]>
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 | 19 | #include <sched.h> |
db1b32f6 | 20 | #include <stdarg.h> |
29a73c2f | 21 | #include <stdbool.h> |
0ecddf02 | 22 | #include <stdint.h> |
29a73c2f CB |
23 | #include <stdio.h> |
24 | #include <stdlib.h> | |
25 | #include <string.h> | |
26 | #include <time.h> | |
27 | #include <unistd.h> | |
28 | #include <wait.h> | |
d89504c4 | 29 | #include <linux/magic.h> |
237e200e | 30 | #include <linux/sched.h> |
29a73c2f CB |
31 | #include <sys/epoll.h> |
32 | #include <sys/mman.h> | |
33 | #include <sys/mount.h> | |
237e200e SH |
34 | #include <sys/param.h> |
35 | #include <sys/socket.h> | |
29a73c2f | 36 | #include <sys/syscall.h> |
0ecddf02 | 37 | #include <sys/sysinfo.h> |
d89504c4 | 38 | #include <sys/vfs.h> |
237e200e | 39 | |
237e200e | 40 | #include "bindings.h" |
5fbea8a6 CB |
41 | #include "cgroups/cgroup.h" |
42 | #include "cgroups/cgroup_utils.h" | |
c9236032 HY |
43 | #include "memory_utils.h" |
44 | #include "config.h" | |
237e200e | 45 | |
29a73c2f CB |
46 | /* Define pivot_root() if missing from the C library */ |
47 | #ifndef HAVE_PIVOT_ROOT | |
48 | static int pivot_root(const char * new_root, const char * put_old) | |
49 | { | |
50 | #ifdef __NR_pivot_root | |
51 | return syscall(__NR_pivot_root, new_root, put_old); | |
52 | #else | |
53 | errno = ENOSYS; | |
54 | return -1; | |
55 | #endif | |
56 | } | |
57 | #else | |
58 | extern int pivot_root(const char * new_root, const char * put_old); | |
59 | #endif | |
60 | ||
8be92dd1 JS |
61 | struct cpuacct_usage { |
62 | uint64_t user; | |
63 | uint64_t system; | |
056adcef | 64 | uint64_t idle; |
77005a6c | 65 | bool online; |
8be92dd1 JS |
66 | }; |
67 | ||
0e47acaa | 68 | /* The function of hash table.*/ |
69 | #define LOAD_SIZE 100 /*the size of hash_table */ | |
6db4f7a3 | 70 | #define FLUSH_TIME 5 /*the flush rate */ |
71 | #define DEPTH_DIR 3 /*the depth of per cgroup */ | |
72 | /* The function of calculate loadavg .*/ | |
73 | #define FSHIFT 11 /* nr of bits of precision */ | |
74 | #define FIXED_1 (1<<FSHIFT) /* 1.0 as fixed-point */ | |
75 | #define EXP_1 1884 /* 1/exp(5sec/1min) as fixed-point */ | |
76 | #define EXP_5 2014 /* 1/exp(5sec/5min) */ | |
77 | #define EXP_15 2037 /* 1/exp(5sec/15min) */ | |
78 | #define LOAD_INT(x) ((x) >> FSHIFT) | |
79 | #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) | |
beb5024e | 80 | /* |
6db4f7a3 | 81 | * This parameter is used for proc_loadavg_read(). |
82 | * 1 means use loadavg, 0 means not use. | |
83 | */ | |
84 | static int loadavg = 0; | |
a83618e2 | 85 | static volatile sig_atomic_t loadavg_stop = 0; |
056adcef | 86 | static int calc_hash(const char *name) |
0e47acaa | 87 | { |
88 | unsigned int hash = 0; | |
89 | unsigned int x = 0; | |
90 | /* ELFHash algorithm. */ | |
91 | while (*name) { | |
92 | hash = (hash << 4) + *name++; | |
93 | x = hash & 0xf0000000; | |
94 | if (x != 0) | |
95 | hash ^= (x >> 24); | |
96 | hash &= ~x; | |
97 | } | |
b077527b | 98 | return (hash & 0x7fffffff); |
0e47acaa | 99 | } |
100 | ||
101 | struct load_node { | |
102 | char *cg; /*cg */ | |
103 | unsigned long avenrun[3]; /* Load averages */ | |
104 | unsigned int run_pid; | |
105 | unsigned int total_pid; | |
106 | unsigned int last_pid; | |
107 | int cfd; /* The file descriptor of the mounted cgroup */ | |
108 | struct load_node *next; | |
109 | struct load_node **pre; | |
110 | }; | |
111 | ||
112 | struct load_head { | |
113 | /* | |
114 | * The lock is about insert load_node and refresh load_node.To the first | |
115 | * load_node of each hash bucket, insert and refresh in this hash bucket is | |
116 | * mutually exclusive. | |
117 | */ | |
118 | pthread_mutex_t lock; | |
119 | /* | |
120 | * The rdlock is about read loadavg and delete load_node.To each hash | |
121 | * bucket, read and delete is mutually exclusive. But at the same time, we | |
122 | * allow paratactic read operation. This rdlock is at list level. | |
123 | */ | |
124 | pthread_rwlock_t rdlock; | |
125 | /* | |
126 | * The rilock is about read loadavg and insert load_node.To the first | |
127 | * load_node of each hash bucket, read and insert is mutually exclusive. | |
128 | * But at the same time, we allow paratactic read operation. | |
129 | */ | |
130 | pthread_rwlock_t rilock; | |
131 | struct load_node *next; | |
132 | }; | |
133 | ||
134 | static struct load_head load_hash[LOAD_SIZE]; /* hash table */ | |
135 | /* | |
136 | * init_load initialize the hash table. | |
137 | * Return 0 on success, return -1 on failure. | |
138 | */ | |
139 | static int init_load(void) | |
140 | { | |
141 | int i; | |
142 | int ret; | |
143 | ||
144 | for (i = 0; i < LOAD_SIZE; i++) { | |
145 | load_hash[i].next = NULL; | |
146 | ret = pthread_mutex_init(&load_hash[i].lock, NULL); | |
147 | if (ret != 0) { | |
148 | lxcfs_error("%s\n", "Failed to initialize lock"); | |
149 | goto out3; | |
150 | } | |
151 | ret = pthread_rwlock_init(&load_hash[i].rdlock, NULL); | |
152 | if (ret != 0) { | |
153 | lxcfs_error("%s\n", "Failed to initialize rdlock"); | |
154 | goto out2; | |
155 | } | |
156 | ret = pthread_rwlock_init(&load_hash[i].rilock, NULL); | |
157 | if (ret != 0) { | |
158 | lxcfs_error("%s\n", "Failed to initialize rilock"); | |
159 | goto out1; | |
160 | } | |
161 | } | |
162 | return 0; | |
163 | out1: | |
164 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
165 | out2: | |
166 | pthread_mutex_destroy(&load_hash[i].lock); | |
167 | out3: | |
168 | while (i > 0) { | |
169 | i--; | |
170 | pthread_mutex_destroy(&load_hash[i].lock); | |
171 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
172 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
173 | } | |
174 | return -1; | |
175 | } | |
176 | ||
177 | static void insert_node(struct load_node **n, int locate) | |
178 | { | |
179 | struct load_node *f; | |
180 | ||
181 | pthread_mutex_lock(&load_hash[locate].lock); | |
182 | pthread_rwlock_wrlock(&load_hash[locate].rilock); | |
183 | f = load_hash[locate].next; | |
184 | load_hash[locate].next = *n; | |
185 | ||
186 | (*n)->pre = &(load_hash[locate].next); | |
187 | if (f) | |
188 | f->pre = &((*n)->next); | |
189 | (*n)->next = f; | |
190 | pthread_mutex_unlock(&load_hash[locate].lock); | |
191 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
192 | } | |
193 | /* | |
194 | * locate_node() finds special node. Not return NULL means success. | |
195 | * It should be noted that rdlock isn't unlocked at the end of code | |
196 | * because this function is used to read special node. Delete is not | |
197 | * allowed before read has ended. | |
198 | * unlock rdlock only in proc_loadavg_read(). | |
199 | */ | |
200 | static struct load_node *locate_node(char *cg, int locate) | |
201 | { | |
202 | struct load_node *f = NULL; | |
203 | int i = 0; | |
204 | ||
205 | pthread_rwlock_rdlock(&load_hash[locate].rilock); | |
206 | pthread_rwlock_rdlock(&load_hash[locate].rdlock); | |
207 | if (load_hash[locate].next == NULL) { | |
208 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
209 | return f; | |
210 | } | |
211 | f = load_hash[locate].next; | |
212 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
213 | while (f && ((i = strcmp(f->cg, cg)) != 0)) | |
214 | f = f->next; | |
215 | return f; | |
216 | } | |
54a6d46a | 217 | |
0e47acaa | 218 | /* Delete the load_node n and return the next node of it. */ |
219 | static struct load_node *del_node(struct load_node *n, int locate) | |
220 | { | |
221 | struct load_node *g; | |
222 | ||
223 | pthread_rwlock_wrlock(&load_hash[locate].rdlock); | |
224 | if (n->next == NULL) { | |
225 | *(n->pre) = NULL; | |
226 | } else { | |
227 | *(n->pre) = n->next; | |
228 | n->next->pre = n->pre; | |
229 | } | |
230 | g = n->next; | |
54a6d46a CB |
231 | free_disarm(n->cg); |
232 | free_disarm(n); | |
0e47acaa | 233 | pthread_rwlock_unlock(&load_hash[locate].rdlock); |
234 | return g; | |
235 | } | |
236 | ||
a83618e2 | 237 | static void load_free(void) |
9c480eb7 | 238 | { |
9c480eb7 | 239 | struct load_node *f, *p; |
240 | ||
54a6d46a | 241 | for (int i = 0; i < LOAD_SIZE; i++) { |
9c480eb7 | 242 | pthread_mutex_lock(&load_hash[i].lock); |
243 | pthread_rwlock_wrlock(&load_hash[i].rilock); | |
244 | pthread_rwlock_wrlock(&load_hash[i].rdlock); | |
245 | if (load_hash[i].next == NULL) { | |
246 | pthread_mutex_unlock(&load_hash[i].lock); | |
247 | pthread_mutex_destroy(&load_hash[i].lock); | |
248 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
249 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
250 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
251 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
252 | continue; | |
253 | } | |
54a6d46a CB |
254 | |
255 | for (f = load_hash[i].next; f;) { | |
256 | free_disarm(f->cg); | |
9c480eb7 | 257 | p = f->next; |
54a6d46a | 258 | free_disarm(f); |
9c480eb7 | 259 | f = p; |
260 | } | |
54a6d46a | 261 | |
9c480eb7 | 262 | pthread_mutex_unlock(&load_hash[i].lock); |
263 | pthread_mutex_destroy(&load_hash[i].lock); | |
264 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
265 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
266 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
267 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
268 | } | |
269 | } | |
056adcef JS |
270 | |
271 | /* Data for CPU view */ | |
272 | struct cg_proc_stat { | |
273 | char *cg; | |
274 | struct cpuacct_usage *usage; // Real usage as read from the host's /proc/stat | |
275 | struct cpuacct_usage *view; // Usage stats reported to the container | |
276 | int cpu_count; | |
2f49b662 | 277 | pthread_mutex_t lock; // For node manipulation |
056adcef JS |
278 | struct cg_proc_stat *next; |
279 | }; | |
280 | ||
281 | struct cg_proc_stat_head { | |
282 | struct cg_proc_stat *next; | |
951acc94 | 283 | time_t lastcheck; |
2f49b662 JS |
284 | |
285 | /* | |
286 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
287 | */ | |
288 | pthread_rwlock_t lock; | |
056adcef JS |
289 | }; |
290 | ||
291 | #define CPUVIEW_HASH_SIZE 100 | |
292 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
293 | ||
294 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
295 | { | |
296 | *head = malloc(sizeof(struct cg_proc_stat_head)); | |
297 | if (!(*head)) { | |
298 | lxcfs_error("%s\n", strerror(errno)); | |
299 | return false; | |
300 | } | |
301 | ||
951acc94 | 302 | (*head)->lastcheck = time(NULL); |
056adcef | 303 | (*head)->next = NULL; |
2f49b662 JS |
304 | |
305 | if (pthread_rwlock_init(&(*head)->lock, NULL) != 0) { | |
306 | lxcfs_error("%s\n", "Failed to initialize list lock"); | |
54a6d46a | 307 | free_disarm(*head); |
2f49b662 JS |
308 | return false; |
309 | } | |
310 | ||
056adcef JS |
311 | return true; |
312 | } | |
313 | ||
314 | static bool init_cpuview() | |
315 | { | |
316 | int i; | |
317 | ||
318 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
319 | proc_stat_history[i] = NULL; | |
320 | ||
321 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
322 | if (!cpuview_init_head(&proc_stat_history[i])) | |
323 | goto err; | |
324 | } | |
325 | ||
326 | return true; | |
327 | ||
328 | err: | |
329 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
54a6d46a CB |
330 | if (proc_stat_history[i]) |
331 | free_disarm(proc_stat_history[i]); | |
056adcef JS |
332 | } |
333 | ||
334 | return false; | |
335 | } | |
336 | ||
951acc94 JS |
337 | static void free_proc_stat_node(struct cg_proc_stat *node) |
338 | { | |
2f49b662 | 339 | pthread_mutex_destroy(&node->lock); |
54a6d46a CB |
340 | free_disarm(node->cg); |
341 | free_disarm(node->usage); | |
342 | free_disarm(node->view); | |
343 | free_disarm(node); | |
951acc94 JS |
344 | } |
345 | ||
056adcef JS |
346 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
347 | { | |
348 | struct cg_proc_stat *node, *tmp; | |
349 | ||
350 | if (head->next) { | |
351 | node = head->next; | |
352 | ||
353 | for (;;) { | |
354 | tmp = node; | |
355 | node = node->next; | |
951acc94 | 356 | free_proc_stat_node(tmp); |
056adcef JS |
357 | |
358 | if (!node) | |
359 | break; | |
360 | } | |
361 | } | |
362 | ||
2f49b662 | 363 | pthread_rwlock_destroy(&head->lock); |
54a6d46a | 364 | free_disarm(head); |
056adcef JS |
365 | } |
366 | ||
367 | static void free_cpuview() | |
368 | { | |
369 | int i; | |
370 | ||
371 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
372 | if (proc_stat_history[i]) | |
373 | cpuview_free_head(proc_stat_history[i]); | |
374 | } | |
375 | } | |
376 | ||
237e200e SH |
377 | /* |
378 | * A table caching which pid is init for a pid namespace. | |
379 | * When looking up which pid is init for $qpid, we first | |
380 | * 1. Stat /proc/$qpid/ns/pid. | |
381 | * 2. Check whether the ino_t is in our store. | |
382 | * a. if not, fork a child in qpid's ns to send us | |
383 | * ucred.pid = 1, and read the initpid. Cache | |
384 | * initpid and creation time for /proc/initpid | |
385 | * in a new store entry. | |
386 | * b. if so, verify that /proc/initpid still matches | |
387 | * what we have saved. If not, clear the store | |
388 | * entry and go back to a. If so, return the | |
389 | * cached initpid. | |
390 | */ | |
391 | struct pidns_init_store { | |
392 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
393 | pid_t initpid; // the pid of nit in that ns | |
394 | long int ctime; // the time at which /proc/$initpid was created | |
395 | struct pidns_init_store *next; | |
396 | long int lastcheck; | |
397 | }; | |
398 | ||
399 | /* lol - look at how they are allocated in the kernel */ | |
400 | #define PIDNS_HASH_SIZE 4096 | |
401 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
402 | ||
403 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
404 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
405 | static void lock_mutex(pthread_mutex_t *l) | |
406 | { | |
407 | int ret; | |
408 | ||
409 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
b8defc3d | 410 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
411 | exit(1); |
412 | } | |
413 | } | |
414 | ||
77f4399a | 415 | struct cgroup_ops *cgroup_ops; |
29a73c2f | 416 | |
237e200e SH |
417 | static void unlock_mutex(pthread_mutex_t *l) |
418 | { | |
419 | int ret; | |
420 | ||
421 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
b8defc3d | 422 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
423 | exit(1); |
424 | } | |
425 | } | |
426 | ||
427 | static void store_lock(void) | |
428 | { | |
429 | lock_mutex(&pidns_store_mutex); | |
430 | } | |
431 | ||
432 | static void store_unlock(void) | |
433 | { | |
434 | unlock_mutex(&pidns_store_mutex); | |
435 | } | |
436 | ||
437 | /* Must be called under store_lock */ | |
438 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
439 | { | |
440 | struct stat initsb; | |
441 | char fnam[100]; | |
442 | ||
443 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
444 | if (stat(fnam, &initsb) < 0) | |
445 | return false; | |
7dd6560a CB |
446 | |
447 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
448 | initsb.st_ctime, e->initpid); | |
449 | ||
237e200e SH |
450 | if (e->ctime != initsb.st_ctime) |
451 | return false; | |
452 | return true; | |
453 | } | |
454 | ||
455 | /* Must be called under store_lock */ | |
456 | static void remove_initpid(struct pidns_init_store *e) | |
457 | { | |
458 | struct pidns_init_store *tmp; | |
459 | int h; | |
460 | ||
7dd6560a CB |
461 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
462 | ||
237e200e SH |
463 | h = HASH(e->ino); |
464 | if (pidns_hash_table[h] == e) { | |
465 | pidns_hash_table[h] = e->next; | |
54a6d46a | 466 | free_disarm(e); |
237e200e SH |
467 | return; |
468 | } | |
469 | ||
470 | tmp = pidns_hash_table[h]; | |
471 | while (tmp) { | |
472 | if (tmp->next == e) { | |
473 | tmp->next = e->next; | |
54a6d46a | 474 | free_disarm(e); |
237e200e SH |
475 | return; |
476 | } | |
477 | tmp = tmp->next; | |
478 | } | |
479 | } | |
480 | ||
481 | #define PURGE_SECS 5 | |
482 | /* Must be called under store_lock */ | |
483 | static void prune_initpid_store(void) | |
484 | { | |
485 | static long int last_prune = 0; | |
486 | struct pidns_init_store *e, *prev, *delme; | |
487 | long int now, threshold; | |
488 | int i; | |
489 | ||
490 | if (!last_prune) { | |
491 | last_prune = time(NULL); | |
492 | return; | |
493 | } | |
494 | now = time(NULL); | |
495 | if (now < last_prune + PURGE_SECS) | |
496 | return; | |
7dd6560a CB |
497 | |
498 | lxcfs_debug("%s\n", "Pruning."); | |
499 | ||
237e200e SH |
500 | last_prune = now; |
501 | threshold = now - 2 * PURGE_SECS; | |
502 | ||
503 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
504 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
505 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
506 | |
507 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
508 | ||
237e200e SH |
509 | delme = e; |
510 | if (prev) | |
511 | prev->next = e->next; | |
512 | else | |
513 | pidns_hash_table[i] = e->next; | |
514 | e = e->next; | |
54a6d46a | 515 | free_disarm(delme); |
237e200e SH |
516 | } else { |
517 | prev = e; | |
518 | e = e->next; | |
519 | } | |
520 | } | |
521 | } | |
522 | } | |
523 | ||
524 | /* Must be called under store_lock */ | |
525 | static void save_initpid(struct stat *sb, pid_t pid) | |
526 | { | |
527 | struct pidns_init_store *e; | |
528 | char fpath[100]; | |
529 | struct stat procsb; | |
530 | int h; | |
531 | ||
7dd6560a CB |
532 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
533 | ||
237e200e SH |
534 | snprintf(fpath, 100, "/proc/%d", pid); |
535 | if (stat(fpath, &procsb) < 0) | |
536 | return; | |
537 | do { | |
538 | e = malloc(sizeof(*e)); | |
539 | } while (!e); | |
540 | e->ino = sb->st_ino; | |
541 | e->initpid = pid; | |
542 | e->ctime = procsb.st_ctime; | |
543 | h = HASH(e->ino); | |
544 | e->next = pidns_hash_table[h]; | |
545 | e->lastcheck = time(NULL); | |
546 | pidns_hash_table[h] = e; | |
547 | } | |
548 | ||
549 | /* | |
550 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
551 | * entry for the inode number and creation time. Verify that the init pid | |
552 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
553 | * otherwise. | |
554 | * Must be called under store_lock | |
555 | */ | |
556 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
557 | { | |
558 | int h = HASH(sb->st_ino); | |
559 | struct pidns_init_store *e = pidns_hash_table[h]; | |
560 | ||
561 | while (e) { | |
562 | if (e->ino == sb->st_ino) { | |
