]>
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 | 18 | |
f834b6bf SP |
19 | #include "../config.h" |
20 | ||
21 | #ifdef HAVE_FUSE3 | |
22 | #ifndef FUSE_USE_VERSION | |
23 | #define FUSE_USE_VERSION 30 | |
24 | #endif | |
25 | #else | |
1f5596dd CB |
26 | #ifndef FUSE_USE_VERSION |
27 | #define FUSE_USE_VERSION 26 | |
28 | #endif | |
f834b6bf | 29 | #endif |
1f5596dd CB |
30 | |
31 | #define _FILE_OFFSET_BITS 64 | |
32 | ||
5fbea8a6 CB |
33 | #include <ctype.h> |
34 | #include <dirent.h> | |
35 | #include <errno.h> | |
36 | #include <grp.h> | |
37 | #include <linux/kdev_t.h> | |
38 | #include <linux/types.h> | |
39 | #include <poll.h> | |
40 | #include <signal.h> | |
41 | #include <stdint.h> | |
42 | #include <stdio.h> | |
43 | #include <stdlib.h> | |
44 | #include <string.h> | |
45 | #include <sys/mount.h> | |
46 | #include <sys/types.h> | |
47 | #include <unistd.h> | |
48 | ||
1f5596dd CB |
49 | #include "../macro.h" |
50 | #include "../memory_utils.h" | |
757a63e7 | 51 | #include "../utils.h" |
5fbea8a6 CB |
52 | #include "cgroup.h" |
53 | #include "cgroup2_devices.h" | |
54 | #include "cgroup_utils.h" | |
5fbea8a6 | 55 | |
5fbea8a6 CB |
56 | /* Given a pointer to a null-terminated array of pointers, realloc to add one |
57 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
58 | * second-to-last entry - that is, the one which is now available for use | |
59 | * (keeping the list null-terminated). | |
60 | */ | |
61 | static int append_null_to_list(void ***list) | |
62 | { | |
63 | int newentry = 0; | |
64 | ||
65 | if (*list) | |
66 | for (; (*list)[newentry]; newentry++) | |
67 | ; | |
68 | ||
69 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
70 | (*list)[newentry + 1] = NULL; | |
71 | return newentry; | |
72 | } | |
73 | ||
74 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
75 | * strings. | |
76 | */ | |
77 | static bool string_in_list(char **list, const char *entry) | |
78 | { | |
79 | int i; | |
80 | ||
81 | if (!list) | |
82 | return false; | |
83 | ||
84 | for (i = 0; list[i]; i++) | |
85 | if (strcmp(list[i], entry) == 0) | |
86 | return true; | |
87 | ||
88 | return false; | |
89 | } | |
90 | ||
91 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into | |
92 | * "name=systemd". Do not fail. | |
93 | */ | |
94 | static char *cg_legacy_must_prefix_named(char *entry) | |
95 | { | |
96 | size_t len; | |
97 | char *prefixed; | |
98 | ||
99 | len = strlen(entry); | |
100 | prefixed = must_realloc(NULL, len + 6); | |
101 | ||
102 | memcpy(prefixed, "name=", STRLITERALLEN("name=")); | |
103 | memcpy(prefixed + STRLITERALLEN("name="), entry, len); | |
104 | prefixed[len + 5] = '\0'; | |
105 | ||
106 | return prefixed; | |
107 | } | |
108 | ||
109 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time | |
110 | * we are called. | |
111 | * | |
112 | * We also handle named subsystems here. Any controller which is not a kernel | |
113 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
114 | * we refuse to use because we're not sure which we have here. | |
115 | * (TODO: We could work around this in some cases by just remounting to be | |
116 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
117 | * | |
118 | * The last entry will always be NULL. | |
119 | */ | |
120 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
121 | char *entry) | |
122 | { | |
123 | int newentry; | |
124 | char *copy; | |
125 | ||
126 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
127 | return; | |
128 | ||
129 | newentry = append_null_to_list((void ***)clist); | |
130 | ||
131 | if (strncmp(entry, "name=", 5) == 0) | |
132 | copy = must_copy_string(entry); | |
133 | else if (string_in_list(klist, entry)) | |
134 | copy = must_copy_string(entry); | |
135 | else | |
136 | copy = cg_legacy_must_prefix_named(entry); | |
137 | ||
138 | (*clist)[newentry] = copy; | |
139 | } | |
140 | ||
141 | /* Given a handler's cgroup data, return the struct hierarchy for the controller | |
142 | * @c, or NULL if there is none. | |
143 | */ | |
144 | static struct hierarchy *cgfsng_get_hierarchy(struct cgroup_ops *ops, | |
145 | const char *controller) | |
146 | { | |
147 | int i; | |
148 | ||
149 | errno = ENOENT; | |
150 | ||
151 | if (!ops->hierarchies) | |
152 | return NULL; | |
