]>
Commit | Line | Data |
---|---|---|
7ac43d3d CB |
1 | /* pam-cgfs |
2 | * | |
3 | * Copyright © 2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * Author: Christian Brauner <christian.brauner@ubuntu.com> | |
6 | * | |
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. | |
14 | * The cgroup created will be "user/$user/0" for the first session, | |
15 | * "user/$user/1" for the second, etc. | |
16 | * | |
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/$user/n) and chown it to our uid. | |
26 | * The same holds for cgroupfs v2 where this assumptions becomes crucial: | |
27 | * We __have to__ be placed in our under the cgroup systemd created for us on | |
28 | * login, otherwise things like starting an xserver or similar will not work. | |
29 | * | |
30 | * All requested cgroups must be mounted under /sys/fs/cgroup/$controller, | |
31 | * no messing around with finding mountpoints. | |
32 | * | |
33 | * See COPYING file for details. | |
34 | */ | |
35 | ||
d38dd64a CB |
36 | #ifndef _GNU_SOURCE |
37 | #define _GNU_SOURCE 1 | |
38 | #endif | |
7ac43d3d CB |
39 | #include <ctype.h> |
40 | #include <dirent.h> | |
41 | #include <errno.h> | |
42 | #include <fcntl.h> | |
d38dd64a | 43 | #include <linux/unistd.h> |
7ac43d3d CB |
44 | #include <pwd.h> |
45 | #include <stdarg.h> | |
46 | #include <stdbool.h> | |
47 | #include <stdint.h> | |
48 | #include <stdio.h> | |
49 | #include <stdlib.h> | |
50 | #include <string.h> | |
7ac43d3d CB |
51 | #include <sys/mount.h> |
52 | #include <sys/param.h> | |
53 | #include <sys/stat.h> | |
54 | #include <sys/types.h> | |
55 | #include <sys/vfs.h> | |
d38dd64a CB |
56 | #include <syslog.h> |
57 | #include <unistd.h> | |
58 | ||
59 | #include "config.h" | |
60 | #include "macro.h" | |
61 | #include "utils.h" | |
7ac43d3d CB |
62 | |
63 | #define PAM_SM_SESSION | |
64 | #include <security/_pam_macros.h> | |
65 | #include <security/pam_modules.h> | |
66 | ||
43f984ea DJ |
67 | #ifndef HAVE_STRLCPY |
68 | #include "include/strlcpy.h" | |
69 | #endif | |
70 | ||
9c8b2b13 DJ |
71 | #ifndef HAVE_STRLCAT |
72 | #include "include/strlcat.h" | |
73 | #endif | |
74 | ||
7ac43d3d CB |
75 | #define pam_cgfs_debug_stream(stream, format, ...) \ |
76 | do { \ | |
77 | fprintf(stream, "%s: %d: %s: " format, __FILE__, __LINE__, \ | |
78 | __func__, __VA_ARGS__); \ | |
79 | } while (false) | |
80 | ||
81 | #define pam_cgfs_error(format, ...) pam_cgfs_debug_stream(stderr, format, __VA_ARGS__) | |
82 | ||
83 | #ifdef DEBUG | |
84 | #define pam_cgfs_debug(format, ...) pam_cgfs_error(format, __VA_ARGS__) | |
85 | #else | |
86 | #define pam_cgfs_debug(format, ...) | |
87 | #endif /* DEBUG */ | |
88 | ||
89 | /* Taken over modified from the kernel sources. */ | |
90 | #define NBITS 32 /* bits in uint32_t */ | |
91 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
92 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
93 | ||
94 | static enum cg_mount_mode { | |
95 | CGROUP_UNKNOWN = -1, | |
96 | CGROUP_MIXED = 0, | |
97 | CGROUP_PURE_V1 = 1, | |
98 | CGROUP_PURE_V2 = 2, | |
99 | CGROUP_UNINITIALIZED = 3, | |
100 | } cg_mount_mode = CGROUP_UNINITIALIZED; | |
101 | ||
102 | /* Common helper functions. Most of these have been taken from LXC. */ | |
103 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen); | |
104 | static int append_null_to_list(void ***list); | |
105 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen); | |
106 | static inline void clear_bit(unsigned bit, uint32_t *bitarr) | |
107 | { | |
108 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
109 | } | |
110 | static char *copy_to_eol(char *s); | |
111 | static void free_string_list(char **list); | |
112 | static char *get_mountpoint(char *line); | |
113 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid); | |
114 | static int handle_login(const char *user, uid_t uid, gid_t gid); | |
115 | static inline bool is_set(unsigned bit, uint32_t *bitarr) | |
116 | { | |
117 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
118 | } | |
119 | static bool is_lxcfs(const char *line); | |
120 | static bool is_cgv1(char *line); | |
121 | static bool is_cgv2(char *line); | |
7ac43d3d CB |
122 | static void must_add_to_list(char ***clist, char *entry); |
123 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
124 | char *entry); | |
125 | static void must_append_string(char ***list, char *entry); | |
126 | static void mysyslog(int err, const char *format, ...) __attribute__((sentinel)); | |
127 | static char *read_file(char *fnam); | |
7ac43d3d CB |
128 | static int recursive_rmdir(char *dirname); |
129 | static inline void set_bit(unsigned bit, uint32_t *bitarr) | |
130 | { | |
131 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
132 | } | |
133 | static bool string_in_list(char **list, const char *entry); | |
134 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix); | |
135 | static void trim(char *s); | |
136 | static bool write_int(char *path, int v); | |
7ac43d3d CB |
137 | |
138 | /* cgroupfs prototypes. */ | |
139 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid); | |
140 | static uint32_t *cg_cpumask(char *buf, size_t nbits); | |
141 | static bool cg_copy_parent_file(char *path, char *file); | |
142 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits); | |
143 | static bool cg_enter(const char *cgroup); | |
144 | static void cg_escape(void); | |
145 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized); | |
146 | static ssize_t cg_get_max_cpus(char *cpulist); | |
147 | static int cg_get_version_of_mntpt(const char *path); | |
148 | static bool cg_init(uid_t uid, gid_t gid); | |
149 | static void cg_mark_to_make_rw(char **list); | |
150 | static void cg_prune_empty_cgroups(const char *user); | |
151 | static bool cg_systemd_created_user_slice(const char *base_cgroup, | |
152 | const char *init_cgroup, | |
153 | const char *in, uid_t uid); | |
154 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
155 | const char *base_cgroup, uid_t uid, | |
156 | gid_t gid, | |
157 | bool systemd_user_slice); | |
158 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid); | |
159 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
160 | const char *init_cgroup, uid_t uid); | |
161 | static void cg_systemd_prune_init_scope(char *cg); | |
162 | static bool is_lxcfs(const char *line); | |
163 | ||
164 | /* cgroupfs v1 prototypes. */ | |
165 | struct cgv1_hierarchy { | |
166 | char **controllers; | |
167 | char *mountpoint; | |
168 | char *base_cgroup; | |
169 | char *fullcgpath; | |
170 | char *init_cgroup; | |
171 | bool create_rw_cgroup; | |
172 | bool systemd_user_slice; | |
173 | }; | |
174 | ||
175 | static struct cgv1_hierarchy **cgv1_hierarchies; | |
176 | ||
177 | static void cgv1_add_controller(char **clist, char *mountpoint, | |
178 | char *base_cgroup, char *init_cgroup); | |
179 | static bool cgv1_controller_in_clist(char *cgline, char *c); | |
180 | static bool cgv1_controller_lists_intersect(char **l1, char **l2); | |
181 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, | |
182 | char **clist); | |
183 | static bool cgv1_create(const char *cgroup, uid_t uid, gid_t gid, | |
184 | bool *existed); | |
185 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, | |
186 | uid_t uid, gid_t gid, bool *existed); | |
187 | static bool cgv1_enter(const char *cgroup); | |
188 | static void cgv1_escape(void); | |
189 | static bool cgv1_get_controllers(char ***klist, char ***nlist); | |
190 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller); | |
191 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, | |
192 | char *line); | |
193 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, | |
194 | const char *cgroup); | |
195 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h); | |
196 | static bool cgv1_init(uid_t uid, gid_t gid); | |
197 | static void cgv1_mark_to_make_rw(char **clist); | |
198 | static char *cgv1_must_prefix_named(char *entry); | |
199 | static bool cgv1_prune_empty_cgroups(const char *user); | |
200 | static bool cgv1_remove_one(struct cgv1_hierarchy *h, const char *cgroup); | |
201 | static bool is_cgv1(char *line); | |
202 | ||
203 | /* cgroupfs v2 prototypes. */ | |
204 | struct cgv2_hierarchy { | |
205 | char **controllers; | |
206 | char *mountpoint; | |
207 | char *base_cgroup; | |
208 | char *fullcgpath; | |
209 | char *init_cgroup; | |
210 | bool create_rw_cgroup; | |
211 | bool systemd_user_slice; | |
212 | }; | |
213 | ||
214 | /* Actually this should only be a single hierarchy. But for the sake of | |
215 | * parallelism and because the layout of the cgroupfs v2 is still somewhat | |
216 | * changing, we'll leave it as an array of structs. | |
217 | */ | |
218 | static struct cgv2_hierarchy **cgv2_hierarchies; | |
219 | ||
220 | static void cgv2_add_controller(char **clist, char *mountpoint, | |
221 | char *base_cgroup, char *init_cgroup, | |
222 | bool systemd_user_slice); | |
223 | static bool cgv2_create(const char *cgroup, uid_t uid, gid_t gid, | |
224 | bool *existed); | |
225 | static bool cgv2_enter(const char *cgroup); | |
226 | static void cgv2_escape(void); | |
227 | static char *cgv2_get_current_cgroup(int pid); | |
228 | static bool cgv2_init(uid_t uid, gid_t gid); | |
229 | static void cgv2_mark_to_make_rw(char **clist); | |
230 | static bool cgv2_prune_empty_cgroups(const char *user); | |
231 | static bool cgv2_remove(const char *cgroup); | |
232 | static bool is_cgv2(char *line); | |
233 | ||
c4a4578f JC |
234 | static int do_mkdir(const char *path, mode_t mode) |
235 | { | |
236 | int saved_errno; | |
237 | mode_t mask; | |
238 | int r; | |
239 | ||
240 | mask = umask(0); | |
241 | r = mkdir(path, mode); | |
242 | saved_errno = errno; | |
243 | umask(mask); | |
244 | errno = saved_errno; | |
245 | return (r); | |
246 | } | |
247 | ||
7ac43d3d CB |
248 | /* Create directory and (if necessary) its parents. */ |
249 | static bool mkdir_parent(const char *root, char *path) | |
250 | { | |
251 | char *b, orig, *e; | |
252 | ||
253 | if (strlen(path) < strlen(root)) | |
254 | return false; | |
255 | ||
256 | if (strlen(path) == strlen(root)) | |
257 | return true; | |
258 | ||
259 | b = path + strlen(root) + 1; | |
260 | while (true) { | |
261 | while (*b && (*b == '/')) | |
262 | b++; | |
263 | if (!*b) | |
264 | return true; | |
265 | ||
266 | e = b + 1; | |
267 | while (*e && *e != '/') | |
268 | e++; | |
269 | ||
270 | orig = *e; | |
271 | if (orig) | |
272 | *e = '\0'; | |
273 | ||
274 | if (file_exists(path)) | |
275 | goto next; | |
276 | ||
c4a4578f | 277 | if (do_mkdir(path, 0755) < 0) { |
2082ac62 | 278 | pam_cgfs_debug("Failed to create %s: %s\n", path, strerror(errno)); |
7ac43d3d CB |
279 | return false; |
280 | } | |
281 | ||
282 | next: | |
283 | if (!orig) | |
284 | return true; | |
285 | ||
286 | *e = orig; | |
287 | b = e + 1; | |
288 | } | |
289 | ||
290 | return false; | |
291 | } | |
292 | ||
293 | /* Common helper functions. Most of these have been taken from LXC. */ | |
294 | static void mysyslog(int err, const char *format, ...) | |
295 | { | |
296 | va_list args; | |
297 | ||
298 | va_start(args, format); | |
299 | openlog("PAM-CGFS", LOG_CONS | LOG_PID, LOG_AUTH); | |
300 | vsyslog(err, format, args); | |
301 | va_end(args); | |
302 | closelog(); | |
303 | } | |
304 | ||
305 | /* realloc() pointer in batch sizes; do not fail. */ | |
306 | #define BATCH_SIZE 50 | |
