]>
Commit | Line | Data |
---|---|---|
df54106a SH |
1 | /* pam-cgfs |
2 | * | |
3 | * Copyright © 2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
e65cfafc | 5 | * Author: Christian Brauner <christian.brauner@ubuntu.com> |
df54106a | 6 | * |
e65cfafc CB |
7 | * When a user logs in, this pam module will create cgroups which the user may |
8 | * administer. It handles both pure cgroupfs v1 and pure cgroupfs v2, as well as | |
9 | * mixed mounts, where some controllers are mounted in a standard cgroupfs v1 | |
10 | * hierarchy location (/sys/fs/cgroup/<controller>) and others are in the | |
11 | * cgroupfs v2 hierarchy. | |
12 | * Writeable cgroups are either created for all controllers or, if specified, | |
13 | * for any controllers listed on the command line. | |
df54106a SH |
14 | * The cgroup created will be "user/$user/0" for the first session, |
15 | * "user/$user/1" for the second, etc. | |
16 | * | |
e65cfafc CB |
17 | * Systems with a systemd init system are treated specially, both with respect |
18 | * to cgroupfs v1 and cgroupfs v2. For both, cgroupfs v1 and cgroupfs v2, We | |
19 | * check whether systemd already placed us in a cgroup it created: | |
20 | * | |
21 | * user.slice/user-uid.slice/session-n.scope | |
22 | * | |
23 | * by checking whether uid == our uid. If it did, we simply chown the last | |
24 | * part (session-n.scope). If it did not we create a cgroup as outlined above | |
25 | * (user/$user/n) and chown it to our uid. | |
26 | * The same holds for cgroupfs v2 where this assumptions becomes crucial: | |
27 | * We __have to__ be placed in our under the cgroup systemd created for us on | |
28 | * login, otherwise things like starting an xserver or similar will not work. | |
edd25678 | 29 | * |
df54106a SH |
30 | * All requested cgroups must be mounted under /sys/fs/cgroup/$controller, |
31 | * no messing around with finding mountpoints. | |
32 | * | |
33 | * See COPYING file for details. | |
34 | */ | |
35 | ||
04742595 | 36 | #include <ctype.h> |
e65cfafc CB |
37 | #include <dirent.h> |
38 | #include <errno.h> | |
04742595 | 39 | #include <fcntl.h> |
e65cfafc CB |
40 | #include <pwd.h> |
41 | #include <stdarg.h> | |
42 | #include <stdbool.h> | |
04742595 | 43 | #include <stdint.h> |
df54106a SH |
44 | #include <stdio.h> |
45 | #include <stdlib.h> | |
e65cfafc | 46 | #include <string.h> |
df54106a | 47 | #include <syslog.h> |
e65cfafc CB |
48 | #include <unistd.h> |
49 | #include <linux/unistd.h> | |
df54106a | 50 | #include <sys/mount.h> |
df54106a | 51 | #include <sys/param.h> |
e65cfafc CB |
52 | #include <sys/stat.h> |
53 | #include <sys/types.h> | |
54 | #include <sys/vfs.h> | |
df54106a SH |
55 | |
56 | #define PAM_SM_SESSION | |
57 | #include <security/_pam_macros.h> | |
58 | #include <security/pam_modules.h> | |
59 | ||
e65cfafc CB |
60 | #include "macro.h" |
61 | ||
62 | #ifndef CGROUP_SUPER_MAGIC | |
63 | #define CGROUP_SUPER_MAGIC 0x27e0eb | |
64 | #endif | |
df54106a | 65 | |
e65cfafc CB |
66 | #ifndef CGROUP2_SUPER_MAGIC |
67 | #define CGROUP2_SUPER_MAGIC 0x63677270 | |
68 | #endif | |
df54106a | 69 | |
04742595 CB |
70 | /* Taken over modified from the kernel sources. */ |
71 | #define NBITS 32 /* bits in uint32_t */ | |
72 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
73 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
74 | ||
e65cfafc CB |
75 | static enum cg_mount_mode { |
76 | CGROUP_UNKNOWN = -1, | |
77 | CGROUP_MIXED = 0, | |
78 | CGROUP_PURE_V1 = 1, | |
79 | CGROUP_PURE_V2 = 2, | |
80 | CGROUP_UNINITIALIZED = 3, | |
81 | } cg_mount_mode = CGROUP_UNINITIALIZED; | |
82 | ||
7c029b0f | 83 | /* Common helper functions. Most of these have been taken from LXC. */ |
e65cfafc CB |
84 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen); |
85 | static int append_null_to_list(void ***list); | |
86 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen); | |
04742595 CB |
87 | static inline void clear_bit(unsigned bit, uint32_t *bitarr) |
88 | { | |
89 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
90 | } | |
e65cfafc CB |
91 | static char *copy_to_eol(char *s); |
92 | static bool file_exists(const char *f); | |
93 | static void free_string_list(char **list); | |
94 | static char *get_mountpoint(char *line); | |
95 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid); | |
96 | static int handle_login(const char *user, uid_t uid, gid_t gid); | |
04742595 CB |
97 | static inline bool is_set(unsigned bit, uint32_t *bitarr) |
98 | { | |
99 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
100 | } | |
e65cfafc CB |
101 | /* __typeof__ should be safe to use with all compilers. */ |
102 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
103 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val); | |
104 | static bool is_lxcfs(const char *line); | |
105 | static bool is_cgv1(char *line); | |
106 | static bool is_cgv2(char *line); | |
107 | static bool mkdir_p(const char *root, char *path); | |
108 | static void *must_alloc(size_t sz); | |
109 | static void must_add_to_list(char ***clist, char *entry); | |
110 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
111 | char *entry); | |
112 | static void must_append_string(char ***list, char *entry); | |
113 | static char *must_copy_string(const char *entry); | |
114 | static char *must_make_path(const char *first, ...) __attribute__((sentinel)); | |
115 | static void *must_realloc(void *orig, size_t sz); | |
116 | static void mysyslog(int err, const char *format, ...) __attribute__((sentinel)); | |
117 | static char *read_file(char *fnam); | |
04742595 | 118 | static int read_from_file(const char *filename, void* buf, size_t count); |
e65cfafc | 119 | static int recursive_rmdir(char *dirname); |
04742595 CB |
120 | static inline void set_bit(unsigned bit, uint32_t *bitarr) |
121 | { | |
122 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
123 | } | |
e65cfafc | 124 | static bool string_in_list(char **list, const char *entry); |
808fd1ef | 125 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix); |
e65cfafc CB |
126 | static void trim(char *s); |
127 | static bool write_int(char *path, int v); | |
808fd1ef CB |
128 | static ssize_t write_nointr(int fd, const void* buf, size_t count); |
129 | static int write_to_file(const char *filename, const void *buf, size_t count, | |
130 | bool add_newline); | |
e65cfafc CB |
131 | |
132 | /* cgroupfs prototypes. */ | |
04742595 CB |
133 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid); |
134 | static uint32_t *cg_cpumask(char *buf, size_t nbits); | |
135 | static bool cg_copy_parent_file(char *path, char *file); | |
136 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits); | |
137 | static bool cg_enter(const char *cgroup); | |
138 | static void cg_escape(void); | |
139 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized); | |
140 | static ssize_t cg_get_max_cpus(char *cpulist); | |
141 | static int cg_get_version_of_mntpt(const char *path); | |
142 | static bool cg_init(uid_t uid, gid_t gid); | |
e65cfafc | 143 | static void cg_mark_to_make_rw(const char *cstring); |
04742595 | 144 | static void cg_prune_empty_cgroups(const char *user); |
e65cfafc CB |
145 | static bool cg_systemd_created_user_slice(const char *base_cgroup, |
146 | const char *init_cgroup, | |
147 | const char *in, uid_t uid); | |
148 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
149 | const char *base_cgroup, uid_t uid, | |
150 | gid_t gid, | |
151 | bool systemd_user_slice); | |
04742595 CB |
152 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid); |
153 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
154 | const char *init_cgroup, uid_t uid); | |
e65cfafc | 155 | static void cg_systemd_prune_init_scope(char *cg); |
e65cfafc CB |
156 | static bool is_lxcfs(const char *line); |
157 | ||
158 | /* cgroupfs v1 prototypes. */ | |
159 | struct cgv1_hierarchy { | |
160 | char **controllers; | |
161 | char *mountpoint; | |
162 | char *base_cgroup; | |
163 | char *fullcgpath; | |
164 | char *init_cgroup; | |
165 | bool create_rw_cgroup; | |
166 | bool systemd_user_slice; | |
167 | }; | |
168 | ||
169 | static struct cgv1_hierarchy **cgv1_hierarchies; | |
170 | ||
171 | static void cgv1_add_controller(char **clist, char *mountpoint, | |
172 | char *base_cgroup, char *init_cgroup); | |
173 | static bool cgv1_controller_in_clist(char *cgline, char *c); | |
174 | static bool cgv1_controller_lists_intersect(char **l1, char **l2); | |
175 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, | |
176 | char **clist); | |
177 | static bool cgv1_create(const char *cgroup, uid_t uid, gid_t gid, | |
178 | bool *existed); | |
179 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, | |
180 | uid_t uid, gid_t gid, bool *existed); | |
181 | static bool cgv1_enter(const char *cgroup); | |
182 | static void cgv1_escape(void); | |
183 | static bool cgv1_get_controllers(char ***klist, char ***nlist); | |
184 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller); | |
185 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, | |
186 | char *line); | |
04742595 CB |
187 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, |
188 | const char *cgroup); | |
189 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h); | |
e65cfafc CB |
190 | static bool cgv1_init(uid_t uid, gid_t gid); |
191 | static void cgv1_mark_to_make_rw(char **clist); | |
192 | static char *cgv1_must_prefix_named(char *entry); | |
193 | static bool cgv1_prune_empty_cgroups(const char *user); | |
194 | static bool cgv1_remove_one(struct cgv1_hierarchy *h, const char *cgroup); | |
195 | static bool is_cgv1(char *line); | |
196 | ||
197 | /* cgroupfs v2 prototypes. */ | |
198 | struct cgv2_hierarchy { | |
199 | char **controllers; | |
200 | char *mountpoint; | |
201 | char *base_cgroup; | |
202 | char *fullcgpath; | |
203 | char *init_cgroup; | |
204 | bool create_rw_cgroup; | |
205 | bool systemd_user_slice; | |
206 | }; | |
207 | ||
208 | /* Actually this should only be a single hierarchy. But for the sake of | |
209 | * parallelism and because the layout of the cgroupfs v2 is still somewhat | |
210 | * changing, we'll leave it as an array of structs. | |
211 | */ | |
212 | static struct cgv2_hierarchy **cgv2_hierarchies; | |
213 | ||
214 | static void cgv2_add_controller(char **clist, char *mountpoint, | |
215 | char *base_cgroup, char *init_cgroup, | |
216 | bool systemd_user_slice); | |
217 | static bool cgv2_create(const char *cgroup, uid_t uid, gid_t gid, | |
218 | bool *existed); | |
219 | static bool cgv2_enter(const char *cgroup); | |
220 | static void cgv2_escape(void); | |
221 | static char *cgv2_get_current_cgroup(int pid); | |
222 | static bool cgv2_init(uid_t uid, gid_t gid); | |
223 | static void cgv2_mark_to_make_rw(char **clist); | |
224 | static bool cgv2_prune_empty_cgroups(const char *user); | |
225 | static bool cgv2_remove(const char *cgroup); | |
226 | static bool is_cgv2(char *line); | |
227 | ||
7c029b0f | 228 | /* Common helper functions. Most of these have been taken from LXC. */ |
df54106a SH |
229 | static void mysyslog(int err, const char *format, ...) |
230 | { | |
231 | va_list args; | |
232 | ||
233 | va_start(args, format); | |
04742595 | 234 | openlog("PAM-CGFS", LOG_CONS | LOG_PID, LOG_AUTH); |
df54106a SH |
235 | vsyslog(err, format, args); |
236 | va_end(args); | |
237 | closelog(); | |
238 | } | |
239 | ||
e65cfafc CB |
240 | /* realloc() pointer; do not fail. */ |
241 | static void *must_realloc(void *orig, size_t sz) | |
242 | { | |
243 | void *ret; | |
df54106a | 244 | |
e65cfafc CB |
245 | do { |
