summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/platform/ring0/pagetables
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/platform/ring0/pagetables')
-rw-r--r--pkg/sentry/platform/ring0/pagetables/BUILD112
-rwxr-xr-x[-rw-r--r--]pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go0
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go5
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go5
-rw-r--r--pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go75
-rwxr-xr-x[-rw-r--r--]pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go0
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go5
-rw-r--r--pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go80
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go5
-rw-r--r--pkg/sentry/platform/ring0/pagetables/pagetables_test.go156
-rwxr-xr-x[-rw-r--r--]pkg/sentry/platform/ring0/pagetables/pcids.go0
-rwxr-xr-x[-rw-r--r--]pkg/sentry/platform/ring0/pagetables/walker_amd64.go0
-rwxr-xr-x[-rw-r--r--]pkg/sentry/platform/ring0/pagetables/walker_arm64.go0
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/walker_empty.go255
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/walker_lookup.go255
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/walker_map.go255
-rwxr-xr-xpkg/sentry/platform/ring0/pagetables/walker_unmap.go255
17 files changed, 1040 insertions, 423 deletions
diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD
deleted file mode 100644
index 581841555..000000000
--- a/pkg/sentry/platform/ring0/pagetables/BUILD
+++ /dev/null
@@ -1,112 +0,0 @@
-load("//tools:defs.bzl", "go_library", "go_test", "select_arch")
-load("//tools/go_generics:defs.bzl", "go_template", "go_template_instance")
-
-package(licenses = ["notice"])
-
-go_template(
- name = "generic_walker",
- srcs = select_arch(
- amd64 = ["walker_amd64.go"],
- arm64 = ["walker_arm64.go"],
- ),
- opt_types = [
- "Visitor",
- ],
- visibility = [":__pkg__"],
-)
-
-go_template_instance(
- name = "walker_map",
- out = "walker_map.go",
- package = "pagetables",
- prefix = "map",
- template = ":generic_walker",
- types = {
- "Visitor": "mapVisitor",
- },
-)
-
-go_template_instance(
- name = "walker_unmap",
- out = "walker_unmap.go",
- package = "pagetables",
- prefix = "unmap",
- template = ":generic_walker",
- types = {
- "Visitor": "unmapVisitor",
- },
-)
-
-go_template_instance(
- name = "walker_lookup",
- out = "walker_lookup.go",
- package = "pagetables",
- prefix = "lookup",
- template = ":generic_walker",
- types = {
- "Visitor": "lookupVisitor",
- },
-)
-
-go_template_instance(
- name = "walker_empty",
- out = "walker_empty.go",
- package = "pagetables",
- prefix = "empty",
- template = ":generic_walker",
- types = {
- "Visitor": "emptyVisitor",
- },
-)
-
-go_template_instance(
- name = "walker_check",
- out = "walker_check.go",
- package = "pagetables",
- prefix = "check",
- template = ":generic_walker",
- types = {
- "Visitor": "checkVisitor",
- },
-)
-
-go_library(
- name = "pagetables",
- srcs = [
- "allocator.go",
- "allocator_unsafe.go",
- "pagetables.go",
- "pagetables_aarch64.go",
- "pagetables_amd64.go",
- "pagetables_arm64.go",
- "pagetables_x86.go",
- "pcids.go",
- "walker_amd64.go",
- "walker_arm64.go",
- "walker_empty.go",
- "walker_lookup.go",
- "walker_map.go",
- "walker_unmap.go",
- ],
- visibility = [
- "//pkg/sentry/platform/kvm:__subpackages__",
- "//pkg/sentry/platform/ring0:__subpackages__",
- ],
- deps = [
- "//pkg/sync",
- "//pkg/usermem",
- ],
-)
-
-go_test(
- name = "pagetables_test",
- size = "small",
- srcs = [
- "pagetables_amd64_test.go",
- "pagetables_arm64_test.go",
- "pagetables_test.go",
- "walker_check.go",
- ],
- library = ":pagetables",
- deps = ["//pkg/usermem"],
-)
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go
index 78510ebed..78510ebed 100644..100755
--- a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64.go
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go
new file mode 100755
index 000000000..ae9d2b272
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_aarch64_state_autogen.go
@@ -0,0 +1,5 @@
+// automatically generated by stateify.
+
+// +build arm64
+
+package pagetables
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go
new file mode 100755
index 000000000..f48a8acd1
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_state_autogen.go
@@ -0,0 +1,5 @@
+// automatically generated by stateify.
