]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/nspawn/nspawn-oci.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nspawn: add support for executing OCI runtime bundles with nspawn
[thirdparty/systemd.git] / src / nspawn / nspawn-oci.c
CommitLineData
de40a303
LP		 1/* SPDX-License-Identifier: LGPL-2.1+ */
2
3#include <linux/seccomp.h>
4#include <linux/oom.h>
5#include <seccomp.h>
6
7#include "bus-util.h"
8#include "cap-list.h"
9#include "cpu-set-util.h"
10#include "env-util.h"
11#include "fs-util.h"
12#include "hostname-util.h"
13#include "json.h"
14#include "missing_sched.h"
15#include "nspawn-oci.h"
16#include "path-util.h"
17#include "rlimit-util.h"
18#include "seccomp-util.h"
19#include "stat-util.h"
20#include "stdio-util.h"
21#include "string-util.h"
22#include "strv.h"
23#include "user-util.h"
24
25/* TODO:
26 * OCI runtime tool implementation
27 * hooks
28 *
29 * Spec issues:
30 *
31 * How is RLIM_INFINITY supposed to be encoded?
32 * configured effective caps is bullshit, as execv() corrupts it anyway
33 * pipes bind mounted is *very* different from pipes newly created, comments regarding bind mount or not are bogus
34 * annotation values structured? or string?
35 * configurable file system namespace path, but then also root path? wtf?
36 * apply sysctl inside of the container? or outside?
37 * how is unlimited pids tasks limit to be encoded?
38 * what are the defaults for caps if not specified?
39 * what are the default uid/gid mappings if one is missing but the other set, or when user ns is on but no namespace configured
40 * the source field of "mounts" is really weird, as it cannot realistically be relative to the bundle, since we never know if that's what the fs wants
41 * spec contradicts itself on the mount "type" field, as the example uses "bind" as type, but it's not listed in /proc/filesystem, and is something made up by /bin/mount
42 * if type of mount is left out, what shall be assumed? "bind"?
43 * readonly mounts is entirely redundant?
44 * should escaping be applied when joining mount options with ","?
45 * devices cgroup support is bogus, "allow" and "deny" on the kernel level is about adding/removing entries, not about access
46 * spec needs to say that "rwm" devices cgroup combination can't be the empty string
47 * cgrouspv1 crap: kernel, kernelTCP, swapiness, disableOOMKiller, swap, devices, leafWeight
48 * general: it shouldn't leak lower level abstractions this obviously
49 * unmanagable cgroups stuff: realtimeRuntime/realtimePeriod
50 * needs to say what happense when some option is not specified, i.e. which defautls apply
51 * no architecture? no personality?
52 * seccomp example and logic is simply broken: there's no constant "SCMP_ACT_ERRNO".
53 * spec should say what to do with unknown props
54 * /bin/mount regarding NFS and FUSE required?
55 * what does terminal=false mean?
56 * sysctl inside or outside? whitelisting?
57 *
58 * Unsupported:
59 *
60 * apparmorProfile
61 * selinuxLabel + mountLabel
62 * hugepageLimits
63 * network
64 * rdma
65 * intelRdt
66 * swappiness, disableOOMKiller, kernel, kernelTCP, leafWeight (because it's dead, cgroupsv2 can't do it and hence systemd neither)
67 *
68 * Non-slice cgroup paths
69 * Propagation that is not slave + shared
70 * more than one uid/gid mapping, mappings with a container base != 0, or non-matching uid/gid mappings
71 * device cgroups access = false items that are not catchall
72 * device cgroups matches where minor is specified, but major isn't. similar where major is specified but char/block is not. also, any match that only has a type set that has less than "rwm" set. also, any entry that has none of rwm set.
73 *
74 */
75
76static int oci_unexpected(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
77 json_log(v, flags, 0, "Unexpected OCI element '%s' of type '%s'.", name, json_variant_type_to_string(json_variant_type(v)));
78 return -EINVAL;
79}
80
81static int oci_unsupported(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
82 json_log(v, flags, 0, "Unsupported OCI element '%s' of type '%s'.", name, json_variant_type_to_string(json_variant_type(v)));
83 return -EOPNOTSUPP;
84}
85
86static int oci_terminal(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
87 Settings *s = userdata;
88
89 /* If not specified, or set to true, we'll default to either an interactive or a read-only
90 * console. If specifiy as false, we'll forcibly move to "pipe" mode though. */
91 s->console_mode = json_variant_boolean(v) ? _CONSOLE_MODE_INVALID : CONSOLE_PIPE;
92 return 0;
93}
94
95static int oci_console_dimension(const char *name, JsonVariant *variant, JsonDispatchFlags flags, void *userdata) {
96 unsigned *u = userdata;
97 uintmax_t k;
98
99 assert(u);
100
101 k = json_variant_unsigned(variant);
102 if (k == 0) {
103 json_log(variant, flags, 0, "Console size field '%s' is too small.", strna(name));
104 return -ERANGE;
105 }
106 if (k > USHRT_MAX) { /* TIOCSWINSZ's struct winsize uses "unsigned short" for width and height */
107 json_log(variant, flags, 0, "Console size field '%s' is too large.", strna(name));
108 return -ERANGE;
109 }
110
111 *u = (unsigned) k;
112 return 0;
113}
114
115static int oci_console_size(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
116
117 static const JsonDispatch table[] = {
118 { "height", JSON_VARIANT_UNSIGNED, oci_console_dimension, offsetof(Settings, console_height), JSON_MANDATORY },
119 { "width", JSON_VARIANT_UNSIGNED, oci_console_dimension, offsetof(Settings, console_width), JSON_MANDATORY },
120 {}
121 };
122
123 return json_dispatch(v, table, oci_unexpected, flags, userdata);
124}
125
126static int oci_absolute_path(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
127 char **p = userdata;
128 const char *n;
129 int r;
130
131 assert(p);
132
133 n = json_variant_string(v);
134
135 if (!path_is_absolute(n)) {
136 json_log(v, flags, 0, "Path in JSON field '%s' is not absolute: %s", strna(name), n);
137 return -EINVAL;
138 }
139
140 r = free_and_strdup(p, n);
141 if (r < 0)
142 return log_oom();
143
144 return 0;
145}
146
147static int oci_env(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
148 char ***l = userdata;
149 JsonVariant *e;
150 int r;
151
152 assert(l);
153
154 JSON_VARIANT_ARRAY_FOREACH(e, v) {
155 const char *n;
156
157 if (!json_variant_is_string(e)) {
158 json_log(e, flags, 0, "Environment array contains non-string.");
159 return -EINVAL;
160 }
161
162 assert_se(n = json_variant_string(e));
163
164 if (!env_assignment_is_valid(n)) {
165 json_log(e, flags, 0, "Environment assignment not valid: %s", n);
166 return -EINVAL;
167 }
168
169 r = strv_extend(l, n);
170 if (r < 0)
171 return log_oom();
172 }
173
174 return 0;
175}
176
177static int oci_args(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
178 _cleanup_strv_free_ char **l = NULL;
179 char ***value = userdata;
180 JsonVariant *e;
181 int r;
182
183 assert(value);
184
185 JSON_VARIANT_ARRAY_FOREACH(e, v) {
186 const char *n;
187
188 if (!json_variant_is_string(e)) {
189 json_log(v, flags, 0, "Argument is not a string.");
190 return -EINVAL;
191 }
192
193 assert_se(n = json_variant_string(e));
194
195 r = strv_extend(&l, n);
196 if (r < 0)
197 return log_oom();
198 }
199
200 if (strv_isempty(l)) {
201 json_log(v, flags, 0, "Argument list empty, refusing.");
202 return -EINVAL;
203 }
204
205 if (isempty(l[0])) {
206 json_log(v, flags, 0, "Executable name is empty, refusing.");
207 return -EINVAL;
208 }
209
210 strv_free_and_replace(*value, l);
211 return 0;
212}
213
214static int oci_rlimit_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
215 const char *z;
216 int t, *type = userdata;
217
218 assert_se(type);
219
220 z = startswith(json_variant_string(v), "RLIMIT_");
221 if (!z) {
222 json_log(v, flags, 0, "rlimit entry's name does not begin with 'RLIMIT_', refusing: %s", json_variant_string(v));
223 return -EINVAL;
224 }
225
226 t = rlimit_from_string(z);
227 if (t < 0) {
228 json_log(v, flags, 0, "rlimit name unknown: %s", json_variant_string(v));
229 return -EINVAL;
230 }
231
232 *type = t;
233 return 0;
234}
235
236static int oci_rlimit_value(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
237 rlim_t z, *value = userdata;
238
239 assert(value);
240
241 if (json_variant_is_negative(v))
242 z = RLIM_INFINITY;
243 else {
244 if (!json_variant_is_unsigned(v)) {
245 json_log(v, flags, 0, "rlimits limit not unsigned, refusing.");
246 return -ERANGE;
247 }
248
249 z = (rlim_t) json_variant_unsigned(v);
250
251 if ((uintmax_t) z != json_variant_unsigned(v)) {
252 json_log(v, flags, 0, "rlimits limit out of range, refusing.");
