]>
Commit | Line | Data |
---|---|---|
237e200e SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014-2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * See COPYING file for details. | |
7 | */ | |
8 | ||
9 | #define FUSE_USE_VERSION 26 | |
10 | ||
237e200e | 11 | #include <dirent.h> |
29a73c2f | 12 | #include <errno.h> |
237e200e SH |
13 | #include <fcntl.h> |
14 | #include <fuse.h> | |
237e200e | 15 | #include <libgen.h> |
237e200e | 16 | #include <pthread.h> |
29a73c2f CB |
17 | #include <sched.h> |
18 | #include <stdbool.h> | |
19 | #include <stdio.h> | |
20 | #include <stdlib.h> | |
21 | #include <string.h> | |
22 | #include <time.h> | |
23 | #include <unistd.h> | |
24 | #include <wait.h> | |
237e200e | 25 | #include <linux/sched.h> |
29a73c2f CB |
26 | #include <sys/epoll.h> |
27 | #include <sys/mman.h> | |
28 | #include <sys/mount.h> | |
237e200e SH |
29 | #include <sys/param.h> |
30 | #include <sys/socket.h> | |
29a73c2f | 31 | #include <sys/syscall.h> |
237e200e | 32 | |
237e200e | 33 | #include "bindings.h" |
237e200e SH |
34 | #include "config.h" // for VERSION |
35 | ||
29a73c2f CB |
36 | /* Define pivot_root() if missing from the C library */ |
37 | #ifndef HAVE_PIVOT_ROOT | |
38 | static int pivot_root(const char * new_root, const char * put_old) | |
39 | { | |
40 | #ifdef __NR_pivot_root | |
41 | return syscall(__NR_pivot_root, new_root, put_old); | |
42 | #else | |
43 | errno = ENOSYS; | |
44 | return -1; | |
45 | #endif | |
46 | } | |
47 | #else | |
48 | extern int pivot_root(const char * new_root, const char * put_old); | |
49 | #endif | |
50 | ||
7dd6560a CB |
51 | #ifdef DEBUG |
52 | #define lxcfs_debug(format, ...) \ | |
53 | do { \ | |
54 | fprintf(stderr, "%s: %d: %s: " format, __FILE__, __LINE__, \ | |
55 | __func__, __VA_ARGS__); \ | |
56 | } while (false) | |
57 | #else | |
58 | #define lxcfs_debug(format, ...) | |
59 | #endif /* DEBUG */ | |
60 | ||
237e200e SH |
61 | enum { |
62 | LXC_TYPE_CGDIR, | |
63 | LXC_TYPE_CGFILE, | |
64 | LXC_TYPE_PROC_MEMINFO, | |
65 | LXC_TYPE_PROC_CPUINFO, | |
66 | LXC_TYPE_PROC_UPTIME, | |
67 | LXC_TYPE_PROC_STAT, | |
68 | LXC_TYPE_PROC_DISKSTATS, | |
70dcc12e | 69 | LXC_TYPE_PROC_SWAPS, |
237e200e SH |
70 | }; |
71 | ||
72 | struct file_info { | |
73 | char *controller; | |
74 | char *cgroup; | |
75 | char *file; | |
76 | int type; | |
77 | char *buf; // unused as of yet | |
78 | int buflen; | |
79 | int size; //actual data size | |
80 | int cached; | |
81 | }; | |
82 | ||
83 | /* reserve buffer size, for cpuall in /proc/stat */ | |
84 | #define BUF_RESERVE_SIZE 256 | |
85 | ||
86 | /* | |
87 | * A table caching which pid is init for a pid namespace. | |
88 | * When looking up which pid is init for $qpid, we first | |
89 | * 1. Stat /proc/$qpid/ns/pid. | |
90 | * 2. Check whether the ino_t is in our store. | |
91 | * a. if not, fork a child in qpid's ns to send us | |
92 | * ucred.pid = 1, and read the initpid. Cache | |
93 | * initpid and creation time for /proc/initpid | |
94 | * in a new store entry. | |
95 | * b. if so, verify that /proc/initpid still matches | |
96 | * what we have saved. If not, clear the store | |
97 | * entry and go back to a. If so, return the | |
98 | * cached initpid. | |
99 | */ | |
100 | struct pidns_init_store { | |
101 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
102 | pid_t initpid; // the pid of nit in that ns | |
103 | long int ctime; // the time at which /proc/$initpid was created | |
104 | struct pidns_init_store *next; | |
105 | long int lastcheck; | |
106 | }; | |
107 | ||
108 | /* lol - look at how they are allocated in the kernel */ | |
109 | #define PIDNS_HASH_SIZE 4096 | |
110 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
111 | ||
112 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
113 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
114 | static void lock_mutex(pthread_mutex_t *l) | |
115 | { | |
116 | int ret; | |
117 | ||
118 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
119 | fprintf(stderr, "pthread_mutex_lock returned:%d %s\n", ret, strerror(ret)); | |
120 | exit(1); | |
121 | } | |
122 | } | |
123 | ||
29a73c2f CB |
124 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
125 | * Number of hierarchies mounted. */ | |
126 | static int num_hierarchies; | |
127 | ||
128 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
129 | * Hierachies mounted {cpuset, blkio, ...}: | |
130 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
131 | static char **hierarchies; | |
132 | ||
133 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
134 | * Open file descriptors: | |
135 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
136 | * private mount namespace. | |
137 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
138 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
139 | * mounts and respective files in the private namespace even when located in | |
140 | * another namespace using the *at() family of functions | |
141 | * {openat(), fchownat(), ...}. */ | |
142 | static int *fd_hierarchies; | |
143 | ||
237e200e SH |
144 | static void unlock_mutex(pthread_mutex_t *l) |
145 | { | |
146 | int ret; | |
147 | ||
148 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
149 | fprintf(stderr, "pthread_mutex_unlock returned:%d %s\n", ret, strerror(ret)); | |
150 | exit(1); | |
151 | } | |
152 | } | |
153 | ||
154 | static void store_lock(void) | |
155 | { | |
156 | lock_mutex(&pidns_store_mutex); | |
157 | } | |
158 | ||
159 | static void store_unlock(void) | |
160 | { | |
161 | unlock_mutex(&pidns_store_mutex); | |
162 | } | |
163 | ||
164 | /* Must be called under store_lock */ | |
165 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
166 | { | |
167 | struct stat initsb; | |
168 | char fnam[100]; | |
169 | ||
170 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
171 | if (stat(fnam, &initsb) < 0) | |
172 | return false; | |
7dd6560a CB |
173 | |
174 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
175 | initsb.st_ctime, e->initpid); | |
176 | ||
237e200e SH |
177 | if (e->ctime != initsb.st_ctime) |
178 | return false; | |
179 | return true; | |
180 | } | |
181 | ||
182 | /* Must be called under store_lock */ | |
183 | static void remove_initpid(struct pidns_init_store *e) | |
184 | { | |
185 | struct pidns_init_store *tmp; | |
186 | int h; | |
187 | ||
7dd6560a CB |
188 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
189 | ||
237e200e SH |
190 | h = HASH(e->ino); |
191 | if (pidns_hash_table[h] == e) { | |
192 | pidns_hash_table[h] = e->next; | |
193 | free(e); | |
194 | return; | |
195 | } | |
196 | ||
197 | tmp = pidns_hash_table[h]; | |
198 | while (tmp) { | |
199 | if (tmp->next == e) { | |
200 | tmp->next = e->next; | |
201 | free(e); | |
202 | return; | |
203 | } | |
204 | tmp = tmp->next; | |
205 | } | |
206 | } | |
207 | ||
208 | #define PURGE_SECS 5 | |
209 | /* Must be called under store_lock */ | |
210 | static void prune_initpid_store(void) | |
211 | { | |
212 | static long int last_prune = 0; | |
213 | struct pidns_init_store *e, *prev, *delme; | |
214 | long int now, threshold; | |
215 | int i; | |
216 | ||
217 | if (!last_prune) { | |
218 | last_prune = time(NULL); | |
219 | return; | |
220 | } | |
221 | now = time(NULL); | |
222 | if (now < last_prune + PURGE_SECS) | |
223 | return; | |
7dd6560a CB |
224 | |
225 | lxcfs_debug("%s\n", "Pruning."); | |
226 | ||
237e200e SH |
227 | last_prune = now; |
228 | threshold = now - 2 * PURGE_SECS; | |
229 | ||
230 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
231 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
232 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
233 | |
234 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
235 | ||
237e200e SH |
236 | delme = e; |
237 | if (prev) | |
238 | prev->next = e->next; | |
239 | else | |
240 | pidns_hash_table[i] = e->next; | |
241 | e = e->next; | |
242 | free(delme); | |
243 | } else { | |
244 | prev = e; | |
245 | e = e->next; | |
246 | } | |
247 | } | |
248 | } | |
249 | } | |
250 | ||
251 | /* Must be called under store_lock */ | |
252 | static void save_initpid(struct stat *sb, pid_t pid) | |
253 | { | |
254 | struct pidns_init_store *e; | |
255 | char fpath[100]; | |
256 | struct stat procsb; | |
257 | int h; | |
258 | ||
7dd6560a CB |
259 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
260 | ||
237e200e SH |
261 | snprintf(fpath, 100, "/proc/%d", pid); |
262 | if (stat(fpath, &procsb) < 0) | |
263 | return; | |
264 | do { | |
265 | e = malloc(sizeof(*e)); | |
266 | } while (!e); | |
267 | e->ino = sb->st_ino; | |
268 | e->initpid = pid; | |
269 | e->ctime = procsb.st_ctime; | |
270 | h = HASH(e->ino); | |
271 | e->next = pidns_hash_table[h]; | |
272 | e->lastcheck = time(NULL); | |
273 | pidns_hash_table[h] = e; | |
274 | } | |
275 | ||
276 | /* | |
277 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
278 | * entry for the inode number and creation time. Verify that the init pid | |
279 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
280 | * otherwise. | |
281 | * Must be called under store_lock | |
282 | */ | |
283 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
284 | { | |
285 | int h = HASH(sb->st_ino); | |
286 | struct pidns_init_store *e = pidns_hash_table[h]; | |
287 | ||
288 | while (e) { | |
289 | if (e->ino == sb->st_ino) { | |
290 | if (initpid_still_valid(e, sb)) { | |
291 | e->lastcheck = time(NULL); | |
292 | return e; | |
293 | } | |
294 | remove_initpid(e); | |
295 | return NULL; | |
296 | } | |
297 | e = e->next; | |
298 | } | |
299 | ||
300 | return NULL; | |
301 | } | |
302 | ||
0f657ce3 | 303 | static int is_dir(const char *path, int fd) |
237e200e SH |
304 | { |
305 | struct stat statbuf; | |
0f657ce3 | 306 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
307 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
308 | return 1; | |
309 | return 0; | |
310 | } | |
311 | ||
312 | static char *must_copy_string(const char *str) | |
313 | { | |
314 | char *dup = NULL; | |
315 | if (!str) | |
316 | return NULL; | |
317 | do { | |
318 | dup = strdup(str); | |
319 | } while (!dup); | |
320 | ||
321 | return dup; | |
322 | } | |
323 | ||
324 | static inline void drop_trailing_newlines(char *s) | |
325 | { | |
326 | int l; | |
327 | ||
328 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
329 | s[l-1] = '\0'; | |
330 | } | |
331 | ||
332 | #define BATCH_SIZE 50 | |
333 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
334 | { | |
335 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
336 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
337 | ||
338 | if (!*mem || newbatches > oldbatches) { | |
339 | char *tmp; | |
340 | do { | |
341 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
342 | } while (!tmp); | |
343 | *mem = tmp; | |
344 | } | |
345 | } | |
346 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
347 | { | |
348 | size_t newlen = *len + linelen; | |
349 | dorealloc(contents, *len, newlen + 1); | |
350 | memcpy(*contents + *len, line, linelen+1); | |
351 | *len = newlen; | |
352 | } | |
353 | ||
60f2ae53 | 354 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
355 | { |
356 | char *line = NULL; | |
357 | char *contents = NULL; | |
60f2ae53 | 358 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
359 | size_t len = 0, fulllen = 0; |
360 | ssize_t linelen; | |
361 | ||
362 | if (!f) | |
363 | return NULL; | |
364 | ||
365 | while ((linelen = getline(&line, &len, f)) != -1) { | |
366 | append_line(&contents, &fulllen, line, linelen); | |
367 | } | |
368 | fclose(f); | |
369 | ||
370 | if (contents) | |
371 | drop_trailing_newlines(contents); | |
372 | free(line); | |
373 | return contents; | |
374 | } | |
375 | ||
ba59ea09 | 376 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
377 | { |
378 | FILE *f; | |
379 | size_t len, ret; | |
380 | ||
ba59ea09 | 381 | if (!(f = fdopen(fd, "w"))) |
237e200e SH |
382 | return false; |
383 | len = strlen(string); | |
384 | ret = fwrite(string, 1, len, f); | |
385 | if (ret != len) { | |
386 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
387 | fclose(f); | |
388 | return false; | |
389 | } | |
390 | if (fclose(f) < 0) { | |
391 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
392 | return false; | |
393 | } | |
394 | return true; | |
395 | } | |
396 | ||
237e200e SH |
397 | struct cgfs_files { |
398 | char *name; | |
399 | uint32_t uid, gid; | |
400 | uint32_t mode; | |
401 | }; | |
402 | ||
0619767c | 403 | #define ALLOC_NUM 20 |
237e200e SH |
404 | static bool store_hierarchy(char *stridx, char *h) |
405 | { | |
0619767c SH |
406 | if (num_hierarchies % ALLOC_NUM == 0) { |
407 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
408 | n *= ALLOC_NUM; | |
409 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c SH |
410 | if (!tmp) { |
411 | fprintf(stderr, "Out of memory\n"); | |
412 | exit(1); | |
413 | } | |
237e200e | 414 | hierarchies = tmp; |
237e200e | 415 | } |
f676eb79 | 416 | |
0619767c | 417 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
418 | return true; |
419 | } | |
420 | ||
421 | static void print_subsystems(void) | |
422 | { | |
423 | int i; | |
424 | ||
cc97d34c | 425 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
426 | for (i = 0; i < num_hierarchies; i++) { |
427 | if (hierarchies[i]) | |
428 | fprintf(stderr, " %d: %s\n", i, hierarchies[i]); | |
429 | } | |
430 | } | |
431 | ||
432 | static bool in_comma_list(const char *needle, const char *haystack) | |
433 | { | |
434 | const char *s = haystack, *e; | |
435 | size_t nlen = strlen(needle); | |
436 | ||
06081b29 | 437 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
438 | if (nlen != e - s) { |
439 | s = e + 1; | |
440 | continue; | |
441 | } | |
442 | if (strncmp(needle, s, nlen) == 0) | |
443 | return true; | |
444 | s = e + 1; | |
445 | } | |
446 | if (strcmp(needle, s) == 0) | |
447 | return true; | |
448 | return false; | |
449 | } | |
450 | ||
451 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
452 | /* Return the mounted controller and store the corresponding open file descriptor |
453 | * referring to the controller mountpoint in the private lxcfs namespace in | |
454 | * @cfd. | |
455 | */ | |
456 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
457 | { |
458 | int i; | |
459 | ||
460 | for (i = 0; i < num_hierarchies; i++) { | |
461 | if (!hierarchies[i]) | |
462 | continue; | |
5dd3e6fd CB |
463 | if (strcmp(hierarchies[i], controller) == 0) { |
464 | *cfd = fd_hierarchies[i]; | |
237e200e | 465 | return hierarchies[i]; |
5dd3e6fd CB |
466 | } |
467 | if (in_comma_list(controller, hierarchies[i])) { | |
468 | *cfd = fd_hierarchies[i]; | |
237e200e | 469 | return hierarchies[i]; |
5dd3e6fd | 470 | } |
237e200e SH |
471 | } |
472 | ||
473 | return NULL; | |
474 | } | |
475 | ||
476 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
477 | const char *value) | |
478 | { | |
ba59ea09 | 479 | int ret, fd, cfd; |
237e200e | 480 | size_t len; |
f5a6d92e | 481 | char *fnam, *tmpc; |
237e200e | 482 | |
f5a6d92e | 483 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
484 | if (!tmpc) |
485 | return false; | |
f5a6d92e CB |
486 | |
487 | /* Make sure we pass a relative path to *at() family of functions. | |
488 | * . + /cgroup + / + file + \0 | |
489 | */ | |
ba59ea09 | 490 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 491 | fnam = alloca(len); |
ba59ea09 CB |
492 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
493 | if (ret < 0 || (size_t)ret >= len) | |
494 | return false; | |
495 | ||
496 | fd = openat(cfd, fnam, O_WRONLY); | |
497 | if (fd < 0) | |
498 | return false; | |
f676eb79 | 499 | |
ba59ea09 | 500 | return write_string(fnam, value, fd); |
237e200e SH |
501 | } |
502 | ||
503 | // Chown all the files in the cgroup directory. We do this when we create | |
504 | // a cgroup on behalf of a user. | |
f23fe717 | 505 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 506 | { |
f23fe717 | 507 | struct dirent *direntp; |
237e200e SH |
508 | char path[MAXPATHLEN]; |
509 | size_t len; | |
510 | DIR *d; | |
f23fe717 | 511 | int fd1, ret; |
237e200e SH |
512 | |
513 | len = strlen(dirname); | |
514 | if (len >= MAXPATHLEN) { | |
515 | fprintf(stderr, "chown_all_cgroup_files: pathname too long: %s\n", dirname); | |
516 | return; | |
517 | } | |
518 | ||
f23fe717 CB |
519 | fd1 = openat(fd, dirname, O_DIRECTORY); |
520 | if (fd1 < 0) | |
521 | return; | |
522 | ||
523 | d = fdopendir(fd1); | |
237e200e SH |
524 | if (!d) { |
525 | fprintf(stderr, "chown_all_cgroup_files: failed to open %s\n", dirname); | |
526 | return; | |
527 | } | |
528 | ||
f23fe717 | 529 | while ((direntp = readdir(d))) { |
237e200e SH |
530 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
531 | continue; | |
532 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
533 | if (ret < 0 || ret >= MAXPATHLEN) { | |
534 | fprintf(stderr, "chown_all_cgroup_files: pathname too long under %s\n", dirname); | |
535 | continue; | |
536 | } | |
f23fe717 | 537 | if (fchownat(fd, path, uid, gid, 0) < 0) |
237e200e SH |
538 | fprintf(stderr, "Failed to chown file %s to %u:%u", path, uid, gid); |
539 | } | |
540 | closedir(d); | |
541 | } | |
542 | ||
543 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
544 | { | |
5dd3e6fd | 545 | int cfd; |
237e200e | 546 | size_t len; |
f5a6d92e | 547 | char *dirnam, *tmpc; |
237e200e | 548 | |
f5a6d92e | 549 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
