diff options
Diffstat (limited to 'pkg/sentry/platform')
-rw-r--r-- | pkg/sentry/platform/kvm/BUILD | 1 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/address_space.go | 6 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/context.go | 10 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/kvm_const_arm64.go | 11 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/machine_arm64.go | 72 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/testutil/testutil_arm64.go | 4 | ||||
-rw-r--r-- | pkg/sentry/platform/kvm/testutil/testutil_arm64.s | 14 | ||||
-rw-r--r-- | pkg/sentry/platform/platform.go | 44 | ||||
-rw-r--r-- | pkg/sentry/platform/ptrace/BUILD | 1 | ||||
-rw-r--r-- | pkg/sentry/platform/ptrace/ptrace.go | 10 | ||||
-rw-r--r-- | pkg/sentry/platform/ptrace/subprocess.go | 6 | ||||
-rw-r--r-- | pkg/sentry/platform/ring0/kernel_arm64.go | 2 | ||||
-rw-r--r-- | pkg/sentry/platform/ring0/lib_arm64.go | 6 | ||||
-rw-r--r-- | pkg/sentry/platform/ring0/lib_arm64.s | 86 |
14 files changed, 224 insertions, 49 deletions
diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index b5d27a72a..3970dd81d 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -41,6 +41,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/cpuid", "//pkg/log", "//pkg/procid", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index 98a3e539d..af5c5e191 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -248,3 +248,9 @@ func (as *addressSpace) Release() { // Drop all cached machine references. as.machine.dropPageTables(as.pageTables) } + +// PreFork implements platform.AddressSpace.PreFork. +func (as *addressSpace) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (as *addressSpace) PostFork() {} diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index 6507121ea..eb92721fb 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -15,6 +15,7 @@ package kvm import ( + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -37,7 +38,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() localAS := as.(*addressSpace) // Grab a vCPU. @@ -88,3 +90,9 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} + +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 6f0539c29..fdc599477 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -116,6 +116,17 @@ const ( // Arm64: Exception Syndrome Register EL1. const ( + _ESR_ELx_EC_SHIFT = 26 + _ESR_ELx_EC_MASK = 0x3F << _ESR_ELx_EC_SHIFT + + _ESR_ELx_EC_IMP_DEF = 0x1f + _ESR_ELx_EC_IABT_LOW = 0x20 + _ESR_ELx_EC_IABT_CUR = 0x21 + _ESR_ELx_EC_PC_ALIGN = 0x22 + + _ESR_ELx_CM = 1 << 8 + _ESR_ELx_WNR = 1 << 6 + _ESR_ELx_FSC = 0x3F _ESR_SEGV_MAPERR_L0 = 0x4 diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index f3bf973de..9db171af9 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -125,71 +125,59 @@ func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.Acc return usermem.NoAccess, platform.ErrContextSignal } +// isInstructionAbort returns true if it is an instruction abort. +// +//go:nosplit +func isInstructionAbort(code uint64) bool { + value := (code & _ESR_ELx_EC_MASK) >> _ESR_ELx_EC_SHIFT + return value == _ESR_ELx_EC_IABT_LOW +} + +// isWriteFault returns whether it is a write fault. +// +//go:nosplit +func isWriteFault(code uint64) bool { + if isInstructionAbort(code) { + return false + } + + return (code & _ESR_ELx_WNR) != 0 +} + // fault generates an appropriate fault return. // //go:nosplit func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + bluepill(c) // Probably no-op, but may not be. faultAddr := c.GetFaultAddr() code, user := c.ErrorCode() + if !user { + // The last fault serviced by this CPU was not a user + // fault, so we can't reliably trust the faultAddr or + // the code provided here. We need to re-execute. + return usermem.NoAccess, platform.ErrContextInterrupt + } + // Reset the pointed SignalInfo. *info = arch.SignalInfo{Signo: signal} info.SetAddr(uint64(faultAddr)) - read := true - write := false - execute := true - ret := code & _ESR_ELx_FSC switch ret { case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3: info.Code = 1 //SEGV_MAPERR - read = false - write = true - execute = false case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3: info.Code = 2 // SEGV_ACCERR. - read = true - write = false - execute = false default: info.Code = 2 } - if !user { - read = true - write = false - execute = true - - } accessType := usermem.AccessType{ - Read: read, - Write: write, - Execute: execute, + Read: !isWriteFault(uint64(code)), + Write: isWriteFault(uint64(code)), + Execute: isInstructionAbort(uint64(code)), } return accessType, platform.ErrContextSignal } - -// retryInGuest runs the given function in guest mode. -// -// If the function does not complete in guest mode (due to execution of a -// system call due to a GC stall, for example), then it will be retried. The -// given function must be idempotent as a result of the retry mechanism. -func (m *machine) retryInGuest(fn func()) { - c := m.Get() - defer m.Put(c) - for { - c.ClearErrorCode() // See below. - bluepill(c) // Force guest mode. - fn() // Execute the given function. - _, user := c.ErrorCode() - if user { - // If user is set, then we haven't bailed back to host - // mode via a kernel exception or system call. We - // consider the full function to have executed in guest - // mode and we can return. - break - } - } -} diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go index ca902c8c1..4dad877ba 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go @@ -56,5 +56,9 @@ func CheckTestRegs(regs *arch.Registers, full bool) (err error) { err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need) } } + // Check tls. + if need := ^uint64(11); regs.TPIDR_EL0 != need { + err = addRegisterMismatch(err, "tpdir_el0", regs.TPIDR_EL0, need) + } return } diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s index 07658144e..6caf7282d 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s @@ -52,6 +52,8 @@ start: TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 NO_LOCAL_POINTERS + // gc will touch fpsimd, so we should test it. + // such as in <runtime.deductSweepCredit>. FMOVD $(9.9), F0 MOVD $SYS_GETPID, R8 // getpid SVC @@ -102,11 +104,15 @@ isNaN: TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_svc. SVC RET // never reached TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0 - TWIDDLE_REGS() - // Branch to Register branches unconditionally to an address in <Rn>. - JMP (R4) // <=> br x4, must fault - RET // never reached + TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_ia. + // Branch to Register branches unconditionally to an address in <Rn>. + JMP (R6) // <=> br x6, must fault + RET // never reached diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 4b13eec30..3f99afdd1 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -22,6 +22,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -114,6 +115,17 @@ func (NoCPUPreemptionDetection) PreemptAllCPUs() error { panic("This platform does not support CPU preemption detection") } +// MemoryManager represents an abstraction above the platform address space +// which manages memory mappings and their contents. +type MemoryManager interface { + //usermem.IO provides access to the contents of a virtual memory space. + usermem.IO + // MMap establishes a memory mapping. + MMap(ctx context.Context, opts memmap.MMapOpts) (usermem.Addr, error) + // AddressSpace returns the AddressSpace bound to mm. + AddressSpace() AddressSpace +} + // Context represents the execution context for a single thread. type Context interface { // Switch resumes execution of the thread specified by the arch.Context @@ -143,7 +155,30 @@ type Context interface { // concurrent call to Switch(). // // - ErrContextCPUPreempted: See the definition of that error for details. - Switch(as AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + Switch(ctx context.Context, mm MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + + // PullFullState() pulls a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state + // which includes registers and a floating point state. + // + // For example, when the Sentry handles a system call, it may have only + // syscall arguments without other registers and a floating point + // state. And in this case, if the Sentry will need to construct a + // signal frame to call a signal handler, it will need to call + // PullFullState() to load all registers and FPU state. + // + // Preconditions: The caller must be running on the task goroutine. + PullFullState(as AddressSpace, ac arch.Context) + + // FloatingPointStateChanged forces restoring a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state. + // This means that if the Sentry has not changed a thread state, + // the platform may not restore it. + // + // Preconditions: The caller must be running on the task goroutine. + FloatingPointStateChanged() // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. @@ -218,6 +253,13 @@ type AddressSpace interface { // must