253 return -EINVAL;
254 }
255 }
256
257 *value = z;
258 return 0;
259}
260
261static int oci_rlimits(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
262
263 Settings *s = userdata;
264 JsonVariant *e;
265 int r;
266
267 assert(s);
268
269 JSON_VARIANT_ARRAY_FOREACH(e, v) {
270
271 struct rlimit_data {
272 int type;
273 rlim_t soft;
274 rlim_t hard;
275 } data = {
276 .type = -1,
277 .soft = RLIM_INFINITY,
278 .hard = RLIM_INFINITY,
279 };
280
281 static const JsonDispatch table[] = {
282 { "soft", JSON_VARIANT_NUMBER, oci_rlimit_value, offsetof(struct rlimit_data, soft), JSON_MANDATORY },
283 { "hard", JSON_VARIANT_NUMBER, oci_rlimit_value, offsetof(struct rlimit_data, hard), JSON_MANDATORY },
284 { "type", JSON_VARIANT_STRING, oci_rlimit_type, offsetof(struct rlimit_data, type), JSON_MANDATORY },
285 {}
286 };
287
288
289 r = json_dispatch(e, table, oci_unexpected, flags, &data);
290 if (r < 0)
291 return r;
292
293 assert(data.type >= 0);
294 assert(data.type < _RLIMIT_MAX);
295
296 if (s->rlimit[data.type]) {
297 json_log(v, flags, 0, "rlimits array contains duplicate entry, refusing.");
298 return -EINVAL;
299 }
300
301 s->rlimit[data.type] = new(struct rlimit, 1);
302 if (!s->rlimit[data.type])
303 return log_oom();
304
305 *s->rlimit[data.type] = (struct rlimit) {
306 .rlim_cur = data.soft,
307 .rlim_max = data.hard,
308 };
309
310 }
311 return 0;
312}
313
314static int oci_capability_array(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
315 uint64_t *mask = userdata, m = 0;
316 JsonVariant *e;
317
318 JSON_VARIANT_ARRAY_FOREACH(e, v) {
319 const char *n;
320 int cap;
321
322 if (!json_variant_is_string(e)) {
323 json_log(v, flags, 0, "Entry in capabilities array is not a string.");
324 return -EINVAL;
325 }
326
327 assert_se(n = json_variant_string(e));
328
329 cap = capability_from_name(n);
330 if (cap < 0) {
331 json_log(v, flags, 0, "Unknown capability: %s", n);
332 return -EINVAL;
333 }
334
335 m |= UINT64_C(1) << cap;
336 }
337
338 if (*mask == (uint64_t) -1)
339 *mask = m;
340 else
341 *mask |= m;
342
343 return 0;
344}
345
346static int oci_capabilities(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
347
348 static const JsonDispatch table[] = {
349 { "effective", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, effective) },
350 { "bounding", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, bounding) },
351 { "inheritable", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, inheritable) },
352 { "permitted", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, permitted) },
353 { "ambient", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, ambient) },
354 {}
355 };
356
357 Settings *s = userdata;
358 int r;
359
360 assert(s);
361
362 r = json_dispatch(v, table, oci_unexpected, flags, &s->full_capabilities);
363 if (r < 0)
364 return r;
365
366 if (s->full_capabilities.bounding != (uint64_t) -1) {
367 s->capability = s->full_capabilities.bounding;
368 s->drop_capability = ~s->full_capabilities.bounding;
369 }
370
371 return 0;
372}
373
374static int oci_oom_score_adj(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
375 Settings *s = userdata;
376 intmax_t k;
377
378 assert(s);
379
380 k = json_variant_integer(v);
381 if (k < OOM_SCORE_ADJ_MIN || k > OOM_SCORE_ADJ_MAX) {
382 json_log(v, flags, 0, "oomScoreAdj value out of range: %ji", k);
383 return -EINVAL;
384 }
385
386 s->oom_score_adjust = (int) k;
387 s->oom_score_adjust_set = true;
388
389 return 0;
390}
391
392static int oci_uid_gid(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
393 uid_t *uid = userdata, u;
394 uintmax_t k;
395
396 assert(uid);
397 assert_cc(sizeof(uid_t) == sizeof(gid_t));
398
399 k = json_variant_unsigned(v);
400 u = (uid_t) k;
401 if ((uintmax_t) u != k) {
402 json_log(v, flags, 0, "UID/GID out of range: %ji", k);
403 return -EINVAL;
404 }
405
406 if (!uid_is_valid(u)) {
407 json_log(v, flags, 0, "UID/GID is not valid: " UID_FMT, u);
408 return -EINVAL;
409 }
410
411 *uid = u;
412 return 0;
413}
414
415static int oci_supplementary_gids(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
416 Settings *s = userdata;
417 JsonVariant *e;
418 int r;
419
420 assert(s);
421
422 JSON_VARIANT_ARRAY_FOREACH(e, v) {
423 gid_t gid, *a;
424
425 if (!json_variant_is_unsigned(e)) {
426 json_log(v, flags, 0, "Supplementary GID entry is not a UID.");
427 return -EINVAL;
428 }
429
430 r = oci_uid_gid(name, e, flags, &gid);
431 if (r < 0)
432 return r;
433
434 a = reallocarray(s->supplementary_gids, s->n_supplementary_gids + 1, sizeof(gid_t));
435 if (!a)
436 return log_oom();
437
438 s->supplementary_gids = a;
439 s->supplementary_gids[s->n_supplementary_gids++] = gid;
440 }
441
442 return 0;
443}
444
445static int oci_user(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
446 static const JsonDispatch table[] = {
447 { "uid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(Settings, uid), JSON_MANDATORY },
448 { "gid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(Settings, gid), JSON_MANDATORY },
449 { "additionalGids", JSON_VARIANT_ARRAY, oci_supplementary_gids, 0, 0 },
450 {}
451 };
452
453 return json_dispatch(v, table, oci_unexpected, flags, userdata);
454}
455
456static int oci_process(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
457
458 static const JsonDispatch table[] = {
459 { "terminal", JSON_VARIANT_BOOLEAN, oci_terminal, 0, 0 },
460 { "consoleSize", JSON_VARIANT_OBJECT, oci_console_size, 0, 0 },
461 { "cwd", JSON_VARIANT_STRING, oci_absolute_path, offsetof(Settings, working_directory), 0 },
462 { "env", JSON_VARIANT_ARRAY, oci_env, offsetof(Settings, environment), 0 },
463 { "args", JSON_VARIANT_ARRAY, oci_args, offsetof(Settings, parameters), 0 },
464 { "rlimits", JSON_VARIANT_ARRAY, oci_rlimits, 0, 0 },
465 { "apparmorProfile", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
466 { "capabilities", JSON_VARIANT_OBJECT, oci_capabilities, 0, 0 },
467 { "noNewPrivileges", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(Settings, no_new_privileges), 0 },
468 { "oomScoreAdj", JSON_VARIANT_INTEGER, oci_oom_score_adj, 0, 0 },
469 { "selinuxLabel", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
470 { "user", JSON_VARIANT_OBJECT, oci_user, 0, 0 },
471 {}
472 };
473
474 return json_dispatch(v, table, oci_unexpected, flags, userdata);
475}
476
477static int oci_root(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
478
479 static const JsonDispatch table[] = {
480 { "path", JSON_VARIANT_STRING, json_dispatch_string, offsetof(Settings, root) },
481 { "readonly", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(Settings, read_only) },
482 {}
483 };
484
485 return json_dispatch(v, table, oci_unexpected, flags, userdata);
486}
487
488static int oci_hostname(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
489 Settings *s = userdata;
490 const char *n;
491
492 assert(s);
493
494 assert_se(n = json_variant_string(v));
495
496 if (!hostname_is_valid(n, false)) {
497 json_log(v, flags, 0, "Hostname string is not a valid hostname: %s", n);
498 return -EINVAL;
499 }
500
501 if (free_and_strdup(&s->hostname, n) < 0)
502 return log_oom();
503
504 return 0;
505}
506
507static bool oci_exclude_mount(const char *path) {
508
509 /* Returns "true" for all mounts we insist to mount on our own, and hence ignore the OCI data. */
510
511 if (PATH_IN_SET(path,
512 "/dev",
513 "/dev/mqueue",
514 "/dev/pts",
515 "/dev/shm",
516 "/proc",
517 "/proc/acpi",
518 "/proc/apm",
519 "/proc/asound",
520 "/proc/bus",
521 "/proc/fs",
522 "/proc/irq",
523 "/proc/kallsyms",
524 "/proc/kcore",
525 "/proc/keys",
526 "/proc/scsi",
527 "/proc/sys",
528 "/proc/sys/net",
529 "/proc/sysrq-trigger",
530 "/proc/timer_list",
531 "/run",
532 "/sys",
533 "/sys",
534 "/sys/fs/selinux",
535 "/tmp"))
536 return true;
537
538 /* Similar, skip the whole /sys/fs/cgroups subtree */
539 if (path_startswith(path, "/sys/fs/cgroup"))
540 return true;
541
542 return false;
543}
544
545static int oci_mounts(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
546 Settings *s = userdata;
547 JsonVariant *e;
548 int r;
549
550 assert(s);
551
552 JSON_VARIANT_ARRAY_FOREACH(e, v) {
553
554 struct mount_data {
555 char *destination;
556 char *source;
557 char *type;
558 char **options;
559 } data = {};
560
561 static const JsonDispatch table[] = {
562 { "destination", JSON_VARIANT_STRING, oci_absolute_path, offsetof(struct mount_data, destination), JSON_MANDATORY },
563 { "source", JSON_VARIANT_STRING, json_dispatch_string, offsetof(struct mount_data, source), 0 },
564 { "options", JSON_VARIANT_ARRAY, json_dispatch_strv, offsetof(struct mount_data, options), 0, },
