Commit | Line | Data |
---|---|---|
88840f02 | 1 | ;;; GNU Guix --- Functional package management for GNU |
b97c95eb | 2 | ;;; Copyright © 2013, 2014 Ludovic Courtès <ludo@gnu.org> |
88840f02 LC |
3 | ;;; |
4 | ;;; This file is part of GNU Guix. | |
5 | ;;; | |
6 | ;;; GNU Guix is free software; you can redistribute it and/or modify it | |
7 | ;;; under the terms of the GNU General Public License as published by | |
8 | ;;; the Free Software Foundation; either version 3 of the License, or (at | |
9 | ;;; your option) any later version. | |
10 | ;;; | |
11 | ;;; GNU Guix is distributed in the hope that it will be useful, but | |
12 | ;;; WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | ;;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | ;;; GNU General Public License for more details. | |
15 | ;;; | |
16 | ;;; You should have received a copy of the GNU General Public License | |
17 | ;;; along with GNU Guix. If not, see <http://www.gnu.org/licenses/>. | |
18 | ||
19 | (define-module (guix build linux-initrd) | |
20 | #:use-module (rnrs io ports) | |
21 | #:use-module (system foreign) | |
d4254711 LC |
22 | #:autoload (system repl repl) (start-repl) |
23 | #:autoload (system base compile) (compile-file) | |
24 | #:use-module (srfi srfi-1) | |
25 | #:use-module (srfi srfi-26) | |
26 | #:use-module (ice-9 match) | |
44ddf33e | 27 | #:use-module (ice-9 ftw) |
d4254711 | 28 | #:use-module (guix build utils) |
88840f02 LC |
29 | #:export (mount-essential-file-systems |
30 | linux-command-line | |
d91712ee | 31 | make-essential-device-nodes |
88840f02 | 32 | configure-qemu-networking |
83bcd0b8 | 33 | mount-file-system |
89bf140b | 34 | bind-mount |
88840f02 | 35 | load-linux-module* |
d4254711 LC |
36 | device-number |
37 | boot-system)) | |
88840f02 LC |
38 | |
39 | ;;; Commentary: | |
40 | ;;; | |
41 | ;;; Utility procedures useful in a Linux initial RAM disk (initrd). Note that | |
42 | ;;; many of these use procedures not yet available in vanilla Guile (`mount', | |
43 | ;;; `load-linux-module', etc.); these are provided by a Guile patch used in | |
44 | ;;; the GNU distribution. | |
45 | ;;; | |
46 | ;;; Code: | |
47 | ||
48 | (define* (mount-essential-file-systems #:key (root "/")) | |
49 | "Mount /proc and /sys under ROOT." | |
50 | (define (scope dir) | |
51 | (string-append root | |
52 | (if (string-suffix? "/" root) | |
53 | "" | |
54 | "/") | |
55 | dir)) | |
56 | ||
57 | (unless (file-exists? (scope "proc")) | |
58 | (mkdir (scope "proc"))) | |
59 | (mount "none" (scope "proc") "proc") | |
60 | ||
61 | (unless (file-exists? (scope "sys")) | |
62 | (mkdir (scope "sys"))) | |
63 | (mount "none" (scope "sys") "sysfs")) | |
64 | ||
1d462832 LC |
65 | (define (move-essential-file-systems root) |
66 | "Move currently mounted essential file systems to ROOT." | |
67 | (for-each (lambda (dir) | |
68 | (let ((target (string-append root dir))) | |
69 | (unless (file-exists? target) | |
70 | (mkdir target)) | |
71 | (mount dir target "" MS_MOVE))) | |
72 | '("/proc" "/sys"))) | |
73 | ||
88840f02 LC |
74 | (define (linux-command-line) |
75 | "Return the Linux kernel command line as a list of strings." | |
76 | (string-tokenize | |