307 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
308 | { | |
309 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
310 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
311 | ||
312 | if (!*mem || newbatches > oldbatches) | |
313 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
314 | } | |
315 | ||
316 | /* Append lines as is to pointer; do not fail. */ | |
317 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
318 | { | |
319 | size_t full = oldlen + newlen; | |
320 | ||
321 | batch_realloc(dest, oldlen, full + 1); | |
322 | ||
323 | memcpy(*dest + oldlen, new, newlen + 1); | |
324 | } | |
325 | ||
326 | /* Read in whole file and return allocated pointer. */ | |
327 | static char *read_file(char *fnam) | |
328 | { | |
329 | FILE *f; | |
330 | int linelen; | |
331 | char *line = NULL, *buf = NULL; | |
332 | size_t len = 0, fulllen = 0; | |
333 | ||
334 | f = fopen(fnam, "r"); | |
335 | if (!f) | |
336 | return NULL; | |
337 | ||
338 | while ((linelen = getline(&line, &len, f)) != -1) { | |
339 | append_line(&buf, fulllen, line, linelen); | |
340 | fulllen += linelen; | |
341 | } | |
342 | ||
343 | fclose(f); | |
344 | free(line); | |
345 | ||
346 | return buf; | |
347 | } | |
348 | ||
349 | /* Given a pointer to a null-terminated array of pointers, realloc to add one | |
350 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
351 | * second-to-last entry - that is, the one which is now available for use | |
352 | * (keeping the list null-terminated). | |
353 | */ | |
354 | static int append_null_to_list(void ***list) | |
355 | { | |
356 | int newentry = 0; | |
357 | ||
358 | if (*list) | |
2082ac62 | 359 | for (; (*list)[newentry]; newentry++) |
7ac43d3d | 360 | ; |
7ac43d3d CB |
361 | |
362 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
363 | (*list)[newentry + 1] = NULL; | |
364 | ||
365 | return newentry; | |
366 | } | |
367 | ||
368 | /* Append new entry to null-terminated array of pointer; make sure that array of | |
369 | * pointers will still be null-terminated. | |
370 | */ | |
371 | static void must_append_string(char ***list, char *entry) | |
372 | { | |
373 | int newentry; | |
374 | char *copy; | |
375 | ||
376 | newentry = append_null_to_list((void ***)list); | |
377 | copy = must_copy_string(entry); | |
378 | (*list)[newentry] = copy; | |
379 | } | |
380 | ||
381 | /* Remove newlines from string. */ | |
382 | static void trim(char *s) | |
383 | { | |
384 | size_t len = strlen(s); | |
385 | ||
386 | while ((len > 0) && s[len - 1] == '\n') | |
387 | s[--len] = '\0'; | |
388 | } | |
389 | ||
7ac43d3d CB |
390 | /* Make allocated copy of string. End of string is taken to be '\n'. */ |
391 | static char *copy_to_eol(char *s) | |
392 | { | |
393 | char *newline, *sret; | |
394 | size_t len; | |
395 | ||
396 | newline = strchr(s, '\n'); | |
397 | if (!newline) | |
398 | return NULL; | |
399 | ||
400 | len = newline - s; | |
f25a2044 | 401 | sret = must_realloc(NULL, len + 1); |
7ac43d3d CB |
402 | memcpy(sret, s, len); |
403 | sret[len] = '\0'; | |
404 | ||
405 | return sret; | |
406 | } | |
407 | ||
408 | /* Check if given entry under /proc/<pid>/mountinfo is a fuse.lxcfs mount. */ | |
409 | static bool is_lxcfs(const char *line) | |
410 | { | |
411 | char *p = strstr(line, " - "); | |
412 | if (!p) | |
413 | return false; | |
414 | ||
415 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; | |
416 | } | |
417 | ||
418 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v1 mount. */ | |
419 | static bool is_cgv1(char *line) | |
420 | { | |
421 | char *p = strstr(line, " - "); | |
422 | if (!p) | |
423 | return false; | |
424 | ||
425 | return strncmp(p, " - cgroup ", 10) == 0; | |
426 | } | |
427 | ||
428 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v2 mount. */ | |
429 | static bool is_cgv2(char *line) | |
430 | { | |
431 | char *p = strstr(line, " - "); | |
432 | if (!p) | |
433 | return false; | |
434 | ||
435 | return strncmp(p, " - cgroup2 ", 11) == 0; | |
436 | } | |
437 | ||
438 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
439 | * strings | |
440 | */ | |
441 | static bool string_in_list(char **list, const char *entry) | |
442 | { | |
443 | char **it; | |
444 | ||
445 | for (it = list; it && *it; it++) | |
446 | if (strcmp(*it, entry) == 0) | |
447 | return true; | |
448 | ||
449 | return false; | |
450 | } | |
451 | ||
452 | /* | |
453 | * Creates a null-terminated array of strings, made by splitting the entries in | |
454 | * @str on each @sep. Caller is responsible for calling free_string_list. | |
455 | */ | |
456 | static char **make_string_list(const char *str, const char *sep) | |
457 | { | |
458 | char *copy, *tok; | |
459 | char *saveptr = NULL; | |
460 | char **clist = NULL; | |
461 | ||
462 | copy = must_copy_string(str); | |
463 | ||
464 | for (tok = strtok_r(copy, sep, &saveptr); tok; | |
465 | tok = strtok_r(NULL, sep, &saveptr)) | |
466 | must_add_to_list(&clist, tok); | |
467 | ||
468 | free(copy); | |
469 | ||
470 | return clist; | |
471 | } | |
472 | ||
473 | /* Gets the length of a null-terminated array of strings. */ | |
474 | static size_t string_list_length(char **list) | |
475 | { | |
476 | size_t len = 0; | |
477 | char **it; | |
478 | ||
479 | for (it = list; it && *it; it++) | |
480 | len++; | |
481 | ||
482 | return len; | |
483 | } | |
484 | ||
485 | /* Free null-terminated array of strings. */ | |
486 | static void free_string_list(char **list) | |
487 | { | |
488 | char **it; | |
489 | ||
490 | for (it = list; it && *it; it++) | |
491 | free(*it); | |
492 | free(list); | |
493 | } | |
494 | ||
495 | /* Write single integer to file. */ | |
496 | static bool write_int(char *path, int v) | |
497 | { | |
498 | FILE *f; | |
499 | bool ret = true; | |
500 | ||
501 | f = fopen(path, "w"); | |
502 | if (!f) | |
503 | return false; | |
504 | ||
505 | if (fprintf(f, "%d\n", v) < 0) | |
506 | ret = false; | |
507 | ||
508 | if (fclose(f) != 0) | |
509 | ret = false; | |
510 | ||
511 | return ret; | |
512 | } | |
513 | ||
514 | /* Recursively remove directory and its parents. */ | |
515 | static int recursive_rmdir(char *dirname) | |
516 | { | |
517 | struct dirent *direntp; | |
518 | DIR *dir; | |
519 | int r = 0; | |
520 | ||
521 | dir = opendir(dirname); | |
522 | if (!dir) | |
523 | return -ENOENT; | |
524 | ||
525 | while ((direntp = readdir(dir))) { | |
526 | struct stat st; | |
527 | char *pathname; | |
528 | ||
7ac43d3d CB |
529 | if (!strcmp(direntp->d_name, ".") || |
530 | !strcmp(direntp->d_name, "..")) | |
531 | continue; | |
532 | ||
533 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
534 | ||
535 | if (lstat(pathname, &st)) { | |
536 | if (!r) | |
2082ac62 | 537 | pam_cgfs_debug("Failed to stat %s\n", pathname); |
7ac43d3d CB |
538 | r = -1; |
539 | goto next; | |
540 | } | |
541 | ||
542 | if (!S_ISDIR(st.st_mode)) | |
543 | goto next; | |
544 | ||
545 | if (recursive_rmdir(pathname) < 0) | |
546 | r = -1; | |
2082ac62 | 547 | |
7ac43d3d CB |
548 | next: |
549 | free(pathname); | |
550 | } | |
551 | ||
552 | if (rmdir(dirname) < 0) { | |
553 | if (!r) | |
2082ac62 | 554 | pam_cgfs_debug("Failed to delete %s: %s\n", dirname, strerror(errno)); |
7ac43d3d CB |
555 | r = -1; |
556 | } | |
557 | ||
558 | if (closedir(dir) < 0) { | |
559 | if (!r) | |
2082ac62 | 560 | pam_cgfs_debug("Failed to delete %s: %s\n", dirname, strerror(errno)); |
7ac43d3d CB |
561 | r = -1; |
562 | } | |
563 | ||
564 | return r; | |
565 | } | |
566 | ||
567 | /* Add new entry to null-terminated array of pointers. Make sure array is still | |
568 | * null-terminated. | |
569 | */ | |
570 | static void must_add_to_list(char ***clist, char *entry) | |
571 | { | |
572 | int newentry; | |
573 | ||
574 | newentry = append_null_to_list((void ***)clist); | |
575 | (*clist)[newentry] = must_copy_string(entry); | |
576 | } | |
577 | ||
578 | /* Get mountpoint from a /proc/<pid>/mountinfo line. */ | |
579 | static char *get_mountpoint(char *line) | |
580 | { | |
581 | int i; | |
582 | char *p, *sret, *p2; | |
583 | size_t len; | |
584 | ||
585 | p = line; | |
586 | ||
587 | for (i = 0; i < 4; i++) { | |
588 | p = strchr(p, ' '); | |
589 | if (!p) | |
590 | return NULL; | |
591 | p++; | |
592 | } | |
593 | ||
594 | p2 = strchr(p, ' '); | |
595 | if (p2) | |
596 | *p2 = '\0'; | |
597 | ||
598 | len = strlen(p); | |
f25a2044 | 599 | sret = must_realloc(NULL, len + 1); |
7ac43d3d CB |
600 | memcpy(sret, p, len); |
601 | sret[len] = '\0'; | |
602 | ||
603 | return sret; | |
604 | } | |
605 | ||
606 | /* Create list of cgroupfs v1 controller found under /proc/self/cgroup. Skips | |
607 | * the 0::/some/path cgroupfs v2 hierarchy listed. Splits controllers into | |
608 | * kernel controllers (@klist) and named controllers (@nlist). | |
609 | */ | |
610 | static bool cgv1_get_controllers(char ***klist, char ***nlist) | |
611 | { | |
612 | FILE *f; | |
613 | char *line = NULL; | |
614 | size_t len = 0; | |
615 | ||
616 | f = fopen("/proc/self/cgroup", "r"); | |
617 | if (!f) | |
618 | return false; | |
619 | ||
620 | while (getline(&line, &len, f) != -1) { | |
621 | char *p, *p2, *tok; | |
622 | char *saveptr = NULL; | |
623 | ||
624 | p = strchr(line, ':'); | |
625 | if (!p) | |
626 | continue; | |
627 | p++; | |
628 | ||
629 | p2 = strchr(p, ':'); | |
630 | if (!p2) | |
631 | continue; | |
632 | *p2 = '\0'; | |
633 | ||
634 | /* Skip the v2 hierarchy. */ | |
635 | if ((p2 - p) == 0) | |
636 | continue; | |
637 | ||
638 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
639 | tok = strtok_r(NULL, ",", &saveptr)) { | |
640 | if (strncmp(tok, "name=", 5) == 0) | |
641 | must_append_string(nlist, tok); | |
642 | else | |
643 | must_append_string(klist, tok); | |
644 | } | |
645 | } | |
646 | ||
647 | free(line); | |
648 | fclose(f); | |
649 | ||
650 | return true; | |
651 | } | |
652 | ||
653 | /* Get list of controllers for cgroupfs v2 hierarchy by looking at | |
654 | * cgroup.controllers and/or cgroup.subtree_control of a given (parent) cgroup. | |
655 | static bool cgv2_get_controllers(char ***klist) | |
656 | { | |
657 | return -ENOSYS; | |
658 | } | |
659 | */ | |
660 | ||
661 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
662 | static char *cgv2_get_current_cgroup(int pid) | |
663 | { | |
664 | int ret; | |
665 | char *cgroups_v2; | |
666 | char *current_cgroup; | |
667 | char *copy = NULL; | |
668 | /* The largest integer that can fit into long int is 2^64. This is a | |
669 | * 20-digit number. */ | |
670 | #define __PIDLEN /* /proc */ 5 + /* /pid-to-str */ 21 + /* /cgroup */ 7 + /* \0 */ 1 | |
671 | char path[__PIDLEN]; | |
672 | ||
673 | ret = snprintf(path, __PIDLEN, "/proc/%d/cgroup", pid); | |
674 | if (ret < 0 || ret >= __PIDLEN) | |
675 | return NULL; | |
676 | ||
677 | cgroups_v2 = read_file(path); | |
678 | if (!cgroups_v2) | |
679 | return NULL; | |
680 | ||
681 | current_cgroup = strstr(cgroups_v2, "0::/"); | |
682 | if (!current_cgroup) | |
683 | goto cleanup_on_err; | |
684 | ||
685 | current_cgroup = current_cgroup + 3; | |
686 | copy = copy_to_eol(current_cgroup); | |
687 | if (!copy) | |
688 | goto cleanup_on_err; | |
689 | ||
690 | cleanup_on_err: | |
691 | free(cgroups_v2); | |
692 | if (copy) | |
693 | trim(copy); | |
694 | ||
695 | return copy; | |
696 | } | |
697 | ||
698 | /* Given two null-terminated lists of strings, return true if any string is in | |
699 | * both. | |
700 | */ | |
