]>
Commit | Line | Data |
---|---|---|
ccb4cabe SH |
1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * Copyright © 2016 Canonical Ltd. | |
5 | * | |
6 | * Authors: | |
7 | * Serge Hallyn <serge.hallyn@ubuntu.com> | |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | */ | |
23 | ||
24 | /* | |
25 | * cgfs-ng.c: this is a new, simplified implementation of a filesystem | |
26 | * cgroup backend. The original cgfs.c was designed to be as flexible | |
27 | * as possible. It would try to find cgroup filesystems no matter where | |
28 | * or how you had them mounted, and deduce the most usable mount for | |
29 | * each controller. It also was not designed for unprivileged use, as | |
30 | * that was reserved for cgmanager. | |
31 | * | |
32 | * This new implementation assumes that cgroup filesystems are mounted | |
33 | * under /sys/fs/cgroup/clist where clist is either the controller, or | |
34 | * a comman-separated list of controllers. | |
35 | */ | |
a54694f8 | 36 | |
ccb4cabe | 37 | #include "config.h" |
a54694f8 CB |
38 | |
39 | #include <ctype.h> | |
40 | #include <dirent.h> | |
41 | #include <errno.h> | |
42 | #include <grp.h> | |
43 | #include <stdint.h> | |
ccb4cabe SH |
44 | #include <stdio.h> |
45 | #include <stdlib.h> | |
a54694f8 | 46 | #include <string.h> |
ccb4cabe | 47 | #include <unistd.h> |
a54694f8 | 48 | #include <sys/types.h> |
ccb4cabe | 49 | |
c8bf519d | 50 | #include <linux/types.h> |
51 | #include <linux/kdev_t.h> | |
52 | ||
b635e92d | 53 | #include "caps.h" |
ccb4cabe | 54 | #include "cgroup.h" |
6328fd9c | 55 | #include "cgroup_utils.h" |
ccb4cabe | 56 | #include "commands.h" |
43654d34 | 57 | #include "conf.h" |
a54694f8 | 58 | #include "log.h" |
43654d34 | 59 | #include "storage/storage.h" |
a54694f8 | 60 | #include "utils.h" |
ccb4cabe SH |
61 | |
62 | lxc_log_define(lxc_cgfsng, lxc); | |
63 | ||
64 | static struct cgroup_ops cgfsng_ops; | |
65 | ||
ccb4cabe SH |
66 | /* |
67 | * A descriptor for a mounted hierarchy | |
68 | * @controllers: either NULL, or a null-terminated list of all | |
69 | * the co-mounted controllers | |
70 | * @mountpoint: the mountpoint we will use. It will be either | |
71 | * /sys/fs/cgroup/controller or /sys/fs/cgroup/controllerlist | |
72 | * @base_cgroup: the cgroup under which the container cgroup path | |
73 | is created. This will be either the caller's cgroup (if not | |
74 | root), or init's cgroup (if root). | |
75 | */ | |
76 | struct hierarchy { | |
77 | char **controllers; | |
78 | char *mountpoint; | |
79 | char *base_cgroup; | |
80 | char *fullcgpath; | |
6328fd9c | 81 | bool is_cgroup_v2; |
ccb4cabe SH |
82 | }; |
83 | ||
84 | /* | |
85 | * The cgroup data which is attached to the lxc_handler. | |
43654d34 CB |
86 | * @cgroup_pattern : A copy of the lxc.cgroup.pattern |
87 | * @container_cgroup : If not null, the cgroup which was created for the | |
88 | * container. For each hierarchy, it is created under the | |
89 | * @hierarchy->base_cgroup directory. Relative to the | |
90 | * base_cgroup it is the same for all hierarchies. | |
91 | * @name : The name of the container. | |
92 | * @cgroup_meta : A copy of the container's cgroup information. This | |
93 | * overrides @cgroup_pattern. | |
ccb4cabe SH |
94 | */ |
95 | struct cgfsng_handler_data { | |
ccb4cabe | 96 | char *cgroup_pattern; |
1a0e70ac CB |
97 | char *container_cgroup; /* cgroup we created for the container */ |
98 | char *name; /* container name */ | |
43654d34 CB |
99 | /* per-container cgroup information */ |
100 | struct lxc_cgroup cgroup_meta; | |
ccb4cabe SH |
101 | }; |
102 | ||
457ca9aa SH |
103 | /* |
104 | * @hierarchies - a NULL-terminated array of struct hierarchy, one per | |
105 | * hierarchy. No duplicates. First sufficient, writeable mounted | |
106 | * hierarchy wins | |
107 | */ | |
108 | struct hierarchy **hierarchies; | |
109 | ||
110 | /* | |
111 | * @cgroup_use - a copy of the lxc.cgroup.use | |
112 | */ | |
113 | char *cgroup_use; | |
114 | ||
e4aeecf5 CB |
115 | /* |
116 | * @lxc_cgfsng_debug - whether to print debug info to stdout for the cgfsng | |
117 | * driver | |
118 | */ | |
119 | static bool lxc_cgfsng_debug; | |
120 | ||
ccb4cabe SH |
121 | static void free_string_list(char **clist) |
122 | { | |
123 | if (clist) { | |
124 | int i; | |
125 | ||
126 | for (i = 0; clist[i]; i++) | |
127 | free(clist[i]); | |
128 | free(clist); | |
129 | } | |
130 | } | |
131 | ||
ccb4cabe SH |
132 | /* Allocate a pointer, do not fail */ |
133 | static void *must_alloc(size_t sz) | |
134 | { | |
135 | return must_realloc(NULL, sz); | |
136 | } | |
137 | ||
ccb4cabe SH |
138 | /* |
139 | * This is a special case - return a copy of @entry | |
140 | * prepending 'name='. I.e. turn systemd into name=systemd. | |
141 | * Do not fail. | |
142 | */ | |
143 | static char *must_prefix_named(char *entry) | |
144 | { | |
145 | char *ret; | |
146 | size_t len = strlen(entry); | |
147 | ||
148 | ret = must_alloc(len + 6); | |
149 | snprintf(ret, len + 6, "name=%s", entry); | |
150 | return ret; | |
151 | } | |
152 | ||
153 | /* | |
154 | * Given a pointer to a null-terminated array of pointers, realloc to | |
155 | * add one entry, and point the new entry to NULL. Do not fail. Return | |
156 | * the index to the second-to-last entry - that is, the one which is | |
157 | * now available for use (keeping the list null-terminated). | |
158 | */ | |
159 | static int append_null_to_list(void ***list) | |
160 | { | |
161 | int newentry = 0; | |
162 | ||
163 | if (*list) | |
164 | for (; (*list)[newentry]; newentry++); | |
165 | ||
166 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
167 | (*list)[newentry + 1] = NULL; | |
168 | return newentry; | |
169 | } | |
170 | ||
171 | /* | |
172 | * Given a null-terminated array of strings, check whether @entry | |
173 | * is one of the strings | |
174 | */ | |
175 | static bool string_in_list(char **list, const char *entry) | |
176 | { | |
177 | int i; | |
178 | ||
179 | if (!list) | |
180 | return false; | |
181 | for (i = 0; list[i]; i++) | |
182 | if (strcmp(list[i], entry) == 0) | |
183 | return true; | |
184 | ||
185 | return false; | |
186 | } | |
187 | ||
188 | /* | |
189 | * append an entry to the clist. Do not fail. | |
190 | * *clist must be NULL the first time we are called. | |
191 | * | |
192 | * We also handle named subsystems here. Any controller which is not a | |
193 | * kernel subsystem, we prefix 'name='. Any which is both a kernel and | |
194 | * named subsystem, we refuse to use because we're not sure which we | |
195 | * have here. (TODO - we could work around this in some cases by just | |
196 | * remounting to be unambiguous, or by comparing mountpoint contents | |
197 | * with current cgroup) | |
198 | * | |
199 | * The last entry will always be NULL. | |
200 | */ | |
201 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
202 | { | |
203 | int newentry; | |
204 | char *copy; | |
205 | ||
206 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
207 | ERROR("Refusing to use ambiguous controller '%s'", entry); | |
208 | ERROR("It is both a named and kernel subsystem"); | |
209 | return; | |
210 | } | |
211 | ||
212 | newentry = append_null_to_list((void ***)clist); | |
213 | ||
214 | if (strncmp(entry, "name=", 5) == 0) | |
215 | copy = must_copy_string(entry); | |
216 | else if (string_in_list(klist, entry)) | |
217 | copy = must_copy_string(entry); | |
218 | else | |
219 | copy = must_prefix_named(entry); | |
220 | ||
221 | (*clist)[newentry] = copy; | |
222 | } | |
223 | ||
ccb4cabe SH |
224 | static void free_handler_data(struct cgfsng_handler_data *d) |
225 | { | |
ccb4cabe SH |
226 | free(d->cgroup_pattern); |
227 | free(d->container_cgroup); | |
228 | free(d->name); | |
43654d34 CB |
229 | if (d->cgroup_meta.dir) |
230 | free(d->cgroup_meta.dir); | |
231 | if (d->cgroup_meta.controllers) | |
232 | free(d->cgroup_meta.controllers); | |
ccb4cabe SH |
233 | free(d); |
234 | } | |
235 | ||
236 | /* | |
237 | * Given a handler's cgroup data, return the struct hierarchy for the | |
238 | * controller @c, or NULL if there is none. | |
239 | */ | |
457ca9aa | 240 | struct hierarchy *get_hierarchy(const char *c) |
ccb4cabe SH |
241 | { |
242 | int i; | |
243 | ||
457ca9aa | 244 | if (!hierarchies) |
ccb4cabe | 245 | return NULL; |
457ca9aa SH |
246 | for (i = 0; hierarchies[i]; i++) { |
247 | if (string_in_list(hierarchies[i]->controllers, c)) | |
248 | return hierarchies[i]; | |
ccb4cabe SH |
249 | } |
250 | return NULL; | |
251 | } | |
252 | ||
a54694f8 CB |
253 | #define BATCH_SIZE 50 |
254 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
255 | { | |
256 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
257 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
258 | ||
259 | if (!*mem || newbatches > oldbatches) { | |
260 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
261 | } | |
262 | } | |
263 | ||
264 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
265 | { | |
266 | size_t full = oldlen + newlen; | |
267 | ||
268 | batch_realloc(dest, oldlen, full + 1); | |
269 | ||
270 | memcpy(*dest + oldlen, new, newlen + 1); | |
271 | } | |
272 | ||
273 | /* Slurp in a whole file */ | |
274 | static char *read_file(char *fnam) | |
275 | { | |
276 | FILE *f; | |
277 | char *line = NULL, *buf = NULL; | |
278 | size_t len = 0, fulllen = 0; | |
279 | int linelen; | |
280 | ||
281 | f = fopen(fnam, "r"); | |
282 | if (!f) | |
