]>
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 | ||
237e200e SH |
51 | enum { |
52 | LXC_TYPE_CGDIR, | |
53 | LXC_TYPE_CGFILE, | |
54 | LXC_TYPE_PROC_MEMINFO, | |
55 | LXC_TYPE_PROC_CPUINFO, | |
56 | LXC_TYPE_PROC_UPTIME, | |
57 | LXC_TYPE_PROC_STAT, | |
58 | LXC_TYPE_PROC_DISKSTATS, | |
70dcc12e | 59 | LXC_TYPE_PROC_SWAPS, |
237e200e SH |
60 | }; |
61 | ||
62 | struct file_info { | |
63 | char *controller; | |
64 | char *cgroup; | |
65 | char *file; | |
66 | int type; | |
67 | char *buf; // unused as of yet | |
68 | int buflen; | |
69 | int size; //actual data size | |
70 | int cached; | |
71 | }; | |
72 | ||
73 | /* reserve buffer size, for cpuall in /proc/stat */ | |
74 | #define BUF_RESERVE_SIZE 256 | |
75 | ||
76 | /* | |
77 | * A table caching which pid is init for a pid namespace. | |
78 | * When looking up which pid is init for $qpid, we first | |
79 | * 1. Stat /proc/$qpid/ns/pid. | |
80 | * 2. Check whether the ino_t is in our store. | |
81 | * a. if not, fork a child in qpid's ns to send us | |
82 | * ucred.pid = 1, and read the initpid. Cache | |
83 | * initpid and creation time for /proc/initpid | |
84 | * in a new store entry. | |
85 | * b. if so, verify that /proc/initpid still matches | |
86 | * what we have saved. If not, clear the store | |
87 | * entry and go back to a. If so, return the | |
88 | * cached initpid. | |
89 | */ | |
90 | struct pidns_init_store { | |
91 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
92 | pid_t initpid; // the pid of nit in that ns | |
93 | long int ctime; // the time at which /proc/$initpid was created | |
94 | struct pidns_init_store *next; | |
95 | long int lastcheck; | |
96 | }; | |
97 | ||
98 | /* lol - look at how they are allocated in the kernel */ | |
99 | #define PIDNS_HASH_SIZE 4096 | |
100 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
101 | ||
102 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
103 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
104 | static void lock_mutex(pthread_mutex_t *l) | |
105 | { | |
106 | int ret; | |
107 | ||
108 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
109 | fprintf(stderr, "pthread_mutex_lock returned:%d %s\n", ret, strerror(ret)); | |
110 | exit(1); | |
111 | } | |
112 | } | |
113 | ||
29a73c2f CB |
114 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
115 | * Number of hierarchies mounted. */ | |
116 | static int num_hierarchies; | |
117 | ||
118 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
119 | * Hierachies mounted {cpuset, blkio, ...}: | |
120 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
121 | static char **hierarchies; | |
122 | ||
123 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
124 | * Open file descriptors: | |
125 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
126 | * private mount namespace. | |
127 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
128 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
129 | * mounts and respective files in the private namespace even when located in | |
130 | * another namespace using the *at() family of functions | |
131 | * {openat(), fchownat(), ...}. */ | |
132 | static int *fd_hierarchies; | |
133 | ||
237e200e SH |
134 | static void unlock_mutex(pthread_mutex_t *l) |
135 | { | |
136 | int ret; | |
137 | ||
138 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
139 | fprintf(stderr, "pthread_mutex_unlock returned:%d %s\n", ret, strerror(ret)); | |
140 | exit(1); | |
141 | } | |
142 | } | |
143 | ||
144 | static void store_lock(void) | |
145 | { | |
146 | lock_mutex(&pidns_store_mutex); | |
147 | } | |
148 | ||
149 | static void store_unlock(void) | |
150 | { | |
151 | unlock_mutex(&pidns_store_mutex); | |
152 | } | |
153 | ||
154 | /* Must be called under store_lock */ | |
155 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
156 | { | |
157 | struct stat initsb; | |
158 | char fnam[100]; | |
159 | ||
160 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
161 | if (stat(fnam, &initsb) < 0) | |
162 | return false; | |
163 | #if DEBUG | |
164 | fprintf(stderr, "comparing ctime %ld %ld for pid %d\n", | |
165 | e->ctime, initsb.st_ctime, e->initpid); | |
166 | #endif | |
167 | if (e->ctime != initsb.st_ctime) | |
168 | return false; | |
169 | return true; | |
170 | } | |
171 | ||
172 | /* Must be called under store_lock */ | |
173 | static void remove_initpid(struct pidns_init_store *e) | |
174 | { | |
175 | struct pidns_init_store *tmp; | |
176 | int h; | |
177 | ||
178 | #if DEBUG | |
179 | fprintf(stderr, "remove_initpid: removing entry for %d\n", e->initpid); | |
180 | #endif | |
181 | h = HASH(e->ino); | |
182 | if (pidns_hash_table[h] == e) { | |
183 | pidns_hash_table[h] = e->next; | |
184 | free(e); | |
185 | return; | |
186 | } | |
187 | ||
188 | tmp = pidns_hash_table[h]; | |
189 | while (tmp) { | |
190 | if (tmp->next == e) { | |
191 | tmp->next = e->next; | |
192 | free(e); | |
193 | return; | |
194 | } | |
195 | tmp = tmp->next; | |
196 | } | |
197 | } | |
198 | ||
199 | #define PURGE_SECS 5 | |
200 | /* Must be called under store_lock */ | |
201 | static void prune_initpid_store(void) | |
202 | { | |
203 | static long int last_prune = 0; | |
204 | struct pidns_init_store *e, *prev, *delme; | |
205 | long int now, threshold; | |
206 | int i; | |
207 | ||
208 | if (!last_prune) { | |
209 | last_prune = time(NULL); | |
210 | return; | |
211 | } | |
212 | now = time(NULL); | |
213 | if (now < last_prune + PURGE_SECS) | |
214 | return; | |
215 | #if DEBUG | |
216 | fprintf(stderr, "pruning\n"); | |
217 | #endif | |
218 | last_prune = now; | |
219 | threshold = now - 2 * PURGE_SECS; | |
220 | ||
221 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
222 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
223 | if (e->lastcheck < threshold) { | |
224 | #if DEBUG | |
225 | fprintf(stderr, "Removing cached entry for %d\n", e->initpid); | |
226 | #endif | |
227 | delme = e; | |
228 | if (prev) | |
229 | prev->next = e->next; | |
230 | else | |
231 | pidns_hash_table[i] = e->next; | |
232 | e = e->next; | |
233 | free(delme); | |
234 | } else { | |
235 | prev = e; | |
236 | e = e->next; | |
237 | } | |
238 | } | |
239 | } | |
240 | } | |
241 | ||
242 | /* Must be called under store_lock */ | |
243 | static void save_initpid(struct stat *sb, pid_t pid) | |
244 | { | |
245 | struct pidns_init_store *e; | |
246 | char fpath[100]; | |
247 | struct stat procsb; | |
248 | int h; | |
249 | ||
250 | #if DEBUG | |
251 | fprintf(stderr, "save_initpid: adding entry for %d\n", pid); | |
252 | #endif | |
253 | snprintf(fpath, 100, "/proc/%d", pid); | |
254 | if (stat(fpath, &procsb) < 0) | |
255 | return; | |
256 | do { | |
257 | e = malloc(sizeof(*e)); | |
258 | } while (!e); | |
259 | e->ino = sb->st_ino; | |
260 | e->initpid = pid; | |
261 | e->ctime = procsb.st_ctime; | |
262 | h = HASH(e->ino); | |
263 | e->next = pidns_hash_table[h]; | |
264 | e->lastcheck = time(NULL); | |
265 | pidns_hash_table[h] = e; | |
266 | } | |
267 | ||
268 | /* | |
269 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
270 | * entry for the inode number and creation time. Verify that the init pid | |
271 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
272 | * otherwise. | |
273 | * Must be called under store_lock | |
274 | */ | |
275 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
276 | { | |
277 | int h = HASH(sb->st_ino); | |
278 | struct pidns_init_store *e = pidns_hash_table[h]; | |
279 | ||
280 | while (e) { | |
281 | if (e->ino == sb->st_ino) { | |
282 | if (initpid_still_valid(e, sb)) { | |
283 | e->lastcheck = time(NULL); | |
284 | return e; | |
285 | } | |
286 | remove_initpid(e); | |
287 | return NULL; | |
288 | } | |
289 | e = e->next; | |
290 | } | |
291 | ||
292 | return NULL; | |
293 | } | |
294 | ||
0f657ce3 | 295 | static int is_dir(const char *path, int fd) |
237e200e SH |
296 | { |
297 | struct stat statbuf; | |
0f657ce3 | 298 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
299 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
300 | return 1; | |
301 | return 0; | |
302 | } | |
303 | ||
304 | static char *must_copy_string(const char *str) | |
305 | { | |
306 | char *dup = NULL; | |
307 | if (!str) | |
308 | return NULL; | |
309 | do { | |
310 | dup = strdup(str); | |
311 | } while (!dup); | |
312 | ||
313 | return dup; | |
314 | } | |
315 | ||
316 | static inline void drop_trailing_newlines(char *s) | |
317 | { | |
318 | int l; | |
319 | ||
320 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
321 | s[l-1] = '\0'; | |
322 | } | |
323 | ||
324 | #define BATCH_SIZE 50 | |
325 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
326 | { | |
327 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
328 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
329 | ||
330 | if (!*mem || newbatches > oldbatches) { | |
331 | char *tmp; | |
332 | do { | |
333 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
334 | } while (!tmp); | |
335 | *mem = tmp; | |
336 | } | |
337 | } | |
338 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
339 | { | |
340 | size_t newlen = *len + linelen; | |
341 | dorealloc(contents, *len, newlen + 1); | |
342 | memcpy(*contents + *len, line, linelen+1); | |
343 | *len = newlen; | |
344 | } | |
345 | ||
60f2ae53 | 346 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
347 | { |
348 | char *line = NULL; | |
349 | char *contents = NULL; | |
60f2ae53 | 350 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
351 | size_t len = 0, fulllen = 0; |
352 | ssize_t linelen; | |
353 | ||
354 | if (!f) | |
355 | return NULL; | |
356 | ||
357 | while ((linelen = getline(&line, &len, f)) != -1) { | |
358 | append_line(&contents, &fulllen, line, linelen); | |
359 | } | |
360 | fclose(f); | |
361 | ||
362 | if (contents) | |
363 | drop_trailing_newlines(contents); | |
364 | free(line); | |
365 | return contents; | |
366 | } | |
367 | ||
ba59ea09 | 368 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
369 | { |
370 | FILE *f; | |
371 | size_t len, ret; | |
372 | ||
ba59ea09 | 373 | if (!(f = fdopen(fd, "w"))) |
237e200e SH |
374 | return false; |
375 | len = strlen(string); | |
376 | ret = fwrite(string, 1, len, f); | |
377 | if (ret != len) { | |
378 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
379 | fclose(f); | |
380 | return false; | |
381 | } | |
382 | if (fclose(f) < 0) { | |
383 | fprintf(stderr, "Error writing to file: %s\n", strerror(errno)); | |
384 | return false; | |
385 | } | |
386 | return true; | |
387 | } | |
388 | ||
237e200e SH |
389 | struct cgfs_files { |
390 | char *name; | |
391 | uint32_t uid, gid; | |
392 | uint32_t mode; | |
393 | }; | |
394 | ||
0619767c | 395 | #define ALLOC_NUM 20 |
237e200e SH |
396 | static bool store_hierarchy(char *stridx, char *h) |
397 | { | |
0619767c SH |
398 | if (num_hierarchies % ALLOC_NUM == 0) { |
399 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
400 | n *= ALLOC_NUM; | |
401 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c SH |
402 | if (!tmp) { |
403 | fprintf(stderr, "Out of memory\n"); | |
404 | exit(1); | |
405 | } | |
237e200e | 406 | hierarchies = tmp; |
237e200e | 407 | } |
f676eb79 | 408 | |
0619767c | 409 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
410 | return true; |
411 | } | |
412 | ||
413 | static void print_subsystems(void) | |
414 | { | |
415 | int i; | |
416 | ||
cc97d34c | 417 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
418 | for (i = 0; i < num_hierarchies; i++) { |
419 | if (hierarchies[i]) | |
420 | fprintf(stderr, " %d: %s\n", i, hierarchies[i]); | |
421 | } | |
422 | } | |
423 | ||
424 | static bool in_comma_list(const char *needle, const char *haystack) | |
425 | { | |
426 | const char *s = haystack, *e; | |
427 | size_t nlen = strlen(needle); | |
428 | ||
06081b29 | 429 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
430 | if (nlen != e - s) { |
431 | s = e + 1; | |
432 | continue; | |
433 | } | |
434 | if (strncmp(needle, s, nlen) == 0) | |
435 | return true; | |
436 | s = e + 1; | |
437 | } | |
438 | if (strcmp(needle, s) == 0) | |
439 | return true; | |
440 | return false; | |
441 | } | |
442 | ||
443 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
444 | /* Return the mounted controller and store the corresponding open file descriptor |
445 | * referring to the controller mountpoint in the private lxcfs namespace in | |
446 | * @cfd. | |
447 | */ | |
448 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
449 | { |
450 | int i; | |
451 | ||
452 | for (i = 0; i < num_hierarchies; i++) { | |
453 | if (!hierarchies[i]) | |
454 | continue; | |
5dd3e6fd CB |
455 | if (strcmp(hierarchies[i], controller) == 0) { |
456 | *cfd = fd_hierarchies[i]; | |
237e200e | 457 | return hierarchies[i]; |
5dd3e6fd CB |
458 | } |
459 | if (in_comma_list(controller, hierarchies[i])) { | |
460 | *cfd = fd_hierarchies[i]; | |
237e200e | 461 | return hierarchies[i]; |
5dd3e6fd | 462 | } |
237e200e SH |
463 | } |
464 | ||
465 | return NULL; | |
466 | } | |
467 | ||
468 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
469 | const char *value) | |
470 | { | |
ba59ea09 | 471 | int ret, fd, cfd; |
237e200e | 472 | size_t len; |
f5a6d92e | 473 | char *fnam, *tmpc; |
237e200e | 474 | |
f5a6d92e | 475 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
476 | if (!tmpc) |
477 | return false; | |
f5a6d92e CB |
478 | |
479 | /* Make sure we pass a relative path to *at() family of functions. | |
480 | * . + /cgroup + / + file + \0 | |
481 | */ | |
ba59ea09 | 482 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 483 | fnam = alloca(len); |
ba59ea09 CB |
484 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
485 | if (ret < 0 || (size_t)ret >= len) | |
486 | return false; | |
487 | ||
488 | fd = openat(cfd, fnam, O_WRONLY); | |
489 | if (fd < 0) | |
490 | return false; | |
f676eb79 | 491 | |
ba59ea09 | 492 | return write_string(fnam, value, fd); |
237e200e SH |
493 | } |
494 | ||
495 | // Chown all the files in the cgroup directory. We do this when we create | |
496 | // a cgroup on behalf of a user. | |
f23fe717 | 497 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 498 | { |
f23fe717 | 499 | struct dirent *direntp; |
237e200e SH |
500 | char path[MAXPATHLEN]; |
501 | size_t len; | |
502 | DIR *d; | |
f23fe717 | 503 | int fd1, ret; |
237e200e SH |
504 | |
505 | len = strlen(dirname); | |
506 | if (len >= MAXPATHLEN) { | |
507 | fprintf(stderr, "chown_all_cgroup_files: pathname too long: %s\n", dirname); | |
508 | return; | |
509 | } | |
510 | ||
f23fe717 CB |
511 | fd1 = openat(fd, dirname, O_DIRECTORY); |
512 | if (fd1 < 0) | |
513 | return; | |
514 | ||
515 | d = fdopendir(fd1); | |
237e200e SH |
516 | if (!d) { |
517 | fprintf(stderr, "chown_all_cgroup_files: failed to open %s\n", dirname); | |
518 | return; | |
519 | } | |
520 | ||
f23fe717 | 521 | while ((direntp = readdir(d))) { |
237e200e SH |
522 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
523 | continue; | |
524 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
525 | if (ret < 0 || ret >= MAXPATHLEN) { | |
526 | fprintf(stderr, "chown_all_cgroup_files: pathname too long under %s\n", dirname); | |
527 | continue; | |
528 | } | |
f23fe717 | 529 | if (fchownat(fd, path, uid, gid, 0) < 0) |
237e200e SH |
530 | fprintf(stderr, "Failed to chown file %s to %u:%u", path, uid, gid); |
531 | } | |
532 | closedir(d); | |
533 | } | |
534 | ||
535 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
536 | { | |
5dd3e6fd | 537 | int cfd; |
237e200e | 538 | size_t len; |
f5a6d92e | 539 | char *dirnam, *tmpc; |
237e200e | 540 | |
f5a6d92e | 541 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
542 | if (!tmpc) |
543 | return -EINVAL; | |
f5a6d92e CB |
544 | |
545 | /* Make sure we pass a relative path to *at() family of functions. | |
546 | * . + /cg + \0 | |
547 | */ | |
f23fe717 | 548 | len = strlen(cg) + 2; |
237e200e | 549 | dirnam = alloca(len); |
f23fe717 | 550 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 551 | |
f23fe717 | 552 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
553 | return -errno; |
554 | ||
555 | if (uid == 0 && gid == 0) | |
556 | return 0; | |
557 | ||
f23fe717 | 558 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
559 | return -errno; |
560 | ||
f23fe717 | 561 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
562 | |
563 | return 0; | |
564 | } | |
565 | ||
b7672ded | 566 | static bool recursive_rmdir(const char *dirname, int fd) |
237e200e | 567 | { |
b7672ded | 568 | struct dirent *direntp; |
237e200e SH |
569 | DIR *dir; |
570 | bool ret = false; | |
571 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
572 | int dupfd; |
573 | ||
574 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
575 | if (dupfd < 0) | |
576 | return false; | |
237e200e | 577 | |
b7672ded | 578 | dir = fdopendir(dupfd); |
237e200e SH |
579 | if (!dir) { |
580 | #if DEBUG | |
581 | fprintf(stderr, "%s: failed to open %s: %s\n", __func__, dirname, strerror(errno)); | |
582 | #endif | |
583 | return false; | |
584 | } | |
585 | ||
b7672ded | 586 | while ((direntp = readdir(dir))) { |
237e200e SH |
587 | struct stat mystat; |
588 | int rc; | |
589 | ||
590 | if (!direntp) | |
591 | break; | |
592 | ||
593 | if (!strcmp(direntp->d_name, ".") || | |
