]>
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 | ||
36 | #include <ctype.h> | |
37 | #include <dirent.h> | |
38 | #include <errno.h> | |
39 | #include <fcntl.h> | |
40 | #include <pwd.h> | |
41 | #include <stdarg.h> | |
42 | #include <stdbool.h> | |
43 | #include <stdint.h> | |
44 | #include <stdio.h> | |
45 | #include <stdlib.h> | |
46 | #include <string.h> | |
47 | #include <syslog.h> | |
48 | #include <unistd.h> | |
49 | #include <linux/unistd.h> | |
50 | #include <sys/mount.h> | |
51 | #include <sys/param.h> | |
52 | #include <sys/stat.h> | |
53 | #include <sys/types.h> | |
54 | #include <sys/vfs.h> | |
55 | ||
56 | #define PAM_SM_SESSION | |
57 | #include <security/_pam_macros.h> | |
58 | #include <security/pam_modules.h> | |
59 | ||
60 | #include "utils.h" | |
61 | ||
62 | #define pam_cgfs_debug_stream(stream, format, ...) \ | |
63 | do { \ | |
64 | fprintf(stream, "%s: %d: %s: " format, __FILE__, __LINE__, \ | |
65 | __func__, __VA_ARGS__); \ | |
66 | } while (false) | |
67 | ||
68 | #define pam_cgfs_error(format, ...) pam_cgfs_debug_stream(stderr, format, __VA_ARGS__) | |
69 | ||
70 | #ifdef DEBUG | |
71 | #define pam_cgfs_debug(format, ...) pam_cgfs_error(format, __VA_ARGS__) | |
72 | #else | |
73 | #define pam_cgfs_debug(format, ...) | |
74 | #endif /* DEBUG */ | |
75 | ||
76 | /* Taken over modified from the kernel sources. */ | |
77 | #define NBITS 32 /* bits in uint32_t */ | |
78 | #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d)) | |
79 | #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, NBITS) | |
80 | ||
81 | static enum cg_mount_mode { | |
82 | CGROUP_UNKNOWN = -1, | |
83 | CGROUP_MIXED = 0, | |
84 | CGROUP_PURE_V1 = 1, | |
85 | CGROUP_PURE_V2 = 2, | |
86 | CGROUP_UNINITIALIZED = 3, | |
87 | } cg_mount_mode = CGROUP_UNINITIALIZED; | |
88 | ||
89 | /* Common helper functions. Most of these have been taken from LXC. */ | |
90 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen); | |
91 | static int append_null_to_list(void ***list); | |
92 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen); | |
93 | static inline void clear_bit(unsigned bit, uint32_t *bitarr) | |
94 | { | |
95 | bitarr[bit / NBITS] &= ~(1 << (bit % NBITS)); | |
96 | } | |
97 | static char *copy_to_eol(char *s); | |
98 | static void free_string_list(char **list); | |
99 | static char *get_mountpoint(char *line); | |
100 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid); | |
101 | static int handle_login(const char *user, uid_t uid, gid_t gid); | |
102 | static inline bool is_set(unsigned bit, uint32_t *bitarr) | |
103 | { | |
104 | return (bitarr[bit / NBITS] & (1 << (bit % NBITS))) != 0; | |
105 | } | |
106 | static bool is_lxcfs(const char *line); | |
107 | static bool is_cgv1(char *line); | |
108 | static bool is_cgv2(char *line); | |
109 | static void *must_alloc(size_t sz); | |
110 | static void must_add_to_list(char ***clist, char *entry); | |
111 | static void must_append_controller(char **klist, char **nlist, char ***clist, | |
112 | char *entry); | |
113 | static void must_append_string(char ***list, char *entry); | |
114 | static void mysyslog(int err, const char *format, ...) __attribute__((sentinel)); | |
115 | static char *read_file(char *fnam); | |
116 | static int read_from_file(const char *filename, void* buf, size_t count); | |
117 | static int recursive_rmdir(char *dirname); | |
118 | static inline void set_bit(unsigned bit, uint32_t *bitarr) | |
119 | { | |
120 | bitarr[bit / NBITS] |= (1 << (bit % NBITS)); | |
121 | } | |
122 | static bool string_in_list(char **list, const char *entry); | |
123 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix); | |
124 | static void trim(char *s); | |
125 | static bool write_int(char *path, int v); | |
126 | static ssize_t write_nointr(int fd, const void* buf, size_t count); | |
127 | static int write_to_file(const char *filename, const void *buf, size_t count, | |
128 | bool add_newline); | |
129 | ||
130 | /* cgroupfs prototypes. */ | |
131 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid); | |
132 | static uint32_t *cg_cpumask(char *buf, size_t nbits); | |
133 | static bool cg_copy_parent_file(char *path, char *file); | |
134 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits); | |
135 | static bool cg_enter(const char *cgroup); | |
136 | static void cg_escape(void); | |
137 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized); | |
138 | static ssize_t cg_get_max_cpus(char *cpulist); | |
139 | static int cg_get_version_of_mntpt(const char *path); | |
140 | static bool cg_init(uid_t uid, gid_t gid); | |
141 | static void cg_mark_to_make_rw(char **list); | |
142 | static void cg_prune_empty_cgroups(const char *user); | |
143 | static bool cg_systemd_created_user_slice(const char *base_cgroup, | |
144 | const char *init_cgroup, | |
145 | const char *in, uid_t uid); | |
146 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
147 | const char *base_cgroup, uid_t uid, | |
148 | gid_t gid, | |
149 | bool systemd_user_slice); | |
150 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid); | |
151 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
152 | const char *init_cgroup, uid_t uid); | |
153 | static void cg_systemd_prune_init_scope(char *cg); | |
154 | static bool is_lxcfs(const char *line); | |
155 | ||
156 | /* cgroupfs v1 prototypes. */ | |
157 | struct cgv1_hierarchy { | |
158 | char **controllers; | |
159 | char *mountpoint; | |
160 | char *base_cgroup; | |
161 | char *fullcgpath; | |
162 | char *init_cgroup; | |
163 | bool create_rw_cgroup; | |
164 | bool systemd_user_slice; | |
165 | }; | |
166 | ||
167 | static struct cgv1_hierarchy **cgv1_hierarchies; | |
168 | ||
169 | static void cgv1_add_controller(char **clist, char *mountpoint, | |
170 | char *base_cgroup, char *init_cgroup); | |
171 | static bool cgv1_controller_in_clist(char *cgline, char *c); | |
172 | static bool cgv1_controller_lists_intersect(char **l1, char **l2); | |
173 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, | |
174 | char **clist); | |
175 | static bool cgv1_create(const char *cgroup, uid_t uid, gid_t gid, | |
176 | bool *existed); | |
177 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, | |
178 | uid_t uid, gid_t gid, bool *existed); | |
179 | static bool cgv1_enter(const char *cgroup); | |
180 | static void cgv1_escape(void); | |
181 | static bool cgv1_get_controllers(char ***klist, char ***nlist); | |
182 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller); | |
183 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, | |
184 | char *line); | |
185 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, | |
186 | const char *cgroup); | |
187 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h); | |
188 | static bool cgv1_init(uid_t uid, gid_t gid); | |
189 | static void cgv1_mark_to_make_rw(char **clist); | |
190 | static char *cgv1_must_prefix_named(char *entry); | |
191 | static bool cgv1_prune_empty_cgroups(const char *user); | |
192 | static bool cgv1_remove_one(struct cgv1_hierarchy *h, const char *cgroup); | |
193 | static bool is_cgv1(char *line); | |
194 | ||
195 | /* cgroupfs v2 prototypes. */ | |
196 | struct cgv2_hierarchy { | |
197 | char **controllers; | |
198 | char *mountpoint; | |
199 | char *base_cgroup; | |
200 | char *fullcgpath; | |
201 | char *init_cgroup; | |
202 | bool create_rw_cgroup; | |
203 | bool systemd_user_slice; | |
204 | }; | |
205 | ||
206 | /* Actually this should only be a single hierarchy. But for the sake of | |
207 | * parallelism and because the layout of the cgroupfs v2 is still somewhat | |
208 | * changing, we'll leave it as an array of structs. | |
209 | */ | |
210 | static struct cgv2_hierarchy **cgv2_hierarchies; | |
211 | ||
212 | static void cgv2_add_controller(char **clist, char *mountpoint, | |
213 | char *base_cgroup, char *init_cgroup, | |
214 | bool systemd_user_slice); | |
215 | static bool cgv2_create(const char *cgroup, uid_t uid, gid_t gid, | |
216 | bool *existed); | |
217 | static bool cgv2_enter(const char *cgroup); | |
218 | static void cgv2_escape(void); | |
219 | static char *cgv2_get_current_cgroup(int pid); | |
220 | static bool cgv2_init(uid_t uid, gid_t gid); | |
221 | static void cgv2_mark_to_make_rw(char **clist); | |
222 | static bool cgv2_prune_empty_cgroups(const char *user); | |
223 | static bool cgv2_remove(const char *cgroup); | |
224 | static bool is_cgv2(char *line); | |
225 | ||
c4a4578f JC |
226 | static int do_mkdir(const char *path, mode_t mode) |
227 | { | |
228 | int saved_errno; | |
229 | mode_t mask; | |
230 | int r; | |
231 | ||
232 | mask = umask(0); | |
233 | r = mkdir(path, mode); | |
234 | saved_errno = errno; | |
235 | umask(mask); | |
236 | errno = saved_errno; | |
237 | return (r); | |
238 | } | |
239 | ||
7ac43d3d CB |
240 | /* Create directory and (if necessary) its parents. */ |
241 | static bool mkdir_parent(const char *root, char *path) | |
242 | { | |
243 | char *b, orig, *e; | |
244 | ||
245 | if (strlen(path) < strlen(root)) | |
246 | return false; | |
247 | ||
248 | if (strlen(path) == strlen(root)) | |
249 | return true; | |
250 | ||
251 | b = path + strlen(root) + 1; | |
252 | while (true) { | |
253 | while (*b && (*b == '/')) | |
254 | b++; | |
255 | if (!*b) | |
256 | return true; | |
257 | ||
258 | e = b + 1; | |
259 | while (*e && *e != '/') | |
260 | e++; | |
261 | ||
262 | orig = *e; | |
263 | if (orig) | |
264 | *e = '\0'; | |
265 | ||
266 | if (file_exists(path)) | |
267 | goto next; | |
268 | ||
c4a4578f | 269 | if (do_mkdir(path, 0755) < 0) { |
7ac43d3d CB |
270 | pam_cgfs_debug("Failed to create %s: %s.\n", path, strerror(errno)); |
271 | return false; | |
272 | } | |
273 | ||
274 | next: | |
275 | if (!orig) | |
276 | return true; | |
277 | ||
278 | *e = orig; | |
279 | b = e + 1; | |
280 | } | |
281 | ||
282 | return false; | |
283 | } | |
284 | ||
285 | /* Common helper functions. Most of these have been taken from LXC. */ | |
286 | static void mysyslog(int err, const char *format, ...) | |
287 | { | |
288 | va_list args; | |
289 | ||
290 | va_start(args, format); | |
291 | openlog("PAM-CGFS", LOG_CONS | LOG_PID, LOG_AUTH); | |
292 | vsyslog(err, format, args); | |
293 | va_end(args); | |
294 | closelog(); | |
295 | } | |
296 | ||
297 | /* realloc() pointer in batch sizes; do not fail. */ | |
298 | #define BATCH_SIZE 50 | |
