]>
Commit | Line | Data |
---|---|---|
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 | ||
11 | #define __STDC_FORMAT_MACROS | |
12 | #include <dirent.h> | |
13 | #include <errno.h> | |
14 | #include <fcntl.h> | |
15 | #include <fuse.h> | |
16 | #include <inttypes.h> | |
17 | #include <libgen.h> | |
18 | #include <pthread.h> | |
19 | #include <sched.h> | |
20 | #include <stdbool.h> | |
21 | #include <stdint.h> | |
22 | #include <stdio.h> | |
23 | #include <stdlib.h> | |
24 | #include <string.h> | |
25 | #include <time.h> | |
26 | #include <unistd.h> | |
27 | #include <wait.h> | |
28 | #include <linux/magic.h> | |
29 | #include <linux/sched.h> | |
30 | #include <sys/epoll.h> | |
31 | #include <sys/mman.h> | |
32 | #include <sys/mount.h> | |
33 | #include <sys/param.h> | |
34 | #include <sys/socket.h> | |
35 | #include <sys/syscall.h> | |
36 | #include <sys/sysinfo.h> | |
37 | #include <sys/vfs.h> | |
38 | ||
39 | #include "bindings.h" | |
40 | #include "config.h" // for VERSION | |
41 | ||
42 | /* Maximum number for 64 bit integer is a string with 21 digits: 2^64 - 1 = 21 */ | |
43 | #define LXCFS_NUMSTRLEN64 21 | |
44 | ||
45 | /* Define pivot_root() if missing from the C library */ | |
46 | #ifndef HAVE_PIVOT_ROOT | |
47 | static int pivot_root(const char * new_root, const char * put_old) | |
48 | { | |
49 | #ifdef __NR_pivot_root | |
50 | return syscall(__NR_pivot_root, new_root, put_old); | |
51 | #else | |
52 | errno = ENOSYS; | |
53 | return -1; | |
54 | #endif | |
55 | } | |
56 | #else | |
57 | extern int pivot_root(const char * new_root, const char * put_old); | |
58 | #endif | |
59 | ||
60 | enum { | |
61 | LXC_TYPE_CGDIR, | |
62 | LXC_TYPE_CGFILE, | |
63 | LXC_TYPE_PROC_MEMINFO, | |
64 | LXC_TYPE_PROC_CPUINFO, | |
65 | LXC_TYPE_PROC_UPTIME, | |
66 | LXC_TYPE_PROC_STAT, | |
67 | LXC_TYPE_PROC_DISKSTATS, | |
68 | LXC_TYPE_PROC_SWAPS, | |
69 | }; | |
70 | ||
71 | struct file_info { | |
72 | char *controller; | |
73 | char *cgroup; | |
74 | char *file; | |
75 | int type; | |
76 | char *buf; // unused as of yet | |
77 | int buflen; | |
78 | int size; //actual data size | |
79 | int cached; | |
80 | }; | |
81 | ||
82 | /* Reserve buffer size to account for file size changes. */ | |
83 | #define BUF_RESERVE_SIZE 512 | |
84 | ||
85 | /* | |
86 | * A table caching which pid is init for a pid namespace. | |
87 | * When looking up which pid is init for $qpid, we first | |
88 | * 1. Stat /proc/$qpid/ns/pid. | |
89 | * 2. Check whether the ino_t is in our store. | |
90 | * a. if not, fork a child in qpid's ns to send us | |
91 | * ucred.pid = 1, and read the initpid. Cache | |
92 | * initpid and creation time for /proc/initpid | |
93 | * in a new store entry. | |
94 | * b. if so, verify that /proc/initpid still matches | |
95 | * what we have saved. If not, clear the store | |
96 | * entry and go back to a. If so, return the | |
97 | * cached initpid. | |
98 | */ | |
99 | struct pidns_init_store { | |
100 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
101 | pid_t initpid; // the pid of nit in that ns | |
102 | long int ctime; // the time at which /proc/$initpid was created | |
103 | struct pidns_init_store *next; | |
104 | long int lastcheck; | |
105 | }; | |
106 | ||
107 | /* lol - look at how they are allocated in the kernel */ | |
108 | #define PIDNS_HASH_SIZE 4096 | |
109 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
110 | ||
111 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
112 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
113 | static void lock_mutex(pthread_mutex_t *l) | |
114 | { | |
115 | int ret; | |
116 | ||
117 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
118 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); | |
119 | exit(1); | |
120 | } | |
121 | } | |
122 | ||
123 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
124 | * Number of hierarchies mounted. */ | |
125 | static int num_hierarchies; | |
126 | ||
127 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
128 | * Hierachies mounted {cpuset, blkio, ...}: | |
129 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
130 | static char **hierarchies; | |
131 | ||
132 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
133 | * Open file descriptors: | |
134 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
135 | * private mount namespace. | |
136 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
137 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
138 | * mounts and respective files in the private namespace even when located in | |
139 | * another namespace using the *at() family of functions | |
140 | * {openat(), fchownat(), ...}. */ | |
141 | static int *fd_hierarchies; | |
142 | static int cgroup_mount_ns_fd = -1; | |
143 | ||
144 | static void unlock_mutex(pthread_mutex_t *l) | |
145 | { | |
146 | int ret; | |
147 | ||
148 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
149 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); | |
150 | exit(1); | |
151 | } | |
152 | } | |
153 | ||
154 | static void store_lock(void) | |
155 | { | |
156 | lock_mutex(&pidns_store_mutex); | |
157 | } | |
158 | ||
159 | static void store_unlock(void) | |
160 | { | |
161 | unlock_mutex(&pidns_store_mutex); | |
162 | } | |
163 | ||
164 | /* Must be called under store_lock */ | |
165 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
166 | { | |
167 | struct stat initsb; | |
168 | char fnam[100]; | |
169 | ||
170 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
171 | if (stat(fnam, &initsb) < 0) | |
172 | return false; | |
173 | ||
174 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
175 | initsb.st_ctime, e->initpid); | |
176 | ||
177 | if (e->ctime != initsb.st_ctime) | |
178 | return false; | |
179 | return true; | |
180 | } | |
181 | ||
182 | /* Must be called under store_lock */ | |
183 | static void remove_initpid(struct pidns_init_store *e) | |
184 | { | |
185 | struct pidns_init_store *tmp; | |
186 | int h; | |
187 | ||
188 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); | |
189 | ||
190 | h = HASH(e->ino); | |
191 | if (pidns_hash_table[h] == e) { | |
192 | pidns_hash_table[h] = e->next; | |
193 | free(e); | |
194 | return; | |
195 | } | |
196 | ||
197 | tmp = pidns_hash_table[h]; | |
198 | while (tmp) { | |
199 | if (tmp->next == e) { | |
200 | tmp->next = e->next; | |
201 | free(e); | |
202 | return; | |
203 | } | |
204 | tmp = tmp->next; | |
205 | } | |
206 | } | |
207 | ||
208 | #define PURGE_SECS 5 | |
209 | /* Must be called under store_lock */ | |
210 | static void prune_initpid_store(void) | |
211 | { | |
212 | static long int last_prune = 0; | |
213 | struct pidns_init_store *e, *prev, *delme; | |
214 | long int now, threshold; | |
215 | int i; | |
216 | ||
217 | if (!last_prune) { | |
218 | last_prune = time(NULL); | |
219 | return; | |
220 | } | |
221 | now = time(NULL); | |
222 | if (now < last_prune + PURGE_SECS) | |
223 | return; | |
224 | ||
225 | lxcfs_debug("%s\n", "Pruning."); | |
226 | ||
227 | last_prune = now; | |
228 | threshold = now - 2 * PURGE_SECS; | |
229 | ||
230 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
231 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
232 | if (e->lastcheck < threshold) { | |
233 | ||
234 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
235 | ||
236 | delme = e; | |
237 | if (prev) | |
238 | prev->next = e->next; | |
239 | else | |
240 | pidns_hash_table[i] = e->next; | |
241 | e = e->next; | |
242 | free(delme); | |
243 | } else { | |
244 | prev = e; | |
245 | e = e->next; | |
246 | } | |
247 | } | |
248 | } | |
249 | } | |
250 | ||
251 | /* Must be called under store_lock */ | |
252 | static void save_initpid(struct stat *sb, pid_t pid) | |
253 | { | |
254 | struct pidns_init_store *e; | |
255 | char fpath[100]; | |
256 | struct stat procsb; | |
257 | int h; | |
258 | ||
259 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); | |
260 | ||
261 | snprintf(fpath, 100, "/proc/%d", pid); | |
262 | if (stat(fpath, &procsb) < 0) | |
263 | return; | |
264 | do { | |
265 | e = malloc(sizeof(*e)); | |
266 | } while (!e); | |
267 | e->ino = sb->st_ino; | |
268 | e->initpid = pid; | |
269 | e->ctime = procsb.st_ctime; | |
270 | h = HASH(e->ino); | |
271 | e->next = pidns_hash_table[h]; | |
272 | e->lastcheck = time(NULL); | |
273 | pidns_hash_table[h] = e; | |
274 | } | |
275 | ||
276 | /* | |
277 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
278 | * entry for the inode number and creation time. Verify that the init pid | |
279 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
280 | * otherwise. | |
281 | * Must be called under store_lock | |
282 | */ | |
283 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
284 | { | |
285 | int h = HASH(sb->st_ino); | |
286 | struct pidns_init_store *e = pidns_hash_table[h]; | |
287 | ||
288 | while (e) { | |
289 | if (e->ino == sb->st_ino) { | |
290 | if (initpid_still_valid(e, sb)) { | |
291 | e->lastcheck = time(NULL); | |
292 | return e; | |
293 | } | |
294 | remove_initpid(e); | |
295 | return NULL; | |
296 | } | |
297 | e = e->next; | |
298 | } | |
299 | ||
300 | return NULL; | |
301 | } | |
302 | ||
303 | static int is_dir(const char *path, int fd) | |
304 | { | |
305 | struct stat statbuf; | |
306 | int ret = fstatat(fd, path, &statbuf, fd); | |
307 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) | |
308 | return 1; | |
309 | return 0; | |
310 | } | |
311 | ||
312 | static char *must_copy_string(const char *str) | |
313 | { | |
314 | char *dup = NULL; | |
315 | if (!str) | |
316 | return NULL; | |
317 | do { | |
318 | dup = strdup(str); | |
319 | } while (!dup); | |
320 | ||
321 | return dup; | |
322 | } | |
323 | ||
324 | static inline void drop_trailing_newlines(char *s) | |
325 | { | |
326 | int l; | |
327 | ||
328 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
329 | s[l-1] = '\0'; | |
330 | } | |
331 | ||
332 | #define BATCH_SIZE 50 | |
333 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
334 | { | |
335 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
336 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
337 | ||
338 | if (!*mem || newbatches > oldbatches) { | |
339 | char *tmp; | |
340 | do { | |
341 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
342 | } while (!tmp); | |
343 | *mem = tmp; | |
344 | } | |
345 | } | |
346 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
347 | { | |
348 | size_t newlen = *len + linelen; | |
349 | dorealloc(contents, *len, newlen + 1); | |
350 | memcpy(*contents + *len, line, linelen+1); | |
351 | *len = newlen; | |
352 | } | |
353 | ||
354 | static char *slurp_file(const char *from, int fd) | |
355 | { | |
356 | char *line = NULL; | |
357 | char *contents = NULL; | |
358 | FILE *f = fdopen(fd, "r"); | |
359 | size_t len = 0, fulllen = 0; | |
360 | ssize_t linelen; | |
361 | ||
362 | if (!f) | |
363 | return NULL; | |
364 | ||
365 | while ((linelen = getline(&line, &len, f)) != -1) { | |
366 | append_line(&contents, &fulllen, line, linelen); | |
367 | } | |
368 | fclose(f); | |
369 | ||
370 | if (contents) | |
371 | drop_trailing_newlines(contents); | |
372 | free(line); | |
373 | return contents; | |
374 | } | |
375 | ||
376 | static bool write_string(const char *fnam, const char *string, int fd) | |
377 | { | |
378 | FILE *f; | |
379 | size_t len, ret; | |
380 | ||
381 | if (!(f = fdopen(fd, "w"))) | |
382 | return false; | |
383 | len = strlen(string); | |
384 | ret = fwrite(string, 1, len, f); | |
385 | if (ret != len) { | |
386 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); | |
387 | fclose(f); | |
388 | return false; | |
389 | } | |
390 | if (fclose(f) < 0) { | |
391 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); | |
392 | return false; | |
393 | } | |
394 | return true; | |
395 | } | |
396 | ||
397 | struct cgfs_files { | |
398 | char *name; | |
399 | uint32_t uid, gid; | |
400 | uint32_t mode; | |
401 | }; | |
402 | ||
403 | #define ALLOC_NUM 20 | |
404 | static bool store_hierarchy(char *stridx, char *h) | |
405 | { | |
406 | if (num_hierarchies % ALLOC_NUM == 0) { | |
407 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
408 | n *= ALLOC_NUM; | |
409 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
410 | if (!tmp) { | |
411 | lxcfs_error("%s\n", strerror(errno)); | |
412 | exit(1); | |
413 | } | |
414 | hierarchies = tmp; | |
415 | } | |
416 | ||
417 | hierarchies[num_hierarchies++] = must_copy_string(h); | |
418 | return true; | |
419 | } | |
420 | ||
421 | static void print_subsystems(void) | |
422 | { | |
423 | int i; | |
424 | ||
425 | fprintf(stderr, "mount namespace: %d\n", cgroup_mount_ns_fd); | |
426 | fprintf(stderr, "hierarchies:\n"); | |
427 | for (i = 0; i < num_hierarchies; i++) { | |
428 | if (hierarchies[i]) | |
429 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, | |
430 | fd_hierarchies[i], hierarchies[i]); | |
431 | } | |
432 | } | |
433 | ||
434 | static bool in_comma_list(const char *needle, const char *haystack) | |
435 | { | |
436 | const char *s = haystack, *e; | |
437 | size_t nlen = strlen(needle); | |
438 | ||
439 | while (*s && (e = strchr(s, ','))) { | |
440 | if (nlen != e - s) { | |
441 | s = e + 1; | |
442 | continue; | |
443 | } | |
444 | if (strncmp(needle, s, nlen) == 0) | |
445 | return true; | |
446 | s = e + 1; | |
447 | } | |
448 | if (strcmp(needle, s) == 0) | |
449 | return true; | |
450 | return false; | |
451 | } | |
452 | ||
453 | /* do we need to do any massaging here? I'm not sure... */ | |
454 | /* Return the mounted controller and store the corresponding open file descriptor | |
455 | * referring to the controller mountpoint in the private lxcfs namespace in | |
456 | * @cfd. | |
457 | */ | |
458 | static char *find_mounted_controller(const char *controller, int *cfd) | |
459 | { | |
460 | int i; | |
461 | ||
462 | for (i = 0; i < num_hierarchies; i++) { | |
463 | if (!hierarchies[i]) | |
464 | continue; | |
465 | if (strcmp(hierarchies[i], controller) == 0) { | |
466 | *cfd = fd_hierarchies[i]; | |
467 | return hierarchies[i]; | |
468 | } | |
469 | if (in_comma_list(controller, hierarchies[i])) { | |
470 | *cfd = fd_hierarchies[i]; | |
471 | return hierarchies[i]; | |
472 | } | |
473 | } | |
474 | ||
475 | return NULL; | |
476 | } | |
477 | ||
478 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
479 | const char *value) | |
480 | { | |
481 | int ret, fd, cfd; | |
482 | size_t len; | |
483 | char *fnam, *tmpc; | |
484 | ||
485 | tmpc = find_mounted_controller(controller, &cfd); | |
486 | if (!tmpc) | |
487 | return false; | |
488 | ||
489 | /* Make sure we pass a relative path to *at() family of functions. | |
490 | * . + /cgroup + / + file + \0 | |
491 | */ | |
492 | len = strlen(cgroup) + strlen(file) + 3; | |
493 | fnam = alloca(len); | |
494 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); | |
495 | if (ret < 0 || (size_t)ret >= len) | |
496 | return false; | |
497 | ||
498 | fd = openat(cfd, fnam, O_WRONLY); | |
499 | if (fd < 0) | |
500 | return false; | |
501 | ||
502 | return write_string(fnam, value, fd); | |
503 | } | |
504 | ||
505 | // Chown all the files in the cgroup directory. We do this when we create | |
506 | // a cgroup on behalf of a user. | |
507 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) | |
508 | { | |
509 | struct dirent *direntp; | |
510 | char path[MAXPATHLEN]; | |
511 | size_t len; | |
512 | DIR *d; | |
513 | int fd1, ret; | |
514 | ||
515 | len = strlen(dirname); | |
516 | if (len >= MAXPATHLEN) { | |
517 | lxcfs_error("Pathname too long: %s\n", dirname); | |
518 | return; | |
519 | } | |
520 | ||
521 | fd1 = openat(fd, dirname, O_DIRECTORY); | |
522 | if (fd1 < 0) | |
523 | return; | |
524 | ||
525 | d = fdopendir(fd1); | |
526 | if (!d) { | |
527 | lxcfs_error("Failed to open %s\n", dirname); | |
528 | return; | |
529 | } | |
530 | ||
531 | while ((direntp = readdir(d))) { | |
532 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) | |
533 | continue; | |
534 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
535 | if (ret < 0 || ret >= MAXPATHLEN) { | |
536 | lxcfs_error("Pathname too long under %s\n", dirname); | |
537 | continue; | |
538 | } | |
539 | if (fchownat(fd, path, uid, gid, 0) < 0) | |
540 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); | |
541 | } | |
542 | closedir(d); | |
543 | } | |
544 | ||
545 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
546 | { | |
547 | int cfd; | |
548 | size_t len; | |
549 | char *dirnam, *tmpc; | |
550 | ||
551 | tmpc = find_mounted_controller(controller, &cfd); | |
552 | if (!tmpc) | |
553 | return -EINVAL; | |
554 | ||
555 | /* Make sure we pass a relative path to *at() family of functions. | |
556 | * . + /cg + \0 | |
557 | */ | |
558 | len = strlen(cg) + 2; | |
559 | dirnam = alloca(len); | |
560 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); | |
561 | ||
562 | if (mkdirat(cfd, dirnam, 0755) < 0) | |
563 | return -errno; | |
564 | ||
565 | if (uid == 0 && gid == 0) | |
566 | return 0; | |
567 | ||
568 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) | |
569 | return -errno; | |
570 | ||
571 | chown_all_cgroup_files(dirnam, uid, gid, cfd); | |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
576 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) | |
577 | { | |
578 | struct dirent *direntp; | |
579 | DIR *dir; | |
580 | bool ret = false; | |