594 | !strcmp(direntp->d_name, "..")) | |
595 | continue; | |
596 | ||
597 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
598 | if (rc < 0 || rc >= MAXPATHLEN) { | |
599 | fprintf(stderr, "pathname too long\n"); | |
600 | continue; | |
601 | } | |
602 | ||
b7672ded | 603 | ret = fstatat(fd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e SH |
604 | if (ret) { |
605 | #if DEBUG | |
606 | fprintf(stderr, "%s: failed to stat %s: %s\n", __func__, pathname, strerror(errno)); | |
607 | #endif | |
608 | continue; | |
609 | } | |
610 | if (S_ISDIR(mystat.st_mode)) { | |
b7672ded | 611 | if (!recursive_rmdir(pathname, fd)) { |
237e200e SH |
612 | #if DEBUG |
613 | fprintf(stderr, "Error removing %s\n", pathname); | |
614 | #endif | |
615 | } | |
616 | } | |
617 | } | |
618 | ||
619 | ret = true; | |
620 | if (closedir(dir) < 0) { | |
621 | fprintf(stderr, "%s: failed to close directory %s: %s\n", __func__, dirname, strerror(errno)); | |
622 | ret = false; | |
623 | } | |
624 | ||
b7672ded | 625 | if (unlinkat(fd, dirname, AT_REMOVEDIR) < 0) { |
237e200e SH |
626 | #if DEBUG |
627 | fprintf(stderr, "%s: failed to delete %s: %s\n", __func__, dirname, strerror(errno)); | |
628 | #endif | |
629 | ret = false; | |
630 | } | |
b7672ded | 631 | close(fd); |
237e200e SH |
632 | |
633 | return ret; | |
634 | } | |
635 | ||
636 | bool cgfs_remove(const char *controller, const char *cg) | |
637 | { | |
b7672ded | 638 | int fd, cfd; |
237e200e | 639 | size_t len; |
f5a6d92e | 640 | char *dirnam, *tmpc; |
237e200e | 641 | |
f5a6d92e | 642 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
643 | if (!tmpc) |
644 | return false; | |
f5a6d92e CB |
645 | |
646 | /* Make sure we pass a relative path to *at() family of functions. | |
647 | * . + /cg + \0 | |
648 | */ | |
b7672ded | 649 | len = strlen(cg) + 2; |
237e200e | 650 | dirnam = alloca(len); |
b7672ded CB |
651 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
652 | ||
653 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
654 | if (fd < 0) | |
655 | return false; | |
656 | ||
657 | return recursive_rmdir(dirnam, fd); | |
237e200e SH |
658 | } |
659 | ||
660 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
661 | { | |
5dd3e6fd | 662 | int cfd; |
237e200e | 663 | size_t len; |
f5a6d92e | 664 | char *pathname, *tmpc; |
237e200e | 665 | |
f5a6d92e | 666 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
667 | if (!tmpc) |
668 | return false; | |
f5a6d92e CB |
669 | |
670 | /* Make sure we pass a relative path to *at() family of functions. | |
671 | * . + /file + \0 | |
672 | */ | |
534690b4 | 673 | len = strlen(file) + 2; |
237e200e | 674 | pathname = alloca(len); |
534690b4 CB |
675 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
676 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
677 | return false; |
678 | return true; | |
679 | } | |
680 | ||
0f657ce3 | 681 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
682 | { |
683 | size_t len; | |
684 | char *fname; | |
685 | ||
686 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
687 | fname = alloca(len); | |
688 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 689 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
690 | return -errno; |
691 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 692 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
693 | return -errno; |
694 | return 0; | |
695 | } | |
696 | ||
697 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
698 | { | |
5dd3e6fd | 699 | int cfd; |
237e200e | 700 | size_t len; |
f5a6d92e | 701 | char *pathname, *tmpc; |
237e200e | 702 | |
f5a6d92e | 703 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
704 | if (!tmpc) |
705 | return -EINVAL; | |
f5a6d92e CB |
706 | |
707 | /* Make sure we pass a relative path to *at() family of functions. | |
708 | * . + /file + \0 | |
709 | */ | |
0f657ce3 | 710 | len = strlen(file) + 2; |
237e200e | 711 | pathname = alloca(len); |
0f657ce3 CB |
712 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
713 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
714 | return -errno; |
715 | ||
0f657ce3 | 716 | if (is_dir(pathname, cfd)) |
237e200e | 717 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 718 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
719 | |
720 | return 0; | |
721 | } | |
722 | ||
723 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
724 | { | |
3ffd08ee | 725 | int fd, cfd; |
237e200e | 726 | size_t len; |
f5a6d92e | 727 | char *pathname, *tmpc; |
237e200e | 728 | |
f5a6d92e | 729 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
730 | if (!tmpc) |
731 | return NULL; | |
f5a6d92e CB |
732 | |
733 | /* Make sure we pass a relative path to *at() family of functions. | |
734 | * . + /cgroup + / "cgroup.procs" + \0 | |
735 | */ | |
3ffd08ee | 736 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 737 | pathname = alloca(len); |
3ffd08ee CB |
738 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
739 | ||
740 | fd = openat(cfd, pathname, O_WRONLY); | |
741 | if (fd < 0) | |
742 | return NULL; | |
743 | ||
744 | return fdopen(fd, "w"); | |
237e200e SH |
745 | } |
746 | ||
f366da65 WB |
747 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
748 | void ***list, size_t typesize, | |
749 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 750 | { |
4ea38a4c | 751 | int cfd, fd, ret; |
237e200e | 752 | size_t len; |
4ea38a4c | 753 | char *cg, *tmpc; |
237e200e | 754 | char pathname[MAXPATHLEN]; |
f366da65 | 755 | size_t sz = 0, asz = 0; |
4ea38a4c | 756 | struct dirent *dirent; |
237e200e | 757 | DIR *dir; |
237e200e | 758 | |
4ea38a4c | 759 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 760 | *list = NULL; |
237e200e | 761 | if (!tmpc) |
e97c834b | 762 | return false; |
237e200e | 763 | |
f5a6d92e | 764 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
765 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
766 | cg = alloca(len); | |
767 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
768 | if (ret < 0 || (size_t)ret >= len) { | |
769 | fprintf(stderr, "%s: pathname too long under %s\n", __func__, cgroup); | |
770 | return false; | |
771 | } | |
237e200e | 772 | |
4ea38a4c CB |
773 | fd = openat(cfd, cg, O_DIRECTORY); |
774 | if (fd < 0) | |
775 | return false; | |
776 | ||
777 | dir = fdopendir(fd); | |
237e200e SH |
778 | if (!dir) |
779 | return false; | |
780 | ||
4ea38a4c | 781 | while ((dirent = readdir(dir))) { |
237e200e | 782 | struct stat mystat; |
237e200e | 783 | |
4ea38a4c CB |
784 | if (!strcmp(dirent->d_name, ".") || |
785 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
786 | continue; |
787 | ||
4ea38a4c CB |
788 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
789 | if (ret < 0 || ret >= MAXPATHLEN) { | |
790 | fprintf(stderr, "%s: pathname too long under %s\n", __func__, cg); | |
237e200e SH |
791 | continue; |
792 | } | |
793 | ||
4ea38a4c | 794 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e SH |
795 | if (ret) { |
796 | fprintf(stderr, "%s: failed to stat %s: %s\n", __func__, pathname, strerror(errno)); | |
797 | continue; | |
798 | } | |
f366da65 WB |
799 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
800 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
801 | continue; |
802 | ||
803 | if (sz+2 >= asz) { | |
f366da65 | 804 | void **tmp; |
237e200e SH |
805 | asz += BATCH_SIZE; |
806 | do { | |
f366da65 | 807 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
808 | } while (!tmp); |
809 | *list = tmp; | |
810 | } | |
4ea38a4c | 811 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
812 | (*list)[sz+1] = NULL; |
813 | sz++; | |
814 | } | |
815 | if (closedir(dir) < 0) { | |
4ea38a4c | 816 | fprintf(stderr, "%s: failed closedir for %s: %s\n", __func__, cgroup, strerror(errno)); |
237e200e SH |
817 | return false; |
818 | } | |
819 | return true; | |
820 | } | |
821 | ||
f366da65 WB |
822 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
823 | { | |
824 | char *dup; | |
825 | do { | |
826 | dup = strdup(dir_entry); | |
827 | } while (!dup); | |
828 | return dup; | |
829 | } | |
830 | ||
831 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
832 | { | |
833 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
834 | } | |
835 | ||
237e200e SH |
836 | void free_key(struct cgfs_files *k) |
837 | { | |
838 | if (!k) | |
839 | return; | |
840 | free(k->name); | |
841 | free(k); | |
842 | } | |
843 | ||
844 | void free_keys(struct cgfs_files **keys) | |
845 | { | |
846 | int i; | |
847 | ||
848 | if (!keys) | |
849 | return; | |
850 | for (i = 0; keys[i]; i++) { | |
851 | free_key(keys[i]); | |
852 | } | |
853 | free(keys); | |
854 | } | |
855 | ||
856 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
857 | { | |
60f2ae53 | 858 | int ret, fd, cfd; |
237e200e | 859 | size_t len; |
f5a6d92e | 860 | char *fnam, *tmpc; |
237e200e | 861 | |
f5a6d92e | 862 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
863 | if (!tmpc) |
864 | return false; | |
f5a6d92e CB |
865 | |
866 | /* Make sure we pass a relative path to *at() family of functions. | |
867 | * . + /cgroup + / + file + \0 | |
868 | */ | |
60f2ae53 | 869 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 870 | fnam = alloca(len); |
60f2ae53 CB |
871 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
872 | if (ret < 0 || (size_t)ret >= len) | |
873 | return NULL; | |
874 | ||
875 | fd = openat(cfd, fnam, O_RDONLY); | |
876 | if (fd < 0) | |
877 | return NULL; | |
237e200e | 878 | |
60f2ae53 | 879 | *value = slurp_file(fnam, fd); |
237e200e SH |
880 | return *value != NULL; |
881 | } | |
882 | ||
883 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
884 | { | |
4ea38a4c | 885 | int ret, cfd; |
237e200e | 886 | size_t len; |
f5a6d92e | 887 | char *fnam, *tmpc; |
237e200e SH |
888 | struct stat sb; |
889 | struct cgfs_files *newkey; | |
237e200e | 890 | |
f5a6d92e | 891 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
892 | if (!tmpc) |
893 | return false; | |
894 | ||
895 | if (file && *file == '/') | |
896 | file++; | |
897 | ||
06081b29 | 898 | if (file && strchr(file, '/')) |
237e200e SH |
899 | return NULL; |
900 | ||
f5a6d92e CB |
901 | /* Make sure we pass a relative path to *at() family of functions. |
902 | * . + /cgroup + / + file + \0 | |
903 | */ | |
4ea38a4c | 904 | len = strlen(cgroup) + 3; |
237e200e SH |
905 | if (file) |
906 | len += strlen(file) + 1; | |
907 | fnam = alloca(len); | |
4ea38a4c CB |
908 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
909 | file ? "/" : "", file ? file : ""); | |
237e200e | 910 | |
4ea38a4c | 911 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
912 | if (ret < 0) |
913 | return NULL; | |
914 | ||
915 | do { | |
916 | newkey = malloc(sizeof(struct cgfs_files)); | |
917 | } while (!newkey); | |
918 | if (file) | |
919 | newkey->name = must_copy_string(file); | |
06081b29 CB |
920 | else if (strrchr(cgroup, '/')) |
921 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
922 | else |
923 | newkey->name = must_copy_string(cgroup); | |
924 | newkey->uid = sb.st_uid; | |
925 | newkey->gid = sb.st_gid; | |
926 | newkey->mode = sb.st_mode; | |
927 | ||
928 | return newkey; | |
929 | } | |
930 | ||
f366da65 | 931 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 932 | { |
f366da65 WB |
933 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
934 | if (!entry) { | |
935 | fprintf(stderr, "%s: Error getting files under %s:%s\n", | |
936 | __func__, controller, cgroup); | |
237e200e | 937 | } |
f366da65 WB |
938 | return entry; |
939 | } | |
940 | ||
941 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
942 | { | |
943 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
944 | } |
945 | ||
946 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
947 | { |
948 | int cfd; | |
949 | size_t len; | |
f5a6d92e | 950 | char *fnam, *tmpc; |
237e200e SH |
951 | int ret; |
952 | struct stat sb; | |
953 | ||
f5a6d92e | 954 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
955 | if (!tmpc) |
956 | return false; | |
f5a6d92e CB |
957 | |
958 | /* Make sure we pass a relative path to *at() family of functions. | |
959 | * . + /cgroup + / + f + \0 | |
960 | */ | |
d04232f2 | 961 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 962 | fnam = alloca(len); |
d04232f2 CB |
963 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
964 | if (ret < 0 || (size_t)ret >= len) | |
965 | return false; | |
237e200e | 966 | |
d04232f2 | 967 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
968 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
969 | return false; | |
f5a6d92e | 970 | |
237e200e SH |
971 | return true; |
972 | } | |
973 | ||
974 | #define SEND_CREDS_OK 0 | |
975 | #define SEND_CREDS_NOTSK 1 | |
976 | #define SEND_CREDS_FAIL 2 | |
977 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
978 | static int wait_for_pid(pid_t pid); | |
979 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 980 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
981 | |
982 | /* | |
b10bdd6c | 983 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
984 | * over a unix sock so we can read the task's reaper's pid in our |
985 | * namespace | |
b10bdd6c FG |
986 | * |
987 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
988 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
989 | * the pidns and the parent pid outside are identical. Using clone prevents | |
990 | * this issue. | |
237e200e SH |
991 | */ |
992 | static void write_task_init_pid_exit(int sock, pid_t target) | |
993 | { | |
237e200e SH |
994 | char fnam[100]; |
995 | pid_t pid; | |
237e200e | 996 | int fd, ret; |
b10bdd6c FG |
997 | size_t stack_size = sysconf(_SC_PAGESIZE); |
998 | void *stack = alloca(stack_size); | |
237e200e SH |
999 | |
1000 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1001 | if (ret < 0 || ret >= sizeof(fnam)) | |
1002 | _exit(1); | |
1003 | ||
1004 | fd = open(fnam, O_RDONLY); | |
1005 | if (fd < 0) { | |
1006 | perror("write_task_init_pid_exit open of ns/pid"); | |
1007 | _exit(1); | |
1008 | } | |
1009 | if (setns(fd, 0)) { | |
1010 | perror("write_task_init_pid_exit setns 1"); | |
1011 | close(fd); | |
1012 | _exit(1); | |
1013 | } | |
b10bdd6c | 1014 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1015 | if (pid < 0) |
1016 | _exit(1); | |
1017 | if (pid != 0) { | |
1018 | if (!wait_for_pid(pid)) | |
1019 | _exit(1); | |
1020 | _exit(0); | |
1021 | } | |
b10bdd6c FG |
1022 | } |
1023 | ||
1024 | static int send_creds_clone_wrapper(void *arg) { | |
1025 | struct ucred cred; | |
1026 | char v; | |
1027 | int sock = *(int *)arg; | |
237e200e SH |
1028 | |
1029 | /* we are the child */ | |
1030 | cred.uid = 0; | |
1031 | cred.gid = 0; | |
1032 | cred.pid = 1; | |
1033 | v = '1'; | |
1034 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1035 | return 1; |
1036 | return 0; | |
237e200e SH |
1037 | } |
1038 | ||
1039 | static pid_t get_init_pid_for_task(pid_t task) | |
1040 | { | |
1041 | int sock[2]; | |
1042 | pid_t pid; | |
1043 | pid_t ret = -1; | |
1044 | char v = '0'; | |
1045 | struct ucred cred; | |
1046 | ||
1047 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1048 | perror("socketpair"); | |
1049 | return -1; | |
1050 | } | |
1051 | ||
1052 | pid = fork(); | |
1053 | if (pid < 0) | |
1054 | goto out; | |
1055 | if (!pid) { | |
1056 | close(sock[1]); | |
1057 | write_task_init_pid_exit(sock[0], task); | |
1058 | _exit(0); | |
1059 | } | |
1060 | ||
1061 | if (!recv_creds(sock[1], &cred, &v)) | |
1062 | goto out; | |
1063 | ret = cred.pid; | |
1064 | ||
1065 | out: | |
1066 | close(sock[0]); | |
1067 | close(sock[1]); | |
1068 | if (pid > 0) | |
1069 | wait_for_pid(pid); | |
1070 | return ret; | |
1071 | } | |
1072 | ||
1073 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1074 | { | |
1075 | pid_t answer = 0; | |
1076 | struct stat sb; | |
1077 | struct pidns_init_store *e; | |
1078 | char fnam[100]; | |
1079 | ||
1080 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1081 | store_lock(); | |
1082 | if (stat(fnam, &sb) < 0) | |
1083 | goto out; | |
1084 | e = lookup_verify_initpid(&sb); | |
1085 | if (e) { | |
1086 | answer = e->initpid; | |
1087 | goto out; | |
1088 | } | |
1089 | answer = get_init_pid_for_task(qpid); | |
1090 | if (answer > 0) | |
1091 | save_initpid(&sb, answer); | |
1092 | ||
1093 | out: | |
1094 | /* we prune at end in case we are returning | |
1095 | * the value we were about to return */ | |
1096 | prune_initpid_store(); | |
1097 | store_unlock(); | |
1098 | return answer; | |
1099 | } | |
1100 | ||
1101 | static int wait_for_pid(pid_t pid) | |
1102 | { | |
1103 | int status, ret; | |
1104 | ||
1105 | if (pid <= 0) | |
1106 | return -1; | |
1107 | ||
1108 | again: | |
1109 | ret = waitpid(pid, &status, 0); | |
1110 | if (ret == -1) { | |
1111 | if (errno == EINTR) | |
1112 | goto again; | |
1113 | return -1; | |
1114 | } | |
1115 | if (ret != pid) | |
1116 | goto again; | |
1117 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1118 | return -1; | |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | ||
1123 | /* | |
1124 | * append pid to *src. | |
1125 | * src: a pointer to a char* in which ot append the pid. | |
1126 | * sz: the number of characters printed so far, minus trailing \0. | |