550 | if (!tmpc) |
551 | return -EINVAL; | |
f5a6d92e CB |
552 | |
553 | /* Make sure we pass a relative path to *at() family of functions. | |
554 | * . + /cg + \0 | |
555 | */ | |
f23fe717 | 556 | len = strlen(cg) + 2; |
237e200e | 557 | dirnam = alloca(len); |
f23fe717 | 558 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 559 | |
f23fe717 | 560 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
561 | return -errno; |
562 | ||
563 | if (uid == 0 && gid == 0) | |
564 | return 0; | |
565 | ||
f23fe717 | 566 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
567 | return -errno; |
568 | ||
f23fe717 | 569 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
570 | |
571 | return 0; | |
572 | } | |
573 | ||
2e81a5e3 | 574 | static bool recursive_rmdir(const char *dirname, int fd, int cfd) |
237e200e | 575 | { |
b7672ded | 576 | struct dirent *direntp; |
237e200e SH |
577 | DIR *dir; |
578 | bool ret = false; | |
579 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
580 | int dupfd; |
581 | ||
582 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
583 | if (dupfd < 0) | |
584 | return false; | |
237e200e | 585 | |
b7672ded | 586 | dir = fdopendir(dupfd); |
237e200e | 587 | if (!dir) { |
7dd6560a | 588 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
589 | return false; |
590 | } | |
591 | ||
b7672ded | 592 | while ((direntp = readdir(dir))) { |
237e200e SH |
593 | struct stat mystat; |
594 | int rc; | |
595 | ||
237e200e SH |
596 | if (!strcmp(direntp->d_name, ".") || |
597 | !strcmp(direntp->d_name, "..")) | |
598 | continue; | |
599 | ||
600 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
601 | if (rc < 0 || rc >= MAXPATHLEN) { | |
602 | fprintf(stderr, "pathname too long\n"); | |
603 | continue; | |
604 | } | |
605 | ||
2e81a5e3 CB |
606 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
607 | if (rc) { | |
7dd6560a | 608 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
609 | continue; |
610 | } | |
7dd6560a | 611 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 612 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 613 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
614 | } |
615 | ||
616 | ret = true; | |
617 | if (closedir(dir) < 0) { | |
618 | fprintf(stderr, "%s: failed to close directory %s: %s\n", __func__, dirname, strerror(errno)); | |
619 | ret = false; | |
620 | } | |
621 | ||
2e81a5e3 | 622 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 623 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
624 | ret = false; |
625 | } | |
b7672ded | 626 | close(fd); |
237e200e SH |
627 | |
628 | return ret; | |
629 | } | |
630 | ||
631 | bool cgfs_remove(const char *controller, const char *cg) | |
632 | { | |
b7672ded | 633 | int fd, cfd; |
237e200e | 634 | size_t len; |
f5a6d92e | 635 | char *dirnam, *tmpc; |
237e200e | 636 | |
f5a6d92e | 637 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
638 | if (!tmpc) |
639 | return false; | |
f5a6d92e CB |
640 | |
641 | /* Make sure we pass a relative path to *at() family of functions. | |
642 | * . + /cg + \0 | |
643 | */ | |
b7672ded | 644 | len = strlen(cg) + 2; |
237e200e | 645 | dirnam = alloca(len); |
b7672ded CB |
646 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
647 | ||
648 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
649 | if (fd < 0) | |
650 | return false; | |
651 | ||
2e81a5e3 | 652 | return recursive_rmdir(dirnam, fd, cfd); |
237e200e SH |
653 | } |
654 | ||
655 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
656 | { | |
5dd3e6fd | 657 | int cfd; |
237e200e | 658 | size_t len; |
f5a6d92e | 659 | char *pathname, *tmpc; |
237e200e | 660 | |
f5a6d92e | 661 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
662 | if (!tmpc) |
663 | return false; | |
f5a6d92e CB |
664 | |
665 | /* Make sure we pass a relative path to *at() family of functions. | |
666 | * . + /file + \0 | |
667 | */ | |
534690b4 | 668 | len = strlen(file) + 2; |
237e200e | 669 | pathname = alloca(len); |
534690b4 CB |
670 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
671 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
672 | return false; |
673 | return true; | |
674 | } | |
675 | ||
0f657ce3 | 676 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
677 | { |
678 | size_t len; | |
679 | char *fname; | |
680 | ||
681 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
682 | fname = alloca(len); | |
683 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 684 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
685 | return -errno; |
686 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 687 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
688 | return -errno; |
689 | return 0; | |
690 | } | |
691 | ||
692 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
693 | { | |
5dd3e6fd | 694 | int cfd; |
237e200e | 695 | size_t len; |
f5a6d92e | 696 | char *pathname, *tmpc; |
237e200e | 697 | |
f5a6d92e | 698 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
699 | if (!tmpc) |
700 | return -EINVAL; | |
f5a6d92e CB |
701 | |
702 | /* Make sure we pass a relative path to *at() family of functions. | |
703 | * . + /file + \0 | |
704 | */ | |
0f657ce3 | 705 | len = strlen(file) + 2; |
237e200e | 706 | pathname = alloca(len); |
0f657ce3 CB |
707 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
708 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
709 | return -errno; |
710 | ||
0f657ce3 | 711 | if (is_dir(pathname, cfd)) |
237e200e | 712 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 713 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
714 | |
715 | return 0; | |
716 | } | |
717 | ||
718 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
719 | { | |
3ffd08ee | 720 | int fd, cfd; |
237e200e | 721 | size_t len; |
f5a6d92e | 722 | char *pathname, *tmpc; |
237e200e | 723 | |
f5a6d92e | 724 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
725 | if (!tmpc) |
726 | return NULL; | |
f5a6d92e CB |
727 | |
728 | /* Make sure we pass a relative path to *at() family of functions. | |
729 | * . + /cgroup + / "cgroup.procs" + \0 | |
730 | */ | |
3ffd08ee | 731 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 732 | pathname = alloca(len); |
3ffd08ee CB |
733 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
734 | ||
735 | fd = openat(cfd, pathname, O_WRONLY); | |
736 | if (fd < 0) | |
737 | return NULL; | |
738 | ||
739 | return fdopen(fd, "w"); | |
237e200e SH |
740 | } |
741 | ||
f366da65 WB |
742 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
743 | void ***list, size_t typesize, | |
744 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 745 | { |
4ea38a4c | 746 | int cfd, fd, ret; |
237e200e | 747 | size_t len; |
4ea38a4c | 748 | char *cg, *tmpc; |
237e200e | 749 | char pathname[MAXPATHLEN]; |
f366da65 | 750 | size_t sz = 0, asz = 0; |
4ea38a4c | 751 | struct dirent *dirent; |
237e200e | 752 | DIR *dir; |
237e200e | 753 | |
4ea38a4c | 754 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 755 | *list = NULL; |
237e200e | 756 | if (!tmpc) |
e97c834b | 757 | return false; |
237e200e | 758 | |
f5a6d92e | 759 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
760 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
761 | cg = alloca(len); | |
762 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
763 | if (ret < 0 || (size_t)ret >= len) { | |
764 | fprintf(stderr, "%s: pathname too long under %s\n", __func__, cgroup); | |
765 | return false; | |
766 | } | |
237e200e | 767 | |
4ea38a4c CB |
768 | fd = openat(cfd, cg, O_DIRECTORY); |
769 | if (fd < 0) | |
770 | return false; | |
771 | ||
772 | dir = fdopendir(fd); | |
237e200e SH |
773 | if (!dir) |
774 | return false; | |
775 | ||
4ea38a4c | 776 | while ((dirent = readdir(dir))) { |
237e200e | 777 | struct stat mystat; |
237e200e | 778 | |
4ea38a4c CB |
779 | if (!strcmp(dirent->d_name, ".") || |
780 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
781 | continue; |
782 | ||
4ea38a4c CB |
783 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
784 | if (ret < 0 || ret >= MAXPATHLEN) { | |
785 | fprintf(stderr, "%s: pathname too long under %s\n", __func__, cg); | |
237e200e SH |
786 | continue; |
787 | } | |
788 | ||
4ea38a4c | 789 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e SH |
790 | if (ret) { |
791 | fprintf(stderr, "%s: failed to stat %s: %s\n", __func__, pathname, strerror(errno)); | |
792 | continue; | |
793 | } | |
f366da65 WB |
794 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
795 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
796 | continue; |
797 | ||
798 | if (sz+2 >= asz) { | |
f366da65 | 799 | void **tmp; |
237e200e SH |
800 | asz += BATCH_SIZE; |
801 | do { | |
f366da65 | 802 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
803 | } while (!tmp); |
804 | *list = tmp; | |
805 | } | |
4ea38a4c | 806 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
807 | (*list)[sz+1] = NULL; |
808 | sz++; | |
809 | } | |
810 | if (closedir(dir) < 0) { | |
4ea38a4c | 811 | fprintf(stderr, "%s: failed closedir for %s: %s\n", __func__, cgroup, strerror(errno)); |
237e200e SH |
812 | return false; |
813 | } | |
814 | return true; | |
815 | } | |
816 | ||
f366da65 WB |
817 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
818 | { | |
819 | char *dup; | |
820 | do { | |
821 | dup = strdup(dir_entry); | |
822 | } while (!dup); | |
823 | return dup; | |
824 | } | |
825 | ||
826 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
827 | { | |
828 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
829 | } | |
830 | ||
237e200e SH |
831 | void free_key(struct cgfs_files *k) |
832 | { | |
833 | if (!k) | |
834 | return; | |
835 | free(k->name); | |
836 | free(k); | |
837 | } | |
838 | ||
839 | void free_keys(struct cgfs_files **keys) | |
840 | { | |
841 | int i; | |
842 | ||
843 | if (!keys) | |
844 | return; | |
845 | for (i = 0; keys[i]; i++) { | |
846 | free_key(keys[i]); | |
847 | } | |
848 | free(keys); | |
849 | } | |
850 | ||
851 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
852 | { | |
60f2ae53 | 853 | int ret, fd, cfd; |
237e200e | 854 | size_t len; |
f5a6d92e | 855 | char *fnam, *tmpc; |
237e200e | 856 | |
f5a6d92e | 857 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
858 | if (!tmpc) |
859 | return false; | |
f5a6d92e CB |
860 | |
861 | /* Make sure we pass a relative path to *at() family of functions. | |
862 | * . + /cgroup + / + file + \0 | |
863 | */ | |
60f2ae53 | 864 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 865 | fnam = alloca(len); |
60f2ae53 CB |
866 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
867 | if (ret < 0 || (size_t)ret >= len) | |
868 | return NULL; | |
869 | ||
870 | fd = openat(cfd, fnam, O_RDONLY); | |
871 | if (fd < 0) | |
872 | return NULL; | |
237e200e | 873 | |
60f2ae53 | 874 | *value = slurp_file(fnam, fd); |
237e200e SH |
875 | return *value != NULL; |
876 | } | |
877 | ||
878 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
879 | { | |
4ea38a4c | 880 | int ret, cfd; |
237e200e | 881 | size_t len; |
f5a6d92e | 882 | char *fnam, *tmpc; |
237e200e SH |
883 | struct stat sb; |
884 | struct cgfs_files *newkey; | |
237e200e | 885 | |
f5a6d92e | 886 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
887 | if (!tmpc) |
888 | return false; | |
889 | ||
890 | if (file && *file == '/') | |
891 | file++; | |
892 | ||
06081b29 | 893 | if (file && strchr(file, '/')) |
237e200e SH |
894 | return NULL; |
895 | ||
f5a6d92e CB |
896 | /* Make sure we pass a relative path to *at() family of functions. |
897 | * . + /cgroup + / + file + \0 | |
898 | */ | |
4ea38a4c | 899 | len = strlen(cgroup) + 3; |
237e200e SH |
900 | if (file) |
901 | len += strlen(file) + 1; | |
902 | fnam = alloca(len); | |
4ea38a4c CB |
903 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
904 | file ? "/" : "", file ? file : ""); | |
237e200e | 905 | |
4ea38a4c | 906 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
907 | if (ret < 0) |
908 | return NULL; | |
909 | ||
910 | do { | |
911 | newkey = malloc(sizeof(struct cgfs_files)); | |
912 | } while (!newkey); | |
913 | if (file) | |
914 | newkey->name = must_copy_string(file); | |
06081b29 CB |
915 | else if (strrchr(cgroup, '/')) |
916 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
917 | else |
918 | newkey->name = must_copy_string(cgroup); | |
919 | newkey->uid = sb.st_uid; | |
920 | newkey->gid = sb.st_gid; | |
921 | newkey->mode = sb.st_mode; | |
922 | ||
923 | return newkey; | |
924 | } | |
925 | ||
f366da65 | 926 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 927 | { |
f366da65 WB |
928 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
929 | if (!entry) { | |
930 | fprintf(stderr, "%s: Error getting files under %s:%s\n", | |
931 | __func__, controller, cgroup); | |
237e200e | 932 | } |
f366da65 WB |
933 | return entry; |
934 | } | |
935 | ||
936 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
937 | { | |
938 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
939 | } |
940 | ||
941 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
942 | { |
943 | int cfd; | |
944 | size_t len; | |
f5a6d92e | 945 | char *fnam, *tmpc; |
237e200e SH |
946 | int ret; |
947 | struct stat sb; | |
948 | ||
f5a6d92e | 949 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
950 | if (!tmpc) |
951 | return false; | |
f5a6d92e CB |
952 | |
953 | /* Make sure we pass a relative path to *at() family of functions. | |
954 | * . + /cgroup + / + f + \0 | |
955 | */ | |
d04232f2 | 956 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 957 | fnam = alloca(len); |
d04232f2 CB |
958 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
959 | if (ret < 0 || (size_t)ret >= len) | |
960 | return false; | |
237e200e | 961 | |
d04232f2 | 962 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
963 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
964 | return false; | |
f5a6d92e | 965 | |
237e200e SH |
966 | return true; |
967 | } | |
968 | ||
969 | #define SEND_CREDS_OK 0 | |
970 | #define SEND_CREDS_NOTSK 1 | |
971 | #define SEND_CREDS_FAIL 2 | |
972 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
973 | static int wait_for_pid(pid_t pid); | |
974 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 975 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
976 | |
977 | /* | |
b10bdd6c | 978 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
979 | * over a unix sock so we can read the task's reaper's pid in our |
980 | * namespace | |
b10bdd6c FG |
981 | * |
982 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
983 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
984 | * the pidns and the parent pid outside are identical. Using clone prevents | |
985 | * this issue. | |
237e200e SH |
986 | */ |
987 | static void write_task_init_pid_exit(int sock, pid_t target) | |
988 | { | |
237e200e SH |
989 | char fnam[100]; |
990 | pid_t pid; | |
237e200e | 991 | int fd, ret; |
b10bdd6c FG |
992 | size_t stack_size = sysconf(_SC_PAGESIZE); |
993 | void *stack = alloca(stack_size); | |
237e200e SH |
994 | |
995 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
996 | if (ret < 0 || ret >= sizeof(fnam)) | |
997 | _exit(1); | |
998 | ||
999 | fd = open(fnam, O_RDONLY); | |
1000 | if (fd < 0) { | |
1001 | perror("write_task_init_pid_exit open of ns/pid"); | |
1002 | _exit(1); | |
1003 | } | |
1004 | if (setns(fd, 0)) { | |
1005 | perror("write_task_init_pid_exit setns 1"); | |
1006 | close(fd); | |
1007 | _exit(1); | |
1008 | } | |
b10bdd6c | 1009 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1010 | if (pid < 0) |
1011 | _exit(1); | |
1012 | if (pid != 0) { | |
1013 | if (!wait_for_pid(pid)) | |
1014 | _exit(1); | |
1015 | _exit(0); | |
1016 | } | |
b10bdd6c FG |
1017 | } |
1018 | ||
1019 | static int send_creds_clone_wrapper(void *arg) { | |
1020 | struct ucred cred; | |
1021 | char v; | |
1022 | int sock = *(int *)arg; | |
237e200e SH |
1023 | |
1024 | /* we are the child */ | |
1025 | cred.uid = 0; | |
1026 | cred.gid = 0; | |
1027 | cred.pid = 1; | |
1028 | v = '1'; | |
1029 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1030 | return 1; |
1031 | return 0; | |
237e200e SH |
1032 | } |
1033 | ||
1034 | static pid_t get_init_pid_for_task(pid_t task) | |
1035 | { | |
1036 | int sock[2]; | |
1037 | pid_t pid; | |
1038 | pid_t ret = -1; | |
1039 | char v = '0'; | |
1040 | struct ucred cred; | |
1041 | ||
1042 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1043 | perror("socketpair"); | |
1044 | return -1; | |
1045 | } | |
1046 | ||
1047 | pid = fork(); | |
1048 | if (pid < 0) | |
1049 | goto out; | |
1050 | if (!pid) { | |
1051 | close(sock[1]); | |
1052 | write_task_init_pid_exit(sock[0], task); | |
1053 | _exit(0); | |
1054 | } | |
1055 | ||
1056 | if (!recv_creds(sock[1], &cred, &v)) | |
1057 | goto out; | |
1058 | ret = cred.pid; | |
1059 | ||
1060 | out: | |
1061 | close(sock[0]); | |
1062 | close(sock[1]); | |
1063 | if (pid > 0) | |
1064 | wait_for_pid(pid); | |
1065 | return ret; | |
1066 | } | |
1067 | ||
1068 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1069 | { | |
1070 | pid_t answer = 0; | |
1071 | struct stat sb; | |
1072 | struct pidns_init_store *e; | |
1073 | char fnam[100]; | |
1074 | ||
1075 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1076 | store_lock(); | |
1077 | if (stat(fnam, &sb) < 0) | |
1078 | goto out; | |
1079 | e = lookup_verify_initpid(&sb); | |
1080 | if (e) { | |