581 | char pathname[MAXPATHLEN]; | |
582 | int dupfd; | |
583 | ||
584 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
585 | if (dupfd < 0) | |
586 | return false; | |
587 | ||
588 | dir = fdopendir(dupfd); | |
589 | if (!dir) { | |
590 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); | |
591 | close(dupfd); | |
592 | return false; | |
593 | } | |
594 | ||
595 | while ((direntp = readdir(dir))) { | |
596 | struct stat mystat; | |
597 | int rc; | |
598 | ||
599 | if (!strcmp(direntp->d_name, ".") || | |
600 | !strcmp(direntp->d_name, "..")) | |
601 | continue; | |
602 | ||
603 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
604 | if (rc < 0 || rc >= MAXPATHLEN) { | |
605 | lxcfs_error("%s\n", "Pathname too long."); | |
606 | continue; | |
607 | } | |
608 | ||
609 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); | |
610 | if (rc) { | |
611 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); | |
612 | continue; | |
613 | } | |
614 | if (S_ISDIR(mystat.st_mode)) | |
615 | if (!recursive_rmdir(pathname, fd, cfd)) | |
616 | lxcfs_debug("Error removing %s.\n", pathname); | |
617 | } | |
618 | ||
619 | ret = true; | |
620 | if (closedir(dir) < 0) { | |
621 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); | |
622 | ret = false; | |
623 | } | |
624 | ||
625 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { | |
626 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); | |
627 | ret = false; | |
628 | } | |
629 | ||
630 | close(dupfd); | |
631 | ||
632 | return ret; | |
633 | } | |
634 | ||
635 | bool cgfs_remove(const char *controller, const char *cg) | |
636 | { | |
637 | int fd, cfd; | |
638 | size_t len; | |
639 | char *dirnam, *tmpc; | |
640 | bool bret; | |
641 | ||
642 | tmpc = find_mounted_controller(controller, &cfd); | |
643 | if (!tmpc) | |
644 | return false; | |
645 | ||
646 | /* Make sure we pass a relative path to *at() family of functions. | |
647 | * . + /cg + \0 | |
648 | */ | |
649 | len = strlen(cg) + 2; | |
650 | dirnam = alloca(len); | |
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 | bret = recursive_rmdir(dirnam, fd, cfd); | |
658 | close(fd); | |
659 | return bret; | |
660 | } | |
661 | ||
662 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
663 | { | |
664 | int cfd; | |
665 | size_t len; | |
666 | char *pathname, *tmpc; | |
667 | ||
668 | tmpc = find_mounted_controller(controller, &cfd); | |
669 | if (!tmpc) | |
670 | return false; | |
671 | ||
672 | /* Make sure we pass a relative path to *at() family of functions. | |
673 | * . + /file + \0 | |
674 | */ | |
675 | len = strlen(file) + 2; | |
676 | pathname = alloca(len); | |
677 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); | |
678 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
679 | return false; | |
680 | return true; | |
681 | } | |
682 | ||
683 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) | |
684 | { | |
685 | size_t len; | |
686 | char *fname; | |
687 | ||
688 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
689 | fname = alloca(len); | |
690 | snprintf(fname, len, "%s/tasks", dirname); | |
691 | if (fchownat(fd, fname, uid, gid, 0) != 0) | |
692 | return -errno; | |
693 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
694 | if (fchownat(fd, fname, uid, gid, 0) != 0) | |
695 | return -errno; | |
696 | return 0; | |
697 | } | |
698 | ||
699 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
700 | { | |
701 | int cfd; | |
702 | size_t len; | |
703 | char *pathname, *tmpc; | |
704 | ||
705 | tmpc = find_mounted_controller(controller, &cfd); | |
706 | if (!tmpc) | |
707 | return -EINVAL; | |
708 | ||
709 | /* Make sure we pass a relative path to *at() family of functions. | |
710 | * . + /file + \0 | |
711 | */ | |
712 | len = strlen(file) + 2; | |
713 | pathname = alloca(len); | |
714 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); | |
715 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
716 | return -errno; | |
717 | ||
718 | if (is_dir(pathname, cfd)) | |
719 | // like cgmanager did, we want to chown the tasks file as well | |
720 | return chown_tasks_files(pathname, uid, gid, cfd); | |
721 | ||
722 | return 0; | |
723 | } | |
724 | ||
725 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
726 | { | |
727 | int fd, cfd; | |
728 | size_t len; | |
729 | char *pathname, *tmpc; | |
730 | ||
731 | tmpc = find_mounted_controller(controller, &cfd); | |
732 | if (!tmpc) | |
733 | return NULL; | |
734 | ||
735 | /* Make sure we pass a relative path to *at() family of functions. | |
736 | * . + /cgroup + / "cgroup.procs" + \0 | |
737 | */ | |
738 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; | |
739 | pathname = alloca(len); | |
740 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); | |
741 | ||
742 | fd = openat(cfd, pathname, O_WRONLY); | |
743 | if (fd < 0) | |
744 | return NULL; | |
745 | ||
746 | return fdopen(fd, "w"); | |
747 | } | |
748 | ||
749 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, | |
750 | void ***list, size_t typesize, | |
751 | void* (*iterator)(const char*, const char*, const char*)) | |
752 | { | |
753 | int cfd, fd, ret; | |
754 | size_t len; | |
755 | char *cg, *tmpc; | |
756 | char pathname[MAXPATHLEN]; | |
757 | size_t sz = 0, asz = 0; | |
758 | struct dirent *dirent; | |
759 | DIR *dir; | |
760 | ||
761 | tmpc = find_mounted_controller(controller, &cfd); | |
762 | *list = NULL; | |
763 | if (!tmpc) | |
764 | return false; | |
765 | ||
766 | /* Make sure we pass a relative path to *at() family of functions. */ | |
767 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; | |
768 | cg = alloca(len); | |
769 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
770 | if (ret < 0 || (size_t)ret >= len) { | |
771 | lxcfs_error("Pathname too long under %s\n", cgroup); | |
772 | return false; | |
773 | } | |
774 | ||
775 | fd = openat(cfd, cg, O_DIRECTORY); | |
776 | if (fd < 0) | |
777 | return false; | |
778 | ||
779 | dir = fdopendir(fd); | |
780 | if (!dir) | |
781 | return false; | |
782 | ||
783 | while ((dirent = readdir(dir))) { | |
784 | struct stat mystat; | |
785 | ||
786 | if (!strcmp(dirent->d_name, ".") || | |
787 | !strcmp(dirent->d_name, "..")) | |
788 | continue; | |
789 | ||
790 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); | |
791 | if (ret < 0 || ret >= MAXPATHLEN) { | |
792 | lxcfs_error("Pathname too long under %s\n", cg); | |
793 | continue; | |
794 | } | |
795 | ||
796 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); | |
797 | if (ret) { | |
798 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); | |
799 | continue; | |
800 | } | |
801 | if ((!directories && !S_ISREG(mystat.st_mode)) || | |
802 | (directories && !S_ISDIR(mystat.st_mode))) | |
803 | continue; | |
804 | ||
805 | if (sz+2 >= asz) { | |
806 | void **tmp; | |
807 | asz += BATCH_SIZE; | |
808 | do { | |
809 | tmp = realloc(*list, asz * typesize); | |
810 | } while (!tmp); | |
811 | *list = tmp; | |
812 | } | |
813 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); | |
814 | (*list)[sz+1] = NULL; | |
815 | sz++; | |
816 | } | |
817 | if (closedir(dir) < 0) { | |
818 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); | |
819 | return false; | |
820 | } | |
821 | return true; | |
822 | } | |
823 | ||
824 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) | |
825 | { | |
826 | char *dup; | |
827 | do { | |
828 | dup = strdup(dir_entry); | |
829 | } while (!dup); | |
830 | return dup; | |
831 | } | |
832 | ||
833 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
834 | { | |
835 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
836 | } | |
837 | ||
838 | void free_key(struct cgfs_files *k) | |
839 | { | |
840 | if (!k) | |
841 | return; | |
842 | free(k->name); | |
843 | free(k); | |
844 | } | |
845 | ||
846 | void free_keys(struct cgfs_files **keys) | |
847 | { | |
848 | int i; | |
849 | ||
850 | if (!keys) | |
851 | return; | |
852 | for (i = 0; keys[i]; i++) { | |
853 | free_key(keys[i]); | |
854 | } | |
855 | free(keys); | |
856 | } | |
857 | ||
858 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
859 | { | |
860 | int ret, fd, cfd; | |
861 | size_t len; | |
862 | char *fnam, *tmpc; | |
863 | ||
864 | tmpc = find_mounted_controller(controller, &cfd); | |
865 | if (!tmpc) | |
866 | return false; | |
867 | ||
868 | /* Make sure we pass a relative path to *at() family of functions. | |
869 | * . + /cgroup + / + file + \0 | |
870 | */ | |
871 | len = strlen(cgroup) + strlen(file) + 3; | |
872 | fnam = alloca(len); | |
873 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); | |
874 | if (ret < 0 || (size_t)ret >= len) | |
875 | return false; | |
876 | ||
877 | fd = openat(cfd, fnam, O_RDONLY); | |
878 | if (fd < 0) | |
879 | return false; | |
880 | ||
881 | *value = slurp_file(fnam, fd); | |
882 | return *value != NULL; | |
883 | } | |
884 | ||
885 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
886 | { | |
887 | int ret, cfd; | |
888 | size_t len; | |
889 | char *fnam, *tmpc; | |
890 | struct stat sb; | |
891 | struct cgfs_files *newkey; | |
892 | ||
893 | tmpc = find_mounted_controller(controller, &cfd); | |
894 | if (!tmpc) | |
895 | return false; | |
896 | ||
897 | if (file && *file == '/') | |
898 | file++; | |
899 | ||
900 | if (file && strchr(file, '/')) | |
901 | return NULL; | |
902 | ||
903 | /* Make sure we pass a relative path to *at() family of functions. | |
904 | * . + /cgroup + / + file + \0 | |
905 | */ | |
906 | len = strlen(cgroup) + 3; | |
907 | if (file) | |
908 | len += strlen(file) + 1; | |
909 | fnam = alloca(len); | |
910 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, | |
911 | file ? "/" : "", file ? file : ""); | |
912 | ||
913 | ret = fstatat(cfd, fnam, &sb, 0); | |
914 | if (ret < 0) | |
915 | return NULL; | |
916 | ||
917 | do { | |
918 | newkey = malloc(sizeof(struct cgfs_files)); | |
919 | } while (!newkey); | |
920 | if (file) | |
921 | newkey->name = must_copy_string(file); | |
922 | else if (strrchr(cgroup, '/')) | |
923 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
924 | else | |
925 | newkey->name = must_copy_string(cgroup); | |
926 | newkey->uid = sb.st_uid; | |
927 | newkey->gid = sb.st_gid; | |
928 | newkey->mode = sb.st_mode; | |
929 | ||
930 | return newkey; | |
931 | } | |
932 | ||
933 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) | |
934 | { | |
935 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); | |
936 | if (!entry) { | |
937 | lxcfs_error("Error getting files under %s:%s\n", controller, | |
938 | cgroup); | |
939 | } | |
940 | return entry; | |
941 | } | |
942 | ||
943 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
944 | { | |
945 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
946 | } | |
947 | ||
948 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
949 | { | |
950 | int cfd; | |
951 | size_t len; | |
952 | char *fnam, *tmpc; | |
953 | int ret; | |
954 | struct stat sb; | |
955 | ||
956 | tmpc = find_mounted_controller(controller, &cfd); | |
957 | if (!tmpc) | |
958 | return false; | |
959 | ||
960 | /* Make sure we pass a relative path to *at() family of functions. | |
961 | * . + /cgroup + / + f + \0 | |
962 | */ | |
963 | len = strlen(cgroup) + strlen(f) + 3; | |
964 | fnam = alloca(len); | |
965 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); | |
966 | if (ret < 0 || (size_t)ret >= len) | |
967 | return false; | |
968 | ||
969 | ret = fstatat(cfd, fnam, &sb, 0); | |
970 | if (ret < 0 || !S_ISDIR(sb.st_mode)) | |
971 | return false; | |
972 | ||
973 | return true; | |
974 | } | |
975 | ||
976 | #define SEND_CREDS_OK 0 | |
977 | #define SEND_CREDS_NOTSK 1 | |
978 | #define SEND_CREDS_FAIL 2 | |
979 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
980 | static int wait_for_pid(pid_t pid); | |
981 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
982 | static int send_creds_clone_wrapper(void *arg); | |
983 | ||
984 | /* | |
985 | * clone a task which switches to @task's namespace and writes '1'. | |
986 | * over a unix sock so we can read the task's reaper's pid in our | |
987 | * namespace | |
988 | * | |
989 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
990 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
991 | * the pidns and the parent pid outside are identical. Using clone prevents | |
992 | * this issue. | |
993 | */ | |
994 | static void write_task_init_pid_exit(int sock, pid_t target) | |
995 | { | |
996 | char fnam[100]; | |
997 | pid_t pid; | |
998 | int fd, ret; | |
999 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
1000 | void *stack = alloca(stack_size); | |
1001 | ||
1002 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1003 | if (ret < 0 || ret >= sizeof(fnam)) | |
1004 | _exit(1); | |
1005 | ||
1006 | fd = open(fnam, O_RDONLY); | |
1007 | if (fd < 0) { | |
1008 | perror("write_task_init_pid_exit open of ns/pid"); | |
1009 | _exit(1); | |
1010 | } | |
1011 | if (setns(fd, 0)) { | |
1012 | perror("write_task_init_pid_exit setns 1"); | |
1013 | close(fd); | |
1014 | _exit(1); | |
1015 | } | |
1016 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); | |
1017 | if (pid < 0) | |
1018 | _exit(1); | |
1019 | if (pid != 0) { | |
1020 | if (!wait_for_pid(pid)) | |
1021 | _exit(1); | |
1022 | _exit(0); | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | static int send_creds_clone_wrapper(void *arg) { | |
1027 | struct ucred cred; | |
1028 | char v; | |
1029 | int sock = *(int *)arg; | |
1030 | ||
1031 | /* we are the child */ | |
1032 | cred.uid = 0; | |
1033 | cred.gid = 0; | |
1034 | cred.pid = 1; | |
1035 | v = '1'; | |
1036 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
1037 | return 1; | |
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | static pid_t get_init_pid_for_task(pid_t task) | |
1042 | { | |
1043 | int sock[2]; | |
1044 | pid_t pid; | |
1045 | pid_t ret = -1; | |
1046 | char v = '0'; | |
1047 | struct ucred cred; | |
1048 | ||
1049 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1050 | perror("socketpair"); | |
1051 | return -1; | |
1052 | } | |
1053 | ||
1054 | pid = fork(); | |
1055 | if (pid < 0) | |
1056 | goto out; | |
1057 | if (!pid) { | |
1058 | close(sock[1]); | |
1059 | write_task_init_pid_exit(sock[0], task); | |
1060 | _exit(0); | |
1061 | } | |
1062 | ||
1063 | if (!recv_creds(sock[1], &cred, &v)) | |
1064 | goto out; | |
1065 | ret = cred.pid; | |
1066 | ||
1067 | out: | |
1068 | close(sock[0]); | |
1069 | close(sock[1]); | |
1070 | if (pid > 0) | |
1071 | wait_for_pid(pid); | |
1072 | return ret; | |
1073 | } | |
1074 | ||
1075 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1076 | { | |
1077 | pid_t answer = 0; | |
1078 | struct stat sb; | |
1079 | struct pidns_init_store *e; | |
1080 | char fnam[100]; | |
1081 | ||
1082 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1083 | store_lock(); | |
1084 | if (stat(fnam, &sb) < 0) | |
1085 | goto out; | |
1086 | e = lookup_verify_initpid(&sb); | |
1087 | if (e) { | |
1088 | answer = e->initpid; | |
1089 | goto out; | |
1090 | } | |
1091 | answer = get_init_pid_for_task(qpid); | |
1092 | if (answer > 0) | |
1093 | save_initpid(&sb, answer); | |
1094 | ||
1095 | out: | |
1096 | /* we prune at end in case we are returning | |
1097 | * the value we were about to return */ | |
1098 | prune_initpid_store(); | |
1099 | store_unlock(); | |
1100 | return answer; | |
1101 | } | |
1102 | ||
1103 | static int wait_for_pid(pid_t pid) | |
1104 | { | |
1105 | int status, ret; | |
1106 | ||
1107 | if (pid <= 0) | |
1108 | return -1; | |
1109 | ||
1110 | again: | |
1111 | ret = waitpid(pid, &status, 0); | |
1112 | if (ret == -1) { | |
1113 | if (errno == EINTR) | |
1114 | goto again; | |
1115 | return -1; | |
1116 | } | |
1117 | if (ret != pid) | |
1118 | goto again; | |
1119 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1120 | return -1; | |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | ||
1125 | /* | |
1126 | * append pid to *src. | |
1127 | * src: a pointer to a char* in which ot append the pid. | |
1128 | * sz: the number of characters printed so far, minus trailing \0. | |
1129 | * asz: the allocated size so far | |
1130 | * pid: the pid to append | |
1131 | */ | |
1132 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1133 | { | |
1134 | char tmp[30]; | |
1135 | ||
1136 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1137 | ||
1138 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1139 | char *tmp; | |
1140 | do { | |
1141 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1142 | } while (!tmp); | |
1143 | *src = tmp; | |
1144 | *asz += BUF_RESERVE_SIZE; | |
1145 | } | |
1146 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ | |
1147 | *sz += tmplen; | |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1152 | * valid in the caller's namespace, return the id mapped into | |
1153 | * pid's namespace. | |
1154 | * Returns the mapped id, or -1 on error. | |
1155 | */ | |
1156 | unsigned int | |
1157 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1158 | { | |
1159 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1160 | hostuid, // base id for a range in the caller's namespace | |
1161 | count; // number of ids in this range | |
1162 | char line[400]; | |
1163 | int ret; | |
1164 | ||
1165 | fseek(idfile, 0L, SEEK_SET); | |
1166 | while (fgets(line, 400, idfile)) { | |