1127 | * asz: the allocated size so far | |
1128 | * pid: the pid to append | |
1129 | */ | |
1130 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1131 | { | |
1132 | char tmp[30]; | |
1133 | ||
1134 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1135 | ||
1136 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1137 | char *tmp; | |
1138 | do { | |
1139 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1140 | } while (!tmp); | |
1141 | *src = tmp; | |
1142 | *asz += BUF_RESERVE_SIZE; | |
1143 | } | |
bbfd0e33 | 1144 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1145 | *sz += tmplen; |
237e200e SH |
1146 | } |
1147 | ||
1148 | /* | |
1149 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1150 | * valid in the caller's namespace, return the id mapped into | |
1151 | * pid's namespace. | |
1152 | * Returns the mapped id, or -1 on error. | |
1153 | */ | |
1154 | unsigned int | |
1155 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1156 | { | |
1157 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1158 | hostuid, // base id for a range in the caller's namespace | |
1159 | count; // number of ids in this range | |
1160 | char line[400]; | |
1161 | int ret; | |
1162 | ||
1163 | fseek(idfile, 0L, SEEK_SET); | |
1164 | while (fgets(line, 400, idfile)) { | |
1165 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1166 | if (ret != 3) | |
1167 | continue; | |
1168 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1169 | /* | |
1170 | * uids wrapped around - unexpected as this is a procfile, | |
1171 | * so just bail. | |
1172 | */ | |
1173 | fprintf(stderr, "pid wrapparound at entry %u %u %u in %s\n", | |
1174 | nsuid, hostuid, count, line); | |
1175 | return -1; | |
1176 | } | |
1177 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1178 | /* | |
1179 | * now since hostuid <= in_id < hostuid+count, and | |
1180 | * hostuid+count and nsuid+count do not wrap around, | |
1181 | * we know that nsuid+(in_id-hostuid) which must be | |
1182 | * less that nsuid+(count) must not wrap around | |
1183 | */ | |
1184 | return (in_id - hostuid) + nsuid; | |
1185 | } | |
1186 | } | |
1187 | ||
1188 | // no answer found | |
1189 | return -1; | |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * for is_privileged_over, | |
1194 | * specify whether we require the calling uid to be root in his | |
1195 | * namespace | |
1196 | */ | |
1197 | #define NS_ROOT_REQD true | |
1198 | #define NS_ROOT_OPT false | |
1199 | ||
1200 | #define PROCLEN 100 | |
1201 | ||
1202 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1203 | { | |
1204 | char fpath[PROCLEN]; | |
1205 | int ret; | |
1206 | bool answer = false; | |
1207 | uid_t nsuid; | |
1208 | ||
1209 | if (victim == -1 || uid == -1) | |
1210 | return false; | |
1211 | ||
1212 | /* | |
1213 | * If the request is one not requiring root in the namespace, | |
1214 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1215 | * access to files owned by uid 1000 | |
1216 | */ | |
1217 | if (!req_ns_root && uid == victim) | |
1218 | return true; | |
1219 | ||
1220 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1221 | if (ret < 0 || ret >= PROCLEN) | |
1222 | return false; | |
1223 | FILE *f = fopen(fpath, "r"); | |
1224 | if (!f) | |
1225 | return false; | |
1226 | ||
1227 | /* if caller's not root in his namespace, reject */ | |
1228 | nsuid = convert_id_to_ns(f, uid); | |
1229 | if (nsuid) | |
1230 | goto out; | |
1231 | ||
1232 | /* | |
1233 | * If victim is not mapped into caller's ns, reject. | |
1234 | * XXX I'm not sure this check is needed given that fuse | |
1235 | * will be sending requests where the vfs has converted | |
1236 | */ | |
1237 | nsuid = convert_id_to_ns(f, victim); | |
1238 | if (nsuid == -1) | |
1239 | goto out; | |
1240 | ||
1241 | answer = true; | |
1242 | ||
1243 | out: | |
1244 | fclose(f); | |
1245 | return answer; | |
1246 | } | |
1247 | ||
1248 | static bool perms_include(int fmode, mode_t req_mode) | |
1249 | { | |
1250 | mode_t r; | |
1251 | ||
1252 | switch (req_mode & O_ACCMODE) { | |
1253 | case O_RDONLY: | |
1254 | r = S_IROTH; | |
1255 | break; | |
1256 | case O_WRONLY: | |
1257 | r = S_IWOTH; | |
1258 | break; | |
1259 | case O_RDWR: | |
1260 | r = S_IROTH | S_IWOTH; | |
1261 | break; | |
1262 | default: | |
1263 | return false; | |
1264 | } | |
1265 | return ((fmode & r) == r); | |
1266 | } | |
1267 | ||
1268 | ||
1269 | /* | |
1270 | * taskcg is a/b/c | |
1271 | * querycg is /a/b/c/d/e | |
1272 | * we return 'd' | |
1273 | */ | |
1274 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1275 | { | |
1276 | char *start, *end; | |
1277 | ||
1278 | if (strlen(taskcg) <= strlen(querycg)) { | |
1279 | fprintf(stderr, "%s: I was fed bad input\n", __func__); | |
1280 | return NULL; | |
1281 | } | |
1282 | ||
06081b29 | 1283 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1284 | start = strdup(taskcg + 1); |
1285 | else | |
1286 | start = strdup(taskcg + strlen(querycg) + 1); | |
1287 | if (!start) | |
1288 | return NULL; | |
1289 | end = strchr(start, '/'); | |
1290 | if (end) | |
1291 | *end = '\0'; | |
1292 | return start; | |
1293 | } | |
1294 | ||
1295 | static void stripnewline(char *x) | |
1296 | { | |
1297 | size_t l = strlen(x); | |
1298 | if (l && x[l-1] == '\n') | |
1299 | x[l-1] = '\0'; | |
1300 | } | |
1301 | ||
1302 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1303 | { | |
5dd3e6fd | 1304 | int cfd; |
237e200e SH |
1305 | char fnam[PROCLEN]; |
1306 | FILE *f; | |
1307 | char *answer = NULL; | |
1308 | char *line = NULL; | |
1309 | size_t len = 0; | |
1310 | int ret; | |
5dd3e6fd | 1311 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1312 | if (!h) |
1313 | return NULL; | |
1314 | ||
1315 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1316 | if (ret < 0 || ret >= PROCLEN) | |
1317 | return NULL; | |
1318 | if (!(f = fopen(fnam, "r"))) | |
1319 | return NULL; | |
1320 | ||
1321 | while (getline(&line, &len, f) != -1) { | |
1322 | char *c1, *c2; | |
1323 | if (!line[0]) | |
1324 | continue; | |
1325 | c1 = strchr(line, ':'); | |
1326 | if (!c1) | |
1327 | goto out; | |
1328 | c1++; | |
1329 | c2 = strchr(c1, ':'); | |
1330 | if (!c2) | |
1331 | goto out; | |
1332 | *c2 = '\0'; | |
1333 | if (strcmp(c1, h) != 0) | |
1334 | continue; | |
1335 | c2++; | |
1336 | stripnewline(c2); | |
1337 | do { | |
1338 | answer = strdup(c2); | |
1339 | } while (!answer); | |
1340 | break; | |
1341 | } | |
1342 | ||
1343 | out: | |
1344 | fclose(f); | |
1345 | free(line); | |
1346 | return answer; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * check whether a fuse context may access a cgroup dir or file | |
1351 | * | |
1352 | * If file is not null, it is a cgroup file to check under cg. | |
1353 | * If file is null, then we are checking perms on cg itself. | |
1354 | * | |
1355 | * For files we can check the mode of the list_keys result. | |
1356 | * For cgroups, we must make assumptions based on the files under the | |
1357 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1358 | * yet. | |
1359 | */ | |
1360 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1361 | { | |
1362 | struct cgfs_files *k = NULL; | |
1363 | bool ret = false; | |
1364 | ||
1365 | k = cgfs_get_key(contrl, cg, file); | |
1366 | if (!k) | |
1367 | return false; | |
1368 | ||
1369 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1370 | if (perms_include(k->mode >> 6, mode)) { | |
1371 | ret = true; | |
1372 | goto out; | |
1373 | } | |
1374 | } | |
1375 | if (fc->gid == k->gid) { | |
1376 | if (perms_include(k->mode >> 3, mode)) { | |
1377 | ret = true; | |
1378 | goto out; | |
1379 | } | |
1380 | } | |
1381 | ret = perms_include(k->mode, mode); | |
1382 | ||
1383 | out: | |
1384 | free_key(k); | |
1385 | return ret; | |
1386 | } | |
1387 | ||
1388 | #define INITSCOPE "/init.scope" | |
1389 | static void prune_init_slice(char *cg) | |
1390 | { | |
1391 | char *point; | |
1392 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1393 | ||
1394 | if (cg_len < initscope_len) | |
1395 | return; | |
1396 | ||
1397 | point = cg + cg_len - initscope_len; | |
1398 | if (strcmp(point, INITSCOPE) == 0) { | |
1399 | if (point == cg) | |
1400 | *(point+1) = '\0'; | |
1401 | else | |
1402 | *point = '\0'; | |
1403 | } | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1408 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1409 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1410 | * to a string containing the next cgroup directory under cg, which must be | |
1411 | * freed by the caller. | |
1412 | */ | |
1413 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1414 | { | |
1415 | bool answer = false; | |
1416 | char *c2 = get_pid_cgroup(pid, contrl); | |
1417 | char *linecmp; | |
1418 | ||
1419 | if (!c2) | |
1420 | return false; | |
1421 | prune_init_slice(c2); | |
1422 | ||
1423 | /* | |
12c31268 CB |
1424 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1425 | * they pass in a cgroup without leading '/' | |
1426 | * | |
1427 | * The original line here was: | |
1428 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1429 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1430 | * Serge, do you know? | |
237e200e | 1431 | */ |
12c31268 CB |
1432 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1433 | linecmp = c2; | |
1434 | else | |
1435 | linecmp = c2 + 1; | |
237e200e SH |
1436 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1437 | if (nextcg) { | |
1438 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1439 | } | |
1440 | goto out; | |
1441 | } | |
1442 | answer = true; | |
1443 | ||
1444 | out: | |
1445 | free(c2); | |
1446 | return answer; | |
1447 | } | |
1448 | ||
1449 | /* | |
1450 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1451 | */ | |
1452 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1453 | { | |
1454 | bool answer = false; | |
1455 | char *c2, *task_cg; | |
1456 | size_t target_len, task_len; | |
1457 | ||
f7bff426 | 1458 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1459 | return true; |
1460 | ||
1461 | c2 = get_pid_cgroup(pid, contrl); | |
1462 | if (!c2) | |
1463 | return false; | |
1464 | prune_init_slice(c2); | |
1465 | ||
1466 | task_cg = c2 + 1; | |
1467 | target_len = strlen(cg); | |
1468 | task_len = strlen(task_cg); | |
1469 | if (task_len == 0) { | |
1470 | /* Task is in the root cg, it can see everything. This case is | |
1471 | * not handled by the strmcps below, since they test for the | |
1472 | * last /, but that is the first / that we've chopped off | |
1473 | * above. | |
1474 | */ | |
1475 | answer = true; | |
1476 | goto out; | |
1477 | } | |
1478 | if (strcmp(cg, task_cg) == 0) { | |
1479 | answer = true; | |
1480 | goto out; | |
1481 | } | |
1482 | if (target_len < task_len) { | |
1483 | /* looking up a parent dir */ | |
1484 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1485 | answer = true; | |
1486 | goto out; | |
1487 | } | |
1488 | if (target_len > task_len) { | |
1489 | /* looking up a child dir */ | |
1490 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1491 | answer = true; | |
1492 | goto out; | |
1493 | } | |
1494 | ||
1495 | out: | |
1496 | free(c2); | |
1497 | return answer; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * given /cgroup/freezer/a/b, return "freezer". | |
1502 | * the returned char* should NOT be freed. | |
1503 | */ | |
1504 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1505 | { | |
1506 | const char *p1; | |
1507 | char *contr, *slash; | |
1508 | ||
1509 | if (strlen(path) < 9) | |
1510 | return NULL; | |
1511 | if (*(path+7) != '/') | |
1512 | return NULL; | |
1513 | p1 = path+8; | |
1514 | contr = strdupa(p1); | |
1515 | if (!contr) | |
1516 | return NULL; | |
1517 | slash = strstr(contr, "/"); | |
1518 | if (slash) | |
1519 | *slash = '\0'; | |
1520 | ||
1521 | int i; | |
1522 | for (i = 0; i < num_hierarchies; i++) { | |
1523 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) | |
1524 | return hierarchies[i]; | |
1525 | } | |
1526 | return NULL; | |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1531 | * Note that the returned value may include files (keynames) etc | |
1532 | */ | |
1533 | static const char *find_cgroup_in_path(const char *path) | |
1534 | { | |
1535 | const char *p1; | |
1536 | ||
1537 | if (strlen(path) < 9) | |
1538 | return NULL; | |
1539 | p1 = strstr(path+8, "/"); | |
1540 | if (!p1) | |
1541 | return NULL; | |
1542 | return p1+1; | |
1543 | } | |
1544 | ||
1545 | /* | |
1546 | * split the last path element from the path in @cg. | |
1547 | * @dir is newly allocated and should be freed, @last not | |
1548 | */ | |
1549 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1550 | { | |
1551 | char *p; | |
1552 | ||
1553 | do { | |
1554 | *dir = strdup(cg); | |
1555 | } while (!*dir); | |
1556 | *last = strrchr(cg, '/'); | |
1557 | if (!*last) { | |
1558 | *last = NULL; | |
1559 | return; | |
1560 | } | |
1561 | p = strrchr(*dir, '/'); | |
1562 | *p = '\0'; | |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * FUSE ops for /cgroup | |
1567 | */ | |
1568 | ||
1569 | int cg_getattr(const char *path, struct stat *sb) | |
1570 | { | |
1571 | struct timespec now; | |
1572 | struct fuse_context *fc = fuse_get_context(); | |
1573 | char * cgdir = NULL; | |
1574 | char *last = NULL, *path1, *path2; | |
1575 | struct cgfs_files *k = NULL; | |
1576 | const char *cgroup; | |
1577 | const char *controller = NULL; | |
1578 | int ret = -ENOENT; | |
1579 | ||
1580 | ||
1581 | if (!fc) | |
1582 | return -EIO; | |
1583 | ||
1584 | memset(sb, 0, sizeof(struct stat)); | |
1585 | ||
1586 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1587 | return -EINVAL; | |
1588 | ||
1589 | sb->st_uid = sb->st_gid = 0; | |
1590 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1591 | sb->st_size = 0; | |
1592 | ||
1593 | if (strcmp(path, "/cgroup") == 0) { | |
1594 | sb->st_mode = S_IFDIR | 00755; | |
1595 | sb->st_nlink = 2; | |
1596 | return 0; | |
1597 | } | |
1598 | ||
1599 | controller = pick_controller_from_path(fc, path); | |
1600 | if (!controller) | |
1601 | return -EIO; | |
1602 | cgroup = find_cgroup_in_path(path); | |
1603 | if (!cgroup) { | |
1604 | /* this is just /cgroup/controller, return it as a dir */ | |
1605 | sb->st_mode = S_IFDIR | 00755; | |
1606 | sb->st_nlink = 2; | |
1607 | return 0; | |
1608 | } | |
1609 | ||
1610 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1611 | ||
1612 | if (!last) { | |
1613 | path1 = "/"; | |
1614 | path2 = cgdir; | |
1615 | } else { | |
1616 | path1 = cgdir; | |
1617 | path2 = last; | |
1618 | } | |
1619 | ||
1620 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1621 | if (initpid <= 0) | |
1622 | initpid = fc->pid; | |
1623 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1624 | * Then check that caller's cgroup is under path if last is a child | |
1625 | * cgroup, or cgdir if last is a file */ | |
1626 | ||
1627 | if (is_child_cgroup(controller, path1, path2)) { | |
1628 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1629 | ret = -ENOENT; | |
1630 | goto out; | |
1631 | } | |
1632 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1633 | /* this is just /cgroup/controller, return it as a dir */ | |
1634 | sb->st_mode = S_IFDIR | 00555; | |
1635 | sb->st_nlink = 2; | |
1636 | ret = 0; | |
1637 | goto out; | |
1638 | } | |
1639 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1640 | ret = -EACCES; | |
1641 | goto out; | |
1642 | } | |
1643 | ||
1644 | // get uid, gid, from '/tasks' file and make up a mode | |
1645 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1646 | sb->st_mode = S_IFDIR | 00755; | |
1647 | k = cgfs_get_key(controller, cgroup, NULL); | |
1648 | if (!k) { | |
1649 | sb->st_uid = sb->st_gid = 0; | |
1650 | } else { | |
1651 | sb->st_uid = k->uid; | |
1652 | sb->st_gid = k->gid; | |
1653 | } | |
1654 | free_key(k); | |
1655 | sb->st_nlink = 2; | |
1656 | ret = 0; | |
1657 | goto out; | |
1658 | } | |
1659 | ||
1660 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1661 | sb->st_mode = S_IFREG | k->mode; | |
1662 | sb->st_nlink = 1; | |
1663 | sb->st_uid = k->uid; | |
1664 | sb->st_gid = k->gid; | |
1665 | sb->st_size = 0; | |
1666 | free_key(k); | |
1667 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1668 | ret = -ENOENT; | |
1669 | goto out; | |
1670 | } | |
1671 | if (!fc_may_access(fc, controller, path1, path2, O_RDONLY)) { | |
1672 | ret = -EACCES; | |
1673 | goto out; | |
1674 | } | |
1675 | ||
1676 | ret = 0; | |
1677 | } | |
1678 | ||
1679 | out: | |
1680 | free(cgdir); | |
1681 | return ret; | |
1682 | } | |
1683 | ||
1684 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1685 | { | |
1686 | struct fuse_context *fc = fuse_get_context(); | |
1687 | const char *cgroup; | |
1688 | struct file_info *dir_info; | |
1689 | char *controller = NULL; | |
1690 | ||
1691 | if (!fc) | |
1692 | return -EIO; | |
1693 | ||
1694 | if (strcmp(path, "/cgroup") == 0) { | |
1695 | cgroup = NULL; | |
1696 | controller = NULL; | |
1697 | } else { | |
1698 | // return list of keys for the controller, and list of child cgroups | |
1699 | controller = pick_controller_from_path(fc, path); | |
1700 | if (!controller) | |
1701 | return -EIO; | |
1702 | ||
1703 | cgroup = find_cgroup_in_path(path); | |
