]>
Commit | Line | Data |
---|---|---|
ccb4cabe SH |
1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * Copyright © 2016 Canonical Ltd. | |
5 | * | |
6 | * Authors: | |
7 | * Serge Hallyn <serge.hallyn@ubuntu.com> | |
3fd0de4d | 8 | * Christian Brauner <christian.brauner@ubuntu.com> |
ccb4cabe SH |
9 | * |
10 | * This library is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU Lesser General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2.1 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This library is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * Lesser General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU Lesser General Public | |
21 | * License along with this library; if not, write to the Free Software | |
22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
23 | */ | |
24 | ||
25 | /* | |
26 | * cgfs-ng.c: this is a new, simplified implementation of a filesystem | |
27 | * cgroup backend. The original cgfs.c was designed to be as flexible | |
28 | * as possible. It would try to find cgroup filesystems no matter where | |
29 | * or how you had them mounted, and deduce the most usable mount for | |
0e7ff52c | 30 | * each controller. |
ccb4cabe SH |
31 | * |
32 | * This new implementation assumes that cgroup filesystems are mounted | |
33 | * under /sys/fs/cgroup/clist where clist is either the controller, or | |
34 | * a comman-separated list of controllers. | |
35 | */ | |
a54694f8 | 36 | |
ccb4cabe | 37 | #include "config.h" |
a54694f8 CB |
38 | |
39 | #include <ctype.h> | |
40 | #include <dirent.h> | |
41 | #include <errno.h> | |
42 | #include <grp.h> | |
43 | #include <stdint.h> | |
ccb4cabe SH |
44 | #include <stdio.h> |
45 | #include <stdlib.h> | |
a54694f8 | 46 | #include <string.h> |
ccb4cabe | 47 | #include <unistd.h> |
c8bf519d | 48 | #include <linux/kdev_t.h> |
438c4581 CB |
49 | #include <linux/types.h> |
50 | #include <sys/types.h> | |
c8bf519d | 51 | |
b635e92d | 52 | #include "caps.h" |
ccb4cabe | 53 | #include "cgroup.h" |
6328fd9c | 54 | #include "cgroup_utils.h" |
ccb4cabe | 55 | #include "commands.h" |
43654d34 | 56 | #include "conf.h" |
a54694f8 | 57 | #include "log.h" |
43654d34 | 58 | #include "storage/storage.h" |
a54694f8 | 59 | #include "utils.h" |
ccb4cabe SH |
60 | |
61 | lxc_log_define(lxc_cgfsng, lxc); | |
62 | ||
63 | static struct cgroup_ops cgfsng_ops; | |
64 | ||
9e288301 | 65 | /* A descriptor for a mounted hierarchy |
16a2cde9 CB |
66 | * |
67 | * @controllers | |
68 | * - legacy hierarchy | |
9e288301 | 69 | * Either NULL, or a null-terminated list of all the co-mounted controllers. |
16a2cde9 | 70 | * - unified hierarchy |
9e288301 | 71 | * Either NULL, or a null-terminated list of all enabled controllers. |
16a2cde9 CB |
72 | * |
73 | * @mountpoint | |
9e288301 | 74 | * - The mountpoint we will use. |
16a2cde9 | 75 | * - legacy hierarchy |
9e288301 CB |
76 | * It will be either /sys/fs/cgroup/controller or |
77 | * /sys/fs/cgroup/controllerlist. | |
16a2cde9 | 78 | * - unified hierarchy |
9e288301 CB |
79 | * It will either be /sys/fs/cgroup or /sys/fs/cgroup/<mountpoint-name> |
80 | * depending on whether this is a hybrid cgroup layout (mix of legacy and | |
81 | * unified hierarchies) or a pure unified cgroup layout. | |
16a2cde9 CB |
82 | * |
83 | * @base_cgroup | |
9e288301 CB |
84 | * - The cgroup under which the container cgroup path |
85 | * is created. This will be either the caller's cgroup (if not root), or | |
86 | * init's cgroup (if root). | |
16a2cde9 CB |
87 | * |
88 | * @fullcgpath | |
9e288301 | 89 | * - The full path to the containers cgroup. |
16a2cde9 CB |
90 | * |
91 | * @version | |
92 | * - legacy hierarchy | |
9e288301 CB |
93 | * If the hierarchy is a legacy hierarchy this will be set to |
94 | * CGROUP_SUPER_MAGIC. | |
16a2cde9 | 95 | * - unified hierarchy |
9e288301 CB |
96 | * If the hierarchy is a legacy hierarchy this will be set to |
97 | * CGROUP2_SUPER_MAGIC. | |
ccb4cabe SH |
98 | */ |
99 | struct hierarchy { | |
100 | char **controllers; | |
101 | char *mountpoint; | |
102 | char *base_cgroup; | |
103 | char *fullcgpath; | |
d6337a5f | 104 | int version; |
ccb4cabe SH |
105 | }; |
106 | ||
16a2cde9 CB |
107 | /* The cgroup data which is attached to the lxc_handler. |
108 | * | |
109 | * @cgroup_pattern | |
110 | * - A copy of lxc.cgroup.pattern. | |
111 | * | |
112 | * @container_cgroup | |
113 | * - If not null, the cgroup which was created for the container. For each | |
114 | * hierarchy, it is created under the @hierarchy->base_cgroup directory. | |
115 | * Relative to the base_cgroup it is the same for all hierarchies. | |
116 | * | |
117 | * @name | |
118 | * - The name of the container. | |
119 | * | |
120 | * @cgroup_meta | |
121 | * - A copy of the container's cgroup information. This overrides | |
122 | * @cgroup_pattern. | |
123 | * | |
09f3bb13 CB |
124 | * @cgroup_layout |
125 | * - What cgroup layout the container is running with. | |
16a2cde9 CB |
126 | * - CGROUP_LAYOUT_UNKNOWN |
127 | * The cgroup layout could not be determined. This should be treated as an | |
128 | * error condition. | |
129 | * - CGROUP_LAYOUT_LEGACY | |
130 | * The container is running with all controllers mounted into legacy cgroup | |
131 | * hierarchies. | |
132 | * - CGROUP_LAYOUT_HYBRID | |
133 | * The container is running with at least one controller mounted into a | |
134 | * legacy cgroup hierarchy and a mountpoint for the unified hierarchy. The | |
135 | * unified hierarchy can be empty (no controllers enabled) or non-empty | |
136 | * (controllers enabled). | |
137 | * - CGROUP_LAYOUT_UNIFIED | |
138 | * The container is running on a pure unified cgroup hierarchy. The unified | |
139 | * hierarchy can be empty (no controllers enabled) or non-empty (controllers | |
140 | * enabled). | |
ccb4cabe SH |
141 | */ |
142 | struct cgfsng_handler_data { | |
ccb4cabe | 143 | char *cgroup_pattern; |
1a0e70ac CB |
144 | char *container_cgroup; /* cgroup we created for the container */ |
145 | char *name; /* container name */ | |
43654d34 CB |
146 | /* per-container cgroup information */ |
147 | struct lxc_cgroup cgroup_meta; | |
d6337a5f | 148 | cgroup_layout_t cgroup_layout; |
ccb4cabe SH |
149 | }; |
150 | ||
09f3bb13 CB |
151 | /* @hierarchies |
152 | * - A NULL-terminated array of struct hierarchy, one per legacy hierarchy. No | |
153 | * duplicates. First sufficient, writeable mounted hierarchy wins. | |
457ca9aa SH |
154 | */ |
155 | struct hierarchy **hierarchies; | |
09f3bb13 CB |
156 | /* Pointer to the unified hierarchy in the null terminated list @hierarchies. |
157 | * This is merely a convenience for hybrid cgroup layouts to easily retrieve the | |
158 | * unified hierarchy without iterating throught @hierarchies. | |
159 | */ | |
d6337a5f | 160 | struct hierarchy *unified; |
457ca9aa | 161 | /* |
09f3bb13 CB |
162 | * @cgroup_layout |
163 | * - What cgroup layout the container is running with. | |
164 | * - CGROUP_LAYOUT_UNKNOWN | |
165 | * The cgroup layout could not be determined. This should be treated as an | |
166 | * error condition. | |
167 | * - CGROUP_LAYOUT_LEGACY | |
168 | * The container is running with all controllers mounted into legacy cgroup | |
169 | * hierarchies. | |
170 | * - CGROUP_LAYOUT_HYBRID | |
171 | * The container is running with at least one controller mounted into a | |
172 | * legacy cgroup hierarchy and a mountpoint for the unified hierarchy. The | |
173 | * unified hierarchy can be empty (no controllers enabled) or non-empty | |
174 | * (controllers enabled). | |
175 | * - CGROUP_LAYOUT_UNIFIED | |
176 | * The container is running on a pure unified cgroup hierarchy. The unified | |
177 | * hierarchy can be empty (no controllers enabled) or non-empty (controllers | |
178 | * enabled). | |
457ca9aa | 179 | */ |
09f3bb13 CB |
180 | cgroup_layout_t cgroup_layout; |
181 | /* What controllers is the container supposed to use. */ | |
457ca9aa SH |
182 | char *cgroup_use; |
183 | ||
09f3bb13 CB |
184 | /* @lxc_cgfsng_debug |
185 | * - Whether to print debug info to stdout for the cgfsng driver. | |
e4aeecf5 CB |
186 | */ |
187 | static bool lxc_cgfsng_debug; | |
188 | ||
09f3bb13 CB |
189 | #define CGFSNG_DEBUG(format, ...) \ |
190 | do { \ | |
191 | if (lxc_cgfsng_debug) \ | |
192 | printf("cgfsng: " format, ##__VA_ARGS__); \ | |
193 | } while (0) | |
65d78313 | 194 | |
ccb4cabe SH |
195 | static void free_string_list(char **clist) |
196 | { | |
2d5fe5ba | 197 | int i; |
ccb4cabe | 198 | |
2d5fe5ba CB |
199 | if (!clist) |
200 | return; | |
201 | ||
202 | for (i = 0; clist[i]; i++) | |
203 | free(clist[i]); | |
204 | ||
205 | free(clist); | |
ccb4cabe SH |
206 | } |
207 | ||
7745483d | 208 | /* Allocate a pointer, do not fail. */ |
ccb4cabe SH |
209 | static void *must_alloc(size_t sz) |
210 | { | |
211 | return must_realloc(NULL, sz); | |
212 | } | |
213 | ||
8b8db2f6 CB |
214 | /* Given a pointer to a null-terminated array of pointers, realloc to add one |
215 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
216 | * second-to-last entry - that is, the one which is now available for use | |
217 | * (keeping the list null-terminated). | |
ccb4cabe SH |
218 | */ |
219 | static int append_null_to_list(void ***list) | |
220 | { | |
221 | int newentry = 0; | |
222 | ||
223 | if (*list) | |
8b8db2f6 CB |
224 | for (; (*list)[newentry]; newentry++) |
225 | ; | |
ccb4cabe SH |
226 | |
227 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
228 | (*list)[newentry + 1] = NULL; | |
229 | return newentry; | |
230 | } | |
231 | ||
8073018d CB |
232 | /* Given a null-terminated array of strings, check whether @entry is one of the |
233 | * strings. | |
ccb4cabe SH |
234 | */ |
235 | static bool string_in_list(char **list, const char *entry) | |
236 | { | |
237 | int i; | |
238 | ||
239 | if (!list) | |
240 | return false; | |
d6337a5f | 241 | |
ccb4cabe SH |
242 | for (i = 0; list[i]; i++) |
243 | if (strcmp(list[i], entry) == 0) | |
244 | return true; | |
245 | ||
246 | return false; | |
247 | } | |
248 | ||
ac010944 CB |
249 | /* Return a copy of @entry prepending "name=", i.e. turn "systemd" into |
250 | * "name=systemd". Do not fail. | |
251 | */ | |
252 | static char *cg_legacy_must_prefix_named(char *entry) | |
253 | { | |
254 | size_t len; | |
255 | char *prefixed; | |
256 | ||
257 | len = strlen(entry); | |
258 | prefixed = must_alloc(len + 6); | |
259 | ||
260 | memcpy(prefixed, "name=", sizeof("name=")); | |
261 | memcpy(prefixed + sizeof("name="), entry, len); | |
262 | prefixed[len + 5] = '\0'; | |
263 | return prefixed; | |
264 | } | |
265 | ||
42a993b4 CB |
266 | /* Append an entry to the clist. Do not fail. @clist must be NULL the first time |
267 | * we are called. | |
ccb4cabe | 268 | * |
42a993b4 CB |
269 | * We also handle named subsystems here. Any controller which is not a kernel |
270 | * subsystem, we prefix "name=". Any which is both a kernel and named subsystem, | |
271 | * we refuse to use because we're not sure which we have here. | |
272 | * (TODO: We could work around this in some cases by just remounting to be | |
273 | * unambiguous, or by comparing mountpoint contents with current cgroup.) | |
ccb4cabe SH |
274 | * |
275 | * The last entry will always be NULL. | |
276 | */ | |
42a993b4 CB |
277 | static void must_append_controller(char **klist, char **nlist, char ***clist, |
278 | char *entry) | |
ccb4cabe SH |
279 | { |
280 | int newentry; | |
281 | char *copy; | |
282 | ||
283 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) { | |
c2712f64 | 284 | ERROR("Refusing to use ambiguous controller \"%s\"", entry); |
ccb4cabe SH |
285 | ERROR("It is both a named and kernel subsystem"); |
286 | return; | |
287 | } | |
288 | ||
289 | newentry = append_null_to_list((void ***)clist); | |
290 | ||
291 | if (strncmp(entry, "name=", 5) == 0) | |
292 | copy = must_copy_string(entry); | |
293 | else if (string_in_list(klist, entry)) | |
294 | copy = must_copy_string(entry); | |
295 | else | |
7745483d | 296 | copy = cg_legacy_must_prefix_named(entry); |
ccb4cabe SH |
297 | |
298 | (*clist)[newentry] = copy; | |
299 | } | |
300 | ||
ccb4cabe SH |
301 | static void free_handler_data(struct cgfsng_handler_data *d) |
302 | { | |
ccb4cabe SH |
303 | free(d->cgroup_pattern); |
304 | free(d->container_cgroup); | |
305 | free(d->name); | |
43654d34 CB |
306 | if (d->cgroup_meta.dir) |
307 | free(d->cgroup_meta.dir); | |
308 | if (d->cgroup_meta.controllers) | |
309 | free(d->cgroup_meta.controllers); | |
ccb4cabe SH |
310 | free(d); |
311 | } | |
312 | ||
5ae0207c CB |
313 | /* Given a handler's cgroup data, return the struct hierarchy for the controller |