701 | static bool cgv1_controller_lists_intersect(char **l1, char **l2) | |
702 | { | |
703 | char **it; | |
704 | ||
705 | if (!l2) | |
706 | return false; | |
707 | ||
708 | for (it = l1; it && *it; it++) | |
709 | if (string_in_list(l2, *it)) | |
710 | return true; | |
711 | ||
712 | return false; | |
713 | } | |
714 | ||
715 | /* For a null-terminated list of controllers @clist, return true if any of those | |
716 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
717 | * Realistically, if one is present, all must be present. | |
718 | */ | |
719 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, char **clist) | |
720 | { | |
721 | struct cgv1_hierarchy **it; | |
722 | ||
723 | for (it = hlist; it && *it; it++) | |
724 | if ((*it)->controllers) | |
725 | if (cgv1_controller_lists_intersect((*it)->controllers, clist)) | |
726 | return true; | |
2082ac62 | 727 | |
7ac43d3d CB |
728 | return false; |
729 | ||
730 | } | |
731 | ||
732 | /* Set boolean to mark controllers under which we are supposed create a | |
733 | * writeable cgroup. | |
734 | */ | |
735 | static void cgv1_mark_to_make_rw(char **clist) | |
736 | { | |
737 | struct cgv1_hierarchy **it; | |
738 | ||
739 | for (it = cgv1_hierarchies; it && *it; it++) | |
740 | if ((*it)->controllers) | |
741 | if (cgv1_controller_lists_intersect((*it)->controllers, clist) || | |
2082ac62 | 742 | string_in_list(clist, "all")) |
7ac43d3d CB |
743 | (*it)->create_rw_cgroup = true; |
744 | } | |
745 | ||
746 | /* Set boolean to mark whether we are supposed to create a writeable cgroup in | |
747 | * the cgroupfs v2 hierarchy. | |
748 | */ | |
749 | static void cgv2_mark_to_make_rw(char **clist) | |
750 | { | |
751 | if (string_in_list(clist, "unified") || string_in_list(clist, "all")) | |
752 | if (cgv2_hierarchies) | |
753 | (*cgv2_hierarchies)->create_rw_cgroup = true; | |
754 | } | |
755 | ||
756 | /* Wrapper around cgv{1,2}_mark_to_make_rw(). */ | |
757 | static void cg_mark_to_make_rw(char **clist) | |
758 | { | |
759 | cgv1_mark_to_make_rw(clist); | |
760 | cgv2_mark_to_make_rw(clist); | |
761 | } | |
762 | ||
763 | /* Prefix any named controllers with "name=", e.g. "name=systemd". */ | |
764 | static char *cgv1_must_prefix_named(char *entry) | |
765 | { | |
766 | char *s; | |
767 | int ret; | |
768 | size_t len; | |
769 | ||
770 | len = strlen(entry); | |
f25a2044 | 771 | s = must_realloc(NULL, len + 6); |
7ac43d3d CB |
772 | |
773 | ret = snprintf(s, len + 6, "name=%s", entry); | |
8ddce7df | 774 | if (ret < 0 || (size_t)ret >= (len + 6)) { |
775 | free(s); | |
7ac43d3d | 776 | return NULL; |
8ddce7df | 777 | } |
7ac43d3d CB |
778 | |
779 | return s; | |
780 | } | |
781 | ||
782 | /* Append kernel controller in @klist or named controller in @nlist to @clist */ | |
783 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
784 | { | |
785 | int newentry; | |
786 | char *copy; | |
787 | ||
788 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
789 | return; | |
790 | ||
791 | newentry = append_null_to_list((void ***)clist); | |
792 | ||
793 | if (strncmp(entry, "name=", 5) == 0) | |
794 | copy = must_copy_string(entry); | |
795 | else if (string_in_list(klist, entry)) | |
796 | copy = must_copy_string(entry); | |
797 | else | |
798 | copy = cgv1_must_prefix_named(entry); | |
799 | ||
800 | (*clist)[newentry] = copy; | |
801 | } | |
802 | ||
803 | /* Get the controllers from a mountinfo line. There are other ways we could get | |
804 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
805 | * could parse the mount options. But we simply assume that the mountpoint must | |
806 | * be /sys/fs/cgroup/controller-list | |
807 | */ | |
808 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, char *line) | |
809 | { | |
810 | int i; | |
811 | char *p, *p2, *tok; | |
812 | char *saveptr = NULL; | |
813 | char **aret = NULL; | |
814 | ||
815 | p = line; | |
816 | ||
817 | for (i = 0; i < 4; i++) { | |
818 | p = strchr(p, ' '); | |
819 | if (!p) | |
820 | return NULL; | |
821 | p++; | |
822 | } | |
7ac43d3d CB |
823 | |
824 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) | |
825 | return NULL; | |
826 | ||
827 | p += 15; | |
828 | ||
829 | p2 = strchr(p, ' '); | |
830 | if (!p2) | |
831 | return NULL; | |
832 | *p2 = '\0'; | |
833 | ||
834 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
835 | tok = strtok_r(NULL, ",", &saveptr)) | |
836 | must_append_controller(klist, nlist, &aret, tok); | |
837 | ||
838 | return aret; | |
839 | } | |
840 | ||
841 | /* Check if a cgroupfs v2 controller is present in the string @cgline. */ | |
842 | static bool cgv1_controller_in_clist(char *cgline, char *c) | |
843 | { | |
844 | size_t len; | |
845 | char *tok, *eol, *tmp; | |
846 | char *saveptr = NULL; | |
847 | ||
848 | eol = strchr(cgline, ':'); | |
849 | if (!eol) | |
850 | return false; | |
851 | ||
852 | len = eol - cgline; | |
853 | tmp = alloca(len + 1); | |
854 | memcpy(tmp, cgline, len); | |
855 | tmp[len] = '\0'; | |
856 | ||
857 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
858 | tok = strtok_r(NULL, ",", &saveptr)) { | |
859 | if (strcmp(tok, c) == 0) | |
860 | return true; | |
861 | } | |
2082ac62 | 862 | |
7ac43d3d CB |
863 | return false; |
864 | } | |
865 | ||
866 | /* Get current cgroup from the /proc/<pid>/cgroup file passed in via @basecginfo | |
867 | * of a given cgv1 controller passed in via @controller. | |
868 | */ | |
869 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller) | |
870 | { | |
871 | char *p; | |
872 | ||
873 | p = basecginfo; | |
874 | ||
875 | while (true) { | |
876 | p = strchr(p, ':'); | |
877 | if (!p) | |
878 | return NULL; | |
879 | p++; | |
880 | ||
881 | if (cgv1_controller_in_clist(p, controller)) { | |
882 | p = strchr(p, ':'); | |
883 | if (!p) | |
884 | return NULL; | |
885 | p++; | |
886 | ||
887 | return copy_to_eol(p); | |
888 | } | |
889 | ||
890 | p = strchr(p, '\n'); | |
891 | if (!p) | |
892 | return NULL; | |
893 | p++; | |
894 | } | |
895 | ||
896 | return NULL; | |
897 | } | |
898 | ||
899 | /* Remove /init.scope from string @cg. This will mostly affect systemd-based | |
900 | * systems. | |
901 | */ | |
902 | #define INIT_SCOPE "/init.scope" | |
903 | static void cg_systemd_prune_init_scope(char *cg) | |
904 | { | |
905 | char *point; | |
906 | ||
907 | if (!cg) | |
908 | return; | |
909 | ||
910 | point = cg + strlen(cg) - strlen(INIT_SCOPE); | |
911 | if (point < cg) | |
912 | return; | |
913 | ||
914 | if (strcmp(point, INIT_SCOPE) == 0) { | |
915 | if (point == cg) | |
916 | *(point + 1) = '\0'; | |
917 | else | |
918 | *point = '\0'; | |
919 | } | |
920 | } | |
921 | ||
922 | /* Add new info about a mounted cgroupfs v1 hierarchy. Includes the controllers | |
923 | * mounted into that hierarchy (e.g. cpu,cpuacct), the mountpoint of that | |
924 | * hierarchy (/sys/fs/cgroup/<controller>, the base cgroup of the current | |
925 | * process gathered from /proc/self/cgroup, and the init cgroup of PID1 gathered | |
926 | * from /proc/1/cgroup. | |
927 | */ | |
928 | static void cgv1_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup) | |
929 | { | |
930 | struct cgv1_hierarchy *new; | |
931 | int newentry; | |
932 | ||
f25a2044 | 933 | new = must_realloc(NULL, sizeof(*new)); |
2082ac62 | 934 | |
7ac43d3d CB |
935 | new->controllers = clist; |
936 | new->mountpoint = mountpoint; | |
937 | new->base_cgroup = base_cgroup; | |
938 | new->fullcgpath = NULL; | |
939 | new->create_rw_cgroup = false; | |
940 | new->init_cgroup = init_cgroup; | |
941 | new->systemd_user_slice = false; | |
942 | ||
943 | newentry = append_null_to_list((void ***)&cgv1_hierarchies); | |
944 | cgv1_hierarchies[newentry] = new; | |
945 | } | |
946 | ||
947 | /* Add new info about the mounted cgroupfs v2 hierarchy. Can (but doesn't | |
948 | * currently) include the controllers mounted into the hierarchy (e.g. memory, | |
949 | * pids, blkio), the mountpoint of that hierarchy (Should usually be | |
950 | * /sys/fs/cgroup but some init systems seems to think it might be a good idea | |
951 | * to also mount empty cgroupfs v2 hierarchies at /sys/fs/cgroup/systemd.), the | |
952 | * base cgroup of the current process gathered from /proc/self/cgroup, and the | |
953 | * init cgroup of PID1 gathered from /proc/1/cgroup. | |
954 | */ | |
955 | static void cgv2_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup, bool systemd_user_slice) | |
956 | { | |
957 | struct cgv2_hierarchy *new; | |
958 | int newentry; | |
959 | ||
f25a2044 | 960 | new = must_realloc(NULL, sizeof(*new)); |
2082ac62 | 961 | |
7ac43d3d CB |
962 | new->controllers = clist; |
963 | new->mountpoint = mountpoint; | |
964 | new->base_cgroup = base_cgroup; | |
965 | new->fullcgpath = NULL; | |
966 | new->create_rw_cgroup = false; | |
967 | new->init_cgroup = init_cgroup; | |
968 | new->systemd_user_slice = systemd_user_slice; | |
969 | ||
970 | newentry = append_null_to_list((void ***)&cgv2_hierarchies); | |
971 | cgv2_hierarchies[newentry] = new; | |
972 | } | |
973 | ||
974 | /* In Ubuntu 14.04, the paths created for us were | |
975 | * '/user/$uid.user/$something.session' This can be merged better with | |
976 | * systemd_created_slice_for_us(), but keeping it separate makes it easier to | |
977 | * reason about the correctness. | |
978 | */ | |
979 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid) | |
980 | { | |
981 | char *p; | |
982 | size_t len; | |
983 | int id; | |
984 | char *copy = NULL; | |
985 | bool bret = false; | |
986 | ||
987 | copy = must_copy_string(in); | |
988 | if (strlen(copy) < strlen("/user/1.user/1.session")) | |
989 | goto cleanup; | |
990 | p = copy + strlen(copy) - 1; | |
991 | ||
992 | /* skip any trailing '/' (shouldn't be any, but be sure) */ | |
993 | while (p >= copy && *p == '/') | |
994 | *(p--) = '\0'; | |
995 | if (p < copy) | |
996 | goto cleanup; | |
997 | ||
998 | /* Get last path element */ | |
999 | while (p >= copy && *p != '/') | |
1000 | p--; | |
1001 | if (p < copy) | |
1002 | goto cleanup; | |
2082ac62 | 1003 | |
7ac43d3d CB |
1004 | /* make sure it is something.session */ |
1005 | len = strlen(p + 1); | |
1006 | if (len < strlen("1.session") || | |
1007 | strncmp(p + 1 + len - 8, ".session", 8) != 0) | |
1008 | goto cleanup; | |
1009 | ||
1010 | /* ok last path piece checks out, now check the second to last */ | |
1011 | *(p + 1) = '\0'; | |
1012 | while (p >= copy && *(--p) != '/') | |
1013 | ; | |
2082ac62 | 1014 | |
7ac43d3d CB |
1015 | if (sscanf(p + 1, "%d.user/", &id) != 1) |
1016 | goto cleanup; | |
1017 | ||
1018 | if (id != (int)uid) | |
1019 | goto cleanup; | |
1020 | ||
1021 | bret = true; | |
1022 | ||
1023 | cleanup: | |
1024 | free(copy); | |
1025 | return bret; | |
1026 | } | |
1027 | ||
1028 | /* So long as our path relative to init starts with /user.slice/user-$uid.slice, | |
1029 | * assume it belongs to $uid and chown it | |
1030 | */ | |
1031 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
1032 | const char *init_cgroup, uid_t uid) | |
1033 | { | |
1034 | int ret; | |
1035 | char buf[100]; | |
1036 | size_t curlen, initlen; | |
1037 | ||
1038 | curlen = strlen(base_cgroup); | |
1039 | initlen = strlen(init_cgroup); | |
1040 | if (curlen <= initlen) | |
1041 | return false; | |
1042 | ||
1043 | if (strncmp(base_cgroup, init_cgroup, initlen) != 0) | |
1044 | return false; | |
1045 | ||
1046 | ret = snprintf(buf, 100, "/user.slice/user-%d.slice/", (int)uid); | |