246 | ret = realloc(orig, sz); | |
247 | } while (!ret); | |
248 | ||
249 | return ret; | |
250 | } | |
251 | ||
252 | /* realloc() pointer in batch sizes; do not fail. */ | |
253 | #define BATCH_SIZE 50 | |
254 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
df54106a | 255 | { |
e65cfafc CB |
256 | int newbatches = (newlen / BATCH_SIZE) + 1; |
257 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
258 | ||
259 | if (!*mem || newbatches > oldbatches) | |
260 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
261 | } | |
262 | ||
263 | /* Append lines as is to pointer; do not fail. */ | |
264 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
265 | { | |
266 | size_t full = oldlen + newlen; | |
267 | ||
268 | batch_realloc(dest, oldlen, full + 1); | |
269 | ||
270 | memcpy(*dest + oldlen, new, newlen + 1); | |
271 | } | |
272 | ||
273 | /* Read in whole file and return allocated pointer. */ | |
274 | static char *read_file(char *fnam) | |
275 | { | |
276 | FILE *f; | |
277 | int linelen; | |
278 | char *line = NULL, *buf = NULL; | |
279 | size_t len = 0, fulllen = 0; | |
280 | ||
281 | f = fopen(fnam, "r"); | |
282 | if (!f) | |
283 | return NULL; | |
284 | ||
285 | while ((linelen = getline(&line, &len, f)) != -1) { | |
286 | append_line(&buf, fulllen, line, linelen); | |
287 | fulllen += linelen; | |
288 | } | |
289 | ||
290 | fclose(f); | |
291 | free(line); | |
292 | ||
293 | return buf; | |
294 | } | |
295 | ||
296 | /* Given a pointer to a null-terminated array of pointers, realloc to add one | |
297 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
298 | * second-to-last entry - that is, the one which is now available for use | |
299 | * (keeping the list null-terminated). | |
300 | */ | |
301 | static int append_null_to_list(void ***list) | |
302 | { | |
303 | int newentry = 0; | |
304 | ||
305 | if (*list) | |
306 | for (; (*list)[newentry]; newentry++) { | |
307 | ; | |
308 | } | |
309 | ||
310 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
311 | (*list)[newentry + 1] = NULL; | |
312 | ||
313 | return newentry; | |
314 | } | |
315 | ||
316 | /* Make allocated copy of string; do not fail. */ | |
317 | static char *must_copy_string(const char *entry) | |
318 | { | |
319 | char *ret; | |
320 | ||
321 | if (!entry) | |
322 | return NULL; | |
df54106a SH |
323 | |
324 | do { | |
e65cfafc CB |
325 | ret = strdup(entry); |
326 | } while (!ret); | |
df54106a | 327 | |
e65cfafc CB |
328 | return ret; |
329 | } | |
330 | ||
331 | /* Append new entry to null-terminated array of pointer; make sure that array of | |
332 | * pointers will still be null-terminated. | |
333 | */ | |
334 | static void must_append_string(char ***list, char *entry) | |
335 | { | |
336 | int newentry; | |
337 | char *copy; | |
338 | ||
339 | newentry = append_null_to_list((void ***)list); | |
340 | copy = must_copy_string(entry); | |
341 | (*list)[newentry] = copy; | |
342 | } | |
343 | ||
344 | /* Remove newlines from string. */ | |
345 | static void trim(char *s) | |
346 | { | |
347 | size_t len = strlen(s); | |
348 | ||
fa00d59a | 349 | while ((len > 0) && s[len - 1] == '\n') |
e65cfafc CB |
350 | s[--len] = '\0'; |
351 | } | |
352 | ||
353 | /* Allocate pointer; do not fail. */ | |
354 | static void *must_alloc(size_t sz) | |
355 | { | |
356 | return must_realloc(NULL, sz); | |
357 | } | |
358 | ||
359 | /* Make allocated copy of string. End of string is taken to be '\n'. */ | |
360 | static char *copy_to_eol(char *s) | |
361 | { | |
362 | char *newline, *sret; | |
363 | size_t len; | |
364 | ||
365 | newline = strchr(s, '\n'); | |
366 | if (!newline) | |
367 | return NULL; | |
368 | ||
369 | len = newline - s; | |
370 | sret = must_alloc(len + 1); | |
371 | memcpy(sret, s, len); | |
372 | sret[len] = '\0'; | |
373 | ||
374 | return sret; | |
375 | } | |
376 | ||
377 | /* Check if given entry under /proc/<pid>/mountinfo is a fuse.lxcfs mount. */ | |
378 | static bool is_lxcfs(const char *line) | |
379 | { | |
380 | char *p = strstr(line, " - "); | |
381 | if (!p) | |
382 | return false; | |
383 | ||
384 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; | |
385 | } | |
386 | ||
387 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v1 mount. */ | |
388 | static bool is_cgv1(char *line) | |
389 | { | |
390 | char *p = strstr(line, " - "); | |
391 | if (!p) | |
392 | return false; | |
393 | ||
394 | return strncmp(p, " - cgroup ", 10) == 0; | |
395 | } | |
396 | ||
397 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v2 mount. */ | |
398 | static bool is_cgv2(char *line) | |
399 | { | |
400 | char *p = strstr(line, " - "); | |
401 | if (!p) | |
402 | return false; | |
403 | ||
404 | return strncmp(p, " - cgroup2 ", 11) == 0; | |
405 | } | |
406 | ||
407 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
408 | * strings | |
409 | */ | |
410 | static bool string_in_list(char **list, const char *entry) | |
411 | { | |
412 | char **it; | |
413 | ||
414 | for (it = list; it && *it; it++) | |
415 | if (strcmp(*it, entry) == 0) | |
416 | return true; | |
417 | ||
418 | return false; | |
419 | } | |
420 | ||
421 | /* Free null-terminated array of strings. */ | |
422 | static void free_string_list(char **list) | |
423 | { | |
424 | char **it; | |
425 | ||
426 | for (it = list; it && *it; it++) | |
427 | free(*it); | |
428 | free(list); | |
429 | } | |
430 | ||
431 | /* Concatenate all passed-in strings into one path. Do not fail. If any piece | |
432 | * is not prefixed with '/', add a '/'. Does not remove duplicate '///' from the | |
433 | * created path. | |
434 | */ | |
435 | static char *must_make_path(const char *first, ...) | |
436 | { | |
437 | va_list args; | |
438 | char *cur, *dest; | |
439 | size_t full_len; | |
440 | ||
441 | full_len = strlen(first); | |
442 | ||
443 | dest = must_copy_string(first); | |
df54106a | 444 | |
e65cfafc | 445 | va_start(args, first); |
df54106a | 446 | while ((cur = va_arg(args, char *)) != NULL) { |
e65cfafc CB |
447 | full_len += strlen(cur); |
448 | ||
449 | if (cur[0] != '/') | |
450 | full_len++; | |
451 | ||
452 | dest = must_realloc(dest, full_len + 1); | |
453 | ||
454 | if (cur[0] != '/') | |
455 | strcat(dest, "/"); | |
456 | ||
df54106a | 457 | strcat(dest, cur); |
df54106a SH |
458 | } |
459 | va_end(args); | |
460 | ||
461 | return dest; | |
462 | } | |
463 | ||
e65cfafc CB |
464 | /* Write single integer to file. */ |
465 | static bool write_int(char *path, int v) | |
df54106a | 466 | { |
e65cfafc CB |
467 | FILE *f; |
468 | bool ret = true; | |
469 | ||
470 | f = fopen(path, "w"); | |
471 | if (!f) | |
472 | return false; | |
df54106a | 473 | |
e65cfafc CB |
474 | if (fprintf(f, "%d\n", v) < 0) |
475 | ret = false; | |
476 | ||
477 | if (fclose(f) != 0) | |
478 | ret = false; | |
479 | ||
480 | return ret; | |
df54106a SH |
481 | } |
482 | ||
e65cfafc CB |
483 | /* Check if a given file exists. */ |
484 | static bool file_exists(const char *f) | |
df54106a | 485 | { |
e65cfafc | 486 | struct stat statbuf; |
df54106a | 487 | |
e65cfafc | 488 | return stat(f, &statbuf) == 0; |
df54106a SH |
489 | } |
490 | ||
e65cfafc | 491 | /* Create directory and (if necessary) its parents. */ |
df54106a SH |
492 | static bool mkdir_p(const char *root, char *path) |
493 | { | |
494 | char *b, orig, *e; | |
495 | ||
496 | if (strlen(path) < strlen(root)) | |
497 | return false; | |
e65cfafc | 498 | |
df54106a SH |
499 | if (strlen(path) == strlen(root)) |
500 | return true; | |
501 | ||
502 | b = path + strlen(root) + 1; | |
e65cfafc CB |
503 | while (true) { |
504 | while (*b && (*b == '/')) | |
df54106a SH |
505 | b++; |
506 | if (!*b) | |
507 | return true; | |
e65cfafc | 508 | |
df54106a SH |
509 | e = b + 1; |
510 | while (*e && *e != '/') | |
511 | e++; | |
e65cfafc | 512 | |
df54106a SH |
513 | orig = *e; |
514 | if (orig) | |
515 | *e = '\0'; | |
e65cfafc CB |
516 | |
517 | if (file_exists(path)) | |
df54106a | 518 | goto next; |
e65cfafc | 519 | |
df54106a | 520 | if (mkdir(path, 0755) < 0) { |
b36273fa | 521 | lxcfs_debug("Failed to create %s: %s.\n", path, strerror(errno)); |
df54106a SH |
522 | return false; |
523 | } | |
e65cfafc CB |
524 | |
525 | next: | |
df54106a SH |
526 | if (!orig) |
527 | return true; | |
e65cfafc | 528 | |
df54106a SH |
529 | *e = orig; |
530 | b = e + 1; | |
531 | } | |
78a2a9f3 | 532 | |
e65cfafc | 533 | return false; |
df54106a SH |
534 | } |
535 | ||
e65cfafc CB |
536 | /* Recursively remove directory and its parents. */ |
537 | static int recursive_rmdir(char *dirname) | |
538 | { | |
539 | struct dirent *direntp; | |
540 | DIR *dir; | |
541 | int r = 0; | |
df54106a | 542 | |
e65cfafc CB |
543 | dir = opendir(dirname); |
544 | if (!dir) | |
545 | return -ENOENT; | |
df54106a | 546 | |
e65cfafc CB |
547 | while ((direntp = readdir(dir))) { |
548 | struct stat st; | |
549 | char *pathname; | |
550 | ||
551 | if (!direntp) | |
552 | break; | |
553 | ||
554 | if (!strcmp(direntp->d_name, ".") || | |
555 | !strcmp(direntp->d_name, "..")) | |
556 | continue; | |
557 | ||
558 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
559 | ||
560 | if (lstat(pathname, &st)) { | |
561 | if (!r) | |
562 | lxcfs_debug("Failed to stat %s.\n", pathname); | |
563 | r = -1; | |
564 | goto next; | |
edd25678 | 565 | } |
e65cfafc CB |
566 | |
567 | if (!S_ISDIR(st.st_mode)) | |
568 | goto next; | |
569 | ||
570 | if (recursive_rmdir(pathname) < 0) | |
571 | r = -1; | |
572 | next: | |
573 | free(pathname); | |
2be80971 | 574 | } |
e65cfafc CB |
575 | |
576 | if (rmdir(dirname) < 0) { | |
577 | if (!r) | |
b36273fa | 578 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
e65cfafc CB |
579 | r = -1; |
580 | } | |
581 | ||
582 | if (closedir(dir) < 0) { | |
583 | if (!r) | |
b36273fa | 584 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
e65cfafc CB |
585 | r = -1; |
586 | } | |
587 | ||
588 | return r; | |
589 | } | |
590 | ||
591 | /* Add new entry to null-terminated array of pointers. Make sure array is still | |
592 | * null-terminated. | |
593 | */ | |
594 | static void must_add_to_list(char ***clist, char *entry) | |
595 | { | |
596 | int newentry; | |
597 | ||
598 | newentry = append_null_to_list((void ***)clist); | |
599 | (*clist)[newentry] = must_copy_string(entry); | |
2be80971 SH |
600 | } |
601 | ||
e65cfafc CB |
602 | /* Get mountpoint from a /proc/<pid>/mountinfo line. */ |
603 | static char *get_mountpoint(char *line) | |
df54106a SH |
604 | { |
605 | int i; | |
e65cfafc CB |
606 | char *p, *sret, *p2; |
607 | size_t len; | |
608 | ||
609 | p = line; | |
610 | ||
611 | for (i = 0; i < 4; i++) { | |
612 | p = strchr(p, ' '); | |
613 | if (!p) | |
614 | return NULL; | |
615 | p++; | |
616 | } | |
df54106a | 617 | |
e65cfafc CB |
618 | p2 = strchr(p, ' '); |
619 | if (p2) | |
620 | *p2 = '\0'; | |
621 | ||
622 | len = strlen(p); | |
623 | sret = must_alloc(len + 1); | |
624 | memcpy(sret, p, len); | |
625 | sret[len] = '\0'; | |
626 | ||
627 | return sret; | |
628 | } | |
629 | ||
630 | /* Create list of cgroupfs v1 controller found under /proc/self/cgroup. Skips | |
631 | * the 0::/some/path cgroupfs v2 hierarchy listed. Splits controllers into | |
632 | * kernel controllers (@klist) and named controllers (@nlist). | |
633 | */ | |
634 | static bool cgv1_get_controllers(char ***klist, char ***nlist) | |
635 | { | |
636 | FILE *f; | |
637 | char *line = NULL; | |
638 | size_t len = 0; | |
639 | ||
640 | f = fopen("/proc/self/cgroup", "r"); | |
641 | if (!f) | |
642 | return false; | |
643 | ||
644 | while (getline(&line, &len, f) != -1) { | |
645 | char *p, *p2, *tok; | |