1704 | if (!cgroup) { | |
1705 | /* this is just /cgroup/controller, return its contents */ | |
1706 | cgroup = "/"; | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1711 | if (initpid <= 0) | |
1712 | initpid = fc->pid; | |
1713 | if (cgroup) { | |
1714 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1715 | return -ENOENT; | |
1716 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1717 | return -EACCES; | |
1718 | } | |
1719 | ||
1720 | /* we'll free this at cg_releasedir */ | |
1721 | dir_info = malloc(sizeof(*dir_info)); | |
1722 | if (!dir_info) | |
1723 | return -ENOMEM; | |
1724 | dir_info->controller = must_copy_string(controller); | |
1725 | dir_info->cgroup = must_copy_string(cgroup); | |
1726 | dir_info->type = LXC_TYPE_CGDIR; | |
1727 | dir_info->buf = NULL; | |
1728 | dir_info->file = NULL; | |
1729 | dir_info->buflen = 0; | |
1730 | ||
1731 | fi->fh = (unsigned long)dir_info; | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1736 | struct fuse_file_info *fi) | |
1737 | { | |
1738 | struct file_info *d = (struct file_info *)fi->fh; | |
1739 | struct cgfs_files **list = NULL; | |
1740 | int i, ret; | |
1741 | char *nextcg = NULL; | |
1742 | struct fuse_context *fc = fuse_get_context(); | |
1743 | char **clist = NULL; | |
1744 | ||
1745 | if (d->type != LXC_TYPE_CGDIR) { | |
1746 | fprintf(stderr, "Internal error: file cache info used in readdir\n"); | |
1747 | return -EIO; | |
1748 | } | |
1749 | if (!d->cgroup && !d->controller) { | |
1750 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1751 | int i; | |
1752 | ||
1753 | for (i = 0; i < num_hierarchies; i++) { | |
1754 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1755 | return -EIO; | |
1756 | } | |
1757 | } | |
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1762 | // not a valid cgroup | |
1763 | ret = -EINVAL; | |
1764 | goto out; | |
1765 | } | |
1766 | ||
1767 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1768 | if (initpid <= 0) | |
1769 | initpid = fc->pid; | |
1770 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1771 | if (nextcg) { | |
1772 | ret = filler(buf, nextcg, NULL, 0); | |
1773 | free(nextcg); | |
1774 | if (ret != 0) { | |
1775 | ret = -EIO; | |
1776 | goto out; | |
1777 | } | |
1778 | } | |
1779 | ret = 0; | |
1780 | goto out; | |
1781 | } | |
1782 | ||
1783 | for (i = 0; list[i]; i++) { | |
1784 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
1785 | ret = -EIO; | |
1786 | goto out; | |
1787 | } | |
1788 | } | |
1789 | ||
1790 | // now get the list of child cgroups | |
1791 | ||
1792 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
1793 | ret = 0; | |
1794 | goto out; | |
1795 | } | |
f366da65 WB |
1796 | if (clist) { |
1797 | for (i = 0; clist[i]; i++) { | |
1798 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
1799 | ret = -EIO; | |
1800 | goto out; | |
1801 | } | |
237e200e SH |
1802 | } |
1803 | } | |
1804 | ret = 0; | |
1805 | ||
1806 | out: | |
1807 | free_keys(list); | |
1808 | if (clist) { | |
1809 | for (i = 0; clist[i]; i++) | |
1810 | free(clist[i]); | |
1811 | free(clist); | |
1812 | } | |
1813 | return ret; | |
1814 | } | |
1815 | ||
43215927 | 1816 | static void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 1817 | { |
43215927 SH |
1818 | struct file_info *f = (struct file_info *)fi->fh; |
1819 | ||
237e200e SH |
1820 | if (!f) |
1821 | return; | |
43215927 SH |
1822 | |
1823 | fi->fh = 0; | |
1824 | ||
237e200e | 1825 | free(f->controller); |
43215927 | 1826 | f->controller = NULL; |
237e200e | 1827 | free(f->cgroup); |
43215927 | 1828 | f->cgroup = NULL; |
237e200e | 1829 | free(f->file); |
43215927 | 1830 | f->file = NULL; |
237e200e | 1831 | free(f->buf); |
43215927 | 1832 | f->buf = NULL; |
237e200e SH |
1833 | free(f); |
1834 | } | |
1835 | ||
1836 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
1837 | { | |
43215927 | 1838 | do_release_file_info(fi); |
237e200e SH |
1839 | return 0; |
1840 | } | |
1841 | ||
1842 | int cg_open(const char *path, struct fuse_file_info *fi) | |
1843 | { | |
1844 | const char *cgroup; | |
1845 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1846 | struct cgfs_files *k = NULL; | |
1847 | struct file_info *file_info; | |
1848 | struct fuse_context *fc = fuse_get_context(); | |
1849 | int ret; | |
1850 | ||
1851 | if (!fc) | |
1852 | return -EIO; | |
1853 | ||
1854 | controller = pick_controller_from_path(fc, path); | |
1855 | if (!controller) | |
1856 | return -EIO; | |
1857 | cgroup = find_cgroup_in_path(path); | |
1858 | if (!cgroup) | |
1859 | return -EINVAL; | |
1860 | ||
1861 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1862 | if (!last) { | |
1863 | path1 = "/"; | |
1864 | path2 = cgdir; | |
1865 | } else { | |
1866 | path1 = cgdir; | |
1867 | path2 = last; | |
1868 | } | |
1869 | ||
1870 | k = cgfs_get_key(controller, path1, path2); | |
1871 | if (!k) { | |
1872 | ret = -EINVAL; | |
1873 | goto out; | |
1874 | } | |
1875 | free_key(k); | |
1876 | ||
1877 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1878 | if (initpid <= 0) | |
1879 | initpid = fc->pid; | |
1880 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1881 | ret = -ENOENT; | |
1882 | goto out; | |
1883 | } | |
1884 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
1885 | ret = -EACCES; |
1886 | goto out; | |
1887 | } | |
1888 | ||
1889 | /* we'll free this at cg_release */ | |
1890 | file_info = malloc(sizeof(*file_info)); | |
1891 | if (!file_info) { | |
1892 | ret = -ENOMEM; | |
1893 | goto out; | |
1894 | } | |
1895 | file_info->controller = must_copy_string(controller); | |
1896 | file_info->cgroup = must_copy_string(path1); | |
1897 | file_info->file = must_copy_string(path2); | |
1898 | file_info->type = LXC_TYPE_CGFILE; | |
1899 | file_info->buf = NULL; | |
1900 | file_info->buflen = 0; | |
1901 | ||
1902 | fi->fh = (unsigned long)file_info; | |
1903 | ret = 0; | |
1904 | ||
1905 | out: | |
1906 | free(cgdir); | |
1907 | return ret; | |
1908 | } | |
1909 | ||
bddbb106 SH |
1910 | int cg_access(const char *path, int mode) |
1911 | { | |
1912 | const char *cgroup; | |
1913 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1914 | struct cgfs_files *k = NULL; | |
1915 | struct fuse_context *fc = fuse_get_context(); | |
1916 | int ret; | |
1917 | ||
1918 | if (!fc) | |
1919 | return -EIO; | |
1920 | ||
1921 | controller = pick_controller_from_path(fc, path); | |
1922 | if (!controller) | |
1923 | return -EIO; | |
1924 | cgroup = find_cgroup_in_path(path); | |
575316c4 SH |
1925 | if (!cgroup) { |
1926 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
1927 | if ((mode & W_OK) == 0) |
1928 | return 0; | |
1929 | return -EACCES; | |
575316c4 | 1930 | } |
bddbb106 SH |
1931 | |
1932 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1933 | if (!last) { | |
1934 | path1 = "/"; | |
1935 | path2 = cgdir; | |
1936 | } else { | |
1937 | path1 = cgdir; | |
1938 | path2 = last; | |
1939 | } | |
1940 | ||
1941 | k = cgfs_get_key(controller, path1, path2); | |
1942 | if (!k) { | |
3f441bc7 SH |
1943 | if ((mode & W_OK) == 0) |
1944 | ret = 0; | |
1945 | else | |
1946 | ret = -EACCES; | |
bddbb106 SH |
1947 | goto out; |
1948 | } | |
1949 | free_key(k); | |
1950 | ||
1951 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1952 | if (initpid <= 0) | |
1953 | initpid = fc->pid; | |
1954 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1955 | ret = -ENOENT; | |
1956 | goto out; | |
1957 | } | |
1958 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
1959 | ret = -EACCES; | |
1960 | goto out; | |
1961 | } | |
1962 | ||
1963 | ret = 0; | |
1964 | ||
1965 | out: | |
1966 | free(cgdir); | |
1967 | return ret; | |
1968 | } | |
1969 | ||
237e200e SH |
1970 | int cg_release(const char *path, struct fuse_file_info *fi) |
1971 | { | |
43215927 | 1972 | do_release_file_info(fi); |
237e200e SH |
1973 | return 0; |
1974 | } | |
1975 | ||
1976 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
1977 | ||
1978 | static bool wait_for_sock(int sock, int timeout) | |
1979 | { | |
1980 | struct epoll_event ev; | |
1981 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
1982 | ||
1983 | if ((starttime = time(NULL)) < 0) | |
1984 | return false; | |
1985 | ||
1986 | if ((epfd = epoll_create(1)) < 0) { | |
1987 | fprintf(stderr, "Failed to create epoll socket: %m\n"); | |
1988 | return false; | |
1989 | } | |
1990 | ||
1991 | ev.events = POLLIN_SET; | |
1992 | ev.data.fd = sock; | |
1993 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
1994 | fprintf(stderr, "Failed adding socket to epoll: %m\n"); | |
1995 | close(epfd); | |
1996 | return false; | |
1997 | } | |
1998 | ||
1999 | again: | |
2000 | if ((now = time(NULL)) < 0) { | |
2001 | close(epfd); | |
2002 | return false; | |
2003 | } | |
2004 | ||
2005 | deltatime = (starttime + timeout) - now; | |
2006 | if (deltatime < 0) { // timeout | |
2007 | errno = 0; | |
2008 | close(epfd); | |
2009 | return false; | |
2010 | } | |
2011 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2012 | if (ret < 0 && errno == EINTR) | |
2013 | goto again; | |
2014 | saved_errno = errno; | |
2015 | close(epfd); | |
2016 | ||
2017 | if (ret <= 0) { | |
2018 | errno = saved_errno; | |
2019 | return false; | |
2020 | } | |
2021 | return true; | |
2022 | } | |
2023 | ||
2024 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2025 | { | |
2026 | if (!wait_for_sock(sockfd, 2)) | |
2027 | return -1; | |
2028 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2029 | } | |
2030 | ||
2031 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2032 | { | |
2033 | struct msghdr msg = { 0 }; | |
2034 | struct iovec iov; | |
2035 | struct cmsghdr *cmsg; | |
2036 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2037 | char buf[1]; | |
2038 | buf[0] = 'p'; | |
2039 | ||
2040 | if (pingfirst) { | |
2041 | if (msgrecv(sock, buf, 1) != 1) { | |
2042 | fprintf(stderr, "%s: Error getting reply from server over socketpair\n", | |
2043 | __func__); | |
2044 | return SEND_CREDS_FAIL; | |
2045 | } | |
2046 | } | |
2047 | ||
2048 | msg.msg_control = cmsgbuf; | |
2049 | msg.msg_controllen = sizeof(cmsgbuf); | |
2050 | ||
2051 | cmsg = CMSG_FIRSTHDR(&msg); | |
2052 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2053 | cmsg->cmsg_level = SOL_SOCKET; | |
2054 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2055 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2056 | ||
2057 | msg.msg_name = NULL; | |
2058 | msg.msg_namelen = 0; | |
2059 | ||
2060 | buf[0] = v; | |
2061 | iov.iov_base = buf; | |
2062 | iov.iov_len = sizeof(buf); | |
2063 | msg.msg_iov = &iov; | |
2064 | msg.msg_iovlen = 1; | |
2065 | ||
2066 | if (sendmsg(sock, &msg, 0) < 0) { | |
2067 | fprintf(stderr, "%s: failed at sendmsg: %s\n", __func__, | |
2068 | strerror(errno)); | |
2069 | if (errno == 3) | |
2070 | return SEND_CREDS_NOTSK; | |
2071 | return SEND_CREDS_FAIL; | |
2072 | } | |
2073 | ||
2074 | return SEND_CREDS_OK; | |
2075 | } | |
2076 | ||
2077 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2078 | { | |
2079 | struct msghdr msg = { 0 }; | |
2080 | struct iovec iov; | |
2081 | struct cmsghdr *cmsg; | |
2082 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2083 | char buf[1]; | |
2084 | int ret; | |
2085 | int optval = 1; | |
2086 | ||
2087 | *v = '1'; | |
2088 | ||
2089 | cred->pid = -1; | |
2090 | cred->uid = -1; | |
2091 | cred->gid = -1; | |
2092 | ||
2093 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
2094 | fprintf(stderr, "Failed to set passcred: %s\n", strerror(errno)); | |
2095 | return false; | |
2096 | } | |
2097 | buf[0] = '1'; | |
2098 | if (write(sock, buf, 1) != 1) { | |
2099 | fprintf(stderr, "Failed to start write on scm fd: %s\n", strerror(errno)); | |
2100 | return false; | |
2101 | } | |
2102 | ||
2103 | msg.msg_name = NULL; | |
2104 | msg.msg_namelen = 0; | |
2105 | msg.msg_control = cmsgbuf; | |
2106 | msg.msg_controllen = sizeof(cmsgbuf); | |
2107 | ||
2108 | iov.iov_base = buf; | |
2109 | iov.iov_len = sizeof(buf); | |
2110 | msg.msg_iov = &iov; | |
2111 | msg.msg_iovlen = 1; | |
2112 | ||
2113 | if (!wait_for_sock(sock, 2)) { | |
2114 | fprintf(stderr, "Timed out waiting for scm_cred: %s\n", | |
2115 | strerror(errno)); | |
2116 | return false; | |
2117 | } | |
2118 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2119 | if (ret < 0) { | |
2120 | fprintf(stderr, "Failed to receive scm_cred: %s\n", | |
2121 | strerror(errno)); | |
2122 | return false; | |
2123 | } | |
2124 | ||
2125 | cmsg = CMSG_FIRSTHDR(&msg); | |
2126 | ||
2127 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2128 | cmsg->cmsg_level == SOL_SOCKET && | |
2129 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2130 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2131 | } | |
2132 | *v = buf[0]; | |
2133 | ||
2134 | return true; | |
2135 | } | |
2136 | ||
35174b0f FG |
2137 | struct pid_ns_clone_args { |
2138 | int *cpipe; | |
2139 | int sock; | |
2140 | pid_t tpid; | |
2141 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2142 | }; | |
2143 | ||
2144 | /* | |
2145 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2146 | * with clone(). This simply writes '1' as ACK back to the parent | |
2147 | * before calling the actual wrapped function. | |
2148 | */ | |
2149 | static int pid_ns_clone_wrapper(void *arg) { | |
2150 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2151 | char b = '1'; | |
2152 | ||
2153 | close(args->cpipe[0]); | |
2154 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) { | |
2155 | fprintf(stderr, "%s (child): error on write: %s\n", | |
2156 | __func__, strerror(errno)); | |
2157 | } | |
2158 | close(args->cpipe[1]); | |
2159 | return args->wrapped(args->sock, args->tpid); | |
2160 | } | |
237e200e SH |
2161 | |
2162 | /* | |
2163 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2164 | * int value back over the socket. This shifts the pid from the | |
2165 | * sender's pidns into tpid's pidns. | |
2166 | */ | |
35174b0f | 2167 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2168 | { |
2169 | char v = '0'; | |
2170 | struct ucred cred; | |
2171 | ||
2172 | while (recv_creds(sock, &cred, &v)) { | |
2173 | if (v == '1') | |
35174b0f | 2174 | return 0; |
237e200e | 2175 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2176 | return 1; |
237e200e | 2177 | } |
35174b0f | 2178 | return 0; |
237e200e SH |
2179 | } |
2180 | ||
35174b0f | 2181 | |
237e200e SH |
2182 | /* |
2183 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2184 | * in your old pidns. Only children which you clone will be in the target |
2185 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2186 | * actually convert pids. | |
2187 | * | |
2188 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2189 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2190 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2191 | * this issue. | |
237e200e SH |
2192 | */ |
2193 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2194 | { | |
2195 | int newnsfd = -1, ret, cpipe[2]; | |
2196 | char fnam[100]; | |
2197 | pid_t cpid; | |
2198 | char v; | |
2199 | ||
2200 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2201 | if (ret < 0 || ret >= sizeof(fnam)) | |
2202 | _exit(1); | |
2203 | newnsfd = open(fnam, O_RDONLY); | |
2204 | if (newnsfd < 0) | |
2205 | _exit(1); | |
2206 | if (setns(newnsfd, 0) < 0) | |
2207 | _exit(1); | |
2208 | close(newnsfd); | |
2209 | ||
2210 | if (pipe(cpipe) < 0) | |
2211 | _exit(1); | |
2212 | ||
35174b0f FG |
2213 | struct pid_ns_clone_args args = { |
2214 | .cpipe = cpipe, | |
2215 | .sock = sock, | |
2216 | .tpid = tpid, | |
2217 | .wrapped = &pid_to_ns | |
2218 | }; | |
2219 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2220 | void *stack = alloca(stack_size); | |
2221 | ||
2222 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2223 | if (cpid < 0) |
2224 | _exit(1); | |
2225 | ||
237e200e SH |
2226 | // give the child 1 second to be done forking and |
2227 | // write its ack | |
2228 | if (!wait_for_sock(cpipe[0], 1)) | |
2229 | _exit(1); | |
2230 | ret = read(cpipe[0], &v, 1); | |
2231 | if (ret != sizeof(char) || v != '1') | |
2232 | _exit(1); | |
2233 | ||
2234 | if (!wait_for_pid(cpid)) | |
2235 | _exit(1); | |
2236 | _exit(0); | |
2237 | } | |
2238 | ||
2239 | /* | |
2240 | * To read cgroup files with a particular pid, we will setns into the child | |
2241 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2242 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2243 | */ | |
2244 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2245 | { | |
2246 | int sock[2] = {-1, -1}; | |
2247 | char *tmpdata = NULL; | |
2248 | int ret; | |
2249 | pid_t qpid, cpid = -1; | |
2250 | bool answer = false; | |
2251 | char v = '0'; | |
2252 | struct ucred cred; | |
2253 | size_t sz = 0, asz = 0; | |
2254 | ||
2255 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2256 | return false; | |
2257 | ||
2258 | /* | |
2259 | * Now we read the pids from returned data one by one, pass | |
2260 | * them into a child in the target namespace, read back the | |
2261 | * translated pids, and put them into our to-return data | |
2262 | */ | |
2263 | ||
2264 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2265 | perror("socketpair"); | |
2266 | free(tmpdata); | |
