]>
Commit | Line | Data |
---|---|---|
5fbea8a6 CB |
1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
2 | ||
3 | /* | |
4 | * cgfs-ng.c: this is a new, simplified implementation of a filesystem | |
5 | * cgroup backend. The original cgfs.c was designed to be as flexible | |
6 | * as possible. It would try to find cgroup filesystems no matter where | |
7 | * or how you had them mounted, and deduce the most usable mount for | |
8 | * each controller. | |
9 | * | |
10 | * This new implementation assumes that cgroup filesystems are mounted | |
11 | * under /sys/fs/cgroup/clist where clist is either the controller, or | |
12 | * a comma-separated list of controllers. | |
13 | */ | |
14 | ||
15 | #ifndef _GNU_SOURCE | |
1f5596dd | 16 | #define _GNU_SOURCE |
5fbea8a6 | 17 | #endif |
1f5596dd CB |
18 | |
19 | #ifndef FUSE_USE_VERSION | |
20 | #define FUSE_USE_VERSION 26 | |
21 | #endif | |
22 | ||
23 | #define _FILE_OFFSET_BITS 64 | |
24 | ||
5fbea8a6 CB |
25 | #include <ctype.h> |
26 | #include <dirent.h> | |
27 | #include <errno.h> | |
28 | #include <grp.h> | |
29 | #include <linux/kdev_t.h> | |
30 | #include <linux/types.h> | |
31 | #include <poll.h> | |
32 | #include <signal.h> | |
33 | #include <stdint.h> | |
34 | #include <stdio.h> | |
35 | #include <stdlib.h> | |
36 | #include <string.h> | |
37 | #include <sys/mount.h> | |
38 | #include <sys/types.h> | |
39 | #include <unistd.h> | |
40 | ||
1f5596dd CB |
41 | #include "../config.h" |
42 | #include "../macro.h" | |
43 | #include "../memory_utils.h" | |
5fbea8a6 CB |
44 | #include "cgroup.h" |
45 | #include "cgroup2_devices.h" | |
46 | #include "cgroup_utils.h" | |
5fbea8a6 CB |
47 | |
48 | static void free_string_list(char **clist) | |
49 | { | |
50 | int i; | |
51 | ||
52 | if (!clist) | |
53 | return; | |
54 | ||
55 | for (i = 0; clist[i]; i++) | |
56 | free(clist[i]); | |
57 | ||
58 | free(clist); | |
59 | } | |
60 | ||
61 | /* Given a pointer to a null-terminated array of pointers, realloc to add one | |
62 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
63 | * second-to-last entry - that is, the one which is now available for use | |
64 | * (keeping the list null-terminated). | |
65 | */ | |
66 | static int append_null_to_list(void ***list) | |
67 | { | |
68 | int newentry = 0; | |
69 | ||
70 | if (*list) | |
71 | for (; (*list)[newentry]; newentry++) | |
72 | ; | |
73 | ||
74 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
75 | (*list)[newentry + 1] = NULL; | |
76 | return newentry; | |
77 | } | |
78 | ||
79 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
80 | * strings. | |
81 | */ | |
82 | static bool string_in_list(char **list, const char *entry) | |
83 | { | |
84 | int i; | |
85 | ||
86 | if (!list) | |
87 | return false; | |
88 | ||
89 | for (i = 0; list[i]; i++) | |
90 | if (strcmp(list[i], entry) == 0) | |
91 | return true; | |
92 | ||
93 | return false; | |
94 | } | |
95 | ||
96 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into | |
97 | * "name=systemd". Do not fail. | |
98 | */ | |
99 | static char *cg_legacy_must_prefix_named(char *entry) | |
100 | { | |
101 | size_t len; | |
102 | char *prefixed; | |
103 | ||
104 | len = strlen(entry); | |
105 | prefixed = must_realloc(NULL, len + 6); | |
106 | ||
107 | memcpy(prefixed, "name=", STRLITERALLEN("name=")); | |
108 | memcpy(prefixed + STRLITERALLEN("name="), entry, len); | |
109 | prefixed[len + 5] = '\0'; | |
110 | ||
111 | return prefixed; | |
112 | } | |
113 | ||
114 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time | |
115 | * we are called. | |
116 | * | |
117 | * We also handle named subsystems here. Any controller which is not a kernel | |
118 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
119 | * we refuse to use because we're not sure which we have here. | |
120 | * (TODO: We could work around this in some cases by just remounting to be | |
121 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
122 | * | |
123 | * The last entry will always be NULL. | |
124 | */ | |