+
+// +build amd64
+
+package pagetables
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go
deleted file mode 100644
index 54e8e554f..000000000
--- a/pkg/sentry/platform/ring0/pagetables/pagetables_amd64_test.go
+++ /dev/null
@@ -1,75 +0,0 @@
-// Copyright 2018 The gVisor Authors.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-// +build amd64
-
-package pagetables
-
-import (
- "testing"
-
- "gvisor.dev/gvisor/pkg/usermem"
-)
-
-func Test2MAnd4K(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a small page and a huge page.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
- pt.Map(0x00007f0000000000, pmdSize, MapOpts{AccessType: usermem.Read}, pmdSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}},
- {0x00007f0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read}},
- })
-}
-
-func Test1GAnd4K(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a small page and a super page.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
- pt.Map(0x00007f0000000000, pudSize, MapOpts{AccessType: usermem.Read}, pudSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}},
- {0x00007f0000000000, pudSize, pudSize * 47, MapOpts{AccessType: usermem.Read}},
- })
-}
-
-func TestSplit1GPage(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a super page and knock out the middle.
- pt.Map(0x00007f0000000000, pudSize, MapOpts{AccessType: usermem.Read}, pudSize*42)
- pt.Unmap(usermem.Addr(0x00007f0000000000+pteSize), pudSize-(2*pteSize))
-
- checkMappings(t, pt, []mapping{
- {0x00007f0000000000, pteSize, pudSize * 42, MapOpts{AccessType: usermem.Read}},
- {0x00007f0000000000 + pudSize - pteSize, pteSize, pudSize*42 + pudSize - pteSize, MapOpts{AccessType: usermem.Read}},
- })
-}
-
-func TestSplit2MPage(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a huge page and knock out the middle.
- pt.Map(0x00007f0000000000, pmdSize, MapOpts{AccessType: usermem.Read}, pmdSize*42)
- pt.Unmap(usermem.Addr(0x00007f0000000000+pteSize), pmdSize-(2*pteSize))
-
- checkMappings(t, pt, []mapping{
- {0x00007f0000000000, pteSize, pmdSize * 42, MapOpts{AccessType: usermem.Read}},
- {0x00007f0000000000 + pmdSize - pteSize, pteSize, pmdSize*42 + pmdSize - pteSize, MapOpts{AccessType: usermem.Read}},
- })
-}
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go
index 1a49f12a2..1a49f12a2 100644..100755
--- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64.go
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go
new file mode 100755
index 000000000..ae9d2b272
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_state_autogen.go
@@ -0,0 +1,5 @@
+// automatically generated by stateify.
+
+// +build arm64
+
+package pagetables
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go
deleted file mode 100644
index 2f73d424f..000000000
--- a/pkg/sentry/platform/ring0/pagetables/pagetables_arm64_test.go
+++ /dev/null
@@ -1,80 +0,0 @@
-// Copyright 2019 The gVisor Authors.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-// +build arm64
-
-package pagetables
-
-import (
- "testing"
-
- "gvisor.dev/gvisor/pkg/usermem"
-)
-
-func Test2MAnd4K(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a small page and a huge page.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: true}, pteSize*42)
- pt.Map(0x0000ff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: true}, pmdSize*47)
-
- pt.Map(0xffff000000400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: false}, pteSize*42)
- pt.Map(0xffffff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: false}, pmdSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: true}},
- {0x0000ff0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read, User: true}},
- {0xffff000000400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: false}},
- {0xffffff0000000000, pmdSize, pmdSize * 47, MapOpts{AccessType: usermem.Read, User: false}},
- })
-}
-
-func Test1GAnd4K(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a small page and a super page.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite, User: true}, pteSize*42)
- pt.Map(0x0000ff0000000000, pudSize, MapOpts{AccessType: usermem.Read, User: true}, pudSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite, User: true}},
- {0x0000ff0000000000, pudSize, pudSize * 47, MapOpts{AccessType: usermem.Read, User: true}},
- })
-}
-
-func TestSplit1GPage(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a super page and knock out the middle.
- pt.Map(0x0000ff0000000000, pudSize, MapOpts{AccessType: usermem.Read, User: true}, pudSize*42)
- pt.Unmap(usermem.Addr(0x0000ff0000000000+pteSize), pudSize-(2*pteSize))
-
- checkMappings(t, pt, []mapping{
- {0x0000ff0000000000, pteSize, pudSize * 42, MapOpts{AccessType: usermem.Read, User: true}},
- {0x0000ff0000000000 + pudSize - pteSize, pteSize, pudSize*42 + pudSize - pteSize, MapOpts{AccessType: usermem.Read, User: true}},
- })
-}
-
-func TestSplit2MPage(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map a huge page and knock out the middle.
- pt.Map(0x0000ff0000000000, pmdSize, MapOpts{AccessType: usermem.Read, User: true}, pmdSize*42)
- pt.Unmap(usermem.Addr(0x0000ff0000000000+pteSize), pmdSize-(2*pteSize))
-
- checkMappings(t, pt, []mapping{
- {0x0000ff0000000000, pteSize, pmdSize * 42, MapOpts{AccessType: usermem.Read, User: true}},
- {0x0000ff0000000000 + pmdSize - pteSize, pteSize, pmdSize*42 + pmdSize - pteSize, MapOpts{AccessType: usermem.Read, User: true}},
- })
-}
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go b/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go
new file mode 100755
index 000000000..52bab66fe
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/pagetables_state_autogen.go
@@ -0,0 +1,5 @@
+// automatically generated by stateify.