1047 | if (ret < 0 || ret >= 100) | |
1048 | return false; | |
1049 | ||
1050 | if (initlen == 1) | |
1051 | initlen = 0; // skip the '/' | |
1052 | ||
1053 | return strncmp(base_cgroup + initlen, buf, strlen(buf)) == 0; | |
1054 | } | |
1055 | ||
1056 | /* The systemd-created path is: user-$uid.slice/session-c$session.scope. If that | |
1057 | * is not the end of our systemd path, then we're not part of the PAM call that | |
1058 | * created that path. | |
1059 | * | |
1060 | * The last piece is chowned to $uid, the user- part not. | |
1061 | * Note: If the user creates paths that look like what we're looking for to | |
1062 | * 'fool' us, either | |
1063 | * - they fool us, we create new cgroups, and they get auto-logged-out. | |
1064 | * - they fool a root sudo, systemd cgroup is not changed but chowned, and they | |
1065 | * lose ownership of their cgroups | |
1066 | */ | |
1067 | static bool cg_systemd_created_user_slice(const char *base_cgroup, | |
1068 | const char *init_cgroup, | |
1069 | const char *in, uid_t uid) | |
1070 | { | |
1071 | char *p; | |
1072 | size_t len; | |
1073 | int id; | |
1074 | char *copy = NULL; | |
1075 | bool bret = false; | |
1076 | ||
1077 | copy = must_copy_string(in); | |
1078 | ||
1079 | /* An old version of systemd has already created a cgroup for us. */ | |
1080 | if (cg_systemd_under_user_slice_1(in, uid)) | |
1081 | goto succeed; | |
1082 | ||
1083 | /* A new version of systemd has already created a cgroup for us. */ | |
1084 | if (cg_systemd_under_user_slice_2(base_cgroup, init_cgroup, uid)) | |
1085 | goto succeed; | |
1086 | ||
1087 | if (strlen(copy) < strlen("/user-0.slice/session-0.scope")) | |
1088 | goto cleanup; | |
1089 | ||
1090 | p = copy + strlen(copy) - 1; | |
1091 | /* Skip any trailing '/' (shouldn't be any, but be sure). */ | |
1092 | while (p >= copy && *p == '/') | |
1093 | *(p--) = '\0'; | |
1094 | ||
1095 | if (p < copy) | |
1096 | goto cleanup; | |
1097 | ||
1098 | /* Get last path element */ | |
1099 | while (p >= copy && *p != '/') | |
1100 | p--; | |
1101 | ||
1102 | if (p < copy) | |
1103 | goto cleanup; | |
1104 | ||
1105 | /* Make sure it is session-something.scope. */ | |
1106 | len = strlen(p + 1); | |
1107 | if (strncmp(p + 1, "session-", strlen("session-")) != 0 || | |
1108 | strncmp(p + 1 + len - 6, ".scope", 6) != 0) | |
1109 | goto cleanup; | |
1110 | ||
1111 | /* Ok last path piece checks out, now check the second to last. */ | |
1112 | *(p + 1) = '\0'; | |
1113 | while (p >= copy && *(--p) != '/') | |
1114 | ; | |
1115 | ||
1116 | if (sscanf(p + 1, "user-%d.slice/", &id) != 1) | |
1117 | goto cleanup; | |
1118 | ||
1119 | if (id != (int)uid) | |
1120 | goto cleanup; | |
1121 | ||
1122 | succeed: | |
1123 | bret = true; | |
2082ac62 | 1124 | |
7ac43d3d CB |
1125 | cleanup: |
1126 | free(copy); | |
1127 | return bret; | |
1128 | } | |
1129 | ||
1130 | /* Chown existing cgroup that systemd has already created for us. */ | |
1131 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
1132 | const char *base_cgroup, uid_t uid, | |
1133 | gid_t gid, bool systemd_user_slice) | |
1134 | { | |
1135 | char *path; | |
1136 | ||
1137 | if (!systemd_user_slice) | |
1138 | return false; | |
1139 | ||
1140 | path = must_make_path(mountpoint, base_cgroup, NULL); | |
1141 | ||
1142 | /* A cgroup within name=systemd has already been created. So we only | |
1143 | * need to chown it. | |
1144 | */ | |
1145 | if (chown(path, uid, gid) < 0) | |
2082ac62 | 1146 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s\n", |
1147 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
1148 | pam_cgfs_debug("Chowned %s to %d:%d\n", path, (int)uid, (int)gid); | |
7ac43d3d CB |
1149 | |
1150 | free(path); | |
1151 | return true; | |
1152 | } | |
1153 | ||
1154 | /* Detect and store information about cgroupfs v1 hierarchies. */ | |
1155 | static bool cgv1_init(uid_t uid, gid_t gid) | |
1156 | { | |
1157 | FILE *f; | |
1158 | struct cgv1_hierarchy **it; | |
1159 | char *basecginfo; | |
1160 | char *line = NULL; | |
1161 | char **klist = NULL, **nlist = NULL; | |
1162 | size_t len = 0; | |
1163 | ||
1164 | basecginfo = read_file("/proc/self/cgroup"); | |
1165 | if (!basecginfo) | |
1166 | return false; | |
1167 | ||
1168 | f = fopen("/proc/self/mountinfo", "r"); | |
1169 | if (!f) { | |
1170 | free(basecginfo); | |
1171 | return false; | |
1172 | } | |
1173 | ||
1174 | cgv1_get_controllers(&klist, &nlist); | |
1175 | ||
1176 | while (getline(&line, &len, f) != -1) { | |
1177 | char **controller_list = NULL; | |
1178 | char *mountpoint, *base_cgroup; | |
1179 | ||
1180 | if (is_lxcfs(line) || !is_cgv1(line)) | |
1181 | continue; | |
1182 | ||
1183 | controller_list = cgv1_get_proc_mountinfo_controllers(klist, nlist, line); | |
1184 | if (!controller_list) | |
1185 | continue; | |
1186 | ||
9159b38c | 1187 | if (cgv1_controller_list_is_dup(cgv1_hierarchies, controller_list)) { |
7ac43d3d CB |
1188 | free(controller_list); |
1189 | continue; | |
1190 | } | |
1191 | ||
1192 | mountpoint = get_mountpoint(line); | |
1193 | if (!mountpoint) { | |
1194 | free_string_list(controller_list); | |
1195 | continue; | |
1196 | } | |
1197 | ||
1198 | base_cgroup = cgv1_get_current_cgroup(basecginfo, controller_list[0]); | |
1199 | if (!base_cgroup) { | |
1200 | free_string_list(controller_list); | |
1201 | free(mountpoint); | |
1202 | continue; | |
1203 | } | |
9159b38c | 1204 | |
7ac43d3d CB |
1205 | trim(base_cgroup); |
1206 | pam_cgfs_debug("Detected cgroupfs v1 controller \"%s\" with " | |
9159b38c | 1207 | "mountpoint \"%s\" and cgroup \"%s\"\n", |
1208 | controller_list[0], mountpoint, base_cgroup); | |
1209 | cgv1_add_controller(controller_list, mountpoint, base_cgroup, NULL); | |
7ac43d3d | 1210 | } |
9159b38c | 1211 | |
7ac43d3d CB |
1212 | free_string_list(klist); |
1213 | free_string_list(nlist); | |
1214 | free(basecginfo); | |
1215 | fclose(f); | |
1216 | free(line); | |
1217 | ||
1218 | /* Retrieve init cgroup path for all controllers. */ | |
1219 | basecginfo = read_file("/proc/1/cgroup"); | |
1220 | if (!basecginfo) | |
1221 | return false; | |
1222 | ||
1223 | for (it = cgv1_hierarchies; it && *it; it++) { | |
1224 | if ((*it)->controllers) { | |
1225 | char *init_cgroup, *user_slice; | |
9159b38c | 1226 | |
7ac43d3d CB |
1227 | /* We've already stored the controller and received its |
1228 | * current cgroup. If we now fail to retrieve its init | |
1229 | * cgroup, we should probably fail. | |
1230 | */ | |
1231 | init_cgroup = cgv1_get_current_cgroup(basecginfo, (*it)->controllers[0]); | |
1232 | if (!init_cgroup) { | |
1233 | free(basecginfo); | |
1234 | return false; | |
1235 | } | |
9159b38c | 1236 | |
7ac43d3d CB |
1237 | cg_systemd_prune_init_scope(init_cgroup); |
1238 | (*it)->init_cgroup = init_cgroup; | |
1239 | pam_cgfs_debug("cgroupfs v1 controller \"%s\" has init " | |
2082ac62 | 1240 | "cgroup \"%s\"\n", |
1241 | (*(*it)->controllers), init_cgroup); | |
1242 | ||
7ac43d3d CB |
1243 | /* Check whether systemd has already created a cgroup |
1244 | * for us. | |
1245 | */ | |
1246 | user_slice = must_make_path((*it)->mountpoint, (*it)->base_cgroup, NULL); | |
1247 | if (cg_systemd_created_user_slice((*it)->base_cgroup, (*it)->init_cgroup, user_slice, uid)) | |
1248 | (*it)->systemd_user_slice = true; | |
9159b38c | 1249 | |
1250 | free(user_slice); | |
7ac43d3d CB |
1251 | } |
1252 | } | |
1253 | free(basecginfo); | |
1254 | ||
1255 | return true; | |
1256 | } | |
1257 | ||
1258 | /* Check whether @path is a cgroupfs v1 or cgroupfs v2 mount. Returns -1 if | |
1259 | * statfs fails. If @path is null /sys/fs/cgroup is checked. | |
1260 | */ | |
1261 | static inline int cg_get_version_of_mntpt(const char *path) | |
1262 | { | |
1263 | if (has_fs_type(path, CGROUP_SUPER_MAGIC)) | |
1264 | return 1; | |
1265 | ||
1266 | if (has_fs_type(path, CGROUP2_SUPER_MAGIC)) | |
1267 | return 2; | |
1268 | ||
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | /* Detect and store information about the cgroupfs v2 hierarchy. Currently only | |
d0d68468 | 1273 | * deals with the empty v2 hierarchy as we do not retrieve enabled controllers. |
7ac43d3d CB |
1274 | */ |
1275 | static bool cgv2_init(uid_t uid, gid_t gid) | |
1276 | { | |
1277 | char *mountpoint; | |
1278 | FILE *f = NULL; | |
1279 | char *current_cgroup = NULL, *init_cgroup = NULL; | |
1280 | char * line = NULL; | |
1281 | size_t len = 0; | |
1282 | int ret = false; | |
1283 | ||
1284 | current_cgroup = cgv2_get_current_cgroup(getpid()); | |
1285 | if (!current_cgroup) { | |
1286 | /* No v2 hierarchy present. We're done. */ | |
1287 | ret = true; | |
1288 | goto cleanup; | |
1289 | } | |
1290 | ||
1291 | init_cgroup = cgv2_get_current_cgroup(1); | |
1292 | if (!init_cgroup) { | |
1293 | /* If we're here and didn't fail already above, then something's | |
1294 | * certainly wrong, so error this time. | |
1295 | */ | |
1296 | goto cleanup; | |
1297 | } | |
90a170d8 | 1298 | |
7ac43d3d CB |
1299 | cg_systemd_prune_init_scope(init_cgroup); |
1300 | ||
1301 | /* Check if the v2 hierarchy is mounted at its standard location. | |
1302 | * If so we can skip the rest of the work here. Although the unified | |
1303 | * hierarchy can be mounted multiple times, each of those mountpoints | |
1304 | * will expose identical information. | |
1305 | */ | |
1306 | if (cg_get_version_of_mntpt("/sys/fs/cgroup") == 2) { | |
1307 | char *user_slice; | |
1308 | bool has_user_slice = false; | |
1309 | ||
1310 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
1311 | if (!mountpoint) | |
1312 | goto cleanup; | |
1313 | ||
1314 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); | |
1315 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1316 | has_user_slice = true; | |
1317 | free(user_slice); | |
1318 | ||
1319 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); | |
1320 | ||
1321 | ret = true; | |
1322 | goto cleanup; | |
1323 | } | |
1324 | ||
1325 | f = fopen("/proc/self/mountinfo", "r"); | |
1326 | if (!f) | |
1327 | goto cleanup; | |
1328 | ||
1329 | /* we support simple cgroup mounts and lxcfs mounts */ | |
1330 | while (getline(&line, &len, f) != -1) { | |
1331 | char *user_slice; | |
1332 | bool has_user_slice = false; | |
90a170d8 | 1333 | |
7ac43d3d CB |
1334 | if (!is_cgv2(line)) |
1335 | continue; | |
1336 | ||
1337 | mountpoint = get_mountpoint(line); | |
1338 | if (!mountpoint) | |
1339 | continue; | |
1340 | ||
1341 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); | |
1342 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1343 | has_user_slice = true; | |
1344 | free(user_slice); | |
1345 | ||
1346 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); | |
90a170d8 | 1347 | |
7ac43d3d CB |
1348 | /* Although the unified hierarchy can be mounted multiple times, |
1349 | * each of those mountpoints will expose identical information. | |
1350 | * So let the first mountpoint we find, win. | |
1351 | */ | |
1352 | ret = true; | |
1353 | break; | |
1354 | } | |
1355 | ||
1356 | pam_cgfs_debug("Detected cgroupfs v2 hierarchy at mountpoint \"%s\" with " | |
d97c3a34 | 1357 | "current cgroup \"%s\" and init cgroup \"%s\"\n", |
1358 | mountpoint, current_cgroup, init_cgroup); | |
7ac43d3d CB |
1359 | |
1360 | cleanup: | |
1361 | if (f) | |
1362 | fclose(f); | |
1363 | free(line); | |
1364 | ||
90a170d8 | 1365 | if (!ret) { |
1366 | free(init_cgroup); | |