2267 | return false; | |
2268 | } | |
2269 | ||
2270 | cpid = fork(); | |
2271 | if (cpid == -1) | |
2272 | goto out; | |
2273 | ||
2274 | if (!cpid) // child - exits when done | |
2275 | pid_to_ns_wrapper(sock[1], tpid); | |
2276 | ||
2277 | char *ptr = tmpdata; | |
2278 | cred.uid = 0; | |
2279 | cred.gid = 0; | |
2280 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2281 | cred.pid = qpid; | |
2282 | ret = send_creds(sock[0], &cred, v, true); | |
2283 | ||
2284 | if (ret == SEND_CREDS_NOTSK) | |
2285 | goto next; | |
2286 | if (ret == SEND_CREDS_FAIL) | |
2287 | goto out; | |
2288 | ||
2289 | // read converted results | |
2290 | if (!wait_for_sock(sock[0], 2)) { | |
2291 | fprintf(stderr, "%s: timed out waiting for pid from child: %s\n", | |
2292 | __func__, strerror(errno)); | |
2293 | goto out; | |
2294 | } | |
2295 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2296 | fprintf(stderr, "%s: error reading pid from child: %s\n", | |
2297 | __func__, strerror(errno)); | |
2298 | goto out; | |
2299 | } | |
2300 | must_strcat_pid(d, &sz, &asz, qpid); | |
2301 | next: | |
2302 | ptr = strchr(ptr, '\n'); | |
2303 | if (!ptr) | |
2304 | break; | |
2305 | ptr++; | |
2306 | } | |
2307 | ||
2308 | cred.pid = getpid(); | |
2309 | v = '1'; | |
2310 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2311 | // failed to ask child to exit | |
2312 | fprintf(stderr, "%s: failed to ask child to exit: %s\n", | |
2313 | __func__, strerror(errno)); | |
2314 | goto out; | |
2315 | } | |
2316 | ||
2317 | answer = true; | |
2318 | ||
2319 | out: | |
2320 | free(tmpdata); | |
2321 | if (cpid != -1) | |
2322 | wait_for_pid(cpid); | |
2323 | if (sock[0] != -1) { | |
2324 | close(sock[0]); | |
2325 | close(sock[1]); | |
2326 | } | |
2327 | return answer; | |
2328 | } | |
2329 | ||
2330 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2331 | struct fuse_file_info *fi) | |
2332 | { | |
2333 | struct fuse_context *fc = fuse_get_context(); | |
2334 | struct file_info *f = (struct file_info *)fi->fh; | |
2335 | struct cgfs_files *k = NULL; | |
2336 | char *data = NULL; | |
2337 | int ret, s; | |
2338 | bool r; | |
2339 | ||
2340 | if (f->type != LXC_TYPE_CGFILE) { | |
2341 | fprintf(stderr, "Internal error: directory cache info used in cg_read\n"); | |
2342 | return -EIO; | |
2343 | } | |
2344 | ||
2345 | if (offset) | |
2346 | return 0; | |
2347 | ||
2348 | if (!fc) | |
2349 | return -EIO; | |
2350 | ||
2351 | if (!f->controller) | |
2352 | return -EINVAL; | |
2353 | ||
2354 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2355 | return -EINVAL; | |
2356 | } | |
2357 | free_key(k); | |
2358 | ||
2359 | ||
888f8f3c | 2360 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2361 | ret = -EACCES; |
2362 | goto out; | |
2363 | } | |
2364 | ||
2365 | if (strcmp(f->file, "tasks") == 0 || | |
2366 | strcmp(f->file, "/tasks") == 0 || | |
2367 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2368 | strcmp(f->file, "cgroup.procs") == 0) | |
2369 | // special case - we have to translate the pids | |
2370 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2371 | else | |
2372 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2373 | ||
2374 | if (!r) { | |
2375 | ret = -EINVAL; | |
2376 | goto out; | |
2377 | } | |
2378 | ||
2379 | if (!data) { | |
2380 | ret = 0; | |
2381 | goto out; | |
2382 | } | |
2383 | s = strlen(data); | |
2384 | if (s > size) | |
2385 | s = size; | |
2386 | memcpy(buf, data, s); | |
2387 | if (s > 0 && s < size && data[s-1] != '\n') | |
2388 | buf[s++] = '\n'; | |
2389 | ||
2390 | ret = s; | |
2391 | ||
2392 | out: | |
2393 | free(data); | |
2394 | return ret; | |
2395 | } | |
2396 | ||
35174b0f | 2397 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2398 | { |
2399 | pid_t vpid; | |
2400 | struct ucred cred; | |
2401 | char v; | |
2402 | int ret; | |
2403 | ||
2404 | cred.uid = 0; | |
2405 | cred.gid = 0; | |
2406 | while (1) { | |
2407 | if (!wait_for_sock(sock, 2)) { | |
2408 | fprintf(stderr, "%s: timeout reading from parent\n", __func__); | |
35174b0f | 2409 | return 1; |
237e200e SH |
2410 | } |
2411 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
2412 | fprintf(stderr, "%s: bad read from parent: %s\n", | |
2413 | __func__, strerror(errno)); | |
35174b0f | 2414 | return 1; |
237e200e SH |
2415 | } |
2416 | if (vpid == -1) // done | |
2417 | break; | |
2418 | v = '0'; | |
2419 | cred.pid = vpid; | |
2420 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2421 | v = '1'; | |
2422 | cred.pid = getpid(); | |
2423 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2424 | return 1; |
237e200e SH |
2425 | } |
2426 | } | |
35174b0f | 2427 | return 0; |
237e200e SH |
2428 | } |
2429 | ||
2430 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2431 | { | |
2432 | int newnsfd = -1, ret, cpipe[2]; | |
2433 | char fnam[100]; | |
2434 | pid_t cpid; | |
2435 | char v; | |
2436 | ||
2437 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2438 | if (ret < 0 || ret >= sizeof(fnam)) | |
2439 | _exit(1); | |
2440 | newnsfd = open(fnam, O_RDONLY); | |
2441 | if (newnsfd < 0) | |
2442 | _exit(1); | |
2443 | if (setns(newnsfd, 0) < 0) | |
2444 | _exit(1); | |
2445 | close(newnsfd); | |
2446 | ||
2447 | if (pipe(cpipe) < 0) | |
2448 | _exit(1); | |
2449 | ||
35174b0f FG |
2450 | struct pid_ns_clone_args args = { |
2451 | .cpipe = cpipe, | |
2452 | .sock = sock, | |
2453 | .tpid = tpid, | |
2454 | .wrapped = &pid_from_ns | |
2455 | }; | |
f0f8b851 SH |
2456 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2457 | void *stack = alloca(stack_size); | |
35174b0f FG |
2458 | |
2459 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2460 | if (cpid < 0) |
2461 | _exit(1); | |
2462 | ||
237e200e SH |
2463 | // give the child 1 second to be done forking and |
2464 | // write its ack | |
2465 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2466 | _exit(1); |
237e200e | 2467 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2468 | if (ret != sizeof(char) || v != '1') |
2469 | _exit(1); | |
237e200e SH |
2470 | |
2471 | if (!wait_for_pid(cpid)) | |
2472 | _exit(1); | |
2473 | _exit(0); | |
237e200e SH |
2474 | } |
2475 | ||
2476 | /* | |
2477 | * Given host @uid, return the uid to which it maps in | |
2478 | * @pid's user namespace, or -1 if none. | |
2479 | */ | |
2480 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2481 | { | |
2482 | FILE *f; | |
2483 | char line[400]; | |
2484 | ||
2485 | sprintf(line, "/proc/%d/uid_map", pid); | |
2486 | if ((f = fopen(line, "r")) == NULL) { | |
2487 | return false; | |
2488 | } | |
2489 | ||
2490 | *answer = convert_id_to_ns(f, uid); | |
2491 | fclose(f); | |
2492 | ||
2493 | if (*answer == -1) | |
2494 | return false; | |
2495 | return true; | |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * get_pid_creds: get the real uid and gid of @pid from | |
2500 | * /proc/$$/status | |
2501 | * (XXX should we use euid here?) | |
2502 | */ | |
2503 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2504 | { | |
2505 | char line[400]; | |
2506 | uid_t u; | |
2507 | gid_t g; | |
2508 | FILE *f; | |
2509 | ||
2510 | *uid = -1; | |
2511 | *gid = -1; | |
2512 | sprintf(line, "/proc/%d/status", pid); | |
2513 | if ((f = fopen(line, "r")) == NULL) { | |
2514 | fprintf(stderr, "Error opening %s: %s\n", line, strerror(errno)); | |
2515 | return; | |
2516 | } | |
2517 | while (fgets(line, 400, f)) { | |
2518 | if (strncmp(line, "Uid:", 4) == 0) { | |
2519 | if (sscanf(line+4, "%u", &u) != 1) { | |
2520 | fprintf(stderr, "bad uid line for pid %u\n", pid); | |
2521 | fclose(f); | |
2522 | return; | |
2523 | } | |
2524 | *uid = u; | |
2525 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2526 | if (sscanf(line+4, "%u", &g) != 1) { | |
2527 | fprintf(stderr, "bad gid line for pid %u\n", pid); | |
2528 | fclose(f); | |
2529 | return; | |
2530 | } | |
2531 | *gid = g; | |
2532 | } | |
2533 | } | |
2534 | fclose(f); | |
2535 | } | |
2536 | ||
2537 | /* | |
2538 | * May the requestor @r move victim @v to a new cgroup? | |
2539 | * This is allowed if | |
2540 | * . they are the same task | |
2541 | * . they are ownedy by the same uid | |
2542 | * . @r is root on the host, or | |
2543 | * . @v's uid is mapped into @r's where @r is root. | |
2544 | */ | |
2545 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2546 | { | |
2547 | uid_t v_uid, tmpuid; | |
2548 | gid_t v_gid; | |
2549 | ||
2550 | if (r == v) | |
2551 | return true; | |
2552 | if (r_uid == 0) | |
2553 | return true; | |
2554 | get_pid_creds(v, &v_uid, &v_gid); | |
2555 | if (r_uid == v_uid) | |
2556 | return true; | |
2557 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2558 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2559 | return true; | |
2560 | return false; | |
2561 | } | |
2562 | ||
2563 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2564 | const char *file, const char *buf) | |
2565 | { | |
2566 | int sock[2] = {-1, -1}; | |
2567 | pid_t qpid, cpid = -1; | |
2568 | FILE *pids_file = NULL; | |
2569 | bool answer = false, fail = false; | |
2570 | ||
2571 | pids_file = open_pids_file(contrl, cg); | |
2572 | if (!pids_file) | |
2573 | return false; | |
2574 | ||
2575 | /* | |
2576 | * write the pids to a socket, have helper in writer's pidns | |
2577 | * call movepid for us | |
2578 | */ | |
2579 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2580 | perror("socketpair"); | |
2581 | goto out; | |
2582 | } | |
2583 | ||
2584 | cpid = fork(); | |
2585 | if (cpid == -1) | |
2586 | goto out; | |
2587 | ||
2588 | if (!cpid) { // child | |
2589 | fclose(pids_file); | |
2590 | pid_from_ns_wrapper(sock[1], tpid); | |
2591 | } | |
2592 | ||
2593 | const char *ptr = buf; | |
2594 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2595 | struct ucred cred; | |
2596 | char v; | |
2597 | ||
2598 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2599 | fprintf(stderr, "%s: error writing pid to child: %s\n", | |
2600 | __func__, strerror(errno)); | |
2601 | goto out; | |
2602 | } | |
2603 | ||
2604 | if (recv_creds(sock[0], &cred, &v)) { | |
2605 | if (v == '0') { | |
2606 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2607 | fail = true; | |
2608 | break; | |
2609 | } | |
2610 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2611 | fail = true; | |
2612 | } | |
2613 | } | |
2614 | ||
2615 | ptr = strchr(ptr, '\n'); | |
2616 | if (!ptr) | |
2617 | break; | |
2618 | ptr++; | |
2619 | } | |
2620 | ||
2621 | /* All good, write the value */ | |
2622 | qpid = -1; | |
2623 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
2624 | fprintf(stderr, "Warning: failed to ask child to exit\n"); | |
2625 | ||
2626 | if (!fail) | |
2627 | answer = true; | |
2628 | ||
2629 | out: | |
2630 | if (cpid != -1) | |
2631 | wait_for_pid(cpid); | |
2632 | if (sock[0] != -1) { | |
2633 | close(sock[0]); | |
2634 | close(sock[1]); | |
2635 | } | |
2636 | if (pids_file) { | |
2637 | if (fclose(pids_file) != 0) | |
2638 | answer = false; | |
2639 | } | |
2640 | return answer; | |
2641 | } | |
2642 | ||
2643 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2644 | struct fuse_file_info *fi) | |
2645 | { | |
2646 | struct fuse_context *fc = fuse_get_context(); | |
2647 | char *localbuf = NULL; | |
2648 | struct cgfs_files *k = NULL; | |
2649 | struct file_info *f = (struct file_info *)fi->fh; | |
2650 | bool r; | |
2651 | ||
2652 | if (f->type != LXC_TYPE_CGFILE) { | |
2653 | fprintf(stderr, "Internal error: directory cache info used in cg_write\n"); | |
2654 | return -EIO; | |
2655 | } | |
2656 | ||
2657 | if (offset) | |
2658 | return 0; | |
2659 | ||
2660 | if (!fc) | |
2661 | return -EIO; | |
2662 | ||
2663 | localbuf = alloca(size+1); | |
2664 | localbuf[size] = '\0'; | |
2665 | memcpy(localbuf, buf, size); | |
2666 | ||
2667 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2668 | size = -EINVAL; | |
2669 | goto out; | |
2670 | } | |
2671 | ||
2672 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2673 | size = -EACCES; | |
2674 | goto out; | |
2675 | } | |
2676 | ||
2677 | if (strcmp(f->file, "tasks") == 0 || | |
2678 | strcmp(f->file, "/tasks") == 0 || | |
2679 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2680 | strcmp(f->file, "cgroup.procs") == 0) | |
2681 | // special case - we have to translate the pids | |
2682 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2683 | else | |
2684 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2685 | ||
2686 | if (!r) | |
2687 | size = -EINVAL; | |
2688 | ||
2689 | out: | |
2690 | free_key(k); | |
2691 | return size; | |
2692 | } | |
2693 | ||
2694 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2695 | { | |
2696 | struct fuse_context *fc = fuse_get_context(); | |
2697 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2698 | struct cgfs_files *k = NULL; | |
2699 | const char *cgroup; | |
2700 | int ret; | |
2701 | ||
2702 | if (!fc) | |
2703 | return -EIO; | |
2704 | ||
2705 | if (strcmp(path, "/cgroup") == 0) | |
2706 | return -EINVAL; | |
2707 | ||
2708 | controller = pick_controller_from_path(fc, path); | |
2709 | if (!controller) | |
2710 | return -EINVAL; | |
2711 | cgroup = find_cgroup_in_path(path); | |
2712 | if (!cgroup) | |
2713 | /* this is just /cgroup/controller */ | |
2714 | return -EINVAL; | |
2715 | ||
2716 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2717 | ||
2718 | if (!last) { | |
2719 | path1 = "/"; | |
2720 | path2 = cgdir; | |
2721 | } else { | |
2722 | path1 = cgdir; | |
2723 | path2 = last; | |
2724 | } | |
2725 | ||
2726 | if (is_child_cgroup(controller, path1, path2)) { | |
2727 | // get uid, gid, from '/tasks' file and make up a mode | |
2728 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2729 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2730 | ||
2731 | } else | |
2732 | k = cgfs_get_key(controller, path1, path2); | |
2733 | ||
2734 | if (!k) { | |
2735 | ret = -EINVAL; | |
2736 | goto out; | |
2737 | } | |
2738 | ||
2739 | /* | |
2740 | * This being a fuse request, the uid and gid must be valid | |
2741 | * in the caller's namespace. So we can just check to make | |
2742 | * sure that the caller is root in his uid, and privileged | |
2743 | * over the file's current owner. | |
2744 | */ | |
2745 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2746 | ret = -EACCES; | |
2747 | goto out; | |
2748 | } | |
2749 | ||
2750 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2751 | ||
2752 | out: | |
2753 | free_key(k); | |
2754 | free(cgdir); | |
2755 | ||
2756 | return ret; | |
2757 | } | |
2758 | ||
2759 | int cg_chmod(const char *path, mode_t mode) | |
2760 | { | |
2761 | struct fuse_context *fc = fuse_get_context(); | |
2762 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2763 | struct cgfs_files *k = NULL; | |
2764 | const char *cgroup; | |
2765 | int ret; | |
2766 | ||
2767 | if (!fc) | |
2768 | return -EIO; | |
2769 | ||
2770 | if (strcmp(path, "/cgroup") == 0) | |
2771 | return -EINVAL; | |
2772 | ||
2773 | controller = pick_controller_from_path(fc, path); | |
2774 | if (!controller) | |
2775 | return -EINVAL; | |
2776 | cgroup = find_cgroup_in_path(path); | |
2777 | if (!cgroup) | |
2778 | /* this is just /cgroup/controller */ | |
2779 | return -EINVAL; | |
2780 | ||
2781 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2782 | ||
2783 | if (!last) { | |
2784 | path1 = "/"; | |
2785 | path2 = cgdir; | |
2786 | } else { | |
2787 | path1 = cgdir; | |
2788 | path2 = last; | |
2789 | } | |
2790 | ||
2791 | if (is_child_cgroup(controller, path1, path2)) { | |
2792 | // get uid, gid, from '/tasks' file and make up a mode | |
2793 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2794 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2795 | ||
2796 | } else | |
2797 | k = cgfs_get_key(controller, path1, path2); | |
2798 | ||
2799 | if (!k) { | |
2800 | ret = -EINVAL; | |
2801 | goto out; | |
2802 | } | |
2803 | ||
2804 | /* | |
2805 | * This being a fuse request, the uid and gid must be valid | |
2806 | * in the caller's namespace. So we can just check to make | |
2807 | * sure that the caller is root in his uid, and privileged | |
2808 | * over the file's current owner. | |
2809 | */ | |
2810 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
2811 | ret = -EPERM; | |
2812 | goto out; | |
2813 | } | |
2814 | ||
2815 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
2816 | ret = -EINVAL; | |
2817 | goto out; | |
2818 | } | |
2819 | ||
2820 | ret = 0; | |
2821 | out: | |
2822 | free_key(k); | |
2823 | free(cgdir); | |
2824 | return ret; | |
2825 | } | |
2826 | ||
2827 | int cg_mkdir(const char *path, mode_t mode) | |
2828 | { | |
2829 | struct fuse_context *fc = fuse_get_context(); | |
2830 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
2831 | const char *cgroup; | |
2832 | int ret; | |
2833 | ||
2834 | if (!fc) | |
2835 | return -EIO; | |
2836 | ||
2837 | ||
2838 | controller = pick_controller_from_path(fc, path); | |
2839 | if (!controller) | |
2840 | return -EINVAL; | |
2841 | ||
2842 | cgroup = find_cgroup_in_path(path); | |
2843 | if (!cgroup) | |
2844 | return -EINVAL; | |
2845 | ||
2846 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2847 | if (!last) | |
2848 | path1 = "/"; | |
2849 | else | |
2850 | path1 = cgdir; | |
2851 | ||
