visor/pkg/sentry/fsimpl/proc/tasks_files.go

// 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.

package proc

import (
	"bytes"
	"fmt"
	"runtime"
	"strconv"

	"gvisor.dev/gvisor/pkg/abi/linux"
	"gvisor.dev/gvisor/pkg/context"
	"gvisor.dev/gvisor/pkg/errors/linuxerr"
	"gvisor.dev/gvisor/pkg/hostarch"
	"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
	"gvisor.dev/gvisor/pkg/sentry/kernel"
	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
	"gvisor.dev/gvisor/pkg/sentry/kernel/version"
	"gvisor.dev/gvisor/pkg/sentry/ktime"
	"gvisor.dev/gvisor/pkg/sentry/usage"
	"gvisor.dev/gvisor/pkg/sentry/vfs"
)

// +stateify savable
type selfSymlink struct {
	implStatFS
	kernfs.InodeAttrs
	kernfs.InodeNoopRefCount
	kernfs.InodeNotAnonymous
	kernfs.InodeSymlink
	kernfs.InodeWatches
	kernfs.InodeFSOwned

	pidns *kernel.PIDNamespace
}

var _ kernfs.Inode = (*selfSymlink)(nil)

func (i *tasksInode) newSelfSymlink(ctx context.Context, creds *auth.Credentials) kernfs.Inode {
	inode := &selfSymlink{pidns: i.pidns}
	inode.Init(ctx, creds, linux.UNNAMED_MAJOR, i.fs.devMinor, i.fs.NextIno(), linux.ModeSymlink|0777)
	return inode
}

func (s *selfSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) {
	t := kernel.TaskFromContext(ctx)
	if t == nil {
		// Who is reading this link?
		return "", linuxerr.EINVAL
	}
	tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup())
	if tgid == 0 {
		return "", linuxerr.ENOENT
	}
	return strconv.FormatUint(uint64(tgid), 10), nil
}

func (s *selfSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) {
	target, err := s.Readlink(ctx, mnt)
	return vfs.VirtualDentry{}, target, err
}

// SetStat implements kernfs.Inode.SetStat not allowing inode attributes to be changed.
func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error {
	return linuxerr.EPERM
}

// +stateify savable
type threadSelfSymlink struct {
	implStatFS
	kernfs.InodeAttrs
	kernfs.InodeNoopRefCount
	kernfs.InodeNotAnonymous
	kernfs.InodeSymlink
	kernfs.InodeWatches
	kernfs.InodeFSOwned

	pidns *kernel.PIDNamespace
}

var _ kernfs.Inode = (*threadSelfSymlink)(nil)

func (i *tasksInode) newThreadSelfSymlink(ctx context.Context, creds *auth.Credentials) kernfs.Inode {
	inode := &threadSelfSymlink{pidns: i.pidns}
	inode.Init(ctx, creds, linux.UNNAMED_MAJOR, i.fs.devMinor, i.fs.NextIno(), linux.ModeSymlink|0777)
	return inode
}

func (s *threadSelfSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) {
	t := kernel.TaskFromContext(ctx)
	if t == nil {
		// Who is reading this link?
		return "", linuxerr.EINVAL
	}
	tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup())
	tid := s.pidns.IDOfTask(t)
	if tid == 0 || tgid == 0 {
		return "", linuxerr.ENOENT
	}
	return fmt.Sprintf("%d/task/%d", tgid, tid), nil
}

func (s *threadSelfSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) {
	target, err := s.Readlink(ctx, mnt)
	return vfs.VirtualDentry{}, target, err
}

// SetStat implements kernfs.Inode.SetStat not allowing inode attributes to be changed.
func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error {
	return linuxerr.EPERM
}

// dynamicBytesFileSetAttr implements a special file that allows inode
// attributes to be set. This is to support /proc files that are readonly, but
// allow attributes to be set.
//
// +stateify savable
type dynamicBytesFileSetAttr struct {
	kernfs.DynamicBytesFile
}

