal {
+			// Filtering out reloTypeIDLocal here makes our lives a lot easier
+			// down the line, since it doesn't have a target at all.
+			if len(relo.accessor) > 1 || relo.accessor[0] != 0 {
+				return nil, fmt.Errorf("%s: unexpected accessor %v", relo.kind, relo.accessor)
+			}
+
+			result[uint64(relo.insnOff)] = COREFixup{
+				relo.kind,
+				uint32(relo.typeID),
+				uint32(relo.typeID),
+				false,
+			}
+			continue
+		}
+
+		relos, ok := relosByID[relo.typeID]
+		if !ok {
+			ids = append(ids, relo.typeID)
+		}
+		relosByID[relo.typeID] = append(relos, relo)
+	}
+
+	// Ensure we work on relocations in a deterministic order.
+	sort.Slice(ids, func(i, j int) bool {
+		return ids[i] < ids[j]
+	})
+
+	for _, id := range ids {
+		if int(id) >= len(local.types) {
+			return nil, fmt.Errorf("invalid type id %d", id)
+		}
+
+		localType := local.types[id]
+		named, ok := localType.(namedType)
+		if !ok || named.name() == "" {
+			return nil, fmt.Errorf("relocate unnamed or anonymous type %s: %w", localType, ErrNotSupported)
+		}
+
+		relos := relosByID[id]
+		targets := target.namedTypes[named.essentialName()]
+		fixups, err := coreCalculateFixups(localType, targets, relos)
+		if err != nil {
+			return nil, fmt.Errorf("relocate %s: %w", localType, err)
+		}
+
+		for i, relo := range relos {
+			result[uint64(relo.insnOff)] = fixups[i]
+		}
+	}
+
+	return result, nil
+}
+
+var errAmbiguousRelocation = errors.New("ambiguous relocation")
+var errImpossibleRelocation = errors.New("impossible relocation")
+
+// coreCalculateFixups calculates the fixups for the given relocations using
+// the "best" target.
+//
+// The best target is determined by scoring: the less poisoning we have to do
+// the better the target is.
+func coreCalculateFixups(local Type, targets []namedType, relos coreRelos) ([]COREFixup, error) {
+	localID := local.ID()
+	local, err := copyType(local, skipQualifierAndTypedef)
 	if err != nil {
 		return nil, err
 	}
 
-		accessor, err := parseCoreAccessor(accessorStr)
+	bestScore := len(relos)
+	var bestFixups []COREFixup
+	for i := range targets {
+		targetID := targets[i].ID()
+		target, err := copyType(targets[i], skipQualifierAndTypedef)
 		if err != nil {
-			return nil, fmt.Errorf("accessor %q: %s", accessorStr, err)
+			return nil, err
 		}
 
-		if int(relo.TypeID) >= len(local.types) {
+		score := 0 // lower is better
-			return nil, fmt.Errorf("invalid type id %d", relo.TypeID)
+		fixups := make([]COREFixup, 0, len(relos))
+		for _, relo := range relos {
+			fixup, err := coreCalculateFixup(local, localID, target, targetID, relo)
+			if err != nil {
+				return nil, fmt.Errorf("target %s: %w", target, err)
+			}
+			if fixup.Poison || fixup.isNonExistant() {
+				score++
+			}
+			fixups = append(fixups, fixup)
 		}
 
-		typ := local.types[relo.TypeID]
+		if score > bestScore {
-
+			// We have a better target already, ignore this one.
-		if relo.ReloKind == reloTypeIDLocal {
-			relocations[uint64(relo.InsnOff)] = Relocation{
-				uint32(typ.ID()),
-				uint32(typ.ID()),
-			}
 			continue
 		}
 
-		named, ok := typ.(namedType)
+		if score < bestScore {
-		if !ok || named.name() == "" {
+			// This is the best target yet, use it.
-			return nil, fmt.Errorf("relocate anonymous type %s: %w", typ.String(), ErrNotSupported)
+			bestScore = score
+			bestFixups = fixups
+			continue
 		}
 
-		name := essentialName(named.name())
+		// Some other target has the same score as the current one. Make sure
-		res, err := coreCalculateRelocation(typ, target.namedTypes[name], relo.ReloKind, accessor)
+		// the fixups agree with each other.
-		if err != nil {
+		for i, fixup := range bestFixups {
-			return nil, fmt.Errorf("relocate %s: %w", name, err)
+			if !fixup.equal(fixups[i]) {
+				return nil, fmt.Errorf("%s: multiple types match: %w", fixup.Kind, errAmbiguousRelocation)
+			}
+		}
 	}
 
-		relocations[uint64(relo.InsnOff)] = res
+	if bestFixups == nil {
+		// Nothing at all matched, probably because there are no suitable
+		// targets at all. Poison everything!
+		bestFixups = make([]COREFixup, len(relos))
+		for i, relo := range relos {
+			bestFixups[i] = COREFixup{Kind: relo.kind, Poison: true}
+		}
 	}
 
-	return relocations, nil
+	return bestFixups, nil
 }
 
-var errAmbiguousRelocation = errors.New("ambiguous relocation")
+// coreCalculateFixup calculates the fixup for a single local type, target type
+// and relocation.
+func coreCalculateFixup(local Type, localID TypeID, target Type, targetID TypeID, relo coreRelo) (COREFixup, error) {
+	fixup := func(local, target uint32) (COREFixup, error) {
+		return COREFixup{relo.kind, local, target, false}, nil
+	}
+	poison := func() (COREFixup, error) {
+		if relo.kind.checksForExistence() {
+			return fixup(1, 0)
+		}
+		return COREFixup{relo.kind, 0, 0, true}, nil
+	}
+	zero := COREFixup{}
+
+	switch relo.kind {
+	case reloTypeIDTarget, reloTypeSize, reloTypeExists:
+		if len(relo.accessor) > 1 || relo.accessor[0] != 0 {
+			return zero, fmt.Errorf("%s: unexpected accessor %v", relo.kind, relo.accessor)
+		}
+
+		err := coreAreTypesCompatible(local, target)
+		if errors.Is(err, errImpossibleRelocation) {
+			return poison()
+		}
+		if err != nil {
+			return zero, fmt.Errorf("relocation %s: %w", relo.kind, err)
+		}
+
+		switch relo.kind {
+		case reloTypeExists:
+			return fixup(1, 1)
 
-func coreCalculateRelocation(local Type, targets []namedType, kind coreReloKind, localAccessor coreAccessor) (Relocation, error) {
-	var relos []Relocation
-	var matches []Type
-	for _, target := range targets {
-		switch kind {
 		case reloTypeIDTarget:
-			if localAccessor[0] != 0 {
+			return fixup(uint32(localID), uint32(targetID))
-				return Relocation{}, fmt.Errorf("%s: unexpected non-zero accessor", kind)
+
+		case reloTypeSize:
+			localSize, err := Sizeof(local)
+			if err != nil {
+				return zero, err
 			}
 
-			if compat, err := coreAreTypesCompatible(local, target); err != nil {
+			targetSize, err := Sizeof(target)
-				return Relocation{}, fmt.Errorf("%s: %s", kind, err)
+			if err != nil {
-			} else if !compat {
+				return zero, err
-				continue
 			}
 
-			relos = append(relos, Relocation{uint32(target.ID()), uint32(target.ID())})
+			return fixup(uint32(localSize), uint32(targetSize))
-
-		default:
-			return Relocation{}, fmt.Errorf("relocation %s: %w", kind, ErrNotSupported)
-		}
-		matches = append(matches, target)
 		}
 
-	if len(relos) == 0 {
+	case reloEnumvalValue, reloEnumvalExists:
-		// TODO: Add switch for existence checks like reloEnumvalExists here.
+		localValue, targetValue, err := coreFindEnumValue(local, relo.accessor, target)
-
+		if errors.Is(err, errImpossibleRelocation) {
-		// TODO: This might have to be poisoned.
+			return poison()
-		return Relocation{}, fmt.Errorf("no relocation found, tried %v", targets)
+		}
+		if err != nil {
+			return zero, fmt.Errorf("relocation %s: %w", relo.kind, err)
 		}
 
-	relo := relos[0]
+		switch relo.kind {
-	for _, altRelo := range relos[1:] {
+		case reloEnumvalExists:
-		if !altRelo.equal(relo) {
+			return fixup(1, 1)
-			return Relocation{}, fmt.Errorf("multiple types %v match: %w", matches, errAmbiguousRelocation)
+
+		case reloEnumvalValue:
+			return fixup(uint32(localValue.Value), uint32(targetValue.Value))
+		}
+
+	case reloFieldByteOffset, reloFieldByteSize, reloFieldExists:
+		if _, ok := target.(*Fwd); ok {
+			// We can't relocate fields using a forward declaration, so
+			// skip it. If a non-forward declaration is present in the BTF
+			// we'll find it in one of the other iterations.
+			return poison()
+		}
+
+		localField, targetField, err := coreFindField(local, relo.accessor, target)
+		if errors.Is(err, errImpossibleRelocation) {
+			return poison()
+		}
+		if err != nil {
+			return zero, fmt.Errorf("target %s: %w", target, err)
+		}
+
+		switch relo.kind {
+		case reloFieldExists:
+			return fixup(1, 1)
+
+		case reloFieldByteOffset:
+			return fixup(localField.offset/8, targetField.offset/8)
+
+		case reloFieldByteSize:
+			localSize, err := Sizeof(localField.Type)
+			if err != nil {
+				return zero, err
+			}
+
+			targetSize, err := Sizeof(targetField.Type)
+			if err != nil {
+				return zero, err
+			}
+
+			return fixup(uint32(localSize), uint32(targetSize))
+
 		}
 	}
 
-	return relo, nil
+	return zero, fmt.Errorf("relocation %s: %w", relo.kind, ErrNotSupported)
 }
 
 /* coreAccessor contains a path through a struct. It contains at least one index.
@@ -219,6 +482,240 @@ func parseCoreAccessor(accessor string) (coreAccessor, error) {
 	return result, nil
 }
 
+func (ca coreAccessor) String() string {
+	strs := make([]string, 0, len(ca))
+	for _, i := range ca {
+		strs = append(strs, strconv.Itoa(i))
+	}
+	return strings.Join(strs, ":")
+}
+
+func (ca coreAccessor) enumValue(t Type) (*EnumValue, error) {
+	e, ok := t.(*Enum)
+	if !ok {
+		return nil, fmt.Errorf("not an enum: %s", t)
+	}
+
+	if len(ca) > 1 {
+		return nil, fmt.Errorf("invalid accessor %s for enum", ca)
+	}
+
+	i := ca[0]
+	if i >= len(e.Values) {
+		return nil, fmt.Errorf("invalid index %d for %s", i, e)
+	}
+
+	return &e.Values[i], nil
+}
+
+type coreField struct {
+	Type   Type
+	offset uint32
+}
+
+func adjustOffset(base uint32, t Type, n int) (uint32, error) {
+	size, err := Sizeof(t)
+	if err != nil {
+		return 0, err
+	}
+
+	return base + (uint32(n) * uint32(size) * 8), nil
+}
+
+// coreFindField descends into the local type using the accessor and tries to
+// find an equivalent field in target at each step.
+//
+// Returns the field and the offset of the field from the start of
+// target in bits.
+func coreFindField(local Type, localAcc coreAccessor, target Type) (_, _ coreField, _ error) {
+	// The first index is used to offset a pointer of the base type like
+	// when accessing an array.
+	localOffset, err := adjustOffset(0, local, localAcc[0])
+	if err != nil {
+		return coreField{}, coreField{}, err
+	}
+
+	targetOffset, err := adjustOffset(0, target, localAcc[0])
+	if err != nil {
+		return coreField{}, coreField{}, err
+	}
+
+	if err := coreAreMembersCompatible(local, target); err != nil {
+		return coreField{}, coreField{}, fmt.Errorf("fields: %w", err)
+	}
+
+	var localMaybeFlex, targetMaybeFlex bool
+	for _, acc := range localAcc[1:] {
+		switch localType := local.(type) {
+		case composite:
+			// For composite types acc is used to find the field in the local type,
+			// and then we try to find a field in target with the same name.
+			localMembers := localType.members()
+			if acc >= len(localMembers) {
+				return coreField{}, coreField{}, fmt.Errorf("invalid accessor %d for %s", acc, local)
+			}
+
+			localMember := localMembers[acc]
+			if localMember.Name == "" {
+				_, ok := localMember.Type.(composite)
+				if !ok {
+					return coreField{}, coreField{}, fmt.Errorf("unnamed field with type %s: %s", localMember.Type, ErrNotSupported)
+				}
+
+				// This is an anonymous struct or union, ignore it.
+				local = localMember.Type
+				localOffset += localMember.Offset
+				localMaybeFlex = false
+				continue
+			}
+
+			targetType, ok := target.(composite)
+			if !ok {
+				return coreField{}, coreField{}, fmt.Errorf("target not composite: %w", errImpossibleRelocation)
+			}
+
+			targetMember, last, err := coreFindMember(targetType, localMember.Name)
+			if err != nil {
+				return coreField{}, coreField{}, err
+			}
+
+			if targetMember.BitfieldSize > 0 {
+				return coreField{}, coreField{}, fmt.Errorf("field %q is a bitfield: %w", targetMember.Name, ErrNotSupported)
+			}
+
+			local = localMember.Type
+			localMaybeFlex = acc == len(localMembers)-1
+			localOffset += localMember.Offset
+			target = targetMember.Type
+			targetMaybeFlex = last
+			targetOffset += targetMember.Offset
+
+		case *Array:
+			// For arrays, acc is the index in the target.
+			targetType, ok := target.(*Array)
+			if !ok {
+				return coreField{}, coreField{}, fmt.Errorf("target not array: %w", errImpossibleRelocation)
+			}
+
+			if localType.Nelems == 0 && !localMaybeFlex {
+				return coreField{}, coreField{}, fmt.Errorf("local type has invalid flexible array")
+			}
+			if targetType.Nelems == 0 && !targetMaybeFlex {
+				return coreField{}, coreField{}, fmt.Errorf("target type has invalid flexible array")
+			}
+
+			if localType.Nelems > 0 && acc >= int(localType.Nelems) {
+				return coreField{}, coreField{}, fmt.Errorf("invalid access of %s at index %d", localType, acc)
+			}
+			if targetType.Nelems > 0 && acc >= int(targetType.Nelems) {
+				return coreField{}, coreField{}, fmt.Errorf("out of bounds access of target: %w", errImpossibleRelocation)
+			}
+
+			local = localType.Type
+			localMaybeFlex = false
+			localOffset, err = adjustOffset(localOffset, local, acc)
+			if err != nil {
+				return coreField{}, coreField{}, err
+			}
+
+			target = targetType.Type
+			targetMaybeFlex = false
+			targetOffset, err = adjustOffset(targetOffset, target, acc)
+			if err != nil {
+				return coreField{}, coreField{}, err
+			}
+
+		default:
+			return coreField{}, coreField{}, fmt.Errorf("relocate field of %T: %w", localType, ErrNotSupported)
+		}
+
+		if err := coreAreMembersCompatible(local, target); err != nil {
+			return coreField{}, coreField{}, err
+		}
+	}
+
+	return coreField{local, localOffset}, coreField{target, targetOffset}, nil
+}
+
+// coreFindMember finds a member in a composite type while handling anonymous
+// structs and unions.
+func coreFindMember(typ composite, name Name) (Member, bool, error) {
+	if name == "" {
+		return Member{}, false, errors.New("can't search for anonymous member")
+	}
+
+	type offsetTarget struct {
+		composite
+		offset uint32
+	}
+
+	targets := []offsetTarget{{typ, 0}}
+	visited := make(map[composite]bool)
+
+	for i := 0; i < len(targets); i++ {
+		target := targets[i]
+
+		// Only visit targets once to prevent infinite recursion.
+		if visited[target] {
+			continue
+		}
+		if len(visited) >= maxTypeDepth {
+			// This check is different than libbpf, which restricts the entire
+			// path to BPF_CORE_SPEC_MAX_LEN items.
+			return Member{}, false, fmt.Errorf("type is nested too deep")
+		}
+		visited[target] = true
+
+		members := target.members()
+		for j, member := range members {
+			if member.Name == name {
+				// NB: This is safe because member is a copy.
+				member.Offset += target.offset
+				return member, j == len(members)-1, nil
+			}
+
+			// The names don't match, but this member could be an anonymous struct
+			// or union.
+			if member.Name != "" {
+				continue
+			}
+
+			comp, ok := member.Type.(composite)
+			if !ok {
+				return Member{}, false, fmt.Errorf("anonymous non-composite type %T not allowed", member.Type)
+			}
+
+			targets = append(targets, offsetTarget{comp, target.offset + member.Offset})
+		}
+	}
+
+	return Member{}, false, fmt.Errorf("no matching member: %w", errImpossibleRelocation)
+}
+
+// coreFindEnumValue follows localAcc to find the equivalent enum value in target.
+func coreFindEnumValue(local Type, localAcc coreAccessor, target Type) (localValue, targetValue *EnumValue, _ error) {
+	localValue, err := localAcc.enumValue(local)
+	if err != nil {
+		return nil, nil, err
+	}
+
+	targetEnum, ok := target.(*Enum)
+	if !ok {
+		return nil, nil, errImpossibleRelocation
+	}
+
+	localName := localValue.Name.essentialName()
+	for i, targetValue := range targetEnum.Values {
+		if targetValue.Name.essentialName() != localName {
+			continue
+		}
+
+		return localValue, &targetEnum.Values[i], nil
+	}
+
+	return nil, nil, errImpossibleRelocation
+}
+
 /* The comment below is from bpf_core_types_are_compat in libbpf.c:
  *
  * Check local and target types for compatibility. This check is used for
@@ -239,8 +736,10 @@ func parseCoreAccessor(accessor string) (coreAccessor, error) {
  *     number of input args and compatible return and argument types.
  * These rules are not set in stone and probably will be adjusted as we get
  * more experience with using BPF CO-RE relocations.
+ *
+ * Returns errImpossibleRelocation if types are not compatible.
  */
-func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
+func coreAreTypesCompatible(localType Type, targetType Type) error {
 	var (
 		localTs, targetTs typeDeque
 		l, t              = &localType, &targetType
@@ -249,14 +748,14 @@ func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
 
 	for ; l != nil && t != nil; l, t = localTs.shift(), targetTs.shift() {
 		if depth >= maxTypeDepth {
-			return false, errors.New("types are nested too deep")
+			return errors.New("types are nested too deep")
 		}
 
-		localType = skipQualifierAndTypedef(*l)
+		localType = *l
-		targetType = skipQualifierAndTypedef(*t)
+		targetType = *t
 
 		if reflect.TypeOf(localType) != reflect.TypeOf(targetType) {
-			return false, nil
+			return fmt.Errorf("type mismatch: %w", errImpossibleRelocation)
 		}
 
 		switch lv := (localType).(type) {
@@ -266,7 +765,7 @@ func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
 		case *Int:
 			tv := targetType.(*Int)
 			if lv.isBitfield() || tv.isBitfield() {
-				return false, nil
+				return fmt.Errorf("bitfield: %w", errImpossibleRelocation)
 			}
 
 		case *Pointer, *Array:
@@ -277,7 +776,7 @@ func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
 		case *FuncProto:
 			tv := targetType.(*FuncProto)
 			if len(lv.Params) != len(tv.Params) {
-				return false, nil
+				return fmt.Errorf("function param mismatch: %w", errImpossibleRelocation)
 			}
 
 			depth++
@@ -285,22 +784,24 @@ func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
 			targetType.walk(&targetTs)
 
 		default:
-			return false, fmt.Errorf("unsupported type %T", localType)
+			return fmt.Errorf("unsupported type %T", localType)
 		}
 	}
 
 	if l != nil {
-		return false, fmt.Errorf("dangling local type %T", *l)
+		return fmt.Errorf("dangling local type %T", *l)
 	}
 
 	if t != nil {
-		return false, fmt.Errorf("dangling target type %T", *t)
+		return fmt.Errorf("dangling target type %T", *t)
 	}
 
-	return true, nil
+	return nil
 }
 
-/* The comment below is from bpf_core_fields_are_compat in libbpf.c:
+/* coreAreMembersCompatible checks two types for field-based relocation compatibility.
+ *
+ * The comment below is from bpf_core_fields_are_compat in libbpf.c:
  *
  * Check two types for compatibility for the purpose of field access
  * relocation. const/volatile/restrict and typedefs are skipped to ensure we
@@ -314,65 +815,63 @@ func coreAreTypesCompatible(localType Type, targetType Type) (bool, error) {
  *   - for INT, size and signedness are ignored;
  *   - for ARRAY, dimensionality is ignored, element types are checked for
  *     compatibility recursively;
+ *     [ NB: coreAreMembersCompatible doesn't recurse, this check is done
+ *       by coreFindField. ]
  *   - everything else shouldn't be ever a target of relocation.
  * These rules are not set in stone and probably will be adjusted as we get
  * more experience with using BPF CO-RE relocations.
+ *
+ * Returns errImpossibleRelocation if the members are not compatible.
  */
-func coreAreMembersCompatible(localType Type, targetType Type) (bool, error) {
+func coreAreMembersCompatible(localType Type, targetType Type) error {
-	doNamesMatch := func(a, b string) bool {
+	doNamesMatch := func(a, b string) error {
 		if a == "" || b == "" {
 			// allow anonymous and named type to match
-			return true
+			return nil
 		}
 
-		return essentialName(a) == essentialName(b)
+		if essentialName(a) == essentialName(b) {
+			return nil
 		}
 
-	for depth := 0; depth <= maxTypeDepth; depth++ {
+		return fmt.Errorf("names don't match: %w", errImpossibleRelocation)
-		localType = skipQualifierAndTypedef(localType)
+	}
-		targetType = skipQualifierAndTypedef(targetType)
 
 	_, lok := localType.(composite)
 	_, tok := targetType.(composite)
 	if lok && tok {
-			return true, nil
+		return nil
 	}
 
 	if reflect.TypeOf(localType) != reflect.TypeOf(targetType) {
-			return false, nil
+		return fmt.Errorf("type mismatch: %w", errImpossibleRelocation)
 	}
 
 	switch lv := localType.(type) {
-		case *Pointer:
+	case *Array, *Pointer:
-			return true, nil
+		return nil
 
 	case *Enum:
 		tv := targetType.(*Enum)
-			return doNamesMatch(lv.name(), tv.name()), nil
+		return doNamesMatch(lv.name(), tv.name())
 
 	case *Fwd:
 		tv := targetType.(*Fwd)
-			return doNamesMatch(lv.name(), tv.name()), nil
+		return doNamesMatch(lv.name(), tv.name())
 
 	case *Int:
 		tv := targetType.(*Int)
-			return !lv.isBitfield() && !tv.isBitfield(), nil
+		if lv.isBitfield() || tv.isBitfield() {
-
+			return fmt.Errorf("bitfield: %w", errImpossibleRelocation)
-		case *Array:
+		}
-			tv := targetType.(*Array)
+		return nil
-
-			localType = lv.Type
-			targetType = tv.Type
 
 	default:
-			return false, fmt.Errorf("unsupported type %T", localType)
+		return fmt.Errorf("type %s: %w", localType, ErrNotSupported)
 	}
-	}
-
-	return false, errors.New("types are nested too deep")
 }
 
-func skipQualifierAndTypedef(typ Type) Type {
+func skipQualifierAndTypedef(typ Type) (Type, error) {
 	result := typ
 	for depth := 0; depth <= maxTypeDepth; depth++ {
 		switch v := (result).(type) {
@@ -381,8 +880,8 @@ func skipQualifierAndTypedef(typ Type) Type {
 		case *Typedef:
 			result = v.Type
 		default:
-			return result
+			return result, nil
 		}
 	}
-	return typ
+	return nil, errors.New("exceeded type depth")
 }

vendor/github.com/cilium/ebpf/internal/btf/ext_info.go

@@ -30,7 +30,7 @@ type btfExtCoreHeader struct {
 	CoreReloLen uint32
 }
 
-func parseExtInfos(r io.ReadSeeker, bo binary.ByteOrder, strings stringTable) (funcInfo, lineInfo map[string]extInfo, coreRelos map[string]bpfCoreRelos, err error) {
+func parseExtInfos(r io.ReadSeeker, bo binary.ByteOrder, strings stringTable) (funcInfo, lineInfo map[string]extInfo, relos map[string]coreRelos, err error) {
 	var header btfExtHeader
 	var coreHeader btfExtCoreHeader
 	if err := binary.Read(r, bo, &header); err != nil {
@@ -94,13 +94,13 @@ func parseExtInfos(r io.ReadSeeker, bo binary.ByteOrder, strings stringTable) (f
 			return nil, nil, nil, fmt.Errorf("can't seek to CO-RE relocation section: %v", err)
 		}
 
-		coreRelos, err = parseExtInfoRelos(io.LimitReader(r, int64(coreHeader.CoreReloLen)), bo, strings)
+		relos, err = parseExtInfoRelos(io.LimitReader(r, int64(coreHeader.CoreReloLen)), bo, strings)
 		if err != nil {
 			return nil, nil, nil, fmt.Errorf("CO-RE relocation info: %w", err)
 		}
 	}
 
-	return funcInfo, lineInfo, coreRelos, nil
+	return funcInfo, lineInfo, relos, nil
 }
 
 type btfExtInfoSec struct {
@@ -208,18 +208,25 @@ type bpfCoreRelo struct {
 	InsnOff      uint32
 	TypeID       TypeID
 	AccessStrOff uint32
-	ReloKind     coreReloKind
+	Kind         COREKind
 }
 