283 | return NULL; | |
284 | while ((linelen = getline(&line, &len, f)) != -1) { | |
285 | append_line(&buf, fulllen, line, linelen); | |
286 | fulllen += linelen; | |
287 | } | |
288 | fclose(f); | |
289 | free(line); | |
290 | return buf; | |
291 | } | |
292 | ||
293 | /* Taken over modified from the kernel sources. */ | |
294 | #define NBITS 32 /* bits in uint32_t */ | |
295 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
296 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
297 | ||
298 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
299 | { | |
300 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
301 | } | |
302 | ||
303 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
304 | { | |
305 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
306 | } | |
307 | ||
308 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
309 | { | |
310 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
311 | } | |
312 | ||
313 | /* Create cpumask from cpulist aka turn: | |
314 | * | |
315 | * 0,2-3 | |
316 | * | |
317 | * into bit array | |
318 | * | |
319 | * 1 0 1 1 | |
320 | */ | |
321 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
322 | { | |
323 | char *token; | |
324 | char *saveptr = NULL; | |
325 | size_t arrlen = BITS_TO_LONGS(nbits); | |
326 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
327 | if (!bitarr) | |
328 | return NULL; | |
329 | ||
330 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
331 | errno = 0; | |
332 | unsigned start = strtoul(token, NULL, 0); | |
333 | unsigned end = start; | |
334 | ||
335 | char *range = strchr(token, '-'); | |
336 | if (range) | |
337 | end = strtoul(range + 1, NULL, 0); | |
338 | if (!(start <= end)) { | |
339 | free(bitarr); | |
340 | return NULL; | |
341 | } | |
342 | ||
343 | if (end >= nbits) { | |
344 | free(bitarr); | |
345 | return NULL; | |
346 | } | |
347 | ||
348 | while (start <= end) | |
349 | set_bit(start++, bitarr); | |
350 | } | |
351 | ||
352 | return bitarr; | |
353 | } | |
354 | ||
a54694f8 CB |
355 | /* Turn cpumask into simple, comma-separated cpulist. */ |
356 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
357 | { | |
358 | size_t i; | |
359 | int ret; | |
eab15c1e | 360 | char numstr[LXC_NUMSTRLEN64] = {0}; |
a54694f8 CB |
361 | char **cpulist = NULL; |
362 | ||
363 | for (i = 0; i <= nbits; i++) { | |
364 | if (is_set(i, bitarr)) { | |
eab15c1e CB |
365 | ret = snprintf(numstr, LXC_NUMSTRLEN64, "%zu", i); |
366 | if (ret < 0 || (size_t)ret >= LXC_NUMSTRLEN64) { | |
a54694f8 CB |
367 | lxc_free_array((void **)cpulist, free); |
368 | return NULL; | |
369 | } | |
370 | if (lxc_append_string(&cpulist, numstr) < 0) { | |
371 | lxc_free_array((void **)cpulist, free); | |
372 | return NULL; | |
373 | } | |
374 | } | |
375 | } | |
376 | return lxc_string_join(",", (const char **)cpulist, false); | |
377 | } | |
378 | ||
379 | static ssize_t get_max_cpus(char *cpulist) | |
380 | { | |
381 | char *c1, *c2; | |
382 | char *maxcpus = cpulist; | |
383 | size_t cpus = 0; | |
384 | ||
385 | c1 = strrchr(maxcpus, ','); | |
386 | if (c1) | |
387 | c1++; | |
388 | ||
389 | c2 = strrchr(maxcpus, '-'); | |
390 | if (c2) | |
391 | c2++; | |
392 | ||
393 | if (!c1 && !c2) | |
394 | c1 = maxcpus; | |
395 | else if (c1 > c2) | |
396 | c2 = c1; | |
397 | else if (c1 < c2) | |
398 | c1 = c2; | |
1a0e70ac | 399 | else if (!c1 && c2) /* The reverse case is obvs. not needed. */ |
a54694f8 CB |
400 | c1 = c2; |
401 | ||
402 | /* If the above logic is correct, c1 should always hold a valid string | |
403 | * here. | |
404 | */ | |
405 | ||
406 | errno = 0; | |
407 | cpus = strtoul(c1, NULL, 0); | |
408 | if (errno != 0) | |
409 | return -1; | |
410 | ||
411 | return cpus; | |
412 | } | |
413 | ||
6f9584d8 | 414 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a54694f8 CB |
415 | static bool filter_and_set_cpus(char *path, bool am_initialized) |
416 | { | |
417 | char *lastslash, *fpath, oldv; | |
418 | int ret; | |
419 | ssize_t i; | |
420 | ||
421 | ssize_t maxposs = 0, maxisol = 0; | |
422 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
423 | uint32_t *possmask = NULL, *isolmask = NULL; | |
6f9584d8 | 424 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
425 | |
426 | lastslash = strrchr(path, '/'); | |
1a0e70ac | 427 | if (!lastslash) { /* bug... this shouldn't be possible */ |
6f9584d8 | 428 | ERROR("Invalid path: %s.", path); |
a54694f8 CB |
429 | return bret; |
430 | } | |
431 | oldv = *lastslash; | |
432 | *lastslash = '\0'; | |
433 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
434 | posscpus = read_file(fpath); | |
6f9584d8 CB |
435 | if (!posscpus) { |
436 | SYSERROR("Could not read file: %s.\n", fpath); | |
437 | goto on_error; | |
438 | } | |
a54694f8 CB |
439 | |
440 | /* Get maximum number of cpus found in possible cpuset. */ | |
441 | maxposs = get_max_cpus(posscpus); | |
442 | if (maxposs < 0) | |
6f9584d8 | 443 | goto on_error; |
a54694f8 | 444 | |
6f9584d8 CB |
445 | if (!file_exists(__ISOL_CPUS)) { |
446 | /* This system doesn't expose isolated cpus. */ | |
447 | DEBUG("Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
65d29cbc CB |
448 | cpulist = posscpus; |
449 | /* No isolated cpus but we weren't already initialized by | |
450 | * someone. We should simply copy the parents cpuset.cpus | |
451 | * values. | |
452 | */ | |
453 | if (!am_initialized) { | |
454 | DEBUG("Copying cpuset of parent cgroup."); | |
455 | goto copy_parent; | |
456 | } | |
457 | /* No isolated cpus but we were already initialized by someone. | |
458 | * Nothing more to do for us. | |
459 | */ | |
6f9584d8 CB |
460 | goto on_success; |
461 | } | |
462 | ||
463 | isolcpus = read_file(__ISOL_CPUS); | |
464 | if (!isolcpus) { | |
465 | SYSERROR("Could not read file "__ISOL_CPUS); | |
466 | goto on_error; | |
467 | } | |
a54694f8 | 468 | if (!isdigit(isolcpus[0])) { |
6f9584d8 | 469 | DEBUG("No isolated cpus detected."); |
a54694f8 CB |
470 | cpulist = posscpus; |
471 | /* No isolated cpus but we weren't already initialized by | |
472 | * someone. We should simply copy the parents cpuset.cpus | |
473 | * values. | |
474 | */ | |
6f9584d8 CB |
475 | if (!am_initialized) { |
476 | DEBUG("Copying cpuset of parent cgroup."); | |
a54694f8 | 477 | goto copy_parent; |
6f9584d8 | 478 | } |
a54694f8 CB |
479 | /* No isolated cpus but we were already initialized by someone. |
480 | * Nothing more to do for us. | |
481 | */ | |
6f9584d8 | 482 | goto on_success; |
a54694f8 CB |
483 | } |
484 | ||
485 | /* Get maximum number of cpus found in isolated cpuset. */ | |
486 | maxisol = get_max_cpus(isolcpus); | |
487 | if (maxisol < 0) | |
6f9584d8 | 488 | goto on_error; |
a54694f8 CB |
489 | |
490 | if (maxposs < maxisol) | |
491 | maxposs = maxisol; | |
492 | maxposs++; | |
493 | ||
494 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 CB |
495 | if (!possmask) { |
496 | ERROR("Could not create cpumask for all possible cpus.\n"); | |
497 | goto on_error; | |
498 | } | |
a54694f8 CB |
499 | |
500 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 CB |
501 | if (!isolmask) { |
502 | ERROR("Could not create cpumask for all isolated cpus.\n"); | |
503 | goto on_error; | |
504 | } | |
a54694f8 CB |
505 | |
506 | for (i = 0; i <= maxposs; i++) { | |
507 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
6f9584d8 | 508 | flipped_bit = true; |
a54694f8 CB |
509 | clear_bit(i, possmask); |
510 | } | |
511 | } | |
512 | ||
6f9584d8 CB |
513 | if (!flipped_bit) { |
514 | DEBUG("No isolated cpus present in cpuset."); | |
515 | goto on_success; | |
516 | } | |
517 | DEBUG("Removed isolated cpus from cpuset."); | |
518 | ||
a54694f8 | 519 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 CB |
520 | if (!cpulist) { |
521 | ERROR("Could not create cpu list.\n"); | |
522 | goto on_error; | |
523 | } | |
a54694f8 CB |
524 | |
525 | copy_parent: | |
526 | *lastslash = oldv; | |
527 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
528 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false); | |
6f9584d8 CB |
529 | if (ret < 0) { |
530 | SYSERROR("Could not write cpu list to: %s.\n", fpath); | |
531 | goto on_error; | |
532 | } | |
533 | ||
534 | on_success: | |
535 | bret = true; | |
a54694f8 | 536 | |
6f9584d8 | 537 | on_error: |
a54694f8 CB |
538 | free(fpath); |
539 | ||
540 | free(isolcpus); | |
541 | free(isolmask); | |
542 | ||
543 | if (posscpus != cpulist) | |
544 | free(posscpus); | |
545 | free(possmask); | |
546 | ||
547 | free(cpulist); | |
548 | return bret; | |
549 | } | |
550 | ||
e3a3fecf SH |
551 | /* Copy contents of parent(@path)/@file to @path/@file */ |
552 | static bool copy_parent_file(char *path, char *file) | |
553 | { | |
554 | char *lastslash, *value = NULL, *fpath, oldv; | |
555 | int len = 0; | |
556 | int ret; | |
557 | ||
558 | lastslash = strrchr(path, '/'); | |
1a0e70ac | 559 | if (!lastslash) { /* bug... this shouldn't be possible */ |
e3a3fecf SH |
560 | ERROR("cgfsng:copy_parent_file: bad path %s", path); |
561 | return false; | |
562 | } | |
563 | oldv = *lastslash; | |
564 | *lastslash = '\0'; | |
565 | fpath = must_make_path(path, file, NULL); | |
566 | len = lxc_read_from_file(fpath, NULL, 0); | |
567 | if (len <= 0) | |
568 | goto bad; | |
569 | value = must_alloc(len + 1); | |
570 | if (lxc_read_from_file(fpath, value, len) != len) | |
571 | goto bad; | |
572 | free(fpath); | |
573 | *lastslash = oldv; | |
574 | fpath = must_make_path(path, file, NULL); | |