125 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
126 | char *entry) | |
127 | { | |
128 | int newentry; | |
129 | char *copy; | |
130 | ||
131 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
132 | return; | |
133 | ||
134 | newentry = append_null_to_list((void ***)clist); | |
135 | ||
136 | if (strncmp(entry, "name=", 5) == 0) | |
137 | copy = must_copy_string(entry); | |
138 | else if (string_in_list(klist, entry)) | |
139 | copy = must_copy_string(entry); | |
140 | else | |
141 | copy = cg_legacy_must_prefix_named(entry); | |
142 | ||
143 | (*clist)[newentry] = copy; | |
144 | } | |
145 | ||
146 | /* Given a handler's cgroup data, return the struct hierarchy for the controller | |
147 | * @c, or NULL if there is none. | |
148 | */ | |
149 | static struct hierarchy *cgfsng_get_hierarchy(struct cgroup_ops *ops, | |
150 | const char *controller) | |
151 | { | |
152 | int i; | |
153 | ||
154 | errno = ENOENT; | |
155 | ||
156 | if (!ops->hierarchies) | |
157 | return NULL; | |
158 | ||
159 | for (i = 0; ops->hierarchies[i]; i++) { | |
160 | if (!controller) { | |
161 | /* This is the empty unified hierarchy. */ | |
162 | if (ops->hierarchies[i]->controllers && | |
163 | !ops->hierarchies[i]->controllers[0]) | |
164 | return ops->hierarchies[i]; | |
165 | continue; | |
166 | } else if (pure_unified_layout(ops) && | |
167 | strcmp(controller, "devices") == 0) { | |
168 | if (ops->unified->bpf_device_controller) | |
169 | return ops->unified; | |
170 | break; | |
171 | } | |
172 | ||
173 | if (string_in_list(ops->hierarchies[i]->controllers, controller)) | |
174 | return ops->hierarchies[i]; | |
175 | } | |
176 | ||
177 | return NULL; | |
178 | } | |
179 | ||
180 | static inline struct hierarchy *get_hierarchy(struct cgroup_ops *ops, | |
181 | const char *controller) | |
182 | { | |
183 | return cgfsng_get_hierarchy(ops, controller); | |
184 | } | |
185 | ||
186 | /* Given two null-terminated lists of strings, return true if any string is in | |
187 | * both. | |
188 | */ | |
189 | static bool controller_lists_intersect(char **l1, char **l2) | |
190 | { | |
191 | int i; | |
192 | ||
193 | if (!l1 || !l2) | |
194 | return false; | |
195 | ||
196 | for (i = 0; l1[i]; i++) { | |
197 | if (string_in_list(l2, l1[i])) | |
198 | return true; | |
199 | } | |
200 | ||
201 | return false; | |
202 | } | |
203 | ||
204 | /* For a null-terminated list of controllers @clist, return true if any of those | |
205 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
206 | * Realistically, if one is present, all must be present. | |
207 | */ | |
208 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
209 | { | |
210 | int i; | |
211 | ||
212 | if (!hlist) | |
213 | return false; | |
214 | ||
215 | for (i = 0; hlist[i]; i++) | |
216 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
217 | return true; | |
218 | ||
219 | return false; | |
220 | } | |
221 | ||
222 | /* Get the controllers from a mountinfo line There are other ways we could get | |
223 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
224 | * could parse the mount options. But we simply assume that the mountpoint must | |
225 | * be /sys/fs/cgroup/controller-list | |
226 | */ | |
227 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, | |
228 | int type, char **controllers) | |
229 | { | |
230 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list | |
231 | * for legacy hierarchies. | |
232 | */ | |
233 | int i; | |
234 | char *p2, *tok; | |
235 | char *p = line, *sep = ","; | |
236 | char **aret = NULL; | |
237 | ||
238 | for (i = 0; i < 4; i++) { | |
239 | p = strchr(p, ' '); | |
240 | if (!p) | |
241 | return NULL; | |
242 | p++; | |
243 | } | |
244 | ||
245 | /* Note, if we change how mountinfo works, then our caller will need to | |
246 | * verify /sys/fs/cgroup/ in this field. | |
247 | */ | |
248 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
249 | return NULL; | |
250 | ||
251 | p += 15; | |
252 | p2 = strchr(p, ' '); | |
253 | if (!p2) | |
254 | return NULL; | |
255 | *p2 = '\0'; | |
256 | ||