563 | if (initpid_still_valid(e, sb)) { | |
564 | e->lastcheck = time(NULL); | |
565 | return e; | |
566 | } | |
567 | remove_initpid(e); | |
568 | return NULL; | |
569 | } | |
570 | e = e->next; | |
571 | } | |
572 | ||
573 | return NULL; | |
574 | } | |
575 | ||
0f657ce3 | 576 | static int is_dir(const char *path, int fd) |
237e200e SH |
577 | { |
578 | struct stat statbuf; | |
0f657ce3 | 579 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
580 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
581 | return 1; | |
582 | return 0; | |
583 | } | |
584 | ||
c9236032 HY |
585 | static int preserve_ns(const int pid, const char *ns) |
586 | { | |
587 | int ret; | |
588 | /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */ | |
589 | #define __NS_PATH_LEN 50 | |
590 | char path[__NS_PATH_LEN]; | |
591 | ||
592 | /* This way we can use this function to also check whether namespaces | |
593 | * are supported by the kernel by passing in the NULL or the empty | |
594 | * string. | |
595 | */ | |
596 | ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid, | |
597 | !ns || strcmp(ns, "") == 0 ? "" : "/", | |
598 | !ns || strcmp(ns, "") == 0 ? "" : ns); | |
599 | if (ret < 0 || (size_t)ret >= __NS_PATH_LEN) { | |
600 | errno = EFBIG; | |
601 | return -1; | |
602 | } | |
603 | ||
604 | return open(path, O_RDONLY | O_CLOEXEC); | |
605 | } | |
606 | ||
607 | /** | |
608 | * in_same_namespace - Check whether two processes are in the same namespace. | |
609 | * @pid1 - PID of the first process. | |
610 | * @pid2 - PID of the second process. | |
611 | * @ns - Name of the namespace to check. Must correspond to one of the names | |
612 | * for the namespaces as shown in /proc/<pid/ns/ | |
613 | * | |
614 | * If the two processes are not in the same namespace returns an fd to the | |
615 | * namespace of the second process identified by @pid2. If the two processes are | |
616 | * in the same namespace returns -EINVAL, -1 if an error occurred. | |
617 | */ | |
618 | static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns) | |
619 | { | |
620 | __do_close_prot_errno int ns_fd1 = -1, ns_fd2 = -1; | |
621 | int ret = -1; | |
622 | struct stat ns_st1, ns_st2; | |
623 | ||
624 | ns_fd1 = preserve_ns(pid1, ns); | |
625 | if (ns_fd1 < 0) { | |
626 | /* The kernel does not support this namespace. This is not an | |
627 | * error. | |
628 | */ | |
629 | if (errno == ENOENT) | |
630 | return -EINVAL; | |
631 | ||
632 | return -1; | |
633 | } | |
634 | ||
635 | ns_fd2 = preserve_ns(pid2, ns); | |
636 | if (ns_fd2 < 0) | |
637 | return -1; | |
638 | ||
639 | ret = fstat(ns_fd1, &ns_st1); | |
640 | if (ret < 0) | |
641 | return -1; | |
642 | ||
643 | ret = fstat(ns_fd2, &ns_st2); | |
644 | if (ret < 0) | |
645 | return -1; | |
646 | ||
647 | /* processes are in the same namespace */ | |
648 | if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino)) | |
649 | return -EINVAL; | |
650 | ||
651 | /* processes are in different namespaces */ | |
652 | return move_fd(ns_fd2); | |
653 | } | |
654 | ||
655 | static bool is_shared_pidns(pid_t pid) | |
656 | { | |
657 | if (pid != 1) | |
658 | return false; | |
659 | ||
6e3637bb | 660 | if (in_same_namespace(pid, getpid(), "pid") == -EINVAL) |
c9236032 HY |
661 | return true; |
662 | ||
663 | return false; | |
664 | } | |
665 | ||
ba59ea09 | 666 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
667 | { |
668 | FILE *f; | |
669 | size_t len, ret; | |
670 | ||
beb5024e CB |
671 | f = fdopen(fd, "w"); |
672 | if (!f) | |
237e200e | 673 | return false; |
beb5024e | 674 | |
237e200e SH |
675 | len = strlen(string); |
676 | ret = fwrite(string, 1, len, f); | |
677 | if (ret != len) { | |
beb5024e CB |
678 | lxcfs_error("%s - Error writing \"%s\" to \"%s\"\n", |
679 | strerror(errno), string, fnam); | |
237e200e SH |
680 | fclose(f); |
681 | return false; | |
682 | } | |
beb5024e | 683 | |
237e200e | 684 | if (fclose(f) < 0) { |
beb5024e | 685 | lxcfs_error("%s - Failed to close \"%s\"\n", strerror(errno), fnam); |
237e200e SH |
686 | return false; |
687 | } | |
beb5024e | 688 | |
237e200e SH |
689 | return true; |
690 | } | |
691 | ||
237e200e SH |
692 | struct cgfs_files { |
693 | char *name; | |
694 | uint32_t uid, gid; | |
695 | uint32_t mode; | |
696 | }; | |
697 | ||
237e200e SH |
698 | static void print_subsystems(void) |
699 | { | |
5fbea8a6 | 700 | int i = 0; |
237e200e | 701 | |
0646f250 | 702 | fprintf(stderr, "mount namespace: %d\n", cgroup_ops->mntns_fd); |
cc97d34c | 703 | fprintf(stderr, "hierarchies:\n"); |
5fbea8a6 CB |
704 | for (struct hierarchy **h = cgroup_ops->hierarchies; h && *h; h++, i++) { |
705 | __do_free char *controllers = lxc_string_join(",", (const char **)(*h)->controllers, false); | |
706 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, (*h)->fd, controllers ?: ""); | |
237e200e SH |
707 | } |
708 | } | |
709 | ||
237e200e SH |
710 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, |
711 | const char *value) | |
712 | { | |
ba59ea09 | 713 | int ret, fd, cfd; |
237e200e | 714 | size_t len; |
5fbea8a6 | 715 | char *fnam; |
237e200e | 716 | |
d298bba1 | 717 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 718 | if (cfd < 0) |
237e200e | 719 | return false; |
f5a6d92e CB |
720 | |
721 | /* Make sure we pass a relative path to *at() family of functions. | |
722 | * . + /cgroup + / + file + \0 | |
723 | */ | |
ba59ea09 | 724 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 725 | fnam = alloca(len); |
075387cd | 726 | ret = snprintf(fnam, len, "%s%s/%s", dot_or_empty(cgroup), cgroup, file); |
ba59ea09 CB |
727 | if (ret < 0 || (size_t)ret >= len) |
728 | return false; | |
729 | ||
730 | fd = openat(cfd, fnam, O_WRONLY); | |
731 | if (fd < 0) | |
732 | return false; | |
f676eb79 | 733 | |
ba59ea09 | 734 | return write_string(fnam, value, fd); |
237e200e SH |
735 | } |
736 | ||
737 | // Chown all the files in the cgroup directory. We do this when we create | |
738 | // a cgroup on behalf of a user. | |
f23fe717 | 739 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 740 | { |
f23fe717 | 741 | struct dirent *direntp; |
237e200e SH |
742 | char path[MAXPATHLEN]; |
743 | size_t len; | |
744 | DIR *d; | |
f23fe717 | 745 | int fd1, ret; |
237e200e SH |
746 | |
747 | len = strlen(dirname); | |
748 | if (len >= MAXPATHLEN) { | |
b8defc3d | 749 | lxcfs_error("Pathname too long: %s\n", dirname); |
237e200e SH |
750 | return; |
751 | } | |
752 | ||
f23fe717 CB |
753 | fd1 = openat(fd, dirname, O_DIRECTORY); |
754 | if (fd1 < 0) | |
755 | return; | |
756 | ||
757 | d = fdopendir(fd1); | |
237e200e | 758 | if (!d) { |
b8defc3d | 759 | lxcfs_error("Failed to open %s\n", dirname); |
237e200e SH |
760 | return; |
761 | } | |
762 | ||
f23fe717 | 763 | while ((direntp = readdir(d))) { |
237e200e SH |
764 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
765 | continue; | |
766 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
767 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 768 | lxcfs_error("Pathname too long under %s\n", dirname); |
237e200e SH |
769 | continue; |
770 | } | |
f23fe717 | 771 | if (fchownat(fd, path, uid, gid, 0) < 0) |
b8defc3d | 772 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); |
237e200e SH |
773 | } |
774 | closedir(d); | |
775 | } | |
776 | ||
777 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
778 | { | |
5dd3e6fd | 779 | int cfd; |
237e200e | 780 | size_t len; |
5fbea8a6 | 781 | char *dirnam; |
237e200e | 782 | |
d298bba1 | 783 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 784 | if (cfd < 0) |
237e200e | 785 | return -EINVAL; |
f5a6d92e CB |
786 | |
787 | /* Make sure we pass a relative path to *at() family of functions. | |
788 | * . + /cg + \0 | |
789 | */ | |
f23fe717 | 790 | len = strlen(cg) + 2; |
237e200e | 791 | dirnam = alloca(len); |
075387cd | 792 | snprintf(dirnam, len, "%s%s", dot_or_empty(cg), cg); |
237e200e | 793 | |
f23fe717 | 794 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
795 | return -errno; |
796 | ||
797 | if (uid == 0 && gid == 0) | |
798 | return 0; | |
799 | ||
f23fe717 | 800 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
801 | return -errno; |
802 | ||
f23fe717 | 803 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
804 | |
805 | return 0; | |
806 | } | |
807 | ||
7213ec5c | 808 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) |
237e200e | 809 | { |
b7672ded | 810 | struct dirent *direntp; |
237e200e SH |
811 | DIR *dir; |
812 | bool ret = false; | |
813 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
814 | int dupfd; |
815 | ||
816 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
817 | if (dupfd < 0) | |
818 | return false; | |
237e200e | 819 | |
b7672ded | 820 | dir = fdopendir(dupfd); |
237e200e | 821 | if (!dir) { |
7dd6560a | 822 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
7213ec5c | 823 | close(dupfd); |
237e200e SH |
824 | return false; |
825 | } | |
826 | ||
b7672ded | 827 | while ((direntp = readdir(dir))) { |
237e200e SH |
828 | struct stat mystat; |
829 | int rc; | |
830 | ||
237e200e SH |
831 | if (!strcmp(direntp->d_name, ".") || |
832 | !strcmp(direntp->d_name, "..")) | |
833 | continue; | |
834 | ||
835 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
836 | if (rc < 0 || rc >= MAXPATHLEN) { | |
b8defc3d | 837 | lxcfs_error("%s\n", "Pathname too long."); |
237e200e SH |
838 | continue; |
839 | } | |
840 | ||
2e81a5e3 CB |
841 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
842 | if (rc) { | |
7dd6560a | 843 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
844 | continue; |
845 | } | |
7dd6560a | 846 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 847 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 848 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
849 | } |
850 | ||
851 | ret = true; | |
852 | if (closedir(dir) < 0) { | |
b8defc3d | 853 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); |
237e200e SH |
854 | ret = false; |
855 | } | |
856 | ||
2e81a5e3 | 857 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 858 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
859 | ret = false; |
860 | } | |
7213ec5c CB |
861 | |
862 | close(dupfd); | |
237e200e SH |
863 | |
864 | return ret; | |
865 | } | |
866 | ||
867 | bool cgfs_remove(const char *controller, const char *cg) | |
868 | { | |
b7672ded | 869 | int fd, cfd; |
237e200e | 870 | size_t len; |
5fbea8a6 | 871 | char *dirnam; |
7213ec5c | 872 | bool bret; |
237e200e | 873 | |
d298bba1 | 874 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 875 | if (cfd < 0) |
237e200e | 876 | return false; |
f5a6d92e CB |
877 | |
878 | /* Make sure we pass a relative path to *at() family of functions. | |
879 | * . + /cg + \0 | |
880 | */ | |
b7672ded | 881 | len = strlen(cg) + 2; |
237e200e | 882 | dirnam = alloca(len); |
075387cd | 883 | snprintf(dirnam, len, "%s%s", dot_or_empty(cg), cg); |
b7672ded CB |
884 | |
885 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
886 | if (fd < 0) | |
887 | return false; | |
888 | ||
7213ec5c CB |
889 | bret = recursive_rmdir(dirnam, fd, cfd); |
890 | close(fd); | |
891 | return bret; | |
237e200e SH |
892 | } |
893 | ||
894 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
895 | { | |
5dd3e6fd | 896 | int cfd; |
237e200e | 897 | size_t len; |
5fbea8a6 | 898 | char *pathname; |
237e200e | 899 | |
d298bba1 | 900 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 901 | if (cfd < 0) |
237e200e | 902 | return false; |
f5a6d92e CB |
903 | |
904 | /* Make sure we pass a relative path to *at() family of functions. | |
905 | * . + /file + \0 | |
906 | */ | |
534690b4 | 907 | len = strlen(file) + 2; |
237e200e | 908 | pathname = alloca(len); |
075387cd | 909 | snprintf(pathname, len, "%s%s", dot_or_empty(file), file); |
534690b4 | 910 | if (fchmodat(cfd, pathname, mode, 0) < 0) |
237e200e SH |
911 | return false; |
912 | return true; | |
913 | } | |
914 | ||
0f657ce3 | 915 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
916 | { |
917 | size_t len; | |
918 | char *fname; | |
919 | ||
920 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
921 | fname = alloca(len); | |
922 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 923 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
924 | return -errno; |
925 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 926 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
927 | return -errno; |
928 | return 0; | |
929 | } | |
930 | ||
931 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
932 | { | |
5dd3e6fd | 933 | int cfd; |
237e200e | 934 | size_t len; |
5fbea8a6 | 935 | char *pathname; |
237e200e | 936 | |
d298bba1 | 937 | cfd = get_cgroup_fd(controller); |
5fbea8a6 CB |
938 | if (cfd < 0) |
939 | return false; | |
f5a6d92e CB |
940 | |
941 | /* Make sure we pass a relative path to *at() family of functions. | |
942 | * . + /file + \0 | |
943 | */ | |
0f657ce3 | 944 | len = strlen(file) + 2; |
237e200e | 945 | pathname = alloca(len); |
075387cd | 946 | snprintf(pathname, len, "%s%s", dot_or_empty(file), file); |
0f657ce3 | 947 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) |
237e200e SH |
948 | return -errno; |
949 | ||
0f657ce3 | 950 | if (is_dir(pathname, cfd)) |
237e200e | 951 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 952 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
953 | |
954 | return 0; | |
955 | } | |
956 | ||
957 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
958 | { | |
3ffd08ee | 959 | int fd, cfd; |
237e200e | 960 | size_t len; |
5fbea8a6 | 961 | char *pathname; |
237e200e | 962 | |
d298bba1 | 963 | cfd = get_cgroup_fd(controller); |
5fbea8a6 CB |
964 | if (cfd < 0) |
965 | return false; | |
f5a6d92e CB |
966 | |
967 | /* Make sure we pass a relative path to *at() family of functions. | |
968 | * . + /cgroup + / "cgroup.procs" + \0 | |
969 | */ | |
3ffd08ee | 970 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 971 | pathname = alloca(len); |
075387cd | 972 | snprintf(pathname, len, "%s%s/cgroup.procs", dot_or_empty(cgroup), cgroup); |
3ffd08ee CB |
973 | |
974 | fd = openat(cfd, pathname, O_WRONLY); | |
975 | if (fd < 0) | |
976 | return NULL; | |
977 | ||
978 | return fdopen(fd, "w"); | |
237e200e SH |
979 | } |
980 | ||
f366da65 WB |
981 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
982 | void ***list, size_t typesize, | |
983 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 984 | { |
4ea38a4c | 985 | int cfd, fd, ret; |
237e200e | 986 | size_t len; |
5fbea8a6 | 987 | char *cg; |
237e200e | 988 | char pathname[MAXPATHLEN]; |
f366da65 | 989 | size_t sz = 0, asz = 0; |
4ea38a4c | 990 | struct dirent *dirent; |
237e200e | 991 | DIR *dir; |
237e200e | 992 | |
d298bba1 | 993 | cfd = get_cgroup_fd(controller); |
f366da65 | 994 | *list = NULL; |
5fbea8a6 | 995 | if (cfd < 0) |
e97c834b | 996 | return false; |
237e200e | 997 | |
f5a6d92e | 998 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
999 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
1000 | cg = alloca(len); | |
075387cd | 1001 | ret = snprintf(cg, len, "%s%s", dot_or_empty(cgroup), cgroup); |
4ea38a4c | 1002 | if (ret < 0 || (size_t)ret >= len) { |
b8defc3d | 1003 | lxcfs_error("Pathname too long under %s\n", cgroup); |
4ea38a4c CB |
1004 | return false; |
1005 | } | |
237e200e | 1006 | |
4ea38a4c CB |
1007 | fd = openat(cfd, cg, O_DIRECTORY); |
1008 | if (fd < 0) | |
1009 | return false; | |
1010 | ||
1011 | dir = fdopendir(fd); | |
237e200e SH |
1012 | if (!dir) |
1013 | return false; | |
1014 | ||
4ea38a4c | 1015 | while ((dirent = readdir(dir))) { |
237e200e | 1016 | struct stat mystat; |
237e200e | 1017 | |
4ea38a4c CB |
1018 | if (!strcmp(dirent->d_name, ".") || |
1019 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
1020 | continue; |
1021 | ||
4ea38a4c CB |
1022 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
1023 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 1024 | lxcfs_error("Pathname too long under %s\n", cg); |
237e200e SH |
1025 | continue; |
1026 | } | |
1027 | ||
4ea38a4c | 1028 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e | 1029 | if (ret) { |
b8defc3d | 1030 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); |
237e200e SH |
1031 | continue; |
1032 | } | |
f366da65 WB |
1033 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
1034 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
1035 | continue; |
1036 | ||
1037 | if (sz+2 >= asz) { | |
f366da65 | 1038 | void **tmp; |
237e200e SH |
1039 | asz += BATCH_SIZE; |
1040 | do { | |
f366da65 | 1041 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
1042 | } while (!tmp); |
1043 | *list = tmp; | |
1044 | } | |
4ea38a4c | 1045 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
1046 | (*list)[sz+1] = NULL; |
1047 | sz++; | |
1048 | } | |
1049 | if (closedir(dir) < 0) { | |
b8defc3d | 1050 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); |
237e200e SH |
1051 | return false; |
1052 | } | |
1053 | return true; | |
1054 | } | |
1055 | ||
f366da65 WB |
1056 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
1057 | { | |
1058 | char *dup; | |
1059 | do { | |
1060 | dup = strdup(dir_entry); | |
1061 | } while (!dup); | |
1062 | return dup; | |
1063 | } | |
1064 | ||
1065 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
1066 | { | |
1067 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
1068 | } | |
1069 | ||
237e200e SH |
1070 | void free_key(struct cgfs_files *k) |
1071 | { | |
1072 | if (!k) | |
1073 | return; | |
54a6d46a CB |
1074 | free_disarm(k->name); |
1075 | free_disarm(k); | |
237e200e SH |
1076 | } |
1077 | ||
1078 | void free_keys(struct cgfs_files **keys) | |
1079 | { | |
1080 | int i; | |
1081 | ||
1082 | if (!keys) | |
1083 | return; | |
1084 | for (i = 0; keys[i]; i++) { | |
1085 | free_key(keys[i]); | |
1086 | } | |
54a6d46a | 1087 | free_disarm(keys); |
237e200e SH |
1088 | } |
1089 | ||
951acc94 JS |
1090 | bool cgfs_param_exist(const char *controller, const char *cgroup, const char *file) |
1091 | { | |
1092 | int ret, cfd; | |
1093 | size_t len; | |
5fbea8a6 | 1094 | char *fnam; |
951acc94 | 1095 | |
d298bba1 | 1096 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 1097 | if (cfd < 0) |
951acc94 JS |
1098 | return false; |
1099 | ||
1100 | /* Make sure we pass a relative path to *at() family of functions. | |
1101 | * . + /cgroup + / + file + \0 | |
1102 | */ | |
1103 | len = strlen(cgroup) + strlen(file) + 3; | |
1104 | fnam = alloca(len); | |