314 | * @c, or NULL if there is none. | |
ccb4cabe | 315 | */ |
457ca9aa | 316 | struct hierarchy *get_hierarchy(const char *c) |
ccb4cabe SH |
317 | { |
318 | int i; | |
319 | ||
457ca9aa | 320 | if (!hierarchies) |
ccb4cabe | 321 | return NULL; |
d6337a5f | 322 | |
457ca9aa | 323 | for (i = 0; hierarchies[i]; i++) { |
d6337a5f CB |
324 | if (!c) { |
325 | /* This is the empty unified hierarchy. */ | |
326 | if (hierarchies[i]->controllers && | |
327 | !hierarchies[i]->controllers[0]) | |
328 | return hierarchies[i]; | |
329 | ||
330 | return NULL; | |
331 | } | |
332 | ||
457ca9aa SH |
333 | if (string_in_list(hierarchies[i]->controllers, c)) |
334 | return hierarchies[i]; | |
ccb4cabe | 335 | } |
d6337a5f | 336 | |
ccb4cabe SH |
337 | return NULL; |
338 | } | |
339 | ||
a54694f8 CB |
340 | #define BATCH_SIZE 50 |
341 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
342 | { | |
343 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
344 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
345 | ||
346 | if (!*mem || newbatches > oldbatches) { | |
347 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
348 | } | |
349 | } | |
350 | ||
351 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
352 | { | |
353 | size_t full = oldlen + newlen; | |
354 | ||
355 | batch_realloc(dest, oldlen, full + 1); | |
356 | ||
357 | memcpy(*dest + oldlen, new, newlen + 1); | |
358 | } | |
359 | ||
360 | /* Slurp in a whole file */ | |
d6337a5f | 361 | static char *read_file(const char *fnam) |
a54694f8 CB |
362 | { |
363 | FILE *f; | |
364 | char *line = NULL, *buf = NULL; | |
365 | size_t len = 0, fulllen = 0; | |
366 | int linelen; | |
367 | ||
368 | f = fopen(fnam, "r"); | |
369 | if (!f) | |
370 | return NULL; | |
371 | while ((linelen = getline(&line, &len, f)) != -1) { | |
372 | append_line(&buf, fulllen, line, linelen); | |
373 | fulllen += linelen; | |
374 | } | |
375 | fclose(f); | |
376 | free(line); | |
377 | return buf; | |
378 | } | |
379 | ||
380 | /* Taken over modified from the kernel sources. */ | |
381 | #define NBITS 32 /* bits in uint32_t */ | |
382 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
383 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
384 | ||
385 | static void set_bit(unsigned bit, uint32_t *bitarr) | |
386 | { | |
387 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
388 | } | |
389 | ||
390 | static void clear_bit(unsigned bit, uint32_t *bitarr) | |
391 | { | |
392 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
393 | } | |
394 | ||
395 | static bool is_set(unsigned bit, uint32_t *bitarr) | |
396 | { | |
397 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
398 | } | |
399 | ||
400 | /* Create cpumask from cpulist aka turn: | |
401 | * | |
402 | * 0,2-3 | |
403 | * | |
d5d468f6 | 404 | * into bit array |
a54694f8 CB |
405 | * |
406 | * 1 0 1 1 | |
407 | */ | |
408 | static uint32_t *lxc_cpumask(char *buf, size_t nbits) | |
409 | { | |
410 | char *token; | |
d5d468f6 CB |
411 | size_t arrlen; |
412 | uint32_t *bitarr; | |
a54694f8 | 413 | char *saveptr = NULL; |
d5d468f6 CB |
414 | |
415 | arrlen = BITS_TO_LONGS(nbits); | |
416 | bitarr = calloc(arrlen, sizeof(uint32_t)); | |
a54694f8 CB |
417 | if (!bitarr) |
418 | return NULL; | |
419 | ||
420 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
421 | errno = 0; | |
d5d468f6 CB |
422 | unsigned end, start; |
423 | char *range; | |
a54694f8 | 424 | |
d5d468f6 CB |
425 | start = strtoul(token, NULL, 0); |
426 | end = start; | |
427 | range = strchr(token, '-'); | |
a54694f8 CB |
428 | if (range) |
429 | end = strtoul(range + 1, NULL, 0); | |
d5d468f6 | 430 | |
a54694f8 CB |
431 | if (!(start <= end)) { |
432 | free(bitarr); | |
433 | return NULL; | |
434 | } | |
435 | ||
436 | if (end >= nbits) { | |
437 | free(bitarr); | |
438 | return NULL; | |
439 | } | |
440 | ||
441 | while (start <= end) | |
442 | set_bit(start++, bitarr); | |
443 | } | |
444 | ||
445 | return bitarr; | |
446 | } | |
447 | ||
a54694f8 CB |
448 | /* Turn cpumask into simple, comma-separated cpulist. */ |
449 | static char *lxc_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
450 | { | |
451 | size_t i; | |
452 | int ret; | |
eab15c1e | 453 | char numstr[LXC_NUMSTRLEN64] = {0}; |
a54694f8 CB |
454 | char **cpulist = NULL; |
455 | ||
456 | for (i = 0; i <= nbits; i++) { | |
457 | if (is_set(i, bitarr)) { | |
eab15c1e CB |
458 | ret = snprintf(numstr, LXC_NUMSTRLEN64, "%zu", i); |
459 | if (ret < 0 || (size_t)ret >= LXC_NUMSTRLEN64) { | |
a54694f8 CB |
460 | lxc_free_array((void **)cpulist, free); |
461 | return NULL; | |
462 | } | |
463 | if (lxc_append_string(&cpulist, numstr) < 0) { | |
464 | lxc_free_array((void **)cpulist, free); | |
465 | return NULL; | |
466 | } | |
467 | } | |
468 | } | |
469 | return lxc_string_join(",", (const char **)cpulist, false); | |
470 | } | |
471 | ||
472 | static ssize_t get_max_cpus(char *cpulist) | |
473 | { | |
474 | char *c1, *c2; | |
475 | char *maxcpus = cpulist; | |
476 | size_t cpus = 0; | |
477 | ||
478 | c1 = strrchr(maxcpus, ','); | |
479 | if (c1) | |
480 | c1++; | |
481 | ||
482 | c2 = strrchr(maxcpus, '-'); | |
483 | if (c2) | |
484 | c2++; | |
485 | ||
486 | if (!c1 && !c2) | |
487 | c1 = maxcpus; | |
488 | else if (c1 > c2) | |
489 | c2 = c1; | |
490 | else if (c1 < c2) | |
491 | c1 = c2; | |
1a0e70ac | 492 | else if (!c1 && c2) /* The reverse case is obvs. not needed. */ |
a54694f8 CB |
493 | c1 = c2; |
494 | ||
495 | /* If the above logic is correct, c1 should always hold a valid string | |
496 | * here. | |
497 | */ | |
498 | ||
499 | errno = 0; | |
500 | cpus = strtoul(c1, NULL, 0); | |
501 | if (errno != 0) | |
502 | return -1; | |
503 | ||
504 | return cpus; | |
505 | } | |
506 | ||
6f9584d8 | 507 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
a3926f6a | 508 | static bool cg_legacy_filter_and_set_cpus(char *path, bool am_initialized) |
a54694f8 CB |
509 | { |
510 | char *lastslash, *fpath, oldv; | |
511 | int ret; | |
512 | ssize_t i; | |
513 | ||
514 | ssize_t maxposs = 0, maxisol = 0; | |
515 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
516 | uint32_t *possmask = NULL, *isolmask = NULL; | |
6f9584d8 | 517 | bool bret = false, flipped_bit = false; |
a54694f8 CB |
518 | |
519 | lastslash = strrchr(path, '/'); | |
1a0e70ac | 520 | if (!lastslash) { /* bug... this shouldn't be possible */ |
6f9584d8 | 521 | ERROR("Invalid path: %s.", path); |
a54694f8 CB |
522 | return bret; |
523 | } | |
524 | oldv = *lastslash; | |
525 | *lastslash = '\0'; | |
526 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
527 | posscpus = read_file(fpath); | |
6f9584d8 CB |
528 | if (!posscpus) { |
529 | SYSERROR("Could not read file: %s.\n", fpath); | |
530 | goto on_error; | |
531 | } | |
a54694f8 CB |
532 | |
533 | /* Get maximum number of cpus found in possible cpuset. */ | |
534 | maxposs = get_max_cpus(posscpus); | |
535 | if (maxposs < 0) | |
6f9584d8 | 536 | goto on_error; |
a54694f8 | 537 | |
6f9584d8 CB |
538 | if (!file_exists(__ISOL_CPUS)) { |
539 | /* This system doesn't expose isolated cpus. */ | |
540 | DEBUG("Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
65d29cbc CB |
541 | cpulist = posscpus; |
542 | /* No isolated cpus but we weren't already initialized by | |
543 | * someone. We should simply copy the parents cpuset.cpus | |
544 | * values. | |
545 | */ | |
546 | if (!am_initialized) { | |
547 | DEBUG("Copying cpuset of parent cgroup."); | |
548 | goto copy_parent; | |
549 | } | |
550 | /* No isolated cpus but we were already initialized by someone. | |
551 | * Nothing more to do for us. | |
552 | */ | |
6f9584d8 CB |
553 | goto on_success; |
554 | } | |
555 | ||
556 | isolcpus = read_file(__ISOL_CPUS); | |
557 | if (!isolcpus) { | |
558 | SYSERROR("Could not read file "__ISOL_CPUS); | |
559 | goto on_error; | |
560 | } | |
a54694f8 | 561 | if (!isdigit(isolcpus[0])) { |
6f9584d8 | 562 | DEBUG("No isolated cpus detected."); |
a54694f8 CB |
563 | cpulist = posscpus; |
564 | /* No isolated cpus but we weren't already initialized by | |
565 | * someone. We should simply copy the parents cpuset.cpus | |
566 | * values. | |
567 | */ | |
6f9584d8 CB |
568 | if (!am_initialized) { |
569 | DEBUG("Copying cpuset of parent cgroup."); | |
a54694f8 | 570 | goto copy_parent; |
6f9584d8 | 571 | } |
a54694f8 CB |
572 | /* No isolated cpus but we were already initialized by someone. |
573 | * Nothing more to do for us. | |
574 | */ | |
6f9584d8 | 575 | goto on_success; |
a54694f8 CB |
576 | } |
577 | ||
578 | /* Get maximum number of cpus found in isolated cpuset. */ | |
579 | maxisol = get_max_cpus(isolcpus); | |
580 | if (maxisol < 0) | |
6f9584d8 | 581 | goto on_error; |
a54694f8 CB |
582 | |
583 | if (maxposs < maxisol) | |
584 | maxposs = maxisol; | |
585 | maxposs++; | |
586 | ||
587 | possmask = lxc_cpumask(posscpus, maxposs); | |
6f9584d8 CB |
588 | if (!possmask) { |
589 | ERROR("Could not create cpumask for all possible cpus.\n"); | |
590 | goto on_error; | |
591 | } | |
a54694f8 CB |
592 | |
593 | isolmask = lxc_cpumask(isolcpus, maxposs); | |
6f9584d8 CB |
594 | if (!isolmask) { |
595 | ERROR("Could not create cpumask for all isolated cpus.\n"); | |
596 | goto on_error; | |
597 | } | |
a54694f8 CB |
598 | |
599 | for (i = 0; i <= maxposs; i++) { | |
600 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
6f9584d8 | 601 | flipped_bit = true; |
a54694f8 CB |
602 | clear_bit(i, possmask); |
603 | } | |
604 | } | |
605 | ||
6f9584d8 CB |
606 | if (!flipped_bit) { |
607 | DEBUG("No isolated cpus present in cpuset."); | |
608 | goto on_success; | |
609 | } | |
610 | DEBUG("Removed isolated cpus from cpuset."); | |
611 | ||
a54694f8 | 612 | cpulist = lxc_cpumask_to_cpulist(possmask, maxposs); |
6f9584d8 CB |
613 | if (!cpulist) { |
614 | ERROR("Could not create cpu list.\n"); | |
615 | goto on_error; | |
616 | } | |
a54694f8 CB |
617 | |
618 | copy_parent: | |
619 | *lastslash = oldv; | |
dcbc861e | 620 | free(fpath); |
a54694f8 CB |
621 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
622 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false); | |
6f9584d8 CB |
623 | if (ret < 0) { |
624 | SYSERROR("Could not write cpu list to: %s.\n", fpath); | |
625 | goto on_error; | |
626 | } | |
627 | ||
628 | on_success: | |
629 | bret = true; | |
a54694f8 | 630 | |
6f9584d8 | 631 | on_error: |
a54694f8 CB |
632 | free(fpath); |
633 | ||
634 | free(isolcpus); | |
635 | free(isolmask); | |
636 | ||
637 | if (posscpus != cpulist) | |
638 | free(posscpus); | |
639 | free(possmask); | |
640 | ||
641 | free(cpulist); | |
642 | return bret; | |
643 | } | |
644 | ||
e3a3fecf SH |
645 | /* Copy contents of parent(@path)/@file to @path/@file */ |
646 | static bool copy_parent_file(char *path, char *file) | |
647 | { | |
648 | char *lastslash, *value = NULL, *fpath, oldv; | |
649 | int len = 0; | |
650 | int ret; | |
651 | ||
652 | lastslash = strrchr(path, '/'); | |
1a0e70ac | 653 | if (!lastslash) { /* bug... this shouldn't be possible */ |
e3a3fecf SH |
654 | ERROR("cgfsng:copy_parent_file: bad path %s", path); |
655 | return false; | |
656 | } | |
657 | oldv = *lastslash; | |
658 | *lastslash = '\0'; | |
659 | fpath = must_make_path(path, file, NULL); | |
660 | len = lxc_read_from_file(fpath, NULL, 0); | |
661 | if (len <= 0) | |
662 | goto bad; | |
663 | value = must_alloc(len + 1); | |
664 | if (lxc_read_from_file(fpath, value, len) != len) | |
665 | goto bad; | |
666 | free(fpath); | |
667 | *lastslash = oldv; | |
668 | fpath = must_make_path(path, file, NULL); | |
669 | ret = lxc_write_to_file(fpath, value, len, false); | |
670 | if (ret < 0) | |
671 | SYSERROR("Unable to write %s to %s", value, fpath); | |
672 | free(fpath); | |
673 | free(value); | |
674 | return ret >= 0; | |
675 | ||
676 | bad: | |
677 | SYSERROR("Error reading '%s'", fpath); | |
678 | free(fpath); | |
679 | free(value); | |
680 | return false; | |
681 | } | |
682 | ||
683 | /* | |
684 | * Initialize the cpuset hierarchy in first directory of @gname and | |
685 | * set cgroup.clone_children so that children inherit settings. | |
686 | * Since the h->base_path is populated by init or ourselves, we know | |
687 | * it is already initialized. | |
688 | */ | |
a3926f6a | 689 | static bool cg_legacy_handle_cpuset_hierarchy(struct hierarchy *h, char *cgname) |
e3a3fecf SH |
690 | { |
691 | char *cgpath, *clonechildrenpath, v, *slash; | |
692 | ||
693 | if (!string_in_list(h->controllers, "cpuset")) | |
694 | return true; | |
695 | ||
696 | if (*cgname == '/') | |
697 | cgname++; | |
698 | slash = strchr(cgname, '/'); | |
699 | if (slash) | |
700 | *slash = '\0'; | |
701 | ||