257 | if (type == CGROUP_SUPER_MAGIC) { | |
258 | __do_free char *dup = NULL; | |
259 | ||
260 | /* strdup() here for v1 hierarchies. Otherwise | |
261 | * lxc_iterate_parts() will destroy mountpoints such as | |
262 | * "/sys/fs/cgroup/cpu,cpuacct". | |
263 | */ | |
264 | dup = must_copy_string(p); | |
265 | if (!dup) | |
266 | return NULL; | |
267 | ||
268 | lxc_iterate_parts (tok, dup, sep) | |
269 | must_append_controller(klist, nlist, &aret, tok); | |
270 | *controllers = move_ptr(dup); | |
271 | } | |
272 | *p2 = ' '; | |
273 | ||
274 | return aret; | |
275 | } | |
276 | ||
277 | static char **cg_unified_make_empty_controller(void) | |
278 | { | |
279 | int newentry; | |
280 | char **aret = NULL; | |
281 | ||
282 | newentry = append_null_to_list((void ***)&aret); | |
283 | aret[newentry] = NULL; | |
284 | return aret; | |
285 | } | |
286 | ||
287 | static char **cg_unified_get_controllers(const char *file) | |
288 | { | |
289 | __do_free char *buf = NULL; | |
290 | char *sep = " \t\n"; | |
291 | char **aret = NULL; | |
292 | char *tok; | |
293 | ||
294 | buf = read_file(file); | |
295 | if (!buf) | |
296 | return NULL; | |
297 | ||
298 | lxc_iterate_parts(tok, buf, sep) { | |
299 | int newentry; | |
300 | char *copy; | |
301 | ||
302 | newentry = append_null_to_list((void ***)&aret); | |
303 | copy = must_copy_string(tok); | |
304 | aret[newentry] = copy; | |
305 | } | |
306 | ||
307 | return aret; | |
308 | } | |
309 | ||
310 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, | |
0fd1b770 | 311 | char *base_path, int type) |
5fbea8a6 CB |
312 | { |
313 | struct hierarchy *new; | |
314 | int newentry; | |
315 | ||
316 | new = zalloc(sizeof(*new)); | |
317 | new->controllers = clist; | |
318 | new->mountpoint = mountpoint; | |
0fd1b770 | 319 | new->base_path = base_path; |
5fbea8a6 CB |
320 | new->version = type; |
321 | ||
322 | newentry = append_null_to_list((void ***)h); | |
323 | (*h)[newentry] = new; | |
324 | return new; | |
325 | } | |
326 | ||
327 | /* Get a copy of the mountpoint from @line, which is a line from | |
328 | * /proc/self/mountinfo. | |
329 | */ | |
330 | static char *cg_hybrid_get_mountpoint(char *line) | |
331 | { | |
332 | int i; | |
333 | size_t len; | |
334 | char *p2; | |
335 | char *p = line, *sret = NULL; | |
336 | ||
337 | for (i = 0; i < 4; i++) { | |
338 | p = strchr(p, ' '); | |
339 | if (!p) | |
340 | return NULL; | |
341 | p++; | |
342 | } | |
343 | ||
344 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
345 | return NULL; | |
346 | ||
347 | p2 = strchr(p + 15, ' '); | |
348 | if (!p2) | |
349 | return NULL; | |
350 | *p2 = '\0'; | |
351 | ||
352 | len = strlen(p); | |
353 | sret = must_realloc(NULL, len + 1); | |
354 | memcpy(sret, p, len); | |
355 | sret[len] = '\0'; | |
356 | return sret; | |
357 | } | |
358 | ||
359 | static void must_append_string(char ***list, char *entry) | |
360 | { | |
361 | int newentry; | |
362 | char *copy; | |
363 | ||
364 | newentry = append_null_to_list((void ***)list); | |
365 | copy = must_copy_string(entry); | |
366 | (*list)[newentry] = copy; | |
367 | } | |
368 | ||
369 | static int get_existing_subsystems(char ***klist, char ***nlist) | |
370 | { | |
371 | __do_free char *line = NULL; | |
372 | __do_fclose FILE *f = NULL; | |
373 | size_t len = 0; | |
374 | ||
dbb1f822 | 375 | f = fopen("/proc/self/cgroup", "re"); |
5fbea8a6 CB |
376 | if (!f) |
377 | return -1; | |
378 | ||
379 | while (getline(&line, &len, f) != -1) { | |
380 | char *p, *p2, *tok; | |
381 | p = strchr(line, ':'); | |
382 | if (!p) | |
383 | continue; | |
384 | p++; | |
385 | p2 = strchr(p, ':'); | |
386 | if (!p2) | |
387 | continue; | |
388 | *p2 = '\0'; | |
389 | ||
390 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup | |
391 | * contains an entry of the form: | |
392 | * | |
393 | * 0::/some/path | |
394 | * | |
395 | * In this case we use "cgroup2" as controller name. | |
396 | */ | |
397 | if ((p2 - p) == 0) { | |
398 | must_append_string(klist, "cgroup2"); | |
399 | continue; | |
400 | } | |
