]>
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 | |
16 | #define _GNU_SOURCE 1 | |
17 | #endif | |
18 | #include <ctype.h> | |
19 | #include <dirent.h> | |
20 | #include <errno.h> | |
21 | #include <grp.h> | |
22 | #include <linux/kdev_t.h> | |
23 | #include <linux/types.h> | |
24 | #include <poll.h> | |
25 | #include <signal.h> | |
26 | #include <stdint.h> | |
27 | #include <stdio.h> | |
28 | #include <stdlib.h> | |
29 | #include <string.h> | |
30 | #include <sys/mount.h> | |
31 | #include <sys/types.h> | |
32 | #include <unistd.h> | |
33 | ||
34 | #include "cgroup.h" | |
35 | #include "cgroup2_devices.h" | |
36 | #include "cgroup_utils.h" | |
37 | #include "macro.h" | |
38 | #include "memory_utils.h" | |
39 | ||
40 | static void free_string_list(char **clist) | |
41 | { | |
42 | int i; | |
43 | ||
44 | if (!clist) | |
45 | return; | |
46 | ||
47 | for (i = 0; clist[i]; i++) | |
48 | free(clist[i]); | |
49 | ||
50 | free(clist); | |
51 | } | |
52 | ||
53 | /* Given a pointer to a null-terminated array of pointers, realloc to add one | |
54 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
55 | * second-to-last entry - that is, the one which is now available for use | |
56 | * (keeping the list null-terminated). | |
57 | */ | |
58 | static int append_null_to_list(void ***list) | |
59 | { | |
60 | int newentry = 0; | |
61 | ||
62 | if (*list) | |
63 | for (; (*list)[newentry]; newentry++) | |
64 | ; | |
65 | ||
66 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
67 | (*list)[newentry + 1] = NULL; | |
68 | return newentry; | |
69 | } | |
70 | ||
71 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
72 | * strings. | |
73 | */ | |
74 | static bool string_in_list(char **list, const char *entry) | |
75 | { | |
76 | int i; | |
77 | ||
78 | if (!list) | |
79 | return false; | |
80 | ||
81 | for (i = 0; list[i]; i++) | |
82 | if (strcmp(list[i], entry) == 0) | |
83 | return true; | |
84 | ||
85 | return false; | |
86 | } | |
87 | ||
88 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into | |
89 | * "name=systemd". Do not fail. | |
90 | */ | |
91 | static char *cg_legacy_must_prefix_named(char *entry) | |
92 | { | |
93 | size_t len; | |
94 | char *prefixed; | |
95 | ||
96 | len = strlen(entry); | |
97 | prefixed = must_realloc(NULL, len + 6); | |
98 | ||
99 | memcpy(prefixed, "name=", STRLITERALLEN("name=")); | |
100 | memcpy(prefixed + STRLITERALLEN("name="), entry, len); | |
101 | prefixed[len + 5] = '\0'; | |
102 | ||
103 | return prefixed; | |
104 | } | |
105 | ||
106 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time | |
107 | * we are called. | |
108 | * | |
109 | * We also handle named subsystems here. Any controller which is not a kernel | |
110 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
111 | * we refuse to use because we're not sure which we have here. | |
112 | * (TODO: We could work around this in some cases by just remounting to be | |
113 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
114 | * | |
115 | * The last entry will always be NULL. | |
116 | */ | |
117 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
118 | char *entry) | |
119 | { | |
120 | int newentry; | |
121 | char *copy; | |
122 | ||
123 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
124 | return; | |
125 | ||
126 | newentry = append_null_to_list((void ***)clist); | |
127 | ||
128 | if (strncmp(entry, "name=", 5) == 0) | |
129 | copy = must_copy_string(entry); | |
130 | else if (string_in_list(klist, entry)) | |
131 | copy = must_copy_string(entry); | |
132 | else | |
133 | copy = cg_legacy_must_prefix_named(entry); | |
134 | ||
135 | (*clist)[newentry] = copy; | |
136 | } | |
137 | ||
138 | /* Given a handler's cgroup data, return the struct hierarchy for the controller | |
139 | * @c, or NULL if there is none. | |
140 | */ | |
141 | static struct hierarchy *cgfsng_get_hierarchy(struct cgroup_ops *ops, | |
142 | const char *controller) | |