+
+// +build i386 amd64
+
+package pagetables
diff --git a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go b/pkg/sentry/platform/ring0/pagetables/pagetables_test.go
deleted file mode 100644
index 5c88d087d..000000000
--- a/pkg/sentry/platform/ring0/pagetables/pagetables_test.go
+++ /dev/null
@@ -1,156 +0,0 @@
-// Copyright 2018 The gVisor Authors.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-// http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-package pagetables
-
-import (
- "testing"
-
- "gvisor.dev/gvisor/pkg/usermem"
-)
-
-type mapping struct {
- start uintptr
- length uintptr
- addr uintptr
- opts MapOpts
-}
-
-type checkVisitor struct {
- expected []mapping // Input.
- current int // Temporary.
- found []mapping // Output.
- failed string // Output.
-}
-
-func (v *checkVisitor) visit(start uintptr, pte *PTE, align uintptr) {
- v.found = append(v.found, mapping{
- start: start,
- length: align + 1,
- addr: pte.Address(),
- opts: pte.Opts(),
- })
- if v.failed != "" {
- // Don't keep looking for errors.
- return
- }
-
- if v.current >= len(v.expected) {
- v.failed = "more mappings than expected"
- } else if v.expected[v.current].start != start {
- v.failed = "start didn't match expected"
- } else if v.expected[v.current].length != (align + 1) {
- v.failed = "end didn't match expected"
- } else if v.expected[v.current].addr != pte.Address() {
- v.failed = "address didn't match expected"
- } else if v.expected[v.current].opts != pte.Opts() {
- v.failed = "opts didn't match"
- }
- v.current++
-}
-
-func (*checkVisitor) requiresAlloc() bool { return false }
-
-func (*checkVisitor) requiresSplit() bool { return false }
-
-func checkMappings(t *testing.T, pt *PageTables, m []mapping) {
- // Iterate over all the mappings.
- w := checkWalker{
- pageTables: pt,
- visitor: checkVisitor{
- expected: m,
- },
- }
- w.iterateRange(0, ^uintptr(0))
-
- // Were we expected additional mappings?
- if w.visitor.failed == "" && w.visitor.current != len(w.visitor.expected) {
- w.visitor.failed = "insufficient mappings found"
- }
-
- // Emit a meaningful error message on failure.
- if w.visitor.failed != "" {
- t.Errorf("%s; got %#v, wanted %#v", w.visitor.failed, w.visitor.found, w.visitor.expected)
- }
-}
-
-func TestUnmap(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map and unmap one entry.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
- pt.Unmap(0x400000, pteSize)
-
- checkMappings(t, pt, nil)
-}
-
-func TestReadOnly(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map one entry.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.Read}, pteSize*42)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.Read}},
- })
-}
-
-func TestReadWrite(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map one entry.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}},
- })
-}
-
-func TestSerialEntries(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map two sequential entries.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
- pt.Map(0x401000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}},
- {0x401000, pteSize, pteSize * 47, MapOpts{AccessType: usermem.ReadWrite}},
- })
-}
-
-func TestSpanningEntries(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Span a pgd with two pages.
- pt.Map(0x00007efffffff000, 2*pteSize, MapOpts{AccessType: usermem.Read}, pteSize*42)
-
- checkMappings(t, pt, []mapping{
- {0x00007efffffff000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.Read}},
- {0x00007f0000000000, pteSize, pteSize * 43, MapOpts{AccessType: usermem.Read}},
- })
-}
-
-func TestSparseEntries(t *testing.T) {
- pt := New(NewRuntimeAllocator())
-
- // Map two entries in different pgds.
- pt.Map(0x400000, pteSize, MapOpts{AccessType: usermem.ReadWrite}, pteSize*42)
- pt.Map(0x00007f0000000000, pteSize, MapOpts{AccessType: usermem.Read}, pteSize*47)
-
- checkMappings(t, pt, []mapping{
- {0x400000, pteSize, pteSize * 42, MapOpts{AccessType: usermem.ReadWrite}},
- {0x00007f0000000000, pteSize, pteSize * 47, MapOpts{AccessType: usermem.Read}},
- })
-}
diff --git a/pkg/sentry/platform/ring0/pagetables/pcids.go b/pkg/sentry/platform/ring0/pagetables/pcids.go
index 9206030bf..9206030bf 100644..100755
--- a/pkg/sentry/platform/ring0/pagetables/pcids.go
+++ b/pkg/sentry/platform/ring0/pagetables/pcids.go
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_amd64.go b/pkg/sentry/platform/ring0/pagetables/walker_amd64.go
index 8f9dacd93..8f9dacd93 100644..100755
--- a/pkg/sentry/platform/ring0/pagetables/walker_amd64.go
+++ b/pkg/sentry/platform/ring0/pagetables/walker_amd64.go
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go
index c261d393a..c261d393a 100644..100755
--- a/pkg/sentry/platform/ring0/pagetables/walker_arm64.go
+++ b/pkg/sentry/platform/ring0/pagetables/walker_arm64.go
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_empty.go b/pkg/sentry/platform/ring0/pagetables/walker_empty.go
new file mode 100755
index 000000000..417784e17
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/walker_empty.go
@@ -0,0 +1,255 @@
+package pagetables
+
+// Walker walks page tables.