2852 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2853 | if (initpid <= 0) | |
2854 | initpid = fc->pid; | |
2855 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
2856 | if (!next) | |
2857 | ret = -EINVAL; | |
2858 | else if (last && strcmp(next, last) == 0) | |
2859 | ret = -EEXIST; | |
2860 | else | |
2861 | ret = -ENOENT; | |
2862 | goto out; | |
2863 | } | |
2864 | ||
2865 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
2866 | ret = -EACCES; | |
2867 | goto out; | |
2868 | } | |
2869 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2870 | ret = -EACCES; | |
2871 | goto out; | |
2872 | } | |
2873 | ||
2874 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
2875 | ||
2876 | out: | |
2877 | free(cgdir); | |
2878 | free(next); | |
2879 | return ret; | |
2880 | } | |
2881 | ||
2882 | int cg_rmdir(const char *path) | |
2883 | { | |
2884 | struct fuse_context *fc = fuse_get_context(); | |
2885 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
2886 | const char *cgroup; | |
2887 | int ret; | |
2888 | ||
2889 | if (!fc) | |
2890 | return -EIO; | |
2891 | ||
2892 | controller = pick_controller_from_path(fc, path); | |
2893 | if (!controller) | |
2894 | return -EINVAL; | |
2895 | ||
2896 | cgroup = find_cgroup_in_path(path); | |
2897 | if (!cgroup) | |
2898 | return -EINVAL; | |
2899 | ||
2900 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2901 | if (!last) { | |
2902 | ret = -EINVAL; | |
2903 | goto out; | |
2904 | } | |
2905 | ||
2906 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2907 | if (initpid <= 0) | |
2908 | initpid = fc->pid; | |
2909 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
2910 | if (!last || strcmp(next, last) == 0) | |
2911 | ret = -EBUSY; | |
2912 | else | |
2913 | ret = -ENOENT; | |
2914 | goto out; | |
2915 | } | |
2916 | ||
2917 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
2918 | ret = -EACCES; | |
2919 | goto out; | |
2920 | } | |
2921 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2922 | ret = -EACCES; | |
2923 | goto out; | |
2924 | } | |
2925 | ||
2926 | if (!cgfs_remove(controller, cgroup)) { | |
2927 | ret = -EINVAL; | |
2928 | goto out; | |
2929 | } | |
2930 | ||
2931 | ret = 0; | |
2932 | ||
2933 | out: | |
2934 | free(cgdir); | |
2935 | free(next); | |
2936 | return ret; | |
2937 | } | |
2938 | ||
2939 | static bool startswith(const char *line, const char *pref) | |
2940 | { | |
2941 | if (strncmp(line, pref, strlen(pref)) == 0) | |
2942 | return true; | |
2943 | return false; | |
2944 | } | |
2945 | ||
c6095b08 SH |
2946 | static void parse_memstat(char *memstat, unsigned long *cached, |
2947 | unsigned long *active_anon, unsigned long *inactive_anon, | |
2948 | unsigned long *active_file, unsigned long *inactive_file, | |
2949 | unsigned long *unevictable) | |
237e200e SH |
2950 | { |
2951 | char *eol; | |
2952 | ||
237e200e | 2953 | while (*memstat) { |
c6095b08 SH |
2954 | if (startswith(memstat, "cache")) { |
2955 | sscanf(memstat + 11, "%lu", cached); | |
2956 | *cached /= 1024; | |
2957 | } else if (startswith(memstat, "active_anon")) { | |
2958 | sscanf(memstat + 11, "%lu", active_anon); | |
2959 | *active_anon /= 1024; | |
2960 | } else if (startswith(memstat, "inactive_anon")) { | |
2961 | sscanf(memstat + 11, "%lu", inactive_anon); | |
2962 | *inactive_anon /= 1024; | |
2963 | } else if (startswith(memstat, "active_file")) { | |
2964 | sscanf(memstat + 11, "%lu", active_file); | |
2965 | *active_file /= 1024; | |
2966 | } else if (startswith(memstat, "inactive_file")) { | |
2967 | sscanf(memstat + 11, "%lu", inactive_file); | |
2968 | *inactive_file /= 1024; | |
2969 | } else if (startswith(memstat, "unevictable")) { | |
2970 | sscanf(memstat + 11, "%lu", unevictable); | |
2971 | *unevictable /= 1024; | |
237e200e SH |
2972 | } |
2973 | eol = strchr(memstat, '\n'); | |
2974 | if (!eol) | |
2975 | return; | |
2976 | memstat = eol+1; | |
2977 | } | |
2978 | } | |
2979 | ||
2980 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
2981 | { | |
2982 | char *eol; | |
2983 | char key[32]; | |
2984 | ||
2985 | memset(key, 0, 32); | |
2986 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2987 | ||
2988 | size_t len = strlen(key); | |
2989 | *v = 0; | |
2990 | ||
2991 | while (*str) { | |
2992 | if (startswith(str, key)) { | |
2993 | sscanf(str + len, "%lu", v); | |
2994 | return; | |
2995 | } | |
2996 | eol = strchr(str, '\n'); | |
2997 | if (!eol) | |
2998 | return; | |
2999 | str = eol+1; | |
3000 | } | |
3001 | } | |
3002 | ||
3003 | static int read_file(const char *path, char *buf, size_t size, | |
3004 | struct file_info *d) | |
3005 | { | |
3006 | size_t linelen = 0, total_len = 0, rv = 0; | |
3007 | char *line = NULL; | |
3008 | char *cache = d->buf; | |
3009 | size_t cache_size = d->buflen; | |
3010 | FILE *f = fopen(path, "r"); | |
3011 | if (!f) | |
3012 | return 0; | |
3013 | ||
3014 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3015 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3016 | if (l < 0) { |
3017 | perror("Error writing to cache"); | |
3018 | rv = 0; | |
3019 | goto err; | |
3020 | } | |
3021 | if (l >= cache_size) { | |
3022 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3023 | rv = 0; | |
3024 | goto err; | |
3025 | } | |
3026 | cache += l; | |
3027 | cache_size -= l; | |
3028 | total_len += l; | |
3029 | } | |
3030 | ||
3031 | d->size = total_len; | |
a262ddb7 CB |
3032 | if (total_len > size) |
3033 | total_len = size; | |
237e200e SH |
3034 | |
3035 | /* read from off 0 */ | |
3036 | memcpy(buf, d->buf, total_len); | |
3037 | rv = total_len; | |
3038 | err: | |
3039 | fclose(f); | |
3040 | free(line); | |
3041 | return rv; | |
3042 | } | |
3043 | ||
3044 | /* | |
3045 | * FUSE ops for /proc | |
3046 | */ | |
3047 | ||
3048 | static unsigned long get_memlimit(const char *cgroup) | |
3049 | { | |
3050 | char *memlimit_str = NULL; | |
3051 | unsigned long memlimit = -1; | |
3052 | ||
3053 | if (cgfs_get_value("memory", cgroup, "memory.limit_in_bytes", &memlimit_str)) | |
3054 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3055 | ||
3056 | free(memlimit_str); | |
3057 | ||
3058 | return memlimit; | |
3059 | } | |
3060 | ||
3061 | static unsigned long get_min_memlimit(const char *cgroup) | |
3062 | { | |
3063 | char *copy = strdupa(cgroup); | |
3064 | unsigned long memlimit = 0, retlimit; | |
3065 | ||
3066 | retlimit = get_memlimit(copy); | |
3067 | ||
3068 | while (strcmp(copy, "/") != 0) { | |
3069 | copy = dirname(copy); | |
3070 | memlimit = get_memlimit(copy); | |
3071 | if (memlimit != -1 && memlimit < retlimit) | |
3072 | retlimit = memlimit; | |
3073 | }; | |
3074 | ||
3075 | return retlimit; | |
3076 | } | |
3077 | ||
3078 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3079 | struct fuse_file_info *fi) | |
3080 | { | |
3081 | struct fuse_context *fc = fuse_get_context(); | |
3082 | struct file_info *d = (struct file_info *)fi->fh; | |
3083 | char *cg; | |
3084 | char *memusage_str = NULL, *memstat_str = NULL, | |
3085 | *memswlimit_str = NULL, *memswusage_str = NULL, | |
3086 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
3087 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, | |
c6095b08 SH |
3088 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
3089 | active_file = 0, inactive_file = 0, unevictable = 0; | |
237e200e SH |
3090 | char *line = NULL; |
3091 | size_t linelen = 0, total_len = 0, rv = 0; | |
3092 | char *cache = d->buf; | |
3093 | size_t cache_size = d->buflen; | |
3094 | FILE *f = NULL; | |
3095 | ||
3096 | if (offset){ | |
3097 | if (offset > d->size) | |
3098 | return -EINVAL; | |
3099 | if (!d->cached) | |
3100 | return 0; | |
3101 | int left = d->size - offset; | |
3102 | total_len = left > size ? size: left; | |
3103 | memcpy(buf, cache + offset, total_len); | |
3104 | return total_len; | |
3105 | } | |
3106 | ||
3107 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3108 | if (initpid <= 0) | |
3109 | initpid = fc->pid; | |
3110 | cg = get_pid_cgroup(initpid, "memory"); | |
3111 | if (!cg) | |
3112 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3113 | prune_init_slice(cg); |
237e200e SH |
3114 | |
3115 | memlimit = get_min_memlimit(cg); | |
3116 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3117 | goto err; | |
3118 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3119 | goto err; | |
3120 | ||
3121 | // Following values are allowed to fail, because swapaccount might be turned | |
3122 | // off for current kernel | |
3123 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3124 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3125 | { | |
3126 | /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */ | |
3127 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
3128 | goto err; | |
3129 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
3130 | goto err; | |
3131 | ||
3132 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
3133 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3134 | ||
3135 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
3136 | memswlimit = 0; | |
3137 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
3138 | memswusage = 0; | |
3139 | ||
3140 | memswlimit = memswlimit / 1024; | |
3141 | memswusage = memswusage / 1024; | |
3142 | } | |
3143 | ||
3144 | memusage = strtoul(memusage_str, NULL, 10); | |
3145 | memlimit /= 1024; | |
3146 | memusage /= 1024; | |
3147 | ||
c6095b08 SH |
3148 | parse_memstat(memstat_str, &cached, &active_anon, |
3149 | &inactive_anon, &active_file, &inactive_file, | |
3150 | &unevictable); | |
237e200e SH |
3151 | |
3152 | f = fopen("/proc/meminfo", "r"); | |
3153 | if (!f) | |
3154 | goto err; | |
3155 | ||
3156 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3157 | ssize_t l; |
237e200e SH |
3158 | char *printme, lbuf[100]; |
3159 | ||
3160 | memset(lbuf, 0, 100); | |
3161 | if (startswith(line, "MemTotal:")) { | |
3162 | sscanf(line+14, "%lu", &hosttotal); | |
3163 | if (hosttotal < memlimit) | |
3164 | memlimit = hosttotal; | |
3165 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3166 | printme = lbuf; | |
3167 | } else if (startswith(line, "MemFree:")) { | |
3168 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3169 | printme = lbuf; | |
3170 | } else if (startswith(line, "MemAvailable:")) { | |
3171 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
3172 | printme = lbuf; | |
3173 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
3174 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit - memlimit); | |
3175 | printme = lbuf; | |
3176 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
b4665ce0 SH |
3177 | unsigned long swaptotal = memswlimit - memlimit, |
3178 | swapusage = memswusage - memusage, | |
3179 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3180 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3181 | printme = lbuf; |
da35d72a SH |
3182 | } else if (startswith(line, "Slab:")) { |
3183 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3184 | printme = lbuf; | |
237e200e SH |
3185 | } else if (startswith(line, "Buffers:")) { |
3186 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3187 | printme = lbuf; | |
3188 | } else if (startswith(line, "Cached:")) { | |
3189 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3190 | printme = lbuf; | |
3191 | } else if (startswith(line, "SwapCached:")) { | |
3192 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3193 | printme = lbuf; | |
c6095b08 SH |
3194 | } else if (startswith(line, "Active")) { |
3195 | snprintf(lbuf, 100, "Active: %8lu kB\n", | |
3196 | active_anon + active_file); | |
3197 | printme = lbuf; | |
3198 | } else if (startswith(line, "Inactive")) { | |
3199 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", | |
3200 | inactive_anon + inactive_file); | |
3201 | printme = lbuf; | |
3202 | } else if (startswith(line, "Active(anon)")) { | |
3203 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3204 | printme = lbuf; | |
3205 | } else if (startswith(line, "Inactive(anon)")) { | |
3206 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3207 | printme = lbuf; | |
3208 | } else if (startswith(line, "Active(file)")) { | |
3209 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3210 | printme = lbuf; | |
3211 | } else if (startswith(line, "Inactive(file)")) { | |
3212 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3213 | printme = lbuf; | |
3214 | } else if (startswith(line, "Unevictable")) { | |
3215 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3216 | printme = lbuf; | |
3217 | } else if (startswith(line, "SReclaimable")) { | |
3218 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3219 | printme = lbuf; | |
3220 | } else if (startswith(line, "SUnreclaim")) { | |
3221 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3222 | printme = lbuf; | |
237e200e SH |
3223 | } else |
3224 | printme = line; | |
3225 | ||
3226 | l = snprintf(cache, cache_size, "%s", printme); | |
3227 | if (l < 0) { | |
3228 | perror("Error writing to cache"); | |
3229 | rv = 0; | |
3230 | goto err; | |
3231 | ||
3232 | } | |
3233 | if (l >= cache_size) { | |
3234 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3235 | rv = 0; | |
3236 | goto err; | |
3237 | } | |
3238 | ||
3239 | cache += l; | |
3240 | cache_size -= l; | |
3241 | total_len += l; | |
3242 | } | |
3243 | ||
3244 | d->cached = 1; | |
3245 | d->size = total_len; | |
3246 | if (total_len > size ) total_len = size; | |
3247 | memcpy(buf, d->buf, total_len); | |
3248 | ||
3249 | rv = total_len; | |
3250 | err: | |
3251 | if (f) | |
3252 | fclose(f); | |
3253 | free(line); | |
3254 | free(cg); | |
3255 | free(memusage_str); | |
3256 | free(memswlimit_str); | |
3257 | free(memswusage_str); | |
3258 | free(memstat_str); | |
3259 | free(memswlimit_default_str); | |
3260 | free(memswusage_default_str); | |
3261 | return rv; | |
3262 | } | |
3263 | ||
3264 | /* | |
3265 | * Read the cpuset.cpus for cg | |
3266 | * Return the answer in a newly allocated string which must be freed | |
3267 | */ | |
3268 | static char *get_cpuset(const char *cg) | |
3269 | { | |
3270 | char *answer; | |
3271 | ||
3272 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3273 | return NULL; | |
3274 | return answer; | |
3275 | } | |
3276 | ||
3277 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3278 | ||
3279 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3280 | { | |
3281 | int cpu; | |
3282 | ||
3283 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3284 | return false; | |
3285 | return cpu_in_cpuset(cpu, cpuset); | |
3286 | } | |
3287 | ||
3288 | /* | |
3289 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3290 | */ | |
3291 | static bool is_processor_line(const char *line) | |
3292 | { | |
3293 | int cpu; | |
3294 | ||
3295 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3296 | return true; | |
3297 | return false; | |
3298 | } | |
3299 | ||
3300 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3301 | struct fuse_file_info *fi) | |
3302 | { | |
3303 | struct fuse_context *fc = fuse_get_context(); | |
3304 | struct file_info *d = (struct file_info *)fi->fh; | |
3305 | char *cg; | |
3306 | char *cpuset = NULL; | |
3307 | char *line = NULL; | |
3308 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 SH |
3309 | bool am_printing = false, firstline = true, is_s390x = false; |
3310 | int curcpu = -1, cpu; | |
237e200e SH |
3311 | char *cache = d->buf; |
3312 | size_t cache_size = d->buflen; | |
3313 | FILE *f = NULL; | |
3314 | ||
3315 | if (offset){ | |
3316 | if (offset > d->size) | |
3317 | return -EINVAL; | |
3318 | if (!d->cached) | |
3319 | return 0; | |
3320 | int left = d->size - offset; | |
3321 | total_len = left > size ? size: left; | |
3322 | memcpy(buf, cache + offset, total_len); | |
3323 | return total_len; | |
3324 | } | |
3325 | ||
3326 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3327 | if (initpid <= 0) | |
3328 | initpid = fc->pid; | |
3329 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3330 | if (!cg) | |
3331 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3332 | prune_init_slice(cg); |
237e200e SH |
3333 | |
3334 | cpuset = get_cpuset(cg); | |
3335 | if (!cpuset) | |
3336 | goto err; | |
3337 | ||
3338 | f = fopen("/proc/cpuinfo", "r"); | |
3339 | if (!f) | |
3340 | goto err; | |
3341 | ||
3342 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3343 | ssize_t l; |
f676eb79 SH |
3344 | if (firstline) { |
3345 | firstline = false; | |
3346 | if (strstr(line, "IBM/S390") != NULL) { | |
3347 | is_s390x = true; | |
3348 | am_printing = true; | |
5ed9d4e2 | 3349 | continue; |
f676eb79 SH |
3350 | } |
3351 | } | |
5ed9d4e2 SH |
3352 | if (strncmp(line, "# processors:", 12) == 0) |
3353 | continue; | |
237e200e SH |
3354 | if (is_processor_line(line)) { |
3355 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3356 | if (am_printing) { | |
3357 | curcpu ++; | |
3358 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3359 | if (l < 0) { | |
3360 | perror("Error writing to cache"); | |
3361 | rv = 0; | |
3362 | goto err; | |
3363 | } | |
3364 | if (l >= cache_size) { | |
3365 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3366 | rv = 0; | |
3367 | goto err; | |
3368 | } | |
3369 | cache += l; | |
3370 | cache_size -= l; | |