77 | (call-with-input-file "/proc/cmdline" | |
78 | get-string-all))) | |
79 | ||
d91712ee LC |
80 | (define* (make-essential-device-nodes #:key (root "/")) |
81 | "Make essential device nodes under ROOT/dev." | |
82 | ;; The hand-made udev! | |
83 | ||
84 | (define (scope dir) | |
85 | (string-append root | |
86 | (if (string-suffix? "/" root) | |
87 | "" | |
88 | "/") | |
89 | dir)) | |
90 | ||
91 | (unless (file-exists? (scope "dev")) | |
92 | (mkdir (scope "dev"))) | |
93 | ||
fc4bc4b6 LC |
94 | ;; Make the device nodes for SCSI disks. |
95 | (mknod (scope "dev/sda") 'block-special #o644 (device-number 8 0)) | |
96 | (mknod (scope "dev/sda1") 'block-special #o644 (device-number 8 1)) | |
97 | (mknod (scope "dev/sda2") 'block-special #o644 (device-number 8 2)) | |
98 | ||
99 | ;; The virtio (para-virtualized) block devices, as supported by QEMU/KVM. | |
100 | (mknod (scope "dev/vda") 'block-special #o644 (device-number 252 0)) | |
101 | (mknod (scope "dev/vda1") 'block-special #o644 (device-number 252 1)) | |
102 | (mknod (scope "dev/vda2") 'block-special #o644 (device-number 252 2)) | |
d91712ee | 103 | |
c04c6ff6 LC |
104 | ;; Memory (used by Xorg's VESA driver.) |
105 | (mknod (scope "dev/mem") 'char-special #o640 (device-number 1 1)) | |
106 | (mknod (scope "dev/kmem") 'char-special #o640 (device-number 1 2)) | |
107 | ||
1c221510 LC |
108 | ;; Inputs (used by Xorg.) |
109 | (unless (file-exists? (scope "dev/input")) | |
110 | (mkdir (scope "dev/input"))) | |
111 | (mknod (scope "dev/input/mice") 'char-special #o640 (device-number 13 63)) | |
112 | (mknod (scope "dev/input/mouse0") 'char-special #o640 (device-number 13 32)) | |
113 | (mknod (scope "dev/input/event0") 'char-special #o640 (device-number 13 64)) | |
114 | ||
d91712ee | 115 | ;; TTYs. |
29804e6e LC |
116 | (mknod (scope "dev/tty") 'char-special #o600 |
117 | (device-number 5 0)) | |
289773c1 | 118 | (chmod (scope "dev/tty") #o666) |
d91712ee LC |
119 | (let loop ((n 0)) |
120 | (and (< n 50) | |
121 | (let ((name (format #f "dev/tty~a" n))) | |
29804e6e | 122 | (mknod (scope name) 'char-special #o600 |
d91712ee LC |
123 | (device-number 4 n)) |
124 | (loop (+ 1 n))))) | |
125 | ||
c9c88118 LC |
126 | ;; Pseudo ttys. |
127 | (mknod (scope "dev/ptmx") 'char-special #o666 | |
128 | (device-number 5 2)) | |
289773c1 | 129 | (chmod (scope "dev/ptmx") #o666) |
c9c88118 | 130 | |
c865a878 | 131 | ;; Create /dev/pts; it will be mounted later, at boot time. |
c9c88118 LC |
132 | (unless (file-exists? (scope "dev/pts")) |
133 | (mkdir (scope "dev/pts"))) | |
c9c88118 | 134 | |
37c825eb LC |
135 | ;; Rendez-vous point for syslogd. |
136 | (mknod (scope "dev/log") 'socket #o666 0) | |
137 | (mknod (scope "dev/kmsg") 'char-special #o600 (device-number 1 11)) | |
138 | ||
289773c1 LC |
139 | ;; Other useful nodes, notably relied on by guix-daemon. |
140 | (for-each (match-lambda | |
141 | ((file major minor) | |
142 | (mknod (scope file) 'char-special #o666 | |
143 | (device-number major minor)) | |
144 | (chmod (scope file) #o666))) | |
145 | '(("dev/null" 1 3) | |
146 | ("dev/zero" 1 5) | |
147 | ("dev/full" 1 7) | |
148 | ("dev/random" 1 8) | |
149 | ("dev/urandom" 1 9))) | |
150 | ||