702 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); | |
703 | if (slash) | |
704 | *slash = '/'; | |
705 | if (mkdir(cgpath, 0755) < 0 && errno != EEXIST) { | |
706 | SYSERROR("Failed to create '%s'", cgpath); | |
707 | free(cgpath); | |
708 | return false; | |
709 | } | |
6f9584d8 | 710 | |
e3a3fecf | 711 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
6328fd9c CB |
712 | /* unified hierarchy doesn't have clone_children */ |
713 | if (!file_exists(clonechildrenpath)) { | |
e3a3fecf SH |
714 | free(clonechildrenpath); |
715 | free(cgpath); | |
716 | return true; | |
717 | } | |
718 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { | |
719 | SYSERROR("Failed to read '%s'", clonechildrenpath); | |
720 | free(clonechildrenpath); | |
721 | free(cgpath); | |
722 | return false; | |
723 | } | |
724 | ||
a54694f8 | 725 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ |
a3926f6a | 726 | if (!cg_legacy_filter_and_set_cpus(cgpath, v == '1')) { |
6f9584d8 CB |
727 | SYSERROR("Failed to remove isolated cpus."); |
728 | free(clonechildrenpath); | |
729 | free(cgpath); | |
a54694f8 | 730 | return false; |
6f9584d8 | 731 | } |
a54694f8 | 732 | |
e3a3fecf | 733 | if (v == '1') { /* already set for us by someone else */ |
6f9584d8 | 734 | DEBUG("\"cgroup.clone_children\" was already set to \"1\"."); |
e3a3fecf SH |
735 | free(clonechildrenpath); |
736 | free(cgpath); | |
737 | return true; | |
738 | } | |
739 | ||
740 | /* copy parent's settings */ | |
a54694f8 | 741 | if (!copy_parent_file(cgpath, "cpuset.mems")) { |
6f9584d8 | 742 | SYSERROR("Failed to copy \"cpuset.mems\" settings."); |
e3a3fecf SH |
743 | free(cgpath); |
744 | free(clonechildrenpath); | |
745 | return false; | |
746 | } | |
747 | free(cgpath); | |
748 | ||
749 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
750 | /* Set clone_children so children inherit our settings */ | |
751 | SYSERROR("Failed to write 1 to %s", clonechildrenpath); | |
752 | free(clonechildrenpath); | |
753 | return false; | |
754 | } | |
755 | free(clonechildrenpath); | |
756 | return true; | |
757 | } | |
758 | ||
ccb4cabe SH |
759 | /* |
760 | * Given two null-terminated lists of strings, return true if any string | |
761 | * is in both. | |
762 | */ | |
763 | static bool controller_lists_intersect(char **l1, char **l2) | |
764 | { | |
765 | int i; | |
766 | ||
767 | if (!l1 || !l2) | |
768 | return false; | |
769 | ||
770 | for (i = 0; l1[i]; i++) { | |
771 | if (string_in_list(l2, l1[i])) | |
772 | return true; | |
773 | } | |
774 | return false; | |
775 | } | |
776 | ||
777 | /* | |
778 | * For a null-terminated list of controllers @clist, return true if any of | |
779 | * those controllers is already listed the null-terminated list of | |
780 | * hierarchies @hlist. Realistically, if one is present, all must be present. | |
781 | */ | |
782 | static bool controller_list_is_dup(struct hierarchy **hlist, char **clist) | |
783 | { | |
784 | int i; | |
785 | ||
786 | if (!hlist) | |
787 | return false; | |
788 | for (i = 0; hlist[i]; i++) | |
789 | if (controller_lists_intersect(hlist[i]->controllers, clist)) | |
790 | return true; | |
791 | return false; | |
792 | ||
793 | } | |
794 | ||
795 | /* | |
796 | * Return true if the controller @entry is found in the null-terminated | |
797 | * list of hierarchies @hlist | |
798 | */ | |
799 | static bool controller_found(struct hierarchy **hlist, char *entry) | |
800 | { | |
801 | int i; | |
d6337a5f | 802 | |
ccb4cabe SH |
803 | if (!hlist) |
804 | return false; | |
805 | ||
806 | for (i = 0; hlist[i]; i++) | |
807 | if (string_in_list(hlist[i]->controllers, entry)) | |
808 | return true; | |
d6337a5f | 809 | |
ccb4cabe SH |
810 | return false; |
811 | } | |
812 | ||
813 | /* | |
c30b61c3 SH |
814 | * Return true if all of the controllers which we require have been found. |
815 | * The required list is freezer and anything in * lxc.cgroup.use. | |
ccb4cabe | 816 | */ |
457ca9aa | 817 | static bool all_controllers_found(void) |
ccb4cabe SH |
818 | { |
819 | char *p, *saveptr = NULL; | |
457ca9aa | 820 | struct hierarchy ** hlist = hierarchies; |
ccb4cabe | 821 | |
ccb4cabe | 822 | if (!controller_found(hlist, "freezer")) { |
65d78313 | 823 | CGFSNG_DEBUG("No freezer controller mountpoint found\n"); |
ccb4cabe SH |
824 | return false; |
825 | } | |
826 | ||
457ca9aa | 827 | if (!cgroup_use) |
ccb4cabe | 828 | return true; |
c2712f64 | 829 | |
457ca9aa | 830 | for (p = strtok_r(cgroup_use, ",", &saveptr); p; |
ccb4cabe SH |
831 | p = strtok_r(NULL, ",", &saveptr)) { |
832 | if (!controller_found(hlist, p)) { | |
65d78313 | 833 | CGFSNG_DEBUG("No %s controller mountpoint found\n", p); |
ccb4cabe SH |
834 | return false; |
835 | } | |
836 | } | |
c2712f64 | 837 | |
ccb4cabe SH |
838 | return true; |
839 | } | |
840 | ||
ccb4cabe SH |
841 | /* |
842 | * Get the controllers from a mountinfo line | |
843 | * There are other ways we could get this info. For lxcfs, field 3 | |
844 | * is /cgroup/controller-list. For cgroupfs, we could parse the mount | |
845 | * options. But we simply assume that the mountpoint must be | |
846 | * /sys/fs/cgroup/controller-list | |
847 | */ | |
a3926f6a CB |
848 | static char **cg_hybrid_get_controllers(char **klist, char **nlist, char *line, |
849 | int type) | |
ccb4cabe | 850 | { |
6328fd9c | 851 | /* the fourth field is /sys/fs/cgroup/comma-delimited-controller-list */ |
ccb4cabe | 852 | int i; |
411ac6d8 | 853 | char *dup, *p2, *tok; |
d6337a5f | 854 | char *p = line, *saveptr = NULL, *sep = ","; |
411ac6d8 | 855 | char **aret = NULL; |
6328fd9c | 856 | |
ccb4cabe | 857 | for (i = 0; i < 4; i++) { |
235f1815 | 858 | p = strchr(p, ' '); |
ccb4cabe SH |
859 | if (!p) |
860 | return NULL; | |
861 | p++; | |
862 | } | |
a55f31bd | 863 | |
ccb4cabe SH |
864 | /* note - if we change how mountinfo works, then our caller |
865 | * will need to verify /sys/fs/cgroup/ in this field */ | |
c2712f64 | 866 | if (strncmp(p, "/sys/fs/cgroup/", 15)) { |
65d78313 | 867 | CGFSNG_DEBUG("Found hierarchy not under /sys/fs/cgroup: \"%s\"\n", p); |
ccb4cabe | 868 | return NULL; |
5059aae9 | 869 | } |
d6337a5f | 870 | |
ccb4cabe | 871 | p += 15; |
235f1815 | 872 | p2 = strchr(p, ' '); |
ccb4cabe | 873 | if (!p2) { |
65d78313 | 874 | CGFSNG_DEBUG("Corrupt mountinfo\n"); |
ccb4cabe SH |
875 | return NULL; |
876 | } | |
877 | *p2 = '\0'; | |
6328fd9c | 878 | |
d6337a5f CB |
879 | if (type == CGROUP_SUPER_MAGIC) { |
880 | /* strdup() here for v1 hierarchies. Otherwise strtok_r() will | |
881 | * destroy mountpoints such as "/sys/fs/cgroup/cpu,cpuacct". | |
882 | */ | |
883 | dup = strdup(p); | |
884 | if (!dup) | |
885 | return NULL; | |
886 | ||
887 | for (tok = strtok_r(dup, sep, &saveptr); tok; | |
888 | tok = strtok_r(NULL, sep, &saveptr)) | |
889 | must_append_controller(klist, nlist, &aret, tok); | |
890 | ||
891 | free(dup); | |
411ac6d8 | 892 | } |
d6337a5f CB |
893 | *p2 = ' '; |
894 | return aret; | |
895 | } | |
411ac6d8 | 896 | |
d6337a5f CB |
897 | static char **cg_unified_make_empty_controller(void) |
898 | { | |
899 | int newentry; | |
900 | char **aret = NULL; | |
901 | ||
902 | newentry = append_null_to_list((void ***)&aret); | |
903 | aret[newentry] = NULL; | |
904 | return aret; | |
905 | } | |
906 | ||
907 | static char **cg_unified_get_controllers(const char *file) | |
908 | { | |
909 | char *buf, *tok; | |
910 | char *saveptr = NULL, *sep = " \t\n"; | |
911 | char **aret = NULL; | |
912 | ||
913 | buf = read_file(file); | |
914 | if (!buf) | |
411ac6d8 | 915 | return NULL; |
6328fd9c | 916 | |
d6337a5f CB |
917 | for (tok = strtok_r(buf, sep, &saveptr); tok; |
918 | tok = strtok_r(NULL, sep, &saveptr)) { | |
919 | int newentry; | |
920 | char *copy; | |
921 | ||
922 | newentry = append_null_to_list((void ***)&aret); | |
923 | copy = must_copy_string(tok); | |
924 | aret[newentry] = copy; | |
ccb4cabe SH |
925 | } |
926 | ||
d6337a5f | 927 | free(buf); |
ccb4cabe SH |
928 | return aret; |
929 | } | |
930 | ||
d6337a5f CB |
931 | static struct hierarchy *add_hierarchy(char **clist, char *mountpoint, |
932 | char *base_cgroup, int type) | |
ccb4cabe SH |
933 | { |
934 | struct hierarchy *new; | |
935 | int newentry; | |
936 | ||
937 | new = must_alloc(sizeof(*new)); | |
938 | new->controllers = clist; | |
939 | new->mountpoint = mountpoint; | |
940 | new->base_cgroup = base_cgroup; | |
941 | new->fullcgpath = NULL; | |
d6337a5f | 942 | new->version = type; |
6328fd9c | 943 | |
457ca9aa SH |
944 | newentry = append_null_to_list((void ***)&hierarchies); |
945 | hierarchies[newentry] = new; | |
d6337a5f | 946 | return new; |
ccb4cabe SH |
947 | } |
948 | ||
949 | /* | |
950 | * Get a copy of the mountpoint from @line, which is a line from | |
951 | * /proc/self/mountinfo | |
952 | */ | |
a3926f6a | 953 | static char *cg_hybrid_get_mountpoint(char *line) |
ccb4cabe SH |
954 | { |
955 | int i; | |
d6337a5f | 956 | char *p2; |
ccb4cabe | 957 | size_t len; |
d6337a5f CB |
958 | char *p = line; |
959 | char *sret = NULL; | |
ccb4cabe SH |
960 | |
961 | for (i = 0; i < 4; i++) { | |
235f1815 | 962 | p = strchr(p, ' '); |
ccb4cabe SH |
963 | if (!p) |
964 | return NULL; | |
965 | p++; | |
966 | } | |
d6337a5f CB |
967 | |
968 | if (strncmp(p, "/sys/fs/cgroup/", 15)) | |
969 | return NULL; | |
970 | ||
971 | p2 = strchr(p + 15, ' '); | |
972 | if (!p2) | |
973 | return NULL; | |
974 | *p2 = '\0'; | |
975 | ||
ccb4cabe SH |
976 | len = strlen(p); |
977 | sret = must_alloc(len + 1); | |
978 | memcpy(sret, p, len); | |
979 | sret[len] = '\0'; | |
980 | return sret; | |
981 | } | |
982 | ||
983 | /* | |
984 | * Given a multi-line string, return a null-terminated copy of the | |
985 | * current line. | |
986 | */ | |
987 | static char *copy_to_eol(char *p) | |
988 | { | |
235f1815 | 989 | char *p2 = strchr(p, '\n'), *sret; |
ccb4cabe SH |
990 | size_t len; |
991 | ||
992 | if (!p2) | |
993 | return NULL; | |
994 | ||
995 | len = p2 - p; | |
996 | sret = must_alloc(len + 1); | |
997 | memcpy(sret, p, len); | |
998 | sret[len] = '\0'; | |
999 | return sret; | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * cgline: pointer to character after the first ':' in a line in a | |
1004 | * \n-terminated /proc/self/cgroup file. Check whether * controller c is | |
1005 | * present. | |
1006 | */ | |
1007 | static bool controller_in_clist(char *cgline, char *c) | |
1008 | { | |
1009 | char *tok, *saveptr = NULL, *eol, *tmp; | |
1010 | size_t len; | |
1011 | ||
235f1815 | 1012 | eol = strchr(cgline, ':'); |
ccb4cabe SH |
1013 | if (!eol) |
1014 | return false; | |
1015 | ||
1016 | len = eol - cgline; | |
1017 | tmp = alloca(len + 1); | |
1018 | memcpy(tmp, cgline, len); | |
1019 | tmp[len] = '\0'; | |
1020 | ||
1021 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
d6337a5f | 1022 | tok = strtok_r(NULL, ",", &saveptr)) { |
ccb4cabe SH |
1023 | if (strcmp(tok, c) == 0) |
1024 | return true; | |
1025 | } | |
d6337a5f | 1026 | |
ccb4cabe SH |
1027 | return false; |
1028 | } | |
1029 | ||
1030 | /* | |
1031 | * @basecginfo is a copy of /proc/$$/cgroup. Return the current | |
1032 | * cgroup for @controller | |
1033 | */ | |
a3926f6a | 1034 | static char *cg_hybrid_get_current_cgroup(char *basecginfo, char *controller, int type) |
ccb4cabe SH |
1035 | { |
1036 | char *p = basecginfo; | |
6328fd9c | 1037 | |
d6337a5f CB |
1038 | for (;;) { |
1039 | bool is_cgv2_base_cgroup = false; | |
1040 | ||
6328fd9c | 1041 | /* cgroup v2 entry in "/proc/<pid>/cgroup": "0::/some/path" */ |
d6337a5f CB |
1042 | if ((type == CGROUP2_SUPER_MAGIC) && (*p == '0')) |
1043 | is_cgv2_base_cgroup = true; | |
ccb4cabe | 1044 | |
235f1815 | 1045 | p = strchr(p, ':'); |
ccb4cabe SH |
1046 | if (!p) |
1047 | return NULL; | |
1048 | p++; | |
d6337a5f CB |
1049 | |
1050 | if (is_cgv2_base_cgroup || (controller && controller_in_clist(p, controller))) { | |
235f1815 | 1051 | p = strchr(p, ':'); |
ccb4cabe SH |
1052 | if (!p) |
1053 | return NULL; | |
1054 | p++; | |
1055 | return copy_to_eol(p); | |
1056 | } | |
1057 | ||
235f1815 | 1058 | p = strchr(p, '\n'); |
ccb4cabe SH |
1059 | if (!p) |
1060 | return NULL; | |
1061 | p++; | |
1062 | } | |
1063 | } | |
1064 | ||
ccb4cabe SH |
1065 | static void must_append_string(char ***list, char *entry) |
1066 | { | |
1067 | int newentry = append_null_to_list((void ***)list); | |
1068 | char *copy; | |
1069 | ||
1070 | copy = must_copy_string(entry); | |
1071 | (*list)[newentry] = copy; | |
1072 | } | |
1073 | ||
d6337a5f | 1074 | static int get_existing_subsystems(char ***klist, char ***nlist) |
ccb4cabe SH |
1075 | { |
1076 | FILE *f; | |
1077 | char *line = NULL; | |
1078 | size_t len = 0; | |
