diff options
Diffstat (limited to 'pkg/sentry/control')
-rw-r--r-- | pkg/sentry/control/BUILD | 14 | ||||
-rw-r--r-- | pkg/sentry/control/pprof.go | 51 | ||||
-rw-r--r-- | pkg/sentry/control/proc.go | 181 | ||||
-rw-r--r-- | pkg/sentry/control/proc_test.go | 10 |
4 files changed, 212 insertions, 44 deletions
diff --git a/pkg/sentry/control/BUILD b/pkg/sentry/control/BUILD index 5522cecd0..d16d78aa5 100644 --- a/pkg/sentry/control/BUILD +++ b/pkg/sentry/control/BUILD @@ -1,5 +1,4 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") package(licenses = ["notice"]) @@ -12,16 +11,18 @@ go_library( "proc.go", "state.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/control", visibility = [ - "//pkg/sentry:internal", + "//:sandbox", ], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/fd", + "//pkg/fspath", "//pkg/log", "//pkg/sentry/fs", "//pkg/sentry/fs/host", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", @@ -29,7 +30,10 @@ go_library( "//pkg/sentry/state", "//pkg/sentry/strace", "//pkg/sentry/usage", + "//pkg/sentry/vfs", "//pkg/sentry/watchdog", + "//pkg/sync", + "//pkg/syserror", "//pkg/tcpip/link/sniffer", "//pkg/urpc", ], @@ -39,7 +43,7 @@ go_test( name = "control_test", size = "small", srcs = ["proc_test.go"], - embed = [":control"], + library = ":control", deps = [ "//pkg/log", "//pkg/sentry/kernel/time", diff --git a/pkg/sentry/control/pprof.go b/pkg/sentry/control/pprof.go index 1f78d54a2..663e51989 100644 --- a/pkg/sentry/control/pprof.go +++ b/pkg/sentry/control/pprof.go @@ -19,9 +19,10 @@ import ( "runtime" "runtime/pprof" "runtime/trace" - "sync" "gvisor.dev/gvisor/pkg/fd" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/urpc" ) @@ -56,6 +57,9 @@ type Profile struct { // traceFile is the current execution trace output file. traceFile *fd.FD + + // Kernel is the kernel under profile. + Kernel *kernel.Kernel } // StartCPUProfile is an RPC stub which starts recording the CPU profile in a @@ -113,9 +117,9 @@ func (p *Profile) HeapProfile(o *ProfileOpts, _ *struct{}) error { return nil } -// Goroutine is an RPC stub which dumps out the stack trace for all running -// goroutines. -func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error { +// GoroutineProfile is an RPC stub which dumps out the stack trace for all +// running goroutines. +func (p *Profile) GoroutineProfile(o *ProfileOpts, _ *struct{}) error { if len(o.FilePayload.Files) < 1 { return errNoOutput } @@ -127,6 +131,34 @@ func (p *Profile) Goroutine(o *ProfileOpts, _ *struct{}) error { return nil } +// BlockProfile is an RPC stub which dumps out the stack trace that led to +// blocking on synchronization primitives. +func (p *Profile) BlockProfile(o *ProfileOpts, _ *struct{}) error { + if len(o.FilePayload.Files) < 1 { + return errNoOutput + } + output := o.FilePayload.Files[0] + defer output.Close() + if err := pprof.Lookup("block").WriteTo(output, 0); err != nil { + return err + } + return nil +} + +// MutexProfile is an RPC stub which dumps out the stack trace of holders of +// contended mutexes. +func (p *Profile) MutexProfile(o *ProfileOpts, _ *struct{}) error { + if len(o.FilePayload.Files) < 1 { + return errNoOutput + } + output := o.FilePayload.Files[0] + defer output.Close() + if err := pprof.Lookup("mutex").WriteTo(output, 0); err != nil { + return err + } + return nil +} + // StartTrace is an RPC stub which starts collection of an execution trace. func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error { if len(o.FilePayload.Files) < 1 { @@ -147,6 +179,9 @@ func (p *Profile) StartTrace(o *ProfileOpts, _ *struct{}) error { return err } + // Ensure all trace contexts are registered. + p.Kernel.RebuildTraceContexts() + p.traceFile = output return nil } @@ -158,9 +193,15 @@ func (p *Profile) StopTrace(_, _ *struct{}) error { defer p.mu.Unlock() if p.traceFile == nil { - return errors.New("Execution tracing not start") + return errors.New("Execution tracing not started") } + // Similarly to the case above, if tasks have not ended traces, we will + // lose information. Thus we need to rebuild the tasks in order to have + // complete information. This will not lose information if multiple + // traces are overlapping. + p.Kernel.RebuildTraceContexts() + trace.Stop() p.traceFile.Close() p.traceFile = nil diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index c35faeb4c..b51fb3959 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -18,19 +18,26 @@ import ( "bytes" "encoding/json" "fmt" + "path" "sort" "strings" "text/tabwriter" "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/urpc" ) @@ -60,6 +67,12 @@ type ExecArgs struct { // process's MountNamespace. MountNamespace *fs.MountNamespace + // MountNamespaceVFS2 is the mount namespace to execute the new process in. + // A reference on MountNamespace must be held for the lifetime of the + // ExecArgs. If MountNamespace is nil, it will default to the init + // process's MountNamespace. + MountNamespaceVFS2 *vfs.MountNamespace + // WorkingDirectory defines the working directory for the new process. WorkingDirectory string `json:"wd"` @@ -150,6 +163,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI Envv: args.Envv, WorkingDirectory: args.WorkingDirectory, MountNamespace: args.MountNamespace, + MountNamespaceVFS2: args.MountNamespaceVFS2, Credentials: creds, FDTable: fdTable, Umask: 0022, @@ -166,24 +180,51 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // be donated to the new process in CreateProcess. initArgs.MountNamespace.IncRef() } + if initArgs.MountNamespaceVFS2 != nil { + // initArgs must hold a reference on MountNamespaceVFS2, which will + // be donated to the new process in CreateProcess. + initArgs.MountNamespaceVFS2.IncRef() + } ctx := initArgs.NewContext(proc.Kernel) if initArgs.Filename == "" { - // Get the full path to the filename from the PATH env variable. - paths := fs.GetPath(initArgs.Envv) - mns := initArgs.MountNamespace - if mns == nil { - mns = proc.Kernel.GlobalInit().Leader().MountNamespace() - } - f, err := mns.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) - if err != nil { - return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + if kernel.VFS2Enabled { + // Get the full path to the filename from the PATH env variable. + if initArgs.MountNamespaceVFS2 == nil { + // Set initArgs so that 'ctx' returns the namespace. + // + // MountNamespaceVFS2 adds a reference to the namespace, which is + // transferred to the new process. + initArgs.MountNamespaceVFS2 = proc.Kernel.GlobalInit().Leader().MountNamespaceVFS2() + } + + paths := fs.GetPath(initArgs.Envv) + vfsObj := proc.Kernel.VFS() + file, err := ResolveExecutablePath(ctx, vfsObj, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + if err != nil { + return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + } + initArgs.File = fsbridge.NewVFSFile(file) + } else { + // Get the full path to the filename from the PATH env variable. + paths := fs.GetPath(initArgs.Envv) + if initArgs.MountNamespace == nil { + // Set initArgs so that 'ctx' returns the namespace. + initArgs.MountNamespace = proc.Kernel.GlobalInit().Leader().MountNamespace() + + // initArgs must hold a reference on MountNamespace, which will + // be donated to the new