]>
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" |
c19ad94b | 58 | #include "macro.h" |
43654d34 | 59 | #include "storage/storage.h" |
a54694f8 | 60 | #include "utils.h" |
ccb4cabe | 61 | |
64e82f8b DJ |
62 | #ifndef HAVE_STRLCPY |
63 | #include "include/strlcpy.h" | |
64 | #endif | |
65 | ||
3ebe2fbd DJ |
66 | #ifndef HAVE_STRLCAT |
67 | #include "include/strlcat.h" | |
68 | #endif | |
69 | ||
fb55e009 CB |
70 | #define __cgfsng_ops__ |
71 | ||
ac2cecc4 | 72 | lxc_log_define(cgfsng, cgroup); |
ccb4cabe | 73 | |
ccb4cabe SH |
74 | static void free_string_list(char **clist) |
75 | { | |
2d5fe5ba | 76 | int i; |
ccb4cabe | 77 | |
2d5fe5ba CB |
78 | if (!clist) |
79 | return; | |
80 | ||
81 | for (i = 0; clist[i]; i++) | |
82 | free(clist[i]); | |
83 | ||
84 | free(clist); | |
ccb4cabe SH |
85 | } |
86 | ||
7745483d | 87 | /* Allocate a pointer, do not fail. */ |
ccb4cabe SH |
88 | static void *must_alloc(size_t sz) |
89 | { | |
90 | return must_realloc(NULL, sz); | |
91 | } | |
92 | ||
8b8db2f6 CB |
93 | /* Given a pointer to a null-terminated array of pointers, realloc to add one |
94 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
95 | * second-to-last entry - that is, the one which is now available for use | |
96 | * (keeping the list null-terminated). | |
ccb4cabe SH |
97 | */ |
98 | static int append_null_to_list(void ***list) | |
99 | { | |
100 | int newentry = 0; | |
101 | ||
102 | if (*list) | |
8b8db2f6 CB |
103 | for (; (*list)[newentry]; newentry++) |
104 | ; | |
ccb4cabe SH |
105 | |
106 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
107 | (*list)[newentry + 1] = NULL; | |
108 | return newentry; | |
109 | } | |
110 | ||
8073018d CB |
111 | /* Given a null-terminated array of strings, check whether @entry is one of the |
112 | * strings. | |
ccb4cabe SH |
113 | */ |
114 | static bool string_in_list(char **list, const char *entry) | |
115 | { | |
116 | int i; | |
117 | ||
118 | if (!list) | |
119 | return false; | |
d6337a5f | 120 | |
ccb4cabe SH |
121 | for (i = 0; list[i]; i++) |
122 | if (strcmp(list[i], entry) == 0) | |
123 | return true; | |
124 | ||
125 | return false; | |
126 | } | |
127 | ||
ac010944 CB |
128 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into |
129 | * "name=systemd". Do not fail. | |
130 | */ | |
131 | static char *cg_legacy_must_prefix_named(char *entry) | |
132 | { | |
133 | size_t len; | |
134 | char *prefixed; | |
135 | ||
136 | len = strlen(entry); | |
137 | prefixed = must_alloc(len + 6); | |
138 | ||
cbe2185b CB |
139 | memcpy(prefixed, "name=", sizeof("name=") - 1); |
140 | memcpy(prefixed + sizeof("name=") - 1, entry, len); | |
ac010944 CB |
141 | prefixed[len + 5] = '\0'; |
142 | return prefixed; | |
143 | } | |
144 | ||
42a993b4 CB |
145 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time |
146 | * we are called. | |
ccb4cabe | 147 | * |
42a993b4 CB |
148 | * We also handle named subsystems here. Any controller which is not a kernel |
149 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
150 | * we refuse to use because we're not sure which we have here. | |
151 | * (TODO: We could work around this in some cases by just remounting to be | |
152 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
ccb4cabe SH |
153 | * |
154 | * The last entry will always be NULL. | |
155 | */ | |
42a993b4 CB |
156 | static void must_append_controller(char **klist, char **nlist, char ***clist, |
157 | char *entry) | |
ccb4cabe SH |
158 | { |
159 | int newentry; | |
160 | char *copy; | |
161 | ||
162 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
c2712f64 | 163 | ERROR("Refusing to use ambiguous controller \"%s\"", entry); |
ccb4cabe SH |
164 | ERROR("It is both a named and kernel subsystem"); |
165 | return; | |
166 | } | |
167 | ||
168 | newentry = append_null_to_list((void ***)clist); | |
169 | ||
170 | if (strncmp(entry, "name=", 5) == 0) | |
171 | copy = must_copy_string(entry); | |
172 | else if (string_in_list(klist, entry)) | |
173 | copy = must_copy_string(entry); | |
174 | else | |
7745483d | 175 | copy = cg_legacy_must_prefix_named(entry); |
ccb4cabe SH |
176 | |
177 | (*clist)[newentry] = copy; | |
178 | } | |
179 | ||
5ae0207c CB |
180 | /* Given a handler's cgroup data, return the struct hierarchy for the controller |
181 | * @c, or NULL if there is none. | |
ccb4cabe | 182 | */ |
2202afc9 | 183 | struct hierarchy *get_hierarchy(struct cgroup_ops *ops, const char *c) |
ccb4cabe SH |
184 | { |
185 | int i; | |
186 | ||
2202afc9 | 187 | if (!ops->hierarchies) |
ccb4cabe | 188 | return NULL; |
d6337a5f | 189 | |
2202afc9 | 190 | for (i = 0; ops->hierarchies[i]; i++) { |
d6337a5f CB |
191 | if (!c) { |
192 | /* This is the empty unified hierarchy. */ | |
2202afc9 CB |
193 | if (ops->hierarchies[i]->controllers && |
194 | !ops->hierarchies[i]->controllers[0]) | |
195 | return ops->hierarchies[i]; | |
d6337a5f | 196 | |
106f1f38 | 197 | continue; |
d6337a5f CB |
198 | } |
199 | ||
2202afc9 CB |
200 | if (string_in_list(ops->hierarchies[i]->controllers, c)) |
201 | return ops->hierarchies[i]; | |
ccb4cabe | 202 | } |
d6337a5f | 203 | |
ccb4cabe SH |
204 | return NULL; |
205 | } | |
206 | ||
a54694f8 CB |
207 | #define BATCH_SIZE 50 |
208 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
209 | { | |
210 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
211 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
212 | ||
213 | if (!*mem || newbatches > oldbatches) { | |
214 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
215 | } | |
216 | } | |
217 | ||
218 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
219 | { | |
220 | size_t full = oldlen + newlen; | |
221 | ||
222 | batch_realloc(dest, oldlen, full + 1); | |
223 | ||
224 | memcpy(*dest + oldlen, new, newlen + 1); | |
225 | } | |
226 | ||
227 | /* Slurp in a whole file */ | |
d6337a5f | 228 | static char *read_file(const char *fnam) |
a54694f8 CB |
229 | { |
230 | FILE *f; | |
231 | char *line = NULL, *buf = NULL; | |
232 | size_t len = 0, fulllen = 0; | |
233 | int linelen; | |
234 | ||
235 | f = fopen(fnam, "r"); | |
236 | if (!f) | |
237 | return NULL; | |
238 | while ((linelen = getline(&line, &len, f)) != -1) { | |
239 | append_line(&buf, fulllen, line, linelen); | |
240 | fulllen += linelen; | |
241 | } | |
242 | fclose(f); | |
243 | free(line); | |
244 | return buf; | |
245 | } | |
246 | ||
247 | /* Taken over modified from the kernel sources. */ | |
248 | #define NBITS 32 /* bits in uint32_t */ | |
249 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
250 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
251 | ||
252 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
253 | { | |
254 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
255 | } | |
256 | ||
257 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
258 | { | |
259 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
260 | } | |
261 | ||
262 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
263 | { | |
264 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
265 | } | |
266 | ||
267 | /* Create cpumask from cpulist aka turn: | |
268 | * | |
269 | * 0,2-3 | |
270 | * | |
d5d468f6 | 271 | * into bit array |
a54694f8 CB |
272 | * |
273 | * 1 0 1 1 | |
274 | */ | |
275 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
276 | { | |
277 | char *token; | |
d5d468f6 CB |
278 | size_t arrlen; |
279 | uint32_t *bitarr; | |
d5d468f6 CB |
280 | |
281 | arrlen = BITS_TO_LONGS(nbits); | |
282 | bitarr = calloc(arrlen, sizeof(uint32_t)); | |
a54694f8 CB |
283 | if (!bitarr) |
284 | return NULL; | |
285 | ||
0be0d78f | 286 | lxc_iterate_parts(token, buf, ",") { |
a54694f8 | 287 | errno = 0; |
d5d468f6 CB |
288 | unsigned end, start; |
289 | char *range; | |
a54694f8 | 290 | |
d5d468f6 CB |
291 | start = strtoul(token, NULL, 0); |
292 | end = start; | |
293 | range = strchr(token, '-'); | |
a54694f8 CB |
294 | if (range) |
295 | end = strtoul(range + 1, NULL, 0); | |
d5d468f6 | 296 | |
a54694f8 CB |
297 | if (!(start <= end)) { |
298 | free(bitarr); | |
299 | return NULL; | |
300 | } | |
301 | ||
302 | if (end >= nbits) { | |
303 | free(bitarr); | |
304 | return NULL; | |
305 | } | |
306 | ||
307 | while (start <= end) | |
308 | set_bit(start++, bitarr); | |
309 | } | |
310 | ||
311 | return bitarr; | |
312 | } | |
313 | ||
a54694f8 CB |
314 | /* Turn cpumask into simple, comma-separated cpulist. */ |
315 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
316 | { | |
a54694f8 | 317 | int ret; |
414c6719 | 318 | size_t i; |
a54694f8 | 319 | char **cpulist = NULL; |
c19ad94b | 320 | char numstr[INTTYPE_TO_STRLEN(size_t)] = {0}; |
a54694f8 CB |
321 | |
322 | for (i = 0; i <= nbits; i++) { | |
414c6719 CB |
323 | if (!is_set(i, bitarr)) |
324 | continue; | |
325 | ||
979a0d93 CB |
326 | ret = snprintf(numstr, sizeof(numstr), "%zu", i); |
327 | if (ret < 0 || (size_t)ret >= sizeof(numstr)) { | |
414c6719 CB |
328 | lxc_free_array((void **)cpulist, free); |
329 | return NULL; | |
330 | } | |
331 | ||
332 | ret = lxc_append_string(&cpulist, numstr); | |
333 | if (ret < 0) { | |
334 | lxc_free_array((void **)cpulist, free); | |
335 | return NULL; | |
a54694f8 CB |
336 | } |
337 | } | |
414c6719 CB |
338 | |
339 | if (!cpulist) | |
340 | return NULL; | |
341 | ||
a54694f8 CB |
342 | return lxc_string_join(",", (const char **)cpulist, false); |
343 | } | |
344 | ||
345 | static ssize_t get_max_cpus(char *cpulist) | |
346 | { | |
347 | char *c1, *c2; | |
348 | char *maxcpus = cpulist; | |
349 | size_t cpus = 0; | |
350 | ||
351 | c1 = strrchr(maxcpus, ','); | |
352 | if (c1) | |
353 | c1++; | |
354 | ||
355 | c2 = strrchr(maxcpus, '-'); | |
356 | if (c2) | |
357 | c2++; | |
358 | ||
359 | if (!c1 && !c2) | |
360 | c1 = maxcpus; | |
361 | else if (c1 > c2) | |
362 | c2 = c1; | |
363 | else if (c1 < c2) | |
364 | c1 = c2; | |
333987b9 | 365 | else if (!c1 && c2) |
a54694f8 CB |
366 | c1 = c2; |
367 | ||
a54694f8 CB |
368 | errno = 0; |
369 | cpus = strtoul(c1, NULL, 0); | |
370 | if (errno != 0) | |
371 | return -1; | |
372 | ||
373 | return cpus; | |
374 | } | |
375 | ||
6f9584d8 | 376 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a3926f6a | 377 | static bool cg_legacy_filter_and_set_cpus(char *path, bool am_initialized) |
a54694f8 | 378 | { |
a54694f8 CB |
379 | int ret; |
380 | ssize_t i; | |
59ac3b88 CB |
381 | char *lastslash, *fpath, oldv; |
382 | ssize_t maxisol = 0, maxposs = 0; | |
383 | char *cpulist = NULL, *isolcpus = NULL, *posscpus = NULL; | |
384 | uint32_t *isolmask = NULL, *possmask = NULL; | |
6f9584d8 | 385 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
386 | |
387 | lastslash = strrchr(path, '/'); | |
59ac3b88 CB |
388 | if (!lastslash) { |
389 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
a54694f8 CB |
390 | return bret; |
391 | } | |
392 | oldv = *lastslash; | |
393 | *lastslash = '\0'; | |
394 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
395 | posscpus = read_file(fpath); | |
6f9584d8 | 396 | if (!posscpus) { |
59ac3b88 | 397 | SYSERROR("Failed to read file \"%s\"", fpath); |
6f9584d8 CB |
398 | goto on_error; |
399 | } | |
a54694f8 CB |
400 | |
401 | /* Get maximum number of cpus found in possible cpuset. */ | |
402 | maxposs = get_max_cpus(posscpus); | |
92d5ea57 | 403 | if (maxposs < 0 || maxposs >= INT_MAX - 1) |
6f9584d8 | 404 | goto on_error; |