153 | ||
154 | for (i = 0; ops->hierarchies[i]; i++) { | |
155 | if (!controller) { | |
156 | /* This is the empty unified hierarchy. */ | |
157 | if (ops->hierarchies[i]->controllers && | |
158 | !ops->hierarchies[i]->controllers[0]) | |
159 | return ops->hierarchies[i]; | |
160 | continue; | |
161 | } else if (pure_unified_layout(ops) && | |
162 | strcmp(controller, "devices") == 0) { | |
163 | if (ops->unified->bpf_device_controller) | |
164 | return ops->unified; | |
165 | break; | |
166 | } | |
167 | ||
168 | if (string_in_list(ops->hierarchies[i]->controllers, controller)) | |
169 | return ops->hierarchies[i]; | |
170 | } | |
171 | ||
172 | return NULL; | |
173 | } | |
174 | ||
5fbea8a6 CB |
175 | /* Given two null-terminated lists of strings, return true if any string is in |
176 | * both. | |
177 | */ | |
178 | static bool controller_lists_intersect(char **l1, char **l2) | |
179 | { | |
180 | int i; | |
181 | ||
182 | if (!l1 || !l2) | |
183 | return false; | |
184 | ||
185 | for (i = 0; l1[i]; i++) { | |
186 | if (string_in_list(l2, l1[i])) | |
187 | return true; | |
188 | } | |
189 | ||
190 | return false; | |
191 | } | |
192 | ||
193 | /* For a null-terminated list of controllers @clist, return true if any of those | |
194 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
195 | * Realistically, if one is present, all must be present. | |
196 | */ | |
197 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
198 | { | |
199 | int i; | |
200 | ||
201 | if (!hlist) | |
202 | return false; | |
203 | ||
204 | for (i = 0; hlist[i]; i++) | |
205 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
206 | return true; | |
207 | ||
208 | return false; | |
209 | } | |
210 | ||
211 | /* Get the controllers from a mountinfo line There are other ways we could get | |
212 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
213 | * could parse the mount options. But we simply assume that the mountpoint must | |
214 | * be /sys/fs/cgroup/controller-list | |
215 | */ | |
216 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, | |
217 | int type, char **controllers) | |
218 | { | |
219 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list | |
220 | * for legacy hierarchies. | |
221 | */ | |
222 | int i; | |
223 | char *p2, *tok; | |
224 | char *p = line, *sep = ","; | |
225 | char **aret = NULL; | |
226 | ||
227 | for (i = 0; i < 4; i++) { | |
228 | p = strchr(p, ' '); | |
229 | if (!p) | |
230 | return NULL; | |
231 | p++; | |
232 | } | |
233 | ||
234 | /* Note, if we change how mountinfo works, then our caller will need to | |
235 | * verify /sys/fs/cgroup/ in this field. | |
236 | */ | |
237 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
238 | return NULL; | |
239 | ||
240 | p += 15; | |
241 | p2 = strchr(p, ' '); | |
242 | if (!p2) | |
243 | return NULL; | |
244 | *p2 = '\0'; | |
245 | ||
246 | if (type == CGROUP_SUPER_MAGIC) { | |
247 | __do_free char *dup = NULL; | |
248 | ||
249 | /* strdup() here for v1 hierarchies. Otherwise | |
250 | * lxc_iterate_parts() will destroy mountpoints such as | |
251 | * "/sys/fs/cgroup/cpu,cpuacct". | |
252 | */ | |
253 | dup = must_copy_string(p); | |
254 | if (!dup) | |
255 | return NULL; | |
256 | ||
257 | lxc_iterate_parts (tok, dup, sep) | |
258 | must_append_controller(klist, nlist, &aret, tok); | |
259 | *controllers = move_ptr(dup); | |
260 | } | |
261 | *p2 = ' '; | |
262 | ||
263 | return aret; | |
264 | } | |
265 | ||
266 | static char **cg_unified_make_empty_controller(void) | |
267 | { | |
268 | int newentry; | |
269 | char **aret = NULL; | |
270 | ||
271 | newentry = append_null_to_list((void ***)&aret); | |
272 | aret[newentry] = NULL; | |
273 | return aret; | |
274 | } | |
275 | ||
276 | static char **cg_unified_get_controllers(const char *file) | |
277 | { | |
278 | __do_free char *buf = NULL; | |
279 | char *sep = " \t\n"; | |
280 | char **aret = NULL; | |
281 | char *tok; | |
282 | ||
283 | buf = read_file(file); | |
284 | if (!buf) | |
285 | return NULL; | |
286 | ||
287 | lxc_iterate_parts(tok, buf, sep) { | |
288 | int newentry; | |
289 | char *copy; | |
290 | ||
291 | newentry = append_null_to_list((void ***)&aret); | |
292 | copy = must_copy_string(tok); | |
293 | aret[newentry] = copy; | |
294 | } | |