575 | ret = lxc_write_to_file(fpath, value, len, false); | |
576 | if (ret < 0) | |
577 | SYSERROR("Unable to write %s to %s", value, fpath); | |
578 | free(fpath); | |
579 | free(value); | |
580 | return ret >= 0; | |
581 | ||
582 | bad: | |
583 | SYSERROR("Error reading '%s'", fpath); | |
584 | free(fpath); | |
585 | free(value); | |
586 | return false; | |
587 | } | |
588 | ||
589 | /* | |
590 | * Initialize the cpuset hierarchy in first directory of @gname and | |
591 | * set cgroup.clone_children so that children inherit settings. | |
592 | * Since the h->base_path is populated by init or ourselves, we know | |
593 | * it is already initialized. | |
594 | */ | |
a54694f8 | 595 | static bool handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf SH |
596 | { |
597 | char *cgpath, *clonechildrenpath, v, *slash; | |
598 | ||
599 | if (!string_in_list(h->controllers, "cpuset")) | |
600 | return true; | |
601 | ||
602 | if (*cgname == '/') | |
603 | cgname++; | |
604 | slash = strchr(cgname, '/'); | |
605 | if (slash) | |
606 | *slash = '\0'; | |
607 | ||
608 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); | |
609 | if (slash) | |
610 | *slash = '/'; | |
611 | if (mkdir(cgpath, 0755) < 0 && errno != EEXIST) { | |
612 | SYSERROR("Failed to create '%s'", cgpath); | |
613 | free(cgpath); | |
614 | return false; | |
615 | } | |
6f9584d8 | 616 | |
e3a3fecf | 617 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
6328fd9c CB |
618 | /* unified hierarchy doesn't have clone_children */ |
619 | if (!file_exists(clonechildrenpath)) { | |
e3a3fecf SH |
620 | free(clonechildrenpath); |
621 | free(cgpath); | |
622 | return true; | |
623 | } | |
624 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { | |
625 | SYSERROR("Failed to read '%s'", clonechildrenpath); | |
626 | free(clonechildrenpath); | |
627 | free(cgpath); | |
628 | return false; | |
629 | } | |
630 | ||
a54694f8 | 631 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
6f9584d8 CB |
632 | if (!filter_and_set_cpus(cgpath, v == '1')) { |
633 | SYSERROR("Failed to remove isolated cpus."); | |
634 | free(clonechildrenpath); | |
635 | free(cgpath); | |
a54694f8 | 636 | return false; |
6f9584d8 | 637 | } |
a54694f8 | 638 | |
e3a3fecf | 639 | if (v == '1') { /* already set for us by someone else */ |
6f9584d8 | 640 | DEBUG("\"cgroup.clone_children\" was already set to \"1\"."); |
e3a3fecf SH |
641 | free(clonechildrenpath); |
642 | free(cgpath); | |
643 | return true; | |
644 | } | |
645 | ||
646 | /* copy parent's settings */ | |
a54694f8 | 647 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
6f9584d8 | 648 | SYSERROR("Failed to copy \"cpuset.mems\" settings."); |
e3a3fecf SH |
649 | free(cgpath); |
650 | free(clonechildrenpath); | |
651 | return false; | |
652 | } | |
653 | free(cgpath); | |
654 | ||
655 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
656 | /* Set clone_children so children inherit our settings */ | |
657 | SYSERROR("Failed to write 1 to %s", clonechildrenpath); | |
658 | free(clonechildrenpath); | |
659 | return false; | |
660 | } | |
661 | free(clonechildrenpath); | |
662 | return true; | |
663 | } | |
664 | ||
ccb4cabe SH |
665 | /* |
666 | * Given two null-terminated lists of strings, return true if any string | |
667 | * is in both. | |
668 | */ | |
669 | static bool controller_lists_intersect(char **l1, char **l2) | |
670 | { | |
671 | int i; | |
672 | ||
673 | if (!l1 || !l2) | |
674 | return false; | |
675 | ||
676 | for (i = 0; l1[i]; i++) { | |
677 | if (string_in_list(l2, l1[i])) | |
678 | return true; | |
679 | } | |
680 | return false; | |
681 | } | |
682 | ||
683 | /* | |
684 | * For a null-terminated list of controllers @clist, return true if any of | |
685 | * those controllers is already listed the null-terminated list of | |
686 | * hierarchies @hlist. Realistically, if one is present, all must be present. | |
687 | */ | |
688 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
689 | { | |
690 | int i; | |
691 | ||
692 | if (!hlist) | |
693 | return false; | |
694 | for (i = 0; hlist[i]; i++) | |
695 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
696 | return true; | |
697 | return false; | |
698 | ||
699 | } | |
700 | ||
701 | /* | |
702 | * Return true if the controller @entry is found in the null-terminated | |
703 | * list of hierarchies @hlist | |
704 | */ | |
705 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
706 | { | |
707 | int i; | |
708 | if (!hlist) | |
709 | return false; | |
710 | ||
711 | for (i = 0; hlist[i]; i++) | |
712 | if (string_in_list(hlist[i]->controllers, entry)) | |
713 | return true; | |
714 | return false; | |
715 | } | |
716 | ||
717 | /* | |
c30b61c3 SH |
718 | * Return true if all of the controllers which we require have been found. |
719 | * The required list is freezer and anything in * lxc.cgroup.use. | |
ccb4cabe | 720 | */ |
457ca9aa | 721 | static bool all_controllers_found(void) |
ccb4cabe SH |
722 | { |
723 | char *p, *saveptr = NULL; | |
457ca9aa | 724 | struct hierarchy ** hlist = hierarchies; |
ccb4cabe | 725 | |
ccb4cabe SH |
726 | if (!controller_found(hlist, "freezer")) { |
727 | ERROR("no freezer controller mountpoint found"); | |
728 | return false; | |
729 | } | |
730 | ||
457ca9aa | 731 | if (!cgroup_use) |
ccb4cabe | 732 | return true; |
457ca9aa | 733 | for (p = strtok_r(cgroup_use, ",", &saveptr); p; |
ccb4cabe SH |
734 | p = strtok_r(NULL, ",", &saveptr)) { |
735 | if (!controller_found(hlist, p)) { | |
736 | ERROR("no %s controller mountpoint found", p); | |
737 | return false; | |
738 | } | |
739 | } | |
740 | return true; | |
741 | } | |
742 | ||
743 | /* Return true if the fs type is fuse.lxcfs */ | |
744 | static bool is_lxcfs(const char *line) | |
745 | { | |
746 | char *p = strstr(line, " - "); | |
747 | if (!p) | |
748 | return false; | |
2f62fb00 | 749 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; |
ccb4cabe SH |
750 | } |
751 | ||
752 | /* | |
753 | * Get the controllers from a mountinfo line | |
754 | * There are other ways we could get this info. For lxcfs, field 3 | |
755 | * is /cgroup/controller-list. For cgroupfs, we could parse the mount | |
756 | * options. But we simply assume that the mountpoint must be | |
757 | * /sys/fs/cgroup/controller-list | |
758 | */ | |
759 | static char **get_controllers(char **klist, char **nlist, char *line) | |
760 | { | |
6328fd9c | 761 | /* the fourth field is /sys/fs/cgroup/comma-delimited-controller-list */ |
ccb4cabe | 762 | int i; |
411ac6d8 | 763 | char *dup, *p2, *tok; |
6328fd9c | 764 | bool is_cgroup_v2; |
411ac6d8 CB |
765 | char *p = line, *saveptr = NULL; |
766 | char **aret = NULL; | |
6328fd9c CB |
767 | |
768 | /* handle cgroup v2 */ | |
769 | is_cgroup_v2 = is_cgroupfs_v2(line); | |
ccb4cabe SH |
770 | |
771 | for (i = 0; i < 4; i++) { | |
235f1815 | 772 | p = strchr(p, ' '); |
ccb4cabe SH |
773 | if (!p) |
774 | return NULL; | |
775 | p++; | |
776 | } | |
777 | if (!p) | |
778 | return NULL; | |
779 | /* note - if we change how mountinfo works, then our caller | |
780 | * will need to verify /sys/fs/cgroup/ in this field */ | |
5059aae9 SH |
781 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { |
782 | INFO("cgfsng: found hierarchy not under /sys/fs/cgroup: \"%s\"", p); | |
ccb4cabe | 783 | return NULL; |
5059aae9 | 784 | } |
ccb4cabe | 785 | p += 15; |
235f1815 | 786 | p2 = strchr(p, ' '); |
ccb4cabe SH |
787 | if (!p2) { |
788 | ERROR("corrupt mountinfo"); | |
789 | return NULL; | |
790 | } | |
791 | *p2 = '\0'; | |
6328fd9c CB |
792 | |
793 | /* cgroup v2 does not have separate mountpoints for controllers */ | |
794 | if (is_cgroup_v2) { | |
795 | must_append_controller(klist, nlist, &aret, "cgroup2"); | |
411ac6d8 CB |
796 | return NULL; |
797 | } | |
798 | ||
799 | /* strdup() here for v1 hierarchies. Otherwise strtok_r() will destroy | |
800 | * mountpoints such as "/sys/fs/cgroup/cpu,cpuacct". | |
801 | */ | |
802 | dup = strdup(p); | |
803 | if (!dup) { | |
804 | SYSERROR("Failed to duplicate string"); | |
805 | return NULL; | |
6328fd9c CB |
806 | } |
807 | ||
411ac6d8 | 808 | for (tok = strtok_r(dup, ",", &saveptr); tok; |
ccb4cabe SH |
809 | tok = strtok_r(NULL, ",", &saveptr)) { |
810 | must_append_controller(klist, nlist, &aret, tok); | |
811 | } | |
812 | ||
411ac6d8 | 813 | free(dup); |
ccb4cabe SH |
814 | return aret; |
815 | } | |
816 | ||
ccb4cabe | 817 | /* Add a controller to our list of hierarchies */ |
457ca9aa | 818 | static void add_controller(char **clist, char *mountpoint, char *base_cgroup) |
ccb4cabe SH |
819 | { |
820 | struct hierarchy *new; | |
821 | int newentry; | |
822 | ||
823 | new = must_alloc(sizeof(*new)); | |
824 | new->controllers = clist; | |
825 | new->mountpoint = mountpoint; | |
826 | new->base_cgroup = base_cgroup; | |
827 | new->fullcgpath = NULL; | |
828 | ||
6328fd9c | 829 | /* record if this is the cgroup v2 hierarchy */ |
cdfe90a4 | 830 | if (clist && !strcmp(*clist, "cgroup2")) |
6328fd9c CB |
831 | new->is_cgroup_v2 = true; |
832 | else | |
833 | new->is_cgroup_v2 = false; | |
834 | ||
457ca9aa SH |
835 | newentry = append_null_to_list((void ***)&hierarchies); |
836 | hierarchies[newentry] = new; | |
ccb4cabe SH |
837 | } |
838 | ||
839 | /* | |
840 | * Get a copy of the mountpoint from @line, which is a line from | |
841 | * /proc/self/mountinfo | |
842 | */ | |
843 | static char *get_mountpoint(char *line) | |