1081 | answer = e->initpid; | |
1082 | goto out; | |
1083 | } | |
1084 | answer = get_init_pid_for_task(qpid); | |
1085 | if (answer > 0) | |
1086 | save_initpid(&sb, answer); | |
1087 | ||
1088 | out: | |
1089 | /* we prune at end in case we are returning | |
1090 | * the value we were about to return */ | |
1091 | prune_initpid_store(); | |
1092 | store_unlock(); | |
1093 | return answer; | |
1094 | } | |
1095 | ||
1096 | static int wait_for_pid(pid_t pid) | |
1097 | { | |
1098 | int status, ret; | |
1099 | ||
1100 | if (pid <= 0) | |
1101 | return -1; | |
1102 | ||
1103 | again: | |
1104 | ret = waitpid(pid, &status, 0); | |
1105 | if (ret == -1) { | |
1106 | if (errno == EINTR) | |
1107 | goto again; | |
1108 | return -1; | |
1109 | } | |
1110 | if (ret != pid) | |
1111 | goto again; | |
1112 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1113 | return -1; | |
1114 | return 0; | |
1115 | } | |
1116 | ||
1117 | ||
1118 | /* | |
1119 | * append pid to *src. | |
1120 | * src: a pointer to a char* in which ot append the pid. | |
1121 | * sz: the number of characters printed so far, minus trailing \0. | |
1122 | * asz: the allocated size so far | |
1123 | * pid: the pid to append | |
1124 | */ | |
1125 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1126 | { | |
1127 | char tmp[30]; | |
1128 | ||
1129 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1130 | ||
1131 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1132 | char *tmp; | |
1133 | do { | |
1134 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1135 | } while (!tmp); | |
1136 | *src = tmp; | |
1137 | *asz += BUF_RESERVE_SIZE; | |
1138 | } | |
bbfd0e33 | 1139 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1140 | *sz += tmplen; |
237e200e SH |
1141 | } |
1142 | ||
1143 | /* | |
1144 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1145 | * valid in the caller's namespace, return the id mapped into | |
1146 | * pid's namespace. | |
1147 | * Returns the mapped id, or -1 on error. | |
1148 | */ | |
1149 | unsigned int | |
1150 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1151 | { | |
1152 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1153 | hostuid, // base id for a range in the caller's namespace | |
1154 | count; // number of ids in this range | |
1155 | char line[400]; | |
1156 | int ret; | |
1157 | ||
1158 | fseek(idfile, 0L, SEEK_SET); | |
1159 | while (fgets(line, 400, idfile)) { | |
1160 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1161 | if (ret != 3) | |
1162 | continue; | |
1163 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1164 | /* | |
1165 | * uids wrapped around - unexpected as this is a procfile, | |
1166 | * so just bail. | |
1167 | */ | |
1168 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s\n", | |
1169 | nsuid, hostuid, count, line); | |
1170 | return -1; | |
1171 | } | |
1172 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1173 | /* | |
1174 | * now since hostuid <= in_id < hostuid+count, and | |
1175 | * hostuid+count and nsuid+count do not wrap around, | |
1176 | * we know that nsuid+(in_id-hostuid) which must be | |
1177 | * less that nsuid+(count) must not wrap around | |
1178 | */ | |
1179 | return (in_id - hostuid) + nsuid; | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | // no answer found | |
1184 | return -1; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * for is_privileged_over, | |
1189 | * specify whether we require the calling uid to be root in his | |
1190 | * namespace | |
1191 | */ | |
1192 | #define NS_ROOT_REQD true | |
1193 | #define NS_ROOT_OPT false | |
1194 | ||
1195 | #define PROCLEN 100 | |
1196 | ||
1197 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1198 | { | |
1199 | char fpath[PROCLEN]; | |
1200 | int ret; | |
1201 | bool answer = false; | |
1202 | uid_t nsuid; | |
1203 | ||
1204 | if (victim == -1 || uid == -1) | |
1205 | return false; | |
1206 | ||
1207 | /* | |
1208 | * If the request is one not requiring root in the namespace, | |
1209 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1210 | * access to files owned by uid 1000 | |
1211 | */ | |
1212 | if (!req_ns_root && uid == victim) | |
1213 | return true; | |
1214 | ||
1215 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1216 | if (ret < 0 || ret >= PROCLEN) | |
1217 | return false; | |
1218 | FILE *f = fopen(fpath, "r"); | |
1219 | if (!f) | |
1220 | return false; | |
1221 | ||
1222 | /* if caller's not root in his namespace, reject */ | |
1223 | nsuid = convert_id_to_ns(f, uid); | |
1224 | if (nsuid) | |
1225 | goto out; | |
1226 | ||
1227 | /* | |
1228 | * If victim is not mapped into caller's ns, reject. | |
1229 | * XXX I'm not sure this check is needed given that fuse | |
1230 | * will be sending requests where the vfs has converted | |
1231 | */ | |
1232 | nsuid = convert_id_to_ns(f, victim); | |
1233 | if (nsuid == -1) | |
1234 | goto out; | |
1235 | ||
1236 | answer = true; | |
1237 | ||
1238 | out: | |
1239 | fclose(f); | |
1240 | return answer; | |
1241 | } | |
1242 | ||
1243 | static bool perms_include(int fmode, mode_t req_mode) | |
1244 | { | |
1245 | mode_t r; | |
1246 | ||
1247 | switch (req_mode & O_ACCMODE) { | |
1248 | case O_RDONLY: | |
1249 | r = S_IROTH; | |
1250 | break; | |
1251 | case O_WRONLY: | |
1252 | r = S_IWOTH; | |
1253 | break; | |
1254 | case O_RDWR: | |
1255 | r = S_IROTH | S_IWOTH; | |
1256 | break; | |
1257 | default: | |
1258 | return false; | |
1259 | } | |
1260 | return ((fmode & r) == r); | |
1261 | } | |
1262 | ||
1263 | ||
1264 | /* | |
1265 | * taskcg is a/b/c | |
1266 | * querycg is /a/b/c/d/e | |
1267 | * we return 'd' | |
1268 | */ | |
1269 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1270 | { | |
1271 | char *start, *end; | |
1272 | ||
1273 | if (strlen(taskcg) <= strlen(querycg)) { | |
1274 | fprintf(stderr, "%s: I was fed bad input\n", __func__); | |
1275 | return NULL; | |
1276 | } | |
1277 | ||
06081b29 | 1278 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1279 | start = strdup(taskcg + 1); |
1280 | else | |
1281 | start = strdup(taskcg + strlen(querycg) + 1); | |
1282 | if (!start) | |
1283 | return NULL; | |
1284 | end = strchr(start, '/'); | |
1285 | if (end) | |
1286 | *end = '\0'; | |
1287 | return start; | |
1288 | } | |
1289 | ||
1290 | static void stripnewline(char *x) | |
1291 | { | |
1292 | size_t l = strlen(x); | |
1293 | if (l && x[l-1] == '\n') | |
1294 | x[l-1] = '\0'; | |
1295 | } | |
1296 | ||
1297 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1298 | { | |
5dd3e6fd | 1299 | int cfd; |
237e200e SH |
1300 | char fnam[PROCLEN]; |
1301 | FILE *f; | |
1302 | char *answer = NULL; | |
1303 | char *line = NULL; | |
1304 | size_t len = 0; | |
1305 | int ret; | |
5dd3e6fd | 1306 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1307 | if (!h) |
1308 | return NULL; | |
1309 | ||
1310 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1311 | if (ret < 0 || ret >= PROCLEN) | |
1312 | return NULL; | |
1313 | if (!(f = fopen(fnam, "r"))) | |
1314 | return NULL; | |
1315 | ||
1316 | while (getline(&line, &len, f) != -1) { | |
1317 | char *c1, *c2; | |
1318 | if (!line[0]) | |
1319 | continue; | |
1320 | c1 = strchr(line, ':'); | |
1321 | if (!c1) | |
1322 | goto out; | |
1323 | c1++; | |
1324 | c2 = strchr(c1, ':'); | |
1325 | if (!c2) | |
1326 | goto out; | |
1327 | *c2 = '\0'; | |
1328 | if (strcmp(c1, h) != 0) | |
1329 | continue; | |
1330 | c2++; | |
1331 | stripnewline(c2); | |
1332 | do { | |
1333 | answer = strdup(c2); | |
1334 | } while (!answer); | |
1335 | break; | |
1336 | } | |
1337 | ||
1338 | out: | |
1339 | fclose(f); | |
1340 | free(line); | |
1341 | return answer; | |
1342 | } | |
1343 | ||
1344 | /* | |
1345 | * check whether a fuse context may access a cgroup dir or file | |
1346 | * | |
1347 | * If file is not null, it is a cgroup file to check under cg. | |
1348 | * If file is null, then we are checking perms on cg itself. | |
1349 | * | |
1350 | * For files we can check the mode of the list_keys result. | |
1351 | * For cgroups, we must make assumptions based on the files under the | |
1352 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1353 | * yet. | |
1354 | */ | |
1355 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1356 | { | |
1357 | struct cgfs_files *k = NULL; | |
1358 | bool ret = false; | |
1359 | ||
1360 | k = cgfs_get_key(contrl, cg, file); | |
1361 | if (!k) | |
1362 | return false; | |
1363 | ||
1364 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1365 | if (perms_include(k->mode >> 6, mode)) { | |
1366 | ret = true; | |
1367 | goto out; | |
1368 | } | |
1369 | } | |
1370 | if (fc->gid == k->gid) { | |
1371 | if (perms_include(k->mode >> 3, mode)) { | |
1372 | ret = true; | |
1373 | goto out; | |
1374 | } | |
1375 | } | |
1376 | ret = perms_include(k->mode, mode); | |
1377 | ||
1378 | out: | |
1379 | free_key(k); | |
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | #define INITSCOPE "/init.scope" | |
1384 | static void prune_init_slice(char *cg) | |
1385 | { | |
1386 | char *point; | |
1387 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1388 | ||
1389 | if (cg_len < initscope_len) | |
1390 | return; | |
1391 | ||
1392 | point = cg + cg_len - initscope_len; | |
1393 | if (strcmp(point, INITSCOPE) == 0) { | |
1394 | if (point == cg) | |
1395 | *(point+1) = '\0'; | |
1396 | else | |
1397 | *point = '\0'; | |
1398 | } | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1403 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1404 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1405 | * to a string containing the next cgroup directory under cg, which must be | |
1406 | * freed by the caller. | |
1407 | */ | |
1408 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1409 | { | |
1410 | bool answer = false; | |
1411 | char *c2 = get_pid_cgroup(pid, contrl); | |
1412 | char *linecmp; | |
1413 | ||
1414 | if (!c2) | |
1415 | return false; | |
1416 | prune_init_slice(c2); | |
1417 | ||
1418 | /* | |
12c31268 CB |
1419 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1420 | * they pass in a cgroup without leading '/' | |
1421 | * | |
1422 | * The original line here was: | |
1423 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1424 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1425 | * Serge, do you know? | |
237e200e | 1426 | */ |
12c31268 CB |
1427 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1428 | linecmp = c2; | |
1429 | else | |
1430 | linecmp = c2 + 1; | |
237e200e SH |
1431 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1432 | if (nextcg) { | |
1433 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1434 | } | |
1435 | goto out; | |
1436 | } | |
1437 | answer = true; | |
1438 | ||
1439 | out: | |
1440 | free(c2); | |
1441 | return answer; | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1446 | */ | |
1447 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1448 | { | |
1449 | bool answer = false; | |
1450 | char *c2, *task_cg; | |
1451 | size_t target_len, task_len; | |
1452 | ||
f7bff426 | 1453 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1454 | return true; |
1455 | ||
1456 | c2 = get_pid_cgroup(pid, contrl); | |
1457 | if (!c2) | |
1458 | return false; | |
1459 | prune_init_slice(c2); | |
1460 | ||
1461 | task_cg = c2 + 1; | |
1462 | target_len = strlen(cg); | |
1463 | task_len = strlen(task_cg); | |
1464 | if (task_len == 0) { | |
1465 | /* Task is in the root cg, it can see everything. This case is | |
1466 | * not handled by the strmcps below, since they test for the | |
1467 | * last /, but that is the first / that we've chopped off | |
1468 | * above. | |
1469 | */ | |
1470 | answer = true; | |
1471 | goto out; | |
1472 | } | |
1473 | if (strcmp(cg, task_cg) == 0) { | |
1474 | answer = true; | |
1475 | goto out; | |
1476 | } | |
1477 | if (target_len < task_len) { | |
1478 | /* looking up a parent dir */ | |
1479 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1480 | answer = true; | |
1481 | goto out; | |
1482 | } | |
1483 | if (target_len > task_len) { | |
1484 | /* looking up a child dir */ | |
1485 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1486 | answer = true; | |
1487 | goto out; | |
1488 | } | |
1489 | ||
1490 | out: | |
1491 | free(c2); | |
1492 | return answer; | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * given /cgroup/freezer/a/b, return "freezer". | |
1497 | * the returned char* should NOT be freed. | |
1498 | */ | |
1499 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1500 | { | |
1501 | const char *p1; | |
1502 | char *contr, *slash; | |
1503 | ||
1504 | if (strlen(path) < 9) | |
1505 | return NULL; | |
1506 | if (*(path+7) != '/') | |
1507 | return NULL; | |
1508 | p1 = path+8; | |
1509 | contr = strdupa(p1); | |
1510 | if (!contr) | |
1511 | return NULL; | |
1512 | slash = strstr(contr, "/"); | |
1513 | if (slash) | |
1514 | *slash = '\0'; | |
1515 | ||
1516 | int i; | |
1517 | for (i = 0; i < num_hierarchies; i++) { | |
1518 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) | |
1519 | return hierarchies[i]; | |
1520 | } | |
1521 | return NULL; | |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1526 | * Note that the returned value may include files (keynames) etc | |
1527 | */ | |
1528 | static const char *find_cgroup_in_path(const char *path) | |
1529 | { | |
1530 | const char *p1; | |
1531 | ||
1532 | if (strlen(path) < 9) | |
1533 | return NULL; | |
1534 | p1 = strstr(path+8, "/"); | |
1535 | if (!p1) | |
1536 | return NULL; | |
1537 | return p1+1; | |
1538 | } | |
1539 | ||
1540 | /* | |
1541 | * split the last path element from the path in @cg. | |
1542 | * @dir is newly allocated and should be freed, @last not | |
1543 | */ | |
1544 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1545 | { | |
1546 | char *p; | |
1547 | ||
1548 | do { | |
1549 | *dir = strdup(cg); | |
1550 | } while (!*dir); | |
1551 | *last = strrchr(cg, '/'); | |
1552 | if (!*last) { | |
1553 | *last = NULL; | |
1554 | return; | |
1555 | } | |
1556 | p = strrchr(*dir, '/'); | |
1557 | *p = '\0'; | |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * FUSE ops for /cgroup | |
1562 | */ | |
1563 | ||
1564 | int cg_getattr(const char *path, struct stat *sb) | |
1565 | { | |
1566 | struct timespec now; | |
1567 | struct fuse_context *fc = fuse_get_context(); | |
1568 | char * cgdir = NULL; | |
1569 | char *last = NULL, *path1, *path2; | |
1570 | struct cgfs_files *k = NULL; | |
1571 | const char *cgroup; | |
1572 | const char *controller = NULL; | |
1573 | int ret = -ENOENT; | |
1574 | ||
1575 | ||
1576 | if (!fc) | |
1577 | return -EIO; | |
1578 | ||
1579 | memset(sb, 0, sizeof(struct stat)); | |
1580 | ||
1581 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1582 | return -EINVAL; | |
1583 | ||
1584 | sb->st_uid = sb->st_gid = 0; | |
1585 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1586 | sb->st_size = 0; | |
1587 | ||
1588 | if (strcmp(path, "/cgroup") == 0) { | |
1589 | sb->st_mode = S_IFDIR | 00755; | |
1590 | sb->st_nlink = 2; | |
1591 | return 0; | |
1592 | } | |
1593 | ||
1594 | controller = pick_controller_from_path(fc, path); | |
1595 | if (!controller) | |
1596 | return -EIO; | |
1597 | cgroup = find_cgroup_in_path(path); | |
1598 | if (!cgroup) { | |
1599 | /* this is just /cgroup/controller, return it as a dir */ | |
1600 | sb->st_mode = S_IFDIR | 00755; | |
1601 | sb->st_nlink = 2; | |
1602 | return 0; | |
1603 | } | |
1604 | ||
1605 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1606 | ||
1607 | if (!last) { | |
1608 | path1 = "/"; | |
1609 | path2 = cgdir; | |
1610 | } else { | |
1611 | path1 = cgdir; | |
1612 | path2 = last; | |
1613 | } | |
1614 | ||
1615 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1616 | if (initpid <= 0) | |
1617 | initpid = fc->pid; | |
1618 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1619 | * Then check that caller's cgroup is under path if last is a child | |
1620 | * cgroup, or cgdir if last is a file */ | |
1621 | ||
1622 | if (is_child_cgroup(controller, path1, path2)) { | |
1623 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1624 | ret = -ENOENT; | |
1625 | goto out; | |
1626 | } | |
1627 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1628 | /* this is just /cgroup/controller, return it as a dir */ | |
1629 | sb->st_mode = S_IFDIR | 00555; | |
1630 | sb->st_nlink = 2; | |
1631 | ret = 0; | |
1632 | goto out; | |
1633 | } | |
1634 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1635 | ret = -EACCES; | |
1636 | goto out; | |
1637 | } | |
1638 | ||
1639 | // get uid, gid, from '/tasks' file and make up a mode | |
1640 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1641 | sb->st_mode = S_IFDIR | 00755; | |
1642 | k = cgfs_get_key(controller, cgroup, NULL); | |
1643 | if (!k) { | |
1644 | sb->st_uid = sb->st_gid = 0; | |
1645 | } else { | |
1646 | sb->st_uid = k->uid; | |
1647 | sb->st_gid = k->gid; | |
1648 | } | |
1649 | free_key(k); | |
1650 | sb->st_nlink = 2; | |
1651 | ret = 0; | |
1652 | goto out; | |
1653 | } | |
1654 | ||
1655 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1656 | sb->st_mode = S_IFREG | k->mode; | |
1657 | sb->st_nlink = 1; | |
1658 | sb->st_uid = k->uid; | |
1659 | sb->st_gid = k->gid; | |
1660 | sb->st_size = 0; | |
1661 | free_key(k); | |
1662 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1663 | ret = -ENOENT; | |
1664 | goto out; | |
1665 | } | |
