]>
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 | |
18406e5a | 34 | * a comma-separated list of controllers. |
ccb4cabe | 35 | */ |
a54694f8 | 36 | |
d38dd64a CB |
37 | #ifndef _GNU_SOURCE |
38 | #define _GNU_SOURCE 1 | |
39 | #endif | |
a54694f8 CB |
40 | #include <ctype.h> |
41 | #include <dirent.h> | |
42 | #include <errno.h> | |
43 | #include <grp.h> | |
d38dd64a CB |
44 | #include <linux/kdev_t.h> |
45 | #include <linux/types.h> | |
a54694f8 | 46 | #include <stdint.h> |
ccb4cabe SH |
47 | #include <stdio.h> |
48 | #include <stdlib.h> | |
a54694f8 | 49 | #include <string.h> |
438c4581 | 50 | #include <sys/types.h> |
d38dd64a | 51 | #include <unistd.h> |
c8bf519d | 52 | |
b635e92d | 53 | #include "caps.h" |
ccb4cabe | 54 | #include "cgroup.h" |
6328fd9c | 55 | #include "cgroup_utils.h" |
ccb4cabe | 56 | #include "commands.h" |
43654d34 | 57 | #include "conf.h" |
d38dd64a | 58 | #include "config.h" |
a54694f8 | 59 | #include "log.h" |
c19ad94b | 60 | #include "macro.h" |
43654d34 | 61 | #include "storage/storage.h" |
a54694f8 | 62 | #include "utils.h" |
ccb4cabe | 63 | |
64e82f8b DJ |
64 | #ifndef HAVE_STRLCPY |
65 | #include "include/strlcpy.h" | |
66 | #endif | |
67 | ||
3ebe2fbd DJ |
68 | #ifndef HAVE_STRLCAT |
69 | #include "include/strlcat.h" | |
70 | #endif | |
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 | ||
8b8db2f6 CB |
87 | /* Given a pointer to a null-terminated array of pointers, realloc to add one |
88 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
89 | * second-to-last entry - that is, the one which is now available for use | |
90 | * (keeping the list null-terminated). | |
ccb4cabe SH |
91 | */ |
92 | static int append_null_to_list(void ***list) | |
93 | { | |
94 | int newentry = 0; | |
95 | ||
96 | if (*list) | |
8b8db2f6 CB |
97 | for (; (*list)[newentry]; newentry++) |
98 | ; | |
ccb4cabe SH |
99 | |
100 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
101 | (*list)[newentry + 1] = NULL; | |
102 | return newentry; | |
103 | } | |
104 | ||
8073018d CB |
105 | /* Given a null-terminated array of strings, check whether @entry is one of the |
106 | * strings. | |
ccb4cabe SH |
107 | */ |
108 | static bool string_in_list(char **list, const char *entry) | |
109 | { | |
110 | int i; | |
111 | ||
112 | if (!list) | |
113 | return false; | |
d6337a5f | 114 | |
ccb4cabe SH |
115 | for (i = 0; list[i]; i++) |
116 | if (strcmp(list[i], entry) == 0) | |
117 | return true; | |
118 | ||
119 | return false; | |
120 | } | |
121 | ||
ac010944 CB |
122 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into |
123 | * "name=systemd". Do not fail. | |
124 | */ | |
125 | static char *cg_legacy_must_prefix_named(char *entry) | |
126 | { | |
127 | size_t len; | |
128 | char *prefixed; | |
129 | ||
130 | len = strlen(entry); | |
f25a2044 | 131 | prefixed = must_realloc(NULL, len + 6); |
ac010944 | 132 | |
6333c915 CB |
133 | memcpy(prefixed, "name=", STRLITERALLEN("name=")); |
134 | memcpy(prefixed + STRLITERALLEN("name="), entry, len); | |
ac010944 | 135 | prefixed[len + 5] = '\0'; |
99bb3fa8 | 136 | |
ac010944 CB |
137 | return prefixed; |
138 | } | |
139 | ||
42a993b4 CB |
140 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time |
141 | * we are called. | |
ccb4cabe | 142 | * |
42a993b4 CB |
143 | * We also handle named subsystems here. Any controller which is not a kernel |
144 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
145 | * we refuse to use because we're not sure which we have here. | |
146 | * (TODO: We could work around this in some cases by just remounting to be | |
147 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
ccb4cabe SH |
148 | * |
149 | * The last entry will always be NULL. | |
150 | */ | |
42a993b4 CB |
151 | static void must_append_controller(char **klist, char **nlist, char ***clist, |
152 | char *entry) | |
ccb4cabe SH |
153 | { |
154 | int newentry; | |
155 | char *copy; | |
156 | ||
157 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
c2712f64 | 158 | ERROR("Refusing to use ambiguous controller \"%s\"", entry); |
ccb4cabe SH |
159 | ERROR("It is both a named and kernel subsystem"); |
160 | return; | |
161 | } | |
162 | ||
163 | newentry = append_null_to_list((void ***)clist); | |
164 | ||
165 | if (strncmp(entry, "name=", 5) == 0) | |
166 | copy = must_copy_string(entry); | |
167 | else if (string_in_list(klist, entry)) | |
168 | copy = must_copy_string(entry); | |
169 | else | |
7745483d | 170 | copy = cg_legacy_must_prefix_named(entry); |
ccb4cabe SH |
171 | |
172 | (*clist)[newentry] = copy; | |
173 | } | |
174 | ||
5ae0207c CB |
175 | /* Given a handler's cgroup data, return the struct hierarchy for the controller |
176 | * @c, or NULL if there is none. | |
ccb4cabe | 177 | */ |
27a5132c | 178 | struct hierarchy *get_hierarchy(struct cgroup_ops *ops, const char *controller) |
ccb4cabe SH |
179 | { |
180 | int i; | |
181 | ||
27a5132c CB |
182 | errno = ENOENT; |
183 | ||
184 | if (!ops->hierarchies) { | |
185 | TRACE("There are no useable cgroup controllers"); | |
ccb4cabe | 186 | return NULL; |
27a5132c | 187 | } |
d6337a5f | 188 | |
2202afc9 | 189 | for (i = 0; ops->hierarchies[i]; i++) { |
27a5132c | 190 | if (!controller) { |
d6337a5f | 191 | /* This is the empty unified hierarchy. */ |
2202afc9 CB |
192 | if (ops->hierarchies[i]->controllers && |
193 | !ops->hierarchies[i]->controllers[0]) | |
194 | return ops->hierarchies[i]; | |
d6337a5f | 195 | |
106f1f38 | 196 | continue; |
d6337a5f CB |
197 | } |
198 | ||
27a5132c | 199 | if (string_in_list(ops->hierarchies[i]->controllers, controller)) |
2202afc9 | 200 | return ops->hierarchies[i]; |
ccb4cabe | 201 | } |
d6337a5f | 202 | |
27a5132c CB |
203 | if (controller) |
204 | WARN("There is no useable %s controller", controller); | |
205 | else | |
206 | WARN("There is no empty unified cgroup hierarchy"); | |
207 | ||
ccb4cabe SH |
208 | return NULL; |
209 | } | |
210 | ||
a54694f8 CB |
211 | #define BATCH_SIZE 50 |
212 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
213 | { | |
214 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
215 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
216 | ||
217 | if (!*mem || newbatches > oldbatches) { | |
218 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
219 | } | |
220 | } | |
221 | ||
222 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
223 | { | |
224 | size_t full = oldlen + newlen; | |
225 | ||
226 | batch_realloc(dest, oldlen, full + 1); | |
227 | ||
228 | memcpy(*dest + oldlen, new, newlen + 1); | |
229 | } | |
230 | ||
231 | /* Slurp in a whole file */ | |
d6337a5f | 232 | static char *read_file(const char *fnam) |
a54694f8 CB |
233 | { |
234 | FILE *f; | |
235 | char *line = NULL, *buf = NULL; | |
236 | size_t len = 0, fulllen = 0; | |
237 | int linelen; | |
238 | ||
239 | f = fopen(fnam, "r"); | |
240 | if (!f) | |
241 | return NULL; | |
242 | while ((linelen = getline(&line, &len, f)) != -1) { | |
243 | append_line(&buf, fulllen, line, linelen); | |
244 | fulllen += linelen; | |
245 | } | |
246 | fclose(f); | |
247 | free(line); | |
248 | return buf; | |
249 | } | |
250 | ||
251 | /* Taken over modified from the kernel sources. */ | |
252 | #define NBITS 32 /* bits in uint32_t */ | |
253 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
254 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
255 | ||
256 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
257 | { | |
258 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
259 | } | |
260 | ||
261 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
262 | { | |
263 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
264 | } | |
265 | ||
266 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
267 | { | |
268 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
269 | } | |
270 | ||
271 | /* Create cpumask from cpulist aka turn: | |
272 | * | |
273 | * 0,2-3 | |
274 | * | |
d5d468f6 | 275 | * into bit array |
a54694f8 CB |
276 | * |
277 | * 1 0 1 1 | |
278 | */ | |
279 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
280 | { | |
281 | char *token; | |
d5d468f6 CB |
282 | size_t arrlen; |
283 | uint32_t *bitarr; | |
d5d468f6 CB |
284 | |
285 | arrlen = BITS_TO_LONGS(nbits); | |
286 | bitarr = calloc(arrlen, sizeof(uint32_t)); | |
a54694f8 CB |
287 | if (!bitarr) |
288 | return NULL; | |
289 | ||
0be0d78f | 290 | lxc_iterate_parts(token, buf, ",") { |
a54694f8 | 291 | errno = 0; |
d5d468f6 CB |
292 | unsigned end, start; |
293 | char *range; | |
a54694f8 | 294 | |
d5d468f6 CB |
295 | start = strtoul(token, NULL, 0); |
296 | end = start; | |
297 | range = strchr(token, '-'); | |
a54694f8 CB |
298 | if (range) |
299 | end = strtoul(range + 1, NULL, 0); | |
d5d468f6 | 300 | |
a54694f8 CB |
301 | if (!(start <= end)) { |
302 | free(bitarr); | |
303 | return NULL; | |
304 | } | |
305 | ||
306 | if (end >= nbits) { | |
307 | free(bitarr); | |
308 | return NULL; | |
309 | } | |
310 | ||
311 | while (start <= end) | |
312 | set_bit(start++, bitarr); | |
313 | } | |
314 | ||
315 | return bitarr; | |
316 | } | |
317 | ||
a54694f8 CB |
318 | /* Turn cpumask into simple, comma-separated cpulist. */ |
319 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
320 | { | |
a54694f8 | 321 | int ret; |
414c6719 | 322 | size_t i; |
a54694f8 | 323 | char **cpulist = NULL; |
c19ad94b | 324 | char numstr[INTTYPE_TO_STRLEN(size_t)] = {0}; |
a54694f8 CB |
325 | |
326 | for (i = 0; i <= nbits; i++) { | |
414c6719 CB |
327 | if (!is_set(i, bitarr)) |
328 | continue; | |
329 | ||
979a0d93 CB |
330 | ret = snprintf(numstr, sizeof(numstr), "%zu", i); |
331 | if (ret < 0 || (size_t)ret >= sizeof(numstr)) { | |
414c6719 CB |
332 | lxc_free_array((void **)cpulist, free); |
333 | return NULL; | |
334 | } | |
335 | ||
336 | ret = lxc_append_string(&cpulist, numstr); | |
337 | if (ret < 0) { | |
338 | lxc_free_array((void **)cpulist, free); | |
339 | return NULL; | |
a54694f8 CB |
340 | } |
341 | } | |
414c6719 CB |
342 | |
343 | if (!cpulist) | |
344 | return NULL; | |
345 | ||
a54694f8 CB |
346 | return lxc_string_join(",", (const char **)cpulist, false); |
347 | } | |
348 | ||
349 | static ssize_t get_max_cpus(char *cpulist) | |
350 | { | |
351 | char *c1, *c2; | |
352 | char *maxcpus = cpulist; | |
353 | size_t cpus = 0; | |
354 | ||
355 | c1 = strrchr(maxcpus, ','); | |
356 | if (c1) | |
357 | c1++; | |
358 | ||
359 | c2 = strrchr(maxcpus, '-'); | |
360 | if (c2) | |
361 | c2++; | |
362 | ||
363 | if (!c1 && !c2) | |
364 | c1 = maxcpus; | |
365 | else if (c1 > c2) | |
366 | c2 = c1; | |
367 | else if (c1 < c2) | |
368 | c1 = c2; | |
333987b9 | 369 | else if (!c1 && c2) |
a54694f8 CB |
370 | c1 = c2; |
371 | ||
a54694f8 CB |
372 | errno = 0; |
373 | cpus = strtoul(c1, NULL, 0); | |
374 | if (errno != 0) | |
375 | return -1; | |
376 | ||
377 | return cpus; | |
378 | } | |
379 | ||
6f9584d8 | 380 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a3926f6a | 381 | static bool cg_legacy_filter_and_set_cpus(char *path, bool am_initialized) |
a54694f8 | 382 | { |
a54694f8 CB |
383 | int ret; |
384 | ssize_t i; | |
59ac3b88 CB |
385 | char *lastslash, *fpath, oldv; |
386 | ssize_t maxisol = 0, maxposs = 0; | |
387 | char *cpulist = NULL, *isolcpus = NULL, *posscpus = NULL; | |
388 | uint32_t *isolmask = NULL, *possmask = NULL; | |
6f9584d8 | 389 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
390 | |
391 | lastslash = strrchr(path, '/'); | |
59ac3b88 CB |
392 | if (!lastslash) { |
393 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
a54694f8 CB |
394 | return bret; |
395 | } | |
396 | oldv = *lastslash; | |
397 | *lastslash = '\0'; | |
398 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
399 | posscpus = read_file(fpath); | |
6f9584d8 | 400 | if (!posscpus) { |
59ac3b88 | 401 | SYSERROR("Failed to read file \"%s\"", fpath); |
6f9584d8 CB |
402 | goto on_error; |
403 | } | |
a54694f8 CB |
404 | |
405 | /* Get maximum number of cpus found in possible cpuset. */ | |
406 | maxposs = get_max_cpus(posscpus); | |
92d5ea57 | 407 | if (maxposs < 0 || maxposs >= INT_MAX - 1) |
6f9584d8 | 408 | goto on_error; |
a54694f8 | 409 | |
6f9584d8 CB |
410 | if (!file_exists(__ISOL_CPUS)) { |
411 | /* This system doesn't expose isolated cpus. */ | |
59ac3b88 | 412 | DEBUG("The path \""__ISOL_CPUS"\" to read isolated cpus from does not exist"); |
65d29cbc CB |
413 | cpulist = posscpus; |
414 | /* No isolated cpus but we weren't already initialized by | |
415 | * someone. We should simply copy the parents cpuset.cpus | |
416 | * values. | |
417 | */ | |
418 | if (!am_initialized) { | |
59ac3b88 | 419 | DEBUG("Copying cpu settings of parent cgroup"); |
65d29cbc CB |
420 | goto copy_parent; |
421 | } | |