// SetStat implements kernfs.Inode.SetStat.
func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error {
	return d.DynamicBytesFile.InodeAttrs.SetStat(ctx, fs, creds, opts)
}

// cpuStats contains the breakdown of CPU time for /proc/stat.
//
// +stateify savable
type cpuStats struct {
	// user is time spent in userspace tasks with non-positive niceness.
	user uint64

	// nice is time spent in userspace tasks with positive niceness.
	nice uint64

	// system is time spent in non-interrupt kernel context.
	system uint64

	// idle is time spent idle.
	idle uint64

	// ioWait is time spent waiting for IO.
	ioWait uint64

	// irq is time spent in interrupt context.
	irq uint64

	// softirq is time spent in software interrupt context.
	softirq uint64

	// steal is involuntary wait time.
	steal uint64

	// guest is time spent in guests with non-positive niceness.
	guest uint64

	// guestNice is time spent in guests with positive niceness.
	guestNice uint64
}

// String implements fmt.Stringer.
func (c cpuStats) String() string {
	return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice)
}

// statData implements vfs.DynamicBytesSource for /proc/stat.
//
// +stateify savable
type statData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*statData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*statData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	k := kernel.KernelFromContext(ctx)

	// gVisor does not track per-CPU usage. Instead, report the kernel-wide
	// aggregate user and system CPU time (maintained by the CPU clock ticker)
	// and approximate idle time from uptime. This allows utilities like top(1)
	// and htop(1) to compute non-zero CPU usage from the delta between
	// successive reads of /proc/stat.
	cpu := k.CPUStats()

	cores := k.ApplicationCores()
	uptime := ktime.NowFromContext(ctx).Sub(k.Timekeeper().BootTime())

	userTicks := uint64(linux.ClockTFromDuration(cpu.UserTime))
	sysTicks := uint64(linux.ClockTFromDuration(cpu.SysTime))

	// Idle time is the total available CPU time (one uptime's worth per core)
	// minus the time spent doing work.
	var idleTicks uint64
	if total := uint64(linux.ClockTFromDuration(uptime)) * uint64(cores); total > userTicks+sysTicks {
		idleTicks = total - userTicks - sysTicks
	}

	fmt.Fprintf(buf, "cpu  %s\n", cpuStats{user: userTicks, system: sysTicks, idle: idleTicks})

	// gVisor has no per-CPU accounting, so distribute the aggregate stats
	// evenly across the application cores, assigning any remainder to cpu0.
	// The per-CPU lines therefore sum back to the aggregate "cpu" line. This
	// makes the aggregate statistic a reasonable approximation of the total CPU
	// usage across all cores.
	for c := uint(0); c < cores; c++ {
		per := cpuStats{
			user:   userTicks / uint64(cores),
			system: sysTicks / uint64(cores),
			idle:   idleTicks / uint64(cores),
		}
		if c == 0 {
			per.user += userTicks % uint64(cores)
			per.system += sysTicks % uint64(cores)
			per.idle += idleTicks % uint64(cores)
		}
		fmt.Fprintf(buf, "cpu%d %s\n", c, per)
	}

	// The total number of interrupts is dependent on the CPUs and PCI
	// devices on the system. See arch_probe_nr_irqs.
	//
	// Since we don't report real interrupt stats, just choose an arbitrary
	// value from a representative VM.
	const numInterrupts = 256

	// The Kernel doesn't handle real interrupts, so report all zeroes.
	// TODO(b/37226836): We could count page faults as #PF.
	fmt.Fprintf(buf, "intr 0") // total
	for i := 0; i < numInterrupts; i++ {
		fmt.Fprintf(buf, " 0")
	}
	fmt.Fprintf(buf, "\n")

	// Total number of context switches.
	// TODO(b/37226836): Count this.
	fmt.Fprintf(buf, "ctxt 0\n")

	// CLOCK_REALTIME timestamp from boot, in seconds.
	fmt.Fprintf(buf, "btime %d\n", k.Timekeeper().BootTime().Seconds())

	// Total number of clones.
	// TODO(b/37226836): Count this.
	fmt.Fprintf(buf, "processes 0\n")