a54694f8 | 405 | |
6f9584d8 CB |
406 | if (!file_exists(__ISOL_CPUS)) { |
407 | /* This system doesn't expose isolated cpus. */ | |
59ac3b88 | 408 | DEBUG("The path \""__ISOL_CPUS"\" to read isolated cpus from does not exist"); |
65d29cbc CB |
409 | cpulist = posscpus; |
410 | /* No isolated cpus but we weren't already initialized by | |
411 | * someone. We should simply copy the parents cpuset.cpus | |
412 | * values. | |
413 | */ | |
414 | if (!am_initialized) { | |
59ac3b88 | 415 | DEBUG("Copying cpu settings of parent cgroup"); |
65d29cbc CB |
416 | goto copy_parent; |
417 | } | |
418 | /* No isolated cpus but we were already initialized by someone. | |
419 | * Nothing more to do for us. | |
420 | */ | |
6f9584d8 CB |
421 | goto on_success; |
422 | } | |
423 | ||
424 | isolcpus = read_file(__ISOL_CPUS); | |
425 | if (!isolcpus) { | |
59ac3b88 | 426 | SYSERROR("Failed to read file \""__ISOL_CPUS"\""); |
6f9584d8 CB |
427 | goto on_error; |
428 | } | |
a54694f8 | 429 | if (!isdigit(isolcpus[0])) { |
59ac3b88 | 430 | TRACE("No isolated cpus detected"); |
a54694f8 CB |
431 | cpulist = posscpus; |
432 | /* No isolated cpus but we weren't already initialized by | |
433 | * someone. We should simply copy the parents cpuset.cpus | |
434 | * values. | |
435 | */ | |
6f9584d8 | 436 | if (!am_initialized) { |
59ac3b88 | 437 | DEBUG("Copying cpu settings of parent cgroup"); |
a54694f8 | 438 | goto copy_parent; |
6f9584d8 | 439 | } |
a54694f8 CB |
440 | /* No isolated cpus but we were already initialized by someone. |
441 | * Nothing more to do for us. | |
442 | */ | |
6f9584d8 | 443 | goto on_success; |
a54694f8 CB |
444 | } |
445 | ||
446 | /* Get maximum number of cpus found in isolated cpuset. */ | |
447 | maxisol = get_max_cpus(isolcpus); | |
92d5ea57 | 448 | if (maxisol < 0 || maxisol >= INT_MAX - 1) |
6f9584d8 | 449 | goto on_error; |
a54694f8 CB |
450 | |
451 | if (maxposs < maxisol) | |
452 | maxposs = maxisol; | |
453 | maxposs++; | |
454 | ||
455 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 | 456 | if (!possmask) { |
59ac3b88 | 457 | ERROR("Failed to create cpumask for possible cpus"); |
6f9584d8 CB |
458 | goto on_error; |
459 | } | |
a54694f8 CB |
460 | |
461 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 | 462 | if (!isolmask) { |
59ac3b88 | 463 | ERROR("Failed to create cpumask for isolated cpus"); |
6f9584d8 CB |
464 | goto on_error; |
465 | } | |
a54694f8 CB |
466 | |
467 | for (i = 0; i <= maxposs; i++) { | |
59ac3b88 CB |
468 | if (!is_set(i, isolmask) || !is_set(i, possmask)) |
469 | continue; | |
470 | ||
471 | flipped_bit = true; | |
472 | clear_bit(i, possmask); | |
a54694f8 CB |
473 | } |
474 | ||
6f9584d8 | 475 | if (!flipped_bit) { |
59ac3b88 | 476 | DEBUG("No isolated cpus present in cpuset"); |
6f9584d8 CB |
477 | goto on_success; |
478 | } | |
59ac3b88 | 479 | DEBUG("Removed isolated cpus from cpuset"); |
6f9584d8 | 480 | |
a54694f8 | 481 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 | 482 | if (!cpulist) { |
59ac3b88 | 483 | ERROR("Failed to create cpu list"); |
6f9584d8 CB |
484 | goto on_error; |
485 | } | |
a54694f8 CB |
486 | |
487 | copy_parent: | |
488 | *lastslash = oldv; | |
dcbc861e | 489 | free(fpath); |
a54694f8 | 490 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
7cea5905 | 491 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false, 0666); |
6f9584d8 | 492 | if (ret < 0) { |
59ac3b88 | 493 | SYSERROR("Failed to write cpu list to \"%s\"", fpath); |
6f9584d8 CB |
494 | goto on_error; |
495 | } | |
496 | ||
497 | on_success: | |
498 | bret = true; | |
a54694f8 | 499 | |
6f9584d8 | 500 | on_error: |
a54694f8 CB |
501 | free(fpath); |
502 | ||
503 | free(isolcpus); | |
504 | free(isolmask); | |
505 | ||
506 | if (posscpus != cpulist) | |
507 | free(posscpus); | |
508 | free(possmask); | |
509 | ||
510 | free(cpulist); | |
511 | return bret; | |
512 | } | |
513 | ||
e3a3fecf SH |
514 | /* Copy contents of parent(@path)/@file to @path/@file */ |
515 | static bool copy_parent_file(char *path, char *file) | |
516 | { | |
e3a3fecf | 517 | int ret; |
b095a8eb CB |
518 | char *fpath, *lastslash, oldv; |
519 | int len = 0; | |
520 | char *value = NULL; | |
e3a3fecf SH |
521 | |
522 | lastslash = strrchr(path, '/'); | |
b095a8eb CB |
523 | if (!lastslash) { |
524 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
e3a3fecf SH |
525 | return false; |
526 | } | |
527 | oldv = *lastslash; | |
528 | *lastslash = '\0'; | |
529 | fpath = must_make_path(path, file, NULL); | |
530 | len = lxc_read_from_file(fpath, NULL, 0); | |
531 | if (len <= 0) | |
b095a8eb CB |
532 | goto on_error; |
533 | ||
e3a3fecf | 534 | value = must_alloc(len + 1); |
b095a8eb CB |
535 | ret = lxc_read_from_file(fpath, value, len); |
536 | if (ret != len) | |
537 | goto on_error; | |
e3a3fecf | 538 | free(fpath); |
b095a8eb | 539 | |
e3a3fecf SH |
540 | *lastslash = oldv; |
541 | fpath = must_make_path(path, file, NULL); | |
7cea5905 | 542 | ret = lxc_write_to_file(fpath, value, len, false, 0666); |
e3a3fecf | 543 | if (ret < 0) |
b095a8eb | 544 | SYSERROR("Failed to write \"%s\" to file \"%s\"", value, fpath); |
e3a3fecf SH |
545 | free(fpath); |
546 | free(value); | |
547 | return ret >= 0; | |
548 | ||
b095a8eb CB |
549 | on_error: |
550 | SYSERROR("Failed to read file \"%s\"", fpath); | |
e3a3fecf SH |
551 | free(fpath); |
552 | free(value); | |
553 | return false; | |
554 | } | |
555 | ||
7793add3 CB |
556 | /* Initialize the cpuset hierarchy in first directory of @gname and set |
557 | * cgroup.clone_children so that children inherit settings. Since the | |
558 | * h->base_path is populated by init or ourselves, we know it is already | |
559 | * initialized. | |
e3a3fecf | 560 | */ |
a3926f6a | 561 | static bool cg_legacy_handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf | 562 | { |
7793add3 CB |
563 | int ret; |
564 | char v; | |
565 | char *cgpath, *clonechildrenpath, *slash; | |
e3a3fecf SH |
566 | |
567 | if (!string_in_list(h->controllers, "cpuset")) | |
568 | return true; | |
569 | ||
570 | if (*cgname == '/') | |
571 | cgname++; | |
572 | slash = strchr(cgname, '/'); | |
573 | if (slash) | |
574 | *slash = '\0'; | |
575 | ||
bb221ad1 | 576 | cgpath = must_make_path(h->mountpoint, h->container_base_path, cgname, NULL); |
e3a3fecf SH |
577 | if (slash) |
578 | *slash = '/'; | |
7793add3 CB |
579 | |
580 | ret = mkdir(cgpath, 0755); | |
581 | if (ret < 0) { | |
582 | if (errno != EEXIST) { | |
583 | SYSERROR("Failed to create directory \"%s\"", cgpath); | |
584 | free(cgpath); | |
585 | return false; | |
586 | } | |
e3a3fecf | 587 | } |
6f9584d8 | 588 | |
7793add3 CB |
589 | clonechildrenpath = |
590 | must_make_path(cgpath, "cgroup.clone_children", NULL); | |
6328fd9c CB |
591 | /* unified hierarchy doesn't have clone_children */ |
592 | if (!file_exists(clonechildrenpath)) { | |
e3a3fecf SH |
593 | free(clonechildrenpath); |
594 | free(cgpath); | |
595 | return true; | |
596 | } | |
7793add3 CB |
597 | |
598 | ret = lxc_read_from_file(clonechildrenpath, &v, 1); | |
599 | if (ret < 0) { | |
600 | SYSERROR("Failed to read file \"%s\"", clonechildrenpath); | |
e3a3fecf SH |
601 | free(clonechildrenpath); |
602 | free(cgpath); | |
603 | return false; | |
604 | } | |
605 | ||
a54694f8 | 606 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
a3926f6a | 607 | if (!cg_legacy_filter_and_set_cpus(cgpath, v == '1')) { |
7793add3 | 608 | SYSERROR("Failed to remove isolated cpus"); |
6f9584d8 CB |
609 | free(clonechildrenpath); |
610 | free(cgpath); | |
a54694f8 | 611 | return false; |
6f9584d8 | 612 | } |
a54694f8 | 613 | |
7793add3 CB |
614 | /* Already set for us by someone else. */ |
615 | if (v == '1') { | |
616 | DEBUG("\"cgroup.clone_children\" was already set to \"1\""); | |
e3a3fecf SH |
617 | free(clonechildrenpath); |
618 | free(cgpath); | |
619 | return true; | |
620 | } | |
621 | ||
622 | /* copy parent's settings */ | |
a54694f8 | 623 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
7793add3 | 624 | SYSERROR("Failed to copy \"cpuset.mems\" settings"); |
e3a3fecf SH |
625 | free(cgpath); |
626 | free(clonechildrenpath); | |
627 | return false; | |
628 | } | |
629 | free(cgpath); | |
630 | ||
7cea5905 | 631 | ret = lxc_write_to_file(clonechildrenpath, "1", 1, false, 0666); |
7793add3 | 632 | if (ret < 0) { |
e3a3fecf | 633 | /* Set clone_children so children inherit our settings */ |
7793add3 | 634 | SYSERROR("Failed to write 1 to \"%s\"", clonechildrenpath); |
e3a3fecf SH |
635 | free(clonechildrenpath); |
636 | return false; | |
637 | } | |
638 | free(clonechildrenpath); | |
639 | return true; | |
640 | } | |
641 | ||
5c0089ae CB |
642 | /* Given two null-terminated lists of strings, return true if any string is in |
643 | * both. | |
ccb4cabe SH |
644 | */ |
645 | static bool controller_lists_intersect(char **l1, char **l2) | |
646 | { | |
647 | int i; | |
648 | ||
649 | if (!l1 || !l2) | |
650 | return false; | |
651 | ||
652 | for (i = 0; l1[i]; i++) { | |
653 | if (string_in_list(l2, l1[i])) | |
654 | return true; | |
655 | } | |
5c0089ae | 656 | |
ccb4cabe SH |
657 | return false; |
658 | } | |
659 | ||
258449e5 CB |
660 | /* For a null-terminated list of controllers @clist, return true if any of those |
661 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
662 | * Realistically, if one is present, all must be present. | |
ccb4cabe SH |
663 | */ |
664 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
665 | { | |
666 | int i; | |
667 | ||
668 | if (!hlist) | |
669 | return false; | |
258449e5 | 670 | |
ccb4cabe SH |
671 | for (i = 0; hlist[i]; i++) |
672 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
673 | return true; | |
ccb4cabe | 674 | |
258449e5 | 675 | return false; |
ccb4cabe SH |
676 | } |
677 | ||
f57ac67f CB |
678 | /* Return true if the controller @entry is found in the null-terminated list of |
679 | * hierarchies @hlist. | |
ccb4cabe SH |
680 | */ |
681 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
682 | { | |
683 | int i; | |
d6337a5f | 684 | |
ccb4cabe SH |
685 | if (!hlist) |
686 | return false; | |
687 | ||
688 | for (i = 0; hlist[i]; i++) | |
689 | if (string_in_list(hlist[i]->controllers, entry)) | |
690 | return true; | |
d6337a5f | 691 | |
ccb4cabe SH |
692 | return false; |
693 | } | |
694 | ||
e1c27ab0 CB |
695 | /* Return true if all of the controllers which we require have been found. The |
696 | * required list is freezer and anything in lxc.cgroup.use. | |
ccb4cabe | 697 | */ |
2202afc9 | 698 | static bool all_controllers_found(struct cgroup_ops *ops) |
ccb4cabe | 699 | { |
b7b18fc5 | 700 | char **cur; |
2202afc9 | 701 | struct hierarchy **hlist = ops->hierarchies; |
ccb4cabe | 702 | |
ccb4cabe | 703 | if (!controller_found(hlist, "freezer")) { |
2202afc9 | 704 | ERROR("No freezer controller mountpoint found"); |
ccb4cabe SH |
705 | return false; |
706 | } | |
707 | ||
2202afc9 | 708 | if (!ops->cgroup_use) |
ccb4cabe | 709 | return true; |
c2712f64 | 710 | |
b7b18fc5 CB |
711 | for (cur = ops->cgroup_use; cur && *cur; cur++) |
712 | if (!controller_found(hlist, *cur)) { | |
713 | ERROR("No %s controller mountpoint found", *cur); | |
ccb4cabe SH |
714 | return false; |
715 | } | |
c2712f64 | 716 | |
ccb4cabe SH |
717 | return true; |
718 | } | |
719 | ||
f205f10c CB |