422 | /* No isolated cpus but we were already initialized by someone. | |
423 | * Nothing more to do for us. | |
424 | */ | |
6f9584d8 CB |
425 | goto on_success; |
426 | } | |
427 | ||
428 | isolcpus = read_file(__ISOL_CPUS); | |
429 | if (!isolcpus) { | |
59ac3b88 | 430 | SYSERROR("Failed to read file \""__ISOL_CPUS"\""); |
6f9584d8 CB |
431 | goto on_error; |
432 | } | |
a54694f8 | 433 | if (!isdigit(isolcpus[0])) { |
59ac3b88 | 434 | TRACE("No isolated cpus detected"); |
a54694f8 CB |
435 | cpulist = posscpus; |
436 | /* No isolated cpus but we weren't already initialized by | |
437 | * someone. We should simply copy the parents cpuset.cpus | |
438 | * values. | |
439 | */ | |
6f9584d8 | 440 | if (!am_initialized) { |
59ac3b88 | 441 | DEBUG("Copying cpu settings of parent cgroup"); |
a54694f8 | 442 | goto copy_parent; |
6f9584d8 | 443 | } |
a54694f8 CB |
444 | /* No isolated cpus but we were already initialized by someone. |
445 | * Nothing more to do for us. | |
446 | */ | |
6f9584d8 | 447 | goto on_success; |
a54694f8 CB |
448 | } |
449 | ||
450 | /* Get maximum number of cpus found in isolated cpuset. */ | |
451 | maxisol = get_max_cpus(isolcpus); | |
92d5ea57 | 452 | if (maxisol < 0 || maxisol >= INT_MAX - 1) |
6f9584d8 | 453 | goto on_error; |
a54694f8 CB |
454 | |
455 | if (maxposs < maxisol) | |
456 | maxposs = maxisol; | |
457 | maxposs++; | |
458 | ||
459 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 | 460 | if (!possmask) { |
59ac3b88 | 461 | ERROR("Failed to create cpumask for possible cpus"); |
6f9584d8 CB |
462 | goto on_error; |
463 | } | |
a54694f8 CB |
464 | |
465 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 | 466 | if (!isolmask) { |
59ac3b88 | 467 | ERROR("Failed to create cpumask for isolated cpus"); |
6f9584d8 CB |
468 | goto on_error; |
469 | } | |
a54694f8 CB |
470 | |
471 | for (i = 0; i <= maxposs; i++) { | |
59ac3b88 CB |
472 | if (!is_set(i, isolmask) || !is_set(i, possmask)) |
473 | continue; | |
474 | ||
475 | flipped_bit = true; | |
476 | clear_bit(i, possmask); | |
a54694f8 CB |
477 | } |
478 | ||
6f9584d8 | 479 | if (!flipped_bit) { |
59ac3b88 | 480 | DEBUG("No isolated cpus present in cpuset"); |
6f9584d8 CB |
481 | goto on_success; |
482 | } | |
59ac3b88 | 483 | DEBUG("Removed isolated cpus from cpuset"); |
6f9584d8 | 484 | |
a54694f8 | 485 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 | 486 | if (!cpulist) { |
59ac3b88 | 487 | ERROR("Failed to create cpu list"); |
6f9584d8 CB |
488 | goto on_error; |
489 | } | |
a54694f8 CB |
490 | |
491 | copy_parent: | |
492 | *lastslash = oldv; | |
dcbc861e | 493 | free(fpath); |
a54694f8 | 494 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
7cea5905 | 495 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false, 0666); |
6f9584d8 | 496 | if (ret < 0) { |
59ac3b88 | 497 | SYSERROR("Failed to write cpu list to \"%s\"", fpath); |
6f9584d8 CB |
498 | goto on_error; |
499 | } | |
500 | ||
501 | on_success: | |
502 | bret = true; | |
a54694f8 | 503 | |
6f9584d8 | 504 | on_error: |
a54694f8 CB |
505 | free(fpath); |
506 | ||
507 | free(isolcpus); | |
508 | free(isolmask); | |
509 | ||
510 | if (posscpus != cpulist) | |
511 | free(posscpus); | |
512 | free(possmask); | |
513 | ||
514 | free(cpulist); | |
515 | return bret; | |
516 | } | |
517 | ||
e3a3fecf SH |
518 | /* Copy contents of parent(@path)/@file to @path/@file */ |
519 | static bool copy_parent_file(char *path, char *file) | |
520 | { | |
e3a3fecf | 521 | int ret; |
b095a8eb CB |
522 | char *fpath, *lastslash, oldv; |
523 | int len = 0; | |
524 | char *value = NULL; | |
e3a3fecf SH |
525 | |
526 | lastslash = strrchr(path, '/'); | |
b095a8eb CB |
527 | if (!lastslash) { |
528 | ERROR("Failed to detect \"/\" in \"%s\"", path); | |
e3a3fecf SH |
529 | return false; |
530 | } | |
531 | oldv = *lastslash; | |
532 | *lastslash = '\0'; | |
533 | fpath = must_make_path(path, file, NULL); | |
534 | len = lxc_read_from_file(fpath, NULL, 0); | |
535 | if (len <= 0) | |
b095a8eb CB |
536 | goto on_error; |
537 | ||
f25a2044 | 538 | value = must_realloc(NULL, len + 1); |
b095a8eb CB |
539 | ret = lxc_read_from_file(fpath, value, len); |
540 | if (ret != len) | |
541 | goto on_error; | |
e3a3fecf | 542 | free(fpath); |
b095a8eb | 543 | |
e3a3fecf SH |
544 | *lastslash = oldv; |
545 | fpath = must_make_path(path, file, NULL); | |
7cea5905 | 546 | ret = lxc_write_to_file(fpath, value, len, false, 0666); |
e3a3fecf | 547 | if (ret < 0) |
b095a8eb | 548 | SYSERROR("Failed to write \"%s\" to file \"%s\"", value, fpath); |
e3a3fecf SH |
549 | free(fpath); |
550 | free(value); | |
551 | return ret >= 0; | |
552 | ||
b095a8eb CB |
553 | on_error: |
554 | SYSERROR("Failed to read file \"%s\"", fpath); | |
e3a3fecf SH |
555 | free(fpath); |
556 | free(value); | |
557 | return false; | |
558 | } | |
559 | ||
7793add3 CB |
560 | /* Initialize the cpuset hierarchy in first directory of @gname and set |
561 | * cgroup.clone_children so that children inherit settings. Since the | |
562 | * h->base_path is populated by init or ourselves, we know it is already | |
563 | * initialized. | |
e3a3fecf | 564 | */ |
a3926f6a | 565 | static bool cg_legacy_handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf | 566 | { |
7793add3 CB |
567 | int ret; |
568 | char v; | |
569 | char *cgpath, *clonechildrenpath, *slash; | |
e3a3fecf SH |
570 | |
571 | if (!string_in_list(h->controllers, "cpuset")) | |
572 | return true; | |
573 | ||
574 | if (*cgname == '/') | |
575 | cgname++; | |
576 | slash = strchr(cgname, '/'); | |
577 | if (slash) | |
578 | *slash = '\0'; | |
579 | ||
bb221ad1 | 580 | cgpath = must_make_path(h->mountpoint, h->container_base_path, cgname, NULL); |
e3a3fecf SH |
581 | if (slash) |
582 | *slash = '/'; | |
7793add3 CB |
583 | |
584 | ret = mkdir(cgpath, 0755); | |
585 | if (ret < 0) { | |
586 | if (errno != EEXIST) { | |
587 | SYSERROR("Failed to create directory \"%s\"", cgpath); | |
588 | free(cgpath); | |
589 | return false; | |
590 | } | |
e3a3fecf | 591 | } |
6f9584d8 | 592 | |
f8390327 | 593 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
6328fd9c CB |
594 | /* unified hierarchy doesn't have clone_children */ |
595 | if (!file_exists(clonechildrenpath)) { | |
e3a3fecf SH |
596 | free(clonechildrenpath); |
597 | free(cgpath); | |
598 | return true; | |
599 | } | |
7793add3 CB |
600 | |
601 | ret = lxc_read_from_file(clonechildrenpath, &v, 1); | |
602 | if (ret < 0) { | |
603 | SYSERROR("Failed to read file \"%s\"", clonechildrenpath); | |
e3a3fecf SH |
604 | free(clonechildrenpath); |
605 | free(cgpath); | |
606 | return false; | |
607 | } | |
608 | ||
a54694f8 | 609 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
a3926f6a | 610 | if (!cg_legacy_filter_and_set_cpus(cgpath, v == '1')) { |
7793add3 | 611 | SYSERROR("Failed to remove isolated cpus"); |
6f9584d8 CB |
612 | free(clonechildrenpath); |
613 | free(cgpath); | |
a54694f8 | 614 | return false; |
6f9584d8 | 615 | } |
a54694f8 | 616 | |
7793add3 CB |
617 | /* Already set for us by someone else. */ |
618 | if (v == '1') { | |
619 | DEBUG("\"cgroup.clone_children\" was already set to \"1\""); | |
e3a3fecf SH |
620 | free(clonechildrenpath); |
621 | free(cgpath); | |
622 | return true; | |
623 | } | |
624 | ||
625 | /* copy parent's settings */ | |
a54694f8 | 626 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
7793add3 | 627 | SYSERROR("Failed to copy \"cpuset.mems\" settings"); |
e3a3fecf SH |
628 | free(cgpath); |
629 | free(clonechildrenpath); | |
630 | return false; | |
631 | } | |
632 | free(cgpath); | |
633 | ||
7cea5905 | 634 | ret = lxc_write_to_file(clonechildrenpath, "1", 1, false, 0666); |
7793add3 | 635 | if (ret < 0) { |
e3a3fecf | 636 | /* Set clone_children so children inherit our settings */ |
7793add3 | 637 | SYSERROR("Failed to write 1 to \"%s\"", clonechildrenpath); |
e3a3fecf SH |
638 | free(clonechildrenpath); |
639 | return false; | |
640 | } | |
641 | free(clonechildrenpath); | |
642 | return true; | |
643 | } | |
644 | ||
5c0089ae CB |
645 | /* Given two null-terminated lists of strings, return true if any string is in |
646 | * both. | |
ccb4cabe SH |
647 | */ |
648 | static bool controller_lists_intersect(char **l1, char **l2) | |
649 | { | |
650 | int i; | |
651 | ||
652 | if (!l1 || !l2) | |
653 | return false; | |
654 | ||
655 | for (i = 0; l1[i]; i++) { | |
656 | if (string_in_list(l2, l1[i])) | |
657 | return true; | |
658 | } | |
5c0089ae | 659 | |
ccb4cabe SH |
660 | return false; |
661 | } | |
662 | ||
258449e5 CB |
663 | /* For a null-terminated list of controllers @clist, return true if any of those |
664 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
665 | * Realistically, if one is present, all must be present. | |
ccb4cabe SH |
666 | */ |
667 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
668 | { | |
669 | int i; | |
670 | ||
671 | if (!hlist) | |
672 | return false; | |
258449e5 | 673 | |
ccb4cabe SH |
674 | for (i = 0; hlist[i]; i++) |
675 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
676 | return true; | |
ccb4cabe | 677 | |
258449e5 | 678 | return false; |
ccb4cabe SH |
679 | } |
680 | ||
f57ac67f CB |
681 | /* Return true if the controller @entry is found in the null-terminated list of |
682 | * hierarchies @hlist. | |
ccb4cabe SH |
683 | */ |
684 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
685 | { | |
686 | int i; | |
d6337a5f | 687 | |
ccb4cabe SH |
688 | if (!hlist) |
689 | return false; | |
690 | ||
691 | for (i = 0; hlist[i]; i++) | |
692 | if (string_in_list(hlist[i]->controllers, entry)) | |
693 | return true; | |
d6337a5f | 694 | |
ccb4cabe SH |
695 | return false; |
696 | } | |
697 | ||
e1c27ab0 CB |
698 | /* Return true if all of the controllers which we require have been found. The |
699 | * required list is freezer and anything in lxc.cgroup.use. | |
ccb4cabe | 700 | */ |
2202afc9 | 701 | static bool all_controllers_found(struct cgroup_ops *ops) |
ccb4cabe | 702 | { |
b7b18fc5 | 703 | char **cur; |
2202afc9 | 704 | struct hierarchy **hlist = ops->hierarchies; |
ccb4cabe | 705 | |
ccb4cabe | 706 | if (!controller_found(hlist, "freezer")) { |
2202afc9 | 707 | ERROR("No freezer controller mountpoint found"); |
ccb4cabe SH |
708 | return false; |
709 | } | |
710 | ||
2202afc9 | 711 | if (!ops->cgroup_use) |
ccb4cabe | 712 | return true; |
c2712f64 | 713 | |
b7b18fc5 CB |
714 | for (cur = ops->cgroup_use; cur && *cur; cur++) |
715 | if (!controller_found(hlist, *cur)) { | |
716 | ERROR("No %s controller mountpoint found", *cur); | |
ccb4cabe SH |
717 | return false; |
718 | } | |
c2712f64 | 719 | |
ccb4cabe SH |
720 | return true; |
721 | } | |
722 | ||
f205f10c CB |
723 | /* Get the controllers from a mountinfo line There are other ways we could get |
724 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
725 | * could parse the mount options. But we simply assume that the mountpoint must | |
726 | * be /sys/fs/cgroup/controller-list | |
ccb4cabe | 727 | */ |
a3926f6a CB |
728 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, |
729 | int type) | |
ccb4cabe | 730 | { |
f205f10c CB |
731 | /* The fourth field is /sys/fs/cgroup/comma-delimited-controller-list |
732 | * for legacy hierarchies. | |
733 | */ | |
ccb4cabe | 734 | int i; |
411ac6d8 | 735 | char *dup, *p2, *tok; |
0be0d78f | 736 | char *p = line, *sep = ","; |
411ac6d8 | 737 | char **aret = NULL; |
6328fd9c | 738 | |
ccb4cabe | 739 | for (i = 0; i < 4; i++) { |
235f1815 | 740 | p = strchr(p, ' '); |
ccb4cabe SH |
741 | if (!p) |
742 | return NULL; | |
743 | p++; | |
744 | } | |
a55f31bd | 745 | |
f205f10c CB |
746 | /* Note, if we change how mountinfo works, then our caller will need to |
747 | * verify /sys/fs/cgroup/ in this field. | |
748 | */ | |
749 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) { | |
2202afc9 | 750 | ERROR("Found hierarchy not under /sys/fs/cgroup: \"%s\"", p); |
ccb4cabe | 751 | return NULL; |
5059aae9 | 752 | } |
d6337a5f | 753 | |
ccb4cabe | 754 | p += 15; |
235f1815 | 755 | p2 = strchr(p, ' '); |
ccb4cabe | 756 | if (!p2) { |
2202afc9 | 757 | ERROR("Corrupt mountinfo"); |
ccb4cabe SH |
758 | return NULL; |
759 | } | |
760 | *p2 = '\0'; | |
6328fd9c | 761 | |
d6337a5f | 762 | if (type == CGROUP_SUPER_MAGIC) { |
0be0d78f CB |
763 | /* strdup() here for v1 hierarchies. Otherwise |
764 | * lxc_iterate_parts() will destroy mountpoints such as | |
765 | * "/sys/fs/cgroup/cpu,cpuacct". | |
d6337a5f CB |
766 | */ |
767 | dup = strdup(p); | |
768 | if (!dup) | |
769 | return NULL; | |
770 | ||
0be0d78f | 771 | lxc_iterate_parts(tok, dup, sep) { |
d6337a5f | 772 | must_append_controller(klist, nlist, &aret, tok); |
0be0d78f | 773 | } |
d6337a5f CB |
