vm: 'expression->derivation-in-linux-vm' always returns a native build.
[jackhill/guix/guix.git] / gnu / build / linux-boot.scm
1 ;;; GNU Guix --- Functional package management for GNU
2 ;;; Copyright © 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020 Ludovic Courtès <ludo@gnu.org>
3 ;;; Copyright © 2017 Mathieu Othacehe <m.othacehe@gmail.com>
4 ;;; Copyright © 2019 Guillaume Le Vaillant <glv@posteo.net>
5 ;;;
6 ;;; This file is part of GNU Guix.
7 ;;;
8 ;;; GNU Guix is free software; you can redistribute it and/or modify it
9 ;;; under the terms of the GNU General Public License as published by
10 ;;; the Free Software Foundation; either version 3 of the License, or (at
11 ;;; your option) any later version.
12 ;;;
13 ;;; GNU Guix is distributed in the hope that it will be useful, but
14 ;;; WITHOUT ANY WARRANTY; without even the implied warranty of
15 ;;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 ;;; GNU General Public License for more details.
17 ;;;
18 ;;; You should have received a copy of the GNU General Public License
19 ;;; along with GNU Guix. If not, see <http://www.gnu.org/licenses/>.
20
21 (define-module (gnu build linux-boot)
22 #:use-module (rnrs io ports)
23 #:use-module (system repl error-handling)
24 #:autoload (system repl repl) (start-repl)
25 #:use-module (srfi srfi-1)
26 #:use-module (srfi srfi-9)
27 #:use-module (srfi srfi-26)
28 #:use-module (ice-9 match)
29 #:use-module (ice-9 rdelim)
30 #:use-module (ice-9 regex)
31 #:use-module (ice-9 ftw)
32 #:use-module (guix build utils)
33 #:use-module ((guix build syscalls)
34 #:hide (file-system-type))
35 #:use-module (gnu build linux-modules)
36 #:use-module (gnu build file-systems)
37 #:use-module (gnu system file-systems)
38 #:export (mount-essential-file-systems
39 linux-command-line
40 find-long-option
41 find-long-options
42 make-essential-device-nodes
43 make-hurd-device-nodes
44 make-static-device-nodes
45 configure-qemu-networking
46
47 device-number
48 boot-system))
49
50 ;;; Commentary:
51 ;;;
52 ;;; Utility procedures useful in a Linux initial RAM disk (initrd). Note that
53 ;;; many of these use procedures not yet available in vanilla Guile (`mount',
54 ;;; `load-linux-module', etc.); these are provided by a Guile patch used in
55 ;;; the GNU distribution.
56 ;;;
57 ;;; Code:
58
59 (define* (mount-essential-file-systems #:key (root "/"))
60 "Mount /dev, /proc, and /sys under ROOT."
61 (define (scope dir)
62 (string-append root
63 (if (string-suffix? "/" root)
64 ""
65 "/")
66 dir))
67
68 (unless (file-exists? (scope "proc"))
69 (mkdir (scope "proc")))
70 (mount "none" (scope "proc") "proc")
71
72 (unless (file-exists? (scope "dev"))
73 (mkdir (scope "dev")))
74 (mount "none" (scope "dev") "devtmpfs")
75
76 (unless (file-exists? (scope "sys"))
77 (mkdir (scope "sys")))
78 (mount "none" (scope "sys") "sysfs"))
79
80 (define (move-essential-file-systems root)
81 "Move currently mounted essential file systems to ROOT."
82 (for-each (lambda (dir)
83 (let ((target (string-append root dir)))
84 (unless (file-exists? target)
85 (mkdir target))
86 (mount dir target "" MS_MOVE)))
87 '("/dev" "/proc" "/sys")))
88
89 (define (linux-command-line)
90 "Return the Linux kernel command line as a list of strings."
91 (string-tokenize
92 (call-with-input-file "/proc/cmdline"
93 get-string-all)))
94
95 (define (find-long-option option arguments)
96 "Find OPTION among ARGUMENTS, where OPTION is something like \"--load\".