1079 | ||
d6337a5f CB |
1080 | f = fopen("/proc/self/cgroup", "r"); |
1081 | if (!f) | |
1082 | return -1; | |
1083 | ||
ccb4cabe SH |
1084 | while (getline(&line, &len, f) != -1) { |
1085 | char *p, *p2, *tok, *saveptr = NULL; | |
235f1815 | 1086 | p = strchr(line, ':'); |
ccb4cabe SH |
1087 | if (!p) |
1088 | continue; | |
1089 | p++; | |
235f1815 | 1090 | p2 = strchr(p, ':'); |
ccb4cabe SH |
1091 | if (!p2) |
1092 | continue; | |
1093 | *p2 = '\0'; | |
ff8d6ee9 | 1094 | |
6328fd9c CB |
1095 | /* If the kernel has cgroup v2 support, then /proc/self/cgroup |
1096 | * contains an entry of the form: | |
ff8d6ee9 CB |
1097 | * |
1098 | * 0::/some/path | |
1099 | * | |
6328fd9c | 1100 | * In this case we use "cgroup2" as controller name. |
ff8d6ee9 | 1101 | */ |
6328fd9c CB |
1102 | if ((p2 - p) == 0) { |
1103 | must_append_string(klist, "cgroup2"); | |
ff8d6ee9 | 1104 | continue; |
6328fd9c | 1105 | } |
ff8d6ee9 | 1106 | |
ccb4cabe | 1107 | for (tok = strtok_r(p, ",", &saveptr); tok; |
d6337a5f | 1108 | tok = strtok_r(NULL, ",", &saveptr)) { |
ccb4cabe SH |
1109 | if (strncmp(tok, "name=", 5) == 0) |
1110 | must_append_string(nlist, tok); | |
1111 | else | |
1112 | must_append_string(klist, tok); | |
1113 | } | |
1114 | } | |
1115 | ||
1116 | free(line); | |
1117 | fclose(f); | |
d6337a5f | 1118 | return 0; |
ccb4cabe SH |
1119 | } |
1120 | ||
1121 | static void trim(char *s) | |
1122 | { | |
1123 | size_t len = strlen(s); | |
2c28d76b | 1124 | while ((len > 1) && (s[len - 1] == '\n')) |
ccb4cabe SH |
1125 | s[--len] = '\0'; |
1126 | } | |
1127 | ||
e4aeecf5 CB |
1128 | static void lxc_cgfsng_print_handler_data(const struct cgfsng_handler_data *d) |
1129 | { | |
1130 | printf("Cgroup information:\n"); | |
1131 | printf(" container name: %s\n", d->name ? d->name : "(null)"); | |
1132 | printf(" lxc.cgroup.use: %s\n", cgroup_use ? cgroup_use : "(null)"); | |
43654d34 CB |
1133 | printf(" lxc.cgroup.pattern: %s\n", |
1134 | d->cgroup_pattern ? d->cgroup_pattern : "(null)"); | |
1135 | printf(" lxc.cgroup.dir: %s\n", | |
1136 | d->cgroup_meta.dir ? d->cgroup_meta.dir : "(null)"); | |
1137 | printf(" cgroup: %s\n", | |
1138 | d->container_cgroup ? d->container_cgroup : "(null)"); | |
e4aeecf5 CB |
1139 | } |
1140 | ||
1141 | static void lxc_cgfsng_print_hierarchies() | |
ccb4cabe | 1142 | { |
a7b0cc4c | 1143 | struct hierarchy **it; |
ccb4cabe | 1144 | int i; |
41c33dbe | 1145 | |
457ca9aa | 1146 | if (!hierarchies) { |
c2712f64 | 1147 | printf(" No hierarchies found\n"); |
ccb4cabe SH |
1148 | return; |
1149 | } | |
e4aeecf5 | 1150 | printf(" Hierarchies:\n"); |
a7b0cc4c CB |
1151 | for (i = 0, it = hierarchies; it && *it; it++, i++) { |
1152 | char **cit; | |
ccb4cabe | 1153 | int j; |
c2712f64 CB |
1154 | printf(" %d: base_cgroup: %s\n", i, (*it)->base_cgroup ? (*it)->base_cgroup : "(null)"); |
1155 | printf(" mountpoint: %s\n", (*it)->mountpoint ? (*it)->mountpoint : "(null)"); | |
e4aeecf5 | 1156 | printf(" controllers:\n"); |
a7b0cc4c | 1157 | for (j = 0, cit = (*it)->controllers; cit && *cit; cit++, j++) |
e4aeecf5 | 1158 | printf(" %d: %s\n", j, *cit); |
ccb4cabe SH |
1159 | } |
1160 | } | |
41c33dbe | 1161 | |
a3926f6a CB |
1162 | static void lxc_cgfsng_print_basecg_debuginfo(char *basecginfo, char **klist, |
1163 | char **nlist) | |
41c33dbe SH |
1164 | { |
1165 | int k; | |
a7b0cc4c | 1166 | char **it; |
41c33dbe | 1167 | |
a7b0cc4c CB |
1168 | printf("basecginfo is:\n"); |
1169 | printf("%s\n", basecginfo); | |
41c33dbe | 1170 | |
a7b0cc4c CB |
1171 | for (k = 0, it = klist; it && *it; it++, k++) |
1172 | printf("kernel subsystem %d: %s\n", k, *it); | |
1173 | for (k = 0, it = nlist; it && *it; it++, k++) | |
1174 | printf("named subsystem %d: %s\n", k, *it); | |
41c33dbe | 1175 | } |
ccb4cabe | 1176 | |
e4aeecf5 CB |
1177 | static void lxc_cgfsng_print_debuginfo(const struct cgfsng_handler_data *d) |
1178 | { | |
1179 | lxc_cgfsng_print_handler_data(d); | |
1180 | lxc_cgfsng_print_hierarchies(); | |
1181 | } | |
1182 | ||
ccb4cabe SH |
1183 | /* |
1184 | * At startup, parse_hierarchies finds all the info we need about | |
1185 | * cgroup mountpoints and current cgroups, and stores it in @d. | |
1186 | */ | |
a3926f6a | 1187 | static bool cg_hybrid_init(void) |
ccb4cabe | 1188 | { |
d6337a5f CB |
1189 | int ret; |
1190 | char *basecginfo; | |
1191 | bool will_escape; | |
ccb4cabe | 1192 | FILE *f; |
ccb4cabe | 1193 | size_t len = 0; |
d6337a5f CB |
1194 | char *line = NULL; |
1195 | char **klist = NULL, **nlist = NULL; | |
ccb4cabe | 1196 | |
d30ec4cb SH |
1197 | /* |
1198 | * Root spawned containers escape the current cgroup, so use init's | |
1199 | * cgroups as our base in that case. | |
1200 | */ | |
d6337a5f CB |
1201 | will_escape = (geteuid() == 0); |
1202 | if (will_escape) | |
ccb4cabe | 1203 | basecginfo = read_file("/proc/1/cgroup"); |
d6337a5f CB |
1204 | else |
1205 | basecginfo = read_file("/proc/self/cgroup"); | |
ccb4cabe SH |
1206 | if (!basecginfo) |
1207 | return false; | |
1208 | ||
d6337a5f CB |
1209 | ret = get_existing_subsystems(&klist, &nlist); |
1210 | if (ret < 0) { | |
1211 | CGFSNG_DEBUG("Failed to retrieve available cgroup v1 controllers\n"); | |
1212 | free(basecginfo); | |
ccb4cabe SH |
1213 | return false; |
1214 | } | |
1215 | ||
d6337a5f CB |
1216 | f = fopen("/proc/self/mountinfo", "r"); |
1217 | if (!f) { | |
1218 | CGFSNG_DEBUG("Failed to open \"/proc/self/mountinfo\"\n"); | |
bd01b7d5 | 1219 | free(basecginfo); |
d6337a5f CB |
1220 | return false; |
1221 | } | |
41c33dbe | 1222 | |
e4aeecf5 CB |
1223 | if (lxc_cgfsng_debug) |
1224 | lxc_cgfsng_print_basecg_debuginfo(basecginfo, klist, nlist); | |
ccb4cabe | 1225 | |
ccb4cabe | 1226 | while (getline(&line, &len, f) != -1) { |
49ff3958 | 1227 | int type; |
d6337a5f CB |
1228 | bool writeable; |
1229 | struct hierarchy *new; | |
1230 | char *mountpoint = NULL, *base_cgroup = NULL; | |
1231 | char **controller_list = NULL; | |
ccb4cabe | 1232 | |
49ff3958 | 1233 | type = get_cgroup_version(line); |
d6337a5f | 1234 | if (type == 0) |
ccb4cabe SH |
1235 | continue; |
1236 | ||
d6337a5f | 1237 | if (type == CGROUP2_SUPER_MAGIC && unified) |
ccb4cabe SH |
1238 | continue; |
1239 | ||
d6337a5f CB |
1240 | if (cgroup_layout == CGROUP_LAYOUT_UNKNOWN) { |
1241 | if (type == CGROUP2_SUPER_MAGIC) | |
1242 | cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
1243 | else if (type == CGROUP_SUPER_MAGIC) | |
1244 | cgroup_layout = CGROUP_LAYOUT_LEGACY; | |
1245 | } else if (cgroup_layout == CGROUP_LAYOUT_UNIFIED) { | |
1246 | if (type == CGROUP_SUPER_MAGIC) | |
1247 | cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
1248 | } else if (cgroup_layout == CGROUP_LAYOUT_LEGACY) { | |
1249 | if (type == CGROUP2_SUPER_MAGIC) | |
1250 | cgroup_layout = CGROUP_LAYOUT_HYBRID; | |
ccb4cabe SH |
1251 | } |
1252 | ||
a3926f6a | 1253 | controller_list = cg_hybrid_get_controllers(klist, nlist, line, type); |
d6337a5f CB |
1254 | if (!controller_list && type == CGROUP_SUPER_MAGIC) |
1255 | continue; | |
1256 | ||
1257 | if (type == CGROUP_SUPER_MAGIC) | |
1258 | if (controller_list_is_dup(hierarchies, controller_list)) | |
1259 | goto next; | |
1260 | ||
a3926f6a | 1261 | mountpoint = cg_hybrid_get_mountpoint(line); |
ccb4cabe | 1262 | if (!mountpoint) { |
65d78313 | 1263 | CGFSNG_DEBUG("Failed parsing mountpoint from \"%s\"\n", line); |
d6337a5f | 1264 | goto next; |
ccb4cabe SH |
1265 | } |
1266 | ||
d6337a5f | 1267 | if (type == CGROUP_SUPER_MAGIC) |
a3926f6a | 1268 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, controller_list[0], CGROUP_SUPER_MAGIC); |
d6337a5f | 1269 | else |
a3926f6a | 1270 | base_cgroup = cg_hybrid_get_current_cgroup(basecginfo, NULL, CGROUP2_SUPER_MAGIC); |
ccb4cabe | 1271 | if (!base_cgroup) { |
d6337a5f CB |
1272 | CGFSNG_DEBUG("Failed to find current cgroup\n"); |
1273 | goto next; | |
ccb4cabe | 1274 | } |
6328fd9c | 1275 | |
ccb4cabe SH |
1276 | trim(base_cgroup); |
1277 | prune_init_scope(base_cgroup); | |
d6337a5f | 1278 | if (type == CGROUP2_SUPER_MAGIC) |
6328fd9c CB |
1279 | writeable = test_writeable_v2(mountpoint, base_cgroup); |
1280 | else | |
1281 | writeable = test_writeable_v1(mountpoint, base_cgroup); | |
d6337a5f CB |
1282 | if (!writeable) |
1283 | goto next; | |
1284 | ||
1285 | if (type == CGROUP2_SUPER_MAGIC) { | |
1286 | char *cgv2_ctrl_path; | |
1287 | ||
1288 | cgv2_ctrl_path = must_make_path(mountpoint, base_cgroup, | |
1289 | "cgroup.controllers", | |
1290 | NULL); | |
1291 | ||
1292 | controller_list = cg_unified_get_controllers(cgv2_ctrl_path); | |
1293 | free(cgv2_ctrl_path); | |
1294 | if (!controller_list) | |
1295 | controller_list = cg_unified_make_empty_controller(); | |
ccb4cabe | 1296 | } |
d6337a5f CB |
1297 | new = add_hierarchy(controller_list, mountpoint, base_cgroup, type); |
1298 | if (type == CGROUP2_SUPER_MAGIC && !unified) | |
1299 | unified = new; | |
1300 | ||
1301 | continue; | |
1302 | ||
1303 | next: | |
1304 | free_string_list(controller_list); | |
1305 | free(mountpoint); | |
1306 | free(base_cgroup); | |
ccb4cabe SH |
1307 | } |
1308 | ||
1309 | free_string_list(klist); | |
1310 | free_string_list(nlist); | |
1311 | ||
1312 | free(basecginfo); | |
1313 | ||
1314 | fclose(f); | |
1315 | free(line); | |
1316 | ||
e4aeecf5 CB |
1317 | if (lxc_cgfsng_debug) { |
1318 | printf("writeable subsystems:\n"); | |
1319 | lxc_cgfsng_print_hierarchies(); | |
1320 | } | |
1321 | ||
ccb4cabe SH |
1322 | /* verify that all controllers in cgroup.use and all crucial |
1323 | * controllers are accounted for | |
1324 | */ | |
c2712f64 | 1325 | if (!all_controllers_found()) |
ccb4cabe SH |
1326 | return false; |
1327 | ||
1328 | return true; | |
1329 | } | |
1330 | ||
d6337a5f CB |
1331 | static int cg_is_pure_unified(void) { |
1332 | ||
1333 | int ret; | |
1334 | struct statfs fs; | |
1335 | ||
1336 | ret = statfs("/sys/fs/cgroup", &fs); | |
1337 | if (ret < 0) | |
1338 | return -ENOMEDIUM; | |
1339 | ||
1340 | if (is_fs_type(&fs, CGROUP2_SUPER_MAGIC)) | |
1341 | return CGROUP2_SUPER_MAGIC; | |
1342 | ||
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
a3926f6a | 1347 | static char *cg_unified_get_current_cgroup(void) |
457ca9aa | 1348 | { |
d6337a5f CB |
1349 | char *basecginfo; |
1350 | char *base_cgroup; | |
1351 | bool will_escape; | |
1352 | char *copy = NULL; | |
1353 | ||
1354 | will_escape = (geteuid() == 0); | |
1355 | if (will_escape) | |
1356 | basecginfo = read_file("/proc/1/cgroup"); | |
1357 | else | |
1358 | basecginfo = read_file("/proc/self/cgroup"); | |
1359 | if (!basecginfo) | |
1360 | return NULL; | |
1361 | ||
1362 | base_cgroup = strstr(basecginfo, "0::/"); | |
1363 | if (!base_cgroup) | |
1364 | goto cleanup_on_err; | |
1365 | ||
1366 | base_cgroup = base_cgroup + 3; | |
1367 | copy = copy_to_eol(base_cgroup); | |
1368 | if (!copy) | |
1369 | goto cleanup_on_err; | |
1370 | ||
1371 | cleanup_on_err: | |
1372 | free(basecginfo); | |
1373 | if (copy) | |
1374 | trim(copy); | |
1375 | ||
1376 | return copy; | |
1377 | } | |
1378 | ||
a3926f6a | 1379 | static int cg_unified_init(void) |
d6337a5f CB |
1380 | { |
1381 | int ret; | |
1382 | char *mountpoint, *subtree_path; | |
1383 | char **delegatable; | |
1384 | char *base_cgroup = NULL; | |
1385 | ||
1386 | ret = cg_is_pure_unified(); | |
1387 | if (ret == -ENOMEDIUM) | |
1388 | return -ENOMEDIUM; | |
1389 | ||
1390 | if (ret != CGROUP2_SUPER_MAGIC) | |
1391 | return 0; | |
1392 | ||
a3926f6a | 1393 | base_cgroup = cg_unified_get_current_cgroup(); |
d6337a5f CB |
1394 | if (!base_cgroup) |
1395 | return -EINVAL; | |
1396 | prune_init_scope(base_cgroup); | |
1397 | ||
1398 | /* We assume that we have already been given controllers to delegate | |
1399 | * further down the hierarchy. If not it is up to the user to delegate | |
1400 | * them to us. | |
1401 | */ | |
1402 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
1403 | subtree_path = must_make_path(mountpoint, base_cgroup, | |
1404 | "cgroup.subtree_control", NULL); | |
1405 | delegatable = cg_unified_get_controllers(subtree_path); | |
1406 | free(subtree_path); | |
1407 | if (!delegatable) | |
1408 | delegatable = cg_unified_make_empty_controller(); | |
1409 | if (!delegatable[0]) | |
1410 | CGFSNG_DEBUG("No controllers are enabled for delegation\n"); | |
1411 | ||
1412 | /* TODO: If the user requested specific controllers via lxc.cgroup.use | |
1413 | * we should verify here. The reason I'm not doing it right is that I'm | |
1414 | * not convinced that lxc.cgroup.use will be the future since it is a | |
1415 | * global property. I much rather have an option that lets you request | |
1416 | * controllers per container. | |
