]>
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> | |
3fd0de4d | 8 | * Christian Brauner <christian.brauner@ubuntu.com> |
ccb4cabe SH |
9 | * |
10 | * This library is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU Lesser General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2.1 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This library is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * Lesser General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU Lesser General Public | |
21 | * License along with this library; if not, write to the Free Software | |
22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
23 | */ | |
24 | ||
25 | /* | |
26 | * cgfs-ng.c: this is a new, simplified implementation of a filesystem | |
27 | * cgroup backend. The original cgfs.c was designed to be as flexible | |
28 | * as possible. It would try to find cgroup filesystems no matter where | |
29 | * or how you had them mounted, and deduce the most usable mount for | |
0e7ff52c | 30 | * each controller. |
ccb4cabe SH |
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> |
c8bf519d | 48 | #include <linux/kdev_t.h> |
438c4581 CB |
49 | #include <linux/types.h> |
50 | #include <sys/types.h> | |
c8bf519d | 51 | |
b635e92d | 52 | #include "caps.h" |
ccb4cabe | 53 | #include "cgroup.h" |
6328fd9c | 54 | #include "cgroup_utils.h" |
ccb4cabe | 55 | #include "commands.h" |
43654d34 | 56 | #include "conf.h" |
a54694f8 | 57 | #include "log.h" |
43654d34 | 58 | #include "storage/storage.h" |
a54694f8 | 59 | #include "utils.h" |
ccb4cabe | 60 | |
64e82f8b DJ |
61 | #ifndef HAVE_STRLCPY |
62 | #include "include/strlcpy.h" | |
63 | #endif | |
64 | ||
3ebe2fbd DJ |
65 | #ifndef HAVE_STRLCAT |
66 | #include "include/strlcat.h" | |
67 | #endif | |
68 | ||
ccb4cabe SH |
69 | lxc_log_define(lxc_cgfsng, lxc); |
70 | ||
ccb4cabe SH |
71 | static void free_string_list(char **clist) |
72 | { | |
2d5fe5ba | 73 | int i; |
ccb4cabe | 74 | |
2d5fe5ba CB |
75 | if (!clist) |
76 | return; | |
77 | ||
78 | for (i = 0; clist[i]; i++) | |
79 | free(clist[i]); | |
80 | ||
81 | free(clist); | |
ccb4cabe SH |
82 | } |
83 | ||
7745483d | 84 | /* Allocate a pointer, do not fail. */ |
ccb4cabe SH |
85 | static void *must_alloc(size_t sz) |
86 | { | |
87 | return must_realloc(NULL, sz); | |
88 | } | |
89 | ||
8b8db2f6 CB |
90 | /* Given a pointer to a null-terminated array of pointers, realloc to add one |
91 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
92 | * second-to-last entry - that is, the one which is now available for use | |
93 | * (keeping the list null-terminated). | |
ccb4cabe SH |
94 | */ |
95 | static int append_null_to_list(void ***list) | |
96 | { | |
97 | int newentry = 0; | |
98 | ||
99 | if (*list) | |
8b8db2f6 CB |
100 | for (; (*list)[newentry]; newentry++) |
101 | ; | |
ccb4cabe SH |
102 | |
103 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
104 | (*list)[newentry + 1] = NULL; | |
105 | return newentry; | |
106 | } | |
107 | ||
8073018d CB |
108 | /* Given a null-terminated array of strings, check whether @entry is one of the |
109 | * strings. | |
ccb4cabe SH |
110 | */ |
111 | static bool string_in_list(char **list, const char *entry) | |
112 | { | |
113 | int i; | |
114 | ||
115 | if (!list) | |
116 | return false; | |
d6337a5f | 117 | |
ccb4cabe SH |
118 | for (i = 0; list[i]; i++) |
119 | if (strcmp(list[i], entry) == 0) | |
120 | return true; | |
121 | ||
122 | return false; | |
123 | } | |
124 | ||
ac010944 CB |
125 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into |
126 | * "name=systemd". Do not fail. | |
127 | */ | |
128 | static char *cg_legacy_must_prefix_named(char *entry) | |
129 | { | |
130 | size_t len; | |
131 | char *prefixed; | |
132 | ||
133 | len = strlen(entry); | |
134 | prefixed = must_alloc(len + 6); | |
135 | ||
cbe2185b CB |
136 | memcpy(prefixed, "name=", sizeof("name=") - 1); |
137 | memcpy(prefixed + sizeof("name=") - 1, entry, len); | |
ac010944 CB |
138 | prefixed[len + 5] = '\0'; |
139 | return prefixed; | |
140 | } | |
141 | ||
42a993b4 CB |
142 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time |
143 | * we are called. | |
ccb4cabe | 144 | * |
42a993b4 CB |
145 | * We also handle named subsystems here. Any controller which is not a kernel |
146 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
147 | * we refuse to use because we're not sure which we have here. | |
148 | * (TODO: We could work around this in some cases by just remounting to be | |
149 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
ccb4cabe SH |
150 | * |
151 | * The last entry will always be NULL. | |
152 | */ | |
42a993b4 CB |
153 | static void must_append_controller(char **klist, char **nlist, char ***clist, |
154 | char *entry) | |
ccb4cabe SH |
155 | { |
156 | int newentry; | |
157 | char *copy; | |
158 | ||
159 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
c2712f64 | 160 | ERROR("Refusing to use ambiguous controller \"%s\"", entry); |
ccb4cabe SH |
161 | ERROR("It is both a named and kernel subsystem"); |
162 | return; | |
163 | } | |
164 | ||
165 | newentry = append_null_to_list((void ***)clist); | |
166 | ||
167 | if (strncmp(entry, "name=", 5) == 0) | |
168 | copy = must_copy_string(entry); | |
169 | else if (string_in_list(klist, entry)) | |
170 | copy = must_copy_string(entry); | |
171 | else | |
7745483d | 172 | copy = cg_legacy_must_prefix_named(entry); |
ccb4cabe SH |
173 | |
174 | (*clist)[newentry] = copy; | |
175 | } | |
176 | ||
5ae0207c CB |
177 | /* Given a handler's cgroup data, return the struct hierarchy for the controller |
178 | * @c, or NULL if there is none. | |
ccb4cabe | 179 | */ |
2202afc9 | 180 | struct hierarchy *get_hierarchy(struct cgroup_ops *ops, const char *c) |
ccb4cabe SH |
181 | { |
182 | int i; | |
183 | ||
2202afc9 | 184 | if (!ops->hierarchies) |
ccb4cabe | 185 | return NULL; |
d6337a5f | 186 | |
2202afc9 | 187 | for (i = 0; ops->hierarchies[i]; i++) { |
d6337a5f CB |
188 | if (!c) { |
189 | /* This is the empty unified hierarchy. */ | |
2202afc9 CB |
190 | if (ops->hierarchies[i]->controllers && |
191 | !ops->hierarchies[i]->controllers[0]) | |
192 | return ops->hierarchies[i]; | |
d6337a5f | 193 | |
106f1f38 | 194 | continue; |
d6337a5f CB |
195 | } |
196 | ||
2202afc9 CB |
197 | if (string_in_list(ops->hierarchies[i]->controllers, c)) |
198 | return ops->hierarchies[i]; | |
ccb4cabe | 199 | } |
d6337a5f | 200 | |
ccb4cabe SH |
201 | return NULL; |
202 | } | |
203 | ||
a54694f8 CB |
204 | #define BATCH_SIZE 50 |
205 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
206 | { | |
207 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
208 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
209 | ||
210 | if (!*mem || newbatches > oldbatches) { | |
211 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
212 | } | |
213 | } | |
214 | ||
215 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
216 | { | |
217 | size_t full = oldlen + newlen; | |
218 | ||
219 | batch_realloc(dest, oldlen, full + 1); | |
220 | ||
221 | memcpy(*dest + oldlen, new, newlen + 1); | |
222 | } | |
223 | ||
224 | /* Slurp in a whole file */ | |
d6337a5f | 225 | static char *read_file(const char *fnam) |
a54694f8 CB |
226 | { |
227 | FILE *f; | |
228 | char *line = NULL, *buf = NULL; | |
229 | size_t len = 0, fulllen = 0; | |
230 | int linelen; | |
231 | ||
232 | f = fopen(fnam, "r"); | |
233 | if (!f) | |
234 | return NULL; | |
235 | while ((linelen = getline(&line, &len, f)) != -1) { | |
236 | append_line(&buf, fulllen, line, linelen); | |
237 | fulllen += linelen; | |
238 | } | |
239 | fclose(f); | |
240 | free(line); | |
241 | return buf; | |
242 | } | |
243 | ||
244 | /* Taken over modified from the kernel sources. */ | |
245 | #define NBITS 32 /* bits in uint32_t */ | |
246 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
247 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
248 | ||
249 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
250 | { | |
251 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
252 | } | |
253 | ||
254 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
255 | { | |
256 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
257 | } | |
258 | ||
259 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
260 | { | |
261 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
262 | } | |
263 | ||
264 | /* Create cpumask from cpulist aka turn: | |
265 | * | |
266 | * 0,2-3 | |
267 | * | |
d5d468f6 | 268 | * into bit array |
a54694f8 CB |
269 | * |
270 | * 1 0 1 1 | |
271 | */ | |
272 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
273 | { | |
274 | char *token; | |
d5d468f6 CB |
275 | size_t arrlen; |
276 | uint32_t *bitarr; | |
a54694f8 | 277 | char *saveptr = NULL; |
d5d468f6 CB |
278 | |
279 | arrlen = BITS_TO_LONGS(nbits); | |
280 | bitarr = calloc(arrlen, sizeof(uint32_t)); | |
a54694f8 CB |
281 | if (!bitarr) |
282 | return NULL; | |
283 | ||
284 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
285 | errno = 0; | |
d5d468f6 CB |
286 | unsigned end, start; |
287 | char *range; | |
a54694f8 | 288 | |
d5d468f6 CB |
289 | start = strtoul(token, NULL, 0); |
290 | end = start; | |
291 | range = strchr(token, '-'); | |
a54694f8 CB |
292 | if (range) |
293 | end = strtoul(range + 1, NULL, 0); | |
d5d468f6 | 294 | |
a54694f8 CB |
295 | if (!(start <= end)) { |
296 | free(bitarr); | |
297 | return NULL; | |
298 | } | |
299 | ||
300 | if (end >= nbits) { | |
301 | free(bitarr); | |
302 | return NULL; | |
303 | } | |
304 | ||
305 | while (start <= end) | |
306 | set_bit(start++, bitarr); | |
307 | } | |
308 | ||
309 | return bitarr; | |
310 | } | |
311 | ||
a54694f8 CB |
312 | /* Turn cpumask into simple, comma-separated cpulist. */ |
313 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
314 | { | |
a54694f8 | 315 | int ret; |
414c6719 | 316 | size_t i; |
a54694f8 | 317 | char **cpulist = NULL; |
414c6719 | 318 | char numstr[LXC_NUMSTRLEN64] = {0}; |
a54694f8 CB |
319 | |
320 | for (i = 0; i <= nbits; i++) { | |
414c6719 CB |
321 | if (!is_set(i, bitarr)) |
322 | continue; | |
323 | ||
324 | ret = snprintf(numstr, LXC_NUMSTRLEN64, "%zu", i); | |
325 | if (ret < 0 || (size_t)ret >= LXC_NUMSTRLEN64) { | |
326 | lxc_free_array((void **)cpulist, free); | |
327 | return NULL; | |
328 | } | |
329 | ||
330 | ret = lxc_append_string(&cpulist, numstr); | |
331 | if (ret < 0) { | |
332 | lxc_free_array((void **)cpulist, free); | |
333 | return NULL; | |
a54694f8 CB |
334 | } |
335 | } | |
414c6719 CB |
336 | |
337 | if (!cpulist) | |
338 | return NULL; | |
339 | ||
a54694f8 CB |
340 | return lxc_string_join(",", (const char **)cpulist, false); |
341 | } | |
342 | ||
343 | static ssize_t get_max_cpus(char *cpulist) | |
344 | { | |
345 | char *c1, *c2; | |
346 | char *maxcpus = cpulist; | |
347 | size_t cpus = 0; | |
348 | ||
349 | c1 = strrchr(maxcpus, ','); | |
350 | if (c1) | |
351 | c1++; | |
352 | ||
353 | c2 = strrchr(maxcpus, '-'); | |
354 | if (c2) | |
355 | c2++; | |
356 | ||
357 | if (!c1 && !c2) | |
358 | c1 = maxcpus; | |
359 | else if (c1 > c2) | |
360 | c2 = c1; | |
361 | else if (c1 < c2) | |
362 | c1 = c2; | |
333987b9 | 363 | else if (!c1 && c2) |
a54694f8 CB |
364 | c1 = c2; |
365 | ||
a54694f8 CB |
366 | errno = 0; |
367 | cpus = strtoul(c1, NULL, 0); | |
368 | if (errno != 0) | |
369 | return -1; | |
370 | ||
371 | return cpus; | |
372 | } | |
373 | ||
6f9584d8 | 374 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a3926f6a | 375 | static bool cg_legacy_filter_and_set_cpus(char *path, bool am_initialized) |
a54694f8 | 376 | { |
a54694f8 CB |