295 | ||
296 | return aret; | |
297 | } | |
298 | ||
299 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, | |
0fd1b770 | 300 | char *base_path, int type) |
5fbea8a6 CB |
301 | { |
302 | struct hierarchy *new; | |
303 | int newentry; | |
304 | ||
305 | new = zalloc(sizeof(*new)); | |
306 | new->controllers = clist; | |
307 | new->mountpoint = mountpoint; | |
0fd1b770 | 308 | new->base_path = base_path; |
5fbea8a6 CB |
309 | new->version = type; |
310 | ||
311 | newentry = append_null_to_list((void ***)h); | |
312 | (*h)[newentry] = new; | |
313 | return new; | |
314 | } | |
315 | ||
316 | /* Get a copy of the mountpoint from @line, which is a line from | |
317 | * /proc/self/mountinfo. | |
318 | */ | |
319 | static char *cg_hybrid_get_mountpoint(char *line) | |
320 | { | |
321 | int i; | |
322 | size_t len; | |
323 | char *p2; | |
324 | char *p = line, *sret = NULL; | |
325 | ||
326 | for (i = 0; i < 4; i++) { | |
327 | p = strchr(p, ' '); | |
328 | if (!p) | |
329 | return NULL; | |
330 | p++; | |
331 | } | |
332 | ||
333 | if (strncmp(p, DEFAULT_CGROUP_MOUNTPOINT "/", 15) != 0) | |
334 | return NULL; | |
335 | ||
336 | p2 = strchr(p + 15, ' '); | |
337 | if (!p2) | |
338 | return NULL; | |
339 | *p2 = '\0'; | |
340 | ||
341 | len = strlen(p); | |
342 | sret = must_realloc(NULL, len + 1); | |
343 | memcpy(sret, p, len); | |
344 | sret[len] = '\0'; | |
345 | return sret; | |
346 | } | |
347 | ||
348 | static void must_append_string(char ***list, char *entry) | |
349 | { | |
350 | int newentry; | |
351 | char *copy; | |
352 | ||
353 | newentry = append_null_to_list((void ***)list); | |
354 | copy = must_copy_string(entry); | |
355 | (*list)[newentry] = copy; | |
356 | } | |
357 | ||
358 | static int get_existing_subsystems(char ***klist, char ***nlist) | |
359 | { | |
360 | __do_free char *line = NULL; | |
361 | __do_fclose FILE *f = NULL; | |
362 | size_t len = 0; | |
363 | ||
dbb1f822 | 364 | f = fopen("/proc/self/cgroup", "re"); |
5fbea8a6 CB |
365 | if (!f) |
366 | return -1; | |
367 | ||
368 | while (getline(&line, &len, f) != -1) { | |
369 | char *p, *p2, *tok; | |
370 | p = strchr(line, ':'); | |
371 | if (!p) | |
372 | continue; | |
373 | p++; | |
374 | p2 = strchr(p, ':'); | |
375 | if (!p2) | |
376 | continue; | |
377 | *p2 = '\0'; | |
378 | ||
379 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup | |
380 | * contains an entry of the form: | |
381 | * | |
382 | * 0::/some/path | |
383 | * | |
384 | * In this case we use "cgroup2" as controller name. | |
385 | */ | |
386 | if ((p2 - p) == 0) { | |
387 | must_append_string(klist, "cgroup2"); | |
388 | continue; | |
389 | } | |
390 | ||
391 | lxc_iterate_parts(tok, p, ",") { | |
392 | if (strncmp(tok, "name=", 5) == 0) | |
393 | must_append_string(nlist, tok); | |
394 | else | |
395 | must_append_string(klist, tok); | |
396 | } | |
397 | } | |
398 | ||
399 | return 0; | |
400 | } | |
401 | ||
402 | static void trim(char *s) | |
403 | { | |
404 | size_t len; | |
405 | ||
406 | len = strlen(s); | |
407 | while ((len > 1) && (s[len - 1] == '\n')) | |
408 | s[--len] = '\0'; | |
409 | } | |
410 | ||
411 | /* __cg_mount_direct | |
412 | * | |
413 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
414 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
415 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
416 | */ | |
417 | static int __cg_mount_direct(struct hierarchy *h, const char *controllerpath) | |
418 | { | |
419 | __do_free char *controllers = NULL; | |
420 | char *fstype = "cgroup2"; | |
421 | unsigned long flags = 0; | |
422 | int ret; | |
423 | ||
424 | flags |= MS_NOSUID; | |
425 | flags |= MS_NOEXEC; | |
426 | flags |= MS_NODEV; | |
427 | flags |= MS_RELATIME; | |
428 | ||
429 | if (h->version != CGROUP2_SUPER_MAGIC) { | |
430 | controllers = lxc_string_join(",", (const char **)h->controllers, false); | |
431 | if (!controllers) | |
432 | return -ENOMEM; | |
433 | fstype = "cgroup"; | |
434 | } | |
435 | ||
436 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); | |
437 | if (ret < 0) | |
438 | return -1; | |
439 | ||
440 | return 0; | |
441 | } | |
442 | ||
443 | static inline int cg_mount_cgroup_full(struct hierarchy *h, | |
444 | const char *controllerpath) | |
445 | { | |