97 Return the value associated with OPTION, or #f on failure."
98 (let ((opt (string-append option "=")))
99 (and=> (find (cut string-prefix? opt <>)
100 arguments)
101 (lambda (arg)
102 (substring arg (+ 1 (string-index arg #\=)))))))
103
104 (define (find-long-options option arguments)
105 "Find OPTIONs among ARGUMENTS, where OPTION is something like \"console\".
106 Return the values associated with OPTIONs as a list, or the empty list if
107 OPTION doesn't appear in ARGUMENTS."
108 (let ((opt (string-append option "=")))
109 (filter-map (lambda (arg)
110 (and (string-prefix? opt arg)
111 (substring arg (+ 1 (string-index arg #\=)))))
112 arguments)))
113
114 (define* (make-disk-device-nodes base major #:optional (minor 0))
115 "Make the block device nodes around BASE (something like \"/root/dev/sda\")
116 with the given MAJOR number, starting with MINOR."
117 (mknod base 'block-special #o644 (device-number major minor))
118 (let loop ((i 1))
119 (when (< i 16)
120 (mknod (string-append base (number->string i))
121 'block-special #o644 (device-number major (+ minor i)))
122 (loop (+ i 1)))))
123
124 ;; Representation of a /dev node.
125 (define-record-type <device-node>
126 (device-node name type major minor module)
127 device-node?
128 (name device-node-name)
129 (type device-node-type)
130 (major device-node-major)
131 (minor device-node-minor)
132 (module device-node-module))
133
134 (define (read-static-device-nodes port)
135 "Read from PORT a list of <device-node> written in the format used by
136 /lib/modules/*/*.devname files."
137 (let loop ((line (read-line port)))
138 (if (eof-object? line)
139 '()
140 (match (string-split line #\space)
141 (((? (cut string-prefix? "#" <>)) _ ...)
142 (loop (read-line port)))
143 ((module-name device-name device-spec)
144 (let* ((device-parts
145 (string-match "([bc])([0-9][0-9]*):([0-9][0-9]*)"
146 device-spec))
147 (type-string (match:substring device-parts 1))
148 (type (match type-string
149 ("c" 'char-special)
150 ("b" 'block-special)))
151 (major-string (match:substring device-parts 2))
152 (major (string->number major-string 10))
153 (minor-string (match:substring device-parts 3))
154 (minor (string->number minor-string 10)))
155 (cons (device-node device-name type major minor module-name)
156 (loop (read-line port)))))
157 (_
158 (begin
159 (format (current-error-port)
160 "read-static-device-nodes: ignored devname line '~a'~%" line)
161 (loop (read-line port))))))))
162
163 (define* (mkdir-p* dir #:optional (mode #o755))
164 "This is a variant of 'mkdir-p' that works around
165 <http://bugs.gnu.org/24659> by passing MODE explicitly in each 'mkdir' call."
166 (define absolute?
167 (string-prefix? "/" dir))
168
169 (define not-slash
170 (char-set-complement (char-set #\/)))
171
172 (let loop ((components (string-tokenize dir not-slash))
173 (root (if absolute?
174 ""
175 ".")))
176 (match components
177 ((head tail ...)
178 (let ((path (string-append root "/" head)))
179 (catch 'system-error
180 (lambda ()
181 (mkdir path mode)
182 (loop tail path))
183 (lambda args
184 (if (= EEXIST (system-error-errno args))
185 (loop tail path)
186 (apply throw args))))))
187 (() #t))))
188
189 (define (report-system-error name . args)
190 "Report a system error for the file NAME."
191 (let ((errno (system-error-errno args)))
192 (format (current-error-port) "could not create '~a': ~a~%" name
193 (strerror errno))))
194
195 ;; Catch a system-error, log it and don't die from it.
196 (define-syntax-rule (catch-system-error name exp)
197 (catch 'system-error
198 (lambda ()
199 exp)
200 (lambda args
201 (apply report-system-error name args))))
202
203 ;; Create a device node like the <device-node> passed here on the file system.