377 | int ret; |
378 | ssize_t i; | |
59ac3b88 CB |
379 | char *lastslash, *fpath, oldv; |
380 | ssize_t maxisol = 0, maxposs = 0; | |
381 | char *cpulist = NULL, *isolcpus = NULL, *posscpus = NULL; | |
382 | uint32_t *isolmask = NULL, *possmask = NULL; | |
6f9584d8 | 383 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
384 | |
385 | lastslash = strrchr(path, '/'); | |
59ac3b88 CB |
386 | if (!lastslash) { |
387 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
a54694f8 CB |
388 | return bret; |
389 | } | |
390 | oldv = *lastslash; | |
391 | *lastslash = '\0'; | |
392 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
393 | posscpus = read_file(fpath); | |
6f9584d8 | 394 | if (!posscpus) { |
59ac3b88 | 395 | SYSERROR("Failed to read file \"%s\"", fpath); |
6f9584d8 CB |
396 | goto on_error; |
397 | } | |
a54694f8 CB |
398 | |
399 | /* Get maximum number of cpus found in possible cpuset. */ | |
400 | maxposs = get_max_cpus(posscpus); | |
401 | if (maxposs < 0) | |
6f9584d8 | 402 | goto on_error; |
a54694f8 | 403 | |
6f9584d8 CB |
404 | if (!file_exists(__ISOL_CPUS)) { |
405 | /* This system doesn't expose isolated cpus. */ | |
59ac3b88 | 406 | DEBUG("The path \""__ISOL_CPUS"\" to read isolated cpus from does not exist"); |
65d29cbc CB |
407 | cpulist = posscpus; |
408 | /* No isolated cpus but we weren't already initialized by | |
409 | * someone. We should simply copy the parents cpuset.cpus | |
410 | * values. | |
411 | */ | |
412 | if (!am_initialized) { | |
59ac3b88 | 413 | DEBUG("Copying cpu settings of parent cgroup"); |
65d29cbc CB |
414 | goto copy_parent; |
415 | } | |
416 | /* No isolated cpus but we were already initialized by someone. | |
417 | * Nothing more to do for us. | |
418 | */ | |
6f9584d8 CB |
419 | goto on_success; |
420 | } | |
421 | ||
422 | isolcpus = read_file(__ISOL_CPUS); | |
423 | if (!isolcpus) { | |
59ac3b88 | 424 | SYSERROR("Failed to read file \""__ISOL_CPUS"\""); |
6f9584d8 CB |
425 | goto on_error; |
426 | } | |
a54694f8 | 427 | if (!isdigit(isolcpus[0])) { |
59ac3b88 | 428 | TRACE("No isolated cpus detected"); |
a54694f8 CB |
429 | cpulist = posscpus; |
430 | /* No isolated cpus but we weren't already initialized by | |
431 | * someone. We should simply copy the parents cpuset.cpus | |
432 | * values. | |
433 | */ | |
6f9584d8 | 434 | if (!am_initialized) { |
59ac3b88 | 435 | DEBUG("Copying cpu settings of parent cgroup"); |
a54694f8 | 436 | goto copy_parent; |
6f9584d8 | 437 | } |
a54694f8 CB |
438 | /* No isolated cpus but we were already initialized by someone. |
439 | * Nothing more to do for us. | |
440 | */ | |
6f9584d8 | 441 | goto on_success; |
a54694f8 CB |
442 | } |
443 | ||
444 | /* Get maximum number of cpus found in isolated cpuset. */ | |
445 | maxisol = get_max_cpus(isolcpus); | |
446 | if (maxisol < 0) | |
6f9584d8 | 447 | goto on_error; |
a54694f8 CB |
448 | |
449 | if (maxposs < maxisol) | |
450 | maxposs = maxisol; | |
451 | maxposs++; | |
452 | ||
453 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 | 454 | if (!possmask) { |
59ac3b88 | 455 | ERROR("Failed to create cpumask for possible cpus"); |
6f9584d8 CB |
456 | goto on_error; |
457 | } | |
a54694f8 CB |
458 | |
459 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 | 460 | if (!isolmask) { |
59ac3b88 | 461 | ERROR("Failed to create cpumask for isolated cpus"); |
6f9584d8 CB |
462 | goto on_error; |
463 | } | |
a54694f8 CB |
464 | |
465 | for (i = 0; i <= maxposs; i++) { | |
59ac3b88 CB |
466 | if (!is_set(i, isolmask) || !is_set(i, possmask)) |
467 | continue; | |
468 | ||
469 | flipped_bit = true; | |
470 | clear_bit(i, possmask); | |
a54694f8 CB |
471 | } |
472 | ||
6f9584d8 | 473 | if (!flipped_bit) { |
59ac3b88 | 474 | DEBUG("No isolated cpus present in cpuset"); |
6f9584d8 CB |
475 | goto on_success; |
476 | } | |
59ac3b88 | 477 | DEBUG("Removed isolated cpus from cpuset"); |
6f9584d8 | 478 | |
a54694f8 | 479 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 | 480 | if (!cpulist) { |
59ac3b88 | 481 | ERROR("Failed to create cpu list"); |
6f9584d8 CB |
482 | goto on_error; |
483 | } | |
a54694f8 CB |
484 | |
485 | copy_parent: | |
486 | *lastslash = oldv; | |
dcbc861e | 487 | free(fpath); |
a54694f8 | 488 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
7cea5905 | 489 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false, 0666); |
6f9584d8 | 490 | if (ret < 0) { |
59ac3b88 | 491 | SYSERROR("Failed to write cpu list to \"%s\"", fpath); |
6f9584d8 CB |
492 | goto on_error; |
493 | } | |
494 | ||
495 | on_success: | |
496 | bret = true; | |
a54694f8 | 497 | |
6f9584d8 | 498 | on_error: |
a54694f8 CB |
499 | free(fpath); |
500 | ||
501 | free(isolcpus); | |
502 | free(isolmask); | |
503 | ||
504 | if (posscpus != cpulist) | |
505 | free(posscpus); | |
506 | free(possmask); | |
507 | ||
508 | free(cpulist); | |
509 | return bret; | |
510 | } | |
511 | ||
e3a3fecf SH |
512 | /* Copy contents of parent(@path)/@file to @path/@file */ |
513 | static bool copy_parent_file(char *path, char *file) | |
514 | { | |
e3a3fecf | 515 | int ret; |
b095a8eb CB |
516 | char *fpath, *lastslash, oldv; |
517 | int len = 0; | |
518 | char *value = NULL; | |
e3a3fecf SH |
519 | |
520 | lastslash = strrchr(path, '/'); | |
b095a8eb CB |
521 | if (!lastslash) { |
522 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
e3a3fecf SH |
523 | return false; |
524 | } | |
525 | oldv = *lastslash; | |
526 | *lastslash = '\0'; | |
527 | fpath = must_make_path(path, file, NULL); | |
528 | len = lxc_read_from_file(fpath, NULL, 0); | |
529 | if (len <= 0) | |
b095a8eb CB |
530 | goto on_error; |
531 | ||
e3a3fecf | 532 | value = must_alloc(len + 1); |
b095a8eb CB |
533 | ret = lxc_read_from_file(fpath, value, len); |
534 | if (ret != len) | |
535 | goto on_error; | |
e3a3fecf | 536 | free(fpath); |
b095a8eb | 537 | |
e3a3fecf SH |
538 | *lastslash = oldv; |
539 | fpath = must_make_path(path, file, NULL); | |
7cea5905 | 540 | ret = lxc_write_to_file(fpath, value, len, false, 0666); |
e3a3fecf | 541 | if (ret < 0) |
b095a8eb | 542 | SYSERROR("Failed to write \"%s\" to file \"%s\"", value, fpath); |
e3a3fecf SH |
543 | free(fpath); |
544 | free(value); | |
545 | return ret >= 0; | |
546 | ||
b095a8eb CB |
547 | on_error: |
548 | SYSERROR("Failed to read file \"%s\"", fpath); | |
e3a3fecf SH |
549 | free(fpath); |
550 | free(value); | |
551 | return false; | |
552 | } | |
553 | ||
7793add3 CB |
554 | /* Initialize the cpuset hierarchy in first directory of @gname and set |
555 | * cgroup.clone_children so that children inherit settings. Since the | |
556 | * h->base_path is populated by init or ourselves, we know it is already | |
557 | * initialized. | |
e3a3fecf | 558 | */ |
a3926f6a | 559 | static bool cg_legacy_handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf | 560 | { |
7793add3 CB |
561 | int ret; |
562 | char v; | |
563 | char *cgpath, *clonechildrenpath, *slash; | |
e3a3fecf SH |
564 | |
565 | if (!string_in_list(h->controllers, "cpuset")) | |
566 | return true; | |
567 | ||
568 | if (*cgname == '/') | |
569 | cgname++; | |
570 | slash = strchr(cgname, '/'); | |
571 | if (slash) | |
572 | *slash = '\0'; | |
573 | ||
574 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); | |
575 | if (slash) | |
576 | *slash = '/'; | |
7793add3 CB |
577 | |
578 | ret = mkdir(cgpath, 0755); | |
579 | if (ret < 0) { | |
580 | if (errno != EEXIST) { | |
581 | SYSERROR("Failed to create directory \"%s\"", cgpath); | |
582 | free(cgpath); | |
583 | return false; | |
584 | } | |
e3a3fecf | 585 | } |
6f9584d8 | 586 | |
7793add3 CB |
587 | clonechildrenpath = |
588 | must_make_path(cgpath, "cgroup.clone_children", NULL); | |
6328fd9c CB |
589 | /* unified hierarchy doesn't have clone_children */ |
590 | if (!file_exists(clonechildrenpath)) { | |
e3a3fecf SH |
591 | free(clonechildrenpath); |
592 | free(cgpath); | |
593 | return true; | |
594 | } | |
7793add3 CB |
595 | |
596 | ret = lxc_read_from_file(clonechildrenpath, &v, 1); | |
597 | if (ret < 0) { | |
598 | SYSERROR("Failed to read file \"%s\"", clonechildrenpath); | |
e3a3fecf SH |
599 | free(clonechildrenpath); |
600 | free(cgpath); | |
601 | return false; | |
602 | } | |
603 | ||
a54694f8 | 604 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
a3926f6a | 605 | if (!cg_legacy_filter_and_set_cpus(cgpath, v == '1')) { |
7793add3 | 606 | SYSERROR("Failed to remove isolated cpus"); |
6f9584d8 CB |
607 | free(clonechildrenpath); |
608 | free(cgpath); | |
a54694f8 | 609 | return false; |
6f9584d8 | 610 | } |
a54694f8 | 611 | |
7793add3 CB |
612 | /* Already set for us by someone else. */ |
613 | if (v == '1') { | |
614 | DEBUG("\"cgroup.clone_children\" was already set to \"1\""); | |
e3a3fecf SH |
615 | free(clonechildrenpath); |
616 | free(cgpath); | |
617 | return true; | |
618 | } | |
619 | ||
620 | /* copy parent's settings */ | |
a54694f8 | 621 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
7793add3 | 622 | SYSERROR("Failed to copy \"cpuset.mems\" settings"); |
e3a3fecf SH |
623 | free(cgpath); |
624 | free(clonechildrenpath); | |
625 | return false; | |
626 | } | |
627 | free(cgpath); | |
628 | ||
7cea5905 | 629 | ret = lxc_write_to_file(clonechildrenpath, "1", 1, false, 0666); |
7793add3 | 630 | if (ret < 0) { |
e3a3fecf | 631 | /* Set clone_children so children inherit our settings */ |
7793add3 | 632 | SYSERROR("Failed to write 1 to \"%s\"", clonechildrenpath); |
e3a3fecf SH |
633 | free(clonechildrenpath); |
634 | return false; | |
635 | } | |
636 | free(clonechildrenpath); | |
637 | return true; | |
638 | } | |
639 | ||
5c0089ae CB |
640 | /* Given two null-terminated lists of strings, return true if any string is in |
641 | * both. | |
ccb4cabe SH |
642 | */ |
643 | static bool controller_lists_intersect(char **l1, char **l2) | |
644 | { | |
645 | int i; | |
646 | ||
647 | if (!l1 || !l2) | |
648 | return false; | |
649 | ||
650 | for (i = 0; l1[i]; i++) { | |
651 | if (string_in_list(l2, l1[i])) | |
652 | return true; | |
653 | } | |
5c0089ae | 654 | |
ccb4cabe SH |
655 | return false; |
656 | } | |
657 | ||
258449e5 CB |
658 | /* For a null-terminated list of controllers @clist, return true if any of those |
659 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
660 | * Realistically, if one is present, all must be present. | |
ccb4cabe SH |
661 | */ |
662 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
663 | { | |
664 | int i; | |
665 | ||
666 | if (!hlist) | |
667 | return false; | |
258449e5 | 668 | |
ccb4cabe SH |
669 | for (i = 0; hlist[i]; i++) |
670 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
671 | return true; | |
ccb4cabe | 672 | |
258449e5 | 673 | return false; |
ccb4cabe SH |
674 | } |
675 | ||
f57ac67f CB |
676 | /* Return true if the controller @entry is found in the null-terminated list of |
677 | * hierarchies @hlist. | |
ccb4cabe SH |
678 | */ |
679 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
680 | { | |
681 | int i; | |
d6337a5f | 682 | |
ccb4cabe SH |
683 | if (!hlist) |
684 | return false; | |
685 | ||
686 | for (i = 0; hlist[i]; i++) | |
687 | if (string_in_list(hlist[i]->controllers, entry)) | |
688 | return true; | |
d6337a5f | 689 | |
ccb4cabe SH |
690 | return false; |
691 | } | |
692 | ||
e1c27ab0 CB |
693 | /* Return true if all of the controllers which we require have been found. The |
694 | * required list is freezer and anything in lxc.cgroup.use. | |
ccb4cabe | 695 | */ |
2202afc9 | 696 | static bool all_controllers_found(struct cgroup_ops *ops) |
ccb4cabe | 697 | { |
e1c27ab0 CB |
698 | char *p; |
699 | char *saveptr = NULL; | |
