]>
Commit | Line | Data |
---|---|---|
237e200e SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014-2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * See COPYING file for details. | |
7 | */ | |
8 | ||
9 | #define FUSE_USE_VERSION 26 | |
10 | ||
0ecddf02 | 11 | #define __STDC_FORMAT_MACROS |
237e200e | 12 | #include <dirent.h> |
29a73c2f | 13 | #include <errno.h> |
237e200e SH |
14 | #include <fcntl.h> |
15 | #include <fuse.h> | |
0ecddf02 | 16 | #include <inttypes.h> |
237e200e | 17 | #include <libgen.h> |
237e200e | 18 | #include <pthread.h> |
29a73c2f CB |
19 | #include <sched.h> |
20 | #include <stdbool.h> | |
0ecddf02 | 21 | #include <stdint.h> |
29a73c2f CB |
22 | #include <stdio.h> |
23 | #include <stdlib.h> | |
24 | #include <string.h> | |
25 | #include <time.h> | |
26 | #include <unistd.h> | |
27 | #include <wait.h> | |
d89504c4 | 28 | #include <linux/magic.h> |
237e200e | 29 | #include <linux/sched.h> |
29a73c2f CB |
30 | #include <sys/epoll.h> |
31 | #include <sys/mman.h> | |
32 | #include <sys/mount.h> | |
237e200e SH |
33 | #include <sys/param.h> |
34 | #include <sys/socket.h> | |
29a73c2f | 35 | #include <sys/syscall.h> |
0ecddf02 | 36 | #include <sys/sysinfo.h> |
d89504c4 | 37 | #include <sys/vfs.h> |
237e200e | 38 | |
237e200e | 39 | #include "bindings.h" |
237e200e SH |
40 | #include "config.h" // for VERSION |
41 | ||
0ecddf02 CB |
42 | /* Maximum number for 64 bit integer is a string with 21 digits: 2^64 - 1 = 21 */ |
43 | #define LXCFS_NUMSTRLEN64 21 | |
44 | ||
29a73c2f CB |
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 | ||
237e200e SH |
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, | |
70dcc12e | 68 | LXC_TYPE_PROC_SWAPS, |
46be8eed | 69 | LXC_TYPE_PROC_LOADAVG, |
237e200e SH |
70 | }; |
71 | ||
72 | struct file_info { | |
73 | char *controller; | |
74 | char *cgroup; | |
75 | char *file; | |
76 | int type; | |
77 | char *buf; // unused as of yet | |
78 | int buflen; | |
79 | int size; //actual data size | |
80 | int cached; | |
81 | }; | |
82 | ||
8be92dd1 JS |
83 | struct cpuacct_usage { |
84 | uint64_t user; | |
85 | uint64_t system; | |
86 | }; | |
87 | ||
0e47acaa | 88 | /* The function of hash table.*/ |
89 | #define LOAD_SIZE 100 /*the size of hash_table */ | |
6db4f7a3 | 90 | #define FLUSH_TIME 5 /*the flush rate */ |
91 | #define DEPTH_DIR 3 /*the depth of per cgroup */ | |
92 | /* The function of calculate loadavg .*/ | |
93 | #define FSHIFT 11 /* nr of bits of precision */ | |
94 | #define FIXED_1 (1<<FSHIFT) /* 1.0 as fixed-point */ | |
95 | #define EXP_1 1884 /* 1/exp(5sec/1min) as fixed-point */ | |
96 | #define EXP_5 2014 /* 1/exp(5sec/5min) */ | |
97 | #define EXP_15 2037 /* 1/exp(5sec/15min) */ | |
98 | #define LOAD_INT(x) ((x) >> FSHIFT) | |
99 | #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) | |
100 | /* | |
101 | * This parameter is used for proc_loadavg_read(). | |
102 | * 1 means use loadavg, 0 means not use. | |
103 | */ | |
104 | static int loadavg = 0; | |
a83618e2 | 105 | static volatile sig_atomic_t loadavg_stop = 0; |
0e47acaa | 106 | static int calc_hash(char *name) |
107 | { | |
108 | unsigned int hash = 0; | |
109 | unsigned int x = 0; | |
110 | /* ELFHash algorithm. */ | |
111 | while (*name) { | |
112 | hash = (hash << 4) + *name++; | |
113 | x = hash & 0xf0000000; | |
114 | if (x != 0) | |
115 | hash ^= (x >> 24); | |
116 | hash &= ~x; | |
117 | } | |
118 | return ((hash & 0x7fffffff) % LOAD_SIZE); | |
119 | } | |
120 | ||
121 | struct load_node { | |
122 | char *cg; /*cg */ | |
123 | unsigned long avenrun[3]; /* Load averages */ | |
124 | unsigned int run_pid; | |
125 | unsigned int total_pid; | |
126 | unsigned int last_pid; | |
127 | int cfd; /* The file descriptor of the mounted cgroup */ | |
128 | struct load_node *next; | |
129 | struct load_node **pre; | |
130 | }; | |
131 | ||
132 | struct load_head { | |
133 | /* | |
134 | * The lock is about insert load_node and refresh load_node.To the first | |
135 | * load_node of each hash bucket, insert and refresh in this hash bucket is | |
136 | * mutually exclusive. | |
137 | */ | |
138 | pthread_mutex_t lock; | |
139 | /* | |
140 | * The rdlock is about read loadavg and delete load_node.To each hash | |
141 | * bucket, read and delete is mutually exclusive. But at the same time, we | |
142 | * allow paratactic read operation. This rdlock is at list level. | |
143 | */ | |
144 | pthread_rwlock_t rdlock; | |
145 | /* | |
146 | * The rilock is about read loadavg and insert load_node.To the first | |
147 | * load_node of each hash bucket, read and insert is mutually exclusive. | |
148 | * But at the same time, we allow paratactic read operation. | |
149 | */ | |
150 | pthread_rwlock_t rilock; | |
151 | struct load_node *next; | |
152 | }; | |
153 | ||
154 | static struct load_head load_hash[LOAD_SIZE]; /* hash table */ | |
155 | /* | |
156 | * init_load initialize the hash table. | |
157 | * Return 0 on success, return -1 on failure. | |
158 | */ | |
159 | static int init_load(void) | |
160 | { | |
161 | int i; | |
162 | int ret; | |
163 | ||
164 | for (i = 0; i < LOAD_SIZE; i++) { | |
165 | load_hash[i].next = NULL; | |
166 | ret = pthread_mutex_init(&load_hash[i].lock, NULL); | |
167 | if (ret != 0) { | |
168 | lxcfs_error("%s\n", "Failed to initialize lock"); | |
169 | goto out3; | |
170 | } | |
171 | ret = pthread_rwlock_init(&load_hash[i].rdlock, NULL); | |
172 | if (ret != 0) { | |
173 | lxcfs_error("%s\n", "Failed to initialize rdlock"); | |
174 | goto out2; | |
175 | } | |
176 | ret = pthread_rwlock_init(&load_hash[i].rilock, NULL); | |
177 | if (ret != 0) { | |
178 | lxcfs_error("%s\n", "Failed to initialize rilock"); | |
179 | goto out1; | |
180 | } | |
181 | } | |
182 | return 0; | |
183 | out1: | |
184 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
185 | out2: | |
186 | pthread_mutex_destroy(&load_hash[i].lock); | |
187 | out3: | |
188 | while (i > 0) { | |
189 | i--; | |
190 | pthread_mutex_destroy(&load_hash[i].lock); | |
191 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
192 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
193 | } | |
194 | return -1; | |
195 | } | |
196 | ||
197 | static void insert_node(struct load_node **n, int locate) | |
198 | { | |
199 | struct load_node *f; | |
200 | ||
201 | pthread_mutex_lock(&load_hash[locate].lock); | |
202 | pthread_rwlock_wrlock(&load_hash[locate].rilock); | |
203 | f = load_hash[locate].next; | |
204 | load_hash[locate].next = *n; | |
205 | ||
206 | (*n)->pre = &(load_hash[locate].next); | |
207 | if (f) | |
208 | f->pre = &((*n)->next); | |
209 | (*n)->next = f; | |
210 | pthread_mutex_unlock(&load_hash[locate].lock); | |
211 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
212 | } | |
213 | /* | |
214 | * locate_node() finds special node. Not return NULL means success. | |
215 | * It should be noted that rdlock isn't unlocked at the end of code | |
216 | * because this function is used to read special node. Delete is not | |
217 | * allowed before read has ended. | |
218 | * unlock rdlock only in proc_loadavg_read(). | |
219 | */ | |
220 | static struct load_node *locate_node(char *cg, int locate) | |
221 | { | |
222 | struct load_node *f = NULL; | |
223 | int i = 0; | |
224 | ||
225 | pthread_rwlock_rdlock(&load_hash[locate].rilock); | |
226 | pthread_rwlock_rdlock(&load_hash[locate].rdlock); | |
227 | if (load_hash[locate].next == NULL) { | |
228 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
229 | return f; | |
230 | } | |
231 | f = load_hash[locate].next; | |
232 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
233 | while (f && ((i = strcmp(f->cg, cg)) != 0)) | |
234 | f = f->next; | |
235 | return f; | |
236 | } | |
237 | /* Delete the load_node n and return the next node of it. */ | |
238 | static struct load_node *del_node(struct load_node *n, int locate) | |
239 | { | |
240 | struct load_node *g; | |
241 | ||
242 | pthread_rwlock_wrlock(&load_hash[locate].rdlock); | |
243 | if (n->next == NULL) { | |
244 | *(n->pre) = NULL; | |
245 | } else { | |
246 | *(n->pre) = n->next; | |
247 | n->next->pre = n->pre; | |
248 | } | |
249 | g = n->next; | |
250 | free(n->cg); | |
251 | free(n); | |
252 | pthread_rwlock_unlock(&load_hash[locate].rdlock); | |
253 | return g; | |
254 | } | |
255 | ||
a83618e2 | 256 | static void load_free(void) |
9c480eb7 | 257 | { |
258 | int i; | |
259 | struct load_node *f, *p; | |
260 | ||
261 | for (i = 0; i < LOAD_SIZE; i++) { | |
262 | pthread_mutex_lock(&load_hash[i].lock); | |
263 | pthread_rwlock_wrlock(&load_hash[i].rilock); | |
264 | pthread_rwlock_wrlock(&load_hash[i].rdlock); | |
265 | if (load_hash[i].next == NULL) { | |
266 | pthread_mutex_unlock(&load_hash[i].lock); | |
267 | pthread_mutex_destroy(&load_hash[i].lock); | |
268 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
269 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
270 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
271 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
272 | continue; | |
273 | } | |
274 | for (f = load_hash[i].next; f; ) { | |
275 | free(f->cg); | |
276 | p = f->next; | |
277 | free(f); | |
278 | f = p; | |
279 | } | |
280 | pthread_mutex_unlock(&load_hash[i].lock); | |
281 | pthread_mutex_destroy(&load_hash[i].lock); | |
282 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
283 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
284 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
285 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
286 | } | |
287 | } | |
f34de69a CB |
288 | /* Reserve buffer size to account for file size changes. */ |
289 | #define BUF_RESERVE_SIZE 512 | |
237e200e SH |
290 | |
291 | /* | |
292 | * A table caching which pid is init for a pid namespace. | |
293 | * When looking up which pid is init for $qpid, we first | |
294 | * 1. Stat /proc/$qpid/ns/pid. | |
295 | * 2. Check whether the ino_t is in our store. | |
296 | * a. if not, fork a child in qpid's ns to send us | |
297 | * ucred.pid = 1, and read the initpid. Cache | |
298 | * initpid and creation time for /proc/initpid | |
299 | * in a new store entry. | |
300 | * b. if so, verify that /proc/initpid still matches | |
301 | * what we have saved. If not, clear the store | |
302 | * entry and go back to a. If so, return the | |
303 | * cached initpid. | |
304 | */ | |
305 | struct pidns_init_store { | |
306 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
307 | pid_t initpid; // the pid of nit in that ns | |
308 | long int ctime; // the time at which /proc/$initpid was created | |
309 | struct pidns_init_store *next; | |
310 | long int lastcheck; | |
311 | }; | |
312 | ||
313 | /* lol - look at how they are allocated in the kernel */ | |
314 | #define PIDNS_HASH_SIZE 4096 | |
315 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
316 | ||
317 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
318 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
319 | static void lock_mutex(pthread_mutex_t *l) | |
320 | { | |
321 | int ret; | |
322 | ||
323 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
b8defc3d | 324 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
325 | exit(1); |
326 | } | |
327 | } | |
328 | ||
29a73c2f CB |
329 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
330 | * Number of hierarchies mounted. */ | |
331 | static int num_hierarchies; | |
332 | ||
333 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
334 | * Hierachies mounted {cpuset, blkio, ...}: | |
335 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
336 | static char **hierarchies; | |
337 | ||
338 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
339 | * Open file descriptors: | |
340 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
341 | * private mount namespace. | |
342 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
343 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
344 | * mounts and respective files in the private namespace even when located in | |
345 | * another namespace using the *at() family of functions | |
346 | * {openat(), fchownat(), ...}. */ | |
347 | static int *fd_hierarchies; | |
a257a8ee | 348 | static int cgroup_mount_ns_fd = -1; |
29a73c2f | 349 | |
237e200e SH |
350 | static void unlock_mutex(pthread_mutex_t *l) |
351 | { | |
352 | int ret; | |
353 | ||
354 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
b8defc3d | 355 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
356 | exit(1); |
357 | } | |
358 | } | |
359 | ||
360 | static void store_lock(void) | |
361 | { | |
362 | lock_mutex(&pidns_store_mutex); | |
363 | } | |
364 | ||
365 | static void store_unlock(void) | |
366 | { | |
367 | unlock_mutex(&pidns_store_mutex); | |
368 | } | |
369 | ||
370 | /* Must be called under store_lock */ | |
371 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
372 | { | |
373 | struct stat initsb; | |
374 | char fnam[100]; | |
375 | ||
376 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
377 | if (stat(fnam, &initsb) < 0) | |
378 | return false; | |
7dd6560a CB |
379 | |
380 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
381 | initsb.st_ctime, e->initpid); | |
382 | ||
237e200e SH |
383 | if (e->ctime != initsb.st_ctime) |
384 | return false; | |
385 | return true; | |
386 | } | |
387 | ||
388 | /* Must be called under store_lock */ | |
389 | static void remove_initpid(struct pidns_init_store *e) | |
390 | { | |
391 | struct pidns_init_store *tmp; | |
392 | int h; | |
393 | ||
7dd6560a CB |
394 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
395 | ||
237e200e SH |
396 | h = HASH(e->ino); |
397 | if (pidns_hash_table[h] == e) { | |
398 | pidns_hash_table[h] = e->next; | |
399 | free(e); | |
400 | return; | |
401 | } | |
402 | ||
403 | tmp = pidns_hash_table[h]; | |
404 | while (tmp) { | |
405 | if (tmp->next == e) { | |
406 | tmp->next = e->next; | |
407 | free(e); | |
408 | return; | |
409 | } | |
410 | tmp = tmp->next; | |
411 | } | |
412 | } | |
413 | ||
414 | #define PURGE_SECS 5 | |
415 | /* Must be called under store_lock */ | |
416 | static void prune_initpid_store(void) | |
417 | { | |
418 | static long int last_prune = 0; | |
419 | struct pidns_init_store *e, *prev, *delme; | |
420 | long int now, threshold; | |
421 | int i; | |
422 | ||
423 | if (!last_prune) { | |
424 | last_prune = time(NULL); | |
425 | return; | |
426 | } | |
427 | now = time(NULL); | |
428 | if (now < last_prune + PURGE_SECS) | |
429 | return; | |
7dd6560a CB |
430 | |
431 | lxcfs_debug("%s\n", "Pruning."); | |
432 | ||
237e200e SH |
433 | last_prune = now; |
434 | threshold = now - 2 * PURGE_SECS; | |
435 | ||
436 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
437 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
438 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
439 | |
440 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
441 | ||
237e200e SH |
442 | delme = e; |
443 | if (prev) | |
444 | prev->next = e->next; | |
445 | else | |
446 | pidns_hash_table[i] = e->next; | |
447 | e = e->next; | |
448 | free(delme); | |
449 | } else { | |
450 | prev = e; | |
451 | e = e->next; | |
452 | } | |
453 | } | |
454 | } | |
455 | } | |
456 | ||
457 | /* Must be called under store_lock */ | |
458 | static void save_initpid(struct stat *sb, pid_t pid) | |
459 | { | |
460 | struct pidns_init_store *e; | |
461 | char fpath[100]; | |
462 | struct stat procsb; | |
463 | int h; | |
464 | ||
7dd6560a CB |
465 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
466 | ||
237e200e SH |
467 | snprintf(fpath, 100, "/proc/%d", pid); |
468 | if (stat(fpath, &procsb) < 0) | |
469 | return; | |
470 | do { | |
471 | e = malloc(sizeof(*e)); | |
472 | } while (!e); | |
473 | e->ino = sb->st_ino; | |
474 | e->initpid = pid; | |
475 | e->ctime = procsb.st_ctime; | |
476 | h = HASH(e->ino); | |
477 | e->next = pidns_hash_table[h]; | |
478 | e->lastcheck = time(NULL); | |
479 | pidns_hash_table[h] = e; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
484 | * entry for the inode number and creation time. Verify that the init pid | |
485 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
486 | * otherwise. | |
487 | * Must be called under store_lock | |
488 | */ | |
489 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
490 | { | |
491 | int h = HASH(sb->st_ino); | |
492 | struct pidns_init_store *e = pidns_hash_table[h]; | |
493 | ||
494 | while (e) { | |
495 | if (e->ino == sb->st_ino) { | |
496 | if (initpid_still_valid(e, sb)) { | |
497 | e->lastcheck = time(NULL); | |
498 | return e; | |
499 | } | |
500 | remove_initpid(e); | |
501 | return NULL; | |
502 | } | |
503 | e = e->next; | |
504 | } | |
505 | ||
506 | return NULL; | |
507 | } | |
508 | ||
0f657ce3 | 509 | static int is_dir(const char *path, int fd) |
237e200e SH |
510 | { |
511 | struct stat statbuf; | |
0f657ce3 | 512 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
513 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
514 | return 1; | |
515 | return 0; | |
516 | } | |
517 | ||
518 | static char *must_copy_string(const char *str) | |
519 | { | |
520 | char *dup = NULL; | |
521 | if (!str) | |
522 | return NULL; | |
523 | do { | |
524 | dup = strdup(str); | |
525 | } while (!dup); | |
526 | ||
527 | return dup; | |
528 | } | |
529 | ||
530 | static inline void drop_trailing_newlines(char *s) | |
531 | { | |
532 | int l; | |
533 | ||
534 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
535 | s[l-1] = '\0'; | |
536 | } | |
537 | ||
538 | #define BATCH_SIZE 50 | |
539 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
540 | { | |
541 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
542 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
543 | ||
544 | if (!*mem || newbatches > oldbatches) { | |
545 | char *tmp; | |
546 | do { | |
547 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
548 | } while (!tmp); | |
549 | *mem = tmp; | |
550 | } | |
551 | } | |
552 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
553 | { | |
554 | size_t newlen = *len + linelen; | |
555 | dorealloc(contents, *len, newlen + 1); | |
556 | memcpy(*contents + *len, line, linelen+1); | |
557 | *len = newlen; | |
558 | } | |
559 | ||
60f2ae53 | 560 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
561 | { |
562 | char *line = NULL; | |
563 | char *contents = NULL; | |
60f2ae53 | 564 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
565 | size_t len = 0, fulllen = 0; |
566 | ssize_t linelen; | |
567 | ||
568 | if (!f) | |
569 | return NULL; | |
570 | ||
571 | while ((linelen = getline(&line, &len, f)) != -1) { | |
572 | append_line(&contents, &fulllen, line, linelen); | |
573 | } | |
574 | fclose(f); | |
575 | ||
576 | if (contents) | |
577 | drop_trailing_newlines(contents); | |
578 | free(line); | |
579 | return contents; | |
580 | } | |
581 | ||
ba59ea09 | 582 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
583 | { |
584 | FILE *f; | |
585 | size_t len, ret; | |
586 | ||
ba59ea09 | 587 | if (!(f = fdopen(fd, "w"))) |
237e200e SH |
588 | return false; |
589 | len = strlen(string); | |
590 | ret = fwrite(string, 1, len, f); | |
591 | if (ret != len) { | |
b8defc3d | 592 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); |
237e200e SH |
593 | fclose(f); |
594 | return false; | |
595 | } | |
596 | if (fclose(f) < 0) { | |
b8defc3d | 597 | lxcfs_error("Error writing to file: %s\n", strerror(errno)); |
237e200e SH |
598 | return false; |
599 | } | |
600 | return true; | |
601 | } | |
602 | ||
237e200e SH |
603 | struct cgfs_files { |
604 | char *name; | |
605 | uint32_t uid, gid; | |
606 | uint32_t mode; | |
607 | }; | |
608 | ||
0619767c | 609 | #define ALLOC_NUM 20 |
237e200e SH |
610 | static bool store_hierarchy(char *stridx, char *h) |
611 | { | |
0619767c SH |
612 | if (num_hierarchies % ALLOC_NUM == 0) { |
613 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
614 | n *= ALLOC_NUM; | |
615 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c | 616 | if (!tmp) { |
b8defc3d | 617 | lxcfs_error("%s\n", strerror(errno)); |
0619767c SH |
618 | exit(1); |
619 | } | |
237e200e | 620 | hierarchies = tmp; |
237e200e | 621 | } |
f676eb79 | 622 | |
0619767c | 623 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
624 | return true; |
625 | } | |
626 | ||
627 | static void print_subsystems(void) | |
628 | { | |
629 | int i; | |
630 | ||
a257a8ee | 631 | fprintf(stderr, "mount namespace: %d\n", cgroup_mount_ns_fd); |
cc97d34c | 632 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
633 | for (i = 0; i < num_hierarchies; i++) { |
634 | if (hierarchies[i]) | |
b8defc3d CB |
635 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, |
636 | fd_hierarchies[i], hierarchies[i]); | |
237e200e SH |
637 | } |
638 | } | |
639 | ||
640 | static bool in_comma_list(const char *needle, const char *haystack) | |
641 | { | |
642 | const char *s = haystack, *e; | |
643 | size_t nlen = strlen(needle); | |
644 | ||
06081b29 | 645 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
646 | if (nlen != e - s) { |
647 | s = e + 1; | |
648 | continue; | |
649 | } | |
650 | if (strncmp(needle, s, nlen) == 0) | |
651 | return true; | |
652 | s = e + 1; | |
653 | } | |
654 | if (strcmp(needle, s) == 0) | |
655 | return true; | |
656 | return false; | |
657 | } | |
658 | ||
659 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
660 | /* Return the mounted controller and store the corresponding open file descriptor |
661 | * referring to the controller mountpoint in the private lxcfs namespace in | |
662 | * @cfd. | |
663 | */ | |
664 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
665 | { |
666 | int i; | |
667 | ||
668 | for (i = 0; i < num_hierarchies; i++) { | |
669 | if (!hierarchies[i]) | |
670 | continue; | |
5dd3e6fd CB |
671 | if (strcmp(hierarchies[i], controller) == 0) { |
672 | *cfd = fd_hierarchies[i]; | |
237e200e | 673 | return hierarchies[i]; |
5dd3e6fd CB |
674 | } |
675 | if (in_comma_list(controller, hierarchies[i])) { | |
676 | *cfd = fd_hierarchies[i]; | |