1417 | */ | |
1418 | ||
1419 | add_hierarchy(delegatable, mountpoint, base_cgroup, CGROUP2_SUPER_MAGIC); | |
1420 | unified = hierarchies[0]; | |
1421 | ||
1422 | cgroup_layout = CGROUP_LAYOUT_UNIFIED; | |
1423 | return CGROUP2_SUPER_MAGIC; | |
1424 | } | |
1425 | ||
1426 | static bool cg_init(void) | |
1427 | { | |
1428 | int ret; | |
457ca9aa | 1429 | const char *tmp; |
d6337a5f | 1430 | |
457ca9aa SH |
1431 | errno = 0; |
1432 | tmp = lxc_global_config_value("lxc.cgroup.use"); | |
1a0e70ac | 1433 | if (!cgroup_use && errno != 0) { /* lxc.cgroup.use can be NULL */ |
65d78313 | 1434 | CGFSNG_DEBUG("Failed to retrieve list of cgroups to use\n"); |
457ca9aa SH |
1435 | return false; |
1436 | } | |
1437 | cgroup_use = must_copy_string(tmp); | |
1438 | ||
a3926f6a | 1439 | ret = cg_unified_init(); |
d6337a5f CB |
1440 | if (ret < 0) |
1441 | return false; | |
1442 | ||
1443 | if (ret == CGROUP2_SUPER_MAGIC) | |
1444 | return true; | |
1445 | ||
a3926f6a | 1446 | return cg_hybrid_init(); |
457ca9aa SH |
1447 | } |
1448 | ||
43654d34 | 1449 | static void *cgfsng_init(struct lxc_handler *handler) |
ccb4cabe | 1450 | { |
457ca9aa | 1451 | const char *cgroup_pattern; |
43654d34 | 1452 | struct cgfsng_handler_data *d; |
ccb4cabe SH |
1453 | |
1454 | d = must_alloc(sizeof(*d)); | |
1455 | memset(d, 0, sizeof(*d)); | |
1456 | ||
43654d34 CB |
1457 | /* copy container name */ |
1458 | d->name = must_copy_string(handler->name); | |
1459 | ||
1460 | /* copy per-container cgroup information */ | |
ae5e6c08 CB |
1461 | d->cgroup_meta.dir = NULL; |
1462 | d->cgroup_meta.controllers = NULL; | |
9b5396f9 CB |
1463 | if (handler->conf) { |
1464 | d->cgroup_meta.dir = must_copy_string(handler->conf->cgroup_meta.dir); | |
1465 | d->cgroup_meta.controllers = must_copy_string(handler->conf->cgroup_meta.controllers); | |
1466 | } | |
ccb4cabe | 1467 | |
43654d34 | 1468 | /* copy system-wide cgroup information */ |
ccb4cabe | 1469 | cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern"); |
43654d34 CB |
1470 | if (!cgroup_pattern) { |
1471 | /* lxc.cgroup.pattern is only NULL on error. */ | |
ccb4cabe SH |
1472 | ERROR("Error getting cgroup pattern"); |
1473 | goto out_free; | |
1474 | } | |
1475 | d->cgroup_pattern = must_copy_string(cgroup_pattern); | |
1476 | ||
d6337a5f CB |
1477 | d->cgroup_layout = cgroup_layout; |
1478 | if (d->cgroup_layout == CGROUP_LAYOUT_LEGACY) | |
1479 | TRACE("Running with legacy cgroup layout"); | |
1480 | else if (d->cgroup_layout == CGROUP_LAYOUT_HYBRID) | |
1481 | TRACE("Running with hybrid cgroup layout"); | |
1482 | else if (d->cgroup_layout == CGROUP_LAYOUT_UNIFIED) | |
1483 | TRACE("Running with unified cgroup layout"); | |
1484 | else | |
1485 | WARN("Running with unknown cgroup layout"); | |
1486 | ||
e4aeecf5 CB |
1487 | if (lxc_cgfsng_debug) |
1488 | lxc_cgfsng_print_debuginfo(d); | |
ccb4cabe SH |
1489 | |
1490 | return d; | |
1491 | ||
1492 | out_free: | |
1493 | free_handler_data(d); | |
1494 | return NULL; | |
1495 | } | |
1496 | ||
bd8ef4e4 | 1497 | static int recursive_destroy(char *dirname) |
ccb4cabe | 1498 | { |
a17f8b3f | 1499 | int ret; |
74f96976 | 1500 | struct dirent *direntp; |
ccb4cabe SH |
1501 | DIR *dir; |
1502 | int r = 0; | |
1503 | ||
1504 | dir = opendir(dirname); | |
1505 | if (!dir) | |
1506 | return -1; | |
1507 | ||
74f96976 | 1508 | while ((direntp = readdir(dir))) { |
ccb4cabe | 1509 | char *pathname; |
a17f8b3f | 1510 | struct stat mystat; |
ccb4cabe | 1511 | |
ccb4cabe SH |
1512 | if (!strcmp(direntp->d_name, ".") || |
1513 | !strcmp(direntp->d_name, "..")) | |
1514 | continue; | |
1515 | ||
1516 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
1517 | ||
a17f8b3f CB |
1518 | ret = lstat(pathname, &mystat); |
1519 | if (ret < 0) { | |
ccb4cabe | 1520 | if (!r) |
a17f8b3f | 1521 | WARN("Failed to stat %s", pathname); |
ccb4cabe SH |
1522 | r = -1; |
1523 | goto next; | |
1524 | } | |
1525 | ||
1526 | if (!S_ISDIR(mystat.st_mode)) | |
1527 | goto next; | |
a17f8b3f | 1528 | |
bd8ef4e4 | 1529 | ret = recursive_destroy(pathname); |
a17f8b3f | 1530 | if (ret < 0) |
ccb4cabe | 1531 | r = -1; |
bd8ef4e4 | 1532 | next: |
ccb4cabe SH |
1533 | free(pathname); |
1534 | } | |
1535 | ||
a17f8b3f CB |
1536 | ret = rmdir(dirname); |
1537 | if (ret < 0) { | |
ccb4cabe | 1538 | if (!r) |
bd8ef4e4 CB |
1539 | WARN("%s - Failed to delete \"%s\"", strerror(errno), |
1540 | dirname); | |
ccb4cabe SH |
1541 | r = -1; |
1542 | } | |
1543 | ||
a17f8b3f CB |
1544 | ret = closedir(dir); |
1545 | if (ret < 0) { | |
ccb4cabe | 1546 | if (!r) |
bd8ef4e4 CB |
1547 | WARN("%s - Failed to delete \"%s\"", strerror(errno), |
1548 | dirname); | |
ccb4cabe SH |
1549 | r = -1; |
1550 | } | |
a17f8b3f | 1551 | |
ccb4cabe SH |
1552 | return r; |
1553 | } | |
1554 | ||
bd8ef4e4 CB |
1555 | static int cgroup_rmdir(char *container_cgroup) |
1556 | { | |
1557 | int i; | |
1558 | ||
1559 | if (!container_cgroup || !hierarchies) | |
1560 | return 0; | |
1561 | ||
1562 | for (i = 0; hierarchies[i]; i++) { | |
1563 | int ret; | |
1564 | struct hierarchy *h = hierarchies[i]; | |
1565 | ||
1566 | if (!h->fullcgpath) | |
1567 | continue; | |
1568 | ||
1569 | ret = recursive_destroy(h->fullcgpath); | |
1570 | if (ret < 0) | |
1571 | WARN("Failed to destroy \"%s\"", h->fullcgpath); | |
1572 | ||
1573 | free(h->fullcgpath); | |
1574 | h->fullcgpath = NULL; | |
1575 | } | |
1576 | ||
1577 | return 0; | |
1578 | } | |
1579 | ||
4160c3a0 CB |
1580 | struct generic_userns_exec_data { |
1581 | struct cgfsng_handler_data *d; | |
1582 | struct lxc_conf *conf; | |
1583 | uid_t origuid; /* target uid in parent namespace */ | |
1584 | char *path; | |
1585 | }; | |
1586 | ||
bd8ef4e4 | 1587 | static int cgroup_rmdir_wrapper(void *data) |
ccb4cabe | 1588 | { |
6efacf80 | 1589 | int ret; |
4160c3a0 CB |
1590 | struct generic_userns_exec_data *arg = data; |
1591 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1592 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
ccb4cabe | 1593 | |
6efacf80 CB |
1594 | ret = setresgid(nsgid, nsgid, nsgid); |
1595 | if (ret < 0) { | |
1596 | SYSERROR("Failed to setresgid(%d, %d, %d)", (int)nsgid, | |
1597 | (int)nsgid, (int)nsgid); | |
1598 | return -1; | |
1599 | } | |
1600 | ||
1601 | ret = setresuid(nsuid, nsuid, nsuid); | |
1602 | if (ret < 0) { | |
1603 | SYSERROR("Failed to setresuid(%d, %d, %d)", (int)nsuid, | |
1604 | (int)nsuid, (int)nsuid); | |
1605 | return -1; | |
1606 | } | |
1607 | ||
1608 | ret = setgroups(0, NULL); | |
1609 | if (ret < 0 && errno != EPERM) { | |
1610 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1611 | return -1; | |
1612 | } | |
ccb4cabe | 1613 | |
bd8ef4e4 | 1614 | return cgroup_rmdir(arg->d->container_cgroup); |
ccb4cabe SH |
1615 | } |
1616 | ||
bd8ef4e4 | 1617 | static void cgfsng_destroy(void *hdata, struct lxc_conf *conf) |
ccb4cabe | 1618 | { |
bd8ef4e4 CB |
1619 | int ret; |
1620 | struct cgfsng_handler_data *d = hdata; | |
4160c3a0 CB |
1621 | struct generic_userns_exec_data wrap; |
1622 | ||
bd8ef4e4 CB |
1623 | if (!d) |
1624 | return; | |
1625 | ||
4160c3a0 | 1626 | wrap.origuid = 0; |
bd8ef4e4 | 1627 | wrap.d = hdata; |
4160c3a0 CB |
1628 | wrap.conf = conf; |
1629 | ||
ccb4cabe | 1630 | if (conf && !lxc_list_empty(&conf->id_map)) |
bd8ef4e4 CB |
1631 | ret = userns_exec_1(conf, cgroup_rmdir_wrapper, &wrap, |
1632 | "cgroup_rmdir_wrapper"); | |
ccb4cabe | 1633 | else |
bd8ef4e4 CB |
1634 | ret = cgroup_rmdir(d->container_cgroup); |
1635 | if (ret < 0) { | |
1636 | WARN("Failed to destroy cgroups"); | |
ccb4cabe | 1637 | return; |
ccb4cabe SH |
1638 | } |
1639 | ||
1640 | free_handler_data(d); | |
1641 | } | |
1642 | ||
1643 | struct cgroup_ops *cgfsng_ops_init(void) | |
1644 | { | |
e4aeecf5 CB |
1645 | if (getenv("LXC_DEBUG_CGFSNG")) |
1646 | lxc_cgfsng_debug = true; | |
1647 | ||
d6337a5f | 1648 | if (!cg_init()) |
457ca9aa | 1649 | return NULL; |
e4aeecf5 | 1650 | |
ccb4cabe SH |
1651 | return &cgfsng_ops; |
1652 | } | |
1653 | ||
a3926f6a | 1654 | static bool cg_unified_create_cgroup(struct hierarchy *h, char *cgname) |
0c3deb94 CB |
1655 | { |
1656 | char **it; | |
1657 | size_t i, parts_len; | |
1658 | size_t full_len = 0; | |
1659 | char *add_controllers = NULL, *cgroup = NULL; | |
1660 | char **parts = NULL; | |
1661 | bool bret = false; | |
1662 | ||
1663 | if (h->version != CGROUP2_SUPER_MAGIC) | |
1664 | return true; | |
1665 | ||
1666 | if (!h->controllers) | |
1667 | return true; | |
1668 | ||
1669 | /* For now we simply enable all controllers that we have detected by | |
1670 | * creating a string like "+memory +pids +cpu +io". | |
1671 | * TODO: In the near future we might want to support "-<controller>" | |
1672 | * etc. but whether supporting semantics like this make sense will need | |
1673 | * some thinking. | |
1674 | */ | |
1675 | for (it = h->controllers; it && *it; it++) { | |
1676 | full_len += strlen(*it) + 2; | |
1677 | add_controllers = must_realloc(add_controllers, full_len + 1); | |
1678 | if (h->controllers[0] == *it) | |
1679 | add_controllers[0] = '\0'; | |
1680 | strcat(add_controllers, "+"); | |
1681 | strcat(add_controllers, *it); | |
1682 | if ((it + 1) && *(it + 1)) | |
1683 | strcat(add_controllers, " "); | |
1684 | } | |
1685 | ||
1686 | parts = lxc_string_split(cgname, '/'); | |
1687 | if (!parts) | |
1688 | goto on_error; | |
1689 | parts_len = lxc_array_len((void **)parts); | |
1690 | if (parts_len > 0) | |
1691 | parts_len--; | |
1692 | ||
1693 | cgroup = must_make_path(h->mountpoint, h->base_cgroup, NULL); | |
1694 | for (i = 0; i < parts_len; i++) { | |
1695 | int ret; | |
1696 | char *target; | |
1697 | ||
1698 | cgroup = must_append_path(cgroup, parts[i], NULL); | |
1699 | target = must_make_path(cgroup, "cgroup.subtree_control", NULL); | |
1700 | ret = lxc_write_to_file(target, add_controllers, full_len, false); | |
1701 | free(target); | |
1702 | if (ret < 0) { | |
1703 | SYSERROR("Could not enable \"%s\" controllers in the " | |
1704 | "unified cgroup \"%s\"", add_controllers, cgroup); | |
1705 | goto on_error; | |
1706 | } | |
1707 | } | |
1708 | ||
1709 | bret = true; | |
1710 | ||
1711 | on_error: | |
1712 | lxc_free_array((void **)parts, free); | |
1713 | free(add_controllers); | |
1714 | free(cgroup); | |
1715 | return bret; | |
1716 | } | |
1717 | ||
ccb4cabe SH |
1718 | static bool create_path_for_hierarchy(struct hierarchy *h, char *cgname) |
1719 | { | |
0c3deb94 CB |
1720 | int ret; |
1721 | ||
e3a3fecf | 1722 | h->fullcgpath = must_make_path(h->mountpoint, h->base_cgroup, cgname, NULL); |
1a0e70ac | 1723 | if (dir_exists(h->fullcgpath)) { /* it must not already exist */ |
0c3deb94 | 1724 | ERROR("cgroup \"%s\" already existed", h->fullcgpath); |
d8da679e | 1725 | return false; |
6f9584d8 | 1726 | } |
0c3deb94 | 1727 | |
a3926f6a | 1728 | if (!cg_legacy_handle_cpuset_hierarchy(h, cgname)) { |
0c3deb94 CB |
1729 | ERROR("Failed to handle cgroupfs v1 cpuset controller"); |
1730 | return false; | |
1731 | } | |
1732 | ||
1733 | ret = mkdir_p(h->fullcgpath, 0755); | |
1734 | if (ret < 0) { | |
1735 | ERROR("Failed to create cgroup \"%s\"", h->fullcgpath); | |
e3a3fecf | 1736 | return false; |
6f9584d8 | 1737 | } |
0c3deb94 | 1738 | |
a3926f6a | 1739 | return cg_unified_create_cgroup(h, cgname); |
ccb4cabe SH |
1740 | } |
1741 | ||
1742 | static void remove_path_for_hierarchy(struct hierarchy *h, char *cgname) | |
1743 | { | |
1744 | if (rmdir(h->fullcgpath) < 0) | |
1745 | SYSERROR("Failed to clean up cgroup %s from failed creation attempt", h->fullcgpath); | |
1746 | free(h->fullcgpath); | |
1747 | h->fullcgpath = NULL; | |
1748 | } | |
1749 | ||
1750 | /* | |
d30ec4cb | 1751 | * Try to create the same cgroup in all hierarchies. |
ccb4cabe SH |
1752 | * Start with cgroup_pattern; next cgroup_pattern-1, -2, ..., -999 |
1753 | */ | |
1754 | static inline bool cgfsng_create(void *hdata) | |
1755 | { | |
bb30b52a | 1756 | int i; |
ccb4cabe | 1757 | size_t len; |
0c3deb94 | 1758 | char *container_cgroup, *offset, *tmp; |
7d531e9b CB |
1759 | int idx = 0; |
1760 | struct cgfsng_handler_data *d = hdata; | |
ccb4cabe SH |
1761 | |
1762 | if (!d) | |
1763 | return false; | |
43654d34 | 1764 | |
ccb4cabe SH |
1765 | if (d->container_cgroup) { |
1766 | WARN("cgfsng_create called a second time"); | |
1767 | return false; | |
1768 | } | |
1769 | ||
43654d34 | 1770 | if (d->cgroup_meta.dir) |
7d531e9b | 1771 | tmp = lxc_string_join("/", (const char *[]){d->cgroup_meta.dir, d->name, NULL}, false); |
43654d34 CB |
1772 | else |
1773 | tmp = lxc_string_replace("%n", d->name, d->cgroup_pattern); | |
ccb4cabe SH |
1774 | if (!tmp) { |