2202afc9 | 700 | struct hierarchy **hlist = ops->hierarchies; |
ccb4cabe | 701 | |
ccb4cabe | 702 | if (!controller_found(hlist, "freezer")) { |
2202afc9 | 703 | ERROR("No freezer controller mountpoint found"); |
ccb4cabe SH |
704 | return false; |
705 | } | |
706 | ||
2202afc9 | 707 | if (!ops->cgroup_use) |
ccb4cabe | 708 | return true; |
c2712f64 | 709 | |
2202afc9 | 710 | for (; (p = strtok_r(ops->cgroup_use, ",", &saveptr)); ops->cgroup_use = NULL) |
ccb4cabe | 711 | if (!controller_found(hlist, p)) { |
2202afc9 | 712 | ERROR("No %s controller mountpoint found", p); |
ccb4cabe SH |
713 | return false; |
714 | } | |
c2712f64 | 715 | |
ccb4cabe SH |
716 | return true; |
717 | } | |
718 | ||
f205f10c CB |
719 | /* Get the controllers from a mountinfo line There are other ways we could get |
720 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
721 | * could parse the mount options. But we simply assume that the mountpoint must | |
722 | * be /sys/fs/cgroup/controller-list | |
ccb4cabe | 723 | */ |
a3926f6a CB |
724 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, |
725 | int type) | |
ccb4cabe | 726 | { |
f205f10c CB |
727 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list |
728 | * for legacy hierarchies. | |
729 | */ | |
ccb4cabe | 730 | int i; |
411ac6d8 | 731 | char *dup, *p2, *tok; |
d6337a5f | 732 | char *p = line, *saveptr = NULL, *sep = ","; |
411ac6d8 | 733 | char **aret = NULL; |
6328fd9c | 734 | |
ccb4cabe | 735 | for (i = 0; i < 4; i++) { |
235f1815 | 736 | p = strchr(p, ' '); |
ccb4cabe SH |
737 | if (!p) |
738 | return NULL; | |
739 | p++; | |
740 | } | |
a55f31bd | 741 | |
f205f10c CB |
742 | /* Note, if we change how mountinfo works, then our caller will need to |
743 | * verify /sys/fs/cgroup/ in this field. | |
744 | */ | |
745 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { | |
2202afc9 | 746 | ERROR("Found hierarchy not under /sys/fs/cgroup: \"%s\"", p); |
ccb4cabe | 747 | return NULL; |
5059aae9 | 748 | } |
d6337a5f | 749 | |
ccb4cabe | 750 | p += 15; |
235f1815 | 751 | p2 = strchr(p, ' '); |
ccb4cabe | 752 | if (!p2) { |
2202afc9 | 753 | ERROR("Corrupt mountinfo"); |
ccb4cabe SH |
754 | return NULL; |
755 | } | |
756 | *p2 = '\0'; | |
6328fd9c | 757 | |
d6337a5f CB |
758 | if (type == CGROUP_SUPER_MAGIC) { |
759 | /* strdup() here for v1 hierarchies. Otherwise strtok_r() will | |
760 | * destroy mountpoints such as "/sys/fs/cgroup/cpu,cpuacct". | |
761 | */ | |
762 | dup = strdup(p); | |
763 | if (!dup) | |
764 | return NULL; | |
765 | ||
766 | for (tok = strtok_r(dup, sep, &saveptr); tok; | |
767 | tok = strtok_r(NULL, sep, &saveptr)) | |
768 | must_append_controller(klist, nlist, &aret, tok); | |
769 | ||
770 | free(dup); | |
411ac6d8 | 771 | } |
d6337a5f | 772 | *p2 = ' '; |
f205f10c | 773 | |
d6337a5f CB |
774 | return aret; |
775 | } | |
411ac6d8 | 776 | |
d6337a5f CB |
777 | static char **cg_unified_make_empty_controller(void) |
778 | { | |
779 | int newentry; | |
780 | char **aret = NULL; | |
781 | ||
782 | newentry = append_null_to_list((void ***)&aret); | |
783 | aret[newentry] = NULL; | |
784 | return aret; | |
785 | } | |
786 | ||
787 | static char **cg_unified_get_controllers(const char *file) | |
788 | { | |
789 | char *buf, *tok; | |
790 | char *saveptr = NULL, *sep = " \t\n"; | |
791 | char **aret = NULL; | |
792 | ||
793 | buf = read_file(file); | |
794 | if (!buf) | |
411ac6d8 | 795 | return NULL; |
6328fd9c | 796 | |
d6337a5f CB |
797 | for (tok = strtok_r(buf, sep, &saveptr); tok; |
798 | tok = strtok_r(NULL, sep, &saveptr)) { | |
799 | int newentry; | |
800 | char *copy; | |
801 | ||
802 | newentry = append_null_to_list((void ***)&aret); | |
803 | copy = must_copy_string(tok); | |
804 | aret[newentry] = copy; | |
ccb4cabe SH |
805 | } |
806 | ||
d6337a5f | 807 | free(buf); |
ccb4cabe SH |
808 | return aret; |
809 | } | |
810 | ||
2202afc9 | 811 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, |
d6337a5f | 812 | char *base_cgroup, int type) |
ccb4cabe SH |
813 | { |
814 | struct hierarchy *new; | |
815 | int newentry; | |
816 | ||
817 | new = must_alloc(sizeof(*new)); | |
818 | new->controllers = clist; | |
819 | new->mountpoint = mountpoint; | |
820 | new->base_cgroup = base_cgroup; | |
821 | new->fullcgpath = NULL; | |
d6337a5f | 822 | new->version = type; |
6328fd9c | 823 | |
2202afc9 CB |
824 | newentry = append_null_to_list((void ***)h); |
825 | (*h)[newentry] = new; | |
d6337a5f | 826 | return new; |
ccb4cabe SH |
827 | } |
828 | ||
798c3b33 CB |
829 | /* Get a copy of the mountpoint from @line, which is a line from |
830 | * /proc/self/mountinfo. | |
ccb4cabe | 831 | */ |
a3926f6a | 832 | static char *cg_hybrid_get_mountpoint(char *line) |
ccb4cabe SH |
833 | { |
834 | int i; | |
ccb4cabe | 835 | size_t len; |
798c3b33 CB |
836 | char *p2; |
837 | char *p = line, *sret = NULL; | |
ccb4cabe SH |
838 | |
839 | for (i = 0; i < 4; i++) { | |
235f1815 | 840 | p = strchr(p, ' '); |
ccb4cabe SH |
841 | if (!p) |
842 | return NULL; | |
843 | p++; | |
844 | } | |
d6337a5f | 845 | |
798c3b33 | 846 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) |
d6337a5f CB |
847 | return NULL; |
848 | ||
849 | p2 = strchr(p + 15, ' '); | |
850 | if (!p2) | |
851 | return NULL; | |
852 | *p2 = '\0'; | |
853 | ||
ccb4cabe SH |
854 | len = strlen(p); |
855 | sret = must_alloc(len + 1); | |
856 | memcpy(sret, p, len); | |
857 | sret[len] = '\0'; | |
858 | return sret; | |
859 | } | |
860 | ||
f523291e | 861 | /* Given a multi-line string, return a null-terminated copy of the current line. */ |
ccb4cabe SH |
862 | static char *copy_to_eol(char *p) |
863 | { | |
235f1815 | 864 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
865 | size_t len; |
866 | ||
867 | if (!p2) | |
868 | return NULL; | |
869 | ||
870 | len = p2 - p; | |
871 | sret = must_alloc(len + 1); | |
872 | memcpy(sret, p, len); | |
873 | sret[len] = '\0'; | |
874 | return sret; | |
875 | } | |
876 | ||
bced39de CB |
877 | /* cgline: pointer to character after the first ':' in a line in a \n-terminated |
878 | * /proc/self/cgroup file. Check whether controller c is present. | |
ccb4cabe SH |
879 | */ |
880 | static bool controller_in_clist(char *cgline, char *c) | |
881 | { | |
882 | char *tok, *saveptr = NULL, *eol, *tmp; | |
883 | size_t len; | |
884 | ||
235f1815 | 885 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
886 | if (!eol) |
887 | return false; | |
888 | ||
889 | len = eol - cgline; | |
890 | tmp = alloca(len + 1); | |
891 | memcpy(tmp, cgline, len); | |
892 | tmp[len] = '\0'; | |
893 | ||
894 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
d6337a5f | 895 | tok = strtok_r(NULL, ",", &saveptr)) { |
ccb4cabe SH |
896 | if (strcmp(tok, c) == 0) |
897 | return true; | |
898 | } | |
d6337a5f | 899 | |
ccb4cabe SH |
900 | return false; |
901 | } | |
902 | ||
c3ef912e CB |
903 | /* @basecginfo is a copy of /proc/$$/cgroup. Return the current cgroup for |
904 | * @controller. | |
ccb4cabe | 905 | */ |
c3ef912e CB |
906 | static char *cg_hybrid_get_current_cgroup(char *basecginfo, char *controller, |
907 | int type) | |
ccb4cabe SH |
908 | { |
909 | char *p = basecginfo; | |
6328fd9c | 910 | |
d6337a5f CB |
911 | for (;;) { |
912 | bool is_cgv2_base_cgroup = false; | |
913 | ||
6328fd9c | 914 | /* cgroup v2 entry in "/proc/<pid>/cgroup": "0::/some/path" */ |
d6337a5f CB |
915 | if ((type == CGROUP2_SUPER_MAGIC) && (*p == '0')) |
916 | is_cgv2_base_cgroup = true; | |
ccb4cabe | 917 | |
235f1815 | 918 | p = strchr(p, ':'); |
ccb4cabe SH |
919 | if (!p) |
920 | return NULL; | |
921 | p++; | |
d6337a5f CB |
922 | |
923 | if (is_cgv2_base_cgroup || (controller && controller_in_clist(p, controller))) { | |
235f1815 | 924 | p = strchr(p, ':'); |
ccb4cabe SH |
925 | if (!p) |
926 | return NULL; | |
927 | p++; | |
928 | return copy_to_eol(p); | |
929 | } | |
930 | ||
235f1815 | 931 | p = strchr(p, '\n'); |
ccb4cabe SH |
932 | if (!p) |
933 | return NULL; | |
934 | p++; | |
935 | } | |
936 | } | |
937 | ||
ccb4cabe SH |
938 | static void must_append_string(char ***list, char *entry) |
939 | { | |
6dfb18bf | 940 | int newentry; |
ccb4cabe SH |
941 | char *copy; |
942 | ||
6dfb18bf | 943 | newentry = append_null_to_list((void ***)list); |
ccb4cabe SH |
944 | copy = must_copy_string(entry); |
945 | (*list)[newentry] = copy; | |
946 | } | |
947 | ||
d6337a5f | 948 | static int get_existing_subsystems(char ***klist, char ***nlist) |
ccb4cabe SH |
949 | { |
950 | FILE *f; | |
951 | char *line = NULL; | |
952 | size_t len = 0; | |
953 | ||
d6337a5f CB |
954 | f = fopen("/proc/self/cgroup", "r"); |
955 | if (!f) | |
956 | return -1; | |
957 | ||
ccb4cabe SH |
958 | while (getline(&line, &len, f) != -1) { |
959 | char *p, *p2, *tok, *saveptr = NULL; | |
235f1815 | 960 | p = strchr(line, ':'); |
ccb4cabe SH |
961 | if (!p) |
962 | continue; | |
963 | p++; | |
235f1815 | 964 | p2 = strchr(p, ':'); |
ccb4cabe SH |
965 | if (!p2) |
966 | continue; | |
967 | *p2 = '\0'; | |
ff8d6ee9 | 968 | |
6328fd9c CB |
969 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup |
970 | * contains an entry of the form: | |
ff8d6ee9 CB |
971 | * |
972 | * 0::/some/path | |
973 | * | |
6328fd9c | 974 | * In this case we use "cgroup2" as controller name. |
ff8d6ee9 | 975 | */ |
6328fd9c CB |
976 | if ((p2 - p) == 0) { |
977 | must_append_string(klist, "cgroup2"); | |
ff8d6ee9 | 978 | continue; |
6328fd9c | 979 | } |
ff8d6ee9 | 980 | |
ccb4cabe | 981 | for (tok = strtok_r(p, ",", &saveptr); tok; |
d6337a5f | 982 | tok = strtok_r(NULL, ",", &saveptr)) { |
ccb4cabe SH |
983 | if (strncmp(tok, "name=", 5) == 0) |
984 | must_append_string(nlist, tok); | |
985 | else | |
986 | must_append_string(klist, tok); | |
987 | } | |
988 | } | |
989 | ||
990 | free(line); | |
991 | fclose(f); | |
d6337a5f | 992 | return 0; |
ccb4cabe SH |
993 | } |
994 | ||
995 | static void trim(char *s) | |
996 | { | |
7689dfd7 CB |
997 | size_t len; |
998 | ||
999 | len = strlen(s); | |
2c28d76b | 1000 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
1001 | s[--len] = '\0'; |
1002 | } | |
1003 | ||
2202afc9 | 1004 | static void lxc_cgfsng_print_hierarchies(struct cgroup_ops *ops) |
ccb4cabe SH |
1005 | { |
1006 | int i; | |
27d84737 | 1007 | struct hierarchy **it; |
41c33dbe | 1008 | |
2202afc9 CB |
1009 | if (!ops->hierarchies) { |
1010 | TRACE(" No hierarchies found"); | |
ccb4cabe SH |
1011 | return; |
1012 | } | |
27d84737 | 1013 | |
2202afc9 CB |
1014 | TRACE(" Hierarchies:"); |
1015 | for (i = 0, it = ops->hierarchies; it && *it; it++, i++) { | |
ccb4cabe | 1016 | int j; |
27d84737 CB |
1017 | char **cit; |
1018 | ||
2202afc9 CB |
1019 | TRACE(" %d: base_cgroup: %s", i, (*it)->base_cgroup ? (*it)->base_cgroup : "(null)"); |
1020 | TRACE(" mountpoint: %s", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); | |
1021 | TRACE(" controllers:"); | |
a7b0cc4c | 1022 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
2202afc9 | 1023 | TRACE(" %d: %s", j, *cit); |
ccb4cabe SH |
1024 | } |
1025 | } | |
41c33dbe | 1026 | |
a3926f6a CB |
1027 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, |
1028 | char **nlist) | |
41c33dbe SH |
1029 | { |
1030 | int k; | |
a7b0cc4c | 1031 | char **it; |
41c33dbe | 1032 | |
2202afc9 CB |
1033 | TRACE("basecginfo is:"); |
1034 | TRACE("%s", basecginfo); | |
41c33dbe | 1035 | |
a7b0cc4c | 1036 | for (k = 0, it = klist; it && *it; it++, k++) |
2202afc9 | 1037 | TRACE("kernel subsystem %d: %s", k, *it); |
0f71dd9b | 1038 | |
a7b0cc4c | 1039 | for (k = 0, it = nlist; it && *it; it++, k++) |
2202afc9 | 1040 | TRACE("named subsystem %d: %s", k, *it); |
41c33dbe | 1041 | } |