151 | (symlink "/proc/self/fd" (scope "dev/fd")) | |
152 | (symlink "/proc/self/fd/0" (scope "dev/stdin")) | |
153 | (symlink "/proc/self/fd/1" (scope "dev/stdout")) | |
1c96c1bb LC |
154 | (symlink "/proc/self/fd/2" (scope "dev/stderr")) |
155 | ||
156 | ;; File systems in user space (FUSE). | |
157 | (mknod (scope "dev/fuse") 'char-special #o666 (device-number 10 229))) | |
d91712ee | 158 | |
88840f02 LC |
159 | (define %host-qemu-ipv4-address |
160 | (inet-pton AF_INET "10.0.2.10")) | |
161 | ||
162 | (define* (configure-qemu-networking #:optional (interface "eth0")) | |
163 | "Setup the INTERFACE network interface and /etc/resolv.conf according to | |
164 | QEMU's default networking settings (see net/slirp.c in QEMU for default | |
165 | networking values.) Return #t if INTERFACE is up, #f otherwise." | |
166 | (display "configuring QEMU networking...\n") | |
167 | (let* ((sock (socket AF_INET SOCK_STREAM 0)) | |
168 | (address (make-socket-address AF_INET %host-qemu-ipv4-address 0)) | |
169 | (flags (network-interface-flags sock interface))) | |
170 | (set-network-interface-address sock interface address) | |
171 | (set-network-interface-flags sock interface (logior flags IFF_UP)) | |
172 | ||
173 | (unless (file-exists? "/etc") | |
174 | (mkdir "/etc")) | |
175 | (call-with-output-file "/etc/resolv.conf" | |
176 | (lambda (p) | |
177 | (display "nameserver 10.0.2.3\n" p))) | |
178 | ||
179 | (logand (network-interface-flags sock interface) IFF_UP))) | |
180 | ||
83bcd0b8 LC |
181 | ;; Linux mount flags, from libc's <sys/mount.h>. |
182 | (define MS_RDONLY 1) | |
183 | (define MS_BIND 4096) | |
1d462832 | 184 | (define MS_MOVE 8192) |
4919d684 | 185 | |
89bf140b LC |
186 | (define (bind-mount source target) |
187 | "Bind-mount SOURCE at TARGET." | |
89bf140b LC |
188 | (mount source target "" MS_BIND)) |
189 | ||
88840f02 LC |
190 | (define (load-linux-module* file) |
191 | "Load Linux module from FILE, the name of a `.ko' file." | |
192 | (define (slurp module) | |
193 | (call-with-input-file file get-bytevector-all)) | |
194 | ||
195 | (load-linux-module (slurp file))) | |
196 | ||
197 | (define (device-number major minor) | |
198 | "Return the device number for the device with MAJOR and MINOR, for use as | |
199 | the last argument of `mknod'." | |
200 | (+ (* major 256) minor)) | |
201 | ||
83bcd0b8 | 202 | (define* (mount-root-file-system root type |
3c05b4bc | 203 | #:key volatile-root? (unionfs "unionfs")) |
83bcd0b8 LC |
204 | "Mount the root file system of type TYPE at device ROOT. If VOLATILE-ROOT? |
205 | is true, mount ROOT read-only and make it a union with a writable tmpfs using | |
206 | UNIONFS." | |
207 | (catch #t | |
208 | (lambda () | |
209 | (if volatile-root? | |
210 | (begin | |
211 | (mkdir-p "/real-root") | |
212 | (mount root "/real-root" type MS_RDONLY) | |
213 | (mkdir-p "/rw-root") | |
214 | (mount "none" "/rw-root" "tmpfs") | |
215 | ||
216 | ;; We want read-write /dev nodes. | |
217 | (make-essential-device-nodes #:root "/rw-root") | |
218 | ||
219 | ;; Make /root a union of the tmpfs and the actual root. | |
220 | (unless (zero? (system* unionfs "-o" | |
221 | "cow,allow_other,use_ino,suid,dev" | |