+type emptyWalker struct {
+ // pageTables are the tables to walk.
+ pageTables *PageTables
+
+ // Visitor is the set of arguments.
+ visitor emptyVisitor
+}
+
+// iterateRange iterates over all appropriate levels of page tables for the given range.
+//
+// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The
+// exception is super pages. If a valid super page (huge or jumbo) cannot be
+// installed, then the walk will continue to individual entries.
+//
+// This algorithm will attempt to maximize the use of super pages whenever
+// possible. Whether a super page is provided will be clear through the range
+// provided in the callback.
+//
+// Note that if requiresAlloc is true, then no gaps will be present. However,
+// if alloc is not set, then the iteration will likely be full of gaps.
+//
+// Note that this function should generally be avoided in favor of Map, Unmap,
+// etc. when not necessary.
+//
+// Precondition: start must be page-aligned.
+//
+// Precondition: start must be less than end.
+//
+// Precondition: If requiresAlloc is true, then start and end should not span
+// non-canonical ranges. If they do, a panic will result.
+//
+//go:nosplit
+func (w *emptyWalker) iterateRange(start, end uintptr) {
+ if start%pteSize != 0 {
+ panic("unaligned start")
+ }
+ if end < start {
+ panic("start > end")
+ }
+ if start < lowerTop {
+ if end <= lowerTop {
+ w.iterateRangeCanonical(start, end)
+ } else if end > lowerTop && end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else if start < upperBottom {
+ if end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else {
+ w.iterateRangeCanonical(start, end)
+ }
+}
+
+// next returns the next address quantized by the given size.
+//
+//go:nosplit
+func emptynext(start uintptr, size uintptr) uintptr {
+ start &= ^(size - 1)
+ start += size
+ return start
+}
+
+// iterateRangeCanonical walks a canonical range.
+//
+//go:nosplit
+func (w *emptyWalker) iterateRangeCanonical(start, end uintptr) {
+ for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ {
+ var (
+ pgdEntry = &w.pageTables.root[pgdIndex]
+ pudEntries *PTEs
+ )
+ if !pgdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ start = emptynext(start, pgdSize)
+ continue
+ }
+
+ pudEntries = w.pageTables.Allocator.NewPTEs()
+ pgdEntry.setPageTable(w.pageTables, pudEntries)
+ } else {
+ pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address())
+ }
+
+ clearPUDEntries := uint16(0)
+
+ for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ {
+ var (
+ pudEntry = &pudEntries[pudIndex]
+ pmdEntries *PTEs
+ )
+ if !pudEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPUDEntries++
+ start = emptynext(start, pudSize)
+ continue
+ }
+
+ if start&(pudSize-1) == 0 && end-start >= pudSize {
+ pudEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+ if pudEntry.Valid() {
+ start = emptynext(start, pudSize)
+ continue
+ }
+ }
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+
+ } else if pudEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < emptynext(start, pudSize)) {
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pmdEntries[index].SetSuper()
+ pmdEntries[index].Set(
+ pudEntry.Address()+(pmdSize*uintptr(index)),
+ pudEntry.Opts())
+ }
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+
+ if !pudEntry.Valid() {
+ clearPUDEntries++
+ }
+
+ start = emptynext(start, pudSize)
+ continue
+ }
+ } else {
+ pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address())
+ }
+
+ clearPMDEntries := uint16(0)
+
+ for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ {
+ var (
+ pmdEntry = &pmdEntries[pmdIndex]
+ pteEntries *PTEs
+ )
+ if !pmdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPMDEntries++
+ start = emptynext(start, pmdSize)
+ continue
+ }
+
+ if start&(pmdSize-1) == 0 && end-start >= pmdSize {
+ pmdEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+ if pmdEntry.Valid() {
+ start = emptynext(start, pmdSize)
+ continue
+ }
+ }
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+
+ } else if pmdEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < emptynext(start, pmdSize)) {
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pteEntries[index].Set(
+ pmdEntry.Address()+(pteSize*uintptr(index)),
+ pmdEntry.Opts())
+ }
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+
+ if !pmdEntry.Valid() {
+ clearPMDEntries++
+ }
+
+ start = emptynext(start, pmdSize)
+ continue
+ }
+ } else {
+ pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address())
+ }
+
+ clearPTEEntries := uint16(0)
+
+ for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ {
+ var (
+ pteEntry = &pteEntries[pteIndex]
+ )
+ if !pteEntry.Valid() && !w.visitor.requiresAlloc() {
+ clearPTEEntries++
+ start += pteSize
+ continue
+ }
+
+ w.visitor.visit(uintptr(start), pteEntry, pteSize-1)
+ if !pteEntry.Valid() {
+ if w.visitor.requiresAlloc() {
+ panic("PTE not set after iteration with requiresAlloc!")