844 | { | |
845 | int i; | |
846 | char *p = line, *sret; | |
847 | size_t len; | |
848 | ||
849 | for (i = 0; i < 4; i++) { | |
235f1815 | 850 | p = strchr(p, ' '); |
ccb4cabe SH |
851 | if (!p) |
852 | return NULL; | |
853 | p++; | |
854 | } | |
855 | /* we've already stuck a \0 after the mountpoint */ | |
856 | len = strlen(p); | |
857 | sret = must_alloc(len + 1); | |
858 | memcpy(sret, p, len); | |
859 | sret[len] = '\0'; | |
860 | return sret; | |
861 | } | |
862 | ||
863 | /* | |
864 | * Given a multi-line string, return a null-terminated copy of the | |
865 | * current line. | |
866 | */ | |
867 | static char *copy_to_eol(char *p) | |
868 | { | |
235f1815 | 869 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
870 | size_t len; |
871 | ||
872 | if (!p2) | |
873 | return NULL; | |
874 | ||
875 | len = p2 - p; | |
876 | sret = must_alloc(len + 1); | |
877 | memcpy(sret, p, len); | |
878 | sret[len] = '\0'; | |
879 | return sret; | |
880 | } | |
881 | ||
882 | /* | |
883 | * cgline: pointer to character after the first ':' in a line in a | |
884 | * \n-terminated /proc/self/cgroup file. Check whether * controller c is | |
885 | * present. | |
886 | */ | |
887 | static bool controller_in_clist(char *cgline, char *c) | |
888 | { | |
889 | char *tok, *saveptr = NULL, *eol, *tmp; | |
890 | size_t len; | |
891 | ||
235f1815 | 892 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
893 | if (!eol) |
894 | return false; | |
895 | ||
896 | len = eol - cgline; | |
897 | tmp = alloca(len + 1); | |
898 | memcpy(tmp, cgline, len); | |
899 | tmp[len] = '\0'; | |
900 | ||
901 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
902 | tok = strtok_r(NULL, ",", &saveptr)) { | |
903 | if (strcmp(tok, c) == 0) | |
904 | return true; | |
905 | } | |
906 | return false; | |
907 | } | |
908 | ||
909 | /* | |
910 | * @basecginfo is a copy of /proc/$$/cgroup. Return the current | |
911 | * cgroup for @controller | |
912 | */ | |
913 | static char *get_current_cgroup(char *basecginfo, char *controller) | |
914 | { | |
915 | char *p = basecginfo; | |
6328fd9c CB |
916 | bool is_cgroup_v2; |
917 | bool is_cgroup_v2_base_cgroup; | |
918 | ||
919 | is_cgroup_v2 = !strcmp(controller, "cgroup2"); | |
920 | while (true) { | |
921 | is_cgroup_v2_base_cgroup = false; | |
922 | /* cgroup v2 entry in "/proc/<pid>/cgroup": "0::/some/path" */ | |
923 | if (is_cgroup_v2 && (*p == '0')) | |
924 | is_cgroup_v2_base_cgroup = true; | |
ccb4cabe | 925 | |
235f1815 | 926 | p = strchr(p, ':'); |
ccb4cabe SH |
927 | if (!p) |
928 | return NULL; | |
929 | p++; | |
6328fd9c | 930 | if (is_cgroup_v2_base_cgroup || controller_in_clist(p, controller)) { |
235f1815 | 931 | p = strchr(p, ':'); |
ccb4cabe SH |
932 | if (!p) |
933 | return NULL; | |
934 | p++; | |
935 | return copy_to_eol(p); | |
936 | } | |
937 | ||
235f1815 | 938 | p = strchr(p, '\n'); |
ccb4cabe SH |
939 | if (!p) |
940 | return NULL; | |
941 | p++; | |
942 | } | |
943 | } | |
944 | ||
ccb4cabe SH |
945 | static void must_append_string(char ***list, char *entry) |
946 | { | |
947 | int newentry = append_null_to_list((void ***)list); | |
948 | char *copy; | |
949 | ||
950 | copy = must_copy_string(entry); | |
951 | (*list)[newentry] = copy; | |
952 | } | |
953 | ||
954 | static void get_existing_subsystems(char ***klist, char ***nlist) | |
955 | { | |
956 | FILE *f; | |
957 | char *line = NULL; | |
958 | size_t len = 0; | |
959 | ||
960 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) | |
961 | return; | |
962 | while (getline(&line, &len, f) != -1) { | |
963 | char *p, *p2, *tok, *saveptr = NULL; | |
235f1815 | 964 | p = strchr(line, ':'); |
ccb4cabe SH |
965 | if (!p) |
966 | continue; | |
967 | p++; | |
235f1815 | 968 | p2 = strchr(p, ':'); |
ccb4cabe SH |
969 | if (!p2) |
970 | continue; | |
971 | *p2 = '\0'; | |
ff8d6ee9 | 972 | |
6328fd9c CB |
973 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup |
974 | * contains an entry of the form: | |
ff8d6ee9 CB |
975 | * |
976 | * 0::/some/path | |
977 | * | |
6328fd9c | 978 | * In this case we use "cgroup2" as controller name. |
ff8d6ee9 | 979 | */ |
6328fd9c CB |
980 | if ((p2 - p) == 0) { |
981 | must_append_string(klist, "cgroup2"); | |
ff8d6ee9 | 982 | continue; |
6328fd9c | 983 | } |
ff8d6ee9 | 984 | |
ccb4cabe SH |
985 | for (tok = strtok_r(p, ",", &saveptr); tok; |
986 | tok = strtok_r(NULL, ",", &saveptr)) { | |
987 | if (strncmp(tok, "name=", 5) == 0) | |
988 | must_append_string(nlist, tok); | |
989 | else | |
990 | must_append_string(klist, tok); | |
991 | } | |
992 | } | |
993 | ||
994 | free(line); | |
995 | fclose(f); | |
996 | } | |
997 | ||
998 | static void trim(char *s) | |
999 | { | |
1000 | size_t len = strlen(s); | |
2c28d76b | 1001 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
1002 | s[--len] = '\0'; |
1003 | } | |
1004 | ||
e4aeecf5 CB |
1005 | static void lxc_cgfsng_print_handler_data(const struct cgfsng_handler_data *d) |
1006 | { | |
1007 | printf("Cgroup information:\n"); | |
1008 | printf(" container name: %s\n", d->name ? d->name : "(null)"); | |
1009 | printf(" lxc.cgroup.use: %s\n", cgroup_use ? cgroup_use : "(null)"); | |
43654d34 CB |
1010 | printf(" lxc.cgroup.pattern: %s\n", |
1011 | d->cgroup_pattern ? d->cgroup_pattern : "(null)"); | |
1012 | printf(" lxc.cgroup.dir: %s\n", | |
1013 | d->cgroup_meta.dir ? d->cgroup_meta.dir : "(null)"); | |
1014 | printf(" cgroup: %s\n", | |
1015 | d->container_cgroup ? d->container_cgroup : "(null)"); | |
e4aeecf5 CB |
1016 | } |
1017 | ||
1018 | static void lxc_cgfsng_print_hierarchies() | |
ccb4cabe | 1019 | { |
a7b0cc4c | 1020 | struct hierarchy **it; |
ccb4cabe | 1021 | int i; |
41c33dbe | 1022 | |
457ca9aa | 1023 | if (!hierarchies) { |
e4aeecf5 | 1024 | printf(" No hierarchies found."); |
ccb4cabe SH |
1025 | return; |
1026 | } | |
e4aeecf5 | 1027 | printf(" Hierarchies:\n"); |
a7b0cc4c CB |
1028 | for (i = 0, it = hierarchies; it && *it; it++, i++) { |
1029 | char **cit; | |
ccb4cabe | 1030 | int j; |
e4aeecf5 CB |
1031 | printf(" %d: base_cgroup %s\n", i, (*it)->base_cgroup ? (*it)->base_cgroup : "(null)"); |
1032 | printf(" mountpoint %s\n", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); | |
1033 | printf(" controllers:\n"); | |
a7b0cc4c | 1034 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
e4aeecf5 | 1035 | printf(" %d: %s\n", j, *cit); |
ccb4cabe SH |
1036 | } |
1037 | } | |
41c33dbe | 1038 | |
e4aeecf5 | 1039 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, char **nlist) |
41c33dbe SH |
1040 | { |
1041 | int k; | |
a7b0cc4c | 1042 | char **it; |
41c33dbe | 1043 | |
a7b0cc4c CB |
1044 | printf("basecginfo is:\n"); |
1045 | printf("%s\n", basecginfo); | |
41c33dbe | 1046 | |
a7b0cc4c CB |
1047 | for (k = 0, it = klist; it && *it; it++, k++) |
1048 | printf("kernel subsystem %d: %s\n", k, *it); | |
1049 | for (k = 0, it = nlist; it && *it; it++, k++) | |
1050 | printf("named subsystem %d: %s\n", k, *it); | |
41c33dbe | 1051 | } |
ccb4cabe | 1052 | |
e4aeecf5 CB |
1053 | static void lxc_cgfsng_print_debuginfo(const struct cgfsng_handler_data *d) |
1054 | { | |
1055 | lxc_cgfsng_print_handler_data(d); | |
1056 | lxc_cgfsng_print_hierarchies(); | |
1057 | } | |
1058 | ||
ccb4cabe SH |
1059 | /* |
1060 | * At startup, parse_hierarchies finds all the info we need about | |
1061 | * cgroup mountpoints and current cgroups, and stores it in @d. | |
1062 | */ | |
457ca9aa | 1063 | static bool parse_hierarchies(void) |
ccb4cabe SH |
1064 | { |
1065 | FILE *f; | |
1066 | char * line = NULL, *basecginfo; | |
1067 | char **klist = NULL, **nlist = NULL; | |
1068 | size_t len = 0; | |
1069 | ||
d30ec4cb SH |
1070 | /* |
1071 | * Root spawned containers escape the current cgroup, so use init's | |
1072 | * cgroups as our base in that case. | |
1073 | */ | |
ccb4cabe SH |
1074 | if (geteuid()) |
1075 | basecginfo = read_file("/proc/self/cgroup"); | |
1076 | else | |
1077 | basecginfo = read_file("/proc/1/cgroup"); | |
1078 | if (!basecginfo) | |
1079 | return false; | |
1080 | ||
1081 | if ((f = fopen("/proc/self/mountinfo", "r")) == NULL) { | |
d3b00a8f | 1082 | SYSERROR("Failed opening /proc/self/mountinfo"); |
ccb4cabe SH |
1083 | return false; |
1084 | } | |
1085 | ||
1086 | get_existing_subsystems(&klist, &nlist); | |
41c33dbe | 1087 | |
e4aeecf5 CB |
1088 | if (lxc_cgfsng_debug) |
1089 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
ccb4cabe SH |
1090 | |
1091 | /* we support simple cgroup mounts and lxcfs mounts */ | |
1092 | while (getline(&line, &len, f) != -1) { | |
1093 | char **controller_list = NULL; | |
1094 | char *mountpoint, *base_cgroup; | |
6328fd9c | 1095 | bool is_cgroup_v2, writeable; |
ccb4cabe | 1096 | |
6328fd9c CB |
1097 | is_cgroup_v2 = is_cgroupfs_v2(line); |
1098 | if (!is_lxcfs(line) && !is_cgroupfs_v1(line) && !is_cgroup_v2) | |
ccb4cabe SH |
1099 | continue; |
1100 | ||
1101 | controller_list = get_controllers(klist, nlist, line); | |
1102 | if (!controller_list) | |
1103 | continue; | |
1104 | ||
457ca9aa | 1105 | if (controller_list_is_dup(hierarchies, controller_list)) { |
ccb4cabe SH |
1106 | free(controller_list); |
1107 | continue; | |
1108 | } | |
1109 | ||
1110 | mountpoint = get_mountpoint(line); | |
1111 | if (!mountpoint) { | |
1112 | ERROR("Error reading mountinfo: bad line '%s'", line); | |
1113 | free_string_list(controller_list); | |
1114 | continue; | |
1115 | } | |
1116 | ||
1117 | base_cgroup = get_current_cgroup(basecginfo, controller_list[0]); | |