299 | static void batch_realloc(char **mem, size_t oldlen, size_t newlen) | |
300 | { | |
301 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
302 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
303 | ||
304 | if (!*mem || newbatches > oldbatches) | |
305 | *mem = must_realloc(*mem, newbatches * BATCH_SIZE); | |
306 | } | |
307 | ||
308 | /* Append lines as is to pointer; do not fail. */ | |
309 | static void append_line(char **dest, size_t oldlen, char *new, size_t newlen) | |
310 | { | |
311 | size_t full = oldlen + newlen; | |
312 | ||
313 | batch_realloc(dest, oldlen, full + 1); | |
314 | ||
315 | memcpy(*dest + oldlen, new, newlen + 1); | |
316 | } | |
317 | ||
318 | /* Read in whole file and return allocated pointer. */ | |
319 | static char *read_file(char *fnam) | |
320 | { | |
321 | FILE *f; | |
322 | int linelen; | |
323 | char *line = NULL, *buf = NULL; | |
324 | size_t len = 0, fulllen = 0; | |
325 | ||
326 | f = fopen(fnam, "r"); | |
327 | if (!f) | |
328 | return NULL; | |
329 | ||
330 | while ((linelen = getline(&line, &len, f)) != -1) { | |
331 | append_line(&buf, fulllen, line, linelen); | |
332 | fulllen += linelen; | |
333 | } | |
334 | ||
335 | fclose(f); | |
336 | free(line); | |
337 | ||
338 | return buf; | |
339 | } | |
340 | ||
341 | /* Given a pointer to a null-terminated array of pointers, realloc to add one | |
342 | * entry, and point the new entry to NULL. Do not fail. Return the index to the | |
343 | * second-to-last entry - that is, the one which is now available for use | |
344 | * (keeping the list null-terminated). | |
345 | */ | |
346 | static int append_null_to_list(void ***list) | |
347 | { | |
348 | int newentry = 0; | |
349 | ||
350 | if (*list) | |
351 | for (; (*list)[newentry]; newentry++) { | |
352 | ; | |
353 | } | |
354 | ||
355 | *list = must_realloc(*list, (newentry + 2) * sizeof(void **)); | |
356 | (*list)[newentry + 1] = NULL; | |
357 | ||
358 | return newentry; | |
359 | } | |
360 | ||
361 | /* Append new entry to null-terminated array of pointer; make sure that array of | |
362 | * pointers will still be null-terminated. | |
363 | */ | |
364 | static void must_append_string(char ***list, char *entry) | |
365 | { | |
366 | int newentry; | |
367 | char *copy; | |
368 | ||
369 | newentry = append_null_to_list((void ***)list); | |
370 | copy = must_copy_string(entry); | |
371 | (*list)[newentry] = copy; | |
372 | } | |
373 | ||
374 | /* Remove newlines from string. */ | |
375 | static void trim(char *s) | |
376 | { | |
377 | size_t len = strlen(s); | |
378 | ||
379 | while ((len > 0) && s[len - 1] == '\n') | |
380 | s[--len] = '\0'; | |
381 | } | |
382 | ||
383 | /* Allocate pointer; do not fail. */ | |
384 | static void *must_alloc(size_t sz) | |
385 | { | |
386 | return must_realloc(NULL, sz); | |
387 | } | |
388 | ||
389 | /* Make allocated copy of string. End of string is taken to be '\n'. */ | |
390 | static char *copy_to_eol(char *s) | |
391 | { | |
392 | char *newline, *sret; | |
393 | size_t len; | |
394 | ||
395 | newline = strchr(s, '\n'); | |
396 | if (!newline) | |
397 | return NULL; | |
398 | ||
399 | len = newline - s; | |
400 | sret = must_alloc(len + 1); | |
401 | memcpy(sret, s, len); | |
402 | sret[len] = '\0'; | |
403 | ||
404 | return sret; | |
405 | } | |
406 | ||
407 | /* Check if given entry under /proc/<pid>/mountinfo is a fuse.lxcfs mount. */ | |
408 | static bool is_lxcfs(const char *line) | |
409 | { | |
410 | char *p = strstr(line, " - "); | |
411 | if (!p) | |
412 | return false; | |
413 | ||
414 | return strncmp(p, " - fuse.lxcfs ", 14) == 0; | |
415 | } | |
416 | ||
417 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v1 mount. */ | |
418 | static bool is_cgv1(char *line) | |
419 | { | |
420 | char *p = strstr(line, " - "); | |
421 | if (!p) | |
422 | return false; | |
423 | ||
424 | return strncmp(p, " - cgroup ", 10) == 0; | |
425 | } | |
426 | ||
427 | /* Check if given entry under /proc/<pid>/mountinfo is a cgroupfs v2 mount. */ | |
428 | static bool is_cgv2(char *line) | |
429 | { | |
430 | char *p = strstr(line, " - "); | |
431 | if (!p) | |
432 | return false; | |
433 | ||
434 | return strncmp(p, " - cgroup2 ", 11) == 0; | |
435 | } | |
436 | ||
437 | /* Given a null-terminated array of strings, check whether @entry is one of the | |
438 | * strings | |
439 | */ | |
440 | static bool string_in_list(char **list, const char *entry) | |
441 | { | |
442 | char **it; | |
443 | ||
444 | for (it = list; it && *it; it++) | |
445 | if (strcmp(*it, entry) == 0) | |
446 | return true; | |
447 | ||
448 | return false; | |
449 | } | |
450 | ||
451 | /* | |
452 | * Creates a null-terminated array of strings, made by splitting the entries in | |
453 | * @str on each @sep. Caller is responsible for calling free_string_list. | |
454 | */ | |
455 | static char **make_string_list(const char *str, const char *sep) | |
456 | { | |
457 | char *copy, *tok; | |
458 | char *saveptr = NULL; | |
459 | char **clist = NULL; | |
460 | ||
461 | copy = must_copy_string(str); | |
462 | ||
463 | for (tok = strtok_r(copy, sep, &saveptr); tok; | |
464 | tok = strtok_r(NULL, sep, &saveptr)) | |
465 | must_add_to_list(&clist, tok); | |
466 | ||
467 | free(copy); | |
468 | ||
469 | return clist; | |
470 | } | |
471 | ||
472 | /* Gets the length of a null-terminated array of strings. */ | |
473 | static size_t string_list_length(char **list) | |
474 | { | |
475 | size_t len = 0; | |
476 | char **it; | |
477 | ||
478 | for (it = list; it && *it; it++) | |
479 | len++; | |
480 | ||
481 | return len; | |
482 | } | |
483 | ||
484 | /* Free null-terminated array of strings. */ | |
485 | static void free_string_list(char **list) | |
486 | { | |
487 | char **it; | |
488 | ||
489 | for (it = list; it && *it; it++) | |
490 | free(*it); | |
491 | free(list); | |
492 | } | |
493 | ||
494 | /* Write single integer to file. */ | |
495 | static bool write_int(char *path, int v) | |
496 | { | |
497 | FILE *f; | |
498 | bool ret = true; | |
499 | ||
500 | f = fopen(path, "w"); | |
501 | if (!f) | |
502 | return false; | |
503 | ||
504 | if (fprintf(f, "%d\n", v) < 0) | |
505 | ret = false; | |
506 | ||
507 | if (fclose(f) != 0) | |
508 | ret = false; | |
509 | ||
510 | return ret; | |
511 | } | |
512 | ||
513 | /* Recursively remove directory and its parents. */ | |
514 | static int recursive_rmdir(char *dirname) | |
515 | { | |
516 | struct dirent *direntp; | |
517 | DIR *dir; | |
518 | int r = 0; | |
519 | ||
520 | dir = opendir(dirname); | |
521 | if (!dir) | |
522 | return -ENOENT; | |
523 | ||
524 | while ((direntp = readdir(dir))) { | |
525 | struct stat st; | |
526 | char *pathname; | |
527 | ||
7ac43d3d CB |
528 | if (!strcmp(direntp->d_name, ".") || |
529 | !strcmp(direntp->d_name, "..")) | |
530 | continue; | |
531 | ||
532 | pathname = must_make_path(dirname, direntp->d_name, NULL); | |
533 | ||
534 | if (lstat(pathname, &st)) { | |
535 | if (!r) | |
536 | pam_cgfs_debug("Failed to stat %s.\n", pathname); | |
537 | r = -1; | |
538 | goto next; | |
539 | } | |
540 | ||
541 | if (!S_ISDIR(st.st_mode)) | |
542 | goto next; | |
543 | ||
544 | if (recursive_rmdir(pathname) < 0) | |
545 | r = -1; | |
546 | next: | |
547 | free(pathname); | |
548 | } | |
549 | ||
550 | if (rmdir(dirname) < 0) { | |
551 | if (!r) | |
552 | pam_cgfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); | |
553 | r = -1; | |
554 | } | |
555 | ||
556 | if (closedir(dir) < 0) { | |
557 | if (!r) | |
558 | pam_cgfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); | |
559 | r = -1; | |
560 | } | |
561 | ||
562 | return r; | |
563 | } | |
564 | ||
565 | /* Add new entry to null-terminated array of pointers. Make sure array is still | |
566 | * null-terminated. | |
567 | */ | |
568 | static void must_add_to_list(char ***clist, char *entry) | |
569 | { | |
570 | int newentry; | |
571 | ||
572 | newentry = append_null_to_list((void ***)clist); | |
573 | (*clist)[newentry] = must_copy_string(entry); | |
574 | } | |
575 | ||
576 | /* Get mountpoint from a /proc/<pid>/mountinfo line. */ | |
577 | static char *get_mountpoint(char *line) | |
578 | { | |
579 | int i; | |
580 | char *p, *sret, *p2; | |
581 | size_t len; | |
582 | ||
583 | p = line; | |
584 | ||
585 | for (i = 0; i < 4; i++) { | |
586 | p = strchr(p, ' '); | |
587 | if (!p) | |
588 | return NULL; | |
589 | p++; | |
590 | } | |
591 | ||
592 | p2 = strchr(p, ' '); | |
593 | if (p2) | |
594 | *p2 = '\0'; | |
595 | ||
596 | len = strlen(p); | |
597 | sret = must_alloc(len + 1); | |
598 | memcpy(sret, p, len); | |
599 | sret[len] = '\0'; | |
600 | ||
601 | return sret; | |
602 | } | |
603 | ||
604 | /* Create list of cgroupfs v1 controller found under /proc/self/cgroup. Skips | |
605 | * the 0::/some/path cgroupfs v2 hierarchy listed. Splits controllers into | |
606 | * kernel controllers (@klist) and named controllers (@nlist). | |
607 | */ | |
608 | static bool cgv1_get_controllers(char ***klist, char ***nlist) | |
609 | { | |
610 | FILE *f; | |
611 | char *line = NULL; | |
612 | size_t len = 0; | |
613 | ||
614 | f = fopen("/proc/self/cgroup", "r"); | |
615 | if (!f) | |
616 | return false; | |
617 | ||
618 | while (getline(&line, &len, f) != -1) { | |
619 | char *p, *p2, *tok; | |
620 | char *saveptr = NULL; | |
621 | ||
622 | p = strchr(line, ':'); | |
623 | if (!p) | |
624 | continue; | |
625 | p++; | |
626 | ||
627 | p2 = strchr(p, ':'); | |
628 | if (!p2) | |
629 | continue; | |
630 | *p2 = '\0'; | |
631 | ||
632 | /* Skip the v2 hierarchy. */ | |
633 | if ((p2 - p) == 0) | |
634 | continue; | |
635 | ||
636 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
637 | tok = strtok_r(NULL, ",", &saveptr)) { | |
638 | if (strncmp(tok, "name=", 5) == 0) | |
639 | must_append_string(nlist, tok); | |
640 | else | |
641 | must_append_string(klist, tok); | |
642 | } | |
643 | } | |
644 | ||
645 | free(line); | |
646 | fclose(f); | |
647 | ||
648 | return true; | |
649 | } | |
650 | ||
651 | /* Get list of controllers for cgroupfs v2 hierarchy by looking at | |
652 | * cgroup.controllers and/or cgroup.subtree_control of a given (parent) cgroup. | |
653 | static bool cgv2_get_controllers(char ***klist) | |
654 | { | |
655 | return -ENOSYS; | |
656 | } | |
657 | */ | |
658 | ||
659 | /* Get current cgroup from /proc/self/cgroup for the cgroupfs v2 hierarchy. */ | |
660 | static char *cgv2_get_current_cgroup(int pid) | |
661 | { | |
662 | int ret; | |