222 | "/rw-root=RW:/real-root=RO" | |
223 | "/root")) | |
224 | (error "unionfs failed"))) | |
3c05b4bc LC |
225 | (begin |
226 | (check-file-system root type) | |
227 | (mount root "/root" type)))) | |
83bcd0b8 LC |
228 | (lambda args |
229 | (format (current-error-port) "exception while mounting '~a': ~s~%" | |
230 | root args) | |
231 | (start-repl)))) | |
232 | ||
3c05b4bc LC |
233 | (define (check-file-system device type) |
234 | "Run a file system check of TYPE on DEVICE." | |
235 | (define fsck | |
236 | (string-append "fsck." type)) | |
237 | ||
238 | (let ((status (system* fsck "-v" "-p" device))) | |
239 | (match (status:exit-val status) | |
240 | (0 | |
241 | #t) | |
242 | (1 | |
243 | (format (current-error-port) "'~a' corrected errors on ~a; continuing~%" | |
244 | fsck device)) | |
245 | (2 | |
246 | (format (current-error-port) "'~a' corrected errors on ~a; rebooting~%" | |
247 | fsck device) | |
248 | (sleep 3) | |
249 | (reboot)) | |
250 | (code | |
251 | (format (current-error-port) "'~a' exited with code ~a on ~a; spawning REPL~%" | |
252 | fsck code device) | |
253 | (start-repl))))) | |
254 | ||
83bcd0b8 LC |
255 | (define* (mount-file-system spec #:key (root "/root")) |
256 | "Mount the file system described by SPEC under ROOT. SPEC must have the | |
257 | form: | |
258 | ||
3c05b4bc | 259 | (DEVICE MOUNT-POINT TYPE (FLAGS ...) OPTIONS CHECK?) |
83bcd0b8 LC |
260 | |
261 | DEVICE, MOUNT-POINT, and TYPE must be strings; OPTIONS can be a string or #f; | |
3c05b4bc LC |
262 | FLAGS must be a list of symbols. CHECK? is a Boolean indicating whether to |
263 | run a file system check." | |
83bcd0b8 LC |
264 | (define flags->bit-mask |
265 | (match-lambda | |
266 | (('read-only rest ...) | |
267 | (or MS_RDONLY (flags->bit-mask rest))) | |
268 | (('bind-mount rest ...) | |
269 | (or MS_BIND (flags->bit-mask rest))) | |
270 | (() | |
271 | 0))) | |
272 | ||
273 | (match spec | |
3c05b4bc | 274 | ((source mount-point type (flags ...) options check?) |
83bcd0b8 | 275 | (let ((mount-point (string-append root "/" mount-point))) |
3c05b4bc LC |
276 | (when check? |
277 | (check-file-system source type)) | |
83bcd0b8 LC |
278 | (mkdir-p mount-point) |
279 | (mount source mount-point type (flags->bit-mask flags) | |
280 | (if options | |
281 | (string->pointer options) | |
282 | %null-pointer)))))) | |
283 | ||
1d462832 LC |
284 | (define (switch-root root) |
285 | "Switch to ROOT as the root file system, in a way similar to what | |
286 | util-linux' switch_root(8) does." | |
287 | (move-essential-file-systems root) | |
288 | (chdir root) | |
26a728eb LC |
289 | |
290 | ;; Since we're about to 'rm -rf /', try to make sure we're on an initrd. | |
291 | ;; TODO: Use 'statfs' to check the fs type, like klibc does. | |
292 | (when (or (not (file-exists? "/init")) (directory-exists? "/home")) | |
293 | (format (current-error-port) | |
294 | "The root file system is probably not an initrd; \ | |
295 | bailing out.~%root contents: ~s~%" (scandir "/")) | |
296 | (force-output (current-error-port)) | |
297 | (exit 1)) | |
298 | ||
299 | ;; Delete files from the old root, without crossing mount points (assuming | |
300 | ;; there are no mount points in sub-directories.) That means we're leaving | |