process in CreateProcess. + initArgs.MountNamespaceVFS2.IncRef() + } + f, err := initArgs.MountNamespace.ResolveExecutablePath(ctx, initArgs.WorkingDirectory, initArgs.Argv[0], paths) + if err != nil { + return nil, 0, nil, fmt.Errorf("error finding executable %q in PATH %v: %v", initArgs.Argv[0], paths, err) + } + initArgs.Filename = f } - initArgs.Filename = f } - mounter := fs.FileOwnerFromContext(ctx) - + // TODO(gvisor.dev/issue/1623): Use host FD when supported in VFS2. var ttyFile *fs.File for appFD, hostFile := range args.FilePayload.Files { var appFile *fs.File @@ -192,7 +233,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI // Import the file as a host TTY file. if ttyFile == nil { var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, true /* isTTY */) + appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), true /* isTTY */) if err != nil { return nil, 0, nil, err } @@ -211,7 +252,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } else { // Import the file as a regular host file. var err error - appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), mounter, false /* isTTY */) + appFile, err = host.ImportFile(ctx, int(hostFile.Fd()), false /* isTTY */) if err != nil { return nil, 0, nil, err } @@ -268,14 +309,17 @@ func (proc *Proc) Ps(args *PsArgs, out *string) error { } // Process contains information about a single process in a Sandbox. -// TODO(b/117881927): Implement TTY field. type Process struct { UID auth.KUID `json:"uid"` PID kernel.ThreadID `json:"pid"` // Parent PID - PPID kernel.ThreadID `json:"ppid"` + PPID kernel.ThreadID `json:"ppid"` + Threads []kernel.ThreadID `json:"threads"` // Processor utilization C int32 `json:"c"` + // TTY name of the process. Will be of the form "pts/N" if there is a + // TTY, or "?" if there is not. + TTY string `json:"tty"` // Start time STime string `json:"stime"` // CPU time @@ -285,18 +329,19 @@ type Process struct { } // ProcessListToTable prints a table with the following format: -// UID PID PPID C STIME TIME CMD -// 0 1 0 0 14:04 505262ns tail +// UID PID PPID C TTY STIME TIME CMD +// 0 1 0 0 pty/4 14:04 505262ns tail func ProcessListToTable(pl []*Process) string { var buf bytes.Buffer tw := tabwriter.NewWriter(&buf, 10, 1, 3, ' ', 0) - fmt.Fprint(tw, "UID\tPID\tPPID\tC\tSTIME\tTIME\tCMD") + fmt.Fprint(tw, "UID\tPID\tPPID\tC\tTTY\tSTIME\tTIME\tCMD") for _, d := range pl { - fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s", + fmt.Fprintf(tw, "\n%d\t%d\t%d\t%d\t%s\t%s\t%s\t%s", d.UID, d.PID, d.PPID, d.C, + d.TTY, d.STime, d.Time, d.Cmd) @@ -307,7 +352,7 @@ func ProcessListToTable(pl []*Process) string { // ProcessListToJSON will return the JSON representation of ps. func ProcessListToJSON(pl []*Process) (string, error) { - b, err := json.Marshal(pl) + b, err := json.MarshalIndent(pl, "", " ") if err != nil { return "", fmt.Errorf("couldn't marshal process list %v: %v", pl, err) } @@ -334,7 +379,9 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ts := k.TaskSet() now := k.RealtimeClock().Now() for _, tg := range ts.Root.ThreadGroups() { - pid := tg.PIDNamespace().IDOfThreadGroup(tg) + pidns := tg.PIDNamespace() + pid := pidns.IDOfThreadGroup(tg) + // If tg has already been reaped ignore it. if pid == 0 { continue @@ -345,16 +392,19 @@ func Processes(k *kernel.Kernel, containerID string, out *[]*Process) error { ppid := kernel.ThreadID(0) if p := tg.Leader().Parent(); p != nil { - ppid = p.PIDNamespace().IDOfThreadGroup(p.ThreadGroup()) + ppid = pidns.IDOfThreadGroup(p.ThreadGroup()) } + threads := tg.MemberIDs(pidns) *out = append(*out, &Process{ - UID: tg.Leader().Credentials().EffectiveKUID, - PID: pid, - PPID: ppid, - STime: formatStartTime(now, tg.Leader().StartTime()), - C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), - Time: tg.CPUStats().SysTime.String(), - Cmd: tg.Leader().Name(), + UID: tg.Leader().Credentials().EffectiveKUID, + PID: pid, + PPID: ppid, + Threads: threads, + STime: formatStartTime(now, tg.Leader().StartTime()), + C: percentCPU(tg.CPUStats(), tg.Leader().StartTime(), now), + Time: tg.CPUStats().SysTime.String(), + Cmd: tg.Leader().Name(), + TTY: ttyName(tg.TTY()), }) } sort.Slice(*out, func(i, j int) bool { return (*out)[i].PID < (*out)[j].PID }) @@ -395,3 +445,74 @@ func percentCPU(stats usage.CPUStats, startTime, now ktime.Time) int32 { } return int32(percentCPU) } + +func ttyName(tty *kernel.TTY) string { + if tty == nil { + return "?" + } + return fmt.Sprintf("pts/%d", tty.Index) +} + +// ResolveExecutablePath resolves the given executable name given a set of +// paths that might contain it. +func ResolveExecutablePath(ctx context.Context, vfsObj *vfs.VirtualFilesystem, wd, name string, paths []string) (*vfs.FileDescription, error) { + root := vfs.RootFromContext(ctx) + defer root.DecRef() + creds := auth.CredentialsFromContext(ctx) + + // Absolute paths can be used directly. + if path.IsAbs(name) { + return openExecutable(ctx, vfsObj, creds, root, name) + } + + // Paths with '/' in them should be joined to the working directory, or + // to the root if working directory is not set. + if strings.IndexByte(name, '/') > 0 { + if len(wd) == 0 { + wd = "/" + } + if !path.IsAbs(wd) { + return nil, fmt.Errorf("working directory %q must be absolute", wd) + } + return openExecutable(ctx, vfsObj, creds, root, path.Join(wd, name)) + } + + // Otherwise, we must lookup the name in the paths, starting from the + // calling context's root directory. + for _, p := range paths { + if !path.IsAbs(p) { + // Relative paths aren't safe, no one should be using them. + log.Warningf("Skipping relative path %q in $PATH", p) + continue + } + + binPath := path.Join(p, name) + f, err := openExecutable(ctx, vfsObj, creds, root, binPath) + if err != nil { + return nil, err + } + if f == nil { + continue // Not found/no access. + } + return f, nil + } + return nil, syserror.ENOENT +} + +func openExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry, path string) (*vfs.FileDescription, error) { + pop := vfs.PathOperation{ + Root: root, + Start: root, // binPath is absolute, Start can be anything. + Path: fspath.Parse(path), + FollowFinalSymlink: true, + } + opts := &vfs.OpenOptions{ + Flags: linux.O_RDONLY, + FileExec: true, + } + f, err := vfsObj.OpenAt(ctx, creds, &pop, opts) + if err == syserror.ENOENT || err == syserror.EACCES { + return nil, nil + } + return f, err +} diff --git a/pkg/sentry/control/proc_test.go b/pkg/sentry/control/proc_test.go index d8ada2694..0a88459b2 100644 --- a/pkg/sentry/control/proc_test.go +++ b/pkg/sentry/control/proc_test.go @@ -34,7 +34,7 @@ func TestProcessListTable(t *testing.T) { }{ { pl: []*Process{}, - expected: "UID PID PPID C STIME TIME CMD", + expected: "UID PID PPID C TTY STIME TIME CMD", }, { pl: []*Process{ @@ -43,6 +43,7 @@ func TestProcessListTable(t *testing.T) { PID: 0, PPID: 0, C: 0, + TTY: "?", STime: "0", Time: "0", Cmd: "zero", @@ -52,14 +53,15 @@ func TestProcessListTable(t *testing.T) { PID: 1, PPID: 1, C: 1, + TTY: "pts/4", STime: "1", Time: "1", Cmd: "one", }, }, - expected: `UID PID PPID C STIME TIME CMD -0 0 0 0 0 0 zero -1 1 1 1 1 1 one`, + expected: `UID PID PPID C TTY STIME TIME CMD +0 0 0 0 ? 0 0 zero +1 1 1 1 pts/4 1 1 one`, }, } |