237e200e | 677 | return hierarchies[i]; |
5dd3e6fd | 678 | } |
237e200e SH |
679 | } |
680 | ||
681 | return NULL; | |
682 | } | |
683 | ||
684 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
685 | const char *value) | |
686 | { | |
ba59ea09 | 687 | int ret, fd, cfd; |
237e200e | 688 | size_t len; |
f5a6d92e | 689 | char *fnam, *tmpc; |
237e200e | 690 | |
f5a6d92e | 691 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
692 | if (!tmpc) |
693 | return false; | |
f5a6d92e CB |
694 | |
695 | /* Make sure we pass a relative path to *at() family of functions. | |
696 | * . + /cgroup + / + file + \0 | |
697 | */ | |
ba59ea09 | 698 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 699 | fnam = alloca(len); |
ba59ea09 CB |
700 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
701 | if (ret < 0 || (size_t)ret >= len) | |
702 | return false; | |
703 | ||
704 | fd = openat(cfd, fnam, O_WRONLY); | |
705 | if (fd < 0) | |
706 | return false; | |
f676eb79 | 707 | |
ba59ea09 | 708 | return write_string(fnam, value, fd); |
237e200e SH |
709 | } |
710 | ||
711 | // Chown all the files in the cgroup directory. We do this when we create | |
712 | // a cgroup on behalf of a user. | |
f23fe717 | 713 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 714 | { |
f23fe717 | 715 | struct dirent *direntp; |
237e200e SH |
716 | char path[MAXPATHLEN]; |
717 | size_t len; | |
718 | DIR *d; | |
f23fe717 | 719 | int fd1, ret; |
237e200e SH |
720 | |
721 | len = strlen(dirname); | |
722 | if (len >= MAXPATHLEN) { | |
b8defc3d | 723 | lxcfs_error("Pathname too long: %s\n", dirname); |
237e200e SH |
724 | return; |
725 | } | |
726 | ||
f23fe717 CB |
727 | fd1 = openat(fd, dirname, O_DIRECTORY); |
728 | if (fd1 < 0) | |
729 | return; | |
730 | ||
731 | d = fdopendir(fd1); | |
237e200e | 732 | if (!d) { |
b8defc3d | 733 | lxcfs_error("Failed to open %s\n", dirname); |
237e200e SH |
734 | return; |
735 | } | |
736 | ||
f23fe717 | 737 | while ((direntp = readdir(d))) { |
237e200e SH |
738 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
739 | continue; | |
740 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
741 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 742 | lxcfs_error("Pathname too long under %s\n", dirname); |
237e200e SH |
743 | continue; |
744 | } | |
f23fe717 | 745 | if (fchownat(fd, path, uid, gid, 0) < 0) |
b8defc3d | 746 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); |
237e200e SH |
747 | } |
748 | closedir(d); | |
749 | } | |
750 | ||
751 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
752 | { | |
5dd3e6fd | 753 | int cfd; |
237e200e | 754 | size_t len; |
f5a6d92e | 755 | char *dirnam, *tmpc; |
237e200e | 756 | |
f5a6d92e | 757 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
758 | if (!tmpc) |
759 | return -EINVAL; | |
f5a6d92e CB |
760 | |
761 | /* Make sure we pass a relative path to *at() family of functions. | |
762 | * . + /cg + \0 | |
763 | */ | |
f23fe717 | 764 | len = strlen(cg) + 2; |
237e200e | 765 | dirnam = alloca(len); |
f23fe717 | 766 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 767 | |
f23fe717 | 768 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
769 | return -errno; |
770 | ||
771 | if (uid == 0 && gid == 0) | |
772 | return 0; | |
773 | ||
f23fe717 | 774 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
775 | return -errno; |
776 | ||
f23fe717 | 777 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
778 | |
779 | return 0; | |
780 | } | |
781 | ||
7213ec5c | 782 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) |
237e200e | 783 | { |
b7672ded | 784 | struct dirent *direntp; |
237e200e SH |
785 | DIR *dir; |
786 | bool ret = false; | |
787 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
788 | int dupfd; |
789 | ||
790 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
791 | if (dupfd < 0) | |
792 | return false; | |
237e200e | 793 | |
b7672ded | 794 | dir = fdopendir(dupfd); |
237e200e | 795 | if (!dir) { |
7dd6560a | 796 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
7213ec5c | 797 | close(dupfd); |
237e200e SH |
798 | return false; |
799 | } | |
800 | ||
b7672ded | 801 | while ((direntp = readdir(dir))) { |
237e200e SH |
802 | struct stat mystat; |
803 | int rc; | |
804 | ||
237e200e SH |
805 | if (!strcmp(direntp->d_name, ".") || |
806 | !strcmp(direntp->d_name, "..")) | |
807 | continue; | |
808 | ||
809 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
810 | if (rc < 0 || rc >= MAXPATHLEN) { | |
b8defc3d | 811 | lxcfs_error("%s\n", "Pathname too long."); |
237e200e SH |
812 | continue; |
813 | } | |
814 | ||
2e81a5e3 CB |
815 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
816 | if (rc) { | |
7dd6560a | 817 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
818 | continue; |
819 | } | |
7dd6560a | 820 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 821 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 822 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
823 | } |
824 | ||
825 | ret = true; | |
826 | if (closedir(dir) < 0) { | |
b8defc3d | 827 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); |
237e200e SH |
828 | ret = false; |
829 | } | |
830 | ||
2e81a5e3 | 831 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 832 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
833 | ret = false; |
834 | } | |
7213ec5c CB |
835 | |
836 | close(dupfd); | |
237e200e SH |
837 | |
838 | return ret; | |
839 | } | |
840 | ||
841 | bool cgfs_remove(const char *controller, const char *cg) | |
842 | { | |
b7672ded | 843 | int fd, cfd; |
237e200e | 844 | size_t len; |
f5a6d92e | 845 | char *dirnam, *tmpc; |
7213ec5c | 846 | bool bret; |
237e200e | 847 | |
f5a6d92e | 848 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
849 | if (!tmpc) |
850 | return false; | |
f5a6d92e CB |
851 | |
852 | /* Make sure we pass a relative path to *at() family of functions. | |
853 | * . + /cg + \0 | |
854 | */ | |
b7672ded | 855 | len = strlen(cg) + 2; |
237e200e | 856 | dirnam = alloca(len); |
b7672ded CB |
857 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
858 | ||
859 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
860 | if (fd < 0) | |
861 | return false; | |
862 | ||
7213ec5c CB |
863 | bret = recursive_rmdir(dirnam, fd, cfd); |
864 | close(fd); | |
865 | return bret; | |
237e200e SH |
866 | } |
867 | ||
868 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
869 | { | |
5dd3e6fd | 870 | int cfd; |
237e200e | 871 | size_t len; |
f5a6d92e | 872 | char *pathname, *tmpc; |
237e200e | 873 | |
f5a6d92e | 874 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
875 | if (!tmpc) |
876 | return false; | |
f5a6d92e CB |
877 | |
878 | /* Make sure we pass a relative path to *at() family of functions. | |
879 | * . + /file + \0 | |
880 | */ | |
534690b4 | 881 | len = strlen(file) + 2; |
237e200e | 882 | pathname = alloca(len); |
534690b4 CB |
883 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
884 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
885 | return false; |
886 | return true; | |
887 | } | |
888 | ||
0f657ce3 | 889 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
890 | { |
891 | size_t len; | |
892 | char *fname; | |
893 | ||
894 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
895 | fname = alloca(len); | |
896 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 897 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
898 | return -errno; |
899 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 900 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
901 | return -errno; |
902 | return 0; | |
903 | } | |
904 | ||
905 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
906 | { | |
5dd3e6fd | 907 | int cfd; |
237e200e | 908 | size_t len; |
f5a6d92e | 909 | char *pathname, *tmpc; |
237e200e | 910 | |
f5a6d92e | 911 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
912 | if (!tmpc) |
913 | return -EINVAL; | |
f5a6d92e CB |
914 | |
915 | /* Make sure we pass a relative path to *at() family of functions. | |
916 | * . + /file + \0 | |
917 | */ | |
0f657ce3 | 918 | len = strlen(file) + 2; |
237e200e | 919 | pathname = alloca(len); |
0f657ce3 CB |
920 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
921 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
922 | return -errno; |
923 | ||
0f657ce3 | 924 | if (is_dir(pathname, cfd)) |
237e200e | 925 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 926 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
927 | |
928 | return 0; | |
929 | } | |
930 | ||
931 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
932 | { | |
3ffd08ee | 933 | int fd, cfd; |
237e200e | 934 | size_t len; |
f5a6d92e | 935 | char *pathname, *tmpc; |
237e200e | 936 | |
f5a6d92e | 937 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
938 | if (!tmpc) |
939 | return NULL; | |
f5a6d92e CB |
940 | |
941 | /* Make sure we pass a relative path to *at() family of functions. | |
942 | * . + /cgroup + / "cgroup.procs" + \0 | |
943 | */ | |
3ffd08ee | 944 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 945 | pathname = alloca(len); |
3ffd08ee CB |
946 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
947 | ||
948 | fd = openat(cfd, pathname, O_WRONLY); | |
949 | if (fd < 0) | |
950 | return NULL; | |
951 | ||
952 | return fdopen(fd, "w"); | |
237e200e SH |
953 | } |
954 | ||
f366da65 WB |
955 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
956 | void ***list, size_t typesize, | |
957 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 958 | { |
4ea38a4c | 959 | int cfd, fd, ret; |
237e200e | 960 | size_t len; |
4ea38a4c | 961 | char *cg, *tmpc; |
237e200e | 962 | char pathname[MAXPATHLEN]; |
f366da65 | 963 | size_t sz = 0, asz = 0; |
4ea38a4c | 964 | struct dirent *dirent; |
237e200e | 965 | DIR *dir; |
237e200e | 966 | |
4ea38a4c | 967 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 968 | *list = NULL; |
237e200e | 969 | if (!tmpc) |
e97c834b | 970 | return false; |
237e200e | 971 | |
f5a6d92e | 972 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
973 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
974 | cg = alloca(len); | |
975 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
976 | if (ret < 0 || (size_t)ret >= len) { | |
b8defc3d | 977 | lxcfs_error("Pathname too long under %s\n", cgroup); |
4ea38a4c CB |
978 | return false; |
979 | } | |
237e200e | 980 | |
4ea38a4c CB |
981 | fd = openat(cfd, cg, O_DIRECTORY); |
982 | if (fd < 0) | |
983 | return false; | |
984 | ||
985 | dir = fdopendir(fd); | |
237e200e SH |
986 | if (!dir) |
987 | return false; | |
988 | ||
4ea38a4c | 989 | while ((dirent = readdir(dir))) { |
237e200e | 990 | struct stat mystat; |
237e200e | 991 | |
4ea38a4c CB |
992 | if (!strcmp(dirent->d_name, ".") || |
993 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
994 | continue; |
995 | ||
4ea38a4c CB |
996 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
997 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 998 | lxcfs_error("Pathname too long under %s\n", cg); |
237e200e SH |
999 | continue; |
1000 | } | |
1001 | ||
4ea38a4c | 1002 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e | 1003 | if (ret) { |
b8defc3d | 1004 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); |
237e200e SH |
1005 | continue; |
1006 | } | |
f366da65 WB |
1007 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
1008 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
1009 | continue; |
1010 | ||
1011 | if (sz+2 >= asz) { | |
f366da65 | 1012 | void **tmp; |
237e200e SH |
1013 | asz += BATCH_SIZE; |
1014 | do { | |
f366da65 | 1015 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
1016 | } while (!tmp); |
1017 | *list = tmp; | |
1018 | } | |
4ea38a4c | 1019 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
1020 | (*list)[sz+1] = NULL; |
1021 | sz++; | |
1022 | } | |
1023 | if (closedir(dir) < 0) { | |
b8defc3d | 1024 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); |
237e200e SH |
1025 | return false; |
1026 | } | |
1027 | return true; | |
1028 | } | |
1029 | ||
f366da65 WB |
1030 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
1031 | { | |
1032 | char *dup; | |
1033 | do { | |
1034 | dup = strdup(dir_entry); | |
1035 | } while (!dup); | |
1036 | return dup; | |
1037 | } | |
1038 | ||
1039 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
1040 | { | |
1041 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
1042 | } | |
1043 | ||
237e200e SH |
1044 | void free_key(struct cgfs_files *k) |
1045 | { | |
1046 | if (!k) | |
1047 | return; | |
1048 | free(k->name); | |
1049 | free(k); | |
1050 | } | |
1051 | ||
1052 | void free_keys(struct cgfs_files **keys) | |
1053 | { | |
1054 | int i; | |
1055 | ||
1056 | if (!keys) | |
1057 | return; | |
1058 | for (i = 0; keys[i]; i++) { | |
1059 | free_key(keys[i]); | |
1060 | } | |
1061 | free(keys); | |
1062 | } | |
1063 | ||
1064 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
1065 | { | |
60f2ae53 | 1066 | int ret, fd, cfd; |
237e200e | 1067 | size_t len; |
f5a6d92e | 1068 | char *fnam, *tmpc; |
237e200e | 1069 | |
f5a6d92e | 1070 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1071 | if (!tmpc) |
1072 | return false; | |
f5a6d92e CB |
1073 | |
1074 | /* Make sure we pass a relative path to *at() family of functions. | |
1075 | * . + /cgroup + / + file + \0 | |
1076 | */ | |
60f2ae53 | 1077 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 1078 | fnam = alloca(len); |
60f2ae53 CB |
1079 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
1080 | if (ret < 0 || (size_t)ret >= len) | |
234a820c | 1081 | return false; |
60f2ae53 CB |
1082 | |
1083 | fd = openat(cfd, fnam, O_RDONLY); | |
1084 | if (fd < 0) | |
234a820c | 1085 | return false; |
237e200e | 1086 | |
60f2ae53 | 1087 | *value = slurp_file(fnam, fd); |
237e200e SH |
1088 | return *value != NULL; |
1089 | } | |
1090 | ||
1091 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
1092 | { | |
4ea38a4c | 1093 | int ret, cfd; |
237e200e | 1094 | size_t len; |
f5a6d92e | 1095 | char *fnam, *tmpc; |
237e200e SH |
1096 | struct stat sb; |
1097 | struct cgfs_files *newkey; | |
237e200e | 1098 | |
f5a6d92e | 1099 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1100 | if (!tmpc) |
1101 | return false; | |
1102 | ||
1103 | if (file && *file == '/') | |
1104 | file++; | |
1105 | ||
06081b29 | 1106 | if (file && strchr(file, '/')) |
237e200e SH |
1107 | return NULL; |
1108 | ||
f5a6d92e CB |
1109 | /* Make sure we pass a relative path to *at() family of functions. |
1110 | * . + /cgroup + / + file + \0 | |
1111 | */ | |
4ea38a4c | 1112 | len = strlen(cgroup) + 3; |
237e200e SH |
1113 | if (file) |
1114 | len += strlen(file) + 1; | |
1115 | fnam = alloca(len); | |
4ea38a4c CB |
1116 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
1117 | file ? "/" : "", file ? file : ""); | |
237e200e | 1118 | |
4ea38a4c | 1119 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1120 | if (ret < 0) |
1121 | return NULL; | |
1122 | ||
1123 | do { | |
1124 | newkey = malloc(sizeof(struct cgfs_files)); | |
1125 | } while (!newkey); | |
1126 | if (file) | |
1127 | newkey->name = must_copy_string(file); | |
06081b29 CB |
1128 | else if (strrchr(cgroup, '/')) |
1129 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
1130 | else |
1131 | newkey->name = must_copy_string(cgroup); | |
1132 | newkey->uid = sb.st_uid; | |
1133 | newkey->gid = sb.st_gid; | |
1134 | newkey->mode = sb.st_mode; | |
1135 | ||
1136 | return newkey; | |
1137 | } | |
1138 | ||
f366da65 | 1139 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 1140 | { |
f366da65 WB |
1141 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
1142 | if (!entry) { | |
b8defc3d CB |
1143 | lxcfs_error("Error getting files under %s:%s\n", controller, |
1144 | cgroup); | |
237e200e | 1145 | } |
f366da65 WB |
1146 | return entry; |
1147 | } | |
1148 | ||
1149 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
1150 | { | |
1151 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
1152 | } |
1153 | ||
1154 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
1155 | { |
1156 | int cfd; | |
1157 | size_t len; | |
f5a6d92e | 1158 | char *fnam, *tmpc; |
237e200e SH |
1159 | int ret; |
1160 | struct stat sb; | |
1161 | ||
f5a6d92e | 1162 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1163 | if (!tmpc) |
1164 | return false; | |
f5a6d92e CB |
1165 | |
1166 | /* Make sure we pass a relative path to *at() family of functions. | |
1167 | * . + /cgroup + / + f + \0 | |
1168 | */ | |
d04232f2 | 1169 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 1170 | fnam = alloca(len); |
d04232f2 CB |
1171 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
1172 | if (ret < 0 || (size_t)ret >= len) | |
1173 | return false; | |
237e200e | 1174 | |
d04232f2 | 1175 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1176 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
1177 | return false; | |
f5a6d92e | 1178 | |
237e200e SH |
1179 | return true; |
1180 | } | |
1181 | ||
1182 | #define SEND_CREDS_OK 0 | |
1183 | #define SEND_CREDS_NOTSK 1 | |
1184 | #define SEND_CREDS_FAIL 2 | |
1185 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
1186 | static int wait_for_pid(pid_t pid); | |
1187 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 1188 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
1189 | |
1190 | /* | |
b10bdd6c | 1191 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
1192 | * over a unix sock so we can read the task's reaper's pid in our |
1193 | * namespace | |
b10bdd6c FG |
1194 | * |
1195 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
1196 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
1197 | * the pidns and the parent pid outside are identical. Using clone prevents | |
1198 | * this issue. | |
237e200e SH |
1199 | */ |
1200 | static void write_task_init_pid_exit(int sock, pid_t target) | |
1201 | { | |
237e200e SH |
1202 | char fnam[100]; |
1203 | pid_t pid; | |
237e200e | 1204 | int fd, ret; |
b10bdd6c FG |
1205 | size_t stack_size = sysconf(_SC_PAGESIZE); |
1206 | void *stack = alloca(stack_size); | |
237e200e SH |
1207 | |
1208 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1209 | if (ret < 0 || ret >= sizeof(fnam)) | |
1210 | _exit(1); | |
1211 | ||
1212 | fd = open(fnam, O_RDONLY); | |
1213 | if (fd < 0) { | |
1214 | perror("write_task_init_pid_exit open of ns/pid"); | |
1215 | _exit(1); | |
1216 | } | |
1217 | if (setns(fd, 0)) { | |
1218 | perror("write_task_init_pid_exit setns 1"); | |
1219 | close(fd); | |
1220 | _exit(1); | |
1221 | } | |
b10bdd6c | 1222 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1223 | if (pid < 0) |
1224 | _exit(1); | |
1225 | if (pid != 0) { | |
1226 | if (!wait_for_pid(pid)) | |
1227 | _exit(1); | |
1228 | _exit(0); | |
1229 | } | |
b10bdd6c FG |
1230 | } |
1231 | ||
1232 | static int send_creds_clone_wrapper(void *arg) { | |
1233 | struct ucred cred; | |
1234 | char v; | |
1235 | int sock = *(int *)arg; | |
237e200e SH |
1236 | |
1237 | /* we are the child */ | |
1238 | cred.uid = 0; | |
1239 | cred.gid = 0; | |
1240 | cred.pid = 1; | |
1241 | v = '1'; | |
1242 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1243 | return 1; |
1244 | return 0; | |
237e200e SH |
1245 | } |
1246 | ||
1247 | static pid_t get_init_pid_for_task(pid_t task) | |
1248 | { | |
1249 | int sock[2]; | |
1250 | pid_t pid; | |
1251 | pid_t ret = -1; | |
1252 | char v = '0'; | |
1253 | struct ucred cred; | |
1254 | ||
1255 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1256 | perror("socketpair"); | |
1257 | return -1; | |
1258 | } | |
1259 | ||
1260 | pid = fork(); | |
1261 | if (pid < 0) | |
1262 | goto out; | |
1263 | if (!pid) { | |
1264 | close(sock[1]); | |
1265 | write_task_init_pid_exit(sock[0], task); | |
1266 | _exit(0); | |
1267 | } | |
1268 | ||
1269 | if (!recv_creds(sock[1], &cred, &v)) | |
1270 | goto out; | |
1271 | ret = cred.pid; | |
1272 | ||
1273 | out: | |
1274 | close(sock[0]); | |
1275 | close(sock[1]); | |
1276 | if (pid > 0) | |
1277 | wait_for_pid(pid); | |
1278 | return ret; | |
1279 | } | |
1280 | ||
1281 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1282 | { | |
1283 | pid_t answer = 0; | |
1284 | struct stat sb; | |
1285 | struct pidns_init_store *e; | |
1286 | char fnam[100]; | |
1287 | ||
1288 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1289 | store_lock(); | |
1290 | if (stat(fnam, &sb) < 0) | |
1291 | goto out; | |
1292 | e = lookup_verify_initpid(&sb); | |
1293 | if (e) { | |
1294 | answer = e->initpid; | |
1295 | goto out; | |
1296 | } | |
1297 | answer = get_init_pid_for_task(qpid); | |
1298 | if (answer > 0) | |
1299 | save_initpid(&sb, answer); | |
1300 | ||
1301 | out: | |
1302 | /* we prune at end in case we are returning | |
1303 | * the value we were about to return */ | |
1304 | prune_initpid_store(); | |
1305 | store_unlock(); | |
1306 | return answer; | |
1307 | } | |
1308 | ||
1309 | static int wait_for_pid(pid_t pid) | |
1310 | { | |
1311 | int status, ret; | |
1312 | ||
1313 | if (pid <= 0) | |
1314 | return -1; | |
1315 | ||
1316 | again: | |
1317 | ret = waitpid(pid, &status, 0); | |
1318 | if (ret == -1) { | |
1319 | if (errno == EINTR) | |
1320 | goto again; | |
1321 | return -1; | |
1322 | } | |
1323 | if (ret != pid) | |
1324 | goto again; | |
1325 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1326 | return -1; | |
1327 | return 0; | |
1328 | } | |
1329 | ||
1330 | ||
1331 | /* | |
1332 | * append pid to *src. | |
1333 | * src: a pointer to a char* in which ot append the pid. | |
1334 | * sz: the number of characters printed so far, minus trailing \0. | |
1335 | * asz: the allocated size so far | |
1336 | * pid: the pid to append | |
1337 | */ | |
1338 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1339 | { | |
1340 | char tmp[30]; | |
1341 | ||
1342 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1343 | ||
1344 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1345 | char *tmp; | |
1346 | do { | |