1167 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1168 | if (ret != 3) | |
1169 | continue; | |
1170 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1171 | /* | |
1172 | * uids wrapped around - unexpected as this is a procfile, | |
1173 | * so just bail. | |
1174 | */ | |
1175 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", | |
1176 | nsuid, hostuid, count, line); | |
1177 | return -1; | |
1178 | } | |
1179 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1180 | /* | |
1181 | * now since hostuid <= in_id < hostuid+count, and | |
1182 | * hostuid+count and nsuid+count do not wrap around, | |
1183 | * we know that nsuid+(in_id-hostuid) which must be | |
1184 | * less that nsuid+(count) must not wrap around | |
1185 | */ | |
1186 | return (in_id - hostuid) + nsuid; | |
1187 | } | |
1188 | } | |
1189 | ||
1190 | // no answer found | |
1191 | return -1; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * for is_privileged_over, | |
1196 | * specify whether we require the calling uid to be root in his | |
1197 | * namespace | |
1198 | */ | |
1199 | #define NS_ROOT_REQD true | |
1200 | #define NS_ROOT_OPT false | |
1201 | ||
1202 | #define PROCLEN 100 | |
1203 | ||
1204 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1205 | { | |
1206 | char fpath[PROCLEN]; | |
1207 | int ret; | |
1208 | bool answer = false; | |
1209 | uid_t nsuid; | |
1210 | ||
1211 | if (victim == -1 || uid == -1) | |
1212 | return false; | |
1213 | ||
1214 | /* | |
1215 | * If the request is one not requiring root in the namespace, | |
1216 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1217 | * access to files owned by uid 1000 | |
1218 | */ | |
1219 | if (!req_ns_root && uid == victim) | |
1220 | return true; | |
1221 | ||
1222 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1223 | if (ret < 0 || ret >= PROCLEN) | |
1224 | return false; | |
1225 | FILE *f = fopen(fpath, "r"); | |
1226 | if (!f) | |
1227 | return false; | |
1228 | ||
1229 | /* if caller's not root in his namespace, reject */ | |
1230 | nsuid = convert_id_to_ns(f, uid); | |
1231 | if (nsuid) | |
1232 | goto out; | |
1233 | ||
1234 | /* | |
1235 | * If victim is not mapped into caller's ns, reject. | |
1236 | * XXX I'm not sure this check is needed given that fuse | |
1237 | * will be sending requests where the vfs has converted | |
1238 | */ | |
1239 | nsuid = convert_id_to_ns(f, victim); | |
1240 | if (nsuid == -1) | |
1241 | goto out; | |
1242 | ||
1243 | answer = true; | |
1244 | ||
1245 | out: | |
1246 | fclose(f); | |
1247 | return answer; | |
1248 | } | |
1249 | ||
1250 | static bool perms_include(int fmode, mode_t req_mode) | |
1251 | { | |
1252 | mode_t r; | |
1253 | ||
1254 | switch (req_mode & O_ACCMODE) { | |
1255 | case O_RDONLY: | |
1256 | r = S_IROTH; | |
1257 | break; | |
1258 | case O_WRONLY: | |
1259 | r = S_IWOTH; | |
1260 | break; | |
1261 | case O_RDWR: | |
1262 | r = S_IROTH | S_IWOTH; | |
1263 | break; | |
1264 | default: | |
1265 | return false; | |
1266 | } | |
1267 | return ((fmode & r) == r); | |
1268 | } | |
1269 | ||
1270 | ||
1271 | /* | |
1272 | * taskcg is a/b/c | |
1273 | * querycg is /a/b/c/d/e | |
1274 | * we return 'd' | |
1275 | */ | |
1276 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1277 | { | |
1278 | char *start, *end; | |
1279 | ||
1280 | if (strlen(taskcg) <= strlen(querycg)) { | |
1281 | lxcfs_error("%s\n", "I was fed bad input."); | |
1282 | return NULL; | |
1283 | } | |
1284 | ||
1285 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) | |
1286 | start = strdup(taskcg + 1); | |
1287 | else | |
1288 | start = strdup(taskcg + strlen(querycg) + 1); | |
1289 | if (!start) | |
1290 | return NULL; | |
1291 | end = strchr(start, '/'); | |
1292 | if (end) | |
1293 | *end = '\0'; | |
1294 | return start; | |
1295 | } | |
1296 | ||
1297 | static void stripnewline(char *x) | |
1298 | { | |
1299 | size_t l = strlen(x); | |
1300 | if (l && x[l-1] == '\n') | |
1301 | x[l-1] = '\0'; | |
1302 | } | |
1303 | ||
1304 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1305 | { | |
1306 | int cfd; | |
1307 | char fnam[PROCLEN]; | |
1308 | FILE *f; | |
1309 | char *answer = NULL; | |
1310 | char *line = NULL; | |
1311 | size_t len = 0; | |
1312 | int ret; | |
1313 | const char *h = find_mounted_controller(contrl, &cfd); | |
1314 | if (!h) | |
1315 | return NULL; | |
1316 | ||
1317 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1318 | if (ret < 0 || ret >= PROCLEN) | |
1319 | return NULL; | |
1320 | if (!(f = fopen(fnam, "r"))) | |
1321 | return NULL; | |
1322 | ||
1323 | while (getline(&line, &len, f) != -1) { | |
1324 | char *c1, *c2; | |
1325 | if (!line[0]) | |
1326 | continue; | |
1327 | c1 = strchr(line, ':'); | |
1328 | if (!c1) | |
1329 | goto out; | |
1330 | c1++; | |
1331 | c2 = strchr(c1, ':'); | |
1332 | if (!c2) | |
1333 | goto out; | |
1334 | *c2 = '\0'; | |
1335 | if (strcmp(c1, h) != 0) | |
1336 | continue; | |
1337 | c2++; | |
1338 | stripnewline(c2); | |
1339 | do { | |
1340 | answer = strdup(c2); | |
1341 | } while (!answer); | |
1342 | break; | |
1343 | } | |
1344 | ||
1345 | out: | |
1346 | fclose(f); | |
1347 | free(line); | |
1348 | return answer; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * check whether a fuse context may access a cgroup dir or file | |
1353 | * | |
1354 | * If file is not null, it is a cgroup file to check under cg. | |
1355 | * If file is null, then we are checking perms on cg itself. | |
1356 | * | |
1357 | * For files we can check the mode of the list_keys result. | |
1358 | * For cgroups, we must make assumptions based on the files under the | |
1359 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1360 | * yet. | |
1361 | */ | |
1362 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1363 | { | |
1364 | struct cgfs_files *k = NULL; | |
1365 | bool ret = false; | |
1366 | ||
1367 | k = cgfs_get_key(contrl, cg, file); | |
1368 | if (!k) | |
1369 | return false; | |
1370 | ||
1371 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1372 | if (perms_include(k->mode >> 6, mode)) { | |
1373 | ret = true; | |
1374 | goto out; | |
1375 | } | |
1376 | } | |
1377 | if (fc->gid == k->gid) { | |
1378 | if (perms_include(k->mode >> 3, mode)) { | |
1379 | ret = true; | |
1380 | goto out; | |
1381 | } | |
1382 | } | |
1383 | ret = perms_include(k->mode, mode); | |
1384 | ||
1385 | out: | |
1386 | free_key(k); | |
1387 | return ret; | |
1388 | } | |
1389 | ||
1390 | #define INITSCOPE "/init.scope" | |
1391 | static void prune_init_slice(char *cg) | |
1392 | { | |
1393 | char *point; | |
1394 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1395 | ||
1396 | if (cg_len < initscope_len) | |
1397 | return; | |
1398 | ||
1399 | point = cg + cg_len - initscope_len; | |
1400 | if (strcmp(point, INITSCOPE) == 0) { | |
1401 | if (point == cg) | |
1402 | *(point+1) = '\0'; | |
1403 | else | |
1404 | *point = '\0'; | |
1405 | } | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1410 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1411 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1412 | * to a string containing the next cgroup directory under cg, which must be | |
1413 | * freed by the caller. | |
1414 | */ | |
1415 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1416 | { | |
1417 | bool answer = false; | |
1418 | char *c2 = get_pid_cgroup(pid, contrl); | |
1419 | char *linecmp; | |
1420 | ||
1421 | if (!c2) | |
1422 | return false; | |
1423 | prune_init_slice(c2); | |
1424 | ||
1425 | /* | |
1426 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise | |
1427 | * they pass in a cgroup without leading '/' | |
1428 | * | |
1429 | * The original line here was: | |
1430 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1431 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1432 | * Serge, do you know? | |
1433 | */ | |
1434 | if (*cg == '/' || !strncmp(cg, "./", 2)) | |
1435 | linecmp = c2; | |
1436 | else | |
1437 | linecmp = c2 + 1; | |
1438 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { | |
1439 | if (nextcg) { | |
1440 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1441 | } | |
1442 | goto out; | |
1443 | } | |
1444 | answer = true; | |
1445 | ||
1446 | out: | |
1447 | free(c2); | |
1448 | return answer; | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1453 | */ | |
1454 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1455 | { | |
1456 | bool answer = false; | |
1457 | char *c2, *task_cg; | |
1458 | size_t target_len, task_len; | |
1459 | ||
1460 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) | |
1461 | return true; | |
1462 | ||
1463 | c2 = get_pid_cgroup(pid, contrl); | |
1464 | if (!c2) | |
1465 | return false; | |
1466 | prune_init_slice(c2); | |
1467 | ||
1468 | task_cg = c2 + 1; | |
1469 | target_len = strlen(cg); | |
1470 | task_len = strlen(task_cg); | |
1471 | if (task_len == 0) { | |
1472 | /* Task is in the root cg, it can see everything. This case is | |
1473 | * not handled by the strmcps below, since they test for the | |
1474 | * last /, but that is the first / that we've chopped off | |
1475 | * above. | |
1476 | */ | |
1477 | answer = true; | |
1478 | goto out; | |
1479 | } | |
1480 | if (strcmp(cg, task_cg) == 0) { | |
1481 | answer = true; | |
1482 | goto out; | |
1483 | } | |
1484 | if (target_len < task_len) { | |
1485 | /* looking up a parent dir */ | |
1486 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1487 | answer = true; | |
1488 | goto out; | |
1489 | } | |
1490 | if (target_len > task_len) { | |
1491 | /* looking up a child dir */ | |
1492 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1493 | answer = true; | |
1494 | goto out; | |
1495 | } | |
1496 | ||
1497 | out: | |
1498 | free(c2); | |
1499 | return answer; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * given /cgroup/freezer/a/b, return "freezer". | |
1504 | * the returned char* should NOT be freed. | |
1505 | */ | |
1506 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1507 | { | |
1508 | const char *p1; | |
1509 | char *contr, *slash; | |
1510 | ||
1511 | if (strlen(path) < 9) { | |
1512 | errno = EACCES; | |
1513 | return NULL; | |
1514 | } | |
1515 | if (*(path + 7) != '/') { | |
1516 | errno = EINVAL; | |
1517 | return NULL; | |
1518 | } | |
1519 | p1 = path + 8; | |
1520 | contr = strdupa(p1); | |
1521 | if (!contr) { | |
1522 | errno = ENOMEM; | |
1523 | return NULL; | |
1524 | } | |
1525 | slash = strstr(contr, "/"); | |
1526 | if (slash) | |
1527 | *slash = '\0'; | |
1528 | ||
1529 | int i; | |
1530 | for (i = 0; i < num_hierarchies; i++) { | |
1531 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) | |
1532 | return hierarchies[i]; | |
1533 | } | |
1534 | errno = ENOENT; | |
1535 | return NULL; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1540 | * Note that the returned value may include files (keynames) etc | |
1541 | */ | |
1542 | static const char *find_cgroup_in_path(const char *path) | |
1543 | { | |
1544 | const char *p1; | |
1545 | ||
1546 | if (strlen(path) < 9) { | |
1547 | errno = EACCES; | |
1548 | return NULL; | |
1549 | } | |
1550 | p1 = strstr(path + 8, "/"); | |
1551 | if (!p1) { | |
1552 | errno = EINVAL; | |
1553 | return NULL; | |
1554 | } | |
1555 | errno = 0; | |
1556 | return p1 + 1; | |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * split the last path element from the path in @cg. | |
1561 | * @dir is newly allocated and should be freed, @last not | |
1562 | */ | |
1563 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1564 | { | |
1565 | char *p; | |
1566 | ||
1567 | do { | |
1568 | *dir = strdup(cg); | |
1569 | } while (!*dir); | |
1570 | *last = strrchr(cg, '/'); | |
1571 | if (!*last) { | |
1572 | *last = NULL; | |
1573 | return; | |
1574 | } | |
1575 | p = strrchr(*dir, '/'); | |
1576 | *p = '\0'; | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * FUSE ops for /cgroup | |
1581 | */ | |
1582 | ||
1583 | int cg_getattr(const char *path, struct stat *sb) | |
1584 | { | |
1585 | struct timespec now; | |
1586 | struct fuse_context *fc = fuse_get_context(); | |
1587 | char * cgdir = NULL; | |
1588 | char *last = NULL, *path1, *path2; | |
1589 | struct cgfs_files *k = NULL; | |
1590 | const char *cgroup; | |
1591 | const char *controller = NULL; | |
1592 | int ret = -ENOENT; | |
1593 | ||
1594 | ||
1595 | if (!fc) | |
1596 | return -EIO; | |
1597 | ||
1598 | memset(sb, 0, sizeof(struct stat)); | |
1599 | ||
1600 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1601 | return -EINVAL; | |
1602 | ||
1603 | sb->st_uid = sb->st_gid = 0; | |
1604 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1605 | sb->st_size = 0; | |
1606 | ||
1607 | if (strcmp(path, "/cgroup") == 0) { | |
1608 | sb->st_mode = S_IFDIR | 00755; | |
1609 | sb->st_nlink = 2; | |
1610 | return 0; | |
1611 | } | |
1612 | ||
1613 | controller = pick_controller_from_path(fc, path); | |
1614 | if (!controller) | |
1615 | return -errno; | |
1616 | cgroup = find_cgroup_in_path(path); | |
1617 | if (!cgroup) { | |
1618 | /* this is just /cgroup/controller, return it as a dir */ | |
1619 | sb->st_mode = S_IFDIR | 00755; | |
1620 | sb->st_nlink = 2; | |
1621 | return 0; | |
1622 | } | |
1623 | ||
1624 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1625 | ||
1626 | if (!last) { | |
1627 | path1 = "/"; | |
1628 | path2 = cgdir; | |
1629 | } else { | |
1630 | path1 = cgdir; | |
1631 | path2 = last; | |
1632 | } | |
1633 | ||
1634 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1635 | if (initpid <= 0) | |
1636 | initpid = fc->pid; | |
1637 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1638 | * Then check that caller's cgroup is under path if last is a child | |
1639 | * cgroup, or cgdir if last is a file */ | |
1640 | ||
1641 | if (is_child_cgroup(controller, path1, path2)) { | |
1642 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1643 | ret = -ENOENT; | |
1644 | goto out; | |
1645 | } | |
1646 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1647 | /* this is just /cgroup/controller, return it as a dir */ | |
1648 | sb->st_mode = S_IFDIR | 00555; | |
1649 | sb->st_nlink = 2; | |
1650 | ret = 0; | |
1651 | goto out; | |
1652 | } | |
1653 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1654 | ret = -EACCES; | |
1655 | goto out; | |
1656 | } | |
1657 | ||
1658 | // get uid, gid, from '/tasks' file and make up a mode | |
1659 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1660 | sb->st_mode = S_IFDIR | 00755; | |
1661 | k = cgfs_get_key(controller, cgroup, NULL); | |
1662 | if (!k) { | |
1663 | sb->st_uid = sb->st_gid = 0; | |
1664 | } else { | |
1665 | sb->st_uid = k->uid; | |
1666 | sb->st_gid = k->gid; | |
1667 | } | |
1668 | free_key(k); | |
1669 | sb->st_nlink = 2; | |
1670 | ret = 0; | |
1671 | goto out; | |
1672 | } | |
1673 | ||
1674 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1675 | sb->st_mode = S_IFREG | k->mode; | |
1676 | sb->st_nlink = 1; | |
1677 | sb->st_uid = k->uid; | |
1678 | sb->st_gid = k->gid; | |
1679 | sb->st_size = 0; | |
1680 | free_key(k); | |
1681 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1682 | ret = -ENOENT; | |
1683 | goto out; | |
1684 | } | |
1685 | ret = 0; | |
1686 | } | |
1687 | ||
1688 | out: | |
1689 | free(cgdir); | |
1690 | return ret; | |
1691 | } | |
1692 | ||
1693 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1694 | { | |
1695 | struct fuse_context *fc = fuse_get_context(); | |
1696 | const char *cgroup; | |
1697 | struct file_info *dir_info; | |
1698 | char *controller = NULL; | |
1699 | ||
1700 | if (!fc) | |
1701 | return -EIO; | |
1702 | ||
1703 | if (strcmp(path, "/cgroup") == 0) { | |
1704 | cgroup = NULL; | |
1705 | controller = NULL; | |
1706 | } else { | |
1707 | // return list of keys for the controller, and list of child cgroups | |
1708 | controller = pick_controller_from_path(fc, path); | |
1709 | if (!controller) | |
1710 | return -errno; | |
1711 | ||
1712 | cgroup = find_cgroup_in_path(path); | |
1713 | if (!cgroup) { | |
1714 | /* this is just /cgroup/controller, return its contents */ | |
1715 | cgroup = "/"; | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1720 | if (initpid <= 0) | |
1721 | initpid = fc->pid; | |
1722 | if (cgroup) { | |
1723 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1724 | return -ENOENT; | |
1725 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1726 | return -EACCES; | |
1727 | } | |
1728 | ||
1729 | /* we'll free this at cg_releasedir */ | |
1730 | dir_info = malloc(sizeof(*dir_info)); | |
1731 | if (!dir_info) | |
1732 | return -ENOMEM; | |
1733 | dir_info->controller = must_copy_string(controller); | |
1734 | dir_info->cgroup = must_copy_string(cgroup); | |
1735 | dir_info->type = LXC_TYPE_CGDIR; | |
1736 | dir_info->buf = NULL; | |
1737 | dir_info->file = NULL; | |
1738 | dir_info->buflen = 0; | |
1739 | ||
1740 | fi->fh = (unsigned long)dir_info; | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1745 | struct fuse_file_info *fi) | |
1746 | { | |
1747 | struct file_info *d = (struct file_info *)fi->fh; | |
1748 | struct cgfs_files **list = NULL; | |
1749 | int i, ret; | |
1750 | char *nextcg = NULL; | |
1751 | struct fuse_context *fc = fuse_get_context(); | |
1752 | char **clist = NULL; | |
1753 | ||
1754 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) | |
1755 | return -EIO; | |
1756 | ||
1757 | if (d->type != LXC_TYPE_CGDIR) { | |