1666 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) { | |
1667 | ret = -EACCES; | |
1668 | goto out; | |
1669 | } | |
1670 | ||
1671 | ret = 0; | |
1672 | } | |
1673 | ||
1674 | out: | |
1675 | free(cgdir); | |
1676 | return ret; | |
1677 | } | |
1678 | ||
1679 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1680 | { | |
1681 | struct fuse_context *fc = fuse_get_context(); | |
1682 | const char *cgroup; | |
1683 | struct file_info *dir_info; | |
1684 | char *controller = NULL; | |
1685 | ||
1686 | if (!fc) | |
1687 | return -EIO; | |
1688 | ||
1689 | if (strcmp(path, "/cgroup") == 0) { | |
1690 | cgroup = NULL; | |
1691 | controller = NULL; | |
1692 | } else { | |
1693 | // return list of keys for the controller, and list of child cgroups | |
1694 | controller = pick_controller_from_path(fc, path); | |
1695 | if (!controller) | |
1696 | return -EIO; | |
1697 | ||
1698 | cgroup = find_cgroup_in_path(path); | |
1699 | if (!cgroup) { | |
1700 | /* this is just /cgroup/controller, return its contents */ | |
1701 | cgroup = "/"; | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1706 | if (initpid <= 0) | |
1707 | initpid = fc->pid; | |
1708 | if (cgroup) { | |
1709 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1710 | return -ENOENT; | |
1711 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1712 | return -EACCES; | |
1713 | } | |
1714 | ||
1715 | /* we'll free this at cg_releasedir */ | |
1716 | dir_info = malloc(sizeof(*dir_info)); | |
1717 | if (!dir_info) | |
1718 | return -ENOMEM; | |
1719 | dir_info->controller = must_copy_string(controller); | |
1720 | dir_info->cgroup = must_copy_string(cgroup); | |
1721 | dir_info->type = LXC_TYPE_CGDIR; | |
1722 | dir_info->buf = NULL; | |
1723 | dir_info->file = NULL; | |
1724 | dir_info->buflen = 0; | |
1725 | ||
1726 | fi->fh = (unsigned long)dir_info; | |
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1731 | struct fuse_file_info *fi) | |
1732 | { | |
1733 | struct file_info *d = (struct file_info *)fi->fh; | |
1734 | struct cgfs_files **list = NULL; | |
1735 | int i, ret; | |
1736 | char *nextcg = NULL; | |
1737 | struct fuse_context *fc = fuse_get_context(); | |
1738 | char **clist = NULL; | |
1739 | ||
1740 | if (d->type != LXC_TYPE_CGDIR) { | |
1741 | fprintf(stderr, "Internal error: file cache info used in readdir\n"); | |
1742 | return -EIO; | |
1743 | } | |
1744 | if (!d->cgroup && !d->controller) { | |
1745 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1746 | int i; | |
1747 | ||
1748 | for (i = 0; i < num_hierarchies; i++) { | |
1749 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1750 | return -EIO; | |
1751 | } | |
1752 | } | |
1753 | return 0; | |
1754 | } | |
1755 | ||
1756 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1757 | // not a valid cgroup | |
1758 | ret = -EINVAL; | |
1759 | goto out; | |
1760 | } | |
1761 | ||
1762 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1763 | if (initpid <= 0) | |
1764 | initpid = fc->pid; | |
1765 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1766 | if (nextcg) { | |
1767 | ret = filler(buf, nextcg, NULL, 0); | |
1768 | free(nextcg); | |
1769 | if (ret != 0) { | |
1770 | ret = -EIO; | |
1771 | goto out; | |
1772 | } | |
1773 | } | |
1774 | ret = 0; | |
1775 | goto out; | |
1776 | } | |
1777 | ||
1778 | for (i = 0; list[i]; i++) { | |
1779 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
1780 | ret = -EIO; | |
1781 | goto out; | |
1782 | } | |
1783 | } | |
1784 | ||
1785 | // now get the list of child cgroups | |
1786 | ||
1787 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
1788 | ret = 0; | |
1789 | goto out; | |
1790 | } | |
f366da65 WB |
1791 | if (clist) { |
1792 | for (i = 0; clist[i]; i++) { | |
1793 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
1794 | ret = -EIO; | |
1795 | goto out; | |
1796 | } | |
237e200e SH |
1797 | } |
1798 | } | |
1799 | ret = 0; | |
1800 | ||
1801 | out: | |
1802 | free_keys(list); | |
1803 | if (clist) { | |
1804 | for (i = 0; clist[i]; i++) | |
1805 | free(clist[i]); | |
1806 | free(clist); | |
1807 | } | |
1808 | return ret; | |
1809 | } | |
1810 | ||
43215927 | 1811 | static void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 1812 | { |
43215927 SH |
1813 | struct file_info *f = (struct file_info *)fi->fh; |
1814 | ||
237e200e SH |
1815 | if (!f) |
1816 | return; | |
43215927 SH |
1817 | |
1818 | fi->fh = 0; | |
1819 | ||
237e200e | 1820 | free(f->controller); |
43215927 | 1821 | f->controller = NULL; |
237e200e | 1822 | free(f->cgroup); |
43215927 | 1823 | f->cgroup = NULL; |
237e200e | 1824 | free(f->file); |
43215927 | 1825 | f->file = NULL; |
237e200e | 1826 | free(f->buf); |
43215927 | 1827 | f->buf = NULL; |
237e200e SH |
1828 | free(f); |
1829 | } | |
1830 | ||
1831 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
1832 | { | |
43215927 | 1833 | do_release_file_info(fi); |
237e200e SH |
1834 | return 0; |
1835 | } | |
1836 | ||
1837 | int cg_open(const char *path, struct fuse_file_info *fi) | |
1838 | { | |
1839 | const char *cgroup; | |
1840 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1841 | struct cgfs_files *k = NULL; | |
1842 | struct file_info *file_info; | |
1843 | struct fuse_context *fc = fuse_get_context(); | |
1844 | int ret; | |
1845 | ||
1846 | if (!fc) | |
1847 | return -EIO; | |
1848 | ||
1849 | controller = pick_controller_from_path(fc, path); | |
1850 | if (!controller) | |
1851 | return -EIO; | |
1852 | cgroup = find_cgroup_in_path(path); | |
1853 | if (!cgroup) | |
1854 | return -EINVAL; | |
1855 | ||
1856 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1857 | if (!last) { | |
1858 | path1 = "/"; | |
1859 | path2 = cgdir; | |
1860 | } else { | |
1861 | path1 = cgdir; | |
1862 | path2 = last; | |
1863 | } | |
1864 | ||
1865 | k = cgfs_get_key(controller, path1, path2); | |
1866 | if (!k) { | |
1867 | ret = -EINVAL; | |
1868 | goto out; | |
1869 | } | |
1870 | free_key(k); | |
1871 | ||
1872 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1873 | if (initpid <= 0) | |
1874 | initpid = fc->pid; | |
1875 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1876 | ret = -ENOENT; | |
1877 | goto out; | |
1878 | } | |
1879 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
1880 | ret = -EACCES; |
1881 | goto out; | |
1882 | } | |
1883 | ||
1884 | /* we'll free this at cg_release */ | |
1885 | file_info = malloc(sizeof(*file_info)); | |
1886 | if (!file_info) { | |
1887 | ret = -ENOMEM; | |
1888 | goto out; | |
1889 | } | |
1890 | file_info->controller = must_copy_string(controller); | |
1891 | file_info->cgroup = must_copy_string(path1); | |
1892 | file_info->file = must_copy_string(path2); | |
1893 | file_info->type = LXC_TYPE_CGFILE; | |
1894 | file_info->buf = NULL; | |
1895 | file_info->buflen = 0; | |
1896 | ||
1897 | fi->fh = (unsigned long)file_info; | |
1898 | ret = 0; | |
1899 | ||
1900 | out: | |
1901 | free(cgdir); | |
1902 | return ret; | |
1903 | } | |
1904 | ||
bddbb106 SH |
1905 | int cg_access(const char *path, int mode) |
1906 | { | |
1907 | const char *cgroup; | |
1908 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1909 | struct cgfs_files *k = NULL; | |
1910 | struct fuse_context *fc = fuse_get_context(); | |
1911 | int ret; | |
1912 | ||
1913 | if (!fc) | |
1914 | return -EIO; | |
1915 | ||
1916 | controller = pick_controller_from_path(fc, path); | |
1917 | if (!controller) | |
1918 | return -EIO; | |
1919 | cgroup = find_cgroup_in_path(path); | |
575316c4 SH |
1920 | if (!cgroup) { |
1921 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
1922 | if ((mode & W_OK) == 0) |
1923 | return 0; | |
1924 | return -EACCES; | |
575316c4 | 1925 | } |
bddbb106 SH |
1926 | |
1927 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1928 | if (!last) { | |
1929 | path1 = "/"; | |
1930 | path2 = cgdir; | |
1931 | } else { | |
1932 | path1 = cgdir; | |
1933 | path2 = last; | |
1934 | } | |
1935 | ||
1936 | k = cgfs_get_key(controller, path1, path2); | |
1937 | if (!k) { | |
3f441bc7 SH |
1938 | if ((mode & W_OK) == 0) |
1939 | ret = 0; | |
1940 | else | |
1941 | ret = -EACCES; | |
bddbb106 SH |
1942 | goto out; |
1943 | } | |
1944 | free_key(k); | |
1945 | ||
1946 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1947 | if (initpid <= 0) | |
1948 | initpid = fc->pid; | |
1949 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1950 | ret = -ENOENT; | |
1951 | goto out; | |
1952 | } | |
1953 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
1954 | ret = -EACCES; | |
1955 | goto out; | |
1956 | } | |
1957 | ||
1958 | ret = 0; | |
1959 | ||
1960 | out: | |
1961 | free(cgdir); | |
1962 | return ret; | |
1963 | } | |
1964 | ||
237e200e SH |
1965 | int cg_release(const char *path, struct fuse_file_info *fi) |
1966 | { | |
43215927 | 1967 | do_release_file_info(fi); |
237e200e SH |
1968 | return 0; |
1969 | } | |
1970 | ||
1971 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
1972 | ||
1973 | static bool wait_for_sock(int sock, int timeout) | |
1974 | { | |
1975 | struct epoll_event ev; | |
1976 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
1977 | ||
1978 | if ((starttime = time(NULL)) < 0) | |
1979 | return false; | |
1980 | ||
1981 | if ((epfd = epoll_create(1)) < 0) { | |
1982 | fprintf(stderr, "Failed to create epoll socket: %m\n"); | |
1983 | return false; | |
1984 | } | |
1985 | ||
1986 | ev.events = POLLIN_SET; | |
1987 | ev.data.fd = sock; | |
1988 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
1989 | fprintf(stderr, "Failed adding socket to epoll: %m\n"); | |
1990 | close(epfd); | |
1991 | return false; | |
1992 | } | |
1993 | ||
1994 | again: | |
1995 | if ((now = time(NULL)) < 0) { | |
1996 | close(epfd); | |
1997 | return false; | |
1998 | } | |
1999 | ||
2000 | deltatime = (starttime + timeout) - now; | |
2001 | if (deltatime < 0) { // timeout | |
2002 | errno = 0; | |
2003 | close(epfd); | |
2004 | return false; | |
2005 | } | |
2006 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2007 | if (ret < 0 && errno == EINTR) | |
2008 | goto again; | |
2009 | saved_errno = errno; | |
2010 | close(epfd); | |
2011 | ||
2012 | if (ret <= 0) { | |
2013 | errno = saved_errno; | |
2014 | return false; | |
2015 | } | |
2016 | return true; | |
2017 | } | |
2018 | ||
2019 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2020 | { | |
2021 | if (!wait_for_sock(sockfd, 2)) | |
2022 | return -1; | |
2023 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2024 | } | |
2025 | ||
2026 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2027 | { | |
2028 | struct msghdr msg = { 0 }; | |
2029 | struct iovec iov; | |
2030 | struct cmsghdr *cmsg; | |
2031 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2032 | char buf[1]; | |
2033 | buf[0] = 'p'; | |
2034 | ||
2035 | if (pingfirst) { | |
2036 | if (msgrecv(sock, buf, 1) != 1) { | |
2037 | fprintf(stderr, "%s: Error getting reply from server over socketpair\n", | |
2038 | __func__); | |
2039 | return SEND_CREDS_FAIL; | |
2040 | } | |
2041 | } | |
2042 | ||
2043 | msg.msg_control = cmsgbuf; | |
2044 | msg.msg_controllen = sizeof(cmsgbuf); | |
2045 | ||
2046 | cmsg = CMSG_FIRSTHDR(&msg); | |
2047 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2048 | cmsg->cmsg_level = SOL_SOCKET; | |
2049 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2050 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2051 | ||
2052 | msg.msg_name = NULL; | |
2053 | msg.msg_namelen = 0; | |
2054 | ||
2055 | buf[0] = v; | |
2056 | iov.iov_base = buf; | |
2057 | iov.iov_len = sizeof(buf); | |
2058 | msg.msg_iov = &iov; | |
2059 | msg.msg_iovlen = 1; | |
2060 | ||
2061 | if (sendmsg(sock, &msg, 0) < 0) { | |
2062 | fprintf(stderr, "%s: failed at sendmsg: %s\n", __func__, | |
2063 | strerror(errno)); | |
2064 | if (errno == 3) | |
2065 | return SEND_CREDS_NOTSK; | |
2066 | return SEND_CREDS_FAIL; | |
2067 | } | |
2068 | ||
2069 | return SEND_CREDS_OK; | |
2070 | } | |
2071 | ||
2072 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2073 | { | |
2074 | struct msghdr msg = { 0 }; | |
2075 | struct iovec iov; | |
2076 | struct cmsghdr *cmsg; | |
2077 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2078 | char buf[1]; | |
2079 | int ret; | |
2080 | int optval = 1; | |
2081 | ||
2082 | *v = '1'; | |
2083 | ||
2084 | cred->pid = -1; | |
2085 | cred->uid = -1; | |
2086 | cred->gid = -1; | |
2087 | ||
2088 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
2089 | fprintf(stderr, "Failed to set passcred: %s\n", strerror(errno)); | |
2090 | return false; | |
2091 | } | |
2092 | buf[0] = '1'; | |
2093 | if (write(sock, buf, 1) != 1) { | |
2094 | fprintf(stderr, "Failed to start write on scm fd: %s\n", strerror(errno)); | |
2095 | return false; | |
2096 | } | |
2097 | ||
2098 | msg.msg_name = NULL; | |
2099 | msg.msg_namelen = 0; | |
2100 | msg.msg_control = cmsgbuf; | |
2101 | msg.msg_controllen = sizeof(cmsgbuf); | |
2102 | ||
2103 | iov.iov_base = buf; | |
2104 | iov.iov_len = sizeof(buf); | |
2105 | msg.msg_iov = &iov; | |
2106 | msg.msg_iovlen = 1; | |
2107 | ||
2108 | if (!wait_for_sock(sock, 2)) { | |
2109 | fprintf(stderr, "Timed out waiting for scm_cred: %s\n", | |
2110 | strerror(errno)); | |
2111 | return false; | |
2112 | } | |
2113 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2114 | if (ret < 0) { | |
2115 | fprintf(stderr, "Failed to receive scm_cred: %s\n", | |
2116 | strerror(errno)); | |
2117 | return false; | |
2118 | } | |
2119 | ||
2120 | cmsg = CMSG_FIRSTHDR(&msg); | |
2121 | ||
2122 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2123 | cmsg->cmsg_level == SOL_SOCKET && | |
2124 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2125 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2126 | } | |
2127 | *v = buf[0]; | |
2128 | ||
2129 | return true; | |
2130 | } | |
2131 | ||
35174b0f FG |
2132 | struct pid_ns_clone_args { |
2133 | int *cpipe; | |
2134 | int sock; | |
2135 | pid_t tpid; | |
2136 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2137 | }; | |
2138 | ||
2139 | /* | |
2140 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2141 | * with clone(). This simply writes '1' as ACK back to the parent | |
2142 | * before calling the actual wrapped function. | |
2143 | */ | |
2144 | static int pid_ns_clone_wrapper(void *arg) { | |
2145 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2146 | char b = '1'; | |
2147 | ||
2148 | close(args->cpipe[0]); | |
2149 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) { | |
2150 | fprintf(stderr, "%s (child): error on write: %s\n", | |
2151 | __func__, strerror(errno)); | |
2152 | } | |
2153 | close(args->cpipe[1]); | |
2154 | return args->wrapped(args->sock, args->tpid); | |
2155 | } | |
237e200e SH |
2156 | |
2157 | /* | |
2158 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2159 | * int value back over the socket. This shifts the pid from the | |
2160 | * sender's pidns into tpid's pidns. | |
2161 | */ | |
35174b0f | 2162 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2163 | { |
2164 | char v = '0'; | |
2165 | struct ucred cred; | |
2166 | ||
2167 | while (recv_creds(sock, &cred, &v)) { | |
2168 | if (v == '1') | |
35174b0f | 2169 | return 0; |
237e200e | 2170 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2171 | return 1; |
237e200e | 2172 | } |
35174b0f | 2173 | return 0; |
237e200e SH |
2174 | } |
2175 | ||
35174b0f | 2176 | |
237e200e SH |
2177 | /* |
2178 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2179 | * in your old pidns. Only children which you clone will be in the target |
2180 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2181 | * actually convert pids. | |
2182 | * | |
2183 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2184 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2185 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2186 | * this issue. | |
237e200e SH |
2187 | */ |
2188 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2189 | { | |
2190 | int newnsfd = -1, ret, cpipe[2]; | |
2191 | char fnam[100]; | |
2192 | pid_t cpid; | |
2193 | char v; | |
2194 | ||
2195 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2196 | if (ret < 0 || ret >= sizeof(fnam)) | |
2197 | _exit(1); | |
2198 | newnsfd = open(fnam, O_RDONLY); | |
2199 | if (newnsfd < 0) | |
2200 | _exit(1); | |
2201 | if (setns(newnsfd, 0) < 0) | |
2202 | _exit(1); | |
2203 | close(newnsfd); | |
2204 | ||
2205 | if (pipe(cpipe) < 0) | |
2206 | _exit(1); | |
2207 | ||
35174b0f FG |
2208 | struct pid_ns_clone_args args = { |
2209 | .cpipe = cpipe, | |
2210 | .sock = sock, | |
2211 | .tpid = tpid, | |
2212 | .wrapped = &pid_to_ns | |
2213 | }; | |
2214 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2215 | void *stack = alloca(stack_size); | |
2216 | ||
2217 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2218 | if (cpid < 0) |
2219 | _exit(1); | |
2220 | ||
237e200e SH |
2221 | // give the child 1 second to be done forking and |
2222 | // write its ack | |
2223 | if (!wait_for_sock(cpipe[0], 1)) | |
2224 | _exit(1); | |
2225 | ret = read(cpipe[0], &v, 1); | |
2226 | if (ret != sizeof(char) || v != '1') | |
2227 | _exit(1); | |
2228 | ||
2229 | if (!wait_for_pid(cpid)) | |
2230 | _exit(1); | |
2231 | _exit(0); | |
2232 | } | |
2233 | ||
2234 | /* | |
2235 | * To read cgroup files with a particular pid, we will setns into the child | |
2236 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2237 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2238 | */ | |