075387cd | 1105 | ret = snprintf(fnam, len, "%s%s/%s", dot_or_empty(cgroup), cgroup, file); |
951acc94 JS |
1106 | if (ret < 0 || (size_t)ret >= len) |
1107 | return false; | |
1108 | ||
1109 | return (faccessat(cfd, fnam, F_OK, 0) == 0); | |
1110 | } | |
1111 | ||
237e200e SH |
1112 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) |
1113 | { | |
4ea38a4c | 1114 | int ret, cfd; |
237e200e | 1115 | size_t len; |
5fbea8a6 | 1116 | char *fnam; |
237e200e SH |
1117 | struct stat sb; |
1118 | struct cgfs_files *newkey; | |
237e200e | 1119 | |
d298bba1 | 1120 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 1121 | if (cfd < 0) |
237e200e SH |
1122 | return false; |
1123 | ||
1124 | if (file && *file == '/') | |
1125 | file++; | |
1126 | ||
06081b29 | 1127 | if (file && strchr(file, '/')) |
237e200e SH |
1128 | return NULL; |
1129 | ||
f5a6d92e CB |
1130 | /* Make sure we pass a relative path to *at() family of functions. |
1131 | * . + /cgroup + / + file + \0 | |
1132 | */ | |
4ea38a4c | 1133 | len = strlen(cgroup) + 3; |
237e200e SH |
1134 | if (file) |
1135 | len += strlen(file) + 1; | |
1136 | fnam = alloca(len); | |
075387cd | 1137 | snprintf(fnam, len, "%s%s%s%s", dot_or_empty(cgroup), cgroup, |
4ea38a4c | 1138 | file ? "/" : "", file ? file : ""); |
237e200e | 1139 | |
4ea38a4c | 1140 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1141 | if (ret < 0) |
1142 | return NULL; | |
1143 | ||
1144 | do { | |
1145 | newkey = malloc(sizeof(struct cgfs_files)); | |
1146 | } while (!newkey); | |
1147 | if (file) | |
1148 | newkey->name = must_copy_string(file); | |
06081b29 CB |
1149 | else if (strrchr(cgroup, '/')) |
1150 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
1151 | else |
1152 | newkey->name = must_copy_string(cgroup); | |
1153 | newkey->uid = sb.st_uid; | |
1154 | newkey->gid = sb.st_gid; | |
1155 | newkey->mode = sb.st_mode; | |
1156 | ||
1157 | return newkey; | |
1158 | } | |
1159 | ||
f366da65 | 1160 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 1161 | { |
f366da65 WB |
1162 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
1163 | if (!entry) { | |
b8defc3d CB |
1164 | lxcfs_error("Error getting files under %s:%s\n", controller, |
1165 | cgroup); | |
237e200e | 1166 | } |
f366da65 WB |
1167 | return entry; |
1168 | } | |
1169 | ||
1170 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
1171 | { | |
1172 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
1173 | } |
1174 | ||
1175 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
1176 | { |
1177 | int cfd; | |
1178 | size_t len; | |
5fbea8a6 | 1179 | char *fnam; |
237e200e SH |
1180 | int ret; |
1181 | struct stat sb; | |
1182 | ||
d298bba1 | 1183 | cfd = get_cgroup_fd(controller); |
5fbea8a6 | 1184 | if (cfd < 0) |
237e200e | 1185 | return false; |
f5a6d92e CB |
1186 | |
1187 | /* Make sure we pass a relative path to *at() family of functions. | |
1188 | * . + /cgroup + / + f + \0 | |
1189 | */ | |
d04232f2 | 1190 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 1191 | fnam = alloca(len); |
075387cd | 1192 | ret = snprintf(fnam, len, "%s%s/%s", dot_or_empty(cgroup), cgroup, f); |
d04232f2 CB |
1193 | if (ret < 0 || (size_t)ret >= len) |
1194 | return false; | |
237e200e | 1195 | |
d04232f2 | 1196 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1197 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
1198 | return false; | |
f5a6d92e | 1199 | |
237e200e SH |
1200 | return true; |
1201 | } | |
1202 | ||
1203 | #define SEND_CREDS_OK 0 | |
1204 | #define SEND_CREDS_NOTSK 1 | |
1205 | #define SEND_CREDS_FAIL 2 | |
1206 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
1207 | static int wait_for_pid(pid_t pid); | |
1208 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 1209 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
1210 | |
1211 | /* | |
b10bdd6c | 1212 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
1213 | * over a unix sock so we can read the task's reaper's pid in our |
1214 | * namespace | |
b10bdd6c FG |
1215 | * |
1216 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
1217 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
1218 | * the pidns and the parent pid outside are identical. Using clone prevents | |
1219 | * this issue. | |
237e200e SH |
1220 | */ |
1221 | static void write_task_init_pid_exit(int sock, pid_t target) | |
1222 | { | |
237e200e SH |
1223 | char fnam[100]; |
1224 | pid_t pid; | |
237e200e | 1225 | int fd, ret; |
b10bdd6c FG |
1226 | size_t stack_size = sysconf(_SC_PAGESIZE); |
1227 | void *stack = alloca(stack_size); | |
237e200e SH |
1228 | |
1229 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1230 | if (ret < 0 || ret >= sizeof(fnam)) | |
1231 | _exit(1); | |
1232 | ||
1233 | fd = open(fnam, O_RDONLY); | |
1234 | if (fd < 0) { | |
1235 | perror("write_task_init_pid_exit open of ns/pid"); | |
1236 | _exit(1); | |
1237 | } | |
1238 | if (setns(fd, 0)) { | |
1239 | perror("write_task_init_pid_exit setns 1"); | |
1240 | close(fd); | |
1241 | _exit(1); | |
1242 | } | |
b10bdd6c | 1243 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1244 | if (pid < 0) |
1245 | _exit(1); | |
1246 | if (pid != 0) { | |
1247 | if (!wait_for_pid(pid)) | |
1248 | _exit(1); | |
1249 | _exit(0); | |
1250 | } | |
b10bdd6c FG |
1251 | } |
1252 | ||
1253 | static int send_creds_clone_wrapper(void *arg) { | |
1254 | struct ucred cred; | |
1255 | char v; | |
1256 | int sock = *(int *)arg; | |
237e200e SH |
1257 | |
1258 | /* we are the child */ | |
1259 | cred.uid = 0; | |
1260 | cred.gid = 0; | |
1261 | cred.pid = 1; | |
1262 | v = '1'; | |
1263 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1264 | return 1; |
1265 | return 0; | |
237e200e SH |
1266 | } |
1267 | ||
1268 | static pid_t get_init_pid_for_task(pid_t task) | |
1269 | { | |
1270 | int sock[2]; | |
1271 | pid_t pid; | |
1272 | pid_t ret = -1; | |
1273 | char v = '0'; | |
1274 | struct ucred cred; | |
1275 | ||
1276 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1277 | perror("socketpair"); | |
1278 | return -1; | |
1279 | } | |
1280 | ||
1281 | pid = fork(); | |
1282 | if (pid < 0) | |
1283 | goto out; | |
1284 | if (!pid) { | |
1285 | close(sock[1]); | |
1286 | write_task_init_pid_exit(sock[0], task); | |
1287 | _exit(0); | |
1288 | } | |
1289 | ||
1290 | if (!recv_creds(sock[1], &cred, &v)) | |
1291 | goto out; | |
1292 | ret = cred.pid; | |
1293 | ||
1294 | out: | |
1295 | close(sock[0]); | |
1296 | close(sock[1]); | |
1297 | if (pid > 0) | |
1298 | wait_for_pid(pid); | |
1299 | return ret; | |
1300 | } | |
1301 | ||
71f17cd2 | 1302 | pid_t lookup_initpid_in_store(pid_t qpid) |
237e200e SH |
1303 | { |
1304 | pid_t answer = 0; | |
1305 | struct stat sb; | |
1306 | struct pidns_init_store *e; | |
1307 | char fnam[100]; | |
1308 | ||
1309 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1310 | store_lock(); | |
1311 | if (stat(fnam, &sb) < 0) | |
1312 | goto out; | |
1313 | e = lookup_verify_initpid(&sb); | |
1314 | if (e) { | |
1315 | answer = e->initpid; | |
1316 | goto out; | |
1317 | } | |
1318 | answer = get_init_pid_for_task(qpid); | |
1319 | if (answer > 0) | |
1320 | save_initpid(&sb, answer); | |
1321 | ||
1322 | out: | |
1323 | /* we prune at end in case we are returning | |
1324 | * the value we were about to return */ | |
1325 | prune_initpid_store(); | |
1326 | store_unlock(); | |
1327 | return answer; | |
1328 | } | |
1329 | ||
1330 | static int wait_for_pid(pid_t pid) | |
1331 | { | |
1332 | int status, ret; | |
1333 | ||
1334 | if (pid <= 0) | |
1335 | return -1; | |
1336 | ||
1337 | again: | |
1338 | ret = waitpid(pid, &status, 0); | |
1339 | if (ret == -1) { | |
1340 | if (errno == EINTR) | |
1341 | goto again; | |
1342 | return -1; | |
1343 | } | |
1344 | if (ret != pid) | |
1345 | goto again; | |
1346 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1347 | return -1; | |
1348 | return 0; | |
1349 | } | |
1350 | ||
237e200e | 1351 | /* |
db1b32f6 SX |
1352 | * append the given formatted string to *src. |
1353 | * src: a pointer to a char* in which to append the formatted string. | |
237e200e SH |
1354 | * sz: the number of characters printed so far, minus trailing \0. |
1355 | * asz: the allocated size so far | |
db1b32f6 SX |
1356 | * format: string format. See printf for details. |
1357 | * ...: varargs. See printf for details. | |
237e200e | 1358 | */ |
db1b32f6 | 1359 | static void must_strcat(char **src, size_t *sz, size_t *asz, const char *format, ...) |
237e200e | 1360 | { |
db1b32f6 SX |
1361 | char tmp[BUF_RESERVE_SIZE]; |
1362 | va_list args; | |
237e200e | 1363 | |
db1b32f6 SX |
1364 | va_start (args, format); |
1365 | int tmplen = vsnprintf(tmp, BUF_RESERVE_SIZE, format, args); | |
1366 | va_end(args); | |
237e200e SH |
1367 | |
1368 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1369 | char *tmp; | |
1370 | do { | |
1371 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1372 | } while (!tmp); | |
1373 | *src = tmp; | |
1374 | *asz += BUF_RESERVE_SIZE; | |
1375 | } | |
bbfd0e33 | 1376 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1377 | *sz += tmplen; |
237e200e SH |
1378 | } |
1379 | ||
db1b32f6 SX |
1380 | /* |
1381 | * append pid to *src. | |
1382 | * src: a pointer to a char* in which ot append the pid. | |
1383 | * sz: the number of characters printed so far, minus trailing \0. | |
1384 | * asz: the allocated size so far | |
1385 | * pid: the pid to append | |
1386 | */ | |
1387 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1388 | { | |
1389 | must_strcat(src, sz, asz, "%d\n", (int)pid); | |
1390 | } | |
1391 | ||
237e200e SH |
1392 | /* |
1393 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1394 | * valid in the caller's namespace, return the id mapped into | |
1395 | * pid's namespace. | |
1396 | * Returns the mapped id, or -1 on error. | |
1397 | */ | |
1398 | unsigned int | |
1399 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1400 | { | |
1401 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1402 | hostuid, // base id for a range in the caller's namespace | |
1403 | count; // number of ids in this range | |
1404 | char line[400]; | |
1405 | int ret; | |
1406 | ||
1407 | fseek(idfile, 0L, SEEK_SET); | |
1408 | while (fgets(line, 400, idfile)) { | |
1409 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1410 | if (ret != 3) | |
1411 | continue; | |
1412 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1413 | /* | |
1414 | * uids wrapped around - unexpected as this is a procfile, | |
1415 | * so just bail. | |
1416 | */ | |
b8defc3d | 1417 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", |
237e200e SH |
1418 | nsuid, hostuid, count, line); |
1419 | return -1; | |
1420 | } | |
1421 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1422 | /* | |
1423 | * now since hostuid <= in_id < hostuid+count, and | |
1424 | * hostuid+count and nsuid+count do not wrap around, | |
1425 | * we know that nsuid+(in_id-hostuid) which must be | |
1426 | * less that nsuid+(count) must not wrap around | |
1427 | */ | |
1428 | return (in_id - hostuid) + nsuid; | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | // no answer found | |
1433 | return -1; | |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * for is_privileged_over, | |
1438 | * specify whether we require the calling uid to be root in his | |
1439 | * namespace | |
1440 | */ | |
1441 | #define NS_ROOT_REQD true | |
1442 | #define NS_ROOT_OPT false | |
1443 | ||
1444 | #define PROCLEN 100 | |
1445 | ||
1446 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1447 | { | |
1448 | char fpath[PROCLEN]; | |
1449 | int ret; | |
1450 | bool answer = false; | |
1451 | uid_t nsuid; | |
1452 | ||
1453 | if (victim == -1 || uid == -1) | |
1454 | return false; | |
1455 | ||
1456 | /* | |
1457 | * If the request is one not requiring root in the namespace, | |
1458 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1459 | * access to files owned by uid 1000 | |
1460 | */ | |
1461 | if (!req_ns_root && uid == victim) | |
1462 | return true; | |
1463 | ||
1464 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1465 | if (ret < 0 || ret >= PROCLEN) | |
1466 | return false; | |
1467 | FILE *f = fopen(fpath, "r"); | |
1468 | if (!f) | |
1469 | return false; | |
1470 | ||
1471 | /* if caller's not root in his namespace, reject */ | |
1472 | nsuid = convert_id_to_ns(f, uid); | |
1473 | if (nsuid) | |
1474 | goto out; | |
1475 | ||
1476 | /* | |
1477 | * If victim is not mapped into caller's ns, reject. | |
1478 | * XXX I'm not sure this check is needed given that fuse | |
1479 | * will be sending requests where the vfs has converted | |
1480 | */ | |
1481 | nsuid = convert_id_to_ns(f, victim); | |
1482 | if (nsuid == -1) | |
1483 | goto out; | |
1484 | ||
1485 | answer = true; | |
1486 | ||
1487 | out: | |
1488 | fclose(f); | |
1489 | return answer; | |
1490 | } | |
1491 | ||
1492 | static bool perms_include(int fmode, mode_t req_mode) | |
1493 | { | |
1494 | mode_t r; | |
1495 | ||
1496 | switch (req_mode & O_ACCMODE) { | |
1497 | case O_RDONLY: | |
1498 | r = S_IROTH; | |
1499 | break; | |
1500 | case O_WRONLY: | |
1501 | r = S_IWOTH; | |
1502 | break; | |
1503 | case O_RDWR: | |
1504 | r = S_IROTH | S_IWOTH; | |
1505 | break; | |
1506 | default: | |
1507 | return false; | |
1508 | } | |
1509 | return ((fmode & r) == r); | |
1510 | } | |
1511 | ||
1512 | ||
1513 | /* | |
1514 | * taskcg is a/b/c | |
1515 | * querycg is /a/b/c/d/e | |
1516 | * we return 'd' | |
1517 | */ | |
1518 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1519 | { | |
1520 | char *start, *end; | |
1521 | ||
1522 | if (strlen(taskcg) <= strlen(querycg)) { | |
b8defc3d | 1523 | lxcfs_error("%s\n", "I was fed bad input."); |
237e200e SH |
1524 | return NULL; |
1525 | } | |
1526 | ||
06081b29 | 1527 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1528 | start = strdup(taskcg + 1); |
1529 | else | |
1530 | start = strdup(taskcg + strlen(querycg) + 1); | |
1531 | if (!start) | |
1532 | return NULL; | |
1533 | end = strchr(start, '/'); | |
1534 | if (end) | |
1535 | *end = '\0'; | |
1536 | return start; | |
1537 | } | |
1538 | ||
71f17cd2 | 1539 | char *get_pid_cgroup(pid_t pid, const char *contrl) |
237e200e | 1540 | { |
5dd3e6fd | 1541 | int cfd; |
237e200e | 1542 | |
d298bba1 | 1543 | cfd = get_cgroup_fd(contrl); |
5fbea8a6 CB |
1544 | if (cfd < 0) |
1545 | return false; | |
237e200e | 1546 | |
5fbea8a6 CB |
1547 | if (pure_unified_layout(cgroup_ops)) |
1548 | return cg_unified_get_current_cgroup(pid); | |
237e200e | 1549 | |
5fbea8a6 | 1550 | return cg_legacy_get_current_cgroup(pid, contrl); |
237e200e SH |
1551 | } |
1552 | ||
1553 | /* | |
1554 | * check whether a fuse context may access a cgroup dir or file | |
1555 | * | |
1556 | * If file is not null, it is a cgroup file to check under cg. | |
1557 | * If file is null, then we are checking perms on cg itself. | |
1558 | * | |
1559 | * For files we can check the mode of the list_keys result. | |
1560 | * For cgroups, we must make assumptions based on the files under the | |
1561 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1562 | * yet. | |
1563 | */ | |
1564 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1565 | { | |
1566 | struct cgfs_files *k = NULL; | |
1567 | bool ret = false; | |
1568 | ||
1569 | k = cgfs_get_key(contrl, cg, file); | |
1570 | if (!k) | |
1571 | return false; | |
1572 | ||
1573 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1574 | if (perms_include(k->mode >> 6, mode)) { | |
1575 | ret = true; | |
1576 | goto out; | |
1577 | } | |
1578 | } | |
1579 | if (fc->gid == k->gid) { | |
1580 | if (perms_include(k->mode >> 3, mode)) { | |
1581 | ret = true; | |
1582 | goto out; | |
1583 | } | |
1584 | } | |
1585 | ret = perms_include(k->mode, mode); | |
1586 | ||
1587 | out: | |
1588 | free_key(k); | |
1589 | return ret; | |
1590 | } | |
1591 | ||
1592 | #define INITSCOPE "/init.scope" | |
71f17cd2 | 1593 | void prune_init_slice(char *cg) |
237e200e SH |
1594 | { |
1595 | char *point; | |
1596 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1597 | ||
1598 | if (cg_len < initscope_len) | |
1599 | return; | |
1600 | ||
1601 | point = cg + cg_len - initscope_len; | |
1602 | if (strcmp(point, INITSCOPE) == 0) { | |
1603 | if (point == cg) | |
1604 | *(point+1) = '\0'; | |
1605 | else | |
1606 | *point = '\0'; | |
1607 | } | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1612 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1613 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1614 | * to a string containing the next cgroup directory under cg, which must be | |
1615 | * freed by the caller. | |
1616 | */ | |
1617 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1618 | { | |
1619 | bool answer = false; | |
1620 | char *c2 = get_pid_cgroup(pid, contrl); | |
1621 | char *linecmp; | |
1622 | ||
1623 | if (!c2) | |
1624 | return false; | |
1625 | prune_init_slice(c2); | |
1626 | ||
1627 | /* | |
12c31268 CB |
1628 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1629 | * they pass in a cgroup without leading '/' | |
1630 | * | |
1631 | * The original line here was: | |
1632 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1633 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1634 | * Serge, do you know? | |
237e200e | 1635 | */ |
12c31268 CB |
1636 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1637 | linecmp = c2; | |
1638 | else | |
1639 | linecmp = c2 + 1; | |
237e200e SH |
1640 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1641 | if (nextcg) { | |
1642 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1643 | } | |
1644 | goto out; | |
1645 | } | |
1646 | answer = true; | |
1647 | ||
1648 | out: | |
1649 | free(c2); | |
1650 | return answer; | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1655 | */ | |
1656 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1657 | { | |
1658 | bool answer = false; | |
1659 | char *c2, *task_cg; | |
1660 | size_t target_len, task_len; | |
1661 | ||
f7bff426 | 1662 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1663 | return true; |
1664 | ||
1665 | c2 = get_pid_cgroup(pid, contrl); | |
1666 | if (!c2) | |
1667 | return false; | |
1668 | prune_init_slice(c2); | |
1669 | ||
1670 | task_cg = c2 + 1; | |
1671 | target_len = strlen(cg); | |
1672 | task_len = strlen(task_cg); | |
1673 | if (task_len == 0) { | |
1674 | /* Task is in the root cg, it can see everything. This case is | |
1675 | * not handled by the strmcps below, since they test for the | |