204 (define create-device-node
205 (match-lambda
206 (($ <device-node> xname type major minor module)
207 (let ((name (string-append "/dev/" xname)))
208 (mkdir-p* (dirname name))
209 (catch-system-error name
210 (mknod name type #o600 (device-number major minor)))))))
211
212 (define* (make-static-device-nodes linux-release-module-directory)
213 "Create static device nodes required by the given Linux release.
214 This is required in order to solve a chicken-or-egg problem:
215 The Linux kernel has a feature to autoload modules when a device is first
216 accessed.
217 And udev has a feature to set the permissions of static nodes correctly
218 when it is starting up and also to automatically create nodes when hardware
219 is hotplugged. That leaves universal device files which are not linked to
220 one specific hardware device. These we have to create."
221 (let ((devname-name (string-append linux-release-module-directory "/"
222 "modules.devname")))
223 (for-each create-device-node
224 (call-with-input-file devname-name
225 read-static-device-nodes))))
226
227 (define* (make-essential-device-nodes #:optional (root "/"))
228 "Make essential device nodes under ROOT/dev."
229 ;; The hand-made devtmpfs/udev!
230
231 (define (scope dir)
232 (string-append root
233 (if (string-suffix? "/" root)
234 ""
235 "/")
236 dir))
237
238 (unless (file-exists? (scope "dev"))
239 (mkdir (scope "dev")))
240
241 ;; Make the device nodes for SCSI disks.
242 (make-disk-device-nodes (scope "dev/sda") 8)
243 (make-disk-device-nodes (scope "dev/sdb") 8 16)
244 (make-disk-device-nodes (scope "dev/sdc") 8 32)
245 (make-disk-device-nodes (scope "dev/sdd") 8 48)
246
247 ;; SCSI CD-ROM devices (aka. "/dev/sr0" etc.).
248 (mknod (scope "dev/scd0") 'block-special #o644 (device-number 11 0))
249 (mknod (scope "dev/scd1") 'block-special #o644 (device-number 11 1))
250
251 ;; The virtio (para-virtualized) block devices, as supported by QEMU/KVM.
252 (make-disk-device-nodes (scope "dev/vda") 252)
253
254 ;; Memory (used by Xorg's VESA driver.)
255 (mknod (scope "dev/mem") 'char-special #o640 (device-number 1 1))
256 (mknod (scope "dev/kmem") 'char-special #o640 (device-number 1 2))
257
258 ;; Inputs (used by Xorg.)
259 (unless (file-exists? (scope "dev/input"))
260 (mkdir (scope "dev/input")))
261 (mknod (scope "dev/input/mice") 'char-special #o640 (device-number 13 63))
262 (mknod (scope "dev/input/mouse0") 'char-special #o640 (device-number 13 32))
263 (mknod (scope "dev/input/event0") 'char-special #o640 (device-number 13 64))
264
265 ;; System console. This node is magically created by the kernel on the
266 ;; initrd's root, so don't try to create it in that case.
267 (unless (string=? root "/")
268 (mknod (scope "dev/console") 'char-special #o600
269 (device-number 5 1)))
270
271 ;; TTYs.
272 (mknod (scope "dev/tty") 'char-special #o600
273 (device-number 5 0))
274 (chmod (scope "dev/tty") #o666)
275 (let loop ((n 0))
276 (and (< n 50)
277 (let ((name (format #f "dev/tty~a" n)))
278 (mknod (scope name) 'char-special #o600
279 (device-number 4 n))
280 (loop (+ 1 n)))))
281
282 ;; Serial line.
283 (mknod (scope "dev/ttyS0") 'char-special #o660
284 (device-number 4 64))
285
286 ;; Pseudo ttys.
287 (mknod (scope "dev/ptmx") 'char-special #o666
288 (device-number 5 2))
289 (chmod (scope "dev/ptmx") #o666)
290
291 ;; Create /dev/pts; it will be mounted later, at boot time.
292 (unless (file-exists? (scope "dev/pts"))
293 (mkdir (scope "dev/pts")))
294
295 ;; Rendez-vous point for syslogd.