	// Number of runnable tasks, read directly from the kernel's running-task
	// counter, which is maintained as tasks change scheduling state.
	fmt.Fprintf(buf, "procs_running %d\n", k.RunningTasks())

	// Number of tasks in uninterruptible sleep (e.g. waiting on IO).
	fmt.Fprintf(buf, "procs_blocked %d\n", k.BlockedTasks())

	// Number of each softirq handled.
	fmt.Fprintf(buf, "softirq 0") // total
	for i := 0; i < linux.NumSoftIRQ; i++ {
		fmt.Fprintf(buf, " 0")
	}
	fmt.Fprintf(buf, "\n")
	return nil
}

// loadavgData backs /proc/loadavg.
//
// +stateify savable
type loadavgData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*loadavgData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	// TODO(b/62345059): Include real data in fields.
	// Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods.
	// Column 4-5: currently running processes and the total number of processes.
	// Column 6: the last process ID used.
	fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0)
	return nil
}

// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo.
//
// +stateify savable
type meminfoData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*meminfoData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	mf := kernel.KernelFromContext(ctx).MemoryFile()
	_ = mf.UpdateUsage(nil) // Best effort
	snapshot, totalUsage := usage.MemoryAccounting.Copy()
	totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage)
	anon := snapshot.Anonymous + snapshot.Tmpfs
	file := snapshot.PageCache + snapshot.Mapped
	// We don't actually have active/inactive LRUs, so just make up numbers.
	activeFile := (file / 2) &^ (hostarch.PageSize - 1)
	inactiveFile := file - activeFile

	fmt.Fprintf(buf, "MemTotal:       %8d kB\n", totalSize/1024)
	memFree := totalSize - totalUsage
	if memFree > totalSize {
		// Underflow.
		memFree = 0
	}
	dirty, writeback := usage.DirtyMemoryAccounting.Copy()
	// We use MemFree as MemAvailable because we don't swap.
	// TODO(rahat): When reclaim is implemented the value of MemAvailable
	// should change.
	fmt.Fprintf(buf, "MemFree:        %8d kB\n", memFree/1024)
	fmt.Fprintf(buf, "MemAvailable:   %8d kB\n", memFree/1024)
	fmt.Fprintf(buf, "Buffers:               0 kB\n") // memory usage by block devices
	fmt.Fprintf(buf, "Cached:         %8d kB\n", (file+snapshot.Tmpfs)/1024)
	// Emulate a system with no swap, which disables inactivation of anon pages.
	fmt.Fprintf(buf, "SwapCache:             0 kB\n")
	fmt.Fprintf(buf, "Active:         %8d kB\n", (anon+activeFile)/1024)
	fmt.Fprintf(buf, "Inactive:       %8d kB\n", inactiveFile/1024)
	fmt.Fprintf(buf, "Active(anon):   %8d kB\n", anon/1024)
	fmt.Fprintf(buf, "Inactive(anon):        0 kB\n")
	fmt.Fprintf(buf, "Active(file):   %8d kB\n", activeFile/1024)
	fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024)
	fmt.Fprintf(buf, "Unevictable:           0 kB\n") // TODO(b/31823263)
	fmt.Fprintf(buf, "Mlocked:               0 kB\n") // TODO(b/31823263)
	fmt.Fprintf(buf, "SwapTotal:             0 kB\n")
	fmt.Fprintf(buf, "SwapFree:              0 kB\n")
	fmt.Fprintf(buf, "Dirty:          %8d kB\n", dirty/1024)
	fmt.Fprintf(buf, "Writeback:      %8d kB\n", writeback/1024)
	fmt.Fprintf(buf, "AnonPages:      %8d kB\n", anon/1024)
	fmt.Fprintf(buf, "Mapped:         %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know
	fmt.Fprintf(buf, "Shmem:          %8d kB\n", snapshot.Tmpfs/1024)
	return nil
}

// uptimeData implements vfs.DynamicBytesSource for /proc/uptime.
//
// +stateify savable
type uptimeData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*uptimeData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	k := kernel.KernelFromContext(ctx)
	now := ktime.NowFromContext(ctx)