143 | { | |
144 | int i; | |
145 | ||
146 | errno = ENOENT; | |
147 | ||
148 | if (!ops->hierarchies) | |
149 | return NULL; | |
150 | ||
151 | for (i = 0; ops->hierarchies[i]; i++) { | |
152 | if (!controller) { | |
153 | /* This is the empty unified hierarchy. */ | |
154 | if (ops->hierarchies[i]->controllers && | |
155 | !ops->hierarchies[i]->controllers[0]) | |
156 | return ops->hierarchies[i]; | |
157 | continue; | |
158 | } else if (pure_unified_layout(ops) && | |
159 | strcmp(controller, "devices") == 0) { | |
160 | if (ops->unified->bpf_device_controller) | |
161 | return ops->unified; | |
162 | break; | |
163 | } | |
164 | ||
165 | if (string_in_list(ops->hierarchies[i]->controllers, controller)) | |
166 | return ops->hierarchies[i]; | |
167 | } | |
168 | ||
169 | return NULL; | |
170 | } | |
171 | ||
172 | static inline struct hierarchy *get_hierarchy(struct cgroup_ops *ops, | |
173 | const char *controller) | |
174 | { | |
175 | return cgfsng_get_hierarchy(ops, controller); | |
176 | } | |
177 | ||
178 | /* Given two null-terminated lists of strings, return true if any string is in | |
179 | * both. | |
180 | */ | |
181 | static bool controller_lists_intersect(char **l1, char **l2) | |
182 | { | |
183 | int i; | |
184 | ||
185 | if (!l1 || !l2) | |
186 | return false; | |
187 | ||
188 | for (i = 0; l1[i]; i++) { | |
189 | if (string_in_list(l2, l1[i])) | |
190 | return true; | |
191 | } | |
192 | ||
193 | return false; | |
194 | } | |
195 | ||
196 | /* For a null-terminated list of controllers @clist, return true if any of those | |
197 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
198 | * Realistically, if one is present, all must be present. | |
199 | */ | |
200 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
201 | { | |
202 | int i; | |
203 | ||
204 | if (!hlist) | |
205 | return false; | |
206 | ||
207 | for (i = 0; hlist[i]; i++) | |
208 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
209 | return true; | |
210 | ||
211 | return false; | |
212 | } | |
213 | ||
214 | /* Get the controllers from a mountinfo line There are other ways we could get | |
215 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
216 | * could parse the mount options. But we simply assume that the mountpoint must | |
217 | * be /sys/fs/cgroup/controller-list | |
218 | */ | |
219 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, | |
220 | int type, char **controllers) | |
221 | { | |
222 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list | |
223 | * for legacy hierarchies. | |
224 | */ | |
225 | int i; | |
226 | char *p2, *tok; | |
227 | char *p = line, *sep = ","; | |
228 | char **aret = NULL; | |
229 | ||
230 | for (i = 0; i < 4; i++) { | |
231 | p = strchr(p, ' '); | |
232 | if (!p) | |
233 | return NULL; | |
234 | p++; | |
235 | } | |
236 | ||
237 | /* Note, if we change how mountinfo works, then our caller will need to | |
238 | * verify /sys/fs/cgroup/ in this field. | |
239 | */ | |
240 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
241 | return NULL; | |
242 | ||
243 | p += 15; | |
244 | p2 = strchr(p, ' '); | |
245 | if (!p2) | |
246 | return NULL; | |
247 | *p2 = '\0'; | |
248 | ||
249 | if (type == CGROUP_SUPER_MAGIC) { | |
250 | __do_free char *dup = NULL; | |
251 | ||
252 | /* strdup() here for v1 hierarchies. Otherwise | |
253 | * lxc_iterate_parts() will destroy mountpoints such as | |
254 | * "/sys/fs/cgroup/cpu,cpuacct". | |
255 | */ | |
256 | dup = must_copy_string(p); | |
257 | if (!dup) | |
258 | return NULL; | |
259 | ||
260 | lxc_iterate_parts (tok, dup, sep) | |
261 | must_append_controller(klist, nlist, &aret, tok); | |
262 | *controllers = move_ptr(dup); | |
263 | } | |
264 | *p2 = ' '; | |
265 | ||
266 | return aret; | |
267 | } | |
268 | ||
269 | static char **cg_unified_make_empty_controller(void) | |
270 | { | |
271 | int newentry; | |
272 | char **aret = NULL; | |
273 | ||
274 | newentry = append_null_to_list((void ***)&aret); | |