1758 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); | |
1759 | return -EIO; | |
1760 | } | |
1761 | if (!d->cgroup && !d->controller) { | |
1762 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1763 | int i; | |
1764 | ||
1765 | for (i = 0; i < num_hierarchies; i++) { | |
1766 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1767 | return -EIO; | |
1768 | } | |
1769 | } | |
1770 | return 0; | |
1771 | } | |
1772 | ||
1773 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1774 | // not a valid cgroup | |
1775 | ret = -EINVAL; | |
1776 | goto out; | |
1777 | } | |
1778 | ||
1779 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1780 | if (initpid <= 0) | |
1781 | initpid = fc->pid; | |
1782 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1783 | if (nextcg) { | |
1784 | ret = filler(buf, nextcg, NULL, 0); | |
1785 | free(nextcg); | |
1786 | if (ret != 0) { | |
1787 | ret = -EIO; | |
1788 | goto out; | |
1789 | } | |
1790 | } | |
1791 | ret = 0; | |
1792 | goto out; | |
1793 | } | |
1794 | ||
1795 | for (i = 0; list[i]; i++) { | |
1796 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
1797 | ret = -EIO; | |
1798 | goto out; | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | // now get the list of child cgroups | |
1803 | ||
1804 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
1805 | ret = 0; | |
1806 | goto out; | |
1807 | } | |
1808 | if (clist) { | |
1809 | for (i = 0; clist[i]; i++) { | |
1810 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
1811 | ret = -EIO; | |
1812 | goto out; | |
1813 | } | |
1814 | } | |
1815 | } | |
1816 | ret = 0; | |
1817 | ||
1818 | out: | |
1819 | free_keys(list); | |
1820 | if (clist) { | |
1821 | for (i = 0; clist[i]; i++) | |
1822 | free(clist[i]); | |
1823 | free(clist); | |
1824 | } | |
1825 | return ret; | |
1826 | } | |
1827 | ||
1828 | static void do_release_file_info(struct fuse_file_info *fi) | |
1829 | { | |
1830 | struct file_info *f = (struct file_info *)fi->fh; | |
1831 | ||
1832 | if (!f) | |
1833 | return; | |
1834 | ||
1835 | fi->fh = 0; | |
1836 | ||
1837 | free(f->controller); | |
1838 | f->controller = NULL; | |
1839 | free(f->cgroup); | |
1840 | f->cgroup = NULL; | |
1841 | free(f->file); | |
1842 | f->file = NULL; | |
1843 | free(f->buf); | |
1844 | f->buf = NULL; | |
1845 | free(f); | |
1846 | } | |
1847 | ||
1848 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
1849 | { | |
1850 | do_release_file_info(fi); | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1854 | int cg_open(const char *path, struct fuse_file_info *fi) | |
1855 | { | |
1856 | const char *cgroup; | |
1857 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
1858 | struct cgfs_files *k = NULL; | |
1859 | struct file_info *file_info; | |
1860 | struct fuse_context *fc = fuse_get_context(); | |
1861 | int ret; | |
1862 | ||
1863 | if (!fc) | |
1864 | return -EIO; | |
1865 | ||
1866 | controller = pick_controller_from_path(fc, path); | |
1867 | if (!controller) | |
1868 | return -errno; | |
1869 | cgroup = find_cgroup_in_path(path); | |
1870 | if (!cgroup) | |
1871 | return -errno; | |
1872 | ||
1873 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1874 | if (!last) { | |
1875 | path1 = "/"; | |
1876 | path2 = cgdir; | |
1877 | } else { | |
1878 | path1 = cgdir; | |
1879 | path2 = last; | |
1880 | } | |
1881 | ||
1882 | k = cgfs_get_key(controller, path1, path2); | |
1883 | if (!k) { | |
1884 | ret = -EINVAL; | |
1885 | goto out; | |
1886 | } | |
1887 | free_key(k); | |
1888 | ||
1889 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1890 | if (initpid <= 0) | |
1891 | initpid = fc->pid; | |
1892 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1893 | ret = -ENOENT; | |
1894 | goto out; | |
1895 | } | |
1896 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
1897 | ret = -EACCES; | |
1898 | goto out; | |
1899 | } | |
1900 | ||
1901 | /* we'll free this at cg_release */ | |
1902 | file_info = malloc(sizeof(*file_info)); | |
1903 | if (!file_info) { | |
1904 | ret = -ENOMEM; | |
1905 | goto out; | |
1906 | } | |
1907 | file_info->controller = must_copy_string(controller); | |
1908 | file_info->cgroup = must_copy_string(path1); | |
1909 | file_info->file = must_copy_string(path2); | |
1910 | file_info->type = LXC_TYPE_CGFILE; | |
1911 | file_info->buf = NULL; | |
1912 | file_info->buflen = 0; | |
1913 | ||
1914 | fi->fh = (unsigned long)file_info; | |
1915 | ret = 0; | |
1916 | ||
1917 | out: | |
1918 | free(cgdir); | |
1919 | return ret; | |
1920 | } | |
1921 | ||
1922 | int cg_access(const char *path, int mode) | |
1923 | { | |
1924 | int ret; | |
1925 | const char *cgroup; | |
1926 | char *path1, *path2, *controller; | |
1927 | char *last = NULL, *cgdir = NULL; | |
1928 | struct cgfs_files *k = NULL; | |
1929 | struct fuse_context *fc = fuse_get_context(); | |
1930 | ||
1931 | if (strcmp(path, "/cgroup") == 0) | |
1932 | return 0; | |
1933 | ||
1934 | if (!fc) | |
1935 | return -EIO; | |
1936 | ||
1937 | controller = pick_controller_from_path(fc, path); | |
1938 | if (!controller) | |
1939 | return -errno; | |
1940 | cgroup = find_cgroup_in_path(path); | |
1941 | if (!cgroup) { | |
1942 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
1943 | if ((mode & W_OK) == 0) | |
1944 | return 0; | |
1945 | return -EACCES; | |
1946 | } | |
1947 | ||
1948 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1949 | if (!last) { | |
1950 | path1 = "/"; | |
1951 | path2 = cgdir; | |
1952 | } else { | |
1953 | path1 = cgdir; | |
1954 | path2 = last; | |
1955 | } | |
1956 | ||
1957 | k = cgfs_get_key(controller, path1, path2); | |
1958 | if (!k) { | |
1959 | if ((mode & W_OK) == 0) | |
1960 | ret = 0; | |
1961 | else | |
1962 | ret = -EACCES; | |
1963 | goto out; | |
1964 | } | |
1965 | free_key(k); | |
1966 | ||
1967 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1968 | if (initpid <= 0) | |
1969 | initpid = fc->pid; | |
1970 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
1971 | ret = -ENOENT; | |
1972 | goto out; | |
1973 | } | |
1974 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
1975 | ret = -EACCES; | |
1976 | goto out; | |
1977 | } | |
1978 | ||
1979 | ret = 0; | |
1980 | ||
1981 | out: | |
1982 | free(cgdir); | |
1983 | return ret; | |
1984 | } | |
1985 | ||
1986 | int cg_release(const char *path, struct fuse_file_info *fi) | |
1987 | { | |
1988 | do_release_file_info(fi); | |
1989 | return 0; | |
1990 | } | |
1991 | ||
1992 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
1993 | ||
1994 | static bool wait_for_sock(int sock, int timeout) | |
1995 | { | |
1996 | struct epoll_event ev; | |
1997 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
1998 | ||
1999 | if ((starttime = time(NULL)) < 0) | |
2000 | return false; | |
2001 | ||
2002 | if ((epfd = epoll_create(1)) < 0) { | |
2003 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); | |
2004 | return false; | |
2005 | } | |
2006 | ||
2007 | ev.events = POLLIN_SET; | |
2008 | ev.data.fd = sock; | |
2009 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
2010 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); | |
2011 | close(epfd); | |
2012 | return false; | |
2013 | } | |
2014 | ||
2015 | again: | |
2016 | if ((now = time(NULL)) < 0) { | |
2017 | close(epfd); | |
2018 | return false; | |
2019 | } | |
2020 | ||
2021 | deltatime = (starttime + timeout) - now; | |
2022 | if (deltatime < 0) { // timeout | |
2023 | errno = 0; | |
2024 | close(epfd); | |
2025 | return false; | |
2026 | } | |
2027 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2028 | if (ret < 0 && errno == EINTR) | |
2029 | goto again; | |
2030 | saved_errno = errno; | |
2031 | close(epfd); | |
2032 | ||
2033 | if (ret <= 0) { | |
2034 | errno = saved_errno; | |
2035 | return false; | |
2036 | } | |
2037 | return true; | |
2038 | } | |
2039 | ||
2040 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2041 | { | |
2042 | if (!wait_for_sock(sockfd, 2)) | |
2043 | return -1; | |
2044 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2045 | } | |
2046 | ||
2047 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2048 | { | |
2049 | struct msghdr msg = { 0 }; | |
2050 | struct iovec iov; | |
2051 | struct cmsghdr *cmsg; | |
2052 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2053 | char buf[1]; | |
2054 | buf[0] = 'p'; | |
2055 | ||
2056 | if (pingfirst) { | |
2057 | if (msgrecv(sock, buf, 1) != 1) { | |
2058 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); | |
2059 | return SEND_CREDS_FAIL; | |
2060 | } | |
2061 | } | |
2062 | ||
2063 | msg.msg_control = cmsgbuf; | |
2064 | msg.msg_controllen = sizeof(cmsgbuf); | |
2065 | ||
2066 | cmsg = CMSG_FIRSTHDR(&msg); | |
2067 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2068 | cmsg->cmsg_level = SOL_SOCKET; | |
2069 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2070 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2071 | ||
2072 | msg.msg_name = NULL; | |
2073 | msg.msg_namelen = 0; | |
2074 | ||
2075 | buf[0] = v; | |
2076 | iov.iov_base = buf; | |
2077 | iov.iov_len = sizeof(buf); | |
2078 | msg.msg_iov = &iov; | |
2079 | msg.msg_iovlen = 1; | |
2080 | ||
2081 | if (sendmsg(sock, &msg, 0) < 0) { | |
2082 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); | |
2083 | if (errno == 3) | |
2084 | return SEND_CREDS_NOTSK; | |
2085 | return SEND_CREDS_FAIL; | |
2086 | } | |
2087 | ||
2088 | return SEND_CREDS_OK; | |
2089 | } | |
2090 | ||
2091 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2092 | { | |
2093 | struct msghdr msg = { 0 }; | |
2094 | struct iovec iov; | |
2095 | struct cmsghdr *cmsg; | |
2096 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2097 | char buf[1]; | |
2098 | int ret; | |
2099 | int optval = 1; | |
2100 | ||
2101 | *v = '1'; | |
2102 | ||
2103 | cred->pid = -1; | |
2104 | cred->uid = -1; | |
2105 | cred->gid = -1; | |
2106 | ||
2107 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
2108 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); | |
2109 | return false; | |
2110 | } | |
2111 | buf[0] = '1'; | |
2112 | if (write(sock, buf, 1) != 1) { | |
2113 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); | |
2114 | return false; | |
2115 | } | |
2116 | ||
2117 | msg.msg_name = NULL; | |
2118 | msg.msg_namelen = 0; | |
2119 | msg.msg_control = cmsgbuf; | |
2120 | msg.msg_controllen = sizeof(cmsgbuf); | |
2121 | ||
2122 | iov.iov_base = buf; | |
2123 | iov.iov_len = sizeof(buf); | |
2124 | msg.msg_iov = &iov; | |
2125 | msg.msg_iovlen = 1; | |
2126 | ||
2127 | if (!wait_for_sock(sock, 2)) { | |
2128 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); | |
2129 | return false; | |
2130 | } | |
2131 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2132 | if (ret < 0) { | |
2133 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); | |
2134 | return false; | |
2135 | } | |
2136 | ||
2137 | cmsg = CMSG_FIRSTHDR(&msg); | |
2138 | ||
2139 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2140 | cmsg->cmsg_level == SOL_SOCKET && | |
2141 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2142 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2143 | } | |
2144 | *v = buf[0]; | |
2145 | ||
2146 | return true; | |
2147 | } | |
2148 | ||
2149 | struct pid_ns_clone_args { | |
2150 | int *cpipe; | |
2151 | int sock; | |
2152 | pid_t tpid; | |
2153 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2154 | }; | |
2155 | ||
2156 | /* | |
2157 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2158 | * with clone(). This simply writes '1' as ACK back to the parent | |
2159 | * before calling the actual wrapped function. | |
2160 | */ | |
2161 | static int pid_ns_clone_wrapper(void *arg) { | |
2162 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2163 | char b = '1'; | |
2164 | ||
2165 | close(args->cpipe[0]); | |
2166 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) | |
2167 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
2168 | close(args->cpipe[1]); | |
2169 | return args->wrapped(args->sock, args->tpid); | |
2170 | } | |
2171 | ||
2172 | /* | |
2173 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2174 | * int value back over the socket. This shifts the pid from the | |
2175 | * sender's pidns into tpid's pidns. | |
2176 | */ | |
2177 | static int pid_to_ns(int sock, pid_t tpid) | |
2178 | { | |
2179 | char v = '0'; | |
2180 | struct ucred cred; | |
2181 | ||
2182 | while (recv_creds(sock, &cred, &v)) { | |
2183 | if (v == '1') | |
2184 | return 0; | |
2185 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) | |
2186 | return 1; | |
2187 | } | |
2188 | return 0; | |
2189 | } | |
2190 | ||
2191 | ||
2192 | /* | |
2193 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
2194 | * in your old pidns. Only children which you clone will be in the target | |
2195 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2196 | * actually convert pids. | |
2197 | * | |
2198 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2199 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2200 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2201 | * this issue. | |
2202 | */ | |
2203 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2204 | { | |
2205 | int newnsfd = -1, ret, cpipe[2]; | |
2206 | char fnam[100]; | |
2207 | pid_t cpid; | |
2208 | char v; | |
2209 | ||
2210 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2211 | if (ret < 0 || ret >= sizeof(fnam)) | |
2212 | _exit(1); | |
2213 | newnsfd = open(fnam, O_RDONLY); | |
2214 | if (newnsfd < 0) | |
2215 | _exit(1); | |
2216 | if (setns(newnsfd, 0) < 0) | |
2217 | _exit(1); | |
2218 | close(newnsfd); | |
2219 | ||
2220 | if (pipe(cpipe) < 0) | |
2221 | _exit(1); | |
2222 | ||
2223 | struct pid_ns_clone_args args = { | |
2224 | .cpipe = cpipe, | |
2225 | .sock = sock, | |
2226 | .tpid = tpid, | |
2227 | .wrapped = &pid_to_ns | |
2228 | }; | |
2229 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2230 | void *stack = alloca(stack_size); | |
2231 | ||
2232 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
2233 | if (cpid < 0) | |
2234 | _exit(1); | |
2235 | ||
2236 | // give the child 1 second to be done forking and | |
2237 | // write its ack | |
2238 | if (!wait_for_sock(cpipe[0], 1)) | |
2239 | _exit(1); | |
2240 | ret = read(cpipe[0], &v, 1); | |
2241 | if (ret != sizeof(char) || v != '1') | |
2242 | _exit(1); | |
2243 | ||
2244 | if (!wait_for_pid(cpid)) | |
2245 | _exit(1); | |
2246 | _exit(0); | |
2247 | } | |
2248 | ||
2249 | /* | |
2250 | * To read cgroup files with a particular pid, we will setns into the child | |
2251 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2252 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2253 | */ | |
2254 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2255 | { | |
2256 | int sock[2] = {-1, -1}; | |
2257 | char *tmpdata = NULL; | |
2258 | int ret; | |
2259 | pid_t qpid, cpid = -1; | |
2260 | bool answer = false; | |
2261 | char v = '0'; | |
2262 | struct ucred cred; | |
2263 | size_t sz = 0, asz = 0; | |
2264 | ||
2265 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2266 | return false; | |
2267 | ||
2268 | /* | |
2269 | * Now we read the pids from returned data one by one, pass | |
2270 | * them into a child in the target namespace, read back the | |
2271 | * translated pids, and put them into our to-return data | |
2272 | */ | |
2273 | ||
2274 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2275 | perror("socketpair"); | |
2276 | free(tmpdata); | |
2277 | return false; | |
2278 | } | |
2279 | ||
2280 | cpid = fork(); | |
2281 | if (cpid == -1) | |
2282 | goto out; | |
2283 | ||
2284 | if (!cpid) // child - exits when done | |
2285 | pid_to_ns_wrapper(sock[1], tpid); | |
2286 | ||
2287 | char *ptr = tmpdata; | |
2288 | cred.uid = 0; | |
2289 | cred.gid = 0; | |
2290 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2291 | cred.pid = qpid; | |
2292 | ret = send_creds(sock[0], &cred, v, true); | |
2293 | ||
2294 | if (ret == SEND_CREDS_NOTSK) | |
2295 | goto next; | |
2296 | if (ret == SEND_CREDS_FAIL) | |
2297 | goto out; | |
2298 | ||
2299 | // read converted results | |
2300 | if (!wait_for_sock(sock[0], 2)) { | |
2301 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); | |
2302 | goto out; | |
2303 | } | |
2304 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2305 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); | |
2306 | goto out; | |
2307 | } | |
2308 | must_strcat_pid(d, &sz, &asz, qpid); | |
2309 | next: | |
2310 | ptr = strchr(ptr, '\n'); | |
2311 | if (!ptr) | |
2312 | break; | |
2313 | ptr++; | |
2314 | } | |
2315 | ||
2316 | cred.pid = getpid(); | |
2317 | v = '1'; | |
2318 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2319 | // failed to ask child to exit | |
2320 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); | |
2321 | goto out; | |
2322 | } | |
2323 | ||
2324 | answer = true; | |
2325 | ||
2326 | out: | |
2327 | free(tmpdata); | |
2328 | if (cpid != -1) | |
2329 | wait_for_pid(cpid); | |
2330 | if (sock[0] != -1) { | |
2331 | close(sock[0]); | |
2332 | close(sock[1]); | |
2333 | } | |
2334 | return answer; | |
2335 | } | |
2336 | ||
2337 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2338 | struct fuse_file_info *fi) | |
2339 | { | |
2340 | struct fuse_context *fc = fuse_get_context(); | |
2341 | struct file_info *f = (struct file_info *)fi->fh; | |