1676 | * last /, but that is the first / that we've chopped off | |
1677 | * above. | |
1678 | */ | |
1679 | answer = true; | |
1680 | goto out; | |
1681 | } | |
1682 | if (strcmp(cg, task_cg) == 0) { | |
1683 | answer = true; | |
1684 | goto out; | |
1685 | } | |
1686 | if (target_len < task_len) { | |
1687 | /* looking up a parent dir */ | |
1688 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1689 | answer = true; | |
1690 | goto out; | |
1691 | } | |
1692 | if (target_len > task_len) { | |
1693 | /* looking up a child dir */ | |
1694 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1695 | answer = true; | |
1696 | goto out; | |
1697 | } | |
1698 | ||
1699 | out: | |
1700 | free(c2); | |
1701 | return answer; | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * given /cgroup/freezer/a/b, return "freezer". | |
1706 | * the returned char* should NOT be freed. | |
1707 | */ | |
1708 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1709 | { | |
1710 | const char *p1; | |
1711 | char *contr, *slash; | |
1712 | ||
99142521 | 1713 | if (strlen(path) < 9) { |
e254948f | 1714 | errno = EACCES; |
237e200e | 1715 | return NULL; |
99142521 CB |
1716 | } |
1717 | if (*(path + 7) != '/') { | |
1718 | errno = EINVAL; | |
237e200e | 1719 | return NULL; |
99142521 | 1720 | } |
3adc421c | 1721 | p1 = path + 8; |
237e200e | 1722 | contr = strdupa(p1); |
99142521 CB |
1723 | if (!contr) { |
1724 | errno = ENOMEM; | |
237e200e | 1725 | return NULL; |
99142521 | 1726 | } |
237e200e SH |
1727 | slash = strstr(contr, "/"); |
1728 | if (slash) | |
1729 | *slash = '\0'; | |
1730 | ||
5fbea8a6 CB |
1731 | for (struct hierarchy **h = cgroup_ops->hierarchies; h && *h; h++) { |
1732 | if ((*h)->__controllers && strcmp((*h)->__controllers, contr) == 0) | |
1733 | return (*h)->__controllers; | |
237e200e | 1734 | } |
99142521 | 1735 | errno = ENOENT; |
237e200e SH |
1736 | return NULL; |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1741 | * Note that the returned value may include files (keynames) etc | |
1742 | */ | |
1743 | static const char *find_cgroup_in_path(const char *path) | |
1744 | { | |
1745 | const char *p1; | |
1746 | ||
bc70ba9b | 1747 | if (strlen(path) < 9) { |
e254948f | 1748 | errno = EACCES; |
237e200e | 1749 | return NULL; |
bc70ba9b CB |
1750 | } |
1751 | p1 = strstr(path + 8, "/"); | |
1752 | if (!p1) { | |
1753 | errno = EINVAL; | |
237e200e | 1754 | return NULL; |
bc70ba9b CB |
1755 | } |
1756 | errno = 0; | |
1757 | return p1 + 1; | |
237e200e SH |
1758 | } |
1759 | ||
1760 | /* | |
1761 | * split the last path element from the path in @cg. | |
1762 | * @dir is newly allocated and should be freed, @last not | |
1763 | */ | |
1764 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1765 | { | |
1766 | char *p; | |
1767 | ||
1768 | do { | |
1769 | *dir = strdup(cg); | |
1770 | } while (!*dir); | |
1771 | *last = strrchr(cg, '/'); | |
1772 | if (!*last) { | |
1773 | *last = NULL; | |
1774 | return; | |
1775 | } | |
1776 | p = strrchr(*dir, '/'); | |
1777 | *p = '\0'; | |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * FUSE ops for /cgroup | |
1782 | */ | |
1783 | ||
1784 | int cg_getattr(const char *path, struct stat *sb) | |
1785 | { | |
1786 | struct timespec now; | |
1787 | struct fuse_context *fc = fuse_get_context(); | |
1788 | char * cgdir = NULL; | |
1789 | char *last = NULL, *path1, *path2; | |
1790 | struct cgfs_files *k = NULL; | |
1791 | const char *cgroup; | |
1792 | const char *controller = NULL; | |
1793 | int ret = -ENOENT; | |
1794 | ||
1795 | ||
5fbea8a6 | 1796 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
1797 | return -EIO; |
1798 | ||
1799 | memset(sb, 0, sizeof(struct stat)); | |
1800 | ||
1801 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1802 | return -EINVAL; | |
1803 | ||
1804 | sb->st_uid = sb->st_gid = 0; | |
1805 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1806 | sb->st_size = 0; | |
1807 | ||
1808 | if (strcmp(path, "/cgroup") == 0) { | |
1809 | sb->st_mode = S_IFDIR | 00755; | |
1810 | sb->st_nlink = 2; | |
1811 | return 0; | |
1812 | } | |
1813 | ||
1814 | controller = pick_controller_from_path(fc, path); | |
1815 | if (!controller) | |
2f7036d0 | 1816 | return -errno; |
237e200e SH |
1817 | cgroup = find_cgroup_in_path(path); |
1818 | if (!cgroup) { | |
1819 | /* this is just /cgroup/controller, return it as a dir */ | |
1820 | sb->st_mode = S_IFDIR | 00755; | |
1821 | sb->st_nlink = 2; | |
1822 | return 0; | |
1823 | } | |
1824 | ||
1825 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1826 | ||
1827 | if (!last) { | |
1828 | path1 = "/"; | |
1829 | path2 = cgdir; | |
1830 | } else { | |
1831 | path1 = cgdir; | |
1832 | path2 = last; | |
1833 | } | |
1834 | ||
1835 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 1836 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
1837 | initpid = fc->pid; |
1838 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1839 | * Then check that caller's cgroup is under path if last is a child | |
1840 | * cgroup, or cgdir if last is a file */ | |
1841 | ||
1842 | if (is_child_cgroup(controller, path1, path2)) { | |
1843 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1844 | ret = -ENOENT; | |
1845 | goto out; | |
1846 | } | |
1847 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1848 | /* this is just /cgroup/controller, return it as a dir */ | |
1849 | sb->st_mode = S_IFDIR | 00555; | |
1850 | sb->st_nlink = 2; | |
1851 | ret = 0; | |
1852 | goto out; | |
1853 | } | |
1854 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1855 | ret = -EACCES; | |
1856 | goto out; | |
1857 | } | |
1858 | ||
1859 | // get uid, gid, from '/tasks' file and make up a mode | |
1860 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1861 | sb->st_mode = S_IFDIR | 00755; | |
1862 | k = cgfs_get_key(controller, cgroup, NULL); | |
1863 | if (!k) { | |
1864 | sb->st_uid = sb->st_gid = 0; | |
1865 | } else { | |
1866 | sb->st_uid = k->uid; | |
1867 | sb->st_gid = k->gid; | |
1868 | } | |
1869 | free_key(k); | |
1870 | sb->st_nlink = 2; | |
1871 | ret = 0; | |
1872 | goto out; | |
1873 | } | |
1874 | ||
1875 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1876 | sb->st_mode = S_IFREG | k->mode; | |
1877 | sb->st_nlink = 1; | |
1878 | sb->st_uid = k->uid; | |
1879 | sb->st_gid = k->gid; | |
1880 | sb->st_size = 0; | |
1881 | free_key(k); | |
1882 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1883 | ret = -ENOENT; | |
1884 | goto out; | |
1885 | } | |
237e200e SH |
1886 | ret = 0; |
1887 | } | |
1888 | ||
1889 | out: | |
1890 | free(cgdir); | |
1891 | return ret; | |
1892 | } | |
1893 | ||
1894 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1895 | { | |
1896 | struct fuse_context *fc = fuse_get_context(); | |
1897 | const char *cgroup; | |
1898 | struct file_info *dir_info; | |
1899 | char *controller = NULL; | |
1900 | ||
5fbea8a6 | 1901 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
1902 | return -EIO; |
1903 | ||
1904 | if (strcmp(path, "/cgroup") == 0) { | |
1905 | cgroup = NULL; | |
1906 | controller = NULL; | |
1907 | } else { | |
1908 | // return list of keys for the controller, and list of child cgroups | |
1909 | controller = pick_controller_from_path(fc, path); | |
1910 | if (!controller) | |
2f7036d0 | 1911 | return -errno; |
237e200e SH |
1912 | |
1913 | cgroup = find_cgroup_in_path(path); | |
1914 | if (!cgroup) { | |
1915 | /* this is just /cgroup/controller, return its contents */ | |
1916 | cgroup = "/"; | |
1917 | } | |
1918 | } | |
1919 | ||
1920 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 1921 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
1922 | initpid = fc->pid; |
1923 | if (cgroup) { | |
1924 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1925 | return -ENOENT; | |
1926 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1927 | return -EACCES; | |
1928 | } | |
1929 | ||
1930 | /* we'll free this at cg_releasedir */ | |
1931 | dir_info = malloc(sizeof(*dir_info)); | |
1932 | if (!dir_info) | |
1933 | return -ENOMEM; | |
1934 | dir_info->controller = must_copy_string(controller); | |
1935 | dir_info->cgroup = must_copy_string(cgroup); | |
1936 | dir_info->type = LXC_TYPE_CGDIR; | |
1937 | dir_info->buf = NULL; | |
1938 | dir_info->file = NULL; | |
1939 | dir_info->buflen = 0; | |
1940 | ||
1941 | fi->fh = (unsigned long)dir_info; | |
1942 | return 0; | |
1943 | } | |
1944 | ||
1945 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1946 | struct fuse_file_info *fi) | |
1947 | { | |
1948 | struct file_info *d = (struct file_info *)fi->fh; | |
1949 | struct cgfs_files **list = NULL; | |
1950 | int i, ret; | |
1951 | char *nextcg = NULL; | |
1952 | struct fuse_context *fc = fuse_get_context(); | |
1953 | char **clist = NULL; | |
1954 | ||
5fbea8a6 CB |
1955 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
1956 | return -EIO; | |
1957 | ||
d639f863 CB |
1958 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) |
1959 | return -EIO; | |
1960 | ||
237e200e | 1961 | if (d->type != LXC_TYPE_CGDIR) { |
b8defc3d | 1962 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); |
237e200e SH |
1963 | return -EIO; |
1964 | } | |
1965 | if (!d->cgroup && !d->controller) { | |
5fbea8a6 CB |
1966 | /* |
1967 | * ls /var/lib/lxcfs/cgroup - just show list of controllers. | |
1968 | * This only works with the legacy hierarchy. | |
1969 | */ | |
1970 | for (struct hierarchy **h = cgroup_ops->hierarchies; h && *h; h++) { | |
1971 | if (is_unified_hierarchy(*h)) | |
1972 | continue; | |
237e200e | 1973 | |
5fbea8a6 | 1974 | if ((*h)->__controllers && filler(buf, (*h)->__controllers, NULL, 0)) |
237e200e | 1975 | return -EIO; |
237e200e | 1976 | } |
5fbea8a6 | 1977 | |
237e200e SH |
1978 | return 0; |
1979 | } | |
1980 | ||
1981 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1982 | // not a valid cgroup | |
1983 | ret = -EINVAL; | |
1984 | goto out; | |
1985 | } | |
1986 | ||
1987 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 1988 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
1989 | initpid = fc->pid; |
1990 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1991 | if (nextcg) { | |
1992 | ret = filler(buf, nextcg, NULL, 0); | |
1993 | free(nextcg); | |
1994 | if (ret != 0) { | |
1995 | ret = -EIO; | |
1996 | goto out; | |
1997 | } | |
1998 | } | |
1999 | ret = 0; | |
2000 | goto out; | |
2001 | } | |
2002 | ||
b737a54a | 2003 | for (i = 0; list && list[i]; i++) { |
237e200e SH |
2004 | if (filler(buf, list[i]->name, NULL, 0) != 0) { |
2005 | ret = -EIO; | |
2006 | goto out; | |
2007 | } | |
2008 | } | |
2009 | ||
2010 | // now get the list of child cgroups | |
2011 | ||
2012 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
2013 | ret = 0; | |
2014 | goto out; | |
2015 | } | |
f366da65 WB |
2016 | if (clist) { |
2017 | for (i = 0; clist[i]; i++) { | |
2018 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
2019 | ret = -EIO; | |
2020 | goto out; | |
2021 | } | |
237e200e SH |
2022 | } |
2023 | } | |
2024 | ret = 0; | |
2025 | ||
2026 | out: | |
2027 | free_keys(list); | |
2028 | if (clist) { | |
2029 | for (i = 0; clist[i]; i++) | |
2030 | free(clist[i]); | |
2031 | free(clist); | |
2032 | } | |
2033 | return ret; | |
2034 | } | |
2035 | ||
71f17cd2 | 2036 | void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 2037 | { |
43215927 SH |
2038 | struct file_info *f = (struct file_info *)fi->fh; |
2039 | ||
237e200e SH |
2040 | if (!f) |
2041 | return; | |
43215927 SH |
2042 | |
2043 | fi->fh = 0; | |
2044 | ||
54a6d46a CB |
2045 | free_disarm(f->controller); |
2046 | free_disarm(f->cgroup); | |
2047 | free_disarm(f->file); | |
2048 | free_disarm(f->buf); | |
2049 | free_disarm(f); | |
237e200e SH |
2050 | } |
2051 | ||
2052 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
2053 | { | |
43215927 | 2054 | do_release_file_info(fi); |
237e200e SH |
2055 | return 0; |
2056 | } | |
2057 | ||
2058 | int cg_open(const char *path, struct fuse_file_info *fi) | |
2059 | { | |
2060 | const char *cgroup; | |
2061 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
2062 | struct cgfs_files *k = NULL; | |
2063 | struct file_info *file_info; | |
2064 | struct fuse_context *fc = fuse_get_context(); | |
2065 | int ret; | |
2066 | ||
5fbea8a6 | 2067 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
2068 | return -EIO; |
2069 | ||
2070 | controller = pick_controller_from_path(fc, path); | |
2071 | if (!controller) | |
2f7036d0 | 2072 | return -errno; |
237e200e SH |
2073 | cgroup = find_cgroup_in_path(path); |
2074 | if (!cgroup) | |
bc70ba9b | 2075 | return -errno; |
237e200e SH |
2076 | |
2077 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2078 | if (!last) { | |
2079 | path1 = "/"; | |
2080 | path2 = cgdir; | |
2081 | } else { | |
2082 | path1 = cgdir; | |
2083 | path2 = last; | |
2084 | } | |
2085 | ||
2086 | k = cgfs_get_key(controller, path1, path2); | |
2087 | if (!k) { | |
2088 | ret = -EINVAL; | |
2089 | goto out; | |
2090 | } | |
2091 | free_key(k); | |
2092 | ||
2093 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2094 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
2095 | initpid = fc->pid; |
2096 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2097 | ret = -ENOENT; | |
2098 | goto out; | |
2099 | } | |
2100 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
2101 | ret = -EACCES; |
2102 | goto out; | |
2103 | } | |
2104 | ||
2105 | /* we'll free this at cg_release */ | |
2106 | file_info = malloc(sizeof(*file_info)); | |
2107 | if (!file_info) { | |
2108 | ret = -ENOMEM; | |
2109 | goto out; | |
2110 | } | |
2111 | file_info->controller = must_copy_string(controller); | |
2112 | file_info->cgroup = must_copy_string(path1); | |
2113 | file_info->file = must_copy_string(path2); | |
2114 | file_info->type = LXC_TYPE_CGFILE; | |
2115 | file_info->buf = NULL; | |
2116 | file_info->buflen = 0; | |
2117 | ||
2118 | fi->fh = (unsigned long)file_info; | |
2119 | ret = 0; | |
2120 | ||
2121 | out: | |
2122 | free(cgdir); | |
2123 | return ret; | |
2124 | } | |
2125 | ||
bddbb106 SH |
2126 | int cg_access(const char *path, int mode) |
2127 | { | |
6f0f6b83 | 2128 | int ret; |
bddbb106 | 2129 | const char *cgroup; |
6f0f6b83 CB |
2130 | char *path1, *path2, *controller; |
2131 | char *last = NULL, *cgdir = NULL; | |
bddbb106 SH |
2132 | struct cgfs_files *k = NULL; |
2133 | struct fuse_context *fc = fuse_get_context(); | |
6f0f6b83 | 2134 | |
5fbea8a6 CB |
2135 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
2136 | return -EIO; | |
2137 | ||
9873c5e8 | 2138 | if (strcmp(path, "/cgroup") == 0) |
6f0f6b83 | 2139 | return 0; |
bddbb106 | 2140 | |
bddbb106 SH |
2141 | controller = pick_controller_from_path(fc, path); |
2142 | if (!controller) | |
2f7036d0 | 2143 | return -errno; |
bddbb106 | 2144 | cgroup = find_cgroup_in_path(path); |
575316c4 SH |
2145 | if (!cgroup) { |
2146 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
2147 | if ((mode & W_OK) == 0) |
2148 | return 0; | |
2149 | return -EACCES; | |
575316c4 | 2150 | } |
bddbb106 SH |
2151 | |
2152 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2153 | if (!last) { | |
2154 | path1 = "/"; | |
2155 | path2 = cgdir; | |
2156 | } else { | |
2157 | path1 = cgdir; | |
2158 | path2 = last; | |
2159 | } | |
2160 | ||
2161 | k = cgfs_get_key(controller, path1, path2); | |
2162 | if (!k) { | |
3f441bc7 SH |
2163 | if ((mode & W_OK) == 0) |
2164 | ret = 0; | |
2165 | else | |
2166 | ret = -EACCES; | |
bddbb106 SH |
2167 | goto out; |
2168 | } | |
2169 | free_key(k); | |
2170 | ||
2171 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2172 | if (initpid <= 1 || is_shared_pidns(initpid)) |
bddbb106 SH |
2173 | initpid = fc->pid; |
2174 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2175 | ret = -ENOENT; | |
2176 | goto out; | |
2177 | } | |
2178 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
2179 | ret = -EACCES; | |
2180 | goto out; | |
2181 | } | |
2182 | ||
2183 | ret = 0; | |
2184 | ||
2185 | out: | |
2186 | free(cgdir); | |
2187 | return ret; | |
2188 | } | |
2189 | ||
237e200e SH |
2190 | int cg_release(const char *path, struct fuse_file_info *fi) |
2191 | { | |
43215927 | 2192 | do_release_file_info(fi); |
237e200e SH |
2193 | return 0; |
2194 | } | |
2195 | ||
2196 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
2197 | ||
2198 | static bool wait_for_sock(int sock, int timeout) | |
2199 | { | |
2200 | struct epoll_event ev; | |
2201 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
2202 | ||
2203 | if ((starttime = time(NULL)) < 0) | |
2204 | return false; | |
2205 | ||
2206 | if ((epfd = epoll_create(1)) < 0) { | |
b8defc3d | 2207 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); |
237e200e SH |
2208 | return false; |
2209 | } | |
2210 | ||
2211 | ev.events = POLLIN_SET; | |
2212 | ev.data.fd = sock; | |
2213 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
b8defc3d | 2214 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); |
237e200e SH |
2215 | close(epfd); |
2216 | return false; | |
2217 | } | |
2218 | ||
2219 | again: | |
2220 | if ((now = time(NULL)) < 0) { | |
2221 | close(epfd); | |
2222 | return false; | |
2223 | } | |
2224 | ||
2225 | deltatime = (starttime + timeout) - now; | |
2226 | if (deltatime < 0) { // timeout | |
2227 | errno = 0; | |
2228 | close(epfd); | |
2229 | return false; | |
2230 | } | |
2231 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2232 | if (ret < 0 && errno == EINTR) | |
2233 | goto again; | |
2234 | saved_errno = errno; | |
2235 | close(epfd); | |
2236 | ||
2237 | if (ret <= 0) { | |
2238 | errno = saved_errno; | |
2239 | return false; | |
2240 | } | |
2241 | return true; | |
2242 | } | |
2243 | ||
2244 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2245 | { | |
2246 | if (!wait_for_sock(sockfd, 2)) | |
2247 | return -1; | |
2248 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2249 | } | |
2250 | ||
2251 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2252 | { | |
2253 | struct msghdr msg = { 0 }; | |
2254 | struct iovec iov; | |
2255 | struct cmsghdr *cmsg; | |
2256 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2257 | char buf[1]; | |
2258 | buf[0] = 'p'; | |
2259 | ||
2260 | if (pingfirst) { | |
2261 | if (msgrecv(sock, buf, 1) != 1) { | |
b8defc3d | 2262 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); |
237e200e SH |
2263 | return SEND_CREDS_FAIL; |
2264 | } | |
2265 | } | |
2266 | ||