646 | char *saveptr = NULL; | |
647 | ||
648 | p = strchr(line, ':'); | |
649 | if (!p) | |
df54106a | 650 | continue; |
e65cfafc CB |
651 | p++; |
652 | ||
653 | p2 = strchr(p, ':'); | |
654 | if (!p2) | |
655 | continue; | |
656 | *p2 = '\0'; | |
657 | ||
658 | /* Skip the v2 hierarchy. */ | |
659 | if ((p2 - p) == 0) | |
df54106a | 660 | continue; |
e65cfafc CB |
661 | |
662 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
663 | tok = strtok_r(NULL, ",", &saveptr)) { | |
664 | if (strncmp(tok, "name=", 5) == 0) | |
665 | must_append_string(nlist, tok); | |
666 | else | |
667 | must_append_string(klist, tok); | |
df54106a SH |
668 | } |
669 | } | |
e65cfafc CB |
670 | |
671 | free(line); | |
672 | fclose(f); | |
673 | ||
674 | return true; | |
df54106a SH |
675 | } |
676 | ||
e65cfafc CB |
677 | /* Get list of controllers for cgroupfs v2 hierarchy by looking at |
678 | * cgroup.controllers and/or cgroup.subtree_control of a given (parent) cgroup. | |
679 | static bool cgv2_get_controllers(char ***klist) | |
df54106a | 680 | { |
e65cfafc CB |
681 | return -ENOSYS; |
682 | } | |
683 | */ | |
78a2a9f3 | 684 | |
e65cfafc CB |
685 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ |
686 | static char *cgv2_get_current_cgroup(int pid) | |
687 | { | |
688 | int ret; | |
689 | char *cgroups_v2; | |
690 | char *current_cgroup; | |
691 | char *copy = NULL; | |
692 | /* The largest integer that can fit into long int is 2^64. This is a | |
693 | * 20-digit number. */ | |
694 | #define __PIDLEN /* /proc */ 5 + /* /pid-to-str */ 21 + /* /cgroup */ 7 + /* \0 */ 1 | |
695 | char path[__PIDLEN]; | |
696 | ||
697 | ret = snprintf(path, __PIDLEN, "/proc/%d/cgroup", pid); | |
698 | if (ret < 0 || ret >= __PIDLEN) | |
699 | return NULL; | |
700 | ||
701 | cgroups_v2 = read_file(path); | |
702 | if (!cgroups_v2) | |
703 | return NULL; | |
704 | ||
705 | current_cgroup = strstr(cgroups_v2, "0::/"); | |
706 | if (!current_cgroup) | |
707 | goto cleanup_on_err; | |
708 | ||
709 | current_cgroup = current_cgroup + 3; | |
710 | copy = copy_to_eol(current_cgroup); | |
711 | if (!copy) | |
712 | goto cleanup_on_err; | |
713 | ||
714 | cleanup_on_err: | |
715 | free(cgroups_v2); | |
716 | if (copy) | |
717 | trim(copy); | |
718 | ||
719 | return copy; | |
df54106a SH |
720 | } |
721 | ||
e65cfafc CB |
722 | /* Given two null-terminated lists of strings, return true if any string is in |
723 | * both. | |
724 | */ | |
725 | static bool cgv1_controller_lists_intersect(char **l1, char **l2) | |
df54106a | 726 | { |
e65cfafc | 727 | char **it; |
df54106a | 728 | |
e65cfafc CB |
729 | if (!l2) |
730 | return false; | |
731 | ||
732 | for (it = l1; it && *it; it++) | |
733 | if (string_in_list(l2, *it)) | |
734 | return true; | |
735 | ||
736 | return false; | |
df54106a SH |
737 | } |
738 | ||
e65cfafc CB |
739 | /* For a null-terminated list of controllers @clist, return true if any of those |
740 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
741 | * Realistically, if one is present, all must be present. | |
742 | */ | |
743 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, char **clist) | |
df54106a | 744 | { |
e65cfafc CB |
745 | struct cgv1_hierarchy **it; |
746 | ||
747 | for (it = hlist; it && *it; it++) | |
748 | if ((*it)->controllers) | |
749 | if (cgv1_controller_lists_intersect((*it)->controllers, clist)) | |
750 | return true; | |
df54106a | 751 | return false; |
e65cfafc CB |
752 | |
753 | } | |
754 | ||
755 | /* Set boolean to mark controllers under which we are supposed create a | |
756 | * writeable cgroup. | |
757 | */ | |
758 | static void cgv1_mark_to_make_rw(char **clist) | |
759 | { | |
760 | struct cgv1_hierarchy **it; | |
761 | ||
762 | for (it = cgv1_hierarchies; it && *it; it++) | |
763 | if ((*it)->controllers) | |
764 | if (cgv1_controller_lists_intersect((*it)->controllers, clist)) | |
765 | (*it)->create_rw_cgroup = true; | |
766 | } | |
767 | ||
768 | /* Set boolean to mark whether we are supposed to create a writeable cgroup in | |
769 | * the cgroupfs v2 hierarchy. | |
770 | */ | |
771 | static void cgv2_mark_to_make_rw(char **clist) | |
772 | { | |
773 | if (string_in_list(clist, "unified")) | |
774 | if (cgv2_hierarchies) | |
775 | (*cgv2_hierarchies)->create_rw_cgroup = true; | |
df54106a SH |
776 | } |
777 | ||
e65cfafc CB |
778 | /* Wrapper around cgv{1,2}_mark_to_make_rw(). */ |
779 | static void cg_mark_to_make_rw(const char *cstring) | |
df54106a | 780 | { |
e65cfafc CB |
781 | char *copy, *tok; |
782 | char *saveptr = NULL; | |
783 | char **clist = NULL; | |
df54106a | 784 | |
e65cfafc CB |
785 | copy = must_copy_string(cstring); |
786 | ||
787 | for (tok = strtok_r(copy, ",", &saveptr); tok; | |
788 | tok = strtok_r(NULL, ",", &saveptr)) | |
789 | must_add_to_list(&clist, tok); | |
790 | ||
791 | free(copy); | |
792 | ||
793 | cgv1_mark_to_make_rw(clist); | |
794 | cgv2_mark_to_make_rw(clist); | |
795 | ||
796 | free_string_list(clist); | |
797 | } | |
798 | ||
799 | /* Prefix any named controllers with "name=", e.g. "name=systemd". */ | |
800 | static char *cgv1_must_prefix_named(char *entry) | |
801 | { | |
802 | char *s; | |
803 | int ret; | |
804 | size_t len; | |
805 | ||
806 | len = strlen(entry); | |
807 | s = must_alloc(len + 6); | |
808 | ||
809 | ret = snprintf(s, len + 6, "name=%s", entry); | |
810 | if (ret < 0 || (size_t)ret >= (len + 6)) | |
811 | return NULL; | |
812 | ||
813 | return s; | |
814 | } | |
815 | ||
816 | /* Append kernel controller in @klist or named controller in @nlist to @clist */ | |
817 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
818 | { | |
819 | int newentry; | |
820 | char *copy; | |
821 | ||
822 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
823 | return; | |
824 | ||
825 | newentry = append_null_to_list((void ***)clist); | |
826 | ||
827 | if (strncmp(entry, "name=", 5) == 0) | |
828 | copy = must_copy_string(entry); | |
829 | else if (string_in_list(klist, entry)) | |
830 | copy = must_copy_string(entry); | |
831 | else | |
832 | copy = cgv1_must_prefix_named(entry); | |
833 | ||
834 | (*clist)[newentry] = copy; | |
835 | } | |
836 | ||
837 | /* Get the controllers from a mountinfo line. There are other ways we could get | |
838 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
839 | * could parse the mount options. But we simply assume that the mountpoint must | |
840 | * be /sys/fs/cgroup/controller-list | |
841 | */ | |
842 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, char *line) | |
843 | { | |
844 | int i; | |
845 | char *p, *p2, *tok; | |
846 | char *saveptr = NULL; | |
847 | char **aret = NULL; | |
848 | ||
849 | p = line; | |
850 | ||
851 | for (i = 0; i < 4; i++) { | |
852 | p = strchr(p, ' '); | |
853 | if (!p) | |
854 | return NULL; | |
855 | p++; | |
856 | } | |
857 | if (!p) | |
858 | return NULL; | |
859 | ||
860 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) | |
861 | return NULL; | |
862 | ||
863 | p += 15; | |
864 | ||
865 | p2 = strchr(p, ' '); | |
866 | if (!p2) | |
867 | return NULL; | |
868 | *p2 = '\0'; | |
869 | ||
870 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
871 | tok = strtok_r(NULL, ",", &saveptr)) | |
872 | must_append_controller(klist, nlist, &aret, tok); | |
873 | ||
874 | return aret; | |
875 | } | |
876 | ||
877 | /* Check if a cgroupfs v2 controller is present in the string @cgline. */ | |
878 | static bool cgv1_controller_in_clist(char *cgline, char *c) | |
879 | { | |
880 | size_t len; | |
881 | char *tok, *eol, *tmp; | |
882 | char *saveptr = NULL; | |
883 | ||
884 | eol = strchr(cgline, ':'); | |
885 | if (!eol) | |
886 | return false; | |
887 | ||
888 | len = eol - cgline; | |
889 | tmp = alloca(len + 1); | |
890 | memcpy(tmp, cgline, len); | |
891 | tmp[len] = '\0'; | |
892 | ||
893 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
894 | tok = strtok_r(NULL, ",", &saveptr)) { | |
895 | if (strcmp(tok, c) == 0) | |
df54106a SH |
896 | return true; |
897 | } | |
898 | return false; | |
899 | } | |
900 | ||
e65cfafc CB |
901 | /* Get current cgroup from the /proc/<pid>/cgroup file passed in via @basecginfo |
902 | * of a given cgv1 controller passed in via @controller. | |
df54106a | 903 | */ |
e65cfafc | 904 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller) |
df54106a | 905 | { |
e65cfafc CB |
906 | char *p; |
907 | ||
908 | p = basecginfo; | |
909 | ||
910 | while (true) { | |
911 | p = strchr(p, ':'); | |
912 | if (!p) | |
913 | return NULL; | |
914 | p++; | |
915 | ||
916 | if (cgv1_controller_in_clist(p, controller)) { | |
917 | p = strchr(p, ':'); | |
918 | if (!p) | |
919 | return NULL; | |
920 | p++; | |
921 | ||
922 | return copy_to_eol(p); | |
923 | } | |
924 | ||
925 | p = strchr(p, '\n'); | |
926 | if (!p) | |
927 | return NULL; | |
928 | p++; | |
df54106a | 929 | } |
e65cfafc CB |
930 | |
931 | return NULL; | |
df54106a SH |
932 | } |
933 | ||
e65cfafc CB |
934 | /* Remove /init.scope from string @cg. This will mostly affect systemd-based |
935 | * systems. | |
936 | */ | |
df54106a | 937 | #define INIT_SCOPE "/init.scope" |
e65cfafc | 938 | static void cg_systemd_prune_init_scope(char *cg) |
df54106a SH |
939 | { |
940 | char *point; | |
941 | ||
942 | if (!cg) | |
943 | return; | |
944 | ||
e65cfafc CB |
945 | point = cg + strlen(cg) - strlen(INIT_SCOPE); |
946 | if (point < cg) | |
df54106a | 947 | return; |
826297d7 | 948 | |
df54106a SH |
949 | if (strcmp(point, INIT_SCOPE) == 0) { |
950 | if (point == cg) | |
e65cfafc | 951 | *(point + 1) = '\0'; |
df54106a SH |
952 | else |
953 | *point = '\0'; | |
954 | } | |
955 | } | |
956 | ||
e65cfafc CB |
957 | /* Add new info about a mounted cgroupfs v1 hierarchy. Includes the controllers |
958 | * mounted into that hierarchy (e.g. cpu,cpuacct), the mountpoint of that | |
959 | * hierarchy (/sys/fs/cgroup/<controller>, the base cgroup of the current | |
960 | * process gathered from /proc/self/cgroup, and the init cgroup of PID1 gathered | |
961 | * from /proc/1/cgroup. | |
962 | */ | |
963 | static void cgv1_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup) | |
964 | { | |
965 | struct cgv1_hierarchy *new; | |
966 | int newentry; | |
967 | ||
968 | new = must_alloc(sizeof(*new)); | |
969 | new->controllers = clist; | |
970 | new->mountpoint = mountpoint; | |
971 | new->base_cgroup = base_cgroup; | |
972 | new->fullcgpath = NULL; | |
973 | new->create_rw_cgroup = false; | |
974 | new->init_cgroup = init_cgroup; | |
975 | new->systemd_user_slice = false; | |
976 | ||
977 | newentry = append_null_to_list((void ***)&cgv1_hierarchies); | |
978 | cgv1_hierarchies[newentry] = new; | |
979 | } | |
980 | ||
981 | /* Add new info about the mounted cgroupfs v2 hierarchy. Can (but doesn't | |
982 | * currently) include the controllers mounted into the hierarchy (e.g. memory, | |
983 | * pids, blkio), the mountpoint of that hierarchy (Should usually be | |
984 | * /sys/fs/cgroup but some init systems seems to think it might be a good idea | |
985 | * to also mount empty cgroupfs v2 hierarchies at /sys/fs/cgroup/systemd.), the | |
986 | * base cgroup of the current process gathered from /proc/self/cgroup, and the | |
987 | * init cgroup of PID1 gathered from /proc/1/cgroup. | |
988 | */ | |
989 | static void cgv2_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup, bool systemd_user_slice) | |
990 | { | |
991 | struct cgv2_hierarchy *new; | |
992 | int newentry; | |
993 | ||
994 | new = must_alloc(sizeof(*new)); | |
995 | new->controllers = clist; | |
996 | new->mountpoint = mountpoint; | |
997 | new->base_cgroup = base_cgroup; | |
998 | new->fullcgpath = NULL; | |