1775 | ERROR("Failed expanding cgroup name pattern"); | |
1776 | return false; | |
1777 | } | |
1a0e70ac | 1778 | len = strlen(tmp) + 5; /* leave room for -NNN\0 */ |
0c3deb94 CB |
1779 | container_cgroup = must_alloc(len); |
1780 | strcpy(container_cgroup, tmp); | |
ccb4cabe | 1781 | free(tmp); |
0c3deb94 | 1782 | offset = container_cgroup + len - 5; |
ccb4cabe SH |
1783 | |
1784 | again: | |
95adfe93 SH |
1785 | if (idx == 1000) { |
1786 | ERROR("Too many conflicting cgroup names"); | |
ccb4cabe | 1787 | goto out_free; |
95adfe93 | 1788 | } |
66b66624 | 1789 | if (idx) { |
bb30b52a CB |
1790 | int ret; |
1791 | ||
66b66624 CB |
1792 | ret = snprintf(offset, 5, "-%d", idx); |
1793 | if (ret < 0 || (size_t)ret >= 5) { | |
1794 | FILE *f = fopen("/dev/null", "w"); | |
97ebced3 | 1795 | if (f) { |
66b66624 CB |
1796 | fprintf(f, "Workaround for GCC7 bug: " |
1797 | "https://gcc.gnu.org/bugzilla/" | |
1798 | "show_bug.cgi?id=78969"); | |
1799 | fclose(f); | |
1800 | } | |
1801 | } | |
1802 | } | |
457ca9aa | 1803 | for (i = 0; hierarchies[i]; i++) { |
0c3deb94 | 1804 | if (!create_path_for_hierarchy(hierarchies[i], container_cgroup)) { |
ccb4cabe | 1805 | int j; |
1a0e70ac | 1806 | ERROR("Failed to create \"%s\"", hierarchies[i]->fullcgpath); |
457ca9aa SH |
1807 | free(hierarchies[i]->fullcgpath); |
1808 | hierarchies[i]->fullcgpath = NULL; | |
ccb4cabe | 1809 | for (j = 0; j < i; j++) |
0c3deb94 | 1810 | remove_path_for_hierarchy(hierarchies[j], container_cgroup); |
ccb4cabe SH |
1811 | idx++; |
1812 | goto again; | |
1813 | } | |
1814 | } | |
1815 | /* Done */ | |
0c3deb94 | 1816 | d->container_cgroup = container_cgroup; |
ccb4cabe SH |
1817 | return true; |
1818 | ||
1819 | out_free: | |
0c3deb94 | 1820 | free(container_cgroup); |
ccb4cabe SH |
1821 | return false; |
1822 | } | |
1823 | ||
ccb4cabe SH |
1824 | static bool cgfsng_enter(void *hdata, pid_t pid) |
1825 | { | |
ccb4cabe SH |
1826 | char pidstr[25]; |
1827 | int i, len; | |
1828 | ||
1829 | len = snprintf(pidstr, 25, "%d", pid); | |
1830 | if (len < 0 || len > 25) | |
1831 | return false; | |
1832 | ||
457ca9aa SH |
1833 | for (i = 0; hierarchies[i]; i++) { |
1834 | char *fullpath = must_make_path(hierarchies[i]->fullcgpath, | |
ccb4cabe SH |
1835 | "cgroup.procs", NULL); |
1836 | if (lxc_write_to_file(fullpath, pidstr, len, false) != 0) { | |
d3b00a8f | 1837 | SYSERROR("Failed to enter %s", fullpath); |
ccb4cabe SH |
1838 | free(fullpath); |
1839 | return false; | |
1840 | } | |
1841 | free(fullpath); | |
1842 | } | |
1843 | ||
1844 | return true; | |
1845 | } | |
1846 | ||
6efacf80 CB |
1847 | static int chowmod(char *path, uid_t chown_uid, gid_t chown_gid, |
1848 | mode_t chmod_mode) | |
1849 | { | |
1850 | int ret; | |
1851 | ||
1852 | ret = chown(path, chown_uid, chown_gid); | |
1853 | if (ret < 0) { | |
1854 | WARN("%s - Failed to chown(%s, %d, %d)", strerror(errno), path, | |
1855 | (int)chown_uid, (int)chown_gid); | |
1856 | return -1; | |
1857 | } | |
1858 | ||
1859 | ret = chmod(path, chmod_mode); | |
1860 | if (ret < 0) { | |
1861 | WARN("%s - Failed to chmod(%s, %d)", strerror(errno), path, | |
1862 | (int)chmod_mode); | |
1863 | return -1; | |
1864 | } | |
1865 | ||
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | /* chgrp the container cgroups to container group. We leave | |
c0888dfe SH |
1870 | * the container owner as cgroup owner. So we must make the |
1871 | * directories 775 so that the container can create sub-cgroups. | |
43647298 SH |
1872 | * |
1873 | * Also chown the tasks and cgroup.procs files. Those may not | |
1874 | * exist depending on kernel version. | |
c0888dfe | 1875 | */ |
ccb4cabe SH |
1876 | static int chown_cgroup_wrapper(void *data) |
1877 | { | |
6efacf80 | 1878 | int i, ret; |
4160c3a0 CB |
1879 | uid_t destuid; |
1880 | struct generic_userns_exec_data *arg = data; | |
1881 | uid_t nsuid = (arg->conf->root_nsuid_map != NULL) ? 0 : arg->conf->init_uid; | |
1882 | gid_t nsgid = (arg->conf->root_nsgid_map != NULL) ? 0 : arg->conf->init_gid; | |
ccb4cabe | 1883 | |
6efacf80 CB |
1884 | ret = setresgid(nsgid, nsgid, nsgid); |
1885 | if (ret < 0) { | |
1886 | SYSERROR("Failed to setresgid(%d, %d, %d)", | |
1887 | (int)nsgid, (int)nsgid, (int)nsgid); | |
1888 | return -1; | |
1889 | } | |
1890 | ||
1891 | ret = setresuid(nsuid, nsuid, nsuid); | |
1892 | if (ret < 0) { | |
1893 | SYSERROR("Failed to setresuid(%d, %d, %d)", | |
1894 | (int)nsuid, (int)nsuid, (int)nsuid); | |
1895 | return -1; | |
1896 | } | |
1897 | ||
1898 | ret = setgroups(0, NULL); | |
1899 | if (ret < 0 && errno != EPERM) { | |
1900 | SYSERROR("Failed to setgroups(0, NULL)"); | |
1901 | return -1; | |
1902 | } | |
ccb4cabe SH |
1903 | |
1904 | destuid = get_ns_uid(arg->origuid); | |
1905 | ||
457ca9aa | 1906 | for (i = 0; hierarchies[i]; i++) { |
6efacf80 CB |
1907 | char *fullpath; |
1908 | char *path = hierarchies[i]->fullcgpath; | |
43647298 | 1909 | |
63e42fee | 1910 | ret = chowmod(path, destuid, nsgid, 0775); |
6efacf80 | 1911 | if (ret < 0) |
ccb4cabe | 1912 | return -1; |
c0888dfe | 1913 | |
6efacf80 CB |
1914 | /* Failures to chown() these are inconvenient but not |
1915 | * detrimental We leave these owned by the container launcher, | |
1916 | * so that container root can write to the files to attach. We | |
1917 | * chmod() them 664 so that container systemd can write to the | |
1918 | * files (which systemd in wily insists on doing). | |
ab8f5424 | 1919 | */ |
6efacf80 CB |
1920 | |
1921 | if (hierarchies[i]->version == CGROUP_SUPER_MAGIC) { | |
1922 | fullpath = must_make_path(path, "tasks", NULL); | |
1923 | (void)chowmod(fullpath, destuid, nsgid, 0664); | |
1924 | free(fullpath); | |
1925 | } | |
43647298 SH |
1926 | |
1927 | fullpath = must_make_path(path, "cgroup.procs", NULL); | |
6efacf80 | 1928 | (void)chowmod(fullpath, destuid, 0, 0664); |
ccb4cabe | 1929 | free(fullpath); |
0e17357c | 1930 | |
d6337a5f | 1931 | if (hierarchies[i]->version != CGROUP2_SUPER_MAGIC) |
0e17357c CB |
1932 | continue; |
1933 | ||
1934 | fullpath = must_make_path(path, "cgroup.subtree_control", NULL); | |
6efacf80 | 1935 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c CB |
1936 | free(fullpath); |
1937 | ||
1938 | fullpath = must_make_path(path, "cgroup.threads", NULL); | |
6efacf80 | 1939 | (void)chowmod(fullpath, destuid, nsgid, 0664); |
0e17357c | 1940 | free(fullpath); |
ccb4cabe SH |
1941 | } |
1942 | ||
1943 | return 0; | |
1944 | } | |
1945 | ||
058c1cb6 | 1946 | static bool cgfsng_chown(void *hdata, struct lxc_conf *conf) |
ccb4cabe SH |
1947 | { |
1948 | struct cgfsng_handler_data *d = hdata; | |
4160c3a0 | 1949 | struct generic_userns_exec_data wrap; |
ccb4cabe SH |
1950 | |
1951 | if (!d) | |
1952 | return false; | |
1953 | ||
1954 | if (lxc_list_empty(&conf->id_map)) | |
1955 | return true; | |
1956 | ||
ccb4cabe | 1957 | wrap.origuid = geteuid(); |
4160c3a0 CB |
1958 | wrap.path = NULL; |
1959 | wrap.d = d; | |
1960 | wrap.conf = conf; | |
ccb4cabe | 1961 | |
c9b7c33e CB |
1962 | if (userns_exec_1(conf, chown_cgroup_wrapper, &wrap, |
1963 | "chown_cgroup_wrapper") < 0) { | |
ccb4cabe SH |
1964 | ERROR("Error requesting cgroup chown in new namespace"); |
1965 | return false; | |
1966 | } | |
1967 | ||
1968 | return true; | |
1969 | } | |
1970 | ||
8aa1044f SH |
1971 | /* |
1972 | * We've safe-mounted a tmpfs as parent, so we don't need to protect against | |
1973 | * symlinks any more - just use mount | |
1974 | */ | |
1975 | ||
1976 | /* mount cgroup-full if requested */ | |
1977 | static int mount_cgroup_full(int type, struct hierarchy *h, char *dest, | |
a3926f6a | 1978 | char *container_cgroup) |
8aa1044f SH |
1979 | { |
1980 | if (type < LXC_AUTO_CGROUP_FULL_RO || type > LXC_AUTO_CGROUP_FULL_MIXED) | |
1981 | return 0; | |
1982 | if (mount(h->mountpoint, dest, "cgroup", MS_BIND, NULL) < 0) { | |
1983 | SYSERROR("Error bind-mounting %s cgroup onto %s", h->mountpoint, | |
1984 | dest); | |
1985 | return -1; | |
1986 | } | |
1987 | if (type != LXC_AUTO_CGROUP_FULL_RW) { | |
5b6f9369 SH |
1988 | unsigned long flags = MS_BIND | MS_NOSUID | MS_NOEXEC | MS_NODEV | |
1989 | MS_REMOUNT | MS_RDONLY; | |
1990 | if (mount(NULL, dest, "cgroup", flags, NULL) < 0) { | |
8aa1044f SH |
1991 | SYSERROR("Error remounting %s readonly", dest); |
1992 | return -1; | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | INFO("Bind mounted %s onto %s", h->mountpoint, dest); | |
1997 | if (type != LXC_AUTO_CGROUP_FULL_MIXED) | |
1998 | return 0; | |
1999 | ||
2000 | /* mount just the container path rw */ | |
2001 | char *source = must_make_path(h->mountpoint, h->base_cgroup, container_cgroup, NULL); | |
5b6f9369 | 2002 | char *rwpath = must_make_path(dest, h->base_cgroup, container_cgroup, NULL); |
8aa1044f | 2003 | if (mount(source, rwpath, "cgroup", MS_BIND, NULL) < 0) |
13277ec4 | 2004 | WARN("Failed to mount %s read-write: %s", rwpath, |
2005 | strerror(errno)); | |
8aa1044f SH |
2006 | INFO("Made %s read-write", rwpath); |
2007 | free(rwpath); | |
2008 | free(source); | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | /* cgroup-full:* is done, no need to create subdirs */ | |
2013 | static bool cg_mount_needs_subdirs(int type) | |
2014 | { | |
2015 | if (type >= LXC_AUTO_CGROUP_FULL_RO) | |
2016 | return false; | |
a3926f6a | 2017 | |
8aa1044f SH |
2018 | return true; |
2019 | } | |
2020 | ||
886cac86 CB |
2021 | /* After $rootfs/sys/fs/container/controller/the/cg/path has been created, |
2022 | * remount controller ro if needed and bindmount the cgroupfs onto | |
2023 | * controll/the/cg/path. | |
8aa1044f | 2024 | */ |
a3926f6a CB |
2025 | static int do_secondstage_mounts_if_needed(int type, struct hierarchy *h, |
2026 | char *controllerpath, char *cgpath, | |
2027 | const char *container_cgroup) | |
8aa1044f | 2028 | { |
5285689c | 2029 | int ret, remount_flags; |
886cac86 CB |
2030 | char *sourcepath; |
2031 | int flags = MS_BIND; | |
2032 | ||
8aa1044f | 2033 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_MIXED) { |
886cac86 CB |
2034 | ret = mount(controllerpath, controllerpath, "cgroup", MS_BIND, NULL); |
2035 | if (ret < 0) { | |
2036 | SYSERROR("Failed to bind mount \"%s\" onto \"%s\"", | |
2037 | controllerpath, controllerpath); | |
8aa1044f SH |
2038 | return -1; |
2039 | } | |
886cac86 | 2040 | |
5285689c CB |
2041 | remount_flags = add_required_remount_flags(controllerpath, |
2042 | controllerpath, | |
2043 | flags | MS_REMOUNT); | |
886cac86 CB |
2044 | ret = mount(controllerpath, controllerpath, "cgroup", |
2045 | MS_REMOUNT | MS_BIND | MS_RDONLY, NULL); | |
2046 | if (ret < 0) { | |
2047 | SYSERROR("Failed to remount \"%s\" ro", controllerpath); | |
8aa1044f SH |
2048 | return -1; |
2049 | } | |
886cac86 | 2050 | |
8aa1044f SH |
2051 | INFO("Remounted %s read-only", controllerpath); |
2052 | } | |
886cac86 CB |
2053 | |
2054 | sourcepath = must_make_path(h->mountpoint, h->base_cgroup, | |
2055 | container_cgroup, NULL); | |
8aa1044f SH |
2056 | if (type == LXC_AUTO_CGROUP_RO) |
2057 | flags |= MS_RDONLY; | |
886cac86 CB |
2058 | |
2059 | ret = mount(sourcepath, cgpath, "cgroup", flags, NULL); | |
2060 | if (ret < 0) { | |
2061 | SYSERROR("Failed to mount \"%s\" onto \"%s\"", h->controllers[0], cgpath); | |
8aa1044f | 2062 | free(sourcepath); |
8aa1044f SH |
2063 | return -1; |
2064 | } | |
886cac86 | 2065 | INFO("Mounted \"%s\" onto \"%s\"", h->controllers[0], cgpath); |
f8c40ffa L |
2066 | |
2067 | if (flags & MS_RDONLY) { | |
5285689c CB |
2068 | remount_flags = add_required_remount_flags(sourcepath, cgpath, |
2069 | flags | MS_REMOUNT); | |
2070 | ret = mount(sourcepath, cgpath, "cgroup", remount_flags, NULL); | |
886cac86 CB |
2071 | if (ret < 0) { |
2072 | SYSERROR("Failed to remount \"%s\" ro", cgpath); | |
f8c40ffa | 2073 | free(sourcepath); |
f8c40ffa L |
2074 | return -1; |
2075 | } | |
5285689c | 2076 | INFO("Remounted %s read-only", cgpath); |
f8c40ffa L |
2077 | } |
2078 | ||
8aa1044f | 2079 | free(sourcepath); |
886cac86 | 2080 | INFO("Completed second stage cgroup automounts for \"%s\"", cgpath); |
8aa1044f SH |
2081 | return 0; |
2082 | } | |
2083 | ||
5285689c CB |
2084 | static int cg_mount_in_cgroup_namespace(int type, struct hierarchy *h, |
2085 | const char *controllerpath) | |
b635e92d CB |
2086 | { |
2087 | int ret; | |
2088 | char *controllers = NULL; | |
a760603e CB |
2089 | char *fstype = "cgroup2"; |
2090 | unsigned long flags = 0; | |
b635e92d | 2091 | |
a760603e CB |
2092 | flags |= MS_NOSUID; |
2093 | flags |= MS_NOEXEC; | |
2094 | flags |= MS_NODEV; | |
2095 | flags |= MS_RELATIME; | |
2096 | ||
2097 | if (type == LXC_AUTO_CGROUP_RO || type == LXC_AUTO_CGROUP_FULL_RO) | |
2098 | flags |= MS_RDONLY; | |