2267 | msg.msg_control = cmsgbuf; | |
2268 | msg.msg_controllen = sizeof(cmsgbuf); | |
2269 | ||
2270 | cmsg = CMSG_FIRSTHDR(&msg); | |
2271 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2272 | cmsg->cmsg_level = SOL_SOCKET; | |
2273 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2274 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2275 | ||
2276 | msg.msg_name = NULL; | |
2277 | msg.msg_namelen = 0; | |
2278 | ||
2279 | buf[0] = v; | |
2280 | iov.iov_base = buf; | |
2281 | iov.iov_len = sizeof(buf); | |
2282 | msg.msg_iov = &iov; | |
2283 | msg.msg_iovlen = 1; | |
2284 | ||
2285 | if (sendmsg(sock, &msg, 0) < 0) { | |
b8defc3d | 2286 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); |
237e200e SH |
2287 | if (errno == 3) |
2288 | return SEND_CREDS_NOTSK; | |
2289 | return SEND_CREDS_FAIL; | |
2290 | } | |
2291 | ||
2292 | return SEND_CREDS_OK; | |
2293 | } | |
2294 | ||
2295 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2296 | { | |
2297 | struct msghdr msg = { 0 }; | |
2298 | struct iovec iov; | |
2299 | struct cmsghdr *cmsg; | |
2300 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2301 | char buf[1]; | |
2302 | int ret; | |
2303 | int optval = 1; | |
2304 | ||
2305 | *v = '1'; | |
2306 | ||
2307 | cred->pid = -1; | |
2308 | cred->uid = -1; | |
2309 | cred->gid = -1; | |
2310 | ||
2311 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
b8defc3d | 2312 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); |
237e200e SH |
2313 | return false; |
2314 | } | |
2315 | buf[0] = '1'; | |
2316 | if (write(sock, buf, 1) != 1) { | |
b8defc3d | 2317 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); |
237e200e SH |
2318 | return false; |
2319 | } | |
2320 | ||
2321 | msg.msg_name = NULL; | |
2322 | msg.msg_namelen = 0; | |
2323 | msg.msg_control = cmsgbuf; | |
2324 | msg.msg_controllen = sizeof(cmsgbuf); | |
2325 | ||
2326 | iov.iov_base = buf; | |
2327 | iov.iov_len = sizeof(buf); | |
2328 | msg.msg_iov = &iov; | |
2329 | msg.msg_iovlen = 1; | |
2330 | ||
2331 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2332 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2333 | return false; |
2334 | } | |
2335 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2336 | if (ret < 0) { | |
b8defc3d | 2337 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2338 | return false; |
2339 | } | |
2340 | ||
2341 | cmsg = CMSG_FIRSTHDR(&msg); | |
2342 | ||
2343 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2344 | cmsg->cmsg_level == SOL_SOCKET && | |
2345 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2346 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2347 | } | |
2348 | *v = buf[0]; | |
2349 | ||
2350 | return true; | |
2351 | } | |
2352 | ||
35174b0f FG |
2353 | struct pid_ns_clone_args { |
2354 | int *cpipe; | |
2355 | int sock; | |
2356 | pid_t tpid; | |
2357 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2358 | }; | |
2359 | ||
2360 | /* | |
2361 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2362 | * with clone(). This simply writes '1' as ACK back to the parent | |
2363 | * before calling the actual wrapped function. | |
2364 | */ | |
2365 | static int pid_ns_clone_wrapper(void *arg) { | |
2366 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2367 | char b = '1'; | |
2368 | ||
2369 | close(args->cpipe[0]); | |
b8defc3d CB |
2370 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) |
2371 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
35174b0f FG |
2372 | close(args->cpipe[1]); |
2373 | return args->wrapped(args->sock, args->tpid); | |
2374 | } | |
237e200e SH |
2375 | |
2376 | /* | |
2377 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2378 | * int value back over the socket. This shifts the pid from the | |
2379 | * sender's pidns into tpid's pidns. | |
2380 | */ | |
35174b0f | 2381 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2382 | { |
2383 | char v = '0'; | |
2384 | struct ucred cred; | |
2385 | ||
2386 | while (recv_creds(sock, &cred, &v)) { | |
2387 | if (v == '1') | |
35174b0f | 2388 | return 0; |
237e200e | 2389 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2390 | return 1; |
237e200e | 2391 | } |
35174b0f | 2392 | return 0; |
237e200e SH |
2393 | } |
2394 | ||
35174b0f | 2395 | |
237e200e SH |
2396 | /* |
2397 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2398 | * in your old pidns. Only children which you clone will be in the target |
2399 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2400 | * actually convert pids. | |
2401 | * | |
2402 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2403 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2404 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2405 | * this issue. | |
237e200e SH |
2406 | */ |
2407 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2408 | { | |
2409 | int newnsfd = -1, ret, cpipe[2]; | |
2410 | char fnam[100]; | |
2411 | pid_t cpid; | |
2412 | char v; | |
2413 | ||
2414 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2415 | if (ret < 0 || ret >= sizeof(fnam)) | |
2416 | _exit(1); | |
2417 | newnsfd = open(fnam, O_RDONLY); | |
2418 | if (newnsfd < 0) | |
2419 | _exit(1); | |
2420 | if (setns(newnsfd, 0) < 0) | |
2421 | _exit(1); | |
2422 | close(newnsfd); | |
2423 | ||
2424 | if (pipe(cpipe) < 0) | |
2425 | _exit(1); | |
2426 | ||
35174b0f FG |
2427 | struct pid_ns_clone_args args = { |
2428 | .cpipe = cpipe, | |
2429 | .sock = sock, | |
2430 | .tpid = tpid, | |
2431 | .wrapped = &pid_to_ns | |
2432 | }; | |
2433 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2434 | void *stack = alloca(stack_size); | |
2435 | ||
2436 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2437 | if (cpid < 0) |
2438 | _exit(1); | |
2439 | ||
237e200e SH |
2440 | // give the child 1 second to be done forking and |
2441 | // write its ack | |
2442 | if (!wait_for_sock(cpipe[0], 1)) | |
2443 | _exit(1); | |
2444 | ret = read(cpipe[0], &v, 1); | |
2445 | if (ret != sizeof(char) || v != '1') | |
2446 | _exit(1); | |
2447 | ||
2448 | if (!wait_for_pid(cpid)) | |
2449 | _exit(1); | |
2450 | _exit(0); | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * To read cgroup files with a particular pid, we will setns into the child | |
2455 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2456 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2457 | */ | |
2458 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2459 | { | |
2460 | int sock[2] = {-1, -1}; | |
2461 | char *tmpdata = NULL; | |
2462 | int ret; | |
2463 | pid_t qpid, cpid = -1; | |
2464 | bool answer = false; | |
2465 | char v = '0'; | |
2466 | struct ucred cred; | |
2467 | size_t sz = 0, asz = 0; | |
2468 | ||
1ca6a467 | 2469 | if (!cgroup_ops->get(cgroup_ops, contrl, cg, file, &tmpdata)) |
237e200e SH |
2470 | return false; |
2471 | ||
2472 | /* | |
2473 | * Now we read the pids from returned data one by one, pass | |
2474 | * them into a child in the target namespace, read back the | |
2475 | * translated pids, and put them into our to-return data | |
2476 | */ | |
2477 | ||
2478 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2479 | perror("socketpair"); | |
2480 | free(tmpdata); | |
2481 | return false; | |
2482 | } | |
2483 | ||
2484 | cpid = fork(); | |
2485 | if (cpid == -1) | |
2486 | goto out; | |
2487 | ||
2488 | if (!cpid) // child - exits when done | |
2489 | pid_to_ns_wrapper(sock[1], tpid); | |
2490 | ||
2491 | char *ptr = tmpdata; | |
2492 | cred.uid = 0; | |
2493 | cred.gid = 0; | |
2494 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2495 | cred.pid = qpid; | |
2496 | ret = send_creds(sock[0], &cred, v, true); | |
2497 | ||
2498 | if (ret == SEND_CREDS_NOTSK) | |
2499 | goto next; | |
2500 | if (ret == SEND_CREDS_FAIL) | |
2501 | goto out; | |
2502 | ||
2503 | // read converted results | |
2504 | if (!wait_for_sock(sock[0], 2)) { | |
b8defc3d | 2505 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2506 | goto out; |
2507 | } | |
2508 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2509 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2510 | goto out; |
2511 | } | |
2512 | must_strcat_pid(d, &sz, &asz, qpid); | |
2513 | next: | |
2514 | ptr = strchr(ptr, '\n'); | |
2515 | if (!ptr) | |
2516 | break; | |
2517 | ptr++; | |
2518 | } | |
2519 | ||
2520 | cred.pid = getpid(); | |
2521 | v = '1'; | |
2522 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2523 | // failed to ask child to exit | |
b8defc3d | 2524 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); |
237e200e SH |
2525 | goto out; |
2526 | } | |
2527 | ||
2528 | answer = true; | |
2529 | ||
2530 | out: | |
2531 | free(tmpdata); | |
2532 | if (cpid != -1) | |
2533 | wait_for_pid(cpid); | |
2534 | if (sock[0] != -1) { | |
2535 | close(sock[0]); | |
2536 | close(sock[1]); | |
2537 | } | |
2538 | return answer; | |
2539 | } | |
2540 | ||
2541 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2542 | struct fuse_file_info *fi) | |
2543 | { | |
2544 | struct fuse_context *fc = fuse_get_context(); | |
2545 | struct file_info *f = (struct file_info *)fi->fh; | |
2546 | struct cgfs_files *k = NULL; | |
2547 | char *data = NULL; | |
2548 | int ret, s; | |
2549 | bool r; | |
2550 | ||
5fbea8a6 CB |
2551 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
2552 | return -EIO; | |
2553 | ||
237e200e | 2554 | if (f->type != LXC_TYPE_CGFILE) { |
b8defc3d | 2555 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); |
237e200e SH |
2556 | return -EIO; |
2557 | } | |
2558 | ||
2559 | if (offset) | |
2560 | return 0; | |
2561 | ||
237e200e SH |
2562 | if (!f->controller) |
2563 | return -EINVAL; | |
2564 | ||
2565 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2566 | return -EINVAL; | |
2567 | } | |
2568 | free_key(k); | |
2569 | ||
2570 | ||
888f8f3c | 2571 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2572 | ret = -EACCES; |
2573 | goto out; | |
2574 | } | |
2575 | ||
2576 | if (strcmp(f->file, "tasks") == 0 || | |
2577 | strcmp(f->file, "/tasks") == 0 || | |
2578 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2579 | strcmp(f->file, "cgroup.procs") == 0) | |
2580 | // special case - we have to translate the pids | |
2581 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2582 | else | |
1ca6a467 | 2583 | r = cgroup_ops->get(cgroup_ops, f->controller, f->cgroup, f->file, &data); |
237e200e SH |
2584 | |
2585 | if (!r) { | |
2586 | ret = -EINVAL; | |
2587 | goto out; | |
2588 | } | |
2589 | ||
2590 | if (!data) { | |
2591 | ret = 0; | |
2592 | goto out; | |
2593 | } | |
2594 | s = strlen(data); | |
2595 | if (s > size) | |
2596 | s = size; | |
2597 | memcpy(buf, data, s); | |
2598 | if (s > 0 && s < size && data[s-1] != '\n') | |
2599 | buf[s++] = '\n'; | |
2600 | ||
2601 | ret = s; | |
2602 | ||
2603 | out: | |
2604 | free(data); | |
2605 | return ret; | |
2606 | } | |
2607 | ||
35174b0f | 2608 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2609 | { |
2610 | pid_t vpid; | |
2611 | struct ucred cred; | |
2612 | char v; | |
2613 | int ret; | |
2614 | ||
2615 | cred.uid = 0; | |
2616 | cred.gid = 0; | |
2617 | while (1) { | |
2618 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2619 | lxcfs_error("%s\n", "Timeout reading from parent."); |
35174b0f | 2620 | return 1; |
237e200e SH |
2621 | } |
2622 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
b8defc3d | 2623 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); |
35174b0f | 2624 | return 1; |
237e200e SH |
2625 | } |
2626 | if (vpid == -1) // done | |
2627 | break; | |
2628 | v = '0'; | |
2629 | cred.pid = vpid; | |
2630 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2631 | v = '1'; | |
2632 | cred.pid = getpid(); | |
2633 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2634 | return 1; |
237e200e SH |
2635 | } |
2636 | } | |
35174b0f | 2637 | return 0; |
237e200e SH |
2638 | } |
2639 | ||
2640 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2641 | { | |
2642 | int newnsfd = -1, ret, cpipe[2]; | |
2643 | char fnam[100]; | |
2644 | pid_t cpid; | |
2645 | char v; | |
2646 | ||
2647 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2648 | if (ret < 0 || ret >= sizeof(fnam)) | |
2649 | _exit(1); | |
2650 | newnsfd = open(fnam, O_RDONLY); | |
2651 | if (newnsfd < 0) | |
2652 | _exit(1); | |
2653 | if (setns(newnsfd, 0) < 0) | |
2654 | _exit(1); | |
2655 | close(newnsfd); | |
2656 | ||
2657 | if (pipe(cpipe) < 0) | |
2658 | _exit(1); | |
2659 | ||
35174b0f FG |
2660 | struct pid_ns_clone_args args = { |
2661 | .cpipe = cpipe, | |
2662 | .sock = sock, | |
2663 | .tpid = tpid, | |
2664 | .wrapped = &pid_from_ns | |
2665 | }; | |
f0f8b851 SH |
2666 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2667 | void *stack = alloca(stack_size); | |
35174b0f FG |
2668 | |
2669 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2670 | if (cpid < 0) |
2671 | _exit(1); | |
2672 | ||
237e200e SH |
2673 | // give the child 1 second to be done forking and |
2674 | // write its ack | |
2675 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2676 | _exit(1); |
237e200e | 2677 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2678 | if (ret != sizeof(char) || v != '1') |
2679 | _exit(1); | |
237e200e SH |
2680 | |
2681 | if (!wait_for_pid(cpid)) | |
2682 | _exit(1); | |
2683 | _exit(0); | |
237e200e SH |
2684 | } |
2685 | ||
2686 | /* | |
2687 | * Given host @uid, return the uid to which it maps in | |
2688 | * @pid's user namespace, or -1 if none. | |
2689 | */ | |
2690 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2691 | { | |
2692 | FILE *f; | |
2693 | char line[400]; | |
2694 | ||
2695 | sprintf(line, "/proc/%d/uid_map", pid); | |
2696 | if ((f = fopen(line, "r")) == NULL) { | |
2697 | return false; | |
2698 | } | |
2699 | ||
2700 | *answer = convert_id_to_ns(f, uid); | |
2701 | fclose(f); | |
2702 | ||
2703 | if (*answer == -1) | |
2704 | return false; | |
2705 | return true; | |
2706 | } | |
2707 | ||
2708 | /* | |
2709 | * get_pid_creds: get the real uid and gid of @pid from | |
2710 | * /proc/$$/status | |
2711 | * (XXX should we use euid here?) | |
2712 | */ | |
2713 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2714 | { | |
2715 | char line[400]; | |
2716 | uid_t u; | |
2717 | gid_t g; | |
2718 | FILE *f; | |
2719 | ||
2720 | *uid = -1; | |
2721 | *gid = -1; | |
2722 | sprintf(line, "/proc/%d/status", pid); | |
2723 | if ((f = fopen(line, "r")) == NULL) { | |
b8defc3d | 2724 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); |
237e200e SH |
2725 | return; |
2726 | } | |
2727 | while (fgets(line, 400, f)) { | |
2728 | if (strncmp(line, "Uid:", 4) == 0) { | |
2729 | if (sscanf(line+4, "%u", &u) != 1) { | |
b8defc3d | 2730 | lxcfs_error("bad uid line for pid %u\n", pid); |
237e200e SH |
2731 | fclose(f); |
2732 | return; | |
2733 | } | |
2734 | *uid = u; | |
2735 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2736 | if (sscanf(line+4, "%u", &g) != 1) { | |
b8defc3d | 2737 | lxcfs_error("bad gid line for pid %u\n", pid); |
237e200e SH |
2738 | fclose(f); |
2739 | return; | |
2740 | } | |
2741 | *gid = g; | |
2742 | } | |
2743 | } | |
2744 | fclose(f); | |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * May the requestor @r move victim @v to a new cgroup? | |
2749 | * This is allowed if | |
2750 | * . they are the same task | |
2751 | * . they are ownedy by the same uid | |
2752 | * . @r is root on the host, or | |
2753 | * . @v's uid is mapped into @r's where @r is root. | |
2754 | */ | |
2755 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2756 | { | |
2757 | uid_t v_uid, tmpuid; | |
2758 | gid_t v_gid; | |
2759 | ||
2760 | if (r == v) | |
2761 | return true; | |
2762 | if (r_uid == 0) | |
2763 | return true; | |
2764 | get_pid_creds(v, &v_uid, &v_gid); | |
2765 | if (r_uid == v_uid) | |
2766 | return true; | |
2767 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2768 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2769 | return true; | |
2770 | return false; | |
2771 | } | |
2772 | ||
2773 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2774 | const char *file, const char *buf) | |
2775 | { | |
2776 | int sock[2] = {-1, -1}; | |
2777 | pid_t qpid, cpid = -1; | |
2778 | FILE *pids_file = NULL; | |
2779 | bool answer = false, fail = false; | |
2780 | ||
2781 | pids_file = open_pids_file(contrl, cg); | |
2782 | if (!pids_file) | |
2783 | return false; | |
2784 | ||
2785 | /* | |
2786 | * write the pids to a socket, have helper in writer's pidns | |
2787 | * call movepid for us | |
2788 | */ | |
2789 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2790 | perror("socketpair"); | |
2791 | goto out; | |
2792 | } | |
2793 | ||
2794 | cpid = fork(); | |
2795 | if (cpid == -1) | |
2796 | goto out; | |
2797 | ||
2798 | if (!cpid) { // child | |
2799 | fclose(pids_file); | |
2800 | pid_from_ns_wrapper(sock[1], tpid); | |
2801 | } | |
2802 | ||
2803 | const char *ptr = buf; | |
2804 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2805 | struct ucred cred; | |
2806 | char v; | |
2807 | ||
2808 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2809 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); |
237e200e SH |
2810 | goto out; |
2811 | } | |
2812 | ||
2813 | if (recv_creds(sock[0], &cred, &v)) { | |
2814 | if (v == '0') { | |
2815 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2816 | fail = true; | |
2817 | break; | |
2818 | } | |
2819 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2820 | fail = true; | |
2821 | } | |
2822 | } | |
2823 | ||
2824 | ptr = strchr(ptr, '\n'); | |
2825 | if (!ptr) | |
2826 | break; | |
2827 | ptr++; | |
2828 | } | |
2829 | ||
2830 | /* All good, write the value */ | |
2831 | qpid = -1; | |
2832 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
b8defc3d | 2833 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); |
237e200e SH |
2834 | |
2835 | if (!fail) | |
2836 | answer = true; | |
2837 | ||
2838 | out: | |
2839 | if (cpid != -1) | |
2840 | wait_for_pid(cpid); | |
2841 | if (sock[0] != -1) { | |
2842 | close(sock[0]); | |
2843 | close(sock[1]); | |
2844 | } | |
2845 | if (pids_file) { | |
2846 | if (fclose(pids_file) != 0) | |
2847 | answer = false; | |
2848 | } | |
2849 | return answer; | |
2850 | } | |
2851 | ||
2852 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2853 | struct fuse_file_info *fi) | |
2854 | { | |
2855 | struct fuse_context *fc = fuse_get_context(); | |
2856 | char *localbuf = NULL; | |
2857 | struct cgfs_files *k = NULL; | |