-type bpfCoreRelos []bpfCoreRelo
+type coreRelo struct {
+	insnOff  uint32
+	typeID   TypeID
+	accessor coreAccessor
+	kind     COREKind
+}
+
+type coreRelos []coreRelo
 
 // append two slices of extInfoRelo to each other. The InsnOff of b are adjusted
 // by offset.
-func (r bpfCoreRelos) append(other bpfCoreRelos, offset uint64) bpfCoreRelos {
+func (r coreRelos) append(other coreRelos, offset uint64) coreRelos {
-	result := make([]bpfCoreRelo, 0, len(r)+len(other))
+	result := make([]coreRelo, 0, len(r)+len(other))
 	result = append(result, r...)
 	for _, relo := range other {
-		relo.InsnOff += uint32(offset)
+		relo.insnOff += uint32(offset)
 		result = append(result, relo)
 	}
 	return result
@@ -227,7 +234,7 @@ func (r bpfCoreRelos) append(other bpfCoreRelos, offset uint64) bpfCoreRelos {
 
 var extInfoReloSize = binary.Size(bpfCoreRelo{})
 
-func parseExtInfoRelos(r io.Reader, bo binary.ByteOrder, strings stringTable) (map[string]bpfCoreRelos, error) {
+func parseExtInfoRelos(r io.Reader, bo binary.ByteOrder, strings stringTable) (map[string]coreRelos, error) {
 	var recordSize uint32
 	if err := binary.Read(r, bo, &recordSize); err != nil {
 		return nil, fmt.Errorf("read record size: %v", err)
@@ -237,14 +244,14 @@ func parseExtInfoRelos(r io.Reader, bo binary.ByteOrder, strings stringTable) (m
 		return nil, fmt.Errorf("expected record size %d, got %d", extInfoReloSize, recordSize)
 	}
 
-	result := make(map[string]bpfCoreRelos)
+	result := make(map[string]coreRelos)
 	for {
 		secName, infoHeader, err := parseExtInfoHeader(r, bo, strings)
 		if errors.Is(err, io.EOF) {
 			return result, nil
 		}
 
-		var relos []bpfCoreRelo
+		var relos coreRelos
 		for i := uint32(0); i < infoHeader.NumInfo; i++ {
 			var relo bpfCoreRelo
 			if err := binary.Read(r, bo, &relo); err != nil {
@@ -255,7 +262,22 @@ func parseExtInfoRelos(r io.Reader, bo binary.ByteOrder, strings stringTable) (m
 				return nil, fmt.Errorf("section %v: offset %v is not aligned with instruction size", secName, relo.InsnOff)
 			}
 
-			relos = append(relos, relo)
+			accessorStr, err := strings.Lookup(relo.AccessStrOff)
+			if err != nil {
+				return nil, err
+			}
+
+			accessor, err := parseCoreAccessor(accessorStr)
+			if err != nil {
+				return nil, fmt.Errorf("accessor %q: %s", accessorStr, err)
+			}
+
+			relos = append(relos, coreRelo{
+				relo.InsnOff,
+				relo.TypeID,
+				accessor,
+				relo.Kind,
+			})
 		}
 
 		result[secName] = relos

vendor/github.com/cilium/ebpf/internal/btf/types.go

@@ -1,7 +1,6 @@
 package btf
 
 import (
-	"errors"
 	"fmt"
 	"math"
 	"strings"
@@ -37,6 +36,7 @@ type Type interface {
 type namedType interface {
 	Type
 	name() string
+	essentialName() string
 }
 
 // Name identifies a type.
@@ -48,6 +48,10 @@ func (n Name) name() string {
 	return string(n)
 }
 
+func (n Name) essentialName() string {
+	return essentialName(string(n))
+}
+
 // Void is the unit type of BTF.
 type Void struct{}
 
@@ -174,8 +178,7 @@ func (s *Struct) walk(tdq *typeDeque) {
 
 func (s *Struct) copy() Type {
 	cpy := *s
-	cpy.Members = make([]Member, len(s.Members))
+	cpy.Members = copyMembers(s.Members)
-	copy(cpy.Members, s.Members)
 	return &cpy
 }
 
@@ -206,8 +209,7 @@ func (u *Union) walk(tdq *typeDeque) {
 
 func (u *Union) copy() Type {
 	cpy := *u
-	cpy.Members = make([]Member, len(u.Members))
+	cpy.Members = copyMembers(u.Members)
-	copy(cpy.Members, u.Members)
 	return &cpy
 }
 
@@ -215,6 +217,12 @@ func (u *Union) members() []Member {
 	return u.Members
 }
 
+func copyMembers(orig []Member) []Member {
+	cpy := make([]Member, len(orig))
+	copy(cpy, orig)
+	return cpy
+}
+
 type composite interface {
 	members() []Member
 }
@@ -373,10 +381,11 @@ type Func struct {
 	TypeID
 	Name
 	Type    Type
+	Linkage FuncLinkage
 }
 
 func (f *Func) String() string {
-	return fmt.Sprintf("func#%d[%q proto=#%d]", f.TypeID, f.Name, f.Type.ID())
+	return fmt.Sprintf("func#%d[%s %q proto=#%d]", f.TypeID, f.Linkage, f.Name, f.Type.ID())
 }
 
 func (f *Func) walk(tdq *typeDeque) { tdq.push(&f.Type) }
@@ -426,11 +435,11 @@ type Var struct {
 	TypeID
 	Name
 	Type    Type
+	Linkage VarLinkage
 }
 
 func (v *Var) String() string {
-	// TODO: Linkage
+	return fmt.Sprintf("var#%d[%s %q]", v.TypeID, v.Linkage, v.Name)
-	return fmt.Sprintf("var#%d[%q]", v.TypeID, v.Name)
 }
 
 func (v *Var) walk(tdq *typeDeque) { tdq.push(&v.Type) }
@@ -511,7 +520,7 @@ func Sizeof(typ Type) (int, error) {
 		switch v := typ.(type) {
 		case *Array:
 			if n > 0 && int64(v.Nelems) > math.MaxInt64/n {
-				return 0, errors.New("overflow")
+				return 0, fmt.Errorf("type %s: overflow", typ)
 			}
 
 			// Arrays may be of zero length, which allows
@@ -532,28 +541,30 @@ func Sizeof(typ Type) (int, error) {
 			continue
 
 		default:
-			return 0, fmt.Errorf("unrecognized type %T", typ)
+			return 0, fmt.Errorf("unsized type %T", typ)
 		}
 
 		if n > 0 && elem > math.MaxInt64/n {
-			return 0, errors.New("overflow")
+			return 0, fmt.Errorf("type %s: overflow", typ)
 		}
 
 		size := n * elem
 		if int64(int(size)) != size {
-			return 0, errors.New("overflow")
+			return 0, fmt.Errorf("type %s: overflow", typ)
 		}
 
 		return int(size), nil
 	}
 
-	return 0, errors.New("exceeded type depth")
+	return 0, fmt.Errorf("type %s: exceeded type depth", typ)
 }
 
 // copy a Type recursively.
 //
 // typ may form a cycle.
-func copyType(typ Type) Type {
+//
+// Returns any errors from transform verbatim.
+func copyType(typ Type, transform func(Type) (Type, error)) (Type, error) {
 	var (
 		copies = make(map[Type]Type)
 		work   typeDeque
@@ -566,7 +577,17 @@ func copyType(typ Type) Type {
 			continue
 		}
 
-		cpy := (*t).copy()
+		var cpy Type
+		if transform != nil {
+			tf, err := transform(*t)
+			if err != nil {
+				return nil, fmt.Errorf("copy %s: %w", typ, err)
+			}
+			cpy = tf.copy()
+		} else {
+			cpy = (*t).copy()
+		}
+
 		copies[*t] = cpy
 		*t = cpy
 
@@ -574,7 +595,7 @@ func copyType(typ Type) Type {
 		cpy.walk(&work)
 	}
 
-	return typ
+	return typ, nil
 }
 
 // typeDeque keeps track of pointers to types which still
@@ -783,7 +804,7 @@ func inflateRawTypes(rawTypes []rawType, rawStrings stringTable) (types []Type,
 			typ = restrict
 
 		case kindFunc:
-			fn := &Func{id, name, nil}
+			fn := &Func{id, name, nil, raw.Linkage()}
 			fixup(raw.Type(), kindFuncProto, &fn.Type)
 			typ = fn
 
@@ -808,7 +829,8 @@ func inflateRawTypes(rawTypes []rawType, rawStrings stringTable) (types []Type,
 			typ = fp
 
 		case kindVar:
-			v := &Var{id, name, nil}
+			variable := raw.data.(*btfVariable)
+			v := &Var{id, name, nil, VarLinkage(variable.Linkage)}
 			fixup(raw.Type(), kindUnknown, &v.Type)
 			typ = v
 

vendor/github.com/cilium/ebpf/internal/elf.go

@@ -50,3 +50,19 @@ func (se *SafeELFFile) Symbols() (syms []elf.Symbol, err error) {
 	syms, err = se.File.Symbols()
 	return
 }
+
+// DynamicSymbols is the safe version of elf.File.DynamicSymbols.
+func (se *SafeELFFile) DynamicSymbols() (syms []elf.Symbol, err error) {
+	defer func() {
+		r := recover()
+		if r == nil {
+			return
+		}
+
+		syms = nil
+		err = fmt.Errorf("reading ELF dynamic symbols panicked: %s", r)
+	}()
+
+	syms, err = se.File.DynamicSymbols()
+	return
+}

vendor/github.com/cilium/ebpf/internal/endian.go

@@ -9,11 +9,16 @@ import (
 // depending on the host's endianness.
 var NativeEndian binary.ByteOrder
 
+// Clang is set to either "el" or "eb" depending on the host's endianness.
+var ClangEndian string
+
 func init() {
 	if isBigEndian() {
 		NativeEndian = binary.BigEndian
+		ClangEndian = "eb"
 	} else {
 		NativeEndian = binary.LittleEndian
+		ClangEndian = "el"
 	}
 }
 

vendor/github.com/cilium/ebpf/internal/errors.go

@@ -29,6 +29,10 @@ type VerifierError struct {
 	log   string
 }
 
+func (le *VerifierError) Unwrap() error {
+	return le.cause
+}
+
 func (le *VerifierError) Error() string {
 	if le.log == "" {
 		return le.cause.Error()

vendor/github.com/cilium/ebpf/internal/ptr.go

@@ -22,10 +22,6 @@ func NewSlicePointer(buf []byte) Pointer {
 
 // NewStringPointer creates a 64-bit pointer from a string.
 func NewStringPointer(str string) Pointer {
-	if str == "" {
-		return Pointer{}
-	}
-
 	p, err := unix.BytePtrFromString(str)
 	if err != nil {
 		return Pointer{}

vendor/github.com/cilium/ebpf/internal/syscall.go

@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"path/filepath"
 	"runtime"
+	"syscall"
 	"unsafe"
 
 	"github.com/cilium/ebpf/internal/unix"
@@ -61,7 +62,7 @@ func BPF(cmd BPFCmd, attr unsafe.Pointer, size uintptr) (uintptr, error) {
 
 	var err error
 	if errNo != 0 {
-		err = errNo
+		err = wrappedErrno{errNo}
 	}
 
 	return r1, err
@@ -178,3 +179,67 @@ func BPFObjGetInfoByFD(fd *FD, info unsafe.Pointer, size uintptr) error {
 	}
 	return nil
 }
+
+// BPFObjName is a null-terminated string made up of
+// 'A-Za-z0-9_' characters.
+type BPFObjName [unix.BPF_OBJ_NAME_LEN]byte
+
+// NewBPFObjName truncates the result if it is too long.
+func NewBPFObjName(name string) BPFObjName {
+	var result BPFObjName
+	copy(result[:unix.BPF_OBJ_NAME_LEN-1], name)
+	return result
+}
+
+type BPFMapCreateAttr struct {
+	MapType        uint32
+	KeySize        uint32
+	ValueSize      uint32
+	MaxEntries     uint32
+	Flags          uint32
+	InnerMapFd     uint32     // since 4.12 56f668dfe00d
+	NumaNode       uint32     // since 4.14 96eabe7a40aa
+	MapName        BPFObjName // since 4.15 ad5b177bd73f
+	MapIfIndex     uint32
+	BTFFd          uint32
+	BTFKeyTypeID   uint32
+	BTFValueTypeID uint32
+}
+
+func BPFMapCreate(attr *BPFMapCreateAttr) (*FD, error) {
+	fd, err := BPF(BPF_MAP_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
+	if err != nil {
+		return nil, err
+	}
+
+	return NewFD(uint32(fd)), nil
+}
+
+// wrappedErrno wraps syscall.Errno to prevent direct comparisons with
+// syscall.E* or unix.E* constants.
+//
+// You should never export an error of this type.
+type wrappedErrno struct {
+	syscall.Errno
+}
+
+func (we wrappedErrno) Unwrap() error {
+	return we.Errno
+}
+
+type syscallError struct {
+	error
+	errno syscall.Errno
+}
+
+func SyscallError(err error, errno syscall.Errno) error {
+	return &syscallError{err, errno}
+}
+
+func (se *syscallError) Is(target error) bool {
+	return target == se.error
+}
+
+func (se *syscallError) Unwrap() error {
+	return se.errno
+}

vendor/github.com/cilium/ebpf/internal/unix/types_linux.go

@@ -31,6 +31,8 @@ const (
 	BPF_F_RDONLY_PROG        = linux.BPF_F_RDONLY_PROG
 	BPF_F_WRONLY_PROG        = linux.BPF_F_WRONLY_PROG
 	BPF_F_SLEEPABLE          = linux.BPF_F_SLEEPABLE
+	BPF_F_MMAPABLE           = linux.BPF_F_MMAPABLE
+	BPF_F_INNER_MAP          = linux.BPF_F_INNER_MAP
 	BPF_OBJ_NAME_LEN         = linux.BPF_OBJ_NAME_LEN
 	BPF_TAG_SIZE             = linux.BPF_TAG_SIZE
 	SYS_BPF                  = linux.SYS_BPF
@@ -42,6 +44,7 @@ const (
 	PROT_READ                = linux.PROT_READ
 	PROT_WRITE               = linux.PROT_WRITE
 	MAP_SHARED               = linux.MAP_SHARED
+	PERF_ATTR_SIZE_VER1      = linux.PERF_ATTR_SIZE_VER1
 	PERF_TYPE_SOFTWARE       = linux.PERF_TYPE_SOFTWARE
 	PERF_TYPE_TRACEPOINT     = linux.PERF_TYPE_TRACEPOINT
 	PERF_COUNT_SW_BPF_OUTPUT = linux.PERF_COUNT_SW_BPF_OUTPUT

vendor/github.com/cilium/ebpf/internal/unix/types_other.go

@@ -31,6 +31,8 @@ const (
 	BPF_F_RDONLY_PROG        = 0
 	BPF_F_WRONLY_PROG        = 0
 	BPF_F_SLEEPABLE          = 0
+	BPF_F_MMAPABLE           = 0
+	BPF_F_INNER_MAP          = 0
 	BPF_OBJ_NAME_LEN         = 0x10
 	BPF_TAG_SIZE             = 0x8
 	SYS_BPF                  = 321
@@ -43,6 +45,7 @@ const (
 	PROT_READ                = 0x1
 	PROT_WRITE               = 0x2
 	MAP_SHARED               = 0x1
+	PERF_ATTR_SIZE_VER1      = 0
 	PERF_TYPE_SOFTWARE       = 0x1
 	PERF_TYPE_TRACEPOINT     = 0
 	PERF_COUNT_SW_BPF_OUTPUT = 0xa

vendor/github.com/cilium/ebpf/link/iter.go

@@ -3,8 +3,10 @@ package link
 import (
 	"fmt"
 	"io"
+	"unsafe"
 
 	"github.com/cilium/ebpf"
+	"github.com/cilium/ebpf/internal"
 )
 
 type IterOptions struct {
@@ -15,19 +17,45 @@ type IterOptions struct {
 	// AttachTo requires the kernel to include BTF of itself,
 	// and it to be compiled with a recent pahole (>= 1.16).
 	Program *ebpf.Program
+
+	// Map specifies the target map for bpf_map_elem and sockmap iterators.
+	// It may be nil.
+	Map *ebpf.Map
 }
 
 // AttachIter attaches a BPF seq_file iterator.
 func AttachIter(opts IterOptions) (*Iter, error) {
-	link, err := AttachRawLink(RawLinkOptions{
+	if err := haveBPFLink(); err != nil {
-		Program: opts.Program,
+		return nil, err
-		Attach:  ebpf.AttachTraceIter,
+	}
-	})
+
+	progFd := opts.Program.FD()
+	if progFd < 0 {
+		return nil, fmt.Errorf("invalid program: %s", internal.ErrClosedFd)
+	}
+
+	var info bpfIterLinkInfoMap
+	if opts.Map != nil {
+		mapFd := opts.Map.FD()
+		if mapFd < 0 {
+			return nil, fmt.Errorf("invalid map: %w", internal.ErrClosedFd)
+		}
+		info.map_fd = uint32(mapFd)
+	}
+
+	attr := bpfLinkCreateIterAttr{
+		prog_fd:       uint32(progFd),
+		attach_type:   ebpf.AttachTraceIter,
+		iter_info:     internal.NewPointer(unsafe.Pointer(&info)),
+		iter_info_len: uint32(unsafe.Sizeof(info)),
+	}
+
+	fd, err := bpfLinkCreateIter(&attr)
 	if err != nil {
 		return nil, fmt.Errorf("can't link iterator: %w", err)
 	}
 
-	return &Iter{*link}, err
+	return &Iter{RawLink{fd, ""}}, err
 }
 
 // LoadPinnedIter loads a pinned iterator from a bpffs.
@@ -65,3 +93,8 @@ func (it *Iter) Open() (io.ReadCloser, error) {
 
 	return fd.File("bpf_iter"), nil
 }
+
+// union bpf_iter_link_info.map
+type bpfIterLinkInfoMap struct {
+	map_fd uint32
+}

vendor/github.com/cilium/ebpf/link/kprobe.go

@@ -1,12 +1,16 @@
 package link
 
 import (
+	"bytes"
 	"crypto/rand"
 	"errors"
 	"fmt"
+	"io/ioutil"
 	"os"
 	"path/filepath"
 	"runtime"
+	"sync"
+	"unsafe"
 
 	"github.com/cilium/ebpf"
 	"github.com/cilium/ebpf/internal"
@@ -15,13 +19,60 @@ import (
 
 var (
 	kprobeEventsPath = filepath.Join(tracefsPath, "kprobe_events")
+
+	kprobeRetprobeBit = struct {
+		once  sync.Once
+		value uint64
+		err   error
+	}{}
 )
 
+type probeType uint8
+
+const (
+	kprobeType probeType = iota
+	uprobeType
+)
+
+func (pt probeType) String() string {
+	if pt == kprobeType {
+		return "kprobe"
+	}
+	return "uprobe"
+}
+
+func (pt probeType) EventsPath() string {
+	if pt == kprobeType {
+		return kprobeEventsPath
+	}
+	return uprobeEventsPath
+}
+
+func (pt probeType) PerfEventType(ret bool) perfEventType {
+	if pt == kprobeType {
+		if ret {
+			return kretprobeEvent
+		}
+		return kprobeEvent
+	}
+	if ret {
+		return uretprobeEvent
+	}
+	return uprobeEvent
+}
+
+func (pt probeType) RetprobeBit() (uint64, error) {
+	if pt == kprobeType {
+		return kretprobeBit()
+	}
+	return uretprobeBit()
+}
+
 // Kprobe attaches the given eBPF program to a perf event that fires when the
 // given kernel symbol starts executing. See /proc/kallsyms for available
 // symbols. For example, printk():
 //
-//	Kprobe("printk")
+//	Kprobe("printk", prog)
 //
 // The resulting Link must be Closed during program shutdown to avoid leaking
 // system resources.
@@ -44,7 +95,7 @@ func Kprobe(symbol string, prog *ebpf.Program) (Link, error) {
 // before the given kernel symbol exits, with the function stack left intact.
 // See /proc/kallsyms for available symbols. For example, printk():
 //
-//	Kretprobe("printk")
+//	Kretprobe("printk", prog)
 //
 // The resulting Link must be Closed during program shutdown to avoid leaking
 // system resources.
@@ -80,7 +131,10 @@ func kprobe(symbol string, prog *ebpf.Program, ret bool) (*perfEvent, error) {
 	}
 
 	// Use kprobe PMU if the kernel has it available.
-	tp, err := pmuKprobe(symbol, ret)
+	tp, err := pmuKprobe(platformPrefix(symbol), ret)
+	if errors.Is(err, os.ErrNotExist) {
+		tp, err = pmuKprobe(symbol, ret)
+	}
 	if err == nil {
 		return tp, nil
 	}
@@ -89,7 +143,10 @@ func kprobe(symbol string, prog *ebpf.Program, ret bool) (*perfEvent, error) {
 	}
 
 	// Use tracefs if kprobe PMU is missing.
+	tp, err = tracefsKprobe(platformPrefix(symbol), ret)
+	if errors.Is(err, os.ErrNotExist) {
 		tp, err = tracefsKprobe(symbol, ret)
+	}
 	if err != nil {
 		return nil, fmt.Errorf("creating trace event '%s' in tracefs: %w", symbol, err)
 	}
@@ -97,36 +154,70 @@ func kprobe(symbol string, prog *ebpf.Program, ret bool) (*perfEvent, error) {
 	return tp, nil
 }
 
-// pmuKprobe opens a perf event based on a Performance Monitoring Unit.
+// pmuKprobe opens a perf event based on the kprobe PMU.
-// Requires at least 4.17 (e12f03d7031a "perf/core: Implement the
+// Returns os.ErrNotExist if the given symbol does not exist in the kernel.
-// 'perf_kprobe' PMU").
-// Returns ErrNotSupported if the kernel doesn't support perf_kprobe PMU,
-// or os.ErrNotExist if the given symbol does not exist in the kernel.
 func pmuKprobe(symbol string, ret bool) (*perfEvent, error) {
+	return pmuProbe(kprobeType, symbol, "", 0, ret)
+}
 
+// pmuProbe opens a perf event based on a Performance Monitoring Unit.
+//
+// Requires at least a 4.17 kernel.
+// e12f03d7031a "perf/core: Implement the 'perf_kprobe' PMU"
+// 33ea4b24277b "perf/core: Implement the 'perf_uprobe' PMU"
+//
+// Returns ErrNotSupported if the kernel doesn't support perf_[k,u]probe PMU
+func pmuProbe(typ probeType, symbol, path string, offset uint64, ret bool) (*perfEvent, error) {
 	// Getting the PMU type will fail if the kernel doesn't support
-	// the perf_kprobe PMU.
+	// the perf_[k,u]probe PMU.
-	et, err := getPMUEventType("kprobe")
+	et, err := getPMUEventType(typ)
 	if err != nil {
 		return nil, err
 	}
 
+	var config uint64
+	if ret {
+		bit, err := typ.RetprobeBit()
+		if err != nil {
+			return nil, err
+		}
+		config |= 1 << bit
+	}
+
+	var (
+		attr unix.PerfEventAttr
+		sp   unsafe.Pointer
+	)
+	switch typ {
+	case kprobeType:
 		// Create a pointer to a NUL-terminated string for the kernel.
 		sp, err := unsafeStringPtr(symbol)
 		if err != nil {
 			return nil, err
 		}
 
-	// TODO: Parse the position of the bit from /sys/bus/event_source/devices/%s/format/retprobe.
+		attr = unix.PerfEventAttr{
-	config := 0
-	if ret {
-		config = 1
-	}
-
-	attr := unix.PerfEventAttr{
 			Type:   uint32(et),          // PMU event type read from sysfs
 			Ext1:   uint64(uintptr(sp)), // Kernel symbol to trace
-		Config: uint64(config),      // perf_kprobe PMU treats config as flags
+			Config: config,              // Retprobe flag
+		}
+	case uprobeType:
+		sp, err := unsafeStringPtr(path)
+		if err != nil {
+			return nil, err
+		}
+
+		attr = unix.PerfEventAttr{
+			// The minimum size required for PMU uprobes is PERF_ATTR_SIZE_VER1,
+			// since it added the config2 (Ext2) field. The Size field controls the
+			// size of the internal buffer the kernel allocates for reading the
+			// perf_event_attr argument from userspace.
+			Size:   unix.PERF_ATTR_SIZE_VER1,
+			Type:   uint32(et),          // PMU event type read from sysfs
+			Ext1:   uint64(uintptr(sp)), // Uprobe path
+			Ext2:   offset,              // Uprobe offset
+			Config: config,              // Retprobe flag
+		}
 	}
 
 	fd, err := unix.PerfEventOpen(&attr, perfAllThreads, 0, -1, unix.PERF_FLAG_FD_CLOEXEC)
@@ -144,22 +235,27 @@ func pmuKprobe(symbol string, ret bool) (*perfEvent, error) {
 	// Ensure the string pointer is not collected before PerfEventOpen returns.
 	runtime.KeepAlive(sp)
 
-	// Kernel has perf_kprobe PMU available, initialize perf event.
+	// Kernel has perf_[k,u]probe PMU available, initialize perf event.
 	return &perfEvent{
 		fd:    internal.NewFD(uint32(fd)),
 		pmuID: et,
 		name:  symbol,
-		ret:      ret,
+		typ:   typ.PerfEventType(ret),
-		progType: ebpf.Kprobe,
 	}, nil
 }
 