663 | char *cgroups_v2; | |
664 | char *current_cgroup; | |
665 | char *copy = NULL; | |
666 | /* The largest integer that can fit into long int is 2^64. This is a | |
667 | * 20-digit number. */ | |
668 | #define __PIDLEN /* /proc */ 5 + /* /pid-to-str */ 21 + /* /cgroup */ 7 + /* \0 */ 1 | |
669 | char path[__PIDLEN]; | |
670 | ||
671 | ret = snprintf(path, __PIDLEN, "/proc/%d/cgroup", pid); | |
672 | if (ret < 0 || ret >= __PIDLEN) | |
673 | return NULL; | |
674 | ||
675 | cgroups_v2 = read_file(path); | |
676 | if (!cgroups_v2) | |
677 | return NULL; | |
678 | ||
679 | current_cgroup = strstr(cgroups_v2, "0::/"); | |
680 | if (!current_cgroup) | |
681 | goto cleanup_on_err; | |
682 | ||
683 | current_cgroup = current_cgroup + 3; | |
684 | copy = copy_to_eol(current_cgroup); | |
685 | if (!copy) | |
686 | goto cleanup_on_err; | |
687 | ||
688 | cleanup_on_err: | |
689 | free(cgroups_v2); | |
690 | if (copy) | |
691 | trim(copy); | |
692 | ||
693 | return copy; | |
694 | } | |
695 | ||
696 | /* Given two null-terminated lists of strings, return true if any string is in | |
697 | * both. | |
698 | */ | |
699 | static bool cgv1_controller_lists_intersect(char **l1, char **l2) | |
700 | { | |
701 | char **it; | |
702 | ||
703 | if (!l2) | |
704 | return false; | |
705 | ||
706 | for (it = l1; it && *it; it++) | |
707 | if (string_in_list(l2, *it)) | |
708 | return true; | |
709 | ||
710 | return false; | |
711 | } | |
712 | ||
713 | /* For a null-terminated list of controllers @clist, return true if any of those | |
714 | * controllers is already listed the null-terminated list of hierarchies @hlist. | |
715 | * Realistically, if one is present, all must be present. | |
716 | */ | |
717 | static bool cgv1_controller_list_is_dup(struct cgv1_hierarchy **hlist, char **clist) | |
718 | { | |
719 | struct cgv1_hierarchy **it; | |
720 | ||
721 | for (it = hlist; it && *it; it++) | |
722 | if ((*it)->controllers) | |
723 | if (cgv1_controller_lists_intersect((*it)->controllers, clist)) | |
724 | return true; | |
725 | return false; | |
726 | ||
727 | } | |
728 | ||
729 | /* Set boolean to mark controllers under which we are supposed create a | |
730 | * writeable cgroup. | |
731 | */ | |
732 | static void cgv1_mark_to_make_rw(char **clist) | |
733 | { | |
734 | struct cgv1_hierarchy **it; | |
735 | ||
736 | for (it = cgv1_hierarchies; it && *it; it++) | |
737 | if ((*it)->controllers) | |
738 | if (cgv1_controller_lists_intersect((*it)->controllers, clist) || | |
739 | string_in_list(clist, "all")) | |
740 | (*it)->create_rw_cgroup = true; | |
741 | } | |
742 | ||
743 | /* Set boolean to mark whether we are supposed to create a writeable cgroup in | |
744 | * the cgroupfs v2 hierarchy. | |
745 | */ | |
746 | static void cgv2_mark_to_make_rw(char **clist) | |
747 | { | |
748 | if (string_in_list(clist, "unified") || string_in_list(clist, "all")) | |
749 | if (cgv2_hierarchies) | |
750 | (*cgv2_hierarchies)->create_rw_cgroup = true; | |
751 | } | |
752 | ||
753 | /* Wrapper around cgv{1,2}_mark_to_make_rw(). */ | |
754 | static void cg_mark_to_make_rw(char **clist) | |
755 | { | |
756 | cgv1_mark_to_make_rw(clist); | |
757 | cgv2_mark_to_make_rw(clist); | |
758 | } | |
759 | ||
760 | /* Prefix any named controllers with "name=", e.g. "name=systemd". */ | |
761 | static char *cgv1_must_prefix_named(char *entry) | |
762 | { | |
763 | char *s; | |
764 | int ret; | |
765 | size_t len; | |
766 | ||
767 | len = strlen(entry); | |
768 | s = must_alloc(len + 6); | |
769 | ||
770 | ret = snprintf(s, len + 6, "name=%s", entry); | |
771 | if (ret < 0 || (size_t)ret >= (len + 6)) | |
772 | return NULL; | |
773 | ||
774 | return s; | |
775 | } | |
776 | ||
777 | /* Append kernel controller in @klist or named controller in @nlist to @clist */ | |
778 | static void must_append_controller(char **klist, char **nlist, char ***clist, char *entry) | |
779 | { | |
780 | int newentry; | |
781 | char *copy; | |
782 | ||
783 | if (string_in_list(klist, entry) && string_in_list(nlist, entry)) | |
784 | return; | |
785 | ||
786 | newentry = append_null_to_list((void ***)clist); | |
787 | ||
788 | if (strncmp(entry, "name=", 5) == 0) | |
789 | copy = must_copy_string(entry); | |
790 | else if (string_in_list(klist, entry)) | |
791 | copy = must_copy_string(entry); | |
792 | else | |
793 | copy = cgv1_must_prefix_named(entry); | |
794 | ||
795 | (*clist)[newentry] = copy; | |
796 | } | |
797 | ||
798 | /* Get the controllers from a mountinfo line. There are other ways we could get | |
799 | * this info. For lxcfs, field 3 is /cgroup/controller-list. For cgroupfs, we | |
800 | * could parse the mount options. But we simply assume that the mountpoint must | |
801 | * be /sys/fs/cgroup/controller-list | |
802 | */ | |
803 | static char **cgv1_get_proc_mountinfo_controllers(char **klist, char **nlist, char *line) | |
804 | { | |
805 | int i; | |
806 | char *p, *p2, *tok; | |
807 | char *saveptr = NULL; | |
808 | char **aret = NULL; | |
809 | ||
810 | p = line; | |
811 | ||
812 | for (i = 0; i < 4; i++) { | |
813 | p = strchr(p, ' '); | |
814 | if (!p) | |
815 | return NULL; | |
816 | p++; | |
817 | } | |
818 | if (!p) | |
819 | return NULL; | |
820 | ||
821 | if (strncmp(p, "/sys/fs/cgroup/", 15) != 0) | |
822 | return NULL; | |
823 | ||
824 | p += 15; | |
825 | ||
826 | p2 = strchr(p, ' '); | |
827 | if (!p2) | |
828 | return NULL; | |
829 | *p2 = '\0'; | |
830 | ||
831 | for (tok = strtok_r(p, ",", &saveptr); tok; | |
832 | tok = strtok_r(NULL, ",", &saveptr)) | |
833 | must_append_controller(klist, nlist, &aret, tok); | |
834 | ||
835 | return aret; | |
836 | } | |
837 | ||
838 | /* Check if a cgroupfs v2 controller is present in the string @cgline. */ | |
839 | static bool cgv1_controller_in_clist(char *cgline, char *c) | |
840 | { | |
841 | size_t len; | |
842 | char *tok, *eol, *tmp; | |
843 | char *saveptr = NULL; | |
844 | ||
845 | eol = strchr(cgline, ':'); | |
846 | if (!eol) | |
847 | return false; | |
848 | ||
849 | len = eol - cgline; | |
850 | tmp = alloca(len + 1); | |
851 | memcpy(tmp, cgline, len); | |
852 | tmp[len] = '\0'; | |
853 | ||
854 | for (tok = strtok_r(tmp, ",", &saveptr); tok; | |
855 | tok = strtok_r(NULL, ",", &saveptr)) { | |
856 | if (strcmp(tok, c) == 0) | |
857 | return true; | |
858 | } | |
859 | return false; | |
860 | } | |
861 | ||
862 | /* Get current cgroup from the /proc/<pid>/cgroup file passed in via @basecginfo | |
863 | * of a given cgv1 controller passed in via @controller. | |
864 | */ | |
865 | static char *cgv1_get_current_cgroup(char *basecginfo, char *controller) | |
866 | { | |
867 | char *p; | |
868 | ||
869 | p = basecginfo; | |
870 | ||
871 | while (true) { | |
872 | p = strchr(p, ':'); | |
873 | if (!p) | |
874 | return NULL; | |
875 | p++; | |
876 | ||
877 | if (cgv1_controller_in_clist(p, controller)) { | |
878 | p = strchr(p, ':'); | |
879 | if (!p) | |
880 | return NULL; | |
881 | p++; | |
882 | ||
883 | return copy_to_eol(p); | |
884 | } | |
885 | ||
886 | p = strchr(p, '\n'); | |
887 | if (!p) | |
888 | return NULL; | |
889 | p++; | |
890 | } | |
891 | ||
892 | return NULL; | |
893 | } | |
894 | ||
895 | /* Remove /init.scope from string @cg. This will mostly affect systemd-based | |
896 | * systems. | |
897 | */ | |
898 | #define INIT_SCOPE "/init.scope" | |
899 | static void cg_systemd_prune_init_scope(char *cg) | |
900 | { | |
901 | char *point; | |
902 | ||
903 | if (!cg) | |
904 | return; | |
905 | ||
906 | point = cg + strlen(cg) - strlen(INIT_SCOPE); | |
907 | if (point < cg) | |
908 | return; | |
909 | ||
910 | if (strcmp(point, INIT_SCOPE) == 0) { | |
911 | if (point == cg) | |
912 | *(point + 1) = '\0'; | |
913 | else | |
914 | *point = '\0'; | |
915 | } | |
916 | } | |
917 | ||
918 | /* Add new info about a mounted cgroupfs v1 hierarchy. Includes the controllers | |
919 | * mounted into that hierarchy (e.g. cpu,cpuacct), the mountpoint of that | |
920 | * hierarchy (/sys/fs/cgroup/<controller>, the base cgroup of the current | |
921 | * process gathered from /proc/self/cgroup, and the init cgroup of PID1 gathered | |
922 | * from /proc/1/cgroup. | |
923 | */ | |
924 | static void cgv1_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup) | |
925 | { | |
926 | struct cgv1_hierarchy *new; | |
927 | int newentry; | |
928 | ||
929 | new = must_alloc(sizeof(*new)); | |
930 | new->controllers = clist; | |
931 | new->mountpoint = mountpoint; | |
932 | new->base_cgroup = base_cgroup; | |
933 | new->fullcgpath = NULL; | |
934 | new->create_rw_cgroup = false; | |
935 | new->init_cgroup = init_cgroup; | |
936 | new->systemd_user_slice = false; | |
937 | ||
938 | newentry = append_null_to_list((void ***)&cgv1_hierarchies); | |
939 | cgv1_hierarchies[newentry] = new; | |
940 | } | |
941 | ||
942 | /* Add new info about the mounted cgroupfs v2 hierarchy. Can (but doesn't | |
943 | * currently) include the controllers mounted into the hierarchy (e.g. memory, | |
944 | * pids, blkio), the mountpoint of that hierarchy (Should usually be | |
945 | * /sys/fs/cgroup but some init systems seems to think it might be a good idea | |
946 | * to also mount empty cgroupfs v2 hierarchies at /sys/fs/cgroup/systemd.), the | |
947 | * base cgroup of the current process gathered from /proc/self/cgroup, and the | |
948 | * init cgroup of PID1 gathered from /proc/1/cgroup. | |
949 | */ | |
950 | static void cgv2_add_controller(char **clist, char *mountpoint, char *base_cgroup, char *init_cgroup, bool systemd_user_slice) | |
951 | { | |
952 | struct cgv2_hierarchy *new; | |
953 | int newentry; | |
954 | ||
955 | new = must_alloc(sizeof(*new)); | |
956 | new->controllers = clist; | |
957 | new->mountpoint = mountpoint; | |
958 | new->base_cgroup = base_cgroup; | |
959 | new->fullcgpath = NULL; | |
960 | new->create_rw_cgroup = false; | |
961 | new->init_cgroup = init_cgroup; | |
962 | new->systemd_user_slice = systemd_user_slice; | |
963 | ||
964 | newentry = append_null_to_list((void ***)&cgv2_hierarchies); | |
965 | cgv2_hierarchies[newentry] = new; | |
966 | } | |
967 | ||
968 | /* In Ubuntu 14.04, the paths created for us were | |
969 | * '/user/$uid.user/$something.session' This can be merged better with | |
970 | * systemd_created_slice_for_us(), but keeping it separate makes it easier to | |
971 | * reason about the correctness. | |
972 | */ | |
973 | static bool cg_systemd_under_user_slice_1(const char *in, uid_t uid) | |