446 | return __cg_mount_direct(h, controllerpath); | |
447 | } | |
448 | ||
449 | static bool cgfsng_mount(struct cgroup_ops *ops, const char *root) | |
450 | { | |
451 | __do_free char *cgroup_root = NULL; | |
452 | int ret; | |
453 | bool retval = false; | |
454 | ||
455 | if (!ops) | |
456 | return ret_set_errno(false, ENOENT); | |
457 | ||
458 | if (!ops->hierarchies) | |
459 | return true; | |
460 | ||
461 | cgroup_root = must_make_path(root, DEFAULT_CGROUP_MOUNTPOINT, NULL); | |
462 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) | |
463 | return cg_mount_cgroup_full(ops->unified, cgroup_root) == 0; | |
464 | ||
465 | /* mount tmpfs */ | |
466 | ret = safe_mount(NULL, cgroup_root, "tmpfs", | |
467 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
468 | "size=10240k,mode=755", root); | |
469 | if (ret < 0) | |
470 | goto on_error; | |
471 | ||
472 | for (int i = 0; ops->hierarchies[i]; i++) { | |
473 | __do_free char *controllerpath = NULL; | |
474 | struct hierarchy *h = ops->hierarchies[i]; | |
475 | char *controller = strrchr(h->mountpoint, '/'); | |
476 | ||
477 | if (!controller) | |
478 | continue; | |
479 | controller++; | |
480 | ||
481 | controllerpath = must_make_path(cgroup_root, controller, NULL); | |
482 | if (dir_exists(controllerpath)) | |
483 | continue; | |
484 | ||
485 | ret = mkdir(controllerpath, 0755); | |
486 | if (ret < 0) | |
487 | log_error_errno(goto on_error, errno, | |
488 | "Error creating cgroup path: %s", | |
489 | controllerpath); | |
490 | ||
8a03c08b | 491 | ret = cg_mount_cgroup_full(h, controllerpath); |
5fbea8a6 CB |
492 | if (ret < 0) |
493 | goto on_error; | |
494 | } | |
495 | retval = true; | |
496 | ||
497 | on_error: | |
498 | return retval; | |
499 | } | |
500 | ||
5fbea8a6 CB |
501 | static int cgfsng_num_hierarchies(struct cgroup_ops *ops) |
502 | { | |
503 | int i = 0; | |
504 | ||
505 | if (!ops) | |
506 | return ret_set_errno(-1, ENOENT); | |
507 | ||
508 | if (!ops->hierarchies) | |
509 | return 0; | |
510 | ||
511 | for (; ops->hierarchies[i]; i++) | |
512 | ; | |
513 | ||
514 | return i; | |
515 | } | |
516 | ||
517 | static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) | |
518 | { | |
519 | int i; | |
520 | ||
521 | if (!ops) | |
522 | return ret_set_errno(false, ENOENT); | |
523 | ||
524 | if (!ops->hierarchies) | |
525 | return false; | |
526 | ||
527 | /* sanity check n */ | |
528 | for (i = 0; i < n; i++) | |
529 | if (!ops->hierarchies[i]) | |
530 | return ret_set_errno(false, ENOENT); | |
531 | ||
532 | *out = ops->hierarchies[i]->controllers; | |
533 | ||
534 | return true; | |
535 | } | |
536 | ||
1ca6a467 CB |
537 | static bool cgfsng_get(struct cgroup_ops *ops, const char *controller, |
538 | const char *cgroup, const char *file, char **value) | |
539 | { | |
540 | __do_free char *path = NULL; | |
541 | struct hierarchy *h; | |
542 | ||
543 | h = ops->get_hierarchy(ops, controller); | |
544 | if (!h) | |
545 | return false; | |
546 | ||
925d5849 | 547 | path = must_make_path_relative(cgroup, file, NULL); |
1ca6a467 CB |
548 | *value = readat_file(h->fd, path); |
549 | return *value != NULL; | |
550 | } | |
551 | ||
66c5e848 CB |
552 | static int cgfsng_get_memory(struct cgroup_ops *ops, const char *cgroup, |
553 | const char *file, char **value) | |
554 | { | |
555 | __do_free char *path = NULL; | |
556 | struct hierarchy *h; | |
114eb8b8 | 557 | int cgroup2_root_fd, layout, ret; |
66c5e848 CB |
558 | |
559 | h = ops->get_hierarchy(ops, "memory"); | |
560 | if (!h) | |
561 | return -1; | |
562 | ||
563 | if (!is_unified_hierarchy(h)) { | |
564 | if (strcmp(file, "memory.max") == 0) | |
565 | file = "memory.limit_in_bytes"; | |
566 | else if (strcmp(file, "memory.swap.max") == 0) | |
567 | file = "memory.memsw.limit_in_bytes"; | |
568 | else if (strcmp(file, "memory.swap.current") == 0) | |
569 | file = "memory.memsw.usage_in_bytes"; | |
570 | else if (strcmp(file, "memory.current") == 0) | |
571 | file = "memory.usage_in_bytes"; | |
114eb8b8 CB |
572 | layout = CGROUP_SUPER_MAGIC; |
573 | cgroup2_root_fd = -EBADF; | |
66c5e848 | 574 | } else { |
114eb8b8 CB |
575 | layout = CGROUP2_SUPER_MAGIC; |
576 | cgroup2_root_fd = ops->cgroup2_root_fd; | |
66c5e848 CB |
577 | } |
578 | ||
114eb8b8 CB |
