diff options
Diffstat (limited to 'pkg/sentry/platform')
23 files changed, 168 insertions, 54 deletions
diff --git a/pkg/sentry/platform/interrupt/BUILD b/pkg/sentry/platform/interrupt/BUILD index b6d008dbe..85e882df9 100644 --- a/pkg/sentry/platform/interrupt/BUILD +++ b/pkg/sentry/platform/interrupt/BUILD @@ -10,6 +10,7 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/sentry/platform/interrupt", visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/sentry/platform/interrupt/interrupt.go b/pkg/sentry/platform/interrupt/interrupt.go index a4651f500..57be41647 100644 --- a/pkg/sentry/platform/interrupt/interrupt.go +++ b/pkg/sentry/platform/interrupt/interrupt.go @@ -17,7 +17,8 @@ package interrupt import ( "fmt" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // Receiver receives interrupt notifications from a Forwarder. diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index f3afd98da..6a358d1d4 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -55,6 +55,7 @@ go_library( "//pkg/sentry/platform/safecopy", "//pkg/sentry/time", "//pkg/sentry/usermem", + "//pkg/sync", ], ) diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index ea8b9632e..a25f3c449 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -15,13 +15,13 @@ package kvm import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/atomicbitops" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // dirtySet tracks vCPUs for invalidation. diff --git a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go index e5fac0d6a..2f02c03cf 100644 --- a/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go +++ b/pkg/sentry/platform/kvm/bluepill_arm64_unsafe.go @@ -17,8 +17,6 @@ package kvm import ( - "unsafe" - "gvisor.dev/gvisor/pkg/sentry/arch" ) diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index f2c2c059e..a7850faed 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -18,13 +18,13 @@ package kvm import ( "fmt" "os" - "sync" "syscall" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // KVM represents a lightweight VM context. diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 7d02ebf19..e6d912168 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -17,7 +17,6 @@ package kvm import ( "fmt" "runtime" - "sync" "sync/atomic" "syscall" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // machine contains state associated with the VM as a whole. diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index b99fe425e..873e39dc7 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -90,7 +90,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 7ae47f291..3b1f20219 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -97,7 +97,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 0df8cfa0f..cd13390c3 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -33,6 +33,7 @@ go_library( "//pkg/sentry/platform/interrupt", "//pkg/sentry/platform/safecopy", "//pkg/sentry/usermem", + "//pkg/sync", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 7b120a15d..bb0e03880 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -46,13 +46,13 @@ package ptrace import ( "os" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) var ( diff --git a/pkg/sentry/platform/ptrace/stub_amd64.s b/pkg/sentry/platform/ptrace/stub_amd64.s index 64c718d21..16f9c523e 100644 --- a/pkg/sentry/platform/ptrace/stub_amd64.s +++ b/pkg/sentry/platform/ptrace/stub_amd64.s @@ -64,6 +64,8 @@ begin: CMPQ AX, $0 JL error + MOVQ $0, BX + // SIGSTOP to wait for attach. // // The SYSCALL instruction will be used for future syscall injection by @@ -73,23 +75,26 @@ begin: MOVQ $SIGSTOP, SI SYSCALL - // The tracer may "detach" and/or allow code execution here in three cases: - // - // 1. New (traced) stub threads are explicitly detached by the - // goroutine in newSubprocess. However, they are detached while in - // group-stop, so they do not execute code here. - // - // 2. If a tracer thread exits, it implicitly detaches from the stub, - // potentially allowing code execution here. However, the Go runtime - // never exits individual threads, so this case never occurs. - // - // 3. subprocess.createStub clones a new stub process that is untraced, + // The sentry sets BX to 1 when creating stub process. + CMPQ BX, $1 + JE clone + + // Notify the Sentry that syscall exited. +done: + INT $3 + // Be paranoid. + JMP done +clone: + // subprocess.createStub clones a new stub process that is untraced, // thus executing this code. We setup the PDEATHSIG before SIGSTOPing // ourselves for attach by the tracer. // // R15 has been updated with the expected PPID. - JMP begin + CMPQ AX, $0 + JE begin + // The clone syscall returns a non-zero value. + JMP done error: // Exit with -errno. MOVQ AX, DI diff --git a/pkg/sentry/platform/ptrace/stub_arm64.s b/pkg/sentry/platform/ptrace/stub_arm64.s index 2c5e4d5cb..6162df02a 100644 --- a/pkg/sentry/platform/ptrace/stub_arm64.s +++ b/pkg/sentry/platform/ptrace/stub_arm64.s @@ -59,6 +59,8 @@ begin: CMP $0x0, R0 BLT error + MOVD $0, R9 + // SIGSTOP to wait for attach. // // The SYSCALL instruction will be used for future syscall injection by @@ -66,22 +68,26 @@ begin: MOVD $SYS_KILL, R8 MOVD $SIGSTOP, R1 SVC - // The tracer may "detach" and/or allow code execution here in three cases: - // - // 1. New (traced) stub threads are explicitly detached by the - // goroutine in newSubprocess. However, they are detached while in - // group-stop, so they do not execute code here. - // - // 2. If a tracer thread exits, it implicitly detaches from the stub, - // potentially allowing code execution here. However, the Go runtime - // never exits individual threads, so this case never occurs. - // - // 3. subprocess.createStub clones a new stub process that is untraced, + + // The sentry sets R9 to 1 when creating stub process. + CMP $1, R9 + BEQ clone + +done: + // Notify the Sentry that syscall exited. + BRK $3 + B done // Be paranoid. +clone: + // subprocess.createStub clones a new stub process that is untraced, // thus executing this code. We setup the PDEATHSIG before SIGSTOPing // ourselves for attach by the tracer. // // R7 has been updated with the expected PPID. - B begin + CMP $0, R0 + BEQ begin + + // The clone system call returned a non-zero value. + B done error: // Exit with -errno. diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index 821f6848d..15dc46a5b 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -18,7 +18,6 @@ import ( "fmt" "os" "runtime" - "sync" "syscall" "golang.org/x/sys/unix" @@ -27,6 +26,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" ) // Linux kernel errnos which "should never be seen by user programs", but will @@ -430,13 +430,15 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { } for { - // Execute the syscall instruction. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { + // Execute the syscall instruction. The task has to stop on the + // trap instruction which is right after the syscall + // instruction. + if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_CONT, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) } sig := t.wait(stopped) - if sig == (syscallEvent | syscall.SIGTRAP) { + if sig == syscall.SIGTRAP { // Reached syscall-enter-stop. break } else { @@ -448,18 +450,6 @@ func (t *thread) syscall(regs *syscall.PtraceRegs) (uintptr, error) { } } - // Complete the actual system call. - if _, _, errno := syscall.RawSyscall6(syscall.SYS_PTRACE, syscall.PTRACE_SYSCALL, uintptr(t.tid), 0, 0, 0, 0); errno != 0 { - panic(fmt.Sprintf("ptrace syscall-enter failed: %v", errno)) - } - - // Wait for syscall-exit-stop. "[Signal-delivery-stop] never happens - // between syscall-enter-stop and syscall-exit-stop; it happens *after* - // syscall-exit-stop.)" - ptrace(2), "Syscall-stops" - if sig := t.wait(stopped); sig != (syscallEvent | syscall.SIGTRAP) { - t.dumpAndPanic(fmt.Sprintf("wait failed: expected SIGTRAP, got %v [%d]", sig, sig)) - } - // Grab registers. if err := t.getRegs(regs); err != nil { panic(fmt.Sprintf("ptrace get regs failed: %v", err)) diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go index 606dc2b1d..e99798c56 100644 --- a/pkg/sentry/platform/ptrace/subprocess_amd64.go +++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go @@ -141,9 +141,11 @@ func (t *thread) adjustInitRegsRip() { t.initRegs.Rip -= initRegsRipAdjustment } -// Pass the expected PPID to the child via R15 when creating stub process +// Pass the expected PPID to the child via R15 when creating stub process. func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.R15 = uint64(ppid) + // Rbx has to be set to 1 when creating stub process. + initregs.Rbx = 1 } // patchSignalInfo patches the signal info to account for hitting the seccomp diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go index 62a686ee7..7b975137f 100644 --- a/pkg/sentry/platform/ptrace/subprocess_arm64.go +++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go @@ -127,6 +127,8 @@ func (t *thread) adjustInitRegsRip() { // Pass the expected PPID to the child via X7 when creating stub process func initChildProcessPPID(initregs *syscall.PtraceRegs, ppid int32) { initregs.Regs[7] = uint64(ppid) + // R9 has to be set to 1 when creating stub process. + initregs.Regs[9] = 1 } // patchSignalInfo patches the signal info to account for hitting the seccomp diff --git a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go index 2e6fbe488..245b20722 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux_unsafe.go @@ -18,7 +18,6 @@ package ptrace import ( - "sync" "sync/atomic" "syscall" "unsafe" @@ -26,6 +25,7 @@ import ( "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/hostcpu" + "gvisor.dev/gvisor/pkg/sync" ) // maskPool contains reusable CPU