2342 | struct cgfs_files *k = NULL; | |
2343 | char *data = NULL; | |
2344 | int ret, s; | |
2345 | bool r; | |
2346 | ||
2347 | if (f->type != LXC_TYPE_CGFILE) { | |
2348 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); | |
2349 | return -EIO; | |
2350 | } | |
2351 | ||
2352 | if (offset) | |
2353 | return 0; | |
2354 | ||
2355 | if (!fc) | |
2356 | return -EIO; | |
2357 | ||
2358 | if (!f->controller) | |
2359 | return -EINVAL; | |
2360 | ||
2361 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2362 | return -EINVAL; | |
2363 | } | |
2364 | free_key(k); | |
2365 | ||
2366 | ||
2367 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { | |
2368 | ret = -EACCES; | |
2369 | goto out; | |
2370 | } | |
2371 | ||
2372 | if (strcmp(f->file, "tasks") == 0 || | |
2373 | strcmp(f->file, "/tasks") == 0 || | |
2374 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2375 | strcmp(f->file, "cgroup.procs") == 0) | |
2376 | // special case - we have to translate the pids | |
2377 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2378 | else | |
2379 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2380 | ||
2381 | if (!r) { | |
2382 | ret = -EINVAL; | |
2383 | goto out; | |
2384 | } | |
2385 | ||
2386 | if (!data) { | |
2387 | ret = 0; | |
2388 | goto out; | |
2389 | } | |
2390 | s = strlen(data); | |
2391 | if (s > size) | |
2392 | s = size; | |
2393 | memcpy(buf, data, s); | |
2394 | if (s > 0 && s < size && data[s-1] != '\n') | |
2395 | buf[s++] = '\n'; | |
2396 | ||
2397 | ret = s; | |
2398 | ||
2399 | out: | |
2400 | free(data); | |
2401 | return ret; | |
2402 | } | |
2403 | ||
2404 | static int pid_from_ns(int sock, pid_t tpid) | |
2405 | { | |
2406 | pid_t vpid; | |
2407 | struct ucred cred; | |
2408 | char v; | |
2409 | int ret; | |
2410 | ||
2411 | cred.uid = 0; | |
2412 | cred.gid = 0; | |
2413 | while (1) { | |
2414 | if (!wait_for_sock(sock, 2)) { | |
2415 | lxcfs_error("%s\n", "Timeout reading from parent."); | |
2416 | return 1; | |
2417 | } | |
2418 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
2419 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); | |
2420 | return 1; | |
2421 | } | |
2422 | if (vpid == -1) // done | |
2423 | break; | |
2424 | v = '0'; | |
2425 | cred.pid = vpid; | |
2426 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2427 | v = '1'; | |
2428 | cred.pid = getpid(); | |
2429 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
2430 | return 1; | |
2431 | } | |
2432 | } | |
2433 | return 0; | |
2434 | } | |
2435 | ||
2436 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2437 | { | |
2438 | int newnsfd = -1, ret, cpipe[2]; | |
2439 | char fnam[100]; | |
2440 | pid_t cpid; | |
2441 | char v; | |
2442 | ||
2443 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2444 | if (ret < 0 || ret >= sizeof(fnam)) | |
2445 | _exit(1); | |
2446 | newnsfd = open(fnam, O_RDONLY); | |
2447 | if (newnsfd < 0) | |
2448 | _exit(1); | |
2449 | if (setns(newnsfd, 0) < 0) | |
2450 | _exit(1); | |
2451 | close(newnsfd); | |
2452 | ||
2453 | if (pipe(cpipe) < 0) | |
2454 | _exit(1); | |
2455 | ||
2456 | struct pid_ns_clone_args args = { | |
2457 | .cpipe = cpipe, | |
2458 | .sock = sock, | |
2459 | .tpid = tpid, | |
2460 | .wrapped = &pid_from_ns | |
2461 | }; | |
2462 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2463 | void *stack = alloca(stack_size); | |
2464 | ||
2465 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
2466 | if (cpid < 0) | |
2467 | _exit(1); | |
2468 | ||
2469 | // give the child 1 second to be done forking and | |
2470 | // write its ack | |
2471 | if (!wait_for_sock(cpipe[0], 1)) | |
2472 | _exit(1); | |
2473 | ret = read(cpipe[0], &v, 1); | |
2474 | if (ret != sizeof(char) || v != '1') | |
2475 | _exit(1); | |
2476 | ||
2477 | if (!wait_for_pid(cpid)) | |
2478 | _exit(1); | |
2479 | _exit(0); | |
2480 | } | |
2481 | ||
2482 | /* | |
2483 | * Given host @uid, return the uid to which it maps in | |
2484 | * @pid's user namespace, or -1 if none. | |
2485 | */ | |
2486 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2487 | { | |
2488 | FILE *f; | |
2489 | char line[400]; | |
2490 | ||
2491 | sprintf(line, "/proc/%d/uid_map", pid); | |
2492 | if ((f = fopen(line, "r")) == NULL) { | |
2493 | return false; | |
2494 | } | |
2495 | ||
2496 | *answer = convert_id_to_ns(f, uid); | |
2497 | fclose(f); | |
2498 | ||
2499 | if (*answer == -1) | |
2500 | return false; | |
2501 | return true; | |
2502 | } | |
2503 | ||
2504 | /* | |
2505 | * get_pid_creds: get the real uid and gid of @pid from | |
2506 | * /proc/$$/status | |
2507 | * (XXX should we use euid here?) | |
2508 | */ | |
2509 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2510 | { | |
2511 | char line[400]; | |
2512 | uid_t u; | |
2513 | gid_t g; | |
2514 | FILE *f; | |
2515 | ||
2516 | *uid = -1; | |
2517 | *gid = -1; | |
2518 | sprintf(line, "/proc/%d/status", pid); | |
2519 | if ((f = fopen(line, "r")) == NULL) { | |
2520 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); | |
2521 | return; | |
2522 | } | |
2523 | while (fgets(line, 400, f)) { | |
2524 | if (strncmp(line, "Uid:", 4) == 0) { | |
2525 | if (sscanf(line+4, "%u", &u) != 1) { | |
2526 | lxcfs_error("bad uid line for pid %u\n", pid); | |
2527 | fclose(f); | |
2528 | return; | |
2529 | } | |
2530 | *uid = u; | |
2531 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2532 | if (sscanf(line+4, "%u", &g) != 1) { | |
2533 | lxcfs_error("bad gid line for pid %u\n", pid); | |
2534 | fclose(f); | |
2535 | return; | |
2536 | } | |
2537 | *gid = g; | |
2538 | } | |
2539 | } | |
2540 | fclose(f); | |
2541 | } | |
2542 | ||
2543 | /* | |
2544 | * May the requestor @r move victim @v to a new cgroup? | |
2545 | * This is allowed if | |
2546 | * . they are the same task | |
2547 | * . they are ownedy by the same uid | |
2548 | * . @r is root on the host, or | |
2549 | * . @v's uid is mapped into @r's where @r is root. | |
2550 | */ | |
2551 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2552 | { | |
2553 | uid_t v_uid, tmpuid; | |
2554 | gid_t v_gid; | |
2555 | ||
2556 | if (r == v) | |
2557 | return true; | |
2558 | if (r_uid == 0) | |
2559 | return true; | |
2560 | get_pid_creds(v, &v_uid, &v_gid); | |
2561 | if (r_uid == v_uid) | |
2562 | return true; | |
2563 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2564 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2565 | return true; | |
2566 | return false; | |
2567 | } | |
2568 | ||
2569 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2570 | const char *file, const char *buf) | |
2571 | { | |
2572 | int sock[2] = {-1, -1}; | |
2573 | pid_t qpid, cpid = -1; | |
2574 | FILE *pids_file = NULL; | |
2575 | bool answer = false, fail = false; | |
2576 | ||
2577 | pids_file = open_pids_file(contrl, cg); | |
2578 | if (!pids_file) | |
2579 | return false; | |
2580 | ||
2581 | /* | |
2582 | * write the pids to a socket, have helper in writer's pidns | |
2583 | * call movepid for us | |
2584 | */ | |
2585 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2586 | perror("socketpair"); | |
2587 | goto out; | |
2588 | } | |
2589 | ||
2590 | cpid = fork(); | |
2591 | if (cpid == -1) | |
2592 | goto out; | |
2593 | ||
2594 | if (!cpid) { // child | |
2595 | fclose(pids_file); | |
2596 | pid_from_ns_wrapper(sock[1], tpid); | |
2597 | } | |
2598 | ||
2599 | const char *ptr = buf; | |
2600 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2601 | struct ucred cred; | |
2602 | char v; | |
2603 | ||
2604 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
2605 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); | |
2606 | goto out; | |
2607 | } | |
2608 | ||
2609 | if (recv_creds(sock[0], &cred, &v)) { | |
2610 | if (v == '0') { | |
2611 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2612 | fail = true; | |
2613 | break; | |
2614 | } | |
2615 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2616 | fail = true; | |
2617 | } | |
2618 | } | |
2619 | ||
2620 | ptr = strchr(ptr, '\n'); | |
2621 | if (!ptr) | |
2622 | break; | |
2623 | ptr++; | |
2624 | } | |
2625 | ||
2626 | /* All good, write the value */ | |
2627 | qpid = -1; | |
2628 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
2629 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); | |
2630 | ||
2631 | if (!fail) | |
2632 | answer = true; | |
2633 | ||
2634 | out: | |
2635 | if (cpid != -1) | |
2636 | wait_for_pid(cpid); | |
2637 | if (sock[0] != -1) { | |
2638 | close(sock[0]); | |
2639 | close(sock[1]); | |
2640 | } | |
2641 | if (pids_file) { | |
2642 | if (fclose(pids_file) != 0) | |
2643 | answer = false; | |
2644 | } | |
2645 | return answer; | |
2646 | } | |
2647 | ||
2648 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2649 | struct fuse_file_info *fi) | |
2650 | { | |
2651 | struct fuse_context *fc = fuse_get_context(); | |
2652 | char *localbuf = NULL; | |
2653 | struct cgfs_files *k = NULL; | |
2654 | struct file_info *f = (struct file_info *)fi->fh; | |
2655 | bool r; | |
2656 | ||
2657 | if (f->type != LXC_TYPE_CGFILE) { | |
2658 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); | |
2659 | return -EIO; | |
2660 | } | |
2661 | ||
2662 | if (offset) | |
2663 | return 0; | |
2664 | ||
2665 | if (!fc) | |
2666 | return -EIO; | |
2667 | ||
2668 | localbuf = alloca(size+1); | |
2669 | localbuf[size] = '\0'; | |
2670 | memcpy(localbuf, buf, size); | |
2671 | ||
2672 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2673 | size = -EINVAL; | |
2674 | goto out; | |
2675 | } | |
2676 | ||
2677 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2678 | size = -EACCES; | |
2679 | goto out; | |
2680 | } | |
2681 | ||
2682 | if (strcmp(f->file, "tasks") == 0 || | |
2683 | strcmp(f->file, "/tasks") == 0 || | |
2684 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2685 | strcmp(f->file, "cgroup.procs") == 0) | |
2686 | // special case - we have to translate the pids | |
2687 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2688 | else | |
2689 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2690 | ||
2691 | if (!r) | |
2692 | size = -EINVAL; | |
2693 | ||
2694 | out: | |
2695 | free_key(k); | |
2696 | return size; | |
2697 | } | |
2698 | ||
2699 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2700 | { | |
2701 | struct fuse_context *fc = fuse_get_context(); | |
2702 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2703 | struct cgfs_files *k = NULL; | |
2704 | const char *cgroup; | |
2705 | int ret; | |
2706 | ||
2707 | if (!fc) | |
2708 | return -EIO; | |
2709 | ||
2710 | if (strcmp(path, "/cgroup") == 0) | |
2711 | return -EPERM; | |
2712 | ||
2713 | controller = pick_controller_from_path(fc, path); | |
2714 | if (!controller) | |
2715 | return errno == ENOENT ? -EPERM : -errno; | |
2716 | ||
2717 | cgroup = find_cgroup_in_path(path); | |
2718 | if (!cgroup) | |
2719 | /* this is just /cgroup/controller */ | |
2720 | return -EPERM; | |
2721 | ||
2722 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2723 | ||
2724 | if (!last) { | |
2725 | path1 = "/"; | |
2726 | path2 = cgdir; | |
2727 | } else { | |
2728 | path1 = cgdir; | |
2729 | path2 = last; | |
2730 | } | |
2731 | ||
2732 | if (is_child_cgroup(controller, path1, path2)) { | |
2733 | // get uid, gid, from '/tasks' file and make up a mode | |
2734 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2735 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2736 | ||
2737 | } else | |
2738 | k = cgfs_get_key(controller, path1, path2); | |
2739 | ||
2740 | if (!k) { | |
2741 | ret = -EINVAL; | |
2742 | goto out; | |
2743 | } | |
2744 | ||
2745 | /* | |
2746 | * This being a fuse request, the uid and gid must be valid | |
2747 | * in the caller's namespace. So we can just check to make | |
2748 | * sure that the caller is root in his uid, and privileged | |
2749 | * over the file's current owner. | |
2750 | */ | |
2751 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2752 | ret = -EACCES; | |
2753 | goto out; | |
2754 | } | |
2755 | ||
2756 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2757 | ||
2758 | out: | |
2759 | free_key(k); | |
2760 | free(cgdir); | |
2761 | ||
2762 | return ret; | |
2763 | } | |
2764 | ||
2765 | int cg_chmod(const char *path, mode_t mode) | |
2766 | { | |
2767 | struct fuse_context *fc = fuse_get_context(); | |
2768 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2769 | struct cgfs_files *k = NULL; | |
2770 | const char *cgroup; | |
2771 | int ret; | |
2772 | ||
2773 | if (!fc) | |
2774 | return -EIO; | |
2775 | ||
2776 | if (strcmp(path, "/cgroup") == 0) | |
2777 | return -EPERM; | |
2778 | ||
2779 | controller = pick_controller_from_path(fc, path); | |
2780 | if (!controller) | |
2781 | return errno == ENOENT ? -EPERM : -errno; | |
2782 | ||
2783 | cgroup = find_cgroup_in_path(path); | |
2784 | if (!cgroup) | |
2785 | /* this is just /cgroup/controller */ | |
2786 | return -EPERM; | |
2787 | ||
2788 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2789 | ||
2790 | if (!last) { | |
2791 | path1 = "/"; | |
2792 | path2 = cgdir; | |
2793 | } else { | |
2794 | path1 = cgdir; | |
2795 | path2 = last; | |
2796 | } | |
2797 | ||
2798 | if (is_child_cgroup(controller, path1, path2)) { | |
2799 | // get uid, gid, from '/tasks' file and make up a mode | |
2800 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2801 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2802 | ||
2803 | } else | |
2804 | k = cgfs_get_key(controller, path1, path2); | |
2805 | ||
2806 | if (!k) { | |
2807 | ret = -EINVAL; | |
2808 | goto out; | |
2809 | } | |
2810 | ||
2811 | /* | |
2812 | * This being a fuse request, the uid and gid must be valid | |
2813 | * in the caller's namespace. So we can just check to make | |
2814 | * sure that the caller is root in his uid, and privileged | |
2815 | * over the file's current owner. | |
2816 | */ | |
2817 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
2818 | ret = -EPERM; | |
2819 | goto out; | |
2820 | } | |
2821 | ||
2822 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
2823 | ret = -EINVAL; | |
2824 | goto out; | |
2825 | } | |
2826 | ||
2827 | ret = 0; | |
2828 | out: | |
2829 | free_key(k); | |
2830 | free(cgdir); | |
2831 | return ret; | |
2832 | } | |
2833 | ||
2834 | int cg_mkdir(const char *path, mode_t mode) | |
2835 | { | |
2836 | struct fuse_context *fc = fuse_get_context(); | |
2837 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
2838 | const char *cgroup; | |
2839 | int ret; | |
2840 | ||
2841 | if (!fc) | |
2842 | return -EIO; | |
2843 | ||
2844 | controller = pick_controller_from_path(fc, path); | |
2845 | if (!controller) | |
2846 | return errno == ENOENT ? -EPERM : -errno; | |
2847 | ||
2848 | cgroup = find_cgroup_in_path(path); | |
2849 | if (!cgroup) | |
2850 | return -errno; | |
2851 | ||
2852 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2853 | if (!last) | |
2854 | path1 = "/"; | |
2855 | else | |
2856 | path1 = cgdir; | |
2857 | ||
2858 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2859 | if (initpid <= 0) | |
2860 | initpid = fc->pid; | |
2861 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
2862 | if (!next) | |
2863 | ret = -EINVAL; | |
2864 | else if (last && strcmp(next, last) == 0) | |
2865 | ret = -EEXIST; | |
2866 | else | |
2867 | ret = -EPERM; | |
2868 | goto out; | |
2869 | } | |
2870 | ||
2871 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
2872 | ret = -EACCES; | |
2873 | goto out; | |
2874 | } | |
2875 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2876 | ret = -EACCES; | |
2877 | goto out; | |
2878 | } | |
2879 | ||
2880 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
2881 | ||
2882 | out: | |
2883 | free(cgdir); | |
2884 | free(next); | |
2885 | return ret; | |
2886 | } | |
2887 | ||
2888 | int cg_rmdir(const char *path) | |
2889 | { | |
2890 | struct fuse_context *fc = fuse_get_context(); | |
2891 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
2892 | const char *cgroup; | |
2893 | int ret; | |
2894 | ||
2895 | if (!fc) | |
2896 | return -EIO; | |
2897 | ||
2898 | controller = pick_controller_from_path(fc, path); | |
2899 | if (!controller) /* Someone's trying to delete "/cgroup". */ | |
2900 | return -EPERM; | |
2901 | ||
2902 | cgroup = find_cgroup_in_path(path); | |
2903 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ | |
2904 | return -EPERM; | |
2905 | ||
2906 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2907 | if (!last) { | |
2908 | /* Someone's trying to delete a cgroup on the same level as the | |
2909 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
2910 | * rmdir "/cgroup/blkio/init.slice". | |
2911 | */ | |
2912 | ret = -EPERM; | |
2913 | goto out; | |
2914 | } | |
2915 | ||
2916 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2917 | if (initpid <= 0) | |
2918 | initpid = fc->pid; | |
2919 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
2920 | if (!last || (next && (strcmp(next, last) == 0))) | |
2921 | ret = -EBUSY; | |
2922 | else | |
2923 | ret = -ENOENT; | |
2924 | goto out; | |
2925 | } | |
2926 | ||
2927 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
2928 | ret = -EACCES; | |
2929 | goto out; | |
2930 | } | |
2931 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2932 | ret = -EACCES; | |
2933 | goto out; | |
2934 | } | |
2935 | ||
2936 | if (!cgfs_remove(controller, cgroup)) { | |
2937 | ret = -EINVAL; | |