-// tracefsKprobe creates a trace event by writing an entry to <tracefs>/kprobe_events.
+// tracefsKprobe creates a Kprobe tracefs entry.
-// A new trace event group name is generated on every call to support creating
-// multiple trace events for the same kernel symbol. A perf event is then opened
-// on the newly-created trace event and returned to the caller.
 func tracefsKprobe(symbol string, ret bool) (*perfEvent, error) {
+	return tracefsProbe(kprobeType, symbol, "", 0, ret)
+}
 
+// tracefsProbe creates a trace event by writing an entry to <tracefs>/[k,u]probe_events.
+// A new trace event group name is generated on every call to support creating
+// multiple trace events for the same kernel or userspace symbol.
+// Path and offset are only set in the case of uprobe(s) and are used to set
+// the executable/library path on the filesystem and the offset where the probe is inserted.
+// A perf event is then opened on the newly-created trace event and returned to the caller.
+func tracefsProbe(typ probeType, symbol, path string, offset uint64, ret bool) (*perfEvent, error) {
 	// Generate a random string for each trace event we attempt to create.
 	// This value is used as the 'group' token in tracefs to allow creating
 	// multiple kprobe trace events with the same name.
@@ -176,14 +272,13 @@ func tracefsKprobe(symbol string, ret bool) (*perfEvent, error) {
 	if err == nil {
 		return nil, fmt.Errorf("trace event already exists: %s/%s", group, symbol)
 	}
-	// The read is expected to fail with ErrNotSupported due to a non-existing event.
+	if err != nil && !errors.Is(err, os.ErrNotExist) {
-	if err != nil && !errors.Is(err, ErrNotSupported) {
 		return nil, fmt.Errorf("checking trace event %s/%s: %w", group, symbol, err)
 	}
 
-	// Create the kprobe trace event using tracefs.
+	// Create the [k,u]probe trace event using tracefs.
-	if err := createTraceFSKprobeEvent(group, symbol, ret); err != nil {
+	if err := createTraceFSProbeEvent(typ, group, symbol, path, offset, ret); err != nil {
-		return nil, fmt.Errorf("creating kprobe event on tracefs: %w", err)
+		return nil, fmt.Errorf("creating probe entry on tracefs: %w", err)
 	}
 
 	// Get the newly-created trace event's id.
@@ -202,23 +297,26 @@ func tracefsKprobe(symbol string, ret bool) (*perfEvent, error) {
 		fd:        fd,
 		group:     group,
 		name:      symbol,
-		ret:       ret,
 		tracefsID: tid,
-		progType:  ebpf.Kprobe, // kernel only allows attaching kprobe programs to kprobe events
+		typ:       typ.PerfEventType(ret),
 	}, nil
 }
 
-// createTraceFSKprobeEvent creates a new ephemeral trace event by writing to
+// createTraceFSProbeEvent creates a new ephemeral trace event by writing to
-// <tracefs>/kprobe_events. Returns ErrNotSupported if symbol is not a valid
+// <tracefs>/[k,u]probe_events. Returns os.ErrNotExist if symbol is not a valid
-// kernel symbol, or if it is not traceable with kprobes.
+// kernel symbol, or if it is not traceable with kprobes. Returns os.ErrExist
-func createTraceFSKprobeEvent(group, symbol string, ret bool) error {
+// if a probe with the same group and symbol already exists.
+func createTraceFSProbeEvent(typ probeType, group, symbol, path string, offset uint64, ret bool) error {
 	// Open the kprobe_events file in tracefs.
-	f, err := os.OpenFile(kprobeEventsPath, os.O_APPEND|os.O_WRONLY, 0666)
+	f, err := os.OpenFile(typ.EventsPath(), os.O_APPEND|os.O_WRONLY, 0666)
 	if err != nil {
-		return fmt.Errorf("error opening kprobe_events: %w", err)
+		return fmt.Errorf("error opening '%s': %w", typ.EventsPath(), err)
 	}
 	defer f.Close()
 
+	var pe string
+	switch typ {
+	case kprobeType:
 		// The kprobe_events syntax is as follows (see Documentation/trace/kprobetrace.txt):
 		// p[:[GRP/]EVENT] [MOD:]SYM[+offs]|MEMADDR [FETCHARGS] : Set a probe
 		// r[MAXACTIVE][:[GRP/]EVENT] [MOD:]SYM[+0] [FETCHARGS] : Set a return probe
@@ -231,36 +329,51 @@ func createTraceFSKprobeEvent(group, symbol string, ret bool) error {
 		// Leaving the kretprobe's MAXACTIVE set to 0 (or absent) will make the
 		// kernel default to NR_CPUS. This is desired in most eBPF cases since
 		// subsampling or rate limiting logic can be more accurately implemented in
-	// the eBPF program itself. See Documentation/kprobes.txt for more details.
+		// the eBPF program itself.
-	pe := fmt.Sprintf("%s:%s/%s %s", kprobePrefix(ret), group, symbol, symbol)
+		// See Documentation/kprobes.txt for more details.
+		pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(ret), group, symbol, symbol)
+	case uprobeType:
+		// The uprobe_events syntax is as follows:
+		// p[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a probe
+		// r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a return probe
+		// -:[GRP/]EVENT                           : Clear a probe
+		//
+		// Some examples:
+		// r:ebpf_1234/readline /bin/bash:0x12345
+		// p:ebpf_5678/main_mySymbol /bin/mybin:0x12345
+		//
+		// See Documentation/trace/uprobetracer.txt for more details.
+		pathOffset := uprobePathOffset(path, offset)
+		pe = fmt.Sprintf("%s:%s/%s %s", probePrefix(ret), group, symbol, pathOffset)
+	}
 	_, err = f.WriteString(pe)
 	// Since commit 97c753e62e6c, ENOENT is correctly returned instead of EINVAL
 	// when trying to create a kretprobe for a missing symbol. Make sure ENOENT
 	// is returned to the caller.
 	if errors.Is(err, os.ErrNotExist) || errors.Is(err, unix.EINVAL) {
-		return fmt.Errorf("kernel symbol %s not found: %w", symbol, os.ErrNotExist)
+		return fmt.Errorf("symbol %s not found: %w", symbol, os.ErrNotExist)
 	}
 	if err != nil {
-		return fmt.Errorf("writing '%s' to kprobe_events: %w", pe, err)
+		return fmt.Errorf("writing '%s' to '%s': %w", pe, typ.EventsPath(), err)
 	}
 
 	return nil
 }
 
-// closeTraceFSKprobeEvent removes the kprobe with the given group, symbol and kind
+// closeTraceFSProbeEvent removes the [k,u]probe with the given type, group and symbol
-// from <tracefs>/kprobe_events.
+// from <tracefs>/[k,u]probe_events.
-func closeTraceFSKprobeEvent(group, symbol string) error {
+func closeTraceFSProbeEvent(typ probeType, group, symbol string) error {
-	f, err := os.OpenFile(kprobeEventsPath, os.O_APPEND|os.O_WRONLY, 0666)
+	f, err := os.OpenFile(typ.EventsPath(), os.O_APPEND|os.O_WRONLY, 0666)
 	if err != nil {
-		return fmt.Errorf("error opening kprobe_events: %w", err)
+		return fmt.Errorf("error opening %s: %w", typ.EventsPath(), err)
 	}
 	defer f.Close()
 
-	// See kprobe_events syntax above. Kprobe type does not need to be specified
+	// See [k,u]probe_events syntax above. The probe type does not need to be specified
 	// for removals.
 	pe := fmt.Sprintf("-:%s/%s", group, symbol)
 	if _, err = f.WriteString(pe); err != nil {
-		return fmt.Errorf("writing '%s' to kprobe_events: %w", pe, err)
+		return fmt.Errorf("writing '%s' to '%s': %w", pe, typ.EventsPath(), err)
 	}
 
 	return nil
@@ -288,9 +401,38 @@ func randomGroup(prefix string) (string, error) {
 	return group, nil
 }
 
-func kprobePrefix(ret bool) string {
+func probePrefix(ret bool) string {
 	if ret {
 		return "r"
 	}
 	return "p"
 }
+
+// determineRetprobeBit reads a Performance Monitoring Unit's retprobe bit
+// from /sys/bus/event_source/devices/<pmu>/format/retprobe.
+func determineRetprobeBit(typ probeType) (uint64, error) {
+	p := filepath.Join("/sys/bus/event_source/devices/", typ.String(), "/format/retprobe")
+
+	data, err := ioutil.ReadFile(p)
+	if err != nil {
+		return 0, err
+	}
+
+	var rp uint64
+	n, err := fmt.Sscanf(string(bytes.TrimSpace(data)), "config:%d", &rp)
+	if err != nil {
+		return 0, fmt.Errorf("parse retprobe bit: %w", err)
+	}
+	if n != 1 {
+		return 0, fmt.Errorf("parse retprobe bit: expected 1 item, got %d", n)
+	}
+
+	return rp, nil
+}
+
+func kretprobeBit() (uint64, error) {
+	kprobeRetprobeBit.once.Do(func() {
+		kprobeRetprobeBit.value, kprobeRetprobeBit.err = determineRetprobeBit(kprobeType)
+	})
+	return kprobeRetprobeBit.value, kprobeRetprobeBit.err
+}

vendor/github.com/cilium/ebpf/link/perf_event.go

@@ -31,6 +31,10 @@ import (
 //   exported kernel symbols. kprobe-based (tracefs) trace events can be
 //   created system-wide by writing to the <tracefs>/kprobe_events file, or
 //   they can be scoped to the current process by creating PMU perf events.
+// - u(ret)probe: Ephemeral trace events based on user provides ELF binaries
+//   and offsets. uprobe-based (tracefs) trace events can be
+//   created system-wide by writing to the <tracefs>/uprobe_events file, or
+//   they can be scoped to the current process by creating PMU perf events.
 // - perf event: An object instantiated based on an existing trace event or
 //   kernel symbol. Referred to by fd in userspace.
 //   Exactly one eBPF program can be attached to a perf event. Multiple perf
@@ -52,6 +56,16 @@ const (
 	perfAllThreads = -1
 )
 
+type perfEventType uint8
+
+const (
+	tracepointEvent perfEventType = iota
+	kprobeEvent
+	kretprobeEvent
+	uprobeEvent
+	uretprobeEvent
+)
+
 // A perfEvent represents a perf event kernel object. Exactly one eBPF program
 // can be attached to it. It is created based on a tracefs trace event or a
 // Performance Monitoring Unit (PMU).
@@ -66,11 +80,10 @@ type perfEvent struct {
 	// ID of the trace event read from tracefs. Valid IDs are non-zero.
 	tracefsID uint64
 
-	// True for kretprobes/uretprobes.
+	// The event type determines the types of programs that can be attached.
-	ret bool
+	typ perfEventType
 
 	fd *internal.FD
-	progType ebpf.ProgramType
 }
 
 func (pe *perfEvent) isLink() {}
@@ -117,13 +130,18 @@ func (pe *perfEvent) Close() error {
 		return fmt.Errorf("closing perf event fd: %w", err)
 	}
 
-	switch t := pe.progType; t {
+	switch pe.typ {
-	case ebpf.Kprobe:
+	case kprobeEvent, kretprobeEvent:
-		// For kprobes created using tracefs, clean up the <tracefs>/kprobe_events entry.
+		// Clean up kprobe tracefs entry.
 		if pe.tracefsID != 0 {
-			return closeTraceFSKprobeEvent(pe.group, pe.name)
+			return closeTraceFSProbeEvent(kprobeType, pe.group, pe.name)
 		}
-	case ebpf.TracePoint:
+	case uprobeEvent, uretprobeEvent:
+		// Clean up uprobe tracefs entry.
+		if pe.tracefsID != 0 {
+			return closeTraceFSProbeEvent(uprobeType, pe.group, pe.name)
+		}
+	case tracepointEvent:
 		// Tracepoint trace events don't hold any extra resources.
 		return nil
 	}
@@ -141,12 +159,21 @@ func (pe *perfEvent) attach(prog *ebpf.Program) error {
 	if pe.fd == nil {
 		return errors.New("cannot attach to nil perf event")
 	}
-	if t := prog.Type(); t != pe.progType {
-		return fmt.Errorf("invalid program type (expected %s): %s", pe.progType, t)
-	}
 	if prog.FD() < 0 {
 		return fmt.Errorf("invalid program: %w", internal.ErrClosedFd)
 	}
+	switch pe.typ {
+	case kprobeEvent, kretprobeEvent, uprobeEvent, uretprobeEvent:
+		if t := prog.Type(); t != ebpf.Kprobe {
+			return fmt.Errorf("invalid program type (expected %s): %s", ebpf.Kprobe, t)
+		}
+	case tracepointEvent:
+		if t := prog.Type(); t != ebpf.TracePoint {
+			return fmt.Errorf("invalid program type (expected %s): %s", ebpf.TracePoint, t)
+		}
+	default:
+		return fmt.Errorf("unknown perf event type: %d", pe.typ)
+	}
 
 	// The ioctl below will fail when the fd is invalid.
 	kfd, _ := pe.fd.Value()
@@ -180,8 +207,8 @@ func unsafeStringPtr(str string) (unsafe.Pointer, error) {
 // group and name must be alphanumeric or underscore, as required by the kernel.
 func getTraceEventID(group, name string) (uint64, error) {
 	tid, err := uint64FromFile(tracefsPath, "events", group, name, "id")
-	if errors.Is(err, ErrNotSupported) {
+	if errors.Is(err, os.ErrNotExist) {
-		return 0, fmt.Errorf("trace event %s/%s: %w", group, name, ErrNotSupported)
+		return 0, fmt.Errorf("trace event %s/%s: %w", group, name, os.ErrNotExist)
 	}
 	if err != nil {
 		return 0, fmt.Errorf("reading trace event ID of %s/%s: %w", group, name, err)
@@ -192,20 +219,22 @@ func getTraceEventID(group, name string) (uint64, error) {
 
 // getPMUEventType reads a Performance Monitoring Unit's type (numeric identifier)
 // from /sys/bus/event_source/devices/<pmu>/type.
-func getPMUEventType(pmu string) (uint64, error) {
+//
-	et, err := uint64FromFile("/sys/bus/event_source/devices", pmu, "type")
+// Returns ErrNotSupported if the pmu type is not supported.
-	if errors.Is(err, ErrNotSupported) {
+func getPMUEventType(typ probeType) (uint64, error) {
-		return 0, fmt.Errorf("pmu type %s: %w", pmu, ErrNotSupported)
+	et, err := uint64FromFile("/sys/bus/event_source/devices", typ.String(), "type")
+	if errors.Is(err, os.ErrNotExist) {
+		return 0, fmt.Errorf("pmu type %s: %w", typ, ErrNotSupported)
 	}
 	if err != nil {
-		return 0, fmt.Errorf("reading pmu type %s: %w", pmu, err)
+		return 0, fmt.Errorf("reading pmu type %s: %w", typ, err)
 	}
 
 	return et, nil
 }
 
 // openTracepointPerfEvent opens a tracepoint-type perf event. System-wide
-// kprobes created by writing to <tracefs>/kprobe_events are tracepoints
+// [k,u]probes created by writing to <tracefs>/[k,u]probe_events are tracepoints
 // behind the scenes, and can be attached to using these perf events.
 func openTracepointPerfEvent(tid uint64) (*internal.FD, error) {
 	attr := unix.PerfEventAttr{
@@ -228,22 +257,13 @@ func openTracepointPerfEvent(tid uint64) (*internal.FD, error) {
 // and joined onto base. Returns error if base no longer prefixes the path after
 // joining all components.
 func uint64FromFile(base string, path ...string) (uint64, error) {
-
-	// Resolve leaf path separately for error feedback. Makes the join onto
-	// base more readable (can't mix with variadic args).
 	l := filepath.Join(path...)
-
 	p := filepath.Join(base, l)
 	if !strings.HasPrefix(p, base) {
 		return 0, fmt.Errorf("path '%s' attempts to escape base path '%s': %w", l, base, errInvalidInput)
 	}
 
 	data, err := ioutil.ReadFile(p)
-	if os.IsNotExist(err) {
-		// Only echo leaf path, the base path can be prepended at the call site
-		// if more verbosity is required.
-		return 0, fmt.Errorf("symbol %s: %w", l, ErrNotSupported)
-	}
 	if err != nil {
 		return 0, fmt.Errorf("reading file %s: %w", p, err)
 	}

vendor/github.com/cilium/ebpf/link/platform.go

@@ -0,0 +1,25 @@
+package link
+
+import (
+	"fmt"
+	"runtime"
+)
+
+func platformPrefix(symbol string) string {
+
+	prefix := runtime.GOARCH
+
+	// per https://github.com/golang/go/blob/master/src/go/build/syslist.go
+	switch prefix {
+	case "386":
+		prefix = "ia32"
+	case "amd64", "amd64p32":
+		prefix = "x64"
+	case "arm64", "arm64be":
+		prefix = "arm64"
+	default:
+		return symbol
+	}
+
+	return fmt.Sprintf("__%s_%s", prefix, symbol)
+}

vendor/github.com/cilium/ebpf/link/program.go

@@ -43,7 +43,7 @@ func RawAttachProgram(opts RawAttachProgramOptions) error {
 	}
 
 	if err := internal.BPFProgAttach(&attr); err != nil {
-		return fmt.Errorf("can't attach program: %s", err)
+		return fmt.Errorf("can't attach program: %w", err)
 	}
 	return nil
 }
@@ -69,7 +69,7 @@ func RawDetachProgram(opts RawDetachProgramOptions) error {
 		AttachType:  uint32(opts.Attach),
 	}
 	if err := internal.BPFProgDetach(&attr); err != nil {
-		return fmt.Errorf("can't detach program: %s", err)
+		return fmt.Errorf("can't detach program: %w", err)
 	}
 
 	return nil

vendor/github.com/cilium/ebpf/link/syscalls.go

@@ -102,6 +102,23 @@ func bpfLinkCreate(attr *bpfLinkCreateAttr) (*internal.FD, error) {
 	return internal.NewFD(uint32(ptr)), nil
 }
 
+type bpfLinkCreateIterAttr struct {
+	prog_fd       uint32
+	target_fd     uint32
+	attach_type   ebpf.AttachType
+	flags         uint32
+	iter_info     internal.Pointer
+	iter_info_len uint32
+}
+
+func bpfLinkCreateIter(attr *bpfLinkCreateIterAttr) (*internal.FD, error) {
+	ptr, err := internal.BPF(internal.BPF_LINK_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
+	if err != nil {
+		return nil, err
+	}
+	return internal.NewFD(uint32(ptr)), nil
+}
+
 type bpfLinkUpdateAttr struct {
 	linkFd    uint32
 	newProgFd uint32

vendor/github.com/cilium/ebpf/link/tracepoint.go

@@ -11,7 +11,7 @@ import (
 // tracepoints. The top-level directory is the group, the event's subdirectory
 // is the name. Example:
 //
-//	Tracepoint("syscalls", "sys_enter_fork")
+//	Tracepoint("syscalls", "sys_enter_fork", prog)
 //
 // Note that attaching eBPF programs to syscalls (sys_enter_*/sys_exit_*) is
 // only possible as of kernel 4.14 (commit cf5f5ce).
@@ -44,7 +44,7 @@ func Tracepoint(group, name string, prog *ebpf.Program) (Link, error) {
 		tracefsID: tid,
 		group:     group,
 		name:      name,
-		progType:  ebpf.TracePoint,
+		typ:       tracepointEvent,
 	}
 
 	if err := pe.attach(prog); err != nil {

vendor/github.com/cilium/ebpf/link/uprobe.go

@@ -0,0 +1,237 @@
+package link
+
+import (
+	"debug/elf"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"regexp"
+	"sync"
+
+	"github.com/cilium/ebpf"
+	"github.com/cilium/ebpf/internal"
+)
+
+var (
+	uprobeEventsPath = filepath.Join(tracefsPath, "uprobe_events")
+
+	// rgxUprobeSymbol is used to strip invalid characters from the uprobe symbol
+	// as they are not allowed to be used as the EVENT token in tracefs.
+	rgxUprobeSymbol = regexp.MustCompile("[^a-zA-Z0-9]+")
+
+	uprobeRetprobeBit = struct {
+		once  sync.Once
+		value uint64
+		err   error
+	}{}
+)
+
+// Executable defines an executable program on the filesystem.
+type Executable struct {
+	// Path of the executable on the filesystem.
+	path string
+	// Parsed ELF symbols and dynamic symbols.
+	symbols map[string]elf.Symbol
+}
+
+// UprobeOptions defines additional parameters that will be used
+// when loading Uprobes.
+type UprobeOptions struct {
+	// Symbol offset. Must be provided in case of external symbols (shared libs).
+	// If set, overrides the offset eventually parsed from the executable.
+	Offset uint64
+}
+
+// To open a new Executable, use:
+//
+//	OpenExecutable("/bin/bash")
+//
+// The returned value can then be used to open Uprobe(s).
+func OpenExecutable(path string) (*Executable, error) {
+	if path == "" {
+		return nil, fmt.Errorf("path cannot be empty")
+	}
+
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, fmt.Errorf("open file '%s': %w", path, err)
+	}
+	defer f.Close()
+
+	se, err := internal.NewSafeELFFile(f)
+	if err != nil {
+		return nil, fmt.Errorf("parse ELF file: %w", err)
+	}
+
+	var ex = Executable{
+		path:    path,
+		symbols: make(map[string]elf.Symbol),
+	}
+	if err := ex.addSymbols(se.Symbols); err != nil {
+		return nil, err
+	}
+
+	if err := ex.addSymbols(se.DynamicSymbols); err != nil {
+		return nil, err
+	}
+
+	return &ex, nil
+}
+
+func (ex *Executable) addSymbols(f func() ([]elf.Symbol, error)) error {
+	// elf.Symbols and elf.DynamicSymbols return ErrNoSymbols if the section is not found.
+	syms, err := f()
+	if err != nil && !errors.Is(err, elf.ErrNoSymbols) {
+		return err
+	}
+	for _, s := range syms {
+		if elf.ST_TYPE(s.Info) != elf.STT_FUNC {
+			// Symbol not associated with a function or other executable code.
+			continue
+		}
+		ex.symbols[s.Name] = s
+	}
+	return nil
+}
+
+func (ex *Executable) symbol(symbol string) (*elf.Symbol, error) {
+	if s, ok := ex.symbols[symbol]; ok {
+		return &s, nil
+	}
+	return nil, fmt.Errorf("symbol %s not found", symbol)
+}
+
+// Uprobe attaches the given eBPF program to a perf event that fires when the
+// given symbol starts executing in the given Executable.
+// For example, /bin/bash::main():
+//
+//  ex, _ = OpenExecutable("/bin/bash")
+//  ex.Uprobe("main", prog, nil)
+//
+// When using symbols which belongs to shared libraries,
+// an offset must be provided via options:
+//
+//  ex.Uprobe("main", prog, &UprobeOptions{Offset: 0x123})
+//
+// The resulting Link must be Closed during program shutdown to avoid leaking
+// system resources. Functions provided by shared libraries can currently not
+// be traced and will result in an ErrNotSupported.
+func (ex *Executable) Uprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions) (Link, error) {
+	u, err := ex.uprobe(symbol, prog, opts, false)
+	if err != nil {
+		return nil, err
+	}
+
+	err = u.attach(prog)
+	if err != nil {
+		u.Close()
+		return nil, err
+	}
+
+	return u, nil
+}
+
+// Uretprobe attaches the given eBPF program to a perf event that fires right
+// before the given symbol exits. For example, /bin/bash::main():
+//
+//  ex, _ = OpenExecutable("/bin/bash")
+//  ex.Uretprobe("main", prog, nil)
+//
+// When using symbols which belongs to shared libraries,
+// an offset must be provided via options:
+//
+//  ex.Uretprobe("main", prog, &UprobeOptions{Offset: 0x123})
+//
+// The resulting Link must be Closed during program shutdown to avoid leaking
+// system resources. Functions provided by shared libraries can currently not
+// be traced and will result in an ErrNotSupported.
+func (ex *Executable) Uretprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions) (Link, error) {
+	u, err := ex.uprobe(symbol, prog, opts, true)
+	if err != nil {
+		return nil, err
+	}
+
+	err = u.attach(prog)
+	if err != nil {
+		u.Close()
+		return nil, err
+	}
+
+	return u, nil
+}
+
+// uprobe opens a perf event for the given binary/symbol and attaches prog to it.
+// If ret is true, create a uretprobe.
+func (ex *Executable) uprobe(symbol string, prog *ebpf.Program, opts *UprobeOptions, ret bool) (*perfEvent, error) {
+	if prog == nil {
+		return nil, fmt.Errorf("prog cannot be nil: %w", errInvalidInput)
+	}
+	if prog.Type() != ebpf.Kprobe {
+		return nil, fmt.Errorf("eBPF program type %s is not Kprobe: %w", prog.Type(), errInvalidInput)
+	}
+
+	var offset uint64
+	if opts != nil && opts.Offset != 0 {
+		offset = opts.Offset
+	} else {
+		sym, err := ex.symbol(symbol)
+		if err != nil {
+			return nil, fmt.Errorf("symbol '%s' not found: %w", symbol, err)
+		}
+
+		// Symbols with location 0 from section undef are shared library calls and
+		// are relocated before the binary is executed. Dynamic linking is not
+		// implemented by the library, so mark this as unsupported for now.
+		if sym.Section == elf.SHN_UNDEF && sym.Value == 0 {
+			return nil, fmt.Errorf("cannot resolve %s library call '%s', "+
+				"consider providing the offset via options: %w", ex.path, symbol, ErrNotSupported)
+		}
+
+		offset = sym.Value
+	}
+
+	// Use uprobe PMU if the kernel has it available.
+	tp, err := pmuUprobe(symbol, ex.path, offset, ret)
+	if err == nil {
+		return tp, nil
+	}
+	if err != nil && !errors.Is(err, ErrNotSupported) {
+		return nil, fmt.Errorf("creating perf_uprobe PMU: %w", err)
+	}
+
+	// Use tracefs if uprobe PMU is missing.
+	tp, err = tracefsUprobe(uprobeSanitizedSymbol(symbol), ex.path, offset, ret)
+	if err != nil {
+		return nil, fmt.Errorf("creating trace event '%s:%s' in tracefs: %w", ex.path, symbol, err)
+	}
+
+	return tp, nil
+}
+
+// pmuUprobe opens a perf event based on the uprobe PMU.
+func pmuUprobe(symbol, path string, offset uint64, ret bool) (*perfEvent, error) {
+	return pmuProbe(uprobeType, symbol, path, offset, ret)
+}
+
+// tracefsUprobe creates a Uprobe tracefs entry.
+func tracefsUprobe(symbol, path string, offset uint64, ret bool) (*perfEvent, error) {
+	return tracefsProbe(uprobeType, symbol, path, offset, ret)
+}
+
+// uprobeSanitizedSymbol replaces every invalid characted for the tracefs api with an underscore.
+func uprobeSanitizedSymbol(symbol string) string {
+	return rgxUprobeSymbol.ReplaceAllString(symbol, "_")
+}
+
+// uprobePathOffset creates the PATH:OFFSET token for the tracefs api.
+func uprobePathOffset(path string, offset uint64) string {
+	return fmt.Sprintf("%s:%#x", path, offset)
+}
+
+func uretprobeBit() (uint64, error) {
+	uprobeRetprobeBit.once.Do(func() {
+		uprobeRetprobeBit.value, uprobeRetprobeBit.err = determineRetprobeBit(uprobeType)
+	})
+	return uprobeRetprobeBit.value, uprobeRetprobeBit.err
+}

vendor/github.com/cilium/ebpf/linker.go

@@ -108,12 +108,16 @@ func fixupJumpsAndCalls(insns asm.Instructions) error {
 		offset := iter.Offset
 		ins := iter.Ins
 