ccb4cabe | 1042 | |
2202afc9 | 1043 | static int recursive_destroy(char *dirname) |
ccb4cabe | 1044 | { |
d6337a5f | 1045 | int ret; |
2202afc9 CB |
1046 | struct dirent *direntp; |
1047 | DIR *dir; | |
1048 | int r = 0; | |
ccb4cabe | 1049 | |
2202afc9 CB |
1050 | dir = opendir(dirname); |
1051 | if (!dir) | |
1052 | return -1; | |
ccb4cabe | 1053 | |
2202afc9 CB |
1054 | while ((direntp = readdir(dir))) { |
1055 | char *pathname; | |
1056 | struct stat mystat; | |
ccb4cabe | 1057 | |
2202afc9 CB |
1058 | if (!strcmp(direntp->d_name, ".") || |
1059 | !strcmp(direntp->d_name, "..")) | |
d6337a5f CB |
1060 | continue; |
1061 | ||
2202afc9 | 1062 | pathname = must_make_path(dirname, direntp->d_name, NULL); |
ccb4cabe | 1063 | |
2202afc9 CB |
1064 | ret = lstat(pathname, &mystat); |
1065 | if (ret < 0) { | |
1066 | if (!r) | |
1067 | WARN("Failed to stat \"%s\"", pathname); | |
1068 | r = -1; | |
d6337a5f | 1069 | goto next; |
ccb4cabe | 1070 | } |
6328fd9c | 1071 | |
2202afc9 | 1072 | if (!S_ISDIR(mystat.st_mode)) |
d6337a5f CB |
1073 | goto next; |
1074 | ||
2202afc9 CB |
1075 | ret = recursive_destroy(pathname); |
1076 | if (ret < 0) | |
1077 | r = -1; | |
d6337a5f | 1078 | next: |
2202afc9 | 1079 | free(pathname); |
ccb4cabe SH |
1080 | } |
1081 | ||
2202afc9 CB |
1082 | ret = rmdir(dirname); |
1083 | if (ret < 0) { | |
1084 | if (!r) | |
1085 | WARN("%s - Failed to delete \"%s\"", strerror(errno), dirname); | |
1086 | r = -1; | |
e4aeecf5 CB |
1087 | } |
1088 | ||
2202afc9 CB |
1089 | ret = closedir(dir); |
1090 | if (ret < 0) { | |
1091 | if (!r) | |
1092 | WARN("%s - Failed to delete \"%s\"", strerror(errno), dirname); | |
1093 | r = -1; | |
1094 | } | |
ccb4cabe | 1095 | |
2202afc9 | 1096 | return r; |
ccb4cabe SH |
1097 | } |
1098 | ||
2202afc9 CB |
1099 | static int cgroup_rmdir(struct hierarchy **hierarchies, |
1100 | const char *container_cgroup) | |
c71d83e1 | 1101 | { |
2202afc9 | 1102 | int i; |
d6337a5f | 1103 | |
2202afc9 CB |
1104 | if (!container_cgroup || !hierarchies) |
1105 | return 0; | |
d6337a5f | 1106 | |
2202afc9 CB |
1107 | for (i = 0; hierarchies[i]; i++) { |
1108 | int ret; | |
1109 | struct hierarchy *h = hierarchies[i]; | |
d6337a5f | 1110 | |
2202afc9 CB |
1111 | if (!h->fullcgpath) |
1112 | continue; | |
1113 | ||
1114 | ret = recursive_destroy(h->fullcgpath); | |
1115 | if (ret < 0) | |
1116 | WARN("Failed to destroy \"%s\"", h->fullcgpath); | |
1117 | ||
1118 | free(h->fullcgpath); | |
1119 | h->fullcgpath = NULL; | |
1120 | } | |
d6337a5f | 1121 | |
c71d83e1 | 1122 | return 0; |
d6337a5f CB |
1123 | } |
1124 | ||
2202afc9 CB |
1125 | struct generic_userns_exec_data { |
1126 | struct hierarchy **hierarchies; | |
1127 | const char *container_cgroup; | |
1128 | struct lxc_conf *conf; | |
1129 | uid_t origuid; /* target uid in parent namespace */ | |
1130 | char *path; | |
1131 | }; | |
d6337a5f | 1132 | |
2202afc9 CB |
1133 | static int cgroup_rmdir_wrapper(void *data) |
1134 | { | |
1135 | int ret; | |
1136 | struct generic_userns_exec_data *arg = data; | |
1137 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1138 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
d6337a5f | 1139 | |
2202afc9 CB |
1140 | ret = setresgid(nsgid, nsgid, nsgid); |
1141 | if (ret < 0) { | |
1142 | SYSERROR("Failed to setresgid(%d, %d, %d)", (int)nsgid, | |
1143 | (int)nsgid, (int)nsgid); | |
1144 | return -1; | |
1145 | } | |
d6337a5f | 1146 | |
2202afc9 CB |
1147 | ret = setresuid(nsuid, nsuid, nsuid); |
1148 | if (ret < 0) { | |
1149 | SYSERROR("Failed to setresuid(%d, %d, %d)", (int)nsuid, | |
1150 | (int)nsuid, (int)nsuid); | |
1151 | return -1; | |
1152 | } | |
d6337a5f | 1153 | |
2202afc9 CB |
1154 | ret = setgroups(0, NULL); |
1155 | if (ret < 0 && errno != EPERM) { | |
1156 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1157 | return -1; | |
1158 | } | |
d6337a5f | 1159 | |
2202afc9 | 1160 | return cgroup_rmdir(arg->hierarchies, arg->container_cgroup); |
d6337a5f CB |
1161 | } |
1162 | ||
2202afc9 | 1163 | static void cgfsng_destroy(struct cgroup_ops *ops, struct lxc_handler *handler) |
d6337a5f CB |
1164 | { |
1165 | int ret; | |
2202afc9 | 1166 | struct generic_userns_exec_data wrap; |
bd8ef4e4 | 1167 | |
4160c3a0 | 1168 | wrap.origuid = 0; |
2202afc9 CB |
1169 | wrap.container_cgroup = ops->container_cgroup; |
1170 | wrap.hierarchies = ops->hierarchies; | |
1171 | wrap.conf = handler->conf; | |
4160c3a0 | 1172 | |
2202afc9 CB |
1173 | if (handler->conf && !lxc_list_empty(&handler->conf->id_map)) |
1174 | ret = userns_exec_1(handler->conf, cgroup_rmdir_wrapper, &wrap, | |
bd8ef4e4 | 1175 | "cgroup_rmdir_wrapper"); |
ccb4cabe | 1176 | else |
2202afc9 | 1177 | ret = cgroup_rmdir(ops->hierarchies, ops->container_cgroup); |
bd8ef4e4 CB |
1178 | if (ret < 0) { |
1179 | WARN("Failed to destroy cgroups"); | |
ccb4cabe | 1180 | return; |
ccb4cabe | 1181 | } |
ccb4cabe SH |
1182 | } |
1183 | ||
a3926f6a | 1184 | static bool cg_unified_create_cgroup(struct hierarchy *h, char *cgname) |
0c3deb94 | 1185 | { |
0c3deb94 | 1186 | size_t i, parts_len; |
389d44ec | 1187 | char **it; |
0c3deb94 CB |
1188 | size_t full_len = 0; |
1189 | char *add_controllers = NULL, *cgroup = NULL; | |
1190 | char **parts = NULL; | |
1191 | bool bret = false; | |
1192 | ||
1193 | if (h->version != CGROUP2_SUPER_MAGIC) | |
1194 | return true; | |
1195 | ||
1196 | if (!h->controllers) | |
1197 | return true; | |
1198 | ||
1199 | /* For now we simply enable all controllers that we have detected by | |
1200 | * creating a string like "+memory +pids +cpu +io". | |
1201 | * TODO: In the near future we might want to support "-<controller>" | |
1202 | * etc. but whether supporting semantics like this make sense will need | |
1203 | * some thinking. | |
1204 | */ | |
1205 | for (it = h->controllers; it && *it; it++) { | |
64e82f8b DJ |
1206 | full_len += strlen(*it) + 2; |
1207 | add_controllers = must_realloc(add_controllers, full_len + 1); | |
1208 | ||
1209 | if (h->controllers[0] == *it) | |
1210 | add_controllers[0] = '\0'; | |
1211 | ||
3ebe2fbd DJ |
1212 | (void)strlcat(add_controllers, "+", full_len + 1); |
1213 | (void)strlcat(add_controllers, *it, full_len + 1); | |
64e82f8b DJ |
1214 | |
1215 | if ((it + 1) && *(it + 1)) | |
3ebe2fbd | 1216 | (void)strlcat(add_controllers, " ", full_len + 1); |
0c3deb94 CB |
1217 | } |
1218 | ||
1219 | parts = lxc_string_split(cgname, '/'); | |
1220 | if (!parts) | |
1221 | goto on_error; | |
64e82f8b | 1222 | |
0c3deb94 CB |
1223 | parts_len = lxc_array_len((void **)parts); |
1224 | if (parts_len > 0) | |
1225 | parts_len--; | |
1226 | ||
1227 | cgroup = must_make_path(h->mountpoint, h->base_cgroup, NULL); | |
1228 | for (i = 0; i < parts_len; i++) { | |
1229 | int ret; | |
1230 | char *target; | |
1231 | ||
1232 | cgroup = must_append_path(cgroup, parts[i], NULL); | |
1233 | target = must_make_path(cgroup, "cgroup.subtree_control", NULL); | |
7cea5905 | 1234 | ret = lxc_write_to_file(target, add_controllers, full_len, false, 0666); |
0c3deb94 CB |
1235 | free(target); |
1236 | if (ret < 0) { | |
1237 | SYSERROR("Could not enable \"%s\" controllers in the " | |
1238 | "unified cgroup \"%s\"", add_controllers, cgroup); | |
1239 | goto on_error; | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | bret = true; | |
1244 | ||
1245 | on_error: | |
1246 | lxc_free_array((void **)parts, free); | |
1247 | free(add_controllers); | |
1248 | free(cgroup); | |
1249 | return bret; | |
1250 | } | |
1251 | ||
ccb4cabe SH |
1252 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) |
1253 | { | |
0c3deb94 CB |
1254 | int ret; |
1255 | ||
e3a3fecf | 1256 | h->fullcgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); |
4b4205e3 CB |
1257 | if (dir_exists(h->fullcgpath)) { |
1258 | ERROR("The cgroup \"%s\" already existed", h->fullcgpath); | |
d8da679e | 1259 | return false; |
6f9584d8 | 1260 | } |
0c3deb94 | 1261 | |
a3926f6a | 1262 | if (!cg_legacy_handle_cpuset_hierarchy(h, cgname)) { |
4b4205e3 | 1263 | ERROR("Failed to handle legacy cpuset controller"); |
0c3deb94 CB |
1264 | return false; |
1265 | } | |
1266 | ||
1267 | ret = mkdir_p(h->fullcgpath, 0755); | |
1268 | if (ret < 0) { | |
1269 | ERROR("Failed to create cgroup \"%s\"", h->fullcgpath); | |
e3a3fecf | 1270 | return false; |
6f9584d8 | 1271 | } |
0c3deb94 | 1272 | |
a3926f6a | 1273 | return cg_unified_create_cgroup(h, cgname); |
ccb4cabe SH |
1274 | } |
1275 | ||
1276 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1277 | { | |
e56639fb CB |
1278 | int ret; |
1279 | ||
1280 | ret = rmdir(h->fullcgpath); | |
1281 | if (ret < 0) | |
1282 | SYSERROR("Failed to rmdir(\"%s\") from failed creation attempt", h->fullcgpath); | |
1283 | ||
ccb4cabe SH |
1284 | free(h->fullcgpath); |
1285 | h->fullcgpath = NULL; | |
1286 | } | |
1287 | ||
cecad0c1 CB |
1288 | /* Try to create the same cgroup in all hierarchies. Start with cgroup_pattern; |
1289 | * next cgroup_pattern-1, -2, ..., -999. | |
ccb4cabe | 1290 | */ |
2202afc9 CB |
1291 | static inline bool cgfsng_create(struct cgroup_ops *ops, |
1292 | struct lxc_handler *handler) | |
ccb4cabe | 1293 | { |
bb30b52a | 1294 | int i; |
ccb4cabe | 1295 | size_t len; |
0c3deb94 | 1296 | char *container_cgroup, *offset, *tmp; |
7d531e9b | 1297 | int idx = 0; |
2202afc9 | 1298 | struct lxc_conf *conf = handler->conf; |
ccb4cabe | 1299 | |
2202afc9 CB |
1300 | if (ops->container_cgroup) { |
1301 | WARN("cgfsng_create called a second time: %s", ops->container_cgroup); | |
ccb4cabe | 1302 | return false; |
2202afc9 | 1303 | } |
43654d34 | 1304 | |
2202afc9 | 1305 | if (!conf) |
ccb4cabe | 1306 | return false; |
ccb4cabe | 1307 | |
2202afc9 | 1308 | if (conf->cgroup_meta.dir) |
3ec12d39 | 1309 | tmp = lxc_string_join("/", (const char *[]){conf->cgroup_meta.dir, handler->name, NULL}, false); |
43654d34 | 1310 | else |
2202afc9 | 1311 | tmp = lxc_string_replace("%n", handler->name, ops->cgroup_pattern); |
ccb4cabe SH |
1312 | if (!tmp) { |
1313 | ERROR("Failed expanding cgroup name pattern"); | |
1314 | return false; | |
1315 | } | |
64e82f8b | 1316 | |
1a0e70ac | 1317 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
0c3deb94 | 1318 | container_cgroup = must_alloc(len); |
64e82f8b | 1319 | (void)strlcpy(container_cgroup, tmp, len); |
ccb4cabe | 1320 | free(tmp); |
0c3deb94 | 1321 | offset = container_cgroup + len - 5; |
ccb4cabe SH |
1322 | |
1323 | again: | |
95adfe93 SH |
1324 | if (idx == 1000) { |
1325 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1326 | goto out_free; |
95adfe93 | 1327 | } |
cecad0c1 | 1328 | |
66b66624 | 1329 | if (idx) { |
bb30b52a CB |
1330 | int ret; |
1331 | ||
66b66624 CB |
1332 | ret = snprintf(offset, 5, "-%d", idx); |
1333 | if (ret < 0 || (size_t)ret >= 5) { | |
1334 | FILE *f = fopen("/dev/null", "w"); | |
97ebced3 | 1335 | if (f) { |
66b66624 CB |
1336 | fprintf(f, "Workaround for GCC7 bug: " |
1337 | "https://gcc.gnu.org/bugzilla/" | |
1338 | "show_bug.cgi?id=78969"); | |
1339 | fclose(f); | |
1340 | } | |
1341 | } | |
1342 | } | |
cecad0c1 | 1343 | |
2202afc9 CB |
1344 | for (i = 0; ops->hierarchies[i]; i++) { |
1345 | if (!create_path_for_hierarchy(ops->hierarchies[i], container_cgroup)) { | |
ccb4cabe | 1346 | int j; |
2202afc9 CB |
1347 | ERROR("Failed to create cgroup \"%s\"", ops->hierarchies[i]->fullcgpath); |
1348 | free(ops->hierarchies[i]->fullcgpath); | |
1349 | ops->hierarchies[i]->fullcgpath = NULL; | |
ccb4cabe | 1350 | for (j = 0; j < i; j++) |
2202afc9 | 1351 | remove_path_for_hierarchy(ops->hierarchies[j], container_cgroup); |
ccb4cabe SH |
1352 | idx++; |
1353 | goto again; | |
1354 | } | |
1355 | } | |