+ }
+ clearPTEEntries++
+ }
+
+ start += pteSize
+ continue
+ }
+
+ if clearPTEEntries == entriesPerPage {
+ pmdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pteEntries)
+ clearPMDEntries++
+ }
+ }
+
+ if clearPMDEntries == entriesPerPage {
+ pudEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pmdEntries)
+ clearPUDEntries++
+ }
+ }
+
+ if clearPUDEntries == entriesPerPage {
+ pgdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pudEntries)
+ }
+ }
+}
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_lookup.go b/pkg/sentry/platform/ring0/pagetables/walker_lookup.go
new file mode 100755
index 000000000..906c9c50f
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/walker_lookup.go
@@ -0,0 +1,255 @@
+package pagetables
+
+// Walker walks page tables.
+type lookupWalker struct {
+ // pageTables are the tables to walk.
+ pageTables *PageTables
+
+ // Visitor is the set of arguments.
+ visitor lookupVisitor
+}
+
+// iterateRange iterates over all appropriate levels of page tables for the given range.
+//
+// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The
+// exception is super pages. If a valid super page (huge or jumbo) cannot be
+// installed, then the walk will continue to individual entries.
+//
+// This algorithm will attempt to maximize the use of super pages whenever
+// possible. Whether a super page is provided will be clear through the range
+// provided in the callback.
+//
+// Note that if requiresAlloc is true, then no gaps will be present. However,
+// if alloc is not set, then the iteration will likely be full of gaps.
+//
+// Note that this function should generally be avoided in favor of Map, Unmap,
+// etc. when not necessary.
+//
+// Precondition: start must be page-aligned.
+//
+// Precondition: start must be less than end.
+//
+// Precondition: If requiresAlloc is true, then start and end should not span
+// non-canonical ranges. If they do, a panic will result.
+//
+//go:nosplit
+func (w *lookupWalker) iterateRange(start, end uintptr) {
+ if start%pteSize != 0 {
+ panic("unaligned start")
+ }
+ if end < start {
+ panic("start > end")
+ }
+ if start < lowerTop {
+ if end <= lowerTop {
+ w.iterateRangeCanonical(start, end)
+ } else if end > lowerTop && end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else if start < upperBottom {
+ if end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else {
+ w.iterateRangeCanonical(start, end)
+ }
+}
+
+// next returns the next address quantized by the given size.
+//
+//go:nosplit
+func lookupnext(start uintptr, size uintptr) uintptr {
+ start &= ^(size - 1)
+ start += size
+ return start
+}
+
+// iterateRangeCanonical walks a canonical range.
+//
+//go:nosplit
+func (w *lookupWalker) iterateRangeCanonical(start, end uintptr) {
+ for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ {
+ var (
+ pgdEntry = &w.pageTables.root[pgdIndex]
+ pudEntries *PTEs
+ )
+ if !pgdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ start = lookupnext(start, pgdSize)
+ continue
+ }
+
+ pudEntries = w.pageTables.Allocator.NewPTEs()
+ pgdEntry.setPageTable(w.pageTables, pudEntries)
+ } else {
+ pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address())
+ }
+
+ clearPUDEntries := uint16(0)
+
+ for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ {
+ var (
+ pudEntry = &pudEntries[pudIndex]
+ pmdEntries *PTEs
+ )
+ if !pudEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPUDEntries++
+ start = lookupnext(start, pudSize)
+ continue
+ }
+
+ if start&(pudSize-1) == 0 && end-start >= pudSize {
+ pudEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+ if pudEntry.Valid() {
+ start = lookupnext(start, pudSize)
+ continue
+ }
+ }
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+
+ } else if pudEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < lookupnext(start, pudSize)) {
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pmdEntries[index].SetSuper()
+ pmdEntries[index].Set(
+ pudEntry.Address()+(pmdSize*uintptr(index)),
+ pudEntry.Opts())
+ }
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+
+ if !pudEntry.Valid() {
+ clearPUDEntries++
+ }
+
+ start = lookupnext(start, pudSize)
+ continue
+ }
+ } else {
+ pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address())
+ }
+
+ clearPMDEntries := uint16(0)
+
+ for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ {
+ var (
+ pmdEntry = &pmdEntries[pmdIndex]
+ pteEntries *PTEs
+ )
+ if !pmdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPMDEntries++
+ start = lookupnext(start, pmdSize)
+ continue
+ }
+
+ if start&(pmdSize-1) == 0 && end-start >= pmdSize {
+ pmdEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+ if pmdEntry.Valid() {
+ start = lookupnext(start, pmdSize)
+ continue
+ }
+ }
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+
+ } else if pmdEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < lookupnext(start, pmdSize)) {
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pteEntries[index].Set(
+ pmdEntry.Address()+(pteSize*uintptr(index)),
+ pmdEntry.Opts())
+ }
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+
+ if !pmdEntry.Valid() {
+ clearPMDEntries++
+ }
+
+ start = lookupnext(start, pmdSize)
+ continue
+ }
+ } else {
+ pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address())
+ }
+
+ clearPTEEntries := uint16(0)
+
+ for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ {
+ var (
+ pteEntry = &pteEntries[pteIndex]
+ )
+ if !pteEntry.Valid() && !w.visitor.requiresAlloc() {
+ clearPTEEntries++
+ start += pteSize
+ continue
+ }
+
+ w.visitor.visit(uintptr(start), pteEntry, pteSize-1)
+ if !pteEntry.Valid() {
+ if w.visitor.requiresAlloc() {
+ panic("PTE not set after iteration with requiresAlloc!")