1347 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1348 | } while (!tmp); | |
1349 | *src = tmp; | |
1350 | *asz += BUF_RESERVE_SIZE; | |
1351 | } | |
bbfd0e33 | 1352 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1353 | *sz += tmplen; |
237e200e SH |
1354 | } |
1355 | ||
1356 | /* | |
1357 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1358 | * valid in the caller's namespace, return the id mapped into | |
1359 | * pid's namespace. | |
1360 | * Returns the mapped id, or -1 on error. | |
1361 | */ | |
1362 | unsigned int | |
1363 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1364 | { | |
1365 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1366 | hostuid, // base id for a range in the caller's namespace | |
1367 | count; // number of ids in this range | |
1368 | char line[400]; | |
1369 | int ret; | |
1370 | ||
1371 | fseek(idfile, 0L, SEEK_SET); | |
1372 | while (fgets(line, 400, idfile)) { | |
1373 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1374 | if (ret != 3) | |
1375 | continue; | |
1376 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1377 | /* | |
1378 | * uids wrapped around - unexpected as this is a procfile, | |
1379 | * so just bail. | |
1380 | */ | |
b8defc3d | 1381 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", |
237e200e SH |
1382 | nsuid, hostuid, count, line); |
1383 | return -1; | |
1384 | } | |
1385 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1386 | /* | |
1387 | * now since hostuid <= in_id < hostuid+count, and | |
1388 | * hostuid+count and nsuid+count do not wrap around, | |
1389 | * we know that nsuid+(in_id-hostuid) which must be | |
1390 | * less that nsuid+(count) must not wrap around | |
1391 | */ | |
1392 | return (in_id - hostuid) + nsuid; | |
1393 | } | |
1394 | } | |
1395 | ||
1396 | // no answer found | |
1397 | return -1; | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * for is_privileged_over, | |
1402 | * specify whether we require the calling uid to be root in his | |
1403 | * namespace | |
1404 | */ | |
1405 | #define NS_ROOT_REQD true | |
1406 | #define NS_ROOT_OPT false | |
1407 | ||
1408 | #define PROCLEN 100 | |
1409 | ||
1410 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1411 | { | |
1412 | char fpath[PROCLEN]; | |
1413 | int ret; | |
1414 | bool answer = false; | |
1415 | uid_t nsuid; | |
1416 | ||
1417 | if (victim == -1 || uid == -1) | |
1418 | return false; | |
1419 | ||
1420 | /* | |
1421 | * If the request is one not requiring root in the namespace, | |
1422 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1423 | * access to files owned by uid 1000 | |
1424 | */ | |
1425 | if (!req_ns_root && uid == victim) | |
1426 | return true; | |
1427 | ||
1428 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1429 | if (ret < 0 || ret >= PROCLEN) | |
1430 | return false; | |
1431 | FILE *f = fopen(fpath, "r"); | |
1432 | if (!f) | |
1433 | return false; | |
1434 | ||
1435 | /* if caller's not root in his namespace, reject */ | |
1436 | nsuid = convert_id_to_ns(f, uid); | |
1437 | if (nsuid) | |
1438 | goto out; | |
1439 | ||
1440 | /* | |
1441 | * If victim is not mapped into caller's ns, reject. | |
1442 | * XXX I'm not sure this check is needed given that fuse | |
1443 | * will be sending requests where the vfs has converted | |
1444 | */ | |
1445 | nsuid = convert_id_to_ns(f, victim); | |
1446 | if (nsuid == -1) | |
1447 | goto out; | |
1448 | ||
1449 | answer = true; | |
1450 | ||
1451 | out: | |
1452 | fclose(f); | |
1453 | return answer; | |
1454 | } | |
1455 | ||
1456 | static bool perms_include(int fmode, mode_t req_mode) | |
1457 | { | |
1458 | mode_t r; | |
1459 | ||
1460 | switch (req_mode & O_ACCMODE) { | |
1461 | case O_RDONLY: | |
1462 | r = S_IROTH; | |
1463 | break; | |
1464 | case O_WRONLY: | |
1465 | r = S_IWOTH; | |
1466 | break; | |
1467 | case O_RDWR: | |
1468 | r = S_IROTH | S_IWOTH; | |
1469 | break; | |
1470 | default: | |
1471 | return false; | |
1472 | } | |
1473 | return ((fmode & r) == r); | |
1474 | } | |
1475 | ||
1476 | ||
1477 | /* | |
1478 | * taskcg is a/b/c | |
1479 | * querycg is /a/b/c/d/e | |
1480 | * we return 'd' | |
1481 | */ | |
1482 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1483 | { | |
1484 | char *start, *end; | |
1485 | ||
1486 | if (strlen(taskcg) <= strlen(querycg)) { | |
b8defc3d | 1487 | lxcfs_error("%s\n", "I was fed bad input."); |
237e200e SH |
1488 | return NULL; |
1489 | } | |
1490 | ||
06081b29 | 1491 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1492 | start = strdup(taskcg + 1); |
1493 | else | |
1494 | start = strdup(taskcg + strlen(querycg) + 1); | |
1495 | if (!start) | |
1496 | return NULL; | |
1497 | end = strchr(start, '/'); | |
1498 | if (end) | |
1499 | *end = '\0'; | |
1500 | return start; | |
1501 | } | |
1502 | ||
1503 | static void stripnewline(char *x) | |
1504 | { | |
1505 | size_t l = strlen(x); | |
1506 | if (l && x[l-1] == '\n') | |
1507 | x[l-1] = '\0'; | |
1508 | } | |
1509 | ||
1510 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1511 | { | |
5dd3e6fd | 1512 | int cfd; |
237e200e SH |
1513 | char fnam[PROCLEN]; |
1514 | FILE *f; | |
1515 | char *answer = NULL; | |
1516 | char *line = NULL; | |
1517 | size_t len = 0; | |
1518 | int ret; | |
5dd3e6fd | 1519 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1520 | if (!h) |
1521 | return NULL; | |
1522 | ||
1523 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1524 | if (ret < 0 || ret >= PROCLEN) | |
1525 | return NULL; | |
1526 | if (!(f = fopen(fnam, "r"))) | |
1527 | return NULL; | |
1528 | ||
1529 | while (getline(&line, &len, f) != -1) { | |
1530 | char *c1, *c2; | |
1531 | if (!line[0]) | |
1532 | continue; | |
1533 | c1 = strchr(line, ':'); | |
1534 | if (!c1) | |
1535 | goto out; | |
1536 | c1++; | |
1537 | c2 = strchr(c1, ':'); | |
1538 | if (!c2) | |
1539 | goto out; | |
1540 | *c2 = '\0'; | |
1541 | if (strcmp(c1, h) != 0) | |
1542 | continue; | |
1543 | c2++; | |
1544 | stripnewline(c2); | |
1545 | do { | |
1546 | answer = strdup(c2); | |
1547 | } while (!answer); | |
1548 | break; | |
1549 | } | |
1550 | ||
1551 | out: | |
1552 | fclose(f); | |
1553 | free(line); | |
1554 | return answer; | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * check whether a fuse context may access a cgroup dir or file | |
1559 | * | |
1560 | * If file is not null, it is a cgroup file to check under cg. | |
1561 | * If file is null, then we are checking perms on cg itself. | |
1562 | * | |
1563 | * For files we can check the mode of the list_keys result. | |
1564 | * For cgroups, we must make assumptions based on the files under the | |
1565 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1566 | * yet. | |
1567 | */ | |
1568 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1569 | { | |
1570 | struct cgfs_files *k = NULL; | |
1571 | bool ret = false; | |
1572 | ||
1573 | k = cgfs_get_key(contrl, cg, file); | |
1574 | if (!k) | |
1575 | return false; | |
1576 | ||
1577 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1578 | if (perms_include(k->mode >> 6, mode)) { | |
1579 | ret = true; | |
1580 | goto out; | |
1581 | } | |
1582 | } | |
1583 | if (fc->gid == k->gid) { | |
1584 | if (perms_include(k->mode >> 3, mode)) { | |
1585 | ret = true; | |
1586 | goto out; | |
1587 | } | |
1588 | } | |
1589 | ret = perms_include(k->mode, mode); | |
1590 | ||
1591 | out: | |
1592 | free_key(k); | |
1593 | return ret; | |
1594 | } | |
1595 | ||
1596 | #define INITSCOPE "/init.scope" | |
1597 | static void prune_init_slice(char *cg) | |
1598 | { | |
1599 | char *point; | |
1600 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1601 | ||
1602 | if (cg_len < initscope_len) | |
1603 | return; | |
1604 | ||
1605 | point = cg + cg_len - initscope_len; | |
1606 | if (strcmp(point, INITSCOPE) == 0) { | |
1607 | if (point == cg) | |
1608 | *(point+1) = '\0'; | |
1609 | else | |
1610 | *point = '\0'; | |
1611 | } | |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1616 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1617 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1618 | * to a string containing the next cgroup directory under cg, which must be | |
1619 | * freed by the caller. | |
1620 | */ | |
1621 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1622 | { | |
1623 | bool answer = false; | |
1624 | char *c2 = get_pid_cgroup(pid, contrl); | |
1625 | char *linecmp; | |
1626 | ||
1627 | if (!c2) | |
1628 | return false; | |
1629 | prune_init_slice(c2); | |
1630 | ||
1631 | /* | |
12c31268 CB |
1632 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1633 | * they pass in a cgroup without leading '/' | |
1634 | * | |
1635 | * The original line here was: | |
1636 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1637 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1638 | * Serge, do you know? | |
237e200e | 1639 | */ |
12c31268 CB |
1640 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1641 | linecmp = c2; | |
1642 | else | |
1643 | linecmp = c2 + 1; | |
237e200e SH |
1644 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1645 | if (nextcg) { | |
1646 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1647 | } | |
1648 | goto out; | |
1649 | } | |
1650 | answer = true; | |
1651 | ||
1652 | out: | |
1653 | free(c2); | |
1654 | return answer; | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1659 | */ | |
1660 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1661 | { | |
1662 | bool answer = false; | |
1663 | char *c2, *task_cg; | |
1664 | size_t target_len, task_len; | |
1665 | ||
f7bff426 | 1666 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1667 | return true; |
1668 | ||
1669 | c2 = get_pid_cgroup(pid, contrl); | |
1670 | if (!c2) | |
1671 | return false; | |
1672 | prune_init_slice(c2); | |
1673 | ||
1674 | task_cg = c2 + 1; | |
1675 | target_len = strlen(cg); | |
1676 | task_len = strlen(task_cg); | |
1677 | if (task_len == 0) { | |
1678 | /* Task is in the root cg, it can see everything. This case is | |
1679 | * not handled by the strmcps below, since they test for the | |
1680 | * last /, but that is the first / that we've chopped off | |
1681 | * above. | |
1682 | */ | |
1683 | answer = true; | |
1684 | goto out; | |
1685 | } | |
1686 | if (strcmp(cg, task_cg) == 0) { | |
1687 | answer = true; | |
1688 | goto out; | |
1689 | } | |
1690 | if (target_len < task_len) { | |
1691 | /* looking up a parent dir */ | |
1692 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1693 | answer = true; | |
1694 | goto out; | |
1695 | } | |
1696 | if (target_len > task_len) { | |
1697 | /* looking up a child dir */ | |
1698 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1699 | answer = true; | |
1700 | goto out; | |
1701 | } | |
1702 | ||
1703 | out: | |
1704 | free(c2); | |
1705 | return answer; | |
1706 | } | |
1707 | ||
1708 | /* | |
1709 | * given /cgroup/freezer/a/b, return "freezer". | |
1710 | * the returned char* should NOT be freed. | |
1711 | */ | |
1712 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1713 | { | |
1714 | const char *p1; | |
1715 | char *contr, *slash; | |
1716 | ||
99142521 | 1717 | if (strlen(path) < 9) { |
e254948f | 1718 | errno = EACCES; |
237e200e | 1719 | return NULL; |
99142521 CB |
1720 | } |
1721 | if (*(path + 7) != '/') { | |
1722 | errno = EINVAL; | |
237e200e | 1723 | return NULL; |
99142521 | 1724 | } |
3adc421c | 1725 | p1 = path + 8; |
237e200e | 1726 | contr = strdupa(p1); |
99142521 CB |
1727 | if (!contr) { |
1728 | errno = ENOMEM; | |
237e200e | 1729 | return NULL; |
99142521 | 1730 | } |
237e200e SH |
1731 | slash = strstr(contr, "/"); |
1732 | if (slash) | |
1733 | *slash = '\0'; | |
1734 | ||
1735 | int i; | |
3adc421c | 1736 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
1737 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) |
1738 | return hierarchies[i]; | |
1739 | } | |
99142521 | 1740 | errno = ENOENT; |
237e200e SH |
1741 | return NULL; |
1742 | } | |
1743 | ||
1744 | /* | |
1745 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1746 | * Note that the returned value may include files (keynames) etc | |
1747 | */ | |
1748 | static const char *find_cgroup_in_path(const char *path) | |
1749 | { | |
1750 | const char *p1; | |
1751 | ||
bc70ba9b | 1752 | if (strlen(path) < 9) { |
e254948f | 1753 | errno = EACCES; |
237e200e | 1754 | return NULL; |
bc70ba9b CB |
1755 | } |
1756 | p1 = strstr(path + 8, "/"); | |
1757 | if (!p1) { | |
1758 | errno = EINVAL; | |
237e200e | 1759 | return NULL; |
bc70ba9b CB |
1760 | } |
1761 | errno = 0; | |
1762 | return p1 + 1; | |
237e200e SH |
1763 | } |
1764 | ||
1765 | /* | |
1766 | * split the last path element from the path in @cg. | |
1767 | * @dir is newly allocated and should be freed, @last not | |
1768 | */ | |
1769 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1770 | { | |
1771 | char *p; | |
1772 | ||
1773 | do { | |
1774 | *dir = strdup(cg); | |
1775 | } while (!*dir); | |
1776 | *last = strrchr(cg, '/'); | |
1777 | if (!*last) { | |
1778 | *last = NULL; | |
1779 | return; | |
1780 | } | |
1781 | p = strrchr(*dir, '/'); | |
1782 | *p = '\0'; | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * FUSE ops for /cgroup | |
1787 | */ | |
1788 | ||
1789 | int cg_getattr(const char *path, struct stat *sb) | |
1790 | { | |
1791 | struct timespec now; | |
1792 | struct fuse_context *fc = fuse_get_context(); | |
1793 | char * cgdir = NULL; | |
1794 | char *last = NULL, *path1, *path2; | |
1795 | struct cgfs_files *k = NULL; | |
1796 | const char *cgroup; | |
1797 | const char *controller = NULL; | |
1798 | int ret = -ENOENT; | |
1799 | ||
1800 | ||
1801 | if (!fc) | |
1802 | return -EIO; | |
1803 | ||
1804 | memset(sb, 0, sizeof(struct stat)); | |
1805 | ||
1806 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1807 | return -EINVAL; | |
1808 | ||
1809 | sb->st_uid = sb->st_gid = 0; | |
1810 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1811 | sb->st_size = 0; | |
1812 | ||
1813 | if (strcmp(path, "/cgroup") == 0) { | |
1814 | sb->st_mode = S_IFDIR | 00755; | |
1815 | sb->st_nlink = 2; | |
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | controller = pick_controller_from_path(fc, path); | |
1820 | if (!controller) | |
2f7036d0 | 1821 | return -errno; |
237e200e SH |
1822 | cgroup = find_cgroup_in_path(path); |
1823 | if (!cgroup) { | |
1824 | /* this is just /cgroup/controller, return it as a dir */ | |
1825 | sb->st_mode = S_IFDIR | 00755; | |
1826 | sb->st_nlink = 2; | |
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1831 | ||
1832 | if (!last) { | |
1833 | path1 = "/"; | |
1834 | path2 = cgdir; | |
1835 | } else { | |
1836 | path1 = cgdir; | |
1837 | path2 = last; | |
1838 | } | |
1839 | ||
1840 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1841 | if (initpid <= 0) | |
1842 | initpid = fc->pid; | |
1843 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1844 | * Then check that caller's cgroup is under path if last is a child | |
1845 | * cgroup, or cgdir if last is a file */ | |
1846 | ||
1847 | if (is_child_cgroup(controller, path1, path2)) { | |
1848 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1849 | ret = -ENOENT; | |
1850 | goto out; | |
1851 | } | |
1852 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1853 | /* this is just /cgroup/controller, return it as a dir */ | |
1854 | sb->st_mode = S_IFDIR | 00555; | |
1855 | sb->st_nlink = 2; | |
1856 | ret = 0; | |
1857 | goto out; | |
1858 | } | |
1859 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1860 | ret = -EACCES; | |
1861 | goto out; | |
1862 | } | |
1863 | ||
1864 | // get uid, gid, from '/tasks' file and make up a mode | |
1865 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1866 | sb->st_mode = S_IFDIR | 00755; | |
1867 | k = cgfs_get_key(controller, cgroup, NULL); | |
1868 | if (!k) { | |
1869 | sb->st_uid = sb->st_gid = 0; | |
1870 | } else { | |
1871 | sb->st_uid = k->uid; | |
1872 | sb->st_gid = k->gid; | |
1873 | } | |
1874 | free_key(k); | |
1875 | sb->st_nlink = 2; | |
1876 | ret = 0; | |
1877 | goto out; | |
1878 | } | |
1879 | ||
1880 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1881 | sb->st_mode = S_IFREG | k->mode; | |
1882 | sb->st_nlink = 1; | |
1883 | sb->st_uid = k->uid; | |
1884 | sb->st_gid = k->gid; | |
1885 | sb->st_size = 0; | |
1886 | free_key(k); | |
1887 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1888 | ret = -ENOENT; | |
1889 | goto out; | |
1890 | } | |
237e200e SH |
1891 | ret = 0; |
1892 | } | |
1893 | ||
1894 | out: | |
1895 | free(cgdir); | |
1896 | return ret; | |
1897 | } | |
1898 | ||
1899 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1900 | { | |
1901 | struct fuse_context *fc = fuse_get_context(); | |
1902 | const char *cgroup; | |
1903 | struct file_info *dir_info; | |
1904 | char *controller = NULL; | |
1905 | ||
1906 | if (!fc) | |
1907 | return -EIO; | |
1908 | ||
1909 | if (strcmp(path, "/cgroup") == 0) { | |
1910 | cgroup = NULL; | |
1911 | controller = NULL; | |
1912 | } else { | |
1913 | // return list of keys for the controller, and list of child cgroups | |
1914 | controller = pick_controller_from_path(fc, path); | |
1915 | if (!controller) | |
2f7036d0 | 1916 | return -errno; |
237e200e SH |
1917 | |
1918 | cgroup = find_cgroup_in_path(path); | |
1919 | if (!cgroup) { | |
1920 | /* this is just /cgroup/controller, return its contents */ | |
1921 | cgroup = "/"; | |
1922 | } | |
1923 | } | |
1924 | ||
1925 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1926 | if (initpid <= 0) | |
1927 | initpid = fc->pid; | |
1928 | if (cgroup) { | |
1929 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1930 | return -ENOENT; | |
1931 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1932 | return -EACCES; | |
1933 | } | |
1934 | ||
1935 | /* we'll free this at cg_releasedir */ | |
1936 | dir_info = malloc(sizeof(*dir_info)); | |
1937 | if (!dir_info) | |
1938 | return -ENOMEM; | |
1939 | dir_info->controller = must_copy_string(controller); | |
1940 | dir_info->cgroup = must_copy_string(cgroup); | |
1941 | dir_info->type = LXC_TYPE_CGDIR; | |
1942 | dir_info->buf = NULL; | |
1943 | dir_info->file = NULL; | |
1944 | dir_info->buflen = 0; | |
1945 | ||
1946 | fi->fh = (unsigned long)dir_info; | |
1947 | return 0; | |
1948 | } | |
1949 | ||
1950 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1951 | struct fuse_file_info *fi) | |
1952 | { | |
1953 | struct file_info *d = (struct file_info *)fi->fh; | |
1954 | struct cgfs_files **list = NULL; | |
1955 | int i, ret; | |
1956 | char *nextcg = NULL; | |
1957 | struct fuse_context *fc = fuse_get_context(); | |
1958 | char **clist = NULL; | |
1959 | ||
d639f863 CB |
1960 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) |
1961 | return -EIO; | |
1962 | ||
237e200e | 1963 | if (d->type != LXC_TYPE_CGDIR) { |
b8defc3d | 1964 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); |
237e200e SH |
1965 | return -EIO; |
1966 | } | |
1967 | if (!d->cgroup && !d->controller) { | |
1968 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1969 | int i; | |
1970 | ||
1971 | for (i = 0; i < num_hierarchies; i++) { | |
1972 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1973 | return -EIO; | |
1974 | } | |
1975 | } | |
1976 | return 0; | |
1977 | } | |
1978 | ||
1979 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1980 | // not a valid cgroup | |
1981 | ret = -EINVAL; | |
1982 | goto out; | |
1983 | } | |
1984 | ||
1985 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1986 | if (initpid <= 0) | |
1987 | initpid = fc->pid; | |
1988 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1989 | if (nextcg) { | |
1990 | ret = filler(buf, nextcg, NULL, 0); | |
1991 | free(nextcg); | |
1992 | if (ret != 0) { | |
1993 | ret = -EIO; | |
1994 | goto out; | |
1995 | } | |
1996 | } | |
1997 | ret = 0; | |
1998 | goto out; | |
1999 | } | |
2000 | ||
2001 | for (i = 0; list[i]; i++) { | |
2002 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
2003 | ret = -EIO; | |
2004 | goto out; | |
2005 | } | |
2006 | } | |
2007 | ||
2008 | // now get the list of child cgroups | |
2009 | ||
2010 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
2011 | ret = 0; | |
2012 | goto out; | |
2013 | } | |
f366da65 WB |
2014 | if (clist) { |
2015 | for (i = 0; clist[i]; i++) { | |
2016 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
2017 | ret = -EIO; | |
2018 | goto out; | |
2019 | } | |
237e200e SH |
2020 | } |
2021 | } | |
2022 | ret = 0; | |
2023 | ||
2024 | out: | |
2025 | free_keys(list); | |
2026 | if (clist) { | |
2027 | for (i = 0; clist[i]; i++) | |
2028 | free(clist[i]); | |
2029 | free(clist); | |
2030 | } | |
2031 | return ret; | |
2032 | } | |
2033 | ||
43215927 | 2034 | static void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 2035 | { |
43215927 SH |
2036 | struct file_info *f = (struct file_info *)fi->fh; |
2037 | ||
237e200e SH |
2038 | if (!f) |
2039 | return; | |
43215927 SH |
2040 | |
2041 | fi->fh = 0; | |
2042 | ||
237e200e | 2043 | free(f->controller); |
43215927 | 2044 | f->controller = NULL; |
237e200e | 2045 | free(f->cgroup); |
43215927 | 2046 | f->cgroup = NULL; |
237e200e | 2047 | free(f->file); |
43215927 | 2048 | f->file = NULL; |
237e200e | 2049 | free(f->buf); |
43215927 | 2050 | f->buf = NULL; |
237e200e SH |
2051 | free(f); |
2052 | } | |
2053 | ||
2054 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
2055 | { | |
43215927 | 2056 | do_release_file_info(fi); |
237e200e SH |
2057 | return 0; |
2058 | } | |
2059 | ||
2060 | int cg_open(const char *path, struct fuse_file_info *fi) | |
2061 | { | |
2062 | const char *cgroup; | |
2063 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
2064 | struct cgfs_files *k = NULL; | |
2065 | struct file_info *file_info; | |
2066 | struct fuse_context *fc = fuse_get_context(); | |
2067 | int ret; | |
2068 | ||
2069 | if (!fc) | |
2070 | return -EIO; | |
2071 | ||
2072 | controller = pick_controller_from_path(fc, path); | |
2073 | if (!controller) | |