+		if ins.Reference == "" {
+			continue
+		}
+
 		switch {
 		case ins.IsFunctionCall() && ins.Constant == -1:
 			// Rewrite bpf to bpf call
 			callOffset, ok := symbolOffsets[ins.Reference]
 			if !ok {
-				return fmt.Errorf("instruction %d: reference to missing symbol %q", i, ins.Reference)
+				return fmt.Errorf("call at %d: reference to missing symbol %q", i, ins.Reference)
 			}
 
 			ins.Constant = int64(callOffset - offset - 1)
@@ -122,10 +126,13 @@ func fixupJumpsAndCalls(insns asm.Instructions) error {
 			// Rewrite jump to label
 			jumpOffset, ok := symbolOffsets[ins.Reference]
 			if !ok {
-				return fmt.Errorf("instruction %d: reference to missing symbol %q", i, ins.Reference)
+				return fmt.Errorf("jump at %d: reference to missing symbol %q", i, ins.Reference)
 			}
 
 			ins.Offset = int16(jumpOffset - offset - 1)
+
+		case ins.IsLoadFromMap() && ins.MapPtr() == -1:
+			return fmt.Errorf("map %s: %w", ins.Reference, errUnsatisfiedReference)
 		}
 	}
 

vendor/github.com/cilium/ebpf/map.go

@@ -18,6 +18,7 @@ var (
 	ErrKeyNotExist      = errors.New("key does not exist")
 	ErrKeyExist         = errors.New("key already exists")
 	ErrIterationAborted = errors.New("iteration aborted")
+	ErrMapIncompatible  = errors.New("map's spec is incompatible with pinned map")
 )
 
 // MapOptions control loading a map into the kernel.
@@ -87,6 +88,23 @@ func (ms *MapSpec) Copy() *MapSpec {
 	return &cpy
 }
 
+func (ms *MapSpec) clampPerfEventArraySize() error {
+	if ms.Type != PerfEventArray {
+		return nil
+	}
+
+	n, err := internal.PossibleCPUs()
+	if err != nil {
+		return fmt.Errorf("perf event array: %w", err)
+	}
+
+	if n := uint32(n); ms.MaxEntries > n {
+		ms.MaxEntries = n
+	}
+
+	return nil
+}
+
 // MapKV is used to initialize the contents of a Map.
 type MapKV struct {
 	Key   interface{}
@@ -96,19 +114,19 @@ type MapKV struct {
 func (ms *MapSpec) checkCompatibility(m *Map) error {
 	switch {
 	case m.typ != ms.Type:
-		return fmt.Errorf("expected type %v, got %v", ms.Type, m.typ)
+		return fmt.Errorf("expected type %v, got %v: %w", ms.Type, m.typ, ErrMapIncompatible)
 
 	case m.keySize != ms.KeySize:
-		return fmt.Errorf("expected key size %v, got %v", ms.KeySize, m.keySize)
+		return fmt.Errorf("expected key size %v, got %v: %w", ms.KeySize, m.keySize, ErrMapIncompatible)
 
 	case m.valueSize != ms.ValueSize:
-		return fmt.Errorf("expected value size %v, got %v", ms.ValueSize, m.valueSize)
+		return fmt.Errorf("expected value size %v, got %v: %w", ms.ValueSize, m.valueSize, ErrMapIncompatible)
 
 	case m.maxEntries != ms.MaxEntries:
-		return fmt.Errorf("expected max entries %v, got %v", ms.MaxEntries, m.maxEntries)
+		return fmt.Errorf("expected max entries %v, got %v: %w", ms.MaxEntries, m.maxEntries, ErrMapIncompatible)
 
 	case m.flags != ms.Flags:
-		return fmt.Errorf("expected flags %v, got %v", ms.Flags, m.flags)
+		return fmt.Errorf("expected flags %v, got %v: %w", ms.Flags, m.flags, ErrMapIncompatible)
 	}
 	return nil
 }
@@ -171,14 +189,16 @@ func NewMap(spec *MapSpec) (*Map, error) {
 // The caller is responsible for ensuring the process' rlimit is set
 // sufficiently high for locking memory during map creation. This can be done
 // by calling unix.Setrlimit with unix.RLIMIT_MEMLOCK prior to calling NewMapWithOptions.
+//
+// May return an error wrapping ErrMapIncompatible.
 func NewMapWithOptions(spec *MapSpec, opts MapOptions) (*Map, error) {
-	btfs := make(btfHandleCache)
+	handles := newHandleCache()
-	defer btfs.close()
+	defer handles.close()
 
-	return newMapWithOptions(spec, opts, btfs)
+	return newMapWithOptions(spec, opts, handles)
 }
 
-func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *Map, err error) {
+func newMapWithOptions(spec *MapSpec, opts MapOptions, handles *handleCache) (_ *Map, err error) {
 	closeOnError := func(c io.Closer) {
 		if err != nil {
 			c.Close()
@@ -202,7 +222,7 @@ func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *
 		defer closeOnError(m)
 
 		if err := spec.checkCompatibility(m); err != nil {
-			return nil, fmt.Errorf("use pinned map %s: %s", spec.Name, err)
+			return nil, fmt.Errorf("use pinned map %s: %w", spec.Name, err)
 		}
 
 		return m, nil
@@ -211,7 +231,7 @@ func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *
 		// Nothing to do here
 
 	default:
-		return nil, fmt.Errorf("unsupported pin type %d", int(spec.Pinning))
+		return nil, fmt.Errorf("pin type %d: %w", int(spec.Pinning), ErrNotSupported)
 	}
 
 	var innerFd *internal.FD
@@ -224,7 +244,7 @@ func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *
 			return nil, errors.New("inner maps cannot be pinned")
 		}
 
-		template, err := createMap(spec.InnerMap, nil, opts, btfs)
+		template, err := createMap(spec.InnerMap, nil, opts, handles)
 		if err != nil {
 			return nil, err
 		}
@@ -233,7 +253,7 @@ func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *
 		innerFd = template.fd
 	}
 
-	m, err := createMap(spec, innerFd, opts, btfs)
+	m, err := createMap(spec, innerFd, opts, handles)
 	if err != nil {
 		return nil, err
 	}
@@ -249,7 +269,7 @@ func newMapWithOptions(spec *MapSpec, opts MapOptions, btfs btfHandleCache) (_ *
 	return m, nil
 }
 
-func createMap(spec *MapSpec, inner *internal.FD, opts MapOptions, btfs btfHandleCache) (_ *Map, err error) {
+func createMap(spec *MapSpec, inner *internal.FD, opts MapOptions, handles *handleCache) (_ *Map, err error) {
 	closeOnError := func(closer io.Closer) {
 		if err != nil {
 			closer.Close()
@@ -296,44 +316,54 @@ func createMap(spec *MapSpec, inner *internal.FD, opts MapOptions, btfs btfHandl
 			return nil, fmt.Errorf("map create: %w", err)
 		}
 	}
+	if spec.Flags&unix.BPF_F_MMAPABLE > 0 {
+		if err := haveMmapableMaps(); err != nil {
+			return nil, fmt.Errorf("map create: %w", err)
+		}
+	}
+	if spec.Flags&unix.BPF_F_INNER_MAP > 0 {
+		if err := haveInnerMaps(); err != nil {
+			return nil, fmt.Errorf("map create: %w", err)
+		}
+	}
 
-	attr := bpfMapCreateAttr{
+	attr := internal.BPFMapCreateAttr{
-		mapType:    spec.Type,
+		MapType:    uint32(spec.Type),
-		keySize:    spec.KeySize,
+		KeySize:    spec.KeySize,
-		valueSize:  spec.ValueSize,
+		ValueSize:  spec.ValueSize,
-		maxEntries: spec.MaxEntries,
+		MaxEntries: spec.MaxEntries,
-		flags:      spec.Flags,
+		Flags:      spec.Flags,
-		numaNode:   spec.NumaNode,
+		NumaNode:   spec.NumaNode,
 	}
 
 	if inner != nil {
 		var err error
-		attr.innerMapFd, err = inner.Value()
+		attr.InnerMapFd, err = inner.Value()
 		if err != nil {
 			return nil, fmt.Errorf("map create: %w", err)
 		}
 	}
 
 	if haveObjName() == nil {
-		attr.mapName = newBPFObjName(spec.Name)
+		attr.MapName = internal.NewBPFObjName(spec.Name)
 	}
 
 	var btfDisabled bool
 	if spec.BTF != nil {
-		handle, err := btfs.load(btf.MapSpec(spec.BTF))
+		handle, err := handles.btfHandle(btf.MapSpec(spec.BTF))
 		btfDisabled = errors.Is(err, btf.ErrNotSupported)
 		if err != nil && !btfDisabled {
 			return nil, fmt.Errorf("load BTF: %w", err)
 		}
 
 		if handle != nil {
-			attr.btfFd = uint32(handle.FD())
+			attr.BTFFd = uint32(handle.FD())
-			attr.btfKeyTypeID = btf.MapKey(spec.BTF).ID()
+			attr.BTFKeyTypeID = uint32(btf.MapKey(spec.BTF).ID())
-			attr.btfValueTypeID = btf.MapValue(spec.BTF).ID()
+			attr.BTFValueTypeID = uint32(btf.MapValue(spec.BTF).ID())
 		}
 	}
 
-	fd, err := bpfMapCreate(&attr)
+	fd, err := internal.BPFMapCreate(&attr)
 	if err != nil {
 		if errors.Is(err, unix.EPERM) {
 			return nil, fmt.Errorf("map create: RLIMIT_MEMLOCK may be too low: %w", err)

vendor/github.com/cilium/ebpf/prog.go

@@ -5,6 +5,7 @@ import (
 	"encoding/binary"
 	"errors"
 	"fmt"
+	"io"
 	"math"
 	"path/filepath"
 	"strings"
@@ -19,6 +20,8 @@ import (
 // ErrNotSupported is returned whenever the kernel doesn't support a feature.
 var ErrNotSupported = internal.ErrNotSupported
 
+var errUnsatisfiedReference = errors.New("unsatisfied reference")
+
 // ProgramID represents the unique ID of an eBPF program.
 type ProgramID uint32
 
@@ -41,6 +44,12 @@ type ProgramOptions struct {
 	// Controls the output buffer size for the verifier. Defaults to
 	// DefaultVerifierLogSize.
 	LogSize int
+	// An ELF containing the target BTF for this program. It is used both to
+	// find the correct function to trace and to apply CO-RE relocations.
+	// This is useful in environments where the kernel BTF is not available
+	// (containers) or where it is in a non-standard location. Defaults to
+	// use the kernel BTF from a well-known location.
+	TargetBTF io.ReaderAt
 }
 
 // ProgramSpec defines a Program.
@@ -125,21 +134,21 @@ func NewProgram(spec *ProgramSpec) (*Program, error) {
 // Loading a program for the first time will perform
 // feature detection by loading small, temporary programs.
 func NewProgramWithOptions(spec *ProgramSpec, opts ProgramOptions) (*Program, error) {
-	btfs := make(btfHandleCache)
+	handles := newHandleCache()
-	defer btfs.close()
+	defer handles.close()
 
-	return newProgramWithOptions(spec, opts, btfs)
+	prog, err := newProgramWithOptions(spec, opts, handles)
+	if errors.Is(err, errUnsatisfiedReference) {
+		return nil, fmt.Errorf("cannot load program without loading its whole collection: %w", err)
+	}
+	return prog, err
 }
 
-func newProgramWithOptions(spec *ProgramSpec, opts ProgramOptions, btfs btfHandleCache) (*Program, error) {
+func newProgramWithOptions(spec *ProgramSpec, opts ProgramOptions, handles *handleCache) (*Program, error) {
 	if len(spec.Instructions) == 0 {
 		return nil, errors.New("Instructions cannot be empty")
 	}
 
-	if len(spec.License) == 0 {
-		return nil, errors.New("License cannot be empty")
-	}
-
 	if spec.ByteOrder != nil && spec.ByteOrder != internal.NativeEndian {
 		return nil, fmt.Errorf("can't load %s program on %s", spec.ByteOrder, internal.NativeEndian)
 	}
@@ -157,44 +166,36 @@ func newProgramWithOptions(spec *ProgramSpec, opts ProgramOptions, btfs btfHandl
 		kv = v.Kernel()
 	}
 
-	insns := make(asm.Instructions, len(spec.Instructions))
-	copy(insns, spec.Instructions)
-
-	if err := fixupJumpsAndCalls(insns); err != nil {
-		return nil, err
-	}
-
-	buf := bytes.NewBuffer(make([]byte, 0, len(spec.Instructions)*asm.InstructionSize))
-	err := insns.Marshal(buf, internal.NativeEndian)
-	if err != nil {
-		return nil, err
-	}
-
-	bytecode := buf.Bytes()
-	insCount := uint32(len(bytecode) / asm.InstructionSize)
 	attr := &bpfProgLoadAttr{
 		progType:           spec.Type,
 		progFlags:          spec.Flags,
 		expectedAttachType: spec.AttachType,
-		insCount:           insCount,
-		instructions:       internal.NewSlicePointer(bytecode),
 		license:            internal.NewStringPointer(spec.License),
 		kernelVersion:      kv,
 	}
 
 	if haveObjName() == nil {
-		attr.progName = newBPFObjName(spec.Name)
+		attr.progName = internal.NewBPFObjName(spec.Name)
+	}
+
+	var err error
+	var targetBTF *btf.Spec
+	if opts.TargetBTF != nil {
+		targetBTF, err = handles.btfSpec(opts.TargetBTF)
+		if err != nil {
+			return nil, fmt.Errorf("load target BTF: %w", err)
+		}
 	}
 
 	var btfDisabled bool
+	var core btf.COREFixups
 	if spec.BTF != nil {
-		if relos, err := btf.ProgramRelocations(spec.BTF, nil); err != nil {
+		core, err = btf.ProgramFixups(spec.BTF, targetBTF)
-			return nil, fmt.Errorf("CO-RE relocations: %s", err)
+		if err != nil {
-		} else if len(relos) > 0 {
+			return nil, fmt.Errorf("CO-RE relocations: %w", err)
-			return nil, fmt.Errorf("applying CO-RE relocations: %w", ErrNotSupported)
 		}
 
-		handle, err := btfs.load(btf.ProgramSpec(spec.BTF))
+		handle, err := handles.btfHandle(btf.ProgramSpec(spec.BTF))
 		btfDisabled = errors.Is(err, btf.ErrNotSupported)
 		if err != nil && !btfDisabled {
 			return nil, fmt.Errorf("load BTF: %w", err)
@@ -221,8 +222,27 @@ func newProgramWithOptions(spec *ProgramSpec, opts ProgramOptions, btfs btfHandl
 		}
 	}
 
+	insns, err := core.Apply(spec.Instructions)
+	if err != nil {
+		return nil, fmt.Errorf("CO-RE fixup: %w", err)
+	}
+
+	if err := fixupJumpsAndCalls(insns); err != nil {
+		return nil, err
+	}
+
+	buf := bytes.NewBuffer(make([]byte, 0, len(spec.Instructions)*asm.InstructionSize))
+	err = insns.Marshal(buf, internal.NativeEndian)
+	if err != nil {
+		return nil, err
+	}
+
+	bytecode := buf.Bytes()
+	attr.instructions = internal.NewSlicePointer(bytecode)
+	attr.insCount = uint32(len(bytecode) / asm.InstructionSize)
+
 	if spec.AttachTo != "" {
-		target, err := resolveBTFType(spec.AttachTo, spec.Type, spec.AttachType)
+		target, err := resolveBTFType(targetBTF, spec.AttachTo, spec.Type, spec.AttachType)
 		if err != nil {
 			return nil, err
 		}
@@ -250,7 +270,7 @@ func newProgramWithOptions(spec *ProgramSpec, opts ProgramOptions, btfs btfHandl
 	}
 
 	logErr := err
-	if opts.LogLevel == 0 {
+	if opts.LogLevel == 0 && opts.LogSize >= 0 {
 		// Re-run with the verifier enabled to get better error messages.
 		logBuf = make([]byte, logSize)
 		attr.logLevel = 1
@@ -664,52 +684,45 @@ func (p *Program) ID() (ProgramID, error) {
 	return ProgramID(info.id), nil
 }
 
-func findKernelType(name string, typ btf.Type) error {
+func resolveBTFType(kernel *btf.Spec, name string, progType ProgramType, attachType AttachType) (btf.Type, error) {
-	kernel, err := btf.LoadKernelSpec()
-	if err != nil {
-		return fmt.Errorf("can't load kernel spec: %w", err)
-	}
-
-	return kernel.FindType(name, typ)
-}
-
-func resolveBTFType(name string, progType ProgramType, attachType AttachType) (btf.Type, error) {
 	type match struct {
 		p ProgramType
 		a AttachType
 	}
 
-	target := match{progType, attachType}
+	var target btf.Type
-	switch target {
+	var typeName, featureName string
+	switch (match{progType, attachType}) {
 	case match{LSM, AttachLSMMac}:
-		var target btf.Func
+		target = new(btf.Func)
-		err := findKernelType("bpf_lsm_"+name, &target)
+		typeName = "bpf_lsm_" + name
-		if errors.Is(err, btf.ErrNotFound) {
+		featureName = name + " LSM hook"
-			return nil, &internal.UnsupportedFeatureError{
-				Name: name + " LSM hook",
-			}
-		}
-		if err != nil {
-			return nil, fmt.Errorf("resolve BTF for LSM hook %s: %w", name, err)
-		}
-
-		return &target, nil
 
 	case match{Tracing, AttachTraceIter}:
-		var target btf.Func
+		target = new(btf.Func)
-		err := findKernelType("bpf_iter_"+name, &target)
+		typeName = "bpf_iter_" + name
-		if errors.Is(err, btf.ErrNotFound) {
+		featureName = name + " iterator"
-			return nil, &internal.UnsupportedFeatureError{
-				Name: name + " iterator",
-			}
-		}
-		if err != nil {
-			return nil, fmt.Errorf("resolve BTF for iterator %s: %w", name, err)
-		}
-
-		return &target, nil
 
 	default:
 		return nil, nil
 	}
+
+	if kernel == nil {
+		var err error
+		kernel, err = btf.LoadKernelSpec()
+		if err != nil {
+			return nil, fmt.Errorf("load kernel spec: %w", err)
+		}
+	}
+
+	err := kernel.FindType(typeName, target)
+	if errors.Is(err, btf.ErrNotFound) {
+		return nil, &internal.UnsupportedFeatureError{
+			Name: featureName,
+		}
+	}
+	if err != nil {
+		return nil, fmt.Errorf("resolve BTF for %s: %w", featureName, err)
+	}
+	return target, nil
 }

vendor/github.com/cilium/ebpf/run-tests.sh

@@ -1,56 +1,95 @@
 #!/bin/bash
 # Test the current package under a different kernel.
 # Requires virtme and qemu to be installed.
+# Examples:
+#     Run all tests on a 5.4 kernel
+#     $ ./run-tests.sh 5.4
+#     Run a subset of tests:
+#     $ ./run-tests.sh 5.4 go test ./link
 
-set -eu
+set -euo pipefail
-set -o pipefail
 
-if [[ "${1:-}" = "--in-vm" ]]; then
+script="$(realpath "$0")"
+readonly script
+
+# This script is a bit like a Matryoshka doll since it keeps re-executing itself
+# in various different contexts:
+#
+#   1. invoked by the user like run-tests.sh 5.4
+#   2. invoked by go test like run-tests.sh --exec-vm
+#   3. invoked by init in the vm like run-tests.sh --exec-test
+#
+# This allows us to use all available CPU on the host machine to compile our
+# code, and then only use the VM to execute the test. This is because the VM
+# is usually slower at compiling than the host.
+if [[ "${1:-}" = "--exec-vm" ]]; then
+  shift
+
+  input="$1"
+  shift
+
+  # Use sudo if /dev/kvm isn't accessible by the current user.
+  sudo=""
+  if [[ ! -r /dev/kvm || ! -w /dev/kvm ]]; then
+    sudo="sudo"
+  fi
+  readonly sudo
+
+  testdir="$(dirname "$1")"
+  output="$(mktemp -d)"
+  printf -v cmd "%q " "$@"
+
+  if [[ "$(stat -c '%t:%T' -L /proc/$$/fd/0)" == "1:3" ]]; then
+    # stdin is /dev/null, which doesn't play well with qemu. Use a fifo as a
+    # blocking substitute.
+    mkfifo "${output}/fake-stdin"
+    # Open for reading and writing to avoid blocking.
+    exec 0<> "${output}/fake-stdin"
+    rm "${output}/fake-stdin"
+  fi
+
+  $sudo virtme-run --kimg "${input}/bzImage" --memory 768M --pwd \
+  --rwdir="${testdir}=${testdir}" \
+  --rodir=/run/input="${input}" \
+  --rwdir=/run/output="${output}" \
+  --script-sh "PATH=\"$PATH\" \"$script\" --exec-test $cmd" \
+  --qemu-opts -smp 2 # need at least two CPUs for some tests
+
+  if [[ ! -e "${output}/success" ]]; then
+    exit 1
+  fi
+
+  $sudo rm -r "$output"
+  exit 0
+elif [[ "${1:-}" = "--exec-test" ]]; then
   shift
 
   mount -t bpf bpf /sys/fs/bpf
   mount -t tracefs tracefs /sys/kernel/debug/tracing
-  export CGO_ENABLED=0
-  export GOFLAGS=-mod=readonly
-  export GOPATH=/run/go-path
-  export GOPROXY=file:///run/go-path/pkg/mod/cache/download
-  export GOSUMDB=off
-  export GOCACHE=/run/go-cache
 
   if [[ -d "/run/input/bpf" ]]; then
     export KERNEL_SELFTESTS="/run/input/bpf"
   fi
 
-  readonly output="${1}"
+  dmesg -C
-  shift
+  if ! "$@"; then
-
+    dmesg
-  echo Running tests...
+    exit 1
-  go test -v -coverpkg=./... -coverprofile="$output/coverage.txt" -count 1 ./...
+  fi
-  touch "$output/success"
+  touch "/run/output/success"
   exit 0
 fi
 
-# Pull all dependencies, so that we can run tests without the
-# vm having network access.
-go mod download
-
-# Use sudo if /dev/kvm isn't accessible by the current user.
-sudo=""
-if [[ ! -r /dev/kvm || ! -w /dev/kvm ]]; then
-  sudo="sudo"
-fi
-readonly sudo
-
 readonly kernel_version="${1:-}"
 if [[ -z "${kernel_version}" ]]; then
   echo "Expecting kernel version as first argument"
   exit 1
 fi
+shift
 
 readonly kernel="linux-${kernel_version}.bz"
 readonly selftests="linux-${kernel_version}-selftests-bpf.bz"
 readonly input="$(mktemp -d)"
-readonly output="$(mktemp -d)"
 readonly tmp_dir="${TMPDIR:-/tmp}"
 readonly branch="${BRANCH:-master}"
 
@@ -60,6 +99,7 @@ fetch() {
 }
 
 fetch "${kernel}"
+cp "${tmp_dir}/${kernel}" "${input}/bzImage"
 
 if fetch "${selftests}"; then
   mkdir "${input}/bpf"
@@ -68,25 +108,16 @@ else
   echo "No selftests found, disabling"
 fi
 