565 { "type", JSON_VARIANT_STRING, json_dispatch_string, offsetof(struct mount_data, type), 0 },
566 {}
567 };
568
569 _cleanup_free_ char *joined_options = NULL;
570 CustomMount *m;
571
572 r = json_dispatch(e, table, oci_unexpected, flags, &data);
573 if (r < 0)
574 goto fail_item;
575
576 if (!path_is_absolute(data.destination)) {
577 json_log(e, flags, 0, "Mount destination not an absolute path: %s", data.destination);
578 r = -EINVAL;
579 goto fail_item;
580 }
581
582 if (oci_exclude_mount(data.destination))
583 goto skip_item;
584
585 if (data.options) {
586 joined_options = strv_join(data.options, ",");
587 if (!joined_options) {
588 r = log_oom();
589 goto fail_item;
590 }
591 }
592
593 if (!data.type || streq(data.type, "bind")) {
594
595 if (!path_is_absolute(data.source)) {
596 char *joined;
597
598 joined = path_join(s->bundle, data.source);
599 if (!joined) {
600 r = log_oom();
601 goto fail_item;
602 }
603
604 free_and_replace(data.source, joined);
605 }
606
607 data.type = mfree(data.type);
608
609 m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_BIND);
610 } else
611 m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_ARBITRARY);
612 if (!m) {
613 r = log_oom();
614 goto fail_item;
615 }
616
617 m->destination = TAKE_PTR(data.destination);
618 m->source = TAKE_PTR(data.source);
619 m->options = TAKE_PTR(joined_options);
620 m->type_argument = TAKE_PTR(data.type);
621
622 strv_free(data.options);
623 continue;
624
625 fail_item:
626 free(data.destination);
627 free(data.source);
628 strv_free(data.options);
629 free(data.type);
630
631 return r;
632
633 skip_item:
634 free(data.destination);
635 free(data.source);
636 strv_free(data.options);
637 free(data.type);
638 }
639
640 return 0;
641}
642
643static int oci_namespace_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
644 unsigned long *nsflags = userdata;
645 const char *n;
646
647 assert(nsflags);
648 assert_se(n = json_variant_string(v));
649
650 /* We don't use namespace_flags_from_string() here, as the OCI spec uses slightly different names than the
651 * kernel here. */
652 if (streq(n, "pid"))
653 *nsflags = CLONE_NEWPID;
654 else if (streq(n, "network"))
655 *nsflags = CLONE_NEWNET;
656 else if (streq(n, "mount"))
657 *nsflags = CLONE_NEWNS;
658 else if (streq(n, "ipc"))
659 *nsflags = CLONE_NEWIPC;
660 else if (streq(n, "uts"))
661 *nsflags = CLONE_NEWUTS;
662 else if (streq(n, "user"))
663 *nsflags = CLONE_NEWUSER;
664 else if (streq(n, "cgroup"))
665 *nsflags = CLONE_NEWCGROUP;
666 else {
667 json_log(v, flags, 0, "Unknown cgroup type, refusing: %s", n);
668 return -EINVAL;
669 }
670
671 return 0;
672}
673
674static int oci_namespaces(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
675 Settings *s = userdata;
676 unsigned long n = 0;
677 JsonVariant *e;
678 int r;
679
680 assert_se(s);
681
682 JSON_VARIANT_ARRAY_FOREACH(e, v) {
683
684 struct namespace_data {
685 unsigned long type;
686 char *path;
687 } data = {};
688
689 static const JsonDispatch table[] = {
690 { "type", JSON_VARIANT_STRING, oci_namespace_type, offsetof(struct namespace_data, type), JSON_MANDATORY },
691 { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(struct namespace_data, path), 0 },
692 {}
693 };
694
695 r = json_dispatch(e, table, oci_unexpected, flags, &data);
696 if (r < 0) {
697 free(data.path);
698 return r;
699 }
700
701 if (data.path) {
702 if (data.type != CLONE_NEWNET) {
703 json_log(e, flags, 0, "Specifying namespace path for non-network namespace is not supported.");
704 free(data.path);
705 return -EOPNOTSUPP;
706 }
707
708 if (s->network_namespace_path) {
709 json_log(e, flags, 0, "Network namespace path specified more than once, refusing.");
710 free(data.path);
711 return -EINVAL;
712 }
713
714 free(s->network_namespace_path);
715 s->network_namespace_path = data.path;
716 }
717
718 if (FLAGS_SET(n, data.type)) {
719 json_log(e, flags, 0, "Duplicat namespace specification, refusing.");
720 return -EINVAL;
721 }
722
723 n |= data.type;
724 }
725
726 if (!FLAGS_SET(n, CLONE_NEWNS)) {
727 json_log(v, flags, 0, "Containers without file system namespace aren't supported.");
728 return -EOPNOTSUPP;
729 }
730
731 s->private_network = FLAGS_SET(n, CLONE_NEWNET);
732 s->userns_mode = FLAGS_SET(n, CLONE_NEWUSER) ? USER_NAMESPACE_FIXED : USER_NAMESPACE_NO;
733 s->use_cgns = FLAGS_SET(n, CLONE_NEWCGROUP);
734
735 s->clone_ns_flags = n & (CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
736
737 return 0;
738}
739
740static int oci_uid_gid_range(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
741 uid_t *uid = userdata, u;
742 uintmax_t k;
743
744 assert(uid);
745 assert_cc(sizeof(uid_t) == sizeof(gid_t));
746
747 /* This is very much like oci_uid_gid(), except the checks are a bit different, as this is a UID range rather
748 * than a specific UID, and hence (uid_t) -1 has no special significance. OTOH a range of zero makes no
749 * sense. */
750
751 k = json_variant_unsigned(v);
752 u = (uid_t) k;
753 if ((uintmax_t) u != k) {
754 json_log(v, flags, 0, "UID/GID out of range: %ji", k);
755 return -ERANGE;
756 }
757 if (u == 0) {
758 json_log(v, flags, 0, "UID/GID range can't be zero.");
759 return -ERANGE;
760 }
761
762 *uid = u;
763 return 0;
764}
765
766static int oci_uid_gid_mappings(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
767 struct mapping_data {
768 uid_t host_id;
769 uid_t container_id;
770 uid_t range;
771 } data = {
772 .host_id = UID_INVALID,
773 .container_id = UID_INVALID,
774 .range = 0,
775 };
776
777 static const JsonDispatch table[] = {
778 { "containerID", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(struct mapping_data, container_id), JSON_MANDATORY },
779 { "hostID", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(struct mapping_data, host_id), JSON_MANDATORY },
780 { "size", JSON_VARIANT_UNSIGNED, oci_uid_gid_range, offsetof(struct mapping_data, range), JSON_MANDATORY },
781 {}
782 };
783
784 Settings *s = userdata;
785 JsonVariant *e;
786 int r;
787
788 assert(s);
789
790 if (json_variant_elements(v) == 0)
791 return 0;
792
793 if (json_variant_elements(v) > 1) {
794 json_log(v, flags, 0, "UID/GID mappings with more than one entry are not supported.");
795 return -EOPNOTSUPP;
796 }
797
798 assert_se(e = json_variant_by_index(v, 0));
799
800 r = json_dispatch(e, table, oci_unexpected, flags, &data);
801 if (r < 0)
802 return r;
803
804 if (data.host_id + data.range < data.host_id ||
805 data.container_id + data.range < data.container_id) {
806 json_log(v, flags, 0, "UID/GID range goes beyond UID/GID validity range, refusing.");
807 return -EINVAL;
808 }
809
810 if (data.container_id != 0) {
811 json_log(v, flags, 0, "UID/GID mappings with a non-zero container base are not supported.");
812 return -EOPNOTSUPP;
813 }
814
815 if (data.range < 0x10000)
816 json_log(v, flags|JSON_WARNING, 0, "UID/GID mapping with less than 65536 UID/GIDS set up, you are looking for trouble.");
817
818 if (s->uid_range != UID_INVALID &&
819 (s->uid_shift != data.host_id || s->uid_range != data.range)) {
820 json_log(v, flags, 0, "Non-matching UID and GID mappings are not supported.");
821 return -EOPNOTSUPP;
822 }
823
824 s->uid_shift = data.host_id;
825 s->uid_range = data.range;
826
827 return 0;
828}
829
830static int oci_device_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
831 mode_t *mode = userdata;
832 const char *t;
833
834 assert(mode);
835 assert_se(t = json_variant_string(v));
836
837 if (STR_IN_SET(t, "c", "u"))
838 *mode = (*mode & ~S_IFMT) | S_IFCHR;
839 else if (streq(t, "b"))
840 *mode = (*mode & ~S_IFMT) | S_IFBLK;
841 else if (streq(t, "p"))
842 *mode = (*mode & ~S_IFMT) | S_IFIFO;
843 else {
844 json_log(v, flags, 0, "Unknown device type: %s", t);
845 return -EINVAL;
846 }
847
848 return 0;
849}
850
851static int oci_device_major(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
852 unsigned *u = userdata;
853 uintmax_t k;
854
855 assert_se(u);
856
857 k = json_variant_unsigned(v);
858 if (!DEVICE_MAJOR_VALID(k)) {
859 json_log(v, flags, 0, "Device major %ji out of range.", k);
860 return -ERANGE;
861 }
862
863 *u = (unsigned) k;
864 return 0;
865}
866
867static int oci_device_minor(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
868 unsigned *u = userdata;
869 uintmax_t k;
870
871 assert_se(u);
872
873 k = json_variant_unsigned(v);
874 if (!DEVICE_MINOR_VALID(k)) {
875 json_log(v, flags, 0, "Device minor %ji out of range.", k);
876 return -ERANGE;
877 }
878
879 *u = (unsigned) k;
880 return 0;
881}
882
883static int oci_device_file_mode(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