2f7036d0 | 2074 | return -errno; |
237e200e SH |
2075 | cgroup = find_cgroup_in_path(path); |
2076 | if (!cgroup) | |
bc70ba9b | 2077 | return -errno; |
237e200e SH |
2078 | |
2079 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2080 | if (!last) { | |
2081 | path1 = "/"; | |
2082 | path2 = cgdir; | |
2083 | } else { | |
2084 | path1 = cgdir; | |
2085 | path2 = last; | |
2086 | } | |
2087 | ||
2088 | k = cgfs_get_key(controller, path1, path2); | |
2089 | if (!k) { | |
2090 | ret = -EINVAL; | |
2091 | goto out; | |
2092 | } | |
2093 | free_key(k); | |
2094 | ||
2095 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2096 | if (initpid <= 0) | |
2097 | initpid = fc->pid; | |
2098 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2099 | ret = -ENOENT; | |
2100 | goto out; | |
2101 | } | |
2102 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
2103 | ret = -EACCES; |
2104 | goto out; | |
2105 | } | |
2106 | ||
2107 | /* we'll free this at cg_release */ | |
2108 | file_info = malloc(sizeof(*file_info)); | |
2109 | if (!file_info) { | |
2110 | ret = -ENOMEM; | |
2111 | goto out; | |
2112 | } | |
2113 | file_info->controller = must_copy_string(controller); | |
2114 | file_info->cgroup = must_copy_string(path1); | |
2115 | file_info->file = must_copy_string(path2); | |
2116 | file_info->type = LXC_TYPE_CGFILE; | |
2117 | file_info->buf = NULL; | |
2118 | file_info->buflen = 0; | |
2119 | ||
2120 | fi->fh = (unsigned long)file_info; | |
2121 | ret = 0; | |
2122 | ||
2123 | out: | |
2124 | free(cgdir); | |
2125 | return ret; | |
2126 | } | |
2127 | ||
bddbb106 SH |
2128 | int cg_access(const char *path, int mode) |
2129 | { | |
6f0f6b83 | 2130 | int ret; |
bddbb106 | 2131 | const char *cgroup; |
6f0f6b83 CB |
2132 | char *path1, *path2, *controller; |
2133 | char *last = NULL, *cgdir = NULL; | |
bddbb106 SH |
2134 | struct cgfs_files *k = NULL; |
2135 | struct fuse_context *fc = fuse_get_context(); | |
6f0f6b83 | 2136 | |
9873c5e8 | 2137 | if (strcmp(path, "/cgroup") == 0) |
6f0f6b83 | 2138 | return 0; |
bddbb106 SH |
2139 | |
2140 | if (!fc) | |
2141 | return -EIO; | |
2142 | ||
2143 | controller = pick_controller_from_path(fc, path); | |
2144 | if (!controller) | |
2f7036d0 | 2145 | return -errno; |
bddbb106 | 2146 | cgroup = find_cgroup_in_path(path); |
575316c4 SH |
2147 | if (!cgroup) { |
2148 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
2149 | if ((mode & W_OK) == 0) |
2150 | return 0; | |
2151 | return -EACCES; | |
575316c4 | 2152 | } |
bddbb106 SH |
2153 | |
2154 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2155 | if (!last) { | |
2156 | path1 = "/"; | |
2157 | path2 = cgdir; | |
2158 | } else { | |
2159 | path1 = cgdir; | |
2160 | path2 = last; | |
2161 | } | |
2162 | ||
2163 | k = cgfs_get_key(controller, path1, path2); | |
2164 | if (!k) { | |
3f441bc7 SH |
2165 | if ((mode & W_OK) == 0) |
2166 | ret = 0; | |
2167 | else | |
2168 | ret = -EACCES; | |
bddbb106 SH |
2169 | goto out; |
2170 | } | |
2171 | free_key(k); | |
2172 | ||
2173 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2174 | if (initpid <= 0) | |
2175 | initpid = fc->pid; | |
2176 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2177 | ret = -ENOENT; | |
2178 | goto out; | |
2179 | } | |
2180 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
2181 | ret = -EACCES; | |
2182 | goto out; | |
2183 | } | |
2184 | ||
2185 | ret = 0; | |
2186 | ||
2187 | out: | |
2188 | free(cgdir); | |
2189 | return ret; | |
2190 | } | |
2191 | ||
237e200e SH |
2192 | int cg_release(const char *path, struct fuse_file_info *fi) |
2193 | { | |
43215927 | 2194 | do_release_file_info(fi); |
237e200e SH |
2195 | return 0; |
2196 | } | |
2197 | ||
2198 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
2199 | ||
2200 | static bool wait_for_sock(int sock, int timeout) | |
2201 | { | |
2202 | struct epoll_event ev; | |
2203 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
2204 | ||
2205 | if ((starttime = time(NULL)) < 0) | |
2206 | return false; | |
2207 | ||
2208 | if ((epfd = epoll_create(1)) < 0) { | |
b8defc3d | 2209 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); |
237e200e SH |
2210 | return false; |
2211 | } | |
2212 | ||
2213 | ev.events = POLLIN_SET; | |
2214 | ev.data.fd = sock; | |
2215 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
b8defc3d | 2216 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); |
237e200e SH |
2217 | close(epfd); |
2218 | return false; | |
2219 | } | |
2220 | ||
2221 | again: | |
2222 | if ((now = time(NULL)) < 0) { | |
2223 | close(epfd); | |
2224 | return false; | |
2225 | } | |
2226 | ||
2227 | deltatime = (starttime + timeout) - now; | |
2228 | if (deltatime < 0) { // timeout | |
2229 | errno = 0; | |
2230 | close(epfd); | |
2231 | return false; | |
2232 | } | |
2233 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2234 | if (ret < 0 && errno == EINTR) | |
2235 | goto again; | |
2236 | saved_errno = errno; | |
2237 | close(epfd); | |
2238 | ||
2239 | if (ret <= 0) { | |
2240 | errno = saved_errno; | |
2241 | return false; | |
2242 | } | |
2243 | return true; | |
2244 | } | |
2245 | ||
2246 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2247 | { | |
2248 | if (!wait_for_sock(sockfd, 2)) | |
2249 | return -1; | |
2250 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2251 | } | |
2252 | ||
2253 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2254 | { | |
2255 | struct msghdr msg = { 0 }; | |
2256 | struct iovec iov; | |
2257 | struct cmsghdr *cmsg; | |
2258 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2259 | char buf[1]; | |
2260 | buf[0] = 'p'; | |
2261 | ||
2262 | if (pingfirst) { | |
2263 | if (msgrecv(sock, buf, 1) != 1) { | |
b8defc3d | 2264 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); |
237e200e SH |
2265 | return SEND_CREDS_FAIL; |
2266 | } | |
2267 | } | |
2268 | ||
2269 | msg.msg_control = cmsgbuf; | |
2270 | msg.msg_controllen = sizeof(cmsgbuf); | |
2271 | ||
2272 | cmsg = CMSG_FIRSTHDR(&msg); | |
2273 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2274 | cmsg->cmsg_level = SOL_SOCKET; | |
2275 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2276 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2277 | ||
2278 | msg.msg_name = NULL; | |
2279 | msg.msg_namelen = 0; | |
2280 | ||
2281 | buf[0] = v; | |
2282 | iov.iov_base = buf; | |
2283 | iov.iov_len = sizeof(buf); | |
2284 | msg.msg_iov = &iov; | |
2285 | msg.msg_iovlen = 1; | |
2286 | ||
2287 | if (sendmsg(sock, &msg, 0) < 0) { | |
b8defc3d | 2288 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); |
237e200e SH |
2289 | if (errno == 3) |
2290 | return SEND_CREDS_NOTSK; | |
2291 | return SEND_CREDS_FAIL; | |
2292 | } | |
2293 | ||
2294 | return SEND_CREDS_OK; | |
2295 | } | |
2296 | ||
2297 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2298 | { | |
2299 | struct msghdr msg = { 0 }; | |
2300 | struct iovec iov; | |
2301 | struct cmsghdr *cmsg; | |
2302 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2303 | char buf[1]; | |
2304 | int ret; | |
2305 | int optval = 1; | |
2306 | ||
2307 | *v = '1'; | |
2308 | ||
2309 | cred->pid = -1; | |
2310 | cred->uid = -1; | |
2311 | cred->gid = -1; | |
2312 | ||
2313 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
b8defc3d | 2314 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); |
237e200e SH |
2315 | return false; |
2316 | } | |
2317 | buf[0] = '1'; | |
2318 | if (write(sock, buf, 1) != 1) { | |
b8defc3d | 2319 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); |
237e200e SH |
2320 | return false; |
2321 | } | |
2322 | ||
2323 | msg.msg_name = NULL; | |
2324 | msg.msg_namelen = 0; | |
2325 | msg.msg_control = cmsgbuf; | |
2326 | msg.msg_controllen = sizeof(cmsgbuf); | |
2327 | ||
2328 | iov.iov_base = buf; | |
2329 | iov.iov_len = sizeof(buf); | |
2330 | msg.msg_iov = &iov; | |
2331 | msg.msg_iovlen = 1; | |
2332 | ||
2333 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2334 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2335 | return false; |
2336 | } | |
2337 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2338 | if (ret < 0) { | |
b8defc3d | 2339 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2340 | return false; |
2341 | } | |
2342 | ||
2343 | cmsg = CMSG_FIRSTHDR(&msg); | |
2344 | ||
2345 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2346 | cmsg->cmsg_level == SOL_SOCKET && | |
2347 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2348 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2349 | } | |
2350 | *v = buf[0]; | |
2351 | ||
2352 | return true; | |
2353 | } | |
2354 | ||
35174b0f FG |
2355 | struct pid_ns_clone_args { |
2356 | int *cpipe; | |
2357 | int sock; | |
2358 | pid_t tpid; | |
2359 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2360 | }; | |
2361 | ||
2362 | /* | |
2363 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2364 | * with clone(). This simply writes '1' as ACK back to the parent | |
2365 | * before calling the actual wrapped function. | |
2366 | */ | |
2367 | static int pid_ns_clone_wrapper(void *arg) { | |
2368 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2369 | char b = '1'; | |
2370 | ||
2371 | close(args->cpipe[0]); | |
b8defc3d CB |
2372 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) |
2373 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
35174b0f FG |
2374 | close(args->cpipe[1]); |
2375 | return args->wrapped(args->sock, args->tpid); | |
2376 | } | |
237e200e SH |
2377 | |
2378 | /* | |
2379 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2380 | * int value back over the socket. This shifts the pid from the | |
2381 | * sender's pidns into tpid's pidns. | |
2382 | */ | |
35174b0f | 2383 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2384 | { |
2385 | char v = '0'; | |
2386 | struct ucred cred; | |
2387 | ||
2388 | while (recv_creds(sock, &cred, &v)) { | |
2389 | if (v == '1') | |
35174b0f | 2390 | return 0; |
237e200e | 2391 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2392 | return 1; |
237e200e | 2393 | } |
35174b0f | 2394 | return 0; |
237e200e SH |
2395 | } |
2396 | ||
35174b0f | 2397 | |
237e200e SH |
2398 | /* |
2399 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2400 | * in your old pidns. Only children which you clone will be in the target |
2401 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2402 | * actually convert pids. | |
2403 | * | |
2404 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2405 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2406 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2407 | * this issue. | |
237e200e SH |
2408 | */ |
2409 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2410 | { | |
2411 | int newnsfd = -1, ret, cpipe[2]; | |
2412 | char fnam[100]; | |
2413 | pid_t cpid; | |
2414 | char v; | |
2415 | ||
2416 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2417 | if (ret < 0 || ret >= sizeof(fnam)) | |
2418 | _exit(1); | |
2419 | newnsfd = open(fnam, O_RDONLY); | |
2420 | if (newnsfd < 0) | |
2421 | _exit(1); | |
2422 | if (setns(newnsfd, 0) < 0) | |
2423 | _exit(1); | |
2424 | close(newnsfd); | |
2425 | ||
2426 | if (pipe(cpipe) < 0) | |
2427 | _exit(1); | |
2428 | ||
35174b0f FG |
2429 | struct pid_ns_clone_args args = { |
2430 | .cpipe = cpipe, | |
2431 | .sock = sock, | |
2432 | .tpid = tpid, | |
2433 | .wrapped = &pid_to_ns | |
2434 | }; | |
2435 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2436 | void *stack = alloca(stack_size); | |
2437 | ||
2438 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2439 | if (cpid < 0) |
2440 | _exit(1); | |
2441 | ||
237e200e SH |
2442 | // give the child 1 second to be done forking and |
2443 | // write its ack | |
2444 | if (!wait_for_sock(cpipe[0], 1)) | |
2445 | _exit(1); | |
2446 | ret = read(cpipe[0], &v, 1); | |
2447 | if (ret != sizeof(char) || v != '1') | |
2448 | _exit(1); | |
2449 | ||
2450 | if (!wait_for_pid(cpid)) | |
2451 | _exit(1); | |
2452 | _exit(0); | |
2453 | } | |
2454 | ||
2455 | /* | |
2456 | * To read cgroup files with a particular pid, we will setns into the child | |
2457 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2458 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2459 | */ | |
2460 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2461 | { | |
2462 | int sock[2] = {-1, -1}; | |
2463 | char *tmpdata = NULL; | |
2464 | int ret; | |
2465 | pid_t qpid, cpid = -1; | |
2466 | bool answer = false; | |
2467 | char v = '0'; | |
2468 | struct ucred cred; | |
2469 | size_t sz = 0, asz = 0; | |
2470 | ||
2471 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2472 | return false; | |
2473 | ||
2474 | /* | |
2475 | * Now we read the pids from returned data one by one, pass | |
2476 | * them into a child in the target namespace, read back the | |
2477 | * translated pids, and put them into our to-return data | |
2478 | */ | |
2479 | ||
2480 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2481 | perror("socketpair"); | |
2482 | free(tmpdata); | |
2483 | return false; | |
2484 | } | |
2485 | ||
2486 | cpid = fork(); | |
2487 | if (cpid == -1) | |
2488 | goto out; | |
2489 | ||
2490 | if (!cpid) // child - exits when done | |
2491 | pid_to_ns_wrapper(sock[1], tpid); | |
2492 | ||
2493 | char *ptr = tmpdata; | |
2494 | cred.uid = 0; | |
2495 | cred.gid = 0; | |
2496 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2497 | cred.pid = qpid; | |
2498 | ret = send_creds(sock[0], &cred, v, true); | |
2499 | ||
2500 | if (ret == SEND_CREDS_NOTSK) | |
2501 | goto next; | |
2502 | if (ret == SEND_CREDS_FAIL) | |
2503 | goto out; | |
2504 | ||
2505 | // read converted results | |
2506 | if (!wait_for_sock(sock[0], 2)) { | |
b8defc3d | 2507 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2508 | goto out; |
2509 | } | |
2510 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2511 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2512 | goto out; |
2513 | } | |
2514 | must_strcat_pid(d, &sz, &asz, qpid); | |
2515 | next: | |
2516 | ptr = strchr(ptr, '\n'); | |
2517 | if (!ptr) | |
2518 | break; | |
2519 | ptr++; | |
2520 | } | |
2521 | ||
2522 | cred.pid = getpid(); | |
2523 | v = '1'; | |
2524 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2525 | // failed to ask child to exit | |
b8defc3d | 2526 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); |
237e200e SH |
2527 | goto out; |
2528 | } | |
2529 | ||
2530 | answer = true; | |
2531 | ||
2532 | out: | |
2533 | free(tmpdata); | |
2534 | if (cpid != -1) | |
2535 | wait_for_pid(cpid); | |
2536 | if (sock[0] != -1) { | |
2537 | close(sock[0]); | |
2538 | close(sock[1]); | |
2539 | } | |
2540 | return answer; | |
2541 | } | |
2542 | ||
2543 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2544 | struct fuse_file_info *fi) | |
2545 | { | |
2546 | struct fuse_context *fc = fuse_get_context(); | |
2547 | struct file_info *f = (struct file_info *)fi->fh; | |
2548 | struct cgfs_files *k = NULL; | |
2549 | char *data = NULL; | |
2550 | int ret, s; | |
2551 | bool r; | |
2552 | ||
2553 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2554 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); |
237e200e SH |
2555 | return -EIO; |
2556 | } | |
2557 | ||
2558 | if (offset) | |
2559 | return 0; | |
2560 | ||
2561 | if (!fc) | |
2562 | return -EIO; | |
2563 | ||
2564 | if (!f->controller) | |
2565 | return -EINVAL; | |
2566 | ||
2567 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2568 | return -EINVAL; | |
2569 | } | |
2570 | free_key(k); | |
2571 | ||
2572 | ||
888f8f3c | 2573 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2574 | ret = -EACCES; |
2575 | goto out; | |
2576 | } | |
2577 | ||
2578 | if (strcmp(f->file, "tasks") == 0 || | |
2579 | strcmp(f->file, "/tasks") == 0 || | |
2580 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2581 | strcmp(f->file, "cgroup.procs") == 0) | |
2582 | // special case - we have to translate the pids | |
2583 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2584 | else | |
2585 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2586 | ||
2587 | if (!r) { | |
2588 | ret = -EINVAL; | |
2589 | goto out; | |
2590 | } | |
2591 | ||
2592 | if (!data) { | |
2593 | ret = 0; | |
2594 | goto out; | |
2595 | } | |
2596 | s = strlen(data); | |
2597 | if (s > size) | |
2598 | s = size; | |
2599 | memcpy(buf, data, s); | |
2600 | if (s > 0 && s < size && data[s-1] != '\n') | |
2601 | buf[s++] = '\n'; | |
2602 | ||
2603 | ret = s; | |
2604 | ||
2605 | out: | |
2606 | free(data); | |
2607 | return ret; | |
2608 | } | |
2609 | ||
35174b0f | 2610 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2611 | { |
2612 | pid_t vpid; | |
2613 | struct ucred cred; | |
2614 | char v; | |
2615 | int ret; | |
2616 | ||
2617 | cred.uid = 0; | |
2618 | cred.gid = 0; | |
2619 | while (1) { | |
2620 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2621 | lxcfs_error("%s\n", "Timeout reading from parent."); |
35174b0f | 2622 | return 1; |
237e200e SH |
2623 | } |
2624 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
b8defc3d | 2625 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); |
35174b0f | 2626 | return 1; |
237e200e SH |
2627 | } |
2628 | if (vpid == -1) // done | |
2629 | break; | |
2630 | v = '0'; | |
2631 | cred.pid = vpid; | |
2632 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2633 | v = '1'; | |
2634 | cred.pid = getpid(); | |
2635 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2636 | return 1; |
237e200e SH |
2637 | } |
2638 | } | |
35174b0f | 2639 | return 0; |
237e200e SH |
2640 | } |
2641 | ||
2642 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2643 | { | |
2644 | int newnsfd = -1, ret, cpipe[2]; | |
2645 | char fnam[100]; | |
2646 | pid_t cpid; | |
2647 | char v; | |
2648 | ||
2649 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2650 | if (ret < 0 || ret >= sizeof(fnam)) | |
2651 | _exit(1); | |
2652 | newnsfd = open(fnam, O_RDONLY); | |
2653 | if (newnsfd < 0) | |
2654 | _exit(1); | |
2655 | if (setns(newnsfd, 0) < 0) | |
2656 | _exit(1); | |
2657 | close(newnsfd); | |
2658 | ||
2659 | if (pipe(cpipe) < 0) | |
2660 | _exit(1); | |
2661 | ||
35174b0f FG |
2662 | struct pid_ns_clone_args args = { |
2663 | .cpipe = cpipe, | |
2664 | .sock = sock, | |
2665 | .tpid = tpid, | |
2666 | .wrapped = &pid_from_ns | |
2667 | }; | |
f0f8b851 SH |
2668 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2669 | void *stack = alloca(stack_size); | |
35174b0f FG |
2670 | |
2671 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2672 | if (cpid < 0) |
2673 | _exit(1); | |
2674 | ||
237e200e SH |
2675 | // give the child 1 second to be done forking and |
2676 | // write its ack | |
2677 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2678 | _exit(1); |
237e200e | 2679 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2680 | if (ret != sizeof(char) || v != '1') |
2681 | _exit(1); | |
237e200e SH |
2682 | |
2683 | if (!wait_for_pid(cpid)) | |
2684 | _exit(1); | |
2685 | _exit(0); | |
237e200e SH |
2686 | } |
2687 | ||
2688 | /* | |
2689 | * Given host @uid, return the uid to which it maps in | |
2690 | * @pid's user namespace, or -1 if none. | |
2691 | */ | |
2692 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2693 | { | |
2694 | FILE *f; | |
2695 | char line[400]; | |
2696 | ||
2697 | sprintf(line, "/proc/%d/uid_map", pid); | |
2698 | if ((f = fopen(line, "r")) == NULL) { | |
2699 | return false; | |
2700 | } | |
2701 | ||
2702 | *answer = convert_id_to_ns(f, uid); | |
2703 | fclose(f); | |
2704 | ||
2705 | if (*answer == -1) | |
2706 | return false; | |
2707 | return true; | |
2708 | } | |
2709 | ||
2710 | /* | |
2711 | * get_pid_creds: get the real uid and gid of @pid from | |
2712 | * /proc/$$/status | |
2713 | * (XXX should we use euid here?) | |
2714 | */ | |
2715 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2716 | { | |
2717 | char line[400]; | |
2718 | uid_t u; | |
2719 | gid_t g; | |
2720 | FILE *f; | |
2721 | ||
2722 | *uid = -1; | |
2723 | *gid = -1; | |
2724 | sprintf(line, "/proc/%d/status", pid); | |
2725 | if ((f = fopen(line, "r")) == NULL) { | |
b8defc3d | 2726 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); |
237e200e SH |
2727 | return; |
2728 | } | |
2729 | while (fgets(line, 400, f)) { | |
2730 | if (strncmp(line, "Uid:", 4) == 0) { | |
2731 | if (sscanf(line+4, "%u", &u) != 1) { | |
b8defc3d | 2732 | lxcfs_error("bad uid line for pid %u\n", pid); |
237e200e SH |
2733 | fclose(f); |
2734 | return; | |
2735 | } | |
2736 | *uid = u; | |
2737 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2738 | if (sscanf(line+4, "%u", &g) != 1) { | |
b8defc3d | 2739 | lxcfs_error("bad gid line for pid %u\n", pid); |
237e200e SH |
2740 | fclose(f); |
2741 | return; | |
2742 | } | |
2743 | *gid = g; | |
2744 | } | |
2745 | } | |
2746 | fclose(f); | |
2747 | } | |
2748 | ||
2749 | /* | |
2750 | * May the requestor @r move victim @v to a new cgroup? | |
2751 | * This is allowed if | |
2752 | * . they are the same task | |
2753 | * . they are ownedy by the same uid | |
2754 | * . @r is root on the host, or | |
2755 | * . @v's uid is mapped into @r's where @r is root. | |
2756 | */ | |
2757 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2758 | { | |
2759 | uid_t v_uid, tmpuid; | |
2760 | gid_t v_gid; | |
2761 | ||
2762 | if (r == v) | |
2763 | return true; | |
2764 | if (r_uid == 0) | |
2765 | return true; | |
2766 | get_pid_creds(v, &v_uid, &v_gid); | |
2767 | if (r_uid == v_uid) | |
2768 | return true; | |
2769 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2770 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2771 | return true; | |
2772 | return false; | |
2773 | } | |
2774 | ||
2775 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2776 | const char *file, const char *buf) | |
2777 | { | |
2778 | int sock[2] = {-1, -1}; | |
2779 | pid_t qpid, cpid = -1; | |
2780 | FILE *pids_file = NULL; | |
2781 | bool answer = false, fail = false; | |
2782 | ||
2783 | pids_file = open_pids_file(contrl, cg); | |