+ }
+ clearPTEEntries++
+ }
+
+ start += pteSize
+ continue
+ }
+
+ if clearPTEEntries == entriesPerPage {
+ pmdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pteEntries)
+ clearPMDEntries++
+ }
+ }
+
+ if clearPMDEntries == entriesPerPage {
+ pudEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pmdEntries)
+ clearPUDEntries++
+ }
+ }
+
+ if clearPUDEntries == entriesPerPage {
+ pgdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pudEntries)
+ }
+ }
+}
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_map.go b/pkg/sentry/platform/ring0/pagetables/walker_map.go
new file mode 100755
index 000000000..61ee3c825
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/walker_map.go
@@ -0,0 +1,255 @@
+package pagetables
+
+// Walker walks page tables.
+type mapWalker struct {
+ // pageTables are the tables to walk.
+ pageTables *PageTables
+
+ // Visitor is the set of arguments.
+ visitor mapVisitor
+}
+
+// iterateRange iterates over all appropriate levels of page tables for the given range.
+//
+// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The
+// exception is super pages. If a valid super page (huge or jumbo) cannot be
+// installed, then the walk will continue to individual entries.
+//
+// This algorithm will attempt to maximize the use of super pages whenever
+// possible. Whether a super page is provided will be clear through the range
+// provided in the callback.
+//
+// Note that if requiresAlloc is true, then no gaps will be present. However,
+// if alloc is not set, then the iteration will likely be full of gaps.
+//
+// Note that this function should generally be avoided in favor of Map, Unmap,
+// etc. when not necessary.
+//
+// Precondition: start must be page-aligned.
+//
+// Precondition: start must be less than end.
+//
+// Precondition: If requiresAlloc is true, then start and end should not span
+// non-canonical ranges. If they do, a panic will result.
+//
+//go:nosplit
+func (w *mapWalker) iterateRange(start, end uintptr) {
+ if start%pteSize != 0 {
+ panic("unaligned start")
+ }
+ if end < start {
+ panic("start > end")
+ }
+ if start < lowerTop {
+ if end <= lowerTop {
+ w.iterateRangeCanonical(start, end)
+ } else if end > lowerTop && end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else if start < upperBottom {
+ if end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else {
+ w.iterateRangeCanonical(start, end)
+ }
+}
+
+// next returns the next address quantized by the given size.
+//
+//go:nosplit
+func mapnext(start uintptr, size uintptr) uintptr {
+ start &= ^(size - 1)
+ start += size
+ return start
+}
+
+// iterateRangeCanonical walks a canonical range.