884 mode_t *mode = userdata, m;
885 uintmax_t k;
886
887 assert(mode);
888
889 k = json_variant_unsigned(v);
890 m = (mode_t) k;
891
892 if ((m & ~07777) != 0 || (uintmax_t) m != k) {
893 json_log(v, flags, 0, "fileMode out of range, refusing.");
894 return -ERANGE;
895 }
896
897 *mode = m;
898 return 0;
899}
900
901static int oci_devices(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
902 Settings *s = userdata;
903 JsonVariant *e;
904 int r;
905
906 assert(s);
907
908 JSON_VARIANT_ARRAY_FOREACH(e, v) {
909
910 static const JsonDispatch table[] = {
911 { "type", JSON_VARIANT_STRING, oci_device_type, offsetof(DeviceNode, mode), JSON_MANDATORY },
912 { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(DeviceNode, path), JSON_MANDATORY },
913 { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(DeviceNode, major), 0 },
914 { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(DeviceNode, minor), 0 },
915 { "fileMode", JSON_VARIANT_UNSIGNED, oci_device_file_mode, offsetof(DeviceNode, mode), 0 },
916 { "uid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(DeviceNode, uid), 0 },
917 { "gid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(DeviceNode, gid), 0 },
918 {}
919 };
920
921 DeviceNode *node, *nodes;
922
923 nodes = reallocarray(s->extra_nodes, s->n_extra_nodes + 1, sizeof(DeviceNode));
924 if (!nodes)
925 return log_oom();
926
927 s->extra_nodes = nodes;
928
929 node = nodes + s->n_extra_nodes;
930 *node = (DeviceNode) {
931 .uid = UID_INVALID,
932 .gid = GID_INVALID,
933 .major = (unsigned) -1,
934 .minor = (unsigned) -1,
935 .mode = 0644,
936 };
937
938 r = json_dispatch(e, table, oci_unexpected, flags, node);
939 if (r < 0)
940 goto fail_element;
941
942 if (S_ISCHR(node->mode) || S_ISBLK(node->mode)) {
943 _cleanup_free_ char *path = NULL;
944
945 if (node->major == (unsigned) -1 || node->minor == (unsigned) -1) {
946 json_log(e, flags, 0, "Major/minor required when device node is device node");
947 r = -EINVAL;
948 goto fail_element;
949 }
950
951 /* Suppress a couple of implicit device nodes */
952 r = device_path_make_canonical(node->mode, makedev(node->major, node->minor), &path);
953 if (r < 0)
954 json_log(e, flags|JSON_DEBUG, 0, "Failed to resolve device node %u:%u, ignoring: %m", node->major, node->minor);
955 else {
956 if (PATH_IN_SET(path,
957 "/dev/null",
958 "/dev/zero",
959 "/dev/full",
960 "/dev/random",
961 "/dev/urandom",
962 "/dev/tty",
963 "/dev/net/tun",
964 "/dev/ptmx",
965 "/dev/pts/ptmx",
966 "/dev/console")) {
967
968 json_log(e, flags|JSON_DEBUG, 0, "Ignoring devices item for device '%s', as it is implicitly created anyway.", path);
969 free(node->path);
970 continue;
971 }
972 }
973 }
974
975 s->n_extra_nodes++;
976 continue;
977
978 fail_element:
979 free(node->path);
980 return r;
981 }
982
983 return 0;
984}
985
986static int oci_cgroups_path(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
987 _cleanup_free_ char *slice = NULL, *backwards = NULL;
988 Settings *s = userdata;
989 const char *p;
990 int r;
991
992 assert(s);
993
994 assert_se(p = json_variant_string(v));
995
996 r = cg_path_get_slice(p, &slice);
997 if (r < 0)
998 return json_log(v, flags, r, "Couldn't derive slice unit name from path '%s': %m", p);
999
1000 r = cg_slice_to_path(slice, &backwards);
1001 if (r < 0)
1002 return json_log(v, flags, r, "Couldn't convert slice unit name '%s' back to path: %m", slice);
1003
1004 if (!path_equal(backwards, p)) {
1005 json_log(v, flags, 0, "Control group path '%s' does not refer to slice unit, refusing.", p);
1006 return -EINVAL;
1007 }
1008
1009 free_and_replace(s->slice, slice);
1010 return 0;
1011}
1012
1013static int oci_cgroup_device_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1014 mode_t *mode = userdata;
1015 const char *n;
1016
1017 assert_se(n = json_variant_string(v));
1018
1019 if (streq(n, "c"))
1020 *mode = S_IFCHR;
1021 else if (streq(n, "b"))
1022 *mode = S_IFBLK;
1023 else {
1024 json_log(v, flags, 0, "Control group device type unknown: %s", n);
1025 return -EINVAL;
1026 }
1027
1028 return 0;
1029}
1030
1031struct device_data {
1032 bool allow;
1033 bool r;
1034 bool w;
1035 bool m;
1036 mode_t type;
1037 unsigned major;
1038 unsigned minor;
1039};
1040
1041static int oci_cgroup_device_access(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1042 struct device_data *d = userdata;
1043 bool r = false, w = false, m = false;
1044 const char *s;
1045 size_t i;
1046
1047 assert_se(s = json_variant_string(v));
1048
1049 for (i = 0; s[i]; i++) {
1050 if (s[i] == 'r')
1051 r = true;
1052 else if (s[i] == 'w')
1053 w = true;
1054 else if (s[i] == 'm')
1055 m = true;
1056 else {
1057 json_log(v, flags, 0, "Unknown device access character '%c'.", s[i]);
1058 return -EINVAL;
1059 }
1060 }
1061
1062 d->r = r;
1063 d->w = w;
1064 d->m = m;
1065
1066 return 0;
1067}
1068
1069static int oci_cgroup_devices(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1070
1071 _cleanup_free_ struct device_data *list = NULL;
1072 Settings *s = userdata;
1073 size_t n_list = 0, i;
1074 bool noop = false;
1075 JsonVariant *e;
1076 int r;
1077
1078 assert(s);
1079
1080 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1081
1082 struct device_data data = {
1083 .major = (unsigned) -1,
1084 .minor = (unsigned) -1,
1085 }, *a;
1086
1087 static const JsonDispatch table[] = {
1088 { "allow", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(struct device_data, allow), JSON_MANDATORY },
1089 { "type", JSON_VARIANT_STRING, oci_cgroup_device_type, offsetof(struct device_data, type), 0 },
1090 { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), 0 },
1091 { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), 0 },
1092 { "access", JSON_VARIANT_STRING, oci_cgroup_device_access, 0, 0 },
1093 {}
1094 };
1095
1096 r = json_dispatch(e, table, oci_unexpected, flags, &data);
1097 if (r < 0)
1098 return r;
1099
1100 if (!data.allow) {
1101 /* The fact that OCI allows 'deny' entries makes really no sense, as 'allow' vs. 'deny' for the
1102 * devices cgroup controller is really not about whitelisting and blacklisting but about adding
1103 * and removing entries from the whitelist. Since we always start out with an empty whitelist
1104 * we hence ignore the whole thing, as removing entries which don't exist make no sense. We'll
1105 * log about this, since this is really borked in the spec, with one exception: the entry
1106 * that's supposed to drop the kernel's default we ignore silently */
1107
1108 if (!data.r || !data.w || !data.m || data.type != 0 || data.major != (unsigned) -1 || data.minor != (unsigned) -1)
1109 json_log(v, flags|JSON_WARNING, 0, "Devices cgroup whitelist with arbitrary 'allow' entries not supported, ignoring.");
1110
1111 /* We ignore the 'deny' entry as for us that's implied */
1112 continue;
1113 }
1114
1115 if (!data.r && !data.w && !data.m) {
1116 json_log(v, flags|LOG_WARNING, 0, "Device cgroup whitelist entry with no effect found, ignoring.");
1117 continue;
1118 }
1119
1120 if (data.minor != (unsigned) -1 && data.major == (unsigned) -1) {
1121 json_log(v, flags, 0, "Device cgroup whitelist entries with minors but no majors not supported.");
1122 return -EOPNOTSUPP;
1123 }
1124
1125 if (data.major != (unsigned) -1 && data.type == 0) {
1126 json_log(v, flags, 0, "Device cgroup whitelist entries with majors but no device node type not supported.");
1127 return -EOPNOTSUPP;
1128 }
1129
1130 if (data.type == 0) {
1131 if (data.r && data.w && data.m) /* a catchall whitelist entry means we are looking at a noop */
1132 noop = true;
1133 else {
1134 json_log(v, flags, 0, "Device cgroup whitelist entries with no type not supported.");
1135 return -EOPNOTSUPP;
1136 }
1137 }
1138
1139 a = reallocarray(list, n_list + 1, sizeof(struct device_data));
1140 if (!a)
1141 return log_oom();
1142
1143 list = a;
1144 list[n_list++] = data;
1145 }
1146
1147 if (noop)
1148 return 0;
1149
1150 r = settings_allocate_properties(s);
1151 if (r < 0)
1152 return r;
1153
1154 r = sd_bus_message_open_container(s->properties, 'r', "sv");
1155 if (r < 0)
1156 return bus_log_create_error(r);
1157
1158 r = sd_bus_message_append(s->properties, "s", "DeviceAllow");
1159 if (r < 0)
1160 return bus_log_create_error(r);
1161
1162 r = sd_bus_message_open_container(s->properties, 'v', "a(ss)");
1163 if (r < 0)
1164 return bus_log_create_error(r);
1165
1166 r = sd_bus_message_open_container(s->properties, 'a', "(ss)");
1167 if (r < 0)
1168 return bus_log_create_error(r);
1169
1170 for (i = 0; i < n_list; i++) {
1171 _cleanup_free_ char *pattern = NULL;
1172 char access[4];