974 | { | |
975 | char *p; | |
976 | size_t len; | |
977 | int id; | |
978 | char *copy = NULL; | |
979 | bool bret = false; | |
980 | ||
981 | copy = must_copy_string(in); | |
982 | if (strlen(copy) < strlen("/user/1.user/1.session")) | |
983 | goto cleanup; | |
984 | p = copy + strlen(copy) - 1; | |
985 | ||
986 | /* skip any trailing '/' (shouldn't be any, but be sure) */ | |
987 | while (p >= copy && *p == '/') | |
988 | *(p--) = '\0'; | |
989 | if (p < copy) | |
990 | goto cleanup; | |
991 | ||
992 | /* Get last path element */ | |
993 | while (p >= copy && *p != '/') | |
994 | p--; | |
995 | if (p < copy) | |
996 | goto cleanup; | |
997 | /* make sure it is something.session */ | |
998 | len = strlen(p + 1); | |
999 | if (len < strlen("1.session") || | |
1000 | strncmp(p + 1 + len - 8, ".session", 8) != 0) | |
1001 | goto cleanup; | |
1002 | ||
1003 | /* ok last path piece checks out, now check the second to last */ | |
1004 | *(p + 1) = '\0'; | |
1005 | while (p >= copy && *(--p) != '/') | |
1006 | ; | |
1007 | if (sscanf(p + 1, "%d.user/", &id) != 1) | |
1008 | goto cleanup; | |
1009 | ||
1010 | if (id != (int)uid) | |
1011 | goto cleanup; | |
1012 | ||
1013 | bret = true; | |
1014 | ||
1015 | cleanup: | |
1016 | free(copy); | |
1017 | return bret; | |
1018 | } | |
1019 | ||
1020 | /* So long as our path relative to init starts with /user.slice/user-$uid.slice, | |
1021 | * assume it belongs to $uid and chown it | |
1022 | */ | |
1023 | static bool cg_systemd_under_user_slice_2(const char *base_cgroup, | |
1024 | const char *init_cgroup, uid_t uid) | |
1025 | { | |
1026 | int ret; | |
1027 | char buf[100]; | |
1028 | size_t curlen, initlen; | |
1029 | ||
1030 | curlen = strlen(base_cgroup); | |
1031 | initlen = strlen(init_cgroup); | |
1032 | if (curlen <= initlen) | |
1033 | return false; | |
1034 | ||
1035 | if (strncmp(base_cgroup, init_cgroup, initlen) != 0) | |
1036 | return false; | |
1037 | ||
1038 | ret = snprintf(buf, 100, "/user.slice/user-%d.slice/", (int)uid); | |
1039 | if (ret < 0 || ret >= 100) | |
1040 | return false; | |
1041 | ||
1042 | if (initlen == 1) | |
1043 | initlen = 0; // skip the '/' | |
1044 | ||
1045 | return strncmp(base_cgroup + initlen, buf, strlen(buf)) == 0; | |
1046 | } | |
1047 | ||
1048 | /* The systemd-created path is: user-$uid.slice/session-c$session.scope. If that | |
1049 | * is not the end of our systemd path, then we're not part of the PAM call that | |
1050 | * created that path. | |
1051 | * | |
1052 | * The last piece is chowned to $uid, the user- part not. | |
1053 | * Note: If the user creates paths that look like what we're looking for to | |
1054 | * 'fool' us, either | |
1055 | * - they fool us, we create new cgroups, and they get auto-logged-out. | |
1056 | * - they fool a root sudo, systemd cgroup is not changed but chowned, and they | |
1057 | * lose ownership of their cgroups | |
1058 | */ | |
1059 | static bool cg_systemd_created_user_slice(const char *base_cgroup, | |
1060 | const char *init_cgroup, | |
1061 | const char *in, uid_t uid) | |
1062 | { | |
1063 | char *p; | |
1064 | size_t len; | |
1065 | int id; | |
1066 | char *copy = NULL; | |
1067 | bool bret = false; | |
1068 | ||
1069 | copy = must_copy_string(in); | |
1070 | ||
1071 | /* An old version of systemd has already created a cgroup for us. */ | |
1072 | if (cg_systemd_under_user_slice_1(in, uid)) | |
1073 | goto succeed; | |
1074 | ||
1075 | /* A new version of systemd has already created a cgroup for us. */ | |
1076 | if (cg_systemd_under_user_slice_2(base_cgroup, init_cgroup, uid)) | |
1077 | goto succeed; | |
1078 | ||
1079 | if (strlen(copy) < strlen("/user-0.slice/session-0.scope")) | |
1080 | goto cleanup; | |
1081 | ||
1082 | p = copy + strlen(copy) - 1; | |
1083 | /* Skip any trailing '/' (shouldn't be any, but be sure). */ | |
1084 | while (p >= copy && *p == '/') | |
1085 | *(p--) = '\0'; | |
1086 | ||
1087 | if (p < copy) | |
1088 | goto cleanup; | |
1089 | ||
1090 | /* Get last path element */ | |
1091 | while (p >= copy && *p != '/') | |
1092 | p--; | |
1093 | ||
1094 | if (p < copy) | |
1095 | goto cleanup; | |
1096 | ||
1097 | /* Make sure it is session-something.scope. */ | |
1098 | len = strlen(p + 1); | |
1099 | if (strncmp(p + 1, "session-", strlen("session-")) != 0 || | |
1100 | strncmp(p + 1 + len - 6, ".scope", 6) != 0) | |
1101 | goto cleanup; | |
1102 | ||
1103 | /* Ok last path piece checks out, now check the second to last. */ | |
1104 | *(p + 1) = '\0'; | |
1105 | while (p >= copy && *(--p) != '/') | |
1106 | ; | |
1107 | ||
1108 | if (sscanf(p + 1, "user-%d.slice/", &id) != 1) | |
1109 | goto cleanup; | |
1110 | ||
1111 | if (id != (int)uid) | |
1112 | goto cleanup; | |
1113 | ||
1114 | succeed: | |
1115 | bret = true; | |
1116 | cleanup: | |
1117 | free(copy); | |
1118 | return bret; | |
1119 | } | |
1120 | ||
1121 | /* Chown existing cgroup that systemd has already created for us. */ | |
1122 | static bool cg_systemd_chown_existing_cgroup(const char *mountpoint, | |
1123 | const char *base_cgroup, uid_t uid, | |
1124 | gid_t gid, bool systemd_user_slice) | |
1125 | { | |
1126 | char *path; | |
1127 | ||
1128 | if (!systemd_user_slice) | |
1129 | return false; | |
1130 | ||
1131 | path = must_make_path(mountpoint, base_cgroup, NULL); | |
1132 | ||
1133 | /* A cgroup within name=systemd has already been created. So we only | |
1134 | * need to chown it. | |
1135 | */ | |
1136 | if (chown(path, uid, gid) < 0) | |
1137 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", | |
1138 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
1139 | pam_cgfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
1140 | ||
1141 | free(path); | |
1142 | return true; | |
1143 | } | |
1144 | ||
1145 | /* Detect and store information about cgroupfs v1 hierarchies. */ | |
1146 | static bool cgv1_init(uid_t uid, gid_t gid) | |
1147 | { | |
1148 | FILE *f; | |
1149 | struct cgv1_hierarchy **it; | |
1150 | char *basecginfo; | |
1151 | char *line = NULL; | |
1152 | char **klist = NULL, **nlist = NULL; | |
1153 | size_t len = 0; | |
1154 | ||
1155 | basecginfo = read_file("/proc/self/cgroup"); | |
1156 | if (!basecginfo) | |
1157 | return false; | |
1158 | ||
1159 | f = fopen("/proc/self/mountinfo", "r"); | |
1160 | if (!f) { | |
1161 | free(basecginfo); | |
1162 | return false; | |
1163 | } | |
1164 | ||
1165 | cgv1_get_controllers(&klist, &nlist); | |
1166 | ||
1167 | while (getline(&line, &len, f) != -1) { | |
1168 | char **controller_list = NULL; | |
1169 | char *mountpoint, *base_cgroup; | |
1170 | ||
1171 | if (is_lxcfs(line) || !is_cgv1(line)) | |
1172 | continue; | |
1173 | ||
1174 | controller_list = cgv1_get_proc_mountinfo_controllers(klist, nlist, line); | |
1175 | if (!controller_list) | |
1176 | continue; | |
1177 | ||
1178 | if (cgv1_controller_list_is_dup(cgv1_hierarchies, | |
1179 | controller_list)) { | |
1180 | free(controller_list); | |
1181 | continue; | |
1182 | } | |
1183 | ||
1184 | mountpoint = get_mountpoint(line); | |
1185 | if (!mountpoint) { | |
1186 | free_string_list(controller_list); | |
1187 | continue; | |
1188 | } | |
1189 | ||
1190 | base_cgroup = cgv1_get_current_cgroup(basecginfo, controller_list[0]); | |
1191 | if (!base_cgroup) { | |
1192 | free_string_list(controller_list); | |
1193 | free(mountpoint); | |
1194 | continue; | |
1195 | } | |
1196 | trim(base_cgroup); | |
1197 | pam_cgfs_debug("Detected cgroupfs v1 controller \"%s\" with " | |
1198 | "mountpoint \"%s\" and cgroup \"%s\".\n", | |
1199 | controller_list[0], mountpoint, base_cgroup); | |
1200 | cgv1_add_controller(controller_list, mountpoint, base_cgroup, | |
1201 | NULL); | |
1202 | } | |
1203 | free_string_list(klist); | |
1204 | free_string_list(nlist); | |
1205 | free(basecginfo); | |
1206 | fclose(f); | |
1207 | free(line); | |
1208 | ||
1209 | /* Retrieve init cgroup path for all controllers. */ | |
1210 | basecginfo = read_file("/proc/1/cgroup"); | |
1211 | if (!basecginfo) | |
1212 | return false; | |
1213 | ||
1214 | for (it = cgv1_hierarchies; it && *it; it++) { | |
1215 | if ((*it)->controllers) { | |
1216 | char *init_cgroup, *user_slice; | |
1217 | /* We've already stored the controller and received its | |
1218 | * current cgroup. If we now fail to retrieve its init | |
1219 | * cgroup, we should probably fail. | |
1220 | */ | |
1221 | init_cgroup = cgv1_get_current_cgroup(basecginfo, (*it)->controllers[0]); | |
1222 | if (!init_cgroup) { | |
1223 | free(basecginfo); | |
1224 | return false; | |
1225 | } | |
1226 | cg_systemd_prune_init_scope(init_cgroup); | |
1227 | (*it)->init_cgroup = init_cgroup; | |
1228 | pam_cgfs_debug("cgroupfs v1 controller \"%s\" has init " | |
1229 | "cgroup \"%s\".\n", | |
1230 | (*(*it)->controllers), init_cgroup); | |
1231 | /* Check whether systemd has already created a cgroup | |
1232 | * for us. | |
1233 | */ | |
1234 | user_slice = must_make_path((*it)->mountpoint, (*it)->base_cgroup, NULL); | |
1235 | if (cg_systemd_created_user_slice((*it)->base_cgroup, (*it)->init_cgroup, user_slice, uid)) | |
1236 | (*it)->systemd_user_slice = true; | |
1237 | } | |
1238 | } | |
1239 | free(basecginfo); | |
1240 | ||
1241 | return true; | |
1242 | } | |
1243 | ||
1244 | /* Check whether @path is a cgroupfs v1 or cgroupfs v2 mount. Returns -1 if | |
1245 | * statfs fails. If @path is null /sys/fs/cgroup is checked. | |
1246 | */ | |
1247 | static inline int cg_get_version_of_mntpt(const char *path) | |
1248 | { | |
1249 | if (has_fs_type(path, CGROUP_SUPER_MAGIC)) | |
1250 | return 1; | |
1251 | ||
1252 | if (has_fs_type(path, CGROUP2_SUPER_MAGIC)) | |
1253 | return 2; | |
1254 | ||
1255 | return 0; | |
1256 | } | |
1257 | ||
1258 | /* Detect and store information about the cgroupfs v2 hierarchy. Currently only | |
1259 | * deals with the empty v2 hierachy as we do not retrieve enabled controllers. | |
1260 | */ | |
1261 | static bool cgv2_init(uid_t uid, gid_t gid) | |
1262 | { | |
1263 | char *mountpoint; | |
1264 | FILE *f = NULL; | |
1265 | char *current_cgroup = NULL, *init_cgroup = NULL; | |
1266 | char * line = NULL; | |
1267 | size_t len = 0; | |
1268 | int ret = false; | |
1269 | ||
1270 | current_cgroup = cgv2_get_current_cgroup(getpid()); | |
1271 | if (!current_cgroup) { | |
1272 | /* No v2 hierarchy present. We're done. */ | |
1273 | ret = true; | |
1274 | goto cleanup; | |