be acquired via platform.NewAddressSpace(). Release() + // PreFork() is called before creating a copy of AddressSpace. This + // guarantees that this address space will be in a consistent state. + PreFork() + + // PostFork() is called after creating a copy of AddressSpace. + PostFork() + // AddressSpaceIO methods are supported iff the associated platform's // Platform.SupportsAddressSpaceIO() == true. AddressSpaces for which this // does not hold may panic if AddressSpaceIO methods are invoked. diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 29fd23cc3..e04165fbf 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -24,6 +24,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", "//pkg/procid", "//pkg/safecopy", diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 08d055e05..45ff2bcc3 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -48,6 +48,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -95,7 +96,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() s := as.(*subprocess) isSyscall := s.switchToApp(c, ac) @@ -180,6 +182,12 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} + // PTrace represents a collection of ptrace subprocesses. type PTrace struct { platform.MMapMinAddr diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index c990f3454..e1d54d8a2 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -662,3 +662,9 @@ func (s *subprocess) Unmap(addr usermem.Addr, length uint64) { panic(fmt.Sprintf("munmap(%x, %x)) failed: %v", addr, length, err)) } } + +// PreFork implements platform.AddressSpace.PreFork. +func (s *subprocess) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (s *subprocess) PostFork() {} diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index fca3a5478..d483ff03c 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -59,11 +59,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(UserFlagsClear) regs.Pstate |= UserFlagsSet + LoadFloatingPoint(switchOpts.FloatingPointState) SetTLS(regs.TPIDR_EL0) kernelExitToEl0() regs.TPIDR_EL0 = GetTLS() + SaveFloatingPoint(switchOpts.FloatingPointState) vector = c.vecCode diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index a6345010d..00e52c8af 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -38,6 +38,12 @@ func SaveVRegs(*byte) // LoadVRegs loads V0-V31 registers. func LoadVRegs(*byte) +// LoadFloatingPoint loads floating point state. +func LoadFloatingPoint(*byte) + +// SaveFloatingPoint saves floating point state. +func SaveFloatingPoint(*byte) + // GetTLS returns the value of TPIDR_EL0 register. func GetTLS() (value uint64) diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index b63e14b41..86bfbe46f 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -129,3 +129,89 @@ TEXT ·LoadVRegs(SB),NOSPLIT,$0-8 ISB $15 RET + +TEXT ·LoadFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD 0(R0), R1 + MOVD R1, FPSR + MOVD 8(R0), R1 + MOVD R1, NZCV + + FMOVD 16*1(R0), F0 + FMOVD 16*2(R0), F1 + FMOVD 16*3(R0), F2 + FMOVD 16*4(R0), F3 + FMOVD 16*5(R0), F4 + FMOVD 16*6(R0), F5 + FMOVD 16*7(R0), F6 + FMOVD 16*8(R0), F7 + FMOVD 16*9(R0), F8 + FMOVD 16*10(R0), F9 + FMOVD 16*11(R0), F10 + FMOVD 16*12(R0), F11 + FMOVD 16*13(R0), F12 + FMOVD 16*14(R0), F13 + FMOVD 16*15(R0), F14 + FMOVD 16*16(R0), F15 + FMOVD 16*17(R0), F16 + FMOVD 16*18(R0), F17 + FMOVD 16*19(R0), F18 + FMOVD 16*20(R0), F19 + FMOVD 16*21(R0), F20 + FMOVD 16*22(R0), F21 + FMOVD 16*23(R0), F22 + FMOVD 16*24(R0), F23 + FMOVD 16*25(R0), F24 + FMOVD 16*26(R0), F25 + FMOVD 16*27(R0), F26 + FMOVD 16*28(R0), F27 + FMOVD 16*29(R0), F28 + FMOVD 16*30(R0), F29 + FMOVD 16*31(R0), F30 + FMOVD 16*32(R0), F31 + + RET + +TEXT ·SaveFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD FPSR, R1 + MOVD R1, 0(R0) + MOVD NZCV, R1 + MOVD R1, 8(R0) + + FMOVD F0, 16*1(R0) + FMOVD F1, 16*2(R0) + FMOVD F2, 16*3(R0) + FMOVD F3, 16*4(R0) + FMOVD F4, 16*5(R0) + FMOVD F5, 16*6(R0) + FMOVD F6, 16*7(R0) + FMOVD F7, 16*8(R0) + FMOVD F8, 16*9(R0) + FMOVD F9, 16*10(R0) + FMOVD F10, 16*11(R0) + FMOVD F11, 16*12(R0) + FMOVD F12, 16*13(R0) + FMOVD F13, 16*14(R0) + FMOVD F14, 16*15(R0) + FMOVD F15, 16*16(R0) + FMOVD F16, 16*17(R0) + FMOVD F17, 16*18(R0) + FMOVD F18, 16*19(R0) + FMOVD F19, 16*20(R0) + FMOVD F20, 16*21(R0) + FMOVD F21, 16*22(R0) + FMOVD F22, 16*23(R0) + FMOVD F23, 16*24(R0) + FMOVD F24, 16*25(R0) + FMOVD F25, 16*26(R0) + FMOVD F26, 16*27(R0) + FMOVD F27, 16*28(R0) + FMOVD F28, 16*29(R0) + FMOVD F29, 16*30(R0) + FMOVD F30, 16*31(R0) + FMOVD F31, 16*32(R0) + + RET |