720 | /* Get the controllers from a mountinfo line There are other ways we could get |
721 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
722 | * could parse the mount options. But we simply assume that the mountpoint must | |
723 | * be /sys/fs/cgroup/controller-list | |
ccb4cabe | 724 | */ |
a3926f6a CB |
725 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, |
726 | int type) | |
ccb4cabe | 727 | { |
f205f10c CB |
728 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list |
729 | * for legacy hierarchies. | |
730 | */ | |
ccb4cabe | 731 | int i; |
411ac6d8 | 732 | char *dup, *p2, *tok; |
0be0d78f | 733 | char *p = line, *sep = ","; |
411ac6d8 | 734 | char **aret = NULL; |
6328fd9c | 735 | |
ccb4cabe | 736 | for (i = 0; i < 4; i++) { |
235f1815 | 737 | p = strchr(p, ' '); |
ccb4cabe SH |
738 | if (!p) |
739 | return NULL; | |
740 | p++; | |
741 | } | |
a55f31bd | 742 | |
f205f10c CB |
743 | /* Note, if we change how mountinfo works, then our caller will need to |
744 | * verify /sys/fs/cgroup/ in this field. | |
745 | */ | |
746 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { | |
2202afc9 | 747 | ERROR("Found hierarchy not under /sys/fs/cgroup: \"%s\"", p); |
ccb4cabe | 748 | return NULL; |
5059aae9 | 749 | } |
d6337a5f | 750 | |
ccb4cabe | 751 | p += 15; |
235f1815 | 752 | p2 = strchr(p, ' '); |
ccb4cabe | 753 | if (!p2) { |
2202afc9 | 754 | ERROR("Corrupt mountinfo"); |
ccb4cabe SH |
755 | return NULL; |
756 | } | |
757 | *p2 = '\0'; | |
6328fd9c | 758 | |
d6337a5f | 759 | if (type == CGROUP_SUPER_MAGIC) { |
0be0d78f CB |
760 | /* strdup() here for v1 hierarchies. Otherwise |
761 | * lxc_iterate_parts() will destroy mountpoints such as | |
762 | * "/sys/fs/cgroup/cpu,cpuacct". | |
d6337a5f CB |
763 | */ |
764 | dup = strdup(p); | |
765 | if (!dup) | |
766 | return NULL; | |
767 | ||
0be0d78f | 768 | lxc_iterate_parts(tok, dup, sep) { |
d6337a5f | 769 | must_append_controller(klist, nlist, &aret, tok); |
0be0d78f | 770 | } |
d6337a5f CB |
771 | |
772 | free(dup); | |
411ac6d8 | 773 | } |
d6337a5f | 774 | *p2 = ' '; |
f205f10c | 775 | |
d6337a5f CB |
776 | return aret; |
777 | } | |
411ac6d8 | 778 | |
d6337a5f CB |
779 | static char **cg_unified_make_empty_controller(void) |
780 | { | |
781 | int newentry; | |
782 | char **aret = NULL; | |
783 | ||
784 | newentry = append_null_to_list((void ***)&aret); | |
785 | aret[newentry] = NULL; | |
786 | return aret; | |
787 | } | |
788 | ||
789 | static char **cg_unified_get_controllers(const char *file) | |
790 | { | |
791 | char *buf, *tok; | |
0be0d78f | 792 | char *sep = " \t\n"; |
d6337a5f CB |
793 | char **aret = NULL; |
794 | ||
795 | buf = read_file(file); | |
796 | if (!buf) | |
411ac6d8 | 797 | return NULL; |
6328fd9c | 798 | |
0be0d78f | 799 | lxc_iterate_parts(tok, buf, sep) { |
d6337a5f CB |
800 | int newentry; |
801 | char *copy; | |
802 | ||
803 | newentry = append_null_to_list((void ***)&aret); | |
804 | copy = must_copy_string(tok); | |
805 | aret[newentry] = copy; | |
ccb4cabe SH |
806 | } |
807 | ||
d6337a5f | 808 | free(buf); |
ccb4cabe SH |
809 | return aret; |
810 | } | |
811 | ||
2202afc9 | 812 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, |
bb221ad1 | 813 | char *container_base_path, int type) |
ccb4cabe SH |
814 | { |
815 | struct hierarchy *new; | |
816 | int newentry; | |
817 | ||
818 | new = must_alloc(sizeof(*new)); | |
819 | new->controllers = clist; | |
820 | new->mountpoint = mountpoint; | |
bb221ad1 | 821 | new->container_base_path = container_base_path; |
eb697136 | 822 | new->container_full_path = NULL; |
e09b62f9 | 823 | new->monitor_full_path = NULL; |
d6337a5f | 824 | new->version = type; |
6328fd9c | 825 | |
2202afc9 CB |
826 | newentry = append_null_to_list((void ***)h); |
827 | (*h)[newentry] = new; | |
d6337a5f | 828 | return new; |
ccb4cabe SH |
829 | } |
830 | ||
798c3b33 CB |
831 | /* Get a copy of the mountpoint from @line, which is a line from |
832 | * /proc/self/mountinfo. | |
ccb4cabe | 833 | */ |
a3926f6a | 834 | static char *cg_hybrid_get_mountpoint(char *line) |
ccb4cabe SH |
835 | { |
836 | int i; | |
ccb4cabe | 837 | size_t len; |
798c3b33 CB |
838 | char *p2; |
839 | char *p = line, *sret = NULL; | |
ccb4cabe SH |
840 | |
841 | for (i = 0; i < 4; i++) { | |
235f1815 | 842 | p = strchr(p, ' '); |
ccb4cabe SH |
843 | if (!p) |
844 | return NULL; | |
845 | p++; | |
846 | } | |
d6337a5f | 847 | |
798c3b33 | 848 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) |
d6337a5f CB |
849 | return NULL; |
850 | ||
851 | p2 = strchr(p + 15, ' '); | |
852 | if (!p2) | |
853 | return NULL; | |
854 | *p2 = '\0'; | |
855 | ||
ccb4cabe SH |
856 | len = strlen(p); |
857 | sret = must_alloc(len + 1); | |
858 | memcpy(sret, p, len); | |
859 | sret[len] = '\0'; | |
860 | return sret; | |
861 | } | |
862 | ||
f523291e | 863 | /* Given a multi-line string, return a null-terminated copy of the current line. */ |
ccb4cabe SH |
864 | static char *copy_to_eol(char *p) |
865 | { | |
235f1815 | 866 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
867 | size_t len; |
868 | ||
869 | if (!p2) | |
870 | return NULL; | |
871 | ||
872 | len = p2 - p; | |
873 | sret = must_alloc(len + 1); | |
874 | memcpy(sret, p, len); | |
875 | sret[len] = '\0'; | |
876 | return sret; | |
877 | } | |
878 | ||
bced39de CB |
879 | /* cgline: pointer to character after the first ':' in a line in a \n-terminated |
880 | * /proc/self/cgroup file. Check whether controller c is present. | |
ccb4cabe SH |
881 | */ |
882 | static bool controller_in_clist(char *cgline, char *c) | |
883 | { | |
0be0d78f | 884 | char *tok, *eol, *tmp; |
ccb4cabe SH |
885 | size_t len; |
886 | ||
235f1815 | 887 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
888 | if (!eol) |
889 | return false; | |
890 | ||
891 | len = eol - cgline; | |
892 | tmp = alloca(len + 1); | |
893 | memcpy(tmp, cgline, len); | |
894 | tmp[len] = '\0'; | |
895 | ||
0be0d78f | 896 | lxc_iterate_parts(tok, tmp, ",") { |
ccb4cabe SH |
897 | if (strcmp(tok, c) == 0) |
898 | return true; | |
899 | } | |
d6337a5f | 900 | |
ccb4cabe SH |
901 | return false; |
902 | } | |
903 | ||
c3ef912e CB |
904 | /* @basecginfo is a copy of /proc/$$/cgroup. Return the current cgroup for |
905 | * @controller. | |
ccb4cabe | 906 | */ |
c3ef912e CB |
907 | static char *cg_hybrid_get_current_cgroup(char *basecginfo, char *controller, |
908 | int type) | |
ccb4cabe SH |
909 | { |
910 | char *p = basecginfo; | |
6328fd9c | 911 | |
d6337a5f CB |
912 | for (;;) { |
913 | bool is_cgv2_base_cgroup = false; | |
914 | ||
6328fd9c | 915 | /* cgroup v2 entry in "/proc/<pid>/cgroup": "0::/some/path" */ |
d6337a5f CB |
916 | if ((type == CGROUP2_SUPER_MAGIC) && (*p == '0')) |
917 | is_cgv2_base_cgroup = true; | |
ccb4cabe | 918 | |
235f1815 | 919 | p = strchr(p, ':'); |
ccb4cabe SH |
920 | if (!p) |
921 | return NULL; | |
922 | p++; | |
d6337a5f CB |
923 | |
924 | if (is_cgv2_base_cgroup || (controller && controller_in_clist(p, controller))) { | |
235f1815 | 925 | p = strchr(p, ':'); |
ccb4cabe SH |
926 | if (!p) |
927 | return NULL; | |
928 | p++; | |
929 | return copy_to_eol(p); | |
930 | } | |
931 | ||
235f1815 | 932 | p = strchr(p, '\n'); |
ccb4cabe SH |
933 | if (!p) |
934 | return NULL; | |
935 | p++; | |
936 | } | |
937 | } | |
938 | ||
ccb4cabe SH |
939 | static void must_append_string(char ***list, char *entry) |
940 | { | |
6dfb18bf | 941 | int newentry; |
ccb4cabe SH |
942 | char *copy; |
943 | ||
6dfb18bf | 944 | newentry = append_null_to_list((void ***)list); |
ccb4cabe SH |
945 | copy = must_copy_string(entry); |
946 | (*list)[newentry] = copy; | |
947 | } | |
948 | ||
d6337a5f | 949 | static int get_existing_subsystems(char ***klist, char ***nlist) |
ccb4cabe SH |
950 | { |
951 | FILE *f; | |
952 | char *line = NULL; | |
953 | size_t len = 0; | |
954 | ||
d6337a5f CB |
955 | f = fopen("/proc/self/cgroup", "r"); |
956 | if (!f) | |
957 | return -1; | |
958 | ||
ccb4cabe | 959 | while (getline(&line, &len, f) != -1) { |
0be0d78f | 960 | char *p, *p2, *tok; |
235f1815 | 961 | p = strchr(line, ':'); |
ccb4cabe SH |
962 | if (!p) |
963 | continue; | |
964 | p++; | |
235f1815 | 965 | p2 = strchr(p, ':'); |
ccb4cabe SH |
966 | if (!p2) |
967 | continue; | |
968 | *p2 = '\0'; | |
ff8d6ee9 | 969 | |
6328fd9c CB |
970 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup |
971 | * contains an entry of the form: | |
ff8d6ee9 CB |
972 | * |
973 | * 0::/some/path | |
974 | * | |
6328fd9c | 975 | * In this case we use "cgroup2" as controller name. |
ff8d6ee9 | 976 | */ |
6328fd9c CB |
977 | if ((p2 - p) == 0) { |
978 | must_append_string(klist, "cgroup2"); | |
ff8d6ee9 | 979 | continue; |
6328fd9c | 980 | } |
ff8d6ee9 | 981 | |
0be0d78f | 982 | lxc_iterate_parts(tok, p, ",") { |
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 | ||
bb221ad1 | 1019 | TRACE(" %d: base_cgroup: %s", i, (*it)->container_base_path ? (*it)->container_base_path : "(null)"); |
2202afc9 CB |
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 CB |
1043 | static int cgroup_rmdir(struct hierarchy **hierarchies, |
1044 | const char *container_cgroup) | |
c71d83e1 | 1045 | { |
2202afc9 | 1046 | int i; |
d6337a5f | 1047 | |
2202afc9 CB |
1048 | if (!container_cgroup || !hierarchies) |
1049 | return 0; | |
d6337a5f | 1050 | |
2202afc9 CB |
1051 | for (i = 0; hierarchies[i]; i++) { |
1052 | int ret; | |
1053 | struct hierarchy *h = hierarchies[i]; | |
d6337a5f | 1054 | |
eb697136 | 1055 | if (!h->container_full_path) |
2202afc9 CB |
1056 | continue; |
1057 | ||
eb697136 | 1058 | ret = recursive_destroy(h->container_full_path); |
2202afc9 | 1059 | if (ret < 0) |
eb697136 | 1060 | WARN("Failed to destroy \"%s\"", h->container_full_path); |
2202afc9 | 1061 | |
eb697136 CB |
1062 | free(h->container_full_path); |
1063 | h->container_full_path = NULL; | |
2202afc9 | 1064 | } |
d6337a5f | 1065 | |
c71d83e1 | 1066 | return 0; |
d6337a5f CB |
1067 | } |
1068 | ||
2202afc9 CB |
1069 | struct generic_userns_exec_data { |
1070 | struct hierarchy **hierarchies; | |
1071 | const char *container_cgroup; | |
1072 | struct lxc_conf *conf; | |
1073 | uid_t origuid; /* target uid in parent namespace */ | |
1074 | char *path; | |
1075 | }; | |
d6337a5f | 1076 | |
2202afc9 CB |
1077 | static int cgroup_rmdir_wrapper(void *data) |
1078 | { | |
1079 | int ret; | |
1080 | struct generic_userns_exec_data *arg = data; | |
1081 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1082 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
d6337a5f | 1083 | |
2202afc9 CB |
1084 | ret = setresgid(nsgid, nsgid, nsgid); |
1085 | if (ret < 0) { | |
1086 | SYSERROR("Failed to setresgid(%d, %d, %d)", (int)nsgid, | |
1087 | (int)nsgid, (int)nsgid); | |
1088 | return -1; | |
1089 | } | |
d6337a5f | 1090 | |
2202afc9 CB |
1091 | ret = setresuid(nsuid, nsuid, nsuid); |
1092 | if (ret < 0) { | |
1093 | SYSERROR("Failed to setresuid(%d, %d, %d)", (int)nsuid, | |
1094 | (int)nsuid, (int)nsuid); | |
1095 | return -1; | |
1096 | } | |
d6337a5f | 1097 | |
2202afc9 CB |
1098 | ret = setgroups(0, NULL); |
1099 | if (ret < 0 && errno != EPERM) { | |
1100 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1101 | return -1; | |
1102 | } | |
d6337a5f | 1103 | |
2202afc9 | 1104 | return cgroup_rmdir(arg->hierarchies, arg->container_cgroup); |
d6337a5f CB |
1105 | } |
1106 | ||