2099 | ||
d6337a5f | 2100 | if (h->version != CGROUP2_SUPER_MAGIC) { |
a760603e CB |
2101 | controllers = lxc_string_join(",", (const char **)h->controllers, false); |
2102 | if (!controllers) | |
2103 | return -ENOMEM; | |
2104 | fstype = "cgroup"; | |
b635e92d CB |
2105 | } |
2106 | ||
a760603e | 2107 | ret = mount("cgroup", controllerpath, fstype, flags, controllers); |
b635e92d CB |
2108 | free(controllers); |
2109 | if (ret < 0) { | |
a760603e | 2110 | SYSERROR("Failed to mount %s with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
2111 | return -1; |
2112 | } | |
2113 | ||
a760603e | 2114 | DEBUG("Mounted %s with cgroup filesystem type %s", controllerpath, fstype); |
b635e92d CB |
2115 | return 0; |
2116 | } | |
2117 | ||
ccb4cabe SH |
2118 | static bool cgfsng_mount(void *hdata, const char *root, int type) |
2119 | { | |
3f69fb12 | 2120 | int i, ret; |
8aa1044f SH |
2121 | char *tmpfspath = NULL; |
2122 | bool retval = false; | |
b635e92d CB |
2123 | struct lxc_handler *handler = hdata; |
2124 | struct cgfsng_handler_data *d = handler->cgroup_data; | |
3f69fb12 | 2125 | bool has_cgns = false, wants_force_mount = false; |
8aa1044f SH |
2126 | |
2127 | if ((type & LXC_AUTO_CGROUP_MASK) == 0) | |
2128 | return true; | |
2129 | ||
3f69fb12 SY |
2130 | if (type & LXC_AUTO_CGROUP_FORCE) { |
2131 | type &= ~LXC_AUTO_CGROUP_FORCE; | |
2132 | wants_force_mount = true; | |
2133 | } | |
b635e92d | 2134 | |
3f69fb12 SY |
2135 | if (!wants_force_mount){ |
2136 | if (!lxc_list_empty(&handler->conf->keepcaps)) | |
2137 | wants_force_mount = !in_caplist(CAP_SYS_ADMIN, &handler->conf->keepcaps); | |
2138 | else | |
2139 | wants_force_mount = in_caplist(CAP_SYS_ADMIN, &handler->conf->caps); | |
2140 | } | |
8aa1044f | 2141 | |
3f69fb12 SY |
2142 | has_cgns = cgns_supported(); |
2143 | if (has_cgns && !wants_force_mount) | |
2144 | return true; | |
8aa1044f SH |
2145 | |
2146 | if (type == LXC_AUTO_CGROUP_NOSPEC) | |
2147 | type = LXC_AUTO_CGROUP_MIXED; | |
2148 | else if (type == LXC_AUTO_CGROUP_FULL_NOSPEC) | |
2149 | type = LXC_AUTO_CGROUP_FULL_MIXED; | |
2150 | ||
2151 | /* Mount tmpfs */ | |
3f69fb12 SY |
2152 | tmpfspath = must_make_path(root, "/sys/fs/cgroup", NULL); |
2153 | ret = safe_mount("cgroup_root", tmpfspath, "tmpfs", | |
2154 | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_RELATIME, | |
2155 | "size=10240k,mode=755", root); | |
2156 | if (ret < 0) | |
2157 | goto on_error; | |
8aa1044f | 2158 | |
457ca9aa | 2159 | for (i = 0; hierarchies[i]; i++) { |
8aa1044f | 2160 | char *controllerpath, *path2; |
457ca9aa | 2161 | struct hierarchy *h = hierarchies[i]; |
8aa1044f | 2162 | char *controller = strrchr(h->mountpoint, '/'); |
8aa1044f SH |
2163 | |
2164 | if (!controller) | |
2165 | continue; | |
2166 | controller++; | |
2167 | controllerpath = must_make_path(tmpfspath, controller, NULL); | |
2168 | if (dir_exists(controllerpath)) { | |
2169 | free(controllerpath); | |
2170 | continue; | |
2171 | } | |
3f69fb12 SY |
2172 | ret = mkdir(controllerpath, 0755); |
2173 | if (ret < 0) { | |
8aa1044f SH |
2174 | SYSERROR("Error creating cgroup path: %s", controllerpath); |
2175 | free(controllerpath); | |
3f69fb12 | 2176 | goto on_error; |
8aa1044f | 2177 | } |
b635e92d | 2178 | |
3f69fb12 | 2179 | if (has_cgns && wants_force_mount) { |
b635e92d CB |
2180 | /* If cgroup namespaces are supported but the container |
2181 | * will not have CAP_SYS_ADMIN after it has started we | |
2182 | * need to mount the cgroups manually. | |
2183 | */ | |
3f69fb12 | 2184 | ret = cg_mount_in_cgroup_namespace(type, h, controllerpath); |
b635e92d | 2185 | free(controllerpath); |
3f69fb12 SY |
2186 | if (ret < 0) |
2187 | goto on_error; | |
2188 | ||
b635e92d CB |
2189 | continue; |
2190 | } | |
2191 | ||
3f69fb12 SY |
2192 | ret = mount_cgroup_full(type, h, controllerpath, d->container_cgroup); |
2193 | if (ret < 0) { | |
8aa1044f | 2194 | free(controllerpath); |
3f69fb12 | 2195 | goto on_error; |
8aa1044f | 2196 | } |
3f69fb12 | 2197 | |
8aa1044f SH |
2198 | if (!cg_mount_needs_subdirs(type)) { |
2199 | free(controllerpath); | |
2200 | continue; | |
2201 | } | |
3f69fb12 SY |
2202 | |
2203 | path2 = must_make_path(controllerpath, h->base_cgroup, | |
2204 | d->container_cgroup, NULL); | |
2205 | ret = mkdir_p(path2, 0755); | |
2206 | if (ret < 0) { | |
8aa1044f | 2207 | free(controllerpath); |
8e0c6620 | 2208 | free(path2); |
3f69fb12 | 2209 | goto on_error; |
8aa1044f | 2210 | } |
2f62fb00 | 2211 | |
3f69fb12 SY |
2212 | ret = do_secondstage_mounts_if_needed( |
2213 | type, h, controllerpath, path2, d->container_cgroup); | |
8aa1044f SH |
2214 | free(controllerpath); |
2215 | free(path2); | |
3f69fb12 SY |
2216 | if (ret < 0) |
2217 | goto on_error; | |
8aa1044f SH |
2218 | } |
2219 | retval = true; | |
2220 | ||
3f69fb12 | 2221 | on_error: |
8aa1044f SH |
2222 | free(tmpfspath); |
2223 | return retval; | |
ccb4cabe SH |
2224 | } |
2225 | ||
2226 | static int recursive_count_nrtasks(char *dirname) | |
2227 | { | |
74f96976 | 2228 | struct dirent *direntp; |
ccb4cabe SH |
2229 | DIR *dir; |
2230 | int count = 0, ret; | |
2231 | char *path; | |
2232 | ||
2233 | dir = opendir(dirname); | |
2234 | if (!dir) | |
2235 | return 0; | |
2236 | ||
74f96976 | 2237 | while ((direntp = readdir(dir))) { |
ccb4cabe SH |
2238 | struct stat mystat; |
2239 | ||
2240 | if (!direntp) | |
2241 | break; | |
2242 | ||
2243 | if (!strcmp(direntp->d_name, ".") || | |
2244 | !strcmp(direntp->d_name, "..")) | |
2245 | continue; | |
2246 | ||
2247 | path = must_make_path(dirname, direntp->d_name, NULL); | |
2248 | ||
2249 | if (lstat(path, &mystat)) | |
2250 | goto next; | |
2251 | ||
2252 | if (!S_ISDIR(mystat.st_mode)) | |
2253 | goto next; | |
2254 | ||
2255 | count += recursive_count_nrtasks(path); | |
2256 | next: | |
2257 | free(path); | |
2258 | } | |
2259 | ||
2260 | path = must_make_path(dirname, "cgroup.procs", NULL); | |
2261 | ret = lxc_count_file_lines(path); | |
2262 | if (ret != -1) | |
2263 | count += ret; | |
2264 | free(path); | |
2265 | ||
2266 | (void) closedir(dir); | |
2267 | ||
2268 | return count; | |
2269 | } | |
2270 | ||
2271 | static int cgfsng_nrtasks(void *hdata) { | |
2272 | struct cgfsng_handler_data *d = hdata; | |
2273 | char *path; | |
2274 | int count; | |
2275 | ||
457ca9aa | 2276 | if (!d || !d->container_cgroup || !hierarchies) |
ccb4cabe | 2277 | return -1; |
a3926f6a | 2278 | |
457ca9aa | 2279 | path = must_make_path(hierarchies[0]->fullcgpath, NULL); |
ccb4cabe SH |
2280 | count = recursive_count_nrtasks(path); |
2281 | free(path); | |
2282 | return count; | |
2283 | } | |
2284 | ||
2285 | /* Only root needs to escape to the cgroup of its init */ | |
7103fe6f | 2286 | static bool cgfsng_escape() |
ccb4cabe | 2287 | { |
ccb4cabe SH |
2288 | int i; |
2289 | ||
2290 | if (geteuid()) | |
2291 | return true; | |
2292 | ||
457ca9aa SH |
2293 | for (i = 0; hierarchies[i]; i++) { |
2294 | char *fullpath = must_make_path(hierarchies[i]->mountpoint, | |
2295 | hierarchies[i]->base_cgroup, | |
ccb4cabe SH |
2296 | "cgroup.procs", NULL); |
2297 | if (lxc_write_to_file(fullpath, "0", 2, false) != 0) { | |
d3b00a8f | 2298 | SYSERROR("Failed to escape to %s", fullpath); |
ccb4cabe | 2299 | free(fullpath); |
6df334d1 | 2300 | return false; |
ccb4cabe SH |
2301 | } |
2302 | free(fullpath); | |
2303 | } | |
2304 | ||
6df334d1 | 2305 | return true; |
ccb4cabe SH |
2306 | } |
2307 | ||
36662416 TA |
2308 | static int cgfsng_num_hierarchies(void) |
2309 | { | |
2310 | int i; | |
2311 | ||
2312 | for (i = 0; hierarchies[i]; i++) | |
2313 | ; | |
2314 | ||
2315 | return i; | |
2316 | } | |
2317 | ||
2318 | static bool cgfsng_get_hierarchies(int n, char ***out) | |
2319 | { | |
2320 | int i; | |
2321 | ||
2322 | /* sanity check n */ | |
6b38e644 | 2323 | for (i = 0; i < n; i++) |
36662416 TA |
2324 | if (!hierarchies[i]) |
2325 | return false; | |
36662416 TA |
2326 | |
2327 | *out = hierarchies[i]->controllers; | |
2328 | ||
2329 | return true; | |
2330 | } | |
2331 | ||
ccb4cabe SH |
2332 | #define THAWED "THAWED" |
2333 | #define THAWED_LEN (strlen(THAWED)) | |
2334 | ||
d6337a5f CB |
2335 | /* TODO: If the unified cgroup hierarchy grows a freezer controller this needs |
2336 | * to be adapted. | |
2337 | */ | |
ccb4cabe SH |
2338 | static bool cgfsng_unfreeze(void *hdata) |
2339 | { | |
d6337a5f | 2340 | int ret; |
ccb4cabe | 2341 | char *fullpath; |
d6337a5f | 2342 | struct hierarchy *h; |
ccb4cabe | 2343 | |
d6337a5f | 2344 | h = get_hierarchy("freezer"); |
457ca9aa | 2345 | if (!h) |
ccb4cabe | 2346 | return false; |
d6337a5f | 2347 | |
ccb4cabe | 2348 | fullpath = must_make_path(h->fullcgpath, "freezer.state", NULL); |
d6337a5f | 2349 | ret = lxc_write_to_file(fullpath, THAWED, THAWED_LEN, false); |
ccb4cabe | 2350 | free(fullpath); |
d6337a5f CB |
2351 | if (ret < 0) |
2352 | return false; | |
2353 | ||
ccb4cabe SH |
2354 | return true; |
2355 | } | |
2356 | ||
2357 | static const char *cgfsng_get_cgroup(void *hdata, const char *subsystem) | |
2358 | { | |
d6337a5f CB |
2359 | struct hierarchy *h; |
2360 | ||
2361 | h = get_hierarchy(subsystem); | |
ccb4cabe SH |
2362 | if (!h) |
2363 | return NULL; | |
2364 | ||
371f834d SH |
2365 | return h->fullcgpath ? h->fullcgpath + strlen(h->mountpoint) : NULL; |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * Given a cgroup path returned from lxc_cmd_get_cgroup_path, build a | |
2370 | * full path, which must be freed by the caller. | |
2371 | */ | |
2372 | static char *build_full_cgpath_from_monitorpath(struct hierarchy *h, | |
2373 | const char *inpath, | |
2374 | const char *filename) | |
2375 | { | |
371f834d | 2376 | return must_make_path(h->mountpoint, inpath, filename, NULL); |
ccb4cabe SH |
2377 | } |
2378 | ||
c2aed66d CB |
2379 | /* Technically, we're always at a delegation boundary here. (This is especially |
2380 | * true when cgroup namespaces are available.) The reasoning is that in order | |
2381 | * for us to have been able to start a container in the first place the root | |
2382 | * cgroup must have been a leaf node. Now, either the container's init system | |
2383 | * has populated the cgroup and kept it as a leaf node or it has created | |
2384 | * subtrees. In the former case we will simply attach to the leaf node we | |
2385 | * created when we started the container in the latter case we create our own | |
2386 | * cgroup for the attaching process. | |
2387 | */ | |
a3926f6a CB |
2388 | static int __cg_unified_attach(const struct hierarchy *h, const char *name, |
2389 | const char *lxcpath, const char *pidstr, | |
2390 | size_t pidstr_len, const char *controller) | |
c2aed66d CB |
2391 | { |
2392 | int ret; | |
2393 | size_t len; | |
2394 | int fret = -1, idx = 0; | |
2395 | char *base_path = NULL, *container_cgroup = NULL, *full_path = NULL; | |
2396 | ||
2397 | container_cgroup = lxc_cmd_get_cgroup_path(name, lxcpath, controller); | |
2398 | /* not running */ | |
2399 | if (!container_cgroup) | |
2400 | return 0; | |
2401 | ||
2402 | base_path = must_make_path(h->mountpoint, container_cgroup, NULL); | |
2403 | full_path = must_make_path(base_path, "cgroup.procs", NULL); | |
2404 | /* cgroup is populated */ | |
2405 | ret = lxc_write_to_file(full_path, pidstr, pidstr_len, false); | |
2406 | if (ret < 0 && errno != EBUSY) | |
2407 | goto on_error; | |
2408 | ||
2409 | if (ret == 0) | |
2410 | goto on_success; | |
2411 | ||
2412 | free(full_path); | |
2413 | ||
2414 | len = strlen(base_path) + sizeof("/lxc-1000") - 1 + | |
2415 | sizeof("/cgroup-procs") - 1; | |
2416 | full_path = must_alloc(len + 1); | |
2417 | do { | |
2418 | if (idx) | |
2419 | ret = snprintf(full_path, len + 1, "%s/lxc-%d", | |
2420 | base_path, idx); | |
2421 | else | |
2422 | ret = snprintf(full_path, len + 1, "%s/lxc", base_path); | |
2423 | if (ret < 0 || (size_t)ret >= len + 1) | |
2424 | goto on_error; | |
2425 | ||
2426 | ret = mkdir_p(full_path, 0755); | |
2427 | if (ret < 0 && errno != EEXIST) | |
2428 | goto on_error; | |
2429 | ||
2430 | strcat(full_path, "/cgroup.procs"); | |
2431 | ret = lxc_write_to_file(full_path, pidstr, len, false); | |
2432 | if (ret == 0) | |
2433 | goto on_success; | |
2434 | ||
2435 | /* this is a non-leaf node */ | |
2436 | if (errno != EBUSY) | |
2437 | goto on_error; | |
2438 | ||
2439 | } while (++idx > 0 && idx < 1000); | |
2440 | ||
2441 | on_success: | |
2442 | if (idx < 1000) | |
2443 | fret = 0; | |
2444 | ||
2445 | on_error: | |
2446 | free(base_path); | |
2447 | free(container_cgroup); | |
2448 | free(full_path); | |
2449 | ||
2450 | return fret; | |
2451 | } | |
2452 | ||
ccb4cabe SH |