401 | ||
402 | lxc_iterate_parts(tok, p, ",") { | |
403 | if (strncmp(tok, "name=", 5) == 0) | |
404 | must_append_string(nlist, tok); | |
405 | else | |
406 | must_append_string(klist, tok); | |
407 | } | |
408 | } | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
413 | static void trim(char *s) | |
414 | { | |
415 | size_t len; | |
416 | ||
417 | len = strlen(s); | |
418 | while ((len > 1) && (s[len - 1] == '\n')) | |
419 | s[--len] = '\0'; | |
420 | } | |
421 | ||
422 | /* __cg_mount_direct | |
423 | * | |
424 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
425 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
426 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
427 | */ | |
428 | static int __cg_mount_direct(struct hierarchy *h, const char *controllerpath) | |
429 | { | |
430 | __do_free char *controllers = NULL; | |
431 | char *fstype = "cgroup2"; | |
432 | unsigned long flags = 0; | |
433 | int ret; | |
434 | ||
435 | flags |= MS_NOSUID; | |
436 | flags |= MS_NOEXEC; | |
437 | flags |= MS_NODEV; | |
438 | flags |= MS_RELATIME; | |
439 | ||
440 | if (h->version != CGROUP2_SUPER_MAGIC) { | |
441 | controllers = lxc_string_join(",", (const char **)h->controllers, false); | |
442 | if (!controllers) | |
443 | return -ENOMEM; | |
444 | fstype = "cgroup"; | |
445 | } | |
446 | ||
447 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); | |
448 | if (ret < 0) | |
449 | return -1; | |
450 | ||
451 | return 0; | |
452 | } | |
453 | ||
454 | static inline int cg_mount_cgroup_full(struct hierarchy *h, | |
455 | const char *controllerpath) | |
456 | { | |
457 | return __cg_mount_direct(h, controllerpath); | |
458 | } | |
459 | ||
460 | static bool cgfsng_mount(struct cgroup_ops *ops, const char *root) | |
461 | { | |
462 | __do_free char *cgroup_root = NULL; | |
463 | int ret; | |
464 | bool retval = false; | |
465 | ||
466 | if (!ops) | |
467 | return ret_set_errno(false, ENOENT); | |
468 | ||
469 | if (!ops->hierarchies) | |
470 | return true; | |
471 | ||
472 | cgroup_root = must_make_path(root, DEFAULT_CGROUP_MOUNTPOINT, NULL); | |
473 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) | |
474 | return cg_mount_cgroup_full(ops->unified, cgroup_root) == 0; | |
475 | ||
476 | /* mount tmpfs */ | |
477 | ret = safe_mount(NULL, cgroup_root, "tmpfs", | |
478 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
479 | "size=10240k,mode=755", root); | |
480 | if (ret < 0) | |
481 | goto on_error; | |
482 | ||
483 | for (int i = 0; ops->hierarchies[i]; i++) { | |
484 | __do_free char *controllerpath = NULL; | |
485 | struct hierarchy *h = ops->hierarchies[i]; | |
486 | char *controller = strrchr(h->mountpoint, '/'); | |
487 | ||
488 | if (!controller) | |
489 | continue; | |
490 | controller++; | |
491 | ||
492 | controllerpath = must_make_path(cgroup_root, controller, NULL); | |
493 | if (dir_exists(controllerpath)) | |
494 | continue; | |
495 | ||
496 | ret = mkdir(controllerpath, 0755); | |
497 | if (ret < 0) | |
498 | log_error_errno(goto on_error, errno, | |
499 | "Error creating cgroup path: %s", | |
500 | controllerpath); | |
501 | ||
502 | ret = cg_mount_cgroup_full( h, controllerpath); | |
503 | if (ret < 0) | |
504 | goto on_error; | |
505 | } | |
506 | retval = true; | |
507 | ||
508 | on_error: | |
509 | return retval; | |
510 | } | |
511 | ||
5fbea8a6 CB |
512 | static int cgfsng_num_hierarchies(struct cgroup_ops *ops) |
513 | { | |
514 | int i = 0; | |
515 | ||
516 | if (!ops) | |
517 | return ret_set_errno(-1, ENOENT); | |
518 | ||
519 | if (!ops->hierarchies) | |
520 | return 0; | |
521 | ||
522 | for (; ops->hierarchies[i]; i++) | |
523 | ; | |
524 | ||
525 | return i; | |
526 | } | |
527 | ||
528 | static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) | |
529 | { | |
530 | int i; | |
531 | ||
532 | if (!ops) | |
533 | return ret_set_errno(false, ENOENT); | |
534 | ||
535 | if (!ops->hierarchies) | |
536 | return false; | |
537 | ||
538 | /* sanity check n */ | |
539 | for (i = 0; i < n; i++) | |
540 | if (!ops->hierarchies[i]) | |