fb55e009 | 1107 | __cgfsng_ops__ static void cgfsng_destroy(struct cgroup_ops *ops, struct lxc_handler *handler) |
d6337a5f CB |
1108 | { |
1109 | int ret; | |
2202afc9 | 1110 | struct generic_userns_exec_data wrap; |
bd8ef4e4 | 1111 | |
4160c3a0 | 1112 | wrap.origuid = 0; |
2202afc9 CB |
1113 | wrap.container_cgroup = ops->container_cgroup; |
1114 | wrap.hierarchies = ops->hierarchies; | |
1115 | wrap.conf = handler->conf; | |
4160c3a0 | 1116 | |
2202afc9 CB |
1117 | if (handler->conf && !lxc_list_empty(&handler->conf->id_map)) |
1118 | ret = userns_exec_1(handler->conf, cgroup_rmdir_wrapper, &wrap, | |
bd8ef4e4 | 1119 | "cgroup_rmdir_wrapper"); |
ccb4cabe | 1120 | else |
2202afc9 | 1121 | ret = cgroup_rmdir(ops->hierarchies, ops->container_cgroup); |
bd8ef4e4 CB |
1122 | if (ret < 0) { |
1123 | WARN("Failed to destroy cgroups"); | |
ccb4cabe | 1124 | return; |
ccb4cabe | 1125 | } |
ccb4cabe SH |
1126 | } |
1127 | ||
a3926f6a | 1128 | static bool cg_unified_create_cgroup(struct hierarchy *h, char *cgname) |
0c3deb94 | 1129 | { |
0c3deb94 | 1130 | size_t i, parts_len; |
389d44ec | 1131 | char **it; |
0c3deb94 CB |
1132 | size_t full_len = 0; |
1133 | char *add_controllers = NULL, *cgroup = NULL; | |
1134 | char **parts = NULL; | |
1135 | bool bret = false; | |
1136 | ||
1137 | if (h->version != CGROUP2_SUPER_MAGIC) | |
1138 | return true; | |
1139 | ||
1140 | if (!h->controllers) | |
1141 | return true; | |
1142 | ||
1143 | /* For now we simply enable all controllers that we have detected by | |
1144 | * creating a string like "+memory +pids +cpu +io". | |
1145 | * TODO: In the near future we might want to support "-<controller>" | |
1146 | * etc. but whether supporting semantics like this make sense will need | |
1147 | * some thinking. | |
1148 | */ | |
1149 | for (it = h->controllers; it && *it; it++) { | |
64e82f8b DJ |
1150 | full_len += strlen(*it) + 2; |
1151 | add_controllers = must_realloc(add_controllers, full_len + 1); | |
1152 | ||
1153 | if (h->controllers[0] == *it) | |
1154 | add_controllers[0] = '\0'; | |
1155 | ||
3ebe2fbd DJ |
1156 | (void)strlcat(add_controllers, "+", full_len + 1); |
1157 | (void)strlcat(add_controllers, *it, full_len + 1); | |
64e82f8b DJ |
1158 | |
1159 | if ((it + 1) && *(it + 1)) | |
3ebe2fbd | 1160 | (void)strlcat(add_controllers, " ", full_len + 1); |
0c3deb94 CB |
1161 | } |
1162 | ||
1163 | parts = lxc_string_split(cgname, '/'); | |
1164 | if (!parts) | |
1165 | goto on_error; | |
64e82f8b | 1166 | |
0c3deb94 CB |
1167 | parts_len = lxc_array_len((void **)parts); |
1168 | if (parts_len > 0) | |
1169 | parts_len--; | |
1170 | ||
bb221ad1 | 1171 | cgroup = must_make_path(h->mountpoint, h->container_base_path, NULL); |
0c3deb94 CB |
1172 | for (i = 0; i < parts_len; i++) { |
1173 | int ret; | |
1174 | char *target; | |
1175 | ||
1176 | cgroup = must_append_path(cgroup, parts[i], NULL); | |
1177 | target = must_make_path(cgroup, "cgroup.subtree_control", NULL); | |
7cea5905 | 1178 | ret = lxc_write_to_file(target, add_controllers, full_len, false, 0666); |
0c3deb94 CB |
1179 | free(target); |
1180 | if (ret < 0) { | |
1181 | SYSERROR("Could not enable \"%s\" controllers in the " | |
1182 | "unified cgroup \"%s\"", add_controllers, cgroup); | |
1183 | goto on_error; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | bret = true; | |
1188 | ||
1189 | on_error: | |
1190 | lxc_free_array((void **)parts, free); | |
1191 | free(add_controllers); | |
1192 | free(cgroup); | |
1193 | return bret; | |
1194 | } | |
1195 | ||
ccb4cabe SH |
1196 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) |
1197 | { | |
0c3deb94 CB |
1198 | int ret; |
1199 | ||
bb221ad1 | 1200 | h->container_full_path = must_make_path(h->mountpoint, h->container_base_path, cgname, NULL); |
eb697136 CB |
1201 | if (dir_exists(h->container_full_path)) { |
1202 | ERROR("The cgroup \"%s\" already existed", h->container_full_path); | |
d8da679e | 1203 | return false; |
6f9584d8 | 1204 | } |
0c3deb94 | 1205 | |
a3926f6a | 1206 | if (!cg_legacy_handle_cpuset_hierarchy(h, cgname)) { |
4b4205e3 | 1207 | ERROR("Failed to handle legacy cpuset controller"); |
0c3deb94 CB |
1208 | return false; |
1209 | } | |
1210 | ||
eb697136 | 1211 | ret = mkdir_p(h->container_full_path, 0755); |
0c3deb94 | 1212 | if (ret < 0) { |
eb697136 | 1213 | ERROR("Failed to create cgroup \"%s\"", h->container_full_path); |
e3a3fecf | 1214 | return false; |
6f9584d8 | 1215 | } |
0c3deb94 | 1216 | |
a3926f6a | 1217 | return cg_unified_create_cgroup(h, cgname); |
ccb4cabe SH |
1218 | } |
1219 | ||
1220 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1221 | { | |
e56639fb CB |
1222 | int ret; |
1223 | ||
eb697136 | 1224 | ret = rmdir(h->container_full_path); |
e56639fb | 1225 | if (ret < 0) |
eb697136 | 1226 | SYSERROR("Failed to rmdir(\"%s\") from failed creation attempt", h->container_full_path); |
e56639fb | 1227 | |
eb697136 CB |
1228 | free(h->container_full_path); |
1229 | h->container_full_path = NULL; | |
ccb4cabe SH |
1230 | } |
1231 | ||
cecad0c1 CB |
1232 | /* Try to create the same cgroup in all hierarchies. Start with cgroup_pattern; |
1233 | * next cgroup_pattern-1, -2, ..., -999. | |
ccb4cabe | 1234 | */ |
fb55e009 CB |
1235 | __cgfsng_ops__ static inline bool cgfsng_create(struct cgroup_ops *ops, |
1236 | struct lxc_handler *handler) | |
ccb4cabe | 1237 | { |
bb30b52a | 1238 | int i; |
ccb4cabe | 1239 | size_t len; |
0c3deb94 | 1240 | char *container_cgroup, *offset, *tmp; |
7d531e9b | 1241 | int idx = 0; |
2202afc9 | 1242 | struct lxc_conf *conf = handler->conf; |
ccb4cabe | 1243 | |
2202afc9 CB |
1244 | if (ops->container_cgroup) { |
1245 | WARN("cgfsng_create called a second time: %s", ops->container_cgroup); | |
ccb4cabe | 1246 | return false; |
2202afc9 | 1247 | } |
43654d34 | 1248 | |
2202afc9 | 1249 | if (!conf) |
ccb4cabe | 1250 | return false; |
ccb4cabe | 1251 | |
2202afc9 | 1252 | if (conf->cgroup_meta.dir) |
3ec12d39 | 1253 | tmp = lxc_string_join("/", (const char *[]){conf->cgroup_meta.dir, handler->name, NULL}, false); |
43654d34 | 1254 | else |
2202afc9 | 1255 | tmp = lxc_string_replace("%n", handler->name, ops->cgroup_pattern); |
ccb4cabe SH |
1256 | if (!tmp) { |
1257 | ERROR("Failed expanding cgroup name pattern"); | |
1258 | return false; | |
1259 | } | |
64e82f8b | 1260 | |
1a0e70ac | 1261 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
0c3deb94 | 1262 | container_cgroup = must_alloc(len); |
64e82f8b | 1263 | (void)strlcpy(container_cgroup, tmp, len); |
ccb4cabe | 1264 | free(tmp); |
0c3deb94 | 1265 | offset = container_cgroup + len - 5; |
ccb4cabe SH |
1266 | |
1267 | again: | |
95adfe93 SH |
1268 | if (idx == 1000) { |
1269 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1270 | goto out_free; |
95adfe93 | 1271 | } |
cecad0c1 | 1272 | |
66b66624 | 1273 | if (idx) { |
bb30b52a CB |
1274 | int ret; |
1275 | ||
66b66624 CB |
1276 | ret = snprintf(offset, 5, "-%d", idx); |
1277 | if (ret < 0 || (size_t)ret >= 5) { | |
1278 | FILE *f = fopen("/dev/null", "w"); | |
97ebced3 | 1279 | if (f) { |
66b66624 CB |
1280 | fprintf(f, "Workaround for GCC7 bug: " |
1281 | "https://gcc.gnu.org/bugzilla/" | |
1282 | "show_bug.cgi?id=78969"); | |
1283 | fclose(f); | |
1284 | } | |
1285 | } | |
1286 | } | |
cecad0c1 | 1287 | |
2202afc9 CB |
1288 | for (i = 0; ops->hierarchies[i]; i++) { |
1289 | if (!create_path_for_hierarchy(ops->hierarchies[i], container_cgroup)) { | |
ccb4cabe | 1290 | int j; |
eb697136 CB |
1291 | ERROR("Failed to create cgroup \"%s\"", ops->hierarchies[i]->container_full_path); |
1292 | free(ops->hierarchies[i]->container_full_path); | |
1293 | ops->hierarchies[i]->container_full_path = NULL; | |
ccb4cabe | 1294 | for (j = 0; j < i; j++) |
2202afc9 | 1295 | remove_path_for_hierarchy(ops->hierarchies[j], container_cgroup); |
ccb4cabe SH |
1296 | idx++; |
1297 | goto again; | |
1298 | } | |
1299 | } | |
cecad0c1 | 1300 | |
2202afc9 | 1301 | ops->container_cgroup = container_cgroup; |
cecad0c1 | 1302 | |
ccb4cabe SH |
1303 | return true; |
1304 | ||
1305 | out_free: | |
0c3deb94 | 1306 | free(container_cgroup); |
cecad0c1 | 1307 | |
ccb4cabe SH |
1308 | return false; |
1309 | } | |
1310 | ||
fb55e009 | 1311 | __cgfsng_ops__ static bool cgfsng_enter(struct cgroup_ops *ops, pid_t pid) |
ccb4cabe | 1312 | { |
ccb4cabe | 1313 | int i, len; |
08768001 | 1314 | char pidstr[25]; |
ccb4cabe SH |
1315 | |
1316 | len = snprintf(pidstr, 25, "%d", pid); | |
08768001 | 1317 | if (len < 0 || len >= 25) |
ccb4cabe SH |
1318 | return false; |
1319 | ||
2202afc9 | 1320 | for (i = 0; ops->hierarchies[i]; i++) { |
08768001 CB |
1321 | int ret; |
1322 | char *fullpath; | |
1323 | ||
eb697136 | 1324 | fullpath = must_make_path(ops->hierarchies[i]->container_full_path, |
08768001 | 1325 | "cgroup.procs", NULL); |
7cea5905 | 1326 | ret = lxc_write_to_file(fullpath, pidstr, len, false, 0666); |
08768001 CB |
1327 | if (ret != 0) { |
1328 | SYSERROR("Failed to enter cgroup \"%s\"", fullpath); | |
ccb4cabe SH |
1329 | free(fullpath); |
1330 | return false; | |
1331 | } | |
1332 | free(fullpath); | |
1333 | } | |
1334 | ||
1335 | return true; | |
1336 | } | |
1337 | ||
6efacf80 CB |
1338 | static int chowmod(char *path, uid_t chown_uid, gid_t chown_gid, |
1339 | mode_t chmod_mode) | |
1340 | { | |
1341 | int ret; | |
1342 | ||
1343 | ret = chown(path, chown_uid, chown_gid); | |
1344 | if (ret < 0) { | |
a24c5678 | 1345 | SYSWARN("Failed to chown(%s, %d, %d)", path, (int)chown_uid, (int)chown_gid); |
6efacf80 CB |
1346 | return -1; |
1347 | } | |
1348 | ||
1349 | ret = chmod(path, chmod_mode); | |
1350 | if (ret < 0) { | |
a24c5678 | 1351 | SYSWARN("Failed to chmod(%s, %d)", path, (int)chmod_mode); |
6efacf80 CB |
1352 | return -1; |
1353 | } | |
1354 | ||
1355 | return 0; | |
1356 | } | |
1357 | ||
1358 | /* chgrp the container cgroups to container group. We leave | |
c0888dfe SH |
1359 | * the container owner as cgroup owner. So we must make the |
1360 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1361 | * |
1362 | * Also chown the tasks and cgroup.procs files. Those may not | |
1363 | * exist depending on kernel version. | |
c0888dfe | 1364 | */ |
ccb4cabe SH |
1365 | static int chown_cgroup_wrapper(void *data) |
1366 | { | |
6efacf80 | 1367 | int i, ret; |
4160c3a0 CB |
1368 | uid_t destuid; |
1369 | struct generic_userns_exec_data *arg = data; | |
1370 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1371 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
ccb4cabe | 1372 | |
6efacf80 CB |
1373 | ret = setresgid(nsgid, nsgid, nsgid); |
1374 | if (ret < 0) { | |
1375 | SYSERROR("Failed to setresgid(%d, %d, %d)", | |
1376 | (int)nsgid, (int)nsgid, (int)nsgid); | |
1377 | return -1; | |
1378 | } | |
1379 | ||
1380 | ret = setresuid(nsuid, nsuid, nsuid); | |
1381 | if (ret < 0) { | |
1382 | SYSERROR("Failed to setresuid(%d, %d, %d)", | |
1383 | (int)nsuid, (int)nsuid, (int)nsuid); | |
1384 | return -1; | |
1385 | } | |
1386 | ||
1387 | ret = setgroups(0, NULL); | |
1388 | if (ret < 0 && errno != EPERM) { | |
1389 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1390 | return -1; | |
1391 | } | |
ccb4cabe SH |
1392 | |
1393 | destuid = get_ns_uid(arg->origuid); | |
b962868f CB |
1394 | if (destuid == LXC_INVALID_UID) |
1395 | destuid = 0; | |
ccb4cabe | 1396 | |
2202afc9 | 1397 | for (i = 0; arg->hierarchies[i]; i++) { |