774 | |
775 | free(dup); | |
411ac6d8 | 776 | } |
d6337a5f | 777 | *p2 = ' '; |
f205f10c | 778 | |
d6337a5f CB |
779 | return aret; |
780 | } | |
411ac6d8 | 781 | |
d6337a5f CB |
782 | static char **cg_unified_make_empty_controller(void) |
783 | { | |
784 | int newentry; | |
785 | char **aret = NULL; | |
786 | ||
787 | newentry = append_null_to_list((void ***)&aret); | |
788 | aret[newentry] = NULL; | |
789 | return aret; | |
790 | } | |
791 | ||
792 | static char **cg_unified_get_controllers(const char *file) | |
793 | { | |
794 | char *buf, *tok; | |
0be0d78f | 795 | char *sep = " \t\n"; |
d6337a5f CB |
796 | char **aret = NULL; |
797 | ||
798 | buf = read_file(file); | |
799 | if (!buf) | |
411ac6d8 | 800 | return NULL; |
6328fd9c | 801 | |
0be0d78f | 802 | lxc_iterate_parts(tok, buf, sep) { |
d6337a5f CB |
803 | int newentry; |
804 | char *copy; | |
805 | ||
806 | newentry = append_null_to_list((void ***)&aret); | |
807 | copy = must_copy_string(tok); | |
808 | aret[newentry] = copy; | |
ccb4cabe SH |
809 | } |
810 | ||
d6337a5f | 811 | free(buf); |
ccb4cabe SH |
812 | return aret; |
813 | } | |
814 | ||
2202afc9 | 815 | static struct hierarchy *add_hierarchy(struct hierarchy ***h, char **clist, char *mountpoint, |
bb221ad1 | 816 | char *container_base_path, int type) |
ccb4cabe SH |
817 | { |
818 | struct hierarchy *new; | |
819 | int newentry; | |
820 | ||
f25a2044 | 821 | new = must_realloc(NULL, sizeof(*new)); |
ccb4cabe SH |
822 | new->controllers = clist; |
823 | new->mountpoint = mountpoint; | |
bb221ad1 | 824 | new->container_base_path = container_base_path; |
eb697136 | 825 | new->container_full_path = NULL; |
e09b62f9 | 826 | new->monitor_full_path = NULL; |
d6337a5f | 827 | new->version = type; |
6328fd9c | 828 | |
2202afc9 CB |
829 | newentry = append_null_to_list((void ***)h); |
830 | (*h)[newentry] = new; | |
d6337a5f | 831 | return new; |
ccb4cabe SH |
832 | } |
833 | ||
798c3b33 CB |
834 | /* Get a copy of the mountpoint from @line, which is a line from |
835 | * /proc/self/mountinfo. | |
ccb4cabe | 836 | */ |
a3926f6a | 837 | static char *cg_hybrid_get_mountpoint(char *line) |
ccb4cabe SH |
838 | { |
839 | int i; | |
ccb4cabe | 840 | size_t len; |
798c3b33 CB |
841 | char *p2; |
842 | char *p = line, *sret = NULL; | |
ccb4cabe SH |
843 | |
844 | for (i = 0; i < 4; i++) { | |
235f1815 | 845 | p = strchr(p, ' '); |
ccb4cabe SH |
846 | if (!p) |
847 | return NULL; | |
848 | p++; | |
849 | } | |
d6337a5f | 850 | |
798c3b33 | 851 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) |
d6337a5f CB |
852 | return NULL; |
853 | ||
854 | p2 = strchr(p + 15, ' '); | |
855 | if (!p2) | |
856 | return NULL; | |
857 | *p2 = '\0'; | |
858 | ||
ccb4cabe | 859 | len = strlen(p); |
f25a2044 | 860 | sret = must_realloc(NULL, len + 1); |
ccb4cabe SH |
861 | memcpy(sret, p, len); |
862 | sret[len] = '\0'; | |
863 | return sret; | |
864 | } | |
865 | ||
f523291e | 866 | /* Given a multi-line string, return a null-terminated copy of the current line. */ |
ccb4cabe SH |
867 | static char *copy_to_eol(char *p) |
868 | { | |
235f1815 | 869 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
870 | size_t len; |
871 | ||
872 | if (!p2) | |
873 | return NULL; | |
874 | ||
875 | len = p2 - p; | |
f25a2044 | 876 | sret = must_realloc(NULL, len + 1); |
ccb4cabe SH |
877 | memcpy(sret, p, len); |
878 | sret[len] = '\0'; | |
879 | return sret; | |
880 | } | |
881 | ||
bced39de CB |
882 | /* cgline: pointer to character after the first ':' in a line in a \n-terminated |
883 | * /proc/self/cgroup file. Check whether controller c is present. | |
ccb4cabe SH |
884 | */ |
885 | static bool controller_in_clist(char *cgline, char *c) | |
886 | { | |
0be0d78f | 887 | char *tok, *eol, *tmp; |
ccb4cabe SH |
888 | size_t len; |
889 | ||
235f1815 | 890 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
891 | if (!eol) |
892 | return false; | |
893 | ||
894 | len = eol - cgline; | |
895 | tmp = alloca(len + 1); | |
896 | memcpy(tmp, cgline, len); | |
897 | tmp[len] = '\0'; | |
898 | ||
0be0d78f | 899 | lxc_iterate_parts(tok, tmp, ",") { |
ccb4cabe SH |
900 | if (strcmp(tok, c) == 0) |
901 | return true; | |
902 | } | |
d6337a5f | 903 | |
ccb4cabe SH |
904 | return false; |
905 | } | |
906 | ||
c3ef912e CB |
907 | /* @basecginfo is a copy of /proc/$$/cgroup. Return the current cgroup for |
908 | * @controller. | |
ccb4cabe | 909 | */ |
c3ef912e CB |
910 | static char *cg_hybrid_get_current_cgroup(char *basecginfo, char *controller, |
911 | int type) | |
ccb4cabe SH |
912 | { |
913 | char *p = basecginfo; | |
6328fd9c | 914 | |
d6337a5f CB |
915 | for (;;) { |
916 | bool is_cgv2_base_cgroup = false; | |
917 | ||
6328fd9c | 918 | /* cgroup v2 entry in "/proc/<pid>/cgroup": "0::/some/path" */ |
d6337a5f CB |
919 | if ((type == CGROUP2_SUPER_MAGIC) && (*p == '0')) |
920 | is_cgv2_base_cgroup = true; | |
ccb4cabe | 921 | |
235f1815 | 922 | p = strchr(p, ':'); |
ccb4cabe SH |
923 | if (!p) |
924 | return NULL; | |
925 | p++; | |
d6337a5f CB |
926 | |
927 | if (is_cgv2_base_cgroup || (controller && controller_in_clist(p, controller))) { | |
235f1815 | 928 | p = strchr(p, ':'); |
ccb4cabe SH |
929 | if (!p) |
930 | return NULL; | |
931 | p++; | |
932 | return copy_to_eol(p); | |
933 | } | |
934 | ||
235f1815 | 935 | p = strchr(p, '\n'); |
ccb4cabe SH |
936 | if (!p) |
937 | return NULL; | |
938 | p++; | |
939 | } | |
940 | } | |
941 | ||
ccb4cabe SH |
942 | static void must_append_string(char ***list, char *entry) |
943 | { | |
6dfb18bf | 944 | int newentry; |
ccb4cabe SH |
945 | char *copy; |
946 | ||
6dfb18bf | 947 | newentry = append_null_to_list((void ***)list); |
ccb4cabe SH |
948 | copy = must_copy_string(entry); |
949 | (*list)[newentry] = copy; | |
950 | } | |
951 | ||
d6337a5f | 952 | static int get_existing_subsystems(char ***klist, char ***nlist) |
ccb4cabe SH |
953 | { |
954 | FILE *f; | |
955 | char *line = NULL; | |
956 | size_t len = 0; | |
957 | ||
d6337a5f CB |
958 | f = fopen("/proc/self/cgroup", "r"); |
959 | if (!f) | |
960 | return -1; | |
961 | ||
ccb4cabe | 962 | while (getline(&line, &len, f) != -1) { |
0be0d78f | 963 | char *p, *p2, *tok; |
235f1815 | 964 | p = strchr(line, ':'); |
ccb4cabe SH |
965 | if (!p) |
966 | continue; | |
967 | p++; | |
235f1815 | 968 | p2 = strchr(p, ':'); |
ccb4cabe SH |
969 | if (!p2) |
970 | continue; | |
971 | *p2 = '\0'; | |
ff8d6ee9 | 972 | |
6328fd9c CB |
973 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup |
974 | * contains an entry of the form: | |
ff8d6ee9 CB |
975 | * |
976 | * 0::/some/path | |
977 | * | |
6328fd9c | 978 | * In this case we use "cgroup2" as controller name. |
ff8d6ee9 | 979 | */ |
6328fd9c CB |
980 | if ((p2 - p) == 0) { |
981 | must_append_string(klist, "cgroup2"); | |
ff8d6ee9 | 982 | continue; |
6328fd9c | 983 | } |
ff8d6ee9 | 984 | |
0be0d78f | 985 | lxc_iterate_parts(tok, p, ",") { |
ccb4cabe SH |
986 | if (strncmp(tok, "name=", 5) == 0) |
987 | must_append_string(nlist, tok); | |
988 | else | |
989 | must_append_string(klist, tok); | |
990 | } | |
991 | } | |
992 | ||
993 | free(line); | |
994 | fclose(f); | |
d6337a5f | 995 | return 0; |
ccb4cabe SH |
996 | } |
997 | ||
998 | static void trim(char *s) | |
999 | { | |
7689dfd7 CB |
1000 | size_t len; |
1001 | ||
1002 | len = strlen(s); | |
2c28d76b | 1003 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
1004 | s[--len] = '\0'; |
1005 | } | |
1006 | ||
2202afc9 | 1007 | static void lxc_cgfsng_print_hierarchies(struct cgroup_ops *ops) |
ccb4cabe SH |
1008 | { |
1009 | int i; | |
27d84737 | 1010 | struct hierarchy **it; |
41c33dbe | 1011 | |
2202afc9 CB |
1012 | if (!ops->hierarchies) { |
1013 | TRACE(" No hierarchies found"); | |
ccb4cabe SH |
1014 | return; |
1015 | } | |
27d84737 | 1016 | |
2202afc9 CB |
1017 | TRACE(" Hierarchies:"); |
1018 | for (i = 0, it = ops->hierarchies; it && *it; it++, i++) { | |
ccb4cabe | 1019 | int j; |
27d84737 CB |
1020 | char **cit; |
1021 | ||
bb221ad1 | 1022 | TRACE(" %d: base_cgroup: %s", i, (*it)->container_base_path ? (*it)->container_base_path : "(null)"); |
2202afc9 CB |
1023 | TRACE(" mountpoint: %s", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); |
1024 | TRACE(" controllers:"); | |
a7b0cc4c | 1025 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
2202afc9 | 1026 | TRACE(" %d: %s", j, *cit); |
ccb4cabe SH |
1027 | } |
1028 | } | |
41c33dbe | 1029 | |
a3926f6a CB |
1030 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, |
1031 | char **nlist) | |
41c33dbe SH |
1032 | { |
1033 | int k; | |
a7b0cc4c | 1034 | char **it; |
41c33dbe | 1035 | |
2202afc9 CB |
1036 | TRACE("basecginfo is:"); |
1037 | TRACE("%s", basecginfo); | |
41c33dbe | 1038 | |
a7b0cc4c | 1039 | for (k = 0, it = klist; it && *it; it++, k++) |
2202afc9 | 1040 | TRACE("kernel subsystem %d: %s", k, *it); |
0f71dd9b | 1041 | |
a7b0cc4c | 1042 | for (k = 0, it = nlist; it && *it; it++, k++) |
2202afc9 | 1043 | TRACE("named subsystem %d: %s", k, *it); |
41c33dbe | 1044 | } |
ccb4cabe | 1045 | |
2202afc9 CB |
1046 | static int cgroup_rmdir(struct hierarchy **hierarchies, |
1047 | const char *container_cgroup) | |
c71d83e1 | 1048 | { |
2202afc9 | 1049 | int i; |
d6337a5f | 1050 | |
2202afc9 CB |
1051 | if (!container_cgroup || !hierarchies) |
1052 | return 0; | |
d6337a5f | 1053 | |
2202afc9 CB |
1054 | for (i = 0; hierarchies[i]; i++) { |
1055 | int ret; | |
1056 | struct hierarchy *h = hierarchies[i]; | |
d6337a5f | 1057 | |
eb697136 | 1058 | if (!h->container_full_path) |
2202afc9 CB |
1059 | continue; |
1060 | ||
eb697136 | 1061 | ret = recursive_destroy(h->container_full_path); |
2202afc9 | 1062 | if (ret < 0) |
eb697136 | 1063 | WARN("Failed to destroy \"%s\"", h->container_full_path); |
2202afc9 | 1064 | |
eb697136 CB |
1065 | free(h->container_full_path); |
1066 | h->container_full_path = NULL; | |
2202afc9 | 1067 | } |
d6337a5f | 1068 | |
c71d83e1 | 1069 | return 0; |
d6337a5f CB |
1070 | } |
1071 | ||
2202afc9 CB |
1072 | struct generic_userns_exec_data { |
1073 | struct hierarchy **hierarchies; | |
1074 | const char *container_cgroup; | |
1075 | struct lxc_conf *conf; | |
1076 | uid_t origuid; /* target uid in parent namespace */ | |
1077 | char *path; | |
1078 | }; | |
d6337a5f | 1079 | |
2202afc9 CB |
1080 | static int cgroup_rmdir_wrapper(void *data) |
1081 | { | |
1082 | int ret; | |
1083 | struct generic_userns_exec_data *arg = data; | |
1084 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1085 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
d6337a5f | 1086 | |
2202afc9 CB |
1087 | ret = setresgid(nsgid, nsgid, nsgid); |
1088 | if (ret < 0) { | |
1089 | SYSERROR("Failed to setresgid(%d, %d, %d)", (int)nsgid, | |
1090 | (int)nsgid, (int)nsgid); | |
1091 | return -1; | |
1092 | } | |
d6337a5f | 1093 | |
2202afc9 CB |
1094 | ret = setresuid(nsuid, nsuid, nsuid); |
1095 | if (ret < 0) { | |
1096 | SYSERROR("Failed to setresuid(%d, %d, %d)", (int)nsuid, | |
1097 | (int)nsuid, (int)nsuid); | |
1098 | return -1; | |
1099 | } | |
d6337a5f | 1100 | |
2202afc9 CB |
1101 | ret = setgroups(0, NULL); |
1102 | if (ret < 0 && errno != EPERM) { | |
1103 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1104 | return -1; | |
1105 | } | |
d6337a5f | 1106 | |
2202afc9 | 1107 | return cgroup_rmdir(arg->hierarchies, arg->container_cgroup); |
d6337a5f CB |
1108 | } |
1109 | ||
434c8e15 CB |
1110 | __cgfsng_ops static void cgfsng_payload_destroy(struct cgroup_ops *ops, |
1111 | struct lxc_handler *handler) | |
d6337a5f CB |
1112 | { |
1113 | int ret; | |
2202afc9 | 1114 | struct generic_userns_exec_data wrap; |
bd8ef4e4 | 1115 | |
4160c3a0 | 1116 | wrap.origuid = 0; |
2202afc9 CB |
1117 | wrap.container_cgroup = ops->container_cgroup; |
1118 | wrap.hierarchies = ops->hierarchies; | |
1119 | wrap.conf = handler->conf; | |
4160c3a0 | 1120 | |
2202afc9 CB |
1121 | if (handler->conf && !lxc_list_empty(&handler->conf->id_map)) |
1122 | ret = userns_exec_1(handler->conf, cgroup_rmdir_wrapper, &wrap, | |
bd8ef4e4 | 1123 | "cgroup_rmdir_wrapper"); |
ccb4cabe | 1124 | else |
2202afc9 | 1125 | ret = cgroup_rmdir(ops->hierarchies, ops->container_cgroup); |
bd8ef4e4 CB |
1126 | if (ret < 0) { |
1127 | WARN("Failed to destroy cgroups"); | |
ccb4cabe | 1128 | return; |
ccb4cabe | 1129 | } |
ccb4cabe SH |
1130 | } |
1131 | ||
434c8e15 CB |
1132 | __cgfsng_ops static void cgfsng_monitor_destroy(struct cgroup_ops *ops, |
1133 | struct lxc_handler *handler) | |
1134 | { | |
1135 | int len; | |
1136 | char *pivot_path; | |
1137 | struct lxc_conf *conf = handler->conf; | |
1138 | char pidstr[INTTYPE_TO_STRLEN(pid_t)]; | |
1139 | ||
1140 | if (!ops->hierarchies) | |
1141 | return; | |
1142 | ||