579 | path = must_make_path_relative(cgroup, NULL); |
580 | ret = cgroup_walkup_to_root(cgroup2_root_fd, h->fd, path, file, value); | |
581 | if (ret < 0) | |
582 | return ret; | |
583 | if (ret == 1) { | |
584 | *value = strdup(""); | |
585 | if (!*value) | |
586 | return -ENOMEM; | |
587 | } | |
66c5e848 | 588 | |
114eb8b8 | 589 | return layout; |
66c5e848 CB |
590 | } |
591 | ||
acff9786 CB |
592 | static int cgfsng_get_memory_stats_fd(struct cgroup_ops *ops, const char *cgroup) |
593 | { | |
594 | __do_free char *path = NULL; | |
595 | struct hierarchy *h; | |
596 | ||
597 | h = ops->get_hierarchy(ops, "memory"); | |
598 | if (!h) | |
599 | return -1; | |
600 | ||
925d5849 | 601 | path = must_make_path_relative(cgroup, "memory.stat", NULL); |
acff9786 CB |
602 | return openat(h->fd, path, O_RDONLY | O_CLOEXEC | O_NOFOLLOW); |
603 | } | |
604 | ||
66c5e848 CB |
605 | static int cgfsng_get_memory_current(struct cgroup_ops *ops, const char *cgroup, |
606 | char **value) | |
607 | { | |
608 | return cgfsng_get_memory(ops, cgroup, "memory.current", value); | |
609 | } | |
610 | ||
611 | static int cgfsng_get_memory_swap_current(struct cgroup_ops *ops, | |
612 | const char *cgroup, char **value) | |
613 | { | |
614 | return cgfsng_get_memory(ops, cgroup, "memory.swap.current", value); | |
615 | } | |
616 | ||
617 | static int cgfsng_get_memory_max(struct cgroup_ops *ops, const char *cgroup, | |
618 | char **value) | |
619 | { | |
620 | return cgfsng_get_memory(ops, cgroup, "memory.max", value); | |
621 | } | |
622 | ||
4032b735 SG |
623 | static int cgfsng_get_memory_swappiness(struct cgroup_ops *ops, const char *cgroup, |
624 | char **value) | |
625 | { | |
626 | return cgfsng_get_memory(ops, cgroup, "memory.swappiness", value); | |
627 | } | |
628 | ||
66c5e848 CB |
629 | static int cgfsng_get_memory_swap_max(struct cgroup_ops *ops, |
630 | const char *cgroup, char **value) | |
631 | { | |
632 | return cgfsng_get_memory(ops, cgroup, "memory.swap.max", value); | |
633 | } | |
634 | ||
6cc153e6 FS |
635 | static int cgfsng_get_memory_slabinfo_fd(struct cgroup_ops *ops, const char *cgroup) |
636 | { | |
637 | __do_free char *path = NULL; | |
638 | struct hierarchy *h; | |
639 | ||
640 | h = ops->get_hierarchy(ops, "memory"); | |
641 | if (!h) | |
642 | return -1; | |
643 | ||
644 | if (faccessat(h->fd, "memory.kmem.slabinfo", F_OK, 0)) | |
645 | return -1; | |
646 | ||
647 | path = must_make_path_relative(cgroup, "memory.kmem.slabinfo", NULL); | |
648 | return openat(h->fd, path, O_RDONLY | O_CLOEXEC | O_NOFOLLOW); | |
649 | } | |
650 | ||
c6805016 CB |
651 | static bool cgfsng_can_use_swap(struct cgroup_ops *ops) |
652 | { | |
653 | bool has_swap = false; | |
654 | struct hierarchy *h; | |
655 | ||
656 | h = ops->get_hierarchy(ops, "memory"); | |
657 | if (!h) | |
658 | return false; | |
659 | ||
660 | if (is_unified_hierarchy(h)) { | |
661 | if (faccessat(h->fd, "memory.swap.max", F_OK, 0)) | |
662 | return false; | |
663 | ||
664 | if (faccessat(h->fd, "memory.swap.current", F_OK, 0)) | |
665 | return false; | |
666 | ||
667 | has_swap = true; | |
668 | } else { | |
669 | if (faccessat(h->fd, "memory.memsw.limit_in_bytes", F_OK, 0)) | |
670 | return false; | |
671 | ||
672 | if (faccessat(h->fd, "memory.memsw.usage_in_bytes", F_OK, 0)) | |
673 | return false; | |
674 | ||
675 | has_swap = true; | |
676 | } | |
677 | ||
678 | return has_swap; | |
679 | } | |
680 | ||
66c5e848 CB |
681 | static int cgfsng_get_memory_stats(struct cgroup_ops *ops, const char *cgroup, |
682 | char **value) | |
683 | { | |
684 | return cgfsng_get_memory(ops, cgroup, "memory.stat", value); | |
685 | } | |
686 | ||
2c3bcd9e CB |
687 | static char *readat_cpuset(int cgroup_fd) |
688 | { | |
689 | __do_free char *val = NULL; | |
690 | ||
691 | val = readat_file(cgroup_fd, "cpuset.cpus"); | |
692 | if (val && strcmp(val, "") != 0) | |
693 | return move_ptr(val); | |
694 | ||
695 | free_disarm(val); | |
696 | val = readat_file(cgroup_fd, "cpuset.cpus.effective"); | |
697 | if (val && strcmp(val, "") != 0) | |
698 | return move_ptr(val); | |
699 | ||
700 | return NULL; | |
701 | } | |
702 | ||
703 | static int cgfsng_get_cpuset_cpus(struct cgroup_ops *ops, const char *cgroup, | |
704 | char **value) | |
705 | { | |
