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