999 | new->create_rw_cgroup = false; | |
1000 | new->init_cgroup = init_cgroup; | |
1001 | new->systemd_user_slice = systemd_user_slice; | |
1002 | ||
1003 | newentry = append_null_to_list((void ***)&cgv2_hierarchies); | |
1004 | cgv2_hierarchies[newentry] = new; | |
1005 | } | |
1006 | ||
1007 | /* In Ubuntu 14.04, the paths created for us were | |
1008 | * '/user/$uid.user/$something.session' This can be merged better with | |
1009 | * systemd_created_slice_for_us(), but keeping it separate makes it easier to | |
1010 | * reason about the correctness. | |
1011 | */ | |
1012 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid) | |
1013 | { | |
1014 | char *p; | |
1015 | size_t len; | |
1016 | int id; | |
1017 | char *copy = NULL; | |
1018 | bool bret = false; | |
1019 | ||
1020 | copy = must_copy_string(in); | |
1021 | if (strlen(copy) < strlen("/user/1.user/1.session")) | |
1022 | goto cleanup; | |
1023 | p = copy + strlen(copy) - 1; | |
1024 | ||
1025 | /* skip any trailing '/' (shouldn't be any, but be sure) */ | |
1026 | while (p >= copy && *p == '/') | |
1027 | *(p--) = '\0'; | |
1028 | if (p < copy) | |
1029 | goto cleanup; | |
1030 | ||
1031 | /* Get last path element */ | |
1032 | while (p >= copy && *p != '/') | |
1033 | p--; | |
1034 | if (p < copy) | |
1035 | goto cleanup; | |
1036 | /* make sure it is something.session */ | |
1037 | len = strlen(p + 1); | |
1038 | if (len < strlen("1.session") || | |
1039 | strncmp(p + 1 + len - 8, ".session", 8) != 0) | |
1040 | goto cleanup; | |
1041 | ||
1042 | /* ok last path piece checks out, now check the second to last */ | |
1043 | *(p + 1) = '\0'; | |
1044 | while (p >= copy && *(--p) != '/') | |
1045 | ; | |
1046 | if (sscanf(p + 1, "%d.user/", &id) != 1) | |
1047 | goto cleanup; | |
1048 | ||
1049 | if (id != (int)uid) | |
1050 | goto cleanup; | |
1051 | ||
1052 | bret = true; | |
1053 | ||
1054 | cleanup: | |
1055 | free(copy); | |
1056 | return bret; | |
1057 | } | |
1058 | ||
1059 | /* So long as our path relative to init starts with /user.slice/user-$uid.slice, | |
1060 | * assume it belongs to $uid and chown it | |
1061 | */ | |
1062 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
1063 | const char *init_cgroup, uid_t uid) | |
1064 | { | |
1065 | int ret; | |
1066 | char buf[100]; | |
1067 | size_t curlen, initlen; | |
1068 | ||
1069 | curlen = strlen(base_cgroup); | |
1070 | initlen = strlen(init_cgroup); | |
1071 | if (curlen <= initlen) | |
1072 | return false; | |
1073 | ||
1074 | if (strncmp(base_cgroup, init_cgroup, initlen) != 0) | |
1075 | return false; | |
1076 | ||
1077 | ret = snprintf(buf, 100, "/user.slice/user-%d.slice/", (int)uid); | |
1078 | if (ret < 0 || ret >= 100) | |
1079 | return false; | |
1080 | ||
1081 | if (initlen == 1) | |
1082 | initlen = 0; // skip the '/' | |
1083 | ||
1084 | return strncmp(base_cgroup + initlen, buf, strlen(buf)) == 0; | |
1085 | } | |
1086 | ||
1087 | /* The systemd-created path is: user-$uid.slice/session-c$session.scope. If that | |
1088 | * is not the end of our systemd path, then we're not part of the PAM call that | |
1089 | * created that path. | |
1090 | * | |
1091 | * The last piece is chowned to $uid, the user- part not. | |
1092 | * Note: If the user creates paths that look like what we're looking for to | |
1093 | * 'fool' us, either | |
1094 | * - they fool us, we create new cgroups, and they get auto-logged-out. | |
1095 | * - they fool a root sudo, systemd cgroup is not changed but chowned, and they | |
1096 | * lose ownership of their cgroups | |
1097 | */ | |
1098 | static bool cg_systemd_created_user_slice(const char *base_cgroup, | |
1099 | const char *init_cgroup, | |
1100 | const char *in, uid_t uid) | |
1101 | { | |
1102 | char *p; | |
1103 | size_t len; | |
1104 | int id; | |
1105 | char *copy = NULL; | |
1106 | bool bret = false; | |
1107 | ||
1108 | copy = must_copy_string(in); | |
1109 | ||
1110 | /* An old version of systemd has already created a cgroup for us. */ | |
1111 | if (cg_systemd_under_user_slice_1(in, uid)) | |
1112 | goto succeed; | |
1113 | ||
1114 | /* A new version of systemd has already created a cgroup for us. */ | |
1115 | if (cg_systemd_under_user_slice_2(base_cgroup, init_cgroup, uid)) | |
1116 | goto succeed; | |
1117 | ||
1118 | if (strlen(copy) < strlen("/user-0.slice/session-0.scope")) | |
1119 | goto cleanup; | |
1120 | ||
1121 | p = copy + strlen(copy) - 1; | |
1122 | /* Skip any trailing '/' (shouldn't be any, but be sure). */ | |
1123 | while (p >= copy && *p == '/') | |
1124 | *(p--) = '\0'; | |
1125 | ||
1126 | if (p < copy) | |
1127 | goto cleanup; | |
1128 | ||
1129 | /* Get last path element */ | |
1130 | while (p >= copy && *p != '/') | |
1131 | p--; | |
1132 | ||
1133 | if (p < copy) | |
1134 | goto cleanup; | |
1135 | ||
1136 | /* Make sure it is session-something.scope. */ | |
1137 | len = strlen(p + 1); | |
1138 | if (strncmp(p + 1, "session-", strlen("session-")) != 0 || | |
1139 | strncmp(p + 1 + len - 6, ".scope", 6) != 0) | |
1140 | goto cleanup; | |
1141 | ||
1142 | /* Ok last path piece checks out, now check the second to last. */ | |
1143 | *(p + 1) = '\0'; | |
1144 | while (p >= copy && *(--p) != '/') | |
1145 | ; | |
1146 | ||
1147 | if (sscanf(p + 1, "user-%d.slice/", &id) != 1) | |
1148 | goto cleanup; | |
1149 | ||
1150 | if (id != (int)uid) | |
1151 | goto cleanup; | |
1152 | ||
1153 | succeed: | |
1154 | bret = true; | |
1155 | cleanup: | |
1156 | free(copy); | |
1157 | return bret; | |
1158 | } | |
1159 | ||
1160 | /* Chown existing cgroup that systemd has already created for us. */ | |
1161 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
1162 | const char *base_cgroup, uid_t uid, | |
1163 | gid_t gid, bool systemd_user_slice) | |
1164 | { | |
1165 | char *path; | |
1166 | ||
1167 | if (!systemd_user_slice) | |
1168 | return false; | |
1169 | ||
1170 | path = must_make_path(mountpoint, base_cgroup, NULL); | |
1171 | ||
1172 | /* A cgroup within name=systemd has already been created. So we only | |
1173 | * need to chown it. | |
1174 | */ | |
1175 | if (chown(path, uid, gid) < 0) | |
b36273fa CB |
1176 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", |
1177 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
1178 | lxcfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
e65cfafc CB |
1179 | |
1180 | free(path); | |
1181 | return true; | |
1182 | } | |
1183 | ||
1184 | /* Detect and store information about cgroupfs v1 hierarchies. */ | |
1185 | static bool cgv1_init(uid_t uid, gid_t gid) | |
df54106a SH |
1186 | { |
1187 | FILE *f; | |
e65cfafc CB |
1188 | struct cgv1_hierarchy **it; |
1189 | char *basecginfo; | |
df54106a | 1190 | char *line = NULL; |
e65cfafc | 1191 | char **klist = NULL, **nlist = NULL; |
df54106a | 1192 | size_t len = 0; |
df54106a | 1193 | |
e65cfafc CB |
1194 | basecginfo = read_file("/proc/self/cgroup"); |
1195 | if (!basecginfo) | |
1196 | return false; | |
1197 | ||
1198 | f = fopen("/proc/self/mountinfo", "r"); | |
a6e9ec7d CB |
1199 | if (!f) { |
1200 | free(basecginfo); | |
df54106a | 1201 | return false; |
a6e9ec7d | 1202 | } |
e65cfafc CB |
1203 | |
1204 | cgv1_get_controllers(&klist, &nlist); | |
1205 | ||
df54106a | 1206 | while (getline(&line, &len, f) != -1) { |
e65cfafc CB |
1207 | char **controller_list = NULL; |
1208 | char *mountpoint, *base_cgroup; | |
1209 | ||
a6e9ec7d | 1210 | if (is_lxcfs(line) || !is_cgv1(line)) |
e65cfafc CB |
1211 | continue; |
1212 | ||
1213 | controller_list = cgv1_get_proc_mountinfo_controllers(klist, nlist, line); | |
1214 | if (!controller_list) | |
1215 | continue; | |
1216 | ||
1217 | if (cgv1_controller_list_is_dup(cgv1_hierarchies, | |
1218 | controller_list)) { | |
1219 | free(controller_list); | |
1220 | continue; | |
df54106a | 1221 | } |
e65cfafc CB |
1222 | |
1223 | mountpoint = get_mountpoint(line); | |
1224 | if (!mountpoint) { | |
1225 | free_string_list(controller_list); | |
1226 | continue; | |
df54106a | 1227 | } |
e65cfafc CB |
1228 | |
1229 | base_cgroup = cgv1_get_current_cgroup(basecginfo, controller_list[0]); | |
1230 | if (!base_cgroup) { | |
1231 | free_string_list(controller_list); | |
1232 | free(mountpoint); | |
1233 | continue; | |
edd25678 | 1234 | } |
e65cfafc CB |
1235 | trim(base_cgroup); |
1236 | lxcfs_debug("Detected cgroupfs v1 controller \"%s\" with " | |
1237 | "mountpoint \"%s\" and cgroup \"%s\".\n", | |
1238 | controller_list[0], mountpoint, base_cgroup); | |
1239 | cgv1_add_controller(controller_list, mountpoint, base_cgroup, | |
1240 | NULL); | |
df54106a | 1241 | } |
e65cfafc CB |
1242 | free_string_list(klist); |
1243 | free_string_list(nlist); | |
1244 | free(basecginfo); | |
df54106a | 1245 | fclose(f); |
c65c5956 | 1246 | free(line); |
df54106a | 1247 | |
e65cfafc CB |
1248 | /* Retrieve init cgroup path for all controllers. */ |
1249 | basecginfo = read_file("/proc/1/cgroup"); | |
1250 | if (!basecginfo) | |
1251 | return false; | |
df54106a | 1252 | |
e65cfafc CB |
1253 | for (it = cgv1_hierarchies; it && *it; it++) { |
1254 | if ((*it)->controllers) { | |
1255 | char *init_cgroup, *user_slice; | |
a6e9ec7d CB |
1256 | /* We've already stored the controller and received its |
1257 | * current cgroup. If we now fail to retrieve its init | |
1258 | * cgroup, we should probably fail. | |
1259 | */ | |
e65cfafc | 1260 | init_cgroup = cgv1_get_current_cgroup(basecginfo, (*it)->controllers[0]); |
a6e9ec7d CB |
1261 | if (!init_cgroup) { |
1262 | free(basecginfo); | |
1263 | return false; | |
1264 | } | |
e65cfafc CB |
1265 | cg_systemd_prune_init_scope(init_cgroup); |
1266 | (*it)->init_cgroup = init_cgroup; | |
1267 | lxcfs_debug("cgroupfs v1 controller \"%s\" has init " | |
1268 | "cgroup \"%s\".\n", | |
1269 | (*(*it)->controllers), init_cgroup); | |
1270 | /* Check whether systemd has already created a cgroup | |
1271 | * for us. | |
1272 | */ | |
1273 | user_slice = must_make_path((*it)->mountpoint, (*it)->base_cgroup, NULL); | |
1274 | if (cg_systemd_created_user_slice((*it)->base_cgroup, (*it)->init_cgroup, user_slice, uid)) | |
1275 | (*it)->systemd_user_slice = true; | |
df54106a SH |
1276 | } |
1277 | } | |
e65cfafc CB |
1278 | free(basecginfo); |
1279 | ||
1280 | return true; | |
df54106a | 1281 | } |
e65cfafc CB |
1282 | |
1283 | /* __typeof__ should be safe to use with all compilers. */ | |
1284 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
1285 | /* Check whether given mountpoint has mount type specified via @magic_val. */ | |
1286 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
1287 | { | |
1288 | return (fs->f_type == (fs_type_magic)magic_val); | |
1289 | } | |
1290 | ||
1291 | /* Check whether @path is a cgroupfs v1 or cgroupfs v2 mount. Returns -1 if | |
1292 | * statfs fails. If @path is null /sys/fs/cgroup is checked. | |
df54106a | 1293 | */ |
e65cfafc | 1294 | static int cg_get_version_of_mntpt(const char *path) |
df54106a | 1295 | { |
e65cfafc CB |
1296 | int ret; |
1297 | struct statfs sb; | |
1298 | ||
1299 | if (path) | |
1300 | ret = statfs(path, &sb); | |
1301 | else | |
1302 | ret = statfs("/sys/fs/cgroup", &sb); | |
1303 | ||
1304 | if (ret < 0) | |
1305 | return -1; | |
1306 | ||
1307 | if (has_fs_type(&sb, CGROUP_SUPER_MAGIC)) | |
1308 | return 1; | |
1309 | else if (has_fs_type(&sb, CGROUP2_SUPER_MAGIC)) | |
1310 | return 2; | |
1311 | ||
1312 | return 0; | |
1313 | } | |
1314 | ||
1315 | /* Detect and store information about the cgroupfs v2 hierarchy. Currently only | |
1316 | * deals with the empty v2 hierachy as we do not retrieve enabled controllers. | |
1317 | */ | |