541 | return ret_set_errno(false, ENOENT); | |
542 | ||
543 | *out = ops->hierarchies[i]->controllers; | |
544 | ||
545 | return true; | |
546 | } | |
547 | ||
1ca6a467 CB |
548 | static bool cgfsng_get(struct cgroup_ops *ops, const char *controller, |
549 | const char *cgroup, const char *file, char **value) | |
550 | { | |
551 | __do_free char *path = NULL; | |
552 | struct hierarchy *h; | |
553 | ||
554 | h = ops->get_hierarchy(ops, controller); | |
555 | if (!h) | |
556 | return false; | |
557 | ||
075387cd | 558 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); |
1ca6a467 CB |
559 | *value = readat_file(h->fd, path); |
560 | return *value != NULL; | |
561 | } | |
562 | ||
66c5e848 CB |
563 | static int cgfsng_get_memory(struct cgroup_ops *ops, const char *cgroup, |
564 | const char *file, char **value) | |
565 | { | |
566 | __do_free char *path = NULL; | |
567 | struct hierarchy *h; | |
568 | int ret; | |
569 | ||
570 | h = ops->get_hierarchy(ops, "memory"); | |
571 | if (!h) | |
572 | return -1; | |
573 | ||
574 | if (!is_unified_hierarchy(h)) { | |
575 | if (strcmp(file, "memory.max") == 0) | |
576 | file = "memory.limit_in_bytes"; | |
577 | else if (strcmp(file, "memory.swap.max") == 0) | |
578 | file = "memory.memsw.limit_in_bytes"; | |
579 | else if (strcmp(file, "memory.swap.current") == 0) | |
580 | file = "memory.memsw.usage_in_bytes"; | |
581 | else if (strcmp(file, "memory.current") == 0) | |
582 | file = "memory.usage_in_bytes"; | |
583 | ret = CGROUP_SUPER_MAGIC; | |
584 | } else { | |
585 | ret = CGROUP2_SUPER_MAGIC; | |
586 | } | |
587 | ||
075387cd | 588 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); |
66c5e848 CB |
589 | *value = readat_file(h->fd, path); |
590 | if (!*value) | |
591 | ret = -1; | |
592 | ||
593 | return ret; | |
594 | } | |
595 | ||
acff9786 CB |
596 | static int cgfsng_get_memory_stats_fd(struct cgroup_ops *ops, const char *cgroup) |
597 | { | |
598 | __do_free char *path = NULL; | |
599 | struct hierarchy *h; | |
600 | ||
601 | h = ops->get_hierarchy(ops, "memory"); | |
602 | if (!h) | |
603 | return -1; | |
604 | ||
605 | path = must_make_path(dot_or_empty(cgroup), cgroup, "memory.stat", NULL); | |
606 | return openat(h->fd, path, O_RDONLY | O_CLOEXEC | O_NOFOLLOW); | |
607 | } | |
608 | ||
66c5e848 CB |
609 | static int cgfsng_get_memory_current(struct cgroup_ops *ops, const char *cgroup, |
610 | char **value) | |
611 | { | |
612 | return cgfsng_get_memory(ops, cgroup, "memory.current", value); | |
613 | } | |
614 | ||
615 | static int cgfsng_get_memory_swap_current(struct cgroup_ops *ops, | |
616 | const char *cgroup, char **value) | |
617 | { | |
618 | return cgfsng_get_memory(ops, cgroup, "memory.swap.current", value); | |
619 | } | |
620 | ||
621 | static int cgfsng_get_memory_max(struct cgroup_ops *ops, const char *cgroup, | |
622 | char **value) | |
623 | { | |
624 | return cgfsng_get_memory(ops, cgroup, "memory.max", value); | |
625 | } | |
626 | ||
627 | static int cgfsng_get_memory_swap_max(struct cgroup_ops *ops, | |
628 | const char *cgroup, char **value) | |
629 | { | |
630 | return cgfsng_get_memory(ops, cgroup, "memory.swap.max", value); | |
631 | } | |
632 | ||
633 | static int cgfsng_get_memory_stats(struct cgroup_ops *ops, const char *cgroup, | |
634 | char **value) | |
635 | { | |
636 | return cgfsng_get_memory(ops, cgroup, "memory.stat", value); | |
637 | } | |
638 | ||
2c3bcd9e CB |
639 | static char *readat_cpuset(int cgroup_fd) |
640 | { | |
641 | __do_free char *val = NULL; | |
642 | ||
643 | val = readat_file(cgroup_fd, "cpuset.cpus"); | |
644 | if (val && strcmp(val, "") != 0) | |
645 | return move_ptr(val); | |
646 | ||
647 | free_disarm(val); | |
648 | val = readat_file(cgroup_fd, "cpuset.cpus.effective"); | |
649 | if (val && strcmp(val, "") != 0) | |
650 | return move_ptr(val); | |
651 | ||
652 | return NULL; | |
653 | } | |
654 | ||
655 | static int cgfsng_get_cpuset_cpus(struct cgroup_ops *ops, const char *cgroup, | |
656 | char **value) | |
657 | { | |