296 (mknod (scope "dev/log") 'socket #o666 0)
297 (mknod (scope "dev/kmsg") 'char-special #o600 (device-number 1 11))
298
299 ;; Other useful nodes, notably relied on by guix-daemon.
300 (for-each (match-lambda
301 ((file major minor)
302 (mknod (scope file) 'char-special #o666
303 (device-number major minor))
304 (chmod (scope file) #o666)))
305 '(("dev/null" 1 3)
306 ("dev/zero" 1 5)
307 ("dev/full" 1 7)
308 ("dev/random" 1 8)
309 ("dev/urandom" 1 9)))
310
311 (symlink "/proc/self/fd" (scope "dev/fd"))
312 (symlink "/proc/self/fd/0" (scope "dev/stdin"))
313 (symlink "/proc/self/fd/1" (scope "dev/stdout"))
314 (symlink "/proc/self/fd/2" (scope "dev/stderr"))
315
316 ;; Loopback devices.
317 (let loop ((i 0))
318 (when (< i 8)
319 (mknod (scope (string-append "dev/loop" (number->string i)))
320 'block-special #o660
321 (device-number 7 i))
322 (loop (+ 1 i))))
323
324 ;; File systems in user space (FUSE).
325 (mknod (scope "dev/fuse") 'char-special #o666 (device-number 10 229)))
326
327 (define* (make-hurd-device-nodes #:optional (root "/"))
328 "Make some of the nodes needed on GNU/Hurd."
329 (define (scope dir)
330 (string-append root
331 (if (string-suffix? "/" root)
332 ""
333 "/")
334 dir))
335
336 (mkdir (scope "dev"))
337 (for-each (lambda (file)
338 (call-with-output-file (scope file)
339 (lambda (port)
340 (chmod port #o666))))
341 '("dev/null"
342 "dev/zero"
343 "dev/full"
344 "dev/random"
345 "dev/urandom"))
346 ;; Don't create /dev/console, /dev/vcs, etc.: they are created by
347 ;; console-run on first boot.
348
349 (mkdir (scope "servers"))
350 (mkdir (scope "servers/socket"))
351 ;; Don't create /servers/socket/1 & co: runsystem does that on first boot.
352
353 ;; TODO: Set the 'gnu.translator' extended attribute for passive translator
354 ;; settings?
355 )
356
357 (define %host-qemu-ipv4-address
358 (inet-pton AF_INET "10.0.2.10"))
359
360 (define* (configure-qemu-networking #:optional (interface "eth0"))
361 "Setup the INTERFACE network interface and /etc/resolv.conf according to
362 QEMU's default networking settings (see net/slirp.c in QEMU for default
363 networking values.) Return #t if INTERFACE is up, #f otherwise."
364 (display "configuring QEMU networking...\n")
365 (let* ((sock (socket AF_INET SOCK_STREAM 0))
366 (address (make-socket-address AF_INET %host-qemu-ipv4-address 0))
367 (flags (network-interface-flags sock interface)))
368 (set-network-interface-address sock interface address)
369 (set-network-interface-flags sock interface (logior flags IFF_UP))
370
371 ;; Hello! We used to create /etc/resolv.conf here, with "nameserver
372 ;; 10.0.2.3\n". However, with Linux-libre 3.16, we're getting ENOSPC.
373 ;; And since it's actually unnecessary, it's gone.
374
375 (logand (network-interface-flags sock interface) IFF_UP)))
376
377 (define (device-number major minor)
378 "Return the device number for the device with MAJOR and MINOR, for use as
379 the last argument of `mknod'."
380 (+ (* major 256) minor))
381
382 (define (pidof program)
383 "Return the PID of the first presumed instance of PROGRAM."