1173 size_t n = 0;
1174
1175 if (list[i].minor == (unsigned) -1) {
1176 const char *t;
1177
1178 if (list[i].type == S_IFBLK)
1179 t = "block";
1180 else {
1181 assert(list[i].type == S_IFCHR);
1182 t = "char";
1183 }
1184
1185 if (list[i].major == (unsigned) -1) {
1186 pattern = strjoin(t, "-*");
1187 if (!pattern)
1188 return log_oom();
1189 } else {
1190 if (asprintf(&pattern, "%s-%u", t, list[i].major) < 0)
1191 return log_oom();
1192 }
1193
1194 } else {
1195 assert(list[i].major != (unsigned) -1); /* If a minor is specified, then a major also needs to be specified */
1196
1197 r = device_path_make_major_minor(list[i].type, makedev(list[i].major, list[i].minor), &pattern);
1198 if (r < 0)
1199 return log_oom();
1200 }
1201
1202 if (list[i].r)
1203 access[n++] = 'r';
1204 if (list[i].w)
1205 access[n++] = 'w';
1206 if (list[i].m)
1207 access[n++] = 'm';
1208 access[n] = 0;
1209
1210 assert(n > 0);
1211
1212 r = sd_bus_message_append(s->properties, "(ss)", pattern, access);
1213 if (r < 0)
1214 return bus_log_create_error(r);
1215 }
1216
1217 r = sd_bus_message_close_container(s->properties);
1218 if (r < 0)
1219 return bus_log_create_error(r);
1220
1221 r = sd_bus_message_close_container(s->properties);
1222 if (r < 0)
1223 return bus_log_create_error(r);
1224
1225 r = sd_bus_message_close_container(s->properties);
1226 if (r < 0)
1227 return bus_log_create_error(r);
1228
1229 return 0;
1230}
1231
1232static int oci_cgroup_memory_limit(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1233 uint64_t *m = userdata;
1234 uintmax_t k;
1235
1236 assert(m);
1237
1238 if (json_variant_is_negative(v)) {
1239 *m = UINT64_MAX;
1240 return 0;
1241 }
1242
1243 if (!json_variant_is_unsigned(v)) {
1244 json_log(v, flags, 0, "Memory limit is not an unsigned integer");
1245 return -EINVAL;
1246 }
1247
1248 k = json_variant_unsigned(v);
1249 if (k >= UINT64_MAX) {
1250 json_log(v, flags, 0, "Memory limit too large: %ji", k);
1251 return -ERANGE;
1252 }
1253
1254 *m = (uint64_t) k;
1255 return 0;
1256}
1257
1258static int oci_cgroup_memory(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1259
1260 struct memory_data {
1261 uint64_t limit;
1262 uint64_t reservation;
1263 uint64_t swap;
1264 } data = {
1265 .limit = UINT64_MAX,
1266 .reservation = UINT64_MAX,
1267 .swap = UINT64_MAX,
1268 };
1269
1270 static const JsonDispatch table[] = {
1271 { "limit", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, limit), 0 },
1272 { "reservation", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, reservation), 0 },
1273 { "swap", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, swap), 0 },
1274 { "kernel", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
1275 { "kernelTCP", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
1276 { "swapiness", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
1277 { "disableOOMKiller", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
1278 {}
1279 };
1280
1281 Settings *s = userdata;
1282 int r;
1283
1284 r = json_dispatch(v, table, oci_unexpected, flags, &data);
1285 if (r < 0)
1286 return r;
1287
1288 if (data.swap != UINT64_MAX) {
1289 if (data.limit == UINT64_MAX)
1290 json_log(v, flags|LOG_WARNING, 0, "swap limit without memory limit is not supported, ignoring.");
1291 else if (data.swap < data.limit)
1292 json_log(v, flags|LOG_WARNING, 0, "swap limit is below memory limit, ignoring.");
1293 else {
1294 r = settings_allocate_properties(s);
1295 if (r < 0)
1296 return r;
1297
1298 r = sd_bus_message_append(s->properties, "(sv)", "MemorySwapMax", "t", data.swap - data.limit);
1299 if (r < 0)
1300 return bus_log_create_error(r);
1301 }
1302 }
1303
1304 if (data.limit != UINT64_MAX) {
1305 r = settings_allocate_properties(s);
1306 if (r < 0)
1307 return r;
1308
1309 r = sd_bus_message_append(s->properties, "(sv)", "MemoryMax", "t", data.limit);
1310 if (r < 0)
1311 return bus_log_create_error(r);
1312 }
1313
1314 if (data.reservation != UINT64_MAX) {
1315 r = settings_allocate_properties(s);
1316 if (r < 0)
1317 return r;
1318
1319 r = sd_bus_message_append(s->properties, "(sv)", "MemoryLow", "t", data.reservation);
1320 if (r < 0)
1321 return bus_log_create_error(r);
1322 }
1323
1324 return 0;
1325}
1326
1327struct cpu_data {
1328 uint64_t shares;
1329 uint64_t quota;
1330 uint64_t period;
1331 cpu_set_t *cpuset;
1332 unsigned ncpus;
1333};
1334
1335static int oci_cgroup_cpu_shares(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1336 uint64_t *u = userdata;
1337 uintmax_t k;
1338
1339 assert(u);
1340
1341 k = json_variant_unsigned(v);
1342 if (k < CGROUP_CPU_SHARES_MIN || k > CGROUP_CPU_SHARES_MAX) {
1343 json_log(v, flags, 0, "shares value out of range.");
1344 return -ERANGE;
1345 }
1346
1347 *u = (uint64_t) k;
1348 return 0;
1349}
1350
1351static int oci_cgroup_cpu_quota(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1352 uint64_t *u = userdata;
1353 uintmax_t k;
1354
1355 assert(u);
1356
1357 k = json_variant_unsigned(v);
1358 if (k <= 0 || k >= UINT64_MAX) {
1359 json_log(v, flags, 0, "period/quota value out of range.");
1360 return -ERANGE;
1361 }
1362
1363 *u = (uint64_t) k;
1364 return 0;
1365}
1366
1367static int oci_cgroup_cpu_cpus(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1368 struct cpu_data *data = userdata;
1369 cpu_set_t *set;
1370 const char *n;
1371 int ncpus;
1372
1373 assert(data);
1374
1375 assert_se(n = json_variant_string(v));
1376
1377 ncpus = parse_cpu_set(n, &set);
1378 if (ncpus < 0)
1379 return json_log(v, flags, ncpus, "Failed to parse CPU set specification: %s", n);
1380
1381 CPU_FREE(data->cpuset);
1382 data->cpuset = set;
1383 data->ncpus = ncpus;
1384
1385 return 0;
1386}
1387
1388static int oci_cgroup_cpu(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1389
1390 static const JsonDispatch table[] = {
1391 { "shares", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_shares, offsetof(struct cpu_data, shares), 0 },
1392 { "quota", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_quota, offsetof(struct cpu_data, quota), 0 },
1393 { "period", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_quota, offsetof(struct cpu_data, period), 0 },
1394 { "realtimeRuntime", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, 0 },
1395 { "realtimePeriod", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, 0 },
1396 { "cpus", JSON_VARIANT_STRING, oci_cgroup_cpu_cpus, 0, 0 },
1397 { "mems", JSON_VARIANT_STRING, oci_unsupported, 0, 0 },
1398 {}
1399 };
1400
1401 struct cpu_data data = {
1402 .shares = UINT64_MAX,
1403 .quota = UINT64_MAX,
1404 .period = UINT64_MAX,
1405 };
1406
1407 Settings *s = userdata;
1408 int r;
1409
1410 r = json_dispatch(v, table, oci_unexpected, flags, &data);
1411 if (r < 0) {
1412 CPU_FREE(data.cpuset);
1413 return r;
1414 }
1415
1416 CPU_FREE(s->cpuset);
1417 s->cpuset = data.cpuset;
1418 s->cpuset_ncpus = data.ncpus;
1419
1420 if (data.shares != UINT64_MAX) {
1421 r = settings_allocate_properties(s);
1422 if (r < 0)
1423 return r;
1424
1425 r = sd_bus_message_append(s->properties, "(sv)", "CPUShares", "t", data.shares);
1426 if (r < 0)
1427 return bus_log_create_error(r);
1428 }
1429
1430 if (data.quota != UINT64_MAX && data.period != UINT64_MAX) {
1431 r = settings_allocate_properties(s);
1432 if (r < 0)
1433 return r;
1434
1435 r = sd_bus_message_append(s->properties, "(sv)", "CPUQuotaPerSecUSec", "t", (uint64_t) (data.quota * USEC_PER_SEC / data.period));
1436 if (r < 0)
1437 return bus_log_create_error(r);
1438
1439 } else if ((data.quota != UINT64_MAX) != (data.period != UINT64_MAX)) {
1440 json_log(v, flags, 0, "CPU quota and period not used together.");
1441 return -EINVAL;
1442 }
1443
1444 return 0;
1445}
1446
1447static int oci_cgroup_block_io_weight(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1448 Settings *s = userdata;
1449 uintmax_t k;
1450 int r;
1451
1452 assert(s);
1453
1454 k = json_variant_unsigned(v);
1455 if (k < CGROUP_BLKIO_WEIGHT_MIN || k > CGROUP_BLKIO_WEIGHT_MAX) {
1456 json_log(v, flags, 0, "Block I/O weight out of range.");
1457 return -ERANGE;
1458 }
1459
1460 r = settings_allocate_properties(s);
1461 if (r < 0)
1462 return r;
1463
1464 r = sd_bus_message_append(s->properties, "(sv)", "BlockIOWeight", "t", (uint64_t) k);
1465 if (r < 0)
1466 return bus_log_create_error(r);
1467
1468 return 0;
1469}
1470
1471static int oci_cgroup_block_io_weight_device(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1472 Settings *s = userdata;
1473 JsonVariant *e;
1474 int r;
1475
1476 assert(s);
1477
1478 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1479 struct device_data {
1480 unsigned major;