	// Pretend that we've spent zero time sleeping (second number).
	fmt.Fprintf(buf, "%.2f 0.00\n", now.Sub(k.Timekeeper().BootTime()).Seconds())
	return nil
}

// versionData implements vfs.DynamicBytesSource for /proc/version.
//
// +stateify savable
type versionData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*versionData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*versionData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	// /proc/version takes the form:
	//
	// "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST)
	// (COMPILER_VERSION) VERSION"
	//
	// where:
	//	- SYSNAME, RELEASE, and VERSION are the same as returned by
	//		sys_utsname
	//	- COMPILE_USER is the user that build the kernel
	//	- COMPILE_HOST is the hostname of the machine on which the kernel
	//		was built
	//	- COMPILER_VERSION is the version reported by the building compiler
	//
	// Since we don't really want to expose build information to
	// applications, those fields are omitted.
	//
	// FIXME(mpratt): Using Version from the init task SyscallTable
	// disregards the different version a task may have (e.g., in a uts
	// namespace).
	fmt.Fprintf(buf, "%s version %s %s\n", version.LinuxSysname, version.LinuxRelease(), version.LinuxVersion)
	return nil
}

// filesystemsData backs /proc/filesystems.
//
// +stateify savable
type filesystemsData struct {
	kernfs.DynamicBytesFile
}

var _ dynamicInode = (*filesystemsData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (d *filesystemsData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	k := kernel.KernelFromContext(ctx)
	k.VFS().GenerateProcFilesystems(buf)
	return nil
}

// cgroupsData backs /proc/cgroups.
//
// +stateify savable
type cgroupsData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*cgroupsData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*cgroupsData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	r := kernel.KernelFromContext(ctx).CgroupRegistry()
	r.GenerateProcCgroups(buf)
	return nil
}

// cmdLineData backs /proc/cmdline.
//
// +stateify savable
type cmdLineData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*cmdLineData)(nil)

// Generate implements vfs.DynamicByteSource.Generate.
func (*cmdLineData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	fmt.Fprintf(buf, "BOOT_IMAGE=/vmlinuz-%s quiet\n", version.LinuxRelease())
	return nil
}

// devicesData backs /proc/devices.
//
// +stateify savable
type devicesData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*devicesData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*devicesData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	k := kernel.KernelFromContext(ctx)
	return k.VFS().GenerateProcDevices(buf)
}

// sentryMeminfoData implements vfs.DynamicBytesSource for /proc/sentry-meminfo.
//
// +stateify savable
type sentryMeminfoData struct {
	dynamicBytesFileSetAttr
}

var _ dynamicInode = (*sentryMeminfoData)(nil)

// Generate implements vfs.DynamicBytesSource.Generate.
func (*sentryMeminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error {
	var sentryMeminfo runtime.MemStats
	runtime.ReadMemStats(&sentryMeminfo)

	fmt.Fprintf(buf, "Alloc:          %8d kB\n", sentryMeminfo.Alloc/1024)
	fmt.Fprintf(buf, "TotalAlloc:     %8d kB\n", sentryMeminfo.TotalAlloc/1024)
	fmt.Fprintf(buf, "Sys:            %8d kB\n", sentryMeminfo.Sys/1024)
	fmt.Fprintf(buf, "Mallocs:        %8d\n", sentryMeminfo.Mallocs)
	fmt.Fprintf(buf, "Frees:          %8d\n", sentryMeminfo.Frees)
	fmt.Fprintf(buf, "Live Objects:   %8d\n", sentryMeminfo.Mallocs-sentryMeminfo.Frees)
	fmt.Fprintf(buf, "HeapAlloc:      %8d kB\n", sentryMeminfo.HeapAlloc/1024)
	fmt.Fprintf(buf, "HeapSys:        %8d kB\n", sentryMeminfo.HeapSys/1024)
	fmt.Fprintf(buf, "HeapObjects:    %8d\n", sentryMeminfo.HeapObjects)
	return nil
}