1318 | static bool cgv2_init(uid_t uid, gid_t gid) | |
1319 | { | |
1320 | char *mountpoint; | |
e65cfafc CB |
1321 | FILE *f = NULL; |
1322 | char *current_cgroup = NULL, *init_cgroup = NULL; | |
1323 | char * line = NULL; | |
df54106a | 1324 | size_t len = 0; |
e4992b3e | 1325 | int ret = false; |
df54106a | 1326 | |
e65cfafc CB |
1327 | current_cgroup = cgv2_get_current_cgroup(getpid()); |
1328 | if (!current_cgroup) { | |
1329 | /* No v2 hierarchy present. We're done. */ | |
e4992b3e | 1330 | ret = true; |
e65cfafc CB |
1331 | goto cleanup; |
1332 | } | |
1333 | ||
1334 | init_cgroup = cgv2_get_current_cgroup(1); | |
1335 | if (!init_cgroup) { | |
1336 | /* If we're here and didn't fail already above, then something's | |
1337 | * certainly wrong, so error this time. | |
1338 | */ | |
1339 | goto cleanup; | |
1340 | } | |
1341 | cg_systemd_prune_init_scope(init_cgroup); | |
1342 | ||
1343 | /* Check if the v2 hierarchy is mounted at its standard location. | |
1344 | * If so we can skip the rest of the work here. Although the unified | |
1345 | * hierarchy can be mounted multiple times, each of those mountpoints | |
1346 | * will expose identical information. | |
1347 | */ | |
1348 | if (cg_get_version_of_mntpt("/sys/fs/cgroup") == 2) { | |
1349 | char *user_slice; | |
1350 | bool has_user_slice = false; | |
1351 | ||
1352 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
1353 | if (!mountpoint) | |
1354 | goto cleanup; | |
1355 | ||
1356 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); | |
1357 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1358 | has_user_slice = true; | |
1359 | free(user_slice); | |
1360 | ||
1361 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); | |
1362 | ||
e4992b3e | 1363 | ret = true; |
e65cfafc CB |
1364 | goto cleanup; |
1365 | } | |
1366 | ||
1367 | f = fopen("/proc/self/mountinfo", "r"); | |
df54106a | 1368 | if (!f) |
e4992b3e | 1369 | goto cleanup; |
e65cfafc CB |
1370 | |
1371 | /* we support simple cgroup mounts and lxcfs mounts */ | |
df54106a | 1372 | while (getline(&line, &len, f) != -1) { |
e65cfafc CB |
1373 | char *user_slice; |
1374 | bool has_user_slice = false; | |
1375 | if (!is_cgv2(line)) | |
1376 | continue; | |
1377 | ||
1378 | mountpoint = get_mountpoint(line); | |
1379 | if (!mountpoint) | |
1380 | continue; | |
df54106a | 1381 | |
e65cfafc CB |
1382 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); |
1383 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1384 | has_user_slice = true; | |
1385 | free(user_slice); | |
df54106a | 1386 | |
e65cfafc CB |
1387 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); |
1388 | /* Although the unified hierarchy can be mounted multiple times, | |
1389 | * each of those mountpoints will expose identical information. | |
1390 | * So let the first mountpoint we find, win. | |
1391 | */ | |
1392 | break; | |
df54106a SH |
1393 | } |
1394 | ||
e65cfafc CB |
1395 | lxcfs_debug("Detected cgroupfs v2 hierarchy at mountpoint \"%s\" with " |
1396 | "current cgroup \"%s\" and init cgroup \"%s\".\n", | |
1397 | mountpoint, current_cgroup, init_cgroup); | |
df54106a | 1398 | |
e65cfafc CB |
1399 | cleanup: |
1400 | if (f) | |
1401 | fclose(f); | |
1402 | free(line); | |
df54106a | 1403 | |
e4992b3e | 1404 | return ret; |
df54106a SH |
1405 | } |
1406 | ||
e65cfafc CB |
1407 | /* Detect and store information about mounted cgroupfs v1 hierarchies and the |
1408 | * cgroupfs v2 hierarchy. | |
1409 | * Detect whether we are on a pure cgroupfs v1, cgroupfs v2, or mixed system, | |
1410 | * where some controllers are mounted into their standard cgroupfs v1 locations | |
1411 | * (/sys/fs/cgroup/<controller>) and others are mounted into the cgroupfs v2 | |
1412 | * hierarchy (/sys/fs/cgroup). | |
4deb6092 | 1413 | */ |
e65cfafc | 1414 | static bool cg_init(uid_t uid, gid_t gid) |
4deb6092 | 1415 | { |
e65cfafc | 1416 | if (!cgv1_init(uid, gid)) |
4deb6092 SH |
1417 | return false; |
1418 | ||
e65cfafc | 1419 | if (!cgv2_init(uid, gid)) |
4deb6092 SH |
1420 | return false; |
1421 | ||
e65cfafc CB |
1422 | if (cgv1_hierarchies && cgv2_hierarchies) { |
1423 | cg_mount_mode = CGROUP_MIXED; | |
1424 | lxcfs_debug("%s\n", "Detected cgroupfs v1 and v2 hierarchies."); | |
1425 | } else if (cgv1_hierarchies && !cgv2_hierarchies) { | |
1426 | cg_mount_mode = CGROUP_PURE_V1; | |
1427 | lxcfs_debug("%s\n", "Detected cgroupfs v1 hierarchies."); | |
1428 | } else if (cgv2_hierarchies && !cgv1_hierarchies) { | |
1429 | cg_mount_mode = CGROUP_PURE_V2; | |
1430 | lxcfs_debug("%s\n", "Detected cgroupfs v2 hierarchies."); | |
1431 | } else { | |
1432 | cg_mount_mode = CGROUP_UNKNOWN; | |
1433 | mysyslog(LOG_ERR, "Could not detect cgroupfs hierarchy.\n", NULL); | |
1434 | } | |
4deb6092 | 1435 | |
e65cfafc | 1436 | if (cg_mount_mode == CGROUP_UNKNOWN) |
4deb6092 SH |
1437 | return false; |
1438 | ||
1439 | return true; | |
1440 | } | |
1441 | ||
e65cfafc CB |
1442 | /* Try to move/migrate us into @cgroup in a cgroupfs v1 hierarchy. */ |
1443 | static bool cgv1_enter(const char *cgroup) | |
475a859c | 1444 | { |
e65cfafc CB |
1445 | struct cgv1_hierarchy **it; |
1446 | ||
1447 | for (it = cgv1_hierarchies; it && *it; it++) { | |
e65cfafc CB |
1448 | char **controller; |
1449 | bool entered = false; | |
1450 | ||
1451 | if (!(*it)->controllers || !(*it)->mountpoint || | |
1452 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
1453 | continue; | |
1454 | ||
1455 | for (controller = (*it)->controllers; controller && *controller; | |
1456 | controller++) { | |
a6e9ec7d | 1457 | char *path; |
e65cfafc | 1458 | |
a403c004 CB |
1459 | /* We've already been placed in a user slice, so we |
1460 | * don't need to enter the cgroup again. | |
1461 | */ | |
1462 | if ((*it)->systemd_user_slice) { | |
1463 | entered = true; | |
1464 | break; | |
1465 | } | |
e65cfafc CB |
1466 | |
1467 | path = must_make_path((*it)->mountpoint, | |
1468 | (*it)->init_cgroup, | |
1469 | cgroup, | |
1470 | "/cgroup.procs", | |
1471 | NULL); | |
1472 | if (!file_exists(path)) { | |
1473 | free(path); | |
1474 | path = must_make_path((*it)->mountpoint, | |
1475 | (*it)->init_cgroup, | |
1476 | cgroup, | |
1477 | "/tasks", | |
1478 | NULL); | |
1479 | } | |
1480 | lxcfs_debug("Attempting to enter cgroupfs v1 hierarchy in \"%s\" cgroup.\n", path); | |
1481 | entered = write_int(path, (int)getpid()); | |
a6e9ec7d CB |
1482 | if (entered) { |
1483 | free(path); | |
e65cfafc | 1484 | break; |
a6e9ec7d | 1485 | } |
e65cfafc | 1486 | lxcfs_debug("Failed to enter cgroupfs v1 hierarchy in \"%s\" cgroup.\n", path); |
a6e9ec7d | 1487 | free(path); |
e65cfafc CB |
1488 | } |
1489 | if (!entered) | |
1490 | return false; | |
1491 | } | |
475a859c | 1492 | |
e65cfafc | 1493 | return true; |
475a859c SH |
1494 | } |
1495 | ||
e65cfafc CB |
1496 | /* Try to move/migrate us into @cgroup in the cgroupfs v2 hierarchy. */ |
1497 | static bool cgv2_enter(const char *cgroup) | |
78a2a9f3 | 1498 | { |
e65cfafc CB |
1499 | struct cgv2_hierarchy *v2; |
1500 | char *path; | |
1501 | bool entered = false; | |
78a2a9f3 | 1502 | |
e65cfafc | 1503 | if (!cgv2_hierarchies) |
4deb6092 SH |
1504 | return true; |
1505 | ||
e65cfafc | 1506 | v2 = *cgv2_hierarchies; |
475a859c | 1507 | |
e65cfafc | 1508 | if (!v2->mountpoint || !v2->base_cgroup) |
78a2a9f3 SH |
1509 | return false; |
1510 | ||
e65cfafc CB |
1511 | if (!v2->create_rw_cgroup || v2->systemd_user_slice) |
1512 | return true; | |
78a2a9f3 | 1513 | |
423a3b4f | 1514 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, "/cgroup.procs", NULL); |
e65cfafc CB |
1515 | lxcfs_debug("Attempting to enter cgroupfs v2 hierarchy in cgroup \"%s\".\n", path); |
1516 | entered = write_int(path, (int)getpid()); | |
1517 | if (!entered) { | |
1518 | lxcfs_debug("Failed to enter cgroupfs v2 hierarchy in cgroup \"%s\".\n", path); | |
1519 | free(path); | |
78a2a9f3 | 1520 | return false; |
e65cfafc | 1521 | } |
78a2a9f3 | 1522 | |
e65cfafc | 1523 | free(path); |
78a2a9f3 SH |
1524 | |
1525 | return true; | |
1526 | } | |
edd25678 | 1527 | |
e65cfafc CB |
1528 | /* Wrapper around cgv{1,2}_enter(). */ |
1529 | static bool cg_enter(const char *cgroup) | |
1530 | { | |
1531 | if (!cgv1_enter(cgroup)) { | |
1532 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to enter cgroups.\n", NULL); | |
edd25678 | 1533 | return false; |
e65cfafc | 1534 | } |
edd25678 | 1535 | |
e65cfafc CB |
1536 | if (!cgv2_enter(cgroup)) { |
1537 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to enter cgroups.\n", NULL); | |
78a2a9f3 SH |
1538 | return false; |
1539 | } | |
1540 | ||
edd25678 SH |
1541 | return true; |
1542 | } | |
1543 | ||
e65cfafc CB |
1544 | /* Escape to root cgroup in all detected cgroupfs v1 hierarchies. */ |
1545 | static void cgv1_escape(void) | |
df54106a | 1546 | { |
04742595 CB |
1547 | struct cgv1_hierarchy **it; |
1548 | ||
1549 | /* In case systemd hasn't already placed us in a user slice for the | |
1550 | * cpuset v1 controller we will reside in the root cgroup. This means | |
1551 | * that cgroup.clone_children will not have been initialized for us so | |
1552 | * we need to do it. | |
1553 | */ | |
1554 | for (it = cgv1_hierarchies; it && *it; it++) | |
1555 | if (!cgv1_handle_root_cpuset_hierarchy(*it)) | |
1556 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to initialize cpuset.\n", NULL); | |
1557 | ||
e65cfafc CB |
1558 | if (!cgv1_enter("/")) |
1559 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to escape to init's cgroup.\n", NULL); | |
1560 | } | |
edd25678 | 1561 | |
e65cfafc CB |
1562 | /* Escape to root cgroup in the cgroupfs v2 hierarchy. */ |
1563 | static void cgv2_escape(void) | |
1564 | { | |
1565 | if (!cgv2_enter("/")) | |
1566 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to escape to init's cgroup.\n", NULL); | |
1567 | } | |
edd25678 | 1568 | |
e65cfafc CB |
1569 | /* Wrapper around cgv{1,2}_escape(). */ |
1570 | static void cg_escape(void) | |
1571 | { | |
1572 | cgv1_escape(); | |
1573 | cgv2_escape(); | |
df54106a SH |
1574 | } |
1575 | ||
e65cfafc CB |
1576 | /* Get uid and gid for @user. */ |
1577 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid) | |
df54106a | 1578 | { |
e65cfafc | 1579 | struct passwd *pwent; |
df54106a | 1580 | |
e65cfafc CB |
1581 | pwent = getpwnam(user); |
1582 | if (!pwent) | |
1583 | return false; | |
df54106a | 1584 | |
e65cfafc CB |
1585 | *uid = pwent->pw_uid; |
1586 | *gid = pwent->pw_gid; | |
df54106a | 1587 | |
e65cfafc | 1588 | return true; |
df54106a SH |
1589 | } |
1590 | ||
3145d497 CB |
1591 | /* Check if cgroup belongs to our uid and gid. If so, reuse it. */ |
1592 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid) | |
1593 | { | |
1594 | struct stat statbuf; | |
1595 | ||
1596 | if (stat(path, &statbuf) < 0) | |
1597 | return false; | |
1598 | ||
1599 | if (!(statbuf.st_uid == uid) || !(statbuf.st_gid == gid)) | |
1600 | return false; | |
1601 | ||
1602 | return true; | |
1603 | } | |
1604 | ||
04742595 CB |
1605 | /* Create cpumask from cpulist aka turn: |
1606 | * | |
1607 | * 0,2-3 | |
1608 | * | |
1609 | * into bit array | |
1610 | * | |
1611 | * 1 0 1 1 | |
1612 | */ | |
1613 | static uint32_t *cg_cpumask(char *buf, size_t nbits) | |
1614 | { | |
1615 | char *token; | |
1616 | char *saveptr = NULL; | |
1617 | size_t arrlen = BITS_TO_LONGS(nbits); | |
1618 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
1619 | if (!bitarr) | |
1620 | return NULL; | |
1621 | ||
1622 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
1623 | errno = 0; | |
1624 | unsigned start = strtoul(token, NULL, 0); | |