d97c3a34 | 1367 | free(current_cgroup); |
90a170d8 | 1368 | } |
d97c3a34 | 1369 | |
7ac43d3d CB |
1370 | return ret; |
1371 | } | |
1372 | ||
1373 | /* Detect and store information about mounted cgroupfs v1 hierarchies and the | |
1374 | * cgroupfs v2 hierarchy. | |
1375 | * Detect whether we are on a pure cgroupfs v1, cgroupfs v2, or mixed system, | |
1376 | * where some controllers are mounted into their standard cgroupfs v1 locations | |
1377 | * (/sys/fs/cgroup/<controller>) and others are mounted into the cgroupfs v2 | |
1378 | * hierarchy (/sys/fs/cgroup). | |
1379 | */ | |
1380 | static bool cg_init(uid_t uid, gid_t gid) | |
1381 | { | |
1382 | if (!cgv1_init(uid, gid)) | |
1383 | return false; | |
1384 | ||
1385 | if (!cgv2_init(uid, gid)) | |
1386 | return false; | |
1387 | ||
1388 | if (cgv1_hierarchies && cgv2_hierarchies) { | |
1389 | cg_mount_mode = CGROUP_MIXED; | |
2082ac62 | 1390 | pam_cgfs_debug("%s\n", "Detected cgroupfs v1 and v2 hierarchies"); |
7ac43d3d CB |
1391 | } else if (cgv1_hierarchies && !cgv2_hierarchies) { |
1392 | cg_mount_mode = CGROUP_PURE_V1; | |
2082ac62 | 1393 | pam_cgfs_debug("%s\n", "Detected cgroupfs v1 hierarchies"); |
7ac43d3d CB |
1394 | } else if (cgv2_hierarchies && !cgv1_hierarchies) { |
1395 | cg_mount_mode = CGROUP_PURE_V2; | |
2082ac62 | 1396 | pam_cgfs_debug("%s\n", "Detected cgroupfs v2 hierarchies"); |
7ac43d3d CB |
1397 | } else { |
1398 | cg_mount_mode = CGROUP_UNKNOWN; | |
2082ac62 | 1399 | mysyslog(LOG_ERR, "Could not detect cgroupfs hierarchy\n", NULL); |
7ac43d3d CB |
1400 | } |
1401 | ||
1402 | if (cg_mount_mode == CGROUP_UNKNOWN) | |
1403 | return false; | |
1404 | ||
1405 | return true; | |
1406 | } | |
1407 | ||
1408 | /* Try to move/migrate us into @cgroup in a cgroupfs v1 hierarchy. */ | |
1409 | static bool cgv1_enter(const char *cgroup) | |
1410 | { | |
1411 | struct cgv1_hierarchy **it; | |
1412 | ||
1413 | for (it = cgv1_hierarchies; it && *it; it++) { | |
1414 | char **controller; | |
1415 | bool entered = false; | |
1416 | ||
1417 | if (!(*it)->controllers || !(*it)->mountpoint || | |
1418 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
1419 | continue; | |
1420 | ||
1421 | for (controller = (*it)->controllers; controller && *controller; | |
1422 | controller++) { | |
1423 | char *path; | |
1424 | ||
1425 | /* We've already been placed in a user slice, so we | |
1426 | * don't need to enter the cgroup again. | |
1427 | */ | |
1428 | if ((*it)->systemd_user_slice) { | |
1429 | entered = true; | |
1430 | break; | |
1431 | } | |
1432 | ||
1433 | path = must_make_path((*it)->mountpoint, | |
1434 | (*it)->init_cgroup, | |
1435 | cgroup, | |
1436 | "/cgroup.procs", | |
1437 | NULL); | |
1438 | if (!file_exists(path)) { | |
1439 | free(path); | |
1440 | path = must_make_path((*it)->mountpoint, | |
1441 | (*it)->init_cgroup, | |
1442 | cgroup, | |
1443 | "/tasks", | |
1444 | NULL); | |
1445 | } | |
2082ac62 | 1446 | |
1447 | pam_cgfs_debug("Attempting to enter cgroupfs v1 hierarchy in \"%s\" cgroup\n", path); | |
7ac43d3d CB |
1448 | entered = write_int(path, (int)getpid()); |
1449 | if (entered) { | |
1450 | free(path); | |
1451 | break; | |
1452 | } | |
2082ac62 | 1453 | |
1454 | pam_cgfs_debug("Failed to enter cgroupfs v1 hierarchy in \"%s\" cgroup\n", path); | |
7ac43d3d CB |
1455 | free(path); |
1456 | } | |
2082ac62 | 1457 | |
7ac43d3d CB |
1458 | if (!entered) |
1459 | return false; | |
1460 | } | |
1461 | ||
1462 | return true; | |
1463 | } | |
1464 | ||
1465 | /* Try to move/migrate us into @cgroup in the cgroupfs v2 hierarchy. */ | |
1466 | static bool cgv2_enter(const char *cgroup) | |
1467 | { | |
1468 | struct cgv2_hierarchy *v2; | |
1469 | char *path; | |
1470 | bool entered = false; | |
1471 | ||
1472 | if (!cgv2_hierarchies) | |
1473 | return true; | |
1474 | ||
1475 | v2 = *cgv2_hierarchies; | |
1476 | ||
1477 | if (!v2->mountpoint || !v2->base_cgroup) | |
1478 | return false; | |
1479 | ||
1480 | if (!v2->create_rw_cgroup || v2->systemd_user_slice) | |
1481 | return true; | |
1482 | ||
1483 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, "/cgroup.procs", NULL); | |
2082ac62 | 1484 | pam_cgfs_debug("Attempting to enter cgroupfs v2 hierarchy in cgroup \"%s\"\n", path); |
1485 | ||
7ac43d3d CB |
1486 | entered = write_int(path, (int)getpid()); |
1487 | if (!entered) { | |
2082ac62 | 1488 | pam_cgfs_debug("Failed to enter cgroupfs v2 hierarchy in cgroup \"%s\"\n", path); |
7ac43d3d CB |
1489 | free(path); |
1490 | return false; | |
1491 | } | |
1492 | ||
1493 | free(path); | |
1494 | ||
1495 | return true; | |
1496 | } | |
1497 | ||
1498 | /* Wrapper around cgv{1,2}_enter(). */ | |
1499 | static bool cg_enter(const char *cgroup) | |
1500 | { | |
1501 | if (!cgv1_enter(cgroup)) { | |
2082ac62 | 1502 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to enter cgroups\n", NULL); |
7ac43d3d CB |
1503 | return false; |
1504 | } | |
1505 | ||
1506 | if (!cgv2_enter(cgroup)) { | |
2082ac62 | 1507 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to enter cgroups\n", NULL); |
7ac43d3d CB |
1508 | return false; |
1509 | } | |
1510 | ||
1511 | return true; | |
1512 | } | |
1513 | ||
1514 | /* Escape to root cgroup in all detected cgroupfs v1 hierarchies. */ | |
1515 | static void cgv1_escape(void) | |
1516 | { | |
1517 | struct cgv1_hierarchy **it; | |
1518 | ||
1519 | /* In case systemd hasn't already placed us in a user slice for the | |
1520 | * cpuset v1 controller we will reside in the root cgroup. This means | |
1521 | * that cgroup.clone_children will not have been initialized for us so | |
1522 | * we need to do it. | |
1523 | */ | |
1524 | for (it = cgv1_hierarchies; it && *it; it++) | |
1525 | if (!cgv1_handle_root_cpuset_hierarchy(*it)) | |
2082ac62 | 1526 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to initialize cpuset\n", NULL); |
7ac43d3d CB |
1527 | |
1528 | if (!cgv1_enter("/")) | |
2082ac62 | 1529 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to escape to init's cgroup\n", NULL); |
7ac43d3d CB |
1530 | } |
1531 | ||
1532 | /* Escape to root cgroup in the cgroupfs v2 hierarchy. */ | |
1533 | static void cgv2_escape(void) | |
1534 | { | |
1535 | if (!cgv2_enter("/")) | |
2082ac62 | 1536 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to escape to init's cgroup\n", NULL); |
7ac43d3d CB |
1537 | } |
1538 | ||
1539 | /* Wrapper around cgv{1,2}_escape(). */ | |
1540 | static void cg_escape(void) | |
1541 | { | |
1542 | cgv1_escape(); | |
1543 | cgv2_escape(); | |
1544 | } | |
1545 | ||
1546 | /* Get uid and gid for @user. */ | |
1547 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid) | |
1548 | { | |
2b65c74c | 1549 | struct passwd pwent; |
2dce415b DJ |
1550 | struct passwd *pwentp = NULL; |
1551 | char *buf; | |
1552 | size_t bufsize; | |
1553 | int ret; | |
1554 | ||
1555 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1556 | if (bufsize == -1) | |
1557 | bufsize = 1024; | |
1558 | ||
1559 | buf = malloc(bufsize); | |
1560 | if (!buf) | |
1561 | return false; | |
1562 | ||
1563 | ret = getpwnam_r(user, &pwent, buf, bufsize, &pwentp); | |
1564 | if (!pwentp) { | |
1565 | if (ret == 0) | |
2b65c74c | 1566 | mysyslog(LOG_ERR, |
03e7d72a | 1567 | "Could not find matched password record\n", NULL); |
7ac43d3d | 1568 | |
2dce415b | 1569 | free(buf); |
7ac43d3d | 1570 | return false; |
2dce415b | 1571 | } |
7ac43d3d | 1572 | |
2dce415b DJ |
1573 | *uid = pwent.pw_uid; |
1574 | *gid = pwent.pw_gid; | |
1575 | free(buf); | |
7ac43d3d CB |
1576 | |
1577 | return true; | |
1578 | } | |
1579 | ||
1580 | /* Check if cgroup belongs to our uid and gid. If so, reuse it. */ | |
1581 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid) | |
1582 | { | |
1583 | struct stat statbuf; | |
1584 | ||
1585 | if (stat(path, &statbuf) < 0) | |
1586 | return false; | |
1587 | ||
1588 | if (!(statbuf.st_uid == uid) || !(statbuf.st_gid == gid)) | |
1589 | return false; | |
1590 | ||
1591 | return true; | |
1592 | } | |
1593 | ||
1594 | /* Create cpumask from cpulist aka turn: | |
1595 | * | |
1596 | * 0,2-3 | |
1597 | * | |
1598 | * into bit array | |
1599 | * | |
1600 | * 1 0 1 1 | |
1601 | */ | |
1602 | static uint32_t *cg_cpumask(char *buf, size_t nbits) | |
1603 | { | |
1604 | char *token; | |
1605 | char *saveptr = NULL; | |
1606 | size_t arrlen = BITS_TO_LONGS(nbits); | |
1607 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
1608 | if (!bitarr) | |
1609 | return NULL; | |
1610 | ||
1611 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
1612 | errno = 0; | |
1613 | unsigned start = strtoul(token, NULL, 0); | |
1614 | unsigned end = start; | |
1615 | ||
1616 | char *range = strchr(token, '-'); | |
1617 | if (range) | |
1618 | end = strtoul(range + 1, NULL, 0); | |
03e7d72a | 1619 | |
7ac43d3d CB |
1620 | if (!(start <= end)) { |
1621 | free(bitarr); | |
1622 | return NULL; | |
1623 | } | |
1624 | ||
1625 | if (end >= nbits) { | |
1626 | free(bitarr); | |
1627 | return NULL; | |
1628 | } | |
1629 | ||
1630 | while (start <= end) | |
1631 | set_bit(start++, bitarr); | |
1632 | } | |
1633 | ||
1634 | return bitarr; | |
1635 | } | |
1636 | ||
1637 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix) | |
1638 | { | |
1639 | char *result; | |
1640 | char **p; | |
1641 | size_t sep_len = strlen(sep); | |
1642 | size_t result_len = use_as_prefix * sep_len; | |
9c8b2b13 | 1643 | size_t buf_len; |
7ac43d3d CB |
1644 | |
1645 | if (!parts) | |
1646 | return NULL; | |
1647 | ||
1648 | /* calculate new string length */ | |
1649 | for (p = (char **)parts; *p; p++) | |
1650 | result_len += (p > (char **)parts) * sep_len + strlen(*p); | |
1651 | ||
9c8b2b13 DJ |
1652 | buf_len = result_len + 1; |
1653 | result = calloc(buf_len, sizeof(char)); | |
7ac43d3d CB |
1654 | if (!result) |
1655 | return NULL; | |
1656 | ||
1657 | if (use_as_prefix) | |
9c8b2b13 | 1658 | (void)strlcpy(result, sep, buf_len * sizeof(char)); |
43f984ea | 1659 | |
7ac43d3d CB |
1660 | for (p = (char **)parts; *p; p++) { |
1661 | if (p > (char **)parts) | |
9c8b2b13 | 1662 | (void)strlcat(result, sep, buf_len * sizeof(char)); |
2082ac62 | 1663 | |
9c8b2b13 | 1664 | (void)strlcat(result, *p, buf_len * sizeof(char)); |
7ac43d3d CB |
1665 | } |
1666 | ||
1667 | return result; | |
1668 | } | |
1669 | ||
1670 | /* The largest integer that can fit into long int is 2^64. This is a | |
1671 | * 20-digit number. | |
1672 | */ | |
1673 | #define __IN_TO_STR_LEN 21 | |
1674 | /* Turn cpumask into simple, comma-separated cpulist. */ | |
1675 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
1676 | { | |
1677 | size_t i; | |
1678 | int ret; | |
1679 | char numstr[__IN_TO_STR_LEN] = {0}; | |
1680 | char **cpulist = NULL; | |
1681 | ||
1682 | for (i = 0; i <= nbits; i++) { | |
1683 | if (is_set(i, bitarr)) { | |
1684 | ret = snprintf(numstr, __IN_TO_STR_LEN, "%zu", i); | |
1685 | if (ret < 0 || (size_t)ret >= __IN_TO_STR_LEN) { | |
1686 | free_string_list(cpulist); | |
1687 | return NULL; | |
1688 | } | |
03e7d72a | 1689 | |
7ac43d3d CB |
1690 | must_append_string(&cpulist, numstr); |
1691 | } | |
1692 | } | |
03e7d72a | 1693 | |
7ac43d3d CB |
1694 | return string_join(",", (const char **)cpulist, false); |
1695 | } | |
1696 | ||
1697 | static ssize_t cg_get_max_cpus(char *cpulist) | |
1698 | { | |
1699 | char *c1, *c2; | |
1700 | char *maxcpus = cpulist; | |
1701 | size_t cpus = 0; | |
1702 | ||