658 | __do_close_prot_errno int cgroup_fd = -EBADF; | |
659 | __do_free char *path = NULL; | |
660 | char *v; | |
661 | struct hierarchy *h; | |
662 | int ret; | |
663 | ||
664 | h = ops->get_hierarchy(ops, "cpuset"); | |
665 | if (!h) | |
666 | return -1; | |
667 | ||
668 | if (!is_unified_hierarchy(h)) | |
669 | ret = CGROUP_SUPER_MAGIC; | |
670 | else | |
671 | ret = CGROUP2_SUPER_MAGIC; | |
672 | ||
673 | *value = NULL; | |
075387cd | 674 | path = must_make_path(dot_or_empty(cgroup), cgroup, NULL); |
2c3bcd9e CB |
675 | cgroup_fd = openat_safe(h->fd, path); |
676 | if (cgroup_fd < 0) { | |
677 | return -1; | |
678 | } | |
679 | v = readat_cpuset(cgroup_fd); | |
680 | if (v) { | |
681 | *value = v; | |
682 | return ret; | |
683 | } | |
684 | ||
685 | /* | |
686 | * cpuset.cpus and cpuset.cpus.effective are empty so we need to look | |
687 | * the nearest ancestor with a non-empty cpuset.cpus{.effective} file. | |
688 | */ | |
689 | for (;;) { | |
690 | int fd; | |
691 | ||
692 | fd = openat_safe(cgroup_fd, "../"); | |
693 | if (fd < 0 || !is_cgroup_fd(fd)) { | |
694 | fprintf(stderr, "2222: %s\n", strerror(errno)); | |
695 | return -1; | |
696 | } | |
697 | ||
698 | close_prot_errno_replace(cgroup_fd, fd); | |
699 | ||
700 | v = readat_cpuset(fd); | |
701 | if (v) { | |
702 | *value = v; | |
703 | return ret; | |
704 | } | |
705 | } | |
706 | ||
707 | return -1; | |
708 | } | |
709 | ||
9a9484ab CB |
710 | static int cgfsng_get_io(struct cgroup_ops *ops, const char *cgroup, |
711 | const char *file, char **value) | |
712 | { | |
713 | __do_free char *path = NULL; | |
714 | struct hierarchy *h; | |
715 | int ret; | |
716 | ||
717 | h = ops->get_hierarchy(ops, "blkio"); | |
718 | if (!h) | |
719 | return -1; | |
720 | ||
721 | if (!is_unified_hierarchy(h)) | |
722 | ret = CGROUP_SUPER_MAGIC; | |
723 | else | |
724 | ret = CGROUP2_SUPER_MAGIC; | |
725 | ||
726 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); | |
727 | *value = readat_file(h->fd, path); | |
728 | if (!*value) { | |
729 | if (errno == ENOENT) | |
730 | errno = EOPNOTSUPP; | |
731 | return ret_errno(errno); | |
732 | } | |
733 | ||
734 | return ret; | |
735 | } | |
736 | ||
737 | static int cgfsng_get_io_service_bytes(struct cgroup_ops *ops, | |
738 | const char *cgroup, char **value) | |
739 | { | |
740 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_bytes_recursive", value); | |
741 | } | |
742 | ||
743 | static int cgfsng_get_io_service_time(struct cgroup_ops *ops, | |
744 | const char *cgroup, char **value) | |
745 | { | |
746 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_time_recursive", value); | |
747 | } | |
748 | ||
749 | static int cgfsng_get_io_serviced(struct cgroup_ops *ops, const char *cgroup, | |
750 | char **value) | |
751 | { | |
752 | return cgfsng_get_io(ops, cgroup, "blkio.io_serviced_recursive", value); | |
753 | } | |
754 | ||
755 | static int cgfsng_get_io_merged(struct cgroup_ops *ops, const char *cgroup, | |
756 | char **value) | |
757 | { | |
758 | return cgfsng_get_io(ops, cgroup, "blkio.io_merged_recursive", value); | |
759 | } | |
760 | ||
761 | static int cgfsng_get_io_wait_time(struct cgroup_ops *ops, const char *cgroup, | |
762 | char **value) | |
763 | { | |
764 | return cgfsng_get_io(ops, cgroup, "blkio.io_wait_time_recursive", value); | |
765 | } | |
766 | ||
77f4399a CB |
767 | static bool cgfsng_can_use_cpuview(struct cgroup_ops *ops) |
768 | { | |
769 | struct hierarchy *cpu, *cpuacct; | |
770 | ||
771 | if (pure_unified_layout(ops)) | |
772 | return false; | |
773 | ||
774 | cpu = ops->get_hierarchy(ops, "cpu"); | |
775 | if (!cpu || is_unified_hierarchy(cpu)) | |
776 | return false; | |
777 | ||
778 | cpuacct = ops->get_hierarchy(ops, "cpuacct"); | |
779 | if (!cpuacct || is_unified_hierarchy(cpuacct)) | |
780 | return false; | |
781 | ||
782 | return true; | |
783 | } | |
784 | ||
5fbea8a6 CB |
785 | /* At startup, parse_hierarchies finds all the info we need about cgroup |
786 | * mountpoints and current cgroups, and stores it in @d. | |