cecad0c1 | 1356 | |
2202afc9 | 1357 | ops->container_cgroup = container_cgroup; |
cecad0c1 | 1358 | |
ccb4cabe SH |
1359 | return true; |
1360 | ||
1361 | out_free: | |
0c3deb94 | 1362 | free(container_cgroup); |
cecad0c1 | 1363 | |
ccb4cabe SH |
1364 | return false; |
1365 | } | |
1366 | ||
2202afc9 | 1367 | static bool cgfsng_enter(struct cgroup_ops *ops, pid_t pid) |
ccb4cabe | 1368 | { |
ccb4cabe | 1369 | int i, len; |
08768001 | 1370 | char pidstr[25]; |
ccb4cabe SH |
1371 | |
1372 | len = snprintf(pidstr, 25, "%d", pid); | |
08768001 | 1373 | if (len < 0 || len >= 25) |
ccb4cabe SH |
1374 | return false; |
1375 | ||
2202afc9 | 1376 | for (i = 0; ops->hierarchies[i]; i++) { |
08768001 CB |
1377 | int ret; |
1378 | char *fullpath; | |
1379 | ||
2202afc9 | 1380 | fullpath = must_make_path(ops->hierarchies[i]->fullcgpath, |
08768001 | 1381 | "cgroup.procs", NULL); |
7cea5905 | 1382 | ret = lxc_write_to_file(fullpath, pidstr, len, false, 0666); |
08768001 CB |
1383 | if (ret != 0) { |
1384 | SYSERROR("Failed to enter cgroup \"%s\"", fullpath); | |
ccb4cabe SH |
1385 | free(fullpath); |
1386 | return false; | |
1387 | } | |
1388 | free(fullpath); | |
1389 | } | |
1390 | ||
1391 | return true; | |
1392 | } | |
1393 | ||
6efacf80 CB |
1394 | static int chowmod(char *path, uid_t chown_uid, gid_t chown_gid, |
1395 | mode_t chmod_mode) | |
1396 | { | |
1397 | int ret; | |
1398 | ||
1399 | ret = chown(path, chown_uid, chown_gid); | |
1400 | if (ret < 0) { | |
1401 | WARN("%s - Failed to chown(%s, %d, %d)", strerror(errno), path, | |
1402 | (int)chown_uid, (int)chown_gid); | |
1403 | return -1; | |
1404 | } | |
1405 | ||
1406 | ret = chmod(path, chmod_mode); | |
1407 | if (ret < 0) { | |
1408 | WARN("%s - Failed to chmod(%s, %d)", strerror(errno), path, | |
1409 | (int)chmod_mode); | |
1410 | return -1; | |
1411 | } | |
1412 | ||
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | /* chgrp the container cgroups to container group. We leave | |
c0888dfe SH |
1417 | * the container owner as cgroup owner. So we must make the |
1418 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1419 | * |
1420 | * Also chown the tasks and cgroup.procs files. Those may not | |
1421 | * exist depending on kernel version. | |
c0888dfe | 1422 | */ |
ccb4cabe SH |
1423 | static int chown_cgroup_wrapper(void *data) |
1424 | { | |
6efacf80 | 1425 | int i, ret; |
4160c3a0 CB |
1426 | uid_t destuid; |
1427 | struct generic_userns_exec_data *arg = data; | |
1428 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1429 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
ccb4cabe | 1430 | |
6efacf80 CB |
1431 | ret = setresgid(nsgid, nsgid, nsgid); |
1432 | if (ret < 0) { | |
1433 | SYSERROR("Failed to setresgid(%d, %d, %d)", | |
1434 | (int)nsgid, (int)nsgid, (int)nsgid); | |
1435 | return -1; | |
1436 | } | |
1437 | ||
1438 | ret = setresuid(nsuid, nsuid, nsuid); | |
1439 | if (ret < 0) { | |
1440 | SYSERROR("Failed to setresuid(%d, %d, %d)", | |
1441 | (int)nsuid, (int)nsuid, (int)nsuid); | |
1442 | return -1; | |
1443 | } | |
1444 | ||
1445 | ret = setgroups(0, NULL); | |
1446 | if (ret < 0 && errno != EPERM) { | |
1447 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1448 | return -1; | |
1449 | } | |
ccb4cabe SH |
1450 | |
1451 | destuid = get_ns_uid(arg->origuid); | |
1452 | ||
2202afc9 | 1453 | for (i = 0; arg->hierarchies[i]; i++) { |
6efacf80 | 1454 | char *fullpath; |
2202afc9 | 1455 | char *path = arg->hierarchies[i]->fullcgpath; |
43647298 | 1456 | |
63e42fee | 1457 | ret = chowmod(path, destuid, nsgid, 0775); |
6efacf80 | 1458 | if (ret < 0) |
ccb4cabe | 1459 | return -1; |
c0888dfe | 1460 | |
6efacf80 CB |
1461 | /* Failures to chown() these are inconvenient but not |
1462 | * detrimental We leave these owned by the container launcher, | |
1463 | * so that container root can write to the files to attach. We | |
1464 | * chmod() them 664 so that container systemd can write to the | |
1465 | * files (which systemd in wily insists on doing). | |
ab8f5424 | 1466 | */ |
6efacf80 | 1467 | |
2202afc9 | 1468 | if (arg->hierarchies[i]->version == CGROUP_SUPER_MAGIC) { |
6efacf80 CB |
1469 | fullpath = must_make_path(path, "tasks", NULL); |
1470 | (void)chowmod(fullpath, destuid, nsgid, 0664); | |
1471 | free(fullpath); | |
1472 | } | |
43647298 SH |
1473 | |
1474 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
2202afc9 | 1475 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
ccb4cabe | 1476 | free(fullpath); |
0e17357c | 1477 | |
2202afc9 | 1478 | if (arg->hierarchies[i]->version != CGROUP2_SUPER_MAGIC) |
0e17357c CB |
1479 | continue; |
1480 | ||
1481 | fullpath = must_make_path(path, "cgroup.subtree_control", NULL); | |
6efacf80 | 1482 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c CB |
1483 | free(fullpath); |
1484 | ||
1485 | fullpath = must_make_path(path, "cgroup.threads", NULL); | |
6efacf80 | 1486 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c | 1487 | free(fullpath); |
ccb4cabe SH |
1488 | } |
1489 | ||
1490 | return 0; | |
1491 | } | |
1492 | ||
2202afc9 | 1493 | static bool cgfsng_chown(struct cgroup_ops *ops, struct lxc_conf *conf) |
ccb4cabe | 1494 | { |
4160c3a0 | 1495 | struct generic_userns_exec_data wrap; |
ccb4cabe | 1496 | |
ccb4cabe SH |
1497 | if (lxc_list_empty(&conf->id_map)) |
1498 | return true; | |
1499 | ||
ccb4cabe | 1500 | wrap.origuid = geteuid(); |
4160c3a0 | 1501 | wrap.path = NULL; |
2202afc9 | 1502 | wrap.hierarchies = ops->hierarchies; |
4160c3a0 | 1503 | wrap.conf = conf; |
ccb4cabe | 1504 | |
c9b7c33e CB |
1505 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1506 | "chown_cgroup_wrapper") < 0) { | |
f7faba6c | 1507 | ERROR("Error requesting cgroup chown in new user namespace"); |
ccb4cabe SH |
1508 | return false; |
1509 | } | |
1510 | ||
1511 | return true; | |
1512 | } | |
1513 | ||
8aa1044f SH |
1514 | /* cgroup-full:* is done, no need to create subdirs */ |
1515 | static bool cg_mount_needs_subdirs(int type) | |
1516 | { | |
1517 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1518 | return false; | |
a3926f6a | 1519 | |
8aa1044f SH |
1520 | return true; |
1521 | } | |
1522 | ||
886cac86 CB |
1523 | /* After $rootfs/sys/fs/container/controller/the/cg/path has been created, |
1524 | * remount controller ro if needed and bindmount the cgroupfs onto | |
1525 | * controll/the/cg/path. | |
8aa1044f | 1526 | */ |
6812d833 CB |
1527 | static int cg_legacy_mount_controllers(int type, struct hierarchy *h, |
1528 | char *controllerpath, char *cgpath, | |
1529 | const char *container_cgroup) | |
8aa1044f | 1530 | { |
5285689c | 1531 | int ret, remount_flags; |
886cac86 CB |
1532 | char *sourcepath; |
1533 | int flags = MS_BIND; | |
1534 | ||
8aa1044f | 1535 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { |
886cac86 CB |
1536 | ret = mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL); |
1537 | if (ret < 0) { | |
1538 | SYSERROR("Failed to bind mount \"%s\" onto \"%s\"", | |
1539 | controllerpath, controllerpath); | |
8aa1044f SH |
1540 | return -1; |
1541 | } | |
886cac86 | 1542 | |
5285689c CB |
1543 | remount_flags = add_required_remount_flags(controllerpath, |
1544 | controllerpath, | |
1545 | flags | MS_REMOUNT); | |
886cac86 | 1546 | ret = mount(controllerpath, controllerpath, "cgroup", |
8186c5c7 CB |
1547 | remount_flags | MS_REMOUNT | MS_BIND | MS_RDONLY, |
1548 | NULL); | |
886cac86 CB |
1549 | if (ret < 0) { |
1550 | SYSERROR("Failed to remount \"%s\" ro", controllerpath); | |
8aa1044f SH |
1551 | return -1; |
1552 | } | |
886cac86 | 1553 | |
8aa1044f SH |
1554 | INFO("Remounted %s read-only", controllerpath); |
1555 | } | |
886cac86 CB |
1556 | |
1557 | sourcepath = must_make_path(h->mountpoint, h->base_cgroup, | |
1558 | container_cgroup, NULL); | |
8aa1044f SH |
1559 | if (type == LXC_AUTO_CGROUP_RO) |
1560 | flags |= MS_RDONLY; | |
886cac86 CB |
1561 | |
1562 | ret = mount(sourcepath, cgpath, "cgroup", flags, NULL); | |
1563 | if (ret < 0) { | |
1564 | SYSERROR("Failed to mount \"%s\" onto \"%s\"", h->controllers[0], cgpath); | |
8aa1044f | 1565 | free(sourcepath); |
8aa1044f SH |
1566 | return -1; |
1567 | } | |
886cac86 | 1568 | INFO("Mounted \"%s\" onto \"%s\"", h->controllers[0], cgpath); |
f8c40ffa L |
1569 | |
1570 | if (flags & MS_RDONLY) { | |
5285689c CB |
1571 | remount_flags = add_required_remount_flags(sourcepath, cgpath, |
1572 | flags | MS_REMOUNT); | |
1573 | ret = mount(sourcepath, cgpath, "cgroup", remount_flags, NULL); | |
886cac86 CB |
1574 | if (ret < 0) { |
1575 | SYSERROR("Failed to remount \"%s\" ro", cgpath); | |
f8c40ffa | 1576 | free(sourcepath); |
f8c40ffa L |
1577 | return -1; |
1578 | } | |
5285689c | 1579 | INFO("Remounted %s read-only", cgpath); |
f8c40ffa L |
1580 | } |
1581 | ||
8aa1044f | 1582 | free(sourcepath); |
886cac86 | 1583 | INFO("Completed second stage cgroup automounts for \"%s\"", cgpath); |
8aa1044f SH |
1584 | return 0; |
1585 | } | |
1586 | ||
6812d833 CB |
1587 | /* __cg_mount_direct |
1588 | * | |
1589 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
1590 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
1591 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
1592 | */ | |
1593 | static int __cg_mount_direct(int type, struct hierarchy *h, | |
1594 | const char *controllerpath) | |
b635e92d CB |
1595 | { |
1596 | int ret; | |
1597 | char *controllers = NULL; | |
a760603e CB |
1598 | char *fstype = "cgroup2"; |
1599 | unsigned long flags = 0; | |
b635e92d | 1600 | |
a760603e CB |
1601 | flags |= MS_NOSUID; |
1602 | flags |= MS_NOEXEC; | |
1603 | flags |= MS_NODEV; | |
1604 | flags |= MS_RELATIME; | |
1605 | ||
1606 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_FULL_RO) | |
1607 | flags |= MS_RDONLY; | |
1608 | ||
d6337a5f | 1609 | if (h->version != CGROUP2_SUPER_MAGIC) { |
a760603e CB |
1610 | controllers = lxc_string_join(",", (const char **)h->controllers, false); |
1611 | if (!controllers) | |
1612 | return -ENOMEM; | |
1613 | fstype = "cgroup"; | |
b635e92d CB |
1614 | } |
1615 | ||
a760603e | 1616 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); |
b635e92d CB |
1617 | free(controllers); |
1618 | if (ret < 0) { | |
6812d833 | 1619 | SYSERROR("Failed to mount \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1620 | return -1; |
1621 | } | |
1622 | ||
6812d833 | 1623 | DEBUG("Mounted \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1624 | return 0; |
1625 | } | |
1626 | ||
6812d833 CB |
1627 | static inline int cg_mount_in_cgroup_namespace(int type, struct hierarchy *h, |
1628 | const char *controllerpath) | |
1629 | { | |
1630 | return __cg_mount_direct(type, h, controllerpath); | |
1631 | } | |
1632 | ||
1633 | static inline int cg_mount_cgroup_full(int type, struct hierarchy *h, | |
1634 | const char *controllerpath) | |
1635 | { | |
1636 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1637 | return 0; | |
1638 | ||
1639 | return __cg_mount_direct(type, h, controllerpath); | |
1640 | } | |
1641 | ||
2202afc9 CB |
1642 | static bool cgfsng_mount(struct cgroup_ops *ops, struct lxc_handler *handler, |
1643 | const char *root, int type) | |
ccb4cabe | 1644 | { |
3f69fb12 | 1645 | int i, ret; |
8aa1044f | 1646 | char *tmpfspath = NULL; |
affd10fa | 1647 | bool has_cgns = false, retval = false, wants_force_mount = false; |
8aa1044f SH |
1648 | |
1649 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1650 | return true; | |
1651 | ||
3f69fb12 SY |
1652 | if (type & LXC_AUTO_CGROUP_FORCE) { |
1653 | type &= ~LXC_AUTO_CGROUP_FORCE; | |
1654 | wants_force_mount = true; | |
1655 | } | |
b635e92d | 1656 | |