1625 | unsigned end = start; | |
1626 | ||
1627 | char *range = strchr(token, '-'); | |
1628 | if (range) | |
1629 | end = strtoul(range + 1, NULL, 0); | |
1630 | if (!(start <= end)) { | |
1631 | free(bitarr); | |
1632 | return NULL; | |
1633 | } | |
1634 | ||
1635 | if (end >= nbits) { | |
1636 | free(bitarr); | |
1637 | return NULL; | |
1638 | } | |
1639 | ||
1640 | while (start <= end) | |
1641 | set_bit(start++, bitarr); | |
1642 | } | |
1643 | ||
1644 | return bitarr; | |
1645 | } | |
1646 | ||
808fd1ef | 1647 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix) |
04742595 CB |
1648 | { |
1649 | char *result; | |
1650 | char **p; | |
1651 | size_t sep_len = strlen(sep); | |
1652 | size_t result_len = use_as_prefix * sep_len; | |
1653 | ||
bfd723ff CB |
1654 | if (!parts) |
1655 | return NULL; | |
1656 | ||
04742595 CB |
1657 | /* calculate new string length */ |
1658 | for (p = (char **)parts; *p; p++) | |
1659 | result_len += (p > (char **)parts) * sep_len + strlen(*p); | |
1660 | ||
3acf9e94 | 1661 | result = calloc(result_len + 1, sizeof(char)); |
04742595 CB |
1662 | if (!result) |
1663 | return NULL; | |
1664 | ||
1665 | if (use_as_prefix) | |
1666 | strcpy(result, sep); | |
1667 | for (p = (char **)parts; *p; p++) { | |
1668 | if (p > (char **)parts) | |
1669 | strcat(result, sep); | |
1670 | strcat(result, *p); | |
1671 | } | |
1672 | ||
1673 | return result; | |
1674 | } | |
1675 | ||
1676 | /* The largest integer that can fit into long int is 2^64. This is a | |
1677 | * 20-digit number. | |
1678 | */ | |
1679 | #define __IN_TO_STR_LEN 21 | |
1680 | /* Turn cpumask into simple, comma-separated cpulist. */ | |
1681 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
1682 | { | |
1683 | size_t i; | |
1684 | int ret; | |
1685 | char numstr[__IN_TO_STR_LEN] = {0}; | |
1686 | char **cpulist = NULL; | |
1687 | ||
1688 | for (i = 0; i <= nbits; i++) { | |
1689 | if (is_set(i, bitarr)) { | |
1690 | ret = snprintf(numstr, __IN_TO_STR_LEN, "%zu", i); | |
1691 | if (ret < 0 || (size_t)ret >= __IN_TO_STR_LEN) { | |
1692 | free_string_list(cpulist); | |
1693 | return NULL; | |
1694 | } | |
1695 | must_append_string(&cpulist, numstr); | |
1696 | } | |
1697 | } | |
1698 | return string_join(",", (const char **)cpulist, false); | |
1699 | } | |
1700 | ||
1701 | static ssize_t cg_get_max_cpus(char *cpulist) | |
1702 | { | |
1703 | char *c1, *c2; | |
1704 | char *maxcpus = cpulist; | |
1705 | size_t cpus = 0; | |
1706 | ||
1707 | c1 = strrchr(maxcpus, ','); | |
1708 | if (c1) | |
1709 | c1++; | |
1710 | ||
1711 | c2 = strrchr(maxcpus, '-'); | |
1712 | if (c2) | |
1713 | c2++; | |
1714 | ||
1715 | if (!c1 && !c2) | |
1716 | c1 = maxcpus; | |
04742595 CB |
1717 | else if (c1 < c2) |
1718 | c1 = c2; | |
04742595 CB |
1719 | |
1720 | /* If the above logic is correct, c1 should always hold a valid string | |
1721 | * here. | |
1722 | */ | |
1723 | ||
1724 | errno = 0; | |
1725 | cpus = strtoul(c1, NULL, 0); | |
1726 | if (errno != 0) | |
1727 | return -1; | |
1728 | ||
1729 | return cpus; | |
1730 | } | |
1731 | ||
808fd1ef | 1732 | static ssize_t write_nointr(int fd, const void* buf, size_t count) |
04742595 CB |
1733 | { |
1734 | ssize_t ret; | |
1735 | again: | |
1736 | ret = write(fd, buf, count); | |
1737 | if (ret < 0 && errno == EINTR) | |
1738 | goto again; | |
1739 | return ret; | |
1740 | } | |
1741 | ||
808fd1ef | 1742 | static int write_to_file(const char *filename, const void* buf, size_t count, bool add_newline) |
04742595 CB |
1743 | { |
1744 | int fd, saved_errno; | |
1745 | ssize_t ret; | |
1746 | ||
1747 | fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, 0666); | |
1748 | if (fd < 0) | |
1749 | return -1; | |
1750 | ret = write_nointr(fd, buf, count); | |
1751 | if (ret < 0) | |
1752 | goto out_error; | |
1753 | if ((size_t)ret != count) | |
1754 | goto out_error; | |
1755 | if (add_newline) { | |
1756 | ret = write_nointr(fd, "\n", 1); | |
1757 | if (ret != 1) | |
1758 | goto out_error; | |
1759 | } | |
1760 | close(fd); | |
1761 | return 0; | |
1762 | ||
1763 | out_error: | |
1764 | saved_errno = errno; | |
1765 | close(fd); | |
1766 | errno = saved_errno; | |
1767 | return -1; | |
1768 | } | |
1769 | ||
808fd1ef | 1770 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
04742595 CB |
1771 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized) |
1772 | { | |
1773 | char *lastslash, *fpath, oldv; | |
1774 | int ret; | |
1775 | ssize_t i; | |
1776 | ||
1777 | ssize_t maxposs = 0, maxisol = 0; | |
1778 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
1779 | uint32_t *possmask = NULL, *isolmask = NULL; | |
808fd1ef | 1780 | bool bret = false, flipped_bit = false; |
04742595 CB |
1781 | |
1782 | lastslash = strrchr(path, '/'); | |
1783 | if (!lastslash) { // bug... this shouldn't be possible | |
808fd1ef | 1784 | lxcfs_debug("Invalid path: %s.\n", path); |
04742595 CB |
1785 | return bret; |
1786 | } | |
1787 | oldv = *lastslash; | |
1788 | *lastslash = '\0'; | |
1789 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
1790 | posscpus = read_file(fpath); | |
808fd1ef CB |
1791 | if (!posscpus) { |
1792 | lxcfs_debug("Could not read file: %s.\n", fpath); | |
1793 | goto on_error; | |
1794 | } | |
04742595 CB |
1795 | |
1796 | /* Get maximum number of cpus found in possible cpuset. */ | |
1797 | maxposs = cg_get_max_cpus(posscpus); | |
1798 | if (maxposs < 0) | |
808fd1ef | 1799 | goto on_error; |
04742595 | 1800 | |
b25412f5 CB |
1801 | if (!file_exists(__ISOL_CPUS)) { |
1802 | /* This system doesn't expose isolated cpus. */ | |
1803 | lxcfs_debug("%s", "Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
1804 | cpulist = posscpus; | |
1805 | /* No isolated cpus but we weren't already initialized by | |
1806 | * someone. We should simply copy the parents cpuset.cpus | |
1807 | * values. | |
1808 | */ | |
1809 | if (!am_initialized) { | |
1810 | lxcfs_debug("%s", "Copying cpuset of parent cgroup.\n"); | |
1811 | goto copy_parent; | |
1812 | } | |
1813 | /* No isolated cpus but we were already initialized by someone. | |
1814 | * Nothing more to do for us. | |
1815 | */ | |
1816 | goto on_success; | |
1817 | } | |
1818 | ||
808fd1ef CB |
1819 | isolcpus = read_file(__ISOL_CPUS); |
1820 | if (!isolcpus) { | |
1821 | lxcfs_debug("%s", "Could not read file "__ISOL_CPUS"\n"); | |
1822 | goto on_error; | |
1823 | } | |
04742595 | 1824 | if (!isdigit(isolcpus[0])) { |
808fd1ef | 1825 | lxcfs_debug("%s", "No isolated cpus detected.\n"); |
04742595 CB |
1826 | cpulist = posscpus; |
1827 | /* No isolated cpus but we weren't already initialized by | |
1828 | * someone. We should simply copy the parents cpuset.cpus | |
1829 | * values. | |
1830 | */ | |
808fd1ef CB |
1831 | if (!am_initialized) { |
1832 | lxcfs_debug("%s", "Copying cpuset of parent cgroup.\n"); | |
04742595 | 1833 | goto copy_parent; |
808fd1ef | 1834 | } |
04742595 CB |
1835 | /* No isolated cpus but we were already initialized by someone. |
1836 | * Nothing more to do for us. | |
1837 | */ | |
808fd1ef | 1838 | goto on_success; |
04742595 CB |
1839 | } |
1840 | ||
1841 | /* Get maximum number of cpus found in isolated cpuset. */ | |
1842 | maxisol = cg_get_max_cpus(isolcpus); | |
1843 | if (maxisol < 0) | |
808fd1ef | 1844 | goto on_error; |
04742595 CB |
1845 | |
1846 | if (maxposs < maxisol) | |
1847 | maxposs = maxisol; | |
1848 | maxposs++; | |
1849 | ||
1850 | possmask = cg_cpumask(posscpus, maxposs); | |
808fd1ef CB |
1851 | if (!possmask) { |
1852 | lxcfs_debug("%s", "Could not create cpumask for all possible cpus.\n"); | |
1853 | goto on_error; | |
1854 | } | |
04742595 CB |
1855 | |
1856 | isolmask = cg_cpumask(isolcpus, maxposs); | |
808fd1ef CB |
1857 | if (!isolmask) { |
1858 | lxcfs_debug("%s", "Could not create cpumask for all isolated cpus.\n"); | |
1859 | goto on_error; | |
1860 | } | |
04742595 CB |
1861 | |
1862 | for (i = 0; i <= maxposs; i++) { | |
1863 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
808fd1ef | 1864 | flipped_bit = true; |
04742595 CB |
1865 | clear_bit(i, possmask); |
1866 | } | |
1867 | } | |
1868 | ||
808fd1ef CB |
1869 | if (!flipped_bit) { |
1870 | lxcfs_debug("%s", "No isolated cpus present in cpuset.\n"); | |
1871 | goto on_success; | |
1872 | } | |
1873 | lxcfs_debug("%s", "Removed isolated cpus from cpuset.\n"); | |
1874 | ||
04742595 | 1875 | cpulist = cg_cpumask_to_cpulist(possmask, maxposs); |
808fd1ef CB |
1876 | if (!cpulist) { |
1877 | lxcfs_debug("%s", "Could not create cpu list.\n"); | |
1878 | goto on_error; | |
1879 | } | |
04742595 CB |
1880 | |
1881 | copy_parent: | |
1882 | *lastslash = oldv; | |
1883 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
808fd1ef CB |
1884 | ret = write_to_file(fpath, cpulist, strlen(cpulist), false); |
1885 | if (ret < 0) { | |
1886 | lxcfs_debug("Could not write cpu list to: %s.\n", fpath); | |
1887 | goto on_error; | |
1888 | } | |
04742595 | 1889 | |
808fd1ef CB |
1890 | on_success: |
1891 | bret = true; | |
1892 | ||
1893 | on_error: | |
04742595 CB |
1894 | free(fpath); |
1895 | ||
1896 | free(isolcpus); | |
1897 | free(isolmask); | |
1898 | ||
1899 | if (posscpus != cpulist) | |
1900 | free(posscpus); | |
1901 | free(possmask); | |
1902 | ||
1903 | free(cpulist); | |
1904 | return bret; | |
1905 | } | |
1906 | ||
1907 | int read_from_file(const char *filename, void* buf, size_t count) | |
1908 | { | |
1909 | int fd = -1, saved_errno; | |
1910 | ssize_t ret; | |
1911 | ||
1912 | fd = open(filename, O_RDONLY | O_CLOEXEC); | |
1913 | if (fd < 0) | |
1914 | return -1; | |
1915 | ||
1916 | if (!buf || !count) { | |
1917 | char buf2[100]; | |
1918 | size_t count2 = 0; | |
1919 | while ((ret = read(fd, buf2, 100)) > 0) | |
1920 | count2 += ret; | |
1921 | if (ret >= 0) | |
1922 | ret = count2; | |
1923 | } else { | |
1924 | memset(buf, 0, count); | |
1925 | ret = read(fd, buf, count); | |
1926 | } | |
1927 | ||
1928 | if (ret < 0) | |
1929 | lxcfs_debug("read %s: %s", filename, strerror(errno)); | |
1930 | ||
1931 | saved_errno = errno; | |
1932 | close(fd); | |
1933 | errno = saved_errno; | |
1934 | return ret; | |
1935 | } | |
1936 | ||
1937 | /* Copy contents of parent(@path)/@file to @path/@file */ | |
1938 | static bool cg_copy_parent_file(char *path, char *file) | |
1939 | { | |
1940 | char *lastslash, *value = NULL, *fpath, oldv; | |
1941 | int len = 0; | |
1942 | int ret; | |
1943 | ||
1944 | lastslash = strrchr(path, '/'); | |
1945 | if (!lastslash) { // bug... this shouldn't be possible | |
1946 | lxcfs_debug("cgfsng:copy_parent_file: bad path %s", path); | |
1947 | return false; | |
1948 | } | |
1949 | oldv = *lastslash; | |
1950 | *lastslash = '\0'; | |
1951 | fpath = must_make_path(path, file, NULL); | |
1952 | len = read_from_file(fpath, NULL, 0); | |
1953 | if (len <= 0) | |
1954 | goto bad; | |
1955 | value = must_alloc(len + 1); | |
1956 | if (read_from_file(fpath, value, len) != len) | |
1957 | goto bad; | |
1958 | free(fpath); | |
1959 | *lastslash = oldv; | |
1960 | fpath = must_make_path(path, file, NULL); | |
808fd1ef | 1961 | ret = write_to_file(fpath, value, len, false); |
04742595 CB |
1962 | if (ret < 0) |
1963 | lxcfs_debug("Unable to write %s to %s", value, fpath); | |
1964 | free(fpath); | |
1965 | free(value); | |
1966 | return ret >= 0; | |
1967 | ||
1968 | bad: | |
1969 | lxcfs_debug("Error reading '%s'", fpath); | |
1970 | free(fpath); | |
1971 | free(value); | |
1972 | return false; | |
1973 | } | |
1974 | ||
1975 | /* In case systemd hasn't already placed us in a user slice for the cpuset v1 | |
1976 | * controller we will reside in the root cgroup. This means that | |