1481 unsigned minor;
1482 uintmax_t weight;
1483 } data = {
1484 .major = (unsigned) -1,
1485 .minor = (unsigned) -1,
1486 .weight = UINTMAX_MAX,
1487 };
1488
1489 static const JsonDispatch table[] = {
1490 { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), JSON_MANDATORY },
1491 { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), JSON_MANDATORY },
1492 { "weight", JSON_VARIANT_UNSIGNED, json_dispatch_unsigned, offsetof(struct device_data, weight), 0 },
1493 { "leafWeight", JSON_VARIANT_INTEGER, oci_unsupported, 0, JSON_PERMISSIVE },
1494 {}
1495 };
1496
1497 _cleanup_free_ char *path = NULL;
1498
1499 r = json_dispatch(e, table, oci_unexpected, flags, &data);
1500 if (r < 0)
1501 return r;
1502
1503 if (data.weight == UINTMAX_MAX)
1504 continue;
1505
1506 if (data.weight < CGROUP_BLKIO_WEIGHT_MIN || data.weight > CGROUP_BLKIO_WEIGHT_MAX) {
1507 json_log(v, flags, 0, "Block I/O device weight out of range.");
1508 return -ERANGE;
1509 }
1510
1511 r = device_path_make_major_minor(S_IFBLK, makedev(data.major, data.minor), &path);
1512 if (r < 0)
1513 return json_log(v, flags, r, "Failed to build device path: %m");
1514
1515 r = settings_allocate_properties(s);
1516 if (r < 0)
1517 return r;
1518
1519 r = sd_bus_message_append(s->properties, "(sv)", "BlockIODeviceWeight", "a(st)", 1, path, (uint64_t) data.weight);
1520 if (r < 0)
1521 return bus_log_create_error(r);
1522 }
1523
1524 return 0;
1525}
1526
1527static int oci_cgroup_block_io_throttle(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1528 Settings *s = userdata;
1529 const char *pname;
1530 JsonVariant *e;
1531 int r;
1532
1533 assert(s);
1534
1535 pname = streq(name, "throttleReadBpsDevice") ? "IOReadBandwidthMax" :
1536 streq(name, "throttleWriteBpsDevice") ? "IOWriteBandwidthMax" :
1537 streq(name, "throttleReadIOPSDevice") ? "IOReadIOPSMax" :
1538 "IOWriteIOPSMax";
1539
1540 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1541 struct device_data {
1542 unsigned major;
1543 unsigned minor;
1544 uintmax_t rate;
1545 } data = {
1546 .major = (unsigned) -1,
1547 .minor = (unsigned) -1,
1548 };
1549
1550 static const JsonDispatch table[] = {
1551 { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), JSON_MANDATORY },
1552 { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), JSON_MANDATORY },
1553 { "rate", JSON_VARIANT_UNSIGNED, json_dispatch_unsigned, offsetof(struct device_data, rate), JSON_MANDATORY },
1554 {}
1555 };
1556
1557 _cleanup_free_ char *path = NULL;
1558
1559 r = json_dispatch(e, table, oci_unexpected, flags, &data);
1560 if (r < 0)
1561 return r;
1562
1563 if (data.rate >= UINT64_MAX) {
1564 json_log(v, flags, 0, "Block I/O device rate out of range.");
1565 return -ERANGE;
1566 }
1567
1568 r = device_path_make_major_minor(S_IFBLK, makedev(data.major, data.minor), &path);
1569 if (r < 0)
1570 return json_log(v, flags, r, "Failed to build device path: %m");
1571
1572 r = settings_allocate_properties(s);
1573 if (r < 0)
1574 return r;
1575
1576 r = sd_bus_message_append(s->properties, "(sv)", pname, "a(st)", 1, path, (uint64_t) data.rate);
1577 if (r < 0)
1578 return bus_log_create_error(r);
1579 }
1580
1581 return 0;
1582}
1583
1584static int oci_cgroup_block_io(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1585
1586 static const JsonDispatch table[] = {
1587 { "weight", JSON_VARIANT_UNSIGNED, oci_cgroup_block_io_weight, 0, 0 },
1588 { "leafWeight", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, JSON_PERMISSIVE },
1589 { "weightDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_weight_device, 0, 0 },
1590 { "throttleReadBpsDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
1591 { "throttleWriteBpsDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
1592 { "throttleReadIOPSDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
1593 { "throttleWriteIOPSDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
1594 {}
1595 };
1596
1597 return json_dispatch(v, table, oci_unexpected, flags, userdata);
1598}
1599
1600static int oci_cgroup_pids(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1601
1602 static const JsonDispatch table[] = {
1603 { "limit", JSON_VARIANT_NUMBER, json_dispatch_variant, 0, JSON_MANDATORY },
1604 {}
1605 };
1606
1607 _cleanup_(json_variant_unrefp) JsonVariant *k = NULL;
1608 Settings *s = userdata;
1609 uint64_t m;
1610 int r;
1611
1612 assert(s);
1613
1614 r = json_dispatch(v, table, oci_unexpected, flags, &k);
1615 if (r < 0)
1616 return r;
1617
1618 if (json_variant_is_negative(k))
1619 m = UINT64_MAX;
1620 else {
1621 if (!json_variant_is_unsigned(k)) {
1622 json_log(k, flags, 0, "pids limit not unsigned integer, refusing.");
1623 return -EINVAL;
1624 }
1625
1626 m = (uint64_t) json_variant_unsigned(k);
1627
1628 if ((uintmax_t) m != json_variant_unsigned(k)) {
1629 json_log(v, flags, 0, "pids limit out of range, refusing.");
1630 return -EINVAL;
1631 }
1632 }
1633
1634 r = settings_allocate_properties(s);
1635 if (r < 0)
1636 return r;
1637
1638 r = sd_bus_message_append(s->properties, "(sv)", "TasksMax", "t", m);
1639 if (r < 0)
1640 return bus_log_create_error(r);
1641
1642 return 0;
1643}
1644
1645static int oci_resources(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1646
1647 static const JsonDispatch table[] = {
1648 { "devices", JSON_VARIANT_ARRAY, oci_cgroup_devices, 0, 0 },
1649 { "memory", JSON_VARIANT_OBJECT, oci_cgroup_memory, 0, 0 },
1650 { "cpu", JSON_VARIANT_OBJECT, oci_cgroup_cpu, 0, 0 },
1651 { "blockIO", JSON_VARIANT_OBJECT, oci_cgroup_block_io, 0, 0 },
1652 { "hugepageLimits", JSON_VARIANT_ARRAY, oci_unsupported, 0, 0 },
1653 { "network", JSON_VARIANT_OBJECT, oci_unsupported, 0, 0 },
1654 { "pids", JSON_VARIANT_OBJECT, oci_cgroup_pids, 0, 0 },
1655 { "rdma", JSON_VARIANT_OBJECT, oci_unsupported, 0, 0 },
1656 {}
1657 };
1658
1659 return json_dispatch(v, table, oci_unexpected, flags, userdata);
1660}
1661
1662static bool sysctl_key_valid(const char *s) {
1663 bool dot = true;
1664
1665 /* Note that we are a bit stricter here than in systemd-sysctl, as that inherited semantics from the old sysctl
1666 * tool, which were really weird (as it swaps / and . in both ways) */
1667
1668 if (isempty(s))
1669 return false;
1670
1671 for (; *s; s++) {
1672
1673 if (*s <= ' ' || *s >= 127)
1674 return false;
1675 if (*s == '/')
1676 return false;
1677 if (*s == '.') {
1678
1679 if (dot) /* Don't allow two dots next to each other (or at the beginning) */
1680 return false;
1681
1682 dot = true;
1683 } else
1684 dot = false;
1685 }
1686
1687 if (dot) /* don't allow a dot at the end */
1688 return false;
1689
1690 return true;
1691}
1692
1693static int oci_sysctl(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1694 Settings *s = userdata;
1695 JsonVariant *k, *w;
1696 int r;
1697
1698 assert(s);
1699
1700 JSON_VARIANT_OBJECT_FOREACH(k, w, v) {
1701 const char *n, *m;
1702
1703 if (!json_variant_is_string(w)) {
1704 json_log(v, flags, 0, "sysctl parameter is not a string, refusing.");
1705 return -EINVAL;
1706 }
1707
1708 assert_se(n = json_variant_string(k));
1709 assert_se(m = json_variant_string(w));
1710
1711 if (sysctl_key_valid(n)) {
1712 json_log(v, flags, 0, "sysctl key invalid, refusing: %s", n);
1713 return -EINVAL;
1714 }
1715
1716 r = strv_extend_strv(&s->sysctl, STRV_MAKE(n, m), false);
1717 if (r < 0)
1718 return log_oom();
1719 }
1720
1721 return 0;
1722}
1723
1724static int oci_seccomp_action_from_string(const char *name, uint32_t *ret) {
1725
1726 static const struct {
1727 const char *name;
1728 uint32_t action;
1729 } table[] = {
1730 { "SCMP_ACT_ALLOW", SCMP_ACT_ALLOW },
1731 { "SCMP_ACT_ERRNO", SCMP_ACT_ERRNO(EPERM) }, /* the OCI spec doesn't document the error, but it appears EPERM is supposed to be used */
1732 { "SCMP_ACT_KILL", SCMP_ACT_KILL },
1733#ifdef SCMP_ACT_LOG
1734 { "SCMP_ACT_LOG", SCMP_ACT_LOG },
1735#endif
1736 { "SCMP_ACT_TRAP", SCMP_ACT_TRAP },
1737
1738 /* We don't support SCMP_ACT_TRACE because that requires a tracer, and that doesn't really make sense
1739 * here */
1740 };
1741
1742 size_t i;
1743
1744 for (i = 0; i < ELEMENTSOF(table); i++)
1745 if (streq_ptr(name, table[i].name)) {
1746 *ret = table[i].action;
1747 return 0;
1748 }
1749
1750 return -EINVAL;
1751}
1752
1753static int oci_seccomp_arch_from_string(const char *name, uint32_t *ret) {
1754
1755 static const struct {
1756 const char *name;
1757 uint32_t arch;
1758 } table[] = {
1759 { "SCMP_ARCH_AARCH64", SCMP_ARCH_AARCH64 },
1760 { "SCMP_ARCH_ARM", SCMP_ARCH_ARM },
1761 { "SCMP_ARCH_MIPS", SCMP_ARCH_MIPS },
1762 { "SCMP_ARCH_MIPS64", SCMP_ARCH_MIPS64 },
1763 { "SCMP_ARCH_MIPS64N32", SCMP_ARCH_MIPS64N32 },