+//
+//go:nosplit
+func (w *mapWalker) iterateRangeCanonical(start, end uintptr) {
+ for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ {
+ var (
+ pgdEntry = &w.pageTables.root[pgdIndex]
+ pudEntries *PTEs
+ )
+ if !pgdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ start = mapnext(start, pgdSize)
+ continue
+ }
+
+ pudEntries = w.pageTables.Allocator.NewPTEs()
+ pgdEntry.setPageTable(w.pageTables, pudEntries)
+ } else {
+ pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address())
+ }
+
+ clearPUDEntries := uint16(0)
+
+ for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ {
+ var (
+ pudEntry = &pudEntries[pudIndex]
+ pmdEntries *PTEs
+ )
+ if !pudEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPUDEntries++
+ start = mapnext(start, pudSize)
+ continue
+ }
+
+ if start&(pudSize-1) == 0 && end-start >= pudSize {
+ pudEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+ if pudEntry.Valid() {
+ start = mapnext(start, pudSize)
+ continue
+ }
+ }
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+
+ } else if pudEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < mapnext(start, pudSize)) {
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pmdEntries[index].SetSuper()
+ pmdEntries[index].Set(
+ pudEntry.Address()+(pmdSize*uintptr(index)),
+ pudEntry.Opts())
+ }
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+
+ if !pudEntry.Valid() {
+ clearPUDEntries++
+ }
+
+ start = mapnext(start, pudSize)
+ continue
+ }
+ } else {
+ pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address())
+ }
+
+ clearPMDEntries := uint16(0)
+
+ for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ {
+ var (
+ pmdEntry = &pmdEntries[pmdIndex]
+ pteEntries *PTEs
+ )
+ if !pmdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPMDEntries++
+ start = mapnext(start, pmdSize)
+ continue
+ }
+
+ if start&(pmdSize-1) == 0 && end-start >= pmdSize {
+ pmdEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+ if pmdEntry.Valid() {
+ start = mapnext(start, pmdSize)
+ continue
+ }
+ }
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+
+ } else if pmdEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < mapnext(start, pmdSize)) {
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pteEntries[index].Set(
+ pmdEntry.Address()+(pteSize*uintptr(index)),
+ pmdEntry.Opts())
+ }
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+
+ if !pmdEntry.Valid() {
+ clearPMDEntries++
+ }
+
+ start = mapnext(start, pmdSize)
+ continue
+ }
+ } else {
+ pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address())
+ }
+
+ clearPTEEntries := uint16(0)
+
+ for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ {
+ var (
+ pteEntry = &pteEntries[pteIndex]
+ )
+ if !pteEntry.Valid() && !w.visitor.requiresAlloc() {
+ clearPTEEntries++
+ start += pteSize
+ continue
+ }
+
+ w.visitor.visit(uintptr(start), pteEntry, pteSize-1)
+ if !pteEntry.Valid() {
+ if w.visitor.requiresAlloc() {
+ panic("PTE not set after iteration with requiresAlloc!")
+ }
+ clearPTEEntries++
+ }
+
+ start += pteSize
+ continue
+ }
+
+ if clearPTEEntries == entriesPerPage {
+ pmdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pteEntries)
+ clearPMDEntries++
+ }
+ }
+
+ if clearPMDEntries == entriesPerPage {
+ pudEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pmdEntries)
+ clearPUDEntries++
+ }
+ }
+
+ if clearPUDEntries == entriesPerPage {
+ pgdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pudEntries)
+ }
+ }
+}
diff --git a/pkg/sentry/platform/ring0/pagetables/walker_unmap.go b/pkg/sentry/platform/ring0/pagetables/walker_unmap.go
new file mode 100755
index 000000000..be2aa0ce4
--- /dev/null
+++ b/pkg/sentry/platform/ring0/pagetables/walker_unmap.go
@@ -0,0 +1,255 @@
+package pagetables
+
+// Walker walks page tables.
+type unmapWalker struct {
+ // pageTables are the tables to walk.
+ pageTables *PageTables
+
+ // Visitor is the set of arguments.
+ visitor unmapVisitor
+}
+
+// iterateRange iterates over all appropriate levels of page tables for the given range.
+//
+// If requiresAlloc is true, then Set _must_ be called on all given PTEs. The
+// exception is super pages. If a valid super page (huge or jumbo) cannot be
+// installed, then the walk will continue to individual entries.
+//
+// This algorithm will attempt to maximize the use of super pages whenever
+// possible. Whether a super page is provided will be clear through the range
+// provided in the callback.
+//
+// Note that if requiresAlloc is true, then no gaps will be present. However,
+// if alloc is not set, then the iteration will likely be full of gaps.
+//
+// Note that this function should generally be avoided in favor of Map, Unmap,
+// etc. when not necessary.
+//
+// Precondition: start must be page-aligned.
+//
+// Precondition: start must be less than end.
+//
+// Precondition: If requiresAlloc is true, then start and end should not span
+// non-canonical ranges. If they do, a panic will result.
+//
+//go:nosplit
+func (w *unmapWalker) iterateRange(start, end uintptr) {
+ if start%pteSize != 0 {
+ panic("unaligned start")
+ }
+ if end < start {
+ panic("start > end")
+ }
+ if start < lowerTop {
+ if end <= lowerTop {
+ w.iterateRangeCanonical(start, end)
+ } else if end > lowerTop && end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(start, lowerTop)
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else if start < upperBottom {
+ if end <= upperBottom {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ } else {
+ if w.visitor.requiresAlloc() {
+ panic("alloc spans non-canonical range")
+ }
+ w.iterateRangeCanonical(upperBottom, end)
+ }
+ } else {
+ w.iterateRangeCanonical(start, end)
+ }
+}
+
+// next returns the next address quantized by the given size.
+//
+//go:nosplit
+func unmapnext(start uintptr, size uintptr) uintptr {
+ start &= ^(size - 1)
+ start += size
+ return start
+}
+
+// iterateRangeCanonical walks a canonical range.