-echo Testing on "${kernel_version}"
+args=(-v -short -coverpkg=./... -coverprofile=coverage.out -count 1 ./...)
-$sudo virtme-run --kimg "${tmp_dir}/${kernel}" --memory 512M --pwd \
+if (( $# > 0 )); then
-  --rw \
+  args=("$@")
-  --rwdir=/run/input="${input}" \
-  --rwdir=/run/output="${output}" \
-  --rodir=/run/go-path="$(go env GOPATH)" \
-  --rwdir=/run/go-cache="$(go env GOCACHE)" \
-  --script-sh "PATH=\"$PATH\" $(realpath "$0") --in-vm /run/output" \
-  --qemu-opts -smp 2 # need at least two CPUs for some tests
-
-if [[ ! -e "${output}/success" ]]; then
-  echo "Test failed on ${kernel_version}"
-  exit 1
-else
-  echo "Test successful on ${kernel_version}"
-  if [[ -v COVERALLS_TOKEN ]]; then
-    goveralls -coverprofile="${output}/coverage.txt" -service=semaphore -repotoken "$COVERALLS_TOKEN"
-  fi
 fi
 
-$sudo rm -r "${input}"
+export GOFLAGS=-mod=readonly
-$sudo rm -r "${output}"
+export CGO_ENABLED=0
+
+echo Testing on "${kernel_version}"
+go test -exec "$script --exec-vm $input" "${args[@]}"
+echo "Test successful on ${kernel_version}"
+
+rm -r "${input}"

vendor/github.com/cilium/ebpf/syscalls.go

@@ -3,6 +3,7 @@ package ebpf
 import (
 	"errors"
 	"fmt"
+	"os"
 	"unsafe"
 
 	"github.com/cilium/ebpf/internal"
@@ -10,19 +11,10 @@ import (
 	"github.com/cilium/ebpf/internal/unix"
 )
 
-// Generic errors returned by BPF syscalls.
+// ErrNotExist is returned when loading a non-existing map or program.
-var ErrNotExist = errors.New("requested object does not exist")
+//
-
+// Deprecated: use os.ErrNotExist instead.
-// bpfObjName is a null-terminated string made up of
+var ErrNotExist = os.ErrNotExist
-// 'A-Za-z0-9_' characters.
-type bpfObjName [unix.BPF_OBJ_NAME_LEN]byte
-
-// newBPFObjName truncates the result if it is too long.
-func newBPFObjName(name string) bpfObjName {
-	var result bpfObjName
-	copy(result[:unix.BPF_OBJ_NAME_LEN-1], name)
-	return result
-}
 
 // invalidBPFObjNameChar returns true if char may not appear in
 // a BPF object name.
@@ -45,21 +37,6 @@ func invalidBPFObjNameChar(char rune) bool {
 	}
 }
 
-type bpfMapCreateAttr struct {
-	mapType        MapType
-	keySize        uint32
-	valueSize      uint32
-	maxEntries     uint32
-	flags          uint32
-	innerMapFd     uint32     // since 4.12 56f668dfe00d
-	numaNode       uint32     // since 4.14 96eabe7a40aa
-	mapName        bpfObjName // since 4.15 ad5b177bd73f
-	mapIfIndex     uint32
-	btfFd          uint32
-	btfKeyTypeID   btf.TypeID
-	btfValueTypeID btf.TypeID
-}
-
 type bpfMapOpAttr struct {
 	mapFd   uint32
 	padding uint32
@@ -86,7 +63,7 @@ type bpfMapInfo struct {
 	value_size                uint32
 	max_entries               uint32
 	map_flags                 uint32
-	name                      bpfObjName // since 4.15 ad5b177bd73f
+	name                      internal.BPFObjName // since 4.15 ad5b177bd73f
 	ifindex                   uint32              // since 4.16 52775b33bb50
 	btf_vmlinux_value_type_id uint32              // since 5.6  85d33df357b6
 	netns_dev                 uint64              // since 4.16 52775b33bb50
@@ -106,7 +83,7 @@ type bpfProgLoadAttr struct {
 	logBuf             internal.Pointer
 	kernelVersion      uint32              // since 4.1  2541517c32be
 	progFlags          uint32              // since 4.11 e07b98d9bffe
-	progName           bpfObjName // since 4.15 067cae47771c
+	progName           internal.BPFObjName // since 4.15 067cae47771c
 	progIfIndex        uint32              // since 4.15 1f6f4cb7ba21
 	expectedAttachType AttachType          // since 4.17 5e43f899b03a
 	progBTFFd          uint32
@@ -132,7 +109,7 @@ type bpfProgInfo struct {
 	created_by_uid           uint32
 	nr_map_ids               uint32
 	map_ids                  internal.Pointer
-	name                     bpfObjName // since 4.15 067cae47771c
+	name                     internal.BPFObjName // since 4.15 067cae47771c
 	ifindex                  uint32
 	gpl_compatible           uint32
 	netns_dev                uint64
@@ -188,7 +165,7 @@ func bpfProgLoad(attr *bpfProgLoadAttr) (*internal.FD, error) {
 		fd, err := internal.BPF(internal.BPF_PROG_LOAD, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
 		// As of ~4.20 the verifier can be interrupted by a signal,
 		// and returns EAGAIN in that case.
-		if err == unix.EAGAIN {
+		if errors.Is(err, unix.EAGAIN) {
 			continue
 		}
 
@@ -205,23 +182,14 @@ func bpfProgTestRun(attr *bpfProgTestRunAttr) error {
 	return err
 }
 
-func bpfMapCreate(attr *bpfMapCreateAttr) (*internal.FD, error) {
-	fd, err := internal.BPF(internal.BPF_MAP_CREATE, unsafe.Pointer(attr), unsafe.Sizeof(*attr))
-	if err != nil {
-		return nil, err
-	}
-
-	return internal.NewFD(uint32(fd)), nil
-}
-
 var haveNestedMaps = internal.FeatureTest("nested maps", "4.12", func() error {
-	_, err := bpfMapCreate(&bpfMapCreateAttr{
+	_, err := internal.BPFMapCreate(&internal.BPFMapCreateAttr{
-		mapType:    ArrayOfMaps,
+		MapType:    uint32(ArrayOfMaps),
-		keySize:    4,
+		KeySize:    4,
-		valueSize:  4,
+		ValueSize:  4,
-		maxEntries: 1,
+		MaxEntries: 1,
 		// Invalid file descriptor.
-		innerMapFd: ^uint32(0),
+		InnerMapFd: ^uint32(0),
 	})
 	if errors.Is(err, unix.EINVAL) {
 		return internal.ErrNotSupported
@@ -235,12 +203,44 @@ var haveNestedMaps = internal.FeatureTest("nested maps", "4.12", func() error {
 var haveMapMutabilityModifiers = internal.FeatureTest("read- and write-only maps", "5.2", func() error {
 	// This checks BPF_F_RDONLY_PROG and BPF_F_WRONLY_PROG. Since
 	// BPF_MAP_FREEZE appeared in 5.2 as well we don't do a separate check.
-	m, err := bpfMapCreate(&bpfMapCreateAttr{
+	m, err := internal.BPFMapCreate(&internal.BPFMapCreateAttr{
-		mapType:    Array,
+		MapType:    uint32(Array),
-		keySize:    4,
+		KeySize:    4,
-		valueSize:  4,
+		ValueSize:  4,
-		maxEntries: 1,
+		MaxEntries: 1,
-		flags:      unix.BPF_F_RDONLY_PROG,
+		Flags:      unix.BPF_F_RDONLY_PROG,
+	})
+	if err != nil {
+		return internal.ErrNotSupported
+	}
+	_ = m.Close()
+	return nil
+})
+
+var haveMmapableMaps = internal.FeatureTest("mmapable maps", "5.5", func() error {
+	// This checks BPF_F_MMAPABLE, which appeared in 5.5 for array maps.
+	m, err := internal.BPFMapCreate(&internal.BPFMapCreateAttr{
+		MapType:    uint32(Array),
+		KeySize:    4,
+		ValueSize:  4,
+		MaxEntries: 1,
+		Flags:      unix.BPF_F_MMAPABLE,
+	})
+	if err != nil {
+		return internal.ErrNotSupported
+	}
+	_ = m.Close()
+	return nil
+})
+
+var haveInnerMaps = internal.FeatureTest("inner maps", "5.10", func() error {
+	// This checks BPF_F_INNER_MAP, which appeared in 5.10.
+	m, err := internal.BPFMapCreate(&internal.BPFMapCreateAttr{
+		MapType:    uint32(Array),
+		KeySize:    4,
+		ValueSize:  4,
+		MaxEntries: 1,
+		Flags:      unix.BPF_F_INNER_MAP,
 	})
 	if err != nil {
 		return internal.ErrNotSupported
@@ -329,7 +329,7 @@ func objGetNextID(cmd internal.BPFCmd, start uint32) (uint32, error) {
 		startID: start,
 	}
 	_, err := internal.BPF(cmd, unsafe.Pointer(&attr), unsafe.Sizeof(attr))
-	return attr.nextID, wrapObjError(err)
+	return attr.nextID, err
 }
 
 func bpfMapBatch(cmd internal.BPFCmd, m *internal.FD, inBatch, outBatch, keys, values internal.Pointer, count uint32, opts *BatchOptions) (uint32, error) {
@@ -355,32 +355,21 @@ func bpfMapBatch(cmd internal.BPFCmd, m *internal.FD, inBatch, outBatch, keys, v
 	return attr.count, wrapMapError(err)
 }
 
-func wrapObjError(err error) error {
-	if err == nil {
-		return nil
-	}
-	if errors.Is(err, unix.ENOENT) {
-		return fmt.Errorf("%w", ErrNotExist)
-	}
-
-	return errors.New(err.Error())
-}
-
 func wrapMapError(err error) error {
 	if err == nil {
 		return nil
 	}
 
 	if errors.Is(err, unix.ENOENT) {
-		return ErrKeyNotExist
+		return internal.SyscallError(ErrKeyNotExist, unix.ENOENT)
 	}
 
 	if errors.Is(err, unix.EEXIST) {
-		return ErrKeyExist
+		return internal.SyscallError(ErrKeyExist, unix.EEXIST)
 	}
 
 	if errors.Is(err, unix.ENOTSUPP) {
-		return ErrNotSupported
+		return internal.SyscallError(ErrNotSupported, unix.ENOTSUPP)
 	}
 
 	return err
@@ -417,15 +406,15 @@ func bpfGetMapInfoByFD(fd *internal.FD) (*bpfMapInfo, error) {
 }
 
 var haveObjName = internal.FeatureTest("object names", "4.15", func() error {
-	attr := bpfMapCreateAttr{
+	attr := internal.BPFMapCreateAttr{
-		mapType:    Array,
+		MapType:    uint32(Array),
-		keySize:    4,
+		KeySize:    4,
-		valueSize:  4,
+		ValueSize:  4,
-		maxEntries: 1,
+		MaxEntries: 1,
-		mapName:    newBPFObjName("feature_test"),
+		MapName:    internal.NewBPFObjName("feature_test"),
 	}
 
-	fd, err := bpfMapCreate(&attr)
+	fd, err := internal.BPFMapCreate(&attr)
 	if err != nil {
 		return internal.ErrNotSupported
 	}
@@ -439,15 +428,15 @@ var objNameAllowsDot = internal.FeatureTest("dot in object names", "5.2", func()
 		return err
 	}
 
-	attr := bpfMapCreateAttr{
+	attr := internal.BPFMapCreateAttr{
-		mapType:    Array,
+		MapType:    uint32(Array),
-		keySize:    4,
+		KeySize:    4,
-		valueSize:  4,
+		ValueSize:  4,
-		maxEntries: 1,
+		MaxEntries: 1,
-		mapName:    newBPFObjName(".test"),
+		MapName:    internal.NewBPFObjName(".test"),
 	}
 
-	fd, err := bpfMapCreate(&attr)
+	fd, err := internal.BPFMapCreate(&attr)
 	if err != nil {
 		return internal.ErrNotSupported
 	}
@@ -458,14 +447,14 @@ var objNameAllowsDot = internal.FeatureTest("dot in object names", "5.2", func()
 
 var haveBatchAPI = internal.FeatureTest("map batch api", "5.6", func() error {
 	var maxEntries uint32 = 2
-	attr := bpfMapCreateAttr{
+	attr := internal.BPFMapCreateAttr{
-		mapType:    Hash,
+		MapType:    uint32(Hash),
-		keySize:    4,
+		KeySize:    4,
-		valueSize:  4,
+		ValueSize:  4,
-		maxEntries: maxEntries,
+		MaxEntries: maxEntries,
 	}
 
-	fd, err := bpfMapCreate(&attr)
+	fd, err := internal.BPFMapCreate(&attr)
 	if err != nil {
 		return internal.ErrNotSupported
 	}
@@ -487,5 +476,5 @@ func bpfObjGetFDByID(cmd internal.BPFCmd, id uint32) (*internal.FD, error) {
 		id: id,
 	}
 	ptr, err := internal.BPF(cmd, unsafe.Pointer(&attr), unsafe.Sizeof(attr))
-	return internal.NewFD(uint32(ptr)), wrapObjError(err)
+	return internal.NewFD(uint32(ptr)), err
 }

vendor/github.com/cilium/ebpf/types.go

@@ -85,6 +85,10 @@ const (
 	SkStorage
 	// DevMapHash - Hash-based indexing scheme for references to network devices.
 	DevMapHash
+	StructOpts
+	RingBuf
+	InodeStorage
+	TaskStorage
 )
 
 // hasPerCPUValue returns true if the Map stores a value per CPU.

vendor/github.com/cilium/ebpf/types_string.go

@@ -34,11 +34,15 @@ func _() {
 	_ = x[Stack-23]
 	_ = x[SkStorage-24]
 	_ = x[DevMapHash-25]
+	_ = x[StructOpts-26]
+	_ = x[RingBuf-27]
+	_ = x[InodeStorage-28]
+	_ = x[TaskStorage-29]
 }
 
-const _MapType_name = "UnspecifiedMapHashArrayProgramArrayPerfEventArrayPerCPUHashPerCPUArrayStackTraceCGroupArrayLRUHashLRUCPUHashLPMTrieArrayOfMapsHashOfMapsDevMapSockMapCPUMapXSKMapSockHashCGroupStorageReusePortSockArrayPerCPUCGroupStorageQueueStackSkStorageDevMapHash"
+const _MapType_name = "UnspecifiedMapHashArrayProgramArrayPerfEventArrayPerCPUHashPerCPUArrayStackTraceCGroupArrayLRUHashLRUCPUHashLPMTrieArrayOfMapsHashOfMapsDevMapSockMapCPUMapXSKMapSockHashCGroupStorageReusePortSockArrayPerCPUCGroupStorageQueueStackSkStorageDevMapHashStructOptsRingBufInodeStorageTaskStorage"
 
-var _MapType_index = [...]uint8{0, 14, 18, 23, 35, 49, 59, 70, 80, 91, 98, 108, 115, 126, 136, 142, 149, 155, 161, 169, 182, 200, 219, 224, 229, 238, 248}
+var _MapType_index = [...]uint16{0, 14, 18, 23, 35, 49, 59, 70, 80, 91, 98, 108, 115, 126, 136, 142, 149, 155, 161, 169, 182, 200, 219, 224, 229, 238, 248, 258, 265, 277, 288}
 
 func (i MapType) String() string {
 	if i >= MapType(len(_MapType_index)-1) {

vendor/github.com/opencontainers/runc/libcontainer/apparmor/apparmor.go

@@ -0,0 +1,16 @@
+package apparmor
+
+import "errors"
+
+var (
+	// IsEnabled returns true if apparmor is enabled for the host.
+	IsEnabled = isEnabled
+
+	// ApplyProfile will apply the profile with the specified name to the process after
+	// the next exec. It is only supported on Linux and produces an ErrApparmorNotEnabled
+	// on other platforms.
+	ApplyProfile = applyProfile
+
+	// ErrApparmorNotEnabled indicates that AppArmor is not enabled or not supported.
+	ErrApparmorNotEnabled = errors.New("apparmor: config provided but apparmor not supported")
+)

vendor/github.com/opencontainers/runc/libcontainer/apparmor/apparmor_linux.go

@@ -15,8 +15,8 @@ var (
 	checkAppArmor   sync.Once
 )
 
-// IsEnabled returns true if apparmor is enabled for the host.
+// isEnabled returns true if apparmor is enabled for the host.
-func IsEnabled() bool {
+func isEnabled() bool {
 	checkAppArmor.Do(func() {
 		if _, err := os.Stat("/sys/kernel/security/apparmor"); err == nil {
 			buf, err := ioutil.ReadFile("/sys/module/apparmor/parameters/enabled")
@@ -57,9 +57,10 @@ func changeOnExec(name string) error {
 	return nil
 }
 
-// ApplyProfile will apply the profile with the specified name to the process after
+// applyProfile will apply the profile with the specified name to the process after
-// the next exec.
+// the next exec. It is only supported on Linux and produces an error on other
-func ApplyProfile(name string) error {
+// platforms.
+func applyProfile(name string) error {
 	if name == "" {
 		return nil
 	}

vendor/github.com/opencontainers/runc/libcontainer/apparmor/apparmor_unsupported.go

@@ -2,17 +2,11 @@
 
 package apparmor
 
-import (
+func isEnabled() bool {
-	"errors"
-)
-
-var ErrApparmorNotEnabled = errors.New("apparmor: config provided but apparmor not supported")
-
-func IsEnabled() bool {
 	return false
 }
 
-func ApplyProfile(name string) error {
+func applyProfile(name string) error {
 	if name != "" {
 		return ErrApparmorNotEnabled
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/devices/devices_emulator.go

@@ -258,9 +258,9 @@ func (e *Emulator) Apply(rule devices.Rule) error {
 
 	if rule.Allow {
 		return e.allow(innerRule)
-	} else {
-		return e.deny(innerRule)
 	}
+
+	return e.deny(innerRule)
 }
 
 // EmulatorFromList takes a reader to a "devices.list"-like source, and returns
@@ -371,3 +371,12 @@ func (source *Emulator) Transition(target *Emulator) ([]*devices.Rule, error) {
 	}
 	return transitionRules, nil
 }
+
+// Rules returns the minimum set of rules necessary to convert a *deny-all*
+// cgroup to the emulated filter state (note that this is not the same as a
+// default cgroupv1 cgroup -- which is allow-all). This is effectively just a
+// wrapper around Transition() with the source emulator being an empty cgroup.
+func (e *Emulator) Rules() ([]*devices.Rule, error) {
+	defaultCgroup := &Emulator{defaultAllow: false}
+	return defaultCgroup.Transition(e)
+}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/ebpf/devicefilter/devicefilter.go

@@ -11,6 +11,7 @@ import (
 	"strconv"
 
 	"github.com/cilium/ebpf/asm"
+	devicesemulator "github.com/opencontainers/runc/libcontainer/cgroups/devices"
 	"github.com/opencontainers/runc/libcontainer/devices"
 	"github.com/pkg/errors"
 	"golang.org/x/sys/unix"
@@ -22,11 +23,44 @@ const (
 )
 
 // DeviceFilter returns eBPF device filter program and its license string
-func DeviceFilter(devices []*devices.Rule) (asm.Instructions, string, error) {
+func DeviceFilter(rules []*devices.Rule) (asm.Instructions, string, error) {
-	p := &program{}
+	// Generate the minimum ruleset for the device rules we are given. While we
+	// don't care about minimum transitions in cgroupv2, using the emulator
+	// gives us a guarantee that the behaviour of devices filtering is the same
+	// as cgroupv1, including security hardenings to avoid misconfiguration
+	// (such as punching holes in wildcard rules).
+	emu := new(devicesemulator.Emulator)
+	for _, rule := range rules {
+		if err := emu.Apply(*rule); err != nil {
+			return nil, "", err
+		}
+	}
+	cleanRules, err := emu.Rules()
+	if err != nil {
+		return nil, "", err
+	}
+
+	p := &program{
+		defaultAllow: emu.IsBlacklist(),
+	}
 	p.init()
-	for i := len(devices) - 1; i >= 0; i-- {
+
-		if err := p.appendDevice(devices[i]); err != nil {
+	for idx, rule := range cleanRules {
+		if rule.Type == devices.WildcardDevice {
+			// We can safely skip over wildcard entries because there should
+			// only be one (at most) at the very start to instruct cgroupv1 to
+			// go into allow-list mode. However we do double-check this here.
+			if idx != 0 || rule.Allow != emu.IsBlacklist() {
+				return nil, "", errors.Errorf("[internal error] emulated cgroupv2 devices ruleset had bad wildcard at idx %v (%s)", idx, rule.CgroupString())
+			}
+			continue
+		}
+		if rule.Allow == p.defaultAllow {
+			// There should be no rules which have an action equal to the
+			// default action, the emulator removes those.
+			return nil, "", errors.Errorf("[internal error] emulated cgroupv2 devices ruleset had no-op rule at idx %v (%s)", idx, rule.CgroupString())
+		}
+		if err := p.appendRule(rule); err != nil {
 			return nil, "", err
 		}
 	}
@@ -36,7 +70,7 @@ func DeviceFilter(devices []*devices.Rule) (asm.Instructions, string, error) {
 
 type program struct {
 	insts        asm.Instructions
-	hasWildCard bool
+	defaultAllow bool
 	blockID      int
 }
 
@@ -67,39 +101,35 @@ func (p *program) init() {
 		asm.LoadMem(asm.R5, asm.R1, 8, asm.Word))
 }
 
-// appendDevice needs to be called from the last element of OCI linux.resources.devices to the head element.
+// appendRule rule converts an OCI rule to the relevant eBPF block and adds it
-func (p *program) appendDevice(dev *devices.Rule) error {
+// to the in-progress filter program. In order to operate properly, it must be
+// called with a "clean" rule list (generated by devices.Emulator.Rules() --
+// with any "a" rules removed).
+func (p *program) appendRule(rule *devices.Rule) error {
 	if p.blockID < 0 {
 		return errors.New("the program is finalized")
 	}
-	if p.hasWildCard {
-		// All entries after wildcard entry are ignored
-		return nil
-	}
 
-	bpfType := int32(-1)
+	var bpfType int32
-	hasType := true
+	switch rule.Type {
-	switch dev.Type {
+	case devices.CharDevice:
-	case 'c':
 		bpfType = int32(unix.BPF_DEVCG_DEV_CHAR)
-	case 'b':
+	case devices.BlockDevice:
 		bpfType = int32(unix.BPF_DEVCG_DEV_BLOCK)
-	case 'a':
-		hasType = false
 	default:
-		// if not specified in OCI json, typ is set to DeviceTypeAll
+		// We do not permit 'a', nor any other types we don't know about.
-		return errors.Errorf("invalid Type %q", string(dev.Type))
+		return errors.Errorf("invalid type %q", string(rule.Type))
 	}
-	if dev.Major > math.MaxUint32 {
+	if rule.Major > math.MaxUint32 {
-		return errors.Errorf("invalid major %d", dev.Major)
+		return errors.Errorf("invalid major %d", rule.Major)
 	}
-	if dev.Minor > math.MaxUint32 {
+	if rule.Minor > math.MaxUint32 {
-		return errors.Errorf("invalid minor %d", dev.Major)
+		return errors.Errorf("invalid minor %d", rule.Major)
 	}
-	hasMajor := dev.Major >= 0 // if not specified in OCI json, major is set to -1
+	hasMajor := rule.Major >= 0 // if not specified in OCI json, major is set to -1
-	hasMinor := dev.Minor >= 0
+	hasMinor := rule.Minor >= 0
 	bpfAccess := int32(0)
-	for _, r := range dev.Permissions {
+	for _, r := range rule.Permissions {
 		switch r {
 		case 'r':
 			bpfAccess |= unix.BPF_DEVCG_ACC_READ
@@ -119,12 +149,10 @@ func (p *program) appendDevice(dev *devices.Rule) error {
 		nextBlockSym     = "block-" + strconv.Itoa(p.blockID+1)
 		prevBlockLastIdx = len(p.insts) - 1
 	)
-	if hasType {
 	p.insts = append(p.insts,
 		// if (R2 != bpfType) goto next
 		asm.JNE.Imm(asm.R2, bpfType, nextBlockSym),
 	)
-	}
 	if hasAccess {
 		p.insts = append(p.insts,
 			// if (R3 & bpfAccess != R3 /* use R1 as a temp var */) goto next
@@ -136,19 +164,16 @@ func (p *program) appendDevice(dev *devices.Rule) error {
 	if hasMajor {
 		p.insts = append(p.insts,
 			// if (R4 != major) goto next
-			asm.JNE.Imm(asm.R4, int32(dev.Major), nextBlockSym),
+			asm.JNE.Imm(asm.R4, int32(rule.Major), nextBlockSym),
 		)
 	}
 	if hasMinor {
 		p.insts = append(p.insts,
 			// if (R5 != minor) goto next
-			asm.JNE.Imm(asm.R5, int32(dev.Minor), nextBlockSym),
+			asm.JNE.Imm(asm.R5, int32(rule.Minor), nextBlockSym),
 		)
 	}
-	if !hasType && !hasAccess && !hasMajor && !hasMinor {
+	p.insts = append(p.insts, acceptBlock(rule.Allow)...)
-		p.hasWildCard = true
-	}
-	p.insts = append(p.insts, acceptBlock(dev.Allow)...)
 	// set blockSym to the first instruction we added in this iteration
 	p.insts[prevBlockLastIdx+1] = p.insts[prevBlockLastIdx+1].Sym(blockSym)
 	p.blockID++
@@ -156,14 +181,14 @@ func (p *program) appendDevice(dev *devices.Rule) error {
 }
 
 func (p *program) finalize() (asm.Instructions, error) {
-	if p.hasWildCard {
+	var v int32
-		// acceptBlock with asm.Return() is already inserted
+	if p.defaultAllow {
-		return p.insts, nil
+		v = 1
 	}
 	blockSym := "block-" + strconv.Itoa(p.blockID)
 	p.insts = append(p.insts,
-		// R0 <- 0
+		// R0 <- v
-		asm.Mov.Imm32(asm.R0, 0).Sym(blockSym),
+		asm.Mov.Imm32(asm.R0, v).Sym(blockSym),
 		asm.Return(),
 	)
 	p.blockID = -1
@@ -171,7 +196,7 @@ func (p *program) finalize() (asm.Instructions, error) {
 }
 
 func acceptBlock(accept bool) asm.Instructions {
-	v := int32(0)
+	var v int32
 	if accept {
 		v = 1
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/ebpf/ebpf.go