275 | aret[newentry] = NULL; | |
276 | return aret; | |
277 | } | |
278 | ||
279 | static char **cg_unified_get_controllers(const char *file) | |
280 | { | |
281 | __do_free char *buf = NULL; | |
282 | char *sep = " \t\n"; | |
283 | char **aret = NULL; | |
284 | char *tok; | |
285 | ||
286 | buf = read_file(file); | |
287 | if (!buf) | |
288 | return NULL; | |
289 | ||
290 | lxc_iterate_parts(tok, buf, sep) { | |
291 | int newentry; | |
292 | char *copy; | |
293 | ||
294 | newentry = append_null_to_list((void ***)&aret); | |
295 | copy = must_copy_string(tok); | |
296 | aret[newentry] = copy; | |
297 | } | |
298 | ||
299 | return aret; | |
300 | } | |
301 | ||
302 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, | |
303 | char *container_base_path, int type) | |
304 | { | |
305 | struct hierarchy *new; | |
306 | int newentry; | |
307 | ||
308 | new = zalloc(sizeof(*new)); | |
309 | new->controllers = clist; | |
310 | new->mountpoint = mountpoint; | |
311 | new->container_base_path = container_base_path; | |
312 | new->version = type; | |
313 | ||
314 | newentry = append_null_to_list((void ***)h); | |
315 | (*h)[newentry] = new; | |
316 | return new; | |
317 | } | |
318 | ||
319 | /* Get a copy of the mountpoint from @line, which is a line from | |
320 | * /proc/self/mountinfo. | |
321 | */ | |
322 | static char *cg_hybrid_get_mountpoint(char *line) | |
323 | { | |
324 | int i; | |
325 | size_t len; | |
326 | char *p2; | |
327 | char *p = line, *sret = NULL; | |
328 | ||
329 | for (i = 0; i < 4; i++) { | |
330 | p = strchr(p, ' '); | |
331 | if (!p) | |
332 | return NULL; | |
333 | p++; | |
334 | } | |
335 | ||
336 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
337 | return NULL; | |
338 | ||
339 | p2 = strchr(p + 15, ' '); | |
340 | if (!p2) | |
341 | return NULL; | |
342 | *p2 = '\0'; | |
343 | ||
344 | len = strlen(p); | |
345 | sret = must_realloc(NULL, len + 1); | |
346 | memcpy(sret, p, len); | |
347 | sret[len] = '\0'; | |
348 | return sret; | |
349 | } | |
350 | ||
351 | static void must_append_string(char ***list, char *entry) | |
352 | { | |
353 | int newentry; | |
354 | char *copy; | |
355 | ||
356 | newentry = append_null_to_list((void ***)list); | |
357 | copy = must_copy_string(entry); | |
358 | (*list)[newentry] = copy; | |
359 | } | |
360 | ||
361 | static int get_existing_subsystems(char ***klist, char ***nlist) | |
362 | { | |
363 | __do_free char *line = NULL; | |
364 | __do_fclose FILE *f = NULL; | |
365 | size_t len = 0; | |
366 | ||
367 | f = fopen("/proc/self/cgroup", "r"); | |
368 | if (!f) | |
369 | return -1; | |
370 | ||
371 | while (getline(&line, &len, f) != -1) { | |
372 | char *p, *p2, *tok; | |
373 | p = strchr(line, ':'); | |
374 | if (!p) | |
375 | continue; | |
376 | p++; | |
377 | p2 = strchr(p, ':'); | |
378 | if (!p2) | |
379 | continue; | |
380 | *p2 = '\0'; | |
381 | ||
382 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup | |
383 | * contains an entry of the form: | |
384 | * | |
385 | * 0::/some/path | |
386 | * | |
387 | * In this case we use "cgroup2" as controller name. | |
388 | */ | |
389 | if ((p2 - p) == 0) { | |
390 | must_append_string(klist, "cgroup2"); | |
391 | continue; | |
392 | } | |
393 | ||
394 | lxc_iterate_parts(tok, p, ",") { | |
395 | if (strncmp(tok, "name=", 5) == 0) | |
396 | must_append_string(nlist, tok); | |
397 | else | |
398 | must_append_string(klist, tok); | |
399 | } | |
400 | } | |
401 | ||
402 | return 0; | |
403 | } | |
404 | ||
405 | static void trim(char *s) | |
406 | { | |
407 | size_t len; | |
408 | ||
409 | len = strlen(s); | |
410 | while ((len > 1) && (s[len - 1] == '\n')) | |
411 | s[--len] = '\0'; | |
412 | } | |
413 | ||
414 | /* __cg_mount_direct | |
415 | * | |
416 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
417 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
418 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
419 | */ | |