2239 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2240 | { | |
2241 | int sock[2] = {-1, -1}; | |
2242 | char *tmpdata = NULL; | |
2243 | int ret; | |
2244 | pid_t qpid, cpid = -1; | |
2245 | bool answer = false; | |
2246 | char v = '0'; | |
2247 | struct ucred cred; | |
2248 | size_t sz = 0, asz = 0; | |
2249 | ||
2250 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2251 | return false; | |
2252 | ||
2253 | /* | |
2254 | * Now we read the pids from returned data one by one, pass | |
2255 | * them into a child in the target namespace, read back the | |
2256 | * translated pids, and put them into our to-return data | |
2257 | */ | |
2258 | ||
2259 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2260 | perror("socketpair"); | |
2261 | free(tmpdata); | |
2262 | return false; | |
2263 | } | |
2264 | ||
2265 | cpid = fork(); | |
2266 | if (cpid == -1) | |
2267 | goto out; | |
2268 | ||
2269 | if (!cpid) // child - exits when done | |
2270 | pid_to_ns_wrapper(sock[1], tpid); | |
2271 | ||
2272 | char *ptr = tmpdata; | |
2273 | cred.uid = 0; | |
2274 | cred.gid = 0; | |
2275 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2276 | cred.pid = qpid; | |
2277 | ret = send_creds(sock[0], &cred, v, true); | |
2278 | ||
2279 | if (ret == SEND_CREDS_NOTSK) | |
2280 | goto next; | |
2281 | if (ret == SEND_CREDS_FAIL) | |
2282 | goto out; | |
2283 | ||
2284 | // read converted results | |
2285 | if (!wait_for_sock(sock[0], 2)) { | |
2286 | fprintf(stderr, "%s: timed out waiting for pid from child: %s\n", | |
2287 | __func__, strerror(errno)); | |
2288 | goto out; | |
2289 | } | |
2290 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2291 | fprintf(stderr, "%s: error reading pid from child: %s\n", | |
2292 | __func__, strerror(errno)); | |
2293 | goto out; | |
2294 | } | |
2295 | must_strcat_pid(d, &sz, &asz, qpid); | |
2296 | next: | |
2297 | ptr = strchr(ptr, '\n'); | |
2298 | if (!ptr) | |
2299 | break; | |
2300 | ptr++; | |
2301 | } | |
2302 | ||
2303 | cred.pid = getpid(); | |
2304 | v = '1'; | |
2305 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2306 | // failed to ask child to exit | |
2307 | fprintf(stderr, "%s: failed to ask child to exit: %s\n", | |
2308 | __func__, strerror(errno)); | |
2309 | goto out; | |
2310 | } | |
2311 | ||
2312 | answer = true; | |
2313 | ||
2314 | out: | |
2315 | free(tmpdata); | |
2316 | if (cpid != -1) | |
2317 | wait_for_pid(cpid); | |
2318 | if (sock[0] != -1) { | |
2319 | close(sock[0]); | |
2320 | close(sock[1]); | |
2321 | } | |
2322 | return answer; | |
2323 | } | |
2324 | ||
2325 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2326 | struct fuse_file_info *fi) | |
2327 | { | |
2328 | struct fuse_context *fc = fuse_get_context(); | |
2329 | struct file_info *f = (struct file_info *)fi->fh; | |
2330 | struct cgfs_files *k = NULL; | |
2331 | char *data = NULL; | |
2332 | int ret, s; | |
2333 | bool r; | |
2334 | ||
2335 | if (f->type != LXC_TYPE_CGFILE) { | |
2336 | fprintf(stderr, "Internal error: directory cache info used in cg_read\n"); | |
2337 | return -EIO; | |
2338 | } | |
2339 | ||
2340 | if (offset) | |
2341 | return 0; | |
2342 | ||
2343 | if (!fc) | |
2344 | return -EIO; | |
2345 | ||
2346 | if (!f->controller) | |
2347 | return -EINVAL; | |
2348 | ||
2349 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2350 | return -EINVAL; | |
2351 | } | |
2352 | free_key(k); | |
2353 | ||
2354 | ||
888f8f3c | 2355 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2356 | ret = -EACCES; |
2357 | goto out; | |
2358 | } | |
2359 | ||
2360 | if (strcmp(f->file, "tasks") == 0 || | |
2361 | strcmp(f->file, "/tasks") == 0 || | |
2362 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2363 | strcmp(f->file, "cgroup.procs") == 0) | |
2364 | // special case - we have to translate the pids | |
2365 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2366 | else | |
2367 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2368 | ||
2369 | if (!r) { | |
2370 | ret = -EINVAL; | |
2371 | goto out; | |
2372 | } | |
2373 | ||
2374 | if (!data) { | |
2375 | ret = 0; | |
2376 | goto out; | |
2377 | } | |
2378 | s = strlen(data); | |
2379 | if (s > size) | |
2380 | s = size; | |
2381 | memcpy(buf, data, s); | |
2382 | if (s > 0 && s < size && data[s-1] != '\n') | |
2383 | buf[s++] = '\n'; | |
2384 | ||
2385 | ret = s; | |
2386 | ||
2387 | out: | |
2388 | free(data); | |
2389 | return ret; | |
2390 | } | |
2391 | ||
35174b0f | 2392 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2393 | { |
2394 | pid_t vpid; | |
2395 | struct ucred cred; | |
2396 | char v; | |
2397 | int ret; | |
2398 | ||
2399 | cred.uid = 0; | |
2400 | cred.gid = 0; | |
2401 | while (1) { | |
2402 | if (!wait_for_sock(sock, 2)) { | |
2403 | fprintf(stderr, "%s: timeout reading from parent\n", __func__); | |
35174b0f | 2404 | return 1; |
237e200e SH |
2405 | } |
2406 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
2407 | fprintf(stderr, "%s: bad read from parent: %s\n", | |
2408 | __func__, strerror(errno)); | |
35174b0f | 2409 | return 1; |
237e200e SH |
2410 | } |
2411 | if (vpid == -1) // done | |
2412 | break; | |
2413 | v = '0'; | |
2414 | cred.pid = vpid; | |
2415 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2416 | v = '1'; | |
2417 | cred.pid = getpid(); | |
2418 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2419 | return 1; |
237e200e SH |
2420 | } |
2421 | } | |
35174b0f | 2422 | return 0; |
237e200e SH |
2423 | } |
2424 | ||
2425 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2426 | { | |
2427 | int newnsfd = -1, ret, cpipe[2]; | |
2428 | char fnam[100]; | |
2429 | pid_t cpid; | |
2430 | char v; | |
2431 | ||
2432 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2433 | if (ret < 0 || ret >= sizeof(fnam)) | |
2434 | _exit(1); | |
2435 | newnsfd = open(fnam, O_RDONLY); | |
2436 | if (newnsfd < 0) | |
2437 | _exit(1); | |
2438 | if (setns(newnsfd, 0) < 0) | |
2439 | _exit(1); | |
2440 | close(newnsfd); | |
2441 | ||
2442 | if (pipe(cpipe) < 0) | |
2443 | _exit(1); | |
2444 | ||
35174b0f FG |
2445 | struct pid_ns_clone_args args = { |
2446 | .cpipe = cpipe, | |
2447 | .sock = sock, | |
2448 | .tpid = tpid, | |
2449 | .wrapped = &pid_from_ns | |
2450 | }; | |
f0f8b851 SH |
2451 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2452 | void *stack = alloca(stack_size); | |
35174b0f FG |
2453 | |
2454 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2455 | if (cpid < 0) |
2456 | _exit(1); | |
2457 | ||
237e200e SH |
2458 | // give the child 1 second to be done forking and |
2459 | // write its ack | |
2460 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2461 | _exit(1); |
237e200e | 2462 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2463 | if (ret != sizeof(char) || v != '1') |
2464 | _exit(1); | |
237e200e SH |
2465 | |
2466 | if (!wait_for_pid(cpid)) | |
2467 | _exit(1); | |
2468 | _exit(0); | |
237e200e SH |
2469 | } |
2470 | ||
2471 | /* | |
2472 | * Given host @uid, return the uid to which it maps in | |
2473 | * @pid's user namespace, or -1 if none. | |
2474 | */ | |
2475 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2476 | { | |
2477 | FILE *f; | |
2478 | char line[400]; | |
2479 | ||
2480 | sprintf(line, "/proc/%d/uid_map", pid); | |
2481 | if ((f = fopen(line, "r")) == NULL) { | |
2482 | return false; | |
2483 | } | |
2484 | ||
2485 | *answer = convert_id_to_ns(f, uid); | |
2486 | fclose(f); | |
2487 | ||
2488 | if (*answer == -1) | |
2489 | return false; | |
2490 | return true; | |
2491 | } | |
2492 | ||
2493 | /* | |
2494 | * get_pid_creds: get the real uid and gid of @pid from | |
2495 | * /proc/$$/status | |
2496 | * (XXX should we use euid here?) | |
2497 | */ | |
2498 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2499 | { | |
2500 | char line[400]; | |
2501 | uid_t u; | |
2502 | gid_t g; | |
2503 | FILE *f; | |
2504 | ||
2505 | *uid = -1; | |
2506 | *gid = -1; | |
2507 | sprintf(line, "/proc/%d/status", pid); | |
2508 | if ((f = fopen(line, "r")) == NULL) { | |
2509 | fprintf(stderr, "Error opening %s: %s\n", line, strerror(errno)); | |
2510 | return; | |
2511 | } | |
2512 | while (fgets(line, 400, f)) { | |
2513 | if (strncmp(line, "Uid:", 4) == 0) { | |
2514 | if (sscanf(line+4, "%u", &u) != 1) { | |
2515 | fprintf(stderr, "bad uid line for pid %u\n", pid); | |
2516 | fclose(f); | |
2517 | return; | |
2518 | } | |
2519 | *uid = u; | |
2520 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2521 | if (sscanf(line+4, "%u", &g) != 1) { | |
2522 | fprintf(stderr, "bad gid line for pid %u\n", pid); | |
2523 | fclose(f); | |
2524 | return; | |
2525 | } | |
2526 | *gid = g; | |
2527 | } | |
2528 | } | |
2529 | fclose(f); | |
2530 | } | |
2531 | ||
2532 | /* | |
2533 | * May the requestor @r move victim @v to a new cgroup? | |
2534 | * This is allowed if | |
2535 | * . they are the same task | |
2536 | * . they are ownedy by the same uid | |
2537 | * . @r is root on the host, or | |
2538 | * . @v's uid is mapped into @r's where @r is root. | |
2539 | */ | |
2540 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2541 | { | |
2542 | uid_t v_uid, tmpuid; | |
2543 | gid_t v_gid; | |
2544 | ||
2545 | if (r == v) | |
2546 | return true; | |
2547 | if (r_uid == 0) | |
2548 | return true; | |
2549 | get_pid_creds(v, &v_uid, &v_gid); | |
2550 | if (r_uid == v_uid) | |
2551 | return true; | |
2552 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2553 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2554 | return true; | |
2555 | return false; | |
2556 | } | |
2557 | ||
2558 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2559 | const char *file, const char *buf) | |
2560 | { | |
2561 | int sock[2] = {-1, -1}; | |
2562 | pid_t qpid, cpid = -1; | |
2563 | FILE *pids_file = NULL; | |
2564 | bool answer = false, fail = false; | |
2565 | ||
2566 | pids_file = open_pids_file(contrl, cg); | |
2567 | if (!pids_file) | |
2568 | return false; | |
2569 | ||
2570 | /* | |
2571 | * write the pids to a socket, have helper in writer's pidns | |
2572 | * call movepid for us | |
2573 | */ | |
2574 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2575 | perror("socketpair"); | |
2576 | goto out; | |
2577 | } | |
2578 | ||
2579 | cpid = fork(); | |
2580 | if (cpid == -1) | |
2581 | goto out; | |
2582 | ||
2583 | if (!cpid) { // child | |
2584 | fclose(pids_file); | |
2585 | pid_from_ns_wrapper(sock[1], tpid); | |
2586 | } | |
2587 | ||
2588 | const char *ptr = buf; | |
2589 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2590 | struct ucred cred; | |
2591 | char v; | |
2592 | ||
2593 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2594 | fprintf(stderr, "%s: error writing pid to child: %s\n", | |
2595 | __func__, strerror(errno)); | |
2596 | goto out; | |
2597 | } | |
2598 | ||
2599 | if (recv_creds(sock[0], &cred, &v)) { | |
2600 | if (v == '0') { | |
2601 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2602 | fail = true; | |
2603 | break; | |
2604 | } | |
2605 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2606 | fail = true; | |
2607 | } | |
2608 | } | |
2609 | ||
2610 | ptr = strchr(ptr, '\n'); | |
2611 | if (!ptr) | |
2612 | break; | |
2613 | ptr++; | |
2614 | } | |
2615 | ||
2616 | /* All good, write the value */ | |
2617 | qpid = -1; | |
2618 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
2619 | fprintf(stderr, "Warning: failed to ask child to exit\n"); | |
2620 | ||
2621 | if (!fail) | |
2622 | answer = true; | |
2623 | ||
2624 | out: | |
2625 | if (cpid != -1) | |
2626 | wait_for_pid(cpid); | |
2627 | if (sock[0] != -1) { | |
2628 | close(sock[0]); | |
2629 | close(sock[1]); | |
2630 | } | |
2631 | if (pids_file) { | |
2632 | if (fclose(pids_file) != 0) | |
2633 | answer = false; | |
2634 | } | |
2635 | return answer; | |
2636 | } | |
2637 | ||
2638 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2639 | struct fuse_file_info *fi) | |
2640 | { | |
2641 | struct fuse_context *fc = fuse_get_context(); | |
2642 | char *localbuf = NULL; | |
2643 | struct cgfs_files *k = NULL; | |
2644 | struct file_info *f = (struct file_info *)fi->fh; | |
2645 | bool r; | |
2646 | ||
2647 | if (f->type != LXC_TYPE_CGFILE) { | |
2648 | fprintf(stderr, "Internal error: directory cache info used in cg_write\n"); | |
2649 | return -EIO; | |
2650 | } | |
2651 | ||
2652 | if (offset) | |
2653 | return 0; | |
2654 | ||
2655 | if (!fc) | |
2656 | return -EIO; | |
2657 | ||
2658 | localbuf = alloca(size+1); | |
2659 | localbuf[size] = '\0'; | |
2660 | memcpy(localbuf, buf, size); | |
2661 | ||
2662 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2663 | size = -EINVAL; | |
2664 | goto out; | |
2665 | } | |
2666 | ||
2667 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2668 | size = -EACCES; | |
2669 | goto out; | |
2670 | } | |
2671 | ||
2672 | if (strcmp(f->file, "tasks") == 0 || | |
2673 | strcmp(f->file, "/tasks") == 0 || | |
2674 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2675 | strcmp(f->file, "cgroup.procs") == 0) | |
2676 | // special case - we have to translate the pids | |
2677 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2678 | else | |
2679 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2680 | ||
2681 | if (!r) | |
2682 | size = -EINVAL; | |
2683 | ||
2684 | out: | |
2685 | free_key(k); | |
2686 | return size; | |
2687 | } | |
2688 | ||
2689 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2690 | { | |
2691 | struct fuse_context *fc = fuse_get_context(); | |
2692 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2693 | struct cgfs_files *k = NULL; | |
2694 | const char *cgroup; | |
2695 | int ret; | |
2696 | ||
2697 | if (!fc) | |
2698 | return -EIO; | |
2699 | ||
2700 | if (strcmp(path, "/cgroup") == 0) | |
2701 | return -EINVAL; | |
2702 | ||
2703 | controller = pick_controller_from_path(fc, path); | |
2704 | if (!controller) | |
2705 | return -EINVAL; | |
2706 | cgroup = find_cgroup_in_path(path); | |
2707 | if (!cgroup) | |
2708 | /* this is just /cgroup/controller */ | |
2709 | return -EINVAL; | |
2710 | ||
2711 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2712 | ||
2713 | if (!last) { | |
2714 | path1 = "/"; | |
2715 | path2 = cgdir; | |
2716 | } else { | |
2717 | path1 = cgdir; | |
2718 | path2 = last; | |
2719 | } | |
2720 | ||
2721 | if (is_child_cgroup(controller, path1, path2)) { | |
2722 | // get uid, gid, from '/tasks' file and make up a mode | |
2723 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2724 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2725 | ||
2726 | } else | |
2727 | k = cgfs_get_key(controller, path1, path2); | |
2728 | ||
2729 | if (!k) { | |
2730 | ret = -EINVAL; | |
2731 | goto out; | |
2732 | } | |
2733 | ||
2734 | /* | |
2735 | * This being a fuse request, the uid and gid must be valid | |
2736 | * in the caller's namespace. So we can just check to make | |
2737 | * sure that the caller is root in his uid, and privileged | |
2738 | * over the file's current owner. | |
2739 | */ | |
2740 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2741 | ret = -EACCES; | |
2742 | goto out; | |
2743 | } | |
2744 | ||
2745 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2746 | ||
2747 | out: | |
2748 | free_key(k); | |
2749 | free(cgdir); | |
2750 | ||
2751 | return ret; | |
2752 | } | |
2753 | ||
2754 | int cg_chmod(const char *path, mode_t mode) | |
2755 | { | |
2756 | struct fuse_context *fc = fuse_get_context(); | |
2757 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2758 | struct cgfs_files *k = NULL; | |
2759 | const char *cgroup; | |
2760 | int ret; | |
2761 | ||
2762 | if (!fc) | |
2763 | return -EIO; | |
2764 | ||
2765 | if (strcmp(path, "/cgroup") == 0) | |
2766 | return -EINVAL; | |
2767 | ||
2768 | controller = pick_controller_from_path(fc, path); | |
2769 | if (!controller) | |
2770 | return -EINVAL; | |
2771 | cgroup = find_cgroup_in_path(path); | |
2772 | if (!cgroup) | |
2773 | /* this is just /cgroup/controller */ | |
2774 | return -EINVAL; | |
2775 | ||
2776 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2777 | ||
2778 | if (!last) { | |
2779 | path1 = "/"; | |
2780 | path2 = cgdir; | |
2781 | } else { | |
2782 | path1 = cgdir; | |
2783 | path2 = last; | |
2784 | } | |
2785 | ||
2786 | if (is_child_cgroup(controller, path1, path2)) { | |
2787 | // get uid, gid, from '/tasks' file and make up a mode | |
2788 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2789 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2790 | ||
2791 | } else | |
2792 | k = cgfs_get_key(controller, path1, path2); | |
2793 | ||
2794 | if (!k) { | |
2795 | ret = -EINVAL; | |
2796 | goto out; | |
2797 | } | |
2798 | ||
2799 | /* | |
2800 | * This being a fuse request, the uid and gid must be valid | |
2801 | * in the caller's namespace. So we can just check to make | |
2802 | * sure that the caller is root in his uid, and privileged | |
2803 | * over the file's current owner. | |
2804 | */ | |
2805 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
2806 | ret = -EPERM; | |
2807 | goto out; | |
2808 | } | |
2809 | ||
2810 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
2811 | ret = -EINVAL; | |
2812 | goto out; | |
2813 | } | |
2814 | ||
2815 | ret = 0; | |
2816 | out: | |
2817 | free_key(k); | |
2818 | free(cgdir); | |
2819 | return ret; | |
2820 | } | |
2821 | ||
2822 | int cg_mkdir(const char *path, mode_t mode) | |
2823 | { | |
2824 | struct fuse_context *fc = fuse_get_context(); | |