2938 | goto out; | |
2939 | } | |
2940 | ||
2941 | ret = 0; | |
2942 | ||
2943 | out: | |
2944 | free(cgdir); | |
2945 | free(next); | |
2946 | return ret; | |
2947 | } | |
2948 | ||
2949 | static bool startswith(const char *line, const char *pref) | |
2950 | { | |
2951 | if (strncmp(line, pref, strlen(pref)) == 0) | |
2952 | return true; | |
2953 | return false; | |
2954 | } | |
2955 | ||
2956 | static void parse_memstat(char *memstat, unsigned long *cached, | |
2957 | unsigned long *active_anon, unsigned long *inactive_anon, | |
2958 | unsigned long *active_file, unsigned long *inactive_file, | |
2959 | unsigned long *unevictable) | |
2960 | { | |
2961 | char *eol; | |
2962 | ||
2963 | while (*memstat) { | |
2964 | if (startswith(memstat, "total_cache")) { | |
2965 | sscanf(memstat + 11, "%lu", cached); | |
2966 | *cached /= 1024; | |
2967 | } else if (startswith(memstat, "total_active_anon")) { | |
2968 | sscanf(memstat + 17, "%lu", active_anon); | |
2969 | *active_anon /= 1024; | |
2970 | } else if (startswith(memstat, "total_inactive_anon")) { | |
2971 | sscanf(memstat + 19, "%lu", inactive_anon); | |
2972 | *inactive_anon /= 1024; | |
2973 | } else if (startswith(memstat, "total_active_file")) { | |
2974 | sscanf(memstat + 17, "%lu", active_file); | |
2975 | *active_file /= 1024; | |
2976 | } else if (startswith(memstat, "total_inactive_file")) { | |
2977 | sscanf(memstat + 19, "%lu", inactive_file); | |
2978 | *inactive_file /= 1024; | |
2979 | } else if (startswith(memstat, "total_unevictable")) { | |
2980 | sscanf(memstat + 17, "%lu", unevictable); | |
2981 | *unevictable /= 1024; | |
2982 | } | |
2983 | eol = strchr(memstat, '\n'); | |
2984 | if (!eol) | |
2985 | return; | |
2986 | memstat = eol+1; | |
2987 | } | |
2988 | } | |
2989 | ||
2990 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
2991 | { | |
2992 | char *eol; | |
2993 | char key[32]; | |
2994 | ||
2995 | memset(key, 0, 32); | |
2996 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
2997 | ||
2998 | size_t len = strlen(key); | |
2999 | *v = 0; | |
3000 | ||
3001 | while (*str) { | |
3002 | if (startswith(str, key)) { | |
3003 | sscanf(str + len, "%lu", v); | |
3004 | return; | |
3005 | } | |
3006 | eol = strchr(str, '\n'); | |
3007 | if (!eol) | |
3008 | return; | |
3009 | str = eol+1; | |
3010 | } | |
3011 | } | |
3012 | ||
3013 | static int read_file(const char *path, char *buf, size_t size, | |
3014 | struct file_info *d) | |
3015 | { | |
3016 | size_t linelen = 0, total_len = 0, rv = 0; | |
3017 | char *line = NULL; | |
3018 | char *cache = d->buf; | |
3019 | size_t cache_size = d->buflen; | |
3020 | FILE *f = fopen(path, "r"); | |
3021 | if (!f) | |
3022 | return 0; | |
3023 | ||
3024 | while (getline(&line, &linelen, f) != -1) { | |
3025 | ssize_t l = snprintf(cache, cache_size, "%s", line); | |
3026 | if (l < 0) { | |
3027 | perror("Error writing to cache"); | |
3028 | rv = 0; | |
3029 | goto err; | |
3030 | } | |
3031 | if (l >= cache_size) { | |
3032 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3033 | rv = 0; | |
3034 | goto err; | |
3035 | } | |
3036 | cache += l; | |
3037 | cache_size -= l; | |
3038 | total_len += l; | |
3039 | } | |
3040 | ||
3041 | d->size = total_len; | |
3042 | if (total_len > size) | |
3043 | total_len = size; | |
3044 | ||
3045 | /* read from off 0 */ | |
3046 | memcpy(buf, d->buf, total_len); | |
3047 | rv = total_len; | |
3048 | err: | |
3049 | fclose(f); | |
3050 | free(line); | |
3051 | return rv; | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * FUSE ops for /proc | |
3056 | */ | |
3057 | ||
3058 | static unsigned long get_memlimit(const char *cgroup, const char *file) | |
3059 | { | |
3060 | char *memlimit_str = NULL; | |
3061 | unsigned long memlimit = -1; | |
3062 | ||
3063 | if (cgfs_get_value("memory", cgroup, file, &memlimit_str)) | |
3064 | memlimit = strtoul(memlimit_str, NULL, 10); | |
3065 | ||
3066 | free(memlimit_str); | |
3067 | ||
3068 | return memlimit; | |
3069 | } | |
3070 | ||
3071 | static unsigned long get_min_memlimit(const char *cgroup, const char *file) | |
3072 | { | |
3073 | char *copy = strdupa(cgroup); | |
3074 | unsigned long memlimit = 0, retlimit; | |
3075 | ||
3076 | retlimit = get_memlimit(copy, file); | |
3077 | ||
3078 | while (strcmp(copy, "/") != 0) { | |
3079 | copy = dirname(copy); | |
3080 | memlimit = get_memlimit(copy, file); | |
3081 | if (memlimit != -1 && memlimit < retlimit) | |
3082 | retlimit = memlimit; | |
3083 | }; | |
3084 | ||
3085 | return retlimit; | |
3086 | } | |
3087 | ||
3088 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3089 | struct fuse_file_info *fi) | |
3090 | { | |
3091 | struct fuse_context *fc = fuse_get_context(); | |
3092 | struct file_info *d = (struct file_info *)fi->fh; | |
3093 | char *cg; | |
3094 | char *memusage_str = NULL, *memstat_str = NULL, | |
3095 | *memswlimit_str = NULL, *memswusage_str = NULL; | |
3096 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, | |
3097 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, | |
3098 | active_file = 0, inactive_file = 0, unevictable = 0, | |
3099 | hostswtotal = 0; | |
3100 | char *line = NULL; | |
3101 | size_t linelen = 0, total_len = 0, rv = 0; | |
3102 | char *cache = d->buf; | |
3103 | size_t cache_size = d->buflen; | |
3104 | FILE *f = NULL; | |
3105 | ||
3106 | if (offset){ | |
3107 | if (offset > d->size) | |
3108 | return -EINVAL; | |
3109 | if (!d->cached) | |
3110 | return 0; | |
3111 | int left = d->size - offset; | |
3112 | total_len = left > size ? size: left; | |
3113 | memcpy(buf, cache + offset, total_len); | |
3114 | return total_len; | |
3115 | } | |
3116 | ||
3117 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3118 | if (initpid <= 0) | |
3119 | initpid = fc->pid; | |
3120 | cg = get_pid_cgroup(initpid, "memory"); | |
3121 | if (!cg) | |
3122 | return read_file("/proc/meminfo", buf, size, d); | |
3123 | prune_init_slice(cg); | |
3124 | ||
3125 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); | |
3126 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
3127 | goto err; | |
3128 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3129 | goto err; | |
3130 | ||
3131 | // Following values are allowed to fail, because swapaccount might be turned | |
3132 | // off for current kernel | |
3133 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3134 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3135 | { | |
3136 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); | |
3137 | memswusage = strtoul(memswusage_str, NULL, 10); | |
3138 | ||
3139 | memswlimit = memswlimit / 1024; | |
3140 | memswusage = memswusage / 1024; | |
3141 | } | |
3142 | ||
3143 | memusage = strtoul(memusage_str, NULL, 10); | |
3144 | memlimit /= 1024; | |
3145 | memusage /= 1024; | |
3146 | ||
3147 | parse_memstat(memstat_str, &cached, &active_anon, | |
3148 | &inactive_anon, &active_file, &inactive_file, | |
3149 | &unevictable); | |
3150 | ||
3151 | f = fopen("/proc/meminfo", "r"); | |
3152 | if (!f) | |
3153 | goto err; | |
3154 | ||
3155 | while (getline(&line, &linelen, f) != -1) { | |
3156 | ssize_t l; | |
3157 | char *printme, lbuf[100]; | |
3158 | ||
3159 | memset(lbuf, 0, 100); | |
3160 | if (startswith(line, "MemTotal:")) { | |
3161 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); | |
3162 | if (hosttotal < memlimit) | |
3163 | memlimit = hosttotal; | |
3164 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3165 | printme = lbuf; | |
3166 | } else if (startswith(line, "MemFree:")) { | |
3167 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3168 | printme = lbuf; | |
3169 | } else if (startswith(line, "MemAvailable:")) { | |
3170 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage); | |
3171 | printme = lbuf; | |
3172 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
3173 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); | |
3174 | if (hostswtotal < memswlimit) | |
3175 | memswlimit = hostswtotal; | |
3176 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit); | |
3177 | printme = lbuf; | |
3178 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
3179 | unsigned long swaptotal = memswlimit, | |
3180 | swapusage = memswusage - memusage, | |
3181 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3182 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
3183 | printme = lbuf; | |
3184 | } else if (startswith(line, "Slab:")) { | |
3185 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3186 | printme = lbuf; | |
3187 | } else if (startswith(line, "Buffers:")) { | |
3188 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3189 | printme = lbuf; | |
3190 | } else if (startswith(line, "Cached:")) { | |
3191 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3192 | printme = lbuf; | |
3193 | } else if (startswith(line, "SwapCached:")) { | |
3194 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3195 | printme = lbuf; | |
3196 | } else if (startswith(line, "Active:")) { | |
3197 | snprintf(lbuf, 100, "Active: %8lu kB\n", | |
3198 | active_anon + active_file); | |
3199 | printme = lbuf; | |
3200 | } else if (startswith(line, "Inactive:")) { | |
3201 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", | |
3202 | inactive_anon + inactive_file); | |
3203 | printme = lbuf; | |
3204 | } else if (startswith(line, "Active(anon)")) { | |
3205 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3206 | printme = lbuf; | |
3207 | } else if (startswith(line, "Inactive(anon)")) { | |
3208 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3209 | printme = lbuf; | |
3210 | } else if (startswith(line, "Active(file)")) { | |
3211 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3212 | printme = lbuf; | |
3213 | } else if (startswith(line, "Inactive(file)")) { | |
3214 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3215 | printme = lbuf; | |
3216 | } else if (startswith(line, "Unevictable")) { | |
3217 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3218 | printme = lbuf; | |
3219 | } else if (startswith(line, "SReclaimable")) { | |
3220 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3221 | printme = lbuf; | |
3222 | } else if (startswith(line, "SUnreclaim")) { | |
3223 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3224 | printme = lbuf; | |
3225 | } else | |
3226 | printme = line; | |
3227 | ||
3228 | l = snprintf(cache, cache_size, "%s", printme); | |
3229 | if (l < 0) { | |
3230 | perror("Error writing to cache"); | |
3231 | rv = 0; | |
3232 | goto err; | |
3233 | ||
3234 | } | |
3235 | if (l >= cache_size) { | |
3236 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3237 | rv = 0; | |
3238 | goto err; | |
3239 | } | |
3240 | ||
3241 | cache += l; | |
3242 | cache_size -= l; | |
3243 | total_len += l; | |
3244 | } | |
3245 | ||
3246 | d->cached = 1; | |
3247 | d->size = total_len; | |
3248 | if (total_len > size ) total_len = size; | |
3249 | memcpy(buf, d->buf, total_len); | |
3250 | ||
3251 | rv = total_len; | |
3252 | err: | |
3253 | if (f) | |
3254 | fclose(f); | |
3255 | free(line); | |
3256 | free(cg); | |
3257 | free(memusage_str); | |
3258 | free(memswlimit_str); | |
3259 | free(memswusage_str); | |
3260 | free(memstat_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; | |
3309 | bool am_printing = false, firstline = true, is_s390x = false; | |
3310 | int curcpu = -1, cpu; | |
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); | |
3332 | prune_init_slice(cg); | |
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) { | |
3343 | ssize_t l; | |
3344 | if (firstline) { | |
3345 | firstline = false; | |
3346 | if (strstr(line, "IBM/S390") != NULL) { | |
3347 | is_s390x = true; | |
3348 | am_printing = true; | |
3349 | continue; | |
3350 | } | |
3351 | } | |
3352 | if (strncmp(line, "# processors:", 12) == 0) | |
3353 | continue; | |
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 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3366 | rv = 0; | |
3367 | goto err; | |
3368 | } | |
3369 | cache += l; | |
3370 | cache_size -= l; | |
3371 | total_len += l; | |
3372 | } | |
3373 | continue; | |
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++; | |
3383 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); | |
3384 | if (l < 0) { | |
3385 | perror("Error writing to cache"); | |
3386 | rv = 0; | |
3387 | goto err; | |
3388 | } | |
3389 | if (l >= cache_size) { | |
3390 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3391 | rv = 0; | |
3392 | goto err; | |
3393 | } | |
3394 | cache += l; | |
3395 | cache_size -= l; | |
3396 | total_len += l; | |
3397 | continue; | |
3398 | ||
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 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3409 | rv = 0; | |
3410 | goto err; | |
3411 | } | |
3412 | cache += l; | |
3413 | cache_size -= l; | |
3414 | total_len += l; | |
3415 | } | |
3416 | } | |
3417 | ||
3418 | if (is_s390x) { | |
3419 | char *origcache = d->buf; | |
3420 | ssize_t l; | |
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 | ||
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 uint64_t get_reaper_start_time(pid_t pid) | |
3467 | { | |
3468 | int ret; | |
3469 | FILE *f; | |
3470 | uint64_t starttime; | |
3471 | /* strlen("/proc/") = 6 | |
3472 | * + | |
3473 | * LXCFS_NUMSTRLEN64 | |
3474 | * + | |
3475 | * strlen("/stat") = 5 | |
3476 | * + | |
3477 | * \0 = 1 | |
3478 | * */ | |
3479 | #define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1) | |
3480 | char path[__PROC_PID_STAT_LEN]; | |
3481 | pid_t qpid; | |
3482 | ||
3483 | qpid = lookup_initpid_in_store(pid); | |
3484 | if (qpid <= 0) { | |
3485 | /* Caller can check for EINVAL on 0. */ | |
3486 | errno = EINVAL; | |
3487 | return 0; | |
3488 | } | |
3489 | ||
3490 | ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid); | |
3491 | if (ret < 0 || ret >= __PROC_PID_STAT_LEN) { | |
3492 | /* Caller can check for EINVAL on 0. */ | |
3493 | errno = EINVAL; | |
3494 | return 0; | |
3495 | } | |
3496 | ||
3497 | f = fopen(path, "r"); | |
3498 | if (!f) { | |
3499 | /* Caller can check for EINVAL on 0. */ | |
3500 | errno = EINVAL; | |
3501 | return 0; | |
3502 | } | |
3503 | ||
3504 | /* Note that the *scanf() argument supression requires that length | |
3505 | * modifiers such as "l" are omitted. Otherwise some compilers will yell | |
3506 | * at us. It's like telling someone you're not married and then asking | |
3507 | * if you can bring your wife to the party. | |
3508 | */ | |
3509 | ret = fscanf(f, "%*d " /* (1) pid %d */ | |
3510 | "%*s " /* (2) comm %s */ | |
3511 | "%*c " /* (3) state %c */ | |
3512 | "%*d " /* (4) ppid %d */ | |
3513 | "%*d " /* (5) pgrp %d */ | |
3514 | "%*d " /* (6) session %d */ | |
3515 | "%*d " /* (7) tty_nr %d */ | |
3516 | "%*d " /* (8) tpgid %d */ | |
3517 | "%*u " /* (9) flags %u */ | |
3518 | "%*u " /* (10) minflt %lu */ | |
3519 | "%*u " /* (11) cminflt %lu */ | |
3520 | "%*u " /* (12) majflt %lu */ | |
3521 | "%*u " /* (13) cmajflt %lu */ | |
3522 | "%*u " /* (14) utime %lu */ | |
3523 | "%*u " /* (15) stime %lu */ | |
3524 | "%*d " /* (16) cutime %ld */ | |
3525 | "%*d " /* (17) cstime %ld */ | |
3526 | "%*d " /* (18) priority %ld */ | |
3527 | "%*d " /* (19) nice %ld */ | |
3528 | "%*d " /* (20) num_threads %ld */ | |
3529 | "%*d " /* (21) itrealvalue %ld */ | |
3530 | "%" PRIu64, /* (22) starttime %llu */ | |
3531 | &starttime); | |
3532 | if (ret != 1) { | |
3533 | fclose(f); | |
3534 | /* Caller can check for EINVAL on 0. */ | |
3535 | errno = EINVAL; | |
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | fclose(f); | |
3540 | ||
3541 | errno = 0; | |
3542 | return starttime; | |
3543 | } | |
3544 | ||
3545 | static uint64_t get_reaper_start_time_in_sec(pid_t pid) | |
3546 | { | |
3547 | uint64_t clockticks; | |
3548 | int64_t ticks_per_sec; | |
3549 | ||
3550 | clockticks = get_reaper_start_time(pid); | |
3551 | if (clockticks == 0 && errno == EINVAL) { | |
3552 | lxcfs_debug("failed to retrieve start time of pid %d\n", pid); | |
3553 | return 0; | |
3554 | } | |
3555 | ||
3556 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3557 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
3558 | lxcfs_debug( | |
3559 | "%s\n", | |
3560 | "failed to determine number of clock ticks in a second"); | |
3561 | return 0; | |
3562 | } | |
3563 | ||
3564 | return (clockticks /= ticks_per_sec); | |
3565 | } | |
3566 | ||
3567 | static uint64_t get_reaper_age(pid_t pid) | |
3568 | { | |
3569 | uint64_t procstart, uptime, procage; | |
3570 | ||
3571 | /* We need to substract the time the process has started since system | |
3572 | * boot minus the time when the system has started to get the actual | |
3573 | * reaper age. | |
3574 | */ | |
3575 | procstart = get_reaper_start_time_in_sec(pid); | |
3576 | procage = procstart; | |
3577 | if (procstart > 0) { | |
3578 | int ret; | |
3579 | struct timespec spec; | |
3580 | ||
3581 | ret = clock_gettime(CLOCK_BOOTTIME, &spec); | |
3582 | if (ret < 0) | |
3583 | return 0; | |
3584 | /* We could make this more precise here by using the tv_nsec | |
3585 | * field in the timespec struct and convert it to milliseconds | |
3586 | * and then create a double for the seconds and milliseconds but | |
3587 | * that seems more work than it is worth. | |
3588 | */ | |
3589 | uptime = spec.tv_sec; | |
3590 | procage = uptime - procstart; | |