420 | static int __cg_mount_direct(struct hierarchy *h, const char *controllerpath) | |
421 | { | |
422 | __do_free char *controllers = NULL; | |
423 | char *fstype = "cgroup2"; | |
424 | unsigned long flags = 0; | |
425 | int ret; | |
426 | ||
427 | flags |= MS_NOSUID; | |
428 | flags |= MS_NOEXEC; | |
429 | flags |= MS_NODEV; | |
430 | flags |= MS_RELATIME; | |
431 | ||
432 | if (h->version != CGROUP2_SUPER_MAGIC) { | |
433 | controllers = lxc_string_join(",", (const char **)h->controllers, false); | |
434 | if (!controllers) | |
435 | return -ENOMEM; | |
436 | fstype = "cgroup"; | |
437 | } | |
438 | ||
439 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); | |
440 | if (ret < 0) | |
441 | return -1; | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
446 | static inline int cg_mount_cgroup_full(struct hierarchy *h, | |
447 | const char *controllerpath) | |
448 | { | |
449 | return __cg_mount_direct(h, controllerpath); | |
450 | } | |
451 | ||
452 | static bool cgfsng_mount(struct cgroup_ops *ops, const char *root) | |
453 | { | |
454 | __do_free char *cgroup_root = NULL; | |
455 | int ret; | |
456 | bool retval = false; | |
457 | ||
458 | if (!ops) | |
459 | return ret_set_errno(false, ENOENT); | |
460 | ||
461 | if (!ops->hierarchies) | |
462 | return true; | |
463 | ||
464 | cgroup_root = must_make_path(root, DEFAULT_CGROUP_MOUNTPOINT, NULL); | |
465 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) | |
466 | return cg_mount_cgroup_full(ops->unified, cgroup_root) == 0; | |
467 | ||
468 | /* mount tmpfs */ | |
469 | ret = safe_mount(NULL, cgroup_root, "tmpfs", | |
470 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
471 | "size=10240k,mode=755", root); | |
472 | if (ret < 0) | |
473 | goto on_error; | |
474 | ||
475 | for (int i = 0; ops->hierarchies[i]; i++) { | |
476 | __do_free char *controllerpath = NULL; | |
477 | struct hierarchy *h = ops->hierarchies[i]; | |
478 | char *controller = strrchr(h->mountpoint, '/'); | |
479 | ||
480 | if (!controller) | |
481 | continue; | |
482 | controller++; | |
483 | ||
484 | controllerpath = must_make_path(cgroup_root, controller, NULL); | |
485 | if (dir_exists(controllerpath)) | |
486 | continue; | |
487 | ||
488 | ret = mkdir(controllerpath, 0755); | |
489 | if (ret < 0) | |
490 | log_error_errno(goto on_error, errno, | |
491 | "Error creating cgroup path: %s", | |
492 | controllerpath); | |
493 | ||
494 | ret = cg_mount_cgroup_full( h, controllerpath); | |
495 | if (ret < 0) | |
496 | goto on_error; | |
497 | } | |
498 | retval = true; | |
499 | ||
500 | on_error: | |
501 | return retval; | |
502 | } | |
503 | ||
504 | static int recursive_count_nrtasks(char *dirname) | |
505 | { | |
506 | __do_free char *path = NULL; | |
507 | __do_closedir DIR *dir = NULL; | |
508 | struct dirent *direntp; | |
509 | int count = 0, ret; | |
510 | ||
511 | dir = opendir(dirname); | |
512 | if (!dir) | |
513 | return 0; | |
514 | ||
515 | while ((direntp = readdir(dir))) { | |
516 | struct stat mystat; | |
517 | ||
518 | if (!strcmp(direntp->d_name, ".") || | |
519 | !strcmp(direntp->d_name, "..")) | |
520 | continue; | |
521 | ||
522 | path = must_make_path(dirname, direntp->d_name, NULL); | |
523 | ||
524 | if (lstat(path, &mystat)) | |
525 | continue; | |
526 | ||
527 | if (!S_ISDIR(mystat.st_mode)) | |
528 | continue; | |
529 | ||
530 | count += recursive_count_nrtasks(path); | |
531 | } | |
532 | ||
533 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
534 | ret = lxc_count_file_lines(path); | |
535 | if (ret != -1) | |
536 | count += ret; | |
537 | ||
538 | return count; | |
539 | } | |
540 | ||
541 | static int cgfsng_nrtasks(struct cgroup_ops *ops) | |
542 | { | |
543 | __do_free char *path = NULL; | |
544 | ||
545 | if (!ops) | |
546 | return ret_set_errno(-1, ENOENT); | |
547 | ||
548 | if (!ops->container_cgroup || !ops->hierarchies) | |
549 | return ret_set_errno(-1, EINVAL); | |
550 | ||