6efacf80 | 1398 | char *fullpath; |
eb697136 | 1399 | char *path = arg->hierarchies[i]->container_full_path; |
43647298 | 1400 | |
63e42fee | 1401 | ret = chowmod(path, destuid, nsgid, 0775); |
6efacf80 | 1402 | if (ret < 0) |
ccb4cabe | 1403 | return -1; |
c0888dfe | 1404 | |
6efacf80 CB |
1405 | /* Failures to chown() these are inconvenient but not |
1406 | * detrimental We leave these owned by the container launcher, | |
1407 | * so that container root can write to the files to attach. We | |
1408 | * chmod() them 664 so that container systemd can write to the | |
1409 | * files (which systemd in wily insists on doing). | |
ab8f5424 | 1410 | */ |
6efacf80 | 1411 | |
2202afc9 | 1412 | if (arg->hierarchies[i]->version == CGROUP_SUPER_MAGIC) { |
6efacf80 CB |
1413 | fullpath = must_make_path(path, "tasks", NULL); |
1414 | (void)chowmod(fullpath, destuid, nsgid, 0664); | |
1415 | free(fullpath); | |
1416 | } | |
43647298 SH |
1417 | |
1418 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
2202afc9 | 1419 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
ccb4cabe | 1420 | free(fullpath); |
0e17357c | 1421 | |
2202afc9 | 1422 | if (arg->hierarchies[i]->version != CGROUP2_SUPER_MAGIC) |
0e17357c CB |
1423 | continue; |
1424 | ||
1425 | fullpath = must_make_path(path, "cgroup.subtree_control", NULL); | |
6efacf80 | 1426 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c CB |
1427 | free(fullpath); |
1428 | ||
1429 | fullpath = must_make_path(path, "cgroup.threads", NULL); | |
6efacf80 | 1430 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c | 1431 | free(fullpath); |
ccb4cabe SH |
1432 | } |
1433 | ||
1434 | return 0; | |
1435 | } | |
1436 | ||
fb55e009 CB |
1437 | __cgfsng_ops__ static bool cgfsng_chown(struct cgroup_ops *ops, |
1438 | struct lxc_conf *conf) | |
ccb4cabe | 1439 | { |
4160c3a0 | 1440 | struct generic_userns_exec_data wrap; |
ccb4cabe | 1441 | |
ccb4cabe SH |
1442 | if (lxc_list_empty(&conf->id_map)) |
1443 | return true; | |
1444 | ||
ccb4cabe | 1445 | wrap.origuid = geteuid(); |
4160c3a0 | 1446 | wrap.path = NULL; |
2202afc9 | 1447 | wrap.hierarchies = ops->hierarchies; |
4160c3a0 | 1448 | wrap.conf = conf; |
ccb4cabe | 1449 | |
c9b7c33e CB |
1450 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1451 | "chown_cgroup_wrapper") < 0) { | |
f7faba6c | 1452 | ERROR("Error requesting cgroup chown in new user namespace"); |
ccb4cabe SH |
1453 | return false; |
1454 | } | |
1455 | ||
1456 | return true; | |
1457 | } | |
1458 | ||
8aa1044f SH |
1459 | /* cgroup-full:* is done, no need to create subdirs */ |
1460 | static bool cg_mount_needs_subdirs(int type) | |
1461 | { | |
1462 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1463 | return false; | |
a3926f6a | 1464 | |
8aa1044f SH |
1465 | return true; |
1466 | } | |
1467 | ||
886cac86 CB |
1468 | /* After $rootfs/sys/fs/container/controller/the/cg/path has been created, |
1469 | * remount controller ro if needed and bindmount the cgroupfs onto | |
1470 | * controll/the/cg/path. | |
8aa1044f | 1471 | */ |
6812d833 CB |
1472 | static int cg_legacy_mount_controllers(int type, struct hierarchy *h, |
1473 | char *controllerpath, char *cgpath, | |
1474 | const char *container_cgroup) | |
8aa1044f | 1475 | { |
5285689c | 1476 | int ret, remount_flags; |
886cac86 CB |
1477 | char *sourcepath; |
1478 | int flags = MS_BIND; | |
1479 | ||
8aa1044f | 1480 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { |
886cac86 CB |
1481 | ret = mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL); |
1482 | if (ret < 0) { | |
1483 | SYSERROR("Failed to bind mount \"%s\" onto \"%s\"", | |
1484 | controllerpath, controllerpath); | |
8aa1044f SH |
1485 | return -1; |
1486 | } | |
886cac86 | 1487 | |
5285689c CB |
1488 | remount_flags = add_required_remount_flags(controllerpath, |
1489 | controllerpath, | |
1490 | flags | MS_REMOUNT); | |
886cac86 | 1491 | ret = mount(controllerpath, controllerpath, "cgroup", |
8186c5c7 CB |
1492 | remount_flags | MS_REMOUNT | MS_BIND | MS_RDONLY, |
1493 | NULL); | |
886cac86 CB |
1494 | if (ret < 0) { |
1495 | SYSERROR("Failed to remount \"%s\" ro", controllerpath); | |
8aa1044f SH |
1496 | return -1; |
1497 | } | |
886cac86 | 1498 | |
8aa1044f SH |
1499 | INFO("Remounted %s read-only", controllerpath); |
1500 | } | |
886cac86 | 1501 | |
bb221ad1 | 1502 | sourcepath = must_make_path(h->mountpoint, h->container_base_path, |
886cac86 | 1503 | container_cgroup, NULL); |
8aa1044f SH |
1504 | if (type == LXC_AUTO_CGROUP_RO) |
1505 | flags |= MS_RDONLY; | |
886cac86 CB |
1506 | |
1507 | ret = mount(sourcepath, cgpath, "cgroup", flags, NULL); | |
1508 | if (ret < 0) { | |
1509 | SYSERROR("Failed to mount \"%s\" onto \"%s\"", h->controllers[0], cgpath); | |
8aa1044f | 1510 | free(sourcepath); |
8aa1044f SH |
1511 | return -1; |
1512 | } | |
886cac86 | 1513 | INFO("Mounted \"%s\" onto \"%s\"", h->controllers[0], cgpath); |
f8c40ffa L |
1514 | |
1515 | if (flags & MS_RDONLY) { | |
5285689c CB |
1516 | remount_flags = add_required_remount_flags(sourcepath, cgpath, |
1517 | flags | MS_REMOUNT); | |
1518 | ret = mount(sourcepath, cgpath, "cgroup", remount_flags, NULL); | |
886cac86 CB |
1519 | if (ret < 0) { |
1520 | SYSERROR("Failed to remount \"%s\" ro", cgpath); | |
f8c40ffa | 1521 | free(sourcepath); |
f8c40ffa L |
1522 | return -1; |
1523 | } | |
5285689c | 1524 | INFO("Remounted %s read-only", cgpath); |
f8c40ffa L |
1525 | } |
1526 | ||
8aa1044f | 1527 | free(sourcepath); |
886cac86 | 1528 | INFO("Completed second stage cgroup automounts for \"%s\"", cgpath); |
8aa1044f SH |
1529 | return 0; |
1530 | } | |
1531 | ||
6812d833 CB |
1532 | /* __cg_mount_direct |
1533 | * | |
1534 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
1535 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
1536 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
1537 | */ | |
1538 | static int __cg_mount_direct(int type, struct hierarchy *h, | |
1539 | const char *controllerpath) | |
b635e92d CB |
1540 | { |
1541 | int ret; | |
1542 | char *controllers = NULL; | |
a760603e CB |
1543 | char *fstype = "cgroup2"; |
1544 | unsigned long flags = 0; | |
b635e92d | 1545 | |
a760603e CB |
1546 | flags |= MS_NOSUID; |
1547 | flags |= MS_NOEXEC; | |
1548 | flags |= MS_NODEV; | |
1549 | flags |= MS_RELATIME; | |
1550 | ||
1551 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_FULL_RO) | |
1552 | flags |= MS_RDONLY; | |
1553 | ||
d6337a5f | 1554 | if (h->version != CGROUP2_SUPER_MAGIC) { |
a760603e CB |
1555 | controllers = lxc_string_join(",", (const char **)h->controllers, false); |
1556 | if (!controllers) | |
1557 | return -ENOMEM; | |
1558 | fstype = "cgroup"; | |
b635e92d CB |
1559 | } |
1560 | ||
a760603e | 1561 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); |
b635e92d CB |
1562 | free(controllers); |
1563 | if (ret < 0) { | |
6812d833 | 1564 | SYSERROR("Failed to mount \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1565 | return -1; |
1566 | } | |
1567 | ||
6812d833 | 1568 | DEBUG("Mounted \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1569 | return 0; |
1570 | } | |
1571 | ||
6812d833 CB |
1572 | static inline int cg_mount_in_cgroup_namespace(int type, struct hierarchy *h, |
1573 | const char *controllerpath) | |
1574 | { | |
1575 | return __cg_mount_direct(type, h, controllerpath); | |
1576 | } | |
1577 | ||
1578 | static inline int cg_mount_cgroup_full(int type, struct hierarchy *h, | |
1579 | const char *controllerpath) | |
1580 | { | |
1581 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1582 | return 0; | |
1583 | ||
1584 | return __cg_mount_direct(type, h, controllerpath); | |
1585 | } | |
1586 | ||
fb55e009 CB |
1587 | __cgfsng_ops__ static bool cgfsng_mount(struct cgroup_ops *ops, |
1588 | struct lxc_handler *handler, | |
1589 | const char *root, int type) | |
ccb4cabe | 1590 | { |
3f69fb12 | 1591 | int i, ret; |
8aa1044f | 1592 | char *tmpfspath = NULL; |
affd10fa | 1593 | bool has_cgns = false, retval = false, wants_force_mount = false; |
8aa1044f SH |
1594 | |
1595 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1596 | return true; | |
1597 | ||
3f69fb12 SY |
1598 | if (type & LXC_AUTO_CGROUP_FORCE) { |
1599 | type &= ~LXC_AUTO_CGROUP_FORCE; | |
1600 | wants_force_mount = true; | |
1601 | } | |
b635e92d | 1602 | |
3f69fb12 SY |
1603 | if (!wants_force_mount){ |
1604 | if (!lxc_list_empty(&handler->conf->keepcaps)) | |
1605 | wants_force_mount = !in_caplist(CAP_SYS_ADMIN, &handler->conf->keepcaps); | |
1606 | else | |
1607 | wants_force_mount = in_caplist(CAP_SYS_ADMIN, &handler->conf->caps); | |
1608 | } | |
8aa1044f | 1609 | |
3f69fb12 SY |
1610 | has_cgns = cgns_supported(); |
1611 | if (has_cgns && !wants_force_mount) | |
1612 | return true; | |
8aa1044f SH |
1613 | |
1614 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1615 | type = LXC_AUTO_CGROUP_MIXED; | |
1616 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1617 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1618 | ||
1619 | /* Mount tmpfs */ | |
3f69fb12 | 1620 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); |
6812d833 | 1621 | ret = safe_mount(NULL, tmpfspath, "tmpfs", |
3f69fb12 SY |
1622 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, |
1623 | "size=10240k,mode=755", root); | |
1624 | if (ret < 0) | |
1625 | goto on_error; | |
8aa1044f | 1626 | |
2202afc9 | 1627 | for (i = 0; ops->hierarchies[i]; i++) { |
8aa1044f | 1628 | char *controllerpath, *path2; |
2202afc9 | 1629 | struct hierarchy *h = ops->hierarchies[i]; |
8aa1044f | 1630 | char *controller = strrchr(h->mountpoint, '/'); |
8aa1044f SH |
1631 | |
1632 | if (!controller) | |
1633 | continue; | |
1634 | controller++; | |
affd10fa | 1635 | |
8aa1044f SH |
1636 | controllerpath = must_make_path(tmpfspath, controller, NULL); |
1637 | if (dir_exists(controllerpath)) { | |
1638 | free(controllerpath); | |
1639 | continue; | |
1640 | } | |
affd10fa | 1641 | |
3f69fb12 SY |
1642 | ret = mkdir(controllerpath, 0755); |
1643 | if (ret < 0) { | |
8aa1044f SH |
1644 | SYSERROR("Error creating cgroup path: %s", controllerpath); |
1645 | free(controllerpath); | |
3f69fb12 | 1646 | goto on_error; |
8aa1044f | 1647 | } |
b635e92d | 1648 | |
3f69fb12 | 1649 | if (has_cgns && wants_force_mount) { |
b635e92d CB |
1650 | /* If cgroup namespaces are supported but the container |
1651 | * will not have CAP_SYS_ADMIN after it has started we | |
1652 | * need to mount the cgroups manually. | |
1653 | */ | |
3f69fb12 | 1654 | ret = cg_mount_in_cgroup_namespace(type, h, controllerpath); |
b635e92d | 1655 | free(controllerpath); |
3f69fb12 SY |
1656 | if (ret < 0) |
1657 | goto on_error; | |
1658 | ||
b635e92d CB |
1659 | continue; |
1660 | } | |
1661 | ||
6812d833 | 1662 | ret = cg_mount_cgroup_full(type, h, controllerpath); |
3f69fb12 | 1663 | if (ret < 0) { |
8aa1044f | 1664 | free(controllerpath); |
3f69fb12 | 1665 | goto on_error; |
8aa1044f | 1666 | } |
3f69fb12 | 1667 | |
8aa1044f SH |
1668 | if (!cg_mount_needs_subdirs(type)) { |
1669 | free(controllerpath); | |
1670 | continue; | |
1671 | } | |
3f69fb12 | 1672 | |
bb221ad1 | 1673 | path2 = must_make_path(controllerpath, h->container_base_path, |
2202afc9 | 1674 | ops->container_cgroup, NULL); |
3f69fb12 SY |
1675 | ret = mkdir_p(path2, 0755); |
1676 | if (ret < 0) { | |