1703 | c1 = strrchr(maxcpus, ','); | |
1704 | if (c1) | |
1705 | c1++; | |
1706 | ||
1707 | c2 = strrchr(maxcpus, '-'); | |
1708 | if (c2) | |
1709 | c2++; | |
1710 | ||
1711 | if (!c1 && !c2) | |
1712 | c1 = maxcpus; | |
1713 | else if (c1 < c2) | |
1714 | c1 = c2; | |
1715 | ||
03e7d72a | 1716 | if (!c1) |
1717 | return -1; | |
1718 | ||
7ac43d3d CB |
1719 | /* If the above logic is correct, c1 should always hold a valid string |
1720 | * here. | |
1721 | */ | |
7ac43d3d CB |
1722 | errno = 0; |
1723 | cpus = strtoul(c1, NULL, 0); | |
1724 | if (errno != 0) | |
1725 | return -1; | |
1726 | ||
1727 | return cpus; | |
1728 | } | |
1729 | ||
7ac43d3d CB |
1730 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" |
1731 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized) | |
1732 | { | |
1733 | char *lastslash, *fpath, oldv; | |
1734 | int ret; | |
1735 | ssize_t i; | |
1736 | ||
1737 | ssize_t maxposs = 0, maxisol = 0; | |
1738 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
1739 | uint32_t *possmask = NULL, *isolmask = NULL; | |
1740 | bool bret = false, flipped_bit = false; | |
1741 | ||
1742 | lastslash = strrchr(path, '/'); | |
1743 | if (!lastslash) { // bug... this shouldn't be possible | |
2082ac62 | 1744 | pam_cgfs_debug("Invalid path: %s\n", path); |
7ac43d3d CB |
1745 | return bret; |
1746 | } | |
8ae3983e | 1747 | |
7ac43d3d CB |
1748 | oldv = *lastslash; |
1749 | *lastslash = '\0'; | |
8ae3983e | 1750 | |
7ac43d3d CB |
1751 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
1752 | posscpus = read_file(fpath); | |
1753 | if (!posscpus) { | |
2082ac62 | 1754 | pam_cgfs_debug("Could not read file: %s\n", fpath); |
7ac43d3d CB |
1755 | goto on_error; |
1756 | } | |
1757 | ||
1758 | /* Get maximum number of cpus found in possible cpuset. */ | |
1759 | maxposs = cg_get_max_cpus(posscpus); | |
92d5ea57 | 1760 | if (maxposs < 0 || maxposs >= INT_MAX - 1) |
7ac43d3d CB |
1761 | goto on_error; |
1762 | ||
1763 | if (!file_exists(__ISOL_CPUS)) { | |
1764 | /* This system doesn't expose isolated cpus. */ | |
2082ac62 | 1765 | pam_cgfs_debug("%s", "Path: "__ISOL_CPUS" to read isolated cpus from does not exist\n"); |
7ac43d3d | 1766 | cpulist = posscpus; |
8ae3983e | 1767 | |
7ac43d3d CB |
1768 | /* No isolated cpus but we weren't already initialized by |
1769 | * someone. We should simply copy the parents cpuset.cpus | |
1770 | * values. | |
1771 | */ | |
1772 | if (!am_initialized) { | |
2082ac62 | 1773 | pam_cgfs_debug("%s", "Copying cpuset of parent cgroup\n"); |
7ac43d3d CB |
1774 | goto copy_parent; |
1775 | } | |
8ae3983e | 1776 | |
7ac43d3d CB |
1777 | /* No isolated cpus but we were already initialized by someone. |
1778 | * Nothing more to do for us. | |
1779 | */ | |
1780 | goto on_success; | |
1781 | } | |
1782 | ||
1783 | isolcpus = read_file(__ISOL_CPUS); | |
1784 | if (!isolcpus) { | |
1785 | pam_cgfs_debug("%s", "Could not read file "__ISOL_CPUS"\n"); | |
1786 | goto on_error; | |
1787 | } | |
8ae3983e | 1788 | |
7ac43d3d | 1789 | if (!isdigit(isolcpus[0])) { |
2082ac62 | 1790 | pam_cgfs_debug("%s", "No isolated cpus detected\n"); |
7ac43d3d | 1791 | cpulist = posscpus; |
8ae3983e | 1792 | |
7ac43d3d CB |
1793 | /* No isolated cpus but we weren't already initialized by |
1794 | * someone. We should simply copy the parents cpuset.cpus | |
1795 | * values. | |
1796 | */ | |
1797 | if (!am_initialized) { | |
2082ac62 | 1798 | pam_cgfs_debug("%s", "Copying cpuset of parent cgroup\n"); |
7ac43d3d CB |
1799 | goto copy_parent; |
1800 | } | |
8ae3983e | 1801 | |
7ac43d3d CB |
1802 | /* No isolated cpus but we were already initialized by someone. |
1803 | * Nothing more to do for us. | |
1804 | */ | |
1805 | goto on_success; | |
1806 | } | |
1807 | ||
1808 | /* Get maximum number of cpus found in isolated cpuset. */ | |
1809 | maxisol = cg_get_max_cpus(isolcpus); | |
92d5ea57 | 1810 | if (maxisol < 0 || maxisol >= INT_MAX - 1) |
7ac43d3d CB |
1811 | goto on_error; |
1812 | ||
1813 | if (maxposs < maxisol) | |
1814 | maxposs = maxisol; | |
1815 | maxposs++; | |
1816 | ||
1817 | possmask = cg_cpumask(posscpus, maxposs); | |
1818 | if (!possmask) { | |
2082ac62 | 1819 | pam_cgfs_debug("%s", "Could not create cpumask for all possible cpus\n"); |
7ac43d3d CB |
1820 | goto on_error; |
1821 | } | |
1822 | ||
1823 | isolmask = cg_cpumask(isolcpus, maxposs); | |
1824 | if (!isolmask) { | |
2082ac62 | 1825 | pam_cgfs_debug("%s", "Could not create cpumask for all isolated cpus\n"); |
7ac43d3d CB |
1826 | goto on_error; |
1827 | } | |
1828 | ||
1829 | for (i = 0; i <= maxposs; i++) { | |
1830 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
1831 | flipped_bit = true; | |
1832 | clear_bit(i, possmask); | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | if (!flipped_bit) { | |
2082ac62 | 1837 | pam_cgfs_debug("%s", "No isolated cpus present in cpuset\n"); |
7ac43d3d CB |
1838 | goto on_success; |
1839 | } | |
2082ac62 | 1840 | pam_cgfs_debug("%s", "Removed isolated cpus from cpuset\n"); |
7ac43d3d CB |
1841 | |
1842 | cpulist = cg_cpumask_to_cpulist(possmask, maxposs); | |
1843 | if (!cpulist) { | |
2082ac62 | 1844 | pam_cgfs_debug("%s", "Could not create cpu list\n"); |
7ac43d3d CB |
1845 | goto on_error; |
1846 | } | |
1847 | ||
1848 | copy_parent: | |
1849 | *lastslash = oldv; | |
8ae3983e | 1850 | |
6926e3b8 | 1851 | free(fpath); |
8ae3983e | 1852 | |
7ac43d3d | 1853 | fpath = must_make_path(path, "cpuset.cpus", NULL); |
a6de11a7 | 1854 | ret = lxc_write_to_file(fpath, cpulist, strlen(cpulist), false, 0660); |
7ac43d3d | 1855 | if (ret < 0) { |
2082ac62 | 1856 | pam_cgfs_debug("Could not write cpu list to: %s\n", fpath); |
7ac43d3d CB |
1857 | goto on_error; |
1858 | } | |
1859 | ||
1860 | on_success: | |
1861 | bret = true; | |
1862 | ||
1863 | on_error: | |
3a88eb8e | 1864 | *lastslash = oldv; |
7ac43d3d | 1865 | |
3a88eb8e | 1866 | free(fpath); |
7ac43d3d CB |
1867 | free(isolcpus); |
1868 | free(isolmask); | |
1869 | ||
1870 | if (posscpus != cpulist) | |
1871 | free(posscpus); | |
1872 | free(possmask); | |
1873 | ||
1874 | free(cpulist); | |
1875 | return bret; | |
1876 | } | |
1877 | ||
7ac43d3d CB |
1878 | /* Copy contents of parent(@path)/@file to @path/@file */ |
1879 | static bool cg_copy_parent_file(char *path, char *file) | |
1880 | { | |
1881 | char *lastslash, *value = NULL, *fpath, oldv; | |
1882 | int len = 0; | |
1883 | int ret; | |
1884 | ||
1885 | lastslash = strrchr(path, '/'); | |
1886 | if (!lastslash) { // bug... this shouldn't be possible | |
1887 | pam_cgfs_debug("cgfsng:copy_parent_file: bad path %s", path); | |
1888 | return false; | |
1889 | } | |
2082ac62 | 1890 | |
7ac43d3d CB |
1891 | oldv = *lastslash; |
1892 | *lastslash = '\0'; | |
2082ac62 | 1893 | |
7ac43d3d | 1894 | fpath = must_make_path(path, file, NULL); |
a6de11a7 | 1895 | len = lxc_read_from_file(fpath, NULL, 0); |
1896 | if (len <= 0) { | |
1897 | pam_cgfs_debug("Failed to read %s: %s", fpath, strerror(errno)); | |
7ac43d3d | 1898 | goto bad; |
a6de11a7 | 1899 | } |
2082ac62 | 1900 | |
f25a2044 | 1901 | value = must_realloc(NULL, len + 1); |
a6de11a7 | 1902 | if (lxc_read_from_file(fpath, value, len) != len) { |
1903 | pam_cgfs_debug("Failed to read %s: %s", fpath, strerror(errno)); | |
7ac43d3d | 1904 | goto bad; |
a6de11a7 | 1905 | } |
7ac43d3d | 1906 | free(fpath); |
2082ac62 | 1907 | |
7ac43d3d | 1908 | *lastslash = oldv; |
2082ac62 | 1909 | |
7ac43d3d | 1910 | fpath = must_make_path(path, file, NULL); |
a6de11a7 | 1911 | ret = lxc_write_to_file(fpath, value, len, false, 0660); |
7ac43d3d CB |
1912 | if (ret < 0) |
1913 | pam_cgfs_debug("Unable to write %s to %s", value, fpath); | |
2082ac62 | 1914 | |
7ac43d3d CB |
1915 | free(fpath); |
1916 | free(value); | |
1917 | return ret >= 0; | |
1918 | ||
1919 | bad: | |
1920 | pam_cgfs_debug("Error reading '%s'", fpath); | |
1921 | free(fpath); | |
1922 | free(value); | |
1923 | return false; | |
1924 | } | |
1925 | ||
1926 | /* In case systemd hasn't already placed us in a user slice for the cpuset v1 | |
1927 | * controller we will reside in the root cgroup. This means that | |
1928 | * cgroup.clone_children will not have been initialized for us so we need to do | |
1929 | * it. | |
1930 | */ | |
1931 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h) | |
1932 | { | |
1933 | char *clonechildrenpath, v; | |
1934 | ||
1935 | if (!string_in_list(h->controllers, "cpuset")) | |
1936 | return true; | |
1937 | ||
1938 | clonechildrenpath = must_make_path(h->mountpoint, "cgroup.clone_children", NULL); | |
1939 | ||
a6de11a7 | 1940 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { |
1941 | pam_cgfs_debug("Failed to read %s: %s", clonechildrenpath, strerror(errno)); | |
7ac43d3d CB |
1942 | free(clonechildrenpath); |
1943 | return false; | |
1944 | } | |
1945 | ||
1946 | if (v == '1') { /* already set for us by someone else */ | |
1947 | free(clonechildrenpath); | |
1948 | return true; | |
1949 | } | |
1950 | ||
a6de11a7 | 1951 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false, 0660) < 0) { |
7ac43d3d CB |
1952 | /* Set clone_children so children inherit our settings */ |
1953 | pam_cgfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
1954 | free(clonechildrenpath); | |
1955 | return false; | |
1956 | } | |
2082ac62 | 1957 | |
7ac43d3d CB |
1958 | free(clonechildrenpath); |
1959 | return true; | |
1960 | } | |
1961 | ||
1962 | /* | |
1963 | * Initialize the cpuset hierarchy in first directory of @gname and | |
1964 | * set cgroup.clone_children so that children inherit settings. | |
1965 | * Since the h->base_path is populated by init or ourselves, we know | |
1966 | * it is already initialized. | |
1967 | */ | |
1968 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, | |
1969 | const char *cgroup) | |
1970 | { | |
1971 | char *cgpath, *clonechildrenpath, v, *slash; | |
1972 | ||
1973 | if (!string_in_list(h->controllers, "cpuset")) | |
1974 | return true; | |
1975 | ||
1976 | if (*cgroup == '/') | |
1977 | cgroup++; | |
1978 | slash = strchr(cgroup, '/'); | |
1979 | if (slash) | |
1980 | *slash = '\0'; | |
1981 | ||
1982 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgroup, NULL); | |
1983 | if (slash) | |
1984 | *slash = '/'; | |
2082ac62 | 1985 | |
c4a4578f | 1986 | if (do_mkdir(cgpath, 0755) < 0 && errno != EEXIST) { |
7ac43d3d CB |
1987 | pam_cgfs_debug("Failed to create '%s'", cgpath); |
1988 | free(cgpath); | |
1989 | return false; | |
1990 | } | |
2082ac62 | 1991 | |
7ac43d3d CB |
1992 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); |
1993 | if (!file_exists(clonechildrenpath)) { /* unified hierarchy doesn't have clone_children */ | |
1994 | free(clonechildrenpath); | |
1995 | free(cgpath); | |
1996 | return true; | |
1997 | } | |
2082ac62 | 1998 | |
a6de11a7 | 1999 | if (lxc_read_from_file(clonechildrenpath, &v, 1) < 0) { |
2000 | pam_cgfs_debug("Failed to read %s: %s", clonechildrenpath, strerror(errno)); | |
7ac43d3d CB |
2001 | free(clonechildrenpath); |
2002 | free(cgpath); | |
2003 | return false; | |
2004 | } | |
2005 | ||
2006 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ | |
2007 | if (!cg_filter_and_set_cpus(cgpath, v == '1')) { | |
2082ac62 | 2008 | pam_cgfs_debug("%s", "Failed to remove isolated cpus\n"); |
7ac43d3d CB |
2009 | free(clonechildrenpath); |
2010 | free(cgpath); | |
2011 | return false; | |
2012 | } | |
2013 | ||
2014 | if (v == '1') { /* already set for us by someone else */ | |