1764 { "SCMP_ARCH_MIPSEL", SCMP_ARCH_MIPSEL },
1765 { "SCMP_ARCH_MIPSEL64", SCMP_ARCH_MIPSEL64 },
1766 { "SCMP_ARCH_MIPSEL64N32", SCMP_ARCH_MIPSEL64N32 },
1767 { "SCMP_ARCH_NATIVE", SCMP_ARCH_NATIVE },
1768#ifdef SCMP_ARCH_PARISC
1769 { "SCMP_ARCH_PARISC", SCMP_ARCH_PARISC },
1770#endif
1771#ifdef SCMP_ARCH_PARISC64
1772 { "SCMP_ARCH_PARISC64", SCMP_ARCH_PARISC64 },
1773#endif
1774 { "SCMP_ARCH_PPC", SCMP_ARCH_PPC },
1775 { "SCMP_ARCH_PPC64", SCMP_ARCH_PPC64 },
1776 { "SCMP_ARCH_PPC64LE", SCMP_ARCH_PPC64LE },
1777 { "SCMP_ARCH_S390", SCMP_ARCH_S390 },
1778 { "SCMP_ARCH_S390X", SCMP_ARCH_S390X },
1779 { "SCMP_ARCH_X32", SCMP_ARCH_X32 },
1780 { "SCMP_ARCH_X86", SCMP_ARCH_X86 },
1781 { "SCMP_ARCH_X86_64", SCMP_ARCH_X86_64 },
1782 };
1783
1784 size_t i;
1785
1786 for (i = 0; i < ELEMENTSOF(table); i++)
1787 if (streq_ptr(table[i].name, name)) {
1788 *ret = table[i].arch;
1789 return 0;
1790 }
1791
1792 return -EINVAL;
1793}
1794
1795static int oci_seccomp_compare_from_string(const char *name, enum scmp_compare *ret) {
1796
1797 static const struct {
1798 const char *name;
1799 enum scmp_compare op;
1800 } table[] = {
1801 { "SCMP_CMP_NE", SCMP_CMP_NE },
1802 { "SCMP_CMP_LT", SCMP_CMP_LT },
1803 { "SCMP_CMP_LE", SCMP_CMP_LE },
1804 { "SCMP_CMP_EQ", SCMP_CMP_EQ },
1805 { "SCMP_CMP_GE", SCMP_CMP_GE },
1806 { "SCMP_CMP_GT", SCMP_CMP_GT },
1807 { "SCMP_CMP_MASKED_EQ", SCMP_CMP_MASKED_EQ },
1808 };
1809
1810 size_t i;
1811
1812 for (i = 0; i < ELEMENTSOF(table); i++)
1813 if (streq_ptr(table[i].name, name)) {
1814 *ret = table[i].op;
1815 return 0;
1816 }
1817
1818 return -EINVAL;
1819}
1820
1821static int oci_seccomp_archs(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1822 scmp_filter_ctx *sc = userdata;
1823 JsonVariant *e;
1824 int r;
1825
1826 assert(sc);
1827
1828 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1829 uint32_t a;
1830
1831 if (!json_variant_is_string(e)) {
1832 json_log(e, flags, 0, "Architecture entry is not a string");
1833 return -EINVAL;
1834 }
1835
1836 r = oci_seccomp_arch_from_string(json_variant_string(e), &a);
1837 if (r < 0)
1838 return json_log(e, flags, r, "Unknown architecture: %s", json_variant_string(e));
1839
1840 r = seccomp_arch_add(sc, a);
1841 if (r == -EEXIST)
1842 continue;
1843 if (r < 0)
1844 return json_log(e, flags, r, "Failed to add architecture to seccomp filter: %m");
1845 }
1846
1847 return 0;
1848}
1849
1850struct syscall_rule {
1851 char **names;
1852 uint32_t action;
1853 struct scmp_arg_cmp *arguments;
1854 size_t n_arguments;
1855};
1856
1857static void syscall_rule_free(struct syscall_rule *rule) {
1858 assert(rule);
1859
1860 strv_free(rule->names);
1861 free(rule->arguments);
1862};
1863
1864static int oci_seccomp_action(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1865 uint32_t *action = userdata;
1866 int r;
1867
1868 assert(action);
1869
1870 r = oci_seccomp_action_from_string(json_variant_string(v), action);
1871 if (r < 0)
1872 return json_log(v, flags, r, "Unknown system call action '%s': %m", json_variant_string(v));
1873
1874 return 0;
1875}
1876
1877static int oci_seccomp_op(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1878 enum scmp_compare *op = userdata;
1879 int r;
1880
1881 assert(op);
1882
1883 r = oci_seccomp_compare_from_string(json_variant_string(v), op);
1884 if (r < 0)
1885 return json_log(v, flags, r, "Unknown seccomp operator '%s': %m", json_variant_string(v));
1886
1887 return 0;
1888}
1889
1890static int oci_seccomp_args(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1891 struct syscall_rule *rule = userdata;
1892 JsonVariant *e;
1893 int r;
1894
1895 assert(rule);
1896
1897 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1898 static const struct JsonDispatch table[] = {
1899 { "index", JSON_VARIANT_UNSIGNED, json_dispatch_uint32, offsetof(struct scmp_arg_cmp, arg), JSON_MANDATORY },
1900 { "value", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct scmp_arg_cmp, datum_a), JSON_MANDATORY },
1901 { "valueTwo", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct scmp_arg_cmp, datum_b), 0 },
1902 { "op", JSON_VARIANT_STRING, oci_seccomp_op, offsetof(struct scmp_arg_cmp, op), JSON_MANDATORY },
1903 {},
1904 };
1905
1906 struct scmp_arg_cmp *a, *p;
1907 int expected;
1908
1909 a = reallocarray(rule->arguments, rule->n_arguments + 1, sizeof(struct syscall_rule));
1910 if (!a)
1911 return log_oom();
1912
1913 rule->arguments = a;
1914 p = rule->arguments + rule->n_arguments;
1915
1916 *p = (struct scmp_arg_cmp) {
1917 .arg = 0,
1918 .datum_a = 0,
1919 .datum_b = 0,
1920 .op = 0,
1921 };
1922
1923 r = json_dispatch(e, table, oci_unexpected, flags, p);
1924 if (r < 0)
1925 return r;
1926
1927 expected = p->op == SCMP_CMP_MASKED_EQ ? 4 : 3;
1928 if (r != expected)
1929 json_log(e, flags|JSON_WARNING, 0, "Wrong number of system call arguments for JSON data data, ignoring.");
1930
1931 /* Note that we are a bit sloppy here and do not insist that SCMP_CMP_MASKED_EQ gets two datum values,
1932 * and the other only one. That's because buildah for example by default calls things with
1933 * SCMP_CMP_MASKED_EQ but only one argument. We use 0 when the value is not specified. */
1934
1935 rule->n_arguments++;
1936 }
1937
1938 return 0;
1939}
1940
1941static int oci_seccomp_syscalls(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1942 scmp_filter_ctx *sc = userdata;
1943 JsonVariant *e;
1944 int r;
1945
1946 assert(sc);
1947
1948 JSON_VARIANT_ARRAY_FOREACH(e, v) {
1949 static const JsonDispatch table[] = {
1950 { "names", JSON_VARIANT_ARRAY, json_dispatch_strv, offsetof(struct syscall_rule, names), JSON_MANDATORY },
1951 { "action", JSON_VARIANT_STRING, oci_seccomp_action, offsetof(struct syscall_rule, action), JSON_MANDATORY },
1952 { "args", JSON_VARIANT_ARRAY, oci_seccomp_args, 0, 0 },
1953 };
1954 struct syscall_rule rule = {
1955 .action = (uint32_t) -1,
1956 };
1957 char **i;
1958
1959 r = json_dispatch(e, table, oci_unexpected, flags, &rule);
1960 if (r < 0)
1961 goto fail_rule;
1962
1963 if (strv_isempty(rule.names)) {
1964 json_log(e, flags, 0, "System call name list is empty.");
1965 r = -EINVAL;
1966 goto fail_rule;
1967 }
1968
1969 STRV_FOREACH(i, rule.names) {
1970 int nr;
1971
1972 nr = seccomp_syscall_resolve_name(*i);
1973 if (nr == __NR_SCMP_ERROR) {
1974 log_debug("Unknown syscall %s, skipping.", *i);
1975 continue;
1976 }
1977
1978 r = seccomp_rule_add_array(sc, rule.action, nr, rule.n_arguments, rule.arguments);
1979 if (r < 0)
1980 goto fail_rule;
1981 }
1982
1983 syscall_rule_free(&rule);
1984 continue;
1985
1986 fail_rule:
1987 syscall_rule_free(&rule);
1988 return r;
1989 }
1990
1991 return 0;
1992}
1993
1994static int oci_seccomp(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
1995
1996 static const JsonDispatch table[] = {
1997 { "defaultAction", JSON_VARIANT_STRING, NULL, 0, JSON_MANDATORY },
1998 { "architectures", JSON_VARIANT_ARRAY, oci_seccomp_archs, 0, 0 },
1999 { "syscalls", JSON_VARIANT_ARRAY, oci_seccomp_syscalls, 0, 0 },
2000 {}
2001 };
2002
2003 _cleanup_(seccomp_releasep) scmp_filter_ctx sc = NULL;
2004 Settings *s = userdata;
2005 JsonVariant *def;
2006 uint32_t d;
2007 int r;
2008
2009 assert(s);
2010
2011 def = json_variant_by_key(v, "defaultAction");
2012 if (!def) {
2013 json_log(v, flags, 0, "defaultAction element missing.");
2014 return -EINVAL;
2015 }
2016
2017 if (!json_variant_is_string(def)) {
2018 json_log(def, flags, 0, "defaultAction is not a string.");
2019 return -EINVAL;
2020 }
2021
2022 r = oci_seccomp_action_from_string(json_variant_string(def), &d);
2023 if (r < 0)
2024 return json_log(def, flags, r, "Unknown default action: %s", json_variant_string(def));
2025
2026 sc = seccomp_init(d);
2027 if (!sc) {
2028 log_error("Couldn't allocate seccomp object.");
2029 return -ENOMEM;
2030 }
2031
2032 r = json_dispatch(v, table, oci_unexpected, flags, sc);
2033 if (r < 0)
2034 return r;
2035
2036 seccomp_release(s->seccomp);
2037 s->seccomp = TAKE_PTR(sc);
2038
2039 return 0;
2040}
2041
2042static int oci_rootfs_propagation(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2043 const char *s;
2044
2045 s = json_variant_string(v);
2046
2047 if (streq(s, "shared"))
2048 return 0;
2049
2050 json_log(v, flags|JSON_DEBUG, 0, "Ignoring rootfsPropagation setting '%s'.", s);
2051 return 0;
2052}
2053
2054static int oci_masked_paths(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2055 Settings *s = userdata;
2056 JsonVariant *e;
2057
2058 assert(s);
2059
2060 JSON_VARIANT_ARRAY_FOREACH(e, v) {
2061 _cleanup_free_ char *destination = NULL;
2062 CustomMount *m;
2063 const char *p;