1143 | len = snprintf(pidstr, sizeof(pidstr), "%d", handler->monitor_pid); | |
1144 | if (len < 0 || (size_t)len >= sizeof(pidstr)) | |
1145 | return; | |
1146 | ||
1147 | for (int i = 0; ops->hierarchies[i]; i++) { | |
1148 | int ret; | |
23e5c045 | 1149 | char *chop; |
ecedb5de | 1150 | char pivot_cgroup[] = PIVOT_CGROUP; |
434c8e15 CB |
1151 | struct hierarchy *h = ops->hierarchies[i]; |
1152 | ||
1153 | if (!h->monitor_full_path) | |
1154 | continue; | |
1155 | ||
1156 | if (conf && conf->cgroup_meta.dir) | |
1157 | pivot_path = must_make_path(h->mountpoint, | |
1158 | h->container_base_path, | |
1159 | conf->cgroup_meta.dir, | |
625ad37b | 1160 | PIVOT_CGROUP, |
434c8e15 CB |
1161 | "cgroup.procs", NULL); |
1162 | else | |
1163 | pivot_path = must_make_path(h->mountpoint, | |
1164 | h->container_base_path, | |
625ad37b | 1165 | PIVOT_CGROUP, |
434c8e15 CB |
1166 | "cgroup.procs", NULL); |
1167 | ||
23e5c045 CB |
1168 | chop = strrchr(pivot_path, '/'); |
1169 | if (chop) | |
1170 | *chop = '\0'; | |
1171 | ||
ecedb5de CB |
1172 | /* |
1173 | * Make sure not to pass in the ro string literal PIVOT_CGROUP | |
1174 | * here. | |
1175 | */ | |
1176 | if (!cg_legacy_handle_cpuset_hierarchy(h, pivot_cgroup)) { | |
1177 | WARN("Failed to handle legacy cpuset controller"); | |
1178 | goto next; | |
1179 | } | |
1180 | ||
434c8e15 | 1181 | ret = mkdir_p(pivot_path, 0755); |
d5fc4dd4 CB |
1182 | if (ret < 0 && errno != EEXIST) { |
1183 | SYSWARN("Failed to create cgroup \"%s\"\n", pivot_path); | |
434c8e15 | 1184 | goto next; |
d5fc4dd4 | 1185 | } |
434c8e15 | 1186 | |
23e5c045 CB |
1187 | if (chop) |
1188 | *chop = '/'; | |
1189 | ||
434c8e15 CB |
1190 | /* Move ourselves into the pivot cgroup to delete our own |
1191 | * cgroup. | |
1192 | */ | |
1193 | ret = lxc_write_to_file(pivot_path, pidstr, len, false, 0666); | |
d5fc4dd4 CB |
1194 | if (ret != 0) { |
1195 | SYSWARN("Failed to move monitor %s to \"%s\"\n", pidstr, pivot_path); | |
434c8e15 | 1196 | goto next; |
d5fc4dd4 | 1197 | } |
434c8e15 CB |
1198 | |
1199 | ret = recursive_destroy(h->monitor_full_path); | |
1200 | if (ret < 0) | |
1201 | WARN("Failed to destroy \"%s\"", h->monitor_full_path); | |
1202 | ||
1203 | next: | |
1204 | free(pivot_path); | |
1205 | } | |
1206 | } | |
1207 | ||
a3926f6a | 1208 | static bool cg_unified_create_cgroup(struct hierarchy *h, char *cgname) |
0c3deb94 | 1209 | { |
0c3deb94 | 1210 | size_t i, parts_len; |
389d44ec | 1211 | char **it; |
0c3deb94 CB |
1212 | size_t full_len = 0; |
1213 | char *add_controllers = NULL, *cgroup = NULL; | |
1214 | char **parts = NULL; | |
1215 | bool bret = false; | |
1216 | ||
1217 | if (h->version != CGROUP2_SUPER_MAGIC) | |
1218 | return true; | |
1219 | ||
1220 | if (!h->controllers) | |
1221 | return true; | |
1222 | ||
1223 | /* For now we simply enable all controllers that we have detected by | |
1224 | * creating a string like "+memory +pids +cpu +io". | |
1225 | * TODO: In the near future we might want to support "-<controller>" | |
1226 | * etc. but whether supporting semantics like this make sense will need | |
1227 | * some thinking. | |
1228 | */ | |
1229 | for (it = h->controllers; it && *it; it++) { | |
64e82f8b DJ |
1230 | full_len += strlen(*it) + 2; |
1231 | add_controllers = must_realloc(add_controllers, full_len + 1); | |
1232 | ||
1233 | if (h->controllers[0] == *it) | |
1234 | add_controllers[0] = '\0'; | |
1235 | ||
3ebe2fbd DJ |
1236 | (void)strlcat(add_controllers, "+", full_len + 1); |
1237 | (void)strlcat(add_controllers, *it, full_len + 1); | |
64e82f8b DJ |
1238 | |
1239 | if ((it + 1) && *(it + 1)) | |
3ebe2fbd | 1240 | (void)strlcat(add_controllers, " ", full_len + 1); |
0c3deb94 CB |
1241 | } |
1242 | ||
1243 | parts = lxc_string_split(cgname, '/'); | |
1244 | if (!parts) | |
1245 | goto on_error; | |
64e82f8b | 1246 | |
0c3deb94 CB |
1247 | parts_len = lxc_array_len((void **)parts); |
1248 | if (parts_len > 0) | |
1249 | parts_len--; | |
1250 | ||
bb221ad1 | 1251 | cgroup = must_make_path(h->mountpoint, h->container_base_path, NULL); |
0c3deb94 CB |
1252 | for (i = 0; i < parts_len; i++) { |
1253 | int ret; | |
1254 | char *target; | |
1255 | ||
1256 | cgroup = must_append_path(cgroup, parts[i], NULL); | |
1257 | target = must_make_path(cgroup, "cgroup.subtree_control", NULL); | |
7cea5905 | 1258 | ret = lxc_write_to_file(target, add_controllers, full_len, false, 0666); |
0c3deb94 CB |
1259 | free(target); |
1260 | if (ret < 0) { | |
1261 | SYSERROR("Could not enable \"%s\" controllers in the " | |
1262 | "unified cgroup \"%s\"", add_controllers, cgroup); | |
1263 | goto on_error; | |
1264 | } | |
1265 | } | |
1266 | ||
1267 | bret = true; | |
1268 | ||
1269 | on_error: | |
1270 | lxc_free_array((void **)parts, free); | |
1271 | free(add_controllers); | |
1272 | free(cgroup); | |
1273 | return bret; | |
1274 | } | |
1275 | ||
6099dd5a CB |
1276 | static int mkdir_eexist_on_last(const char *dir, mode_t mode) |
1277 | { | |
1278 | const char *tmp = dir; | |
1279 | const char *orig = dir; | |
1280 | size_t orig_len; | |
1281 | ||
1282 | orig_len = strlen(dir); | |
1283 | do { | |
1284 | int ret; | |
1285 | size_t cur_len; | |
1286 | char *makeme; | |
1287 | ||
1288 | dir = tmp + strspn(tmp, "/"); | |
1289 | tmp = dir + strcspn(dir, "/"); | |
1290 | ||
1291 | errno = ENOMEM; | |
1292 | cur_len = dir - orig; | |
1293 | makeme = strndup(orig, cur_len); | |
1294 | if (!makeme) | |
1295 | return -1; | |
1296 | ||
1297 | ret = mkdir(makeme, mode); | |
1298 | if (ret < 0) { | |
1299 | if ((errno != EEXIST) || (orig_len == cur_len)) { | |
1300 | SYSERROR("Failed to create directory \"%s\"", makeme); | |
1301 | free(makeme); | |
1302 | return -1; | |
1303 | } | |
1304 | } | |
1305 | free(makeme); | |
1306 | ||
1307 | } while (tmp != dir); | |
1308 | ||
1309 | return 0; | |
1310 | } | |
1311 | ||
72068e74 CB |
1312 | static bool monitor_create_path_for_hierarchy(struct hierarchy *h, char *cgname) |
1313 | { | |
1314 | int ret; | |
1315 | ||
ef185360 CB |
1316 | if (!cg_legacy_handle_cpuset_hierarchy(h, cgname)) { |
1317 | ERROR("Failed to handle legacy cpuset controller"); | |
1318 | return false; | |
1319 | } | |
1320 | ||
72068e74 | 1321 | h->monitor_full_path = must_make_path(h->mountpoint, h->container_base_path, cgname, NULL); |
6099dd5a CB |
1322 | ret = mkdir_eexist_on_last(h->monitor_full_path, 0755); |
1323 | if (ret < 0) { | |
1324 | ERROR("Failed to create cgroup \"%s\"", h->monitor_full_path); | |
ee455be4 CB |
1325 | return false; |
1326 | } | |
72068e74 | 1327 | |
72068e74 CB |
1328 | return cg_unified_create_cgroup(h, cgname); |
1329 | } | |
1330 | ||
1331 | static bool container_create_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
ccb4cabe | 1332 | { |
0c3deb94 CB |
1333 | int ret; |
1334 | ||
ef185360 CB |
1335 | if (!cg_legacy_handle_cpuset_hierarchy(h, cgname)) { |
1336 | ERROR("Failed to handle legacy cpuset controller"); | |
1337 | return false; | |
1338 | } | |
1339 | ||
bb221ad1 | 1340 | h->container_full_path = must_make_path(h->mountpoint, h->container_base_path, cgname, NULL); |
6099dd5a CB |
1341 | ret = mkdir_eexist_on_last(h->container_full_path, 0755); |
1342 | if (ret < 0) { | |
1343 | ERROR("Failed to create cgroup \"%s\"", h->container_full_path); | |
d8da679e | 1344 | return false; |
6f9584d8 | 1345 | } |
0c3deb94 | 1346 | |
a3926f6a | 1347 | return cg_unified_create_cgroup(h, cgname); |
ccb4cabe SH |
1348 | } |
1349 | ||
72068e74 | 1350 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname, bool monitor) |
ccb4cabe | 1351 | { |
e56639fb | 1352 | int ret; |
72068e74 CB |
1353 | char *full_path; |
1354 | ||
1355 | if (monitor) | |
1356 | full_path = h->monitor_full_path; | |
1357 | else | |
1358 | full_path = h->container_full_path; | |
e56639fb | 1359 | |
72068e74 | 1360 | ret = rmdir(full_path); |
e56639fb | 1361 | if (ret < 0) |
72068e74 CB |
1362 | SYSERROR("Failed to rmdir(\"%s\") from failed creation attempt", full_path); |
1363 | ||
1364 | free(full_path); | |
1365 | ||
1366 | if (monitor) | |
1367 | h->monitor_full_path = NULL; | |
1368 | else | |
1369 | h->container_full_path = NULL; | |
1370 | } | |
1371 | ||
b857f4be | 1372 | __cgfsng_ops static inline bool cgfsng_monitor_create(struct cgroup_ops *ops, |
d28779d9 | 1373 | struct lxc_handler *handler) |
72068e74 | 1374 | { |
5ce03bc0 | 1375 | char *monitor_cgroup, *offset, *tmp; |
ebc10afe | 1376 | int i, idx = 0; |
5ce03bc0 | 1377 | size_t len; |
72068e74 CB |
1378 | bool bret = false; |
1379 | struct lxc_conf *conf = handler->conf; | |
1380 | ||
1381 | if (!conf) | |
1382 | return bret; | |
e56639fb | 1383 | |
72068e74 | 1384 | if (conf->cgroup_meta.dir) |
5ce03bc0 CB |
1385 | tmp = lxc_string_join("/", |
1386 | (const char *[]){conf->cgroup_meta.dir, | |
1387 | ops->monitor_pattern, | |
1388 | handler->name, NULL}, | |
1389 | false); | |
72068e74 | 1390 | else |
5ce03bc0 CB |
1391 | tmp = must_make_path(ops->monitor_pattern, handler->name, NULL); |
1392 | if (!tmp) | |
72068e74 CB |
1393 | return bret; |
1394 | ||
5ce03bc0 | 1395 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
5407d095 | 1396 | monitor_cgroup = must_realloc(tmp, len); |
5ce03bc0 | 1397 | offset = monitor_cgroup + len - 5; |
5407d095 | 1398 | *offset = 0; |
5ce03bc0 CB |
1399 | |
1400 | do { | |
1401 | if (idx) { | |
1402 | int ret = snprintf(offset, 5, "-%d", idx); | |
1403 | if (ret < 0 || (size_t)ret >= 5) | |
1404 | goto on_error; | |
72068e74 | 1405 | } |
72068e74 | 1406 | |
ebc10afe | 1407 | for (i = 0; ops->hierarchies[i]; i++) { |
5ce03bc0 CB |
1408 | if (!monitor_create_path_for_hierarchy(ops->hierarchies[i], monitor_cgroup)) { |
1409 | ERROR("Failed to create cgroup \"%s\"", ops->hierarchies[i]->monitor_full_path); | |
1410 | free(ops->hierarchies[i]->container_full_path); | |
1411 | ops->hierarchies[i]->container_full_path = NULL; | |
1412 | ||
1413 | for (int j = 0; j < i; j++) | |
1414 | remove_path_for_hierarchy(ops->hierarchies[j], monitor_cgroup, true); | |
1415 | ||
1416 | idx++; | |
1417 | break; | |
1418 | } | |
1419 | } | |
ebc10afe | 1420 | } while (ops->hierarchies[i] && idx > 0 && idx < 1000); |
5ce03bc0 | 1421 | |
529822a4 | 1422 | if (idx < 1000) { |
5ce03bc0 | 1423 | bret = true; |
529822a4 CB |
1424 | INFO("The monitor process uses \"%s\" as cgroup", monitor_cgroup); |
1425 | } | |
72068e74 CB |
1426 | |
1427 | on_error: | |
1428 | free(monitor_cgroup); | |
1429 | ||
1430 | return bret; | |
ccb4cabe SH |
1431 | } |
1432 | ||
cecad0c1 CB |
1433 | /* Try to create the same cgroup in all hierarchies. Start with cgroup_pattern; |
1434 | * next cgroup_pattern-1, -2, ..., -999. | |
ccb4cabe | 1435 | */ |
b857f4be | 1436 | __cgfsng_ops static inline bool cgfsng_payload_create(struct cgroup_ops *ops, |
6439f06e | 1437 | struct lxc_handler *handler) |
ccb4cabe | 1438 | { |
bb30b52a | 1439 | int i; |
ccb4cabe | 1440 | size_t len; |
0c3deb94 | 1441 | char *container_cgroup, *offset, *tmp; |
7d531e9b | 1442 | int idx = 0; |
2202afc9 | 1443 | struct lxc_conf *conf = handler->conf; |
ccb4cabe | 1444 | |
2202afc9 CB |
1445 | if (ops->container_cgroup) { |
1446 | WARN("cgfsng_create called a second time: %s", ops->container_cgroup); | |
ccb4cabe | 1447 | return false; |
2202afc9 | 1448 | } |
43654d34 | 1449 | |
2202afc9 | 1450 | if (!conf) |
ccb4cabe | 1451 | return false; |
ccb4cabe | 1452 | |
2202afc9 | 1453 | if (conf->cgroup_meta.dir) |
3ec12d39 | 1454 | tmp = lxc_string_join("/", (const char *[]){conf->cgroup_meta.dir, handler->name, NULL}, false); |
43654d34 | 1455 | else |
2202afc9 | 1456 | tmp = lxc_string_replace("%n", handler->name, ops->cgroup_pattern); |
ccb4cabe SH |
1457 | if (!tmp) { |
1458 | ERROR("Failed expanding cgroup name pattern"); | |
1459 | return false; | |
1460 | } | |
64e82f8b | 1461 | |
1a0e70ac | 1462 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
f25a2044 | 1463 | container_cgroup = must_realloc(NULL, len); |
64e82f8b | 1464 | (void)strlcpy(container_cgroup, tmp, len); |
ccb4cabe | 1465 | free(tmp); |
0c3deb94 | 1466 | offset = container_cgroup + len - 5; |
ccb4cabe SH |
1467 | |
1468 | again: | |
95adfe93 SH |
1469 | if (idx == 1000) { |
1470 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1471 | goto out_free; |
95adfe93 | 1472 | } |
cecad0c1 | 1473 | |
66b66624 | 1474 | if (idx) { |
bb30b52a CB |
1475 | int ret; |
1476 | ||
66b66624 CB |
1477 | ret = snprintf(offset, 5, "-%d", idx); |
1478 | if (ret < 0 || (size_t)ret >= 5) { | |
1479 | FILE *f = fopen("/dev/null", "w"); | |
97ebced3 | 1480 | if (f) { |
66b66624 CB |
1481 | fprintf(f, "Workaround for GCC7 bug: " |
1482 | "https://gcc.gnu.org/bugzilla/" | |
1483 | "show_bug.cgi?id=78969"); | |
1484 | fclose(f); | |
1485 | } | |
1486 | } | |
