blob: 6528ab51fe1b1a2ffd100cdf7eaac541de20a9c0 [file] [log] [blame]
Lorenz Brunae0d90d2019-09-05 17:53:56 +02001// Copyright 2020 The Monogon Project Authors.
2//
3// SPDX-License-Identifier: Apache-2.0
4//
5// Licensed under the Apache License, Version 2.0 (the "License");
6// you may not use this file except in compliance with the License.
7// You may obtain a copy of the License at
8//
9// http://www.apache.org/licenses/LICENSE-2.0
10//
11// Unless required by applicable law or agreed to in writing, software
12// distributed under the License is distributed on an "AS IS" BASIS,
13// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14// See the License for the specific language governing permissions and
15// limitations under the License.
16
17package main
18
19import (
Serge Bazanskicdb8c782020-02-17 12:34:02 +010020 "context"
Lorenz Brundd8c80e2019-10-07 16:19:49 +020021 "fmt"
Serge Bazanskie803fc12022-01-25 14:58:24 +010022 "io"
Serge Bazanski1ebd1e12020-07-13 19:17:16 +020023 "net"
Lorenz Brunae0d90d2019-09-05 17:53:56 +020024 "os"
Lorenz Brunf95909d2019-09-11 19:48:26 +020025 "runtime/debug"
Lorenz Brunae0d90d2019-09-05 17:53:56 +020026
Lorenz Brunae0d90d2019-09-05 17:53:56 +020027 "golang.org/x/sys/unix"
Serge Bazanski1ebd1e12020-07-13 19:17:16 +020028 "google.golang.org/grpc"
Serge Bazanski99f47742021-08-04 20:21:42 +020029
Serge Bazanski31370b02021-01-07 16:31:14 +010030 common "source.monogon.dev/metropolis/node"
31 "source.monogon.dev/metropolis/node/core/cluster"
32 "source.monogon.dev/metropolis/node/core/localstorage"
33 "source.monogon.dev/metropolis/node/core/localstorage/declarative"
34 "source.monogon.dev/metropolis/node/core/network"
Serge Bazanski6dff6d62022-01-28 18:15:14 +010035 "source.monogon.dev/metropolis/node/core/network/hostsfile"
Serge Bazanskif9edf522021-06-17 15:57:13 +020036 "source.monogon.dev/metropolis/node/core/roleserve"
Lorenz Brune306d782021-09-01 13:01:06 +020037 timesvc "source.monogon.dev/metropolis/node/core/time"
Serge Bazanski31370b02021-01-07 16:31:14 +010038 "source.monogon.dev/metropolis/pkg/logtree"
39 "source.monogon.dev/metropolis/pkg/supervisor"
40 "source.monogon.dev/metropolis/pkg/tpm"
41 apb "source.monogon.dev/metropolis/proto/api"
Lorenz Brunae0d90d2019-09-05 17:53:56 +020042)
43
44func main() {
Lorenz Brunf95909d2019-09-11 19:48:26 +020045 defer func() {
46 if r := recover(); r != nil {
Serge Bazanskieac8f732021-10-05 23:30:37 +020047 fmt.Fprintf(os.Stderr, "\n\n")
48 fmt.Fprintf(os.Stderr, " Metropolis encountered an uncorrectable error and this node must be restarted.\n")
49 fmt.Fprintf(os.Stderr, " Core panicked: %v\n\n", r)
Lorenz Brunf95909d2019-09-11 19:48:26 +020050 debug.PrintStack()
51 }
52 unix.Sync()
Serge Bazanskieac8f732021-10-05 23:30:37 +020053 // TODO(lorenz): Switch this to Reboot when init panics are less likely.
Leopold Schabel68c58752019-11-14 21:00:59 +010054 if err := unix.Reboot(unix.LINUX_REBOOT_CMD_POWER_OFF); err != nil {
Serge Bazanskieac8f732021-10-05 23:30:37 +020055 // Best effort, nothing we can do if this fails except printing the error to the
56 // console.
Leopold Schabel68c58752019-11-14 21:00:59 +010057 panic(fmt.Sprintf("failed to halt node: %v\n", err))
58 }
Lorenz Brunf95909d2019-09-11 19:48:26 +020059 }()
Serge Bazanskic7359672020-10-30 16:38:57 +010060
Serge Bazanskie803fc12022-01-25 14:58:24 +010061 // Set up basic mounts (like /dev, /sys...).