1275 | } | |
1276 | ||
1277 | init_cgroup = cgv2_get_current_cgroup(1); | |
1278 | if (!init_cgroup) { | |
1279 | /* If we're here and didn't fail already above, then something's | |
1280 | * certainly wrong, so error this time. | |
1281 | */ | |
1282 | goto cleanup; | |
1283 | } | |
1284 | cg_systemd_prune_init_scope(init_cgroup); | |
1285 | ||
1286 | /* Check if the v2 hierarchy is mounted at its standard location. | |
1287 | * If so we can skip the rest of the work here. Although the unified | |
1288 | * hierarchy can be mounted multiple times, each of those mountpoints | |
1289 | * will expose identical information. | |
1290 | */ | |
1291 | if (cg_get_version_of_mntpt("/sys/fs/cgroup") == 2) { | |
1292 | char *user_slice; | |
1293 | bool has_user_slice = false; | |
1294 | ||
1295 | mountpoint = must_copy_string("/sys/fs/cgroup"); | |
1296 | if (!mountpoint) | |
1297 | goto cleanup; | |
1298 | ||
1299 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); | |
1300 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1301 | has_user_slice = true; | |
1302 | free(user_slice); | |
1303 | ||
1304 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); | |
1305 | ||
1306 | ret = true; | |
1307 | goto cleanup; | |
1308 | } | |
1309 | ||
1310 | f = fopen("/proc/self/mountinfo", "r"); | |
1311 | if (!f) | |
1312 | goto cleanup; | |
1313 | ||
1314 | /* we support simple cgroup mounts and lxcfs mounts */ | |
1315 | while (getline(&line, &len, f) != -1) { | |
1316 | char *user_slice; | |
1317 | bool has_user_slice = false; | |
1318 | if (!is_cgv2(line)) | |
1319 | continue; | |
1320 | ||
1321 | mountpoint = get_mountpoint(line); | |
1322 | if (!mountpoint) | |
1323 | continue; | |
1324 | ||
1325 | user_slice = must_make_path(mountpoint, current_cgroup, NULL); | |
1326 | if (cg_systemd_created_user_slice(current_cgroup, init_cgroup, user_slice, uid)) | |
1327 | has_user_slice = true; | |
1328 | free(user_slice); | |
1329 | ||
1330 | cgv2_add_controller(NULL, mountpoint, current_cgroup, init_cgroup, has_user_slice); | |
1331 | /* Although the unified hierarchy can be mounted multiple times, | |
1332 | * each of those mountpoints will expose identical information. | |
1333 | * So let the first mountpoint we find, win. | |
1334 | */ | |
1335 | ret = true; | |
1336 | break; | |
1337 | } | |
1338 | ||
1339 | pam_cgfs_debug("Detected cgroupfs v2 hierarchy at mountpoint \"%s\" with " | |
1340 | "current cgroup \"%s\" and init cgroup \"%s\".\n", | |
1341 | mountpoint, current_cgroup, init_cgroup); | |
1342 | ||
1343 | cleanup: | |
1344 | if (f) | |
1345 | fclose(f); | |
1346 | free(line); | |
1347 | ||
1348 | return ret; | |
1349 | } | |
1350 | ||
1351 | /* Detect and store information about mounted cgroupfs v1 hierarchies and the | |
1352 | * cgroupfs v2 hierarchy. | |
1353 | * Detect whether we are on a pure cgroupfs v1, cgroupfs v2, or mixed system, | |
1354 | * where some controllers are mounted into their standard cgroupfs v1 locations | |
1355 | * (/sys/fs/cgroup/<controller>) and others are mounted into the cgroupfs v2 | |
1356 | * hierarchy (/sys/fs/cgroup). | |
1357 | */ | |
1358 | static bool cg_init(uid_t uid, gid_t gid) | |
1359 | { | |
1360 | if (!cgv1_init(uid, gid)) | |
1361 | return false; | |
1362 | ||
1363 | if (!cgv2_init(uid, gid)) | |
1364 | return false; | |
1365 | ||
1366 | if (cgv1_hierarchies && cgv2_hierarchies) { | |
1367 | cg_mount_mode = CGROUP_MIXED; | |
1368 | pam_cgfs_debug("%s\n", "Detected cgroupfs v1 and v2 hierarchies."); | |
1369 | } else if (cgv1_hierarchies && !cgv2_hierarchies) { | |
1370 | cg_mount_mode = CGROUP_PURE_V1; | |
1371 | pam_cgfs_debug("%s\n", "Detected cgroupfs v1 hierarchies."); | |
1372 | } else if (cgv2_hierarchies && !cgv1_hierarchies) { | |
1373 | cg_mount_mode = CGROUP_PURE_V2; | |
1374 | pam_cgfs_debug("%s\n", "Detected cgroupfs v2 hierarchies."); | |
1375 | } else { | |
1376 | cg_mount_mode = CGROUP_UNKNOWN; | |
1377 | mysyslog(LOG_ERR, "Could not detect cgroupfs hierarchy.\n", NULL); | |
1378 | } | |
1379 | ||
1380 | if (cg_mount_mode == CGROUP_UNKNOWN) | |
1381 | return false; | |
1382 | ||
1383 | return true; | |
1384 | } | |
1385 | ||
1386 | /* Try to move/migrate us into @cgroup in a cgroupfs v1 hierarchy. */ | |
1387 | static bool cgv1_enter(const char *cgroup) | |
1388 | { | |
1389 | struct cgv1_hierarchy **it; | |
1390 | ||
1391 | for (it = cgv1_hierarchies; it && *it; it++) { | |
1392 | char **controller; | |
1393 | bool entered = false; | |
1394 | ||
1395 | if (!(*it)->controllers || !(*it)->mountpoint || | |
1396 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
1397 | continue; | |
1398 | ||
1399 | for (controller = (*it)->controllers; controller && *controller; | |
1400 | controller++) { | |
1401 | char *path; | |
1402 | ||
1403 | /* We've already been placed in a user slice, so we | |
1404 | * don't need to enter the cgroup again. | |
1405 | */ | |
1406 | if ((*it)->systemd_user_slice) { | |
1407 | entered = true; | |
1408 | break; | |
1409 | } | |
1410 | ||
1411 | path = must_make_path((*it)->mountpoint, | |
1412 | (*it)->init_cgroup, | |
1413 | cgroup, | |
1414 | "/cgroup.procs", | |
1415 | NULL); | |
1416 | if (!file_exists(path)) { | |
1417 | free(path); | |
1418 | path = must_make_path((*it)->mountpoint, | |
1419 | (*it)->init_cgroup, | |
1420 | cgroup, | |
1421 | "/tasks", | |
1422 | NULL); | |
1423 | } | |
1424 | pam_cgfs_debug("Attempting to enter cgroupfs v1 hierarchy in \"%s\" cgroup.\n", path); | |
1425 | entered = write_int(path, (int)getpid()); | |
1426 | if (entered) { | |
1427 | free(path); | |
1428 | break; | |
1429 | } | |
1430 | pam_cgfs_debug("Failed to enter cgroupfs v1 hierarchy in \"%s\" cgroup.\n", path); | |
1431 | free(path); | |
1432 | } | |
1433 | if (!entered) | |
1434 | return false; | |
1435 | } | |
1436 | ||
1437 | return true; | |
1438 | } | |
1439 | ||
1440 | /* Try to move/migrate us into @cgroup in the cgroupfs v2 hierarchy. */ | |
1441 | static bool cgv2_enter(const char *cgroup) | |
1442 | { | |
1443 | struct cgv2_hierarchy *v2; | |
1444 | char *path; | |
1445 | bool entered = false; | |
1446 | ||
1447 | if (!cgv2_hierarchies) | |
1448 | return true; | |
1449 | ||
1450 | v2 = *cgv2_hierarchies; | |
1451 | ||
1452 | if (!v2->mountpoint || !v2->base_cgroup) | |
1453 | return false; | |
1454 | ||
1455 | if (!v2->create_rw_cgroup || v2->systemd_user_slice) | |
1456 | return true; | |
1457 | ||
1458 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, "/cgroup.procs", NULL); | |
1459 | pam_cgfs_debug("Attempting to enter cgroupfs v2 hierarchy in cgroup \"%s\".\n", path); | |
1460 | entered = write_int(path, (int)getpid()); | |
1461 | if (!entered) { | |
1462 | pam_cgfs_debug("Failed to enter cgroupfs v2 hierarchy in cgroup \"%s\".\n", path); | |
1463 | free(path); | |
1464 | return false; | |
1465 | } | |
1466 | ||
1467 | free(path); | |
1468 | ||
1469 | return true; | |
1470 | } | |
1471 | ||
1472 | /* Wrapper around cgv{1,2}_enter(). */ | |
1473 | static bool cg_enter(const char *cgroup) | |
1474 | { | |
1475 | if (!cgv1_enter(cgroup)) { | |
1476 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to enter cgroups.\n", NULL); | |
1477 | return false; | |
1478 | } | |
1479 | ||
1480 | if (!cgv2_enter(cgroup)) { | |
1481 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to enter cgroups.\n", NULL); | |
1482 | return false; | |
1483 | } | |
1484 | ||
1485 | return true; | |
1486 | } | |
1487 | ||
1488 | /* Escape to root cgroup in all detected cgroupfs v1 hierarchies. */ | |
1489 | static void cgv1_escape(void) | |
1490 | { | |
1491 | struct cgv1_hierarchy **it; | |
1492 | ||
1493 | /* In case systemd hasn't already placed us in a user slice for the | |
1494 | * cpuset v1 controller we will reside in the root cgroup. This means | |
1495 | * that cgroup.clone_children will not have been initialized for us so | |
1496 | * we need to do it. | |
1497 | */ | |
1498 | for (it = cgv1_hierarchies; it && *it; it++) | |
1499 | if (!cgv1_handle_root_cpuset_hierarchy(*it)) | |
1500 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to initialize cpuset.\n", NULL); | |
1501 | ||
1502 | if (!cgv1_enter("/")) | |
1503 | mysyslog(LOG_WARNING, "cgroupfs v1: Failed to escape to init's cgroup.\n", NULL); | |
1504 | } | |
1505 | ||
1506 | /* Escape to root cgroup in the cgroupfs v2 hierarchy. */ | |
1507 | static void cgv2_escape(void) | |
1508 | { | |
1509 | if (!cgv2_enter("/")) | |
1510 | mysyslog(LOG_WARNING, "cgroupfs v2: Failed to escape to init's cgroup.\n", NULL); | |
1511 | } | |
1512 | ||
1513 | /* Wrapper around cgv{1,2}_escape(). */ | |
1514 | static void cg_escape(void) | |
1515 | { | |
1516 | cgv1_escape(); | |
1517 | cgv2_escape(); | |
1518 | } | |
1519 | ||
1520 | /* Get uid and gid for @user. */ | |
1521 | static bool get_uid_gid(const char *user, uid_t *uid, gid_t *gid) | |
1522 | { | |
2dce415b DJ |
1523 | strcut passwd pwent; |
1524 | struct passwd *pwentp = NULL; | |
1525 | char *buf; | |
1526 | size_t bufsize; | |
1527 | int ret; | |
1528 | ||
1529 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1530 | if (bufsize == -1) | |
1531 | bufsize = 1024; | |
1532 | ||
1533 | buf = malloc(bufsize); | |
1534 | if (!buf) | |
1535 | return false; | |
1536 | ||
1537 | ret = getpwnam_r(user, &pwent, buf, bufsize, &pwentp); | |
1538 | if (!pwentp) { | |
1539 | if (ret == 0) | |
1540 | mysyslog(LOG_ERR, "Could not find matched password record.\n", NULL); | |
7ac43d3d | 1541 | |
2dce415b | 1542 | free(buf); |
7ac43d3d | 1543 | return false; |
2dce415b | 1544 | } |
7ac43d3d | 1545 | |
2dce415b DJ |
1546 | *uid = pwent.pw_uid; |
1547 | *gid = pwent.pw_gid; | |
1548 | free(buf); | |
7ac43d3d CB |
1549 | |
1550 | return true; | |
1551 | } | |
1552 | ||
1553 | /* Check if cgroup belongs to our uid and gid. If so, reuse it. */ | |
1554 | static bool cg_belongs_to_uid_gid(const char *path, uid_t uid, gid_t gid) | |
1555 | { | |
1556 | struct stat statbuf; | |
1557 | ||
1558 | if (stat(path, &statbuf) < 0) | |
1559 | return false; | |
1560 | ||
1561 | if (!(statbuf.st_uid == uid) || !(statbuf.st_gid == gid)) | |
1562 | return false; | |
1563 | ||
1564 | return true; | |
1565 | } | |
1566 | ||
1567 | /* Create cpumask from cpulist aka turn: | |
1568 | * | |
1569 | * 0,2-3 | |
1570 | * | |
1571 | * into bit array | |
1572 | * | |
1573 | * 1 0 1 1 | |
1574 | */ | |
1575 | static uint32_t *cg_cpumask(char *buf, size_t nbits) | |