1118 | if (!base_cgroup) { | |
1119 | ERROR("Failed to find current cgroup for controller '%s'", controller_list[0]); | |
1120 | free_string_list(controller_list); | |
1121 | free(mountpoint); | |
1122 | continue; | |
1123 | } | |
6328fd9c | 1124 | |
ccb4cabe SH |
1125 | trim(base_cgroup); |
1126 | prune_init_scope(base_cgroup); | |
6328fd9c CB |
1127 | if (is_cgroup_v2) |
1128 | writeable = test_writeable_v2(mountpoint, base_cgroup); | |
1129 | else | |
1130 | writeable = test_writeable_v1(mountpoint, base_cgroup); | |
1131 | if (!writeable) { | |
ccb4cabe SH |
1132 | free_string_list(controller_list); |
1133 | free(mountpoint); | |
1134 | free(base_cgroup); | |
1135 | continue; | |
1136 | } | |
457ca9aa | 1137 | add_controller(controller_list, mountpoint, base_cgroup); |
ccb4cabe SH |
1138 | } |
1139 | ||
1140 | free_string_list(klist); | |
1141 | free_string_list(nlist); | |
1142 | ||
1143 | free(basecginfo); | |
1144 | ||
1145 | fclose(f); | |
1146 | free(line); | |
1147 | ||
e4aeecf5 CB |
1148 | if (lxc_cgfsng_debug) { |
1149 | printf("writeable subsystems:\n"); | |
1150 | lxc_cgfsng_print_hierarchies(); | |
1151 | } | |
1152 | ||
ccb4cabe SH |
1153 | /* verify that all controllers in cgroup.use and all crucial |
1154 | * controllers are accounted for | |
1155 | */ | |
5059aae9 SH |
1156 | if (!all_controllers_found()) { |
1157 | INFO("cgfsng: not all controllers were find, deferring to cgfs driver"); | |
ccb4cabe | 1158 | return false; |
5059aae9 | 1159 | } |
ccb4cabe SH |
1160 | |
1161 | return true; | |
1162 | } | |
1163 | ||
457ca9aa SH |
1164 | static bool collect_hierarchy_info(void) |
1165 | { | |
1166 | const char *tmp; | |
1167 | errno = 0; | |
1168 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
1a0e70ac | 1169 | if (!cgroup_use && errno != 0) { /* lxc.cgroup.use can be NULL */ |
457ca9aa SH |
1170 | SYSERROR("cgfsng: error reading list of cgroups to use"); |
1171 | return false; | |
1172 | } | |
1173 | cgroup_use = must_copy_string(tmp); | |
1174 | ||
1175 | return parse_hierarchies(); | |
1176 | } | |
1177 | ||
43654d34 | 1178 | static void *cgfsng_init(struct lxc_handler *handler) |
ccb4cabe | 1179 | { |
457ca9aa | 1180 | const char *cgroup_pattern; |
43654d34 | 1181 | struct cgfsng_handler_data *d; |
ccb4cabe SH |
1182 | |
1183 | d = must_alloc(sizeof(*d)); | |
1184 | memset(d, 0, sizeof(*d)); | |
1185 | ||
43654d34 CB |
1186 | /* copy container name */ |
1187 | d->name = must_copy_string(handler->name); | |
1188 | ||
1189 | /* copy per-container cgroup information */ | |
ae5e6c08 CB |
1190 | d->cgroup_meta.dir = NULL; |
1191 | d->cgroup_meta.controllers = NULL; | |
9b5396f9 CB |
1192 | if (handler->conf) { |
1193 | d->cgroup_meta.dir = must_copy_string(handler->conf->cgroup_meta.dir); | |
1194 | d->cgroup_meta.controllers = must_copy_string(handler->conf->cgroup_meta.controllers); | |
1195 | } | |
ccb4cabe | 1196 | |
43654d34 | 1197 | /* copy system-wide cgroup information */ |
ccb4cabe | 1198 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
43654d34 CB |
1199 | if (!cgroup_pattern) { |
1200 | /* lxc.cgroup.pattern is only NULL on error. */ | |
ccb4cabe SH |
1201 | ERROR("Error getting cgroup pattern"); |
1202 | goto out_free; | |
1203 | } | |
1204 | d->cgroup_pattern = must_copy_string(cgroup_pattern); | |
1205 | ||
e4aeecf5 CB |
1206 | if (lxc_cgfsng_debug) |
1207 | lxc_cgfsng_print_debuginfo(d); | |
ccb4cabe SH |
1208 | |
1209 | return d; | |
1210 | ||
1211 | out_free: | |
1212 | free_handler_data(d); | |
1213 | return NULL; | |
1214 | } | |
1215 | ||
ccb4cabe SH |
1216 | static int cgroup_rmdir(char *dirname) |
1217 | { | |
a17f8b3f | 1218 | int ret; |
74f96976 | 1219 | struct dirent *direntp; |
ccb4cabe SH |
1220 | DIR *dir; |
1221 | int r = 0; | |
1222 | ||
1223 | dir = opendir(dirname); | |
1224 | if (!dir) | |
1225 | return -1; | |
1226 | ||
74f96976 | 1227 | while ((direntp = readdir(dir))) { |
ccb4cabe | 1228 | char *pathname; |
a17f8b3f | 1229 | struct stat mystat; |
ccb4cabe SH |
1230 | |
1231 | if (!direntp) | |
1232 | break; | |
1233 | ||
1234 | if (!strcmp(direntp->d_name, ".") || | |
1235 | !strcmp(direntp->d_name, "..")) | |
1236 | continue; | |
1237 | ||
1238 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
1239 | ||
a17f8b3f CB |
1240 | ret = lstat(pathname, &mystat); |
1241 | if (ret < 0) { | |
ccb4cabe | 1242 | if (!r) |
a17f8b3f | 1243 | WARN("Failed to stat %s", pathname); |
ccb4cabe SH |
1244 | r = -1; |
1245 | goto next; | |
1246 | } | |
1247 | ||
1248 | if (!S_ISDIR(mystat.st_mode)) | |
1249 | goto next; | |
a17f8b3f CB |
1250 | |
1251 | ret = cgroup_rmdir(pathname); | |
1252 | if (ret < 0) | |
ccb4cabe SH |
1253 | r = -1; |
1254 | next: | |
1255 | free(pathname); | |
1256 | } | |
1257 | ||
a17f8b3f CB |
1258 | ret = rmdir(dirname); |
1259 | if (ret < 0) { | |
ccb4cabe | 1260 | if (!r) |
a17f8b3f CB |
1261 | WARN("Failed to delete \"%s\": %s", dirname, |
1262 | strerror(errno)); | |
ccb4cabe SH |
1263 | r = -1; |
1264 | } | |
1265 | ||
a17f8b3f CB |
1266 | ret = closedir(dir); |
1267 | if (ret < 0) { | |
ccb4cabe | 1268 | if (!r) |
a17f8b3f CB |
1269 | WARN("Failed to delete \"%s\": %s", dirname, |
1270 | strerror(errno)); | |
ccb4cabe SH |
1271 | r = -1; |
1272 | } | |
a17f8b3f | 1273 | |
ccb4cabe SH |
1274 | return r; |
1275 | } | |
1276 | ||
1277 | static int rmdir_wrapper(void *data) | |
1278 | { | |
1279 | char *path = data; | |
1280 | ||
1281 | if (setresgid(0,0,0) < 0) | |
1282 | SYSERROR("Failed to setgid to 0"); | |
1283 | if (setresuid(0,0,0) < 0) | |
1284 | SYSERROR("Failed to setuid to 0"); | |
1285 | if (setgroups(0, NULL) < 0) | |
1286 | SYSERROR("Failed to clear groups"); | |
1287 | ||
1288 | return cgroup_rmdir(path); | |
1289 | } | |
1290 | ||
308a6c94 | 1291 | void recursive_destroy(char *path, struct lxc_conf *conf) |
ccb4cabe SH |
1292 | { |
1293 | int r; | |
1294 | if (conf && !lxc_list_empty(&conf->id_map)) | |
c9b7c33e | 1295 | r = userns_exec_1(conf, rmdir_wrapper, path, "rmdir_wrapper"); |
ccb4cabe SH |
1296 | else |
1297 | r = cgroup_rmdir(path); | |
308a6c94 | 1298 | |
ccb4cabe | 1299 | if (r < 0) |
1c9da8da | 1300 | ERROR("Error destroying %s", path); |
ccb4cabe SH |
1301 | } |
1302 | ||
1303 | static void cgfsng_destroy(void *hdata, struct lxc_conf *conf) | |
1304 | { | |
1305 | struct cgfsng_handler_data *d = hdata; | |
1306 | ||
1307 | if (!d) | |
1308 | return; | |
1309 | ||
308a6c94 CB |
1310 | if (d->container_cgroup && hierarchies) { |
1311 | int i; | |
1312 | for (i = 0; hierarchies[i]; i++) { | |
1313 | struct hierarchy *h = hierarchies[i]; | |
1314 | if (h->fullcgpath) { | |
1315 | recursive_destroy(h->fullcgpath, conf); | |
1316 | free(h->fullcgpath); | |
1317 | h->fullcgpath = NULL; | |
1318 | } | |
ccb4cabe SH |
1319 | } |
1320 | } | |
1321 | ||
1322 | free_handler_data(d); | |
1323 | } | |
1324 | ||
1325 | struct cgroup_ops *cgfsng_ops_init(void) | |
1326 | { | |
e4aeecf5 CB |
1327 | if (getenv("LXC_DEBUG_CGFSNG")) |
1328 | lxc_cgfsng_debug = true; | |
1329 | ||
457ca9aa SH |
1330 | if (!collect_hierarchy_info()) |
1331 | return NULL; | |
e4aeecf5 | 1332 | |
ccb4cabe SH |
1333 | return &cgfsng_ops; |
1334 | } | |
1335 | ||
1336 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1337 | { | |
e3a3fecf | 1338 | h->fullcgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); |
1a0e70ac | 1339 | if (dir_exists(h->fullcgpath)) { /* it must not already exist */ |
6f9584d8 | 1340 | ERROR("Path \"%s\" already existed.", h->fullcgpath); |
d8da679e | 1341 | return false; |
6f9584d8 CB |
1342 | } |
1343 | if (!handle_cpuset_hierarchy(h, cgname)) { | |
1344 | ERROR("Failed to handle cgroupfs v1 cpuset controller."); | |
e3a3fecf | 1345 | return false; |
6f9584d8 | 1346 | } |
e3a3fecf | 1347 | return mkdir_p(h->fullcgpath, 0755) == 0; |
ccb4cabe SH |
1348 | } |
1349 | ||
1350 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1351 | { | |
1352 | if (rmdir(h->fullcgpath) < 0) | |
1353 | SYSERROR("Failed to clean up cgroup %s from failed creation attempt", h->fullcgpath); | |
1354 | free(h->fullcgpath); | |
1355 | h->fullcgpath = NULL; | |
1356 | } | |
1357 | ||
1358 | /* | |
d30ec4cb | 1359 | * Try to create the same cgroup in all hierarchies. |
ccb4cabe SH |
1360 | * Start with cgroup_pattern; next cgroup_pattern-1, -2, ..., -999 |
1361 | */ | |
1362 | static inline bool cgfsng_create(void *hdata) | |
1363 | { | |
bb30b52a | 1364 | int i; |
ccb4cabe | 1365 | size_t len; |
7d531e9b CB |
1366 | char *cgname, *offset, *tmp; |
1367 | int idx = 0; | |
1368 | struct cgfsng_handler_data *d = hdata; | |
ccb4cabe SH |
1369 | |
1370 | if (!d) | |
1371 | return false; | |
43654d34 | 1372 | |
ccb4cabe SH |
1373 | if (d->container_cgroup) { |
1374 | WARN("cgfsng_create called a second time"); | |
1375 | return false; | |
1376 | } | |
1377 | ||
43654d34 | 1378 | if (d->cgroup_meta.dir) |
7d531e9b | 1379 | tmp = lxc_string_join("/", (const char *[]){d->cgroup_meta.dir, d->name, NULL}, false); |
43654d34 CB |
1380 | else |
1381 | tmp = lxc_string_replace("%n", d->name, d->cgroup_pattern); | |
ccb4cabe SH |
1382 | if (!tmp) { |
1383 | ERROR("Failed expanding cgroup name pattern"); | |
1384 | return false; | |
1385 | } | |
1a0e70ac | 1386 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
ccb4cabe SH |
1387 | cgname = must_alloc(len); |
1388 | strcpy(cgname, tmp); | |
1389 | free(tmp); | |
1390 | offset = cgname + len - 5; | |