62 if err := setupMounts(); err != nil {
63 panic(fmt.Errorf("could not set up basic mounts: %w", err))
Lorenz Brunae0d90d2019-09-05 17:53:56 +020064 }
Serge Bazanskie803fc12022-01-25 14:58:24 +010065
66 // Set up logger for Metropolis. Currently logs everything to /dev/tty0 and
67 // /dev/ttyS0.
68 lt := logtree.New()
69 for _, p := range []string{
70 "/dev/tty0", "/dev/ttyS0",
71 } {
72 f, err := os.OpenFile(p, os.O_WRONLY, 0)
73 if err != nil {
74 continue
Serge Bazanskic7359672020-10-30 16:38:57 +010075 }
Serge Bazanskie803fc12022-01-25 14:58:24 +010076 reader, err := lt.Read("", logtree.WithChildren(), logtree.WithStream())
77 if err != nil {
78 panic(fmt.Errorf("could not set up root log reader: %v", err))
79 }
80 go func(path string, f io.Writer) {
81 fmt.Fprintf(f, "\nMetropolis: this is %s. Verbose node logs follow.\n\n", path)
82 for {
83 p := <-reader.Stream
84 fmt.Fprintf(f, "%s\n", p.String())
85 }
86 }(p, f)
87 }
Serge Bazanskic7359672020-10-30 16:38:57 +010088
89 // Initial logger. Used until we get to a supervisor.
90 logger := lt.MustLeveledFor("init")
Serge Bazanski581b0bd2020-03-12 13:36:43 +010091
Serge Bazanski216fe7b2021-05-21 18:36:16 +020092 // Linux kernel default is 4096 which is far too low. Raise it to 1M which
93 // is what gVisor suggests.
Lorenz Brun878f5f92020-05-12 16:15:39 +020094 if err := unix.Setrlimit(unix.RLIMIT_NOFILE, &unix.Rlimit{Cur: 1048576, Max: 1048576}); err != nil {
Serge Bazanskic7359672020-10-30 16:38:57 +010095 logger.Fatalf("Failed to raise rlimits: %v", err)
Lorenz Brun878f5f92020-05-12 16:15:39 +020096 }
97
Serge Bazanski662b5b32020-12-21 13:49:00 +010098 logger.Info("Starting Metropolis node init")
Lorenz Brunae0d90d2019-09-05 17:53:56 +020099
Serge Bazanskic7359672020-10-30 16:38:57 +0100100 if err := tpm.Initialize(logger); err != nil {
Lorenz Brun8b786892022-01-13 14:21:16 +0100101 logger.Warningf("Failed to initialize TPM 2.0, attempting fallback to untrusted: %v", err)
Lorenz Brunae0d90d2019-09-05 17:53:56 +0200102 }
103
Serge Bazanskid8af5bf2021-03-16 13:38:29 +0100104 networkSvc := network.New()
Lorenz Brune306d782021-09-01 13:01:06 +0200105 timeSvc := timesvc.New()
Leopold Schabel68c58752019-11-14 21:00:59 +0100106
Serge Bazanski216fe7b2021-05-21 18:36:16 +0200107 // This function initializes a headless Delve if this is a debug build or
108 // does nothing if it's not
Lorenz Brun70f65b22020-07-08 17:02:47 +0200109 initializeDebugger(networkSvc)
110
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200111 // Prepare local storage.
112 root := &localstorage.Root{}
113 if err := declarative.PlaceFS(root, "/"); err != nil {
114 panic(fmt.Errorf("when placing root FS: %w", err))
115 }
116
Serge Bazanski216fe7b2021-05-21 18:36:16 +0200117 // trapdoor is a channel used to signal to the init service that a very
Serge Bazanskieac8f732021-10-05 23:30:37 +0200118 // low-level, unrecoverable failure occured.
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200119 trapdoor := make(chan struct{})
120
121 // Make context for supervisor. We cancel it when we reach the trapdoor.
122 ctxS, ctxC := context.WithCancel(context.Background())
123
Serge Bazanski216fe7b2021-05-21 18:36:16 +0200124 // Start root initialization code as a supervisor one-shot runnable. This
125 // means waiting for the network, starting the cluster manager, and then
126 // starting all services related to the node's roles.
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200127 // TODO(q3k): move this to a separate 'init' service.