2825 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
2826 | const char *cgroup; | |
2827 | int ret; | |
2828 | ||
2829 | if (!fc) | |
2830 | return -EIO; | |
2831 | ||
2832 | ||
2833 | controller = pick_controller_from_path(fc, path); | |
2834 | if (!controller) | |
2835 | return -EINVAL; | |
2836 | ||
2837 | cgroup = find_cgroup_in_path(path); | |
2838 | if (!cgroup) | |
2839 | return -EINVAL; | |
2840 | ||
2841 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2842 | if (!last) | |
2843 | path1 = "/"; | |
2844 | else | |
2845 | path1 = cgdir; | |
2846 | ||
2847 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2848 | if (initpid <= 0) | |
2849 | initpid = fc->pid; | |
2850 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
2851 | if (!next) | |
2852 | ret = -EINVAL; | |
2853 | else if (last && strcmp(next, last) == 0) | |
2854 | ret = -EEXIST; | |
2855 | else | |
2856 | ret = -ENOENT; | |
2857 | goto out; | |
2858 | } | |
2859 | ||
2860 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
2861 | ret = -EACCES; | |
2862 | goto out; | |
2863 | } | |
2864 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2865 | ret = -EACCES; | |
2866 | goto out; | |
2867 | } | |
2868 | ||
2869 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
2870 | ||
2871 | out: | |
2872 | free(cgdir); | |
2873 | free(next); | |
2874 | return ret; | |
2875 | } | |
2876 | ||
2877 | int cg_rmdir(const char *path) | |
2878 | { | |
2879 | struct fuse_context *fc = fuse_get_context(); | |
2880 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
2881 | const char *cgroup; | |
2882 | int ret; | |
2883 | ||
2884 | if (!fc) | |
2885 | return -EIO; | |
2886 | ||
2887 | controller = pick_controller_from_path(fc, path); | |
2888 | if (!controller) | |
2889 | return -EINVAL; | |
2890 | ||
2891 | cgroup = find_cgroup_in_path(path); | |
2892 | if (!cgroup) | |
2893 | return -EINVAL; | |
2894 | ||
2895 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2896 | if (!last) { | |
2897 | ret = -EINVAL; | |
2898 | goto out; | |
2899 | } | |
2900 | ||
2901 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2902 | if (initpid <= 0) | |
2903 | initpid = fc->pid; | |
2904 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
2905 | if (!last || strcmp(next, last) == 0) | |
2906 | ret = -EBUSY; | |
2907 | else | |
2908 | ret = -ENOENT; | |
2909 | goto out; | |
2910 | } | |
2911 | ||
2912 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
2913 | ret = -EACCES; | |
2914 | goto out; | |
2915 | } | |
2916 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2917 | ret = -EACCES; | |
2918 | goto out; | |
2919 | } | |
2920 | ||
2921 | if (!cgfs_remove(controller, cgroup)) { | |
2922 | ret = -EINVAL; | |
2923 | goto out; | |
2924 | } | |
2925 | ||
2926 | ret = 0; | |
2927 | ||
2928 | out: | |
2929 | free(cgdir); | |
2930 | free(next); | |
2931 | return ret; | |
2932 | } | |
2933 | ||
2934 | static bool startswith(const char *line, const char *pref) | |
2935 | { | |
2936 | if (strncmp(line, pref, strlen(pref)) == 0) | |
2937 | return true; | |
2938 | return false; | |
2939 | } | |
2940 | ||
c6095b08 SH |
2941 | static void parse_memstat(char *memstat, unsigned long *cached, |
2942 | unsigned long *active_anon, unsigned long *inactive_anon, | |
2943 | unsigned long *active_file, unsigned long *inactive_file, | |
2944 | unsigned long *unevictable) | |
237e200e SH |
2945 | { |
2946 | char *eol; | |
2947 | ||
237e200e | 2948 | while (*memstat) { |
c6095b08 SH |
2949 | if (startswith(memstat, "cache")) { |
2950 | sscanf(memstat + 11, "%lu", cached); | |
2951 | *cached /= 1024; | |
2952 | } else if (startswith(memstat, "active_anon")) { | |
2953 | sscanf(memstat + 11, "%lu", active_anon); | |
2954 | *active_anon /= 1024; | |
2955 | } else if (startswith(memstat, "inactive_anon")) { | |
2956 | sscanf(memstat + 11, "%lu", inactive_anon); | |
2957 | *inactive_anon /= 1024; | |
2958 | } else if (startswith(memstat, "active_file")) { | |
2959 | sscanf(memstat + 11, "%lu", active_file); | |
2960 | *active_file /= 1024; | |
2961 | } else if (startswith(memstat, "inactive_file")) { | |
2962 | sscanf(memstat + 11, "%lu", inactive_file); | |
2963 | *inactive_file /= 1024; | |
2964 | } else if (startswith(memstat, "unevictable")) { | |
2965 | sscanf(memstat + 11, "%lu", unevictable); | |
2966 | *unevictable /= 1024; | |
237e200e SH |
2967 | } |
2968 | eol = strchr(memstat, '\n'); | |
2969 | if (!eol) | |
2970 | return; | |
2971 | memstat = eol+1; | |
2972 | } | |
2973 | } | |
2974 | ||
2975 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
2976 | { | |
2977 | char *eol; | |
2978 | char key[32]; | |
2979 | ||
2980 | memset(key, 0, 32); | |
2981 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2982 | ||
2983 | size_t len = strlen(key); | |
2984 | *v = 0; | |
2985 | ||
2986 | while (*str) { | |
2987 | if (startswith(str, key)) { | |
2988 | sscanf(str + len, "%lu", v); | |
2989 | return; | |
2990 | } | |
2991 | eol = strchr(str, '\n'); | |
2992 | if (!eol) | |
2993 | return; | |
2994 | str = eol+1; | |
2995 | } | |
2996 | } | |
2997 | ||
2998 | static int read_file(const char *path, char *buf, size_t size, | |
2999 | struct file_info *d) | |
3000 | { | |
3001 | size_t linelen = 0, total_len = 0, rv = 0; | |
3002 | char *line = NULL; | |
3003 | char *cache = d->buf; | |
3004 | size_t cache_size = d->buflen; | |
3005 | FILE *f = fopen(path, "r"); | |
3006 | if (!f) | |
3007 | return 0; | |
3008 | ||
3009 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3010 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3011 | if (l < 0) { |
3012 | perror("Error writing to cache"); | |
3013 | rv = 0; | |
3014 | goto err; | |
3015 | } | |
3016 | if (l >= cache_size) { | |
3017 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3018 | rv = 0; | |
3019 | goto err; | |
3020 | } | |
3021 | cache += l; | |
3022 | cache_size -= l; | |
3023 | total_len += l; | |
3024 | } | |
3025 | ||
3026 | d->size = total_len; | |
a262ddb7 CB |
3027 | if (total_len > size) |
3028 | total_len = size; | |
237e200e SH |
3029 | |
3030 | /* read from off 0 */ | |
3031 | memcpy(buf, d->buf, total_len); | |
3032 | rv = total_len; | |
3033 | err: | |
3034 | fclose(f); | |
3035 | free(line); | |
3036 | return rv; | |
3037 | } | |
3038 | ||
3039 | /* | |
3040 | * FUSE ops for /proc | |
3041 | */ | |
3042 | ||
3043 | static unsigned long get_memlimit(const char *cgroup) | |
3044 | { | |
3045 | char *memlimit_str = NULL; | |
3046 | unsigned long memlimit = -1; | |
3047 | ||
3048 | if (cgfs_get_value("memory", cgroup, "memory.limit_in_bytes", &memlimit_str)) | |
3049 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3050 | ||
3051 | free(memlimit_str); | |
3052 | ||
3053 | return memlimit; | |
3054 | } | |
3055 | ||
3056 | static unsigned long get_min_memlimit(const char *cgroup) | |
3057 | { | |
3058 | char *copy = strdupa(cgroup); | |
3059 | unsigned long memlimit = 0, retlimit; | |
3060 | ||
3061 | retlimit = get_memlimit(copy); | |
3062 | ||
3063 | while (strcmp(copy, "/") != 0) { | |
3064 | copy = dirname(copy); | |
3065 | memlimit = get_memlimit(copy); | |
3066 | if (memlimit != -1 && memlimit < retlimit) | |
3067 | retlimit = memlimit; | |
3068 | }; | |
3069 | ||
3070 | return retlimit; | |
3071 | } | |
3072 | ||
3073 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3074 | struct fuse_file_info *fi) | |
3075 | { | |
3076 | struct fuse_context *fc = fuse_get_context(); | |
3077 | struct file_info *d = (struct file_info *)fi->fh; | |
3078 | char *cg; | |
3079 | char *memusage_str = NULL, *memstat_str = NULL, | |
3080 | *memswlimit_str = NULL, *memswusage_str = NULL, | |
3081 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
3082 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, | |
c6095b08 SH |
3083 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
3084 | active_file = 0, inactive_file = 0, unevictable = 0; | |
237e200e SH |
3085 | char *line = NULL; |
3086 | size_t linelen = 0, total_len = 0, rv = 0; | |
3087 | char *cache = d->buf; | |
3088 | size_t cache_size = d->buflen; | |
3089 | FILE *f = NULL; | |
3090 | ||
3091 | if (offset){ | |
3092 | if (offset > d->size) | |
3093 | return -EINVAL; | |
3094 | if (!d->cached) | |
3095 | return 0; | |
3096 | int left = d->size - offset; | |
3097 | total_len = left > size ? size: left; | |
3098 | memcpy(buf, cache + offset, total_len); | |
3099 | return total_len; | |
3100 | } | |
3101 | ||
3102 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3103 | if (initpid <= 0) | |
3104 | initpid = fc->pid; | |
3105 | cg = get_pid_cgroup(initpid, "memory"); | |
3106 | if (!cg) | |
3107 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3108 | prune_init_slice(cg); |
237e200e SH |
3109 | |
3110 | memlimit = get_min_memlimit(cg); | |
3111 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3112 | goto err; | |
3113 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3114 | goto err; | |
3115 | ||
3116 | // Following values are allowed to fail, because swapaccount might be turned | |
3117 | // off for current kernel | |
3118 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3119 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3120 | { | |
3121 | /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */ | |
3122 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
3123 | goto err; | |
3124 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
3125 | goto err; | |
3126 | ||
3127 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
3128 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3129 | ||
3130 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
3131 | memswlimit = 0; | |
3132 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
3133 | memswusage = 0; | |
3134 | ||
3135 | memswlimit = memswlimit / 1024; | |
3136 | memswusage = memswusage / 1024; | |
3137 | } | |
3138 | ||
3139 | memusage = strtoul(memusage_str, NULL, 10); | |
3140 | memlimit /= 1024; | |
3141 | memusage /= 1024; | |
3142 | ||
c6095b08 SH |
3143 | parse_memstat(memstat_str, &cached, &active_anon, |
3144 | &inactive_anon, &active_file, &inactive_file, | |
3145 | &unevictable); | |
237e200e SH |
3146 | |
3147 | f = fopen("/proc/meminfo", "r"); | |
3148 | if (!f) | |
3149 | goto err; | |
3150 | ||
3151 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3152 | ssize_t l; |
237e200e SH |
3153 | char *printme, lbuf[100]; |
3154 | ||
3155 | memset(lbuf, 0, 100); | |
3156 | if (startswith(line, "MemTotal:")) { | |
3157 | sscanf(line+14, "%lu", &hosttotal); | |
3158 | if (hosttotal < memlimit) | |
3159 | memlimit = hosttotal; | |
3160 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3161 | printme = lbuf; | |
3162 | } else if (startswith(line, "MemFree:")) { | |
3163 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3164 | printme = lbuf; | |
3165 | } else if (startswith(line, "MemAvailable:")) { | |
3166 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
3167 | printme = lbuf; | |
3168 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
3169 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit - memlimit); | |
3170 | printme = lbuf; | |
3171 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
b4665ce0 SH |
3172 | unsigned long swaptotal = memswlimit - memlimit, |
3173 | swapusage = memswusage - memusage, | |
3174 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3175 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3176 | printme = lbuf; |
da35d72a SH |
3177 | } else if (startswith(line, "Slab:")) { |
3178 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3179 | printme = lbuf; | |
237e200e SH |
3180 | } else if (startswith(line, "Buffers:")) { |
3181 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3182 | printme = lbuf; | |
3183 | } else if (startswith(line, "Cached:")) { | |
3184 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3185 | printme = lbuf; | |
3186 | } else if (startswith(line, "SwapCached:")) { | |
3187 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3188 | printme = lbuf; | |
c6095b08 SH |
3189 | } else if (startswith(line, "Active")) { |
3190 | snprintf(lbuf, 100, "Active: %8lu kB\n", | |
3191 | active_anon + active_file); | |
3192 | printme = lbuf; | |
3193 | } else if (startswith(line, "Inactive")) { | |
3194 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", | |
3195 | inactive_anon + inactive_file); | |
3196 | printme = lbuf; | |
3197 | } else if (startswith(line, "Active(anon)")) { | |
3198 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3199 | printme = lbuf; | |
3200 | } else if (startswith(line, "Inactive(anon)")) { | |
3201 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3202 | printme = lbuf; | |
3203 | } else if (startswith(line, "Active(file)")) { | |
3204 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3205 | printme = lbuf; | |
3206 | } else if (startswith(line, "Inactive(file)")) { | |
3207 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3208 | printme = lbuf; | |
3209 | } else if (startswith(line, "Unevictable")) { | |
3210 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3211 | printme = lbuf; | |
3212 | } else if (startswith(line, "SReclaimable")) { | |
3213 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3214 | printme = lbuf; | |
3215 | } else if (startswith(line, "SUnreclaim")) { | |
3216 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3217 | printme = lbuf; | |
237e200e SH |
3218 | } else |
3219 | printme = line; | |
3220 | ||
3221 | l = snprintf(cache, cache_size, "%s", printme); | |
3222 | if (l < 0) { | |
3223 | perror("Error writing to cache"); | |
3224 | rv = 0; | |
3225 | goto err; | |
3226 | ||
3227 | } | |
3228 | if (l >= cache_size) { | |
3229 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3230 | rv = 0; | |
3231 | goto err; | |
3232 | } | |
3233 | ||
3234 | cache += l; | |
3235 | cache_size -= l; | |
3236 | total_len += l; | |
3237 | } | |
3238 | ||
3239 | d->cached = 1; | |
3240 | d->size = total_len; | |
3241 | if (total_len > size ) total_len = size; | |
3242 | memcpy(buf, d->buf, total_len); | |
3243 | ||
3244 | rv = total_len; | |
3245 | err: | |
3246 | if (f) | |
3247 | fclose(f); | |
3248 | free(line); | |
3249 | free(cg); | |
3250 | free(memusage_str); | |
3251 | free(memswlimit_str); | |
3252 | free(memswusage_str); | |
3253 | free(memstat_str); | |
3254 | free(memswlimit_default_str); | |
3255 | free(memswusage_default_str); | |
3256 | return rv; | |
3257 | } | |
3258 | ||
3259 | /* | |
3260 | * Read the cpuset.cpus for cg | |
3261 | * Return the answer in a newly allocated string which must be freed | |
3262 | */ | |
3263 | static char *get_cpuset(const char *cg) | |
3264 | { | |
3265 | char *answer; | |
3266 | ||
3267 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3268 | return NULL; | |
3269 | return answer; | |
3270 | } | |
3271 | ||
3272 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3273 | ||
3274 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3275 | { | |
3276 | int cpu; | |
3277 | ||
3278 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3279 | return false; | |
3280 | return cpu_in_cpuset(cpu, cpuset); | |
3281 | } | |
3282 | ||
3283 | /* | |
3284 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3285 | */ | |
3286 | static bool is_processor_line(const char *line) | |
3287 | { | |
3288 | int cpu; | |
3289 | ||
3290 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3291 | return true; | |
3292 | return false; | |
3293 | } | |
3294 | ||
3295 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3296 | struct fuse_file_info *fi) | |
3297 | { | |
3298 | struct fuse_context *fc = fuse_get_context(); | |
3299 | struct file_info *d = (struct file_info *)fi->fh; | |
3300 | char *cg; | |
3301 | char *cpuset = NULL; | |
3302 | char *line = NULL; | |
3303 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 SH |
3304 | bool am_printing = false, firstline = true, is_s390x = false; |
3305 | int curcpu = -1, cpu; | |
237e200e SH |
3306 | char *cache = d->buf; |
3307 | size_t cache_size = d->buflen; | |
3308 | FILE *f = NULL; | |
3309 | ||
3310 | if (offset){ | |
3311 | if (offset > d->size) | |
3312 | return -EINVAL; | |
3313 | if (!d->cached) | |
3314 | return 0; | |
3315 | int left = d->size - offset; | |
3316 | total_len = left > size ? size: left; | |
3317 | memcpy(buf, cache + offset, total_len); | |
3318 | return total_len; | |
3319 | } | |
3320 | ||
3321 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3322 | if (initpid <= 0) | |
3323 | initpid = fc->pid; | |
3324 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3325 | if (!cg) | |
3326 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3327 | prune_init_slice(cg); |
237e200e SH |
3328 | |
3329 | cpuset = get_cpuset(cg); | |
3330 | if (!cpuset) | |
3331 | goto err; | |
3332 | ||
3333 | f = fopen("/proc/cpuinfo", "r"); | |
3334 | if (!f) | |
3335 | goto err; | |
3336 | ||
3337 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3338 | ssize_t l; |
f676eb79 SH |
3339 | if (firstline) { |
3340 | firstline = false; | |
3341 | if (strstr(line, "IBM/S390") != NULL) { | |
3342 | is_s390x = true; | |
3343 | am_printing = true; | |
5ed9d4e2 | 3344 | continue; |
f676eb79 SH |
3345 | } |
3346 | } | |
5ed9d4e2 SH |
3347 | if (strncmp(line, "# processors:", 12) == 0) |
3348 | continue; | |
237e200e SH |
3349 | if (is_processor_line(line)) { |
3350 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3351 | if (am_printing) { | |