8aa1044f | 1677 | free(controllerpath); |
8e0c6620 | 1678 | free(path2); |
3f69fb12 | 1679 | goto on_error; |
8aa1044f | 1680 | } |
2f62fb00 | 1681 | |
6812d833 | 1682 | ret = cg_legacy_mount_controllers(type, h, controllerpath, |
2202afc9 | 1683 | path2, ops->container_cgroup); |
8aa1044f SH |
1684 | free(controllerpath); |
1685 | free(path2); | |
3f69fb12 SY |
1686 | if (ret < 0) |
1687 | goto on_error; | |
8aa1044f SH |
1688 | } |
1689 | retval = true; | |
1690 | ||
3f69fb12 | 1691 | on_error: |
8aa1044f SH |
1692 | free(tmpfspath); |
1693 | return retval; | |
ccb4cabe SH |
1694 | } |
1695 | ||
1696 | static int recursive_count_nrtasks(char *dirname) | |
1697 | { | |
74f96976 | 1698 | struct dirent *direntp; |
ccb4cabe SH |
1699 | DIR *dir; |
1700 | int count = 0, ret; | |
1701 | char *path; | |
1702 | ||
1703 | dir = opendir(dirname); | |
1704 | if (!dir) | |
1705 | return 0; | |
1706 | ||
74f96976 | 1707 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1708 | struct stat mystat; |
1709 | ||
ccb4cabe SH |
1710 | if (!strcmp(direntp->d_name, ".") || |
1711 | !strcmp(direntp->d_name, "..")) | |
1712 | continue; | |
1713 | ||
1714 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1715 | ||
1716 | if (lstat(path, &mystat)) | |
1717 | goto next; | |
1718 | ||
1719 | if (!S_ISDIR(mystat.st_mode)) | |
1720 | goto next; | |
1721 | ||
1722 | count += recursive_count_nrtasks(path); | |
13c49955 | 1723 | next: |
ccb4cabe SH |
1724 | free(path); |
1725 | } | |
1726 | ||
1727 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1728 | ret = lxc_count_file_lines(path); | |
1729 | if (ret != -1) | |
1730 | count += ret; | |
1731 | free(path); | |
1732 | ||
13c49955 | 1733 | (void)closedir(dir); |
ccb4cabe SH |
1734 | |
1735 | return count; | |
1736 | } | |
1737 | ||
fb55e009 | 1738 | __cgfsng_ops__ static int cgfsng_nrtasks(struct cgroup_ops *ops) |
3135c5d4 | 1739 | { |
ccb4cabe | 1740 | int count; |
3135c5d4 | 1741 | char *path; |
ccb4cabe | 1742 | |
2202afc9 | 1743 | if (!ops->container_cgroup || !ops->hierarchies) |
ccb4cabe | 1744 | return -1; |
a3926f6a | 1745 | |
eb697136 | 1746 | path = must_make_path(ops->hierarchies[0]->container_full_path, NULL); |
ccb4cabe SH |
1747 | count = recursive_count_nrtasks(path); |
1748 | free(path); | |
1749 | return count; | |
1750 | } | |
1751 | ||
11c23867 | 1752 | /* Only root needs to escape to the cgroup of its init. */ |
fb55e009 CB |
1753 | __cgfsng_ops__ static bool cgfsng_escape(const struct cgroup_ops *ops, |
1754 | struct lxc_conf *conf) | |
ccb4cabe | 1755 | { |
ccb4cabe SH |
1756 | int i; |
1757 | ||
9caee129 | 1758 | if (conf->cgroup_meta.relative || geteuid()) |
ccb4cabe SH |
1759 | return true; |
1760 | ||
2202afc9 | 1761 | for (i = 0; ops->hierarchies[i]; i++) { |
11c23867 CB |
1762 | int ret; |
1763 | char *fullpath; | |
1764 | ||
2202afc9 | 1765 | fullpath = must_make_path(ops->hierarchies[i]->mountpoint, |
bb221ad1 | 1766 | ops->hierarchies[i]->container_base_path, |
11c23867 | 1767 | "cgroup.procs", NULL); |
7cea5905 | 1768 | ret = lxc_write_to_file(fullpath, "0", 2, false, 0666); |
11c23867 CB |
1769 | if (ret != 0) { |
1770 | SYSERROR("Failed to escape to cgroup \"%s\"", fullpath); | |
ccb4cabe | 1771 | free(fullpath); |
6df334d1 | 1772 | return false; |
ccb4cabe SH |
1773 | } |
1774 | free(fullpath); | |
1775 | } | |
1776 | ||
6df334d1 | 1777 | return true; |
ccb4cabe SH |
1778 | } |
1779 | ||
fb55e009 | 1780 | __cgfsng_ops__ static int cgfsng_num_hierarchies(struct cgroup_ops *ops) |
36662416 TA |
1781 | { |
1782 | int i; | |
1783 | ||
2202afc9 | 1784 | for (i = 0; ops->hierarchies[i]; i++) |
36662416 TA |
1785 | ; |
1786 | ||
1787 | return i; | |
1788 | } | |
1789 | ||
fb55e009 | 1790 | __cgfsng_ops__ static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) |
36662416 TA |
1791 | { |
1792 | int i; | |
1793 | ||
1794 | /* sanity check n */ | |
6b38e644 | 1795 | for (i = 0; i < n; i++) |
2202afc9 | 1796 | if (!ops->hierarchies[i]) |
36662416 | 1797 | return false; |
36662416 | 1798 | |
2202afc9 | 1799 | *out = ops->hierarchies[i]->controllers; |
36662416 TA |
1800 | |
1801 | return true; | |
1802 | } | |
1803 | ||
ccb4cabe SH |
1804 | #define THAWED "THAWED" |
1805 | #define THAWED_LEN (strlen(THAWED)) | |
1806 | ||
d6337a5f CB |
1807 | /* TODO: If the unified cgroup hierarchy grows a freezer controller this needs |
1808 | * to be adapted. | |
1809 | */ | |
fb55e009 | 1810 | __cgfsng_ops__ static bool cgfsng_unfreeze(struct cgroup_ops *ops) |
ccb4cabe | 1811 | { |
d6337a5f | 1812 | int ret; |
ccb4cabe | 1813 | char *fullpath; |
d6337a5f | 1814 | struct hierarchy *h; |
ccb4cabe | 1815 | |
2202afc9 | 1816 | h = get_hierarchy(ops, "freezer"); |
457ca9aa | 1817 | if (!h) |
ccb4cabe | 1818 | return false; |
d6337a5f | 1819 | |
eb697136 | 1820 | fullpath = must_make_path(h->container_full_path, "freezer.state", NULL); |
7cea5905 | 1821 | ret = lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false, 0666); |
ccb4cabe | 1822 | free(fullpath); |
d6337a5f CB |
1823 | if (ret < 0) |
1824 | return false; | |
1825 | ||
ccb4cabe SH |
1826 | return true; |
1827 | } | |
1828 | ||
fb55e009 CB |
1829 | __cgfsng_ops__ static const char *cgfsng_get_cgroup(struct cgroup_ops *ops, |
1830 | const char *controller) | |
ccb4cabe | 1831 | { |
d6337a5f CB |
1832 | struct hierarchy *h; |
1833 | ||
2202afc9 | 1834 | h = get_hierarchy(ops, controller); |
106f1f38 | 1835 | if (!h) { |
2202afc9 CB |
1836 | WARN("Failed to find hierarchy for controller \"%s\"", |
1837 | controller ? controller : "(null)"); | |
ccb4cabe | 1838 | return NULL; |
106f1f38 | 1839 | } |
ccb4cabe | 1840 | |
eb697136 | 1841 | return h->container_full_path ? h->container_full_path + strlen(h->mountpoint) : NULL; |
371f834d SH |
1842 | } |
1843 | ||
c40c8209 CB |
1844 | /* Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a full path, |
1845 | * which must be freed by the caller. | |
371f834d | 1846 | */ |
c40c8209 CB |
1847 | static inline char *build_full_cgpath_from_monitorpath(struct hierarchy *h, |
1848 | const char *inpath, | |
1849 | const char *filename) | |
371f834d | 1850 | { |
371f834d | 1851 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
1852 | } |
1853 | ||
25f66a8f CB |
1854 | /* Technically, we're always at a delegation boundary here (This is especially |
1855 | * true when cgroup namespaces are available.). The reasoning is that in order | |
c2aed66d | 1856 | * for us to have been able to start a container in the first place the root |
25f66a8f | 1857 | * cgroup must have been a leaf node. Now, either the container's init system |
c2aed66d CB |
1858 | * has populated the cgroup and kept it as a leaf node or it has created |
1859 | * subtrees. In the former case we will simply attach to the leaf node we | |
1860 | * created when we started the container in the latter case we create our own | |
1861 | * cgroup for the attaching process. | |
1862 | */ | |
a3926f6a CB |
1863 | static int __cg_unified_attach(const struct hierarchy *h, const char *name, |
1864 | const char *lxcpath, const char *pidstr, | |
1865 | size_t pidstr_len, const char *controller) | |
c2aed66d CB |
1866 | { |
1867 | int ret; | |
1868 | size_t len; | |
1869 | int fret = -1, idx = 0; | |
1870 | char *base_path = NULL, *container_cgroup = NULL, *full_path = NULL; | |
1871 | ||
1872 | container_cgroup = lxc_cmd_get_cgroup_path(name, lxcpath, controller); | |
1873 | /* not running */ | |
1874 | if (!container_cgroup) | |
1875 | return 0; | |
1876 | ||
1877 | base_path = must_make_path(h->mountpoint, container_cgroup, NULL); | |
1878 | full_path = must_make_path(base_path, "cgroup.procs", NULL); | |
1879 | /* cgroup is populated */ | |
7cea5905 | 1880 | ret = lxc_write_to_file(full_path, pidstr, pidstr_len, false, 0666); |
c2aed66d CB |
1881 | if (ret < 0 && errno != EBUSY) |
1882 | goto on_error; | |
1883 | ||
1884 | if (ret == 0) | |
1885 | goto on_success; | |
1886 | ||
1887 | free(full_path); | |
1888 | ||
1889 | len = strlen(base_path) + sizeof("/lxc-1000") - 1 + | |
1890 | sizeof("/cgroup-procs") - 1; | |
1891 | full_path = must_alloc(len + 1); | |
1892 | do { | |
1893 | if (idx) | |
1894 | ret = snprintf(full_path, len + 1, "%s/lxc-%d", | |
1895 | base_path, idx); | |
1896 | else | |
1897 | ret = snprintf(full_path, len + 1, "%s/lxc", base_path); | |
1898 | if (ret < 0 || (size_t)ret >= len + 1) | |
1899 | goto on_error; | |
1900 | ||
1901 | ret = mkdir_p(full_path, 0755); | |
1902 | if (ret < 0 && errno != EEXIST) | |
1903 | goto on_error; | |
1904 | ||
3ebe2fbd | 1905 | (void)strlcat(full_path, "/cgroup.procs", len + 1); |
7cea5905 | 1906 | ret = lxc_write_to_file(full_path, pidstr, len, false, 0666); |
c2aed66d CB |
1907 | if (ret == 0) |
1908 | goto on_success; | |
1909 | ||
1910 | /* this is a non-leaf node */ | |
1911 | if (errno != EBUSY) | |
1912 | goto on_error; | |
1913 | ||
1914 | } while (++idx > 0 && idx < 1000); | |
1915 | ||
1916 | on_success: | |
1917 | if (idx < 1000) | |
1918 | fret = 0; | |
1919 | ||
1920 | on_error: | |
1921 | free(base_path); | |
1922 | free(container_cgroup); | |
1923 | free(full_path); | |
1924 | ||
1925 | return fret; | |
1926 | } | |
1927 | ||
fb55e009 CB |
1928 | __cgfsng_ops__ static bool cgfsng_attach(struct cgroup_ops *ops, const char *name, |
1929 | const char *lxcpath, pid_t pid) | |
ccb4cabe | 1930 | { |
c2aed66d | 1931 | int i, len, ret; |
ccb4cabe | 1932 | char pidstr[25]; |
ccb4cabe SH |
1933 | |
1934 | len = snprintf(pidstr, 25, "%d", pid); | |
0cb10e11 | 1935 | if (len < 0 || len >= 25) |
ccb4cabe SH |
1936 | return false; |
1937 | ||
2202afc9 | 1938 | for (i = 0; ops->hierarchies[i]; i++) { |
c2aed66d CB |
1939 | char *path; |
1940 | char *fullpath = NULL; | |
2202afc9 | 1941 | struct hierarchy *h = ops->hierarchies[i]; |
ccb4cabe | 1942 | |
c2aed66d | 1943 | if (h->version == CGROUP2_SUPER_MAGIC) { |
a3926f6a CB |
1944 | ret = __cg_unified_attach(h, name, lxcpath, pidstr, len, |
1945 | h->controllers[0]); | |
c2aed66d CB |
1946 | if (ret < 0) |
1947 | return false; | |
1948 | ||
1949 | continue; | |
1950 | } | |
1951 | ||
ccb4cabe | 1952 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); |
c2aed66d CB |
1953 | /* not running */ |
1954 | if (!path) | |
ccb4cabe SH |
1955 | continue; |
1956 | ||
371f834d | 1957 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
71cb9afb | 1958 | free(path); |
7cea5905 | 1959 | ret = lxc_write_to_file(fullpath, pidstr, len, false, 0666); |
c2aed66d | 1960 | if (ret < 0) { |
ccb4cabe SH |
1961 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); |
1962 | free(fullpath); | |
ccb4cabe SH |
1963 | return false; |
1964 | } | |
ccb4cabe SH |
1965 | free(fullpath); |
1966 | } | |
1967 | ||
ccb4cabe SH |
1968 | return true; |
1969 | } | |
1970 | ||
e2bd2b13 CB |
1971 | /* Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. Here we |
1972 | * don't have a cgroup_data set up, so we ask the running container through the | |
1973 | * commands API for the cgroup path. | |
ccb4cabe | 1974 | */ |
fb55e009 CB |
1975 | __cgfsng_ops__ static int cgfsng_get(struct cgroup_ops *ops, const char *filename, |
1976 | char *value, size_t len, const char *name, | |
1977 | const char *lxcpath) | |
ccb4cabe | 1978 | { |
ccb4cabe | 1979 | int ret = -1; |
0069cc61 CB |
1980 | size_t controller_len; |
1981 | char *controller, *p, *path; | |