1977 | * cgroup.clone_children will not have been initialized for us so we need to do | |
1978 | * it. | |
1979 | */ | |
1980 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h) | |
1981 | { | |
1982 | char *clonechildrenpath, v; | |
1983 | ||
1984 | if (!string_in_list(h->controllers, "cpuset")) | |
1985 | return true; | |
1986 | ||
1987 | clonechildrenpath = must_make_path(h->mountpoint, "cgroup.clone_children", NULL); | |
1988 | ||
1989 | if (read_from_file(clonechildrenpath, &v, 1) < 0) { | |
1990 | lxcfs_debug("Failed to read '%s'", clonechildrenpath); | |
1991 | free(clonechildrenpath); | |
1992 | return false; | |
1993 | } | |
1994 | ||
1995 | if (v == '1') { /* already set for us by someone else */ | |
1996 | free(clonechildrenpath); | |
1997 | return true; | |
1998 | } | |
1999 | ||
808fd1ef | 2000 | if (write_to_file(clonechildrenpath, "1", 1, false) < 0) { |
04742595 CB |
2001 | /* Set clone_children so children inherit our settings */ |
2002 | lxcfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
2003 | free(clonechildrenpath); | |
2004 | return false; | |
2005 | } | |
2006 | free(clonechildrenpath); | |
2007 | return true; | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * Initialize the cpuset hierarchy in first directory of @gname and | |
2012 | * set cgroup.clone_children so that children inherit settings. | |
2013 | * Since the h->base_path is populated by init or ourselves, we know | |
2014 | * it is already initialized. | |
2015 | */ | |
2016 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, | |
2017 | const char *cgroup) | |
2018 | { | |
2019 | char *cgpath, *clonechildrenpath, v, *slash; | |
2020 | ||
2021 | if (!string_in_list(h->controllers, "cpuset")) | |
2022 | return true; | |
2023 | ||
2024 | if (*cgroup == '/') | |
2025 | cgroup++; | |
2026 | slash = strchr(cgroup, '/'); | |
2027 | if (slash) | |
2028 | *slash = '\0'; | |
2029 | ||
2030 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgroup, NULL); | |
2031 | if (slash) | |
2032 | *slash = '/'; | |
2033 | if (mkdir(cgpath, 0755) < 0 && errno != EEXIST) { | |
2034 | lxcfs_debug("Failed to create '%s'", cgpath); | |
2035 | free(cgpath); | |
2036 | return false; | |
2037 | } | |
2038 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); | |
2039 | if (!file_exists(clonechildrenpath)) { /* unified hierarchy doesn't have clone_children */ | |
2040 | free(clonechildrenpath); | |
2041 | free(cgpath); | |
2042 | return true; | |
2043 | } | |
2044 | if (read_from_file(clonechildrenpath, &v, 1) < 0) { | |
2045 | lxcfs_debug("Failed to read '%s'", clonechildrenpath); | |
2046 | free(clonechildrenpath); | |
2047 | free(cgpath); | |
2048 | return false; | |
2049 | } | |
2050 | ||
2051 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ | |
7c029b0f CB |
2052 | if (!cg_filter_and_set_cpus(cgpath, v == '1')) { |
2053 | lxcfs_debug("%s", "Failed to remove isolated cpus.\n"); | |
2054 | free(clonechildrenpath); | |
2055 | free(cgpath); | |
04742595 | 2056 | return false; |
7c029b0f | 2057 | } |
04742595 CB |
2058 | |
2059 | if (v == '1') { /* already set for us by someone else */ | |
808fd1ef | 2060 | lxcfs_debug("%s", "\"cgroup.clone_children\" was already set to \"1\".\n"); |
04742595 CB |
2061 | free(clonechildrenpath); |
2062 | free(cgpath); | |
2063 | return true; | |
2064 | } | |
2065 | ||
2066 | /* copy parent's settings */ | |
2067 | if (!cg_copy_parent_file(cgpath, "cpuset.mems")) { | |
808fd1ef | 2068 | lxcfs_debug("%s", "Failed to copy \"cpuset.mems\" settings.\n"); |
04742595 CB |
2069 | free(cgpath); |
2070 | free(clonechildrenpath); | |
2071 | return false; | |
2072 | } | |
2073 | free(cgpath); | |
2074 | ||
808fd1ef | 2075 | if (write_to_file(clonechildrenpath, "1", 1, false) < 0) { |
04742595 CB |
2076 | /* Set clone_children so children inherit our settings */ |
2077 | lxcfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
2078 | free(clonechildrenpath); | |
2079 | return false; | |
2080 | } | |
2081 | free(clonechildrenpath); | |
2082 | return true; | |
2083 | } | |
2084 | ||
e65cfafc CB |
2085 | /* Create and chown @cgroup for all given controllers in a cgroupfs v1 hierarchy |
2086 | * (For example, create @cgroup for the cpu and cpuacct controller mounted into | |
2087 | * /sys/fs/cgroup/cpu,cpuacct). Check if the path already exists and report back | |
2088 | * to the caller in @existed. | |
df54106a | 2089 | */ |
e65cfafc CB |
2090 | #define __PAM_CGFS_USER "/user/" |
2091 | #define __PAM_CGFS_USER_LEN 6 | |
2092 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
df54106a | 2093 | { |
e65cfafc CB |
2094 | char *clean_base_cgroup, *path; |
2095 | char **controller; | |
2096 | struct cgv1_hierarchy *it; | |
2097 | bool created = false; | |
2098 | ||
a403c004 | 2099 | *existed = false; |
e65cfafc CB |
2100 | it = h; |
2101 | for (controller = it->controllers; controller && *controller; | |
2102 | controller++) { | |
04742595 CB |
2103 | if (!cgv1_handle_cpuset_hierarchy(it, cgroup)) |
2104 | return false; | |
2105 | ||
e65cfafc CB |
2106 | /* If systemd has already created a cgroup for us, keep using |
2107 | * it. | |
2108 | */ | |
2109 | if (cg_systemd_chown_existing_cgroup(it->mountpoint, | |
2110 | it->base_cgroup, uid, gid, | |
2111 | it->systemd_user_slice)) { | |
2112 | return true; | |
2113 | } | |
2114 | ||
2115 | /* We need to make sure that we do not create an endless chain | |
2116 | * of sub-cgroups. So we check if we have already logged in | |
2117 | * somehow (sudo -i, su, etc.) and have created a | |
2118 | * /user/PAM_user/idx cgroup. If so, we skip that part. For most | |
2119 | * cgroups this is unnecessary since we use the init_cgroup | |
2120 | * anyway, but for controllers which have an existing systemd | |
2121 | * cgroup that does not match the current uid, this is pretty | |
2122 | * useful. | |
2123 | */ | |
2124 | if (strncmp(it->base_cgroup, __PAM_CGFS_USER, __PAM_CGFS_USER_LEN) == 0) { | |
2125 | free(it->base_cgroup); | |
2126 | it->base_cgroup = must_copy_string("/"); | |
2127 | } else { | |
2128 | clean_base_cgroup = | |
2129 | strstr(it->base_cgroup, __PAM_CGFS_USER); | |
2130 | if (clean_base_cgroup) | |
2131 | *clean_base_cgroup = '\0'; | |
2132 | } | |
df54106a | 2133 | |
e65cfafc CB |
2134 | path = must_make_path(it->mountpoint, it->init_cgroup, cgroup, NULL); |
2135 | lxcfs_debug("Constructing path: %s.\n", path); | |
2136 | if (file_exists(path)) { | |
3145d497 | 2137 | bool our_cg = cg_belongs_to_uid_gid(path, uid, gid); |
b36273fa | 2138 | lxcfs_debug("%s existed and does %shave our uid: %d and gid: %d.\n", path, our_cg ? "" : "not ", uid, gid); |
e65cfafc | 2139 | free(path); |
3145d497 CB |
2140 | if (our_cg) |
2141 | *existed = false; | |
2142 | else | |
2143 | *existed = true; | |
2144 | return our_cg; | |
e65cfafc CB |
2145 | } |
2146 | created = mkdir_p(it->mountpoint, path); | |
2147 | if (!created) { | |
df54106a | 2148 | free(path); |
e65cfafc | 2149 | continue; |
df54106a | 2150 | } |
e65cfafc | 2151 | if (chown(path, uid, gid) < 0) |
b36273fa CB |
2152 | mysyslog(LOG_WARNING, |
2153 | "Failed to chown %s to %d:%d: %s.\n", path, | |
2154 | (int)uid, (int)gid, strerror(errno), NULL); | |
2155 | lxcfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
e65cfafc CB |
2156 | free(path); |
2157 | break; | |
df54106a | 2158 | } |
e65cfafc | 2159 | |
eb42e790 | 2160 | return created; |
df54106a SH |
2161 | } |
2162 | ||
e65cfafc CB |
2163 | /* Try to remove @cgroup for all given controllers in a cgroupfs v1 hierarchy |
2164 | * (For example, try to remove @cgroup for the cpu and cpuacct controller | |
2165 | * mounted into /sys/fs/cgroup/cpu,cpuacct). Ignores failures. | |
2166 | */ | |
2167 | static bool cgv1_remove_one(struct cgv1_hierarchy *h, const char *cgroup) | |
df54106a | 2168 | { |
df54106a | 2169 | |
e65cfafc | 2170 | char *path; |
df54106a | 2171 | |
e65cfafc CB |
2172 | /* Better safe than sorry. */ |
2173 | if (!h->controllers) | |
2174 | return true; | |
df54106a | 2175 | |
e65cfafc CB |
2176 | /* Cgroups created by systemd for us which we re-use won't be removed |
2177 | * here, since we're using init_cgroup + cgroup as path instead of | |
2178 | * base_cgroup + cgroup. | |
2179 | */ | |
2180 | path = must_make_path(h->mountpoint, h->init_cgroup, cgroup, NULL); | |
2181 | (void)recursive_rmdir(path); | |
2182 | free(path); | |
df54106a SH |
2183 | |
2184 | return true; | |
2185 | } | |
2186 | ||
e65cfafc CB |
2187 | /* Try to remove @cgroup the cgroupfs v2 hierarchy. */ |
2188 | static bool cgv2_remove(const char *cgroup) | |
df54106a | 2189 | { |
e65cfafc CB |
2190 | struct cgv2_hierarchy *v2; |
2191 | char *path; | |
e9597a70 | 2192 | |
e65cfafc CB |
2193 | if (!cgv2_hierarchies) |
2194 | return true; | |
2195 | ||
2196 | v2 = *cgv2_hierarchies; | |
2197 | ||
2198 | /* If we reused an already existing cgroup, don't bother trying to | |
2199 | * remove (a potentially wrong)/the path. | |
2200 | * Cgroups created by systemd for us which we re-use would be removed | |
2201 | * here, since we're using base_cgroup + cgroup as path. | |
2202 | */ | |
2203 | if (v2->systemd_user_slice) | |
2204 | return true; | |
2205 | ||
2206 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, NULL); | |
2207 | (void)recursive_rmdir(path); | |
2208 | free(path); | |
2209 | ||
2210 | return true; | |
df54106a SH |
2211 | } |
2212 | ||
e65cfafc CB |
2213 | /* Create @cgroup in all detected cgroupfs v1 hierarchy. If the creation fails |
2214 | * for any cgroupfs v1 hierarchy, remove all we have created so far. Report | |
2215 | * back, to the caller if the creation failed due to @cgroup already existing | |
2216 | * via @existed. | |
2217 | */ | |
2218 | static bool cgv1_create(const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
df54106a | 2219 | { |
e65cfafc CB |
2220 | struct cgv1_hierarchy **it, **rev_it; |
2221 | bool all_created = true; | |
df54106a | 2222 | |
e65cfafc CB |
2223 | for (it = cgv1_hierarchies; it && *it; it++) { |
2224 | if (!(*it)->controllers || !(*it)->mountpoint || | |
2225 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
2226 | continue; | |
2227 | ||
2228 | if (!cgv1_create_one(*it, cgroup, uid, gid, existed)) { | |
2229 | all_created = false; | |
2230 | break; | |
df54106a | 2231 | } |
df54106a SH |
2232 | } |
2233 | ||
e65cfafc CB |
2234 | if (all_created) |
2235 | return true; | |
2236 | ||
2237 | for (rev_it = cgv1_hierarchies; rev_it && *rev_it && (*rev_it != *it); | |
2238 | rev_it++) | |
2239 | cgv1_remove_one(*rev_it, cgroup); | |
2240 | ||
df54106a SH |
2241 | return false; |
2242 | } | |
2243 | ||
e65cfafc CB |
2244 | /* Create @cgroup in the cgroupfs v2 hierarchy. Report back, to the caller if |
2245 | * the creation failed due to @cgroup already existing via @existed. | |
2246 | */ | |
2247 | static bool cgv2_create(const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
df54106a | 2248 | { |
e65cfafc CB |
2249 | char *clean_base_cgroup; |
2250 | char *path; | |
2251 | struct cgv2_hierarchy *v2; | |
2252 | bool created = false; | |
df54106a | 2253 | |
a403c004 CB |
2254 | *existed = false; |
2255 | ||
e65cfafc CB |
2256 | if (!cgv2_hierarchies || !(*cgv2_hierarchies)->create_rw_cgroup) |
2257 | return true; | |
df54106a | 2258 | |
e65cfafc | 2259 | v2 = *cgv2_hierarchies; |
df54106a | 2260 | |
e65cfafc CB |
2261 | /* We can't be placed under init's cgroup for the v2 hierarchy. We need |
2262 | * to be placed under our current cgroup. | |
2263 | */ | |
2264 | if (cg_systemd_chown_existing_cgroup(v2->mountpoint, | |
2265 | v2->base_cgroup, uid, gid, | |