1576 | { | |
1577 | char *token; | |
1578 | char *saveptr = NULL; | |
1579 | size_t arrlen = BITS_TO_LONGS(nbits); | |
1580 | uint32_t *bitarr = calloc(arrlen, sizeof(uint32_t)); | |
1581 | if (!bitarr) | |
1582 | return NULL; | |
1583 | ||
1584 | for (; (token = strtok_r(buf, ",", &saveptr)); buf = NULL) { | |
1585 | errno = 0; | |
1586 | unsigned start = strtoul(token, NULL, 0); | |
1587 | unsigned end = start; | |
1588 | ||
1589 | char *range = strchr(token, '-'); | |
1590 | if (range) | |
1591 | end = strtoul(range + 1, NULL, 0); | |
1592 | if (!(start <= end)) { | |
1593 | free(bitarr); | |
1594 | return NULL; | |
1595 | } | |
1596 | ||
1597 | if (end >= nbits) { | |
1598 | free(bitarr); | |
1599 | return NULL; | |
1600 | } | |
1601 | ||
1602 | while (start <= end) | |
1603 | set_bit(start++, bitarr); | |
1604 | } | |
1605 | ||
1606 | return bitarr; | |
1607 | } | |
1608 | ||
1609 | static char *string_join(const char *sep, const char **parts, bool use_as_prefix) | |
1610 | { | |
1611 | char *result; | |
1612 | char **p; | |
1613 | size_t sep_len = strlen(sep); | |
1614 | size_t result_len = use_as_prefix * sep_len; | |
1615 | ||
1616 | if (!parts) | |
1617 | return NULL; | |
1618 | ||
1619 | /* calculate new string length */ | |
1620 | for (p = (char **)parts; *p; p++) | |
1621 | result_len += (p > (char **)parts) * sep_len + strlen(*p); | |
1622 | ||
1623 | result = calloc(result_len + 1, sizeof(char)); | |
1624 | if (!result) | |
1625 | return NULL; | |
1626 | ||
1627 | if (use_as_prefix) | |
1628 | strcpy(result, sep); | |
1629 | for (p = (char **)parts; *p; p++) { | |
1630 | if (p > (char **)parts) | |
1631 | strcat(result, sep); | |
1632 | strcat(result, *p); | |
1633 | } | |
1634 | ||
1635 | return result; | |
1636 | } | |
1637 | ||
1638 | /* The largest integer that can fit into long int is 2^64. This is a | |
1639 | * 20-digit number. | |
1640 | */ | |
1641 | #define __IN_TO_STR_LEN 21 | |
1642 | /* Turn cpumask into simple, comma-separated cpulist. */ | |
1643 | static char *cg_cpumask_to_cpulist(uint32_t *bitarr, size_t nbits) | |
1644 | { | |
1645 | size_t i; | |
1646 | int ret; | |
1647 | char numstr[__IN_TO_STR_LEN] = {0}; | |
1648 | char **cpulist = NULL; | |
1649 | ||
1650 | for (i = 0; i <= nbits; i++) { | |
1651 | if (is_set(i, bitarr)) { | |
1652 | ret = snprintf(numstr, __IN_TO_STR_LEN, "%zu", i); | |
1653 | if (ret < 0 || (size_t)ret >= __IN_TO_STR_LEN) { | |
1654 | free_string_list(cpulist); | |
1655 | return NULL; | |
1656 | } | |
1657 | must_append_string(&cpulist, numstr); | |
1658 | } | |
1659 | } | |
1660 | return string_join(",", (const char **)cpulist, false); | |
1661 | } | |
1662 | ||
1663 | static ssize_t cg_get_max_cpus(char *cpulist) | |
1664 | { | |
1665 | char *c1, *c2; | |
1666 | char *maxcpus = cpulist; | |
1667 | size_t cpus = 0; | |
1668 | ||
1669 | c1 = strrchr(maxcpus, ','); | |
1670 | if (c1) | |
1671 | c1++; | |
1672 | ||
1673 | c2 = strrchr(maxcpus, '-'); | |
1674 | if (c2) | |
1675 | c2++; | |
1676 | ||
1677 | if (!c1 && !c2) | |
1678 | c1 = maxcpus; | |
1679 | else if (c1 < c2) | |
1680 | c1 = c2; | |
1681 | ||
1682 | /* If the above logic is correct, c1 should always hold a valid string | |
1683 | * here. | |
1684 | */ | |
1685 | ||
1686 | errno = 0; | |
1687 | cpus = strtoul(c1, NULL, 0); | |
1688 | if (errno != 0) | |
1689 | return -1; | |
1690 | ||
1691 | return cpus; | |
1692 | } | |
1693 | ||
1694 | static ssize_t write_nointr(int fd, const void* buf, size_t count) | |
1695 | { | |
1696 | ssize_t ret; | |
1697 | again: | |
1698 | ret = write(fd, buf, count); | |
1699 | if (ret < 0 && errno == EINTR) | |
1700 | goto again; | |
1701 | return ret; | |
1702 | } | |
1703 | ||
1704 | static int write_to_file(const char *filename, const void* buf, size_t count, bool add_newline) | |
1705 | { | |
1706 | int fd, saved_errno; | |
1707 | ssize_t ret; | |
1708 | ||
1709 | fd = open(filename, O_WRONLY | O_TRUNC | O_CREAT | O_CLOEXEC, 0666); | |
1710 | if (fd < 0) | |
1711 | return -1; | |
1712 | ret = write_nointr(fd, buf, count); | |
1713 | if (ret < 0) | |
1714 | goto out_error; | |
1715 | if ((size_t)ret != count) | |
1716 | goto out_error; | |
1717 | if (add_newline) { | |
1718 | ret = write_nointr(fd, "\n", 1); | |
1719 | if (ret != 1) | |
1720 | goto out_error; | |
1721 | } | |
1722 | close(fd); | |
1723 | return 0; | |
1724 | ||
1725 | out_error: | |
1726 | saved_errno = errno; | |
1727 | close(fd); | |
1728 | errno = saved_errno; | |
1729 | return -1; | |
1730 | } | |
1731 | ||
1732 | #define __ISOL_CPUS "/sys/devices/system/cpu/isolated" | |
1733 | static bool cg_filter_and_set_cpus(char *path, bool am_initialized) | |
1734 | { | |
1735 | char *lastslash, *fpath, oldv; | |
1736 | int ret; | |
1737 | ssize_t i; | |
1738 | ||
1739 | ssize_t maxposs = 0, maxisol = 0; | |
1740 | char *cpulist = NULL, *posscpus = NULL, *isolcpus = NULL; | |
1741 | uint32_t *possmask = NULL, *isolmask = NULL; | |
1742 | bool bret = false, flipped_bit = false; | |
1743 | ||
1744 | lastslash = strrchr(path, '/'); | |
1745 | if (!lastslash) { // bug... this shouldn't be possible | |
1746 | pam_cgfs_debug("Invalid path: %s.\n", path); | |
1747 | return bret; | |
1748 | } | |
1749 | oldv = *lastslash; | |
1750 | *lastslash = '\0'; | |
1751 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
1752 | posscpus = read_file(fpath); | |
1753 | if (!posscpus) { | |
1754 | pam_cgfs_debug("Could not read file: %s.\n", fpath); | |
1755 | goto on_error; | |
1756 | } | |
1757 | ||
1758 | /* Get maximum number of cpus found in possible cpuset. */ | |
1759 | maxposs = cg_get_max_cpus(posscpus); | |
1760 | if (maxposs < 0) | |
1761 | goto on_error; | |
1762 | ||
1763 | if (!file_exists(__ISOL_CPUS)) { | |
1764 | /* This system doesn't expose isolated cpus. */ | |
1765 | pam_cgfs_debug("%s", "Path: "__ISOL_CPUS" to read isolated cpus from does not exist.\n"); | |
1766 | cpulist = posscpus; | |
1767 | /* No isolated cpus but we weren't already initialized by | |
1768 | * someone. We should simply copy the parents cpuset.cpus | |
1769 | * values. | |
1770 | */ | |
1771 | if (!am_initialized) { | |
1772 | pam_cgfs_debug("%s", "Copying cpuset of parent cgroup.\n"); | |
1773 | goto copy_parent; | |
1774 | } | |
1775 | /* No isolated cpus but we were already initialized by someone. | |
1776 | * Nothing more to do for us. | |
1777 | */ | |
1778 | goto on_success; | |
1779 | } | |
1780 | ||
1781 | isolcpus = read_file(__ISOL_CPUS); | |
1782 | if (!isolcpus) { | |
1783 | pam_cgfs_debug("%s", "Could not read file "__ISOL_CPUS"\n"); | |
1784 | goto on_error; | |
1785 | } | |
1786 | if (!isdigit(isolcpus[0])) { | |
1787 | pam_cgfs_debug("%s", "No isolated cpus detected.\n"); | |
1788 | cpulist = posscpus; | |
1789 | /* No isolated cpus but we weren't already initialized by | |
1790 | * someone. We should simply copy the parents cpuset.cpus | |
1791 | * values. | |
1792 | */ | |
1793 | if (!am_initialized) { | |
1794 | pam_cgfs_debug("%s", "Copying cpuset of parent cgroup.\n"); | |
1795 | goto copy_parent; | |
1796 | } | |
1797 | /* No isolated cpus but we were already initialized by someone. | |
1798 | * Nothing more to do for us. | |
1799 | */ | |
1800 | goto on_success; | |
1801 | } | |
1802 | ||
1803 | /* Get maximum number of cpus found in isolated cpuset. */ | |
1804 | maxisol = cg_get_max_cpus(isolcpus); | |
1805 | if (maxisol < 0) | |
1806 | goto on_error; | |
1807 | ||
1808 | if (maxposs < maxisol) | |
1809 | maxposs = maxisol; | |
1810 | maxposs++; | |
1811 | ||
1812 | possmask = cg_cpumask(posscpus, maxposs); | |
1813 | if (!possmask) { | |
1814 | pam_cgfs_debug("%s", "Could not create cpumask for all possible cpus.\n"); | |
1815 | goto on_error; | |
1816 | } | |
1817 | ||
1818 | isolmask = cg_cpumask(isolcpus, maxposs); | |
1819 | if (!isolmask) { | |
1820 | pam_cgfs_debug("%s", "Could not create cpumask for all isolated cpus.\n"); | |
1821 | goto on_error; | |
1822 | } | |
1823 | ||
1824 | for (i = 0; i <= maxposs; i++) { | |
1825 | if (is_set(i, isolmask) && is_set(i, possmask)) { | |
1826 | flipped_bit = true; | |
1827 | clear_bit(i, possmask); | |
1828 | } | |
1829 | } | |
1830 | ||
1831 | if (!flipped_bit) { | |
1832 | pam_cgfs_debug("%s", "No isolated cpus present in cpuset.\n"); | |
1833 | goto on_success; | |
1834 | } | |
1835 | pam_cgfs_debug("%s", "Removed isolated cpus from cpuset.\n"); | |
1836 | ||
1837 | cpulist = cg_cpumask_to_cpulist(possmask, maxposs); | |
1838 | if (!cpulist) { | |
1839 | pam_cgfs_debug("%s", "Could not create cpu list.\n"); | |
1840 | goto on_error; | |
1841 | } | |
1842 | ||
1843 | copy_parent: | |
1844 | *lastslash = oldv; | |
1845 | fpath = must_make_path(path, "cpuset.cpus", NULL); | |
1846 | ret = write_to_file(fpath, cpulist, strlen(cpulist), false); | |
1847 | if (ret < 0) { | |
1848 | pam_cgfs_debug("Could not write cpu list to: %s.\n", fpath); | |
1849 | goto on_error; | |
1850 | } | |
1851 | ||
1852 | on_success: | |
1853 | bret = true; | |
1854 | ||
1855 | on_error: | |
1856 | free(fpath); | |
1857 | ||
1858 | free(isolcpus); | |
1859 | free(isolmask); | |
1860 | ||
1861 | if (posscpus != cpulist) | |
1862 | free(posscpus); | |
1863 | free(possmask); | |
1864 | ||
1865 | free(cpulist); | |
1866 | return bret; | |
1867 | } | |
1868 | ||
1869 | int read_from_file(const char *filename, void* buf, size_t count) | |
1870 | { | |
1871 | int fd = -1, saved_errno; | |
1872 | ssize_t ret; | |
1873 | ||
1874 | fd = open(filename, O_RDONLY | O_CLOEXEC); | |
1875 | if (fd < 0) | |
1876 | return -1; | |
1877 | ||
1878 | if (!buf || !count) { | |
1879 | char buf2[100]; | |
1880 | size_t count2 = 0; | |
1881 | while ((ret = read(fd, buf2, 100)) > 0) | |
1882 | count2 += ret; | |
1883 | if (ret >= 0) | |
1884 | ret = count2; | |
1885 | } else { | |
1886 | memset(buf, 0, count); | |
1887 | ret = read(fd, buf, count); | |
1888 | } | |
1889 | ||
1890 | if (ret < 0) | |
1891 | pam_cgfs_debug("read %s: %s", filename, strerror(errno)); | |
1892 | ||
1893 | saved_errno = errno; | |
1894 | close(fd); | |
1895 | errno = saved_errno; | |
1896 | return ret; | |
1897 | } | |
1898 | ||
1899 | /* Copy contents of parent(@path)/@file to @path/@file */ | |
1900 | static bool cg_copy_parent_file(char *path, char *file) | |
1901 | { | |
1902 | char *lastslash, *value = NULL, *fpath, oldv; | |
1903 | int len = 0; | |
1904 | int ret; | |
1905 | ||