2453 | static bool cgfsng_attach(const char *name, const char *lxcpath, pid_t pid) |
2454 | { | |
c2aed66d | 2455 | int i, len, ret; |
ccb4cabe | 2456 | char pidstr[25]; |
ccb4cabe SH |
2457 | |
2458 | len = snprintf(pidstr, 25, "%d", pid); | |
2459 | if (len < 0 || len > 25) | |
2460 | return false; | |
2461 | ||
457ca9aa | 2462 | for (i = 0; hierarchies[i]; i++) { |
c2aed66d CB |
2463 | char *path; |
2464 | char *fullpath = NULL; | |
457ca9aa | 2465 | struct hierarchy *h = hierarchies[i]; |
ccb4cabe | 2466 | |
c2aed66d | 2467 | if (h->version == CGROUP2_SUPER_MAGIC) { |
a3926f6a CB |
2468 | ret = __cg_unified_attach(h, name, lxcpath, pidstr, len, |
2469 | h->controllers[0]); | |
c2aed66d CB |
2470 | if (ret < 0) |
2471 | return false; | |
2472 | ||
2473 | continue; | |
2474 | } | |
2475 | ||
ccb4cabe | 2476 | path = lxc_cmd_get_cgroup_path(name, lxcpath, h->controllers[0]); |
c2aed66d CB |
2477 | /* not running */ |
2478 | if (!path) | |
ccb4cabe SH |
2479 | continue; |
2480 | ||
371f834d | 2481 | fullpath = build_full_cgpath_from_monitorpath(h, path, "cgroup.procs"); |
c2aed66d CB |
2482 | ret = lxc_write_to_file(fullpath, pidstr, len, false); |
2483 | if (ret < 0) { | |
ccb4cabe SH |
2484 | SYSERROR("Failed to attach %d to %s", (int)pid, fullpath); |
2485 | free(fullpath); | |
ccb4cabe SH |
2486 | return false; |
2487 | } | |
ccb4cabe SH |
2488 | free(fullpath); |
2489 | } | |
2490 | ||
ccb4cabe SH |
2491 | return true; |
2492 | } | |
2493 | ||
2494 | /* | |
2495 | * Called externally (i.e. from 'lxc-cgroup') to query cgroup limits. | |
2496 | * Here we don't have a cgroup_data set up, so we ask the running | |
2497 | * container through the commands API for the cgroup path | |
2498 | */ | |
0069cc61 CB |
2499 | static int cgfsng_get(const char *filename, char *value, size_t len, |
2500 | const char *name, const char *lxcpath) | |
ccb4cabe | 2501 | { |
ccb4cabe | 2502 | int ret = -1; |
0069cc61 CB |
2503 | size_t controller_len; |
2504 | char *controller, *p, *path; | |
2505 | struct hierarchy *h; | |
ccb4cabe | 2506 | |
0069cc61 CB |
2507 | controller_len = strlen(filename); |
2508 | controller = alloca(controller_len + 1); | |
2509 | strcpy(controller, filename); | |
2510 | p = strchr(controller, '.'); | |
2511 | if (p) | |
ccb4cabe SH |
2512 | *p = '\0'; |
2513 | ||
0069cc61 CB |
2514 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2515 | /* not running */ | |
2516 | if (!path) | |
ccb4cabe SH |
2517 | return -1; |
2518 | ||
0069cc61 | 2519 | h = get_hierarchy(controller); |
ccb4cabe | 2520 | if (h) { |
0069cc61 CB |
2521 | char *fullpath; |
2522 | ||
2523 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
ccb4cabe SH |
2524 | ret = lxc_read_from_file(fullpath, value, len); |
2525 | free(fullpath); | |
2526 | } | |
ccb4cabe SH |
2527 | free(path); |
2528 | ||
2529 | return ret; | |
2530 | } | |
2531 | ||
2532 | /* | |
2533 | * Called externally (i.e. from 'lxc-cgroup') to set new cgroup limits. | |
2534 | * Here we don't have a cgroup_data set up, so we ask the running | |
2535 | * container through the commands API for the cgroup path | |
2536 | */ | |
87777968 CB |
2537 | static int cgfsng_set(const char *filename, const char *value, const char *name, |
2538 | const char *lxcpath) | |
ccb4cabe | 2539 | { |
ccb4cabe | 2540 | int ret = -1; |
87777968 CB |
2541 | size_t controller_len; |
2542 | char *controller, *p, *path; | |
2543 | struct hierarchy *h; | |
ccb4cabe | 2544 | |
87777968 CB |
2545 | controller_len = strlen(filename); |
2546 | controller = alloca(controller_len + 1); | |
2547 | strcpy(controller, filename); | |
2548 | p = strchr(controller, '.'); | |
2549 | if (p) | |
ccb4cabe SH |
2550 | *p = '\0'; |
2551 | ||
87777968 CB |
2552 | path = lxc_cmd_get_cgroup_path(name, lxcpath, controller); |
2553 | /* not running */ | |
2554 | if (!path) | |
ccb4cabe SH |
2555 | return -1; |
2556 | ||
87777968 | 2557 | h = get_hierarchy(controller); |
ccb4cabe | 2558 | if (h) { |
87777968 CB |
2559 | char *fullpath; |
2560 | ||
2561 | fullpath = build_full_cgpath_from_monitorpath(h, path, filename); | |
ccb4cabe SH |
2562 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); |
2563 | free(fullpath); | |
2564 | } | |
ccb4cabe SH |
2565 | free(path); |
2566 | ||
2567 | return ret; | |
2568 | } | |
2569 | ||
72add155 SH |
2570 | /* |
2571 | * take devices cgroup line | |
2572 | * /dev/foo rwx | |
2573 | * and convert it to a valid | |
2574 | * type major:minor mode | |
2575 | * line. Return <0 on error. Dest is a preallocated buffer | |
2576 | * long enough to hold the output. | |
2577 | */ | |
2578 | static int convert_devpath(const char *invalue, char *dest) | |
2579 | { | |
2a06d041 CB |
2580 | int n_parts; |
2581 | char *p, *path, type; | |
72add155 SH |
2582 | struct stat sb; |
2583 | unsigned long minor, major; | |
2a06d041 CB |
2584 | int ret = -EINVAL; |
2585 | char *mode = NULL; | |
72add155 SH |
2586 | |
2587 | path = must_copy_string(invalue); | |
2588 | ||
2589 | /* | |
2590 | * read path followed by mode; ignore any trailing text. | |
2591 | * A ' # comment' would be legal. Technically other text | |
2592 | * is not legal, we could check for that if we cared to | |
2593 | */ | |
2594 | for (n_parts = 1, p = path; *p && n_parts < 3; p++) { | |
2c2d6c49 SH |
2595 | if (*p != ' ') |
2596 | continue; | |
2597 | *p = '\0'; | |
2598 | if (n_parts != 1) | |
2599 | break; | |
2600 | p++; | |
2601 | n_parts++; | |
2602 | while (*p == ' ') | |
2603 | p++; | |
2604 | mode = p; | |
2605 | if (*p == '\0') | |
2606 | goto out; | |
72add155 | 2607 | } |
2c2d6c49 SH |
2608 | |
2609 | if (n_parts == 1) | |
72add155 | 2610 | goto out; |
72add155 SH |
2611 | |
2612 | ret = stat(path, &sb); | |
2613 | if (ret < 0) | |
2614 | goto out; | |
2615 | ||
72add155 SH |
2616 | mode_t m = sb.st_mode & S_IFMT; |
2617 | switch (m) { | |
2618 | case S_IFBLK: | |
2619 | type = 'b'; | |
2620 | break; | |
2621 | case S_IFCHR: | |
2622 | type = 'c'; | |
2623 | break; | |
2c2d6c49 | 2624 | default: |
72add155 SH |
2625 | ERROR("Unsupported device type %i for %s", m, path); |
2626 | ret = -EINVAL; | |
2627 | goto out; | |
2628 | } | |
2c2d6c49 SH |
2629 | |
2630 | major = MAJOR(sb.st_rdev); | |
2631 | minor = MINOR(sb.st_rdev); | |
2632 | ret = snprintf(dest, 50, "%c %lu:%lu %s", type, major, minor, mode); | |
72add155 | 2633 | if (ret < 0 || ret >= 50) { |
2a06d041 CB |
2634 | ERROR("Error on configuration value \"%c %lu:%lu %s\" (max 50 " |
2635 | "chars)", type, major, minor, mode); | |
72add155 SH |
2636 | ret = -ENAMETOOLONG; |
2637 | goto out; | |
2638 | } | |
2639 | ret = 0; | |
2640 | ||
2641 | out: | |
2642 | free(path); | |
2643 | return ret; | |
2644 | } | |
2645 | ||
ccb4cabe SH |
2646 | /* |
2647 | * Called from setup_limits - here we have the container's cgroup_data because | |
2648 | * we created the cgroups | |
2649 | */ | |
a3926f6a CB |
2650 | static int cg_legacy_set_data(const char *filename, const char *value, |
2651 | struct cgfsng_handler_data *d) | |
ccb4cabe | 2652 | { |
b3646d7e | 2653 | char *fullpath, *p; |
ab1a6cac | 2654 | size_t len; |
1a0e70ac CB |
2655 | /* "b|c <2^64-1>:<2^64-1> r|w|m" = 47 chars max */ |
2656 | char converted_value[50]; | |
b3646d7e CB |
2657 | struct hierarchy *h; |
2658 | int ret = 0; | |
2659 | char *controller = NULL; | |
ccb4cabe | 2660 | |
ab1a6cac CB |
2661 | len = strlen(filename); |
2662 | controller = alloca(len + 1); | |
b3646d7e | 2663 | strcpy(controller, filename); |
ab1a6cac CB |
2664 | p = strchr(controller, '.'); |
2665 | if (p) | |
ccb4cabe SH |
2666 | *p = '\0'; |
2667 | ||
c8bf519d | 2668 | if (strcmp("devices.allow", filename) == 0 && value[0] == '/') { |
72add155 SH |
2669 | ret = convert_devpath(value, converted_value); |
2670 | if (ret < 0) | |
c8bf519d | 2671 | return ret; |
72add155 | 2672 | value = converted_value; |
c8bf519d | 2673 | } |
2674 | ||
b3646d7e CB |
2675 | h = get_hierarchy(controller); |
2676 | if (!h) { | |
2677 | ERROR("Failed to setup limits for the \"%s\" controller. " | |
2678 | "The controller seems to be unused by \"cgfsng\" cgroup " | |
2679 | "driver or not enabled on the cgroup hierarchy", | |
2680 | controller); | |
d1953b26 | 2681 | errno = ENOENT; |
ab1a6cac | 2682 | return -ENOENT; |
ccb4cabe | 2683 | } |
b3646d7e CB |
2684 | |
2685 | fullpath = must_make_path(h->fullcgpath, filename, NULL); | |
2686 | ret = lxc_write_to_file(fullpath, value, strlen(value), false); | |
2687 | free(fullpath); | |
ccb4cabe SH |
2688 | return ret; |
2689 | } | |
2690 | ||
a3926f6a CB |
2691 | static bool __cg_legacy_setup_limits(void *hdata, |
2692 | struct lxc_list *cgroup_settings, | |
2693 | bool do_devices) | |
ccb4cabe SH |
2694 | { |
2695 | struct cgfsng_handler_data *d = hdata; | |
2696 | struct lxc_list *iterator, *sorted_cgroup_settings, *next; | |
2697 | struct lxc_cgroup *cg; | |
ccb4cabe SH |
2698 | bool ret = false; |
2699 | ||
2700 | if (lxc_list_empty(cgroup_settings)) | |
2701 | return true; | |
2702 | ||
2703 | sorted_cgroup_settings = sort_cgroup_settings(cgroup_settings); | |
6b38e644 | 2704 | if (!sorted_cgroup_settings) |
ccb4cabe | 2705 | return false; |
ccb4cabe | 2706 | |
ccb4cabe SH |
2707 | lxc_list_for_each(iterator, sorted_cgroup_settings) { |
2708 | cg = iterator->elem; | |
2709 | ||
2710 | if (do_devices == !strncmp("devices", cg->subsystem, 7)) { | |
a3926f6a | 2711 | if (cg_legacy_set_data(cg->subsystem, cg->value, d)) { |
ccb4cabe SH |
2712 | if (do_devices && (errno == EACCES || errno == EPERM)) { |
2713 | WARN("Error setting %s to %s for %s", | |
2714 | cg->subsystem, cg->value, d->name); | |
2715 | continue; | |
2716 | } | |
2717 | SYSERROR("Error setting %s to %s for %s", | |
2718 | cg->subsystem, cg->value, d->name); | |
2719 | goto out; | |
2720 | } | |
6a628f4a | 2721 | DEBUG("cgroup '%s' set to '%s'", cg->subsystem, cg->value); |
ccb4cabe | 2722 | } |
ccb4cabe SH |
2723 | } |
2724 | ||
2725 | ret = true; | |
6b38e644 | 2726 | INFO("Limits for the legacy cgroup hierarchies have been setup"); |
ccb4cabe | 2727 | out: |
ccb4cabe SH |
2728 | lxc_list_for_each_safe(iterator, sorted_cgroup_settings, next) { |
2729 | lxc_list_del(iterator); | |
2730 | free(iterator); | |
2731 | } | |
2732 | free(sorted_cgroup_settings); | |
2733 | return ret; | |
2734 | } | |
2735 | ||
a3926f6a CB |
2736 | static bool __cg_unified_setup_limits(void *hdata, |
2737 | struct lxc_list *cgroup_settings) | |
6b38e644 CB |
2738 | { |
2739 | struct lxc_list *iterator; | |
2740 | struct hierarchy *h = unified; | |
2741 | ||
2742 | if (lxc_list_empty(cgroup_settings)) | |
2743 | return true; | |
2744 | ||
2745 | if (!h) | |
2746 | return false; | |
2747 | ||
2748 | lxc_list_for_each(iterator, cgroup_settings) { | |
2749 | int ret; | |
2750 | char *fullpath; | |
2751 | struct lxc_cgroup *cg = iterator->elem; | |
2752 | ||
2753 | fullpath = must_make_path(h->fullcgpath, cg->subsystem, NULL); | |
2754 | ret = lxc_write_to_file(fullpath, cg->value, strlen(cg->value), false); | |
2755 | free(fullpath); | |
2756 | if (ret < 0) { | |
2757 | SYSERROR("Failed to set \"%s\" to \"%s\"", cg->subsystem, cg->value); | |
2758 | return false; | |
2759 | } | |
2760 | TRACE("Set \"%s\" to \"%s\"", cg->subsystem, cg->value); | |
2761 | } | |
2762 | ||
2763 | INFO("Limits for the unified cgroup hierarchy have been setup"); | |
2764 | return true; | |
2765 | } | |
2766 | ||
2767 | static bool cgfsng_setup_limits(void *hdata, struct lxc_conf *conf, | |
2768 | bool do_devices) | |
2769 | { | |
2770 | bool bret; | |
2771 | ||
a3926f6a | 2772 | bret = __cg_legacy_setup_limits(hdata, &conf->cgroup, do_devices); |
6b38e644 CB |
2773 | if (!bret) |
2774 | return false; | |
2775 | ||
a3926f6a | 2776 | return __cg_unified_setup_limits(hdata, &conf->cgroup2); |
6b38e644 CB |
2777 | } |
2778 | ||
ccb4cabe SH |
2779 | static struct cgroup_ops cgfsng_ops = { |
2780 | .init = cgfsng_init, | |
2781 | .destroy = cgfsng_destroy, | |
2782 | .create = cgfsng_create, | |
2783 | .enter = cgfsng_enter, | |
ccb4cabe | 2784 | .escape = cgfsng_escape, |
36662416 TA |
2785 | .num_hierarchies = cgfsng_num_hierarchies, |
2786 | .get_hierarchies = cgfsng_get_hierarchies, | |
ccb4cabe SH |
2787 | .get_cgroup = cgfsng_get_cgroup, |
2788 | .get = cgfsng_get, | |
2789 | .set = cgfsng_set, | |
2790 | .unfreeze = cgfsng_unfreeze, | |
2791 | .setup_limits = cgfsng_setup_limits, | |
2792 | .name = "cgroupfs-ng", | |
2793 | .attach = cgfsng_attach, | |
058c1cb6 | 2794 | .chown = cgfsng_chown, |
ccb4cabe SH |
2795 | .mount_cgroup = cgfsng_mount, |
2796 | .nrtasks = cgfsng_nrtasks, | |
2797 | .driver = CGFSNG, | |
2798 | ||
2799 | /* unsupported */ | |
2800 | .create_legacy = NULL, | |
2801 | }; |