787 | */ | |
788 | static int cg_hybrid_init(struct cgroup_ops *ops) | |
789 | { | |
790 | __do_free char *basecginfo = NULL; | |
791 | __do_free char *line = NULL; | |
792 | __do_fclose FILE *f = NULL; | |
793 | int ret; | |
794 | size_t len = 0; | |
795 | char **klist = NULL, **nlist = NULL; | |
796 | ||
797 | /* Root spawned containers escape the current cgroup, so use init's | |
798 | * cgroups as our base in that case. | |
799 | */ | |
800 | basecginfo = read_file("/proc/1/cgroup"); | |
801 | if (!basecginfo) | |
802 | return ret_set_errno(-1, ENOMEM); | |
803 | ||
804 | ret = get_existing_subsystems(&klist, &nlist); | |
805 | if (ret < 0) | |
806 | return log_error_errno(-1, errno, "Failed to retrieve available legacy cgroup controllers"); | |
807 | ||
dbb1f822 | 808 | f = fopen("/proc/self/mountinfo", "re"); |
5fbea8a6 CB |
809 | if (!f) |
810 | return log_error_errno(-1, errno, "Failed to open \"/proc/self/mountinfo\""); | |
811 | ||
812 | while (getline(&line, &len, f) != -1) { | |
813 | int type; | |
814 | struct hierarchy *new; | |
815 | char *base_cgroup = NULL, *mountpoint = NULL; | |
816 | char **controller_list = NULL; | |
817 | __do_free char *controllers = NULL; | |
818 | ||
819 | type = get_cgroup_version(line); | |
820 | if (type == 0) | |
821 | continue; | |
822 | ||
823 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
824 | continue; | |
825 | ||
826 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
827 | if (type == CGROUP2_SUPER_MAGIC) | |
828 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
829 | else if (type == CGROUP_SUPER_MAGIC) | |
830 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
831 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
832 | if (type == CGROUP_SUPER_MAGIC) | |
833 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
834 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
835 | if (type == CGROUP2_SUPER_MAGIC) | |
836 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
837 | } | |
838 | ||
839 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, | |
840 | type, &controllers); | |
841 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
842 | continue; | |
843 | ||
844 | if (type == CGROUP_SUPER_MAGIC) | |
845 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
846 | ret_set_errno(goto next, EEXIST); | |
847 | ||
848 | mountpoint = cg_hybrid_get_mountpoint(line); | |
849 | if (!mountpoint) | |
850 | log_error_errno(goto next, EINVAL, "Failed parsing mountpoint from \"%s\"", line); | |
851 | ||
0fd1b770 | 852 | if (type == CGROUP_SUPER_MAGIC) |
5fbea8a6 | 853 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); |
0fd1b770 | 854 | else |
5fbea8a6 | 855 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); |
5fbea8a6 CB |
856 | if (!base_cgroup) |
857 | log_error_errno(goto next, EINVAL, "Failed to find current cgroup %s", mountpoint); | |
858 | ||
859 | trim(base_cgroup); | |
860 | prune_init_scope(base_cgroup); | |
861 | ||
862 | if (type == CGROUP2_SUPER_MAGIC) { | |
863 | char *cgv2_ctrl_path; | |
864 | ||
865 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
866 | "cgroup.controllers", | |
867 | NULL); | |
868 | ||
869 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
870 | free(cgv2_ctrl_path); | |
871 | if (!controller_list) | |
872 | controller_list = cg_unified_make_empty_controller(); | |
873 | } | |
874 | ||
875 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); | |
876 | new->__controllers = move_ptr(controllers); | |
877 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
878 | ops->unified = new; | |
879 | ||
880 | continue; | |
881 | ||
882 | next: | |
883 | free_string_list(controller_list); | |
884 | free(mountpoint); | |
885 | free(base_cgroup); | |
886 | } | |
887 | ||
888 | free_string_list(klist); | |
889 | free_string_list(nlist); | |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
894 | static int cg_unified_init(struct cgroup_ops *ops) | |
895 | { | |
896 | __do_free char *subtree_path = NULL; | |
897 | int ret; | |