3352 | curcpu ++; | |
3353 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3354 | if (l < 0) { | |
3355 | perror("Error writing to cache"); | |
3356 | rv = 0; | |
3357 | goto err; | |
3358 | } | |
3359 | if (l >= cache_size) { | |
3360 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3361 | rv = 0; | |
3362 | goto err; | |
3363 | } | |
3364 | cache += l; | |
3365 | cache_size -= l; | |
3366 | total_len += l; | |
3367 | } | |
3368 | continue; | |
f676eb79 SH |
3369 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3370 | char *p; | |
3371 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3372 | continue; | |
3373 | curcpu ++; | |
3374 | p = strchr(line, ':'); | |
3375 | if (!p || !*p) | |
3376 | goto err; | |
3377 | p++; | |
5ed9d4e2 | 3378 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3379 | if (l < 0) { |
3380 | perror("Error writing to cache"); | |
3381 | rv = 0; | |
3382 | goto err; | |
3383 | } | |
3384 | if (l >= cache_size) { | |
3385 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3386 | rv = 0; | |
3387 | goto err; | |
3388 | } | |
3389 | cache += l; | |
3390 | cache_size -= l; | |
3391 | total_len += l; | |
3392 | continue; | |
3393 | ||
237e200e SH |
3394 | } |
3395 | if (am_printing) { | |
3396 | l = snprintf(cache, cache_size, "%s", line); | |
3397 | if (l < 0) { | |
3398 | perror("Error writing to cache"); | |
3399 | rv = 0; | |
3400 | goto err; | |
3401 | } | |
3402 | if (l >= cache_size) { | |
3403 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3404 | rv = 0; | |
3405 | goto err; | |
3406 | } | |
3407 | cache += l; | |
3408 | cache_size -= l; | |
3409 | total_len += l; | |
3410 | } | |
3411 | } | |
3412 | ||
5ed9d4e2 SH |
3413 | if (is_s390x) { |
3414 | char *origcache = d->buf; | |
a262ddb7 | 3415 | ssize_t l; |
5ed9d4e2 SH |
3416 | do { |
3417 | d->buf = malloc(d->buflen); | |
3418 | } while (!d->buf); | |
3419 | cache = d->buf; | |
3420 | cache_size = d->buflen; | |
3421 | total_len = 0; | |
3422 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3423 | if (l < 0 || l >= cache_size) { | |
3424 | free(origcache); | |
3425 | goto err; | |
3426 | } | |
3427 | cache_size -= l; | |
3428 | cache += l; | |
3429 | total_len += l; | |
3430 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3431 | if (l < 0 || l >= cache_size) { | |
3432 | free(origcache); | |
3433 | goto err; | |
3434 | } | |
3435 | cache_size -= l; | |
3436 | cache += l; | |
3437 | total_len += l; | |
3438 | l = snprintf(cache, cache_size, "%s", origcache); | |
3439 | free(origcache); | |
3440 | if (l < 0 || l >= cache_size) | |
3441 | goto err; | |
3442 | total_len += l; | |
3443 | } | |
3444 | ||
237e200e SH |
3445 | d->cached = 1; |
3446 | d->size = total_len; | |
3447 | if (total_len > size ) total_len = size; | |
3448 | ||
3449 | /* read from off 0 */ | |
3450 | memcpy(buf, d->buf, total_len); | |
3451 | rv = total_len; | |
3452 | err: | |
3453 | if (f) | |
3454 | fclose(f); | |
3455 | free(line); | |
3456 | free(cpuset); | |
3457 | free(cg); | |
3458 | return rv; | |
3459 | } | |
3460 | ||
3461 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
3462 | struct fuse_file_info *fi) | |
3463 | { | |
3464 | struct fuse_context *fc = fuse_get_context(); | |
3465 | struct file_info *d = (struct file_info *)fi->fh; | |
3466 | char *cg; | |
3467 | char *cpuset = NULL; | |
3468 | char *line = NULL; | |
3469 | size_t linelen = 0, total_len = 0, rv = 0; | |
3470 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
3471 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0; | |
3472 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, | |
3473 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0; | |
3474 | #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE | |
3475 | char cpuall[CPUALL_MAX_SIZE]; | |
3476 | /* reserve for cpu all */ | |
3477 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3478 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3479 | FILE *f = NULL; | |
3480 | ||
3481 | if (offset){ | |
3482 | if (offset > d->size) | |
3483 | return -EINVAL; | |
3484 | if (!d->cached) | |
3485 | return 0; | |
3486 | int left = d->size - offset; | |
3487 | total_len = left > size ? size: left; | |
3488 | memcpy(buf, d->buf + offset, total_len); | |
3489 | return total_len; | |
3490 | } | |
3491 | ||
3492 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3493 | if (initpid <= 0) | |
3494 | initpid = fc->pid; | |
3495 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3496 | if (!cg) | |
3497 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 3498 | prune_init_slice(cg); |
237e200e SH |
3499 | |
3500 | cpuset = get_cpuset(cg); | |
3501 | if (!cpuset) | |
3502 | goto err; | |
3503 | ||
3504 | f = fopen("/proc/stat", "r"); | |
3505 | if (!f) | |
3506 | goto err; | |
3507 | ||
3508 | //skip first line | |
3509 | if (getline(&line, &linelen, f) < 0) { | |
3510 | fprintf(stderr, "proc_stat_read read first line failed\n"); | |
3511 | goto err; | |
3512 | } | |
3513 | ||
3514 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3515 | ssize_t l; |
237e200e SH |
3516 | int cpu; |
3517 | char cpu_char[10]; /* That's a lot of cores */ | |
3518 | char *c; | |
3519 | ||
b4665ce0 SH |
3520 | if (strlen(line) == 0) |
3521 | continue; | |
237e200e SH |
3522 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
3523 | /* not a ^cpuN line containing a number N, just print it */ | |
3524 | l = snprintf(cache, cache_size, "%s", line); | |
3525 | if (l < 0) { | |
3526 | perror("Error writing to cache"); | |
3527 | rv = 0; | |
3528 | goto err; | |
3529 | } | |
3530 | if (l >= cache_size) { | |
3531 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3532 | rv = 0; | |
3533 | goto err; | |
3534 | } | |
3535 | cache += l; | |
3536 | cache_size -= l; | |
3537 | total_len += l; | |
3538 | continue; | |
3539 | } | |
3540 | ||
3541 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3542 | continue; | |
3543 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3544 | continue; | |
3545 | curcpu ++; | |
3546 | ||
3547 | c = strchr(line, ' '); | |
3548 | if (!c) | |
3549 | continue; | |
3550 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
3551 | if (l < 0) { | |
3552 | perror("Error writing to cache"); | |
3553 | rv = 0; | |
3554 | goto err; | |
3555 | ||
3556 | } | |
3557 | if (l >= cache_size) { | |
3558 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3559 | rv = 0; | |
3560 | goto err; | |
3561 | } | |
3562 | ||
3563 | cache += l; | |
3564 | cache_size -= l; | |
3565 | total_len += l; | |
3566 | ||
3567 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user, &nice, &system, &idle, &iowait, &irq, | |
3568 | &softirq, &steal, &guest) != 9) | |
3569 | continue; | |
3570 | user_sum += user; | |
3571 | nice_sum += nice; | |
3572 | system_sum += system; | |
3573 | idle_sum += idle; | |
3574 | iowait_sum += iowait; | |
3575 | irq_sum += irq; | |
3576 | softirq_sum += softirq; | |
3577 | steal_sum += steal; | |
3578 | guest_sum += guest; | |
3579 | } | |
3580 | ||
3581 | cache = d->buf; | |
3582 | ||
3583 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3584 | "cpu ", user_sum, nice_sum, system_sum, idle_sum, iowait_sum, irq_sum, softirq_sum, steal_sum, guest_sum); | |
3585 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE){ | |
3586 | memcpy(cache, cpuall, cpuall_len); | |
3587 | cache += cpuall_len; | |
3588 | } else{ | |
3589 | /* shouldn't happen */ | |
3590 | fprintf(stderr, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len); | |
3591 | cpuall_len = 0; | |
3592 | } | |
3593 | ||
3594 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
3595 | total_len += cpuall_len; | |
3596 | d->cached = 1; | |
3597 | d->size = total_len; | |
3598 | if (total_len > size ) total_len = size; | |
3599 | ||
3600 | memcpy(buf, d->buf, total_len); | |
3601 | rv = total_len; | |
3602 | ||
3603 | err: | |
3604 | if (f) | |
3605 | fclose(f); | |
3606 | free(line); | |
3607 | free(cpuset); | |
3608 | free(cg); | |
3609 | return rv; | |
3610 | } | |
3611 | ||
3612 | static long int getreaperage(pid_t pid) | |
3613 | { | |
3614 | char fnam[100]; | |
3615 | struct stat sb; | |
3616 | int ret; | |
3617 | pid_t qpid; | |
3618 | ||
3619 | qpid = lookup_initpid_in_store(pid); | |
3620 | if (qpid <= 0) | |
3621 | return 0; | |
3622 | ||
3623 | ret = snprintf(fnam, 100, "/proc/%d", qpid); | |
3624 | if (ret < 0 || ret >= 100) | |
3625 | return 0; | |
3626 | ||
3627 | if (lstat(fnam, &sb) < 0) | |
3628 | return 0; | |
3629 | ||
3630 | return time(NULL) - sb.st_ctime; | |
3631 | } | |
3632 | ||
3633 | static unsigned long get_reaper_busy(pid_t task) | |
3634 | { | |
3635 | pid_t initpid = lookup_initpid_in_store(task); | |
3636 | char *cgroup = NULL, *usage_str = NULL; | |
3637 | unsigned long usage = 0; | |
3638 | ||
3639 | if (initpid <= 0) | |
3640 | return 0; | |
3641 | ||
3642 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
3643 | if (!cgroup) | |
3644 | goto out; | |
6d2f6996 | 3645 | prune_init_slice(cgroup); |
237e200e SH |
3646 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
3647 | goto out; | |
3648 | usage = strtoul(usage_str, NULL, 10); | |
3649 | usage /= 1000000000; | |
3650 | ||
3651 | out: | |
3652 | free(cgroup); | |
3653 | free(usage_str); | |
3654 | return usage; | |
3655 | } | |
3656 | ||
3657 | #if RELOADTEST | |
3658 | void iwashere(void) | |
3659 | { | |
237e200e SH |
3660 | int fd; |
3661 | ||
ec2b5e7c | 3662 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
3663 | if (fd >= 0) |
3664 | close(fd); | |
3665 | } | |
3666 | #endif | |
3667 | ||
3668 | /* | |
3669 | * We read /proc/uptime and reuse its second field. | |
3670 | * For the first field, we use the mtime for the reaper for | |
3671 | * the calling pid as returned by getreaperage | |
3672 | */ | |
3673 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
3674 | struct fuse_file_info *fi) | |
3675 | { | |
3676 | struct fuse_context *fc = fuse_get_context(); | |
3677 | struct file_info *d = (struct file_info *)fi->fh; | |
3678 | long int reaperage = getreaperage(fc->pid); | |
3679 | unsigned long int busytime = get_reaper_busy(fc->pid), idletime; | |
3680 | char *cache = d->buf; | |
a262ddb7 | 3681 | ssize_t total_len = 0; |
237e200e SH |
3682 | |
3683 | #if RELOADTEST | |
3684 | iwashere(); | |
3685 | #endif | |
3686 | ||
3687 | if (offset){ | |
3688 | if (offset > d->size) | |
3689 | return -EINVAL; | |
3690 | if (!d->cached) | |
3691 | return 0; | |
3692 | int left = d->size - offset; | |
3693 | total_len = left > size ? size: left; | |
3694 | memcpy(buf, cache + offset, total_len); | |
3695 | return total_len; | |
3696 | } | |
3697 | ||
3698 | idletime = reaperage - busytime; | |
3699 | if (idletime > reaperage) | |
3700 | idletime = reaperage; | |
3701 | ||
3702 | total_len = snprintf(d->buf, d->size, "%ld.0 %lu.0\n", reaperage, idletime); | |
3703 | if (total_len < 0){ | |
3704 | perror("Error writing to cache"); | |
3705 | return 0; | |
3706 | } | |
3707 | ||
3708 | d->size = (int)total_len; | |
3709 | d->cached = 1; | |
3710 | ||
3711 | if (total_len > size) total_len = size; | |
3712 | ||
3713 | memcpy(buf, d->buf, total_len); | |
3714 | return total_len; | |
3715 | } | |
3716 | ||
3717 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
3718 | struct fuse_file_info *fi) | |
3719 | { | |
3720 | char dev_name[72]; | |
3721 | struct fuse_context *fc = fuse_get_context(); | |
3722 | struct file_info *d = (struct file_info *)fi->fh; | |
3723 | char *cg; | |
3724 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
3725 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
3726 | unsigned long read = 0, write = 0; | |
3727 | unsigned long read_merged = 0, write_merged = 0; | |
3728 | unsigned long read_sectors = 0, write_sectors = 0; | |
3729 | unsigned long read_ticks = 0, write_ticks = 0; | |
3730 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
3731 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
3732 | char *cache = d->buf; | |
3733 | size_t cache_size = d->buflen; | |
3734 | char *line = NULL; | |
3735 | size_t linelen = 0, total_len = 0, rv = 0; | |
3736 | unsigned int major = 0, minor = 0; | |
3737 | int i = 0; | |
3738 | FILE *f = NULL; | |
3739 | ||
3740 | if (offset){ | |
3741 | if (offset > d->size) | |
3742 | return -EINVAL; | |
3743 | if (!d->cached) | |
3744 | return 0; | |
3745 | int left = d->size - offset; | |
3746 | total_len = left > size ? size: left; | |
3747 | memcpy(buf, cache + offset, total_len); | |
3748 | return total_len; | |
3749 | } | |
3750 | ||
3751 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3752 | if (initpid <= 0) | |
3753 | initpid = fc->pid; | |
3754 | cg = get_pid_cgroup(initpid, "blkio"); | |
3755 | if (!cg) | |
3756 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 3757 | prune_init_slice(cg); |
237e200e | 3758 | |
2209fe50 | 3759 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 3760 | goto err; |
2209fe50 | 3761 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 3762 | goto err; |
2209fe50 | 3763 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 3764 | goto err; |
2209fe50 | 3765 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 3766 | goto err; |
2209fe50 | 3767 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
3768 | goto err; |
3769 | ||
3770 | ||
3771 | f = fopen("/proc/diskstats", "r"); | |
3772 | if (!f) | |
3773 | goto err; | |
3774 | ||
3775 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3776 | ssize_t l; |
2209fe50 | 3777 | char lbuf[256]; |
237e200e SH |
3778 | |
3779 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 3780 | if (i != 3) |
237e200e | 3781 | continue; |
2209fe50 SH |
3782 | |
3783 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
3784 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
3785 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
3786 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
3787 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
3788 | read_sectors = read_sectors/512; | |
3789 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
3790 | write_sectors = write_sectors/512; | |
3791 | ||
3792 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
3793 | rd_svctm = rd_svctm/1000000; | |
3794 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
3795 | rd_wait = rd_wait/1000000; | |
3796 | read_ticks = rd_svctm + rd_wait; | |
3797 | ||
3798 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
3799 | wr_svctm = wr_svctm/1000000; | |
3800 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
3801 | wr_wait = wr_wait/1000000; | |
3802 | write_ticks = wr_svctm + wr_wait; | |
3803 | ||
3804 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
3805 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
3806 | |
3807 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
3808 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
3809 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3810 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
3811 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
3812 | else | |
3813 | continue; | |
237e200e | 3814 | |
2209fe50 | 3815 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
3816 | if (l < 0) { |
3817 | perror("Error writing to fuse buf"); | |
3818 | rv = 0; | |
3819 | goto err; | |
3820 | } | |
3821 | if (l >= cache_size) { | |
3822 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3823 | rv = 0; | |
3824 | goto err; | |
3825 | } | |
3826 | cache += l; | |
3827 | cache_size -= l; | |
3828 | total_len += l; | |
3829 | } | |
3830 | ||
3831 | d->cached = 1; | |
3832 | d->size = total_len; | |
3833 | if (total_len > size ) total_len = size; | |
3834 | memcpy(buf, d->buf, total_len); | |
3835 | ||
3836 | rv = total_len; | |
3837 | err: | |
3838 | free(cg); | |
3839 | if (f) | |
3840 | fclose(f); | |
3841 | free(line); | |
3842 | free(io_serviced_str); | |
3843 | free(io_merged_str); | |
3844 | free(io_service_bytes_str); | |
3845 | free(io_wait_time_str); | |
3846 | free(io_service_time_str); | |
3847 | return rv; | |
3848 | } | |
3849 | ||
70dcc12e SH |
3850 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
3851 | struct fuse_file_info *fi) | |
3852 | { | |
3853 | struct fuse_context *fc = fuse_get_context(); | |
3854 | struct file_info *d = (struct file_info *)fi->fh; | |
3855 | char *cg = NULL; | |
3856 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL, | |
3857 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
3858 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; | |
a262ddb7 CB |
3859 | ssize_t total_len = 0, rv = 0; |
3860 | ssize_t l = 0; | |
70dcc12e SH |
3861 | char *cache = d->buf; |
3862 | ||
3863 | if (offset) { | |
3864 | if (offset > d->size) | |
3865 | return -EINVAL; | |
3866 | if (!d->cached) | |
3867 | return 0; | |
3868 | int left = d->size - offset; | |
3869 | total_len = left > size ? size: left; | |
3870 | memcpy(buf, cache + offset, total_len); | |
3871 | return total_len; | |
3872 | } | |
3873 | ||
3874 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3875 | if (initpid <= 0) | |
3876 | initpid = fc->pid; | |
3877 | cg = get_pid_cgroup(initpid, "memory"); | |
3878 | if (!cg) | |
3879 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 3880 | prune_init_slice(cg); |
70dcc12e SH |
3881 | |
3882 | if (!cgfs_get_value("memory", cg, "memory.limit_in_bytes", &memlimit_str)) | |
3883 | goto err; | |
3884 | ||