2784 | if (!pids_file) | |
2785 | return false; | |
2786 | ||
2787 | /* | |
2788 | * write the pids to a socket, have helper in writer's pidns | |
2789 | * call movepid for us | |
2790 | */ | |
2791 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2792 | perror("socketpair"); | |
2793 | goto out; | |
2794 | } | |
2795 | ||
2796 | cpid = fork(); | |
2797 | if (cpid == -1) | |
2798 | goto out; | |
2799 | ||
2800 | if (!cpid) { // child | |
2801 | fclose(pids_file); | |
2802 | pid_from_ns_wrapper(sock[1], tpid); | |
2803 | } | |
2804 | ||
2805 | const char *ptr = buf; | |
2806 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2807 | struct ucred cred; | |
2808 | char v; | |
2809 | ||
2810 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2811 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); |
237e200e SH |
2812 | goto out; |
2813 | } | |
2814 | ||
2815 | if (recv_creds(sock[0], &cred, &v)) { | |
2816 | if (v == '0') { | |
2817 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2818 | fail = true; | |
2819 | break; | |
2820 | } | |
2821 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2822 | fail = true; | |
2823 | } | |
2824 | } | |
2825 | ||
2826 | ptr = strchr(ptr, '\n'); | |
2827 | if (!ptr) | |
2828 | break; | |
2829 | ptr++; | |
2830 | } | |
2831 | ||
2832 | /* All good, write the value */ | |
2833 | qpid = -1; | |
2834 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
b8defc3d | 2835 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); |
237e200e SH |
2836 | |
2837 | if (!fail) | |
2838 | answer = true; | |
2839 | ||
2840 | out: | |
2841 | if (cpid != -1) | |
2842 | wait_for_pid(cpid); | |
2843 | if (sock[0] != -1) { | |
2844 | close(sock[0]); | |
2845 | close(sock[1]); | |
2846 | } | |
2847 | if (pids_file) { | |
2848 | if (fclose(pids_file) != 0) | |
2849 | answer = false; | |
2850 | } | |
2851 | return answer; | |
2852 | } | |
2853 | ||
2854 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2855 | struct fuse_file_info *fi) | |
2856 | { | |
2857 | struct fuse_context *fc = fuse_get_context(); | |
2858 | char *localbuf = NULL; | |
2859 | struct cgfs_files *k = NULL; | |
2860 | struct file_info *f = (struct file_info *)fi->fh; | |
2861 | bool r; | |
2862 | ||
2863 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2864 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); |
237e200e SH |
2865 | return -EIO; |
2866 | } | |
2867 | ||
2868 | if (offset) | |
2869 | return 0; | |
2870 | ||
2871 | if (!fc) | |
2872 | return -EIO; | |
2873 | ||
2874 | localbuf = alloca(size+1); | |
2875 | localbuf[size] = '\0'; | |
2876 | memcpy(localbuf, buf, size); | |
2877 | ||
2878 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2879 | size = -EINVAL; | |
2880 | goto out; | |
2881 | } | |
2882 | ||
2883 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2884 | size = -EACCES; | |
2885 | goto out; | |
2886 | } | |
2887 | ||
2888 | if (strcmp(f->file, "tasks") == 0 || | |
2889 | strcmp(f->file, "/tasks") == 0 || | |
2890 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2891 | strcmp(f->file, "cgroup.procs") == 0) | |
2892 | // special case - we have to translate the pids | |
2893 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2894 | else | |
2895 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2896 | ||
2897 | if (!r) | |
2898 | size = -EINVAL; | |
2899 | ||
2900 | out: | |
2901 | free_key(k); | |
2902 | return size; | |
2903 | } | |
2904 | ||
2905 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2906 | { | |
2907 | struct fuse_context *fc = fuse_get_context(); | |
2908 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2909 | struct cgfs_files *k = NULL; | |
2910 | const char *cgroup; | |
2911 | int ret; | |
2912 | ||
2913 | if (!fc) | |
2914 | return -EIO; | |
2915 | ||
2916 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2917 | return -EPERM; |
237e200e SH |
2918 | |
2919 | controller = pick_controller_from_path(fc, path); | |
2920 | if (!controller) | |
bc70ba9b CB |
2921 | return errno == ENOENT ? -EPERM : -errno; |
2922 | ||
237e200e SH |
2923 | cgroup = find_cgroup_in_path(path); |
2924 | if (!cgroup) | |
2925 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2926 | return -EPERM; |
237e200e SH |
2927 | |
2928 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2929 | ||
2930 | if (!last) { | |
2931 | path1 = "/"; | |
2932 | path2 = cgdir; | |
2933 | } else { | |
2934 | path1 = cgdir; | |
2935 | path2 = last; | |
2936 | } | |
2937 | ||
2938 | if (is_child_cgroup(controller, path1, path2)) { | |
2939 | // get uid, gid, from '/tasks' file and make up a mode | |
2940 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2941 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2942 | ||
2943 | } else | |
2944 | k = cgfs_get_key(controller, path1, path2); | |
2945 | ||
2946 | if (!k) { | |
2947 | ret = -EINVAL; | |
2948 | goto out; | |
2949 | } | |
2950 | ||
2951 | /* | |
2952 | * This being a fuse request, the uid and gid must be valid | |
2953 | * in the caller's namespace. So we can just check to make | |
2954 | * sure that the caller is root in his uid, and privileged | |
2955 | * over the file's current owner. | |
2956 | */ | |
2957 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2958 | ret = -EACCES; | |
2959 | goto out; | |
2960 | } | |
2961 | ||
2962 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2963 | ||
2964 | out: | |
2965 | free_key(k); | |
2966 | free(cgdir); | |
2967 | ||
2968 | return ret; | |
2969 | } | |
2970 | ||
2971 | int cg_chmod(const char *path, mode_t mode) | |
2972 | { | |
2973 | struct fuse_context *fc = fuse_get_context(); | |
2974 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2975 | struct cgfs_files *k = NULL; | |
2976 | const char *cgroup; | |
2977 | int ret; | |
2978 | ||
2979 | if (!fc) | |
2980 | return -EIO; | |
2981 | ||
2982 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2983 | return -EPERM; |
237e200e SH |
2984 | |
2985 | controller = pick_controller_from_path(fc, path); | |
2986 | if (!controller) | |
bc70ba9b CB |
2987 | return errno == ENOENT ? -EPERM : -errno; |
2988 | ||
237e200e SH |
2989 | cgroup = find_cgroup_in_path(path); |
2990 | if (!cgroup) | |
2991 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2992 | return -EPERM; |
237e200e SH |
2993 | |
2994 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2995 | ||
2996 | if (!last) { | |
2997 | path1 = "/"; | |
2998 | path2 = cgdir; | |
2999 | } else { | |
3000 | path1 = cgdir; | |
3001 | path2 = last; | |
3002 | } | |
3003 | ||
3004 | if (is_child_cgroup(controller, path1, path2)) { | |
3005 | // get uid, gid, from '/tasks' file and make up a mode | |
3006 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
3007 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
3008 | ||
3009 | } else | |
3010 | k = cgfs_get_key(controller, path1, path2); | |
3011 | ||
3012 | if (!k) { | |
3013 | ret = -EINVAL; | |
3014 | goto out; | |
3015 | } | |
3016 | ||
3017 | /* | |
3018 | * This being a fuse request, the uid and gid must be valid | |
3019 | * in the caller's namespace. So we can just check to make | |
3020 | * sure that the caller is root in his uid, and privileged | |
3021 | * over the file's current owner. | |
3022 | */ | |
3023 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
3024 | ret = -EPERM; | |
3025 | goto out; | |
3026 | } | |
3027 | ||
3028 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
3029 | ret = -EINVAL; | |
3030 | goto out; | |
3031 | } | |
3032 | ||
3033 | ret = 0; | |
3034 | out: | |
3035 | free_key(k); | |
3036 | free(cgdir); | |
3037 | return ret; | |
3038 | } | |
3039 | ||
3040 | int cg_mkdir(const char *path, mode_t mode) | |
3041 | { | |
3042 | struct fuse_context *fc = fuse_get_context(); | |
3043 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
3044 | const char *cgroup; | |
3045 | int ret; | |
3046 | ||
3047 | if (!fc) | |
3048 | return -EIO; | |
3049 | ||
237e200e SH |
3050 | controller = pick_controller_from_path(fc, path); |
3051 | if (!controller) | |
2f7036d0 | 3052 | return errno == ENOENT ? -EPERM : -errno; |
237e200e SH |
3053 | |
3054 | cgroup = find_cgroup_in_path(path); | |
3055 | if (!cgroup) | |
bc70ba9b | 3056 | return -errno; |
237e200e SH |
3057 | |
3058 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3059 | if (!last) | |
3060 | path1 = "/"; | |
3061 | else | |
3062 | path1 = cgdir; | |
3063 | ||
3064 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3065 | if (initpid <= 0) | |
3066 | initpid = fc->pid; | |
3067 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
3068 | if (!next) | |
3069 | ret = -EINVAL; | |
3070 | else if (last && strcmp(next, last) == 0) | |
3071 | ret = -EEXIST; | |
3072 | else | |
2f7036d0 | 3073 | ret = -EPERM; |
237e200e SH |
3074 | goto out; |
3075 | } | |
3076 | ||
3077 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
3078 | ret = -EACCES; | |
3079 | goto out; | |
3080 | } | |
3081 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
3082 | ret = -EACCES; | |
3083 | goto out; | |
3084 | } | |
3085 | ||
3086 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
3087 | ||
3088 | out: | |
3089 | free(cgdir); | |
3090 | free(next); | |
3091 | return ret; | |
3092 | } | |
3093 | ||
3094 | int cg_rmdir(const char *path) | |
3095 | { | |
3096 | struct fuse_context *fc = fuse_get_context(); | |
3097 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
3098 | const char *cgroup; | |
3099 | int ret; | |
3100 | ||
3101 | if (!fc) | |
3102 | return -EIO; | |
3103 | ||
3104 | controller = pick_controller_from_path(fc, path); | |
e254948f CB |
3105 | if (!controller) /* Someone's trying to delete "/cgroup". */ |
3106 | return -EPERM; | |
237e200e SH |
3107 | |
3108 | cgroup = find_cgroup_in_path(path); | |
e254948f CB |
3109 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ |
3110 | return -EPERM; | |
237e200e SH |
3111 | |
3112 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3113 | if (!last) { | |
e254948f CB |
3114 | /* Someone's trying to delete a cgroup on the same level as the |
3115 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
3116 | * rmdir "/cgroup/blkio/init.slice". | |
3117 | */ | |
3118 | ret = -EPERM; | |
237e200e SH |
3119 | goto out; |
3120 | } | |
3121 | ||
3122 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3123 | if (initpid <= 0) | |
3124 | initpid = fc->pid; | |
3125 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
de77249b | 3126 | if (!last || (next && (strcmp(next, last) == 0))) |
237e200e SH |
3127 | ret = -EBUSY; |
3128 | else | |
3129 | ret = -ENOENT; | |
3130 | goto out; | |
3131 | } | |
3132 | ||
3133 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
3134 | ret = -EACCES; | |
3135 | goto out; | |
3136 | } | |
3137 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
3138 | ret = -EACCES; | |
3139 | goto out; | |
3140 | } | |
3141 | ||
3142 | if (!cgfs_remove(controller, cgroup)) { | |
3143 | ret = -EINVAL; | |
3144 | goto out; | |
3145 | } | |
3146 | ||
3147 | ret = 0; | |
3148 | ||
3149 | out: | |
3150 | free(cgdir); | |
3151 | free(next); | |
3152 | return ret; | |
3153 | } | |
3154 | ||
3155 | static bool startswith(const char *line, const char *pref) | |
3156 | { | |
3157 | if (strncmp(line, pref, strlen(pref)) == 0) | |
3158 | return true; | |
3159 | return false; | |
3160 | } | |
3161 | ||
c6095b08 SH |
3162 | static void parse_memstat(char *memstat, unsigned long *cached, |
3163 | unsigned long *active_anon, unsigned long *inactive_anon, | |
3164 | unsigned long *active_file, unsigned long *inactive_file, | |
559eaa8f | 3165 | unsigned long *unevictable, unsigned long *shmem) |
237e200e SH |
3166 | { |
3167 | char *eol; | |
3168 | ||
237e200e | 3169 | while (*memstat) { |
4accebfb AS |
3170 | if (startswith(memstat, "total_cache")) { |
3171 | sscanf(memstat + 11, "%lu", cached); | |
c6095b08 | 3172 | *cached /= 1024; |
4accebfb AS |
3173 | } else if (startswith(memstat, "total_active_anon")) { |
3174 | sscanf(memstat + 17, "%lu", active_anon); | |
c6095b08 | 3175 | *active_anon /= 1024; |
4accebfb AS |
3176 | } else if (startswith(memstat, "total_inactive_anon")) { |
3177 | sscanf(memstat + 19, "%lu", inactive_anon); | |
c6095b08 | 3178 | *inactive_anon /= 1024; |
4accebfb AS |
3179 | } else if (startswith(memstat, "total_active_file")) { |
3180 | sscanf(memstat + 17, "%lu", active_file); | |
c6095b08 | 3181 | *active_file /= 1024; |
4accebfb AS |
3182 | } else if (startswith(memstat, "total_inactive_file")) { |
3183 | sscanf(memstat + 19, "%lu", inactive_file); | |
c6095b08 | 3184 | *inactive_file /= 1024; |
4accebfb AS |
3185 | } else if (startswith(memstat, "total_unevictable")) { |
3186 | sscanf(memstat + 17, "%lu", unevictable); | |
c6095b08 | 3187 | *unevictable /= 1024; |
559eaa8f JS |
3188 | } else if (startswith(memstat, "total_shmem")) { |
3189 | sscanf(memstat + 11, "%lu", shmem); | |
3190 | *shmem /= 1024; | |
237e200e SH |
3191 | } |
3192 | eol = strchr(memstat, '\n'); | |
3193 | if (!eol) | |
3194 | return; | |
3195 | memstat = eol+1; | |
3196 | } | |
3197 | } | |
3198 | ||
3199 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
3200 | { | |
3201 | char *eol; | |
3202 | char key[32]; | |
3203 | ||
3204 | memset(key, 0, 32); | |
3205 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
3206 | ||
3207 | size_t len = strlen(key); | |
3208 | *v = 0; | |
3209 | ||
3210 | while (*str) { | |
3211 | if (startswith(str, key)) { | |
3212 | sscanf(str + len, "%lu", v); | |
3213 | return; | |
3214 | } | |
3215 | eol = strchr(str, '\n'); | |
3216 | if (!eol) | |
3217 | return; | |
3218 | str = eol+1; | |
3219 | } | |
3220 | } | |
3221 | ||
3222 | static int read_file(const char *path, char *buf, size_t size, | |
3223 | struct file_info *d) | |
3224 | { | |
3225 | size_t linelen = 0, total_len = 0, rv = 0; | |
3226 | char *line = NULL; | |
3227 | char *cache = d->buf; | |
3228 | size_t cache_size = d->buflen; | |
3229 | FILE *f = fopen(path, "r"); | |
3230 | if (!f) | |
3231 | return 0; | |
3232 | ||
3233 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3234 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3235 | if (l < 0) { |
3236 | perror("Error writing to cache"); | |
3237 | rv = 0; | |
3238 | goto err; | |
3239 | } | |
3240 | if (l >= cache_size) { | |
b8defc3d | 3241 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3242 | rv = 0; |
3243 | goto err; | |
3244 | } | |
3245 | cache += l; | |
3246 | cache_size -= l; | |
3247 | total_len += l; | |
3248 | } | |
3249 | ||
3250 | d->size = total_len; | |
a262ddb7 CB |
3251 | if (total_len > size) |
3252 | total_len = size; | |
237e200e SH |
3253 | |
3254 | /* read from off 0 */ | |
3255 | memcpy(buf, d->buf, total_len); | |
3256 | rv = total_len; | |
3257 | err: | |
3258 | fclose(f); | |
3259 | free(line); | |
3260 | return rv; | |
3261 | } | |
3262 | ||
3263 | /* | |
3264 | * FUSE ops for /proc | |
3265 | */ | |
3266 | ||
018246ff | 3267 | static unsigned long get_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3268 | { |
3269 | char *memlimit_str = NULL; | |
3270 | unsigned long memlimit = -1; | |
3271 | ||
018246ff | 3272 | if (cgfs_get_value("memory", cgroup, file, &memlimit_str)) |
237e200e SH |
3273 | memlimit = strtoul(memlimit_str, NULL, 10); |
3274 | ||
3275 | free(memlimit_str); | |
3276 | ||
3277 | return memlimit; | |
3278 | } | |
3279 | ||
018246ff | 3280 | static unsigned long get_min_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3281 | { |
3282 | char *copy = strdupa(cgroup); | |
3283 | unsigned long memlimit = 0, retlimit; | |
3284 | ||
018246ff | 3285 | retlimit = get_memlimit(copy, file); |
237e200e SH |
3286 | |
3287 | while (strcmp(copy, "/") != 0) { | |
3288 | copy = dirname(copy); | |
018246ff | 3289 | memlimit = get_memlimit(copy, file); |
237e200e SH |
3290 | if (memlimit != -1 && memlimit < retlimit) |
3291 | retlimit = memlimit; | |
3292 | }; | |
3293 | ||
3294 | return retlimit; | |
3295 | } | |
3296 | ||
3297 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3298 | struct fuse_file_info *fi) | |
3299 | { | |
3300 | struct fuse_context *fc = fuse_get_context(); | |
3301 | struct file_info *d = (struct file_info *)fi->fh; | |
3302 | char *cg; | |
3303 | char *memusage_str = NULL, *memstat_str = NULL, | |
018246ff | 3304 | *memswlimit_str = NULL, *memswusage_str = NULL; |
237e200e | 3305 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, |
c6095b08 | 3306 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
559eaa8f | 3307 | active_file = 0, inactive_file = 0, unevictable = 0, shmem = 0, |
594a10e6 | 3308 | hostswtotal = 0; |
237e200e SH |
3309 | char *line = NULL; |
3310 | size_t linelen = 0, total_len = 0, rv = 0; | |
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, "memory"); | |
3330 | if (!cg) | |
3331 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3332 | prune_init_slice(cg); |
237e200e | 3333 | |
018246ff | 3334 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
237e200e SH |
3335 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) |
3336 | goto err; | |
3337 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3338 | goto err; | |
3339 | ||
3340 | // Following values are allowed to fail, because swapaccount might be turned | |
3341 | // off for current kernel | |
3342 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3343 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3344 | { | |
018246ff | 3345 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
237e200e SH |
3346 | memswusage = strtoul(memswusage_str, NULL, 10); |
3347 | ||
237e200e SH |
3348 | memswlimit = memswlimit / 1024; |
3349 | memswusage = memswusage / 1024; | |
3350 | } | |
3351 | ||
3352 | memusage = strtoul(memusage_str, NULL, 10); | |
3353 | memlimit /= 1024; | |
3354 | memusage /= 1024; | |
3355 | ||
c6095b08 SH |
3356 | parse_memstat(memstat_str, &cached, &active_anon, |
3357 | &inactive_anon, &active_file, &inactive_file, | |
559eaa8f | 3358 | &unevictable, &shmem); |
237e200e SH |
3359 | |
3360 | f = fopen("/proc/meminfo", "r"); | |
3361 | if (!f) | |
3362 | goto err; | |
3363 | ||
3364 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3365 | ssize_t l; |
237e200e SH |
3366 | char *printme, lbuf[100]; |
3367 | ||
3368 | memset(lbuf, 0, 100); | |
3369 | if (startswith(line, "MemTotal:")) { | |
594a10e6 | 3370 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); |
237e200e SH |
3371 | if (hosttotal < memlimit) |
3372 | memlimit = hosttotal; | |
3373 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3374 | printme = lbuf; | |
3375 | } else if (startswith(line, "MemFree:")) { | |
3376 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3377 | printme = lbuf; | |
3378 | } else if (startswith(line, "MemAvailable:")) { | |
ad19b86d | 3379 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage + cached); |
237e200e SH |
3380 | printme = lbuf; |
3381 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
594a10e6 | 3382 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); |
4127e51b | 3383 | if (hostswtotal < memswlimit) |
3384 | memswlimit = hostswtotal; | |
3385 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit); | |
237e200e SH |
3386 | printme = lbuf; |
3387 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
4127e51b | 3388 | unsigned long swaptotal = memswlimit, |
b4665ce0 SH |
3389 | swapusage = memswusage - memusage, |
3390 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3391 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3392 | printme = lbuf; |
da35d72a SH |
3393 | } else if (startswith(line, "Slab:")) { |
3394 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3395 | printme = lbuf; | |
237e200e SH |
3396 | } else if (startswith(line, "Buffers:")) { |
3397 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3398 | printme = lbuf; | |
3399 | } else if (startswith(line, "Cached:")) { | |
3400 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3401 | printme = lbuf; | |
3402 | } else if (startswith(line, "SwapCached:")) { | |
3403 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3404 | printme = lbuf; | |
2f306ad3 | 3405 | } else if (startswith(line, "Active:")) { |
c6095b08 SH |
3406 | snprintf(lbuf, 100, "Active: %8lu kB\n", |
3407 | active_anon + active_file); | |
3408 | printme = lbuf; | |
2f306ad3 | 3409 | } else if (startswith(line, "Inactive:")) { |
c6095b08 SH |
3410 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", |
3411 | inactive_anon + inactive_file); | |
3412 | printme = lbuf; | |
3413 | } else if (startswith(line, "Active(anon)")) { | |
3414 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3415 | printme = lbuf; | |
3416 | } else if (startswith(line, "Inactive(anon)")) { | |
3417 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3418 | printme = lbuf; | |
3419 | } else if (startswith(line, "Active(file)")) { | |
3420 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3421 | printme = lbuf; | |
3422 | } else if (startswith(line, "Inactive(file)")) { | |
3423 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3424 | printme = lbuf; | |
3425 | } else if (startswith(line, "Unevictable")) { | |
3426 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3427 | printme = lbuf; | |
3428 | } else if (startswith(line, "SReclaimable")) { | |
3429 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3430 | printme = lbuf; | |
3431 | } else if (startswith(line, "SUnreclaim")) { | |
3432 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3433 | printme = lbuf; | |
559eaa8f JS |
3434 | } else if (startswith(line, "Shmem:")) { |
3435 | snprintf(lbuf, 100, "Shmem: %8lu kB\n", shmem); | |
3436 | printme = lbuf; | |
237e200e SH |
3437 | } else |
3438 | printme = line; | |
3439 | ||
3440 | l = snprintf(cache, cache_size, "%s", printme); | |
3441 | if (l < 0) { | |
3442 | perror("Error writing to cache"); | |
3443 | rv = 0; | |
3444 | goto err; | |
3445 | ||
3446 | } | |
3447 | if (l >= cache_size) { | |