3371 | total_len += l; | |
3372 | } | |
3373 | continue; | |
f676eb79 SH |
3374 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3375 | char *p; | |
3376 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3377 | continue; | |
3378 | curcpu ++; | |
3379 | p = strchr(line, ':'); | |
3380 | if (!p || !*p) | |
3381 | goto err; | |
3382 | p++; | |
5ed9d4e2 | 3383 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3384 | if (l < 0) { |
3385 | perror("Error writing to cache"); | |
3386 | rv = 0; | |
3387 | goto err; | |
3388 | } | |
3389 | if (l >= cache_size) { | |
3390 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3391 | rv = 0; | |
3392 | goto err; | |
3393 | } | |
3394 | cache += l; | |
3395 | cache_size -= l; | |
3396 | total_len += l; | |
3397 | continue; | |
3398 | ||
237e200e SH |
3399 | } |
3400 | if (am_printing) { | |
3401 | l = snprintf(cache, cache_size, "%s", line); | |
3402 | if (l < 0) { | |
3403 | perror("Error writing to cache"); | |
3404 | rv = 0; | |
3405 | goto err; | |
3406 | } | |
3407 | if (l >= cache_size) { | |
3408 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3409 | rv = 0; | |
3410 | goto err; | |
3411 | } | |
3412 | cache += l; | |
3413 | cache_size -= l; | |
3414 | total_len += l; | |
3415 | } | |
3416 | } | |
3417 | ||
5ed9d4e2 SH |
3418 | if (is_s390x) { |
3419 | char *origcache = d->buf; | |
a262ddb7 | 3420 | ssize_t l; |
5ed9d4e2 SH |
3421 | do { |
3422 | d->buf = malloc(d->buflen); | |
3423 | } while (!d->buf); | |
3424 | cache = d->buf; | |
3425 | cache_size = d->buflen; | |
3426 | total_len = 0; | |
3427 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3428 | if (l < 0 || l >= cache_size) { | |
3429 | free(origcache); | |
3430 | goto err; | |
3431 | } | |
3432 | cache_size -= l; | |
3433 | cache += l; | |
3434 | total_len += l; | |
3435 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3436 | if (l < 0 || l >= cache_size) { | |
3437 | free(origcache); | |
3438 | goto err; | |
3439 | } | |
3440 | cache_size -= l; | |
3441 | cache += l; | |
3442 | total_len += l; | |
3443 | l = snprintf(cache, cache_size, "%s", origcache); | |
3444 | free(origcache); | |
3445 | if (l < 0 || l >= cache_size) | |
3446 | goto err; | |
3447 | total_len += l; | |
3448 | } | |
3449 | ||
237e200e SH |
3450 | d->cached = 1; |
3451 | d->size = total_len; | |
3452 | if (total_len > size ) total_len = size; | |
3453 | ||
3454 | /* read from off 0 */ | |
3455 | memcpy(buf, d->buf, total_len); | |
3456 | rv = total_len; | |
3457 | err: | |
3458 | if (f) | |
3459 | fclose(f); | |
3460 | free(line); | |
3461 | free(cpuset); | |
3462 | free(cg); | |
3463 | return rv; | |
3464 | } | |
3465 | ||
3466 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
3467 | struct fuse_file_info *fi) | |
3468 | { | |
3469 | struct fuse_context *fc = fuse_get_context(); | |
3470 | struct file_info *d = (struct file_info *)fi->fh; | |
3471 | char *cg; | |
3472 | char *cpuset = NULL; | |
3473 | char *line = NULL; | |
3474 | size_t linelen = 0, total_len = 0, rv = 0; | |
3475 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
3476 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0; | |
3477 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, | |
3478 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0; | |
3479 | #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE | |
3480 | char cpuall[CPUALL_MAX_SIZE]; | |
3481 | /* reserve for cpu all */ | |
3482 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3483 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3484 | FILE *f = NULL; | |
3485 | ||
3486 | if (offset){ | |
3487 | if (offset > d->size) | |
3488 | return -EINVAL; | |
3489 | if (!d->cached) | |
3490 | return 0; | |
3491 | int left = d->size - offset; | |
3492 | total_len = left > size ? size: left; | |
3493 | memcpy(buf, d->buf + offset, total_len); | |
3494 | return total_len; | |
3495 | } | |
3496 | ||
3497 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3498 | if (initpid <= 0) | |
3499 | initpid = fc->pid; | |
3500 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3501 | if (!cg) | |
3502 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 3503 | prune_init_slice(cg); |
237e200e SH |
3504 | |
3505 | cpuset = get_cpuset(cg); | |
3506 | if (!cpuset) | |
3507 | goto err; | |
3508 | ||
3509 | f = fopen("/proc/stat", "r"); | |
3510 | if (!f) | |
3511 | goto err; | |
3512 | ||
3513 | //skip first line | |
3514 | if (getline(&line, &linelen, f) < 0) { | |
3515 | fprintf(stderr, "proc_stat_read read first line failed\n"); | |
3516 | goto err; | |
3517 | } | |
3518 | ||
3519 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3520 | ssize_t l; |
237e200e SH |
3521 | int cpu; |
3522 | char cpu_char[10]; /* That's a lot of cores */ | |
3523 | char *c; | |
3524 | ||
b4665ce0 SH |
3525 | if (strlen(line) == 0) |
3526 | continue; | |
237e200e SH |
3527 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
3528 | /* not a ^cpuN line containing a number N, just print it */ | |
3529 | l = snprintf(cache, cache_size, "%s", line); | |
3530 | if (l < 0) { | |
3531 | perror("Error writing to cache"); | |
3532 | rv = 0; | |
3533 | goto err; | |
3534 | } | |
3535 | if (l >= cache_size) { | |
3536 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3537 | rv = 0; | |
3538 | goto err; | |
3539 | } | |
3540 | cache += l; | |
3541 | cache_size -= l; | |
3542 | total_len += l; | |
3543 | continue; | |
3544 | } | |
3545 | ||
3546 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3547 | continue; | |
3548 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3549 | continue; | |
3550 | curcpu ++; | |
3551 | ||
3552 | c = strchr(line, ' '); | |
3553 | if (!c) | |
3554 | continue; | |
3555 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
3556 | if (l < 0) { | |
3557 | perror("Error writing to cache"); | |
3558 | rv = 0; | |
3559 | goto err; | |
3560 | ||
3561 | } | |
3562 | if (l >= cache_size) { | |
3563 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3564 | rv = 0; | |
3565 | goto err; | |
3566 | } | |
3567 | ||
3568 | cache += l; | |
3569 | cache_size -= l; | |
3570 | total_len += l; | |
3571 | ||
3572 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user, &nice, &system, &idle, &iowait, &irq, | |
3573 | &softirq, &steal, &guest) != 9) | |
3574 | continue; | |
3575 | user_sum += user; | |
3576 | nice_sum += nice; | |
3577 | system_sum += system; | |
3578 | idle_sum += idle; | |
3579 | iowait_sum += iowait; | |
3580 | irq_sum += irq; | |
3581 | softirq_sum += softirq; | |
3582 | steal_sum += steal; | |
3583 | guest_sum += guest; | |
3584 | } | |
3585 | ||
3586 | cache = d->buf; | |
3587 | ||
3588 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3589 | "cpu ", user_sum, nice_sum, system_sum, idle_sum, iowait_sum, irq_sum, softirq_sum, steal_sum, guest_sum); | |
3590 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE){ | |
3591 | memcpy(cache, cpuall, cpuall_len); | |
3592 | cache += cpuall_len; | |
3593 | } else{ | |
3594 | /* shouldn't happen */ | |
3595 | fprintf(stderr, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len); | |
3596 | cpuall_len = 0; | |
3597 | } | |
3598 | ||
3599 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
3600 | total_len += cpuall_len; | |
3601 | d->cached = 1; | |
3602 | d->size = total_len; | |
3603 | if (total_len > size ) total_len = size; | |
3604 | ||
3605 | memcpy(buf, d->buf, total_len); | |
3606 | rv = total_len; | |
3607 | ||
3608 | err: | |
3609 | if (f) | |
3610 | fclose(f); | |
3611 | free(line); | |
3612 | free(cpuset); | |
3613 | free(cg); | |
3614 | return rv; | |
3615 | } | |
3616 | ||
3617 | static long int getreaperage(pid_t pid) | |
3618 | { | |
3619 | char fnam[100]; | |
3620 | struct stat sb; | |
3621 | int ret; | |
3622 | pid_t qpid; | |
3623 | ||
3624 | qpid = lookup_initpid_in_store(pid); | |
3625 | if (qpid <= 0) | |
3626 | return 0; | |
3627 | ||
3628 | ret = snprintf(fnam, 100, "/proc/%d", qpid); | |
3629 | if (ret < 0 || ret >= 100) | |
3630 | return 0; | |
3631 | ||
3632 | if (lstat(fnam, &sb) < 0) | |
3633 | return 0; | |
3634 | ||
3635 | return time(NULL) - sb.st_ctime; | |
3636 | } | |
3637 | ||
3638 | static unsigned long get_reaper_busy(pid_t task) | |
3639 | { | |
3640 | pid_t initpid = lookup_initpid_in_store(task); | |
3641 | char *cgroup = NULL, *usage_str = NULL; | |
3642 | unsigned long usage = 0; | |
3643 | ||
3644 | if (initpid <= 0) | |
3645 | return 0; | |
3646 | ||
3647 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
3648 | if (!cgroup) | |
3649 | goto out; | |
6d2f6996 | 3650 | prune_init_slice(cgroup); |
237e200e SH |
3651 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
3652 | goto out; | |
3653 | usage = strtoul(usage_str, NULL, 10); | |
3654 | usage /= 1000000000; | |
3655 | ||
3656 | out: | |
3657 | free(cgroup); | |
3658 | free(usage_str); | |
3659 | return usage; | |
3660 | } | |
3661 | ||
3662 | #if RELOADTEST | |
3663 | void iwashere(void) | |
3664 | { | |
237e200e SH |
3665 | int fd; |
3666 | ||
ec2b5e7c | 3667 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
3668 | if (fd >= 0) |
3669 | close(fd); | |
3670 | } | |
3671 | #endif | |
3672 | ||
3673 | /* | |
3674 | * We read /proc/uptime and reuse its second field. | |
3675 | * For the first field, we use the mtime for the reaper for | |
3676 | * the calling pid as returned by getreaperage | |
3677 | */ | |
3678 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
3679 | struct fuse_file_info *fi) | |
3680 | { | |
3681 | struct fuse_context *fc = fuse_get_context(); | |
3682 | struct file_info *d = (struct file_info *)fi->fh; | |
3683 | long int reaperage = getreaperage(fc->pid); | |
3684 | unsigned long int busytime = get_reaper_busy(fc->pid), idletime; | |
3685 | char *cache = d->buf; | |
a262ddb7 | 3686 | ssize_t total_len = 0; |
237e200e SH |
3687 | |
3688 | #if RELOADTEST | |
3689 | iwashere(); | |
3690 | #endif | |
3691 | ||
3692 | if (offset){ | |
3693 | if (offset > d->size) | |
3694 | return -EINVAL; | |
3695 | if (!d->cached) | |
3696 | return 0; | |
3697 | int left = d->size - offset; | |
3698 | total_len = left > size ? size: left; | |
3699 | memcpy(buf, cache + offset, total_len); | |
3700 | return total_len; | |
3701 | } | |
3702 | ||
3703 | idletime = reaperage - busytime; | |
3704 | if (idletime > reaperage) | |
3705 | idletime = reaperage; | |
3706 | ||
3707 | total_len = snprintf(d->buf, d->size, "%ld.0 %lu.0\n", reaperage, idletime); | |
3708 | if (total_len < 0){ | |
3709 | perror("Error writing to cache"); | |
3710 | return 0; | |
3711 | } | |
3712 | ||
3713 | d->size = (int)total_len; | |
3714 | d->cached = 1; | |
3715 | ||
3716 | if (total_len > size) total_len = size; | |
3717 | ||
3718 | memcpy(buf, d->buf, total_len); | |
3719 | return total_len; | |
3720 | } | |
3721 | ||
3722 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
3723 | struct fuse_file_info *fi) | |
3724 | { | |
3725 | char dev_name[72]; | |
3726 | struct fuse_context *fc = fuse_get_context(); | |
3727 | struct file_info *d = (struct file_info *)fi->fh; | |
3728 | char *cg; | |
3729 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
3730 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
3731 | unsigned long read = 0, write = 0; | |
3732 | unsigned long read_merged = 0, write_merged = 0; | |
3733 | unsigned long read_sectors = 0, write_sectors = 0; | |
3734 | unsigned long read_ticks = 0, write_ticks = 0; | |
3735 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
3736 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
3737 | char *cache = d->buf; | |
3738 | size_t cache_size = d->buflen; | |
3739 | char *line = NULL; | |
3740 | size_t linelen = 0, total_len = 0, rv = 0; | |
3741 | unsigned int major = 0, minor = 0; | |
3742 | int i = 0; | |
3743 | FILE *f = NULL; | |
3744 | ||
3745 | if (offset){ | |
3746 | if (offset > d->size) | |
3747 | return -EINVAL; | |
3748 | if (!d->cached) | |
3749 | return 0; | |
3750 | int left = d->size - offset; | |
3751 | total_len = left > size ? size: left; | |
3752 | memcpy(buf, cache + offset, total_len); | |
3753 | return total_len; | |
3754 | } | |
3755 | ||
3756 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3757 | if (initpid <= 0) | |
3758 | initpid = fc->pid; | |
3759 | cg = get_pid_cgroup(initpid, "blkio"); | |
3760 | if (!cg) | |
3761 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 3762 | prune_init_slice(cg); |
237e200e | 3763 | |
2209fe50 | 3764 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 3765 | goto err; |
2209fe50 | 3766 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 3767 | goto err; |
2209fe50 | 3768 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 3769 | goto err; |
2209fe50 | 3770 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 3771 | goto err; |
2209fe50 | 3772 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
3773 | goto err; |
3774 | ||
3775 | ||
3776 | f = fopen("/proc/diskstats", "r"); | |
3777 | if (!f) | |
3778 | goto err; | |
3779 | ||
3780 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3781 | ssize_t l; |
2209fe50 | 3782 | char lbuf[256]; |
237e200e SH |
3783 | |
3784 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 3785 | if (i != 3) |
237e200e | 3786 | continue; |
2209fe50 SH |
3787 | |
3788 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
3789 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
3790 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
3791 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
3792 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
3793 | read_sectors = read_sectors/512; | |
3794 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
3795 | write_sectors = write_sectors/512; | |
3796 | ||
3797 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
3798 | rd_svctm = rd_svctm/1000000; | |
3799 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
3800 | rd_wait = rd_wait/1000000; | |
3801 | read_ticks = rd_svctm + rd_wait; | |
3802 | ||
3803 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
3804 | wr_svctm = wr_svctm/1000000; | |
3805 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
3806 | wr_wait = wr_wait/1000000; | |
3807 | write_ticks = wr_svctm + wr_wait; | |
3808 | ||
3809 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
3810 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
3811 | |
3812 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
3813 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
3814 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3815 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
3816 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
3817 | else | |
3818 | continue; | |
237e200e | 3819 | |
2209fe50 | 3820 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
3821 | if (l < 0) { |
3822 | perror("Error writing to fuse buf"); | |
3823 | rv = 0; | |
3824 | goto err; | |
3825 | } | |
3826 | if (l >= cache_size) { | |
3827 | fprintf(stderr, "Internal error: truncated write to cache\n"); | |
3828 | rv = 0; | |
3829 | goto err; | |
3830 | } | |
3831 | cache += l; | |
3832 | cache_size -= l; | |
3833 | total_len += l; | |
3834 | } | |
3835 | ||
3836 | d->cached = 1; | |
3837 | d->size = total_len; | |
3838 | if (total_len > size ) total_len = size; | |
3839 | memcpy(buf, d->buf, total_len); | |
3840 | ||
3841 | rv = total_len; | |
3842 | err: | |
3843 | free(cg); | |
3844 | if (f) | |
3845 | fclose(f); | |
3846 | free(line); | |
3847 | free(io_serviced_str); | |
3848 | free(io_merged_str); | |
3849 | free(io_service_bytes_str); | |
3850 | free(io_wait_time_str); | |
3851 | free(io_service_time_str); | |
3852 | return rv; | |
3853 | } | |
3854 | ||
70dcc12e SH |
3855 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
3856 | struct fuse_file_info *fi) | |
3857 | { | |
3858 | struct fuse_context *fc = fuse_get_context(); | |
3859 | struct file_info *d = (struct file_info *)fi->fh; | |
3860 | char *cg = NULL; | |
3861 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL, | |
3862 | *memswlimit_default_str = NULL, *memswusage_default_str = NULL; | |
3863 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; | |
a262ddb7 CB |
3864 | ssize_t total_len = 0, rv = 0; |
3865 | ssize_t l = 0; | |
70dcc12e SH |
3866 | char *cache = d->buf; |
3867 | ||
3868 | if (offset) { | |
3869 | if (offset > d->size) | |
3870 | return -EINVAL; | |
3871 | if (!d->cached) | |
3872 | return 0; | |
3873 | int left = d->size - offset; | |
3874 | total_len = left > size ? size: left; | |
3875 | memcpy(buf, cache + offset, total_len); | |
3876 | return total_len; | |
3877 | } | |
3878 | ||
3879 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3880 | if (initpid <= 0) | |
3881 | initpid = fc->pid; | |
3882 | cg = get_pid_cgroup(initpid, "memory"); | |
3883 | if (!cg) | |
3884 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 3885 | prune_init_slice(cg); |
70dcc12e SH |
3886 | |
3887 | if (!cgfs_get_value("memory", cg, "memory.limit_in_bytes", &memlimit_str)) | |
3888 | goto err; | |
3889 | ||
3890 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3891 | goto err; | |
3892 | ||
3893 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3894 | memusage = strtoul(memusage_str, NULL, 10); | |
3895 | ||