2082ac62 | 2015 | pam_cgfs_debug("%s", "\"cgroup.clone_children\" was already set to \"1\"\n"); |
7ac43d3d CB |
2016 | free(clonechildrenpath); |
2017 | free(cgpath); | |
2018 | return true; | |
2019 | } | |
2020 | ||
2021 | /* copy parent's settings */ | |
2022 | if (!cg_copy_parent_file(cgpath, "cpuset.mems")) { | |
2082ac62 | 2023 | pam_cgfs_debug("%s", "Failed to copy \"cpuset.mems\" settings\n"); |
7ac43d3d CB |
2024 | free(cgpath); |
2025 | free(clonechildrenpath); | |
2026 | return false; | |
2027 | } | |
2028 | free(cgpath); | |
2029 | ||
a6de11a7 | 2030 | if (lxc_write_to_file(clonechildrenpath, "1", 1, false, 0660) < 0) { |
7ac43d3d CB |
2031 | /* Set clone_children so children inherit our settings */ |
2032 | pam_cgfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
2033 | free(clonechildrenpath); | |
2034 | return false; | |
2035 | } | |
2036 | free(clonechildrenpath); | |
2037 | return true; | |
2038 | } | |
2039 | ||
2040 | /* Create and chown @cgroup for all given controllers in a cgroupfs v1 hierarchy | |
2041 | * (For example, create @cgroup for the cpu and cpuacct controller mounted into | |
2042 | * /sys/fs/cgroup/cpu,cpuacct). Check if the path already exists and report back | |
2043 | * to the caller in @existed. | |
2044 | */ | |
2045 | #define __PAM_CGFS_USER "/user/" | |
2046 | #define __PAM_CGFS_USER_LEN 6 | |
2047 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
2048 | { | |
2049 | char *clean_base_cgroup, *path; | |
2050 | char **controller; | |
2051 | struct cgv1_hierarchy *it; | |
2052 | bool created = false; | |
2053 | ||
2054 | *existed = false; | |
2055 | it = h; | |
2082ac62 | 2056 | |
7ac43d3d CB |
2057 | for (controller = it->controllers; controller && *controller; |
2058 | controller++) { | |
2059 | if (!cgv1_handle_cpuset_hierarchy(it, cgroup)) | |
2060 | return false; | |
2061 | ||
2062 | /* If systemd has already created a cgroup for us, keep using | |
2063 | * it. | |
2064 | */ | |
2065 | if (cg_systemd_chown_existing_cgroup(it->mountpoint, | |
2066 | it->base_cgroup, uid, gid, | |
2082ac62 | 2067 | it->systemd_user_slice)) |
7ac43d3d | 2068 | return true; |
7ac43d3d CB |
2069 | |
2070 | /* We need to make sure that we do not create an endless chain | |
2071 | * of sub-cgroups. So we check if we have already logged in | |
2072 | * somehow (sudo -i, su, etc.) and have created a | |
2073 | * /user/PAM_user/idx cgroup. If so, we skip that part. For most | |
2074 | * cgroups this is unnecessary since we use the init_cgroup | |
2075 | * anyway, but for controllers which have an existing systemd | |
2076 | * cgroup that does not match the current uid, this is pretty | |
2077 | * useful. | |
2078 | */ | |
2079 | if (strncmp(it->base_cgroup, __PAM_CGFS_USER, __PAM_CGFS_USER_LEN) == 0) { | |
2080 | free(it->base_cgroup); | |
2081 | it->base_cgroup = must_copy_string("/"); | |
2082 | } else { | |
2083 | clean_base_cgroup = | |
2084 | strstr(it->base_cgroup, __PAM_CGFS_USER); | |
2085 | if (clean_base_cgroup) | |
2086 | *clean_base_cgroup = '\0'; | |
2087 | } | |
2088 | ||
2089 | path = must_make_path(it->mountpoint, it->init_cgroup, cgroup, NULL); | |
2082ac62 | 2090 | pam_cgfs_debug("Constructing path: %s\n", path); |
2091 | ||
7ac43d3d CB |
2092 | if (file_exists(path)) { |
2093 | bool our_cg = cg_belongs_to_uid_gid(path, uid, gid); | |
7ac43d3d CB |
2094 | if (our_cg) |
2095 | *existed = false; | |
2096 | else | |
2097 | *existed = true; | |
2082ac62 | 2098 | |
2099 | pam_cgfs_debug("%s existed and does %shave our uid: %d and gid: %d\n", | |
2100 | path, our_cg ? "" : "not ", uid, gid); | |
2101 | free(path); | |
2102 | ||
7ac43d3d CB |
2103 | return our_cg; |
2104 | } | |
2082ac62 | 2105 | |
7ac43d3d CB |
2106 | created = mkdir_parent(it->mountpoint, path); |
2107 | if (!created) { | |
2108 | free(path); | |
2109 | continue; | |
2110 | } | |
2082ac62 | 2111 | |
7ac43d3d CB |
2112 | if (chown(path, uid, gid) < 0) |
2113 | mysyslog(LOG_WARNING, | |
2082ac62 | 2114 | "Failed to chown %s to %d:%d: %s\n", path, |
2115 | (int)uid, (int)gid, strerror(errno), NULL); | |
2116 | ||
2117 | pam_cgfs_debug("Chowned %s to %d:%d\n", path, (int)uid, (int)gid); | |
7ac43d3d CB |
2118 | free(path); |
2119 | break; | |
2120 | } | |
2121 | ||
2122 | return created; | |
2123 | } | |
2124 | ||
2125 | /* Try to remove @cgroup for all given controllers in a cgroupfs v1 hierarchy | |
2126 | * (For example, try to remove @cgroup for the cpu and cpuacct controller | |
2127 | * mounted into /sys/fs/cgroup/cpu,cpuacct). Ignores failures. | |
2128 | */ | |
2129 | static bool cgv1_remove_one(struct cgv1_hierarchy *h, const char *cgroup) | |
2130 | { | |
2131 | ||
2132 | char *path; | |
2133 | ||
2134 | /* Better safe than sorry. */ | |
2135 | if (!h->controllers) | |
2136 | return true; | |
2137 | ||
2138 | /* Cgroups created by systemd for us which we re-use won't be removed | |
2139 | * here, since we're using init_cgroup + cgroup as path instead of | |
2140 | * base_cgroup + cgroup. | |
2141 | */ | |
2142 | path = must_make_path(h->mountpoint, h->init_cgroup, cgroup, NULL); | |
2143 | (void)recursive_rmdir(path); | |
2144 | free(path); | |
2145 | ||
2146 | return true; | |
2147 | } | |
2148 | ||
2149 | /* Try to remove @cgroup the cgroupfs v2 hierarchy. */ | |
2150 | static bool cgv2_remove(const char *cgroup) | |
2151 | { | |
2152 | struct cgv2_hierarchy *v2; | |
2153 | char *path; | |
2154 | ||
2155 | if (!cgv2_hierarchies) | |
2156 | return true; | |
2157 | ||
2158 | v2 = *cgv2_hierarchies; | |
2159 | ||
2160 | /* If we reused an already existing cgroup, don't bother trying to | |
2161 | * remove (a potentially wrong)/the path. | |
2162 | * Cgroups created by systemd for us which we re-use would be removed | |
2163 | * here, since we're using base_cgroup + cgroup as path. | |
2164 | */ | |
2165 | if (v2->systemd_user_slice) | |
2166 | return true; | |
2167 | ||
2168 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, NULL); | |
2169 | (void)recursive_rmdir(path); | |
2170 | free(path); | |
2171 | ||
2172 | return true; | |
2173 | } | |
2174 | ||
2175 | /* Create @cgroup in all detected cgroupfs v1 hierarchy. If the creation fails | |
2176 | * for any cgroupfs v1 hierarchy, remove all we have created so far. Report | |
2177 | * back, to the caller if the creation failed due to @cgroup already existing | |
2178 | * via @existed. | |
2179 | */ | |
2180 | static bool cgv1_create(const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
2181 | { | |
2182 | struct cgv1_hierarchy **it, **rev_it; | |
2183 | bool all_created = true; | |
2184 | ||
2185 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2186 | if (!(*it)->controllers || !(*it)->mountpoint || | |
2187 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
2188 | continue; | |
2189 | ||
2190 | if (!cgv1_create_one(*it, cgroup, uid, gid, existed)) { | |
2191 | all_created = false; | |
2192 | break; | |
2193 | } | |
2194 | } | |
2195 | ||
2196 | if (all_created) | |
2197 | return true; | |
2198 | ||
2199 | for (rev_it = cgv1_hierarchies; rev_it && *rev_it && (*rev_it != *it); | |
2200 | rev_it++) | |
2201 | cgv1_remove_one(*rev_it, cgroup); | |
2202 | ||
2203 | return false; | |
2204 | } | |
2205 | ||
2206 | /* Create @cgroup in the cgroupfs v2 hierarchy. Report back, to the caller if | |
2207 | * the creation failed due to @cgroup already existing via @existed. | |
2208 | */ | |
2209 | static bool cgv2_create(const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
2210 | { | |
2211 | int ret; | |
2212 | char *clean_base_cgroup; | |
2213 | char *path; | |
2214 | struct cgv2_hierarchy *v2; | |
2215 | bool our_cg = false, created = false; | |
2216 | ||
2217 | *existed = false; | |
2218 | ||
2219 | if (!cgv2_hierarchies || !(*cgv2_hierarchies)->create_rw_cgroup) | |
2220 | return true; | |
2221 | ||
2222 | v2 = *cgv2_hierarchies; | |
2223 | ||
2224 | /* We can't be placed under init's cgroup for the v2 hierarchy. We need | |
2225 | * to be placed under our current cgroup. | |
2226 | */ | |
2227 | if (cg_systemd_chown_existing_cgroup(v2->mountpoint, v2->base_cgroup, | |
2228 | uid, gid, v2->systemd_user_slice)) | |
2229 | goto delegate_files; | |
2230 | ||
2231 | /* We need to make sure that we do not create an endless chain of | |
2232 | * sub-cgroups. So we check if we have already logged in somehow (sudo | |
2233 | * -i, su, etc.) and have created a /user/PAM_user/idx cgroup. If so, we | |
2234 | * skip that part. | |
2235 | */ | |
2236 | if (strncmp(v2->base_cgroup, __PAM_CGFS_USER, __PAM_CGFS_USER_LEN) == 0) { | |
2237 | free(v2->base_cgroup); | |
2238 | v2->base_cgroup = must_copy_string("/"); | |
2239 | } else { | |
2240 | clean_base_cgroup = strstr(v2->base_cgroup, __PAM_CGFS_USER); | |
2241 | if (clean_base_cgroup) | |
2242 | *clean_base_cgroup = '\0'; | |
2243 | } | |
2244 | ||
2245 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, NULL); | |
2082ac62 | 2246 | pam_cgfs_debug("Constructing path \"%s\"\n", path); |
2247 | ||
7ac43d3d CB |
2248 | if (file_exists(path)) { |
2249 | our_cg = cg_belongs_to_uid_gid(path, uid, gid); | |
2082ac62 | 2250 | pam_cgfs_debug("%s existed and does %shave our uid: %d and gid: %d\n", |
2251 | path, our_cg ? "" : "not ", uid, gid); | |
7ac43d3d CB |
2252 | free(path); |
2253 | if (our_cg) { | |
2254 | *existed = false; | |
2255 | goto delegate_files; | |
2256 | } else { | |
2257 | *existed = true; | |
2258 | return false; | |
2259 | } | |
2260 | } | |
2261 | ||
2262 | created = mkdir_parent(v2->mountpoint, path); | |
2263 | if (!created) { | |
2264 | free(path); | |
2265 | return false; | |
2266 | } | |
2267 | ||
2268 | /* chown cgroup to user */ | |
2269 | if (chown(path, uid, gid) < 0) | |
2082ac62 | 2270 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s\n", |
2271 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
7ac43d3d | 2272 | else |
2082ac62 | 2273 | pam_cgfs_debug("Chowned %s to %d:%d\n", path, (int)uid, (int)gid); |
7ac43d3d CB |
2274 | free(path); |
2275 | ||
2276 | delegate_files: | |
2277 | /* chown cgroup.procs to user */ | |
2278 | if (v2->systemd_user_slice) | |
2279 | path = must_make_path(v2->mountpoint, v2->base_cgroup, | |
2280 | "/cgroup.procs", NULL); | |
2281 | else | |
2282 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, | |
2283 | "/cgroup.procs", NULL); | |
2082ac62 | 2284 | |
7ac43d3d CB |
2285 | ret = chown(path, uid, gid); |
2286 | if (ret < 0) | |
2082ac62 | 2287 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s\n", |
2288 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
7ac43d3d | 2289 | else |
2082ac62 | 2290 | pam_cgfs_debug("Chowned %s to %d:%d\n", path, (int)uid, (int)gid); |
7ac43d3d CB |
2291 | free(path); |
2292 | ||
2293 | /* chown cgroup.subtree_control to user */ | |
2294 | if (v2->systemd_user_slice) | |
2295 | path = must_make_path(v2->mountpoint, v2->base_cgroup, | |
2296 | "/cgroup.subtree_control", NULL); | |
2297 | else | |
2298 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, | |
2299 | "/cgroup.subtree_control", NULL); | |
2082ac62 | 2300 | |
7ac43d3d CB |
2301 | ret = chown(path, uid, gid); |
2302 | if (ret < 0) | |
2082ac62 | 2303 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s\n", |
7ac43d3d CB |
2304 | path, (int)uid, (int)gid, strerror(errno), NULL); |
2305 | free(path); | |
2306 | ||
2307 | /* chown cgroup.threads to user */ | |