1982 | struct hierarchy *h; | |
ccb4cabe | 1983 | |
0069cc61 CB |
1984 | controller_len = strlen(filename); |
1985 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
1986 | (void)strlcpy(controller, filename, controller_len + 1); |
1987 | ||
0069cc61 CB |
1988 | p = strchr(controller, '.'); |
1989 | if (p) | |
ccb4cabe SH |
1990 | *p = '\0'; |
1991 | ||
0069cc61 CB |
1992 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
1993 | /* not running */ | |
1994 | if (!path) | |
ccb4cabe SH |
1995 | return -1; |
1996 | ||
2202afc9 | 1997 | h = get_hierarchy(ops, controller); |
ccb4cabe | 1998 | if (h) { |
0069cc61 CB |
1999 | char *fullpath; |
2000 | ||
2001 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
ccb4cabe SH |
2002 | ret = lxc_read_from_file(fullpath, value, len); |
2003 | free(fullpath); | |
2004 | } | |
ccb4cabe SH |
2005 | free(path); |
2006 | ||
2007 | return ret; | |
2008 | } | |
2009 | ||
eec533e3 CB |
2010 | /* Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. Here we |
2011 | * don't have a cgroup_data set up, so we ask the running container through the | |
2012 | * commands API for the cgroup path. | |
ccb4cabe | 2013 | */ |
fb55e009 CB |
2014 | __cgfsng_ops__ static int cgfsng_set(struct cgroup_ops *ops, |
2015 | const char *filename, const char *value, | |
2016 | const char *name, const char *lxcpath) | |
ccb4cabe | 2017 | { |
ccb4cabe | 2018 | int ret = -1; |
87777968 CB |
2019 | size_t controller_len; |
2020 | char *controller, *p, *path; | |
2021 | struct hierarchy *h; | |
ccb4cabe | 2022 | |
87777968 CB |
2023 | controller_len = strlen(filename); |
2024 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
2025 | (void)strlcpy(controller, filename, controller_len + 1); |
2026 | ||
87777968 CB |
2027 | p = strchr(controller, '.'); |
2028 | if (p) | |
ccb4cabe SH |
2029 | *p = '\0'; |
2030 | ||
87777968 CB |
2031 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2032 | /* not running */ | |
2033 | if (!path) | |
ccb4cabe SH |
2034 | return -1; |
2035 | ||
2202afc9 | 2036 | h = get_hierarchy(ops, controller); |
ccb4cabe | 2037 | if (h) { |
87777968 CB |
2038 | char *fullpath; |
2039 | ||
2040 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
7cea5905 | 2041 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
ccb4cabe SH |
2042 | free(fullpath); |
2043 | } | |
ccb4cabe SH |
2044 | free(path); |
2045 | ||
2046 | return ret; | |
2047 | } | |
2048 | ||
91d1a13a | 2049 | /* take devices cgroup line |
72add155 SH |
2050 | * /dev/foo rwx |
2051 | * and convert it to a valid | |
2052 | * type major:minor mode | |
91d1a13a CB |
2053 | * line. Return <0 on error. Dest is a preallocated buffer long enough to hold |
2054 | * the output. | |
72add155 SH |
2055 | */ |
2056 | static int convert_devpath(const char *invalue, char *dest) | |
2057 | { | |
2a06d041 CB |
2058 | int n_parts; |
2059 | char *p, *path, type; | |
72add155 | 2060 | unsigned long minor, major; |
91d1a13a | 2061 | struct stat sb; |
2a06d041 CB |
2062 | int ret = -EINVAL; |
2063 | char *mode = NULL; | |
72add155 SH |
2064 | |
2065 | path = must_copy_string(invalue); | |
2066 | ||
91d1a13a CB |
2067 | /* Read path followed by mode. Ignore any trailing text. |
2068 | * A ' # comment' would be legal. Technically other text is not | |
2069 | * legal, we could check for that if we cared to. | |
72add155 SH |
2070 | */ |
2071 | for (n_parts = 1, p = path; *p && n_parts < 3; p++) { | |
2c2d6c49 SH |
2072 | if (*p != ' ') |
2073 | continue; | |
2074 | *p = '\0'; | |
91d1a13a | 2075 | |
2c2d6c49 SH |
2076 | if (n_parts != 1) |
2077 | break; | |
2078 | p++; | |
2079 | n_parts++; | |
91d1a13a | 2080 | |
2c2d6c49 SH |
2081 | while (*p == ' ') |
2082 | p++; | |
91d1a13a | 2083 | |
2c2d6c49 | 2084 | mode = p; |
91d1a13a | 2085 | |
2c2d6c49 SH |
2086 | if (*p == '\0') |
2087 | goto out; | |
72add155 | 2088 | } |
2c2d6c49 SH |
2089 | |
2090 | if (n_parts == 1) | |
72add155 | 2091 | goto out; |
72add155 SH |
2092 | |
2093 | ret = stat(path, &sb); | |
2094 | if (ret < 0) | |
2095 | goto out; | |
2096 | ||
72add155 SH |
2097 | mode_t m = sb.st_mode & S_IFMT; |
2098 | switch (m) { | |
2099 | case S_IFBLK: | |
2100 | type = 'b'; | |
2101 | break; | |
2102 | case S_IFCHR: | |
2103 | type = 'c'; | |
2104 | break; | |
2c2d6c49 | 2105 | default: |
91d1a13a | 2106 | ERROR("Unsupported device type %i for \"%s\"", m, path); |
72add155 SH |
2107 | ret = -EINVAL; |
2108 | goto out; | |
2109 | } | |
2c2d6c49 SH |
2110 | |
2111 | major = MAJOR(sb.st_rdev); | |
2112 | minor = MINOR(sb.st_rdev); | |
2113 | ret = snprintf(dest, 50, "%c %lu:%lu %s", type, major, minor, mode); | |
72add155 | 2114 | if (ret < 0 || ret >= 50) { |
2a06d041 CB |
2115 | ERROR("Error on configuration value \"%c %lu:%lu %s\" (max 50 " |
2116 | "chars)", type, major, minor, mode); | |
72add155 SH |
2117 | ret = -ENAMETOOLONG; |
2118 | goto out; | |
2119 | } | |
2120 | ret = 0; | |
2121 | ||
2122 | out: | |
2123 | free(path); | |
2124 | return ret; | |
2125 | } | |
2126 | ||
90e97284 CB |
2127 | /* Called from setup_limits - here we have the container's cgroup_data because |
2128 | * we created the cgroups. | |
ccb4cabe | 2129 | */ |
2202afc9 CB |
2130 | static int cg_legacy_set_data(struct cgroup_ops *ops, const char *filename, |
2131 | const char *value) | |
ccb4cabe | 2132 | { |
ab1a6cac | 2133 | size_t len; |
90e97284 | 2134 | char *fullpath, *p; |
1a0e70ac CB |
2135 | /* "b|c <2^64-1>:<2^64-1> r|w|m" = 47 chars max */ |
2136 | char converted_value[50]; | |
b3646d7e CB |
2137 | struct hierarchy *h; |
2138 | int ret = 0; | |
2139 | char *controller = NULL; | |
ccb4cabe | 2140 | |
ab1a6cac CB |
2141 | len = strlen(filename); |
2142 | controller = alloca(len + 1); | |
64e82f8b DJ |
2143 | (void)strlcpy(controller, filename, len + 1); |
2144 | ||
ab1a6cac CB |
2145 | p = strchr(controller, '.'); |
2146 | if (p) | |
ccb4cabe SH |
2147 | *p = '\0'; |
2148 | ||
c8bf519d | 2149 | if (strcmp("devices.allow", filename) == 0 && value[0] == '/') { |
72add155 SH |
2150 | ret = convert_devpath(value, converted_value); |
2151 | if (ret < 0) | |
c8bf519d | 2152 | return ret; |
72add155 | 2153 | value = converted_value; |
c8bf519d | 2154 | } |
2155 | ||
2202afc9 | 2156 | h = get_hierarchy(ops, controller); |
b3646d7e CB |
2157 | if (!h) { |
2158 | ERROR("Failed to setup limits for the \"%s\" controller. " | |
2159 | "The controller seems to be unused by \"cgfsng\" cgroup " | |
2160 | "driver or not enabled on the cgroup hierarchy", | |
2161 | controller); | |
d1953b26 | 2162 | errno = ENOENT; |
ab1a6cac | 2163 | return -ENOENT; |
ccb4cabe | 2164 | } |
b3646d7e | 2165 | |
eb697136 | 2166 | fullpath = must_make_path(h->container_full_path, filename, NULL); |
7cea5905 | 2167 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
b3646d7e | 2168 | free(fullpath); |
ccb4cabe SH |
2169 | return ret; |
2170 | } | |
2171 | ||
2202afc9 | 2172 | static bool __cg_legacy_setup_limits(struct cgroup_ops *ops, |
a3926f6a CB |
2173 | struct lxc_list *cgroup_settings, |
2174 | bool do_devices) | |
ccb4cabe | 2175 | { |
c347df58 | 2176 | struct lxc_list *iterator, *next, *sorted_cgroup_settings; |
ccb4cabe | 2177 | struct lxc_cgroup *cg; |
ccb4cabe SH |
2178 | bool ret = false; |
2179 | ||
2180 | if (lxc_list_empty(cgroup_settings)) | |
2181 | return true; | |
2182 | ||
2183 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
6b38e644 | 2184 | if (!sorted_cgroup_settings) |
ccb4cabe | 2185 | return false; |
ccb4cabe | 2186 | |
ccb4cabe SH |
2187 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
2188 | cg = iterator->elem; | |
2189 | ||
2190 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
2202afc9 | 2191 | if (cg_legacy_set_data(ops, cg->subsystem, cg->value)) { |
ccb4cabe | 2192 | if (do_devices && (errno == EACCES || errno == EPERM)) { |
c347df58 CB |
2193 | WARN("Failed to set \"%s\" to \"%s\"", |
2194 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2195 | continue; |
2196 | } | |
c347df58 CB |
2197 | WARN("Failed to set \"%s\" to \"%s\"", |
2198 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2199 | goto out; |
2200 | } | |
c347df58 CB |
2201 | DEBUG("Set controller \"%s\" set to \"%s\"", |
2202 | cg->subsystem, cg->value); | |
ccb4cabe | 2203 | } |
ccb4cabe SH |
2204 | } |
2205 | ||
2206 | ret = true; | |
6b38e644 | 2207 | INFO("Limits for the legacy cgroup hierarchies have been setup"); |
ccb4cabe | 2208 | out: |
ccb4cabe SH |
2209 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
2210 | lxc_list_del(iterator); | |
2211 | free(iterator); | |
2212 | } | |
2213 | free(sorted_cgroup_settings); | |
2214 | return ret; | |
2215 | } | |
2216 | ||
2202afc9 | 2217 | static bool __cg_unified_setup_limits(struct cgroup_ops *ops, |
a3926f6a | 2218 | struct lxc_list *cgroup_settings) |
6b38e644 CB |
2219 | { |
2220 | struct lxc_list *iterator; | |
2202afc9 | 2221 | struct hierarchy *h = ops->unified; |
6b38e644 CB |
2222 | |
2223 | if (lxc_list_empty(cgroup_settings)) | |
2224 | return true; | |
2225 | ||
2226 | if (!h) | |
2227 | return false; | |
2228 | ||
2229 | lxc_list_for_each(iterator, cgroup_settings) { | |
2230 | int ret; | |
2231 | char *fullpath; | |
2232 | struct lxc_cgroup *cg = iterator->elem; | |
2233 | ||
eb697136 | 2234 | fullpath = must_make_path(h->container_full_path, cg->subsystem, NULL); |
7cea5905 | 2235 | ret = lxc_write_to_file(fullpath, cg->value, strlen(cg->value), false, 0666); |
6b38e644 CB |
2236 | free(fullpath); |
2237 | if (ret < 0) { | |
b2ac2cb7 CB |
2238 | SYSERROR("Failed to set \"%s\" to \"%s\"", |
2239 | cg->subsystem, cg->value); | |
6b38e644 CB |
2240 | return false; |
2241 | } | |
2242 | TRACE("Set \"%s\" to \"%s\"", cg->subsystem, cg->value); | |
2243 | } | |
2244 | ||
2245 | INFO("Limits for the unified cgroup hierarchy have been setup"); | |
2246 | return true; | |
2247 | } | |
2248 | ||
fb55e009 CB |
2249 | __cgfsng_ops__ static bool cgfsng_setup_limits(struct cgroup_ops *ops, |
2250 | struct lxc_conf *conf, | |
2251 | bool do_devices) | |
6b38e644 CB |
2252 | { |
2253 | bool bret; | |
2254 | ||
2202afc9 | 2255 | bret = __cg_legacy_setup_limits(ops, &conf->cgroup, do_devices); |
6b38e644 CB |
2256 | if (!bret) |
2257 | return false; | |
2258 | ||
2202afc9 CB |
2259 | return __cg_unified_setup_limits(ops, &conf->cgroup2); |
2260 | } | |
2261 | ||
b7b18fc5 CB |
2262 | static bool cgroup_use_wants_controllers(const struct cgroup_ops *ops, |
2263 | char **controllers) | |
2264 | { | |
2265 | char **cur_ctrl, **cur_use; | |
2266 | ||
2267 | if (!ops->cgroup_use) | |
2268 | return true; | |
2269 | ||
2270 | for (cur_ctrl = controllers; cur_ctrl && *cur_ctrl; cur_ctrl++) { | |
2271 | bool found = false; | |
2272 | ||
2273 | for (cur_use = ops->cgroup_use; cur_use && *cur_use; cur_use++) { | |
2274 | if (strcmp(*cur_use, *cur_ctrl) != 0) | |
2275 | continue; | |
2276 | ||
2277 | found = true; | |
2278 | break; | |
2279 | } | |
2280 | ||
2281 | if (found) | |
2282 | continue; | |
2283 | ||
2284 | return false; | |
2285 | } | |
2286 | ||
2287 | return true; | |
2288 | } | |
2289 | ||
2202afc9 CB |
2290 | /* At startup, parse_hierarchies finds all the info we need about cgroup |
2291 | * mountpoints and current cgroups, and stores it in @d. | |
2292 | */ | |
9caee129 | 2293 | static bool cg_hybrid_init(struct cgroup_ops *ops, bool relative) |