1487 | } | |
cecad0c1 | 1488 | |
2202afc9 | 1489 | for (i = 0; ops->hierarchies[i]; i++) { |
72068e74 | 1490 | if (!container_create_path_for_hierarchy(ops->hierarchies[i], container_cgroup)) { |
eb697136 CB |
1491 | ERROR("Failed to create cgroup \"%s\"", ops->hierarchies[i]->container_full_path); |
1492 | free(ops->hierarchies[i]->container_full_path); | |
1493 | ops->hierarchies[i]->container_full_path = NULL; | |
72068e74 CB |
1494 | for (int j = 0; j < i; j++) |
1495 | remove_path_for_hierarchy(ops->hierarchies[j], container_cgroup, false); | |
ccb4cabe SH |
1496 | idx++; |
1497 | goto again; | |
1498 | } | |
1499 | } | |
cecad0c1 | 1500 | |
2202afc9 | 1501 | ops->container_cgroup = container_cgroup; |
529822a4 | 1502 | INFO("The container uses \"%s\" as cgroup", container_cgroup); |
cecad0c1 | 1503 | |
ccb4cabe SH |
1504 | return true; |
1505 | ||
1506 | out_free: | |
0c3deb94 | 1507 | free(container_cgroup); |
cecad0c1 | 1508 | |
ccb4cabe SH |
1509 | return false; |
1510 | } | |
1511 | ||
b857f4be | 1512 | __cgfsng_ops static bool __do_cgroup_enter(struct cgroup_ops *ops, pid_t pid, |
eeef32bb | 1513 | bool monitor) |
ccb4cabe | 1514 | { |
eeef32bb | 1515 | int len; |
a3650c0c | 1516 | char pidstr[INTTYPE_TO_STRLEN(pid_t)]; |
ccb4cabe | 1517 | |
a3650c0c CB |
1518 | len = snprintf(pidstr, sizeof(pidstr), "%d", pid); |
1519 | if (len < 0 || (size_t)len >= sizeof(pidstr)) | |
ccb4cabe SH |
1520 | return false; |
1521 | ||
eeef32bb | 1522 | for (int i = 0; ops->hierarchies[i]; i++) { |
08768001 | 1523 | int ret; |
eeef32bb | 1524 | char *path; |
08768001 | 1525 | |
eeef32bb CB |
1526 | if (monitor) |
1527 | path = must_make_path(ops->hierarchies[i]->monitor_full_path, | |
1528 | "cgroup.procs", NULL); | |
1529 | else | |
1530 | path = must_make_path(ops->hierarchies[i]->container_full_path, | |
1531 | "cgroup.procs", NULL); | |
1532 | ret = lxc_write_to_file(path, pidstr, len, false, 0666); | |
08768001 | 1533 | if (ret != 0) { |
eeef32bb CB |
1534 | SYSERROR("Failed to enter cgroup \"%s\"", path); |
1535 | free(path); | |
ccb4cabe SH |
1536 | return false; |
1537 | } | |
eeef32bb | 1538 | free(path); |
ccb4cabe SH |
1539 | } |
1540 | ||
1541 | return true; | |
1542 | } | |
1543 | ||
b857f4be | 1544 | __cgfsng_ops static bool cgfsng_monitor_enter(struct cgroup_ops *ops, pid_t pid) |
eeef32bb CB |
1545 | { |
1546 | return __do_cgroup_enter(ops, pid, true); | |
1547 | } | |
1548 | ||
1549 | static bool cgfsng_payload_enter(struct cgroup_ops *ops, pid_t pid) | |
1550 | { | |
1551 | return __do_cgroup_enter(ops, pid, false); | |
1552 | } | |
1553 | ||
6efacf80 CB |
1554 | static int chowmod(char *path, uid_t chown_uid, gid_t chown_gid, |
1555 | mode_t chmod_mode) | |
1556 | { | |
1557 | int ret; | |
1558 | ||
1559 | ret = chown(path, chown_uid, chown_gid); | |
1560 | if (ret < 0) { | |
a24c5678 | 1561 | SYSWARN("Failed to chown(%s, %d, %d)", path, (int)chown_uid, (int)chown_gid); |
6efacf80 CB |
1562 | return -1; |
1563 | } | |
1564 | ||
1565 | ret = chmod(path, chmod_mode); | |
1566 | if (ret < 0) { | |
a24c5678 | 1567 | SYSWARN("Failed to chmod(%s, %d)", path, (int)chmod_mode); |
6efacf80 CB |
1568 | return -1; |
1569 | } | |
1570 | ||
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | /* chgrp the container cgroups to container group. We leave | |
c0888dfe SH |
1575 | * the container owner as cgroup owner. So we must make the |
1576 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1577 | * |
1578 | * Also chown the tasks and cgroup.procs files. Those may not | |
1579 | * exist depending on kernel version. | |
c0888dfe | 1580 | */ |
ccb4cabe SH |
1581 | static int chown_cgroup_wrapper(void *data) |
1582 | { | |
6efacf80 | 1583 | int i, ret; |
4160c3a0 CB |
1584 | uid_t destuid; |
1585 | struct generic_userns_exec_data *arg = data; | |
1586 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1587 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
ccb4cabe | 1588 | |
6efacf80 CB |
1589 | ret = setresgid(nsgid, nsgid, nsgid); |
1590 | if (ret < 0) { | |
1591 | SYSERROR("Failed to setresgid(%d, %d, %d)", | |
1592 | (int)nsgid, (int)nsgid, (int)nsgid); | |
1593 | return -1; | |
1594 | } | |
1595 | ||
1596 | ret = setresuid(nsuid, nsuid, nsuid); | |
1597 | if (ret < 0) { | |
1598 | SYSERROR("Failed to setresuid(%d, %d, %d)", | |
1599 | (int)nsuid, (int)nsuid, (int)nsuid); | |
1600 | return -1; | |
1601 | } | |
1602 | ||
1603 | ret = setgroups(0, NULL); | |
1604 | if (ret < 0 && errno != EPERM) { | |
1605 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1606 | return -1; | |
1607 | } | |
ccb4cabe SH |
1608 | |
1609 | destuid = get_ns_uid(arg->origuid); | |
b962868f CB |
1610 | if (destuid == LXC_INVALID_UID) |
1611 | destuid = 0; | |
ccb4cabe | 1612 | |
2202afc9 | 1613 | for (i = 0; arg->hierarchies[i]; i++) { |
6efacf80 | 1614 | char *fullpath; |
eb697136 | 1615 | char *path = arg->hierarchies[i]->container_full_path; |
43647298 | 1616 | |
63e42fee | 1617 | ret = chowmod(path, destuid, nsgid, 0775); |
6efacf80 | 1618 | if (ret < 0) |
ccb4cabe | 1619 | return -1; |
c0888dfe | 1620 | |
6efacf80 CB |
1621 | /* Failures to chown() these are inconvenient but not |
1622 | * detrimental We leave these owned by the container launcher, | |
1623 | * so that container root can write to the files to attach. We | |
1624 | * chmod() them 664 so that container systemd can write to the | |
1625 | * files (which systemd in wily insists on doing). | |
ab8f5424 | 1626 | */ |
6efacf80 | 1627 | |
2202afc9 | 1628 | if (arg->hierarchies[i]->version == CGROUP_SUPER_MAGIC) { |
6efacf80 CB |
1629 | fullpath = must_make_path(path, "tasks", NULL); |
1630 | (void)chowmod(fullpath, destuid, nsgid, 0664); | |
1631 | free(fullpath); | |
1632 | } | |
43647298 SH |
1633 | |
1634 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
2202afc9 | 1635 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
ccb4cabe | 1636 | free(fullpath); |
0e17357c | 1637 | |
2202afc9 | 1638 | if (arg->hierarchies[i]->version != CGROUP2_SUPER_MAGIC) |
0e17357c CB |
1639 | continue; |
1640 | ||
1641 | fullpath = must_make_path(path, "cgroup.subtree_control", NULL); | |
6efacf80 | 1642 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c CB |
1643 | free(fullpath); |
1644 | ||
1645 | fullpath = must_make_path(path, "cgroup.threads", NULL); | |
6efacf80 | 1646 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c | 1647 | free(fullpath); |
ccb4cabe SH |
1648 | } |
1649 | ||
1650 | return 0; | |
1651 | } | |
1652 | ||
b857f4be | 1653 | __cgfsng_ops static bool cgfsng_chown(struct cgroup_ops *ops, |
fb55e009 | 1654 | struct lxc_conf *conf) |
ccb4cabe | 1655 | { |
4160c3a0 | 1656 | struct generic_userns_exec_data wrap; |
ccb4cabe | 1657 | |
ccb4cabe SH |
1658 | if (lxc_list_empty(&conf->id_map)) |
1659 | return true; | |
1660 | ||
ccb4cabe | 1661 | wrap.origuid = geteuid(); |
4160c3a0 | 1662 | wrap.path = NULL; |
2202afc9 | 1663 | wrap.hierarchies = ops->hierarchies; |
4160c3a0 | 1664 | wrap.conf = conf; |
ccb4cabe | 1665 | |
c9b7c33e CB |
1666 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1667 | "chown_cgroup_wrapper") < 0) { | |
f7faba6c | 1668 | ERROR("Error requesting cgroup chown in new user namespace"); |
ccb4cabe SH |
1669 | return false; |
1670 | } | |
1671 | ||
1672 | return true; | |
1673 | } | |
1674 | ||
8aa1044f SH |
1675 | /* cgroup-full:* is done, no need to create subdirs */ |
1676 | static bool cg_mount_needs_subdirs(int type) | |
1677 | { | |
1678 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
1679 | return false; | |
a3926f6a | 1680 | |
8aa1044f SH |
1681 | return true; |
1682 | } | |
1683 | ||
886cac86 CB |
1684 | /* After $rootfs/sys/fs/container/controller/the/cg/path has been created, |
1685 | * remount controller ro if needed and bindmount the cgroupfs onto | |
25fa6f8c | 1686 | * control/the/cg/path. |
8aa1044f | 1687 | */ |
6812d833 CB |
1688 | static int cg_legacy_mount_controllers(int type, struct hierarchy *h, |
1689 | char *controllerpath, char *cgpath, | |
1690 | const char *container_cgroup) | |
8aa1044f | 1691 | { |
5285689c | 1692 | int ret, remount_flags; |
886cac86 CB |
1693 | char *sourcepath; |
1694 | int flags = MS_BIND; | |
1695 | ||
8aa1044f | 1696 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { |
886cac86 CB |
1697 | ret = mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL); |
1698 | if (ret < 0) { | |
1699 | SYSERROR("Failed to bind mount \"%s\" onto \"%s\"", | |
1700 | controllerpath, controllerpath); | |
8aa1044f SH |
1701 | return -1; |
1702 | } | |
886cac86 | 1703 | |
5285689c CB |
1704 | remount_flags = add_required_remount_flags(controllerpath, |
1705 | controllerpath, | |
1706 | flags | MS_REMOUNT); | |
886cac86 | 1707 | ret = mount(controllerpath, controllerpath, "cgroup", |
8186c5c7 CB |
1708 | remount_flags | MS_REMOUNT | MS_BIND | MS_RDONLY, |
1709 | NULL); | |
886cac86 CB |
1710 | if (ret < 0) { |
1711 | SYSERROR("Failed to remount \"%s\" ro", controllerpath); | |
8aa1044f SH |
1712 | return -1; |
1713 | } | |
886cac86 | 1714 | |
8aa1044f SH |
1715 | INFO("Remounted %s read-only", controllerpath); |
1716 | } | |
886cac86 | 1717 | |
bb221ad1 | 1718 | sourcepath = must_make_path(h->mountpoint, h->container_base_path, |
886cac86 | 1719 | container_cgroup, NULL); |
8aa1044f SH |
1720 | if (type == LXC_AUTO_CGROUP_RO) |
1721 | flags |= MS_RDONLY; | |
886cac86 CB |
1722 | |
1723 | ret = mount(sourcepath, cgpath, "cgroup", flags, NULL); | |
1724 | if (ret < 0) { | |
1725 | SYSERROR("Failed to mount \"%s\" onto \"%s\"", h->controllers[0], cgpath); | |
8aa1044f | 1726 | free(sourcepath); |
8aa1044f SH |
1727 | return -1; |
1728 | } | |
886cac86 | 1729 | INFO("Mounted \"%s\" onto \"%s\"", h->controllers[0], cgpath); |
f8c40ffa L |
1730 | |
1731 | if (flags & MS_RDONLY) { | |
5285689c CB |
1732 | remount_flags = add_required_remount_flags(sourcepath, cgpath, |
1733 | flags | MS_REMOUNT); | |
1734 | ret = mount(sourcepath, cgpath, "cgroup", remount_flags, NULL); | |
886cac86 CB |
1735 | if (ret < 0) { |
1736 | SYSERROR("Failed to remount \"%s\" ro", cgpath); | |
f8c40ffa | 1737 | free(sourcepath); |
f8c40ffa L |
1738 | return -1; |
1739 | } | |
5285689c | 1740 | INFO("Remounted %s read-only", cgpath); |
f8c40ffa L |
1741 | } |
1742 | ||
8aa1044f | 1743 | free(sourcepath); |
886cac86 | 1744 | INFO("Completed second stage cgroup automounts for \"%s\"", cgpath); |
8aa1044f SH |
1745 | return 0; |
1746 | } | |
1747 | ||
6812d833 CB |
1748 | /* __cg_mount_direct |
1749 | * | |
1750 | * Mount cgroup hierarchies directly without using bind-mounts. The main | |
1751 | * uses-cases are mounting cgroup hierarchies in cgroup namespaces and mounting | |
1752 | * cgroups for the LXC_AUTO_CGROUP_FULL option. | |
1753 | */ | |
1754 | static int __cg_mount_direct(int type, struct hierarchy *h, | |
1755 | const char *controllerpath) | |
b635e92d CB |
1756 | { |
1757 | int ret; | |
1758 | char *controllers = NULL; | |
a760603e CB |
1759 | char *fstype = "cgroup2"; |
1760 | unsigned long flags = 0; | |
b635e92d | 1761 | |
a760603e CB |
1762 | flags |= MS_NOSUID; |
1763 | flags |= MS_NOEXEC; | |
1764 | flags |= MS_NODEV; | |
1765 | flags |= MS_RELATIME; | |
1766 | ||
1767 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_FULL_RO) | |
1768 | flags |= MS_RDONLY; | |
1769 | ||
d6337a5f | 1770 | if (h->version != CGROUP2_SUPER_MAGIC) { |
a760603e CB |
1771 | controllers = lxc_string_join(",", (const char **)h->controllers, false); |
1772 | if (!controllers) | |
1773 | return -ENOMEM; | |
1774 | fstype = "cgroup"; | |
b635e92d CB |
1775 | } |
1776 | ||
a760603e | 1777 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); |
b635e92d CB |
1778 | free(controllers); |
1779 | if (ret < 0) { | |
6812d833 | 1780 | SYSERROR("Failed to mount \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1781 | return -1; |
1782 | } | |
1783 | ||
6812d833 | 1784 | DEBUG("Mounted \"%s\" with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
1785 | return 0; |
1786 | } | |
1787 | ||
6812d833 CB |
1788 | static inline int cg_mount_in_cgroup_namespace(int type, struct hierarchy *h, |
1789 | const char *controllerpath) | |
1790 | { | |
1791 | return __cg_mount_direct(type, h, controllerpath); | |
1792 | } | |
1793 | ||
1794 | static inline int cg_mount_cgroup_full(int type, struct hierarchy *h, | |
1795 | const char *controllerpath) | |
1796 | { | |
1797 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1798 | return 0; | |
1799 | ||
1800 | return __cg_mount_direct(type, h, controllerpath); | |
1801 | } | |
1802 | ||
b857f4be | 1803 | __cgfsng_ops static bool cgfsng_mount(struct cgroup_ops *ops, |
fb55e009 CB |
1804 | struct lxc_handler *handler, |