+//
+//go:nosplit
+func (w *unmapWalker) iterateRangeCanonical(start, end uintptr) {
+ for pgdIndex := uint16((start & pgdMask) >> pgdShift); start < end && pgdIndex < entriesPerPage; pgdIndex++ {
+ var (
+ pgdEntry = &w.pageTables.root[pgdIndex]
+ pudEntries *PTEs
+ )
+ if !pgdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ start = unmapnext(start, pgdSize)
+ continue
+ }
+
+ pudEntries = w.pageTables.Allocator.NewPTEs()
+ pgdEntry.setPageTable(w.pageTables, pudEntries)
+ } else {
+ pudEntries = w.pageTables.Allocator.LookupPTEs(pgdEntry.Address())
+ }
+
+ clearPUDEntries := uint16(0)
+
+ for pudIndex := uint16((start & pudMask) >> pudShift); start < end && pudIndex < entriesPerPage; pudIndex++ {
+ var (
+ pudEntry = &pudEntries[pudIndex]
+ pmdEntries *PTEs
+ )
+ if !pudEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPUDEntries++
+ start = unmapnext(start, pudSize)
+ continue
+ }
+
+ if start&(pudSize-1) == 0 && end-start >= pudSize {
+ pudEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+ if pudEntry.Valid() {
+ start = unmapnext(start, pudSize)
+ continue
+ }
+ }
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+
+ } else if pudEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pudSize-1) != 0 || end < unmapnext(start, pudSize)) {
+
+ pmdEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pmdEntries[index].SetSuper()
+ pmdEntries[index].Set(
+ pudEntry.Address()+(pmdSize*uintptr(index)),
+ pudEntry.Opts())
+ }
+ pudEntry.setPageTable(w.pageTables, pmdEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pudEntry, pudSize-1)
+
+ if !pudEntry.Valid() {
+ clearPUDEntries++
+ }
+
+ start = unmapnext(start, pudSize)
+ continue
+ }
+ } else {
+ pmdEntries = w.pageTables.Allocator.LookupPTEs(pudEntry.Address())
+ }
+
+ clearPMDEntries := uint16(0)
+
+ for pmdIndex := uint16((start & pmdMask) >> pmdShift); start < end && pmdIndex < entriesPerPage; pmdIndex++ {
+ var (
+ pmdEntry = &pmdEntries[pmdIndex]
+ pteEntries *PTEs
+ )
+ if !pmdEntry.Valid() {
+ if !w.visitor.requiresAlloc() {
+
+ clearPMDEntries++
+ start = unmapnext(start, pmdSize)
+ continue
+ }
+
+ if start&(pmdSize-1) == 0 && end-start >= pmdSize {
+ pmdEntry.SetSuper()
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+ if pmdEntry.Valid() {
+ start = unmapnext(start, pmdSize)
+ continue
+ }
+ }
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+
+ } else if pmdEntry.IsSuper() {
+
+ if w.visitor.requiresSplit() && (start&(pmdSize-1) != 0 || end < unmapnext(start, pmdSize)) {
+
+ pteEntries = w.pageTables.Allocator.NewPTEs()
+ for index := uint16(0); index < entriesPerPage; index++ {
+ pteEntries[index].Set(
+ pmdEntry.Address()+(pteSize*uintptr(index)),
+ pmdEntry.Opts())
+ }
+ pmdEntry.setPageTable(w.pageTables, pteEntries)
+ } else {
+
+ w.visitor.visit(uintptr(start), pmdEntry, pmdSize-1)
+
+ if !pmdEntry.Valid() {
+ clearPMDEntries++
+ }
+
+ start = unmapnext(start, pmdSize)
+ continue
+ }
+ } else {
+ pteEntries = w.pageTables.Allocator.LookupPTEs(pmdEntry.Address())
+ }
+
+ clearPTEEntries := uint16(0)
+
+ for pteIndex := uint16((start & pteMask) >> pteShift); start < end && pteIndex < entriesPerPage; pteIndex++ {
+ var (
+ pteEntry = &pteEntries[pteIndex]
+ )
+ if !pteEntry.Valid() && !w.visitor.requiresAlloc() {
+ clearPTEEntries++
+ start += pteSize
+ continue
+ }
+
+ w.visitor.visit(uintptr(start), pteEntry, pteSize-1)
+ if !pteEntry.Valid() {
+ if w.visitor.requiresAlloc() {
+ panic("PTE not set after iteration with requiresAlloc!")
+ }
+ clearPTEEntries++
+ }
+
+ start += pteSize
+ continue
+ }
+
+ if clearPTEEntries == entriesPerPage {
+ pmdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pteEntries)
+ clearPMDEntries++
+ }
+ }
+
+ if clearPMDEntries == entriesPerPage {
+ pudEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pmdEntries)
+ clearPUDEntries++
+ }
+ }
+
+ if clearPUDEntries == entriesPerPage {
+ pgdEntry.Clear()
+ w.pageTables.Allocator.FreePTEs(pudEntries)
+ }
+ }
+}