3896 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
3897 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
3898 | ||
3899 | /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */ | |
3900 | if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str)) | |
3901 | goto err; | |
3902 | if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str)) | |
3903 | goto err; | |
3904 | ||
3905 | memswlimit = strtoul(memswlimit_str, NULL, 10); | |
3906 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3907 | ||
3908 | if (!strcmp(memswlimit_str, memswlimit_default_str)) | |
3909 | memswlimit = 0; | |
3910 | if (!strcmp(memswusage_str, memswusage_default_str)) | |
3911 | memswusage = 0; | |
3912 | ||
3913 | swap_total = (memswlimit - memlimit) / 1024; | |
3914 | swap_free = (memswusage - memusage) / 1024; | |
3915 | } | |
3916 | ||
3917 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
3918 | ||
3919 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
3920 | if (!memswlimit) { | |
3921 | char *line = NULL; | |
3922 | size_t linelen = 0; | |
3923 | FILE *f = fopen("/proc/meminfo", "r"); | |
3924 | ||
3925 | if (!f) | |
3926 | goto err; | |
3927 | ||
3928 | while (getline(&line, &linelen, f) != -1) { | |
3929 | if (startswith(line, "SwapTotal:")) { | |
3930 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
3931 | } else if (startswith(line, "SwapFree:")) { | |
3932 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
3933 | } | |
3934 | } | |
3935 | ||
3936 | free(line); | |
3937 | fclose(f); | |
3938 | } | |
3939 | ||
3940 | if (swap_total > 0) { | |
a262ddb7 CB |
3941 | l = snprintf(d->buf + total_len, d->size - total_len, |
3942 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
3943 | swap_total, swap_free); | |
3944 | total_len += l; | |
70dcc12e SH |
3945 | } |
3946 | ||
a262ddb7 | 3947 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
3948 | perror("Error writing to cache"); |
3949 | rv = 0; | |
3950 | goto err; | |
3951 | } | |
3952 | ||
3953 | d->cached = 1; | |
3954 | d->size = (int)total_len; | |
3955 | ||
3956 | if (total_len > size) total_len = size; | |
3957 | memcpy(buf, d->buf, total_len); | |
3958 | rv = total_len; | |
3959 | ||
3960 | err: | |
3961 | free(cg); | |
3962 | free(memswlimit_str); | |
3963 | free(memlimit_str); | |
3964 | free(memusage_str); | |
3965 | free(memswusage_str); | |
3966 | free(memswusage_default_str); | |
3967 | free(memswlimit_default_str); | |
3968 | return rv; | |
3969 | } | |
3970 | ||
237e200e SH |
3971 | static off_t get_procfile_size(const char *which) |
3972 | { | |
3973 | FILE *f = fopen(which, "r"); | |
3974 | char *line = NULL; | |
3975 | size_t len = 0; | |
3976 | ssize_t sz, answer = 0; | |
3977 | if (!f) | |
3978 | return 0; | |
3979 | ||
3980 | while ((sz = getline(&line, &len, f)) != -1) | |
3981 | answer += sz; | |
3982 | fclose (f); | |
3983 | free(line); | |
3984 | ||
3985 | return answer; | |
3986 | } | |
3987 | ||
3988 | int proc_getattr(const char *path, struct stat *sb) | |
3989 | { | |
3990 | struct timespec now; | |
3991 | ||
3992 | memset(sb, 0, sizeof(struct stat)); | |
3993 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
3994 | return -EINVAL; | |
3995 | sb->st_uid = sb->st_gid = 0; | |
3996 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
3997 | if (strcmp(path, "/proc") == 0) { | |
3998 | sb->st_mode = S_IFDIR | 00555; | |
3999 | sb->st_nlink = 2; | |
4000 | return 0; | |
4001 | } | |
4002 | if (strcmp(path, "/proc/meminfo") == 0 || | |
4003 | strcmp(path, "/proc/cpuinfo") == 0 || | |
4004 | strcmp(path, "/proc/uptime") == 0 || | |
4005 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e SH |
4006 | strcmp(path, "/proc/diskstats") == 0 || |
4007 | strcmp(path, "/proc/swaps") == 0) { | |
237e200e SH |
4008 | sb->st_size = 0; |
4009 | sb->st_mode = S_IFREG | 00444; | |
4010 | sb->st_nlink = 1; | |
4011 | return 0; | |
4012 | } | |
4013 | ||
4014 | return -ENOENT; | |
4015 | } | |
4016 | ||
4017 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4018 | struct fuse_file_info *fi) | |
4019 | { | |
4020 | if (filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4021 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4022 | filler(buf, "stat", NULL, 0) != 0 || | |
4023 | filler(buf, "uptime", NULL, 0) != 0 || | |
70dcc12e SH |
4024 | filler(buf, "diskstats", NULL, 0) != 0 || |
4025 | filler(buf, "swaps", NULL, 0) != 0) | |
237e200e SH |
4026 | return -EINVAL; |
4027 | return 0; | |
4028 | } | |
4029 | ||
4030 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4031 | { | |
4032 | int type = -1; | |
4033 | struct file_info *info; | |
4034 | ||
4035 | if (strcmp(path, "/proc/meminfo") == 0) | |
4036 | type = LXC_TYPE_PROC_MEMINFO; | |
4037 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4038 | type = LXC_TYPE_PROC_CPUINFO; | |
4039 | else if (strcmp(path, "/proc/uptime") == 0) | |
4040 | type = LXC_TYPE_PROC_UPTIME; | |
4041 | else if (strcmp(path, "/proc/stat") == 0) | |
4042 | type = LXC_TYPE_PROC_STAT; | |
4043 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4044 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
4045 | else if (strcmp(path, "/proc/swaps") == 0) |
4046 | type = LXC_TYPE_PROC_SWAPS; | |
237e200e SH |
4047 | if (type == -1) |
4048 | return -ENOENT; | |
4049 | ||
4050 | info = malloc(sizeof(*info)); | |
4051 | if (!info) | |
4052 | return -ENOMEM; | |
4053 | ||
4054 | memset(info, 0, sizeof(*info)); | |
4055 | info->type = type; | |
4056 | ||
4057 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4058 | do { | |
4059 | info->buf = malloc(info->buflen); | |
4060 | } while (!info->buf); | |
4061 | memset(info->buf, 0, info->buflen); | |
4062 | /* set actual size to buffer size */ | |
4063 | info->size = info->buflen; | |
4064 | ||
4065 | fi->fh = (unsigned long)info; | |
4066 | return 0; | |
4067 | } | |
4068 | ||
bddbb106 SH |
4069 | int proc_access(const char *path, int mask) |
4070 | { | |
4071 | /* these are all read-only */ | |
4072 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 4073 | return -EACCES; |
bddbb106 SH |
4074 | return 0; |
4075 | } | |
4076 | ||
237e200e SH |
4077 | int proc_release(const char *path, struct fuse_file_info *fi) |
4078 | { | |
43215927 | 4079 | do_release_file_info(fi); |
237e200e SH |
4080 | return 0; |
4081 | } | |
4082 | ||
4083 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4084 | struct fuse_file_info *fi) | |
4085 | { | |
4086 | struct file_info *f = (struct file_info *) fi->fh; | |
4087 | ||
4088 | switch (f->type) { | |
4089 | case LXC_TYPE_PROC_MEMINFO: | |
4090 | return proc_meminfo_read(buf, size, offset, fi); | |
4091 | case LXC_TYPE_PROC_CPUINFO: | |
4092 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4093 | case LXC_TYPE_PROC_UPTIME: | |
4094 | return proc_uptime_read(buf, size, offset, fi); | |
4095 | case LXC_TYPE_PROC_STAT: | |
4096 | return proc_stat_read(buf, size, offset, fi); | |
4097 | case LXC_TYPE_PROC_DISKSTATS: | |
4098 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
4099 | case LXC_TYPE_PROC_SWAPS: |
4100 | return proc_swaps_read(buf, size, offset, fi); | |
237e200e SH |
4101 | default: |
4102 | return -EINVAL; | |
4103 | } | |
4104 | } | |
4105 | ||
29a73c2f CB |
4106 | /* |
4107 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
4108 | */ |
4109 | ||
4110 | static bool mkdir_p(const char *dir, mode_t mode) | |
4111 | { | |
4112 | const char *tmp = dir; | |
4113 | const char *orig = dir; | |
4114 | char *makeme; | |
4115 | ||
4116 | do { | |
4117 | dir = tmp + strspn(tmp, "/"); | |
4118 | tmp = dir + strcspn(dir, "/"); | |
4119 | makeme = strndup(orig, dir - orig); | |
4120 | if (!makeme) | |
4121 | return false; | |
4122 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
4123 | fprintf(stderr, "failed to create directory '%s': %s", | |
4124 | makeme, strerror(errno)); | |
4125 | free(makeme); | |
4126 | return false; | |
4127 | } | |
4128 | free(makeme); | |
4129 | } while(tmp != dir); | |
4130 | ||
4131 | return true; | |
4132 | } | |
4133 | ||
4134 | static bool umount_if_mounted(void) | |
4135 | { | |
4136 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
480262c9 | 4137 | fprintf(stderr, "failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
4138 | return false; |
4139 | } | |
4140 | return true; | |
4141 | } | |
4142 | ||
4143 | static int pivot_enter(void) | |
4144 | { | |
4145 | int ret = -1, oldroot = -1, newroot = -1; | |
4146 | ||
4147 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
4148 | if (oldroot < 0) { | |
4149 | fprintf(stderr, "%s: Failed to open old root for fchdir.\n", __func__); | |
4150 | return ret; | |
4151 | } | |
4152 | ||
4153 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
4154 | if (newroot < 0) { | |
4155 | fprintf(stderr, "%s: Failed to open new root for fchdir.\n", __func__); | |
4156 | goto err; | |
4157 | } | |
4158 | ||
4159 | /* change into new root fs */ | |
4160 | if (fchdir(newroot) < 0) { | |
4161 | fprintf(stderr, "%s: Failed to change directory to new rootfs: %s.\n", __func__, ROOTDIR); | |
4162 | goto err; | |
4163 | } | |
4164 | ||
4165 | /* pivot_root into our new root fs */ | |
4166 | if (pivot_root(".", ".") < 0) { | |
4167 | fprintf(stderr, "%s: pivot_root() syscall failed: %s.\n", __func__, strerror(errno)); | |
4168 | goto err; | |
4169 | } | |
4170 | ||
4171 | /* | |
4172 | * At this point the old-root is mounted on top of our new-root. | |
4173 | * To unmounted it we must not be chdir'd into it, so escape back | |
4174 | * to the old-root. | |
4175 | */ | |
4176 | if (fchdir(oldroot) < 0) { | |
4177 | fprintf(stderr, "%s: Failed to enter old root.\n", __func__); | |
4178 | goto err; | |
4179 | } | |
4180 | if (umount2(".", MNT_DETACH) < 0) { | |
4181 | fprintf(stderr, "%s: Failed to detach old root.\n", __func__); | |
4182 | goto err; | |
4183 | } | |
4184 | ||
4185 | if (fchdir(newroot) < 0) { | |
4186 | fprintf(stderr, "%s: Failed to re-enter new root.\n", __func__); | |
4187 | goto err; | |
4188 | } | |
4189 | ||
4190 | ret = 0; | |
4191 | ||
4192 | err: | |
4193 | if (oldroot > 0) | |
4194 | close(oldroot); | |
4195 | if (newroot > 0) | |
4196 | close(newroot); | |
4197 | return ret; | |
4198 | } | |
4199 | ||
4200 | /* Prepare our new clean root. */ | |
4201 | static int pivot_prepare(void) | |
4202 | { | |
4203 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
4204 | fprintf(stderr, "%s: Failed to create directory for new root.\n", __func__); | |
4205 | return -1; | |
4206 | } | |
4207 | ||
4208 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
4209 | fprintf(stderr, "%s: Failed to bind-mount / for new root: %s.\n", __func__, strerror(errno)); | |
4210 | return -1; | |
4211 | } | |
4212 | ||
4213 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
4214 | fprintf(stderr, "%s: Failed to bind-mount /run into new root: %s.\n", __func__, strerror(errno)); | |
4215 | return -1; | |
4216 | } | |
4217 | ||
4218 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
4219 | printf("%s: failed to move " BASEDIR " into new root: %s.\n", __func__, strerror(errno)); | |
4220 | return -1; | |
4221 | } | |
4222 | ||
4223 | return 0; | |
4224 | } | |
4225 | ||
4226 | static bool pivot_new_root(void) | |
4227 | { | |
4228 | /* Prepare new root. */ | |
4229 | if (pivot_prepare() < 0) | |
4230 | return false; | |
4231 | ||
4232 | /* Pivot into new root. */ | |
4233 | if (pivot_enter() < 0) | |
4234 | return false; | |
4235 | ||
4236 | return true; | |
4237 | } | |
4238 | ||
4239 | static bool setup_cgfs_dir(void) | |
4240 | { | |
4241 | if (!mkdir_p(BASEDIR, 0700)) { | |
480262c9 | 4242 | fprintf(stderr, "Failed to create lxcfs cgroup mountpoint.\n"); |
29a73c2f CB |
4243 | return false; |
4244 | } | |
480262c9 | 4245 | |
29a73c2f | 4246 | if (!umount_if_mounted()) { |
480262c9 CB |
4247 | fprintf(stderr, "Failed to clean up old lxcfs cgroup mountpoint.\n"); |
4248 | return false; | |
4249 | } | |
4250 | ||
4251 | if (unshare(CLONE_NEWNS) < 0) { | |
4252 | fprintf(stderr, "%s: Failed to unshare mount namespace: %s.\n", __func__, strerror(errno)); | |
4253 | return false; | |
4254 | } | |
4255 | ||
4256 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { | |
4257 | fprintf(stderr, "%s: Failed to remount / private: %s.\n", __func__, strerror(errno)); | |
29a73c2f CB |
4258 | return false; |
4259 | } | |
480262c9 | 4260 | |
29a73c2f | 4261 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
480262c9 | 4262 | fprintf(stderr, "Failed to mount tmpfs over lxcfs cgroup mountpoint.\n"); |
29a73c2f CB |
4263 | return false; |
4264 | } | |
480262c9 | 4265 | |
29a73c2f CB |
4266 | return true; |
4267 | } | |
4268 | ||
4269 | static bool do_mount_cgroups(void) | |
4270 | { | |
4271 | char *target; | |
4272 | size_t clen, len; | |
4273 | int i, ret; | |
4274 | ||
4275 | for (i = 0; i < num_hierarchies; i++) { | |
4276 | char *controller = hierarchies[i]; | |
4277 | clen = strlen(controller); | |
4278 | len = strlen(BASEDIR) + clen + 2; | |
4279 | target = malloc(len); | |
4280 | if (!target) | |
4281 | return false; | |
4282 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); | |
4283 | if (ret < 0 || ret >= len) { | |
4284 | free(target); | |
4285 | return false; | |
4286 | } | |
4287 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
4288 | free(target); | |
4289 | return false; | |
4290 | } | |
4291 | if (mount(controller, target, "cgroup", 0, controller) < 0) { | |
4292 | fprintf(stderr, "Failed mounting cgroup %s\n", controller); | |
4293 | free(target); | |
4294 | return false; | |
4295 | } | |
4296 | ||
4297 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
4298 | if (fd_hierarchies[i] < 0) { | |
4299 | free(target); | |
4300 | return false; | |
4301 | } | |
4302 | free(target); | |
4303 | } | |
4304 | return true; | |
4305 | } | |
4306 | ||
480262c9 | 4307 | static bool cgfs_setup_controllers(void) |
29a73c2f | 4308 | { |
480262c9 | 4309 | if (!setup_cgfs_dir()) |
29a73c2f | 4310 | return false; |
29a73c2f CB |
4311 | |
4312 | if (!do_mount_cgroups()) { | |
480262c9 | 4313 | fprintf(stderr, "Failed to set up private lxcfs cgroup mounts.\n"); |
29a73c2f CB |
4314 | return false; |
4315 | } | |
4316 | ||
4317 | if (!pivot_new_root()) | |
4318 | return false; | |
4319 | ||
4320 | return true; | |
4321 | } | |
4322 | ||
480262c9 CB |
4323 | static int preserve_ns(int pid) |
4324 | { | |
4325 | int ret; | |
4326 | size_t len = 5 /* /proc */ + 21 /* /int_as_str */ + 7 /* /ns/mnt */ + 1 /* \0 */; | |
4327 | char path[len]; | |
4328 | ||
4329 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
4330 | if (ret < 0 || (size_t)ret >= len) | |
4331 | return -1; | |
4332 | ||
4333 | return open(path, O_RDONLY | O_CLOEXEC); | |
4334 | } | |
4335 | ||
29a73c2f | 4336 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) |
237e200e SH |
4337 | { |
4338 | FILE *f; | |
4339 | char *line = NULL; | |
4340 | size_t len = 0; | |
480262c9 | 4341 | int i, init_ns = -1; |
237e200e SH |
4342 | |
4343 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
4344 | fprintf(stderr, "Error opening /proc/self/cgroup: %s\n", strerror(errno)); | |
4345 | return; | |
4346 | } | |
4347 | while (getline(&line, &len, f) != -1) { | |
4348 | char *p, *p2; | |
4349 | ||
4350 | p = strchr(line, ':'); | |
4351 | if (!p) | |
4352 | goto out; | |
4353 | *(p++) = '\0'; | |
4354 | ||
4355 | p2 = strrchr(p, ':'); | |
4356 | if (!p2) | |
4357 | goto out; | |
4358 | *p2 = '\0'; | |
4359 | ||
a67719f6 CB |
4360 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
4361 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
4362 | * because it parses out the empty string "" and later on passes | |
4363 | * it to mount(). Let's skip such entries. | |
4364 | */ | |
4365 | if (!strcmp(p, "")) | |
4366 | continue; | |
4367 | ||
237e200e SH |
4368 | if (!store_hierarchy(line, p)) |
4369 | goto out; | |
4370 | } | |
4371 | ||
480262c9 CB |
4372 | /* Preserve initial namespace. */ |
4373 | init_ns = preserve_ns(getpid()); | |
4374 | if (init_ns < 0) | |
4375 | goto out; | |
4376 | ||
4377 | fd_hierarchies = malloc(sizeof(int *) * num_hierarchies); | |
29a73c2f CB |
4378 | if (!fd_hierarchies) |
4379 | goto out; | |
4380 | ||
480262c9 CB |
4381 | for (i = 0; i < num_hierarchies; i++) |
4382 | fd_hierarchies[i] = -1; | |
4383 | ||
4384 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace | |
4385 | * to privately mount lxcfs cgroups. */ | |
4386 | if (!cgfs_setup_controllers()) | |
29a73c2f | 4387 | goto out; |
480262c9 CB |
4388 | |
4389 | if (setns(init_ns, 0) < 0) | |
29a73c2f | 4390 | goto out; |
29a73c2f | 4391 | |
237e200e SH |
4392 | print_subsystems(); |
4393 | ||
4394 | out: | |
4395 | free(line); | |
4396 | fclose(f); | |
480262c9 CB |
4397 | if (init_ns >= 0) |
4398 | close(init_ns); | |
237e200e SH |
4399 | } |
4400 | ||
4401 | static void __attribute__((destructor)) free_subsystems(void) | |
4402 | { | |
4403 | int i; | |
4404 | ||
29a73c2f | 4405 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
4406 | if (hierarchies[i]) |
4407 | free(hierarchies[i]); | |
480262c9 | 4408 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
4409 | close(fd_hierarchies[i]); |
4410 | } | |
237e200e | 4411 | free(hierarchies); |
480262c9 | 4412 | free(fd_hierarchies); |
237e200e | 4413 | } |