551 | path = must_make_path(ops->hierarchies[0]->container_full_path, NULL); | |
552 | return recursive_count_nrtasks(path); | |
553 | } | |
554 | ||
555 | static int cgfsng_num_hierarchies(struct cgroup_ops *ops) | |
556 | { | |
557 | int i = 0; | |
558 | ||
559 | if (!ops) | |
560 | return ret_set_errno(-1, ENOENT); | |
561 | ||
562 | if (!ops->hierarchies) | |
563 | return 0; | |
564 | ||
565 | for (; ops->hierarchies[i]; i++) | |
566 | ; | |
567 | ||
568 | return i; | |
569 | } | |
570 | ||
571 | static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) | |
572 | { | |
573 | int i; | |
574 | ||
575 | if (!ops) | |
576 | return ret_set_errno(false, ENOENT); | |
577 | ||
578 | if (!ops->hierarchies) | |
579 | return false; | |
580 | ||
581 | /* sanity check n */ | |
582 | for (i = 0; i < n; i++) | |
583 | if (!ops->hierarchies[i]) | |
584 | return ret_set_errno(false, ENOENT); | |
585 | ||
586 | *out = ops->hierarchies[i]->controllers; | |
587 | ||
588 | return true; | |
589 | } | |
590 | ||
1ca6a467 CB |
591 | static bool cgfsng_get(struct cgroup_ops *ops, const char *controller, |
592 | const char *cgroup, const char *file, char **value) | |
593 | { | |
594 | __do_free char *path = NULL; | |
595 | struct hierarchy *h; | |
596 | ||
597 | h = ops->get_hierarchy(ops, controller); | |
598 | if (!h) | |
599 | return false; | |
600 | ||
075387cd | 601 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); |
1ca6a467 CB |
602 | *value = readat_file(h->fd, path); |
603 | return *value != NULL; | |
604 | } | |
605 | ||
66c5e848 CB |
606 | static int cgfsng_get_memory(struct cgroup_ops *ops, const char *cgroup, |
607 | const char *file, char **value) | |
608 | { | |
609 | __do_free char *path = NULL; | |
610 | struct hierarchy *h; | |
611 | int ret; | |
612 | ||
613 | h = ops->get_hierarchy(ops, "memory"); | |
614 | if (!h) | |
615 | return -1; | |
616 | ||
617 | if (!is_unified_hierarchy(h)) { | |
618 | if (strcmp(file, "memory.max") == 0) | |
619 | file = "memory.limit_in_bytes"; | |
620 | else if (strcmp(file, "memory.swap.max") == 0) | |
621 | file = "memory.memsw.limit_in_bytes"; | |
622 | else if (strcmp(file, "memory.swap.current") == 0) | |
623 | file = "memory.memsw.usage_in_bytes"; | |
624 | else if (strcmp(file, "memory.current") == 0) | |
625 | file = "memory.usage_in_bytes"; | |
626 | ret = CGROUP_SUPER_MAGIC; | |
627 | } else { | |
628 | ret = CGROUP2_SUPER_MAGIC; | |
629 | } | |
630 | ||
075387cd | 631 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); |
66c5e848 CB |
632 | *value = readat_file(h->fd, path); |
633 | if (!*value) | |
634 | ret = -1; | |
635 | ||
636 | return ret; | |
637 | } | |
638 | ||
639 | static int cgfsng_get_memory_current(struct cgroup_ops *ops, const char *cgroup, | |
640 | char **value) | |
641 | { | |
642 | return cgfsng_get_memory(ops, cgroup, "memory.current", value); | |
643 | } | |
644 | ||
645 | static int cgfsng_get_memory_swap_current(struct cgroup_ops *ops, | |
646 | const char *cgroup, char **value) | |
647 | { | |
648 | return cgfsng_get_memory(ops, cgroup, "memory.swap.current", value); | |
649 | } | |
650 | ||
651 | static int cgfsng_get_memory_max(struct cgroup_ops *ops, const char *cgroup, | |
652 | char **value) | |
653 | { | |
654 | return cgfsng_get_memory(ops, cgroup, "memory.max", value); | |
655 | } | |
656 | ||
657 | static int cgfsng_get_memory_swap_max(struct cgroup_ops *ops, | |
658 | const char *cgroup, char **value) | |
659 | { | |
660 | return cgfsng_get_memory(ops, cgroup, "memory.swap.max", value); | |
661 | } | |
662 | ||
663 | static int cgfsng_get_memory_stats(struct cgroup_ops *ops, const char *cgroup, | |
664 | char **value) | |
665 | { | |
666 | return cgfsng_get_memory(ops, cgroup, "memory.stat", value); | |
667 | } | |
668 | ||
2c3bcd9e CB |
669 | static char *readat_cpuset(int cgroup_fd) |
670 | { | |
671 | __do_free char *val = NULL; | |
672 | ||
673 | val = readat_file(cgroup_fd, "cpuset.cpus"); | |