2266 | v2->systemd_user_slice)) | |
2267 | return true; | |
edd25678 | 2268 | |
423a3b4f | 2269 | /* We need to make sure that we do not create an endless chain of |
e65cfafc CB |
2270 | * sub-cgroups. So we check if we have already logged in somehow (sudo |
2271 | * -i, su, etc.) and have created a /user/PAM_user/idx cgroup. If so, we | |
2272 | * skip that part. | |
2273 | */ | |
2274 | if (strncmp(v2->base_cgroup, __PAM_CGFS_USER, __PAM_CGFS_USER_LEN) == 0) { | |
2275 | free(v2->base_cgroup); | |
2276 | v2->base_cgroup = must_copy_string("/"); | |
2277 | } else { | |
2278 | clean_base_cgroup = strstr(v2->base_cgroup, __PAM_CGFS_USER); | |
2279 | if (clean_base_cgroup) | |
2280 | *clean_base_cgroup = '\0'; | |
df54106a SH |
2281 | } |
2282 | ||
e65cfafc CB |
2283 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, NULL); |
2284 | lxcfs_debug("Constructing path \"%s\".\n", path); | |
2285 | if (file_exists(path)) { | |
3145d497 | 2286 | bool our_cg = cg_belongs_to_uid_gid(path, uid, gid); |
b36273fa | 2287 | lxcfs_debug("%s existed and does %shave our uid: %d and gid: %d.\n", path, our_cg ? "" : "not ", uid, gid); |
e65cfafc | 2288 | free(path); |
3145d497 CB |
2289 | if (our_cg) |
2290 | *existed = false; | |
2291 | else | |
2292 | *existed = true; | |
2293 | return our_cg; | |
df54106a | 2294 | } |
df54106a | 2295 | |
e65cfafc CB |
2296 | created = mkdir_p(v2->mountpoint, path); |
2297 | if (!created) { | |
2298 | free(path); | |
df54106a | 2299 | return false; |
e65cfafc CB |
2300 | } |
2301 | ||
2302 | if (chown(path, uid, gid) < 0) | |
b36273fa CB |
2303 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", |
2304 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
2305 | lxcfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
e65cfafc | 2306 | free(path); |
df54106a SH |
2307 | |
2308 | return true; | |
2309 | } | |
2310 | ||
e65cfafc CB |
2311 | /* Create writeable cgroups for @user at login. Details can be found in the |
2312 | * preamble/license at the top of this file. | |
2313 | */ | |
2314 | static int handle_login(const char *user, uid_t uid, gid_t gid) | |
df54106a SH |
2315 | { |
2316 | int idx = 0, ret; | |
2317 | bool existed; | |
df54106a | 2318 | char cg[MAXPATHLEN]; |
78a2a9f3 | 2319 | |
e65cfafc | 2320 | cg_escape(); |
df54106a SH |
2321 | |
2322 | while (idx >= 0) { | |
2323 | ret = snprintf(cg, MAXPATHLEN, "/user/%s/%d", user, idx); | |
2324 | if (ret < 0 || ret >= MAXPATHLEN) { | |
e65cfafc CB |
2325 | mysyslog(LOG_ERR, "Username too long.\n", NULL); |
2326 | return PAM_SESSION_ERR; | |
2327 | } | |
2328 | ||
2329 | existed = false; | |
2330 | if (!cgv2_create(cg, uid, gid, &existed)) { | |
2331 | if (existed) { | |
2332 | cgv2_remove(cg); | |
2333 | idx++; | |
2334 | continue; | |
2335 | } | |
2336 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s.\n", user, NULL); | |
df54106a SH |
2337 | return PAM_SESSION_ERR; |
2338 | } | |
2339 | ||
56ee748c | 2340 | existed = false; |
e65cfafc | 2341 | if (!cgv1_create(cg, uid, gid, &existed)) { |
56ee748c | 2342 | if (existed) { |
e65cfafc | 2343 | cgv2_remove(cg); |
56ee748c SH |
2344 | idx++; |
2345 | continue; | |
2346 | } | |
e65cfafc | 2347 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s.\n", user, NULL); |
df54106a SH |
2348 | return PAM_SESSION_ERR; |
2349 | } | |
2350 | ||
e65cfafc CB |
2351 | if (!cg_enter(cg)) { |
2352 | mysyslog( LOG_ERR, "Failed to enter user cgroup %s for user %s.\n", cg, user, NULL); | |
df54106a SH |
2353 | return PAM_SESSION_ERR; |
2354 | } | |
2355 | break; | |
2356 | } | |
2357 | ||
2358 | return PAM_SUCCESS; | |
2359 | } | |
2360 | ||
e65cfafc CB |
2361 | /* Try to prune cgroups we created and that now are empty from all cgroupfs v1 |
2362 | * hierarchies. | |
2363 | */ | |
2364 | static bool cgv1_prune_empty_cgroups(const char *user) | |
2365 | { | |
2366 | bool controller_removed = true; | |
2367 | bool all_removed = true; | |
2368 | struct cgv1_hierarchy **it; | |
2369 | ||
2370 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2371 | int ret; | |
2372 | char *path_base, *path_init; | |
2373 | char **controller; | |
2374 | ||
2375 | if (!(*it)->controllers || !(*it)->mountpoint || | |
2376 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
2377 | continue; | |
2378 | ||
2379 | for (controller = (*it)->controllers; controller && *controller; | |
2380 | controller++) { | |
2381 | bool path_base_rm, path_init_rm; | |
2382 | ||
2383 | path_base = must_make_path((*it)->mountpoint, (*it)->base_cgroup, "/user", user, NULL); | |
2384 | lxcfs_debug("cgroupfs v1: Trying to prune \"%s\".\n", path_base); | |
2385 | ret = recursive_rmdir(path_base); | |
2386 | if (ret == -ENOENT || ret >= 0) | |
2387 | path_base_rm = true; | |
2388 | else | |
2389 | path_base_rm = false; | |
2390 | free(path_base); | |
2391 | ||
2392 | path_init = must_make_path((*it)->mountpoint, (*it)->init_cgroup, "/user", user, NULL); | |
2393 | lxcfs_debug("cgroupfs v1: Trying to prune \"%s\".\n", path_init); | |
2394 | ret = recursive_rmdir(path_init); | |
2395 | if (ret == -ENOENT || ret >= 0) | |
2396 | path_init_rm = true; | |
2397 | else | |
2398 | path_init_rm = false; | |
2399 | free(path_init); | |
2400 | ||
2401 | if (!path_base_rm && !path_init_rm) { | |
2402 | controller_removed = false; | |
2403 | continue; | |
2404 | } | |
2405 | ||
2406 | controller_removed = true; | |
2407 | break; | |
2408 | } | |
2409 | if (!controller_removed) | |
2410 | all_removed = false; | |
2411 | } | |
2412 | ||
2413 | return all_removed; | |
2414 | } | |
2415 | ||
2416 | /* Try to prune cgroup we created and that now is empty from the cgroupfs v2 | |
2417 | * hierarchy. | |
2418 | */ | |
2419 | static bool cgv2_prune_empty_cgroups(const char *user) | |
df54106a | 2420 | { |
df54106a | 2421 | int ret; |
e65cfafc CB |
2422 | struct cgv2_hierarchy *v2; |
2423 | char *path_base, *path_init; | |
2424 | bool path_base_rm, path_init_rm; | |
df54106a | 2425 | |
e65cfafc CB |
2426 | if (!cgv2_hierarchies) |
2427 | return true; | |
2428 | ||
2429 | v2 = *cgv2_hierarchies; | |
2430 | ||
2431 | path_base = must_make_path(v2->mountpoint, v2->base_cgroup, "/user", user, NULL); | |
2432 | lxcfs_debug("cgroupfs v2: Trying to prune \"%s\".\n", path_base); | |
2433 | ret = recursive_rmdir(path_base); | |
2434 | if (ret == -ENOENT || ret >= 0) | |
2435 | path_base_rm = true; | |
2436 | else | |
2437 | path_base_rm = false; | |
2438 | free(path_base); | |
2439 | ||
2440 | path_init = must_make_path(v2->mountpoint, v2->init_cgroup, "/user", user, NULL); | |
2441 | lxcfs_debug("cgroupfs v2: Trying to prune \"%s\".\n", path_init); | |
2442 | ret = recursive_rmdir(path_init); | |
2443 | if (ret == -ENOENT || ret >= 0) | |
2444 | path_init_rm = true; | |
2445 | else | |
2446 | path_init_rm = false; | |
2447 | free(path_init); | |
2448 | ||
2449 | if (!path_base_rm && !path_init_rm) | |
2450 | return false; | |
2451 | ||
2452 | return true; | |
2453 | } | |
2454 | ||
2455 | /* Wrapper around cgv{1,2}_prune_empty_cgroups(). */ | |
2456 | static void cg_prune_empty_cgroups(const char *user) | |
2457 | { | |
2458 | (void)cgv1_prune_empty_cgroups(user); | |
2459 | (void)cgv2_prune_empty_cgroups(user); | |
2460 | } | |
2461 | ||
2462 | /* Free allocated information for detected cgroupfs v1 hierarchies. */ | |
2463 | static void cgv1_free_hierarchies(void) | |
2464 | { | |
2465 | struct cgv1_hierarchy **it; | |
2466 | ||
2467 | if (!cgv1_hierarchies) | |
2468 | return; | |
2469 | ||
2470 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2471 | if ((*it)->controllers) { | |
2472 | char **tmp; | |
2473 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) | |
2474 | free(*tmp); | |
2475 | ||
2476 | free((*it)->controllers); | |
2477 | } | |
2478 | free((*it)->mountpoint); | |
2479 | free((*it)->base_cgroup); | |
2480 | free((*it)->fullcgpath); | |
2481 | free((*it)->init_cgroup); | |
df54106a | 2482 | } |
e65cfafc CB |
2483 | free(cgv1_hierarchies); |
2484 | } | |
df54106a | 2485 | |
e65cfafc CB |
2486 | /* Free allocated information for the detected cgroupfs v2 hierarchy. */ |
2487 | static void cgv2_free_hierarchies(void) | |
2488 | { | |
2489 | struct cgv2_hierarchy **it; | |
2490 | ||
2491 | if (!cgv2_hierarchies) | |
2492 | return; | |
2493 | ||
2494 | for (it = cgv2_hierarchies; it && *it; it++) { | |
2495 | if ((*it)->controllers) { | |
2496 | char **tmp; | |
2497 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) | |
2498 | free(*tmp); | |
2499 | ||
2500 | free((*it)->controllers); | |
2501 | } | |
2502 | free((*it)->mountpoint); | |
2503 | free((*it)->base_cgroup); | |
2504 | free((*it)->fullcgpath); | |
2505 | free((*it)->init_cgroup); | |
2506 | } | |
2507 | free(cgv2_hierarchies); | |
2508 | } | |
2509 | ||
2510 | /* Wrapper around cgv{1,2}_free_hierarchies(). */ | |
2511 | static void cg_exit(void) | |
2512 | { | |
2513 | cgv1_free_hierarchies(); | |
2514 | cgv2_free_hierarchies(); | |
2515 | } | |
2516 | ||
2517 | int pam_sm_open_session(pam_handle_t *pamh, int flags, int argc, | |
2518 | const char **argv) | |
2519 | { | |
2520 | int ret; | |
2521 | uid_t uid = 0; | |
2522 | gid_t gid = 0; | |
2523 | const char *PAM_user = NULL; | |
df54106a SH |
2524 | |
2525 | ret = pam_get_user(pamh, &PAM_user, NULL); | |
2526 | if (ret != PAM_SUCCESS) { | |
e65cfafc | 2527 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); |
df54106a SH |
2528 | return PAM_SESSION_ERR; |
2529 | } | |
2530 | ||
e65cfafc CB |
2531 | if (!get_uid_gid(PAM_user, &uid, &gid)) { |
2532 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s.\n", PAM_user, NULL); | |
2533 | return PAM_SESSION_ERR; | |
2534 | } | |
df54106a | 2535 | |
e65cfafc CB |
2536 | if (!cg_init(uid, gid)) { |
2537 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2538 | return PAM_SESSION_ERR; | |
df54106a | 2539 | } |
df54106a | 2540 | |
e65cfafc CB |
2541 | /* Try to prune cgroups, that are actually empty but were still marked |
2542 | * as busy by the kernel so we couldn't remove them on session close. | |
2543 | */ | |
2544 | cg_prune_empty_cgroups(PAM_user); | |
2545 | ||
2546 | if (cg_mount_mode == CGROUP_UNKNOWN) | |
2547 | return PAM_SESSION_ERR; | |
2548 | ||
2549 | if (argc > 1 && strcmp(argv[0], "-c") == 0) | |
2550 | cg_mark_to_make_rw(argv[1]); | |
df54106a | 2551 | |
e65cfafc | 2552 | return handle_login(PAM_user, uid, gid); |
df54106a SH |
2553 | } |
2554 | ||
2555 | int pam_sm_close_session(pam_handle_t *pamh, int flags, int argc, | |
2556 | const char **argv) | |
2557 | { | |
e65cfafc CB |
2558 | int ret; |
2559 | uid_t uid = 0; | |
2560 | gid_t gid = 0; | |
df54106a | 2561 | const char *PAM_user = NULL; |
df54106a | 2562 | |
e65cfafc | 2563 | ret = pam_get_user(pamh, &PAM_user, NULL); |
df54106a | 2564 | if (ret != PAM_SUCCESS) { |
e65cfafc CB |
2565 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); |
2566 | return PAM_SESSION_ERR; | |
2567 | } | |
2568 | ||
2569 | if (!get_uid_gid(PAM_user, &uid, &gid)) { | |
2570 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s.\n", PAM_user, NULL); | |
df54106a SH |
2571 | return PAM_SESSION_ERR; |
2572 | } | |
2573 | ||
e65cfafc CB |
2574 | if (cg_mount_mode == CGROUP_UNINITIALIZED) { |
2575 | if (!cg_init(uid, gid)) | |
2576 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2577 | ||
df54106a | 2578 | if (argc > 1 && strcmp(argv[0], "-c") == 0) |
e65cfafc | 2579 | cg_mark_to_make_rw(argv[1]); |
df54106a SH |
2580 | } |
2581 | ||
e65cfafc CB |
2582 | cg_prune_empty_cgroups(PAM_user); |
2583 | cg_exit(); | |
2584 | ||
df54106a SH |
2585 | return PAM_SUCCESS; |
2586 | } |