05b7a16d | 706 | __do_close int cgroup_fd = -EBADF; |
2c3bcd9e CB |
707 | __do_free char *path = NULL; |
708 | char *v; | |
709 | struct hierarchy *h; | |
710 | int ret; | |
711 | ||
712 | h = ops->get_hierarchy(ops, "cpuset"); | |
713 | if (!h) | |
714 | return -1; | |
715 | ||
716 | if (!is_unified_hierarchy(h)) | |
717 | ret = CGROUP_SUPER_MAGIC; | |
718 | else | |
719 | ret = CGROUP2_SUPER_MAGIC; | |
720 | ||
721 | *value = NULL; | |
925d5849 | 722 | path = must_make_path_relative(cgroup, NULL); |
2c3bcd9e CB |
723 | cgroup_fd = openat_safe(h->fd, path); |
724 | if (cgroup_fd < 0) { | |
725 | return -1; | |
726 | } | |
727 | v = readat_cpuset(cgroup_fd); | |
728 | if (v) { | |
729 | *value = v; | |
730 | return ret; | |
731 | } | |
732 | ||
733 | /* | |
734 | * cpuset.cpus and cpuset.cpus.effective are empty so we need to look | |
735 | * the nearest ancestor with a non-empty cpuset.cpus{.effective} file. | |
736 | */ | |
737 | for (;;) { | |
738 | int fd; | |
739 | ||
740 | fd = openat_safe(cgroup_fd, "../"); | |
227479b5 | 741 | if (fd < 0 || !is_cgroup_fd(fd)) |
2c3bcd9e | 742 | return -1; |
2c3bcd9e CB |
743 | |
744 | close_prot_errno_replace(cgroup_fd, fd); | |
745 | ||
746 | v = readat_cpuset(fd); | |
747 | if (v) { | |
748 | *value = v; | |
749 | return ret; | |
750 | } | |
751 | } | |
752 | ||
753 | return -1; | |
754 | } | |
755 | ||
9a9484ab CB |
756 | static int cgfsng_get_io(struct cgroup_ops *ops, const char *cgroup, |
757 | const char *file, char **value) | |
758 | { | |
759 | __do_free char *path = NULL; | |
760 | struct hierarchy *h; | |
761 | int ret; | |
762 | ||
763 | h = ops->get_hierarchy(ops, "blkio"); | |
764 | if (!h) | |
765 | return -1; | |
766 | ||
767 | if (!is_unified_hierarchy(h)) | |
768 | ret = CGROUP_SUPER_MAGIC; | |
769 | else | |
770 | ret = CGROUP2_SUPER_MAGIC; | |
771 | ||
925d5849 | 772 | path = must_make_path_relative(cgroup, file, NULL); |
9a9484ab CB |
773 | *value = readat_file(h->fd, path); |
774 | if (!*value) { | |
775 | if (errno == ENOENT) | |
776 | errno = EOPNOTSUPP; | |
777 | return ret_errno(errno); | |
778 | } | |
779 | ||
780 | return ret; | |
781 | } | |
782 | ||
783 | static int cgfsng_get_io_service_bytes(struct cgroup_ops *ops, | |
784 | const char *cgroup, char **value) | |
785 | { | |
786 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_bytes_recursive", value); | |
787 | } | |
788 | ||
789 | static int cgfsng_get_io_service_time(struct cgroup_ops *ops, | |
790 | const char *cgroup, char **value) | |
791 | { | |
792 | return cgfsng_get_io(ops, cgroup, "blkio.io_service_time_recursive", value); | |
793 | } | |
794 | ||
795 | static int cgfsng_get_io_serviced(struct cgroup_ops *ops, const char *cgroup, | |
796 | char **value) | |
797 | { | |
798 | return cgfsng_get_io(ops, cgroup, "blkio.io_serviced_recursive", value); | |
799 | } | |
800 | ||
801 | static int cgfsng_get_io_merged(struct cgroup_ops *ops, const char *cgroup, | |
802 | char **value) | |
803 | { | |
804 | return cgfsng_get_io(ops, cgroup, "blkio.io_merged_recursive", value); | |
805 | } | |
806 | ||
807 | static int cgfsng_get_io_wait_time(struct cgroup_ops *ops, const char *cgroup, | |
808 | char **value) | |
809 | { | |
810 | return cgfsng_get_io(ops, cgroup, "blkio.io_wait_time_recursive", value); | |
811 | } | |
812 | ||
77f4399a CB |
813 | static bool cgfsng_can_use_cpuview(struct cgroup_ops *ops) |
814 | { | |
815 | struct hierarchy *cpu, *cpuacct; | |
816 | ||
817 | if (pure_unified_layout(ops)) | |
9844eea7 | 818 | return true; |
77f4399a CB |
819 | |
820 | cpu = ops->get_hierarchy(ops, "cpu"); | |
821 | if (!cpu || is_unified_hierarchy(cpu)) | |
822 | return false; | |
823 | ||
824 | cpuacct = ops->get_hierarchy(ops, "cpuacct"); | |
825 | if (!cpuacct || is_unified_hierarchy(cpuacct)) | |
826 | return false; | |
827 | ||
828 | return true; | |
829 | } | |
830 | ||
5fbea8a6 CB |
831 | /* At startup, parse_hierarchies finds all the info we need about cgroup |
832 | * mountpoints and current cgroups, and stores it in @d. | |
833 | */ | |
834 | static int cg_hybrid_init(struct cgroup_ops *ops) | |
835 | { | |
836 | __do_free char *basecginfo = NULL; | |
837 | __do_free char *line = NULL; | |
757a63e7 | 838 | __do_free void *fopen_cache = NULL; |
5fbea8a6 CB |