b8defc3d | 3448 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3449 | rv = 0; |
3450 | goto err; | |
3451 | } | |
3452 | ||
3453 | cache += l; | |
3454 | cache_size -= l; | |
3455 | total_len += l; | |
3456 | } | |
3457 | ||
3458 | d->cached = 1; | |
3459 | d->size = total_len; | |
3460 | if (total_len > size ) total_len = size; | |
3461 | memcpy(buf, d->buf, total_len); | |
3462 | ||
3463 | rv = total_len; | |
3464 | err: | |
3465 | if (f) | |
3466 | fclose(f); | |
3467 | free(line); | |
3468 | free(cg); | |
3469 | free(memusage_str); | |
3470 | free(memswlimit_str); | |
3471 | free(memswusage_str); | |
3472 | free(memstat_str); | |
237e200e SH |
3473 | return rv; |
3474 | } | |
3475 | ||
3476 | /* | |
3477 | * Read the cpuset.cpus for cg | |
3478 | * Return the answer in a newly allocated string which must be freed | |
3479 | */ | |
3480 | static char *get_cpuset(const char *cg) | |
3481 | { | |
3482 | char *answer; | |
3483 | ||
3484 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3485 | return NULL; | |
3486 | return answer; | |
3487 | } | |
3488 | ||
3489 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3490 | ||
3491 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3492 | { | |
3493 | int cpu; | |
3494 | ||
3495 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3496 | return false; | |
3497 | return cpu_in_cpuset(cpu, cpuset); | |
3498 | } | |
3499 | ||
3500 | /* | |
3501 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3502 | */ | |
3503 | static bool is_processor_line(const char *line) | |
3504 | { | |
3505 | int cpu; | |
3506 | ||
3507 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3508 | return true; | |
3509 | return false; | |
3510 | } | |
3511 | ||
3512 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3513 | struct fuse_file_info *fi) | |
3514 | { | |
3515 | struct fuse_context *fc = fuse_get_context(); | |
3516 | struct file_info *d = (struct file_info *)fi->fh; | |
3517 | char *cg; | |
3518 | char *cpuset = NULL; | |
3519 | char *line = NULL; | |
3520 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 SH |
3521 | bool am_printing = false, firstline = true, is_s390x = false; |
3522 | int curcpu = -1, cpu; | |
237e200e SH |
3523 | char *cache = d->buf; |
3524 | size_t cache_size = d->buflen; | |
3525 | FILE *f = NULL; | |
3526 | ||
3527 | if (offset){ | |
3528 | if (offset > d->size) | |
3529 | return -EINVAL; | |
3530 | if (!d->cached) | |
3531 | return 0; | |
3532 | int left = d->size - offset; | |
3533 | total_len = left > size ? size: left; | |
3534 | memcpy(buf, cache + offset, total_len); | |
3535 | return total_len; | |
3536 | } | |
3537 | ||
3538 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3539 | if (initpid <= 0) | |
3540 | initpid = fc->pid; | |
3541 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3542 | if (!cg) | |
3543 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3544 | prune_init_slice(cg); |
237e200e SH |
3545 | |
3546 | cpuset = get_cpuset(cg); | |
3547 | if (!cpuset) | |
3548 | goto err; | |
3549 | ||
3550 | f = fopen("/proc/cpuinfo", "r"); | |
3551 | if (!f) | |
3552 | goto err; | |
3553 | ||
3554 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3555 | ssize_t l; |
f676eb79 SH |
3556 | if (firstline) { |
3557 | firstline = false; | |
3558 | if (strstr(line, "IBM/S390") != NULL) { | |
3559 | is_s390x = true; | |
3560 | am_printing = true; | |
5ed9d4e2 | 3561 | continue; |
f676eb79 SH |
3562 | } |
3563 | } | |
5ed9d4e2 SH |
3564 | if (strncmp(line, "# processors:", 12) == 0) |
3565 | continue; | |
237e200e SH |
3566 | if (is_processor_line(line)) { |
3567 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3568 | if (am_printing) { | |
3569 | curcpu ++; | |
3570 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3571 | if (l < 0) { | |
3572 | perror("Error writing to cache"); | |
3573 | rv = 0; | |
3574 | goto err; | |
3575 | } | |
3576 | if (l >= cache_size) { | |
b8defc3d | 3577 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3578 | rv = 0; |
3579 | goto err; | |
3580 | } | |
3581 | cache += l; | |
3582 | cache_size -= l; | |
3583 | total_len += l; | |
3584 | } | |
3585 | continue; | |
f676eb79 SH |
3586 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3587 | char *p; | |
3588 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3589 | continue; | |
3590 | curcpu ++; | |
3591 | p = strchr(line, ':'); | |
3592 | if (!p || !*p) | |
3593 | goto err; | |
3594 | p++; | |
5ed9d4e2 | 3595 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3596 | if (l < 0) { |
3597 | perror("Error writing to cache"); | |
3598 | rv = 0; | |
3599 | goto err; | |
3600 | } | |
3601 | if (l >= cache_size) { | |
b8defc3d | 3602 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
f676eb79 SH |
3603 | rv = 0; |
3604 | goto err; | |
3605 | } | |
3606 | cache += l; | |
3607 | cache_size -= l; | |
3608 | total_len += l; | |
3609 | continue; | |
3610 | ||
237e200e SH |
3611 | } |
3612 | if (am_printing) { | |
3613 | l = snprintf(cache, cache_size, "%s", line); | |
3614 | if (l < 0) { | |
3615 | perror("Error writing to cache"); | |
3616 | rv = 0; | |
3617 | goto err; | |
3618 | } | |
3619 | if (l >= cache_size) { | |
b8defc3d | 3620 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3621 | rv = 0; |
3622 | goto err; | |
3623 | } | |
3624 | cache += l; | |
3625 | cache_size -= l; | |
3626 | total_len += l; | |
3627 | } | |
3628 | } | |
3629 | ||
5ed9d4e2 SH |
3630 | if (is_s390x) { |
3631 | char *origcache = d->buf; | |
a262ddb7 | 3632 | ssize_t l; |
5ed9d4e2 SH |
3633 | do { |
3634 | d->buf = malloc(d->buflen); | |
3635 | } while (!d->buf); | |
3636 | cache = d->buf; | |
3637 | cache_size = d->buflen; | |
3638 | total_len = 0; | |
3639 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3640 | if (l < 0 || l >= cache_size) { | |
3641 | free(origcache); | |
3642 | goto err; | |
3643 | } | |
3644 | cache_size -= l; | |
3645 | cache += l; | |
3646 | total_len += l; | |
3647 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3648 | if (l < 0 || l >= cache_size) { | |
3649 | free(origcache); | |
3650 | goto err; | |
3651 | } | |
3652 | cache_size -= l; | |
3653 | cache += l; | |
3654 | total_len += l; | |
3655 | l = snprintf(cache, cache_size, "%s", origcache); | |
3656 | free(origcache); | |
3657 | if (l < 0 || l >= cache_size) | |
3658 | goto err; | |
3659 | total_len += l; | |
3660 | } | |
3661 | ||
237e200e SH |
3662 | d->cached = 1; |
3663 | d->size = total_len; | |
3664 | if (total_len > size ) total_len = size; | |
3665 | ||
3666 | /* read from off 0 */ | |
3667 | memcpy(buf, d->buf, total_len); | |
3668 | rv = total_len; | |
3669 | err: | |
3670 | if (f) | |
3671 | fclose(f); | |
3672 | free(line); | |
3673 | free(cpuset); | |
3674 | free(cg); | |
3675 | return rv; | |
3676 | } | |
3677 | ||
0ecddf02 | 3678 | static uint64_t get_reaper_start_time(pid_t pid) |
9ac264cf | 3679 | { |
9ac264cf | 3680 | int ret; |
0ecddf02 CB |
3681 | FILE *f; |
3682 | uint64_t starttime; | |
3683 | /* strlen("/proc/") = 6 | |
3684 | * + | |
3685 | * LXCFS_NUMSTRLEN64 | |
3686 | * + | |
3687 | * strlen("/stat") = 5 | |
3688 | * + | |
3689 | * \0 = 1 | |
3690 | * */ | |
3691 | #define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1) | |
3692 | char path[__PROC_PID_STAT_LEN]; | |
9ac264cf JB |
3693 | pid_t qpid; |
3694 | ||
3695 | qpid = lookup_initpid_in_store(pid); | |
0ecddf02 CB |
3696 | if (qpid <= 0) { |
3697 | /* Caller can check for EINVAL on 0. */ | |
3698 | errno = EINVAL; | |
9ac264cf | 3699 | return 0; |
0ecddf02 | 3700 | } |
9ac264cf | 3701 | |
0ecddf02 CB |
3702 | ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid); |
3703 | if (ret < 0 || ret >= __PROC_PID_STAT_LEN) { | |
3704 | /* Caller can check for EINVAL on 0. */ | |
3705 | errno = EINVAL; | |
9ac264cf | 3706 | return 0; |
0ecddf02 | 3707 | } |
9ac264cf | 3708 | |
0ecddf02 CB |
3709 | f = fopen(path, "r"); |
3710 | if (!f) { | |
3711 | /* Caller can check for EINVAL on 0. */ | |
3712 | errno = EINVAL; | |
9ac264cf | 3713 | return 0; |
0ecddf02 | 3714 | } |
9ac264cf | 3715 | |
0ecddf02 CB |
3716 | /* Note that the *scanf() argument supression requires that length |
3717 | * modifiers such as "l" are omitted. Otherwise some compilers will yell | |
3718 | * at us. It's like telling someone you're not married and then asking | |
3719 | * if you can bring your wife to the party. | |
3720 | */ | |
3721 | ret = fscanf(f, "%*d " /* (1) pid %d */ | |
3722 | "%*s " /* (2) comm %s */ | |
3723 | "%*c " /* (3) state %c */ | |
3724 | "%*d " /* (4) ppid %d */ | |
3725 | "%*d " /* (5) pgrp %d */ | |
3726 | "%*d " /* (6) session %d */ | |
3727 | "%*d " /* (7) tty_nr %d */ | |
3728 | "%*d " /* (8) tpgid %d */ | |
3729 | "%*u " /* (9) flags %u */ | |
3730 | "%*u " /* (10) minflt %lu */ | |
3731 | "%*u " /* (11) cminflt %lu */ | |
3732 | "%*u " /* (12) majflt %lu */ | |
3733 | "%*u " /* (13) cmajflt %lu */ | |
3734 | "%*u " /* (14) utime %lu */ | |
3735 | "%*u " /* (15) stime %lu */ | |
3736 | "%*d " /* (16) cutime %ld */ | |
3737 | "%*d " /* (17) cstime %ld */ | |
3738 | "%*d " /* (18) priority %ld */ | |
3739 | "%*d " /* (19) nice %ld */ | |
3740 | "%*d " /* (20) num_threads %ld */ | |
3741 | "%*d " /* (21) itrealvalue %ld */ | |
3742 | "%" PRIu64, /* (22) starttime %llu */ | |
3743 | &starttime); | |
3744 | if (ret != 1) { | |
3745 | fclose(f); | |
3746 | /* Caller can check for EINVAL on 0. */ | |
3747 | errno = EINVAL; | |
3748 | return 0; | |
3749 | } | |
3750 | ||
3751 | fclose(f); | |
3752 | ||
3753 | errno = 0; | |
3754 | return starttime; | |
3755 | } | |
3756 | ||
3757 | static uint64_t get_reaper_start_time_in_sec(pid_t pid) | |
3758 | { | |
3759 | uint64_t clockticks; | |
3760 | int64_t ticks_per_sec; | |
3761 | ||
3762 | clockticks = get_reaper_start_time(pid); | |
3763 | if (clockticks == 0 && errno == EINVAL) { | |
3764 | lxcfs_debug("failed to retrieve start time of pid %d\n", pid); | |
3765 | return 0; | |
3766 | } | |
3767 | ||
3768 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3769 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
3770 | lxcfs_debug( | |
3771 | "%s\n", | |
3772 | "failed to determine number of clock ticks in a second"); | |
3773 | return 0; | |
3774 | } | |
3775 | ||
3776 | return (clockticks /= ticks_per_sec); | |
3777 | } | |
3778 | ||
3779 | static uint64_t get_reaper_age(pid_t pid) | |
3780 | { | |
3781 | uint64_t procstart, uptime, procage; | |
3782 | ||
3783 | /* We need to substract the time the process has started since system | |
3784 | * boot minus the time when the system has started to get the actual | |
3785 | * reaper age. | |
3786 | */ | |
3787 | procstart = get_reaper_start_time_in_sec(pid); | |
3788 | procage = procstart; | |
3789 | if (procstart > 0) { | |
3790 | int ret; | |
3791 | struct timespec spec; | |
3792 | ||
3793 | ret = clock_gettime(CLOCK_BOOTTIME, &spec); | |
3794 | if (ret < 0) | |
3795 | return 0; | |
3796 | /* We could make this more precise here by using the tv_nsec | |
3797 | * field in the timespec struct and convert it to milliseconds | |
3798 | * and then create a double for the seconds and milliseconds but | |
3799 | * that seems more work than it is worth. | |
3800 | */ | |
3801 | uptime = spec.tv_sec; | |
3802 | procage = uptime - procstart; | |
3803 | } | |
3804 | ||
3805 | return procage; | |
3806 | } | |
3807 | ||
8be92dd1 JS |
3808 | /* |
3809 | * Returns 0 on success. | |
3810 | * It is the caller's responsibility to free `return_usage`, unless this | |
3811 | * function returns an error. | |
3812 | */ | |
3813 | static int read_cpuacct_usage_all(char *cg, char *cpuset, struct cpuacct_usage **return_usage) | |
3814 | { | |
3815 | int cpucount = get_nprocs(); | |
3816 | struct cpuacct_usage *cpu_usage; | |
3817 | int rv = 0, i, j, ret, read_pos = 0, read_cnt; | |
3818 | int cg_cpu; | |
3819 | uint64_t cg_user, cg_system; | |
3820 | int64_t ticks_per_sec; | |
3821 | char *usage_str = NULL; | |
3822 | ||
3823 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3824 | ||
3825 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
3826 | lxcfs_debug( | |
3827 | "%s\n", | |
3828 | "read_cpuacct_usage_all failed to determine number of clock ticks " | |
3829 | "in a second"); | |
3830 | return -1; | |
3831 | } | |
3832 | ||
3833 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); | |
3834 | if (!cpu_usage) | |
3835 | return -ENOMEM; | |
3836 | ||
3837 | if (!cgfs_get_value("cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
3838 | rv = -1; | |
3839 | goto err; | |
3840 | } | |
3841 | ||
3842 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) { | |
3843 | lxcfs_error("read_cpuacct_usage_all reading first line from " | |
3844 | "%s/cpuacct.usage_all failed.\n", cg); | |
3845 | rv = -1; | |
3846 | goto err; | |
3847 | } | |
3848 | ||
3849 | read_pos += read_cnt; | |
3850 | ||
3851 | for (i = 0, j = 0; i < cpucount; i++) { | |
3852 | ret = sscanf(usage_str + read_pos, "%d %lu %lu\n%n", &cg_cpu, &cg_user, | |
3853 | &cg_system, &read_cnt); | |
3854 | ||
3855 | if (ret == EOF) | |
3856 | break; | |
3857 | ||
3858 | if (ret != 3) { | |
3859 | lxcfs_error("read_cpuacct_usage_all reading from %s/cpuacct.usage_all " | |
3860 | "failed.\n", cg); | |
3861 | rv = -1; | |
3862 | goto err; | |
3863 | } | |
3864 | ||
3865 | read_pos += read_cnt; | |
3866 | ||
3867 | if (!cpu_in_cpuset(i, cpuset)) | |
3868 | continue; | |
3869 | ||
3870 | /* Convert the time from nanoseconds to USER_HZ */ | |
3871 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
3872 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
3873 | j++; | |
3874 | } | |
3875 | ||
3876 | rv = 0; | |
3877 | *return_usage = cpu_usage; | |
3878 | ||
3879 | err: | |
3880 | if (usage_str) | |
3881 | free(usage_str); | |
3882 | ||
3883 | if (rv != 0) { | |
3884 | free(cpu_usage); | |
3885 | *return_usage = NULL; | |
3886 | } | |
3887 | ||
3888 | return rv; | |
3889 | } | |
3890 | ||
f34de69a | 3891 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) |
237e200e SH |
3892 | static int proc_stat_read(char *buf, size_t size, off_t offset, |
3893 | struct fuse_file_info *fi) | |
3894 | { | |
3895 | struct fuse_context *fc = fuse_get_context(); | |
3896 | struct file_info *d = (struct file_info *)fi->fh; | |
3897 | char *cg; | |
3898 | char *cpuset = NULL; | |
3899 | char *line = NULL; | |
3900 | size_t linelen = 0, total_len = 0, rv = 0; | |
3901 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
7144f069 | 3902 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; |
237e200e | 3903 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, |
7144f069 | 3904 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; |
237e200e SH |
3905 | char cpuall[CPUALL_MAX_SIZE]; |
3906 | /* reserve for cpu all */ | |
3907 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3908 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3909 | FILE *f = NULL; | |
8be92dd1 | 3910 | struct cpuacct_usage *cg_cpu_usage = NULL; |
237e200e SH |
3911 | |
3912 | if (offset){ | |
3913 | if (offset > d->size) | |
3914 | return -EINVAL; | |
3915 | if (!d->cached) | |
3916 | return 0; | |
3917 | int left = d->size - offset; | |
3918 | total_len = left > size ? size: left; | |
3919 | memcpy(buf, d->buf + offset, total_len); | |
3920 | return total_len; | |
3921 | } | |
3922 | ||
3923 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3924 | if (initpid <= 0) | |
3925 | initpid = fc->pid; | |
3926 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3927 | if (!cg) | |
3928 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 3929 | prune_init_slice(cg); |
237e200e SH |
3930 | |
3931 | cpuset = get_cpuset(cg); | |
3932 | if (!cpuset) | |
3933 | goto err; | |
3934 | ||
8be92dd1 JS |
3935 | /* |
3936 | * Read cpuacct.usage_all for all CPUs. | |
3937 | * If the cpuacct cgroup is present, it is used to calculate the container's | |
3938 | * CPU usage. If not, values from the host's /proc/stat are used. | |
3939 | */ | |
3940 | if (read_cpuacct_usage_all(cg, cpuset, &cg_cpu_usage) != 0) { | |
3941 | lxcfs_debug("%s\n", "proc_stat_read failed to read from cpuacct, " | |
3942 | "falling back to the host's /proc/stat"); | |
3943 | } | |
3944 | ||
237e200e SH |
3945 | f = fopen("/proc/stat", "r"); |
3946 | if (!f) | |
3947 | goto err; | |
3948 | ||
3949 | //skip first line | |
3950 | if (getline(&line, &linelen, f) < 0) { | |
b8defc3d | 3951 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); |
237e200e SH |
3952 | goto err; |
3953 | } | |
3954 | ||
3955 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3956 | ssize_t l; |
237e200e SH |
3957 | int cpu; |
3958 | char cpu_char[10]; /* That's a lot of cores */ | |
3959 | char *c; | |
8be92dd1 JS |
3960 | uint64_t all_used, cg_used, new_idle; |
3961 | int ret; | |
237e200e | 3962 | |
b4665ce0 SH |
3963 | if (strlen(line) == 0) |
3964 | continue; | |
237e200e SH |
3965 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
3966 | /* not a ^cpuN line containing a number N, just print it */ | |
9502bae2 | 3967 | l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3968 | if (l < 0) { |
3969 | perror("Error writing to cache"); | |
3970 | rv = 0; | |
3971 | goto err; | |
3972 | } | |
3973 | if (l >= cache_size) { | |
b8defc3d | 3974 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3975 | rv = 0; |
3976 | goto err; | |
3977 | } | |
3978 | cache += l; | |
3979 | cache_size -= l; | |
3980 | total_len += l; | |
3981 | continue; | |
3982 | } | |
3983 | ||
3984 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3985 | continue; | |
3986 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3987 | continue; | |
3988 | curcpu ++; | |
3989 | ||
8be92dd1 | 3990 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
7144f069 CB |
3991 | &user, |
3992 | &nice, | |
3993 | &system, | |
3994 | &idle, | |
3995 | &iowait, | |
3996 | &irq, | |
3997 | &softirq, | |
3998 | &steal, | |
3999 | &guest, | |
8be92dd1 JS |
4000 | &guest_nice); |
4001 | ||
4002 | if (ret != 10 || !cg_cpu_usage) { | |
4003 | c = strchr(line, ' '); | |
4004 | if (!c) | |
4005 | continue; | |
4006 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
4007 | if (l < 0) { | |
4008 | perror("Error writing to cache"); | |
4009 | rv = 0; | |
4010 | goto err; | |
4011 | ||
4012 | } | |
4013 | if (l >= cache_size) { | |
4014 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4015 | rv = 0; | |
4016 | goto err; | |
4017 | } | |
4018 | ||
4019 | cache += l; | |
4020 | cache_size -= l; | |
4021 | total_len += l; | |
4022 | ||
4023 | if (ret != 10) | |
4024 | continue; | |
4025 | } | |
4026 | ||
4027 | if (cg_cpu_usage) { | |
4028 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
4029 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
4030 | ||
4031 | if (all_used >= cg_used) { | |
4032 | new_idle = idle + (all_used - cg_used); | |
4033 | ||
4034 | } else { | |
4035 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
4036 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
4037 | curcpu, cg, all_used, cg_used); | |
4038 | new_idle = idle; | |
4039 | } | |
4040 | ||
4041 | l = snprintf(cache, cache_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
4042 | curcpu, cg_cpu_usage[curcpu].user, cg_cpu_usage[curcpu].system, | |
4043 | new_idle); | |
4044 | ||
4045 | if (l < 0) { | |
4046 | perror("Error writing to cache"); | |
4047 | rv = 0; | |
4048 | goto err; | |
4049 | ||
4050 | } | |
4051 | if (l >= cache_size) { | |
4052 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4053 | rv = 0; | |
4054 | goto err; | |
4055 | } | |
4056 | ||
4057 | cache += l; | |
4058 | cache_size -= l; | |
4059 | total_len += l; | |
4060 | ||
4061 | user_sum += cg_cpu_usage[curcpu].user; | |
4062 | system_sum += cg_cpu_usage[curcpu].system; | |
4063 | idle_sum += new_idle; | |
4064 | ||
4065 | } else { | |
4066 | user_sum += user; | |
4067 | nice_sum += nice; | |
4068 | system_sum += system; | |
4069 | idle_sum += idle; | |
4070 | iowait_sum += iowait; | |
4071 | irq_sum += irq; | |
4072 | softirq_sum += softirq; | |
4073 | steal_sum += steal; | |
4074 | guest_sum += guest; | |
4075 | guest_nice_sum += guest_nice; | |
4076 | } | |
237e200e SH |
4077 | } |
4078 | ||
4079 | cache = d->buf; | |
4080 | ||
7144f069 CB |
4081 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
4082 | user_sum, | |
4083 | nice_sum, | |
4084 | system_sum, | |
4085 | idle_sum, | |
4086 | iowait_sum, | |
4087 | irq_sum, | |
4088 | softirq_sum, | |
4089 | steal_sum, | |
4090 | guest_sum, | |
4091 | guest_nice_sum); | |
4092 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
237e200e SH |
4093 | memcpy(cache, cpuall, cpuall_len); |
4094 | cache += cpuall_len; | |
7144f069 | 4095 | } else { |
237e200e | 4096 | /* shouldn't happen */ |
b8defc3d | 4097 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); |
237e200e SH |
4098 | cpuall_len = 0; |
4099 | } | |
4100 | ||
4101 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
4102 | total_len += cpuall_len; | |
4103 | d->cached = 1; | |
4104 | d->size = total_len; | |
7144f069 CB |
4105 | if (total_len > size) |
4106 | total_len = size; | |
237e200e SH |
4107 | |
4108 | memcpy(buf, d->buf, total_len); | |
4109 | rv = total_len; | |
4110 | ||
4111 | err: | |
4112 | if (f) | |
4113 | fclose(f); | |
8be92dd1 JS |
4114 | if (cg_cpu_usage) |
4115 | free(cg_cpu_usage); | |
237e200e SH |
4116 | free(line); |
4117 | free(cpuset); | |
4118 | free(cg); | |
4119 | return rv; | |
4120 | } | |
4121 | ||
0ecddf02 CB |
4122 | /* This function retrieves the busy time of a group of tasks by looking at |
4123 | * cpuacct.usage. Unfortunately, this only makes sense when the container has | |
4124 | * been given it's own cpuacct cgroup. If not, this function will take the busy | |
4125 | * time of all other taks that do not actually belong to the container into | |
4126 | * account as well. If someone has a clever solution for this please send a | |
4127 | * patch! | |
4128 | */ | |
237e200e SH |
4129 | static unsigned long get_reaper_busy(pid_t task) |
4130 | { | |