1805 | const char *root, int type) | |
ccb4cabe | 1806 | { |
3f69fb12 | 1807 | int i, ret; |
8aa1044f | 1808 | char *tmpfspath = NULL; |
affd10fa | 1809 | bool has_cgns = false, retval = false, wants_force_mount = false; |
8aa1044f SH |
1810 | |
1811 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
1812 | return true; | |
1813 | ||
3f69fb12 SY |
1814 | if (type & LXC_AUTO_CGROUP_FORCE) { |
1815 | type &= ~LXC_AUTO_CGROUP_FORCE; | |
1816 | wants_force_mount = true; | |
1817 | } | |
b635e92d | 1818 | |
3f69fb12 SY |
1819 | if (!wants_force_mount){ |
1820 | if (!lxc_list_empty(&handler->conf->keepcaps)) | |
1821 | wants_force_mount = !in_caplist(CAP_SYS_ADMIN, &handler->conf->keepcaps); | |
1822 | else | |
1823 | wants_force_mount = in_caplist(CAP_SYS_ADMIN, &handler->conf->caps); | |
1824 | } | |
8aa1044f | 1825 | |
3f69fb12 SY |
1826 | has_cgns = cgns_supported(); |
1827 | if (has_cgns && !wants_force_mount) | |
1828 | return true; | |
8aa1044f SH |
1829 | |
1830 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
1831 | type = LXC_AUTO_CGROUP_MIXED; | |
1832 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
1833 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
1834 | ||
1835 | /* Mount tmpfs */ | |
3f69fb12 | 1836 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); |
6812d833 | 1837 | ret = safe_mount(NULL, tmpfspath, "tmpfs", |
3f69fb12 SY |
1838 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, |
1839 | "size=10240k,mode=755", root); | |
1840 | if (ret < 0) | |
1841 | goto on_error; | |
8aa1044f | 1842 | |
2202afc9 | 1843 | for (i = 0; ops->hierarchies[i]; i++) { |
8aa1044f | 1844 | char *controllerpath, *path2; |
2202afc9 | 1845 | struct hierarchy *h = ops->hierarchies[i]; |
8aa1044f | 1846 | char *controller = strrchr(h->mountpoint, '/'); |
8aa1044f SH |
1847 | |
1848 | if (!controller) | |
1849 | continue; | |
1850 | controller++; | |
affd10fa | 1851 | |
8aa1044f SH |
1852 | controllerpath = must_make_path(tmpfspath, controller, NULL); |
1853 | if (dir_exists(controllerpath)) { | |
1854 | free(controllerpath); | |
1855 | continue; | |
1856 | } | |
affd10fa | 1857 | |
3f69fb12 SY |
1858 | ret = mkdir(controllerpath, 0755); |
1859 | if (ret < 0) { | |
8aa1044f SH |
1860 | SYSERROR("Error creating cgroup path: %s", controllerpath); |
1861 | free(controllerpath); | |
3f69fb12 | 1862 | goto on_error; |
8aa1044f | 1863 | } |
b635e92d | 1864 | |
3f69fb12 | 1865 | if (has_cgns && wants_force_mount) { |
b635e92d CB |
1866 | /* If cgroup namespaces are supported but the container |
1867 | * will not have CAP_SYS_ADMIN after it has started we | |
1868 | * need to mount the cgroups manually. | |
1869 | */ | |
3f69fb12 | 1870 | ret = cg_mount_in_cgroup_namespace(type, h, controllerpath); |
b635e92d | 1871 | free(controllerpath); |
3f69fb12 SY |
1872 | if (ret < 0) |
1873 | goto on_error; | |
1874 | ||
b635e92d CB |
1875 | continue; |
1876 | } | |
1877 | ||
6812d833 | 1878 | ret = cg_mount_cgroup_full(type, h, controllerpath); |
3f69fb12 | 1879 | if (ret < 0) { |
8aa1044f | 1880 | free(controllerpath); |
3f69fb12 | 1881 | goto on_error; |
8aa1044f | 1882 | } |
3f69fb12 | 1883 | |
8aa1044f SH |
1884 | if (!cg_mount_needs_subdirs(type)) { |
1885 | free(controllerpath); | |
1886 | continue; | |
1887 | } | |
3f69fb12 | 1888 | |
bb221ad1 | 1889 | path2 = must_make_path(controllerpath, h->container_base_path, |
2202afc9 | 1890 | ops->container_cgroup, NULL); |
3f69fb12 SY |
1891 | ret = mkdir_p(path2, 0755); |
1892 | if (ret < 0) { | |
8aa1044f | 1893 | free(controllerpath); |
8e0c6620 | 1894 | free(path2); |
3f69fb12 | 1895 | goto on_error; |
8aa1044f | 1896 | } |
2f62fb00 | 1897 | |
6812d833 | 1898 | ret = cg_legacy_mount_controllers(type, h, controllerpath, |
2202afc9 | 1899 | path2, ops->container_cgroup); |
8aa1044f SH |
1900 | free(controllerpath); |
1901 | free(path2); | |
3f69fb12 SY |
1902 | if (ret < 0) |
1903 | goto on_error; | |
8aa1044f SH |
1904 | } |
1905 | retval = true; | |
1906 | ||
3f69fb12 | 1907 | on_error: |
8aa1044f SH |
1908 | free(tmpfspath); |
1909 | return retval; | |
ccb4cabe SH |
1910 | } |
1911 | ||
1912 | static int recursive_count_nrtasks(char *dirname) | |
1913 | { | |
74f96976 | 1914 | struct dirent *direntp; |
ccb4cabe SH |
1915 | DIR *dir; |
1916 | int count = 0, ret; | |
1917 | char *path; | |
1918 | ||
1919 | dir = opendir(dirname); | |
1920 | if (!dir) | |
1921 | return 0; | |
1922 | ||
74f96976 | 1923 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
1924 | struct stat mystat; |
1925 | ||
ccb4cabe SH |
1926 | if (!strcmp(direntp->d_name, ".") || |
1927 | !strcmp(direntp->d_name, "..")) | |
1928 | continue; | |
1929 | ||
1930 | path = must_make_path(dirname, direntp->d_name, NULL); | |
1931 | ||
1932 | if (lstat(path, &mystat)) | |
1933 | goto next; | |
1934 | ||
1935 | if (!S_ISDIR(mystat.st_mode)) | |
1936 | goto next; | |
1937 | ||
1938 | count += recursive_count_nrtasks(path); | |
13c49955 | 1939 | next: |
ccb4cabe SH |
1940 | free(path); |
1941 | } | |
1942 | ||
1943 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
1944 | ret = lxc_count_file_lines(path); | |
1945 | if (ret != -1) | |
1946 | count += ret; | |
1947 | free(path); | |
1948 | ||
13c49955 | 1949 | (void)closedir(dir); |
ccb4cabe SH |
1950 | |
1951 | return count; | |
1952 | } | |
1953 | ||
b857f4be | 1954 | __cgfsng_ops static int cgfsng_nrtasks(struct cgroup_ops *ops) |
3135c5d4 | 1955 | { |
ccb4cabe | 1956 | int count; |
3135c5d4 | 1957 | char *path; |
ccb4cabe | 1958 | |
2202afc9 | 1959 | if (!ops->container_cgroup || !ops->hierarchies) |
ccb4cabe | 1960 | return -1; |
a3926f6a | 1961 | |
eb697136 | 1962 | path = must_make_path(ops->hierarchies[0]->container_full_path, NULL); |
ccb4cabe SH |
1963 | count = recursive_count_nrtasks(path); |
1964 | free(path); | |
1965 | return count; | |
1966 | } | |
1967 | ||
11c23867 | 1968 | /* Only root needs to escape to the cgroup of its init. */ |
b857f4be | 1969 | __cgfsng_ops static bool cgfsng_escape(const struct cgroup_ops *ops, |
fb55e009 | 1970 | struct lxc_conf *conf) |
ccb4cabe | 1971 | { |
ccb4cabe SH |
1972 | int i; |
1973 | ||
9caee129 | 1974 | if (conf->cgroup_meta.relative || geteuid()) |
ccb4cabe SH |
1975 | return true; |
1976 | ||
2202afc9 | 1977 | for (i = 0; ops->hierarchies[i]; i++) { |
11c23867 CB |
1978 | int ret; |
1979 | char *fullpath; | |
1980 | ||
2202afc9 | 1981 | fullpath = must_make_path(ops->hierarchies[i]->mountpoint, |
bb221ad1 | 1982 | ops->hierarchies[i]->container_base_path, |
11c23867 | 1983 | "cgroup.procs", NULL); |
7cea5905 | 1984 | ret = lxc_write_to_file(fullpath, "0", 2, false, 0666); |
11c23867 CB |
1985 | if (ret != 0) { |
1986 | SYSERROR("Failed to escape to cgroup \"%s\"", fullpath); | |
ccb4cabe | 1987 | free(fullpath); |
6df334d1 | 1988 | return false; |
ccb4cabe SH |
1989 | } |
1990 | free(fullpath); | |
1991 | } | |
1992 | ||
6df334d1 | 1993 | return true; |
ccb4cabe SH |
1994 | } |
1995 | ||
b857f4be | 1996 | __cgfsng_ops static int cgfsng_num_hierarchies(struct cgroup_ops *ops) |
36662416 TA |
1997 | { |
1998 | int i; | |
1999 | ||
2202afc9 | 2000 | for (i = 0; ops->hierarchies[i]; i++) |
36662416 TA |
2001 | ; |
2002 | ||
2003 | return i; | |
2004 | } | |
2005 | ||
b857f4be | 2006 | __cgfsng_ops static bool cgfsng_get_hierarchies(struct cgroup_ops *ops, int n, char ***out) |
36662416 TA |
2007 | { |
2008 | int i; | |
2009 | ||
2010 | /* sanity check n */ | |
6b38e644 | 2011 | for (i = 0; i < n; i++) |
2202afc9 | 2012 | if (!ops->hierarchies[i]) |
36662416 | 2013 | return false; |
36662416 | 2014 | |
2202afc9 | 2015 | *out = ops->hierarchies[i]->controllers; |
36662416 TA |
2016 | |
2017 | return true; | |
2018 | } | |
2019 | ||
ccb4cabe SH |
2020 | #define THAWED "THAWED" |
2021 | #define THAWED_LEN (strlen(THAWED)) | |
2022 | ||
d6337a5f CB |
2023 | /* TODO: If the unified cgroup hierarchy grows a freezer controller this needs |
2024 | * to be adapted. | |
2025 | */ | |
b857f4be | 2026 | __cgfsng_ops static bool cgfsng_unfreeze(struct cgroup_ops *ops) |
ccb4cabe | 2027 | { |
d6337a5f | 2028 | int ret; |
ccb4cabe | 2029 | char *fullpath; |
d6337a5f | 2030 | struct hierarchy *h; |
ccb4cabe | 2031 | |
2202afc9 | 2032 | h = get_hierarchy(ops, "freezer"); |
457ca9aa | 2033 | if (!h) |
ccb4cabe | 2034 | return false; |
d6337a5f | 2035 | |
eb697136 | 2036 | fullpath = must_make_path(h->container_full_path, "freezer.state", NULL); |
7cea5905 | 2037 | ret = lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false, 0666); |
ccb4cabe | 2038 | free(fullpath); |
d6337a5f CB |
2039 | if (ret < 0) |
2040 | return false; | |
2041 | ||
ccb4cabe SH |
2042 | return true; |
2043 | } | |
2044 | ||
b857f4be | 2045 | __cgfsng_ops static const char *cgfsng_get_cgroup(struct cgroup_ops *ops, |
fb55e009 | 2046 | const char *controller) |
ccb4cabe | 2047 | { |
d6337a5f CB |
2048 | struct hierarchy *h; |
2049 | ||
2202afc9 | 2050 | h = get_hierarchy(ops, controller); |
106f1f38 | 2051 | if (!h) { |
2202afc9 CB |
2052 | WARN("Failed to find hierarchy for controller \"%s\"", |
2053 | controller ? controller : "(null)"); | |
ccb4cabe | 2054 | return NULL; |
106f1f38 | 2055 | } |
ccb4cabe | 2056 | |
eb697136 | 2057 | return h->container_full_path ? h->container_full_path + strlen(h->mountpoint) : NULL; |
371f834d SH |
2058 | } |
2059 | ||
c40c8209 CB |
2060 | /* Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a full path, |
2061 | * which must be freed by the caller. | |
371f834d | 2062 | */ |
c40c8209 CB |
2063 | static inline char *build_full_cgpath_from_monitorpath(struct hierarchy *h, |
2064 | const char *inpath, | |
2065 | const char *filename) | |
371f834d | 2066 | { |
371f834d | 2067 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
2068 | } |
2069 | ||
25f66a8f CB |
2070 | /* Technically, we're always at a delegation boundary here (This is especially |
2071 | * true when cgroup namespaces are available.). The reasoning is that in order | |
c2aed66d | 2072 | * for us to have been able to start a container in the first place the root |
25f66a8f | 2073 | * cgroup must have been a leaf node. Now, either the container's init system |
c2aed66d CB |
2074 | * has populated the cgroup and kept it as a leaf node or it has created |
2075 | * subtrees. In the former case we will simply attach to the leaf node we | |
2076 | * created when we started the container in the latter case we create our own | |
2077 | * cgroup for the attaching process. | |
2078 | */ | |
a3926f6a CB |
2079 | static int __cg_unified_attach(const struct hierarchy *h, const char *name, |
2080 | const char *lxcpath, const char *pidstr, | |
2081 | size_t pidstr_len, const char *controller) | |
c2aed66d CB |
2082 | { |
2083 | int ret; | |
2084 | size_t len; | |
2085 | int fret = -1, idx = 0; | |
2086 | char *base_path = NULL, *container_cgroup = NULL, *full_path = NULL; | |
2087 | ||
2088 | container_cgroup = lxc_cmd_get_cgroup_path(name, lxcpath, controller); | |
2089 | /* not running */ | |
2090 | if (!container_cgroup) | |
2091 | return 0; | |
2092 | ||
2093 | base_path = must_make_path(h->mountpoint, container_cgroup, NULL); | |
2094 | full_path = must_make_path(base_path, "cgroup.procs", NULL); | |
2095 | /* cgroup is populated */ | |
7cea5905 | 2096 | ret = lxc_write_to_file(full_path, pidstr, pidstr_len, false, 0666); |
c2aed66d CB |
2097 | if (ret < 0 && errno != EBUSY) |
2098 | goto on_error; | |
2099 | ||
2100 | if (ret == 0) | |
2101 | goto on_success; | |
2102 | ||
2103 | free(full_path); | |
2104 | ||
6333c915 CB |
2105 | len = strlen(base_path) + STRLITERALLEN("/lxc-1000") + |
2106 | STRLITERALLEN("/cgroup-procs"); | |
f25a2044 | 2107 | full_path = must_realloc(NULL, len + 1); |
c2aed66d CB |
2108 | do { |
2109 | if (idx) | |
2110 | ret = snprintf(full_path, len + 1, "%s/lxc-%d", | |
2111 | base_path, idx); | |
2112 | else | |
2113 | ret = snprintf(full_path, len + 1, "%s/lxc", base_path); | |
2114 | if (ret < 0 || (size_t)ret >= len + 1) | |
2115 | goto on_error; | |
2116 | ||
2117 | ret = mkdir_p(full_path, 0755); | |
2118 | if (ret < 0 && errno != EEXIST) | |
2119 | goto on_error; | |
2120 | ||
3ebe2fbd | 2121 | (void)strlcat(full_path, "/cgroup.procs", len + 1); |
7cea5905 | 2122 | ret = lxc_write_to_file(full_path, pidstr, len, false, 0666); |
c2aed66d CB |
2123 | if (ret == 0) |
2124 | goto on_success; | |
2125 | ||
2126 | /* this is a non-leaf node */ | |
2127 | if (errno != EBUSY) | |
2128 | goto on_error; | |
2129 | ||
edae86e9 CB |
2130 | idx++; |
2131 | } while (idx < 1000); | |
c2aed66d CB |
2132 | |
2133 | on_success: | |
2134 | if (idx < 1000) | |
2135 | fret = 0; | |
2136 | ||
2137 | on_error: | |
2138 | free(base_path); | |
2139 | free(container_cgroup); | |