674 | if (val && strcmp(val, "") != 0) | |
675 | return move_ptr(val); | |
676 | ||
677 | free_disarm(val); | |
678 | val = readat_file(cgroup_fd, "cpuset.cpus.effective"); | |
679 | if (val && strcmp(val, "") != 0) | |
680 | return move_ptr(val); | |
681 | ||
682 | return NULL; | |
683 | } | |
684 | ||
685 | static int cgfsng_get_cpuset_cpus(struct cgroup_ops *ops, const char *cgroup, | |
686 | char **value) | |
687 | { | |
688 | __do_close_prot_errno int cgroup_fd = -EBADF; | |
689 | __do_free char *path = NULL; | |
690 | char *v; | |
691 | struct hierarchy *h; | |
692 | int ret; | |
693 | ||
694 | h = ops->get_hierarchy(ops, "cpuset"); | |
695 | if (!h) | |
696 | return -1; | |
697 | ||
698 | if (!is_unified_hierarchy(h)) | |
699 | ret = CGROUP_SUPER_MAGIC; | |
700 | else | |
701 | ret = CGROUP2_SUPER_MAGIC; | |
702 | ||
703 | *value = NULL; | |
075387cd | 704 | path = must_make_path(dot_or_empty(cgroup), cgroup, NULL); |
2c3bcd9e CB |
705 | cgroup_fd = openat_safe(h->fd, path); |
706 | if (cgroup_fd < 0) { | |
707 | return -1; | |
708 | } | |
709 | v = readat_cpuset(cgroup_fd); | |
710 | if (v) { | |
711 | *value = v; | |
712 | return ret; | |
713 | } | |
714 | ||
715 | /* | |
716 | * cpuset.cpus and cpuset.cpus.effective are empty so we need to look | |
717 | * the nearest ancestor with a non-empty cpuset.cpus{.effective} file. | |
718 | */ | |
719 | for (;;) { | |
720 | int fd; | |
721 | ||
722 | fd = openat_safe(cgroup_fd, "../"); | |
723 | if (fd < 0 || !is_cgroup_fd(fd)) { | |
724 | fprintf(stderr, "2222: %s\n", strerror(errno)); | |
725 | return -1; | |
726 | } | |
727 | ||
728 | close_prot_errno_replace(cgroup_fd, fd); | |
729 | ||
730 | v = readat_cpuset(fd); | |
731 | if (v) { | |
732 | *value = v; | |
733 | return ret; | |
734 | } | |
735 | } | |
736 | ||
737 | return -1; | |
738 | } | |
739 | ||
9a9484ab CB |
740 | static int cgfsng_get_io(struct cgroup_ops *ops, const char *cgroup, |
741 | const char *file, char **value) | |
742 | { | |
743 | __do_free char *path = NULL; | |
744 | struct hierarchy *h; | |
745 | int ret; | |
746 | ||
747 | h = ops->get_hierarchy(ops, "blkio"); | |
748 | if (!h) | |
749 | return -1; | |
750 | ||
751 | if (!is_unified_hierarchy(h)) | |
752 | ret = CGROUP_SUPER_MAGIC; | |
753 | else | |
754 | ret = CGROUP2_SUPER_MAGIC; | |
755 | ||
756 | path = must_make_path(dot_or_empty(cgroup), cgroup, file, NULL); | |
757 | *value = readat_file(h->fd, path); | |
758 | if (!*value) { | |
759 | if (errno == ENOENT) | |
760 | errno = EOPNOTSUPP; | |
761 | return ret_errno(errno); | |
762 | } | |
763 | ||
764 | return ret; | |
765 | } | |
766 | ||
767 | static int cgfsng_get_io_service_bytes(struct cgroup_ops *ops, | |
768 | const char *cgroup, char **value) | |
769 | { | |
770 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_bytes_recursive", value); | |
771 | } | |
772 | ||
773 | static int cgfsng_get_io_service_time(struct cgroup_ops *ops, | |
774 | const char *cgroup, char **value) | |
775 | { | |
776 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_time_recursive", value); | |
777 | } | |
778 | ||
779 | static int cgfsng_get_io_serviced(struct cgroup_ops *ops, const char *cgroup, | |
780 | char **value) | |
781 | { | |
782 | return cgfsng_get_io(ops, cgroup, "blkio.io_serviced_recursive", value); | |
783 | } | |
784 | ||
785 | static int cgfsng_get_io_merged(struct cgroup_ops *ops, const char *cgroup, | |
786 | char **value) | |
787 | { | |
788 | return cgfsng_get_io(ops, cgroup, "blkio.io_merged_recursive", value); | |
789 | } | |
790 | ||
791 | static int cgfsng_get_io_wait_time(struct cgroup_ops *ops, const char *cgroup, | |
792 | char **value) | |
793 | { | |
794 | return cgfsng_get_io(ops, cgroup, "blkio.io_wait_time_recursive", value); | |
795 | } | |
796 | ||
5fbea8a6 CB |
797 | /* At startup, parse_hierarchies finds all the info we need about cgroup |
798 | * mountpoints and current cgroups, and stores it in @d. | |