masks for setting affinity. Unfortunately, diff --git a/pkg/sentry/platform/ring0/defs.go b/pkg/sentry/platform/ring0/defs.go index 3f094c2a7..86fd5ed58 100644 --- a/pkg/sentry/platform/ring0/defs.go +++ b/pkg/sentry/platform/ring0/defs.go @@ -17,7 +17,7 @@ package ring0 import ( "syscall" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" ) // Kernel is a global kernel object. diff --git a/pkg/sentry/platform/ring0/defs_amd64.go b/pkg/sentry/platform/ring0/defs_amd64.go index 10dbd381f..9dae0dccb 100644 --- a/pkg/sentry/platform/ring0/defs_amd64.go +++ b/pkg/sentry/platform/ring0/defs_amd64.go @@ -18,6 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/defs_arm64.go b/pkg/sentry/platform/ring0/defs_arm64.go index dc0eeec01..a850ce6cf 100644 --- a/pkg/sentry/platform/ring0/defs_arm64.go +++ b/pkg/sentry/platform/ring0/defs_arm64.go @@ -18,6 +18,7 @@ package ring0 import ( "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 813ef9822..679842288 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -357,6 +357,73 @@ TEXT ·Current(SB),NOSPLIT,$0-8 #define STACK_FRAME_SIZE 16 TEXT ·kernelExitToEl0(SB),NOSPLIT,$0 + // Step1, save sentry context into memory. + REGISTERS_SAVE(RSV_REG, CPU_REGISTERS) + MOVD RSV_REG_APP, CPU_REGISTERS+PTRACE_R9(RSV_REG) + + WORD $0xd5384003 // MRS SPSR_EL1, R3 + MOVD R3, CPU_REGISTERS+PTRACE_PSTATE(RSV_REG) + MOVD R30, CPU_REGISTERS+PTRACE_PC(RSV_REG) + MOVD RSP, R3 + MOVD R3, CPU_REGISTERS+PTRACE_SP(RSV_REG) + + MOVD CPU_REGISTERS+PTRACE_R3(RSV_REG), R3 + + // Step2, save SP_EL1, PSTATE into kernel temporary stack. + // switch to temporary stack. + LOAD_KERNEL_STACK(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + SUB $STACK_FRAME_SIZE, RSP, RSP + MOVD CPU_REGISTERS+PTRACE_SP(RSV_REG), R11 + MOVD CPU_REGISTERS+PTRACE_PSTATE(RSV_REG), R12 + STP (R11, R12), 16*0(RSP) + + MOVD CPU_REGISTERS+PTRACE_R11(RSV_REG), R11 + MOVD CPU_REGISTERS+PTRACE_R12(RSV_REG), R12 + + // Step3, test user pagetable. + // If user pagetable is empty, trapped in el1_ia. + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + SWITCH_TO_APP_PAGETABLE(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + SWITCH_TO_KVM_PAGETABLE(RSV_REG) + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + // If pagetable is not empty, recovery kernel temporary stack. + ADD $STACK_FRAME_SIZE, RSP, RSP + + // Step4, load app context pointer. + MOVD CPU_APP_ADDR(RSV_REG), RSV_REG_APP + + // Step5, prepare the environment for container application. + // set sp_el0. + MOVD PTRACE_SP(RSV_REG_APP), R1 + WORD $0xd5184101 //MSR R1, SP_EL0 + // set pc. + MOVD PTRACE_PC(RSV_REG_APP), R1 + MSR R1, ELR_EL1 + // set pstate. + MOVD PTRACE_PSTATE(RSV_REG_APP), R1 + WORD $0xd5184001 //MSR R1, SPSR_EL1 + + // RSV_REG & RSV_REG_APP will be loaded at the end. + REGISTERS_LOAD(RSV_REG_APP, 0) + + // switch to user pagetable. + MOVD PTRACE_R18(RSV_REG_APP), RSV_REG + MOVD PTRACE_R9(RSV_REG_APP), RSV_REG_APP + + SUB $STACK_FRAME_SIZE, RSP, RSP + STP (RSV_REG, RSV_REG_APP), 16*0(RSP) + + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + SWITCH_TO_APP_PAGETABLE(RSV_REG) + + LDP 16*0(RSP), (RSV_REG, RSV_REG_APP) + ADD $STACK_FRAME_SIZE, RSP, RSP + ERET() TEXT ·kernelExitToEl1(SB),NOSPLIT,$0 @@ -407,6 +474,16 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 B el1_invalid el1_da: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $0, CPU_ERROR_TYPE(RSV_REG) + + MOVD $PageFault, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el1_ia: @@ -467,9 +544,30 @@ TEXT ·El0_sync(SB),NOSPLIT,$0 B el0_invalid el0_svc: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + + MOVD $0, CPU_ERROR_CODE(RSV_REG) // Clear error code. + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $Syscall, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el0_da: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $1, R3 + MOVD R3, CPU_ERROR_TYPE(RSV_REG) // Set error type to user. + + MOVD $PageFault, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el0_ia: diff --git a/pkg/sentry/platform/ring0/pagetables/BUILD b/pkg/sentry/platform/ring0/pagetables/BUILD index e2e15ba5c..387a7f6c3 100644 --- a/pkg/sentry/platform/ring0/pagetables/BUILD +++ b/pkg/sentry/platform/ring0/pagetables/BUILD @@ -96,7 +96,10 @@ go_library( "//pkg/sentry/platform/kvm:__subpackages__", "//pkg/sentry/platform/ring0:__subpackages__", ], - deps = ["//pkg/sentry/usermem"], + deps = [ + "//pkg/sentry/usermem", + "//pkg/sync", + ], ) go_test( diff --git a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go b/pkg/sentry/platform/ring0/pagetables/pcids_x86.go index 0f029f25d..e199bae18 100644 --- a/pkg/sentry/platform/ring0/pagetables/pcids_x86.go +++ b/pkg/sentry/platform/ring0/pagetables/pcids_x86.go @@ -17,7 +17,7 @@ package pagetables import ( - "sync" + "gvisor.dev/gvisor/pkg/sync" ) // limitPCID is the number of valid PCIDs. |