384 (let ((program (basename program)))
385 (find (lambda (pid)
386 (let ((exe (format #f "/proc/~a/exe" pid)))
387 (and=> (false-if-exception (readlink exe))
388 (compose (cut string=? program <>) basename))))
389 (filter-map string->number (scandir "/proc")))))
390
391 (define* (mount-root-file-system root type
392 #:key volatile-root? (flags 0) options)
393 "Mount the root file system of type TYPE at device ROOT. If VOLATILE-ROOT? is
394 true, mount ROOT read-only and make it an overlay with a writable tmpfs using
395 the kernel built-in overlayfs. FLAGS and OPTIONS indicates the options to use
396 to mount ROOT, and behave the same as for the `mount' procedure."
397
398 (if volatile-root?
399 (begin
400 (mkdir-p "/real-root")
401 (mount root "/real-root" type (logior MS_RDONLY flags) options)
402 (mkdir-p "/rw-root")
403 (mount "none" "/rw-root" "tmpfs")
404
405 ;; Create the upperdir and the workdir of the overlayfs
406 (mkdir-p "/rw-root/upper")
407 (mkdir-p "/rw-root/work")
408
409 ;; We want read-write /dev nodes.
410 (mkdir-p "/rw-root/upper/dev")
411 (mount "none" "/rw-root/upper/dev" "devtmpfs")
412
413 ;; Make /root an overlay of the tmpfs and the actual root.
414 (mount "none" "/root" "overlay" 0
415 "lowerdir=/real-root,upperdir=/rw-root/upper,workdir=/rw-root/work"))
416 (begin
417 (check-file-system root type)
418 (mount root "/root" type flags options)))
419
420 ;; Make sure /root/etc/mtab is a symlink to /proc/self/mounts.
421 (false-if-exception
422 (delete-file "/root/etc/mtab"))
423 (mkdir-p "/root/etc")
424 (symlink "/proc/self/mounts" "/root/etc/mtab"))
425
426 (define (switch-root root)
427 "Switch to ROOT as the root file system, in a way similar to what
428 util-linux' switch_root(8) does."
429 (move-essential-file-systems root)
430 (chdir root)
431
432 ;; Since we're about to 'rm -rf /', try to make sure we're on an initrd.
433 ;; TODO: Use 'statfs' to check the fs type, like klibc does.
434 (when (or (not (file-exists? "/init")) (directory-exists? "/home"))
435 (format (current-error-port)
436 "The root file system is probably not an initrd; \
437 bailing out.~%root contents: ~s~%" (scandir "/"))
438 (force-output (current-error-port))
439 (exit 1))
440
441 ;; Delete files from the old root, without crossing mount points (assuming
442 ;; there are no mount points in sub-directories.) That means we're leaving
443 ;; the empty ROOT directory behind us, but that's OK.
444 (let ((root-device (stat:dev (stat "/"))))
445 (for-each (lambda (file)
446 (unless (member file '("." ".."))
447 (let* ((file (string-append "/" file))
448 (device (stat:dev (lstat file))))
449 (when (= device root-device)
450 (delete-file-recursively file)))))
451 (scandir "/")))
452
453 ;; Make ROOT the new root.
454 (mount root "/" "" MS_MOVE)
455 (chroot ".")
456 (chdir "/")
457
458 (when (file-exists? "/dev/console")
459 ;; Close the standard file descriptors since they refer to the old
460 ;; /dev/console, and reopen them.
461 (let ((console (open-file "/dev/console" "r+b0")))
462 (for-each close-fdes '(0 1 2))
463
464 (dup2 (fileno console) 0)
465 (dup2 (fileno console) 1)
466 (dup2 (fileno console) 2)
467
468 (close-port console))))
469
470 \f
471 (define* (boot-system #:key
472 (linux-modules '())
473 linux-module-directory
474 keymap-file
475 qemu-guest-networking?
476 volatile-root?
477 pre-mount
478 (mounts '())
479 (on-error 'debug))
480 "This procedure is meant to be called from an initrd. Boot a system by
481 first loading LINUX-MODULES (a list of module names) from
482 LINUX-MODULE-DIRECTORY, then installing KEYMAP-FILE with 'loadkeys' (if
483 KEYMAP-FILE is true), then setting up QEMU guest networking if
484 QEMU-GUEST-NETWORKING? is true, calling PRE-MOUNT, mounting the file systems
485 specified in MOUNTS, and finally booting into the new root if any. The initrd
486 supports kernel command-line options '--load', '--root', and '--repl'.