Serge Bazanskic7359672020-10-30 16:38:57 +0100128 supervisor.New(ctxS, func(ctx context.Context) error {
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200129 // Start storage and network - we need this to get anything else done.
130 if err := root.Start(ctx); err != nil {
131 return fmt.Errorf("cannot start root FS: %w", err)
132 }
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100133 if err := supervisor.Run(ctx, "network", networkSvc.Run); err != nil {
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200134 return fmt.Errorf("when starting network: %w", err)
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100135 }
Lorenz Brune306d782021-09-01 13:01:06 +0200136 if err := supervisor.Run(ctx, "time", timeSvc.Run); err != nil {
137 return fmt.Errorf("when starting time: %w", err)
138 }
Lorenz Brunf95909d2019-09-11 19:48:26 +0200139
Serge Bazanski6dff6d62022-01-28 18:15:14 +0100140 // Start the role service. The role service connects to the curator and runs
141 // all node-specific role code (eg. Kubernetes services).
142 // supervisor.Logger(ctx).Infof("Starting role service...")
143 rs := roleserve.New(roleserve.Config{
144 StorageRoot: root,
145 Network: networkSvc,
146 })
147 if err := supervisor.Run(ctx, "role", rs.Run); err != nil {
148 close(trapdoor)
149 return fmt.Errorf("failed to start role service: %w", err)
150 }
151
152 // Start the hostsfile service.
153 hostsfileSvc := hostsfile.Service{
154 Config: hostsfile.Config{
155 Roleserver: rs,
156 Network: networkSvc,
157 Ephemeral: &root.Ephemeral,
158 },
159 }
160 if err := supervisor.Run(ctx, "hostsfile", hostsfileSvc.Run); err != nil {
161 close(trapdoor)
162 return fmt.Errorf("failed to start hostsfile service: %w", err)
163 }
164
Serge Bazanski216fe7b2021-05-21 18:36:16 +0200165 // Start cluster manager. This kicks off cluster membership machinery,
166 // which will either start a new cluster, enroll into one or join one.
Serge Bazanski6dff6d62022-01-28 18:15:14 +0100167 m := cluster.NewManager(root, networkSvc, rs)
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200168 if err := supervisor.Run(ctx, "enrolment", m.Run); err != nil {
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200169 close(trapdoor)
Serge Bazanski6dff6d62022-01-28 18:15:14 +0100170 return fmt.Errorf("when starting enrolment: %w", err)
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200171 }
172
173 // Start the node debug service.
Serge Bazanskif9edf522021-06-17 15:57:13 +0200174 supervisor.Logger(ctx).Infof("Starting debug service...")
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200175 dbg := &debugService{
Serge Bazanski158e9a42021-08-17 17:04:54 +0200176 roleserve: rs,
177 logtree: lt,
178 traceLock: make(chan struct{}, 1),
Lorenz Brun9d6c4c72021-07-20 21:16:27 +0200179 ephemeralVolume: &root.Ephemeral.Containerd,
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200180 }
181 dbgSrv := grpc.NewServer()
182 apb.RegisterNodeDebugServiceServer(dbgSrv, dbg)
183 dbgLis, err := net.Listen("tcp", fmt.Sprintf(":%d", common.DebugServicePort))
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100184 if err != nil {
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200185 return fmt.Errorf("failed to listen on debug service: %w", err)
186 }
187 if err := supervisor.Run(ctx, "debug", supervisor.GRPCServer(dbgSrv, dbgLis, false)); err != nil {
188 return fmt.Errorf("failed to start debug service: %w", err)
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100189 }
Lorenz Brunae0d90d2019-09-05 17:53:56 +0200190
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100191 supervisor.Signal(ctx, supervisor.SignalHealthy)
Serge Bazanski1ebd1e12020-07-13 19:17:16 +0200192 supervisor.Signal(ctx, supervisor.SignalDone)
193 return nil
Serge Bazanskic7359672020-10-30 16:38:57 +0100194 }, supervisor.WithExistingLogtree(lt))
Serge Bazanskib1b742f2020-03-24 13:58:19 +0100195
Serge Bazanskieac8f732021-10-05 23:30:37 +0200196 <-trapdoor
197 logger.Infof("Trapdoor closed, exiting core.")
198 ctxC()
Serge Bazanski57b43752020-07-13 19:17:48 +0200199}