4131 | pid_t initpid = lookup_initpid_in_store(task); | |
4132 | char *cgroup = NULL, *usage_str = NULL; | |
4133 | unsigned long usage = 0; | |
4134 | ||
4135 | if (initpid <= 0) | |
4136 | return 0; | |
4137 | ||
4138 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
4139 | if (!cgroup) | |
4140 | goto out; | |
6d2f6996 | 4141 | prune_init_slice(cgroup); |
237e200e SH |
4142 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
4143 | goto out; | |
4144 | usage = strtoul(usage_str, NULL, 10); | |
4145 | usage /= 1000000000; | |
4146 | ||
4147 | out: | |
4148 | free(cgroup); | |
4149 | free(usage_str); | |
4150 | return usage; | |
4151 | } | |
4152 | ||
4153 | #if RELOADTEST | |
4154 | void iwashere(void) | |
4155 | { | |
237e200e SH |
4156 | int fd; |
4157 | ||
ec2b5e7c | 4158 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
4159 | if (fd >= 0) |
4160 | close(fd); | |
4161 | } | |
4162 | #endif | |
4163 | ||
4164 | /* | |
4165 | * We read /proc/uptime and reuse its second field. | |
4166 | * For the first field, we use the mtime for the reaper for | |
4167 | * the calling pid as returned by getreaperage | |
4168 | */ | |
4169 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
4170 | struct fuse_file_info *fi) | |
4171 | { | |
4172 | struct fuse_context *fc = fuse_get_context(); | |
4173 | struct file_info *d = (struct file_info *)fi->fh; | |
0ecddf02 | 4174 | unsigned long int busytime = get_reaper_busy(fc->pid); |
237e200e | 4175 | char *cache = d->buf; |
a262ddb7 | 4176 | ssize_t total_len = 0; |
0ecddf02 | 4177 | uint64_t idletime, reaperage; |
237e200e SH |
4178 | |
4179 | #if RELOADTEST | |
4180 | iwashere(); | |
4181 | #endif | |
4182 | ||
4183 | if (offset){ | |
237e200e SH |
4184 | if (!d->cached) |
4185 | return 0; | |
bbdf646b BM |
4186 | if (offset > d->size) |
4187 | return -EINVAL; | |
237e200e SH |
4188 | int left = d->size - offset; |
4189 | total_len = left > size ? size: left; | |
4190 | memcpy(buf, cache + offset, total_len); | |
4191 | return total_len; | |
4192 | } | |
4193 | ||
0ecddf02 CB |
4194 | reaperage = get_reaper_age(fc->pid); |
4195 | /* To understand why this is done, please read the comment to the | |
4196 | * get_reaper_busy() function. | |
4197 | */ | |
4198 | idletime = reaperage; | |
4199 | if (reaperage >= busytime) | |
4200 | idletime = reaperage - busytime; | |
237e200e | 4201 | |
bbdf646b BM |
4202 | total_len = snprintf(d->buf, d->buflen, "%"PRIu64".00 %"PRIu64".00\n", reaperage, idletime); |
4203 | if (total_len < 0 || total_len >= d->buflen){ | |
0ecddf02 | 4204 | lxcfs_error("%s\n", "failed to write to cache"); |
237e200e SH |
4205 | return 0; |
4206 | } | |
4207 | ||
4208 | d->size = (int)total_len; | |
4209 | d->cached = 1; | |
4210 | ||
4211 | if (total_len > size) total_len = size; | |
4212 | ||
4213 | memcpy(buf, d->buf, total_len); | |
4214 | return total_len; | |
4215 | } | |
4216 | ||
4217 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
4218 | struct fuse_file_info *fi) | |
4219 | { | |
4220 | char dev_name[72]; | |
4221 | struct fuse_context *fc = fuse_get_context(); | |
4222 | struct file_info *d = (struct file_info *)fi->fh; | |
4223 | char *cg; | |
4224 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
4225 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
4226 | unsigned long read = 0, write = 0; | |
4227 | unsigned long read_merged = 0, write_merged = 0; | |
4228 | unsigned long read_sectors = 0, write_sectors = 0; | |
4229 | unsigned long read_ticks = 0, write_ticks = 0; | |
4230 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
4231 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
4232 | char *cache = d->buf; | |
4233 | size_t cache_size = d->buflen; | |
4234 | char *line = NULL; | |
4235 | size_t linelen = 0, total_len = 0, rv = 0; | |
4236 | unsigned int major = 0, minor = 0; | |
4237 | int i = 0; | |
4238 | FILE *f = NULL; | |
4239 | ||
4240 | if (offset){ | |
4241 | if (offset > d->size) | |
4242 | return -EINVAL; | |
4243 | if (!d->cached) | |
4244 | return 0; | |
4245 | int left = d->size - offset; | |
4246 | total_len = left > size ? size: left; | |
4247 | memcpy(buf, cache + offset, total_len); | |
4248 | return total_len; | |
4249 | } | |
4250 | ||
4251 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
4252 | if (initpid <= 0) | |
4253 | initpid = fc->pid; | |
4254 | cg = get_pid_cgroup(initpid, "blkio"); | |
4255 | if (!cg) | |
4256 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 4257 | prune_init_slice(cg); |
237e200e | 4258 | |
2209fe50 | 4259 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 4260 | goto err; |
2209fe50 | 4261 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 4262 | goto err; |
2209fe50 | 4263 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 4264 | goto err; |
2209fe50 | 4265 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 4266 | goto err; |
2209fe50 | 4267 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
4268 | goto err; |
4269 | ||
4270 | ||
4271 | f = fopen("/proc/diskstats", "r"); | |
4272 | if (!f) | |
4273 | goto err; | |
4274 | ||
4275 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 4276 | ssize_t l; |
2209fe50 | 4277 | char lbuf[256]; |
237e200e SH |
4278 | |
4279 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 4280 | if (i != 3) |
237e200e | 4281 | continue; |
2209fe50 SH |
4282 | |
4283 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
4284 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
4285 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
4286 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
4287 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
4288 | read_sectors = read_sectors/512; | |
4289 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
4290 | write_sectors = write_sectors/512; | |
4291 | ||
4292 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
4293 | rd_svctm = rd_svctm/1000000; | |
4294 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
4295 | rd_wait = rd_wait/1000000; | |
4296 | read_ticks = rd_svctm + rd_wait; | |
4297 | ||
4298 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
4299 | wr_svctm = wr_svctm/1000000; | |
4300 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
4301 | wr_wait = wr_wait/1000000; | |
4302 | write_ticks = wr_svctm + wr_wait; | |
4303 | ||
4304 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
4305 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
4306 | |
4307 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
4308 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
4309 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
4310 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
4311 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
4312 | else | |
4313 | continue; | |
237e200e | 4314 | |
2209fe50 | 4315 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
4316 | if (l < 0) { |
4317 | perror("Error writing to fuse buf"); | |
4318 | rv = 0; | |
4319 | goto err; | |
4320 | } | |
4321 | if (l >= cache_size) { | |
b8defc3d | 4322 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
4323 | rv = 0; |
4324 | goto err; | |
4325 | } | |
4326 | cache += l; | |
4327 | cache_size -= l; | |
4328 | total_len += l; | |
4329 | } | |
4330 | ||
4331 | d->cached = 1; | |
4332 | d->size = total_len; | |
4333 | if (total_len > size ) total_len = size; | |
4334 | memcpy(buf, d->buf, total_len); | |
4335 | ||
4336 | rv = total_len; | |
4337 | err: | |
4338 | free(cg); | |
4339 | if (f) | |
4340 | fclose(f); | |
4341 | free(line); | |
4342 | free(io_serviced_str); | |
4343 | free(io_merged_str); | |
4344 | free(io_service_bytes_str); | |
4345 | free(io_wait_time_str); | |
4346 | free(io_service_time_str); | |
4347 | return rv; | |
4348 | } | |
4349 | ||
70dcc12e SH |
4350 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
4351 | struct fuse_file_info *fi) | |
4352 | { | |
4353 | struct fuse_context *fc = fuse_get_context(); | |
4354 | struct file_info *d = (struct file_info *)fi->fh; | |
4355 | char *cg = NULL; | |
018246ff | 4356 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL; |
70dcc12e | 4357 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; |
a262ddb7 CB |
4358 | ssize_t total_len = 0, rv = 0; |
4359 | ssize_t l = 0; | |
70dcc12e SH |
4360 | char *cache = d->buf; |
4361 | ||
4362 | if (offset) { | |
4363 | if (offset > d->size) | |
4364 | return -EINVAL; | |
4365 | if (!d->cached) | |
4366 | return 0; | |
4367 | int left = d->size - offset; | |
4368 | total_len = left > size ? size: left; | |
4369 | memcpy(buf, cache + offset, total_len); | |
4370 | return total_len; | |
4371 | } | |
4372 | ||
4373 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
4374 | if (initpid <= 0) | |
4375 | initpid = fc->pid; | |
4376 | cg = get_pid_cgroup(initpid, "memory"); | |
4377 | if (!cg) | |
4378 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 4379 | prune_init_slice(cg); |
70dcc12e | 4380 | |
018246ff | 4381 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
70dcc12e SH |
4382 | |
4383 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
4384 | goto err; | |
4385 | ||
70dcc12e SH |
4386 | memusage = strtoul(memusage_str, NULL, 10); |
4387 | ||
4388 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
4389 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
4390 | ||
018246ff | 4391 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
70dcc12e SH |
4392 | memswusage = strtoul(memswusage_str, NULL, 10); |
4393 | ||
70dcc12e SH |
4394 | swap_total = (memswlimit - memlimit) / 1024; |
4395 | swap_free = (memswusage - memusage) / 1024; | |
4396 | } | |
4397 | ||
4398 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
4399 | ||
4400 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
4401 | if (!memswlimit) { | |
4402 | char *line = NULL; | |
4403 | size_t linelen = 0; | |
4404 | FILE *f = fopen("/proc/meminfo", "r"); | |
4405 | ||
4406 | if (!f) | |
4407 | goto err; | |
4408 | ||
4409 | while (getline(&line, &linelen, f) != -1) { | |
4410 | if (startswith(line, "SwapTotal:")) { | |
4411 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
4412 | } else if (startswith(line, "SwapFree:")) { | |
4413 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
4414 | } | |
4415 | } | |
4416 | ||
4417 | free(line); | |
4418 | fclose(f); | |
4419 | } | |
4420 | ||
4421 | if (swap_total > 0) { | |
a262ddb7 CB |
4422 | l = snprintf(d->buf + total_len, d->size - total_len, |
4423 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
4424 | swap_total, swap_free); | |
4425 | total_len += l; | |
70dcc12e SH |
4426 | } |
4427 | ||
a262ddb7 | 4428 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
4429 | perror("Error writing to cache"); |
4430 | rv = 0; | |
4431 | goto err; | |
4432 | } | |
4433 | ||
4434 | d->cached = 1; | |
4435 | d->size = (int)total_len; | |
4436 | ||
4437 | if (total_len > size) total_len = size; | |
4438 | memcpy(buf, d->buf, total_len); | |
4439 | rv = total_len; | |
4440 | ||
4441 | err: | |
4442 | free(cg); | |
4443 | free(memswlimit_str); | |
4444 | free(memlimit_str); | |
4445 | free(memusage_str); | |
4446 | free(memswusage_str); | |
70dcc12e SH |
4447 | return rv; |
4448 | } | |
6db4f7a3 | 4449 | /* |
4450 | * Find the process pid from cgroup path. | |
4451 | * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid. | |
4452 | * @pid_buf : put pid to pid_buf. | |
4453 | * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ... | |
4454 | * @depth : the depth of cgroup in container. | |
4455 | * @sum : return the number of pid. | |
4456 | * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu | |
4457 | */ | |
4458 | static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd) | |
4459 | { | |
4460 | DIR *dir; | |
4461 | int fd; | |
4462 | struct dirent *file; | |
4463 | FILE *f = NULL; | |
4464 | size_t linelen = 0; | |
4465 | char *line = NULL; | |
4466 | int pd; | |
4467 | char *path_dir, *path; | |
4468 | char **pid; | |
4469 | ||
4470 | /* path = dpath + "/cgroup.procs" + /0 */ | |
4471 | do { | |
4472 | path = malloc(strlen(dpath) + 20); | |
4473 | } while (!path); | |
4474 | ||
4475 | strcpy(path, dpath); | |
4476 | fd = openat(cfd, path, O_RDONLY); | |
4477 | if (fd < 0) | |
4478 | goto out; | |
4479 | ||
4480 | dir = fdopendir(fd); | |
4481 | if (dir == NULL) { | |
4482 | close(fd); | |
4483 | goto out; | |
4484 | } | |
4485 | ||
4486 | while (((file = readdir(dir)) != NULL) && depth > 0) { | |
4487 | if (strncmp(file->d_name, ".", 1) == 0) | |
4488 | continue; | |
4489 | if (strncmp(file->d_name, "..", 1) == 0) | |
4490 | continue; | |
4491 | if (file->d_type == DT_DIR) { | |
4492 | /* path + '/' + d_name +/0 */ | |
4493 | do { | |
4494 | path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name)); | |
4495 | } while (!path_dir); | |
4496 | strcpy(path_dir, path); | |
4497 | strcat(path_dir, "/"); | |
4498 | strcat(path_dir, file->d_name); | |
4499 | pd = depth - 1; | |
4500 | sum = calc_pid(pid_buf, path_dir, pd, sum, cfd); | |
4501 | free(path_dir); | |
4502 | } | |
4503 | } | |
4504 | closedir(dir); | |
4505 | ||
4506 | strcat(path, "/cgroup.procs"); | |
4507 | fd = openat(cfd, path, O_RDONLY); | |
4508 | if (fd < 0) | |
4509 | goto out; | |
4510 | ||
4511 | f = fdopen(fd, "r"); | |
4512 | if (!f) { | |
4513 | close(fd); | |
4514 | goto out; | |
4515 | } | |
4516 | ||
4517 | while (getline(&line, &linelen, f) != -1) { | |
4518 | do { | |
4519 | pid = realloc(*pid_buf, sizeof(char *) * (sum + 1)); | |
4520 | } while (!pid); | |
4521 | *pid_buf = pid; | |
4522 | do { | |
4523 | *(*pid_buf + sum) = malloc(strlen(line) + 1); | |
4524 | } while (*(*pid_buf + sum) == NULL); | |
4525 | strcpy(*(*pid_buf + sum), line); | |
4526 | sum++; | |
4527 | } | |
4528 | fclose(f); | |
4529 | out: | |
832904c1 JS |
4530 | if (line) |
4531 | free(line); | |
6db4f7a3 | 4532 | free(path); |
4533 | return sum; | |
4534 | } | |
4535 | /* | |
4536 | * calc_load calculates the load according to the following formula: | |
4537 | * load1 = load0 * exp + active * (1 - exp) | |
4538 | * | |
4539 | * @load1: the new loadavg. | |
4540 | * @load0: the former loadavg. | |
4541 | * @active: the total number of running pid at this moment. | |
4542 | * @exp: the fixed-point defined in the beginning. | |
4543 | */ | |
4544 | static unsigned long | |
4545 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
4546 | { | |
4547 | unsigned long newload; | |
4548 | ||
4549 | active = active > 0 ? active * FIXED_1 : 0; | |
4550 | newload = load * exp + active * (FIXED_1 - exp); | |
4551 | if (active >= load) | |
4552 | newload += FIXED_1 - 1; | |
4553 | ||
4554 | return newload / FIXED_1; | |
4555 | } | |
4556 | ||
4557 | /* | |
4558 | * Return 0 means that container p->cg is closed. | |
4559 | * Return -1 means that error occurred in refresh. | |
4560 | * Positive num equals the total number of pid. | |
4561 | */ | |
4562 | static int refresh_load(struct load_node *p, char *path) | |
4563 | { | |
4564 | FILE *f = NULL; | |
4565 | char **idbuf; | |
4566 | char proc_path[256]; | |
4567 | int i, ret, run_pid = 0, total_pid = 0, last_pid = 0; | |
4568 | char *line = NULL; | |
4569 | size_t linelen = 0; | |
4570 | int sum, length; | |
4571 | DIR *dp; | |
4572 | struct dirent *file; | |
4573 | ||
4574 | do { | |
4575 | idbuf = malloc(sizeof(char *)); | |
4576 | } while (!idbuf); | |
4577 | sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd); | |
4578 | /* normal exit */ | |
4579 | if (sum == 0) | |
4580 | goto out; | |
4581 | ||
4582 | for (i = 0; i < sum; i++) { | |
4583 | /*clean up '\n' */ | |
4584 | length = strlen(idbuf[i])-1; | |
4585 | idbuf[i][length] = '\0'; | |
4586 | ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]); | |
4587 | if (ret < 0 || ret > 255) { | |
4588 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
4589 | i = sum; | |
4590 | sum = -1; | |
4591 | goto err_out; | |
4592 | } | |
4593 | ||
4594 | dp = opendir(proc_path); | |
4595 | if (!dp) { | |
4596 | lxcfs_error("%s\n", "Open proc_path failed in refresh_load."); | |
4597 | continue; | |
4598 | } | |
4599 | while ((file = readdir(dp)) != NULL) { | |
4600 | if (strncmp(file->d_name, ".", 1) == 0) | |
4601 | continue; | |
4602 | if (strncmp(file->d_name, "..", 1) == 0) | |
4603 | continue; | |
4604 | total_pid++; | |
4605 | /* We make the biggest pid become last_pid.*/ | |
4606 | ret = atof(file->d_name); | |
4607 | last_pid = (ret > last_pid) ? ret : last_pid; | |
4608 | ||
4609 | ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name); | |
4610 | if (ret < 0 || ret > 255) { | |
4611 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
4612 | i = sum; | |
4613 | sum = -1; | |
4614 | closedir(dp); | |
4615 | goto err_out; | |
4616 | } | |
4617 | f = fopen(proc_path, "r"); | |
4618 | if (f != NULL) { | |
4619 | while (getline(&line, &linelen, f) != -1) { | |
4620 | /* Find State */ | |
4621 | if ((line[0] == 'S') && (line[1] == 't')) | |
4622 | break; | |
4623 | } | |
4624 | if ((line[7] == 'R') || (line[7] == 'D')) | |
4625 | run_pid++; | |
4626 | fclose(f); | |
4627 | } | |
4628 | } | |
4629 | closedir(dp); | |
4630 | } | |
4631 | /*Calculate the loadavg.*/ | |
4632 | p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid); | |
4633 | p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid); | |
4634 | p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid); | |
4635 | p->run_pid = run_pid; | |
4636 | p->total_pid = total_pid; | |
4637 | p->last_pid = last_pid; | |
4638 | ||
4639 | free(line); | |
4640 | err_out: | |
4641 | for (; i > 0; i--) | |
4642 | free(idbuf[i-1]); | |
4643 | out: | |
4644 | free(idbuf); | |
4645 | return sum; | |
4646 | } | |
4647 | /* | |
4648 | * Traverse the hash table and update it. | |
4649 | */ | |
4650 | void *load_begin(void *arg) | |
4651 | { | |
4652 | ||
4653 | char *path = NULL; | |
4654 | int i, sum, length, ret; | |
4655 | struct load_node *f; | |
4656 | int first_node; | |
4657 | clock_t time1, time2; | |
4658 | ||
4659 | while (1) { | |
a83618e2 JS |
4660 | if (loadavg_stop == 1) |
4661 | return NULL; | |
4662 | ||
6db4f7a3 | 4663 | time1 = clock(); |
4664 | for (i = 0; i < LOAD_SIZE; i++) { | |
4665 | pthread_mutex_lock(&load_hash[i].lock); | |
4666 | if (load_hash[i].next == NULL) { | |
4667 | pthread_mutex_unlock(&load_hash[i].lock); | |
4668 | continue; | |
4669 | } | |
4670 | f = load_hash[i].next; | |
4671 | first_node = 1; | |
4672 | while (f) { | |
4673 | length = strlen(f->cg) + 2; | |
4674 | do { | |
4675 | /* strlen(f->cg) + '.' or '' + \0 */ | |
4676 | path = malloc(length); | |
4677 | } while (!path); | |
4678 | ||
4679 | ret = snprintf(path, length, "%s%s", *(f->cg) == '/' ? "." : "", f->cg); | |
4680 | if (ret < 0 || ret > length - 1) { | |
4681 | /* snprintf failed, ignore the node.*/ | |
4682 | lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg); | |
4683 | goto out; | |
4684 | } | |
4685 | sum = refresh_load(f, path); | |
4686 | if (sum == 0) { | |
4687 | f = del_node(f, i); | |
4688 | } else { | |
4689 | out: f = f->next; | |
4690 | } | |
4691 | free(path); | |
4692 | /* load_hash[i].lock locks only on the first node.*/ | |
4693 | if (first_node == 1) { | |
4694 | first_node = 0; | |
4695 | pthread_mutex_unlock(&load_hash[i].lock); | |
4696 | } | |
4697 | } | |
4698 | } | |
a83618e2 JS |
4699 | |
4700 | if (loadavg_stop == 1) | |
4701 | return NULL; | |
4702 | ||
6db4f7a3 | 4703 | time2 = clock(); |
4704 | usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC)); | |
4705 | } | |
4706 | } | |
4707 | ||
4708 | static int proc_loadavg_read(char *buf, size_t size, off_t offset, | |
4709 | struct fuse_file_info *fi) | |
4710 | { | |
4711 | struct fuse_context *fc = fuse_get_context(); | |
4712 | struct file_info *d = (struct file_info *)fi->fh; | |
4713 | pid_t initpid; | |
4714 | char *cg; | |
4715 | size_t total_len = 0; | |
4716 | char *cache = d->buf; | |
4717 | struct load_node *n; | |
4718 | int hash; | |
01d88ede | 4719 | int cfd, rv = 0; |
6db4f7a3 | 4720 | unsigned long a, b, c; |
4721 | ||
4722 | if (offset) { | |
4723 | if (offset > d->size) | |
4724 | return -EINVAL; | |
4725 | if (!d->cached) | |
4726 | return 0; | |
4727 | int left = d->size - offset; | |
4728 | total_len = left > size ? size : left; | |
4729 | memcpy(buf, cache + offset, total_len); | |
4730 | return total_len; | |
4731 | } | |
4732 | if (!loadavg) | |
4733 | return read_file("/proc/loadavg", buf, size, d); | |
4734 | ||
4735 | initpid = lookup_initpid_in_store(fc->pid); | |
4736 | if (initpid <= 0) | |
4737 | initpid = fc->pid; | |
4738 | cg = get_pid_cgroup(initpid, "cpu"); | |
4739 | if (!cg) | |
4740 | return read_file("/proc/loadavg", buf, size, d); | |
4741 | ||
4742 | prune_init_slice(cg); | |
4743 | hash = calc_hash(cg); | |
4744 | n = locate_node(cg, hash); | |
4745 | ||
4746 | /* First time */ | |
4747 | if (n == NULL) { | |
4748 | if (!find_mounted_controller("cpu", &cfd)) { | |
4749 | /* | |
4750 | * In locate_node() above, pthread_rwlock_unlock() isn't used | |
4751 | * because delete is not allowed before read has ended. | |
4752 | */ | |
4753 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
01d88ede JS |
4754 | rv = 0; |
4755 | goto err; | |
6db4f7a3 | 4756 | } |
4757 | do { | |
4758 | n = malloc(sizeof(struct load_node)); | |
4759 | } while (!n); | |
4760 | ||
4761 | do { | |
4762 | n->cg = malloc(strlen(cg)+1); | |
4763 | } while (!n->cg); | |
4764 | strcpy(n->cg, cg); | |
4765 | n->avenrun[0] = 0; | |
4766 | n->avenrun[1] = 0; | |
4767 | n->avenrun[2] = 0; | |
4768 | n->run_pid = 0; | |
4769 | n->total_pid = 1; | |
4770 | n->last_pid = initpid; | |
4771 | n->cfd = cfd; | |
4772 | insert_node(&n, hash); | |
4773 | } | |
4774 | a = n->avenrun[0] + (FIXED_1/200); | |
4775 | b = n->avenrun[1] + (FIXED_1/200); | |
4776 | c = n->avenrun[2] + (FIXED_1/200); | |
4777 | total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n", | |
4778 | LOAD_INT(a), LOAD_FRAC(a), | |
4779 | LOAD_INT(b), LOAD_FRAC(b), | |
4780 | LOAD_INT(c), LOAD_FRAC(c), | |