@@ -1,57 +0,0 @@
-package ebpf
-
-import (
-	"github.com/cilium/ebpf"
-	"github.com/cilium/ebpf/asm"
-	"github.com/cilium/ebpf/link"
-	"github.com/pkg/errors"
-	"golang.org/x/sys/unix"
-)
-
-// LoadAttachCgroupDeviceFilter installs eBPF device filter program to /sys/fs/cgroup/<foo> directory.
-//
-// Requires the system to be running in cgroup2 unified-mode with kernel >= 4.15 .
-//
-// https://github.com/torvalds/linux/commit/ebc614f687369f9df99828572b1d85a7c2de3d92
-func LoadAttachCgroupDeviceFilter(insts asm.Instructions, license string, dirFD int) (func() error, error) {
-	nilCloser := func() error {
-		return nil
-	}
-	// Increase `ulimit -l` limit to avoid BPF_PROG_LOAD error (#2167).
-	// This limit is not inherited into the container.
-	memlockLimit := &unix.Rlimit{
-		Cur: unix.RLIM_INFINITY,
-		Max: unix.RLIM_INFINITY,
-	}
-	_ = unix.Setrlimit(unix.RLIMIT_MEMLOCK, memlockLimit)
-	spec := &ebpf.ProgramSpec{
-		Type:         ebpf.CGroupDevice,
-		Instructions: insts,
-		License:      license,
-	}
-	prog, err := ebpf.NewProgram(spec)
-	if err != nil {
-		return nilCloser, err
-	}
-	err = link.RawAttachProgram(link.RawAttachProgramOptions{
-		Target:  dirFD,
-		Program: prog,
-		Attach:  ebpf.AttachCGroupDevice,
-		Flags:   unix.BPF_F_ALLOW_MULTI,
-	})
-	if err != nil {
-		return nilCloser, errors.Wrap(err, "failed to call BPF_PROG_ATTACH (BPF_CGROUP_DEVICE, BPF_F_ALLOW_MULTI)")
-	}
-	closer := func() error {
-		err = link.RawDetachProgram(link.RawDetachProgramOptions{
-			Target:  dirFD,
-			Program: prog,
-			Attach:  ebpf.AttachCGroupDevice,
-		})
-		if err != nil {
-			return errors.Wrap(err, "failed to call BPF_PROG_DETACH (BPF_CGROUP_DEVICE)")
-		}
-		return nil
-	}
-	return closer, nil
-}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/ebpf/ebpf_linux.go

@@ -0,0 +1,253 @@
+package ebpf
+
+import (
+	"fmt"
+	"os"
+	"runtime"
+	"sync"
+	"unsafe"
+
+	"github.com/cilium/ebpf"
+	"github.com/cilium/ebpf/asm"
+	"github.com/cilium/ebpf/link"
+	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+func nilCloser() error {
+	return nil
+}
+
+func findAttachedCgroupDeviceFilters(dirFd int) ([]*ebpf.Program, error) {
+	type bpfAttrQuery struct {
+		TargetFd    uint32
+		AttachType  uint32
+		QueryType   uint32
+		AttachFlags uint32
+		ProgIds     uint64 // __aligned_u64
+		ProgCnt     uint32
+	}
+
+	// Currently you can only have 64 eBPF programs attached to a cgroup.
+	size := 64
+	retries := 0
+	for retries < 10 {
+		progIds := make([]uint32, size)
+		query := bpfAttrQuery{
+			TargetFd:   uint32(dirFd),
+			AttachType: uint32(unix.BPF_CGROUP_DEVICE),
+			ProgIds:    uint64(uintptr(unsafe.Pointer(&progIds[0]))),
+			ProgCnt:    uint32(len(progIds)),
+		}
+
+		// Fetch the list of program ids.
+		_, _, errno := unix.Syscall(unix.SYS_BPF,
+			uintptr(unix.BPF_PROG_QUERY),
+			uintptr(unsafe.Pointer(&query)),
+			unsafe.Sizeof(query))
+		size = int(query.ProgCnt)
+		runtime.KeepAlive(query)
+		if errno != 0 {
+			// On ENOSPC we get the correct number of programs.
+			if errno == unix.ENOSPC {
+				retries++
+				continue
+			}
+			return nil, fmt.Errorf("bpf_prog_query(BPF_CGROUP_DEVICE) failed: %w", errno)
+		}
+
+		// Convert the ids to program handles.
+		progIds = progIds[:size]
+		programs := make([]*ebpf.Program, 0, len(progIds))
+		for _, progId := range progIds {
+			program, err := ebpf.NewProgramFromID(ebpf.ProgramID(progId))
+			if err != nil {
+				// We skip over programs that give us -EACCES or -EPERM. This
+				// is necessary because there may be BPF programs that have
+				// been attached (such as with --systemd-cgroup) which have an
+				// LSM label that blocks us from interacting with the program.
+				//
+				// Because additional BPF_CGROUP_DEVICE programs only can add
+				// restrictions, there's no real issue with just ignoring these
+				// programs (and stops runc from breaking on distributions with
+				// very strict SELinux policies).
+				if errors.Is(err, os.ErrPermission) {
+					logrus.Debugf("ignoring existing CGROUP_DEVICE program (prog_id=%v) which cannot be accessed by runc -- likely due to LSM policy: %v", progId, err)
+					continue
+				}
+				return nil, fmt.Errorf("cannot fetch program from id: %w", err)
+			}
+			programs = append(programs, program)
+		}
+		runtime.KeepAlive(progIds)
+		return programs, nil
+	}
+
+	return nil, errors.New("could not get complete list of CGROUP_DEVICE programs")
+}
+
+var (
+	haveBpfProgReplaceBool bool
+	haveBpfProgReplaceOnce sync.Once
+)
+
+// Loosely based on the BPF_F_REPLACE support check in
+//   <https://github.com/cilium/ebpf/blob/v0.6.0/link/syscalls.go>.
+//
+// TODO: move this logic to cilium/ebpf
+func haveBpfProgReplace() bool {
+	haveBpfProgReplaceOnce.Do(func() {
+		prog, err := ebpf.NewProgram(&ebpf.ProgramSpec{
+			Type:    ebpf.CGroupDevice,
+			License: "MIT",
+			Instructions: asm.Instructions{
+				asm.Mov.Imm(asm.R0, 0),
+				asm.Return(),
+			},
+		})
+		if err != nil {
+			logrus.Debugf("checking for BPF_F_REPLACE support: ebpf.NewProgram failed: %v", err)
+			return
+		}
+		defer prog.Close()
+
+		devnull, err := os.Open("/dev/null")
+		if err != nil {
+			logrus.Debugf("checking for BPF_F_REPLACE support: open dummy target fd: %v", err)
+			return
+		}
+		defer devnull.Close()
+
+		// We know that we have BPF_PROG_ATTACH since we can load
+		// BPF_CGROUP_DEVICE programs. If passing BPF_F_REPLACE gives us EINVAL
+		// we know that the feature isn't present.
+		err = link.RawAttachProgram(link.RawAttachProgramOptions{
+			// We rely on this fd being checked after attachFlags.
+			Target: int(devnull.Fd()),
+			// Attempt to "replace" bad fds with this program.
+			Program: prog,
+			Attach:  ebpf.AttachCGroupDevice,
+			Flags:   unix.BPF_F_ALLOW_MULTI | unix.BPF_F_REPLACE,
+		})
+		if errors.Is(err, unix.EINVAL) {
+			// not supported
+			return
+		}
+		// attach_flags test succeded.
+		if !errors.Is(err, unix.EBADF) {
+			logrus.Debugf("checking for BPF_F_REPLACE: got unexpected (not EBADF or EINVAL) error: %v", err)
+		}
+		haveBpfProgReplaceBool = true
+	})
+	return haveBpfProgReplaceBool
+}
+
+// LoadAttachCgroupDeviceFilter installs eBPF device filter program to /sys/fs/cgroup/<foo> directory.
+//
+// Requires the system to be running in cgroup2 unified-mode with kernel >= 4.15 .
+//
+// https://github.com/torvalds/linux/commit/ebc614f687369f9df99828572b1d85a7c2de3d92
+func LoadAttachCgroupDeviceFilter(insts asm.Instructions, license string, dirFd int) (func() error, error) {
+	// Increase `ulimit -l` limit to avoid BPF_PROG_LOAD error (#2167).
+	// This limit is not inherited into the container.
+	memlockLimit := &unix.Rlimit{
+		Cur: unix.RLIM_INFINITY,
+		Max: unix.RLIM_INFINITY,
+	}
+	_ = unix.Setrlimit(unix.RLIMIT_MEMLOCK, memlockLimit)
+
+	// Get the list of existing programs.
+	oldProgs, err := findAttachedCgroupDeviceFilters(dirFd)
+	if err != nil {
+		return nilCloser, err
+	}
+	useReplaceProg := haveBpfProgReplace() && len(oldProgs) == 1
+
+	// Generate new program.
+	spec := &ebpf.ProgramSpec{
+		Type:         ebpf.CGroupDevice,
+		Instructions: insts,
+		License:      license,
+	}
+	prog, err := ebpf.NewProgram(spec)
+	if err != nil {
+		return nilCloser, err
+	}
+
+	// If there is only one old program, we can just replace it directly.
+	var (
+		replaceProg *ebpf.Program
+		attachFlags uint32 = unix.BPF_F_ALLOW_MULTI
+	)
+	if useReplaceProg {
+		replaceProg = oldProgs[0]
+		attachFlags |= unix.BPF_F_REPLACE
+	}
+	err = link.RawAttachProgram(link.RawAttachProgramOptions{
+		Target:  dirFd,
+		Program: prog,
+		Replace: replaceProg,
+		Attach:  ebpf.AttachCGroupDevice,
+		Flags:   attachFlags,
+	})
+	if err != nil {
+		return nilCloser, fmt.Errorf("failed to call BPF_PROG_ATTACH (BPF_CGROUP_DEVICE, BPF_F_ALLOW_MULTI): %w", err)
+	}
+	closer := func() error {
+		err = link.RawDetachProgram(link.RawDetachProgramOptions{
+			Target:  dirFd,
+			Program: prog,
+			Attach:  ebpf.AttachCGroupDevice,
+		})
+		if err != nil {
+			return fmt.Errorf("failed to call BPF_PROG_DETACH (BPF_CGROUP_DEVICE): %w", err)
+		}
+		// TODO: Should we attach the old filters back in this case? Otherwise
+		//       we fail-open on a security feature, which is a bit scary.
+		return nil
+	}
+	if !useReplaceProg {
+		logLevel := logrus.DebugLevel
+		// If there was more than one old program, give a warning (since this
+		// really shouldn't happen with runc-managed cgroups) and then detach
+		// all the old programs.
+		if len(oldProgs) > 1 {
+			// NOTE: Ideally this should be a warning but it turns out that
+			//       systemd-managed cgroups trigger this warning (apparently
+			//       systemd doesn't delete old non-systemd programs when
+			//       setting properties).
+			logrus.Infof("found more than one filter (%d) attached to a cgroup -- removing extra filters!", len(oldProgs))
+			logLevel = logrus.InfoLevel
+		}
+		for idx, oldProg := range oldProgs {
+			// Output some extra debug info.
+			if info, err := oldProg.Info(); err == nil {
+				fields := logrus.Fields{
+					"type": info.Type.String(),
+					"tag":  info.Tag,
+					"name": info.Name,
+				}
+				if id, ok := info.ID(); ok {
+					fields["id"] = id
+				}
+				if runCount, ok := info.RunCount(); ok {
+					fields["run_count"] = runCount
+				}
+				if runtime, ok := info.Runtime(); ok {
+					fields["runtime"] = runtime.String()
+				}
+				logrus.WithFields(fields).Logf(logLevel, "removing old filter %d from cgroup", idx)
+			}
+			err = link.RawDetachProgram(link.RawDetachProgramOptions{
+				Target:  dirFd,
+				Program: oldProg,
+				Attach:  ebpf.AttachCGroupDevice,
+			})
+			if err != nil {
+				return closer, fmt.Errorf("failed to call BPF_PROG_DETACH (BPF_CGROUP_DEVICE) on old filter program: %w", err)
+			}
+		}
+	}
+	return closer, nil
+}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/file.go

@@ -1,6 +1,7 @@
-package fscommon
+package cgroups
 
 import (
+	"bytes"
 	"os"
 	"strings"
 	"sync"
@@ -10,6 +11,54 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+// OpenFile opens a cgroup file in a given dir with given flags.
+// It is supposed to be used for cgroup files only.
+func OpenFile(dir, file string, flags int) (*os.File, error) {
+	if dir == "" {
+		return nil, errors.Errorf("no directory specified for %s", file)
+	}
+	return openFile(dir, file, flags)
+}
+
+// ReadFile reads data from a cgroup file in dir.
+// It is supposed to be used for cgroup files only.
+func ReadFile(dir, file string) (string, error) {
+	fd, err := OpenFile(dir, file, unix.O_RDONLY)
+	if err != nil {
+		return "", err
+	}
+	defer fd.Close()
+	var buf bytes.Buffer
+
+	_, err = buf.ReadFrom(fd)
+	return buf.String(), err
+}
+
+// WriteFile writes data to a cgroup file in dir.
+// It is supposed to be used for cgroup files only.
+func WriteFile(dir, file, data string) error {
+	fd, err := OpenFile(dir, file, unix.O_WRONLY)
+	if err != nil {
+		return err
+	}
+	defer fd.Close()
+	if err := retryingWriteFile(fd, data); err != nil {
+		return errors.Wrapf(err, "failed to write %q", data)
+	}
+	return nil
+}
+
+func retryingWriteFile(fd *os.File, data string) error {
+	for {
+		_, err := fd.Write([]byte(data))
+		if errors.Is(err, unix.EINTR) {
+			logrus.Infof("interrupted while writing %s to %s", data, fd.Name())
+			continue
+		}
+		return err
+	}
+}
+
 const (
 	cgroupfsDir    = "/sys/fs/cgroup"
 	cgroupfsPrefix = cgroupfsDir + "/"
@@ -28,7 +77,8 @@ var (
 func prepareOpenat2() error {
 	prepOnce.Do(func() {
 		fd, err := unix.Openat2(-1, cgroupfsDir, &unix.OpenHow{
-			Flags: unix.O_DIRECTORY | unix.O_PATH})
+			Flags: unix.O_DIRECTORY | unix.O_PATH,
+		})
 		if err != nil {
 			prepErr = &os.PathError{Op: "openat2", Path: cgroupfsDir, Err: err}
 			if err != unix.ENOSYS {
@@ -52,7 +102,6 @@ func prepareOpenat2() error {
 			// cgroupv2 has a single mountpoint and no "cpu,cpuacct" symlinks
 			resolveFlags |= unix.RESOLVE_NO_XDEV | unix.RESOLVE_NO_SYMLINKS
 		}
-
 	})
 
 	return prepErr
@@ -60,10 +109,7 @@ func prepareOpenat2() error {
 
 // OpenFile opens a cgroup file in a given dir with given flags.
 // It is supposed to be used for cgroup files only.
-func OpenFile(dir, file string, flags int) (*os.File, error) {
+func openFile(dir, file string, flags int) (*os.File, error) {
-	if dir == "" {
-		return nil, errors.Errorf("no directory specified for %s", file)
-	}
 	mode := os.FileMode(0)
 	if TestMode && flags&os.O_WRONLY != 0 {
 		// "emulate" cgroup fs for unit tests

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/blkio.go

@@ -6,15 +6,17 @@ import (
 	"bufio"
 	"fmt"
 	"os"
+	"path/filepath"
 	"strconv"
 	"strings"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
 type BlkioGroup struct {
+	weightFilename       string
+	weightDeviceFilename string
 }
 
 func (s *BlkioGroup) Name() string {
@@ -26,42 +28,47 @@ func (s *BlkioGroup) Apply(path string, d *cgroupData) error {
 }
 
 func (s *BlkioGroup) Set(path string, r *configs.Resources) error {
+	s.detectWeightFilenames(path)
 	if r.BlkioWeight != 0 {
-		if err := fscommon.WriteFile(path, "blkio.weight", strconv.FormatUint(uint64(r.BlkioWeight), 10)); err != nil {
+		if err := cgroups.WriteFile(path, s.weightFilename, strconv.FormatUint(uint64(r.BlkioWeight), 10)); err != nil {
 			return err
 		}
 	}
 
 	if r.BlkioLeafWeight != 0 {
-		if err := fscommon.WriteFile(path, "blkio.leaf_weight", strconv.FormatUint(uint64(r.BlkioLeafWeight), 10)); err != nil {
+		if err := cgroups.WriteFile(path, "blkio.leaf_weight", strconv.FormatUint(uint64(r.BlkioLeafWeight), 10)); err != nil {
 			return err
 		}
 	}
 	for _, wd := range r.BlkioWeightDevice {
-		if err := fscommon.WriteFile(path, "blkio.weight_device", wd.WeightString()); err != nil {
+		if wd.Weight != 0 {
+			if err := cgroups.WriteFile(path, s.weightDeviceFilename, wd.WeightString()); err != nil {
 				return err
 			}
-		if err := fscommon.WriteFile(path, "blkio.leaf_weight_device", wd.LeafWeightString()); err != nil {
+		}
+		if wd.LeafWeight != 0 {
+			if err := cgroups.WriteFile(path, "blkio.leaf_weight_device", wd.LeafWeightString()); err != nil {
 				return err
 			}
 		}
+	}
 	for _, td := range r.BlkioThrottleReadBpsDevice {
-		if err := fscommon.WriteFile(path, "blkio.throttle.read_bps_device", td.String()); err != nil {
+		if err := cgroups.WriteFile(path, "blkio.throttle.read_bps_device", td.String()); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleWriteBpsDevice {
-		if err := fscommon.WriteFile(path, "blkio.throttle.write_bps_device", td.String()); err != nil {
+		if err := cgroups.WriteFile(path, "blkio.throttle.write_bps_device", td.String()); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleReadIOPSDevice {
-		if err := fscommon.WriteFile(path, "blkio.throttle.read_iops_device", td.String()); err != nil {
+		if err := cgroups.WriteFile(path, "blkio.throttle.read_iops_device", td.String()); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleWriteIOPSDevice {
-		if err := fscommon.WriteFile(path, "blkio.throttle.write_iops_device", td.String()); err != nil {
+		if err := cgroups.WriteFile(path, "blkio.throttle.write_iops_device", td.String()); err != nil {
 			return err
 		}
 	}
@@ -106,7 +113,7 @@ func splitBlkioStatLine(r rune) bool {
 
 func getBlkioStat(dir, file string) ([]cgroups.BlkioStatEntry, error) {
 	var blkioStats []cgroups.BlkioStatEntry
-	f, err := fscommon.OpenFile(dir, file, os.O_RDONLY)
+	f, err := cgroups.OpenFile(dir, file, os.O_RDONLY)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return blkioStats, nil
@@ -161,7 +168,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 		filename            string
 		blkioStatEntriesPtr *[]cgroups.BlkioStatEntry
 	}
-	var bfqDebugStats = []blkioStatInfo{
+	bfqDebugStats := []blkioStatInfo{
 		{
 			filename:            "blkio.bfq.sectors_recursive",
 			blkioStatEntriesPtr: &stats.BlkioStats.SectorsRecursive,
@@ -195,7 +202,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServiceBytesRecursive,
 		},
 	}
-	var bfqStats = []blkioStatInfo{
+	bfqStats := []blkioStatInfo{
 		{
 			filename:            "blkio.bfq.io_serviced_recursive",
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServicedRecursive,
@@ -205,7 +212,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServiceBytesRecursive,
 		},
 	}
-	var cfqStats = []blkioStatInfo{
+	cfqStats := []blkioStatInfo{
 		{
 			filename:            "blkio.sectors_recursive",
 			blkioStatEntriesPtr: &stats.BlkioStats.SectorsRecursive,
@@ -239,7 +246,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServiceBytesRecursive,
 		},
 	}
-	var throttleRecursiveStats = []blkioStatInfo{
+	throttleRecursiveStats := []blkioStatInfo{
 		{
 			filename:            "blkio.throttle.io_serviced_recursive",
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServicedRecursive,
@@ -249,7 +256,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServiceBytesRecursive,
 		},
 	}
-	var baseStats = []blkioStatInfo{
+	baseStats := []blkioStatInfo{
 		{
 			filename:            "blkio.throttle.io_serviced",
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServicedRecursive,
@@ -259,7 +266,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 			blkioStatEntriesPtr: &stats.BlkioStats.IoServiceBytesRecursive,
 		},
 	}
-	var orderedStats = [][]blkioStatInfo{
+	orderedStats := [][]blkioStatInfo{
 		bfqDebugStats,
 		bfqStats,
 		cfqStats,
@@ -280,7 +287,7 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 				return err
 			}
 			*statInfo.blkioStatEntriesPtr = blkioStats
-			//finish if all stats are gathered
+			// finish if all stats are gathered
 			if i == len(statGroup)-1 {
 				return nil
 			}
@@ -288,3 +295,17 @@ func (s *BlkioGroup) GetStats(path string, stats *cgroups.Stats) error {
 	}
 	return nil
 }
+
+func (s *BlkioGroup) detectWeightFilenames(path string) {
+	if s.weightFilename != "" {
+		// Already detected.
+		return
+	}
+	if cgroups.PathExists(filepath.Join(path, "blkio.weight")) {
+		s.weightFilename = "blkio.weight"
+		s.weightDeviceFilename = "blkio.weight_device"
+	} else {
+		s.weightFilename = "blkio.bfq.weight"
+		s.weightDeviceFilename = "blkio.bfq.weight_device"
+	}
+}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/cpu.go

@@ -13,8 +13,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type CpuGroup struct {
+type CpuGroup struct{}
-}
 
 func (s *CpuGroup) Name() string {
 	return "cpu"
@@ -26,7 +25,7 @@ func (s *CpuGroup) Apply(path string, d *cgroupData) error {
 	if path == "" {
 		return nil
 	}
-	if err := os.MkdirAll(path, 0755); err != nil {
+	if err := os.MkdirAll(path, 0o755); err != nil {
 		return err
 	}
 	// We should set the real-Time group scheduling settings before moving
@@ -42,12 +41,12 @@ func (s *CpuGroup) Apply(path string, d *cgroupData) error {
 
 func (s *CpuGroup) SetRtSched(path string, r *configs.Resources) error {
 	if r.CpuRtPeriod != 0 {
-		if err := fscommon.WriteFile(path, "cpu.rt_period_us", strconv.FormatUint(r.CpuRtPeriod, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "cpu.rt_period_us", strconv.FormatUint(r.CpuRtPeriod, 10)); err != nil {
 			return err
 		}
 	}
 	if r.CpuRtRuntime != 0 {
-		if err := fscommon.WriteFile(path, "cpu.rt_runtime_us", strconv.FormatInt(r.CpuRtRuntime, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "cpu.rt_runtime_us", strconv.FormatInt(r.CpuRtRuntime, 10)); err != nil {
 			return err
 		}
 	}
@@ -57,7 +56,7 @@ func (s *CpuGroup) SetRtSched(path string, r *configs.Resources) error {
 func (s *CpuGroup) Set(path string, r *configs.Resources) error {
 	if r.CpuShares != 0 {
 		shares := r.CpuShares
-		if err := fscommon.WriteFile(path, "cpu.shares", strconv.FormatUint(shares, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "cpu.shares", strconv.FormatUint(shares, 10)); err != nil {
 			return err
 		}
 		// read it back
@@ -73,12 +72,12 @@ func (s *CpuGroup) Set(path string, r *configs.Resources) error {
 		}
 	}
 	if r.CpuPeriod != 0 {
-		if err := fscommon.WriteFile(path, "cpu.cfs_period_us", strconv.FormatUint(r.CpuPeriod, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "cpu.cfs_period_us", strconv.FormatUint(r.CpuPeriod, 10)); err != nil {
 			return err
 		}
 	}
 	if r.CpuQuota != 0 {
-		if err := fscommon.WriteFile(path, "cpu.cfs_quota_us", strconv.FormatInt(r.CpuQuota, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "cpu.cfs_quota_us", strconv.FormatInt(r.CpuQuota, 10)); err != nil {
 			return err
 		}
 	}
@@ -86,7 +85,7 @@ func (s *CpuGroup) Set(path string, r *configs.Resources) error {
 }
 
 func (s *CpuGroup) GetStats(path string, stats *cgroups.Stats) error {
-	f, err := fscommon.OpenFile(path, "cpu.stat", os.O_RDONLY)
+	f, err := cgroups.OpenFile(path, "cpu.stat", os.O_RDONLY)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return nil