1906 | lastslash = strrchr(path, '/'); | |
1907 | if (!lastslash) { // bug... this shouldn't be possible | |
1908 | pam_cgfs_debug("cgfsng:copy_parent_file: bad path %s", path); | |
1909 | return false; | |
1910 | } | |
1911 | oldv = *lastslash; | |
1912 | *lastslash = '\0'; | |
1913 | fpath = must_make_path(path, file, NULL); | |
1914 | len = read_from_file(fpath, NULL, 0); | |
1915 | if (len <= 0) | |
1916 | goto bad; | |
1917 | value = must_alloc(len + 1); | |
1918 | if (read_from_file(fpath, value, len) != len) | |
1919 | goto bad; | |
1920 | free(fpath); | |
1921 | *lastslash = oldv; | |
1922 | fpath = must_make_path(path, file, NULL); | |
1923 | ret = write_to_file(fpath, value, len, false); | |
1924 | if (ret < 0) | |
1925 | pam_cgfs_debug("Unable to write %s to %s", value, fpath); | |
1926 | free(fpath); | |
1927 | free(value); | |
1928 | return ret >= 0; | |
1929 | ||
1930 | bad: | |
1931 | pam_cgfs_debug("Error reading '%s'", fpath); | |
1932 | free(fpath); | |
1933 | free(value); | |
1934 | return false; | |
1935 | } | |
1936 | ||
1937 | /* In case systemd hasn't already placed us in a user slice for the cpuset v1 | |
1938 | * controller we will reside in the root cgroup. This means that | |
1939 | * cgroup.clone_children will not have been initialized for us so we need to do | |
1940 | * it. | |
1941 | */ | |
1942 | static bool cgv1_handle_root_cpuset_hierarchy(struct cgv1_hierarchy *h) | |
1943 | { | |
1944 | char *clonechildrenpath, v; | |
1945 | ||
1946 | if (!string_in_list(h->controllers, "cpuset")) | |
1947 | return true; | |
1948 | ||
1949 | clonechildrenpath = must_make_path(h->mountpoint, "cgroup.clone_children", NULL); | |
1950 | ||
1951 | if (read_from_file(clonechildrenpath, &v, 1) < 0) { | |
1952 | pam_cgfs_debug("Failed to read '%s'", clonechildrenpath); | |
1953 | free(clonechildrenpath); | |
1954 | return false; | |
1955 | } | |
1956 | ||
1957 | if (v == '1') { /* already set for us by someone else */ | |
1958 | free(clonechildrenpath); | |
1959 | return true; | |
1960 | } | |
1961 | ||
1962 | if (write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
1963 | /* Set clone_children so children inherit our settings */ | |
1964 | pam_cgfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
1965 | free(clonechildrenpath); | |
1966 | return false; | |
1967 | } | |
1968 | free(clonechildrenpath); | |
1969 | return true; | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * Initialize the cpuset hierarchy in first directory of @gname and | |
1974 | * set cgroup.clone_children so that children inherit settings. | |
1975 | * Since the h->base_path is populated by init or ourselves, we know | |
1976 | * it is already initialized. | |
1977 | */ | |
1978 | static bool cgv1_handle_cpuset_hierarchy(struct cgv1_hierarchy *h, | |
1979 | const char *cgroup) | |
1980 | { | |
1981 | char *cgpath, *clonechildrenpath, v, *slash; | |
1982 | ||
1983 | if (!string_in_list(h->controllers, "cpuset")) | |
1984 | return true; | |
1985 | ||
1986 | if (*cgroup == '/') | |
1987 | cgroup++; | |
1988 | slash = strchr(cgroup, '/'); | |
1989 | if (slash) | |
1990 | *slash = '\0'; | |
1991 | ||
1992 | cgpath = must_make_path(h->mountpoint, h->base_cgroup, cgroup, NULL); | |
1993 | if (slash) | |
1994 | *slash = '/'; | |
c4a4578f | 1995 | if (do_mkdir(cgpath, 0755) < 0 && errno != EEXIST) { |
7ac43d3d CB |
1996 | pam_cgfs_debug("Failed to create '%s'", cgpath); |
1997 | free(cgpath); | |
1998 | return false; | |
1999 | } | |
2000 | clonechildrenpath = must_make_path(cgpath, "cgroup.clone_children", NULL); | |
2001 | if (!file_exists(clonechildrenpath)) { /* unified hierarchy doesn't have clone_children */ | |
2002 | free(clonechildrenpath); | |
2003 | free(cgpath); | |
2004 | return true; | |
2005 | } | |
2006 | if (read_from_file(clonechildrenpath, &v, 1) < 0) { | |
2007 | pam_cgfs_debug("Failed to read '%s'", clonechildrenpath); | |
2008 | free(clonechildrenpath); | |
2009 | free(cgpath); | |
2010 | return false; | |
2011 | } | |
2012 | ||
2013 | /* Make sure any isolated cpus are removed from cpuset.cpus. */ | |
2014 | if (!cg_filter_and_set_cpus(cgpath, v == '1')) { | |
2015 | pam_cgfs_debug("%s", "Failed to remove isolated cpus.\n"); | |
2016 | free(clonechildrenpath); | |
2017 | free(cgpath); | |
2018 | return false; | |
2019 | } | |
2020 | ||
2021 | if (v == '1') { /* already set for us by someone else */ | |
2022 | pam_cgfs_debug("%s", "\"cgroup.clone_children\" was already set to \"1\".\n"); | |
2023 | free(clonechildrenpath); | |
2024 | free(cgpath); | |
2025 | return true; | |
2026 | } | |
2027 | ||
2028 | /* copy parent's settings */ | |
2029 | if (!cg_copy_parent_file(cgpath, "cpuset.mems")) { | |
2030 | pam_cgfs_debug("%s", "Failed to copy \"cpuset.mems\" settings.\n"); | |
2031 | free(cgpath); | |
2032 | free(clonechildrenpath); | |
2033 | return false; | |
2034 | } | |
2035 | free(cgpath); | |
2036 | ||
2037 | if (write_to_file(clonechildrenpath, "1", 1, false) < 0) { | |
2038 | /* Set clone_children so children inherit our settings */ | |
2039 | pam_cgfs_debug("Failed to write 1 to %s", clonechildrenpath); | |
2040 | free(clonechildrenpath); | |
2041 | return false; | |
2042 | } | |
2043 | free(clonechildrenpath); | |
2044 | return true; | |
2045 | } | |
2046 | ||
2047 | /* Create and chown @cgroup for all given controllers in a cgroupfs v1 hierarchy | |
2048 | * (For example, create @cgroup for the cpu and cpuacct controller mounted into | |
2049 | * /sys/fs/cgroup/cpu,cpuacct). Check if the path already exists and report back | |
2050 | * to the caller in @existed. | |
2051 | */ | |
2052 | #define __PAM_CGFS_USER "/user/" | |
2053 | #define __PAM_CGFS_USER_LEN 6 | |
2054 | static bool cgv1_create_one(struct cgv1_hierarchy *h, const char *cgroup, uid_t uid, gid_t gid, bool *existed) | |
2055 | { | |
2056 | char *clean_base_cgroup, *path; | |
2057 | char **controller; | |
2058 | struct cgv1_hierarchy *it; | |
2059 | bool created = false; | |
2060 | ||
2061 | *existed = false; | |
2062 | it = h; | |
2063 | for (controller = it->controllers; controller && *controller; | |
2064 | controller++) { | |
2065 | if (!cgv1_handle_cpuset_hierarchy(it, cgroup)) | |
2066 | return false; | |
2067 | ||
2068 | /* If systemd has already created a cgroup for us, keep using | |
2069 | * it. | |
2070 | */ | |
2071 | if (cg_systemd_chown_existing_cgroup(it->mountpoint, | |
2072 | it->base_cgroup, uid, gid, | |
2073 | it->systemd_user_slice)) { | |
2074 | return true; | |
2075 | } | |
2076 | ||
2077 | /* We need to make sure that we do not create an endless chain | |
2078 | * of sub-cgroups. So we check if we have already logged in | |
2079 | * somehow (sudo -i, su, etc.) and have created a | |
2080 | * /user/PAM_user/idx cgroup. If so, we skip that part. For most | |
2081 | * cgroups this is unnecessary since we use the init_cgroup | |
2082 | * anyway, but for controllers which have an existing systemd | |
2083 | * cgroup that does not match the current uid, this is pretty | |
2084 | * useful. | |
2085 | */ | |
2086 | if (strncmp(it->base_cgroup, __PAM_CGFS_USER, __PAM_CGFS_USER_LEN) == 0) { | |
2087 | free(it->base_cgroup); | |
2088 | it->base_cgroup = must_copy_string("/"); | |
2089 | } else { | |
2090 | clean_base_cgroup = | |
2091 | strstr(it->base_cgroup, __PAM_CGFS_USER); | |
2092 | if (clean_base_cgroup) | |
2093 | *clean_base_cgroup = '\0'; | |
2094 | } | |
2095 | ||
2096 | path = must_make_path(it->mountpoint, it->init_cgroup, cgroup, NULL); | |
2097 | pam_cgfs_debug("Constructing path: %s.\n", path); | |
2098 | if (file_exists(path)) { | |
2099 | bool our_cg = cg_belongs_to_uid_gid(path, uid, gid); | |
2100 | pam_cgfs_debug("%s existed and does %shave our uid: %d and gid: %d.\n", path, our_cg ? "" : "not ", uid, gid); | |
2101 | free(path); | |
2102 | if (our_cg) | |
2103 | *existed = false; | |
2104 | else | |
2105 | *existed = true; | |
2106 | return our_cg; | |
2107 | } | |
2108 | created = mkdir_parent(it->mountpoint, path); | |
2109 | if (!created) { | |
2110 | free(path); | |
2111 | continue; | |
2112 | } | |
2113 | if (chown(path, uid, gid) < 0) | |
2114 | mysyslog(LOG_WARNING, | |
2115 | "Failed to chown %s to %d:%d: %s.\n", path, | |
2116 | (int)uid, (int)gid, strerror(errno), NULL); | |
2117 | pam_cgfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
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); | |
2246 | pam_cgfs_debug("Constructing path \"%s\".\n", path); | |
2247 | if (file_exists(path)) { | |
2248 | our_cg = cg_belongs_to_uid_gid(path, uid, gid); | |
2249 | pam_cgfs_debug( | |
2250 | "%s existed and does %shave our uid: %d and gid: %d.\n", | |
2251 | path, our_cg ? "" : "not ", uid, gid); | |
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) | |
2270 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", | |
2271 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
2272 | else | |
2273 | pam_cgfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
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); | |
2284 | ret = chown(path, uid, gid); | |
2285 | if (ret < 0) | |
2286 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", | |
2287 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
2288 | else | |
2289 | pam_cgfs_debug("Chowned %s to %d:%d.\n", path, (int)uid, (int)gid); | |
2290 | free(path); | |
2291 | ||
2292 | /* chown cgroup.subtree_control to user */ | |
2293 | if (v2->systemd_user_slice) | |
2294 | path = must_make_path(v2->mountpoint, v2->base_cgroup, | |
2295 | "/cgroup.subtree_control", NULL); | |
2296 | else | |
2297 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, | |
2298 | "/cgroup.subtree_control", NULL); | |
2299 | ret = chown(path, uid, gid); | |
2300 | if (ret < 0) | |
2301 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", | |
2302 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
2303 | free(path); | |
2304 | ||
2305 | /* chown cgroup.threads to user */ | |
2306 | if (v2->systemd_user_slice) | |
2307 | path = must_make_path(v2->mountpoint, v2->base_cgroup, | |
2308 | "/cgroup.threads", NULL); | |
2309 | else | |
2310 | path = must_make_path(v2->mountpoint, v2->base_cgroup, cgroup, | |
2311 | "/cgroup.threads", NULL); | |
2312 | ret = chown(path, uid, gid); | |
2313 | if (ret < 0 && errno != ENOENT) | |
2314 | mysyslog(LOG_WARNING, "Failed to chown %s to %d:%d: %s.\n", | |
2315 | path, (int)uid, (int)gid, strerror(errno), NULL); | |