301 | ;; the empty ROOT directory behind us, but that's OK. | |
302 | (let ((root-device (stat:dev (stat "/")))) | |
303 | (for-each (lambda (file) | |
304 | (unless (member file '("." "..")) | |
305 | (let* ((file (string-append "/" file)) | |
306 | (device (stat:dev (lstat file)))) | |
307 | (when (= device root-device) | |
308 | (delete-file-recursively file))))) | |
309 | (scandir "/"))) | |
310 | ||
311 | ;; Make ROOT the new root. | |
1d462832 | 312 | (mount root "/" "" MS_MOVE) |
26a728eb LC |
313 | (chroot ".") |
314 | (chdir "/") | |
315 | ||
316 | (when (file-exists? "/dev/console") | |
317 | ;; Close the standard file descriptors since they refer to the old | |
318 | ;; /dev/console. | |
319 | (for-each close-fdes '(0 1 2)) | |
320 | ||
321 | ;; Reopen them. | |
322 | (let ((in (open-file "/dev/console" "rbl")) | |
323 | (out (open-file "/dev/console" "wbl"))) | |
324 | (dup2 (fileno in) 0) | |
325 | (dup2 (fileno out) 1) | |
326 | (dup2 (fileno out) 2) | |
327 | ||
328 | ;; Safely close IN and OUT. | |
329 | (for-each (lambda (port) | |
330 | (if (memv (fileno port) '(0 1 2)) | |
331 | (set-port-revealed! port 1) | |
332 | (close-port port))) | |
333 | (list in out))))) | |
1d462832 | 334 | |
d4254711 LC |
335 | (define* (boot-system #:key |
336 | (linux-modules '()) | |
337 | qemu-guest-networking? | |
338 | guile-modules-in-chroot? | |
3c05b4bc | 339 | volatile-root? |
d4254711 LC |
340 | (mounts '())) |
341 | "This procedure is meant to be called from an initrd. Boot a system by | |
342 | first loading LINUX-MODULES, then setting up QEMU guest networking if | |
343 | QEMU-GUEST-NETWORKING? is true, mounting the file systems specified in MOUNTS, | |
344 | and finally booting into the new root if any. The initrd supports kernel | |
345 | command-line options '--load', '--root', and '--repl'. | |
346 | ||
3c05b4bc LC |
347 | Mount the root file system, specified by the '--root' command-line argument, |
348 | if any. | |
03ddfaf5 | 349 | |
83bcd0b8 | 350 | MOUNTS must be a list suitable for 'mount-file-system'. |
d4254711 LC |
351 | |
352 | When GUILE-MODULES-IN-CHROOT? is true, make core Guile modules available in | |
44ddf33e LC |
353 | the new root. |
354 | ||
355 | When VOLATILE-ROOT? is true, the root file system is writable but any changes | |
356 | to it are lost." | |
d4254711 LC |
357 | (define (resolve file) |
358 | ;; If FILE is a symlink to an absolute file name, resolve it as if we were | |
359 | ;; under /root. | |
360 | (let ((st (lstat file))) | |
361 | (if (eq? 'symlink (stat:type st)) | |
362 | (let ((target (readlink file))) | |
363 | (resolve (string-append "/root" target))) | |
364 | file))) | |
365 | ||
3c05b4bc LC |
366 | (define root-mount-point? |
367 | (match-lambda | |
368 | ((device "/" _ ...) #t) | |
369 | (_ #f))) | |
370 | ||
371 | (define root-fs-type | |
372 | (or (any (match-lambda | |
373 | ((device "/" type _ ...) type) | |
374 | (_ #f)) | |
375 | mounts) | |
376 | "ext4")) | |
377 | ||
d4254711 LC |
378 | (display "Welcome, this is GNU's early boot Guile.\n") |
379 | (display "Use '--repl' for an initrd REPL.\n\n") | |
380 | ||
381 | (mount-essential-file-systems) | |