2202afc9 CB |
2294 | { |
2295 | int ret; | |
2296 | char *basecginfo; | |
2202afc9 CB |
2297 | FILE *f; |
2298 | size_t len = 0; | |
2299 | char *line = NULL; | |
2300 | char **klist = NULL, **nlist = NULL; | |
2301 | ||
2302 | /* Root spawned containers escape the current cgroup, so use init's | |
2303 | * cgroups as our base in that case. | |
2304 | */ | |
9caee129 | 2305 | if (!relative && (geteuid() == 0)) |
2202afc9 CB |
2306 | basecginfo = read_file("/proc/1/cgroup"); |
2307 | else | |
2308 | basecginfo = read_file("/proc/self/cgroup"); | |
2309 | if (!basecginfo) | |
2310 | return false; | |
2311 | ||
2312 | ret = get_existing_subsystems(&klist, &nlist); | |
2313 | if (ret < 0) { | |
2314 | ERROR("Failed to retrieve available legacy cgroup controllers"); | |
2315 | free(basecginfo); | |
2316 | return false; | |
2317 | } | |
2318 | ||
2319 | f = fopen("/proc/self/mountinfo", "r"); | |
2320 | if (!f) { | |
2321 | ERROR("Failed to open \"/proc/self/mountinfo\""); | |
2322 | free(basecginfo); | |
2323 | return false; | |
2324 | } | |
2325 | ||
2326 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
2327 | ||
2328 | while (getline(&line, &len, f) != -1) { | |
2329 | int type; | |
2330 | bool writeable; | |
2331 | struct hierarchy *new; | |
2332 | char *base_cgroup = NULL, *mountpoint = NULL; | |
2333 | char **controller_list = NULL; | |
2334 | ||
2335 | type = get_cgroup_version(line); | |
2336 | if (type == 0) | |
2337 | continue; | |
2338 | ||
2339 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
2340 | continue; | |
2341 | ||
2342 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
2343 | if (type == CGROUP2_SUPER_MAGIC) | |
2344 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2345 | else if (type == CGROUP_SUPER_MAGIC) | |
2346 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
2347 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
2348 | if (type == CGROUP_SUPER_MAGIC) | |
2349 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2350 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
2351 | if (type == CGROUP2_SUPER_MAGIC) | |
2352 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2353 | } | |
2354 | ||
2355 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, type); | |
2356 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
2357 | continue; | |
2358 | ||
2359 | if (type == CGROUP_SUPER_MAGIC) | |
2360 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
2361 | goto next; | |
2362 | ||
2363 | mountpoint = cg_hybrid_get_mountpoint(line); | |
2364 | if (!mountpoint) { | |
2365 | ERROR("Failed parsing mountpoint from \"%s\"", line); | |
2366 | goto next; | |
2367 | } | |
2368 | ||
2369 | if (type == CGROUP_SUPER_MAGIC) | |
2370 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); | |
2371 | else | |
2372 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); | |
2373 | if (!base_cgroup) { | |
2374 | ERROR("Failed to find current cgroup"); | |
2375 | goto next; | |
2376 | } | |
2377 | ||
2378 | trim(base_cgroup); | |
2379 | prune_init_scope(base_cgroup); | |
2380 | if (type == CGROUP2_SUPER_MAGIC) | |
2381 | writeable = test_writeable_v2(mountpoint, base_cgroup); | |
2382 | else | |
2383 | writeable = test_writeable_v1(mountpoint, base_cgroup); | |
2384 | if (!writeable) | |
2385 | goto next; | |
2386 | ||
2387 | if (type == CGROUP2_SUPER_MAGIC) { | |
2388 | char *cgv2_ctrl_path; | |
2389 | ||
2390 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
2391 | "cgroup.controllers", | |
2392 | NULL); | |
2393 | ||
2394 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
2395 | free(cgv2_ctrl_path); | |
2396 | if (!controller_list) { | |
2397 | controller_list = cg_unified_make_empty_controller(); | |
2398 | TRACE("No controllers are enabled for " | |
2399 | "delegation in the unified hierarchy"); | |
2400 | } | |
2401 | } | |
2402 | ||
b7b18fc5 CB |
2403 | /* Exclude all controllers that cgroup use does not want. */ |
2404 | if (!cgroup_use_wants_controllers(ops, controller_list)) | |
2405 | goto next; | |
2406 | ||
2202afc9 CB |
2407 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); |
2408 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
2409 | ops->unified = new; | |
2410 | ||
2411 | continue; | |
2412 | ||
2413 | next: | |
2414 | free_string_list(controller_list); | |
2415 | free(mountpoint); | |
2416 | free(base_cgroup); | |
2417 | } | |
2418 | ||
2419 | free_string_list(klist); | |
2420 | free_string_list(nlist); | |
2421 | ||
2422 | free(basecginfo); | |
2423 | ||
2424 | fclose(f); | |
2425 | free(line); | |
2426 | ||
2427 | TRACE("Writable cgroup hierarchies:"); | |
2428 | lxc_cgfsng_print_hierarchies(ops); | |
2429 | ||
2430 | /* verify that all controllers in cgroup.use and all crucial | |
2431 | * controllers are accounted for | |
2432 | */ | |
2433 | if (!all_controllers_found(ops)) | |
2434 | return false; | |
2435 | ||
2436 | return true; | |
2437 | } | |
2438 | ||
2439 | static int cg_is_pure_unified(void) | |
2440 | { | |
2441 | ||
2442 | int ret; | |
2443 | struct statfs fs; | |
2444 | ||
2445 | ret = statfs("/sys/fs/cgroup", &fs); | |
2446 | if (ret < 0) | |
2447 | return -ENOMEDIUM; | |
2448 | ||
2449 | if (is_fs_type(&fs, CGROUP2_SUPER_MAGIC)) | |
2450 | return CGROUP2_SUPER_MAGIC; | |
2451 | ||
2452 | return 0; | |
2453 | } | |
2454 | ||
2455 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
9caee129 | 2456 | static char *cg_unified_get_current_cgroup(bool relative) |
2202afc9 CB |
2457 | { |
2458 | char *basecginfo, *base_cgroup; | |
2202afc9 CB |
2459 | char *copy = NULL; |
2460 | ||
9caee129 | 2461 | if (!relative && (geteuid() == 0)) |
2202afc9 CB |
2462 | basecginfo = read_file("/proc/1/cgroup"); |
2463 | else | |
2464 | basecginfo = read_file("/proc/self/cgroup"); | |
2465 | if (!basecginfo) | |
2466 | return NULL; | |
2467 | ||
2468 | base_cgroup = strstr(basecginfo, "0::/"); | |
2469 | if (!base_cgroup) | |
2470 | goto cleanup_on_err; | |
2471 | ||
2472 | base_cgroup = base_cgroup + 3; | |
2473 | copy = copy_to_eol(base_cgroup); | |
2474 | if (!copy) | |
2475 | goto cleanup_on_err; | |
2476 | ||
2477 | cleanup_on_err: | |
2478 | free(basecginfo); | |
2479 | if (copy) | |
2480 | trim(copy); | |
2481 | ||
2482 | return copy; | |
2483 | } | |
2484 | ||
9caee129 | 2485 | static int cg_unified_init(struct cgroup_ops *ops, bool relative) |
2202afc9 CB |
2486 | { |
2487 | int ret; | |
2488 | char *mountpoint, *subtree_path; | |
2489 | char **delegatable; | |
2490 | char *base_cgroup = NULL; | |
2491 | ||
2492 | ret = cg_is_pure_unified(); | |
2493 | if (ret == -ENOMEDIUM) | |
2494 | return -ENOMEDIUM; | |
2495 | ||
2496 | if (ret != CGROUP2_SUPER_MAGIC) | |
2497 | return 0; | |
2498 | ||
9caee129 | 2499 | base_cgroup = cg_unified_get_current_cgroup(relative); |
2202afc9 CB |
2500 | if (!base_cgroup) |
2501 | return -EINVAL; | |
2502 | prune_init_scope(base_cgroup); | |
2503 | ||
2504 | /* We assume that we have already been given controllers to delegate | |
2505 | * further down the hierarchy. If not it is up to the user to delegate | |
2506 | * them to us. | |
2507 | */ | |
2508 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
2509 | subtree_path = must_make_path(mountpoint, base_cgroup, | |
2510 | "cgroup.subtree_control", NULL); | |
2511 | delegatable = cg_unified_get_controllers(subtree_path); | |
2512 | free(subtree_path); | |
2513 | if (!delegatable) | |
2514 | delegatable = cg_unified_make_empty_controller(); | |
2515 | if (!delegatable[0]) | |
2516 | TRACE("No controllers are enabled for delegation"); | |
2517 | ||
2518 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
2519 | * we should verify here. The reason I'm not doing it right is that I'm | |
2520 | * not convinced that lxc.cgroup.use will be the future since it is a | |
2521 | * global property. I much rather have an option that lets you request | |
2522 | * controllers per container. | |
2523 | */ | |
2524 | ||
2525 | add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
2526 | ||
2527 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2528 | return CGROUP2_SUPER_MAGIC; | |
2529 | } | |
2530 | ||
5a087e05 | 2531 | static bool cg_init(struct cgroup_ops *ops, struct lxc_conf *conf) |
2202afc9 CB |
2532 | { |
2533 | int ret; | |
2534 | const char *tmp; | |
9caee129 | 2535 | bool relative = conf->cgroup_meta.relative; |
2202afc9 CB |
2536 | |
2537 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
b7b18fc5 CB |
2538 | if (tmp) { |
2539 | char *chop, *cur, *pin; | |
b7b18fc5 CB |
2540 | |
2541 | pin = must_copy_string(tmp); | |
2542 | chop = pin; | |
2543 | ||
0be0d78f | 2544 | lxc_iterate_parts(cur, chop, ",") { |
b7b18fc5 | 2545 | must_append_string(&ops->cgroup_use, cur); |
0be0d78f | 2546 | } |
b7b18fc5 CB |
2547 | |
2548 | free(pin); | |
2549 | } | |
2202afc9 | 2550 | |
9caee129 | 2551 | ret = cg_unified_init(ops, relative); |
2202afc9 CB |
2552 | if (ret < 0) |
2553 | return false; | |
2554 | ||
2555 | if (ret == CGROUP2_SUPER_MAGIC) | |
2556 | return true; | |
2557 | ||
9caee129 | 2558 | return cg_hybrid_init(ops, relative); |
2202afc9 CB |
2559 | } |
2560 | ||
fb55e009 | 2561 | __cgfsng_ops__ static bool cgfsng_data_init(struct cgroup_ops *ops) |
2202afc9 CB |
2562 | { |
2563 | const char *cgroup_pattern; | |
2564 | ||
2565 | /* copy system-wide cgroup information */ | |
2566 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); | |
2567 | if (!cgroup_pattern) { | |
2568 | /* lxc.cgroup.pattern is only NULL on error. */ | |
2569 | ERROR("Failed to retrieve cgroup pattern"); | |
2570 | return false; | |
2571 | } | |
2572 | ops->cgroup_pattern = must_copy_string(cgroup_pattern); | |
d39e9d8b | 2573 | ops->monitor_pattern = must_copy_string("lxc.monitor"); |
2202afc9 CB |
2574 | |
2575 | return true; | |
2576 | } | |
2577 | ||
5a087e05 | 2578 | struct cgroup_ops *cgfsng_ops_init(struct lxc_conf *conf) |
2202afc9 CB |
2579 | { |
2580 | struct cgroup_ops *cgfsng_ops; | |
2581 | ||
2582 | cgfsng_ops = malloc(sizeof(struct cgroup_ops)); | |
2583 | if (!cgfsng_ops) | |
2584 | return NULL; | |
2585 | ||
2586 | memset(cgfsng_ops, 0, sizeof(struct cgroup_ops)); | |
2587 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; | |
2588 | ||
5a087e05 | 2589 | if (!cg_init(cgfsng_ops, conf)) { |
2202afc9 CB |
2590 | free(cgfsng_ops); |
2591 | return NULL; | |
2592 | } | |
2593 | ||
2594 | cgfsng_ops->data_init = cgfsng_data_init; | |
2595 | cgfsng_ops->destroy = cgfsng_destroy; | |
e8b181f5 CB |
2596 | cgfsng_ops->payload_create = cgfsng_payload_create; |
2597 | cgfsng_ops->payload_enter = cgfsng_payload_enter; | |
2202afc9 CB |
2598 | cgfsng_ops->escape = cgfsng_escape; |
2599 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
2600 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; | |
2601 | cgfsng_ops->get_cgroup = cgfsng_get_cgroup; | |
2602 | cgfsng_ops->get = cgfsng_get; | |
2603 | cgfsng_ops->set = cgfsng_set; | |
2604 | cgfsng_ops->unfreeze = cgfsng_unfreeze; | |
2605 | cgfsng_ops->setup_limits = cgfsng_setup_limits; | |
2606 | cgfsng_ops->driver = "cgfsng"; | |
2607 | cgfsng_ops->version = "1.0.0"; | |
2608 | cgfsng_ops->attach = cgfsng_attach; | |
2609 | cgfsng_ops->chown = cgfsng_chown; | |
2610 | cgfsng_ops->mount = cgfsng_mount; | |
2611 | cgfsng_ops->nrtasks = cgfsng_nrtasks; | |
2612 | ||
2613 | return cgfsng_ops; | |
2614 | } |