1391 | ||
1392 | again: | |
95adfe93 SH |
1393 | if (idx == 1000) { |
1394 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1395 | goto out_free; |
95adfe93 | 1396 | } |
66b66624 | 1397 | if (idx) { |
bb30b52a CB |
1398 | int ret; |
1399 | ||
66b66624 CB |
1400 | ret = snprintf(offset, 5, "-%d", idx); |
1401 | if (ret < 0 || (size_t)ret >= 5) { | |
1402 | FILE *f = fopen("/dev/null", "w"); | |
1403 | if (f >= 0) { | |
1404 | fprintf(f, "Workaround for GCC7 bug: " | |
1405 | "https://gcc.gnu.org/bugzilla/" | |
1406 | "show_bug.cgi?id=78969"); | |
1407 | fclose(f); | |
1408 | } | |
1409 | } | |
1410 | } | |
457ca9aa SH |
1411 | for (i = 0; hierarchies[i]; i++) { |
1412 | if (!create_path_for_hierarchy(hierarchies[i], cgname)) { | |
ccb4cabe | 1413 | int j; |
1a0e70ac | 1414 | ERROR("Failed to create \"%s\"", hierarchies[i]->fullcgpath); |
457ca9aa SH |
1415 | free(hierarchies[i]->fullcgpath); |
1416 | hierarchies[i]->fullcgpath = NULL; | |
ccb4cabe | 1417 | for (j = 0; j < i; j++) |
457ca9aa | 1418 | remove_path_for_hierarchy(hierarchies[j], cgname); |
ccb4cabe SH |
1419 | idx++; |
1420 | goto again; | |
1421 | } | |
1422 | } | |
1423 | /* Done */ | |
1424 | d->container_cgroup = cgname; | |
1425 | return true; | |
1426 | ||
1427 | out_free: | |
1428 | free(cgname); | |
1429 | return false; | |
1430 | } | |
1431 | ||
ccb4cabe SH |
1432 | static bool cgfsng_enter(void *hdata, pid_t pid) |
1433 | { | |
ccb4cabe SH |
1434 | char pidstr[25]; |
1435 | int i, len; | |
1436 | ||
1437 | len = snprintf(pidstr, 25, "%d", pid); | |
1438 | if (len < 0 || len > 25) | |
1439 | return false; | |
1440 | ||
457ca9aa SH |
1441 | for (i = 0; hierarchies[i]; i++) { |
1442 | char *fullpath = must_make_path(hierarchies[i]->fullcgpath, | |
ccb4cabe SH |
1443 | "cgroup.procs", NULL); |
1444 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { | |
d3b00a8f | 1445 | SYSERROR("Failed to enter %s", fullpath); |
ccb4cabe SH |
1446 | free(fullpath); |
1447 | return false; | |
1448 | } | |
1449 | free(fullpath); | |
1450 | } | |
1451 | ||
1452 | return true; | |
1453 | } | |
1454 | ||
1455 | struct chown_data { | |
1456 | struct cgfsng_handler_data *d; | |
1a0e70ac | 1457 | uid_t origuid; /* target uid in parent namespace */ |
ccb4cabe SH |
1458 | }; |
1459 | ||
c0888dfe SH |
1460 | /* |
1461 | * chgrp the container cgroups to container group. We leave | |
1462 | * the container owner as cgroup owner. So we must make the | |
1463 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1464 | * |
1465 | * Also chown the tasks and cgroup.procs files. Those may not | |
1466 | * exist depending on kernel version. | |
c0888dfe | 1467 | */ |
ccb4cabe SH |
1468 | static int chown_cgroup_wrapper(void *data) |
1469 | { | |
1470 | struct chown_data *arg = data; | |
ccb4cabe SH |
1471 | uid_t destuid; |
1472 | int i; | |
1473 | ||
1474 | if (setresgid(0,0,0) < 0) | |
1475 | SYSERROR("Failed to setgid to 0"); | |
1476 | if (setresuid(0,0,0) < 0) | |
1477 | SYSERROR("Failed to setuid to 0"); | |
1478 | if (setgroups(0, NULL) < 0) | |
1479 | SYSERROR("Failed to clear groups"); | |
1480 | ||
1481 | destuid = get_ns_uid(arg->origuid); | |
1482 | ||
457ca9aa SH |
1483 | for (i = 0; hierarchies[i]; i++) { |
1484 | char *fullpath, *path = hierarchies[i]->fullcgpath; | |
43647298 SH |
1485 | |
1486 | if (chown(path, destuid, 0) < 0) { | |
ab8f5424 | 1487 | SYSERROR("Error chowning %s to %d", path, (int) destuid); |
ccb4cabe SH |
1488 | return -1; |
1489 | } | |
c0888dfe | 1490 | |
43647298 | 1491 | if (chmod(path, 0775) < 0) { |
ab8f5424 | 1492 | SYSERROR("Error chmoding %s", path); |
c0888dfe SH |
1493 | return -1; |
1494 | } | |
ccb4cabe | 1495 | |
ab8f5424 SH |
1496 | /* |
1497 | * Failures to chown these are inconvenient but not detrimental | |
1498 | * We leave these owned by the container launcher, so that container | |
1499 | * root can write to the files to attach. We chmod them 664 so that | |
1500 | * container systemd can write to the files (which systemd in wily | |
1501 | * insists on doing) | |
1502 | */ | |
43647298 SH |
1503 | fullpath = must_make_path(path, "tasks", NULL); |
1504 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1505 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1506 | strerror(errno)); | |
ab8f5424 | 1507 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1508 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
43647298 SH |
1509 | free(fullpath); |
1510 | ||
1511 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
1512 | if (chown(fullpath, destuid, 0) < 0 && errno != ENOENT) | |
13277ec4 | 1513 | WARN("Failed chowning %s to %d: %s", fullpath, (int) destuid, |
1514 | strerror(errno)); | |
ab8f5424 | 1515 | if (chmod(fullpath, 0664) < 0) |
13277ec4 | 1516 | WARN("Error chmoding %s: %s", path, strerror(errno)); |
ccb4cabe SH |
1517 | free(fullpath); |
1518 | } | |
1519 | ||
1520 | return 0; | |
1521 | } | |
1522 | ||
058c1cb6 | 1523 | static bool cgfsng_chown(void *hdata, struct lxc_conf *conf) |
ccb4cabe SH |
1524 | { |
1525 | struct cgfsng_handler_data *d = hdata; | |
1526 | struct chown_data wrap; | |
1527 | ||
1528 | if (!d) | |
1529 | return false; | |
1530 | ||
1531 | if (lxc_list_empty(&conf->id_map)) | |
1532 | return true; | |
1533 | ||
1534 | wrap.d = d; | |
1535 | wrap.origuid = geteuid(); | |
1536 | ||
c9b7c33e CB |
1537 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1538 | "chown_cgroup_wrapper") < 0) { | |
ccb4cabe SH |
1539 | ERROR("Error requesting cgroup chown in new namespace"); |
1540 | return false; | |
1541 | } | |
1542 | ||
1543 | return true; | |
1544 | } | |
1545 | ||
8aa1044f SH |
1546 | /* |
1547 | * We've safe-mounted a tmpfs as parent, so we don't need to protect against | |
1548 | * symlinks any more - just use mount | |
1549 | */ | |
1550 | ||
1551 | /* mount cgroup-full if requested */ | |
1552 | static int mount_cgroup_full(int type, struct hierarchy *h, char *dest, | |
1553 | char *container_cgroup) | |
1554 | { | |
1555 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1556 | return 0; | |
1557 | if (mount(h->mountpoint, dest, "cgroup", MS_BIND, NULL) < 0) { | |
1558 | SYSERROR("Error bind-mounting %s cgroup onto %s", h->mountpoint, | |
1559 | dest); | |
1560 | return -1; | |
1561 | } | |
1562 | if (type != LXC_AUTO_CGROUP_FULL_RW) { | |
5b6f9369 SH |
1563 | unsigned long flags = MS_BIND | MS_NOSUID | MS_NOEXEC | MS_NODEV | |
1564 | MS_REMOUNT | MS_RDONLY; | |
1565 | if (mount(NULL, dest, "cgroup", flags, NULL) < 0) { | |
8aa1044f SH |
1566 | SYSERROR("Error remounting %s readonly", dest); |
1567 | return -1; | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | INFO("Bind mounted %s onto %s", h->mountpoint, dest); | |
1572 | if (type != LXC_AUTO_CGROUP_FULL_MIXED) | |
1573 | return 0; | |
1574 | ||
1575 | /* mount just the container path rw */ | |
1576 | char *source = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
5b6f9369 | 1577 | char *rwpath = must_make_path(dest, h->base_cgroup, container_cgroup, NULL); |
8aa1044f | 1578 | if (mount(source, rwpath, "cgroup", MS_BIND, NULL) < 0) |
13277ec4 | 1579 | WARN("Failed to mount %s read-write: %s", rwpath, |
1580 | strerror(errno)); | |
8aa1044f SH |
1581 | INFO("Made %s read-write", rwpath); |
1582 | free(rwpath); | |
1583 | free(source); | |
1584 | return 0; | |
1585 | } | |
1586 | ||
1587 | /* cgroup-full:* is done, no need to create subdirs */ | |
1588 | static bool cg_mount_needs_subdirs(int type) | |
1589 | { | |
1590 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1591 | return false; | |
1592 | return true; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * After $rootfs/sys/fs/container/controller/the/cg/path has been | |
1597 | * created, remount controller ro if needed and bindmount the | |
1598 | * cgroupfs onto controll/the/cg/path | |
1599 | */ | |
1600 | static int | |
1601 | do_secondstage_mounts_if_needed(int type, struct hierarchy *h, | |
1602 | char *controllerpath, char *cgpath, | |
1603 | const char *container_cgroup) | |
1604 | { | |
1605 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { | |
1606 | if (mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL) < 0) { | |
1607 | SYSERROR("Error bind-mounting %s", controllerpath); | |
1608 | return -1; | |
1609 | } | |
1610 | if (mount(controllerpath, controllerpath, "cgroup", | |
1611 | MS_REMOUNT | MS_BIND | MS_RDONLY, NULL) < 0) { | |
1612 | SYSERROR("Error remounting %s read-only", controllerpath); | |
1613 | return -1; | |
1614 | } | |
1615 | INFO("Remounted %s read-only", controllerpath); | |
1616 | } | |
1617 | char *sourcepath = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
1618 | int flags = MS_BIND; | |
1619 | if (type == LXC_AUTO_CGROUP_RO) | |
1620 | flags |= MS_RDONLY; | |
1621 | INFO("Mounting %s onto %s", sourcepath, cgpath); | |
1622 | if (mount(sourcepath, cgpath, "cgroup", flags, NULL) < 0) { | |
1623 | free(sourcepath); | |
1624 | SYSERROR("Error mounting cgroup %s onto %s", h->controllers[0], | |
1625 | cgpath); | |
1626 | return -1; | |
1627 | } | |
1628 | free(sourcepath); | |
1629 | INFO("Completed second stage cgroup automounts for %s", cgpath); | |
1630 | return 0; | |
1631 | } | |
1632 | ||
a760603e | 1633 | static int mount_cgroup_cgns_supported(int type, struct hierarchy *h, const char *controllerpath) |
b635e92d CB |
1634 | { |
1635 | int ret; | |
1636 | char *controllers = NULL; | |
a760603e CB |
1637 | char *fstype = "cgroup2"; |