3f69fb12 SY |
1657 | if (!wants_force_mount){ |
1658 | if (!lxc_list_empty(&handler->conf->keepcaps)) | |
1659 | wants_force_mount = !in_caplist(CAP_SYS_ADMIN, &handler->conf->keepcaps); | |
1660 | else | |
1661 | wants_force_mount = in_caplist(CAP_SYS_ADMIN, &handler->conf->caps); | |
1662 | } | |
8aa1044f | 1663 | |
3f69fb12 SY |
1664 | has_cgns = cgns_supported(); |
1665 | if (has_cgns && !wants_force_mount) | |
1666 | return true; | |
8aa1044f SH |
1667 | |
1668 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1669 | type = LXC_AUTO_CGROUP_MIXED; | |
1670 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1671 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1672 | ||
1673 | /* Mount tmpfs */ | |
3f69fb12 | 1674 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); |
6812d833 | 1675 | ret = safe_mount(NULL, tmpfspath, "tmpfs", |
3f69fb12 SY |
1676 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, |
1677 | "size=10240k,mode=755", root); | |
1678 | if (ret < 0) | |
1679 | goto on_error; | |
8aa1044f | 1680 | |
2202afc9 | 1681 | for (i = 0; ops->hierarchies[i]; i++) { |
8aa1044f | 1682 | char *controllerpath, *path2; |
2202afc9 | 1683 | struct hierarchy *h = ops->hierarchies[i]; |
8aa1044f | 1684 | char *controller = strrchr(h->mountpoint, '/'); |
8aa1044f SH |
1685 | |
1686 | if (!controller) | |
1687 | continue; | |
1688 | controller++; | |
affd10fa | 1689 | |
8aa1044f SH |
1690 | controllerpath = must_make_path(tmpfspath, controller, NULL); |
1691 | if (dir_exists(controllerpath)) { | |
1692 | free(controllerpath); | |
1693 | continue; | |
1694 | } | |
affd10fa | 1695 | |
3f69fb12 SY |
1696 | ret = mkdir(controllerpath, 0755); |
1697 | if (ret < 0) { | |
8aa1044f SH |
1698 | SYSERROR("Error creating cgroup path: %s", controllerpath); |
1699 | free(controllerpath); | |
3f69fb12 | 1700 | goto on_error; |
8aa1044f | 1701 | } |
b635e92d | 1702 | |
3f69fb12 | 1703 | if (has_cgns && wants_force_mount) { |
b635e92d CB |
1704 | /* If cgroup namespaces are supported but the container |
1705 | * will not have CAP_SYS_ADMIN after it has started we | |
1706 | * need to mount the cgroups manually. | |
1707 | */ | |
3f69fb12 | 1708 | ret = cg_mount_in_cgroup_namespace(type, h, controllerpath); |
b635e92d | 1709 | free(controllerpath); |
3f69fb12 SY |
1710 | if (ret < 0) |
1711 | goto on_error; | |
1712 | ||
b635e92d CB |
1713 | continue; |
1714 | } | |
1715 | ||
6812d833 | 1716 | ret = cg_mount_cgroup_full(type, h, controllerpath); |
3f69fb12 | 1717 | if (ret < 0) { |
8aa1044f | 1718 | free(controllerpath); |
3f69fb12 | 1719 | goto on_error; |
8aa1044f | 1720 | } |
3f69fb12 | 1721 | |
8aa1044f SH |
1722 | if (!cg_mount_needs_subdirs(type)) { |
1723 | free(controllerpath); | |
1724 | continue; | |
1725 | } | |
3f69fb12 SY |
1726 | |
1727 | path2 = must_make_path(controllerpath, h->base_cgroup, | |
2202afc9 | 1728 | ops->container_cgroup, NULL); |
3f69fb12 SY |
1729 | ret = mkdir_p(path2, 0755); |
1730 | if (ret < 0) { | |
8aa1044f | 1731 | free(controllerpath); |
8e0c6620 | 1732 | free(path2); |
3f69fb12 | 1733 | goto on_error; |
8aa1044f | 1734 | } |
2f62fb00 | 1735 | |
6812d833 | 1736 | ret = cg_legacy_mount_controllers(type, h, controllerpath, |
2202afc9 | 1737 | path2, ops->container_cgroup); |
8aa1044f SH |
1738 | free(controllerpath); |
1739 | free(path2); | |
3f69fb12 SY |
1740 | if (ret < 0) |
1741 | goto on_error; | |
8aa1044f SH |
1742 | } |
1743 | retval = true; | |
1744 | ||
3f69fb12 | 1745 | on_error: |
8aa1044f SH |
1746 | free(tmpfspath); |
1747 | return retval; | |
ccb4cabe SH |
1748 | } |
1749 | ||
1750 | static int recursive_count_nrtasks(char *dirname) | |
1751 | { | |
74f96976 | 1752 | struct dirent *direntp; |
ccb4cabe SH |
1753 | DIR *dir; |
1754 | int count = 0, ret; | |
1755 | char *path; | |
1756 | ||
1757 | dir = opendir(dirname); | |
1758 | if (!dir) | |
1759 | return 0; | |
1760 | ||
74f96976 | 1761 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1762 | struct stat mystat; |
1763 | ||
ccb4cabe SH |
1764 | if (!strcmp(direntp->d_name, ".") || |
1765 | !strcmp(direntp->d_name, "..")) | |
1766 | continue; | |
1767 | ||
1768 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1769 | ||
1770 | if (lstat(path, &mystat)) | |
1771 | goto next; | |
1772 | ||
1773 | if (!S_ISDIR(mystat.st_mode)) | |
1774 | goto next; | |
1775 | ||
1776 | count += recursive_count_nrtasks(path); | |
13c49955 | 1777 | next: |
ccb4cabe SH |
1778 | free(path); |
1779 | } | |
1780 | ||
1781 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1782 | ret = lxc_count_file_lines(path); | |
1783 | if (ret != -1) | |
1784 | count += ret; | |
1785 | free(path); | |
1786 | ||
13c49955 | 1787 | (void)closedir(dir); |
ccb4cabe SH |
1788 | |
1789 | return count; | |
1790 | } | |
1791 | ||
2202afc9 | 1792 | static int cgfsng_nrtasks(struct cgroup_ops *ops) |
3135c5d4 | 1793 | { |
ccb4cabe | 1794 | int count; |
3135c5d4 | 1795 | char *path; |
ccb4cabe | 1796 | |
2202afc9 | 1797 | if (!ops->container_cgroup || !ops->hierarchies) |
ccb4cabe | 1798 | return -1; |
a3926f6a | 1799 | |
2202afc9 | 1800 | path = must_make_path(ops->hierarchies[0]->fullcgpath, NULL); |
ccb4cabe SH |
1801 | count = recursive_count_nrtasks(path); |
1802 | free(path); | |
1803 | return count; | |
1804 | } | |
1805 | ||
11c23867 | 1806 | /* Only root needs to escape to the cgroup of its init. */ |
2202afc9 | 1807 | static bool cgfsng_escape(const struct cgroup_ops *ops) |
ccb4cabe | 1808 | { |
ccb4cabe SH |
1809 | int i; |
1810 | ||
1811 | if (geteuid()) | |
1812 | return true; | |
1813 | ||
2202afc9 | 1814 | for (i = 0; ops->hierarchies[i]; i++) { |
11c23867 CB |
1815 | int ret; |
1816 | char *fullpath; | |
1817 | ||
2202afc9 CB |
1818 | fullpath = must_make_path(ops->hierarchies[i]->mountpoint, |
1819 | ops->hierarchies[i]->base_cgroup, | |
11c23867 | 1820 | "cgroup.procs", NULL); |
7cea5905 | 1821 | ret = lxc_write_to_file(fullpath, "0", 2, false, 0666); |
11c23867 CB |
1822 | if (ret != 0) { |
1823 | SYSERROR("Failed to escape to cgroup \"%s\"", fullpath); | |
ccb4cabe | 1824 | free(fullpath); |
6df334d1 | 1825 | return false; |
ccb4cabe SH |
1826 | } |
1827 | free(fullpath); | |
1828 | } | |
1829 | ||
6df334d1 | 1830 | return true; |
ccb4cabe SH |
1831 | } |
1832 | ||
2202afc9 | 1833 | static int cgfsng_num_hierarchies(struct cgroup_ops *ops) |
36662416 TA |
1834 | { |
1835 | int i; | |
1836 | ||
2202afc9 | 1837 | for (i = 0; ops->hierarchies[i]; i++) |
36662416 TA |
1838 | ; |
1839 | ||
1840 | return i; | |
1841 | } | |
1842 | ||
2202afc9 | 1843 | static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) |
36662416 TA |
1844 | { |
1845 | int i; | |
1846 | ||
1847 | /* sanity check n */ | |
6b38e644 | 1848 | for (i = 0; i < n; i++) |
2202afc9 | 1849 | if (!ops->hierarchies[i]) |
36662416 | 1850 | return false; |
36662416 | 1851 | |
2202afc9 | 1852 | *out = ops->hierarchies[i]->controllers; |
36662416 TA |
1853 | |
1854 | return true; | |
1855 | } | |
1856 | ||
ccb4cabe SH |
1857 | #define THAWED "THAWED" |
1858 | #define THAWED_LEN (strlen(THAWED)) | |
1859 | ||
d6337a5f CB |
1860 | /* TODO: If the unified cgroup hierarchy grows a freezer controller this needs |
1861 | * to be adapted. | |
1862 | */ | |
2202afc9 | 1863 | static bool cgfsng_unfreeze(struct cgroup_ops *ops) |
ccb4cabe | 1864 | { |
d6337a5f | 1865 | int ret; |
ccb4cabe | 1866 | char *fullpath; |
d6337a5f | 1867 | struct hierarchy *h; |
ccb4cabe | 1868 | |
2202afc9 | 1869 | h = get_hierarchy(ops, "freezer"); |
457ca9aa | 1870 | if (!h) |
ccb4cabe | 1871 | return false; |
d6337a5f | 1872 | |
ccb4cabe | 1873 | fullpath = must_make_path(h->fullcgpath, "freezer.state", NULL); |
7cea5905 | 1874 | ret = lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false, 0666); |
ccb4cabe | 1875 | free(fullpath); |
d6337a5f CB |
1876 | if (ret < 0) |
1877 | return false; | |
1878 | ||
ccb4cabe SH |
1879 | return true; |
1880 | } | |
1881 | ||
2202afc9 CB |
1882 | static const char *cgfsng_get_cgroup(struct cgroup_ops *ops, |
1883 | const char *controller) | |
ccb4cabe | 1884 | { |
d6337a5f CB |
1885 | struct hierarchy *h; |
1886 | ||
2202afc9 | 1887 | h = get_hierarchy(ops, controller); |
106f1f38 | 1888 | if (!h) { |
2202afc9 CB |
1889 | WARN("Failed to find hierarchy for controller \"%s\"", |
1890 | controller ? controller : "(null)"); | |
ccb4cabe | 1891 | return NULL; |
106f1f38 | 1892 | } |
ccb4cabe | 1893 | |
371f834d SH |
1894 | return h->fullcgpath ? h->fullcgpath + strlen(h->mountpoint) : NULL; |
1895 | } | |
1896 | ||
c40c8209 CB |
1897 | /* Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a full path, |
1898 | * which must be freed by the caller. | |
371f834d | 1899 | */ |
c40c8209 CB |
1900 | static inline char *build_full_cgpath_from_monitorpath(struct hierarchy *h, |
1901 | const char *inpath, | |
1902 | const char *filename) | |
371f834d | 1903 | { |
371f834d | 1904 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
1905 | } |
1906 | ||
25f66a8f CB |
1907 | /* Technically, we're always at a delegation boundary here (This is especially |
1908 | * true when cgroup namespaces are available.). The reasoning is that in order | |
c2aed66d | 1909 | * for us to have been able to start a container in the first place the root |
25f66a8f | 1910 | * cgroup must have been a leaf node. Now, either the container's init system |
c2aed66d CB |
1911 | * has populated the cgroup and kept it as a leaf node or it has created |
1912 | * subtrees. In the former case we will simply attach to the leaf node we | |
1913 | * created when we started the container in the latter case we create our own | |
1914 | * cgroup for the attaching process. | |
1915 | */ | |
a3926f6a CB |
1916 | static int __cg_unified_attach(const struct hierarchy *h, const char *name, |
1917 | const char *lxcpath, const char *pidstr, | |
1918 | size_t pidstr_len, const char *controller) | |
c2aed66d CB |
1919 | { |
1920 | int ret; | |
1921 | size_t len; | |
1922 | int fret = -1, idx = 0; | |
1923 | char *base_path = NULL, *container_cgroup = NULL, *full_path = NULL; | |
1924 | ||
1925 | container_cgroup = lxc_cmd_get_cgroup_path(name, lxcpath, controller); | |
1926 | /* not running */ | |
1927 | if (!container_cgroup) | |
1928 | return 0; | |
1929 | ||
1930 | base_path = must_make_path(h->mountpoint, container_cgroup, NULL); | |
1931 | full_path = must_make_path(base_path, "cgroup.procs", NULL); | |
1932 | /* cgroup is populated */ | |
7cea5905 | 1933 | ret = lxc_write_to_file(full_path, pidstr, pidstr_len, false, 0666); |
c2aed66d CB |
1934 | if (ret < 0 && errno != EBUSY) |
1935 | goto on_error; | |
1936 | ||
1937 | if (ret == 0) | |
1938 | goto on_success; | |
1939 | ||
1940 | free(full_path); | |
1941 | ||
1942 | len = strlen(base_path) + sizeof("/lxc-1000") - 1 + | |
1943 | sizeof("/cgroup-procs") - 1; | |
1944 | full_path = must_alloc(len + 1); | |
1945 | do { | |
1946 | if (idx) | |
1947 | ret = snprintf(full_path, len + 1, "%s/lxc-%d", | |
1948 | base_path, idx); | |
1949 | else | |
1950 | ret = snprintf(full_path, len + 1, "%s/lxc", base_path); | |
1951 | if (ret < 0 || (size_t)ret >= len + 1) | |
1952 | goto on_error; | |
1953 | ||
1954 | ret = mkdir_p(full_path, 0755); | |
1955 | if (ret < 0 && errno != EEXIST) | |
1956 | goto on_error; | |
1957 | ||
3ebe2fbd | 1958 | (void)strlcat(full_path, "/cgroup.procs", len + 1); |
7cea5905 | 1959 | ret = lxc_write_to_file(full_path, pidstr, len, false, 0666); |
c2aed66d CB |
1960 | if (ret == 0) |
1961 | goto on_success; | |
1962 | ||
1963 | /* this is a non-leaf node */ | |
1964 | if (errno != EBUSY) | |
1965 | goto on_error; | |
1966 | ||