2064
2065 if (!json_variant_is_string(e)) {
2066 json_log(v, flags, 0, "Path is not a string, refusing.");
2067 return -EINVAL;
2068 }
2069
2070 assert_se(p = json_variant_string(e));
2071
2072 if (!path_is_absolute(p)) {
2073 json_log(v, flags, 0, "Path is not not absolute, refusing: %s", p);
2074 return -EINVAL;
2075 }
2076
2077 if (oci_exclude_mount(p))
2078 continue;
2079
2080 destination = strdup(p);
2081 if (!destination)
2082 return log_oom();
2083
2084 m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_INACCESSIBLE);
2085 if (!m)
2086 return log_oom();
2087
2088 m->destination = TAKE_PTR(destination);
2089
2090 /* The spec doesn't say this, but apparently pre-existing implementations are lenient towards
2091 * non-existing paths to mask. Let's hence be too. */
2092 m->graceful = true;
2093 }
2094
2095 return 0;
2096}
2097
2098static int oci_readonly_paths(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2099 Settings *s = userdata;
2100 JsonVariant *e;
2101
2102 assert(s);
2103
2104 JSON_VARIANT_ARRAY_FOREACH(e, v) {
2105 _cleanup_free_ char *source = NULL, *destination = NULL;
2106 CustomMount *m;
2107 const char *p;
2108
2109 if (!json_variant_is_string(e)) {
2110 json_log(v, flags, 0, "Path is not a string, refusing.");
2111 return -EINVAL;
2112 }
2113
2114 assert_se(p = json_variant_string(e));
2115
2116 if (!path_is_absolute(p)) {
2117 json_log(v, flags, 0, "Path is not not absolute, refusing: %s", p);
2118 return -EINVAL;
2119 }
2120
2121 if (oci_exclude_mount(p))
2122 continue;
2123
2124 source = strjoin("+", p);
2125 if (!source)
2126 return log_oom();
2127
2128 destination = strdup(p);
2129 if (!destination)
2130 return log_oom();
2131
2132 m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_BIND);
2133 if (!m)
2134 return log_oom();
2135
2136 m->source = TAKE_PTR(source);
2137 m->destination = TAKE_PTR(destination);
2138 m->read_only = true;
2139 }
2140
2141 return 0;
2142}
2143
2144static int oci_linux(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2145
2146 static const JsonDispatch table[] = {
2147 { "namespaces", JSON_VARIANT_ARRAY, oci_namespaces, 0, 0 },
2148 { "uidMappings", JSON_VARIANT_ARRAY, oci_uid_gid_mappings, 0, 0 },
2149 { "gidMappings", JSON_VARIANT_ARRAY, oci_uid_gid_mappings, 0, 0 },
2150 { "devices", JSON_VARIANT_ARRAY, oci_devices, 0, 0 },
2151 { "cgroupsPath", JSON_VARIANT_STRING, oci_cgroups_path, 0, 0 },
2152 { "resources", JSON_VARIANT_OBJECT, oci_resources, 0, 0 },
2153 { "intelRdt", JSON_VARIANT_OBJECT, oci_unsupported, 0, JSON_PERMISSIVE },
2154 { "sysctl", JSON_VARIANT_OBJECT, oci_sysctl, 0, 0 },
2155 { "seccomp", JSON_VARIANT_OBJECT, oci_seccomp, 0, 0 },
2156 { "rootfsPropagation", JSON_VARIANT_STRING, oci_rootfs_propagation, 0, 0 },
2157 { "maskedPaths", JSON_VARIANT_ARRAY, oci_masked_paths, 0, 0 },
2158 { "readonlyPaths", JSON_VARIANT_ARRAY, oci_readonly_paths, 0, 0 },
2159 { "mountLabel", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
2160 {}
2161 };
2162
2163 return json_dispatch(v, table, oci_unexpected, flags, userdata);
2164}
2165
2166static int oci_hook_timeout(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2167 usec_t *u = userdata;
2168 uintmax_t k;
2169
2170 k = json_variant_unsigned(v);
2171 if (k == 0) {
2172 json_log(v, flags, 0, "Hook timeout cannot be zero.");
2173 return -EINVAL;
2174 }
2175
2176 if (k > (UINT64_MAX-1/USEC_PER_SEC)) {
2177 json_log(v, flags, 0, "Hook timeout too large.");
2178 return -EINVAL;
2179 }
2180
2181 *u = k * USEC_PER_SEC;
2182 return 0;
2183}
2184
2185static int oci_hooks_array(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2186 Settings *s = userdata;
2187 JsonVariant *e;
2188 int r;
2189
2190 assert(s);
2191
2192 JSON_VARIANT_ARRAY_FOREACH(e, v) {
2193
2194 static const JsonDispatch table[] = {
2195 { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(OciHook, path), JSON_MANDATORY },
2196 { "args", JSON_VARIANT_ARRAY, oci_args, offsetof(OciHook, args), 0 },
2197 { "env", JSON_VARIANT_ARRAY, oci_env, offsetof(OciHook, env), 0 },
2198 { "timeout", JSON_VARIANT_UNSIGNED, oci_hook_timeout, offsetof(OciHook, timeout), 0 },
2199 {}
2200 };
2201
2202 OciHook *a, **array, *new_item;
2203 size_t *n_array;
2204
2205 if (streq(name, "prestart")) {
2206 array = &s->oci_hooks_prestart;
2207 n_array = &s->n_oci_hooks_prestart;
2208 } else if (streq(name, "poststart")) {
2209 array = &s->oci_hooks_poststart;
2210 n_array = &s->n_oci_hooks_poststart;
2211 } else {
2212 assert(streq(name, "poststop"));
2213 array = &s->oci_hooks_poststop;
2214 n_array = &s->n_oci_hooks_poststop;
2215 }
2216
2217 a = reallocarray(*array, *n_array + 1, sizeof(OciHook));
2218 if (!a)
2219 return log_oom();
2220
2221 *array = a;
2222 new_item = a + *n_array;
2223
2224 *new_item = (OciHook) {
2225 .timeout = USEC_INFINITY,
2226 };
2227
2228 r = json_dispatch(e, table, oci_unexpected, flags, userdata);
2229 if (r < 0) {
2230 free(new_item->path);
2231 strv_free(new_item->args);
2232 strv_free(new_item->env);
2233 return r;
2234 }
2235
2236 (*n_array) ++;
2237 }
2238
2239 return 0;
2240}
2241
2242static int oci_hooks(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2243
2244 static const JsonDispatch table[] = {
2245 { "prestart", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
2246 { "poststart", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
2247 { "poststop", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
2248 {}
2249 };
2250
2251 return json_dispatch(v, table, oci_unexpected, flags, userdata);
2252}
2253
2254static int oci_annotations(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
2255 JsonVariant *k, *w;
2256
2257 JSON_VARIANT_OBJECT_FOREACH(k, w, v) {
2258 const char *n;
2259
2260 assert_se(n = json_variant_string(k));
2261
2262 if (isempty(n)) {
2263 json_log(k, flags, 0, "Annotation with empty key, refusing.");
2264 return -EINVAL;
2265 }
2266
2267 if (!json_variant_is_string(w)) {
2268 json_log(w, flags, 0, "Annotation has non-string value, refusing.");
2269 return -EINVAL;
2270 }
2271
2272 json_log(k, flags|JSON_DEBUG, 0, "Ignoring annotation '%s' with value '%s'.", n, json_variant_string(w));
2273 }
2274
2275 return 0;
2276}
2277
2278int oci_load(FILE *f, const char *bundle, Settings **ret) {
2279
2280 static const JsonDispatch table[] = {
2281 { "ociVersion", JSON_VARIANT_STRING, NULL, 0, JSON_MANDATORY },
2282 { "process", JSON_VARIANT_OBJECT, oci_process, 0, 0 },
2283 { "root", JSON_VARIANT_OBJECT, oci_root, 0, 0 },
2284 { "hostname", JSON_VARIANT_STRING, oci_hostname, 0, 0 },
2285 { "mounts", JSON_VARIANT_ARRAY, oci_mounts, 0, 0 },
2286 { "linux", JSON_VARIANT_OBJECT, oci_linux, 0, 0 },
2287 { "hooks", JSON_VARIANT_OBJECT, oci_hooks, 0, 0 },
2288 { "annotations", JSON_VARIANT_OBJECT, oci_annotations, 0, 0 },
2289 {}
2290 };
2291
2292 _cleanup_(json_variant_unrefp) JsonVariant *oci = NULL;
2293 _cleanup_(settings_freep) Settings *s = NULL;
2294 unsigned line = 0, column = 0;
2295 JsonVariant *v;
2296 const char *path;
2297 int r;
2298
2299 assert_se(bundle);
2300
2301 path = strjoina(bundle, "/config.json");
2302
2303 r = json_parse_file(f, path, &oci, &line, &column);
2304 if (r < 0) {
2305 if (line != 0 && column != 0)
2306 return log_error_errno(r, "Failed to parse '%s' at %u:%u: %m", path, line, column);
2307 else
2308 return log_error_errno(r, "Failed to parse '%s': %m", path);
2309 }
2310
2311 v = json_variant_by_key(oci, "ociVersion");
2312 if (!v) {
2313 log_error("JSON file '%s' is not an OCI bundle configuration file. Refusing.", path);
2314 return -EINVAL;
2315 }
2316 if (!streq_ptr(json_variant_string(v), "1.0.0")) {
2317 log_error("OCI bundle version not supported: %s", strna(json_variant_string(v)));
2318 return -EINVAL;
2319 }
2320
2321 // {
2322 // _cleanup_free_ char *formatted = NULL;
2323 // assert_se(json_variant_format(oci, JSON_FORMAT_PRETTY|JSON_FORMAT_COLOR, &formatted) >= 0);
2324 // fputs(formatted, stdout);
2325 // }
2326
2327 s = settings_new();
2328 if (!s)
2329 return log_oom();
2330
2331 s->start_mode = START_PID1;
2332 s->resolv_conf = RESOLV_CONF_OFF;
2333 s->link_journal = LINK_NO;
2334 s->timezone = TIMEZONE_OFF;
2335
2336 s->bundle = strdup(bundle);
2337 if (!s->bundle)
2338 return log_oom();
2339
2340 r = json_dispatch(oci, table, oci_unexpected, 0, s);
2341 if (r < 0)
2342 return r;
2343
2344 if (s->properties) {
2345 r = sd_bus_message_seal(s->properties, 0, 0);
2346 if (r < 0)
2347 return log_error_errno(r, "Cannot seal properties bus message: %m");
2348 }
2349
2350 *ret = TAKE_PTR(s);
2351 return 0;
2352}
Unnamed repository; edit this file 'description' to name the repository.