3591 | } | |
3592 | ||
3593 | return procage; | |
3594 | } | |
3595 | ||
3596 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) | |
3597 | static int proc_stat_read(char *buf, size_t size, off_t offset, | |
3598 | struct fuse_file_info *fi) | |
3599 | { | |
3600 | struct fuse_context *fc = fuse_get_context(); | |
3601 | struct file_info *d = (struct file_info *)fi->fh; | |
3602 | char *cg; | |
3603 | char *cpuset = NULL; | |
3604 | char *line = NULL; | |
3605 | size_t linelen = 0, total_len = 0, rv = 0; | |
3606 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
3607 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
3608 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, | |
3609 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; | |
3610 | char cpuall[CPUALL_MAX_SIZE]; | |
3611 | /* reserve for cpu all */ | |
3612 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3613 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3614 | FILE *f = NULL; | |
3615 | ||
3616 | if (offset){ | |
3617 | if (offset > d->size) | |
3618 | return -EINVAL; | |
3619 | if (!d->cached) | |
3620 | return 0; | |
3621 | int left = d->size - offset; | |
3622 | total_len = left > size ? size: left; | |
3623 | memcpy(buf, d->buf + offset, total_len); | |
3624 | return total_len; | |
3625 | } | |
3626 | ||
3627 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3628 | if (initpid <= 0) | |
3629 | initpid = fc->pid; | |
3630 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3631 | if (!cg) | |
3632 | return read_file("/proc/stat", buf, size, d); | |
3633 | prune_init_slice(cg); | |
3634 | ||
3635 | cpuset = get_cpuset(cg); | |
3636 | if (!cpuset) | |
3637 | goto err; | |
3638 | ||
3639 | f = fopen("/proc/stat", "r"); | |
3640 | if (!f) | |
3641 | goto err; | |
3642 | ||
3643 | //skip first line | |
3644 | if (getline(&line, &linelen, f) < 0) { | |
3645 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); | |
3646 | goto err; | |
3647 | } | |
3648 | ||
3649 | while (getline(&line, &linelen, f) != -1) { | |
3650 | ssize_t l; | |
3651 | int cpu; | |
3652 | char cpu_char[10]; /* That's a lot of cores */ | |
3653 | char *c; | |
3654 | ||
3655 | if (strlen(line) == 0) | |
3656 | continue; | |
3657 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { | |
3658 | /* not a ^cpuN line containing a number N, just print it */ | |
3659 | l = snprintf(cache, cache_size, "%s", line); | |
3660 | if (l < 0) { | |
3661 | perror("Error writing to cache"); | |
3662 | rv = 0; | |
3663 | goto err; | |
3664 | } | |
3665 | if (l >= cache_size) { | |
3666 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3667 | rv = 0; | |
3668 | goto err; | |
3669 | } | |
3670 | cache += l; | |
3671 | cache_size -= l; | |
3672 | total_len += l; | |
3673 | continue; | |
3674 | } | |
3675 | ||
3676 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3677 | continue; | |
3678 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3679 | continue; | |
3680 | curcpu ++; | |
3681 | ||
3682 | c = strchr(line, ' '); | |
3683 | if (!c) | |
3684 | continue; | |
3685 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
3686 | if (l < 0) { | |
3687 | perror("Error writing to cache"); | |
3688 | rv = 0; | |
3689 | goto err; | |
3690 | ||
3691 | } | |
3692 | if (l >= cache_size) { | |
3693 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3694 | rv = 0; | |
3695 | goto err; | |
3696 | } | |
3697 | ||
3698 | cache += l; | |
3699 | cache_size -= l; | |
3700 | total_len += l; | |
3701 | ||
3702 | if (sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", | |
3703 | &user, | |
3704 | &nice, | |
3705 | &system, | |
3706 | &idle, | |
3707 | &iowait, | |
3708 | &irq, | |
3709 | &softirq, | |
3710 | &steal, | |
3711 | &guest, | |
3712 | &guest_nice) != 10) | |
3713 | continue; | |
3714 | user_sum += user; | |
3715 | nice_sum += nice; | |
3716 | system_sum += system; | |
3717 | idle_sum += idle; | |
3718 | iowait_sum += iowait; | |
3719 | irq_sum += irq; | |
3720 | softirq_sum += softirq; | |
3721 | steal_sum += steal; | |
3722 | guest_sum += guest; | |
3723 | guest_nice_sum += guest_nice; | |
3724 | } | |
3725 | ||
3726 | cache = d->buf; | |
3727 | ||
3728 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3729 | user_sum, | |
3730 | nice_sum, | |
3731 | system_sum, | |
3732 | idle_sum, | |
3733 | iowait_sum, | |
3734 | irq_sum, | |
3735 | softirq_sum, | |
3736 | steal_sum, | |
3737 | guest_sum, | |
3738 | guest_nice_sum); | |
3739 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
3740 | memcpy(cache, cpuall, cpuall_len); | |
3741 | cache += cpuall_len; | |
3742 | } else { | |
3743 | /* shouldn't happen */ | |
3744 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); | |
3745 | cpuall_len = 0; | |
3746 | } | |
3747 | ||
3748 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
3749 | total_len += cpuall_len; | |
3750 | d->cached = 1; | |
3751 | d->size = total_len; | |
3752 | if (total_len > size) | |
3753 | total_len = size; | |
3754 | ||
3755 | memcpy(buf, d->buf, total_len); | |
3756 | rv = total_len; | |
3757 | ||
3758 | err: | |
3759 | if (f) | |
3760 | fclose(f); | |
3761 | free(line); | |
3762 | free(cpuset); | |
3763 | free(cg); | |
3764 | return rv; | |
3765 | } | |
3766 | ||
3767 | /* This function retrieves the busy time of a group of tasks by looking at | |
3768 | * cpuacct.usage. Unfortunately, this only makes sense when the container has | |
3769 | * been given it's own cpuacct cgroup. If not, this function will take the busy | |
3770 | * time of all other taks that do not actually belong to the container into | |
3771 | * account as well. If someone has a clever solution for this please send a | |
3772 | * patch! | |
3773 | */ | |
3774 | static unsigned long get_reaper_busy(pid_t task) | |
3775 | { | |
3776 | pid_t initpid = lookup_initpid_in_store(task); | |
3777 | char *cgroup = NULL, *usage_str = NULL; | |
3778 | unsigned long usage = 0; | |
3779 | ||
3780 | if (initpid <= 0) | |
3781 | return 0; | |
3782 | ||
3783 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
3784 | if (!cgroup) | |
3785 | goto out; | |
3786 | prune_init_slice(cgroup); | |
3787 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) | |
3788 | goto out; | |
3789 | usage = strtoul(usage_str, NULL, 10); | |
3790 | usage /= 1000000000; | |
3791 | ||
3792 | out: | |
3793 | free(cgroup); | |
3794 | free(usage_str); | |
3795 | return usage; | |
3796 | } | |
3797 | ||
3798 | #if RELOADTEST | |
3799 | void iwashere(void) | |
3800 | { | |
3801 | int fd; | |
3802 | ||
3803 | fd = creat("/tmp/lxcfs-iwashere", 0644); | |
3804 | if (fd >= 0) | |
3805 | close(fd); | |
3806 | } | |
3807 | #endif | |
3808 | ||
3809 | /* | |
3810 | * We read /proc/uptime and reuse its second field. | |
3811 | * For the first field, we use the mtime for the reaper for | |
3812 | * the calling pid as returned by getreaperage | |
3813 | */ | |
3814 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
3815 | struct fuse_file_info *fi) | |
3816 | { | |
3817 | struct fuse_context *fc = fuse_get_context(); | |
3818 | struct file_info *d = (struct file_info *)fi->fh; | |
3819 | unsigned long int busytime = get_reaper_busy(fc->pid); | |
3820 | char *cache = d->buf; | |
3821 | ssize_t total_len = 0; | |
3822 | uint64_t idletime, reaperage; | |
3823 | ||
3824 | #if RELOADTEST | |
3825 | iwashere(); | |
3826 | #endif | |
3827 | ||
3828 | if (offset){ | |
3829 | if (!d->cached) | |
3830 | return 0; | |
3831 | if (offset > d->size) | |
3832 | return -EINVAL; | |
3833 | int left = d->size - offset; | |
3834 | total_len = left > size ? size: left; | |
3835 | memcpy(buf, cache + offset, total_len); | |
3836 | return total_len; | |
3837 | } | |
3838 | ||
3839 | reaperage = get_reaper_age(fc->pid); | |
3840 | /* To understand why this is done, please read the comment to the | |
3841 | * get_reaper_busy() function. | |
3842 | */ | |
3843 | idletime = reaperage; | |
3844 | if (reaperage >= busytime) | |
3845 | idletime = reaperage - busytime; | |
3846 | ||
3847 | total_len = snprintf(d->buf, d->buflen, "%"PRIu64".00 %"PRIu64".00\n", reaperage, idletime); | |
3848 | if (total_len < 0 || total_len >= d->buflen){ | |
3849 | lxcfs_error("%s\n", "failed to write to cache"); | |
3850 | return 0; | |
3851 | } | |
3852 | ||
3853 | d->size = (int)total_len; | |
3854 | d->cached = 1; | |
3855 | ||
3856 | if (total_len > size) total_len = size; | |
3857 | ||
3858 | memcpy(buf, d->buf, total_len); | |
3859 | return total_len; | |
3860 | } | |
3861 | ||
3862 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
3863 | struct fuse_file_info *fi) | |
3864 | { | |
3865 | char dev_name[72]; | |
3866 | struct fuse_context *fc = fuse_get_context(); | |
3867 | struct file_info *d = (struct file_info *)fi->fh; | |
3868 | char *cg; | |
3869 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
3870 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
3871 | unsigned long read = 0, write = 0; | |
3872 | unsigned long read_merged = 0, write_merged = 0; | |
3873 | unsigned long read_sectors = 0, write_sectors = 0; | |
3874 | unsigned long read_ticks = 0, write_ticks = 0; | |
3875 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
3876 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
3877 | char *cache = d->buf; | |
3878 | size_t cache_size = d->buflen; | |
3879 | char *line = NULL; | |
3880 | size_t linelen = 0, total_len = 0, rv = 0; | |
3881 | unsigned int major = 0, minor = 0; | |
3882 | int i = 0; | |
3883 | FILE *f = NULL; | |
3884 | ||
3885 | if (offset){ | |
3886 | if (offset > d->size) | |
3887 | return -EINVAL; | |
3888 | if (!d->cached) | |
3889 | return 0; | |
3890 | int left = d->size - offset; | |
3891 | total_len = left > size ? size: left; | |
3892 | memcpy(buf, cache + offset, total_len); | |
3893 | return total_len; | |
3894 | } | |
3895 | ||
3896 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3897 | if (initpid <= 0) | |
3898 | initpid = fc->pid; | |
3899 | cg = get_pid_cgroup(initpid, "blkio"); | |
3900 | if (!cg) | |
3901 | return read_file("/proc/diskstats", buf, size, d); | |
3902 | prune_init_slice(cg); | |
3903 | ||
3904 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) | |
3905 | goto err; | |
3906 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) | |
3907 | goto err; | |
3908 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) | |
3909 | goto err; | |
3910 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) | |
3911 | goto err; | |
3912 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) | |
3913 | goto err; | |
3914 | ||
3915 | ||
3916 | f = fopen("/proc/diskstats", "r"); | |
3917 | if (!f) | |
3918 | goto err; | |
3919 | ||
3920 | while (getline(&line, &linelen, f) != -1) { | |
3921 | ssize_t l; | |
3922 | char lbuf[256]; | |
3923 | ||
3924 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
3925 | if (i != 3) | |
3926 | continue; | |
3927 | ||
3928 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
3929 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
3930 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
3931 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
3932 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
3933 | read_sectors = read_sectors/512; | |
3934 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
3935 | write_sectors = write_sectors/512; | |
3936 | ||
3937 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
3938 | rd_svctm = rd_svctm/1000000; | |
3939 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
3940 | rd_wait = rd_wait/1000000; | |
3941 | read_ticks = rd_svctm + rd_wait; | |
3942 | ||
3943 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
3944 | wr_svctm = wr_svctm/1000000; | |
3945 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
3946 | wr_wait = wr_wait/1000000; | |
3947 | write_ticks = wr_svctm + wr_wait; | |
3948 | ||
3949 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
3950 | tot_ticks = tot_ticks/1000000; | |
3951 | ||
3952 | memset(lbuf, 0, 256); | |
3953 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) | |
3954 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
3955 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
3956 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
3957 | else | |
3958 | continue; | |
3959 | ||
3960 | l = snprintf(cache, cache_size, "%s", lbuf); | |
3961 | if (l < 0) { | |
3962 | perror("Error writing to fuse buf"); | |
3963 | rv = 0; | |
3964 | goto err; | |
3965 | } | |
3966 | if (l >= cache_size) { | |
3967 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
3968 | rv = 0; | |
3969 | goto err; | |
3970 | } | |
3971 | cache += l; | |
3972 | cache_size -= l; | |
3973 | total_len += l; | |
3974 | } | |
3975 | ||
3976 | d->cached = 1; | |
3977 | d->size = total_len; | |
3978 | if (total_len > size ) total_len = size; | |
3979 | memcpy(buf, d->buf, total_len); | |
3980 | ||
3981 | rv = total_len; | |
3982 | err: | |
3983 | free(cg); | |
3984 | if (f) | |
3985 | fclose(f); | |
3986 | free(line); | |
3987 | free(io_serviced_str); | |
3988 | free(io_merged_str); | |
3989 | free(io_service_bytes_str); | |
3990 | free(io_wait_time_str); | |
3991 | free(io_service_time_str); | |
3992 | return rv; | |
3993 | } | |
3994 | ||
3995 | static int proc_swaps_read(char *buf, size_t size, off_t offset, | |
3996 | struct fuse_file_info *fi) | |
3997 | { | |
3998 | struct fuse_context *fc = fuse_get_context(); | |
3999 | struct file_info *d = (struct file_info *)fi->fh; | |
4000 | char *cg = NULL; | |
4001 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL; | |
4002 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; | |
4003 | ssize_t total_len = 0, rv = 0; | |
4004 | ssize_t l = 0; | |
4005 | char *cache = d->buf; | |
4006 | ||
4007 | if (offset) { | |
4008 | if (offset > d->size) | |
4009 | return -EINVAL; | |
4010 | if (!d->cached) | |
4011 | return 0; | |
4012 | int left = d->size - offset; | |
4013 | total_len = left > size ? size: left; | |
4014 | memcpy(buf, cache + offset, total_len); | |
4015 | return total_len; | |
4016 | } | |
4017 | ||
4018 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
4019 | if (initpid <= 0) | |
4020 | initpid = fc->pid; | |
4021 | cg = get_pid_cgroup(initpid, "memory"); | |
4022 | if (!cg) | |
4023 | return read_file("/proc/swaps", buf, size, d); | |
4024 | prune_init_slice(cg); | |
4025 | ||
4026 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); | |
4027 | ||
4028 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
4029 | goto err; | |
4030 | ||
4031 | memusage = strtoul(memusage_str, NULL, 10); | |
4032 | ||
4033 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
4034 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
4035 | ||
4036 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); | |
4037 | memswusage = strtoul(memswusage_str, NULL, 10); | |
4038 | ||
4039 | swap_total = (memswlimit - memlimit) / 1024; | |
4040 | swap_free = (memswusage - memusage) / 1024; | |
4041 | } | |
4042 | ||
4043 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
4044 | ||
4045 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
4046 | if (!memswlimit) { | |
4047 | char *line = NULL; | |
4048 | size_t linelen = 0; | |
4049 | FILE *f = fopen("/proc/meminfo", "r"); | |
4050 | ||
4051 | if (!f) | |
4052 | goto err; | |
4053 | ||
4054 | while (getline(&line, &linelen, f) != -1) { | |
4055 | if (startswith(line, "SwapTotal:")) { | |
4056 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
4057 | } else if (startswith(line, "SwapFree:")) { | |
4058 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
4059 | } | |
4060 | } | |
4061 | ||
4062 | free(line); | |
4063 | fclose(f); | |
4064 | } | |
4065 | ||
4066 | if (swap_total > 0) { | |
4067 | l = snprintf(d->buf + total_len, d->size - total_len, | |
4068 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
4069 | swap_total, swap_free); | |
4070 | total_len += l; | |
4071 | } | |
4072 | ||
4073 | if (total_len < 0 || l < 0) { | |
4074 | perror("Error writing to cache"); | |
4075 | rv = 0; | |
4076 | goto err; | |
4077 | } | |
4078 | ||
4079 | d->cached = 1; | |
4080 | d->size = (int)total_len; | |
4081 | ||
4082 | if (total_len > size) total_len = size; | |
4083 | memcpy(buf, d->buf, total_len); | |
4084 | rv = total_len; | |
4085 | ||
4086 | err: | |
4087 | free(cg); | |
4088 | free(memswlimit_str); | |
4089 | free(memlimit_str); | |
4090 | free(memusage_str); | |
4091 | free(memswusage_str); | |
4092 | return rv; | |
4093 | } | |
4094 | ||
4095 | static off_t get_procfile_size(const char *which) | |
4096 | { | |
4097 | FILE *f = fopen(which, "r"); | |
4098 | char *line = NULL; | |
4099 | size_t len = 0; | |
4100 | ssize_t sz, answer = 0; | |
4101 | if (!f) | |
4102 | return 0; | |
4103 | ||
4104 | while ((sz = getline(&line, &len, f)) != -1) | |
4105 | answer += sz; | |
4106 | fclose (f); | |
4107 | free(line); | |
4108 | ||
4109 | return answer; | |
4110 | } | |
4111 | ||
4112 | int proc_getattr(const char *path, struct stat *sb) | |
4113 | { | |
4114 | struct timespec now; | |
4115 | ||
4116 | memset(sb, 0, sizeof(struct stat)); | |
4117 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
4118 | return -EINVAL; | |
4119 | sb->st_uid = sb->st_gid = 0; | |
4120 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
4121 | if (strcmp(path, "/proc") == 0) { | |
4122 | sb->st_mode = S_IFDIR | 00555; | |