1638 | unsigned long flags = 0; | |
b635e92d | 1639 | |
a760603e CB |
1640 | flags |= MS_NOSUID; |
1641 | flags |= MS_NOEXEC; | |
1642 | flags |= MS_NODEV; | |
1643 | flags |= MS_RELATIME; | |
1644 | ||
1645 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_FULL_RO) | |
1646 | flags |= MS_RDONLY; | |
1647 | ||
1648 | if (!h->is_cgroup_v2) { | |
1649 | controllers = lxc_string_join(",", (const char **)h->controllers, false); | |
1650 | if (!controllers) | |
1651 | return -ENOMEM; | |
1652 | fstype = "cgroup"; | |
b635e92d CB |
1653 | } |
1654 | ||
a760603e | 1655 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); |
b635e92d CB |
1656 | free(controllers); |
1657 | if (ret < 0) { | |
a760603e | 1658 | SYSERROR("Failed to mount %s with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1659 | return -1; |
1660 | } | |
1661 | ||
a760603e | 1662 | DEBUG("Mounted %s with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1663 | return 0; |
1664 | } | |
1665 | ||
ccb4cabe SH |
1666 | static bool cgfsng_mount(void *hdata, const char *root, int type) |
1667 | { | |
b635e92d | 1668 | int i; |
8aa1044f SH |
1669 | char *tmpfspath = NULL; |
1670 | bool retval = false; | |
b635e92d CB |
1671 | struct lxc_handler *handler = hdata; |
1672 | struct cgfsng_handler_data *d = handler->cgroup_data; | |
1673 | bool has_cgns = false, has_sys_admin = true; | |
8aa1044f SH |
1674 | |
1675 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1676 | return true; | |
1677 | ||
b635e92d CB |
1678 | has_cgns = cgns_supported(); |
1679 | if (!lxc_list_empty(&handler->conf->keepcaps)) | |
1680 | has_sys_admin = in_caplist(CAP_SYS_ADMIN, &handler->conf->keepcaps); | |
1681 | else | |
1682 | has_sys_admin = !in_caplist(CAP_SYS_ADMIN, &handler->conf->caps); | |
1683 | ||
1684 | if (has_cgns && has_sys_admin) | |
ccb4cabe | 1685 | return true; |
8aa1044f SH |
1686 | |
1687 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); | |
1688 | ||
1689 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1690 | type = LXC_AUTO_CGROUP_MIXED; | |
1691 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1692 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1693 | ||
1694 | /* Mount tmpfs */ | |
1695 | if (safe_mount("cgroup_root", tmpfspath, "tmpfs", | |
1696 | MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_RELATIME, | |
1697 | "size=10240k,mode=755", | |
1698 | root) < 0) | |
1699 | goto bad; | |
1700 | ||
457ca9aa | 1701 | for (i = 0; hierarchies[i]; i++) { |
8aa1044f | 1702 | char *controllerpath, *path2; |
457ca9aa | 1703 | struct hierarchy *h = hierarchies[i]; |
8aa1044f SH |
1704 | char *controller = strrchr(h->mountpoint, '/'); |
1705 | int r; | |
1706 | ||
1707 | if (!controller) | |
1708 | continue; | |
1709 | controller++; | |
1710 | controllerpath = must_make_path(tmpfspath, controller, NULL); | |
1711 | if (dir_exists(controllerpath)) { | |
1712 | free(controllerpath); | |
1713 | continue; | |
1714 | } | |
1715 | if (mkdir(controllerpath, 0755) < 0) { | |
1716 | SYSERROR("Error creating cgroup path: %s", controllerpath); | |
1717 | free(controllerpath); | |
1718 | goto bad; | |
1719 | } | |
b635e92d CB |
1720 | |
1721 | if (has_cgns && !has_sys_admin) { | |
1722 | /* If cgroup namespaces are supported but the container | |
1723 | * will not have CAP_SYS_ADMIN after it has started we | |
1724 | * need to mount the cgroups manually. | |
1725 | */ | |
a760603e | 1726 | r = mount_cgroup_cgns_supported(type, h, controllerpath); |
b635e92d CB |
1727 | free(controllerpath); |
1728 | if (r < 0) | |
1729 | goto bad; | |
1730 | continue; | |
1731 | } | |
1732 | ||
8aa1044f SH |
1733 | if (mount_cgroup_full(type, h, controllerpath, d->container_cgroup) < 0) { |
1734 | free(controllerpath); | |
1735 | goto bad; | |
1736 | } | |
1737 | if (!cg_mount_needs_subdirs(type)) { | |
1738 | free(controllerpath); | |
1739 | continue; | |
1740 | } | |
ef4413fa | 1741 | path2 = must_make_path(controllerpath, h->base_cgroup, d->container_cgroup, NULL); |
8aa1044f SH |
1742 | if (mkdir_p(path2, 0755) < 0) { |
1743 | free(controllerpath); | |
1744 | goto bad; | |
1745 | } | |
2f62fb00 | 1746 | |
8aa1044f SH |
1747 | r = do_secondstage_mounts_if_needed(type, h, controllerpath, path2, |
1748 | d->container_cgroup); | |
1749 | free(controllerpath); | |
1750 | free(path2); | |
1751 | if (r < 0) | |
1752 | goto bad; | |
1753 | } | |
1754 | retval = true; | |
1755 | ||
1756 | bad: | |
1757 | free(tmpfspath); | |
1758 | return retval; | |
ccb4cabe SH |
1759 | } |
1760 | ||
1761 | static int recursive_count_nrtasks(char *dirname) | |
1762 | { | |
74f96976 | 1763 | struct dirent *direntp; |
ccb4cabe SH |
1764 | DIR *dir; |
1765 | int count = 0, ret; | |
1766 | char *path; | |
1767 | ||
1768 | dir = opendir(dirname); | |
1769 | if (!dir) | |
1770 | return 0; | |
1771 | ||
74f96976 | 1772 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1773 | struct stat mystat; |
1774 | ||
1775 | if (!direntp) | |
1776 | break; | |
1777 | ||
1778 | if (!strcmp(direntp->d_name, ".") || | |
1779 | !strcmp(direntp->d_name, "..")) | |
1780 | continue; | |
1781 | ||
1782 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1783 | ||
1784 | if (lstat(path, &mystat)) | |
1785 | goto next; | |
1786 | ||
1787 | if (!S_ISDIR(mystat.st_mode)) | |
1788 | goto next; | |
1789 | ||
1790 | count += recursive_count_nrtasks(path); | |
1791 | next: | |
1792 | free(path); | |
1793 | } | |
1794 | ||
1795 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1796 | ret = lxc_count_file_lines(path); | |
1797 | if (ret != -1) | |
1798 | count += ret; | |
1799 | free(path); | |
1800 | ||
1801 | (void) closedir(dir); | |
1802 | ||
1803 | return count; | |
1804 | } | |
1805 | ||
1806 | static int cgfsng_nrtasks(void *hdata) { | |
1807 | struct cgfsng_handler_data *d = hdata; | |
1808 | char *path; | |
1809 | int count; | |
1810 | ||
457ca9aa | 1811 | if (!d || !d->container_cgroup || !hierarchies) |
ccb4cabe | 1812 | return -1; |
457ca9aa | 1813 | path = must_make_path(hierarchies[0]->fullcgpath, NULL); |
ccb4cabe SH |
1814 | count = recursive_count_nrtasks(path); |
1815 | free(path); | |
1816 | return count; | |
1817 | } | |
1818 | ||
1819 | /* Only root needs to escape to the cgroup of its init */ | |
7103fe6f | 1820 | static bool cgfsng_escape() |
ccb4cabe | 1821 | { |
ccb4cabe SH |
1822 | int i; |
1823 | ||
1824 | if (geteuid()) | |
1825 | return true; | |
1826 | ||
457ca9aa SH |
1827 | for (i = 0; hierarchies[i]; i++) { |
1828 | char *fullpath = must_make_path(hierarchies[i]->mountpoint, | |
1829 | hierarchies[i]->base_cgroup, | |
ccb4cabe SH |
1830 | "cgroup.procs", NULL); |
1831 | if (lxc_write_to_file(fullpath, "0", 2, false) != 0) { | |
d3b00a8f | 1832 | SYSERROR("Failed to escape to %s", fullpath); |
ccb4cabe | 1833 | free(fullpath); |
6df334d1 | 1834 | return false; |
ccb4cabe SH |
1835 | } |
1836 | free(fullpath); | |
1837 | } | |
1838 | ||
6df334d1 | 1839 | return true; |
ccb4cabe SH |
1840 | } |
1841 | ||
36662416 TA |
1842 | static int cgfsng_num_hierarchies(void) |
1843 | { | |
1844 | int i; | |
1845 | ||
1846 | for (i = 0; hierarchies[i]; i++) | |
1847 | ; | |
1848 | ||
1849 | return i; | |
1850 | } | |
1851 | ||
1852 | static bool cgfsng_get_hierarchies(int n, char ***out) | |
1853 | { | |
1854 | int i; | |
1855 | ||
1856 | /* sanity check n */ | |
1857 | for (i = 0; i < n; i++) { | |
1858 | if (!hierarchies[i]) | |
1859 | return false; | |
1860 | } | |
1861 | ||
1862 | *out = hierarchies[i]->controllers; | |
1863 | ||
1864 | return true; | |
1865 | } | |
1866 | ||
ccb4cabe SH |
1867 | #define THAWED "THAWED" |
1868 | #define THAWED_LEN (strlen(THAWED)) | |
1869 | ||
1870 | static bool cgfsng_unfreeze(void *hdata) | |
1871 | { | |
ccb4cabe | 1872 | char *fullpath; |
457ca9aa | 1873 | struct hierarchy *h = get_hierarchy("freezer"); |
ccb4cabe | 1874 | |
457ca9aa | 1875 | if (!h) |
ccb4cabe SH |
1876 | return false; |
1877 | fullpath = must_make_path(h->fullcgpath, "freezer.state", NULL); | |
1878 | if (lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false) != 0) { | |
1879 | free(fullpath); | |
1880 | return false; | |
1881 | } | |
1882 | free(fullpath); | |
1883 | return true; | |
1884 | } | |
1885 | ||
1886 | static const char *cgfsng_get_cgroup(void *hdata, const char *subsystem) | |
1887 | { | |
457ca9aa | 1888 | struct hierarchy *h = get_hierarchy(subsystem); |
ccb4cabe SH |
1889 | if (!h) |
1890 | return NULL; | |
1891 | ||
371f834d SH |
1892 | return h->fullcgpath ? h->fullcgpath + strlen(h->mountpoint) : NULL; |
1893 | } | |
1894 | ||
1895 | /* | |
1896 | * Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a | |
1897 | * full path, which must be freed by the caller. | |
1898 | */ | |
1899 | static char *build_full_cgpath_from_monitorpath(struct hierarchy *h, | |
1900 | const char *inpath, | |
1901 | const char *filename) | |
1902 | { | |
371f834d | 1903 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
1904 | } |
1905 | ||
1906 | static bool cgfsng_attach(const char *name, const char *lxcpath, pid_t pid) | |
1907 | { | |
ccb4cabe SH |
1908 | char pidstr[25]; |
1909 | int i, len; | |
1910 | ||
1911 | len = snprintf(pidstr, 25, "%d", pid); | |
1912 | if (len < 0 || len > 25) | |
1913 | return false; | |
1914 | ||
457ca9aa | 1915 | for (i = 0; hierarchies[i]; i++) { |
ccb4cabe | 1916 | char *path, *fullpath; |
457ca9aa | 1917 | struct hierarchy *h = hierarchies[i]; |
ccb4cabe SH |
1918 | |