2316 | free(path); | |
2317 | ||
2318 | return true; | |
2319 | } | |
2320 | ||
2321 | /* Create writeable cgroups for @user at login. Details can be found in the | |
2322 | * preamble/license at the top of this file. | |
2323 | */ | |
2324 | static int handle_login(const char *user, uid_t uid, gid_t gid) | |
2325 | { | |
2326 | int idx = 0, ret; | |
2327 | bool existed; | |
2328 | char cg[MAXPATHLEN]; | |
2329 | ||
2330 | cg_escape(); | |
2331 | ||
2332 | while (idx >= 0) { | |
2333 | ret = snprintf(cg, MAXPATHLEN, "/user/%s/%d", user, idx); | |
2334 | if (ret < 0 || ret >= MAXPATHLEN) { | |
2335 | mysyslog(LOG_ERR, "Username too long.\n", NULL); | |
2336 | return PAM_SESSION_ERR; | |
2337 | } | |
2338 | ||
2339 | existed = false; | |
2340 | if (!cgv2_create(cg, uid, gid, &existed)) { | |
2341 | if (existed) { | |
2342 | cgv2_remove(cg); | |
2343 | idx++; | |
2344 | continue; | |
2345 | } | |
2346 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s.\n", user, NULL); | |
2347 | return PAM_SESSION_ERR; | |
2348 | } | |
2349 | ||
2350 | existed = false; | |
2351 | if (!cgv1_create(cg, uid, gid, &existed)) { | |
2352 | if (existed) { | |
2353 | cgv2_remove(cg); | |
2354 | idx++; | |
2355 | continue; | |
2356 | } | |
2357 | mysyslog(LOG_ERR, "Failed to create a cgroup for user %s.\n", user, NULL); | |
2358 | return PAM_SESSION_ERR; | |
2359 | } | |
2360 | ||
2361 | if (!cg_enter(cg)) { | |
2362 | mysyslog( LOG_ERR, "Failed to enter user cgroup %s for user %s.\n", cg, user, NULL); | |
2363 | return PAM_SESSION_ERR; | |
2364 | } | |
2365 | break; | |
2366 | } | |
2367 | ||
2368 | return PAM_SUCCESS; | |
2369 | } | |
2370 | ||
2371 | /* Try to prune cgroups we created and that now are empty from all cgroupfs v1 | |
2372 | * hierarchies. | |
2373 | */ | |
2374 | static bool cgv1_prune_empty_cgroups(const char *user) | |
2375 | { | |
2376 | bool controller_removed = true; | |
2377 | bool all_removed = true; | |
2378 | struct cgv1_hierarchy **it; | |
2379 | ||
2380 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2381 | int ret; | |
2382 | char *path_base, *path_init; | |
2383 | char **controller; | |
2384 | ||
2385 | if (!(*it)->controllers || !(*it)->mountpoint || | |
2386 | !(*it)->init_cgroup || !(*it)->create_rw_cgroup) | |
2387 | continue; | |
2388 | ||
2389 | for (controller = (*it)->controllers; controller && *controller; | |
2390 | controller++) { | |
2391 | bool path_base_rm, path_init_rm; | |
2392 | ||
2393 | path_base = must_make_path((*it)->mountpoint, (*it)->base_cgroup, "/user", user, NULL); | |
2394 | pam_cgfs_debug("cgroupfs v1: Trying to prune \"%s\".\n", path_base); | |
2395 | ret = recursive_rmdir(path_base); | |
2396 | if (ret == -ENOENT || ret >= 0) | |
2397 | path_base_rm = true; | |
2398 | else | |
2399 | path_base_rm = false; | |
2400 | free(path_base); | |
2401 | ||
2402 | path_init = must_make_path((*it)->mountpoint, (*it)->init_cgroup, "/user", user, NULL); | |
2403 | pam_cgfs_debug("cgroupfs v1: Trying to prune \"%s\".\n", path_init); | |
2404 | ret = recursive_rmdir(path_init); | |
2405 | if (ret == -ENOENT || ret >= 0) | |
2406 | path_init_rm = true; | |
2407 | else | |
2408 | path_init_rm = false; | |
2409 | free(path_init); | |
2410 | ||
2411 | if (!path_base_rm && !path_init_rm) { | |
2412 | controller_removed = false; | |
2413 | continue; | |
2414 | } | |
2415 | ||
2416 | controller_removed = true; | |
2417 | break; | |
2418 | } | |
2419 | if (!controller_removed) | |
2420 | all_removed = false; | |
2421 | } | |
2422 | ||
2423 | return all_removed; | |
2424 | } | |
2425 | ||
2426 | /* Try to prune cgroup we created and that now is empty from the cgroupfs v2 | |
2427 | * hierarchy. | |
2428 | */ | |
2429 | static bool cgv2_prune_empty_cgroups(const char *user) | |
2430 | { | |
2431 | int ret; | |
2432 | struct cgv2_hierarchy *v2; | |
2433 | char *path_base, *path_init; | |
2434 | bool path_base_rm, path_init_rm; | |
2435 | ||
2436 | if (!cgv2_hierarchies) | |
2437 | return true; | |
2438 | ||
2439 | v2 = *cgv2_hierarchies; | |
2440 | ||
2441 | path_base = must_make_path(v2->mountpoint, v2->base_cgroup, "/user", user, NULL); | |
2442 | pam_cgfs_debug("cgroupfs v2: Trying to prune \"%s\".\n", path_base); | |
2443 | ret = recursive_rmdir(path_base); | |
2444 | if (ret == -ENOENT || ret >= 0) | |
2445 | path_base_rm = true; | |
2446 | else | |
2447 | path_base_rm = false; | |
2448 | free(path_base); | |
2449 | ||
2450 | path_init = must_make_path(v2->mountpoint, v2->init_cgroup, "/user", user, NULL); | |
2451 | pam_cgfs_debug("cgroupfs v2: Trying to prune \"%s\".\n", path_init); | |
2452 | ret = recursive_rmdir(path_init); | |
2453 | if (ret == -ENOENT || ret >= 0) | |
2454 | path_init_rm = true; | |
2455 | else | |
2456 | path_init_rm = false; | |
2457 | free(path_init); | |
2458 | ||
2459 | if (!path_base_rm && !path_init_rm) | |
2460 | return false; | |
2461 | ||
2462 | return true; | |
2463 | } | |
2464 | ||
2465 | /* Wrapper around cgv{1,2}_prune_empty_cgroups(). */ | |
2466 | static void cg_prune_empty_cgroups(const char *user) | |
2467 | { | |
2468 | (void)cgv1_prune_empty_cgroups(user); | |
2469 | (void)cgv2_prune_empty_cgroups(user); | |
2470 | } | |
2471 | ||
2472 | /* Free allocated information for detected cgroupfs v1 hierarchies. */ | |
2473 | static void cgv1_free_hierarchies(void) | |
2474 | { | |
2475 | struct cgv1_hierarchy **it; | |
2476 | ||
2477 | if (!cgv1_hierarchies) | |
2478 | return; | |
2479 | ||
2480 | for (it = cgv1_hierarchies; it && *it; it++) { | |
2481 | if ((*it)->controllers) { | |
2482 | char **tmp; | |
2483 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) | |
2484 | free(*tmp); | |
2485 | ||
2486 | free((*it)->controllers); | |
2487 | } | |
2488 | free((*it)->mountpoint); | |
2489 | free((*it)->base_cgroup); | |
2490 | free((*it)->fullcgpath); | |
2491 | free((*it)->init_cgroup); | |
2492 | } | |
2493 | free(cgv1_hierarchies); | |
2494 | } | |
2495 | ||
2496 | /* Free allocated information for the detected cgroupfs v2 hierarchy. */ | |
2497 | static void cgv2_free_hierarchies(void) | |
2498 | { | |
2499 | struct cgv2_hierarchy **it; | |
2500 | ||
2501 | if (!cgv2_hierarchies) | |
2502 | return; | |
2503 | ||
2504 | for (it = cgv2_hierarchies; it && *it; it++) { | |
2505 | if ((*it)->controllers) { | |
2506 | char **tmp; | |
2507 | for (tmp = (*it)->controllers; tmp && *tmp; tmp++) | |
2508 | free(*tmp); | |
2509 | ||
2510 | free((*it)->controllers); | |
2511 | } | |
2512 | free((*it)->mountpoint); | |
2513 | free((*it)->base_cgroup); | |
2514 | free((*it)->fullcgpath); | |
2515 | free((*it)->init_cgroup); | |
2516 | } | |
2517 | free(cgv2_hierarchies); | |
2518 | } | |
2519 | ||
2520 | /* Wrapper around cgv{1,2}_free_hierarchies(). */ | |
2521 | static void cg_exit(void) | |
2522 | { | |
2523 | cgv1_free_hierarchies(); | |
2524 | cgv2_free_hierarchies(); | |
2525 | } | |
2526 | ||
2527 | int pam_sm_open_session(pam_handle_t *pamh, int flags, int argc, | |
2528 | const char **argv) | |
2529 | { | |
2530 | int ret; | |
2531 | uid_t uid = 0; | |
2532 | gid_t gid = 0; | |
2533 | const char *PAM_user = NULL; | |
2534 | ||
2535 | ret = pam_get_user(pamh, &PAM_user, NULL); | |
2536 | if (ret != PAM_SUCCESS) { | |
2537 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); | |
2538 | return PAM_SESSION_ERR; | |
2539 | } | |
2540 | ||
2541 | if (!get_uid_gid(PAM_user, &uid, &gid)) { | |
2542 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s.\n", PAM_user, NULL); | |
2543 | return PAM_SESSION_ERR; | |
2544 | } | |
2545 | ||
2546 | if (!cg_init(uid, gid)) { | |
2547 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2548 | return PAM_SESSION_ERR; | |
2549 | } | |
2550 | ||
2551 | /* Try to prune cgroups, that are actually empty but were still marked | |
2552 | * as busy by the kernel so we couldn't remove them on session close. | |
2553 | */ | |
2554 | cg_prune_empty_cgroups(PAM_user); | |
2555 | ||
2556 | if (cg_mount_mode == CGROUP_UNKNOWN) | |
2557 | return PAM_SESSION_ERR; | |
2558 | ||
2559 | if (argc > 1 && !strcmp(argv[0], "-c")) { | |
2560 | char **clist = make_string_list(argv[1], ","); | |
2561 | ||
2562 | /* | |
2563 | * We don't allow using "all" and other controllers explicitly because | |
2564 | * that simply doesn't make any sense. | |
2565 | */ | |
2566 | if (string_list_length(clist) > 1 && string_in_list(clist, "all")) { | |
2567 | mysyslog(LOG_ERR, "Invalid -c option, cannot specify individual controllers alongside 'all'.\n", NULL); | |
2568 | free_string_list(clist); | |
2569 | return PAM_SESSION_ERR; | |
2570 | } | |
2571 | ||
2572 | cg_mark_to_make_rw(clist); | |
2573 | free_string_list(clist); | |
2574 | } | |
2575 | ||
2576 | return handle_login(PAM_user, uid, gid); | |
2577 | } | |
2578 | ||
2579 | int pam_sm_close_session(pam_handle_t *pamh, int flags, int argc, | |
2580 | const char **argv) | |
2581 | { | |
2582 | int ret; | |
2583 | uid_t uid = 0; | |
2584 | gid_t gid = 0; | |
2585 | const char *PAM_user = NULL; | |
2586 | ||
2587 | ret = pam_get_user(pamh, &PAM_user, NULL); | |
2588 | if (ret != PAM_SUCCESS) { | |
2589 | mysyslog(LOG_ERR, "PAM-CGFS: couldn't get user\n", NULL); | |
2590 | return PAM_SESSION_ERR; | |
2591 | } | |
2592 | ||
2593 | if (!get_uid_gid(PAM_user, &uid, &gid)) { | |
2594 | mysyslog(LOG_ERR, "Failed to get uid and gid for %s.\n", PAM_user, NULL); | |
2595 | return PAM_SESSION_ERR; | |
2596 | } | |
2597 | ||
2598 | if (cg_mount_mode == CGROUP_UNINITIALIZED) { | |
2599 | if (!cg_init(uid, gid)) | |
2600 | mysyslog(LOG_ERR, "Failed to get list of controllers\n", NULL); | |
2601 | ||
2602 | if (argc > 1 && !strcmp(argv[0], "-c")) { | |
2603 | char **clist = make_string_list(argv[1], ","); | |
2604 | ||
2605 | /* | |
2606 | * We don't allow using "all" and other controllers explicitly because | |
2607 | * that simply doesn't make any sense. | |
2608 | */ | |
2609 | if (string_list_length(clist) > 1 && string_in_list(clist, "all")) { | |
2610 | mysyslog(LOG_ERR, "Invalid -c option, cannot specify individual controllers alongside 'all'.\n", NULL); | |
2611 | free_string_list(clist); | |
2612 | return PAM_SESSION_ERR; | |
2613 | } | |
2614 | ||
2615 | cg_mark_to_make_rw(clist); | |
2616 | free_string_list(clist); | |
2617 | } | |
2618 | } | |
2619 | ||
2620 | cg_prune_empty_cgroups(PAM_user); | |
2621 | cg_exit(); | |
2622 | ||
2623 | return PAM_SUCCESS; | |
2624 | } |