799 | */ | |
800 | static int cg_hybrid_init(struct cgroup_ops *ops) | |
801 | { | |
802 | __do_free char *basecginfo = NULL; | |
803 | __do_free char *line = NULL; | |
804 | __do_fclose FILE *f = NULL; | |
805 | int ret; | |
806 | size_t len = 0; | |
807 | char **klist = NULL, **nlist = NULL; | |
808 | ||
809 | /* Root spawned containers escape the current cgroup, so use init's | |
810 | * cgroups as our base in that case. | |
811 | */ | |
812 | basecginfo = read_file("/proc/1/cgroup"); | |
813 | if (!basecginfo) | |
814 | return ret_set_errno(-1, ENOMEM); | |
815 | ||
816 | ret = get_existing_subsystems(&klist, &nlist); | |
817 | if (ret < 0) | |
818 | return log_error_errno(-1, errno, "Failed to retrieve available legacy cgroup controllers"); | |
819 | ||
820 | f = fopen("/proc/self/mountinfo", "r"); | |
821 | if (!f) | |
822 | return log_error_errno(-1, errno, "Failed to open \"/proc/self/mountinfo\""); | |
823 | ||
824 | while (getline(&line, &len, f) != -1) { | |
825 | int type; | |
826 | struct hierarchy *new; | |
827 | char *base_cgroup = NULL, *mountpoint = NULL; | |
828 | char **controller_list = NULL; | |
829 | __do_free char *controllers = NULL; | |
830 | ||
831 | type = get_cgroup_version(line); | |
832 | if (type == 0) | |
833 | continue; | |
834 | ||
835 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
836 | continue; | |
837 | ||
838 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
839 | if (type == CGROUP2_SUPER_MAGIC) | |
840 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
841 | else if (type == CGROUP_SUPER_MAGIC) | |
842 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
843 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
844 | if (type == CGROUP_SUPER_MAGIC) | |
845 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
846 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
847 | if (type == CGROUP2_SUPER_MAGIC) | |
848 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
849 | } | |
850 | ||
851 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, | |
852 | type, &controllers); | |
853 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
854 | continue; | |
855 | ||
856 | if (type == CGROUP_SUPER_MAGIC) | |
857 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
858 | ret_set_errno(goto next, EEXIST); | |
859 | ||
860 | mountpoint = cg_hybrid_get_mountpoint(line); | |
861 | if (!mountpoint) | |
862 | log_error_errno(goto next, EINVAL, "Failed parsing mountpoint from \"%s\"", line); | |
863 | ||
864 | if (type == CGROUP_SUPER_MAGIC) { | |
865 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); | |
866 | } else { | |
867 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); | |
868 | } | |
869 | if (!base_cgroup) | |
870 | log_error_errno(goto next, EINVAL, "Failed to find current cgroup %s", mountpoint); | |
871 | ||
872 | trim(base_cgroup); | |
873 | prune_init_scope(base_cgroup); | |
874 | ||
875 | if (type == CGROUP2_SUPER_MAGIC) { | |
876 | char *cgv2_ctrl_path; | |
877 | ||
878 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
879 | "cgroup.controllers", | |
880 | NULL); | |
881 | ||
882 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
883 | free(cgv2_ctrl_path); | |
884 | if (!controller_list) | |
885 | controller_list = cg_unified_make_empty_controller(); | |
886 | } | |
887 | ||
888 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); | |
889 | new->__controllers = move_ptr(controllers); | |
890 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
891 | ops->unified = new; | |
892 | ||
893 | continue; | |
894 | ||
895 | next: | |
896 | free_string_list(controller_list); | |
897 | free(mountpoint); | |
898 | free(base_cgroup); | |
899 | } | |
900 | ||
901 | free_string_list(klist); | |
902 | free_string_list(nlist); | |
903 | ||
904 | return 0; | |
905 | } | |
906 | ||
907 | static int cg_unified_init(struct cgroup_ops *ops) | |
908 | { | |
909 | __do_free char *subtree_path = NULL; | |