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/cpuacct.go

@@ -32,8 +32,7 @@ const (
 	clockTicks uint64 = 100
 )
 
-type CpuacctGroup struct {
+type CpuacctGroup struct{}
-}
 
 func (s *CpuacctGroup) Name() string {
 	return "cpuacct"
@@ -91,7 +90,7 @@ func getCpuUsageBreakdown(path string) (uint64, uint64, error) {
 	// Expected format:
 	// user <usage in ticks>
 	// system <usage in ticks>
-	data, err := fscommon.ReadFile(path, cgroupCpuacctStat)
+	data, err := cgroups.ReadFile(path, cgroupCpuacctStat)
 	if err != nil {
 		return 0, 0, err
 	}
@@ -117,7 +116,7 @@ func getCpuUsageBreakdown(path string) (uint64, uint64, error) {
 
 func getPercpuUsage(path string) ([]uint64, error) {
 	percpuUsage := []uint64{}
-	data, err := fscommon.ReadFile(path, "cpuacct.usage_percpu")
+	data, err := cgroups.ReadFile(path, "cpuacct.usage_percpu")
 	if err != nil {
 		return percpuUsage, err
 	}
@@ -135,7 +134,7 @@ func getPercpuUsageInModes(path string) ([]uint64, []uint64, error) {
 	usageKernelMode := []uint64{}
 	usageUserMode := []uint64{}
 
-	file, err := fscommon.OpenFile(path, cgroupCpuacctUsageAll, os.O_RDONLY)
+	file, err := cgroups.OpenFile(path, cgroupCpuacctUsageAll, os.O_RDONLY)
 	if os.IsNotExist(err) {
 		return usageKernelMode, usageUserMode, nil
 	} else if err != nil {
@@ -144,7 +143,7 @@ func getPercpuUsageInModes(path string) ([]uint64, []uint64, error) {
 	defer file.Close()
 
 	scanner := bufio.NewScanner(file)
-	scanner.Scan() //skipping header line
+	scanner.Scan() // skipping header line
 
 	for scanner.Scan() {
 		lineFields := strings.SplitN(scanner.Text(), " ", cuacctUsageAllColumnsNumber+1)

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/cpuset.go

@@ -16,8 +16,7 @@ import (
 	"golang.org/x/sys/unix"
 )
 
-type CpusetGroup struct {
+type CpusetGroup struct{}
-}
 
 func (s *CpusetGroup) Name() string {
 	return "cpuset"
@@ -29,12 +28,12 @@ func (s *CpusetGroup) Apply(path string, d *cgroupData) error {
 
 func (s *CpusetGroup) Set(path string, r *configs.Resources) error {
 	if r.CpusetCpus != "" {
-		if err := fscommon.WriteFile(path, "cpuset.cpus", r.CpusetCpus); err != nil {
+		if err := cgroups.WriteFile(path, "cpuset.cpus", r.CpusetCpus); err != nil {
 			return err
 		}
 	}
 	if r.CpusetMems != "" {
-		if err := fscommon.WriteFile(path, "cpuset.mems", r.CpusetMems); err != nil {
+		if err := cgroups.WriteFile(path, "cpuset.mems", r.CpusetMems); err != nil {
 			return err
 		}
 	}
@@ -156,7 +155,7 @@ func (s *CpusetGroup) ApplyDir(dir string, r *configs.Resources, pid int) error
 	if err := cpusetEnsureParent(filepath.Dir(dir)); err != nil {
 		return err
 	}
-	if err := os.Mkdir(dir, 0755); err != nil && !os.IsExist(err) {
+	if err := os.Mkdir(dir, 0o755); err != nil && !os.IsExist(err) {
 		return err
 	}
 	// We didn't inherit cpuset configs from parent, but we have
@@ -176,10 +175,10 @@ func (s *CpusetGroup) ApplyDir(dir string, r *configs.Resources, pid int) error
 }
 
 func getCpusetSubsystemSettings(parent string) (cpus, mems string, err error) {
-	if cpus, err = fscommon.ReadFile(parent, "cpuset.cpus"); err != nil {
+	if cpus, err = cgroups.ReadFile(parent, "cpuset.cpus"); err != nil {
 		return
 	}
-	if mems, err = fscommon.ReadFile(parent, "cpuset.mems"); err != nil {
+	if mems, err = cgroups.ReadFile(parent, "cpuset.mems"); err != nil {
 		return
 	}
 	return cpus, mems, nil
@@ -206,7 +205,7 @@ func cpusetEnsureParent(current string) error {
 	if err := cpusetEnsureParent(parent); err != nil {
 		return err
 	}
-	if err := os.Mkdir(current, 0755); err != nil && !os.IsExist(err) {
+	if err := os.Mkdir(current, 0o755); err != nil && !os.IsExist(err) {
 		return err
 	}
 	return cpusetCopyIfNeeded(current, parent)
@@ -225,12 +224,12 @@ func cpusetCopyIfNeeded(current, parent string) error {
 	}
 
 	if isEmptyCpuset(currentCpus) {
-		if err := fscommon.WriteFile(current, "cpuset.cpus", string(parentCpus)); err != nil {
+		if err := cgroups.WriteFile(current, "cpuset.cpus", string(parentCpus)); err != nil {
 			return err
 		}
 	}
 	if isEmptyCpuset(currentMems) {
-		if err := fscommon.WriteFile(current, "cpuset.mems", string(parentMems)); err != nil {
+		if err := cgroups.WriteFile(current, "cpuset.mems", string(parentMems)); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/devices.go

@@ -9,7 +9,6 @@ import (
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
 	cgroupdevices "github.com/opencontainers/runc/libcontainer/cgroups/devices"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/opencontainers/runc/libcontainer/devices"
 	"github.com/opencontainers/runc/libcontainer/userns"
@@ -36,7 +35,7 @@ func (s *DevicesGroup) Apply(path string, d *cgroupData) error {
 }
 
 func loadEmulator(path string) (*cgroupdevices.Emulator, error) {
-	list, err := fscommon.ReadFile(path, "devices.list")
+	list, err := cgroups.ReadFile(path, "devices.list")
 	if err != nil {
 		return nil, err
 	}
@@ -81,7 +80,7 @@ func (s *DevicesGroup) Set(path string, r *configs.Resources) error {
 		if rule.Allow {
 			file = "devices.allow"
 		}
-		if err := fscommon.WriteFile(path, file, rule.CgroupString()); err != nil {
+		if err := cgroups.WriteFile(path, file, rule.CgroupString()); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/freezer.go

@@ -10,14 +10,12 @@ import (
 	"time"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 )
 
-type FreezerGroup struct {
+type FreezerGroup struct{}
-}
 
 func (s *FreezerGroup) Name() string {
 	return "freezer"
@@ -35,7 +33,7 @@ func (s *FreezerGroup) Set(path string, r *configs.Resources) (Err error) {
 				// Freezing failed, and it is bad and dangerous
 				// to leave the cgroup in FROZEN or FREEZING
 				// state, so (try to) thaw it back.
-				_ = fscommon.WriteFile(path, "freezer.state", string(configs.Thawed))
+				_ = cgroups.WriteFile(path, "freezer.state", string(configs.Thawed))
 			}
 		}()
 
@@ -68,11 +66,11 @@ func (s *FreezerGroup) Set(path string, r *configs.Resources) (Err error) {
 				// the chances to succeed in freezing
 				// in case new processes keep appearing
 				// in the cgroup.
-				_ = fscommon.WriteFile(path, "freezer.state", string(configs.Thawed))
+				_ = cgroups.WriteFile(path, "freezer.state", string(configs.Thawed))
 				time.Sleep(10 * time.Millisecond)
 			}
 
-			if err := fscommon.WriteFile(path, "freezer.state", string(configs.Frozen)); err != nil {
+			if err := cgroups.WriteFile(path, "freezer.state", string(configs.Frozen)); err != nil {
 				return err
 			}
 
@@ -83,7 +81,7 @@ func (s *FreezerGroup) Set(path string, r *configs.Resources) (Err error) {
 				// system.
 				time.Sleep(10 * time.Microsecond)
 			}
-			state, err := fscommon.ReadFile(path, "freezer.state")
+			state, err := cgroups.ReadFile(path, "freezer.state")
 			if err != nil {
 				return err
 			}
@@ -104,7 +102,7 @@ func (s *FreezerGroup) Set(path string, r *configs.Resources) (Err error) {
 		// Despite our best efforts, it got stuck in FREEZING.
 		return errors.New("unable to freeze")
 	case configs.Thawed:
-		return fscommon.WriteFile(path, "freezer.state", string(configs.Thawed))
+		return cgroups.WriteFile(path, "freezer.state", string(configs.Thawed))
 	case configs.Undefined:
 		return nil
 	default:
@@ -118,7 +116,7 @@ func (s *FreezerGroup) GetStats(path string, stats *cgroups.Stats) error {
 
 func (s *FreezerGroup) GetState(path string) (configs.FreezerState, error) {
 	for {
-		state, err := fscommon.ReadFile(path, "freezer.state")
+		state, err := cgroups.ReadFile(path, "freezer.state")
 		if err != nil {
 			// If the kernel is too old, then we just treat the freezer as
 			// being in an "undefined" state.
@@ -131,7 +129,25 @@ func (s *FreezerGroup) GetState(path string) (configs.FreezerState, error) {
 		case "THAWED":
 			return configs.Thawed, nil
 		case "FROZEN":
+			// Find out whether the cgroup is frozen directly,
+			// or indirectly via an ancestor.
+			self, err := cgroups.ReadFile(path, "freezer.self_freezing")
+			if err != nil {
+				// If the kernel is too old, then we just treat
+				// it as being frozen.
+				if errors.Is(err, os.ErrNotExist) || errors.Is(err, unix.ENODEV) {
+					err = nil
+				}
+				return configs.Frozen, err
+			}
+			switch self {
+			case "0\n":
+				return configs.Thawed, nil
+			case "1\n":
 				return configs.Frozen, nil
+			default:
+				return configs.Undefined, fmt.Errorf(`unknown "freezer.self_freezing" state: %q`, self)
+			}
 		case "FREEZING":
 			// Make sure we get a stable freezer state, so retry if the cgroup
 			// is still undergoing freezing. This should be a temporary delay.

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/fs.go

@@ -64,8 +64,10 @@ func NewManager(cg *configs.Cgroup, paths map[string]string, rootless bool) cgro
 }
 
 // The absolute path to the root of the cgroup hierarchies.
-var cgroupRootLock sync.Mutex
+var (
-var cgroupRoot string
+	cgroupRootLock sync.Mutex
+	cgroupRoot     string
+)
 
 const defaultCgroupRoot = "/sys/fs/cgroup"
 
@@ -393,7 +395,7 @@ func join(path string, pid int) error {
 	if path == "" {
 		return nil
 	}
-	if err := os.MkdirAll(path, 0755); err != nil {
+	if err := os.MkdirAll(path, 0o755); err != nil {
 		return err
 	}
 	return cgroups.WriteCgroupProc(path, pid)

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/hugetlb.go

@@ -11,8 +11,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type HugetlbGroup struct {
+type HugetlbGroup struct{}
-}
 
 func (s *HugetlbGroup) Name() string {
 	return "hugetlb"
@@ -24,7 +23,7 @@ func (s *HugetlbGroup) Apply(path string, d *cgroupData) error {
 
 func (s *HugetlbGroup) Set(path string, r *configs.Resources) error {
 	for _, hugetlb := range r.HugetlbLimit {
-		if err := fscommon.WriteFile(path, "hugetlb."+hugetlb.Pagesize+".limit_in_bytes", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "hugetlb."+hugetlb.Pagesize+".limit_in_bytes", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/memory.go

@@ -25,8 +25,7 @@ const (
 	cgroupMemoryMaxUsage  = "memory.max_usage_in_bytes"
 )
 
-type MemoryGroup struct {
+type MemoryGroup struct{}
-}
 
 func (s *MemoryGroup) Name() string {
 	return "memory"
@@ -41,7 +40,7 @@ func setMemory(path string, val int64) error {
 		return nil
 	}
 
-	err := fscommon.WriteFile(path, cgroupMemoryLimit, strconv.FormatInt(val, 10))
+	err := cgroups.WriteFile(path, cgroupMemoryLimit, strconv.FormatInt(val, 10))
 	if !errors.Is(err, unix.EBUSY) {
 		return err
 	}
@@ -65,7 +64,7 @@ func setSwap(path string, val int64) error {
 		return nil
 	}
 
-	return fscommon.WriteFile(path, cgroupMemorySwapLimit, strconv.FormatInt(val, 10))
+	return cgroups.WriteFile(path, cgroupMemorySwapLimit, strconv.FormatInt(val, 10))
 }
 
 func setMemoryAndSwap(path string, r *configs.Resources) error {
@@ -118,20 +117,20 @@ func (s *MemoryGroup) Set(path string, r *configs.Resources) error {
 	// ignore KernelMemory and KernelMemoryTCP
 
 	if r.MemoryReservation != 0 {
-		if err := fscommon.WriteFile(path, "memory.soft_limit_in_bytes", strconv.FormatInt(r.MemoryReservation, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "memory.soft_limit_in_bytes", strconv.FormatInt(r.MemoryReservation, 10)); err != nil {
 			return err
 		}
 	}
 
 	if r.OomKillDisable {
-		if err := fscommon.WriteFile(path, "memory.oom_control", "1"); err != nil {
+		if err := cgroups.WriteFile(path, "memory.oom_control", "1"); err != nil {
 			return err
 		}
 	}
 	if r.MemorySwappiness == nil || int64(*r.MemorySwappiness) == -1 {
 		return nil
 	} else if *r.MemorySwappiness <= 100 {
-		if err := fscommon.WriteFile(path, "memory.swappiness", strconv.FormatUint(*r.MemorySwappiness, 10)); err != nil {
+		if err := cgroups.WriteFile(path, "memory.swappiness", strconv.FormatUint(*r.MemorySwappiness, 10)); err != nil {
 			return err
 		}
 	} else {
@@ -143,7 +142,7 @@ func (s *MemoryGroup) Set(path string, r *configs.Resources) error {
 
 func (s *MemoryGroup) GetStats(path string, stats *cgroups.Stats) error {
 	// Set stats from memory.stat.
-	statsFile, err := fscommon.OpenFile(path, "memory.stat", os.O_RDONLY)
+	statsFile, err := cgroups.OpenFile(path, "memory.stat", os.O_RDONLY)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return nil
@@ -200,14 +199,6 @@ func (s *MemoryGroup) GetStats(path string, stats *cgroups.Stats) error {
 	return nil
 }
 
-func memoryAssigned(cgroup *configs.Cgroup) bool {
-	return cgroup.Resources.Memory != 0 ||
-		cgroup.Resources.MemoryReservation != 0 ||
-		cgroup.Resources.MemorySwap > 0 ||
-		cgroup.Resources.OomKillDisable ||
-		(cgroup.Resources.MemorySwappiness != nil && int64(*cgroup.Resources.MemorySwappiness) != -1)
-}
-
 func getMemoryData(path, name string) (cgroups.MemoryData, error) {
 	memoryData := cgroups.MemoryData{}
 
@@ -258,12 +249,13 @@ func getPageUsageByNUMA(cgroupPath string) (cgroups.PageUsageByNUMA, error) {
 	)
 	stats := cgroups.PageUsageByNUMA{}
 
-	file, err := fscommon.OpenFile(cgroupPath, filename, os.O_RDONLY)
+	file, err := cgroups.OpenFile(cgroupPath, filename, os.O_RDONLY)
 	if os.IsNotExist(err) {
 		return stats, nil
 	} else if err != nil {
 		return stats, err
 	}
+	defer file.Close()
 
 	// File format is documented in linux/Documentation/cgroup-v1/memory.txt
 	// and it looks like this:

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/net_cls.go

@@ -6,12 +6,10 @@ import (
 	"strconv"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type NetClsGroup struct {
+type NetClsGroup struct{}
-}
 
 func (s *NetClsGroup) Name() string {
 	return "net_cls"
@@ -23,7 +21,7 @@ func (s *NetClsGroup) Apply(path string, d *cgroupData) error {
 
 func (s *NetClsGroup) Set(path string, r *configs.Resources) error {
 	if r.NetClsClassid != 0 {
-		if err := fscommon.WriteFile(path, "net_cls.classid", strconv.FormatUint(uint64(r.NetClsClassid), 10)); err != nil {
+		if err := cgroups.WriteFile(path, "net_cls.classid", strconv.FormatUint(uint64(r.NetClsClassid), 10)); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/net_prio.go

@@ -4,12 +4,10 @@ package fs
 
 import (
 	"github.com/opencontainers/runc/libcontainer/cgroups"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type NetPrioGroup struct {
+type NetPrioGroup struct{}
-}
 
 func (s *NetPrioGroup) Name() string {
 	return "net_prio"
@@ -21,7 +19,7 @@ func (s *NetPrioGroup) Apply(path string, d *cgroupData) error {
 
 func (s *NetPrioGroup) Set(path string, r *configs.Resources) error {
 	for _, prioMap := range r.NetPrioIfpriomap {
-		if err := fscommon.WriteFile(path, "net_prio.ifpriomap", prioMap.CgroupString()); err != nil {
+		if err := cgroups.WriteFile(path, "net_prio.ifpriomap", prioMap.CgroupString()); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/perf_event.go

@@ -7,8 +7,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type PerfEventGroup struct {
+type PerfEventGroup struct{}
-}
 
 func (s *PerfEventGroup) Name() string {
 	return "perf_event"

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/pids.go

@@ -12,8 +12,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-type PidsGroup struct {
+type PidsGroup struct{}
-}
 
 func (s *PidsGroup) Name() string {
 	return "pids"
@@ -32,7 +31,7 @@ func (s *PidsGroup) Set(path string, r *configs.Resources) error {
 			limit = strconv.FormatInt(r.PidsLimit, 10)
 		}
 
-		if err := fscommon.WriteFile(path, "pids.max", limit); err != nil {
+		if err := cgroups.WriteFile(path, "pids.max", limit); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/cpu.go

@@ -23,7 +23,7 @@ func setCpu(dirPath string, r *configs.Resources) error {
 
 	// NOTE: .CpuShares is not used here. Conversion is the caller's responsibility.
 	if r.CpuWeight != 0 {
-		if err := fscommon.WriteFile(dirPath, "cpu.weight", strconv.FormatUint(r.CpuWeight, 10)); err != nil {
+		if err := cgroups.WriteFile(dirPath, "cpu.weight", strconv.FormatUint(r.CpuWeight, 10)); err != nil {
 			return err
 		}
 	}
@@ -40,15 +40,16 @@ func setCpu(dirPath string, r *configs.Resources) error {
 			period = 100000
 		}
 		str += " " + strconv.FormatUint(period, 10)
-		if err := fscommon.WriteFile(dirPath, "cpu.max", str); err != nil {
+		if err := cgroups.WriteFile(dirPath, "cpu.max", str); err != nil {
 			return err
 		}
 	}
 
 	return nil
 }
+
 func statCpu(dirPath string, stats *cgroups.Stats) error {
-	f, err := fscommon.OpenFile(dirPath, "cpu.stat", os.O_RDONLY)
+	f, err := cgroups.OpenFile(dirPath, "cpu.stat", os.O_RDONLY)
 	if err != nil {
 		return err
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/cpuset.go

@@ -3,7 +3,7 @@
 package fs2
 
 import (
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
+	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
@@ -17,12 +17,12 @@ func setCpuset(dirPath string, r *configs.Resources) error {
 	}
 
 	if r.CpusetCpus != "" {
-		if err := fscommon.WriteFile(dirPath, "cpuset.cpus", r.CpusetCpus); err != nil {
+		if err := cgroups.WriteFile(dirPath, "cpuset.cpus", r.CpusetCpus); err != nil {
 			return err
 		}
 	}
 	if r.CpusetMems != "" {
-		if err := fscommon.WriteFile(dirPath, "cpuset.mems", r.CpusetMems); err != nil {
+		if err := cgroups.WriteFile(dirPath, "cpuset.mems", r.CpusetMems); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/create.go

@@ -6,12 +6,12 @@ import (
 	"path/filepath"
 	"strings"
 
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
+	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
 func supportedControllers() (string, error) {
-	return fscommon.ReadFile(UnifiedMountpoint, "/cgroup.controllers")
+	return cgroups.ReadFile(UnifiedMountpoint, "/cgroup.controllers")
 }
 
 // needAnyControllers returns whether we enable some supported controllers or not,
@@ -92,7 +92,7 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 	for i, e := range elements {
 		current = filepath.Join(current, e)
 		if i > 0 {
-			if err := os.Mkdir(current, 0755); err != nil {
+			if err := os.Mkdir(current, 0o755); err != nil {
 				if !os.IsExist(err) {
 					return err
 				}
@@ -105,7 +105,7 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 					}
 				}()
 			}
-			cgType, _ := fscommon.ReadFile(current, cgTypeFile)
+			cgType, _ := cgroups.ReadFile(current, cgTypeFile)
 			cgType = strings.TrimSpace(cgType)
 			switch cgType {
 			// If the cgroup is in an invalid mode (usually this means there's an internal
@@ -122,7 +122,7 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 					// since that means we're a properly delegated cgroup subtree) but in
 					// this case there's not much we can do and it's better than giving an
 					// error.
-					_ = fscommon.WriteFile(current, cgTypeFile, "threaded")
+					_ = cgroups.WriteFile(current, cgTypeFile, "threaded")
 				}
 			// If the cgroup is in (threaded) or (domain threaded) mode, we can only use thread-aware controllers
 			// (and you cannot usually take a cgroup out of threaded mode).
@@ -136,11 +136,11 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 		}
 		// enable all supported controllers
 		if i < len(elements)-1 {
-			if err := fscommon.WriteFile(current, cgStCtlFile, res); err != nil {
+			if err := cgroups.WriteFile(current, cgStCtlFile, res); err != nil {
 				// try write one by one
 				allCtrs := strings.Split(res, " ")
 				for _, ctr := range allCtrs {
-					_ = fscommon.WriteFile(current, cgStCtlFile, ctr)
+					_ = cgroups.WriteFile(current, cgStCtlFile, ctr)
 				}
 			}
 			// Some controllers might not be enabled when rootless or containerized,

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/defaultpath.go

@@ -82,9 +82,7 @@ func parseCgroupFile(path string) (string, error) {
 }
 
 func parseCgroupFromReader(r io.Reader) (string, error) {
-	var (
+	s := bufio.NewScanner(r)
-		s = bufio.NewScanner(r)
-	)
 	for s.Scan() {
 		var (
 			text  = s.Text()

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/devices.go

@@ -58,29 +58,15 @@ func setDevices(dirPath string, r *configs.Resources) error {
 	if r.SkipDevices {
 		return nil
 	}
-	// XXX: This is currently a white-list (but all callers pass a blacklist of
-	//      devices). This is bad for a whole variety of reasons, but will need
-	//      to be fixed with co-ordinated effort with downstreams.
 	insts, license, err := devicefilter.DeviceFilter(r.Devices)
 	if err != nil {
 		return err
 	}
-	dirFD, err := unix.Open(dirPath, unix.O_DIRECTORY|unix.O_RDONLY, 0600)
+	dirFD, err := unix.Open(dirPath, unix.O_DIRECTORY|unix.O_RDONLY, 0o600)
 	if err != nil {
 		return errors.Errorf("cannot get dir FD for %s", dirPath)
 	}
 	defer unix.Close(dirFD)
-	// XXX: This code is currently incorrect when it comes to updating an
-	//      existing cgroup with new rules (new rulesets are just appended to
-	//      the program list because this uses BPF_F_ALLOW_MULTI). If we didn't
-	//      use BPF_F_ALLOW_MULTI we could actually atomically swap the
-	//      programs.
-	//
-	//      The real issue is that BPF_F_ALLOW_MULTI makes it hard to have a
-	//      race-free blacklist because it acts as a whitelist by default, and
-	//      having a deny-everything program cannot be overridden by other
-	//      programs. You could temporarily insert a deny-everything program
-	//      but that would result in spurrious failures during updates.
 	if _, err := ebpf.LoadAttachCgroupDeviceFilter(insts, license, dirFD); err != nil {
 		if !canSkipEBPFError(r) {
 			return err

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/freezer.go

@@ -3,27 +3,20 @@
 package fs2
 
 import (
+	"bufio"
 	stdErrors "errors"
+	"fmt"
 	"os"
 	"strings"
+	"time"
 
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
+	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/pkg/errors"
 	"golang.org/x/sys/unix"
 )
 
 func setFreezer(dirPath string, state configs.FreezerState) error {
-	if err := supportsFreezer(dirPath); err != nil {
-		// We can ignore this request as long as the user didn't ask us to
-		// freeze the container (since without the freezer cgroup, that's a
-		// no-op).
-		if state == configs.Undefined || state == configs.Thawed {
-			return nil
-		}
-		return errors.Wrap(err, "freezer not supported")
-	}
-
 	var stateStr string
 	switch state {
 	case configs.Undefined:
@@ -36,11 +29,23 @@ func setFreezer(dirPath string, state configs.FreezerState) error {
 		return errors.Errorf("invalid freezer state %q requested", state)
 	}
 