4123 | sb->st_nlink = 2; | |
4124 | return 0; | |
4125 | } | |
4126 | if (strcmp(path, "/proc/meminfo") == 0 || | |
4127 | strcmp(path, "/proc/cpuinfo") == 0 || | |
4128 | strcmp(path, "/proc/uptime") == 0 || | |
4129 | strcmp(path, "/proc/stat") == 0 || | |
4130 | strcmp(path, "/proc/diskstats") == 0 || | |
4131 | strcmp(path, "/proc/swaps") == 0) { | |
4132 | sb->st_size = 0; | |
4133 | sb->st_mode = S_IFREG | 00444; | |
4134 | sb->st_nlink = 1; | |
4135 | return 0; | |
4136 | } | |
4137 | ||
4138 | return -ENOENT; | |
4139 | } | |
4140 | ||
4141 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4142 | struct fuse_file_info *fi) | |
4143 | { | |
4144 | if (filler(buf, ".", NULL, 0) != 0 || | |
4145 | filler(buf, "..", NULL, 0) != 0 || | |
4146 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4147 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4148 | filler(buf, "stat", NULL, 0) != 0 || | |
4149 | filler(buf, "uptime", NULL, 0) != 0 || | |
4150 | filler(buf, "diskstats", NULL, 0) != 0 || | |
4151 | filler(buf, "swaps", NULL, 0) != 0) | |
4152 | return -EINVAL; | |
4153 | return 0; | |
4154 | } | |
4155 | ||
4156 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4157 | { | |
4158 | int type = -1; | |
4159 | struct file_info *info; | |
4160 | ||
4161 | if (strcmp(path, "/proc/meminfo") == 0) | |
4162 | type = LXC_TYPE_PROC_MEMINFO; | |
4163 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4164 | type = LXC_TYPE_PROC_CPUINFO; | |
4165 | else if (strcmp(path, "/proc/uptime") == 0) | |
4166 | type = LXC_TYPE_PROC_UPTIME; | |
4167 | else if (strcmp(path, "/proc/stat") == 0) | |
4168 | type = LXC_TYPE_PROC_STAT; | |
4169 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4170 | type = LXC_TYPE_PROC_DISKSTATS; | |
4171 | else if (strcmp(path, "/proc/swaps") == 0) | |
4172 | type = LXC_TYPE_PROC_SWAPS; | |
4173 | if (type == -1) | |
4174 | return -ENOENT; | |
4175 | ||
4176 | info = malloc(sizeof(*info)); | |
4177 | if (!info) | |
4178 | return -ENOMEM; | |
4179 | ||
4180 | memset(info, 0, sizeof(*info)); | |
4181 | info->type = type; | |
4182 | ||
4183 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4184 | do { | |
4185 | info->buf = malloc(info->buflen); | |
4186 | } while (!info->buf); | |
4187 | memset(info->buf, 0, info->buflen); | |
4188 | /* set actual size to buffer size */ | |
4189 | info->size = info->buflen; | |
4190 | ||
4191 | fi->fh = (unsigned long)info; | |
4192 | return 0; | |
4193 | } | |
4194 | ||
4195 | int proc_access(const char *path, int mask) | |
4196 | { | |
4197 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) | |
4198 | return 0; | |
4199 | ||
4200 | /* these are all read-only */ | |
4201 | if ((mask & ~R_OK) != 0) | |
4202 | return -EACCES; | |
4203 | return 0; | |
4204 | } | |
4205 | ||
4206 | int proc_release(const char *path, struct fuse_file_info *fi) | |
4207 | { | |
4208 | do_release_file_info(fi); | |
4209 | return 0; | |
4210 | } | |
4211 | ||
4212 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4213 | struct fuse_file_info *fi) | |
4214 | { | |
4215 | struct file_info *f = (struct file_info *) fi->fh; | |
4216 | ||
4217 | switch (f->type) { | |
4218 | case LXC_TYPE_PROC_MEMINFO: | |
4219 | return proc_meminfo_read(buf, size, offset, fi); | |
4220 | case LXC_TYPE_PROC_CPUINFO: | |
4221 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4222 | case LXC_TYPE_PROC_UPTIME: | |
4223 | return proc_uptime_read(buf, size, offset, fi); | |
4224 | case LXC_TYPE_PROC_STAT: | |
4225 | return proc_stat_read(buf, size, offset, fi); | |
4226 | case LXC_TYPE_PROC_DISKSTATS: | |
4227 | return proc_diskstats_read(buf, size, offset, fi); | |
4228 | case LXC_TYPE_PROC_SWAPS: | |
4229 | return proc_swaps_read(buf, size, offset, fi); | |
4230 | default: | |
4231 | return -EINVAL; | |
4232 | } | |
4233 | } | |
4234 | ||
4235 | /* | |
4236 | * Functions needed to setup cgroups in the __constructor__. | |
4237 | */ | |
4238 | ||
4239 | static bool mkdir_p(const char *dir, mode_t mode) | |
4240 | { | |
4241 | const char *tmp = dir; | |
4242 | const char *orig = dir; | |
4243 | char *makeme; | |
4244 | ||
4245 | do { | |
4246 | dir = tmp + strspn(tmp, "/"); | |
4247 | tmp = dir + strcspn(dir, "/"); | |
4248 | makeme = strndup(orig, dir - orig); | |
4249 | if (!makeme) | |
4250 | return false; | |
4251 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
4252 | lxcfs_error("Failed to create directory '%s': %s.\n", | |
4253 | makeme, strerror(errno)); | |
4254 | free(makeme); | |
4255 | return false; | |
4256 | } | |
4257 | free(makeme); | |
4258 | } while(tmp != dir); | |
4259 | ||
4260 | return true; | |
4261 | } | |
4262 | ||
4263 | static bool umount_if_mounted(void) | |
4264 | { | |
4265 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
4266 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); | |
4267 | return false; | |
4268 | } | |
4269 | return true; | |
4270 | } | |
4271 | ||
4272 | /* __typeof__ should be safe to use with all compilers. */ | |
4273 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
4274 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
4275 | { | |
4276 | return (fs->f_type == (fs_type_magic)magic_val); | |
4277 | } | |
4278 | ||
4279 | /* | |
4280 | * looking at fs/proc_namespace.c, it appears we can | |
4281 | * actually expect the rootfs entry to very specifically contain | |
4282 | * " - rootfs rootfs " | |
4283 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
4284 | * the rootfs entry should always be skipped in mountinfo contents. | |
4285 | */ | |
4286 | static bool is_on_ramfs(void) | |
4287 | { | |
4288 | FILE *f; | |
4289 | char *p, *p2; | |
4290 | char *line = NULL; | |
4291 | size_t len = 0; | |
4292 | int i; | |
4293 | ||
4294 | f = fopen("/proc/self/mountinfo", "r"); | |
4295 | if (!f) | |
4296 | return false; | |
4297 | ||
4298 | while (getline(&line, &len, f) != -1) { | |
4299 | for (p = line, i = 0; p && i < 4; i++) | |
4300 | p = strchr(p + 1, ' '); | |
4301 | if (!p) | |
4302 | continue; | |
4303 | p2 = strchr(p + 1, ' '); | |
4304 | if (!p2) | |
4305 | continue; | |
4306 | *p2 = '\0'; | |
4307 | if (strcmp(p + 1, "/") == 0) { | |
4308 | // this is '/'. is it the ramfs? | |
4309 | p = strchr(p2 + 1, '-'); | |
4310 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
4311 | free(line); | |
4312 | fclose(f); | |
4313 | return true; | |
4314 | } | |
4315 | } | |
4316 | } | |
4317 | free(line); | |
4318 | fclose(f); | |
4319 | return false; | |
4320 | } | |
4321 | ||
4322 | static int pivot_enter() | |
4323 | { | |
4324 | int ret = -1, oldroot = -1, newroot = -1; | |
4325 | ||
4326 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
4327 | if (oldroot < 0) { | |
4328 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
4329 | return ret; | |
4330 | } | |
4331 | ||
4332 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
4333 | if (newroot < 0) { | |
4334 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
4335 | goto err; | |
4336 | } | |
4337 | ||
4338 | /* change into new root fs */ | |
4339 | if (fchdir(newroot) < 0) { | |
4340 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
4341 | goto err; | |
4342 | } | |
4343 | ||
4344 | /* pivot_root into our new root fs */ | |
4345 | if (pivot_root(".", ".") < 0) { | |
4346 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
4347 | goto err; | |
4348 | } | |
4349 | ||
4350 | /* | |
4351 | * At this point the old-root is mounted on top of our new-root. | |
4352 | * To unmounted it we must not be chdir'd into it, so escape back | |
4353 | * to the old-root. | |
4354 | */ | |
4355 | if (fchdir(oldroot) < 0) { | |
4356 | lxcfs_error("%s\n", "Failed to enter old root."); | |
4357 | goto err; | |
4358 | } | |
4359 | ||
4360 | if (umount2(".", MNT_DETACH) < 0) { | |
4361 | lxcfs_error("%s\n", "Failed to detach old root."); | |
4362 | goto err; | |
4363 | } | |
4364 | ||
4365 | if (fchdir(newroot) < 0) { | |
4366 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
4367 | goto err; | |
4368 | } | |
4369 | ||
4370 | ret = 0; | |
4371 | ||
4372 | err: | |
4373 | if (oldroot > 0) | |
4374 | close(oldroot); | |
4375 | if (newroot > 0) | |
4376 | close(newroot); | |
4377 | ||
4378 | return ret; | |
4379 | } | |
4380 | ||
4381 | static int chroot_enter() | |
4382 | { | |
4383 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
4384 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
4385 | return -1; | |
4386 | } | |
4387 | ||
4388 | if (chroot(".") < 0) { | |
4389 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
4390 | return -1; | |
4391 | } | |
4392 | ||
4393 | if (chdir("/") < 0) { | |
4394 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
4395 | return -1; | |
4396 | } | |
4397 | ||
4398 | return 0; | |
4399 | } | |
4400 | ||
4401 | static int permute_and_enter(void) | |
4402 | { | |
4403 | struct statfs sb; | |
4404 | ||
4405 | if (statfs("/", &sb) < 0) { | |
4406 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
4407 | return -1; | |
4408 | } | |
4409 | ||
4410 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
4411 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
4412 | * /proc/1/mountinfo. */ | |
4413 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
4414 | return chroot_enter(); | |
4415 | ||
4416 | if (pivot_enter() < 0) { | |
4417 | lxcfs_error("%s\n", "Could not perform pivot root."); | |
4418 | return -1; | |
4419 | } | |
4420 | ||
4421 | return 0; | |
4422 | } | |
4423 | ||
4424 | /* Prepare our new clean root. */ | |
4425 | static int permute_prepare(void) | |
4426 | { | |
4427 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
4428 | lxcfs_error("%s\n", "Failed to create directory for new root."); | |
4429 | return -1; | |
4430 | } | |
4431 | ||
4432 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
4433 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); | |
4434 | return -1; | |
4435 | } | |
4436 | ||
4437 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
4438 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); | |
4439 | return -1; | |
4440 | } | |
4441 | ||
4442 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
4443 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); | |
4444 | return -1; | |
4445 | } | |
4446 | ||
4447 | return 0; | |
4448 | } | |
4449 | ||
4450 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ | |
4451 | static bool permute_root(void) | |
4452 | { | |
4453 | /* Prepare new root. */ | |
4454 | if (permute_prepare() < 0) | |
4455 | return false; | |
4456 | ||
4457 | /* Pivot into new root. */ | |
4458 | if (permute_and_enter() < 0) | |
4459 | return false; | |
4460 | ||
4461 | return true; | |
4462 | } | |
4463 | ||
4464 | static int preserve_mnt_ns(int pid) | |
4465 | { | |
4466 | int ret; | |
4467 | size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt"); | |
4468 | char path[len]; | |
4469 | ||
4470 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
4471 | if (ret < 0 || (size_t)ret >= len) | |
4472 | return -1; | |
4473 | ||
4474 | return open(path, O_RDONLY | O_CLOEXEC); | |
4475 | } | |
4476 | ||
4477 | static bool cgfs_prepare_mounts(void) | |
4478 | { | |
4479 | if (!mkdir_p(BASEDIR, 0700)) { | |
4480 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); | |
4481 | return false; | |
4482 | } | |
4483 | ||
4484 | if (!umount_if_mounted()) { | |
4485 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); | |
4486 | return false; | |
4487 | } | |
4488 | ||
4489 | if (unshare(CLONE_NEWNS) < 0) { | |
4490 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); | |
4491 | return false; | |
4492 | } | |
4493 | ||
4494 | cgroup_mount_ns_fd = preserve_mnt_ns(getpid()); | |
4495 | if (cgroup_mount_ns_fd < 0) { | |
4496 | lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno)); | |
4497 | return false; | |
4498 | } | |
4499 | ||
4500 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { | |
4501 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); | |
4502 | return false; | |
4503 | } | |
4504 | ||
4505 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { | |
4506 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); | |
4507 | return false; | |
4508 | } | |
4509 | ||
4510 | return true; | |
4511 | } | |
4512 | ||
4513 | static bool cgfs_mount_hierarchies(void) | |
4514 | { | |
4515 | char *target; | |
4516 | size_t clen, len; | |
4517 | int i, ret; | |
4518 | ||
4519 | for (i = 0; i < num_hierarchies; i++) { | |
4520 | char *controller = hierarchies[i]; | |
4521 | ||
4522 | clen = strlen(controller); | |
4523 | len = strlen(BASEDIR) + clen + 2; | |
4524 | target = malloc(len); | |
4525 | if (!target) | |
4526 | return false; | |
4527 | ||
4528 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); | |
4529 | if (ret < 0 || ret >= len) { | |
4530 | free(target); | |
4531 | return false; | |
4532 | } | |
4533 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
4534 | free(target); | |
4535 | return false; | |
4536 | } | |
4537 | if (!strcmp(controller, "unified")) | |
4538 | ret = mount("none", target, "cgroup2", 0, NULL); | |
4539 | else | |
4540 | ret = mount(controller, target, "cgroup", 0, controller); | |
4541 | if (ret < 0) { | |
4542 | lxcfs_error("Failed mounting cgroup %s: %s\n", controller, strerror(errno)); | |
4543 | free(target); | |
4544 | return false; | |
4545 | } | |
4546 | ||
4547 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
4548 | if (fd_hierarchies[i] < 0) { | |
4549 | free(target); | |
4550 | return false; | |
4551 | } | |
4552 | free(target); | |
4553 | } | |
4554 | return true; | |
4555 | } | |
4556 | ||
4557 | static bool cgfs_setup_controllers(void) | |
4558 | { | |
4559 | if (!cgfs_prepare_mounts()) | |
4560 | return false; | |
4561 | ||
4562 | if (!cgfs_mount_hierarchies()) { | |
4563 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); | |
4564 | return false; | |
4565 | } | |
4566 | ||
4567 | if (!permute_root()) | |
4568 | return false; | |
4569 | ||
4570 | return true; | |
4571 | } | |
4572 | ||
4573 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) | |
4574 | { | |
4575 | FILE *f; | |
4576 | char *cret, *line = NULL; | |
4577 | char cwd[MAXPATHLEN]; | |
4578 | size_t len = 0; | |
4579 | int i, init_ns = -1; | |
4580 | bool found_unified = false; | |
4581 | ||
4582 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
4583 | lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno)); | |
4584 | return; | |
4585 | } | |
4586 | ||
4587 | while (getline(&line, &len, f) != -1) { | |
4588 | char *idx, *p, *p2; | |
4589 | ||
4590 | p = strchr(line, ':'); | |
4591 | if (!p) | |
4592 | goto out; | |
4593 | idx = line; | |
4594 | *(p++) = '\0'; | |
4595 | ||
4596 | p2 = strrchr(p, ':'); | |
4597 | if (!p2) | |
4598 | goto out; | |
4599 | *p2 = '\0'; | |
4600 | ||
4601 | /* With cgroupv2 /proc/self/cgroup can contain entries of the | |
4602 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
4603 | * because it parses out the empty string "" and later on passes | |
4604 | * it to mount(). Let's skip such entries. | |
4605 | */ | |
4606 | if (!strcmp(p, "") && !strcmp(idx, "0") && !found_unified) { | |
4607 | found_unified = true; | |
4608 | p = "unified"; | |
4609 | } | |
4610 | ||
4611 | if (!store_hierarchy(line, p)) | |
4612 | goto out; | |
4613 | } | |
4614 | ||
4615 | /* Preserve initial namespace. */ | |
4616 | init_ns = preserve_mnt_ns(getpid()); | |
4617 | if (init_ns < 0) { | |
4618 | lxcfs_error("%s\n", "Failed to preserve initial mount namespace."); | |
4619 | goto out; | |
4620 | } | |
4621 | ||
4622 | fd_hierarchies = malloc(sizeof(int) * num_hierarchies); | |
4623 | if (!fd_hierarchies) { | |
4624 | lxcfs_error("%s\n", strerror(errno)); | |
4625 | goto out; | |
4626 | } | |
4627 | ||
4628 | for (i = 0; i < num_hierarchies; i++) | |
4629 | fd_hierarchies[i] = -1; | |
4630 | ||
4631 | cret = getcwd(cwd, MAXPATHLEN); | |
4632 | if (!cret) | |
4633 | lxcfs_debug("Could not retrieve current working directory: %s.\n", strerror(errno)); | |
4634 | ||
4635 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace | |
4636 | * to privately mount lxcfs cgroups. */ | |
4637 | if (!cgfs_setup_controllers()) { | |
4638 | lxcfs_error("%s\n", "Failed to setup private cgroup mounts for lxcfs."); | |
4639 | goto out; | |
4640 | } | |
4641 | ||
4642 | if (setns(init_ns, 0) < 0) { | |
4643 | lxcfs_error("Failed to switch back to initial mount namespace: %s.\n", strerror(errno)); | |
4644 | goto out; | |
4645 | } | |
4646 | ||
4647 | if (!cret || chdir(cwd) < 0) | |
4648 | lxcfs_debug("Could not change back to original working directory: %s.\n", strerror(errno)); | |
4649 | ||
4650 | print_subsystems(); | |
4651 | ||
4652 | out: | |
4653 | free(line); | |
4654 | fclose(f); | |
4655 | if (init_ns >= 0) | |
4656 | close(init_ns); | |
4657 | } | |
4658 | ||
4659 | static void __attribute__((destructor)) free_subsystems(void) | |
4660 | { | |
4661 | int i; | |
4662 | ||
4663 | lxcfs_debug("%s\n", "Running destructor for liblxcfs."); | |
4664 | ||
4665 | for (i = 0; i < num_hierarchies; i++) { | |
4666 | if (hierarchies[i]) | |
4667 | free(hierarchies[i]); | |
4668 | if (fd_hierarchies && fd_hierarchies[i] >= 0) | |
4669 | close(fd_hierarchies[i]); | |
4670 | } | |
4671 | free(hierarchies); | |
4672 | free(fd_hierarchies); | |
4673 | ||
4674 | if (cgroup_mount_ns_fd >= 0) | |
4675 | close(cgroup_mount_ns_fd); | |
4676 | } |