898 | char *mountpoint; | |
899 | char **delegatable; | |
900 | struct hierarchy *new; | |
901 | char *base_cgroup = NULL; | |
902 | ||
903 | ret = unified_cgroup_hierarchy(); | |
904 | if (ret == -ENOMEDIUM) | |
905 | return ret_errno(ENOMEDIUM); | |
906 | ||
907 | if (ret != CGROUP2_SUPER_MAGIC) | |
908 | return 0; | |
909 | ||
910 | base_cgroup = cg_unified_get_current_cgroup(1); | |
911 | if (!base_cgroup) | |
912 | return ret_errno(EINVAL); | |
913 | prune_init_scope(base_cgroup); | |
914 | ||
915 | /* | |
916 | * We assume that the cgroup we're currently in has been delegated to | |
917 | * us and we are free to further delege all of the controllers listed | |
918 | * in cgroup.controllers further down the hierarchy. | |
919 | */ | |
920 | mountpoint = must_copy_string(DEFAULT_CGROUP_MOUNTPOINT); | |
921 | subtree_path = must_make_path(mountpoint, base_cgroup, "cgroup.controllers", NULL); | |
922 | delegatable = cg_unified_get_controllers(subtree_path); | |
923 | if (!delegatable) | |
924 | delegatable = cg_unified_make_empty_controller(); | |
925 | ||
926 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
927 | * we should verify here. The reason I'm not doing it right is that I'm | |
928 | * not convinced that lxc.cgroup.use will be the future since it is a | |
929 | * global property. I much rather have an option that lets you request | |
930 | * controllers per container. | |
931 | */ | |
932 | ||
933 | new = add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
934 | ||
935 | if (bpf_devices_cgroup_supported()) | |
936 | new->bpf_device_controller = 1; | |
937 | ||
938 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
939 | ops->unified = new; | |
940 | return CGROUP2_SUPER_MAGIC; | |
941 | } | |
942 | ||
943 | static int cg_init(struct cgroup_ops *ops) | |
944 | { | |
945 | int ret; | |
946 | ||
947 | ret = cg_unified_init(ops); | |
948 | if (ret < 0) | |
949 | return -1; | |
950 | ||
951 | if (ret == CGROUP2_SUPER_MAGIC) | |
952 | return 0; | |
953 | ||
954 | return cg_hybrid_init(ops); | |
955 | } | |
956 | ||
957 | struct cgroup_ops *cgfsng_ops_init(void) | |
958 | { | |
959 | __do_free struct cgroup_ops *cgfsng_ops = NULL; | |
960 | ||
961 | cgfsng_ops = malloc(sizeof(struct cgroup_ops)); | |
962 | if (!cgfsng_ops) | |
963 | return ret_set_errno(NULL, ENOMEM); | |
964 | ||
965 | memset(cgfsng_ops, 0, sizeof(struct cgroup_ops)); | |
966 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; | |
967 | ||
968 | if (cg_init(cgfsng_ops)) | |
969 | return NULL; | |
970 | ||
971 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
1ca6a467 | 972 | cgfsng_ops->get = cgfsng_get; |
5fbea8a6 CB |
973 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; |
974 | cgfsng_ops->get_hierarchy = get_hierarchy; | |
975 | cgfsng_ops->driver = "cgfsng"; | |
976 | cgfsng_ops->version = "1.0.0"; | |
977 | cgfsng_ops->mount = cgfsng_mount; | |
5fbea8a6 | 978 | |
66c5e848 | 979 | /* memory */ |
acff9786 | 980 | cgfsng_ops->get_memory_stats_fd = cgfsng_get_memory_stats_fd; |
66c5e848 CB |
981 | cgfsng_ops->get_memory_stats = cgfsng_get_memory_stats; |
982 | cgfsng_ops->get_memory_max = cgfsng_get_memory_max; | |
983 | cgfsng_ops->get_memory_swap_max = cgfsng_get_memory_swap_max; | |
984 | cgfsng_ops->get_memory_current = cgfsng_get_memory_current; | |
985 | cgfsng_ops->get_memory_swap_current = cgfsng_get_memory_swap_current; | |
986 | ||
2c3bcd9e CB |
987 | /* cpuset */ |
988 | cgfsng_ops->get_cpuset_cpus = cgfsng_get_cpuset_cpus; | |
77f4399a | 989 | cgfsng_ops->can_use_cpuview = cgfsng_can_use_cpuview; |
2c3bcd9e | 990 | |
9a9484ab CB |
991 | /* blkio */ |
992 | cgfsng_ops->get_io_service_bytes = cgfsng_get_io_service_bytes; | |
993 | cgfsng_ops->get_io_service_time = cgfsng_get_io_service_time; | |
994 | cgfsng_ops->get_io_serviced = cgfsng_get_io_serviced; | |
995 | cgfsng_ops->get_io_merged = cgfsng_get_io_merged; | |
996 | cgfsng_ops->get_io_wait_time = cgfsng_get_io_wait_time; | |
997 | ||
998 | ||
5fbea8a6 CB |
999 | return move_ptr(cgfsng_ops); |
1000 | } |