2140 | free(full_path); | |
2141 | ||
2142 | return fret; | |
2143 | } | |
2144 | ||
b857f4be | 2145 | __cgfsng_ops static bool cgfsng_attach(struct cgroup_ops *ops, const char *name, |
fb55e009 | 2146 | const char *lxcpath, pid_t pid) |
ccb4cabe | 2147 | { |
c2aed66d | 2148 | int i, len, ret; |
a3650c0c | 2149 | char pidstr[INTTYPE_TO_STRLEN(pid_t)]; |
ccb4cabe | 2150 | |
a3650c0c CB |
2151 | len = snprintf(pidstr, sizeof(pidstr), "%d", pid); |
2152 | if (len < 0 || (size_t)len >= sizeof(pidstr)) | |
ccb4cabe SH |
2153 | return false; |
2154 | ||
2202afc9 | 2155 | for (i = 0; ops->hierarchies[i]; i++) { |
c2aed66d CB |
2156 | char *path; |
2157 | char *fullpath = NULL; | |
2202afc9 | 2158 | struct hierarchy *h = ops->hierarchies[i]; |
ccb4cabe | 2159 | |
c2aed66d | 2160 | if (h->version == CGROUP2_SUPER_MAGIC) { |
a3926f6a CB |
2161 | ret = __cg_unified_attach(h, name, lxcpath, pidstr, len, |
2162 | h->controllers[0]); | |
c2aed66d CB |
2163 | if (ret < 0) |
2164 | return false; | |
2165 | ||
2166 | continue; | |
2167 | } | |
2168 | ||
ccb4cabe | 2169 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); |
c2aed66d CB |
2170 | /* not running */ |
2171 | if (!path) | |
ccb4cabe SH |
2172 | continue; |
2173 | ||
371f834d | 2174 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
71cb9afb | 2175 | free(path); |
7cea5905 | 2176 | ret = lxc_write_to_file(fullpath, pidstr, len, false, 0666); |
c2aed66d | 2177 | if (ret < 0) { |
ccb4cabe SH |
2178 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); |
2179 | free(fullpath); | |
ccb4cabe SH |
2180 | return false; |
2181 | } | |
ccb4cabe SH |
2182 | free(fullpath); |
2183 | } | |
2184 | ||
ccb4cabe SH |
2185 | return true; |
2186 | } | |
2187 | ||
e2bd2b13 CB |
2188 | /* Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. Here we |
2189 | * don't have a cgroup_data set up, so we ask the running container through the | |
2190 | * commands API for the cgroup path. | |
ccb4cabe | 2191 | */ |
b857f4be | 2192 | __cgfsng_ops static int cgfsng_get(struct cgroup_ops *ops, const char *filename, |
fb55e009 CB |
2193 | char *value, size_t len, const char *name, |
2194 | const char *lxcpath) | |
ccb4cabe | 2195 | { |
ccb4cabe | 2196 | int ret = -1; |
0069cc61 CB |
2197 | size_t controller_len; |
2198 | char *controller, *p, *path; | |
2199 | struct hierarchy *h; | |
ccb4cabe | 2200 | |
0069cc61 CB |
2201 | controller_len = strlen(filename); |
2202 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
2203 | (void)strlcpy(controller, filename, controller_len + 1); |
2204 | ||
0069cc61 CB |
2205 | p = strchr(controller, '.'); |
2206 | if (p) | |
ccb4cabe SH |
2207 | *p = '\0'; |
2208 | ||
0069cc61 CB |
2209 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2210 | /* not running */ | |
2211 | if (!path) | |
ccb4cabe SH |
2212 | return -1; |
2213 | ||
2202afc9 | 2214 | h = get_hierarchy(ops, controller); |
ccb4cabe | 2215 | if (h) { |
0069cc61 CB |
2216 | char *fullpath; |
2217 | ||
2218 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
ccb4cabe SH |
2219 | ret = lxc_read_from_file(fullpath, value, len); |
2220 | free(fullpath); | |
2221 | } | |
ccb4cabe SH |
2222 | free(path); |
2223 | ||
2224 | return ret; | |
2225 | } | |
2226 | ||
eec533e3 CB |
2227 | /* Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. Here we |
2228 | * don't have a cgroup_data set up, so we ask the running container through the | |
2229 | * commands API for the cgroup path. | |
ccb4cabe | 2230 | */ |
b857f4be | 2231 | __cgfsng_ops static int cgfsng_set(struct cgroup_ops *ops, |
fb55e009 CB |
2232 | const char *filename, const char *value, |
2233 | const char *name, const char *lxcpath) | |
ccb4cabe | 2234 | { |
ccb4cabe | 2235 | int ret = -1; |
87777968 CB |
2236 | size_t controller_len; |
2237 | char *controller, *p, *path; | |
2238 | struct hierarchy *h; | |
ccb4cabe | 2239 | |
87777968 CB |
2240 | controller_len = strlen(filename); |
2241 | controller = alloca(controller_len + 1); | |
64e82f8b DJ |
2242 | (void)strlcpy(controller, filename, controller_len + 1); |
2243 | ||
87777968 CB |
2244 | p = strchr(controller, '.'); |
2245 | if (p) | |
ccb4cabe SH |
2246 | *p = '\0'; |
2247 | ||
87777968 CB |
2248 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2249 | /* not running */ | |
2250 | if (!path) | |
ccb4cabe SH |
2251 | return -1; |
2252 | ||
2202afc9 | 2253 | h = get_hierarchy(ops, controller); |
ccb4cabe | 2254 | if (h) { |
87777968 CB |
2255 | char *fullpath; |
2256 | ||
2257 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
7cea5905 | 2258 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
ccb4cabe SH |
2259 | free(fullpath); |
2260 | } | |
ccb4cabe SH |
2261 | free(path); |
2262 | ||
2263 | return ret; | |
2264 | } | |
2265 | ||
91d1a13a | 2266 | /* take devices cgroup line |
72add155 SH |
2267 | * /dev/foo rwx |
2268 | * and convert it to a valid | |
2269 | * type major:minor mode | |
91d1a13a CB |
2270 | * line. Return <0 on error. Dest is a preallocated buffer long enough to hold |
2271 | * the output. | |
72add155 SH |
2272 | */ |
2273 | static int convert_devpath(const char *invalue, char *dest) | |
2274 | { | |
2a06d041 CB |
2275 | int n_parts; |
2276 | char *p, *path, type; | |
72add155 | 2277 | unsigned long minor, major; |
91d1a13a | 2278 | struct stat sb; |
2a06d041 CB |
2279 | int ret = -EINVAL; |
2280 | char *mode = NULL; | |
72add155 SH |
2281 | |
2282 | path = must_copy_string(invalue); | |
2283 | ||
91d1a13a CB |
2284 | /* Read path followed by mode. Ignore any trailing text. |
2285 | * A ' # comment' would be legal. Technically other text is not | |
2286 | * legal, we could check for that if we cared to. | |
72add155 SH |
2287 | */ |
2288 | for (n_parts = 1, p = path; *p && n_parts < 3; p++) { | |
2c2d6c49 SH |
2289 | if (*p != ' ') |
2290 | continue; | |
2291 | *p = '\0'; | |
91d1a13a | 2292 | |
2c2d6c49 SH |
2293 | if (n_parts != 1) |
2294 | break; | |
2295 | p++; | |
2296 | n_parts++; | |
91d1a13a | 2297 | |
2c2d6c49 SH |
2298 | while (*p == ' ') |
2299 | p++; | |
91d1a13a | 2300 | |
2c2d6c49 | 2301 | mode = p; |
91d1a13a | 2302 | |
2c2d6c49 SH |
2303 | if (*p == '\0') |
2304 | goto out; | |
72add155 | 2305 | } |
2c2d6c49 SH |
2306 | |
2307 | if (n_parts == 1) | |
72add155 | 2308 | goto out; |
72add155 SH |
2309 | |
2310 | ret = stat(path, &sb); | |
2311 | if (ret < 0) | |
2312 | goto out; | |
2313 | ||
72add155 SH |
2314 | mode_t m = sb.st_mode & S_IFMT; |
2315 | switch (m) { | |
2316 | case S_IFBLK: | |
2317 | type = 'b'; | |
2318 | break; | |
2319 | case S_IFCHR: | |
2320 | type = 'c'; | |
2321 | break; | |
2c2d6c49 | 2322 | default: |
91d1a13a | 2323 | ERROR("Unsupported device type %i for \"%s\"", m, path); |
72add155 SH |
2324 | ret = -EINVAL; |
2325 | goto out; | |
2326 | } | |
2c2d6c49 SH |
2327 | |
2328 | major = MAJOR(sb.st_rdev); | |
2329 | minor = MINOR(sb.st_rdev); | |
2330 | ret = snprintf(dest, 50, "%c %lu:%lu %s", type, major, minor, mode); | |
72add155 | 2331 | if (ret < 0 || ret >= 50) { |
2a06d041 CB |
2332 | ERROR("Error on configuration value \"%c %lu:%lu %s\" (max 50 " |
2333 | "chars)", type, major, minor, mode); | |
72add155 SH |
2334 | ret = -ENAMETOOLONG; |
2335 | goto out; | |
2336 | } | |
2337 | ret = 0; | |
2338 | ||
2339 | out: | |
2340 | free(path); | |
2341 | return ret; | |
2342 | } | |
2343 | ||
90e97284 CB |
2344 | /* Called from setup_limits - here we have the container's cgroup_data because |
2345 | * we created the cgroups. | |
ccb4cabe | 2346 | */ |
2202afc9 CB |
2347 | static int cg_legacy_set_data(struct cgroup_ops *ops, const char *filename, |
2348 | const char *value) | |
ccb4cabe | 2349 | { |
ab1a6cac | 2350 | size_t len; |
90e97284 | 2351 | char *fullpath, *p; |
1a0e70ac CB |
2352 | /* "b|c <2^64-1>:<2^64-1> r|w|m" = 47 chars max */ |
2353 | char converted_value[50]; | |
b3646d7e CB |
2354 | struct hierarchy *h; |
2355 | int ret = 0; | |
2356 | char *controller = NULL; | |
ccb4cabe | 2357 | |
ab1a6cac CB |
2358 | len = strlen(filename); |
2359 | controller = alloca(len + 1); | |
64e82f8b DJ |
2360 | (void)strlcpy(controller, filename, len + 1); |
2361 | ||
ab1a6cac CB |
2362 | p = strchr(controller, '.'); |
2363 | if (p) | |
ccb4cabe SH |
2364 | *p = '\0'; |
2365 | ||
c8bf519d | 2366 | if (strcmp("devices.allow", filename) == 0 && value[0] == '/') { |
72add155 SH |
2367 | ret = convert_devpath(value, converted_value); |
2368 | if (ret < 0) | |
c8bf519d | 2369 | return ret; |
72add155 | 2370 | value = converted_value; |
c8bf519d | 2371 | } |
2372 | ||
2202afc9 | 2373 | h = get_hierarchy(ops, controller); |
b3646d7e CB |
2374 | if (!h) { |
2375 | ERROR("Failed to setup limits for the \"%s\" controller. " | |
2376 | "The controller seems to be unused by \"cgfsng\" cgroup " | |
2377 | "driver or not enabled on the cgroup hierarchy", | |
2378 | controller); | |
d1953b26 | 2379 | errno = ENOENT; |
ab1a6cac | 2380 | return -ENOENT; |
ccb4cabe | 2381 | } |
b3646d7e | 2382 | |
eb697136 | 2383 | fullpath = must_make_path(h->container_full_path, filename, NULL); |
7cea5905 | 2384 | ret = lxc_write_to_file(fullpath, value, strlen(value), false, 0666); |
b3646d7e | 2385 | free(fullpath); |
ccb4cabe SH |
2386 | return ret; |
2387 | } | |
2388 | ||
2202afc9 | 2389 | static bool __cg_legacy_setup_limits(struct cgroup_ops *ops, |
a3926f6a CB |
2390 | struct lxc_list *cgroup_settings, |
2391 | bool do_devices) | |
ccb4cabe | 2392 | { |
c347df58 | 2393 | struct lxc_list *iterator, *next, *sorted_cgroup_settings; |
ccb4cabe | 2394 | struct lxc_cgroup *cg; |
ccb4cabe SH |
2395 | bool ret = false; |
2396 | ||
2397 | if (lxc_list_empty(cgroup_settings)) | |
2398 | return true; | |
2399 | ||
2400 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
6b38e644 | 2401 | if (!sorted_cgroup_settings) |
ccb4cabe | 2402 | return false; |
ccb4cabe | 2403 | |
ccb4cabe SH |
2404 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
2405 | cg = iterator->elem; | |
2406 | ||
2407 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
2202afc9 | 2408 | if (cg_legacy_set_data(ops, cg->subsystem, cg->value)) { |
ccb4cabe | 2409 | if (do_devices && (errno == EACCES || errno == EPERM)) { |
c347df58 CB |
2410 | WARN("Failed to set \"%s\" to \"%s\"", |
2411 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2412 | continue; |
2413 | } | |
c347df58 CB |
2414 | WARN("Failed to set \"%s\" to \"%s\"", |
2415 | cg->subsystem, cg->value); | |
ccb4cabe SH |
2416 | goto out; |
2417 | } | |
c347df58 CB |
2418 | DEBUG("Set controller \"%s\" set to \"%s\"", |
2419 | cg->subsystem, cg->value); | |
ccb4cabe | 2420 | } |
ccb4cabe SH |
2421 | } |
2422 | ||
2423 | ret = true; | |
6b38e644 | 2424 | INFO("Limits for the legacy cgroup hierarchies have been setup"); |
ccb4cabe | 2425 | out: |
ccb4cabe SH |
2426 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
2427 | lxc_list_del(iterator); | |
2428 | free(iterator); | |
2429 | } | |
2430 | free(sorted_cgroup_settings); | |
2431 | return ret; | |
2432 | } | |
2433 | ||
2202afc9 | 2434 | static bool __cg_unified_setup_limits(struct cgroup_ops *ops, |
a3926f6a | 2435 | struct lxc_list *cgroup_settings) |
6b38e644 CB |
2436 | { |
2437 | struct lxc_list *iterator; | |
2202afc9 | 2438 | struct hierarchy *h = ops->unified; |
6b38e644 CB |
2439 | |
2440 | if (lxc_list_empty(cgroup_settings)) | |
2441 | return true; | |
2442 | ||
2443 | if (!h) | |
2444 | return false; | |
2445 | ||
2446 | lxc_list_for_each(iterator, cgroup_settings) { | |
2447 | int ret; | |
2448 | char *fullpath; | |
2449 | struct lxc_cgroup *cg = iterator->elem; | |
2450 | ||
eb697136 | 2451 | fullpath = must_make_path(h->container_full_path, cg->subsystem, NULL); |
7cea5905 | 2452 | ret = lxc_write_to_file(fullpath, cg->value, strlen(cg->value), false, 0666); |
6b38e644 CB |
2453 | free(fullpath); |
2454 | if (ret < 0) { | |
b2ac2cb7 CB |
2455 | SYSERROR("Failed to set \"%s\" to \"%s\"", |
2456 | cg->subsystem, cg->value); | |
6b38e644 CB |
2457 | return false; |
2458 | } | |
2459 | TRACE("Set \"%s\" to \"%s\"", cg->subsystem, cg->value); | |
2460 | } | |
2461 | ||
2462 | INFO("Limits for the unified cgroup hierarchy have been setup"); | |
2463 | return true; | |
2464 | } | |
2465 | ||
b857f4be | 2466 | __cgfsng_ops static bool cgfsng_setup_limits(struct cgroup_ops *ops, |
fb55e009 CB |
2467 | struct lxc_conf *conf, |
2468 | bool do_devices) | |
6b38e644 CB |
2469 | { |
2470 | bool bret; | |
2471 | ||
2202afc9 | 2472 | bret = __cg_legacy_setup_limits(ops, &conf->cgroup, do_devices); |
6b38e644 CB |
2473 | if (!bret) |
2474 | return false; | |
2475 | ||
2202afc9 CB |
2476 | return __cg_unified_setup_limits(ops, &conf->cgroup2); |
2477 | } | |
2478 | ||
b7b18fc5 CB |