487
488 Mount the root file system, specified by the '--root' command-line argument,
489 if any.
490
491 MOUNTS must be a list of <file-system> objects.
492
493 When VOLATILE-ROOT? is true, the root file system is writable but any changes
494 to it are lost.
495
496 ON-ERROR is passed to 'call-with-error-handling'; it determines what happens
497 upon error."
498 (define (root-mount-point? fs)
499 (string=? (file-system-mount-point fs) "/"))
500
501 (define (device-string->file-system-device device-string)
502 ;; The "--root=SPEC" kernel command-line option always provides a
503 ;; string, but the string can represent a device, a UUID, or a
504 ;; label. So check for all three.
505 (cond ((string-prefix? "/" device-string) device-string)
506 ((uuid device-string) => identity)
507 (else (file-system-label device-string))))
508
509 (display "Welcome, this is GNU's early boot Guile.\n")
510 (display "Use '--repl' for an initrd REPL.\n\n")
511
512 (call-with-error-handling
513 (lambda ()
514 (mount-essential-file-systems)
515 (let* ((args (linux-command-line))
516 (to-load (find-long-option "--load" args))
517 (root-fs (find root-mount-point? mounts))
518 (root-fs-type (or (and=> root-fs file-system-type)
519 "ext4"))
520 (root-fs-device (and=> root-fs file-system-device))
521 (root-fs-flags (mount-flags->bit-mask
522 (or (and=> root-fs file-system-flags)
523 '())))
524 (root-options (if root-fs
525 (file-system-options root-fs)
526 #f))
527 ;; --root takes precedence over the 'device' field of the root
528 ;; <file-system> record.
529 (root-device (or (and=> (find-long-option "--root" args)
530 device-string->file-system-device)
531 root-fs-device)))
532
533 (when (member "--repl" args)
534 (start-repl))
535
536 (display "loading kernel modules...\n")
537 (load-linux-modules-from-directory linux-modules
538 linux-module-directory)
539
540 (when keymap-file
541 (let ((status (system* "loadkeys" keymap-file)))
542 (unless (zero? status)
543 ;; Emit a warning rather than abort when we cannot load
544 ;; KEYMAP-FILE.
545 (format (current-error-port)
546 "warning: 'loadkeys' exited with status ~a~%"
547 status))))
548
549 (when qemu-guest-networking?
550 (unless (configure-qemu-networking)
551 (display "network interface is DOWN\n")))
552
553 ;; Prepare the real root file system under /root.
554 (unless (file-exists? "/root")
555 (mkdir "/root"))
556
557 (when (procedure? pre-mount)
558 ;; Do whatever actions are needed before mounting the root file
559 ;; system--e.g., installing device mappings. Error out when the
560 ;; return value is false.
561 (unless (pre-mount)
562 (error "pre-mount actions failed")))
563
564 (setenv "EXT2FS_NO_MTAB_OK" "1")
565
566 (if root-device
567 (mount-root-file-system (canonicalize-device-spec root-device)
568 root-fs-type
569 #:volatile-root? volatile-root?
570 #:flags root-fs-flags
571 #:options root-options)
572 (mount "none" "/root" "tmpfs"))
573
574 ;; Mount the specified file systems.
575 (for-each mount-file-system
576 (remove root-mount-point? mounts))
577
578 (setenv "EXT2FS_NO_MTAB_OK" #f)
579
580 (if to-load
581 (begin
582 (switch-root "/root")
583 (format #t "loading '~a'...\n" to-load)
584
585 (primitive-load to-load)
586
587 (format (current-error-port)
588 "boot program '~a' terminated, rebooting~%"
589 to-load)
590 (sleep 2)
591 (reboot))
592 (begin
593 (display "no boot file passed via '--load'\n")
594 (display "entering a warm and cozy REPL\n")
595 (start-repl)))))
596 #:on-error on-error))
597
598 ;;; linux-initrd.scm ends here