-	if err := fscommon.WriteFile(dirPath, "cgroup.freeze", stateStr); err != nil {
+	fd, err := cgroups.OpenFile(dirPath, "cgroup.freeze", unix.O_RDWR)
+	if err != nil {
+		// We can ignore this request as long as the user didn't ask us to
+		// freeze the container (since without the freezer cgroup, that's a
+		// no-op).
+		if state != configs.Frozen {
+			return nil
+		}
+		return errors.Wrap(err, "freezer not supported")
+	}
+	defer fd.Close()
+
+	if _, err := fd.WriteString(stateStr); err != nil {
 		return err
 	}
 	// Confirm that the cgroup did actually change states.
-	if actualState, err := getFreezer(dirPath); err != nil {
+	if actualState, err := readFreezer(dirPath, fd); err != nil {
 		return err
 	} else if actualState != state {
 		return errors.Errorf(`expected "cgroup.freeze" to be in state %q but was in %q`, state, actualState)
@@ -48,13 +53,8 @@ func setFreezer(dirPath string, state configs.FreezerState) error {
 	return nil
 }
 
-func supportsFreezer(dirPath string) error {
-	_, err := fscommon.ReadFile(dirPath, "cgroup.freeze")
-	return err
-}
-
 func getFreezer(dirPath string) (configs.FreezerState, error) {
-	state, err := fscommon.ReadFile(dirPath, "cgroup.freeze")
+	fd, err := cgroups.OpenFile(dirPath, "cgroup.freeze", unix.O_RDONLY)
 	if err != nil {
 		// If the kernel is too old, then we just treat the freezer as being in
 		// an "undefined" state.
@@ -63,12 +63,67 @@ func getFreezer(dirPath string) (configs.FreezerState, error) {
 		}
 		return configs.Undefined, err
 	}
-	switch strings.TrimSpace(state) {
+	defer fd.Close()
-	case "0":
+
+	return readFreezer(dirPath, fd)
+}
+
+func readFreezer(dirPath string, fd *os.File) (configs.FreezerState, error) {
+	if _, err := fd.Seek(0, 0); err != nil {
+		return configs.Undefined, err
+	}
+	state := make([]byte, 2)
+	if _, err := fd.Read(state); err != nil {
+		return configs.Undefined, err
+	}
+	switch string(state) {
+	case "0\n":
 		return configs.Thawed, nil
-	case "1":
+	case "1\n":
-		return configs.Frozen, nil
+		return waitFrozen(dirPath)
 	default:
 		return configs.Undefined, errors.Errorf(`unknown "cgroup.freeze" state: %q`, state)
 	}
 }
+
+// waitFrozen polls cgroup.events until it sees "frozen 1" in it.
+func waitFrozen(dirPath string) (configs.FreezerState, error) {
+	fd, err := cgroups.OpenFile(dirPath, "cgroup.events", unix.O_RDONLY)
+	if err != nil {
+		return configs.Undefined, err
+	}
+	defer fd.Close()
+
+	// XXX: Simple wait/read/retry is used here. An implementation
+	// based on poll(2) or inotify(7) is possible, but it makes the code
+	// much more complicated. Maybe address this later.
+	const (
+		// Perform maxIter with waitTime in between iterations.
+		waitTime = 10 * time.Millisecond
+		maxIter  = 1000
+	)
+	scanner := bufio.NewScanner(fd)
+	for i := 0; scanner.Scan(); {
+		if i == maxIter {
+			return configs.Undefined, fmt.Errorf("timeout of %s reached waiting for the cgroup to freeze", waitTime*maxIter)
+		}
+		line := scanner.Text()
+		val := strings.TrimPrefix(line, "frozen ")
+		if val != line { // got prefix
+			if val[0] == '1' {
+				return configs.Frozen, nil
+			}
+
+			i++
+			// wait, then re-read
+			time.Sleep(waitTime)
+			_, err := fd.Seek(0, 0)
+			if err != nil {
+				return configs.Undefined, err
+			}
+		}
+	}
+	// Should only reach here either on read error,
+	// or if the file does not contain "frozen " line.
+	return configs.Undefined, scanner.Err()
+}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/fs2.go

@@ -51,7 +51,7 @@ func (m *manager) getControllers() error {
 		return nil
 	}
 
-	data, err := fscommon.ReadFile(m.dirPath, "cgroup.controllers")
+	data, err := cgroups.ReadFile(m.dirPath, "cgroup.controllers")
 	if err != nil {
 		if m.rootless && m.config.Path == "" {
 			return nil
@@ -98,9 +98,7 @@ func (m *manager) GetAllPids() ([]int, error) {
 }
 
 func (m *manager) GetStats() (*cgroups.Stats, error) {
-	var (
+	var errs []error
-		errs []error
-	)
 
 	st := cgroups.NewStats()
 
@@ -199,7 +197,7 @@ func (m *manager) setUnified(res map[string]string) error {
 		if strings.Contains(k, "/") {
 			return fmt.Errorf("unified resource %q must be a file name (no slashes)", k)
 		}
-		if err := fscommon.WriteFile(m.dirPath, k, v); err != nil {
+		if err := cgroups.WriteFile(m.dirPath, k, v); err != nil {
 			errC := errors.Cause(err)
 			// Check for both EPERM and ENOENT since O_CREAT is used by WriteFile.
 			if errors.Is(errC, os.ErrPermission) || errors.Is(errC, os.ErrNotExist) {

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/hugetlb.go

@@ -21,7 +21,7 @@ func setHugeTlb(dirPath string, r *configs.Resources) error {
 		return nil
 	}
 	for _, hugetlb := range r.HugetlbLimit {
-		if err := fscommon.WriteFile(dirPath, "hugetlb."+hugetlb.Pagesize+".max", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
+		if err := cgroups.WriteFile(dirPath, "hugetlb."+hugetlb.Pagesize+".max", strconv.FormatUint(hugetlb.Limit, 10)); err != nil {
 			return err
 		}
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/io.go

@@ -4,60 +4,95 @@ package fs2
 
 import (
 	"bufio"
+	"bytes"
+	"fmt"
 	"os"
 	"strconv"
 	"strings"
 
+	"github.com/sirupsen/logrus"
+
 	"github.com/opencontainers/runc/libcontainer/cgroups"
-	"github.com/opencontainers/runc/libcontainer/cgroups/fscommon"
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
 func isIoSet(r *configs.Resources) bool {
 	return r.BlkioWeight != 0 ||
+		len(r.BlkioWeightDevice) > 0 ||
 		len(r.BlkioThrottleReadBpsDevice) > 0 ||
 		len(r.BlkioThrottleWriteBpsDevice) > 0 ||
 		len(r.BlkioThrottleReadIOPSDevice) > 0 ||
 		len(r.BlkioThrottleWriteIOPSDevice) > 0
 }
 
+// bfqDeviceWeightSupported checks for per-device BFQ weight support (added
+// in kernel v5.4, commit 795fe54c2a8) by reading from "io.bfq.weight".
+func bfqDeviceWeightSupported(bfq *os.File) bool {
+	if bfq == nil {
+		return false
+	}
+	_, _ = bfq.Seek(0, 0)
+	buf := make([]byte, 32)
+	_, _ = bfq.Read(buf)
+	// If only a single number (default weight) if read back, we have older kernel.
+	_, err := strconv.ParseInt(string(bytes.TrimSpace(buf)), 10, 64)
+	return err != nil
+}
+
 func setIo(dirPath string, r *configs.Resources) error {
 	if !isIoSet(r) {
 		return nil
 	}
 
-	if r.BlkioWeight != 0 {
+	// If BFQ IO scheduler is available, use it.
-		filename := "io.bfq.weight"
+	var bfq *os.File
-		if err := fscommon.WriteFile(dirPath, filename,
+	if r.BlkioWeight != 0 || len(r.BlkioWeightDevice) > 0 {
-			strconv.FormatUint(uint64(r.BlkioWeight), 10)); err != nil {
+		var err error
-			// if io.bfq.weight does not exist, then bfq module is not loaded.
+		bfq, err = cgroups.OpenFile(dirPath, "io.bfq.weight", os.O_RDWR)
-			// Fallback to use io.weight with a conversion scheme
+		if err == nil {
-			if !os.IsNotExist(err) {
+			defer bfq.Close()
+		} else if !os.IsNotExist(err) {
 			return err
 		}
-			v := cgroups.ConvertBlkIOToIOWeightValue(r.BlkioWeight)
+	}
-			if err := fscommon.WriteFile(dirPath, "io.weight", strconv.FormatUint(v, 10)); err != nil {
+
+	if r.BlkioWeight != 0 {
+		if bfq != nil { // Use BFQ.
+			if _, err := bfq.WriteString(strconv.FormatUint(uint64(r.BlkioWeight), 10)); err != nil {
 				return err
 			}
+		} else {
+			// Fallback to io.weight with a conversion scheme.
+			v := cgroups.ConvertBlkIOToIOWeightValue(r.BlkioWeight)
+			if err := cgroups.WriteFile(dirPath, "io.weight", strconv.FormatUint(v, 10)); err != nil {
+				return err
+			}
+		}
+	}
+	if bfqDeviceWeightSupported(bfq) {
+		for _, wd := range r.BlkioWeightDevice {
+			if _, err := bfq.WriteString(wd.WeightString() + "\n"); err != nil {
+				return fmt.Errorf("setting device weight %q: %w", wd.WeightString(), err)
+			}
 		}
 	}
 	for _, td := range r.BlkioThrottleReadBpsDevice {
-		if err := fscommon.WriteFile(dirPath, "io.max", td.StringName("rbps")); err != nil {
+		if err := cgroups.WriteFile(dirPath, "io.max", td.StringName("rbps")); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleWriteBpsDevice {
-		if err := fscommon.WriteFile(dirPath, "io.max", td.StringName("wbps")); err != nil {
+		if err := cgroups.WriteFile(dirPath, "io.max", td.StringName("wbps")); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleReadIOPSDevice {
-		if err := fscommon.WriteFile(dirPath, "io.max", td.StringName("riops")); err != nil {
+		if err := cgroups.WriteFile(dirPath, "io.max", td.StringName("riops")); err != nil {
 			return err
 		}
 	}
 	for _, td := range r.BlkioThrottleWriteIOPSDevice {
-		if err := fscommon.WriteFile(dirPath, "io.max", td.StringName("wiops")); err != nil {
+		if err := cgroups.WriteFile(dirPath, "io.max", td.StringName("wiops")); err != nil {
 			return err
 		}
 	}
@@ -67,7 +102,7 @@ func setIo(dirPath string, r *configs.Resources) error {
 
 func readCgroup2MapFile(dirPath string, name string) (map[string][]string, error) {
 	ret := map[string][]string{}
-	f, err := fscommon.OpenFile(dirPath, name, os.O_RDONLY)
+	f, err := cgroups.OpenFile(dirPath, name, os.O_RDONLY)
 	if err != nil {
 		return nil, err
 	}
@@ -88,22 +123,22 @@ func readCgroup2MapFile(dirPath string, name string) (map[string][]string, error
 }
 
 func statIo(dirPath string, stats *cgroups.Stats) error {
-	// more details on the io.stat file format: https://www.kernel.org/doc/Documentation/cgroup-v2.txt
-	var ioServiceBytesRecursive []cgroups.BlkioStatEntry
 	values, err := readCgroup2MapFile(dirPath, "io.stat")
 	if err != nil {
 		return err
 	}
+	// more details on the io.stat file format: https://www.kernel.org/doc/Documentation/cgroup-v2.txt
+	var parsedStats cgroups.BlkioStats
 	for k, v := range values {
 		d := strings.Split(k, ":")
 		if len(d) != 2 {
 			continue
 		}
-		major, err := strconv.ParseUint(d[0], 10, 0)
+		major, err := strconv.ParseUint(d[0], 10, 64)
 		if err != nil {
 			return err
 		}
-		minor, err := strconv.ParseUint(d[1], 10, 0)
+		minor, err := strconv.ParseUint(d[1], 10, 64)
 		if err != nil {
 			return err
 		}
@@ -115,15 +150,32 @@ func statIo(dirPath string, stats *cgroups.Stats) error {
 			}
 			op := d[0]
 
-			// Accommodate the cgroup v1 naming
+			// Map to the cgroupv1 naming and layout (in separate tables).
+			var targetTable *[]cgroups.BlkioStatEntry
 			switch op {
+			// Equivalent to cgroupv1's blkio.io_service_bytes.
 			case "rbytes":
-				op = "read"
+				op = "Read"
+				targetTable = &parsedStats.IoServiceBytesRecursive
 			case "wbytes":
-				op = "write"
+				op = "Write"
+				targetTable = &parsedStats.IoServiceBytesRecursive
+			// Equivalent to cgroupv1's blkio.io_serviced.
+			case "rios":
+				op = "Read"
+				targetTable = &parsedStats.IoServicedRecursive
+			case "wios":
+				op = "Write"
+				targetTable = &parsedStats.IoServicedRecursive
+			default:
+				// Skip over entries we cannot map to cgroupv1 stats for now.
+				// In the future we should expand the stats struct to include
+				// them.
+				logrus.Debugf("cgroupv2 io stats: skipping over unmappable %s entry", item)
+				continue
 			}
 
-			value, err := strconv.ParseUint(d[1], 10, 0)
+			value, err := strconv.ParseUint(d[1], 10, 64)
 			if err != nil {
 				return err
 			}
@@ -134,9 +186,9 @@ func statIo(dirPath string, stats *cgroups.Stats) error {
 				Minor: minor,
 				Value: value,
 			}
-			ioServiceBytesRecursive = append(ioServiceBytesRecursive, entry)
+			*targetTable = append(*targetTable, entry)
 		}
 	}
-	stats.BlkioStats = cgroups.BlkioStats{IoServiceBytesRecursive: ioServiceBytesRecursive}
+	stats.BlkioStats = parsedStats
 	return nil
 }

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/memory.go

@@ -52,13 +52,13 @@ func setMemory(dirPath string, r *configs.Resources) error {
 	}
 	// never write empty string to `memory.swap.max`, it means set to 0.
 	if swapStr != "" {
-		if err := fscommon.WriteFile(dirPath, "memory.swap.max", swapStr); err != nil {
+		if err := cgroups.WriteFile(dirPath, "memory.swap.max", swapStr); err != nil {
 			return err
 		}
 	}
 
 	if val := numToStr(r.Memory); val != "" {
-		if err := fscommon.WriteFile(dirPath, "memory.max", val); err != nil {
+		if err := cgroups.WriteFile(dirPath, "memory.max", val); err != nil {
 			return err
 		}
 	}
@@ -66,7 +66,7 @@ func setMemory(dirPath string, r *configs.Resources) error {
 	// cgroup.Resources.KernelMemory is ignored
 
 	if val := numToStr(r.MemoryReservation); val != "" {
-		if err := fscommon.WriteFile(dirPath, "memory.low", val); err != nil {
+		if err := cgroups.WriteFile(dirPath, "memory.low", val); err != nil {
 			return err
 		}
 	}
@@ -76,7 +76,7 @@ func setMemory(dirPath string, r *configs.Resources) error {
 
 func statMemory(dirPath string, stats *cgroups.Stats) error {
 	// Set stats from memory.stat.
-	statsFile, err := fscommon.OpenFile(dirPath, "memory.stat", os.O_RDONLY)
+	statsFile, err := cgroups.OpenFile(dirPath, "memory.stat", os.O_RDONLY)
 	if err != nil {
 		return err
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/pids.go

@@ -23,7 +23,7 @@ func setPids(dirPath string, r *configs.Resources) error {
 		return nil
 	}
 	if val := numToStr(r.PidsLimit); val != "" {
-		if err := fscommon.WriteFile(dirPath, "pids.max", val); err != nil {
+		if err := cgroups.WriteFile(dirPath, "pids.max", val); err != nil {
 			return err
 		}
 	}
@@ -34,9 +34,9 @@ func setPids(dirPath string, r *configs.Resources) error {
 func statPidsFromCgroupProcs(dirPath string, stats *cgroups.Stats) error {
 	// if the controller is not enabled, let's read PIDS from cgroups.procs
 	// (or threads if cgroup.threads is enabled)
-	contents, err := fscommon.ReadFile(dirPath, "cgroup.procs")
+	contents, err := cgroups.ReadFile(dirPath, "cgroup.procs")
 	if errors.Is(err, unix.ENOTSUP) {
-		contents, err = fscommon.ReadFile(dirPath, "cgroup.threads")
+		contents, err = cgroups.ReadFile(dirPath, "cgroup.threads")
 	}
 	if err != nil {
 		return err

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/fscommon.go

@@ -1,51 +0,0 @@
-// +build linux
-
-package fscommon
-
-import (
-	"bytes"
-	"os"
-
-	"github.com/pkg/errors"
-	"github.com/sirupsen/logrus"
-	"golang.org/x/sys/unix"
-)
-
-// WriteFile writes data to a cgroup file in dir.
-// It is supposed to be used for cgroup files only.
-func WriteFile(dir, file, data string) error {
-	fd, err := OpenFile(dir, file, unix.O_WRONLY)
-	if err != nil {
-		return err
-	}
-	defer fd.Close()
-	if err := retryingWriteFile(fd, data); err != nil {
-		return errors.Wrapf(err, "failed to write %q", data)
-	}
-	return nil
-}
-
-// ReadFile reads data from a cgroup file in dir.
-// It is supposed to be used for cgroup files only.
-func ReadFile(dir, file string) (string, error) {
-	fd, err := OpenFile(dir, file, unix.O_RDONLY)
-	if err != nil {
-		return "", err
-	}
-	defer fd.Close()
-	var buf bytes.Buffer
-
-	_, err = buf.ReadFrom(fd)
-	return buf.String(), err
-}
-
-func retryingWriteFile(fd *os.File, data string) error {
-	for {
-		_, err := fd.Write([]byte(data))
-		if errors.Is(err, unix.EINTR) {
-			logrus.Infof("interrupted while writing %s to %s", data, fd.Name())
-			continue
-		}
-		return err
-	}
-}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fscommon/utils.go

@@ -8,10 +8,19 @@ import (
 	"math"
 	"strconv"
 	"strings"
+
+	"github.com/opencontainers/runc/libcontainer/cgroups"
 )
 
 var (
 	ErrNotValidFormat = errors.New("line is not a valid key value format")
+
+	// Deprecated: use cgroups.OpenFile instead.
+	OpenFile = cgroups.OpenFile
+	// Deprecated: use cgroups.ReadFile instead.
+	ReadFile = cgroups.ReadFile
+	// Deprecated: use cgroups.WriteFile instead.
+	WriteFile = cgroups.WriteFile
 )
 
 // ParseUint converts a string to an uint64 integer.
@@ -57,7 +66,7 @@ func ParseKeyValue(t string) (string, uint64, error) {
 // and returns a value of the specified key. ParseUint is used for value
 // conversion.
 func GetValueByKey(path, file, key string) (uint64, error) {
-	content, err := ReadFile(path, file)
+	content, err := cgroups.ReadFile(path, file)
 	if err != nil {
 		return 0, err
 	}
@@ -95,7 +104,7 @@ func GetCgroupParamUint(path, file string) (uint64, error) {
 // GetCgroupParamInt reads a single int64 value from specified cgroup file.
 // If the value read is "max", the math.MaxInt64 is returned.
 func GetCgroupParamInt(path, file string) (int64, error) {
-	contents, err := ReadFile(path, file)
+	contents, err := cgroups.ReadFile(path, file)
 	if err != nil {
 		return 0, err
 	}
@@ -113,7 +122,7 @@ func GetCgroupParamInt(path, file string) (int64, error) {
 
 // GetCgroupParamString reads a string from the specified cgroup file.
 func GetCgroupParamString(path, file string) (string, error) {
-	contents, err := ReadFile(path, file)
+	contents, err := cgroups.ReadFile(path, file)
 	if err != nil {
 		return "", err
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/systemd/common.go

@@ -158,14 +158,27 @@ func findDeviceGroup(ruleType devices.Type, ruleMajor int64) (string, error) {
 	return "", nil
 }
 
-// generateDeviceProperties takes the configured device rules and generates a
+// DeviceAllow is the dbus type "a(ss)" which means we need a struct
-// corresponding set of systemd properties to configure the devices correctly.
+// to represent it in Go.
-func generateDeviceProperties(rules []*devices.Rule) ([]systemdDbus.Property, error) {
+type deviceAllowEntry struct {
-	// DeviceAllow is the type "a(ss)" which means we need a temporary struct
-	// to represent it in Go.
-	type deviceAllowEntry struct {
 	Path  string
 	Perms string
+}
+
+func allowAllDevices() []systemdDbus.Property {
+	// Setting mode to auto and removing all DeviceAllow rules
+	// results in allowing access to all devices.
+	return []systemdDbus.Property{
+		newProp("DevicePolicy", "auto"),
+		newProp("DeviceAllow", []deviceAllowEntry{}),
+	}
+}
+
+// generateDeviceProperties takes the configured device rules and generates a
+// corresponding set of systemd properties to configure the devices correctly.
+func generateDeviceProperties(r *configs.Resources) ([]systemdDbus.Property, error) {
+	if r.SkipDevices {
+		return nil, nil
 	}
 
 	properties := []systemdDbus.Property{
@@ -177,7 +190,7 @@ func generateDeviceProperties(rules []*devices.Rule) ([]systemdDbus.Property, er
 
 	// Figure out the set of rules.
 	configEmu := &cgroupdevices.Emulator{}
-	for _, rule := range rules {
+	for _, rule := range r.Devices {
 		if err := configEmu.Apply(*rule); err != nil {
 			return nil, errors.Wrap(err, "apply rule for systemd")
 		}
@@ -189,12 +202,7 @@ func generateDeviceProperties(rules []*devices.Rule) ([]systemdDbus.Property, er
 	if configEmu.IsBlacklist() {
 		// However, if we're dealing with an allow-all rule then we can do it.
 		if configEmu.IsAllowAll() {
-			return []systemdDbus.Property{
+			return allowAllDevices(), nil
-				// Run in white-list mode by setting to "auto" and removing all
-				// DeviceAllow rules.
-				newProp("DevicePolicy", "auto"),
-				newProp("DeviceAllow", []deviceAllowEntry{}),
-			}, nil
 		}
 		logrus.Warn("systemd doesn't support blacklist device rules -- applying temporary deny-all rule")
 		return properties, nil
@@ -203,8 +211,7 @@ func generateDeviceProperties(rules []*devices.Rule) ([]systemdDbus.Property, er
 	// Now generate the set of rules we actually need to apply. Unlike the
 	// normal devices cgroup, in "strict" mode systemd defaults to a deny-all
 	// whitelist which is the default for devices.Emulator.
-	baseEmu := &cgroupdevices.Emulator{}
+	finalRules, err := configEmu.Rules()
-	finalRules, err := baseEmu.Transition(configEmu)
 	if err != nil {
 		return nil, errors.Wrap(err, "get simplified rules for systemd")
 	}
@@ -306,7 +313,7 @@ func getUnitName(c *configs.Cgroup) string {
 // isDbusError returns true if the error is a specific dbus error.
 func isDbusError(err error, name string) bool {
 	if err != nil {
-		var derr *dbus.Error
+		var derr dbus.Error
 		if errors.As(err, &derr) {
 			return strings.Contains(derr.Name, name)
 		}
@@ -355,6 +362,9 @@ func stopUnit(cm *dbusConnManager, unitName string) error {
 		return err
 	})
 	if err == nil {
+		timeout := time.NewTimer(30 * time.Second)
+		defer timeout.Stop()
+
 		select {
 		case s := <-statusChan:
 			close(statusChan)
@@ -362,8 +372,8 @@ func stopUnit(cm *dbusConnManager, unitName string) error {
 			if s != "done" {
 				logrus.Warnf("error removing unit `%s`: got `%s`. Continuing...", unitName, s)
 			}
-		case <-time.After(time.Second):
+		case <-timeout.C:
-			logrus.Warnf("Timed out while waiting for StopUnit(%s) completion signal from dbus. Continuing...", unitName)
+			return errors.New("Timed out while waiting for systemd to remove " + unitName)
 		}
 	}
 	return nil
@@ -378,6 +388,15 @@ func resetFailedUnit(cm *dbusConnManager, name string) {
 	}
 }
 
+func getUnitProperty(cm *dbusConnManager, unitName string, propertyName string) (*systemdDbus.Property, error) {
+	var prop *systemdDbus.Property
+	err := cm.retryOnDisconnect(func(c *systemdDbus.Conn) (Err error) {
+		prop, Err = c.GetUnitPropertyContext(context.TODO(), unitName, propertyName)
+		return Err
+	})
+	return prop, err
+}
+
 func setUnitProperties(cm *dbusConnManager, name string, properties ...systemdDbus.Property) error {
 	return cm.retryOnDisconnect(func(c *systemdDbus.Conn) error {
 		return c.SetUnitPropertiesContext(context.TODO(), name, true, properties...)
@@ -476,7 +495,7 @@ func addCpuset(cm *dbusConnManager, props *[]systemdDbus.Property, cpus, mems st
 	}
 
 	if cpus != "" {
-		bits, err := rangeToBits(cpus)
+		bits, err := RangeToBits(cpus)
 		if err != nil {
 			return fmt.Errorf("resources.CPU.Cpus=%q conversion error: %w",
 				cpus, err)
@@ -485,7 +504,7 @@ func addCpuset(cm *dbusConnManager, props *[]systemdDbus.Property, cpus, mems st
 			newProp("AllowedCPUs", bits))
 	}
 	if mems != "" {
-		bits, err := rangeToBits(mems)
+		bits, err := RangeToBits(mems)
 		if err != nil {
 			return fmt.Errorf("resources.CPU.Mems=%q conversion error: %w",
 				mems, err)