1919 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); | |
1a0e70ac | 1920 | if (!path) /* not running */ |
ccb4cabe SH |
1921 | continue; |
1922 | ||
371f834d SH |
1923 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
1924 | free(path); | |
ccb4cabe SH |
1925 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { |
1926 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); | |
1927 | free(fullpath); | |
ccb4cabe SH |
1928 | return false; |
1929 | } | |
ccb4cabe SH |
1930 | free(fullpath); |
1931 | } | |
1932 | ||
ccb4cabe SH |
1933 | return true; |
1934 | } | |
1935 | ||
1936 | /* | |
1937 | * Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. | |
1938 | * Here we don't have a cgroup_data set up, so we ask the running | |
1939 | * container through the commands API for the cgroup path | |
1940 | */ | |
1941 | static int cgfsng_get(const char *filename, char *value, size_t len, const char *name, const char *lxcpath) | |
1942 | { | |
1943 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1944 | struct hierarchy *h; |
1945 | int ret = -1; | |
1946 | ||
1947 | subsystem = alloca(strlen(filename) + 1); | |
1948 | strcpy(subsystem, filename); | |
1949 | if ((p = strchr(subsystem, '.')) != NULL) | |
1950 | *p = '\0'; | |
1951 | ||
1952 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1a0e70ac | 1953 | if (!path) /* not running */ |
ccb4cabe SH |
1954 | return -1; |
1955 | ||
457ca9aa | 1956 | h = get_hierarchy(subsystem); |
ccb4cabe | 1957 | if (h) { |
371f834d | 1958 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1959 | ret = lxc_read_from_file(fullpath, value, len); |
1960 | free(fullpath); | |
1961 | } | |
1962 | ||
ccb4cabe SH |
1963 | free(path); |
1964 | ||
1965 | return ret; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. | |
1970 | * Here we don't have a cgroup_data set up, so we ask the running | |
1971 | * container through the commands API for the cgroup path | |
1972 | */ | |
1973 | static int cgfsng_set(const char *filename, const char *value, const char *name, const char *lxcpath) | |
1974 | { | |
1975 | char *subsystem, *p, *path; | |
ccb4cabe SH |
1976 | struct hierarchy *h; |
1977 | int ret = -1; | |
1978 | ||
1979 | subsystem = alloca(strlen(filename) + 1); | |
1980 | strcpy(subsystem, filename); | |
1981 | if ((p = strchr(subsystem, '.')) != NULL) | |
1982 | *p = '\0'; | |
1983 | ||
1984 | path = lxc_cmd_get_cgroup_path(name, lxcpath, subsystem); | |
1a0e70ac | 1985 | if (!path) /* not running */ |
ccb4cabe SH |
1986 | return -1; |
1987 | ||
457ca9aa | 1988 | h = get_hierarchy(subsystem); |
ccb4cabe | 1989 | if (h) { |
371f834d | 1990 | char *fullpath = build_full_cgpath_from_monitorpath(h, path, filename); |
ccb4cabe SH |
1991 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); |
1992 | free(fullpath); | |
1993 | } | |
1994 | ||
ccb4cabe SH |
1995 | free(path); |
1996 | ||
1997 | return ret; | |
1998 | } | |
1999 | ||
72add155 SH |
2000 | /* |
2001 | * take devices cgroup line | |
2002 | * /dev/foo rwx | |
2003 | * and convert it to a valid | |
2004 | * type major:minor mode | |
2005 | * line. Return <0 on error. Dest is a preallocated buffer | |
2006 | * long enough to hold the output. | |
2007 | */ | |
2008 | static int convert_devpath(const char *invalue, char *dest) | |
2009 | { | |
2a06d041 CB |
2010 | int n_parts; |
2011 | char *p, *path, type; | |
72add155 SH |
2012 | struct stat sb; |
2013 | unsigned long minor, major; | |
2a06d041 CB |
2014 | int ret = -EINVAL; |
2015 | char *mode = NULL; | |
72add155 SH |
2016 | |
2017 | path = must_copy_string(invalue); | |
2018 | ||
2019 | /* | |
2020 | * read path followed by mode; ignore any trailing text. | |
2021 | * A ' # comment' would be legal. Technically other text | |
2022 | * is not legal, we could check for that if we cared to | |
2023 | */ | |
2024 | for (n_parts = 1, p = path; *p && n_parts < 3; p++) { | |
2c2d6c49 SH |
2025 | if (*p != ' ') |
2026 | continue; | |
2027 | *p = '\0'; | |
2028 | if (n_parts != 1) | |
2029 | break; | |
2030 | p++; | |
2031 | n_parts++; | |
2032 | while (*p == ' ') | |
2033 | p++; | |
2034 | mode = p; | |
2035 | if (*p == '\0') | |
2036 | goto out; | |
72add155 | 2037 | } |
2c2d6c49 SH |
2038 | |
2039 | if (n_parts == 1) | |
72add155 | 2040 | goto out; |
72add155 SH |
2041 | |
2042 | ret = stat(path, &sb); | |
2043 | if (ret < 0) | |
2044 | goto out; | |
2045 | ||
72add155 SH |
2046 | mode_t m = sb.st_mode & S_IFMT; |
2047 | switch (m) { | |
2048 | case S_IFBLK: | |
2049 | type = 'b'; | |
2050 | break; | |
2051 | case S_IFCHR: | |
2052 | type = 'c'; | |
2053 | break; | |
2c2d6c49 | 2054 | default: |
72add155 SH |
2055 | ERROR("Unsupported device type %i for %s", m, path); |
2056 | ret = -EINVAL; | |
2057 | goto out; | |
2058 | } | |
2c2d6c49 SH |
2059 | |
2060 | major = MAJOR(sb.st_rdev); | |
2061 | minor = MINOR(sb.st_rdev); | |
2062 | ret = snprintf(dest, 50, "%c %lu:%lu %s", type, major, minor, mode); | |
72add155 | 2063 | if (ret < 0 || ret >= 50) { |
2a06d041 CB |
2064 | ERROR("Error on configuration value \"%c %lu:%lu %s\" (max 50 " |
2065 | "chars)", type, major, minor, mode); | |
72add155 SH |
2066 | ret = -ENAMETOOLONG; |
2067 | goto out; | |
2068 | } | |
2069 | ret = 0; | |
2070 | ||
2071 | out: | |
2072 | free(path); | |
2073 | return ret; | |
2074 | } | |
2075 | ||
ccb4cabe SH |
2076 | /* |
2077 | * Called from setup_limits - here we have the container's cgroup_data because | |
2078 | * we created the cgroups | |
2079 | */ | |
2080 | static int lxc_cgroup_set_data(const char *filename, const char *value, struct cgfsng_handler_data *d) | |
2081 | { | |
b3646d7e | 2082 | char *fullpath, *p; |
1a0e70ac CB |
2083 | /* "b|c <2^64-1>:<2^64-1> r|w|m" = 47 chars max */ |
2084 | char converted_value[50]; | |
b3646d7e CB |
2085 | struct hierarchy *h; |
2086 | int ret = 0; | |
2087 | char *controller = NULL; | |
ccb4cabe | 2088 | |
b3646d7e CB |
2089 | controller = alloca(strlen(filename) + 1); |
2090 | strcpy(controller, filename); | |
2091 | if ((p = strchr(controller, '.')) != NULL) | |
ccb4cabe SH |
2092 | *p = '\0'; |
2093 | ||
c8bf519d | 2094 | if (strcmp("devices.allow", filename) == 0 && value[0] == '/') { |
72add155 SH |
2095 | ret = convert_devpath(value, converted_value); |
2096 | if (ret < 0) | |
c8bf519d | 2097 | return ret; |
72add155 SH |
2098 | value = converted_value; |
2099 | ||
c8bf519d | 2100 | } |
2101 | ||
b3646d7e CB |
2102 | h = get_hierarchy(controller); |
2103 | if (!h) { | |
2104 | ERROR("Failed to setup limits for the \"%s\" controller. " | |
2105 | "The controller seems to be unused by \"cgfsng\" cgroup " | |
2106 | "driver or not enabled on the cgroup hierarchy", | |
2107 | controller); | |
2108 | return -1; | |
ccb4cabe | 2109 | } |
b3646d7e CB |
2110 | |
2111 | fullpath = must_make_path(h->fullcgpath, filename, NULL); | |
2112 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); | |
2113 | free(fullpath); | |
ccb4cabe SH |
2114 | return ret; |
2115 | } | |
2116 | ||
2117 | static bool cgfsng_setup_limits(void *hdata, struct lxc_list *cgroup_settings, | |
2118 | bool do_devices) | |
2119 | { | |
2120 | struct cgfsng_handler_data *d = hdata; | |
2121 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; | |
2122 | struct lxc_cgroup *cg; | |
ccb4cabe SH |
2123 | bool ret = false; |
2124 | ||
2125 | if (lxc_list_empty(cgroup_settings)) | |
2126 | return true; | |
2127 | ||
2128 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
2129 | if (!sorted_cgroup_settings) { | |
2130 | return false; | |
2131 | } | |
2132 | ||
ccb4cabe SH |
2133 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
2134 | cg = iterator->elem; | |
2135 | ||
2136 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
ccb4cabe SH |
2137 | if (lxc_cgroup_set_data(cg->subsystem, cg->value, d)) { |
2138 | if (do_devices && (errno == EACCES || errno == EPERM)) { | |
2139 | WARN("Error setting %s to %s for %s", | |
2140 | cg->subsystem, cg->value, d->name); | |
2141 | continue; | |
2142 | } | |
2143 | SYSERROR("Error setting %s to %s for %s", | |
2144 | cg->subsystem, cg->value, d->name); | |
2145 | goto out; | |
2146 | } | |
6a628f4a | 2147 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); |
ccb4cabe | 2148 | } |
ccb4cabe SH |
2149 | } |
2150 | ||
2151 | ret = true; | |
2152 | INFO("cgroup has been setup"); | |
2153 | out: | |
ccb4cabe SH |
2154 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
2155 | lxc_list_del(iterator); | |
2156 | free(iterator); | |
2157 | } | |
2158 | free(sorted_cgroup_settings); | |
2159 | return ret; | |
2160 | } | |
2161 | ||
2162 | static struct cgroup_ops cgfsng_ops = { | |
2163 | .init = cgfsng_init, | |
2164 | .destroy = cgfsng_destroy, | |
2165 | .create = cgfsng_create, | |
2166 | .enter = cgfsng_enter, | |
ccb4cabe | 2167 | .escape = cgfsng_escape, |
36662416 TA |
2168 | .num_hierarchies = cgfsng_num_hierarchies, |
2169 | .get_hierarchies = cgfsng_get_hierarchies, | |
ccb4cabe SH |
2170 | .get_cgroup = cgfsng_get_cgroup, |
2171 | .get = cgfsng_get, | |
2172 | .set = cgfsng_set, | |
2173 | .unfreeze = cgfsng_unfreeze, | |
2174 | .setup_limits = cgfsng_setup_limits, | |
2175 | .name = "cgroupfs-ng", | |
2176 | .attach = cgfsng_attach, | |
058c1cb6 | 2177 | .chown = cgfsng_chown, |
ccb4cabe SH |
2178 | .mount_cgroup = cgfsng_mount, |
2179 | .nrtasks = cgfsng_nrtasks, | |
2180 | .driver = CGFSNG, | |
2181 | ||
2182 | /* unsupported */ | |
2183 | .create_legacy = NULL, | |
2184 | }; |