2858 | struct file_info *f = (struct file_info *)fi->fh; | |
2859 | bool r; | |
2860 | ||
5fbea8a6 CB |
2861 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
2862 | return -EIO; | |
2863 | ||
237e200e | 2864 | if (f->type != LXC_TYPE_CGFILE) { |
b8defc3d | 2865 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); |
237e200e SH |
2866 | return -EIO; |
2867 | } | |
2868 | ||
2869 | if (offset) | |
2870 | return 0; | |
2871 | ||
237e200e SH |
2872 | localbuf = alloca(size+1); |
2873 | localbuf[size] = '\0'; | |
2874 | memcpy(localbuf, buf, size); | |
2875 | ||
2876 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2877 | size = -EINVAL; | |
2878 | goto out; | |
2879 | } | |
2880 | ||
2881 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2882 | size = -EACCES; | |
2883 | goto out; | |
2884 | } | |
2885 | ||
2886 | if (strcmp(f->file, "tasks") == 0 || | |
2887 | strcmp(f->file, "/tasks") == 0 || | |
2888 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2889 | strcmp(f->file, "cgroup.procs") == 0) | |
2890 | // special case - we have to translate the pids | |
2891 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2892 | else | |
2893 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2894 | ||
2895 | if (!r) | |
2896 | size = -EINVAL; | |
2897 | ||
2898 | out: | |
2899 | free_key(k); | |
2900 | return size; | |
2901 | } | |
2902 | ||
2903 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2904 | { | |
2905 | struct fuse_context *fc = fuse_get_context(); | |
2906 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2907 | struct cgfs_files *k = NULL; | |
2908 | const char *cgroup; | |
2909 | int ret; | |
2910 | ||
5fbea8a6 | 2911 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
2912 | return -EIO; |
2913 | ||
2914 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2915 | return -EPERM; |
237e200e SH |
2916 | |
2917 | controller = pick_controller_from_path(fc, path); | |
2918 | if (!controller) | |
bc70ba9b CB |
2919 | return errno == ENOENT ? -EPERM : -errno; |
2920 | ||
237e200e SH |
2921 | cgroup = find_cgroup_in_path(path); |
2922 | if (!cgroup) | |
2923 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2924 | return -EPERM; |
237e200e SH |
2925 | |
2926 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2927 | ||
2928 | if (!last) { | |
2929 | path1 = "/"; | |
2930 | path2 = cgdir; | |
2931 | } else { | |
2932 | path1 = cgdir; | |
2933 | path2 = last; | |
2934 | } | |
2935 | ||
2936 | if (is_child_cgroup(controller, path1, path2)) { | |
2937 | // get uid, gid, from '/tasks' file and make up a mode | |
2938 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2939 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2940 | ||
2941 | } else | |
2942 | k = cgfs_get_key(controller, path1, path2); | |
2943 | ||
2944 | if (!k) { | |
2945 | ret = -EINVAL; | |
2946 | goto out; | |
2947 | } | |
2948 | ||
2949 | /* | |
2950 | * This being a fuse request, the uid and gid must be valid | |
2951 | * in the caller's namespace. So we can just check to make | |
2952 | * sure that the caller is root in his uid, and privileged | |
2953 | * over the file's current owner. | |
2954 | */ | |
2955 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2956 | ret = -EACCES; | |
2957 | goto out; | |
2958 | } | |
2959 | ||
2960 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2961 | ||
2962 | out: | |
2963 | free_key(k); | |
2964 | free(cgdir); | |
2965 | ||
2966 | return ret; | |
2967 | } | |
2968 | ||
2969 | int cg_chmod(const char *path, mode_t mode) | |
2970 | { | |
2971 | struct fuse_context *fc = fuse_get_context(); | |
2972 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2973 | struct cgfs_files *k = NULL; | |
2974 | const char *cgroup; | |
2975 | int ret; | |
2976 | ||
5fbea8a6 | 2977 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
2978 | return -EIO; |
2979 | ||
2980 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2981 | return -EPERM; |
237e200e SH |
2982 | |
2983 | controller = pick_controller_from_path(fc, path); | |
2984 | if (!controller) | |
bc70ba9b CB |
2985 | return errno == ENOENT ? -EPERM : -errno; |
2986 | ||
237e200e SH |
2987 | cgroup = find_cgroup_in_path(path); |
2988 | if (!cgroup) | |
2989 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2990 | return -EPERM; |
237e200e SH |
2991 | |
2992 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2993 | ||
2994 | if (!last) { | |
2995 | path1 = "/"; | |
2996 | path2 = cgdir; | |
2997 | } else { | |
2998 | path1 = cgdir; | |
2999 | path2 = last; | |
3000 | } | |
3001 | ||
3002 | if (is_child_cgroup(controller, path1, path2)) { | |
3003 | // get uid, gid, from '/tasks' file and make up a mode | |
3004 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
3005 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
3006 | ||
3007 | } else | |
3008 | k = cgfs_get_key(controller, path1, path2); | |
3009 | ||
3010 | if (!k) { | |
3011 | ret = -EINVAL; | |
3012 | goto out; | |
3013 | } | |
3014 | ||
3015 | /* | |
3016 | * This being a fuse request, the uid and gid must be valid | |
3017 | * in the caller's namespace. So we can just check to make | |
3018 | * sure that the caller is root in his uid, and privileged | |
3019 | * over the file's current owner. | |
3020 | */ | |
3021 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
3022 | ret = -EPERM; | |
3023 | goto out; | |
3024 | } | |
3025 | ||
3026 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
3027 | ret = -EINVAL; | |
3028 | goto out; | |
3029 | } | |
3030 | ||
3031 | ret = 0; | |
3032 | out: | |
3033 | free_key(k); | |
3034 | free(cgdir); | |
3035 | return ret; | |
3036 | } | |
3037 | ||
3038 | int cg_mkdir(const char *path, mode_t mode) | |
3039 | { | |
3040 | struct fuse_context *fc = fuse_get_context(); | |
3041 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
3042 | const char *cgroup; | |
3043 | int ret; | |
3044 | ||
5fbea8a6 | 3045 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
3046 | return -EIO; |
3047 | ||
237e200e SH |
3048 | controller = pick_controller_from_path(fc, path); |
3049 | if (!controller) | |
2f7036d0 | 3050 | return errno == ENOENT ? -EPERM : -errno; |
237e200e SH |
3051 | |
3052 | cgroup = find_cgroup_in_path(path); | |
3053 | if (!cgroup) | |
bc70ba9b | 3054 | return -errno; |
237e200e SH |
3055 | |
3056 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3057 | if (!last) | |
3058 | path1 = "/"; | |
3059 | else | |
3060 | path1 = cgdir; | |
3061 | ||
3062 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3063 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3064 | initpid = fc->pid; |
3065 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
3066 | if (!next) | |
3067 | ret = -EINVAL; | |
3068 | else if (last && strcmp(next, last) == 0) | |
3069 | ret = -EEXIST; | |
3070 | else | |
2f7036d0 | 3071 | ret = -EPERM; |
237e200e SH |
3072 | goto out; |
3073 | } | |
3074 | ||
3075 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
3076 | ret = -EACCES; | |
3077 | goto out; | |
3078 | } | |
3079 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
3080 | ret = -EACCES; | |
3081 | goto out; | |
3082 | } | |
3083 | ||
3084 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
3085 | ||
3086 | out: | |
3087 | free(cgdir); | |
3088 | free(next); | |
3089 | return ret; | |
3090 | } | |
3091 | ||
3092 | int cg_rmdir(const char *path) | |
3093 | { | |
3094 | struct fuse_context *fc = fuse_get_context(); | |
3095 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
3096 | const char *cgroup; | |
3097 | int ret; | |
3098 | ||
5fbea8a6 | 3099 | if (!fc || !cgroup_ops || pure_unified_layout(cgroup_ops)) |
237e200e SH |
3100 | return -EIO; |
3101 | ||
3102 | controller = pick_controller_from_path(fc, path); | |
e254948f CB |
3103 | if (!controller) /* Someone's trying to delete "/cgroup". */ |
3104 | return -EPERM; | |
237e200e SH |
3105 | |
3106 | cgroup = find_cgroup_in_path(path); | |
e254948f CB |
3107 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ |
3108 | return -EPERM; | |
237e200e SH |
3109 | |
3110 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3111 | if (!last) { | |
e254948f CB |
3112 | /* Someone's trying to delete a cgroup on the same level as the |
3113 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
3114 | * rmdir "/cgroup/blkio/init.slice". | |
3115 | */ | |
3116 | ret = -EPERM; | |
237e200e SH |
3117 | goto out; |
3118 | } | |
3119 | ||
3120 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3121 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3122 | initpid = fc->pid; |
3123 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
de77249b | 3124 | if (!last || (next && (strcmp(next, last) == 0))) |
237e200e SH |
3125 | ret = -EBUSY; |
3126 | else | |
3127 | ret = -ENOENT; | |
3128 | goto out; | |
3129 | } | |
3130 | ||
3131 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
3132 | ret = -EACCES; | |
3133 | goto out; | |
3134 | } | |
3135 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
3136 | ret = -EACCES; | |
3137 | goto out; | |
3138 | } | |
3139 | ||
3140 | if (!cgfs_remove(controller, cgroup)) { | |
3141 | ret = -EINVAL; | |
3142 | goto out; | |
3143 | } | |
3144 | ||
3145 | ret = 0; | |
3146 | ||
3147 | out: | |
3148 | free(cgdir); | |
3149 | free(next); | |
3150 | return ret; | |
3151 | } | |
3152 | ||
3153 | static bool startswith(const char *line, const char *pref) | |
3154 | { | |
3155 | if (strncmp(line, pref, strlen(pref)) == 0) | |
3156 | return true; | |
3157 | return false; | |
3158 | } | |
3159 | ||
66c5e848 CB |
3160 | /* Note that "memory.stat" in cgroup2 is hierarchical by default. */ |
3161 | static void parse_memstat(int version, | |
3162 | char *memstat, | |
3163 | unsigned long *cached, | |
3164 | unsigned long *active_anon, | |
3165 | unsigned long *inactive_anon, | |
3166 | unsigned long *active_file, | |
3167 | unsigned long *inactive_file, | |
3168 | unsigned long *unevictable, | |
3169 | unsigned long *shmem) | |
237e200e SH |
3170 | { |
3171 | char *eol; | |
3172 | ||
237e200e | 3173 | while (*memstat) { |
66c5e848 CB |
3174 | if (startswith(memstat, is_unified_controller(version) |
3175 | ? "cache" | |
3176 | : "total_cache")) { | |
4accebfb | 3177 | sscanf(memstat + 11, "%lu", cached); |
c6095b08 | 3178 | *cached /= 1024; |
66c5e848 CB |
3179 | } else if (startswith(memstat, is_unified_controller(version) |
3180 | ? "active_anon" | |
3181 | : "total_active_anon")) { | |
4accebfb | 3182 | sscanf(memstat + 17, "%lu", active_anon); |
c6095b08 | 3183 | *active_anon /= 1024; |
66c5e848 CB |
3184 | } else if (startswith(memstat, is_unified_controller(version) |
3185 | ? "inactive_anon" | |
3186 | : "total_inactive_anon")) { | |
4accebfb | 3187 | sscanf(memstat + 19, "%lu", inactive_anon); |
c6095b08 | 3188 | *inactive_anon /= 1024; |
66c5e848 CB |
3189 | } else if (startswith(memstat, is_unified_controller(version) |
3190 | ? "active_file" | |
3191 | : "total_active_file")) { | |
4accebfb | 3192 | sscanf(memstat + 17, "%lu", active_file); |
c6095b08 | 3193 | *active_file /= 1024; |
66c5e848 CB |
3194 | } else if (startswith(memstat, is_unified_controller(version) |
3195 | ? "inactive_file" | |
3196 | : "total_inactive_file")) { | |
4accebfb | 3197 | sscanf(memstat + 19, "%lu", inactive_file); |
c6095b08 | 3198 | *inactive_file /= 1024; |
66c5e848 CB |
3199 | } else if (startswith(memstat, is_unified_controller(version) |
3200 | ? "unevictable" | |
3201 | : "total_unevictable")) { | |
4accebfb | 3202 | sscanf(memstat + 17, "%lu", unevictable); |
c6095b08 | 3203 | *unevictable /= 1024; |
66c5e848 CB |
3204 | } else if (startswith(memstat, is_unified_controller(version) |
3205 | ? "shmem" | |
3206 | : "total_shmem")) { | |
559eaa8f JS |
3207 | sscanf(memstat + 11, "%lu", shmem); |
3208 | *shmem /= 1024; | |
237e200e SH |
3209 | } |
3210 | eol = strchr(memstat, '\n'); | |
3211 | if (!eol) | |
3212 | return; | |
3213 | memstat = eol+1; | |
3214 | } | |
3215 | } | |
3216 | ||
3217 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
3218 | { | |
3219 | char *eol; | |
3220 | char key[32]; | |
3221 | ||
3222 | memset(key, 0, 32); | |
3223 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
3224 | ||
3225 | size_t len = strlen(key); | |
3226 | *v = 0; | |
3227 | ||
3228 | while (*str) { | |
3229 | if (startswith(str, key)) { | |
3230 | sscanf(str + len, "%lu", v); | |
3231 | return; | |
3232 | } | |
3233 | eol = strchr(str, '\n'); | |
3234 | if (!eol) | |
3235 | return; | |
3236 | str = eol+1; | |
3237 | } | |
3238 | } | |
3239 | ||
5fbea8a6 | 3240 | int read_file_fuse(const char *path, char *buf, size_t size, struct file_info *d) |
237e200e | 3241 | { |
54a6d46a CB |
3242 | __do_free char *line = NULL; |
3243 | __do_fclose FILE *f = NULL; | |
3244 | size_t linelen = 0, total_len = 0; | |
237e200e SH |
3245 | char *cache = d->buf; |
3246 | size_t cache_size = d->buflen; | |
54a6d46a CB |
3247 | |
3248 | f = fopen(path, "r"); | |
237e200e SH |
3249 | if (!f) |
3250 | return 0; | |
3251 | ||
3252 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3253 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3254 | if (l < 0) { |
3255 | perror("Error writing to cache"); | |
54a6d46a | 3256 | return 0; |
237e200e SH |
3257 | } |
3258 | if (l >= cache_size) { | |
b8defc3d | 3259 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
54a6d46a | 3260 | return 0; |
237e200e SH |
3261 | } |
3262 | cache += l; | |
3263 | cache_size -= l; | |
3264 | total_len += l; | |
3265 | } | |
3266 | ||
3267 | d->size = total_len; | |
a262ddb7 CB |
3268 | if (total_len > size) |
3269 | total_len = size; | |
237e200e SH |
3270 | |
3271 | /* read from off 0 */ | |
3272 | memcpy(buf, d->buf, total_len); | |
54a6d46a CB |
3273 | |
3274 | if (d->size > total_len) | |
3275 | d->cached = d->size - total_len; | |
3276 | return total_len; | |
237e200e SH |
3277 | } |
3278 | ||
3279 | /* | |
3280 | * FUSE ops for /proc | |
3281 | */ | |
3282 | ||
66c5e848 | 3283 | static unsigned long get_memlimit(const char *cgroup, bool swap) |
237e200e | 3284 | { |
66c5e848 | 3285 | int ret; |
cfb72d52 | 3286 | __do_free char *memlimit_str = NULL; |
237e200e SH |
3287 | unsigned long memlimit = -1; |
3288 | ||
66c5e848 CB |
3289 | if (swap) |
3290 | ret = cgroup_ops->get_memory_swap_max(cgroup_ops, cgroup, &memlimit_str); | |
3291 | else | |
3292 | ret = cgroup_ops->get_memory_max(cgroup_ops, cgroup, &memlimit_str); | |
3293 | if (ret > 0) | |
237e200e SH |
3294 | memlimit = strtoul(memlimit_str, NULL, 10); |
3295 | ||
237e200e SH |
3296 | return memlimit; |
3297 | } | |
3298 | ||
66c5e848 | 3299 | static unsigned long get_min_memlimit(const char *cgroup, bool swap) |
237e200e | 3300 | { |
cfb72d52 CB |
3301 | __do_free char *copy = NULL; |
3302 | unsigned long memlimit = 0; | |
3303 | unsigned long retlimit; | |
237e200e | 3304 | |
66c5e848 CB |
3305 | copy = strdup(cgroup); |
3306 | retlimit = get_memlimit(copy, swap); | |
237e200e SH |
3307 | |
3308 | while (strcmp(copy, "/") != 0) { | |
66c5e848 CB |
3309 | char *it = copy; |
3310 | ||
3311 | it = dirname(it); | |
3312 | memlimit = get_memlimit(it, swap); | |
237e200e SH |
3313 | if (memlimit != -1 && memlimit < retlimit) |
3314 | retlimit = memlimit; | |
3315 | }; | |
3316 | ||
3317 | return retlimit; | |
3318 | } | |
3319 | ||
3320 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
cfb72d52 | 3321 | struct fuse_file_info *fi) |
237e200e | 3322 | { |
66c5e848 | 3323 | __do_free char *cgroup = NULL, *line = NULL, |
cfb72d52 CB |
3324 | *memusage_str = NULL, *memstat_str = NULL, |
3325 | *memswlimit_str = NULL, *memswusage_str = NULL; | |
3326 | __do_fclose FILE *f = NULL; | |
237e200e | 3327 | struct fuse_context *fc = fuse_get_context(); |
7e60aa1b | 3328 | struct lxcfs_opts *opts = (struct lxcfs_opts *) fuse_get_context()->private_data; |
237e200e | 3329 | struct file_info *d = (struct file_info *)fi->fh; |
cfb72d52 CB |
3330 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, |
3331 | memswusage = 0, cached = 0, hosttotal = 0, active_anon = 0, | |
3332 | inactive_anon = 0, active_file = 0, inactive_file = 0, | |
3333 | unevictable = 0, shmem = 0, hostswtotal = 0; | |
66c5e848 | 3334 | size_t linelen = 0, total_len = 0; |
237e200e SH |
3335 | char *cache = d->buf; |
3336 | size_t cache_size = d->buflen; | |
66c5e848 | 3337 | int ret; |
237e200e | 3338 | |
cfb72d52 CB |
3339 | if (offset) { |
3340 | int left; | |
3341 | ||
237e200e SH |
3342 | if (offset > d->size) |
3343 | return -EINVAL; | |
cfb72d52 | 3344 | |
237e200e SH |
3345 | if (!d->cached) |
3346 | return 0; | |
cfb72d52 CB |
3347 | |
3348 | left = d->size - offset; | |
3349 | total_len = left > size ? size : left; | |
237e200e | 3350 | memcpy(buf, cache + offset, total_len); |
cfb72d52 | 3351 | |
237e200e SH |
3352 | return total_len; |
3353 | } | |
3354 | ||
3355 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3356 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e | 3357 | initpid = fc->pid; |
cfb72d52 | 3358 | |
66c5e848 CB |
3359 | cgroup = get_pid_cgroup(initpid, "memory"); |
3360 | if (!cgroup) | |
5fbea8a6 | 3361 | return read_file_fuse("/proc/meminfo", buf, size, d); |
237e200e | 3362 | |
66c5e848 | 3363 | prune_init_slice(cgroup); |
cfb72d52 | 3364 | |
66c5e848 | 3365 | memlimit = get_min_memlimit(cgroup, false); |
cfb72d52 | 3366 | |
66c5e848 CB |
3367 | ret = cgroup_ops->get_memory_current(cgroup_ops, cgroup, &memusage_str); |
3368 | if (ret < 0) | |
cfb72d52 CB |
3369 | return 0; |
3370 | ||
66c5e848 CB |
3371 | ret = cgroup_ops->get_memory_stats(cgroup_ops, cgroup, &memstat_str); |
3372 | if (ret < 0) | |
cfb72d52 | 3373 | return 0; |
66c5e848 CB |
3374 | parse_memstat(ret, memstat_str, &cached, &active_anon, &inactive_anon, |
3375 | &active_file, &inactive_file, &unevictable, &shmem); | |
cfb72d52 CB |
3376 | |
3377 | /* | |
3378 | * Following values are allowed to fail, because swapaccount might be | |
3379 | * turned off for current kernel. | |
3380 | */ | |
66c5e848 CB |
3381 | ret = cgroup_ops->get_memory_swap_max(cgroup_ops, cgroup, &memswlimit_str); |
3382 | if (ret >= 0) | |
3383 | ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str); | |
3384 | if (ret >= 0) { | |
3385 | memswlimit = get_min_memlimit(cgroup, true); | |
cfb72d52 | 3386 | memswusage = strtoul(memswusage_str, NULL, 10); |
237e200e SH |
3387 | memswlimit = memswlimit / 1024; |
3388 | memswusage = memswusage / 1024; | |
3389 | } | |
3390 | ||
3391 | memusage = strtoul(memusage_str, NULL, 10); | |