2479 | static bool cgroup_use_wants_controllers(const struct cgroup_ops *ops, |
2480 | char **controllers) | |
2481 | { | |
2482 | char **cur_ctrl, **cur_use; | |
2483 | ||
2484 | if (!ops->cgroup_use) | |
2485 | return true; | |
2486 | ||
2487 | for (cur_ctrl = controllers; cur_ctrl && *cur_ctrl; cur_ctrl++) { | |
2488 | bool found = false; | |
2489 | ||
2490 | for (cur_use = ops->cgroup_use; cur_use && *cur_use; cur_use++) { | |
2491 | if (strcmp(*cur_use, *cur_ctrl) != 0) | |
2492 | continue; | |
2493 | ||
2494 | found = true; | |
2495 | break; | |
2496 | } | |
2497 | ||
2498 | if (found) | |
2499 | continue; | |
2500 | ||
2501 | return false; | |
2502 | } | |
2503 | ||
2504 | return true; | |
2505 | } | |
2506 | ||
2202afc9 CB |
2507 | /* At startup, parse_hierarchies finds all the info we need about cgroup |
2508 | * mountpoints and current cgroups, and stores it in @d. | |
2509 | */ | |
9caee129 | 2510 | static bool cg_hybrid_init(struct cgroup_ops *ops, bool relative) |
2202afc9 CB |
2511 | { |
2512 | int ret; | |
2513 | char *basecginfo; | |
2202afc9 CB |
2514 | FILE *f; |
2515 | size_t len = 0; | |
2516 | char *line = NULL; | |
2517 | char **klist = NULL, **nlist = NULL; | |
2518 | ||
2519 | /* Root spawned containers escape the current cgroup, so use init's | |
2520 | * cgroups as our base in that case. | |
2521 | */ | |
9caee129 | 2522 | if (!relative && (geteuid() == 0)) |
2202afc9 CB |
2523 | basecginfo = read_file("/proc/1/cgroup"); |
2524 | else | |
2525 | basecginfo = read_file("/proc/self/cgroup"); | |
2526 | if (!basecginfo) | |
2527 | return false; | |
2528 | ||
2529 | ret = get_existing_subsystems(&klist, &nlist); | |
2530 | if (ret < 0) { | |
2531 | ERROR("Failed to retrieve available legacy cgroup controllers"); | |
2532 | free(basecginfo); | |
2533 | return false; | |
2534 | } | |
2535 | ||
2536 | f = fopen("/proc/self/mountinfo", "r"); | |
2537 | if (!f) { | |
2538 | ERROR("Failed to open \"/proc/self/mountinfo\""); | |
2539 | free(basecginfo); | |
2540 | return false; | |
2541 | } | |
2542 | ||
2543 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
2544 | ||
2545 | while (getline(&line, &len, f) != -1) { | |
2546 | int type; | |
2547 | bool writeable; | |
2548 | struct hierarchy *new; | |
2549 | char *base_cgroup = NULL, *mountpoint = NULL; | |
2550 | char **controller_list = NULL; | |
2551 | ||
2552 | type = get_cgroup_version(line); | |
2553 | if (type == 0) | |
2554 | continue; | |
2555 | ||
2556 | if (type == CGROUP2_SUPER_MAGIC && ops->unified) | |
2557 | continue; | |
2558 | ||
2559 | if (ops->cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { | |
2560 | if (type == CGROUP2_SUPER_MAGIC) | |
2561 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2562 | else if (type == CGROUP_SUPER_MAGIC) | |
2563 | ops->cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
2564 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
2565 | if (type == CGROUP_SUPER_MAGIC) | |
2566 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2567 | } else if (ops->cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
2568 | if (type == CGROUP2_SUPER_MAGIC) | |
2569 | ops->cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
2570 | } | |
2571 | ||
2572 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, type); | |
2573 | if (!controller_list && type == CGROUP_SUPER_MAGIC) | |
2574 | continue; | |
2575 | ||
2576 | if (type == CGROUP_SUPER_MAGIC) | |
2577 | if (controller_list_is_dup(ops->hierarchies, controller_list)) | |
2578 | goto next; | |
2579 | ||
2580 | mountpoint = cg_hybrid_get_mountpoint(line); | |
2581 | if (!mountpoint) { | |
2582 | ERROR("Failed parsing mountpoint from \"%s\"", line); | |
2583 | goto next; | |
2584 | } | |
2585 | ||
2586 | if (type == CGROUP_SUPER_MAGIC) | |
2587 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); | |
2588 | else | |
2589 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); | |
2590 | if (!base_cgroup) { | |
2591 | ERROR("Failed to find current cgroup"); | |
2592 | goto next; | |
2593 | } | |
2594 | ||
2595 | trim(base_cgroup); | |
2596 | prune_init_scope(base_cgroup); | |
2597 | if (type == CGROUP2_SUPER_MAGIC) | |
2598 | writeable = test_writeable_v2(mountpoint, base_cgroup); | |
2599 | else | |
2600 | writeable = test_writeable_v1(mountpoint, base_cgroup); | |
2601 | if (!writeable) | |
2602 | goto next; | |
2603 | ||
2604 | if (type == CGROUP2_SUPER_MAGIC) { | |
2605 | char *cgv2_ctrl_path; | |
2606 | ||
2607 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
2608 | "cgroup.controllers", | |
2609 | NULL); | |
2610 | ||
2611 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
2612 | free(cgv2_ctrl_path); | |
2613 | if (!controller_list) { | |
2614 | controller_list = cg_unified_make_empty_controller(); | |
2615 | TRACE("No controllers are enabled for " | |
2616 | "delegation in the unified hierarchy"); | |
2617 | } | |
2618 | } | |
2619 | ||
b7b18fc5 CB |
2620 | /* Exclude all controllers that cgroup use does not want. */ |
2621 | if (!cgroup_use_wants_controllers(ops, controller_list)) | |
2622 | goto next; | |
2623 | ||
2202afc9 CB |
2624 | new = add_hierarchy(&ops->hierarchies, controller_list, mountpoint, base_cgroup, type); |
2625 | if (type == CGROUP2_SUPER_MAGIC && !ops->unified) | |
2626 | ops->unified = new; | |
2627 | ||
2628 | continue; | |
2629 | ||
2630 | next: | |
2631 | free_string_list(controller_list); | |
2632 | free(mountpoint); | |
2633 | free(base_cgroup); | |
2634 | } | |
2635 | ||
2636 | free_string_list(klist); | |
2637 | free_string_list(nlist); | |
2638 | ||
2639 | free(basecginfo); | |
2640 | ||
2641 | fclose(f); | |
2642 | free(line); | |
2643 | ||
2644 | TRACE("Writable cgroup hierarchies:"); | |
2645 | lxc_cgfsng_print_hierarchies(ops); | |
2646 | ||
2647 | /* verify that all controllers in cgroup.use and all crucial | |
2648 | * controllers are accounted for | |
2649 | */ | |
2650 | if (!all_controllers_found(ops)) | |
2651 | return false; | |
2652 | ||
2653 | return true; | |
2654 | } | |
2655 | ||
2656 | static int cg_is_pure_unified(void) | |
2657 | { | |
2658 | ||
2659 | int ret; | |
2660 | struct statfs fs; | |
2661 | ||
2662 | ret = statfs("/sys/fs/cgroup", &fs); | |
2663 | if (ret < 0) | |
2664 | return -ENOMEDIUM; | |
2665 | ||
2666 | if (is_fs_type(&fs, CGROUP2_SUPER_MAGIC)) | |
2667 | return CGROUP2_SUPER_MAGIC; | |
2668 | ||
2669 | return 0; | |
2670 | } | |
2671 | ||
2672 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
9caee129 | 2673 | static char *cg_unified_get_current_cgroup(bool relative) |
2202afc9 CB |
2674 | { |
2675 | char *basecginfo, *base_cgroup; | |
2202afc9 CB |
2676 | char *copy = NULL; |
2677 | ||
9caee129 | 2678 | if (!relative && (geteuid() == 0)) |
2202afc9 CB |
2679 | basecginfo = read_file("/proc/1/cgroup"); |
2680 | else | |
2681 | basecginfo = read_file("/proc/self/cgroup"); | |
2682 | if (!basecginfo) | |
2683 | return NULL; | |
2684 | ||
2685 | base_cgroup = strstr(basecginfo, "0::/"); | |
2686 | if (!base_cgroup) | |
2687 | goto cleanup_on_err; | |
2688 | ||
2689 | base_cgroup = base_cgroup + 3; | |
2690 | copy = copy_to_eol(base_cgroup); | |
2691 | if (!copy) | |
2692 | goto cleanup_on_err; | |
2693 | ||
2694 | cleanup_on_err: | |
2695 | free(basecginfo); | |
2696 | if (copy) | |
2697 | trim(copy); | |
2698 | ||
2699 | return copy; | |
2700 | } | |
2701 | ||
9caee129 | 2702 | static int cg_unified_init(struct cgroup_ops *ops, bool relative) |
2202afc9 CB |
2703 | { |
2704 | int ret; | |
2705 | char *mountpoint, *subtree_path; | |
2706 | char **delegatable; | |
2707 | char *base_cgroup = NULL; | |
2708 | ||
2709 | ret = cg_is_pure_unified(); | |
2710 | if (ret == -ENOMEDIUM) | |
2711 | return -ENOMEDIUM; | |
2712 | ||
2713 | if (ret != CGROUP2_SUPER_MAGIC) | |
2714 | return 0; | |
2715 | ||
9caee129 | 2716 | base_cgroup = cg_unified_get_current_cgroup(relative); |
2202afc9 CB |
2717 | if (!base_cgroup) |
2718 | return -EINVAL; | |
2719 | prune_init_scope(base_cgroup); | |
2720 | ||
2721 | /* We assume that we have already been given controllers to delegate | |
2722 | * further down the hierarchy. If not it is up to the user to delegate | |
2723 | * them to us. | |
2724 | */ | |
2725 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
2726 | subtree_path = must_make_path(mountpoint, base_cgroup, | |
2727 | "cgroup.subtree_control", NULL); | |
2728 | delegatable = cg_unified_get_controllers(subtree_path); | |
2729 | free(subtree_path); | |
2730 | if (!delegatable) | |
2731 | delegatable = cg_unified_make_empty_controller(); | |
2732 | if (!delegatable[0]) | |
2733 | TRACE("No controllers are enabled for delegation"); | |
2734 | ||
2735 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
2736 | * we should verify here. The reason I'm not doing it right is that I'm | |
2737 | * not convinced that lxc.cgroup.use will be the future since it is a | |
2738 | * global property. I much rather have an option that lets you request | |
2739 | * controllers per container. | |
2740 | */ | |
2741 | ||
2742 | add_hierarchy(&ops->hierarchies, delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
2743 | ||
2744 | ops->cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
2745 | return CGROUP2_SUPER_MAGIC; | |
2746 | } | |
2747 | ||
5a087e05 | 2748 | static bool cg_init(struct cgroup_ops *ops, struct lxc_conf *conf) |
2202afc9 CB |
2749 | { |
2750 | int ret; | |
2751 | const char *tmp; | |
9caee129 | 2752 | bool relative = conf->cgroup_meta.relative; |
2202afc9 CB |
2753 | |
2754 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
b7b18fc5 CB |
2755 | if (tmp) { |
2756 | char *chop, *cur, *pin; | |
b7b18fc5 CB |
2757 | |
2758 | pin = must_copy_string(tmp); | |
2759 | chop = pin; | |
2760 | ||
0be0d78f | 2761 | lxc_iterate_parts(cur, chop, ",") { |
b7b18fc5 | 2762 | must_append_string(&ops->cgroup_use, cur); |
0be0d78f | 2763 | } |
b7b18fc5 CB |
2764 | |
2765 | free(pin); | |
2766 | } | |
2202afc9 | 2767 | |
9caee129 | 2768 | ret = cg_unified_init(ops, relative); |
2202afc9 CB |
2769 | if (ret < 0) |
2770 | return false; | |
2771 | ||
2772 | if (ret == CGROUP2_SUPER_MAGIC) | |
2773 | return true; | |
2774 | ||
9caee129 | 2775 | return cg_hybrid_init(ops, relative); |
2202afc9 CB |
2776 | } |
2777 | ||
b857f4be | 2778 | __cgfsng_ops static bool cgfsng_data_init(struct cgroup_ops *ops) |
2202afc9 CB |
2779 | { |
2780 | const char *cgroup_pattern; | |
2781 | ||
2782 | /* copy system-wide cgroup information */ | |
2783 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); | |
2784 | if (!cgroup_pattern) { | |
2785 | /* lxc.cgroup.pattern is only NULL on error. */ | |
2786 | ERROR("Failed to retrieve cgroup pattern"); | |
2787 | return false; | |
2788 | } | |
2789 | ops->cgroup_pattern = must_copy_string(cgroup_pattern); | |
625ad37b | 2790 | ops->monitor_pattern = MONITOR_CGROUP; |
2202afc9 CB |
2791 | |
2792 | return true; | |
2793 | } | |
2794 | ||
5a087e05 | 2795 | struct cgroup_ops *cgfsng_ops_init(struct lxc_conf *conf) |
2202afc9 CB |
2796 | { |
2797 | struct cgroup_ops *cgfsng_ops; | |
2798 | ||
2799 | cgfsng_ops = malloc(sizeof(struct cgroup_ops)); | |
2800 | if (!cgfsng_ops) | |
2801 | return NULL; | |
2802 | ||
2803 | memset(cgfsng_ops, 0, sizeof(struct cgroup_ops)); | |
2804 | cgfsng_ops->cgroup_layout = CGROUP_LAYOUT_UNKNOWN; | |
2805 | ||
5a087e05 | 2806 | if (!cg_init(cgfsng_ops, conf)) { |
2202afc9 CB |
2807 | free(cgfsng_ops); |
2808 | return NULL; | |
2809 | } | |
2810 | ||
2811 | cgfsng_ops->data_init = cgfsng_data_init; | |
434c8e15 CB |
2812 | cgfsng_ops->payload_destroy = cgfsng_payload_destroy; |
2813 | cgfsng_ops->monitor_destroy = cgfsng_monitor_destroy; | |
72068e74 | 2814 | cgfsng_ops->monitor_create = cgfsng_monitor_create; |
eeef32bb | 2815 | cgfsng_ops->monitor_enter = cgfsng_monitor_enter; |
e8b181f5 CB |
2816 | cgfsng_ops->payload_create = cgfsng_payload_create; |
2817 | cgfsng_ops->payload_enter = cgfsng_payload_enter; | |
2202afc9 CB |
2818 | cgfsng_ops->escape = cgfsng_escape; |
2819 | cgfsng_ops->num_hierarchies = cgfsng_num_hierarchies; | |
2820 | cgfsng_ops->get_hierarchies = cgfsng_get_hierarchies; | |
2821 | cgfsng_ops->get_cgroup = cgfsng_get_cgroup; | |
2822 | cgfsng_ops->get = cgfsng_get; | |
2823 | cgfsng_ops->set = cgfsng_set; | |
2824 | cgfsng_ops->unfreeze = cgfsng_unfreeze; | |
2825 | cgfsng_ops->setup_limits = cgfsng_setup_limits; | |
2826 | cgfsng_ops->driver = "cgfsng"; | |
2827 | cgfsng_ops->version = "1.0.0"; | |
2828 | cgfsng_ops->attach = cgfsng_attach; | |
2829 | cgfsng_ops->chown = cgfsng_chown; | |
2830 | cgfsng_ops->mount = cgfsng_mount; | |
2831 | cgfsng_ops->nrtasks = cgfsng_nrtasks; | |
2832 | ||
2833 | return cgfsng_ops; | |
2834 | } |