4781 | n->run_pid, n->total_pid, n->last_pid); | |
4782 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
4783 | if (total_len < 0 || total_len >= d->buflen) { | |
4784 | lxcfs_error("%s\n", "Failed to write to cache"); | |
01d88ede JS |
4785 | rv = 0; |
4786 | goto err; | |
6db4f7a3 | 4787 | } |
4788 | d->size = (int)total_len; | |
4789 | d->cached = 1; | |
4790 | ||
4791 | if (total_len > size) | |
4792 | total_len = size; | |
4793 | memcpy(buf, d->buf, total_len); | |
01d88ede JS |
4794 | rv = total_len; |
4795 | ||
4796 | err: | |
4797 | free(cg); | |
4798 | return rv; | |
6db4f7a3 | 4799 | } |
4800 | /* Return a positive number on success, return 0 on failure.*/ | |
4801 | pthread_t load_daemon(int load_use) | |
4802 | { | |
4803 | int ret; | |
4804 | pthread_t pid; | |
4805 | ||
4806 | ret = init_load(); | |
4807 | if (ret == -1) { | |
4808 | lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!"); | |
4809 | return 0; | |
4810 | } | |
4811 | ret = pthread_create(&pid, NULL, load_begin, NULL); | |
4812 | if (ret != 0) { | |
4813 | lxcfs_error("%s\n", "Create pthread fails in load_daemon!"); | |
4814 | load_free(); | |
4815 | return 0; | |
4816 | } | |
4817 | /* use loadavg, here loadavg = 1*/ | |
4818 | loadavg = load_use; | |
4819 | return pid; | |
4820 | } | |
70dcc12e | 4821 | |
a83618e2 JS |
4822 | /* Returns 0 on success. */ |
4823 | int stop_load_daemon(pthread_t pid) | |
4824 | { | |
4825 | int s; | |
4826 | ||
4827 | /* Signal the thread to gracefully stop */ | |
4828 | loadavg_stop = 1; | |
4829 | ||
4830 | s = pthread_join(pid, NULL); /* Make sure sub thread has been canceled. */ | |
4831 | if (s != 0) { | |
4832 | lxcfs_error("%s\n", "stop_load_daemon error: failed to join"); | |
4833 | return -1; | |
4834 | } | |
4835 | ||
4836 | load_free(); | |
4837 | loadavg_stop = 0; | |
4838 | ||
4839 | return 0; | |
4840 | } | |
4841 | ||
237e200e SH |
4842 | static off_t get_procfile_size(const char *which) |
4843 | { | |
4844 | FILE *f = fopen(which, "r"); | |
4845 | char *line = NULL; | |
4846 | size_t len = 0; | |
4847 | ssize_t sz, answer = 0; | |
4848 | if (!f) | |
4849 | return 0; | |
4850 | ||
4851 | while ((sz = getline(&line, &len, f)) != -1) | |
4852 | answer += sz; | |
4853 | fclose (f); | |
4854 | free(line); | |
4855 | ||
4856 | return answer; | |
4857 | } | |
4858 | ||
4859 | int proc_getattr(const char *path, struct stat *sb) | |
4860 | { | |
4861 | struct timespec now; | |
4862 | ||
4863 | memset(sb, 0, sizeof(struct stat)); | |
4864 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
4865 | return -EINVAL; | |
4866 | sb->st_uid = sb->st_gid = 0; | |
4867 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
4868 | if (strcmp(path, "/proc") == 0) { | |
4869 | sb->st_mode = S_IFDIR | 00555; | |
4870 | sb->st_nlink = 2; | |
4871 | return 0; | |
4872 | } | |
4873 | if (strcmp(path, "/proc/meminfo") == 0 || | |
4874 | strcmp(path, "/proc/cpuinfo") == 0 || | |
4875 | strcmp(path, "/proc/uptime") == 0 || | |
4876 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e | 4877 | strcmp(path, "/proc/diskstats") == 0 || |
46be8eed | 4878 | strcmp(path, "/proc/swaps") == 0 || |
4879 | strcmp(path, "/proc/loadavg") == 0) { | |
237e200e SH |
4880 | sb->st_size = 0; |
4881 | sb->st_mode = S_IFREG | 00444; | |
4882 | sb->st_nlink = 1; | |
4883 | return 0; | |
4884 | } | |
4885 | ||
4886 | return -ENOENT; | |
4887 | } | |
4888 | ||
4889 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4890 | struct fuse_file_info *fi) | |
4891 | { | |
d639f863 CB |
4892 | if (filler(buf, ".", NULL, 0) != 0 || |
4893 | filler(buf, "..", NULL, 0) != 0 || | |
4894 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4895 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4896 | filler(buf, "stat", NULL, 0) != 0 || | |
4897 | filler(buf, "uptime", NULL, 0) != 0 || | |
4898 | filler(buf, "diskstats", NULL, 0) != 0 || | |
46be8eed | 4899 | filler(buf, "swaps", NULL, 0) != 0 || |
4900 | filler(buf, "loadavg", NULL, 0) != 0) | |
237e200e SH |
4901 | return -EINVAL; |
4902 | return 0; | |
4903 | } | |
4904 | ||
4905 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4906 | { | |
4907 | int type = -1; | |
4908 | struct file_info *info; | |
4909 | ||
4910 | if (strcmp(path, "/proc/meminfo") == 0) | |
4911 | type = LXC_TYPE_PROC_MEMINFO; | |
4912 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4913 | type = LXC_TYPE_PROC_CPUINFO; | |
4914 | else if (strcmp(path, "/proc/uptime") == 0) | |
4915 | type = LXC_TYPE_PROC_UPTIME; | |
4916 | else if (strcmp(path, "/proc/stat") == 0) | |
4917 | type = LXC_TYPE_PROC_STAT; | |
4918 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4919 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
4920 | else if (strcmp(path, "/proc/swaps") == 0) |
4921 | type = LXC_TYPE_PROC_SWAPS; | |
46be8eed | 4922 | else if (strcmp(path, "/proc/loadavg") == 0) |
4923 | type = LXC_TYPE_PROC_LOADAVG; | |
237e200e SH |
4924 | if (type == -1) |
4925 | return -ENOENT; | |
4926 | ||
4927 | info = malloc(sizeof(*info)); | |
4928 | if (!info) | |
4929 | return -ENOMEM; | |
4930 | ||
4931 | memset(info, 0, sizeof(*info)); | |
4932 | info->type = type; | |
4933 | ||
4934 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4935 | do { | |
4936 | info->buf = malloc(info->buflen); | |
4937 | } while (!info->buf); | |
4938 | memset(info->buf, 0, info->buflen); | |
4939 | /* set actual size to buffer size */ | |
4940 | info->size = info->buflen; | |
4941 | ||
4942 | fi->fh = (unsigned long)info; | |
4943 | return 0; | |
4944 | } | |
4945 | ||
bddbb106 SH |
4946 | int proc_access(const char *path, int mask) |
4947 | { | |
e7849aa3 CB |
4948 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) |
4949 | return 0; | |
4950 | ||
bddbb106 SH |
4951 | /* these are all read-only */ |
4952 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 4953 | return -EACCES; |
bddbb106 SH |
4954 | return 0; |
4955 | } | |
4956 | ||
237e200e SH |
4957 | int proc_release(const char *path, struct fuse_file_info *fi) |
4958 | { | |
43215927 | 4959 | do_release_file_info(fi); |
237e200e SH |
4960 | return 0; |
4961 | } | |
4962 | ||
4963 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4964 | struct fuse_file_info *fi) | |
4965 | { | |
4966 | struct file_info *f = (struct file_info *) fi->fh; | |
4967 | ||
4968 | switch (f->type) { | |
4969 | case LXC_TYPE_PROC_MEMINFO: | |
4970 | return proc_meminfo_read(buf, size, offset, fi); | |
4971 | case LXC_TYPE_PROC_CPUINFO: | |
4972 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4973 | case LXC_TYPE_PROC_UPTIME: | |
4974 | return proc_uptime_read(buf, size, offset, fi); | |
4975 | case LXC_TYPE_PROC_STAT: | |
4976 | return proc_stat_read(buf, size, offset, fi); | |
4977 | case LXC_TYPE_PROC_DISKSTATS: | |
4978 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
4979 | case LXC_TYPE_PROC_SWAPS: |
4980 | return proc_swaps_read(buf, size, offset, fi); | |
46be8eed | 4981 | case LXC_TYPE_PROC_LOADAVG: |
4982 | return proc_loadavg_read(buf, size, offset, fi); | |
237e200e SH |
4983 | default: |
4984 | return -EINVAL; | |
4985 | } | |
4986 | } | |
4987 | ||
29a73c2f CB |
4988 | /* |
4989 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
4990 | */ |
4991 | ||
4992 | static bool mkdir_p(const char *dir, mode_t mode) | |
4993 | { | |
4994 | const char *tmp = dir; | |
4995 | const char *orig = dir; | |
4996 | char *makeme; | |
4997 | ||
4998 | do { | |
4999 | dir = tmp + strspn(tmp, "/"); | |
5000 | tmp = dir + strcspn(dir, "/"); | |
5001 | makeme = strndup(orig, dir - orig); | |
5002 | if (!makeme) | |
5003 | return false; | |
5004 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
b8defc3d | 5005 | lxcfs_error("Failed to create directory '%s': %s.\n", |
29a73c2f CB |
5006 | makeme, strerror(errno)); |
5007 | free(makeme); | |
5008 | return false; | |
5009 | } | |
5010 | free(makeme); | |
5011 | } while(tmp != dir); | |
5012 | ||
5013 | return true; | |
5014 | } | |
5015 | ||
5016 | static bool umount_if_mounted(void) | |
5017 | { | |
5018 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
b8defc3d | 5019 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
5020 | return false; |
5021 | } | |
5022 | return true; | |
5023 | } | |
5024 | ||
2283e240 CB |
5025 | /* __typeof__ should be safe to use with all compilers. */ |
5026 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
5027 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
5028 | { | |
5029 | return (fs->f_type == (fs_type_magic)magic_val); | |
5030 | } | |
5031 | ||
0a4dea41 CB |
5032 | /* |
5033 | * looking at fs/proc_namespace.c, it appears we can | |
5034 | * actually expect the rootfs entry to very specifically contain | |
5035 | * " - rootfs rootfs " | |
5036 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
5037 | * the rootfs entry should always be skipped in mountinfo contents. | |
5038 | */ | |
5039 | static bool is_on_ramfs(void) | |
5040 | { | |
5041 | FILE *f; | |
5042 | char *p, *p2; | |
5043 | char *line = NULL; | |
5044 | size_t len = 0; | |
5045 | int i; | |
5046 | ||
5047 | f = fopen("/proc/self/mountinfo", "r"); | |
5048 | if (!f) | |
5049 | return false; | |
5050 | ||
5051 | while (getline(&line, &len, f) != -1) { | |
5052 | for (p = line, i = 0; p && i < 4; i++) | |
5053 | p = strchr(p + 1, ' '); | |
5054 | if (!p) | |
5055 | continue; | |
5056 | p2 = strchr(p + 1, ' '); | |
5057 | if (!p2) | |
5058 | continue; | |
5059 | *p2 = '\0'; | |
5060 | if (strcmp(p + 1, "/") == 0) { | |
5061 | // this is '/'. is it the ramfs? | |
5062 | p = strchr(p2 + 1, '-'); | |
5063 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
5064 | free(line); | |
5065 | fclose(f); | |
5066 | return true; | |
5067 | } | |
5068 | } | |
5069 | } | |
5070 | free(line); | |
5071 | fclose(f); | |
5072 | return false; | |
5073 | } | |
5074 | ||
cc309f33 | 5075 | static int pivot_enter() |
0a4dea41 | 5076 | { |
cc309f33 CB |
5077 | int ret = -1, oldroot = -1, newroot = -1; |
5078 | ||
5079 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
5080 | if (oldroot < 0) { | |
5081 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
5082 | return ret; | |
5083 | } | |
5084 | ||
5085 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
5086 | if (newroot < 0) { | |
5087 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
5088 | goto err; | |
5089 | } | |
5090 | ||
5091 | /* change into new root fs */ | |
5092 | if (fchdir(newroot) < 0) { | |
5093 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
5094 | goto err; | |
5095 | } | |
5096 | ||
0a4dea41 CB |
5097 | /* pivot_root into our new root fs */ |
5098 | if (pivot_root(".", ".") < 0) { | |
5099 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
cc309f33 | 5100 | goto err; |
0a4dea41 CB |
5101 | } |
5102 | ||
5103 | /* | |
5104 | * At this point the old-root is mounted on top of our new-root. | |
5105 | * To unmounted it we must not be chdir'd into it, so escape back | |
5106 | * to the old-root. | |
5107 | */ | |
5108 | if (fchdir(oldroot) < 0) { | |
5109 | lxcfs_error("%s\n", "Failed to enter old root."); | |
cc309f33 | 5110 | goto err; |
0a4dea41 CB |
5111 | } |
5112 | ||
5113 | if (umount2(".", MNT_DETACH) < 0) { | |
5114 | lxcfs_error("%s\n", "Failed to detach old root."); | |
cc309f33 | 5115 | goto err; |
0a4dea41 CB |
5116 | } |
5117 | ||
5118 | if (fchdir(newroot) < 0) { | |
5119 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
cc309f33 | 5120 | goto err; |
0a4dea41 CB |
5121 | } |
5122 | ||
cc309f33 CB |
5123 | ret = 0; |
5124 | ||
5125 | err: | |
5126 | if (oldroot > 0) | |
5127 | close(oldroot); | |
5128 | if (newroot > 0) | |
5129 | close(newroot); | |
5130 | ||
5131 | return ret; | |
0a4dea41 CB |
5132 | } |
5133 | ||
5134 | static int chroot_enter() | |
5135 | { | |
5136 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
5137 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
5138 | return -1; | |
5139 | } | |
5140 | ||
5141 | if (chroot(".") < 0) { | |
5142 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
5143 | return -1; | |
5144 | } | |
5145 | ||
5146 | if (chdir("/") < 0) { | |
5147 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
5148 | return -1; | |
5149 | } | |
5150 | ||
5151 | return 0; | |
5152 | } | |
5153 | ||
0232cbac | 5154 | static int permute_and_enter(void) |
29a73c2f | 5155 | { |
0a4dea41 CB |
5156 | struct statfs sb; |
5157 | ||
5158 | if (statfs("/", &sb) < 0) { | |
5159 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
cc309f33 | 5160 | return -1; |
0a4dea41 CB |
5161 | } |
5162 | ||
5163 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
5164 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
5165 | * /proc/1/mountinfo. */ | |
5166 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
5167 | return chroot_enter(); | |
29a73c2f | 5168 | |
cc309f33 | 5169 | if (pivot_enter() < 0) { |
0a4dea41 | 5170 | lxcfs_error("%s\n", "Could not perform pivot root."); |
cc309f33 | 5171 | return -1; |
29a73c2f CB |
5172 | } |
5173 | ||
cc309f33 | 5174 | return 0; |
29a73c2f CB |
5175 | } |
5176 | ||
5177 | /* Prepare our new clean root. */ | |
0232cbac | 5178 | static int permute_prepare(void) |
29a73c2f CB |
5179 | { |
5180 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
b8defc3d | 5181 | lxcfs_error("%s\n", "Failed to create directory for new root."); |
29a73c2f CB |
5182 | return -1; |
5183 | } | |
5184 | ||
5185 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5186 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5187 | return -1; |
5188 | } | |
5189 | ||
5190 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5191 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5192 | return -1; |
5193 | } | |
5194 | ||
5195 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
b8defc3d | 5196 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5197 | return -1; |
5198 | } | |
5199 | ||
5200 | return 0; | |
5201 | } | |
5202 | ||
0232cbac CB |
5203 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ |
5204 | static bool permute_root(void) | |
29a73c2f CB |
5205 | { |
5206 | /* Prepare new root. */ | |
0232cbac | 5207 | if (permute_prepare() < 0) |
29a73c2f CB |
5208 | return false; |
5209 | ||
5210 | /* Pivot into new root. */ | |
0232cbac | 5211 | if (permute_and_enter() < 0) |
29a73c2f CB |
5212 | return false; |
5213 | ||
5214 | return true; | |
5215 | } | |
5216 | ||
a257a8ee CB |
5217 | static int preserve_mnt_ns(int pid) |
5218 | { | |
5219 | int ret; | |
5220 | size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt"); | |
5221 | char path[len]; | |
5222 | ||
5223 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
5224 | if (ret < 0 || (size_t)ret >= len) | |
5225 | return -1; | |
5226 | ||
5227 | return open(path, O_RDONLY | O_CLOEXEC); | |
5228 | } | |
5229 | ||
0a4dea41 | 5230 | static bool cgfs_prepare_mounts(void) |
29a73c2f CB |
5231 | { |
5232 | if (!mkdir_p(BASEDIR, 0700)) { | |
b8defc3d | 5233 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); |
29a73c2f CB |
5234 | return false; |
5235 | } | |
480262c9 | 5236 | |
29a73c2f | 5237 | if (!umount_if_mounted()) { |
b8defc3d | 5238 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); |
480262c9 CB |
5239 | return false; |
5240 | } | |
5241 | ||
5242 | if (unshare(CLONE_NEWNS) < 0) { | |
b8defc3d | 5243 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); |
480262c9 CB |
5244 | return false; |
5245 | } | |
5246 | ||
a257a8ee CB |
5247 | cgroup_mount_ns_fd = preserve_mnt_ns(getpid()); |
5248 | if (cgroup_mount_ns_fd < 0) { | |
5249 | lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno)); | |
5250 | return false; | |
5251 | } | |
5252 | ||
480262c9 | 5253 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { |
b8defc3d | 5254 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); |
29a73c2f CB |
5255 | return false; |
5256 | } | |
480262c9 | 5257 | |
29a73c2f | 5258 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
b8defc3d | 5259 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); |
29a73c2f CB |
5260 | return false; |
5261 | } | |
480262c9 | 5262 | |
29a73c2f CB |
5263 | return true; |
5264 | } | |
5265 | ||
0a4dea41 | 5266 | static bool cgfs_mount_hierarchies(void) |
29a73c2f CB |
5267 | { |
5268 | char *target; | |
5269 | size_t clen, len; | |
5270 | int i, ret; | |
5271 | ||
5272 | for (i = 0; i < num_hierarchies; i++) { | |
5273 | char *controller = hierarchies[i]; | |
51c7ca35 | 5274 | |
29a73c2f CB |
5275 | clen = strlen(controller); |
5276 | len = strlen(BASEDIR) + clen + 2; | |
5277 | target = malloc(len); | |
5278 | if (!target) | |
5279 | return false; | |
51c7ca35 | 5280 | |
29a73c2f CB |
5281 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); |
5282 | if (ret < 0 || ret >= len) { | |
5283 | free(target); | |
5284 | return false; | |
5285 | } | |
5286 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
5287 | free(target); | |
5288 | return false; | |
5289 | } | |
51c7ca35 CB |
5290 | if (!strcmp(controller, "unified")) |
5291 | ret = mount("none", target, "cgroup2", 0, NULL); | |
5292 | else | |
5293 | ret = mount(controller, target, "cgroup", 0, controller); | |
5294 | if (ret < 0) { | |
5295 | lxcfs_error("Failed mounting cgroup %s: %s\n", controller, strerror(errno)); | |
29a73c2f CB |
5296 | free(target); |
5297 | return false; | |
5298 | } | |
5299 | ||
5300 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
5301 | if (fd_hierarchies[i] < 0) { | |
5302 | free(target); | |
5303 | return false; | |
5304 | } | |
5305 | free(target); | |
5306 | } | |
5307 | return true; | |
5308 | } | |
5309 | ||
480262c9 | 5310 | static bool cgfs_setup_controllers(void) |
29a73c2f | 5311 | { |
0a4dea41 | 5312 | if (!cgfs_prepare_mounts()) |
29a73c2f | 5313 | return false; |
29a73c2f | 5314 | |
0a4dea41 | 5315 | if (!cgfs_mount_hierarchies()) { |
b8defc3d | 5316 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); |
29a73c2f CB |
5317 | return false; |
5318 | } | |
5319 | ||
0232cbac | 5320 | if (!permute_root()) |
29a73c2f CB |
5321 | return false; |
5322 | ||
5323 | return true; | |
5324 | } | |
5325 | ||
5326 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) | |
237e200e SH |
5327 | { |
5328 | FILE *f; | |
e58dab00 CB |
5329 | char *cret, *line = NULL; |
5330 | char cwd[MAXPATHLEN]; | |
237e200e | 5331 | size_t len = 0; |
480262c9 | 5332 | int i, init_ns = -1; |
51c7ca35 | 5333 | bool found_unified = false; |
237e200e SH |
5334 | |
5335 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
b8defc3d | 5336 | lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno)); |
237e200e SH |
5337 | return; |
5338 | } | |
e58dab00 | 5339 | |
237e200e | 5340 | while (getline(&line, &len, f) != -1) { |
51c7ca35 | 5341 | char *idx, *p, *p2; |
237e200e SH |
5342 | |
5343 | p = strchr(line, ':'); | |
5344 | if (!p) | |
5345 | goto out; | |
51c7ca35 | 5346 | idx = line; |
237e200e SH |
5347 | *(p++) = '\0'; |
5348 | ||
5349 | p2 = strrchr(p, ':'); | |
5350 | if (!p2) | |
5351 | goto out; | |
5352 | *p2 = '\0'; | |
5353 | ||
a67719f6 CB |
5354 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
5355 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
5356 | * because it parses out the empty string "" and later on passes | |
5357 | * it to mount(). Let's skip such entries. | |
5358 | */ | |
51c7ca35 CB |
5359 | if (!strcmp(p, "") && !strcmp(idx, "0") && !found_unified) { |
5360 | found_unified = true; | |
5361 | p = "unified"; | |
5362 | } | |
a67719f6 | 5363 | |
237e200e SH |
5364 | if (!store_hierarchy(line, p)) |
5365 | goto out; | |
5366 | } | |
5367 | ||
480262c9 | 5368 | /* Preserve initial namespace. */ |
a257a8ee | 5369 | init_ns = preserve_mnt_ns(getpid()); |
b8defc3d CB |
5370 | if (init_ns < 0) { |
5371 | lxcfs_error("%s\n", "Failed to preserve initial mount namespace."); | |
480262c9 | 5372 | goto out; |
b8defc3d | 5373 | } |
480262c9 | 5374 | |
92c3ee11 | 5375 | fd_hierarchies = malloc(sizeof(int) * num_hierarchies); |
b8defc3d CB |
5376 | if (!fd_hierarchies) { |
5377 | lxcfs_error("%s\n", strerror(errno)); | |
29a73c2f | 5378 | goto out; |
b8defc3d | 5379 | } |
29a73c2f | 5380 | |
480262c9 CB |
5381 | for (i = 0; i < num_hierarchies; i++) |
5382 | fd_hierarchies[i] = -1; | |
5383 | ||
e58dab00 CB |
5384 | cret = getcwd(cwd, MAXPATHLEN); |
5385 | if (!cret) | |
5386 | lxcfs_debug("Could not retrieve current working directory: %s.\n", strerror(errno)); | |
5387 | ||
480262c9 CB |
5388 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace |
5389 | * to privately mount lxcfs cgroups. */ | |
b8defc3d CB |
5390 | if (!cgfs_setup_controllers()) { |
5391 | lxcfs_error("%s\n", "Failed to setup private cgroup mounts for lxcfs."); | |
29a73c2f | 5392 | goto out; |
b8defc3d | 5393 | } |
480262c9 | 5394 | |
b8defc3d CB |
5395 | if (setns(init_ns, 0) < 0) { |
5396 | lxcfs_error("Failed to switch back to initial mount namespace: %s.\n", strerror(errno)); | |
29a73c2f | 5397 | goto out; |
b8defc3d | 5398 | } |
29a73c2f | 5399 | |
e58dab00 CB |
5400 | if (!cret || chdir(cwd) < 0) |
5401 | lxcfs_debug("Could not change back to original working directory: %s.\n", strerror(errno)); | |
5402 | ||
237e200e SH |
5403 | print_subsystems(); |
5404 | ||
5405 | out: | |
5406 | free(line); | |
5407 | fclose(f); | |
480262c9 CB |
5408 | if (init_ns >= 0) |
5409 | close(init_ns); | |
237e200e SH |
5410 | } |
5411 | ||
5412 | static void __attribute__((destructor)) free_subsystems(void) | |
5413 | { | |
5414 | int i; | |
5415 | ||
b8defc3d CB |
5416 | lxcfs_debug("%s\n", "Running destructor for liblxcfs."); |
5417 | ||
29a73c2f | 5418 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
5419 | if (hierarchies[i]) |
5420 | free(hierarchies[i]); | |
480262c9 | 5421 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
5422 | close(fd_hierarchies[i]); |
5423 | } | |
237e200e | 5424 | free(hierarchies); |
480262c9 | 5425 | free(fd_hierarchies); |
a257a8ee CB |
5426 | |
5427 | if (cgroup_mount_ns_fd >= 0) | |
5428 | close(cgroup_mount_ns_fd); | |
237e200e | 5429 | } |