910 | int ret; | |
911 | char *mountpoint; | |
912 | char **delegatable; | |
913 | struct hierarchy *new; | |
914 | char *base_cgroup = NULL; | |
915 | ||
916 | ret = unified_cgroup_hierarchy(); | |
917 | if (ret == -ENOMEDIUM) | |
918 | return ret_errno(ENOMEDIUM); | |
919 | ||
920 | if (ret != CGROUP2_SUPER_MAGIC) | |
921 | return 0; | |
922 | ||
923 | base_cgroup = cg_unified_get_current_cgroup(1); | |
924 | if (!base_cgroup) | |
925 | return ret_errno(EINVAL); | |
926 | prune_init_scope(base_cgroup); | |
927 | ||
928 | /* | |
929 | * We assume that the cgroup we're currently in has been delegated to | |
930 | * us and we are free to further delege all of the controllers listed | |
931 | * in cgroup.controllers further down the hierarchy. | |
932 | */ | |
933 | mountpoint = must_copy_string(DEFAULT_CGROUP_MOUNTPOINT); | |
934 | subtree_path = must_make_path(mountpoint, base_cgroup, "cgroup.controllers", NULL); | |
935 | delegatable = cg_unified_get_controllers(subtree_path); | |
936 | if (!delegatable) | |
937 | delegatable = cg_unified_make_empty_controller(); | |
938 | ||
939 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
940 | * we should verify here. The reason I'm not doing it right is that I'm | |
941 | * not convinced that lxc.cgroup.use will be the future since it is a | |
942 | * global property. I much rather have an option that lets you request | |
943 | * controllers per container. | |
944 | */ | |
945 | ||
946 | new = add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
947 | ||
948 | if (bpf_devices_cgroup_supported()) | |
949 | new->bpf_device_controller = 1; | |
950 | ||
951 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
952 | ops->unified = new; | |
953 | return CGROUP2_SUPER_MAGIC; | |
954 | } | |
955 | ||
956 | static int cg_init(struct cgroup_ops *ops) | |
957 | { | |
958 | int ret; | |
959 | ||
960 | ret = cg_unified_init(ops); | |
961 | if (ret < 0) | |
962 | return -1; | |
963 | ||
964 | if (ret == CGROUP2_SUPER_MAGIC) | |
965 | return 0; | |
966 | ||
967 | return cg_hybrid_init(ops); | |
968 | } | |
969 | ||
970 | struct cgroup_ops *cgfsng_ops_init(void) | |
971 | { | |
972 | __do_free struct cgroup_ops *cgfsng_ops = NULL; | |
973 | ||
974 | cgfsng_ops = malloc(sizeof(struct cgroup_ops)); | |
975 | if (!cgfsng_ops) | |
976 | return ret_set_errno(NULL, ENOMEM); | |
977 | ||
978 | memset(cgfsng_ops, 0, sizeof(struct cgroup_ops)); | |
979 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; | |
980 | ||
981 | if (cg_init(cgfsng_ops)) | |
982 | return NULL; | |
983 | ||
984 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
1ca6a467 | 985 | cgfsng_ops->get = cgfsng_get; |
5fbea8a6 CB |
986 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; |
987 | cgfsng_ops->get_hierarchy = get_hierarchy; | |
988 | cgfsng_ops->driver = "cgfsng"; | |
989 | cgfsng_ops->version = "1.0.0"; | |
990 | cgfsng_ops->mount = cgfsng_mount; | |
991 | cgfsng_ops->nrtasks = cgfsng_nrtasks; | |
992 | ||
66c5e848 CB |
993 | /* memory */ |
994 | cgfsng_ops->get_memory_stats = cgfsng_get_memory_stats; | |
995 | cgfsng_ops->get_memory_max = cgfsng_get_memory_max; | |
996 | cgfsng_ops->get_memory_swap_max = cgfsng_get_memory_swap_max; | |
997 | cgfsng_ops->get_memory_current = cgfsng_get_memory_current; | |
998 | cgfsng_ops->get_memory_swap_current = cgfsng_get_memory_swap_current; | |
999 | ||
2c3bcd9e CB |
1000 | /* cpuset */ |
1001 | cgfsng_ops->get_cpuset_cpus = cgfsng_get_cpuset_cpus; | |
1002 | ||
9a9484ab CB |
1003 | /* blkio */ |
1004 | cgfsng_ops->get_io_service_bytes = cgfsng_get_io_service_bytes; | |
1005 | cgfsng_ops->get_io_service_time = cgfsng_get_io_service_time; | |
1006 | cgfsng_ops->get_io_serviced = cgfsng_get_io_serviced; | |
1007 | cgfsng_ops->get_io_merged = cgfsng_get_io_merged; | |
1008 | cgfsng_ops->get_io_wait_time = cgfsng_get_io_wait_time; | |
1009 | ||
1010 | ||
5fbea8a6 CB |
1011 | return move_ptr(cgfsng_ops); |
1012 | } |