839 | __do_fclose FILE *f = NULL; |
840 | int ret; | |
841 | size_t len = 0; | |
842 | char **klist = NULL, **nlist = NULL; | |
843 | ||
844 | /* Root spawned containers escape the current cgroup, so use init's | |
845 | * cgroups as our base in that case. | |
846 | */ | |
847 | basecginfo = read_file("/proc/1/cgroup"); | |
848 | if (!basecginfo) | |
849 | return ret_set_errno(-1, ENOMEM); | |
850 | ||
851 | ret = get_existing_subsystems(&klist, &nlist); | |
852 | if (ret < 0) | |
853 | return log_error_errno(-1, errno, "Failed to retrieve available legacy cgroup controllers"); | |
854 | ||
757a63e7 | 855 | f = fopen_cached("/proc/self/mountinfo", "re", &fopen_cache); |
5fbea8a6 CB |
856 | if (!f) |
857 | return log_error_errno(-1, errno, "Failed to open \"/proc/self/mountinfo\""); | |
858 | ||
859 | while (getline(&line, &len, f) != -1) { | |
860 | int type; | |
861 | struct hierarchy *new; | |
862 | char *base_cgroup = NULL, *mountpoint = NULL; | |
863 | char **controller_list = NULL; | |
864 | __do_free char *controllers = NULL; | |
865 | ||
866 | type = get_cgroup_version(line); | |
867 | if (type == 0) | |
868 | continue; | |
869 | ||
870 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
871 | continue; | |
872 | ||
873 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
874 | if (type == CGROUP2_SUPER_MAGIC) | |
875 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
876 | else if (type == CGROUP_SUPER_MAGIC) | |
877 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
878 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
879 | if (type == CGROUP_SUPER_MAGIC) | |
880 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
881 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
882 | if (type == CGROUP2_SUPER_MAGIC) | |
883 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
884 | } | |
885 | ||
886 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, | |
887 | type, &controllers); | |
888 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
889 | continue; | |
890 | ||
891 | if (type == CGROUP_SUPER_MAGIC) | |
892 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
893 | ret_set_errno(goto next, EEXIST); | |
894 | ||
895 | mountpoint = cg_hybrid_get_mountpoint(line); | |
896 | if (!mountpoint) | |
897 | log_error_errno(goto next, EINVAL, "Failed parsing mountpoint from \"%s\"", line); | |
898 | ||
0fd1b770 | 899 | if (type == CGROUP_SUPER_MAGIC) |
5fbea8a6 | 900 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); |
0fd1b770 | 901 | else |
5fbea8a6 | 902 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); |
5fbea8a6 CB |
903 | if (!base_cgroup) |
904 | log_error_errno(goto next, EINVAL, "Failed to find current cgroup %s", mountpoint); | |
905 | ||
906 | trim(base_cgroup); | |
907 | prune_init_scope(base_cgroup); | |
908 | ||
909 | if (type == CGROUP2_SUPER_MAGIC) { | |
910 | char *cgv2_ctrl_path; | |
911 | ||
912 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
913 | "cgroup.controllers", | |
914 | NULL); | |
915 | ||
916 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
917 | free(cgv2_ctrl_path); | |
918 | if (!controller_list) | |
919 | controller_list = cg_unified_make_empty_controller(); | |
920 | } | |
921 | ||
922 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); | |
923 | new->__controllers = move_ptr(controllers); | |
924 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
925 | ops->unified = new; | |
926 | ||
927 | continue; | |
928 | ||
929 | next: | |
930 | free_string_list(controller_list); | |
931 | free(mountpoint); | |
932 | free(base_cgroup); | |
933 | } | |
934 | ||
935 | free_string_list(klist); | |
936 | free_string_list(nlist); | |
937 | ||
938 | return 0; | |
939 | } | |
940 | ||
941 | static int cg_unified_init(struct cgroup_ops *ops) | |
942 | { | |
943 | __do_free char *subtree_path = NULL; | |
944 | int ret; | |
945 | char *mountpoint; | |
946 | char **delegatable; | |
947 | struct hierarchy *new; | |
948 | char *base_cgroup = NULL; | |
949 | ||
950 | ret = unified_cgroup_hierarchy(); | |
951 | if (ret == -ENOMEDIUM) | |
952 | return ret_errno(ENOMEDIUM); | |
953 | ||
954 | if (ret != CGROUP2_SUPER_MAGIC) | |
955 | return 0; | |
956 | ||
957 | base_cgroup = cg_unified_get_current_cgroup(1); | |
958 | if (!base_cgroup) | |