3392 | memlimit /= 1024; | |
3393 | memusage /= 1024; | |
3394 | ||
237e200e SH |
3395 | f = fopen("/proc/meminfo", "r"); |
3396 | if (!f) | |
cfb72d52 | 3397 | return 0; |
237e200e SH |
3398 | |
3399 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3400 | ssize_t l; |
237e200e SH |
3401 | char *printme, lbuf[100]; |
3402 | ||
3403 | memset(lbuf, 0, 100); | |
3404 | if (startswith(line, "MemTotal:")) { | |
594a10e6 | 3405 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); |
237e200e SH |
3406 | if (hosttotal < memlimit) |
3407 | memlimit = hosttotal; | |
3408 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3409 | printme = lbuf; | |
3410 | } else if (startswith(line, "MemFree:")) { | |
3411 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3412 | printme = lbuf; | |
3413 | } else if (startswith(line, "MemAvailable:")) { | |
ad19b86d | 3414 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage + cached); |
237e200e | 3415 | printme = lbuf; |
cfb72d52 CB |
3416 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0 && |
3417 | opts && opts->swap_off == false) { | |
594a10e6 | 3418 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); |
4127e51b | 3419 | if (hostswtotal < memswlimit) |
3420 | memswlimit = hostswtotal; | |
3421 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit); | |
237e200e | 3422 | printme = lbuf; |
b3aa7f1a | 3423 | } else if (startswith(line, "SwapTotal:") && opts && opts->swap_off == true) { |
7e60aa1b | 3424 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", 0UL); |
3425 | printme = lbuf; | |
cfb72d52 CB |
3426 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && |
3427 | memswusage > 0 && opts && opts->swap_off == false) { | |
4127e51b | 3428 | unsigned long swaptotal = memswlimit, |
cfb72d52 CB |
3429 | swapusage = memusage > memswusage |
3430 | ? 0 | |
3431 | : memswusage - memusage, | |
3432 | swapfree = swapusage < swaptotal | |
3433 | ? swaptotal - swapusage | |
3434 | : 0; | |
b4665ce0 | 3435 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); |
237e200e | 3436 | printme = lbuf; |
b3aa7f1a | 3437 | } else if (startswith(line, "SwapFree:") && opts && opts->swap_off == true) { |
7e60aa1b | 3438 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", 0UL); |
3439 | printme = lbuf; | |
da35d72a SH |
3440 | } else if (startswith(line, "Slab:")) { |
3441 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3442 | printme = lbuf; | |
237e200e SH |
3443 | } else if (startswith(line, "Buffers:")) { |
3444 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3445 | printme = lbuf; | |
3446 | } else if (startswith(line, "Cached:")) { | |
3447 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3448 | printme = lbuf; | |
3449 | } else if (startswith(line, "SwapCached:")) { | |
3450 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3451 | printme = lbuf; | |
2f306ad3 | 3452 | } else if (startswith(line, "Active:")) { |
c6095b08 SH |
3453 | snprintf(lbuf, 100, "Active: %8lu kB\n", |
3454 | active_anon + active_file); | |
3455 | printme = lbuf; | |
2f306ad3 | 3456 | } else if (startswith(line, "Inactive:")) { |
c6095b08 SH |
3457 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", |
3458 | inactive_anon + inactive_file); | |
3459 | printme = lbuf; | |
3460 | } else if (startswith(line, "Active(anon)")) { | |
3461 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3462 | printme = lbuf; | |
3463 | } else if (startswith(line, "Inactive(anon)")) { | |
3464 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3465 | printme = lbuf; | |
3466 | } else if (startswith(line, "Active(file)")) { | |
3467 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3468 | printme = lbuf; | |
3469 | } else if (startswith(line, "Inactive(file)")) { | |
3470 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3471 | printme = lbuf; | |
3472 | } else if (startswith(line, "Unevictable")) { | |
3473 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3474 | printme = lbuf; | |
3475 | } else if (startswith(line, "SReclaimable")) { | |
3476 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3477 | printme = lbuf; | |
3478 | } else if (startswith(line, "SUnreclaim")) { | |
3479 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3480 | printme = lbuf; | |
559eaa8f JS |
3481 | } else if (startswith(line, "Shmem:")) { |
3482 | snprintf(lbuf, 100, "Shmem: %8lu kB\n", shmem); | |
3483 | printme = lbuf; | |
28cdea9b JS |
3484 | } else if (startswith(line, "ShmemHugePages")) { |
3485 | snprintf(lbuf, 100, "ShmemHugePages: %8lu kB\n", 0UL); | |
3486 | printme = lbuf; | |
3487 | } else if (startswith(line, "ShmemPmdMapped")) { | |
3488 | snprintf(lbuf, 100, "ShmemPmdMapped: %8lu kB\n", 0UL); | |
3489 | printme = lbuf; | |
237e200e SH |
3490 | } else |
3491 | printme = line; | |
3492 | ||
3493 | l = snprintf(cache, cache_size, "%s", printme); | |
3494 | if (l < 0) { | |
3495 | perror("Error writing to cache"); | |
cfb72d52 | 3496 | return 0; |
237e200e SH |
3497 | |
3498 | } | |
3499 | if (l >= cache_size) { | |
b8defc3d | 3500 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
cfb72d52 | 3501 | return 0; |
237e200e SH |
3502 | } |
3503 | ||
3504 | cache += l; | |
3505 | cache_size -= l; | |
3506 | total_len += l; | |
3507 | } | |
3508 | ||
3509 | d->cached = 1; | |
3510 | d->size = total_len; | |
3511 | if (total_len > size ) total_len = size; | |
3512 | memcpy(buf, d->buf, total_len); | |
3513 | ||
66c5e848 | 3514 | return total_len; |
237e200e SH |
3515 | } |
3516 | ||
3517 | /* | |
3518 | * Read the cpuset.cpus for cg | |
3519 | * Return the answer in a newly allocated string which must be freed | |
3520 | */ | |
71f17cd2 | 3521 | char *get_cpuset(const char *cg) |
237e200e | 3522 | { |
2c3bcd9e CB |
3523 | char *value = NULL; |
3524 | int ret; | |
237e200e | 3525 | |
2c3bcd9e CB |
3526 | ret = cgroup_ops->get_cpuset_cpus(cgroup_ops, cg, &value); |
3527 | if (ret < 0) | |
237e200e | 3528 | return NULL; |
2c3bcd9e CB |
3529 | |
3530 | return value; | |
237e200e SH |
3531 | } |
3532 | ||
3533 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3534 | ||
3535 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3536 | { | |
3537 | int cpu; | |
3538 | ||
3539 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3540 | return false; | |
3541 | return cpu_in_cpuset(cpu, cpuset); | |
3542 | } | |
3543 | ||
c59d6a55 JS |
3544 | /* |
3545 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or `cpu.cfs_period_us`, | |
3546 | * depending on `param`. Parameter value is returned throuh `value`. | |
3547 | */ | |
3548 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
3549 | { | |
54a6d46a CB |
3550 | __do_free char *str = NULL; |
3551 | char file[11 + 6 + 1]; /* cpu.cfs__us + quota/period + \0 */ | |
c59d6a55 | 3552 | |
54a6d46a | 3553 | snprintf(file, sizeof(file), "cpu.cfs_%s_us", param); |
c59d6a55 | 3554 | |
1ca6a467 | 3555 | if (!cgroup_ops->get(cgroup_ops, "cpu", cg, file, &str)) |
54a6d46a | 3556 | return false; |
c59d6a55 JS |
3557 | |
3558 | if (sscanf(str, "%ld", value) != 1) | |
54a6d46a | 3559 | return false; |
c59d6a55 | 3560 | |
54a6d46a | 3561 | return true; |
c59d6a55 JS |
3562 | } |
3563 | ||
3564 | /* | |
3565 | * Return the maximum number of visible CPUs based on CPU quotas. | |
3566 | * If there is no quota set, zero is returned. | |
3567 | */ | |
3568 | int max_cpu_count(const char *cg) | |
3569 | { | |
3570 | int rv, nprocs; | |
3571 | int64_t cfs_quota, cfs_period; | |
8c23db36 HY |
3572 | int nr_cpus_in_cpuset = 0; |
3573 | char *cpuset = NULL; | |
c59d6a55 JS |
3574 | |
3575 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) | |
3576 | return 0; | |
3577 | ||
3578 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
3579 | return 0; | |
3580 | ||
8c23db36 HY |
3581 | cpuset = get_cpuset(cg); |
3582 | if (cpuset) | |
3583 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
3584 | ||
3585 | if (cfs_quota <= 0 || cfs_period <= 0){ | |
3586 | if (nr_cpus_in_cpuset > 0) | |
3587 | return nr_cpus_in_cpuset; | |
3588 | ||
c59d6a55 | 3589 | return 0; |
8c23db36 | 3590 | } |
c59d6a55 JS |
3591 | |
3592 | rv = cfs_quota / cfs_period; | |
3593 | ||
3594 | /* In case quota/period does not yield a whole number, add one CPU for | |
3595 | * the remainder. | |
3596 | */ | |
3597 | if ((cfs_quota % cfs_period) > 0) | |
3598 | rv += 1; | |
3599 | ||
3600 | nprocs = get_nprocs(); | |
3601 | ||
3602 | if (rv > nprocs) | |
3603 | rv = nprocs; | |
3604 | ||
8c23db36 HY |
3605 | /* use min value in cpu quota and cpuset */ |
3606 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) | |
3607 | rv = nr_cpus_in_cpuset; | |
3608 | ||
c59d6a55 JS |
3609 | return rv; |
3610 | } | |
3611 | ||
db1b32f6 SX |
3612 | /* |
3613 | * Return the exact number of visible CPUs based on CPU quotas. | |
3614 | * If there is no quota set, zero is returned. | |
3615 | */ | |
3616 | static double exact_cpu_count(const char *cg) | |
3617 | { | |
3618 | double rv; | |
3619 | int nprocs; | |
3620 | int64_t cfs_quota, cfs_period; | |
3621 | ||
3622 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) | |
3623 | return 0; | |
3624 | ||
3625 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
3626 | return 0; | |
3627 | ||
3628 | if (cfs_quota <= 0 || cfs_period <= 0) | |
3629 | return 0; | |
3630 | ||
3631 | rv = (double)cfs_quota / (double)cfs_period; | |
3632 | ||
3633 | nprocs = get_nprocs(); | |
3634 | ||
3635 | if (rv > nprocs) | |
3636 | rv = nprocs; | |
3637 | ||
3638 | return rv; | |
3639 | } | |
3640 | ||
237e200e SH |
3641 | /* |
3642 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3643 | */ | |
3644 | static bool is_processor_line(const char *line) | |
3645 | { | |
3646 | int cpu; | |
3647 | ||
3648 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3649 | return true; | |
3650 | return false; | |
3651 | } | |
3652 | ||
3653 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
54a6d46a | 3654 | struct fuse_file_info *fi) |
237e200e | 3655 | { |
54a6d46a CB |
3656 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; |
3657 | __do_fclose FILE *f = NULL; | |
237e200e SH |
3658 | struct fuse_context *fc = fuse_get_context(); |
3659 | struct file_info *d = (struct file_info *)fi->fh; | |
54a6d46a | 3660 | size_t linelen = 0, total_len = 0; |
f676eb79 | 3661 | bool am_printing = false, firstline = true, is_s390x = false; |
c59d6a55 JS |
3662 | int curcpu = -1, cpu, max_cpus = 0; |
3663 | bool use_view; | |
237e200e SH |
3664 | char *cache = d->buf; |
3665 | size_t cache_size = d->buflen; | |
237e200e SH |
3666 | |
3667 | if (offset){ | |
54a6d46a CB |
3668 | int left; |
3669 | ||
237e200e SH |
3670 | if (offset > d->size) |
3671 | return -EINVAL; | |
54a6d46a | 3672 | |
237e200e SH |
3673 | if (!d->cached) |
3674 | return 0; | |
54a6d46a CB |
3675 | |
3676 | left = d->size - offset; | |
237e200e SH |
3677 | total_len = left > size ? size: left; |
3678 | memcpy(buf, cache + offset, total_len); | |
54a6d46a | 3679 | |
237e200e SH |
3680 | return total_len; |
3681 | } | |
3682 | ||
3683 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3684 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3685 | initpid = fc->pid; |
3686 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3687 | if (!cg) | |
5fbea8a6 | 3688 | return read_file_fuse("proc/cpuinfo", buf, size, d); |
6d2f6996 | 3689 | prune_init_slice(cg); |
237e200e SH |
3690 | |
3691 | cpuset = get_cpuset(cg); | |
3692 | if (!cpuset) | |
54a6d46a | 3693 | return 0; |
237e200e | 3694 | |
77f4399a | 3695 | use_view = cgroup_ops->can_use_cpuview(cgroup_ops); |
c59d6a55 JS |
3696 | if (use_view) |
3697 | max_cpus = max_cpu_count(cg); | |
3698 | ||
237e200e SH |
3699 | f = fopen("/proc/cpuinfo", "r"); |
3700 | if (!f) | |
54a6d46a | 3701 | return 0; |
237e200e SH |
3702 | |
3703 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3704 | ssize_t l; |
f676eb79 SH |
3705 | if (firstline) { |
3706 | firstline = false; | |
3707 | if (strstr(line, "IBM/S390") != NULL) { | |
3708 | is_s390x = true; | |
3709 | am_printing = true; | |
5ed9d4e2 | 3710 | continue; |
f676eb79 SH |
3711 | } |
3712 | } | |
5ed9d4e2 SH |
3713 | if (strncmp(line, "# processors:", 12) == 0) |
3714 | continue; | |
237e200e | 3715 | if (is_processor_line(line)) { |
c59d6a55 JS |
3716 | if (use_view && max_cpus > 0 && (curcpu+1) == max_cpus) |
3717 | break; | |
237e200e SH |
3718 | am_printing = cpuline_in_cpuset(line, cpuset); |
3719 | if (am_printing) { | |
3720 | curcpu ++; | |
3721 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3722 | if (l < 0) { | |
3723 | perror("Error writing to cache"); | |
54a6d46a | 3724 | return 0; |
237e200e SH |
3725 | } |
3726 | if (l >= cache_size) { | |
b8defc3d | 3727 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
54a6d46a | 3728 | return 0; |
237e200e SH |
3729 | } |
3730 | cache += l; | |
3731 | cache_size -= l; | |
3732 | total_len += l; | |
3733 | } | |
3734 | continue; | |
f676eb79 SH |
3735 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3736 | char *p; | |
c59d6a55 JS |
3737 | if (use_view && max_cpus > 0 && (curcpu+1) == max_cpus) |
3738 | break; | |
f676eb79 SH |
3739 | if (!cpu_in_cpuset(cpu, cpuset)) |
3740 | continue; | |
3741 | curcpu ++; | |
3742 | p = strchr(line, ':'); | |
3743 | if (!p || !*p) | |
54a6d46a | 3744 | return 0; |
f676eb79 | 3745 | p++; |
5ed9d4e2 | 3746 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3747 | if (l < 0) { |
3748 | perror("Error writing to cache"); | |
54a6d46a | 3749 | return 0; |
f676eb79 SH |
3750 | } |
3751 | if (l >= cache_size) { | |
b8defc3d | 3752 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
54a6d46a | 3753 | return 0; |
f676eb79 SH |
3754 | } |
3755 | cache += l; | |
3756 | cache_size -= l; | |
3757 | total_len += l; | |
3758 | continue; | |
3759 | ||
237e200e SH |
3760 | } |
3761 | if (am_printing) { | |
3762 | l = snprintf(cache, cache_size, "%s", line); | |
3763 | if (l < 0) { | |
3764 | perror("Error writing to cache"); | |
54a6d46a | 3765 | return 0; |
237e200e SH |
3766 | } |
3767 | if (l >= cache_size) { | |
b8defc3d | 3768 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
54a6d46a | 3769 | return 0; |
237e200e SH |
3770 | } |
3771 | cache += l; | |
3772 | cache_size -= l; | |
3773 | total_len += l; | |
3774 | } | |
3775 | } | |
3776 | ||
5ed9d4e2 | 3777 | if (is_s390x) { |
54a6d46a | 3778 | __do_free char *origcache = d->buf; |
a262ddb7 | 3779 | ssize_t l; |
54a6d46a CB |
3780 | |
3781 | d->buf = malloc(d->buflen); | |
3782 | if (!d->buf) { | |
3783 | d->buf = move_ptr(origcache); | |
3784 | return 0; | |
3785 | } | |
3786 | ||
5ed9d4e2 SH |
3787 | cache = d->buf; |
3788 | cache_size = d->buflen; | |
3789 | total_len = 0; | |
3790 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
54a6d46a CB |
3791 | if (l < 0 || l >= cache_size) |
3792 | return 0; | |
3793 | ||
5ed9d4e2 SH |
3794 | cache_size -= l; |
3795 | cache += l; | |
3796 | total_len += l; | |
3797 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
54a6d46a CB |
3798 | if (l < 0 || l >= cache_size) |
3799 | return 0; | |
3800 | ||
5ed9d4e2 SH |
3801 | cache_size -= l; |
3802 | cache += l; | |
3803 | total_len += l; | |
3804 | l = snprintf(cache, cache_size, "%s", origcache); | |
5ed9d4e2 | 3805 | if (l < 0 || l >= cache_size) |
54a6d46a | 3806 | return 0; |
5ed9d4e2 SH |
3807 | total_len += l; |
3808 | } | |
3809 | ||
237e200e SH |
3810 | d->cached = 1; |
3811 | d->size = total_len; | |
3812 | if (total_len > size ) total_len = size; | |
3813 | ||
3814 | /* read from off 0 */ | |
3815 | memcpy(buf, d->buf, total_len); | |
54a6d46a | 3816 | return total_len; |
237e200e SH |
3817 | } |
3818 | ||
0ecddf02 | 3819 | static uint64_t get_reaper_start_time(pid_t pid) |
9ac264cf | 3820 | { |
9ac264cf | 3821 | int ret; |
0ecddf02 CB |
3822 | FILE *f; |
3823 | uint64_t starttime; | |
3824 | /* strlen("/proc/") = 6 | |
3825 | * + | |
3826 | * LXCFS_NUMSTRLEN64 | |
3827 | * + | |
3828 | * strlen("/stat") = 5 | |
3829 | * + | |
3830 | * \0 = 1 | |
3831 | * */ | |
3832 | #define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1) | |
3833 | char path[__PROC_PID_STAT_LEN]; | |
9ac264cf JB |
3834 | pid_t qpid; |
3835 | ||
3836 | qpid = lookup_initpid_in_store(pid); | |
0ecddf02 CB |
3837 | if (qpid <= 0) { |
3838 | /* Caller can check for EINVAL on 0. */ | |
3839 | errno = EINVAL; | |
9ac264cf | 3840 | return 0; |
0ecddf02 | 3841 | } |
9ac264cf | 3842 | |
0ecddf02 CB |
3843 | ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid); |
3844 | if (ret < 0 || ret >= __PROC_PID_STAT_LEN) { | |
3845 | /* Caller can check for EINVAL on 0. */ | |
3846 | errno = EINVAL; | |
9ac264cf | 3847 | return 0; |
0ecddf02 | 3848 | } |
9ac264cf | 3849 | |
0ecddf02 CB |
3850 | f = fopen(path, "r"); |
3851 | if (!f) { | |
3852 | /* Caller can check for EINVAL on 0. */ | |
3853 | errno = EINVAL; | |
9ac264cf | 3854 | return 0; |
0ecddf02 | 3855 | } |
9ac264cf | 3856 | |
0ecddf02 CB |
3857 | /* Note that the *scanf() argument supression requires that length |
3858 | * modifiers such as "l" are omitted. Otherwise some compilers will yell | |
3859 | * at us. It's like telling someone you're not married and then asking | |
3860 | * if you can bring your wife to the party. | |
3861 | */ | |
3862 | ret = fscanf(f, "%*d " /* (1) pid %d */ | |
3863 | "%*s " /* (2) comm %s */ | |
3864 | "%*c " /* (3) state %c */ | |
3865 | "%*d " /* (4) ppid %d */ | |
3866 | "%*d " /* (5) pgrp %d */ | |
3867 | "%*d " /* (6) session %d */ | |
3868 | "%*d " /* (7) tty_nr %d */ | |
3869 | "%*d " /* (8) tpgid %d */ | |
3870 | "%*u " /* (9) flags %u */ | |
3871 | "%*u " /* (10) minflt %lu */ | |
3872 | "%*u " /* (11) cminflt %lu */ | |
3873 | "%*u " /* (12) majflt %lu */ | |
3874 | "%*u " /* (13) cmajflt %lu */ | |
3875 | "%*u " /* (14) utime %lu */ | |
3876 | "%*u " /* (15) stime %lu */ | |
3877 | "%*d " /* (16) cutime %ld */ | |
3878 | "%*d " /* (17) cstime %ld */ | |
3879 | "%*d " /* (18) priority %ld */ | |
3880 | "%*d " /* (19) nice %ld */ | |
3881 | "%*d " /* (20) num_threads %ld */ | |
3882 | "%*d " /* (21) itrealvalue %ld */ | |
3883 | "%" PRIu64, /* (22) starttime %llu */ | |
3884 | &starttime); | |
3885 | if (ret != 1) { | |
3886 | fclose(f); | |
3887 | /* Caller can check for EINVAL on 0. */ | |
3888 | errno = EINVAL; | |
3889 | return 0; | |
3890 | } | |
3891 | ||
3892 | fclose(f); | |
3893 | ||
3894 | errno = 0; | |
3895 | return starttime; | |
3896 | } | |
3897 | ||
1c4b4e38 | 3898 | static double get_reaper_start_time_in_sec(pid_t pid) |