382 | (let* ((args (linux-command-line)) | |
383 | (option (lambda (opt) | |
384 | (let ((opt (string-append opt "="))) | |
385 | (and=> (find (cut string-prefix? opt <>) | |
386 | args) | |
387 | (lambda (arg) | |
388 | (substring arg (+ 1 (string-index arg #\=)))))))) | |
389 | (to-load (option "--load")) | |
390 | (root (option "--root"))) | |
391 | ||
392 | (when (member "--repl" args) | |
393 | (start-repl)) | |
394 | ||
395 | (display "loading kernel modules...\n") | |
396 | (for-each (compose load-linux-module* | |
397 | (cut string-append "/modules/" <>)) | |
398 | linux-modules) | |
399 | ||
400 | (when qemu-guest-networking? | |
401 | (unless (configure-qemu-networking) | |
402 | (display "network interface is DOWN\n"))) | |
403 | ||
404 | ;; Make /dev nodes. | |
405 | (make-essential-device-nodes) | |
406 | ||
407 | ;; Prepare the real root file system under /root. | |
408 | (unless (file-exists? "/root") | |
409 | (mkdir "/root")) | |
410 | (if root | |
83bcd0b8 | 411 | (mount-root-file-system root root-fs-type |
3c05b4bc | 412 | #:volatile-root? volatile-root?) |
d4254711 | 413 | (mount "none" "/root" "tmpfs")) |
44ddf33e | 414 | |
d4254711 LC |
415 | (unless (file-exists? "/root/dev") |
416 | (mkdir "/root/dev") | |
417 | (make-essential-device-nodes #:root "/root")) | |
418 | ||
419 | ;; Mount the specified file systems. | |
3c05b4bc LC |
420 | (for-each mount-file-system |
421 | (remove root-mount-point? mounts)) | |
d4254711 LC |
422 | |
423 | (when guile-modules-in-chroot? | |
424 | ;; Copy the directories that contain .scm and .go files so that the | |
425 | ;; child process in the chroot can load modules (we would bind-mount | |
426 | ;; them but for some reason that fails with EINVAL -- XXX). | |
427 | (mkdir-p "/root/share") | |
428 | (mkdir-p "/root/lib") | |
429 | (mount "none" "/root/share" "tmpfs") | |
430 | (mount "none" "/root/lib" "tmpfs") | |
431 | (copy-recursively "/share" "/root/share" | |
432 | #:log (%make-void-port "w")) | |
433 | (copy-recursively "/lib" "/root/lib" | |
434 | #:log (%make-void-port "w"))) | |
435 | ||
436 | (if to-load | |
437 | (begin | |
1d462832 | 438 | (switch-root "/root") |
26a728eb | 439 | (format #t "loading '~a'...\n" to-load) |
c865a878 LC |
440 | |
441 | ;; Obviously this has to be done each time we boot. Do it from here | |
442 | ;; so that statfs(2) returns DEVPTS_SUPER_MAGIC like libc's getpt(3) | |
443 | ;; expects (and thus openpty(3) and its users, such as xterm.) | |
444 | (mount "none" "/dev/pts" "devpts") | |
445 | ||
d4254711 LC |
446 | ;; TODO: Remove /lib, /share, and /loader.go. |
447 | (catch #t | |
448 | (lambda () | |
449 | (primitive-load to-load)) | |
450 | (lambda args | |
451 | (format (current-error-port) "'~a' raised an exception: ~s~%" | |
452 | to-load args) | |
453 | (start-repl))) | |
454 | (format (current-error-port) | |
455 | "boot program '~a' terminated, rebooting~%" | |
456 | to-load) | |
457 | (sleep 2) | |
458 | (reboot)) | |
459 | (begin | |
460 | (display "no boot file passed via '--load'\n") | |
461 | (display "entering a warm and cozy REPL\n") | |
462 | (start-repl))))) | |
463 | ||
88840f02 | 464 | ;;; linux-initrd.scm ends here |