1967 | } while (++idx > 0 && idx < 1000); | |
1968 | ||
1969 | on_success: | |
1970 | if (idx < 1000) | |
1971 | fret = 0; | |
1972 | ||
1973 | on_error: | |
1974 | free(base_path); | |
1975 | free(container_cgroup); | |
1976 | free(full_path); | |
1977 | ||
1978 | return fret; | |
1979 | } | |
1980 | ||
2202afc9 CB |
1981 | static bool cgfsng_attach(struct cgroup_ops *ops, const char *name, |
1982 | const char *lxcpath, pid_t pid) | |
ccb4cabe | 1983 | { |
c2aed66d | 1984 | int i, len, ret; |
ccb4cabe | 1985 | char pidstr[25]; |
ccb4cabe SH |
1986 | |
1987 | len = snprintf(pidstr, 25, "%d", pid); | |
0cb10e11 | 1988 | if (len < 0 || len >= 25) |
ccb4cabe SH |
1989 | return false; |
1990 | ||
2202afc9 | 1991 | for (i = 0; ops->hierarchies[i]; i++) { |
c2aed66d CB |
1992 | char *path; |
1993 | char *fullpath = NULL; | |
2202afc9 | 1994 | struct hierarchy *h = ops->hierarchies[i]; |
ccb4cabe | 1995 | |
c2aed66d | 1996 | if (h->version == CGROUP2_SUPER_MAGIC) { |
a3926f6a CB |
1997 | ret = __cg_unified_attach(h, name, lxcpath, pidstr, len, |
1998 | h->controllers[0]); | |
c2aed66d CB |
1999 | if (ret < 0) |
2000 | return false; | |
2001 | ||
2002 | continue; | |
2003 | } | |
2004 | ||
ccb4cabe | 2005 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); |
c2aed66d CB |
2006 | /* not running */ |
2007 | if (!path) | |
ccb4cabe SH |
2008 | continue; |
2009 | ||
371f834d | 2010 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
71cb9afb | 2011 | free(path); |
7cea5905 | 2012 | ret = lxc_write_to_file(fullpath, pidstr, len, false, 0666); |
c2aed66d | 2013 | if (ret < 0) { |
ccb4cabe SH |
2014 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); |
2015 | free(fullpath); | |
ccb4cabe SH |
2016 | return false; |
2017 | } | |
ccb4cabe SH |
2018 | free(fullpath); |
2019 | } | |
2020 | ||
ccb4cabe SH |
2021 | return true; |
2022 | } | |
2023 | ||
e2bd2b13 CB |
2024 | /* Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. Here we |
2025 | * don't have a cgroup_data set up, so we ask the running container through the | |
2026 | * commands API for the cgroup path. | |
ccb4cabe | 2027 | */ |
2202afc9 CB |
2028 | static int cgfsng_get(struct cgroup_ops *ops, const char *filename, char *value, |
2029 | size_t len, const char *name, const char *lxcpath) | |
ccb4cabe | 2030 | { |
ccb4cabe | 2031 | int ret = -1; |
0069cc61 CB |
2032 | size_t controller_len; |
2033 | char *controller, *p, *path; | |
2034 | struct hierarchy *h; | |
ccb4cabe | 2035 | |
0069cc61 CB |
2036 | controller_len = strlen(filename); |
2037 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
2038 | (void)strlcpy(controller, filename, controller_len + 1); |
2039 | ||
0069cc61 CB |
2040 | p = strchr(controller, '.'); |
2041 | if (p) | |
ccb4cabe SH |
2042 | *p = '\0'; |
2043 | ||
0069cc61 CB |
2044 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2045 | /* not running */ | |
2046 | if (!path) | |
ccb4cabe SH |
2047 | return -1; |
2048 | ||
2202afc9 | 2049 | h = get_hierarchy(ops, controller); |
ccb4cabe | 2050 | if (h) { |
0069cc61 CB |
2051 | char *fullpath; |
2052 | ||
2053 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
ccb4cabe SH |
2054 | ret = lxc_read_from_file(fullpath, value, len); |
2055 | free(fullpath); | |
2056 | } | |
ccb4cabe SH |
2057 | free(path); |
2058 | ||
2059 | return ret; | |
2060 | } | |
2061 | ||
eec533e3 CB |
2062 | /* Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. Here we |
2063 | * don't have a cgroup_data set up, so we ask the running container through the | |
2064 | * commands API for the cgroup path. | |
ccb4cabe | 2065 | */ |
2202afc9 CB |
2066 | static int cgfsng_set(struct cgroup_ops *ops, const char *filename, |
2067 | const char *value, const char *name, const char *lxcpath) | |
ccb4cabe | 2068 | { |
ccb4cabe | 2069 | int ret = -1; |
87777968 CB |
2070 | size_t controller_len; |
2071 | char *controller, *p, *path; | |
2072 | struct hierarchy *h; | |
ccb4cabe | 2073 | |
87777968 CB |
2074 | controller_len = strlen(filename); |
2075 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
2076 | (void)strlcpy(controller, filename, controller_len + 1); |
2077 | ||
87777968 CB |
2078 | p = strchr(controller, '.'); |
2079 | if (p) | |
ccb4cabe SH |
2080 | *p = '\0'; |
2081 | ||
87777968 CB |
2082 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2083 | /* not running */ | |
2084 | if (!path) | |
ccb4cabe SH |
2085 | return -1; |
2086 | ||
2202afc9 | 2087 | h = get_hierarchy(ops, controller); |
ccb4cabe | 2088 | if (h) { |
87777968 CB |
2089 | char *fullpath; |
2090 | ||
2091 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
7cea5905 | 2092 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
ccb4cabe SH |
2093 | free(fullpath); |
2094 | } | |
ccb4cabe SH |
2095 | free(path); |
2096 | ||
2097 | return ret; | |
2098 | } | |
2099 | ||
91d1a13a | 2100 | /* take devices cgroup line |
72add155 SH |
2101 | * /dev/foo rwx |
2102 | * and convert it to a valid | |
2103 | * type major:minor mode | |
91d1a13a CB |
2104 | * line. Return <0 on error. Dest is a preallocated buffer long enough to hold |
2105 | * the output. | |
72add155 SH |
2106 | */ |
2107 | static int convert_devpath(const char *invalue, char *dest) | |
2108 | { | |
2a06d041 CB |
2109 | int n_parts; |
2110 | char *p, *path, type; | |
72add155 | 2111 | unsigned long minor, major; |
91d1a13a | 2112 | struct stat sb; |
2a06d041 CB |
2113 | int ret = -EINVAL; |
2114 | char *mode = NULL; | |
72add155 SH |
2115 | |
2116 | path = must_copy_string(invalue); | |
2117 | ||
91d1a13a CB |
2118 | /* Read path followed by mode. Ignore any trailing text. |
2119 | * A ' # comment' would be legal. Technically other text is not | |
2120 | * legal, we could check for that if we cared to. | |
72add155 SH |
2121 | */ |
2122 | for (n_parts = 1, p = path; *p && n_parts < 3; p++) { | |
2c2d6c49 SH |
2123 | if (*p != ' ') |
2124 | continue; | |
2125 | *p = '\0'; | |
91d1a13a | 2126 | |
2c2d6c49 SH |
2127 | if (n_parts != 1) |
2128 | break; | |
2129 | p++; | |
2130 | n_parts++; | |
91d1a13a | 2131 | |
2c2d6c49 SH |
2132 | while (*p == ' ') |
2133 | p++; | |
91d1a13a | 2134 | |
2c2d6c49 | 2135 | mode = p; |
91d1a13a | 2136 | |
2c2d6c49 SH |
2137 | if (*p == '\0') |
2138 | goto out; | |
72add155 | 2139 | } |
2c2d6c49 SH |
2140 | |
2141 | if (n_parts == 1) | |
72add155 | 2142 | goto out; |
72add155 SH |
2143 | |
2144 | ret = stat(path, &sb); | |
2145 | if (ret < 0) | |
2146 | goto out; | |
2147 | ||
72add155 SH |
2148 | mode_t m = sb.st_mode & S_IFMT; |
2149 | switch (m) { | |
2150 | case S_IFBLK: | |
2151 | type = 'b'; | |
2152 | break; | |
2153 | case S_IFCHR: | |
2154 | type = 'c'; | |
2155 | break; | |
2c2d6c49 | 2156 | default: |
91d1a13a | 2157 | ERROR("Unsupported device type %i for \"%s\"", m, path); |
72add155 SH |
2158 | ret = -EINVAL; |
2159 | goto out; | |
2160 | } | |
2c2d6c49 SH |
2161 | |
2162 | major = MAJOR(sb.st_rdev); | |
2163 | minor = MINOR(sb.st_rdev); | |
2164 | ret = snprintf(dest, 50, "%c %lu:%lu %s", type, major, minor, mode); | |
72add155 | 2165 | if (ret < 0 || ret >= 50) { |
2a06d041 CB |
2166 | ERROR("Error on configuration value \"%c %lu:%lu %s\" (max 50 " |
2167 | "chars)", type, major, minor, mode); | |
72add155 SH |
2168 | ret = -ENAMETOOLONG; |
2169 | goto out; | |
2170 | } | |
2171 | ret = 0; | |
2172 | ||
2173 | out: | |
2174 | free(path); | |
2175 | return ret; | |
2176 | } | |
2177 | ||
90e97284 CB |
2178 | /* Called from setup_limits - here we have the container's cgroup_data because |
2179 | * we created the cgroups. | |
ccb4cabe | 2180 | */ |
2202afc9 CB |
2181 | static int cg_legacy_set_data(struct cgroup_ops *ops, const char *filename, |
2182 | const char *value) | |
ccb4cabe | 2183 | { |
ab1a6cac | 2184 | size_t len; |
90e97284 | 2185 | char *fullpath, *p; |
1a0e70ac CB |
2186 | /* "b|c <2^64-1>:<2^64-1> r|w|m" = 47 chars max */ |
2187 | char converted_value[50]; | |
b3646d7e CB |
2188 | struct hierarchy *h; |
2189 | int ret = 0; | |
2190 | char *controller = NULL; | |
ccb4cabe | 2191 | |
ab1a6cac CB |
2192 | len = strlen(filename); |
2193 | controller = alloca(len + 1); | |
64e82f8b DJ |
2194 | (void)strlcpy(controller, filename, len + 1); |
2195 | ||
ab1a6cac CB |
2196 | p = strchr(controller, '.'); |
2197 | if (p) | |
ccb4cabe SH |
2198 | *p = '\0'; |
2199 | ||
c8bf519d | 2200 | if (strcmp("devices.allow", filename) == 0 && value[0] == '/') { |
72add155 SH |
2201 | ret = convert_devpath(value, converted_value); |
2202 | if (ret < 0) | |
c8bf519d | 2203 | return ret; |
72add155 | 2204 | value = converted_value; |
c8bf519d | 2205 | } |
2206 | ||
2202afc9 | 2207 | h = get_hierarchy(ops, controller); |
b3646d7e CB |
2208 | if (!h) { |
2209 | ERROR("Failed to setup limits for the \"%s\" controller. " | |
2210 | "The controller seems to be unused by \"cgfsng\" cgroup " | |
2211 | "driver or not enabled on the cgroup hierarchy", | |
2212 | controller); | |
d1953b26 | 2213 | errno = ENOENT; |
ab1a6cac | 2214 | return -ENOENT; |
ccb4cabe | 2215 | } |
b3646d7e CB |
2216 | |
2217 | fullpath = must_make_path(h->fullcgpath, filename, NULL); | |
7cea5905 | 2218 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
b3646d7e | 2219 | free(fullpath); |
ccb4cabe SH |
2220 | return ret; |
2221 | } | |
2222 | ||
2202afc9 | 2223 | static bool __cg_legacy_setup_limits(struct cgroup_ops *ops, |
a3926f6a CB |
2224 | struct lxc_list *cgroup_settings, |
2225 | bool do_devices) | |
ccb4cabe | 2226 | { |
c347df58 | 2227 | struct lxc_list *iterator, *next, *sorted_cgroup_settings; |
ccb4cabe | 2228 | struct lxc_cgroup *cg; |
ccb4cabe SH |
2229 | bool ret = false; |
2230 | ||
2231 | if (lxc_list_empty(cgroup_settings)) | |
2232 | return true; | |
2233 | ||
2234 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
6b38e644 | 2235 | if (!sorted_cgroup_settings) |
ccb4cabe | 2236 | return false; |
ccb4cabe | 2237 | |
ccb4cabe SH |
2238 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
2239 | cg = iterator->elem; | |
2240 | ||
2241 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
2202afc9 | 2242 | if (cg_legacy_set_data(ops, cg->subsystem, cg->value)) { |
ccb4cabe | 2243 | if (do_devices && (errno == EACCES || errno == EPERM)) { |
c347df58 CB |
2244 | WARN("Failed to set \"%s\" to \"%s\"", |
2245 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2246 | continue; |
2247 | } | |
c347df58 CB |
2248 | WARN("Failed to set \"%s\" to \"%s\"", |
2249 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2250 | goto out; |
2251 | } | |
c347df58 CB |
2252 | DEBUG("Set controller \"%s\" set to \"%s\"", |
2253 | cg->subsystem, cg->value); | |
ccb4cabe | 2254 | } |
ccb4cabe SH |
2255 | } |
2256 | ||
2257 | ret = true; | |
6b38e644 | 2258 | INFO("Limits for the legacy cgroup hierarchies have been setup"); |
ccb4cabe | 2259 | out: |
ccb4cabe SH |
2260 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
2261 | lxc_list_del(iterator); | |
2262 | free(iterator); | |
2263 | } | |
2264 | free(sorted_cgroup_settings); | |
2265 | return ret; | |
2266 | } | |
2267 | ||
2202afc9 | 2268 | static bool __cg_unified_setup_limits(struct cgroup_ops *ops, |
a3926f6a | 2269 | struct lxc_list *cgroup_settings) |
6b38e644 CB |
2270 | { |
2271 | struct lxc_list *iterator; | |
2202afc9 | 2272 | struct hierarchy *h = ops->unified; |
6b38e644 CB |
2273 | |
2274 | if (lxc_list_empty(cgroup_settings)) | |
2275 | return true; | |
2276 | ||
2277 | if (!h) | |
2278 | return false; | |
2279 | ||
2280 | lxc_list_for_each(iterator, cgroup_settings) { | |
2281 | int ret; | |
2282 | char *fullpath; | |
2283 | struct lxc_cgroup *cg = iterator->elem; | |
2284 | ||