959 | return ret_errno(EINVAL); | |
960 | prune_init_scope(base_cgroup); | |
961 | ||
962 | /* | |
963 | * We assume that the cgroup we're currently in has been delegated to | |
964 | * us and we are free to further delege all of the controllers listed | |
965 | * in cgroup.controllers further down the hierarchy. | |
966 | */ | |
967 | mountpoint = must_copy_string(DEFAULT_CGROUP_MOUNTPOINT); | |
968 | subtree_path = must_make_path(mountpoint, base_cgroup, "cgroup.controllers", NULL); | |
969 | delegatable = cg_unified_get_controllers(subtree_path); | |
970 | if (!delegatable) | |
971 | delegatable = cg_unified_make_empty_controller(); | |
972 | ||
973 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
974 | * we should verify here. The reason I'm not doing it right is that I'm | |
975 | * not convinced that lxc.cgroup.use will be the future since it is a | |
976 | * global property. I much rather have an option that lets you request | |
977 | * controllers per container. | |
978 | */ | |
979 | ||
980 | new = add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
981 | ||
982 | if (bpf_devices_cgroup_supported()) | |
983 | new->bpf_device_controller = 1; | |
984 | ||
985 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
986 | ops->unified = new; | |
114eb8b8 CB |
987 | |
988 | ops->cgroup2_root_fd = open(DEFAULT_CGROUP_MOUNTPOINT, O_DIRECTORY | O_PATH | O_CLOEXEC); | |
989 | if (ops->cgroup2_root_fd < 0) | |
990 | return -errno; | |
991 | ||
5fbea8a6 CB |
992 | return CGROUP2_SUPER_MAGIC; |
993 | } | |
994 | ||
995 | static int cg_init(struct cgroup_ops *ops) | |
996 | { | |
997 | int ret; | |
998 | ||
999 | ret = cg_unified_init(ops); | |
1000 | if (ret < 0) | |
1001 | return -1; | |
1002 | ||
1003 | if (ret == CGROUP2_SUPER_MAGIC) | |
1004 | return 0; | |
1005 | ||
1006 | return cg_hybrid_init(ops); | |
1007 | } | |
1008 | ||
1009 | struct cgroup_ops *cgfsng_ops_init(void) | |
1010 | { | |
1011 | __do_free struct cgroup_ops *cgfsng_ops = NULL; | |
1012 | ||
114eb8b8 | 1013 | cgfsng_ops = zalloc(sizeof(struct cgroup_ops)); |
5fbea8a6 CB |
1014 | if (!cgfsng_ops) |
1015 | return ret_set_errno(NULL, ENOMEM); | |
1016 | ||
5fbea8a6 | 1017 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; |
114eb8b8 CB |
1018 | cgfsng_ops->mntns_fd = -EBADF; |
1019 | cgfsng_ops->cgroup2_root_fd = -EBADF; | |
5fbea8a6 CB |
1020 | |
1021 | if (cg_init(cgfsng_ops)) | |
1022 | return NULL; | |
1023 | ||
1024 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
1ca6a467 | 1025 | cgfsng_ops->get = cgfsng_get; |
5fbea8a6 | 1026 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; |
8a03c08b | 1027 | cgfsng_ops->get_hierarchy = cgfsng_get_hierarchy; |
5fbea8a6 CB |
1028 | cgfsng_ops->driver = "cgfsng"; |
1029 | cgfsng_ops->version = "1.0.0"; | |
1030 | cgfsng_ops->mount = cgfsng_mount; | |
5fbea8a6 | 1031 | |
66c5e848 | 1032 | /* memory */ |
acff9786 | 1033 | cgfsng_ops->get_memory_stats_fd = cgfsng_get_memory_stats_fd; |
66c5e848 CB |
1034 | cgfsng_ops->get_memory_stats = cgfsng_get_memory_stats; |
1035 | cgfsng_ops->get_memory_max = cgfsng_get_memory_max; | |
4032b735 | 1036 | cgfsng_ops->get_memory_swappiness = cgfsng_get_memory_swappiness; |
66c5e848 CB |
1037 | cgfsng_ops->get_memory_swap_max = cgfsng_get_memory_swap_max; |
1038 | cgfsng_ops->get_memory_current = cgfsng_get_memory_current; | |
1039 | cgfsng_ops->get_memory_swap_current = cgfsng_get_memory_swap_current; | |
6cc153e6 | 1040 | cgfsng_ops->get_memory_slabinfo_fd = cgfsng_get_memory_slabinfo_fd; |
c6805016 | 1041 | cgfsng_ops->can_use_swap = cgfsng_can_use_swap; |
66c5e848 | 1042 | |
2c3bcd9e CB |
1043 | /* cpuset */ |
1044 | cgfsng_ops->get_cpuset_cpus = cgfsng_get_cpuset_cpus; | |
77f4399a | 1045 | cgfsng_ops->can_use_cpuview = cgfsng_can_use_cpuview; |
2c3bcd9e | 1046 | |
9a9484ab CB |
1047 | /* blkio */ |
1048 | cgfsng_ops->get_io_service_bytes = cgfsng_get_io_service_bytes; | |
1049 | cgfsng_ops->get_io_service_time = cgfsng_get_io_service_time; | |
1050 | cgfsng_ops->get_io_serviced = cgfsng_get_io_serviced; | |
1051 | cgfsng_ops->get_io_merged = cgfsng_get_io_merged; | |
1052 | cgfsng_ops->get_io_wait_time = cgfsng_get_io_wait_time; | |
1053 | ||
1054 | ||
5fbea8a6 CB |
1055 | return move_ptr(cgfsng_ops); |
1056 | } |