2308 | if (v2->systemd_user_slice) | |
2309 | path = must_make_path(v2->mountpoint, v2->base_cgroup, | |
2310 | "/cgroup.threads", NULL); | |
2311 | else | |
2312 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, | |
2313 | "/cgroup.threads", NULL); | |
2314 | ret = chown(path, uid, gid); | |
2315 | if (ret < 0 && errno != ENOENT) | |
2082ac62 | 2316 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s\n", |
7ac43d3d CB |
2317 | path, (int)uid, (int)gid, strerror(errno), NULL); |
2318 | free(path); | |
2319 | ||
2320 | return true; | |
2321 | } | |
2322 | ||
2323 | /* Create writeable cgroups for @user at login. Details can be found in the | |
2324 | * preamble/license at the top of this file. | |
2325 | */ | |
2326 | static int handle_login(const char *user, uid_t uid, gid_t gid) | |
2327 | { | |
2328 | int idx = 0, ret; | |
2329 | bool existed; | |
bb3c1e56 | 2330 | char cg[PATH_MAX]; |
7ac43d3d CB |
2331 | |
2332 | cg_escape(); | |
2333 | ||
2334 | while (idx >= 0) { | |
bb3c1e56 CB |
2335 | ret = snprintf(cg, PATH_MAX, "/user/%s/%d", user, idx); |
2336 | if (ret < 0 || ret >= PATH_MAX) { | |
2082ac62 | 2337 | mysyslog(LOG_ERR, "Username too long\n", NULL); |
7ac43d3d CB |
2338 | return PAM_SESSION_ERR; |
2339 | } | |
2340 | ||
2341 | existed = false; | |
2342 | if (!cgv2_create(cg, uid, gid, &existed)) { | |
2343 | if (existed) { | |
2344 | cgv2_remove(cg); | |
2345 | idx++; | |
2346 | continue; | |
2347 | } | |
2082ac62 | 2348 | |
2349 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s\n", user, NULL); | |
7ac43d3d CB |
2350 | return PAM_SESSION_ERR; |
2351 | } | |
2352 | ||
2353 | existed = false; | |
2354 | if (!cgv1_create(cg, uid, gid, &existed)) { | |
2355 | if (existed) { | |
2356 | cgv2_remove(cg); | |
2357 | idx++; | |
2358 | continue; | |
2359 | } | |
2082ac62 | 2360 | |
2361 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s\n", user, NULL); | |
7ac43d3d CB |
2362 | return PAM_SESSION_ERR; |
2363 | } | |
2364 | ||
2365 | if (!cg_enter(cg)) { | |
2082ac62 | 2366 | mysyslog( LOG_ERR, "Failed to enter user cgroup %s for user %s\n", cg, user, NULL); |
7ac43d3d CB |
2367 | return PAM_SESSION_ERR; |
2368 | } | |
2082ac62 | 2369 | |
7ac43d3d CB |
2370 | break; |
2371 | } | |
2372 | ||
2373 | return PAM_SUCCESS; | |
2374 | } | |
2375 | ||
2376 | /* Try to prune cgroups we created and that now are empty from all cgroupfs v1 | |
2377 | * hierarchies. | |
2378 | */ | |
2379 | static bool cgv1_prune_empty_cgroups(const char *user) | |
2380 | { | |
2381 | bool controller_removed = true; | |
2382 | bool all_removed = true; | |
2383 | struct cgv1_hierarchy **it; | |
2384 | ||
2385 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2386 | int ret; | |
2387 | char *path_base, *path_init; | |
2388 | char **controller; | |
2389 | ||
2390 | if (!(*it)->controllers || !(*it)->mountpoint || | |
2391 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
2392 | continue; | |
2393 | ||
2394 | for (controller = (*it)->controllers; controller && *controller; | |
2395 | controller++) { | |
2396 | bool path_base_rm, path_init_rm; | |
2397 | ||
2398 | path_base = must_make_path((*it)->mountpoint, (*it)->base_cgroup, "/user", user, NULL); | |
2082ac62 | 2399 | pam_cgfs_debug("cgroupfs v1: Trying to prune \"%s\"\n", path_base); |
2400 | ||
7ac43d3d CB |
2401 | ret = recursive_rmdir(path_base); |
2402 | if (ret == -ENOENT || ret >= 0) | |
2403 | path_base_rm = true; | |
2404 | else | |
2405 | path_base_rm = false; | |
2406 | free(path_base); | |
2407 | ||
2408 | path_init = must_make_path((*it)->mountpoint, (*it)->init_cgroup, "/user", user, NULL); | |
2082ac62 | 2409 | pam_cgfs_debug("cgroupfs v1: Trying to prune \"%s\"\n", path_init); |
2410 | ||
7ac43d3d CB |
2411 | ret = recursive_rmdir(path_init); |
2412 | if (ret == -ENOENT || ret >= 0) | |
2413 | path_init_rm = true; | |
2414 | else | |
2415 | path_init_rm = false; | |
2416 | free(path_init); | |
2417 | ||
2418 | if (!path_base_rm && !path_init_rm) { | |
2419 | controller_removed = false; | |
2420 | continue; | |
2421 | } | |
2422 | ||
2423 | controller_removed = true; | |
2424 | break; | |
2425 | } | |
2082ac62 | 2426 | |
7ac43d3d CB |
2427 | if (!controller_removed) |
2428 | all_removed = false; | |
2429 | } | |
2430 | ||
2431 | return all_removed; | |
2432 | } | |
2433 | ||
2434 | /* Try to prune cgroup we created and that now is empty from the cgroupfs v2 | |
2435 | * hierarchy. | |
2436 | */ | |
2437 | static bool cgv2_prune_empty_cgroups(const char *user) | |
2438 | { | |
2439 | int ret; | |
2440 | struct cgv2_hierarchy *v2; | |
2441 | char *path_base, *path_init; | |
2442 | bool path_base_rm, path_init_rm; | |
2443 | ||
2444 | if (!cgv2_hierarchies) | |
2445 | return true; | |
2446 | ||
2447 | v2 = *cgv2_hierarchies; | |
2448 | ||
2449 | path_base = must_make_path(v2->mountpoint, v2->base_cgroup, "/user", user, NULL); | |
2082ac62 | 2450 | pam_cgfs_debug("cgroupfs v2: Trying to prune \"%s\"\n", path_base); |
2451 | ||
7ac43d3d CB |
2452 | ret = recursive_rmdir(path_base); |
2453 | if (ret == -ENOENT || ret >= 0) | |
2454 | path_base_rm = true; | |
2455 | else | |
2456 | path_base_rm = false; | |
2457 | free(path_base); | |
2458 | ||
2459 | path_init = must_make_path(v2->mountpoint, v2->init_cgroup, "/user", user, NULL); | |
2082ac62 | 2460 | pam_cgfs_debug("cgroupfs v2: Trying to prune \"%s\"\n", path_init); |
2461 | ||
7ac43d3d CB |
2462 | ret = recursive_rmdir(path_init); |
2463 | if (ret == -ENOENT || ret >= 0) | |
2464 | path_init_rm = true; | |
2465 | else | |
2466 | path_init_rm = false; | |
2467 | free(path_init); | |
2468 | ||
2469 | if (!path_base_rm && !path_init_rm) | |
2470 | return false; | |
2471 | ||
2472 | return true; | |
2473 | } | |
2474 | ||
2475 | /* Wrapper around cgv{1,2}_prune_empty_cgroups(). */ | |
2476 | static void cg_prune_empty_cgroups(const char *user) | |
2477 | { | |
2478 | (void)cgv1_prune_empty_cgroups(user); | |
2479 | (void)cgv2_prune_empty_cgroups(user); | |
2480 | } | |
2481 | ||
2482 | /* Free allocated information for detected cgroupfs v1 hierarchies. */ | |
2483 | static void cgv1_free_hierarchies(void) | |
2484 | { | |
2485 | struct cgv1_hierarchy **it; | |
2486 | ||
2487 | if (!cgv1_hierarchies) | |
2488 | return; | |
2489 | ||
2490 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2491 | if ((*it)->controllers) { | |
2492 | char **tmp; | |
2493 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) | |
2494 | free(*tmp); | |
2495 | ||
2496 | free((*it)->controllers); | |
2497 | } | |
2082ac62 | 2498 | |
7ac43d3d CB |
2499 | free((*it)->mountpoint); |
2500 | free((*it)->base_cgroup); | |
2501 | free((*it)->fullcgpath); | |
2502 | free((*it)->init_cgroup); | |
2503 | } | |
2082ac62 | 2504 | |
7ac43d3d CB |
2505 | free(cgv1_hierarchies); |
2506 | } | |
2507 | ||
2508 | /* Free allocated information for the detected cgroupfs v2 hierarchy. */ | |
2509 | static void cgv2_free_hierarchies(void) | |
2510 | { | |
2511 | struct cgv2_hierarchy **it; | |
2512 | ||
2513 | if (!cgv2_hierarchies) | |
2514 | return; | |
2515 | ||
2516 | for (it = cgv2_hierarchies; it && *it; it++) { | |
2517 | if ((*it)->controllers) { | |
2518 | char **tmp; | |
2082ac62 | 2519 | |
7ac43d3d CB |
2520 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) |
2521 | free(*tmp); | |
2522 | ||
2523 | free((*it)->controllers); | |
2524 | } | |
2082ac62 | 2525 | |
7ac43d3d CB |
2526 | free((*it)->mountpoint); |
2527 | free((*it)->base_cgroup); | |
2528 | free((*it)->fullcgpath); | |
2529 | free((*it)->init_cgroup); | |
2530 | } | |
2082ac62 | 2531 | |
7ac43d3d CB |
2532 | free(cgv2_hierarchies); |
2533 | } | |
2534 | ||
2535 | /* Wrapper around cgv{1,2}_free_hierarchies(). */ | |
2536 | static void cg_exit(void) | |
2537 | { | |
2538 | cgv1_free_hierarchies(); | |
2539 | cgv2_free_hierarchies(); | |
2540 | } | |
2541 | ||
2542 | int pam_sm_open_session(pam_handle_t *pamh, int flags, int argc, | |
2543 | const char **argv) | |
2544 | { | |
2545 | int ret; | |
2546 | uid_t uid = 0; | |
2547 | gid_t gid = 0; | |
2548 | const char *PAM_user = NULL; | |
2549 | ||
2550 | ret = pam_get_user(pamh, &PAM_user, NULL); | |
2551 | if (ret != PAM_SUCCESS) { | |
2552 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); | |
2553 | return PAM_SESSION_ERR; | |
2554 | } | |
2555 | ||
2556 | if (!get_uid_gid(PAM_user, &uid, &gid)) { | |
2082ac62 | 2557 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s\n", PAM_user, NULL); |
7ac43d3d CB |
2558 | return PAM_SESSION_ERR; |
2559 | } | |
2560 | ||
2561 | if (!cg_init(uid, gid)) { | |
2562 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2563 | return PAM_SESSION_ERR; | |
2564 | } | |
2565 | ||
2566 | /* Try to prune cgroups, that are actually empty but were still marked | |
2567 | * as busy by the kernel so we couldn't remove them on session close. | |
2568 | */ | |
2569 | cg_prune_empty_cgroups(PAM_user); | |
2570 | ||
2571 | if (cg_mount_mode == CGROUP_UNKNOWN) | |
2572 | return PAM_SESSION_ERR; | |
2573 | ||
2574 | if (argc > 1 && !strcmp(argv[0], "-c")) { | |
2575 | char **clist = make_string_list(argv[1], ","); | |
2576 | ||
2577 | /* | |
2578 | * We don't allow using "all" and other controllers explicitly because | |
2579 | * that simply doesn't make any sense. | |
2580 | */ | |
2581 | if (string_list_length(clist) > 1 && string_in_list(clist, "all")) { | |
2082ac62 | 2582 | mysyslog(LOG_ERR, "Invalid -c option, cannot specify individual controllers alongside 'all'\n", NULL); |
7ac43d3d CB |
2583 | free_string_list(clist); |
2584 | return PAM_SESSION_ERR; | |
2585 | } | |
2586 | ||
2587 | cg_mark_to_make_rw(clist); | |
2588 | free_string_list(clist); | |
2589 | } | |
2590 | ||
2591 | return handle_login(PAM_user, uid, gid); | |
2592 | } | |
2593 | ||
2594 | int pam_sm_close_session(pam_handle_t *pamh, int flags, int argc, | |
2595 | const char **argv) | |
2596 | { | |
2597 | int ret; | |
2598 | uid_t uid = 0; | |
2599 | gid_t gid = 0; | |
2600 | const char *PAM_user = NULL; | |
2601 | ||
2602 | ret = pam_get_user(pamh, &PAM_user, NULL); | |
2603 | if (ret != PAM_SUCCESS) { | |
2604 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); | |
2605 | return PAM_SESSION_ERR; | |
2606 | } | |
2607 | ||
2608 | if (!get_uid_gid(PAM_user, &uid, &gid)) { | |
2082ac62 | 2609 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s\n", PAM_user, NULL); |
7ac43d3d CB |
2610 | return PAM_SESSION_ERR; |
2611 | } | |
2612 | ||
2613 | if (cg_mount_mode == CGROUP_UNINITIALIZED) { | |
2614 | if (!cg_init(uid, gid)) | |
2615 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2616 | ||
2617 | if (argc > 1 && !strcmp(argv[0], "-c")) { | |
2618 | char **clist = make_string_list(argv[1], ","); | |
2619 | ||
2620 | /* | |
2621 | * We don't allow using "all" and other controllers explicitly because | |
2622 | * that simply doesn't make any sense. | |
2623 | */ | |
2624 | if (string_list_length(clist) > 1 && string_in_list(clist, "all")) { | |
2082ac62 | 2625 | mysyslog(LOG_ERR, "Invalid -c option, cannot specify individual controllers alongside 'all'\n", NULL); |
7ac43d3d CB |
2626 | free_string_list(clist); |
2627 | return PAM_SESSION_ERR; | |
2628 | } | |
2629 | ||
2630 | cg_mark_to_make_rw(clist); | |
2631 | free_string_list(clist); | |
2632 | } | |
2633 | } | |
2634 | ||
2635 | cg_prune_empty_cgroups(PAM_user); | |
2636 | cg_exit(); | |
2637 | ||
2638 | return PAM_SUCCESS; | |
2639 | } |