2285 | fullpath = must_make_path(h->fullcgpath, cg->subsystem, NULL); | |
7cea5905 | 2286 | ret = lxc_write_to_file(fullpath, cg->value, strlen(cg->value), false, 0666); |
6b38e644 CB |
2287 | free(fullpath); |
2288 | if (ret < 0) { | |
b2ac2cb7 CB |
2289 | SYSERROR("Failed to set \"%s\" to \"%s\"", |
2290 | cg->subsystem, cg->value); | |
6b38e644 CB |
2291 | return false; |
2292 | } | |
2293 | TRACE("Set \"%s\" to \"%s\"", cg->subsystem, cg->value); | |
2294 | } | |
2295 | ||
2296 | INFO("Limits for the unified cgroup hierarchy have been setup"); | |
2297 | return true; | |
2298 | } | |
2299 | ||
2202afc9 | 2300 | static bool cgfsng_setup_limits(struct cgroup_ops *ops, struct lxc_conf *conf, |
6b38e644 CB |
2301 | bool do_devices) |
2302 | { | |
2303 | bool bret; | |
2304 | ||
2202afc9 | 2305 | bret = __cg_legacy_setup_limits(ops, &conf->cgroup, do_devices); |
6b38e644 CB |
2306 | if (!bret) |
2307 | return false; | |
2308 | ||
2202afc9 CB |
2309 | return __cg_unified_setup_limits(ops, &conf->cgroup2); |
2310 | } | |
2311 | ||
2312 | /* At startup, parse_hierarchies finds all the info we need about cgroup | |
2313 | * mountpoints and current cgroups, and stores it in @d. | |
2314 | */ | |
2315 | static bool cg_hybrid_init(struct cgroup_ops *ops) | |
2316 | { | |
2317 | int ret; | |
2318 | char *basecginfo; | |
2319 | bool will_escape; | |
2320 | FILE *f; | |
2321 | size_t len = 0; | |
2322 | char *line = NULL; | |
2323 | char **klist = NULL, **nlist = NULL; | |
2324 | ||
2325 | /* Root spawned containers escape the current cgroup, so use init's | |
2326 | * cgroups as our base in that case. | |
2327 | */ | |
2328 | will_escape = (geteuid() == 0); | |
2329 | if (will_escape) | |
2330 | basecginfo = read_file("/proc/1/cgroup"); | |
2331 | else | |
2332 | basecginfo = read_file("/proc/self/cgroup"); | |
2333 | if (!basecginfo) | |
2334 | return false; | |
2335 | ||
2336 | ret = get_existing_subsystems(&klist, &nlist); | |
2337 | if (ret < 0) { | |
2338 | ERROR("Failed to retrieve available legacy cgroup controllers"); | |
2339 | free(basecginfo); | |
2340 | return false; | |
2341 | } | |
2342 | ||
2343 | f = fopen("/proc/self/mountinfo", "r"); | |
2344 | if (!f) { | |
2345 | ERROR("Failed to open \"/proc/self/mountinfo\""); | |
2346 | free(basecginfo); | |
2347 | return false; | |
2348 | } | |
2349 | ||
2350 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
2351 | ||
2352 | while (getline(&line, &len, f) != -1) { | |
2353 | int type; | |
2354 | bool writeable; | |
2355 | struct hierarchy *new; | |
2356 | char *base_cgroup = NULL, *mountpoint = NULL; | |
2357 | char **controller_list = NULL; | |
2358 | ||
2359 | type = get_cgroup_version(line); | |
2360 | if (type == 0) | |
2361 | continue; | |
2362 | ||
2363 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
2364 | continue; | |
2365 | ||
2366 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
2367 | if (type == CGROUP2_SUPER_MAGIC) | |
2368 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2369 | else if (type == CGROUP_SUPER_MAGIC) | |
2370 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
2371 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
2372 | if (type == CGROUP_SUPER_MAGIC) | |
2373 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2374 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
2375 | if (type == CGROUP2_SUPER_MAGIC) | |
2376 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2377 | } | |
2378 | ||
2379 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, type); | |
2380 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
2381 | continue; | |
2382 | ||
2383 | if (type == CGROUP_SUPER_MAGIC) | |
2384 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
2385 | goto next; | |
2386 | ||
2387 | mountpoint = cg_hybrid_get_mountpoint(line); | |
2388 | if (!mountpoint) { | |
2389 | ERROR("Failed parsing mountpoint from \"%s\"", line); | |
2390 | goto next; | |
2391 | } | |
2392 | ||
2393 | if (type == CGROUP_SUPER_MAGIC) | |
2394 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); | |
2395 | else | |
2396 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); | |
2397 | if (!base_cgroup) { | |
2398 | ERROR("Failed to find current cgroup"); | |
2399 | goto next; | |
2400 | } | |
2401 | ||
2402 | trim(base_cgroup); | |
2403 | prune_init_scope(base_cgroup); | |
2404 | if (type == CGROUP2_SUPER_MAGIC) | |
2405 | writeable = test_writeable_v2(mountpoint, base_cgroup); | |
2406 | else | |
2407 | writeable = test_writeable_v1(mountpoint, base_cgroup); | |
2408 | if (!writeable) | |
2409 | goto next; | |
2410 | ||
2411 | if (type == CGROUP2_SUPER_MAGIC) { | |
2412 | char *cgv2_ctrl_path; | |
2413 | ||
2414 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
2415 | "cgroup.controllers", | |
2416 | NULL); | |
2417 | ||
2418 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
2419 | free(cgv2_ctrl_path); | |
2420 | if (!controller_list) { | |
2421 | controller_list = cg_unified_make_empty_controller(); | |
2422 | TRACE("No controllers are enabled for " | |
2423 | "delegation in the unified hierarchy"); | |
2424 | } | |
2425 | } | |
2426 | ||
2427 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); | |
2428 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
2429 | ops->unified = new; | |
2430 | ||
2431 | continue; | |
2432 | ||
2433 | next: | |
2434 | free_string_list(controller_list); | |
2435 | free(mountpoint); | |
2436 | free(base_cgroup); | |
2437 | } | |
2438 | ||
2439 | free_string_list(klist); | |
2440 | free_string_list(nlist); | |
2441 | ||
2442 | free(basecginfo); | |
2443 | ||
2444 | fclose(f); | |
2445 | free(line); | |
2446 | ||
2447 | TRACE("Writable cgroup hierarchies:"); | |
2448 | lxc_cgfsng_print_hierarchies(ops); | |
2449 | ||
2450 | /* verify that all controllers in cgroup.use and all crucial | |
2451 | * controllers are accounted for | |
2452 | */ | |
2453 | if (!all_controllers_found(ops)) | |
2454 | return false; | |
2455 | ||
2456 | return true; | |
2457 | } | |
2458 | ||
2459 | static int cg_is_pure_unified(void) | |
2460 | { | |
2461 | ||
2462 | int ret; | |
2463 | struct statfs fs; | |
2464 | ||
2465 | ret = statfs("/sys/fs/cgroup", &fs); | |
2466 | if (ret < 0) | |
2467 | return -ENOMEDIUM; | |
2468 | ||
2469 | if (is_fs_type(&fs, CGROUP2_SUPER_MAGIC)) | |
2470 | return CGROUP2_SUPER_MAGIC; | |
2471 | ||
2472 | return 0; | |
2473 | } | |
2474 | ||
2475 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
2476 | static char *cg_unified_get_current_cgroup(void) | |
2477 | { | |
2478 | char *basecginfo, *base_cgroup; | |
2479 | bool will_escape; | |
2480 | char *copy = NULL; | |
2481 | ||
2482 | will_escape = (geteuid() == 0); | |
2483 | if (will_escape) | |
2484 | basecginfo = read_file("/proc/1/cgroup"); | |
2485 | else | |
2486 | basecginfo = read_file("/proc/self/cgroup"); | |
2487 | if (!basecginfo) | |
2488 | return NULL; | |
2489 | ||
2490 | base_cgroup = strstr(basecginfo, "0::/"); | |
2491 | if (!base_cgroup) | |
2492 | goto cleanup_on_err; | |
2493 | ||
2494 | base_cgroup = base_cgroup + 3; | |
2495 | copy = copy_to_eol(base_cgroup); | |
2496 | if (!copy) | |
2497 | goto cleanup_on_err; | |
2498 | ||
2499 | cleanup_on_err: | |
2500 | free(basecginfo); | |
2501 | if (copy) | |
2502 | trim(copy); | |
2503 | ||
2504 | return copy; | |
2505 | } | |
2506 | ||
2507 | static int cg_unified_init(struct cgroup_ops *ops) | |
2508 | { | |
2509 | int ret; | |
2510 | char *mountpoint, *subtree_path; | |
2511 | char **delegatable; | |
2512 | char *base_cgroup = NULL; | |
2513 | ||
2514 | ret = cg_is_pure_unified(); | |
2515 | if (ret == -ENOMEDIUM) | |
2516 | return -ENOMEDIUM; | |
2517 | ||
2518 | if (ret != CGROUP2_SUPER_MAGIC) | |
2519 | return 0; | |
2520 | ||
2521 | base_cgroup = cg_unified_get_current_cgroup(); | |
2522 | if (!base_cgroup) | |
2523 | return -EINVAL; | |
2524 | prune_init_scope(base_cgroup); | |
2525 | ||
2526 | /* We assume that we have already been given controllers to delegate | |
2527 | * further down the hierarchy. If not it is up to the user to delegate | |
2528 | * them to us. | |
2529 | */ | |
2530 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
2531 | subtree_path = must_make_path(mountpoint, base_cgroup, | |
2532 | "cgroup.subtree_control", NULL); | |
2533 | delegatable = cg_unified_get_controllers(subtree_path); | |
2534 | free(subtree_path); | |
2535 | if (!delegatable) | |
2536 | delegatable = cg_unified_make_empty_controller(); | |
2537 | if (!delegatable[0]) | |
2538 | TRACE("No controllers are enabled for delegation"); | |
2539 | ||
2540 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
2541 | * we should verify here. The reason I'm not doing it right is that I'm | |
2542 | * not convinced that lxc.cgroup.use will be the future since it is a | |
2543 | * global property. I much rather have an option that lets you request | |
2544 | * controllers per container. | |
2545 | */ | |
2546 | ||
2547 | add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
2548 | ||
2549 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2550 | return CGROUP2_SUPER_MAGIC; | |
2551 | } | |
2552 | ||
2553 | static bool cg_init(struct cgroup_ops *ops) | |
2554 | { | |
2555 | int ret; | |
2556 | const char *tmp; | |
2557 | ||
2558 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
2559 | if (tmp) | |
2560 | ops->cgroup_use = must_copy_string(tmp); | |
2561 | ||
2562 | ret = cg_unified_init(ops); | |
2563 | if (ret < 0) | |
2564 | return false; | |
2565 | ||
2566 | if (ret == CGROUP2_SUPER_MAGIC) | |
2567 | return true; | |
2568 | ||
2569 | return cg_hybrid_init(ops); | |
2570 | } | |
2571 | ||
2572 | static bool cgfsng_data_init(struct cgroup_ops *ops) | |
2573 | { | |
2574 | const char *cgroup_pattern; | |
2575 | ||
2576 | /* copy system-wide cgroup information */ | |
2577 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); | |
2578 | if (!cgroup_pattern) { | |
2579 | /* lxc.cgroup.pattern is only NULL on error. */ | |
2580 | ERROR("Failed to retrieve cgroup pattern"); | |
2581 | return false; | |
2582 | } | |
2583 | ops->cgroup_pattern = must_copy_string(cgroup_pattern); | |
2584 | ||
2585 | return true; | |
2586 | } | |
2587 | ||
2588 | struct cgroup_ops *cgfsng_ops_init(void) | |
2589 | { | |
2590 | struct cgroup_ops *cgfsng_ops; | |
2591 | ||
2592 | cgfsng_ops = malloc(sizeof(struct cgroup_ops)); | |
2593 | if (!cgfsng_ops) | |
2594 | return NULL; | |
2595 | ||
2596 | memset(cgfsng_ops, 0, sizeof(struct cgroup_ops)); | |
2597 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; | |
2598 | ||
2599 | if (!cg_init(cgfsng_ops)) { | |
2600 | free(cgfsng_ops); | |
2601 | return NULL; | |
2602 | } | |
2603 | ||
2604 | cgfsng_ops->data_init = cgfsng_data_init; | |
2605 | cgfsng_ops->destroy = cgfsng_destroy; | |
2606 | cgfsng_ops->create = cgfsng_create; | |
2607 | cgfsng_ops->enter = cgfsng_enter; | |
2608 | cgfsng_ops->escape = cgfsng_escape; | |
2609 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
2610 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; | |
2611 | cgfsng_ops->get_cgroup = cgfsng_get_cgroup; | |
2612 | cgfsng_ops->get = cgfsng_get; | |
2613 | cgfsng_ops->set = cgfsng_set; | |
2614 | cgfsng_ops->unfreeze = cgfsng_unfreeze; | |
2615 | cgfsng_ops->setup_limits = cgfsng_setup_limits; | |
2616 | cgfsng_ops->driver = "cgfsng"; | |
2617 | cgfsng_ops->version = "1.0.0"; | |
2618 | cgfsng_ops->attach = cgfsng_attach; | |
2619 | cgfsng_ops->chown = cgfsng_chown; | |
2620 | cgfsng_ops->mount = cgfsng_mount; | |
2621 | cgfsng_ops->nrtasks = cgfsng_nrtasks; | |
2622 | ||
2623 | return cgfsng_ops; | |
2624 | } |