3885 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3886 | goto err; | |
3887 | ||
3888 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3889 | memusage = strtoul(memusage_str, NULL, 10); | |
3890 | ||
3891 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
3892 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
3893 | ||
3894 | /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */ | |
3895 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
3896 | goto err; | |
3897 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
3898 | goto err; | |
3899 | ||
3900 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
3901 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3902 | ||
3903 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
3904 | memswlimit = 0; | |
3905 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
3906 | memswusage = 0; | |
3907 | ||
3908 | swap_total = (memswlimit - memlimit) / 1024; | |
3909 | swap_free = (memswusage - memusage) / 1024; | |
3910 | } | |
3911 | ||
3912 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
3913 | ||
3914 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
3915 | if (!memswlimit) { | |
3916 | char *line = NULL; | |
3917 | size_t linelen = 0; | |
3918 | FILE *f = fopen("/proc/meminfo", "r"); | |
3919 | ||
3920 | if (!f) | |
3921 | goto err; | |
3922 | ||
3923 | while (getline(&line, &linelen, f) != -1) { | |
3924 | if (startswith(line, "SwapTotal:")) { | |
3925 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
3926 | } else if (startswith(line, "SwapFree:")) { | |
3927 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
3928 | } | |
3929 | } | |
3930 | ||
3931 | free(line); | |
3932 | fclose(f); | |
3933 | } | |
3934 | ||
3935 | if (swap_total > 0) { | |
a262ddb7 CB |
3936 | l = snprintf(d->buf + total_len, d->size - total_len, |
3937 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
3938 | swap_total, swap_free); | |
3939 | total_len += l; | |
70dcc12e SH |
3940 | } |
3941 | ||
a262ddb7 | 3942 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
3943 | perror("Error writing to cache"); |
3944 | rv = 0; | |
3945 | goto err; | |
3946 | } | |
3947 | ||
3948 | d->cached = 1; | |
3949 | d->size = (int)total_len; | |
3950 | ||
3951 | if (total_len > size) total_len = size; | |
3952 | memcpy(buf, d->buf, total_len); | |
3953 | rv = total_len; | |
3954 | ||
3955 | err: | |
3956 | free(cg); | |
3957 | free(memswlimit_str); | |
3958 | free(memlimit_str); | |
3959 | free(memusage_str); | |
3960 | free(memswusage_str); | |
3961 | free(memswusage_default_str); | |
3962 | free(memswlimit_default_str); | |
3963 | return rv; | |
3964 | } | |
3965 | ||
237e200e SH |
3966 | static off_t get_procfile_size(const char *which) |
3967 | { | |
3968 | FILE *f = fopen(which, "r"); | |
3969 | char *line = NULL; | |
3970 | size_t len = 0; | |
3971 | ssize_t sz, answer = 0; | |
3972 | if (!f) | |
3973 | return 0; | |
3974 | ||
3975 | while ((sz = getline(&line, &len, f)) != -1) | |
3976 | answer += sz; | |
3977 | fclose (f); | |
3978 | free(line); | |
3979 | ||
3980 | return answer; | |
3981 | } | |
3982 | ||
3983 | int proc_getattr(const char *path, struct stat *sb) | |
3984 | { | |
3985 | struct timespec now; | |
3986 | ||
3987 | memset(sb, 0, sizeof(struct stat)); | |
3988 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
3989 | return -EINVAL; | |
3990 | sb->st_uid = sb->st_gid = 0; | |
3991 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
3992 | if (strcmp(path, "/proc") == 0) { | |
3993 | sb->st_mode = S_IFDIR | 00555; | |
3994 | sb->st_nlink = 2; | |
3995 | return 0; | |
3996 | } | |
3997 | if (strcmp(path, "/proc/meminfo") == 0 || | |
3998 | strcmp(path, "/proc/cpuinfo") == 0 || | |
3999 | strcmp(path, "/proc/uptime") == 0 || | |
4000 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e SH |
4001 | strcmp(path, "/proc/diskstats") == 0 || |
4002 | strcmp(path, "/proc/swaps") == 0) { | |
237e200e SH |
4003 | sb->st_size = 0; |
4004 | sb->st_mode = S_IFREG | 00444; | |
4005 | sb->st_nlink = 1; | |
4006 | return 0; | |
4007 | } | |
4008 | ||
4009 | return -ENOENT; | |
4010 | } | |
4011 | ||
4012 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4013 | struct fuse_file_info *fi) | |
4014 | { | |
4015 | if (filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4016 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4017 | filler(buf, "stat", NULL, 0) != 0 || | |
4018 | filler(buf, "uptime", NULL, 0) != 0 || | |
70dcc12e SH |
4019 | filler(buf, "diskstats", NULL, 0) != 0 || |
4020 | filler(buf, "swaps", NULL, 0) != 0) | |
237e200e SH |
4021 | return -EINVAL; |
4022 | return 0; | |
4023 | } | |
4024 | ||
4025 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4026 | { | |
4027 | int type = -1; | |
4028 | struct file_info *info; | |
4029 | ||
4030 | if (strcmp(path, "/proc/meminfo") == 0) | |
4031 | type = LXC_TYPE_PROC_MEMINFO; | |
4032 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4033 | type = LXC_TYPE_PROC_CPUINFO; | |
4034 | else if (strcmp(path, "/proc/uptime") == 0) | |
4035 | type = LXC_TYPE_PROC_UPTIME; | |
4036 | else if (strcmp(path, "/proc/stat") == 0) | |
4037 | type = LXC_TYPE_PROC_STAT; | |
4038 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4039 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
4040 | else if (strcmp(path, "/proc/swaps") == 0) |
4041 | type = LXC_TYPE_PROC_SWAPS; | |
237e200e SH |
4042 | if (type == -1) |
4043 | return -ENOENT; | |
4044 | ||
4045 | info = malloc(sizeof(*info)); | |
4046 | if (!info) | |
4047 | return -ENOMEM; | |
4048 | ||
4049 | memset(info, 0, sizeof(*info)); | |
4050 | info->type = type; | |
4051 | ||
4052 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4053 | do { | |
4054 | info->buf = malloc(info->buflen); | |
4055 | } while (!info->buf); | |
4056 | memset(info->buf, 0, info->buflen); | |
4057 | /* set actual size to buffer size */ | |
4058 | info->size = info->buflen; | |
4059 | ||
4060 | fi->fh = (unsigned long)info; | |
4061 | return 0; | |
4062 | } | |
4063 | ||
bddbb106 SH |
4064 | int proc_access(const char *path, int mask) |
4065 | { | |
4066 | /* these are all read-only */ | |
4067 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 4068 | return -EACCES; |
bddbb106 SH |
4069 | return 0; |
4070 | } | |
4071 | ||
237e200e SH |
4072 | int proc_release(const char *path, struct fuse_file_info *fi) |
4073 | { | |
43215927 | 4074 | do_release_file_info(fi); |
237e200e SH |
4075 | return 0; |
4076 | } | |
4077 | ||
4078 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4079 | struct fuse_file_info *fi) | |
4080 | { | |
4081 | struct file_info *f = (struct file_info *) fi->fh; | |
4082 | ||
4083 | switch (f->type) { | |
4084 | case LXC_TYPE_PROC_MEMINFO: | |
4085 | return proc_meminfo_read(buf, size, offset, fi); | |
4086 | case LXC_TYPE_PROC_CPUINFO: | |
4087 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4088 | case LXC_TYPE_PROC_UPTIME: | |
4089 | return proc_uptime_read(buf, size, offset, fi); | |
4090 | case LXC_TYPE_PROC_STAT: | |
4091 | return proc_stat_read(buf, size, offset, fi); | |
4092 | case LXC_TYPE_PROC_DISKSTATS: | |
4093 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
4094 | case LXC_TYPE_PROC_SWAPS: |
4095 | return proc_swaps_read(buf, size, offset, fi); | |
237e200e SH |
4096 | default: |
4097 | return -EINVAL; | |
4098 | } | |
4099 | } | |
4100 | ||
29a73c2f CB |
4101 | /* |
4102 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
4103 | */ |
4104 | ||
4105 | static bool mkdir_p(const char *dir, mode_t mode) | |
4106 | { | |
4107 | const char *tmp = dir; | |
4108 | const char *orig = dir; | |
4109 | char *makeme; | |
4110 | ||
4111 | do { | |
4112 | dir = tmp + strspn(tmp, "/"); | |
4113 | tmp = dir + strcspn(dir, "/"); | |
4114 | makeme = strndup(orig, dir - orig); | |
4115 | if (!makeme) | |
4116 | return false; | |
4117 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
4118 | fprintf(stderr, "failed to create directory '%s': %s", | |
4119 | makeme, strerror(errno)); | |
4120 | free(makeme); | |
4121 | return false; | |
4122 | } | |
4123 | free(makeme); | |
4124 | } while(tmp != dir); | |
4125 | ||
4126 | return true; | |
4127 | } | |
4128 | ||
4129 | static bool umount_if_mounted(void) | |
4130 | { | |
4131 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
480262c9 | 4132 | fprintf(stderr, "failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
4133 | return false; |
4134 | } | |
4135 | return true; | |
4136 | } | |
4137 | ||
4138 | static int pivot_enter(void) | |
4139 | { | |
4140 | int ret = -1, oldroot = -1, newroot = -1; | |
4141 | ||
4142 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
4143 | if (oldroot < 0) { | |
4144 | fprintf(stderr, "%s: Failed to open old root for fchdir.\n", __func__); | |
4145 | return ret; | |
4146 | } | |
4147 | ||
4148 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
4149 | if (newroot < 0) { | |
4150 | fprintf(stderr, "%s: Failed to open new root for fchdir.\n", __func__); | |
4151 | goto err; | |
4152 | } | |
4153 | ||
4154 | /* change into new root fs */ | |
4155 | if (fchdir(newroot) < 0) { | |
4156 | fprintf(stderr, "%s: Failed to change directory to new rootfs: %s.\n", __func__, ROOTDIR); | |
4157 | goto err; | |
4158 | } | |
4159 | ||
4160 | /* pivot_root into our new root fs */ | |
4161 | if (pivot_root(".", ".") < 0) { | |
4162 | fprintf(stderr, "%s: pivot_root() syscall failed: %s.\n", __func__, strerror(errno)); | |
4163 | goto err; | |
4164 | } | |
4165 | ||
4166 | /* | |
4167 | * At this point the old-root is mounted on top of our new-root. | |
4168 | * To unmounted it we must not be chdir'd into it, so escape back | |
4169 | * to the old-root. | |
4170 | */ | |
4171 | if (fchdir(oldroot) < 0) { | |
4172 | fprintf(stderr, "%s: Failed to enter old root.\n", __func__); | |
4173 | goto err; | |
4174 | } | |
4175 | if (umount2(".", MNT_DETACH) < 0) { | |
4176 | fprintf(stderr, "%s: Failed to detach old root.\n", __func__); | |
4177 | goto err; | |
4178 | } | |
4179 | ||
4180 | if (fchdir(newroot) < 0) { | |
4181 | fprintf(stderr, "%s: Failed to re-enter new root.\n", __func__); | |
4182 | goto err; | |
4183 | } | |
4184 | ||
4185 | ret = 0; | |
4186 | ||
4187 | err: | |
4188 | if (oldroot > 0) | |
4189 | close(oldroot); | |
4190 | if (newroot > 0) | |
4191 | close(newroot); | |
4192 | return ret; | |
4193 | } | |
4194 | ||
4195 | /* Prepare our new clean root. */ | |
4196 | static int pivot_prepare(void) | |
4197 | { | |
4198 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
4199 | fprintf(stderr, "%s: Failed to create directory for new root.\n", __func__); | |
4200 | return -1; | |
4201 | } | |
4202 | ||
4203 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
4204 | fprintf(stderr, "%s: Failed to bind-mount / for new root: %s.\n", __func__, strerror(errno)); | |
4205 | return -1; | |
4206 | } | |
4207 | ||
4208 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
4209 | fprintf(stderr, "%s: Failed to bind-mount /run into new root: %s.\n", __func__, strerror(errno)); | |
4210 | return -1; | |
4211 | } | |
4212 | ||
4213 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
4214 | printf("%s: failed to move " BASEDIR " into new root: %s.\n", __func__, strerror(errno)); | |
4215 | return -1; | |
4216 | } | |
4217 | ||
4218 | return 0; | |
4219 | } | |
4220 | ||
4221 | static bool pivot_new_root(void) | |
4222 | { | |
4223 | /* Prepare new root. */ | |
4224 | if (pivot_prepare() < 0) | |
4225 | return false; | |
4226 | ||
4227 | /* Pivot into new root. */ | |
4228 | if (pivot_enter() < 0) | |
4229 | return false; | |
4230 | ||
4231 | return true; | |
4232 | } | |
4233 | ||
4234 | static bool setup_cgfs_dir(void) | |
4235 | { | |
4236 | if (!mkdir_p(BASEDIR, 0700)) { | |
480262c9 | 4237 | fprintf(stderr, "Failed to create lxcfs cgroup mountpoint.\n"); |
29a73c2f CB |
4238 | return false; |
4239 | } | |
480262c9 | 4240 | |
29a73c2f | 4241 | if (!umount_if_mounted()) { |
480262c9 CB |
4242 | fprintf(stderr, "Failed to clean up old lxcfs cgroup mountpoint.\n"); |
4243 | return false; | |
4244 | } | |
4245 | ||
4246 | if (unshare(CLONE_NEWNS) < 0) { | |
4247 | fprintf(stderr, "%s: Failed to unshare mount namespace: %s.\n", __func__, strerror(errno)); | |
4248 | return false; | |
4249 | } | |
4250 | ||
4251 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { | |
4252 | fprintf(stderr, "%s: Failed to remount / private: %s.\n", __func__, strerror(errno)); | |
29a73c2f CB |
4253 | return false; |
4254 | } | |
480262c9 | 4255 | |
29a73c2f | 4256 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
480262c9 | 4257 | fprintf(stderr, "Failed to mount tmpfs over lxcfs cgroup mountpoint.\n"); |
29a73c2f CB |
4258 | return false; |
4259 | } | |
480262c9 | 4260 | |
29a73c2f CB |
4261 | return true; |
4262 | } | |
4263 | ||
4264 | static bool do_mount_cgroups(void) | |
4265 | { | |
4266 | char *target; | |
4267 | size_t clen, len; | |
4268 | int i, ret; | |
4269 | ||
4270 | for (i = 0; i < num_hierarchies; i++) { | |
4271 | char *controller = hierarchies[i]; | |
4272 | clen = strlen(controller); | |
4273 | len = strlen(BASEDIR) + clen + 2; | |
4274 | target = malloc(len); | |
4275 | if (!target) | |
4276 | return false; | |
4277 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); | |
4278 | if (ret < 0 || ret >= len) { | |
4279 | free(target); | |
4280 | return false; | |
4281 | } | |
4282 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
4283 | free(target); | |
4284 | return false; | |
4285 | } | |
4286 | if (mount(controller, target, "cgroup", 0, controller) < 0) { | |
4287 | fprintf(stderr, "Failed mounting cgroup %s\n", controller); | |
4288 | free(target); | |
4289 | return false; | |
4290 | } | |
4291 | ||
4292 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
4293 | if (fd_hierarchies[i] < 0) { | |
4294 | free(target); | |
4295 | return false; | |
4296 | } | |
4297 | free(target); | |
4298 | } | |
4299 | return true; | |
4300 | } | |
4301 | ||
480262c9 | 4302 | static bool cgfs_setup_controllers(void) |
29a73c2f | 4303 | { |
480262c9 | 4304 | if (!setup_cgfs_dir()) |
29a73c2f | 4305 | return false; |
29a73c2f CB |
4306 | |
4307 | if (!do_mount_cgroups()) { | |
480262c9 | 4308 | fprintf(stderr, "Failed to set up private lxcfs cgroup mounts.\n"); |
29a73c2f CB |
4309 | return false; |
4310 | } | |
4311 | ||
4312 | if (!pivot_new_root()) | |
4313 | return false; | |
4314 | ||
4315 | return true; | |
4316 | } | |
4317 | ||
480262c9 CB |
4318 | static int preserve_ns(int pid) |
4319 | { | |
4320 | int ret; | |
4321 | size_t len = 5 /* /proc */ + 21 /* /int_as_str */ + 7 /* /ns/mnt */ + 1 /* \0 */; | |
4322 | char path[len]; | |
4323 | ||
4324 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
4325 | if (ret < 0 || (size_t)ret >= len) | |
4326 | return -1; | |
4327 | ||
4328 | return open(path, O_RDONLY | O_CLOEXEC); | |
4329 | } | |
4330 | ||
29a73c2f | 4331 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) |
237e200e SH |
4332 | { |
4333 | FILE *f; | |
4334 | char *line = NULL; | |
4335 | size_t len = 0; | |
480262c9 | 4336 | int i, init_ns = -1; |
237e200e SH |
4337 | |
4338 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
4339 | fprintf(stderr, "Error opening /proc/self/cgroup: %s\n", strerror(errno)); | |
4340 | return; | |
4341 | } | |
4342 | while (getline(&line, &len, f) != -1) { | |
4343 | char *p, *p2; | |
4344 | ||
4345 | p = strchr(line, ':'); | |
4346 | if (!p) | |
4347 | goto out; | |
4348 | *(p++) = '\0'; | |
4349 | ||
4350 | p2 = strrchr(p, ':'); | |
4351 | if (!p2) | |
4352 | goto out; | |
4353 | *p2 = '\0'; | |
4354 | ||
a67719f6 CB |
4355 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
4356 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
4357 | * because it parses out the empty string "" and later on passes | |
4358 | * it to mount(). Let's skip such entries. | |
4359 | */ | |
4360 | if (!strcmp(p, "")) | |
4361 | continue; | |
4362 | ||
237e200e SH |
4363 | if (!store_hierarchy(line, p)) |
4364 | goto out; | |
4365 | } | |
4366 | ||
480262c9 CB |
4367 | /* Preserve initial namespace. */ |
4368 | init_ns = preserve_ns(getpid()); | |
4369 | if (init_ns < 0) | |
4370 | goto out; | |
4371 | ||
4372 | fd_hierarchies = malloc(sizeof(int *) * num_hierarchies); | |
29a73c2f CB |
4373 | if (!fd_hierarchies) |
4374 | goto out; | |
4375 | ||
480262c9 CB |
4376 | for (i = 0; i < num_hierarchies; i++) |
4377 | fd_hierarchies[i] = -1; | |
4378 | ||
4379 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace | |
4380 | * to privately mount lxcfs cgroups. */ | |
4381 | if (!cgfs_setup_controllers()) | |
29a73c2f | 4382 | goto out; |
480262c9 CB |
4383 | |
4384 | if (setns(init_ns, 0) < 0) | |
29a73c2f | 4385 | goto out; |
29a73c2f | 4386 | |
237e200e SH |
4387 | print_subsystems(); |
4388 | ||
4389 | out: | |
4390 | free(line); | |
4391 | fclose(f); | |
480262c9 CB |
4392 | if (init_ns >= 0) |
4393 | close(init_ns); | |
237e200e SH |
4394 | } |
4395 | ||
4396 | static void __attribute__((destructor)) free_subsystems(void) | |
4397 | { | |
4398 | int i; | |
4399 | ||
29a73c2f | 4400 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
4401 | if (hierarchies[i]) |
4402 | free(hierarchies[i]); | |
480262c9 | 4403 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
4404 | close(fd_hierarchies[i]); |
4405 | } | |
237e200e | 4406 | free(hierarchies); |
480262c9 | 4407 | free(fd_hierarchies); |
237e200e | 4408 | } |