0ecddf02 | 3899 | { |
1c4b4e38 CB |
3900 | uint64_t clockticks, ticks_per_sec; |
3901 | int64_t ret; | |
3902 | double res = 0; | |
0ecddf02 CB |
3903 | |
3904 | clockticks = get_reaper_start_time(pid); | |
3905 | if (clockticks == 0 && errno == EINVAL) { | |
3906 | lxcfs_debug("failed to retrieve start time of pid %d\n", pid); | |
3907 | return 0; | |
3908 | } | |
3909 | ||
1c4b4e38 CB |
3910 | ret = sysconf(_SC_CLK_TCK); |
3911 | if (ret < 0 && errno == EINVAL) { | |
0ecddf02 CB |
3912 | lxcfs_debug( |
3913 | "%s\n", | |
3914 | "failed to determine number of clock ticks in a second"); | |
3915 | return 0; | |
3916 | } | |
3917 | ||
1c4b4e38 CB |
3918 | ticks_per_sec = (uint64_t)ret; |
3919 | res = (double)clockticks / ticks_per_sec; | |
3920 | return res; | |
0ecddf02 CB |
3921 | } |
3922 | ||
1c4b4e38 | 3923 | static double get_reaper_age(pid_t pid) |
0ecddf02 | 3924 | { |
1c4b4e38 CB |
3925 | uint64_t uptime_ms; |
3926 | double procstart, procage; | |
0ecddf02 CB |
3927 | |
3928 | /* We need to substract the time the process has started since system | |
3929 | * boot minus the time when the system has started to get the actual | |
3930 | * reaper age. | |
3931 | */ | |
3932 | procstart = get_reaper_start_time_in_sec(pid); | |
3933 | procage = procstart; | |
3934 | if (procstart > 0) { | |
3935 | int ret; | |
3936 | struct timespec spec; | |
3937 | ||
3938 | ret = clock_gettime(CLOCK_BOOTTIME, &spec); | |
3939 | if (ret < 0) | |
3940 | return 0; | |
1c4b4e38 | 3941 | |
0ecddf02 CB |
3942 | /* We could make this more precise here by using the tv_nsec |
3943 | * field in the timespec struct and convert it to milliseconds | |
3944 | * and then create a double for the seconds and milliseconds but | |
3945 | * that seems more work than it is worth. | |
3946 | */ | |
1c4b4e38 CB |
3947 | uptime_ms = (spec.tv_sec * 1000) + (spec.tv_nsec * 1e-6); |
3948 | procage = (uptime_ms - (procstart * 1000)) / 1000; | |
0ecddf02 CB |
3949 | } |
3950 | ||
3951 | return procage; | |
3952 | } | |
3953 | ||
8be92dd1 JS |
3954 | /* |
3955 | * Returns 0 on success. | |
3956 | * It is the caller's responsibility to free `return_usage`, unless this | |
3957 | * function returns an error. | |
3958 | */ | |
79612c8b | 3959 | static int read_cpuacct_usage_all(char *cg, char *cpuset, struct cpuacct_usage **return_usage, int *size) |
8be92dd1 | 3960 | { |
54a6d46a CB |
3961 | __do_free char *usage_str = NULL; |
3962 | __do_free struct cpuacct_usage *cpu_usage = NULL; | |
77005a6c | 3963 | int cpucount = get_nprocs_conf(); |
54a6d46a CB |
3964 | int read_pos = 0, read_cnt=0; |
3965 | int i, j, ret; | |
8be92dd1 JS |
3966 | int cg_cpu; |
3967 | uint64_t cg_user, cg_system; | |
3968 | int64_t ticks_per_sec; | |
8be92dd1 JS |
3969 | |
3970 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3971 | ||
3972 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
db1b32f6 | 3973 | lxcfs_v( |
8be92dd1 JS |
3974 | "%s\n", |
3975 | "read_cpuacct_usage_all failed to determine number of clock ticks " | |
3976 | "in a second"); | |
3977 | return -1; | |
3978 | } | |
3979 | ||
3980 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); | |
3981 | if (!cpu_usage) | |
3982 | return -ENOMEM; | |
3983 | ||
db1b32f6 | 3984 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); |
1ca6a467 | 3985 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_all", &usage_str)) { |
54a6d46a CB |
3986 | char *data = NULL; |
3987 | int i = 0, read_pos = 0, read_cnt=0; | |
3988 | size_t sz = 0, asz = 0; | |
3989 | ||
3990 | /* read cpuacct.usage_percpu instead. */ | |
db1b32f6 | 3991 | lxcfs_v("failed to read cpuacct.usage_all. reading cpuacct.usage_percpu instead\n%s", ""); |
54a6d46a CB |
3992 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) |
3993 | return -1; | |
db1b32f6 SX |
3994 | lxcfs_v("usage_str: %s\n", usage_str); |
3995 | ||
54a6d46a | 3996 | /* convert cpuacct.usage_percpu into cpuacct.usage_all. */ |
db1b32f6 | 3997 | lxcfs_v("converting cpuacct.usage_percpu into cpuacct.usage_all\n%s", ""); |
1c4b4e38 | 3998 | |
db1b32f6 SX |
3999 | must_strcat(&data, &sz, &asz, "cpu user system\n"); |
4000 | ||
db1b32f6 SX |
4001 | while (sscanf(usage_str + read_pos, "%lu %n", &cg_user, &read_cnt) > 0) { |
4002 | lxcfs_debug("i: %d, cg_user: %lu, read_pos: %d, read_cnt: %d\n", i, cg_user, read_pos, read_cnt); | |
4003 | must_strcat(&data, &sz, &asz, "%d %lu 0\n", i, cg_user); | |
4004 | i++; | |
4005 | read_pos += read_cnt; | |
4006 | } | |
4007 | ||
db1b32f6 SX |
4008 | usage_str = data; |
4009 | ||
4010 | lxcfs_v("usage_str: %s\n", usage_str); | |
8be92dd1 JS |
4011 | } |
4012 | ||
4013 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) { | |
4014 | lxcfs_error("read_cpuacct_usage_all reading first line from " | |
4015 | "%s/cpuacct.usage_all failed.\n", cg); | |
54a6d46a | 4016 | return -1; |
8be92dd1 JS |
4017 | } |
4018 | ||
4019 | read_pos += read_cnt; | |
4020 | ||
4021 | for (i = 0, j = 0; i < cpucount; i++) { | |
4022 | ret = sscanf(usage_str + read_pos, "%d %lu %lu\n%n", &cg_cpu, &cg_user, | |
4023 | &cg_system, &read_cnt); | |
4024 | ||
4025 | if (ret == EOF) | |
4026 | break; | |
4027 | ||
4028 | if (ret != 3) { | |
4029 | lxcfs_error("read_cpuacct_usage_all reading from %s/cpuacct.usage_all " | |
4030 | "failed.\n", cg); | |
54a6d46a | 4031 | return -1; |
8be92dd1 JS |
4032 | } |
4033 | ||
4034 | read_pos += read_cnt; | |
4035 | ||
8be92dd1 JS |
4036 | /* Convert the time from nanoseconds to USER_HZ */ |
4037 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
4038 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
4039 | j++; | |
4040 | } | |
4041 | ||
54a6d46a | 4042 | *return_usage = move_ptr(cpu_usage); |
79612c8b | 4043 | *size = cpucount; |
54a6d46a | 4044 | return 0; |
8be92dd1 JS |
4045 | } |
4046 | ||
056adcef JS |
4047 | static unsigned long diff_cpu_usage(struct cpuacct_usage *older, struct cpuacct_usage *newer, struct cpuacct_usage *diff, int cpu_count) |
4048 | { | |
4049 | int i; | |
4050 | unsigned long sum = 0; | |
4051 | ||
4052 | for (i = 0; i < cpu_count; i++) { | |
77005a6c JS |
4053 | if (!newer[i].online) |
4054 | continue; | |
4055 | ||
056adcef JS |
4056 | /* When cpuset is changed on the fly, the CPUs might get reordered. |
4057 | * We could either reset all counters, or check that the substractions | |
4058 | * below will return expected results. | |
4059 | */ | |
4060 | if (newer[i].user > older[i].user) | |
4061 | diff[i].user = newer[i].user - older[i].user; | |
4062 | else | |
4063 | diff[i].user = 0; | |
4064 | ||
4065 | if (newer[i].system > older[i].system) | |
4066 | diff[i].system = newer[i].system - older[i].system; | |
4067 | else | |
4068 | diff[i].system = 0; | |
4069 | ||
4070 | if (newer[i].idle > older[i].idle) | |
4071 | diff[i].idle = newer[i].idle - older[i].idle; | |
4072 | else | |
4073 | diff[i].idle = 0; | |
4074 | ||
4075 | sum += diff[i].user; | |
4076 | sum += diff[i].system; | |
4077 | sum += diff[i].idle; | |
4078 | } | |
4079 | ||
4080 | return sum; | |
4081 | } | |
4082 | ||
4083 | static void add_cpu_usage(unsigned long *surplus, struct cpuacct_usage *usage, unsigned long *counter, unsigned long threshold) | |
4084 | { | |
4085 | unsigned long free_space, to_add; | |
4086 | ||
4087 | free_space = threshold - usage->user - usage->system; | |
4088 | ||
4089 | if (free_space > usage->idle) | |
4090 | free_space = usage->idle; | |
4091 | ||
4092 | to_add = free_space > *surplus ? *surplus : free_space; | |
4093 | ||
4094 | *counter += to_add; | |
4095 | usage->idle -= to_add; | |
4096 | *surplus -= to_add; | |
4097 | } | |
4098 | ||
951acc94 JS |
4099 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
4100 | { | |
4101 | struct cg_proc_stat *first = NULL, *prev, *tmp; | |
4102 | ||
4103 | for (prev = NULL; node; ) { | |
4104 | if (!cgfs_param_exist("cpu", node->cg, "cpu.shares")) { | |
4105 | tmp = node; | |
4106 | lxcfs_debug("Removing stat node for %s\n", node->cg); | |
4107 | ||
4108 | if (prev) | |
4109 | prev->next = node->next; | |
4110 | else | |
4111 | first = node->next; | |
4112 | ||
4113 | node = node->next; | |
4114 | free_proc_stat_node(tmp); | |
4115 | } else { | |
4116 | if (!first) | |
4117 | first = node; | |
4118 | prev = node; | |
4119 | node = node->next; | |
4120 | } | |
4121 | } | |
4122 | ||
4123 | return first; | |
4124 | } | |
4125 | ||
4126 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
4127 | static void prune_proc_stat_history(void) | |
4128 | { | |
4129 | int i; | |
4130 | time_t now = time(NULL); | |
4131 | ||
4132 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
2f49b662 JS |
4133 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
4134 | ||
4135 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
4136 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
951acc94 | 4137 | return; |
2f49b662 | 4138 | } |
951acc94 | 4139 | |
2f49b662 JS |
4140 | if (proc_stat_history[i]->next) { |
4141 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
4142 | proc_stat_history[i]->lastcheck = now; | |
4143 | } | |
951acc94 | 4144 | |
2f49b662 | 4145 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); |
951acc94 JS |
4146 | } |
4147 | } | |
4148 | ||
2f49b662 | 4149 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, const char *cg) |
056adcef | 4150 | { |
056adcef JS |
4151 | struct cg_proc_stat *node; |
4152 | ||
2f49b662 JS |
4153 | pthread_rwlock_rdlock(&head->lock); |
4154 | ||
4155 | if (!head->next) { | |
4156 | pthread_rwlock_unlock(&head->lock); | |
056adcef | 4157 | return NULL; |
2f49b662 | 4158 | } |
056adcef JS |
4159 | |
4160 | node = head->next; | |
4161 | ||
4162 | do { | |
4163 | if (strcmp(cg, node->cg) == 0) | |
951acc94 | 4164 | goto out; |
056adcef JS |
4165 | } while ((node = node->next)); |
4166 | ||
951acc94 JS |
4167 | node = NULL; |
4168 | ||
4169 | out: | |
2f49b662 | 4170 | pthread_rwlock_unlock(&head->lock); |
951acc94 JS |
4171 | prune_proc_stat_history(); |
4172 | return node; | |
056adcef JS |
4173 | } |
4174 | ||
4175 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) | |
4176 | { | |
4177 | struct cg_proc_stat *node; | |
4178 | int i; | |
4179 | ||
4180 | node = malloc(sizeof(struct cg_proc_stat)); | |
4181 | if (!node) | |
4182 | goto err; | |
4183 | ||
4184 | node->cg = NULL; | |
4185 | node->usage = NULL; | |
4186 | node->view = NULL; | |
4187 | ||
4188 | node->cg = malloc(strlen(cg) + 1); | |
4189 | if (!node->cg) | |
4190 | goto err; | |
4191 | ||
4192 | strcpy(node->cg, cg); | |
4193 | ||
4194 | node->usage = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4195 | if (!node->usage) | |
4196 | goto err; | |
4197 | ||
4198 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
4199 | ||
4200 | node->view = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4201 | if (!node->view) | |
4202 | goto err; | |
4203 | ||
4204 | node->cpu_count = cpu_count; | |
4205 | node->next = NULL; | |
4206 | ||
2f49b662 JS |
4207 | if (pthread_mutex_init(&node->lock, NULL) != 0) { |
4208 | lxcfs_error("%s\n", "Failed to initialize node lock"); | |
4209 | goto err; | |
4210 | } | |
4211 | ||
056adcef JS |
4212 | for (i = 0; i < cpu_count; i++) { |
4213 | node->view[i].user = 0; | |
4214 | node->view[i].system = 0; | |
4215 | node->view[i].idle = 0; | |
4216 | } | |
4217 | ||
4218 | return node; | |
4219 | ||
4220 | err: | |
4221 | if (node && node->cg) | |
4222 | free(node->cg); | |
4223 | if (node && node->usage) | |
4224 | free(node->usage); | |
4225 | if (node && node->view) | |
4226 | free(node->view); | |
4227 | if (node) | |
4228 | free(node); | |
4229 | ||
4230 | return NULL; | |
4231 | } | |
4232 | ||
2f49b662 | 4233 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
056adcef JS |
4234 | { |
4235 | int hash = calc_hash(new_node->cg) % CPUVIEW_HASH_SIZE; | |
4236 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
2f49b662 JS |
4237 | struct cg_proc_stat *node, *rv = new_node; |
4238 | ||
4239 | pthread_rwlock_wrlock(&head->lock); | |
056adcef JS |
4240 | |
4241 | if (!head->next) { | |
4242 | head->next = new_node; | |
2f49b662 | 4243 | goto out; |
056adcef JS |
4244 | } |
4245 | ||
2f49b662 JS |
4246 | node = head->next; |
4247 | ||
056adcef | 4248 | for (;;) { |
2f49b662 JS |
4249 | if (strcmp(node->cg, new_node->cg) == 0) { |
4250 | /* The node is already present, return it */ | |
4251 | free_proc_stat_node(new_node); | |
4252 | rv = node; | |
4253 | goto out; | |
4254 | } | |
056adcef JS |
4255 | |
4256 | if (node->next) { | |
4257 | node = node->next; | |
4258 | continue; | |
4259 | } | |
4260 | ||
4261 | node->next = new_node; | |
2f49b662 JS |
4262 | goto out; |
4263 | } | |
4264 | ||
4265 | out: | |
4266 | pthread_rwlock_unlock(&head->lock); | |
4267 | return rv; | |
4268 | } | |
4269 | ||
895f28e5 JS |
4270 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) |
4271 | { | |
54a6d46a | 4272 | __do_free struct cpuacct_usage *new_usage = NULL, *new_view = NULL; |
895f28e5 JS |
4273 | |
4274 | /* Allocate new memory */ | |
4275 | new_usage = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4276 | if (!new_usage) | |
4277 | return false; | |
4278 | ||
4279 | new_view = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
54a6d46a | 4280 | if (!new_view) |
895f28e5 | 4281 | return false; |
895f28e5 JS |
4282 | |
4283 | /* Copy existing data & initialize new elements */ | |
54a6d46a | 4284 | for (int i = 0; i < cpu_count; i++) { |
895f28e5 JS |
4285 | if (i < node->cpu_count) { |
4286 | new_usage[i].user = node->usage[i].user; | |
4287 | new_usage[i].system = node->usage[i].system; | |
4288 | new_usage[i].idle = node->usage[i].idle; | |
4289 | ||
4290 | new_view[i].user = node->view[i].user; | |
4291 | new_view[i].system = node->view[i].system; | |
4292 | new_view[i].idle = node->view[i].idle; | |
4293 | } else { | |
4294 | new_usage[i].user = 0; | |
4295 | new_usage[i].system = 0; | |
4296 | new_usage[i].idle = 0; | |
4297 | ||
4298 | new_view[i].user = 0; | |
4299 | new_view[i].system = 0; | |
4300 | new_view[i].idle = 0; | |
4301 | } | |
4302 | } | |
4303 | ||
4304 | free(node->usage); | |
54a6d46a | 4305 | node->usage = move_ptr(new_usage); |
895f28e5 | 4306 | |
54a6d46a CB |
4307 | free(node->view); |
4308 | node->view = move_ptr(new_view); | |
895f28e5 JS |
4309 | node->cpu_count = cpu_count; |
4310 | ||
4311 | return true; | |
4312 | } | |
4313 | ||
2f49b662 JS |
4314 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) |
4315 | { | |
4316 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
4317 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
4318 | struct cg_proc_stat *node; | |
4319 | ||
4320 | node = find_proc_stat_node(head, cg); | |
4321 | ||
4322 | if (!node) { | |
4323 | node = new_proc_stat_node(usage, cpu_count, cg); | |
4324 | if (!node) | |
4325 | return NULL; | |
4326 | ||
4327 | node = add_proc_stat_node(node); | |
4328 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
056adcef | 4329 | } |
2f49b662 JS |
4330 | |
4331 | pthread_mutex_lock(&node->lock); | |
895f28e5 JS |
4332 | |
4333 | /* If additional CPUs on the host have been enabled, CPU usage counter | |
4334 | * arrays have to be expanded */ | |
4335 | if (node->cpu_count < cpu_count) { | |
4336 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
4337 | node->cpu_count, cpu_count, cg); | |
4338 | ||
4339 | if (!expand_proc_stat_node(node, cpu_count)) { | |
4340 | pthread_mutex_unlock(&node->lock); | |
4341 | lxcfs_debug("Unable to expand stat node %d->%d for %s\n", | |
4342 | node->cpu_count, cpu_count, cg); | |
4343 | return NULL; | |
4344 | } | |
4345 | } | |
4346 | ||
2f49b662 | 4347 | return node; |
056adcef JS |
4348 | } |
4349 | ||
4350 | static void reset_proc_stat_node(struct cg_proc_stat *node, struct cpuacct_usage *usage, int cpu_count) | |
4351 | { | |
4352 | int i; | |
4353 | ||
4354 | lxcfs_debug("Resetting stat node for %s\n", node->cg); | |
4355 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
4356 | ||
4357 | for (i = 0; i < cpu_count; i++) { | |
4358 | node->view[i].user = 0; | |
4359 | node->view[i].system = 0; | |
4360 | node->view[i].idle = 0; | |
4361 | } | |
4362 | ||
4363 | node->cpu_count = cpu_count; | |
4364 | } | |
4365 | ||
54a6d46a CB |
4366 | static int cpuview_proc_stat(const char *cg, const char *cpuset, |
4367 | struct cpuacct_usage *cg_cpu_usage, | |
4368 | int cg_cpu_usage_size, FILE *f, char *buf, | |
4369 | size_t buf_size) | |
056adcef | 4370 | { |
54a6d46a CB |
4371 | __do_free char *line = NULL; |
4372 | __do_free struct cpuacct_usage *diff = NULL; | |
4373 | size_t linelen = 0, total_len = 0, l; | |
056adcef | 4374 | int curcpu = -1; /* cpu numbering starts at 0 */ |
77005a6c | 4375 | int physcpu, i; |
056adcef | 4376 | int max_cpus = max_cpu_count(cg), cpu_cnt = 0; |
54a6d46a CB |
4377 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, |
4378 | irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
056adcef JS |
4379 | unsigned long user_sum = 0, system_sum = 0, idle_sum = 0; |
4380 | unsigned long user_surplus = 0, system_surplus = 0; | |
4381 | unsigned long total_sum, threshold; | |
4382 | struct cg_proc_stat *stat_node; | |
77005a6c | 4383 | int nprocs = get_nprocs_conf(); |
056adcef | 4384 | |
79612c8b JS |
4385 | if (cg_cpu_usage_size < nprocs) |
4386 | nprocs = cg_cpu_usage_size; | |
4387 | ||
056adcef JS |
4388 | /* Read all CPU stats and stop when we've encountered other lines */ |
4389 | while (getline(&line, &linelen, f) != -1) { | |
77005a6c | 4390 | int ret; |
056adcef JS |
4391 | char cpu_char[10]; /* That's a lot of cores */ |
4392 | uint64_t all_used, cg_used; | |
4393 | ||
4394 | if (strlen(line) == 0) | |
4395 | continue; | |
54a6d46a CB |
4396 | |
4397 | /* not a ^cpuN line containing a number N */ | |
4398 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) | |
056adcef | 4399 | break; |
056adcef | 4400 | |
77005a6c | 4401 | if (sscanf(cpu_char, "%d", &physcpu) != 1) |
056adcef | 4402 | continue; |
77005a6c | 4403 | |
79612c8b JS |
4404 | if (physcpu >= cg_cpu_usage_size) |
4405 | continue; | |
4406 | ||
056adcef JS |
4407 | curcpu ++; |
4408 | cpu_cnt ++; | |
4409 | ||
77005a6c | 4410 | if (!cpu_in_cpuset(physcpu, cpuset)) { |
54a6d46a | 4411 | for (i = curcpu; i <= physcpu; i++) |
77005a6c | 4412 | cg_cpu_usage[i].online = false; |
77005a6c JS |
4413 | continue; |
4414 | } | |
4415 | ||
4416 | if (curcpu < physcpu) { | |
4417 | /* Some CPUs may be disabled */ | |
4418 | for (i = curcpu; i < physcpu; i++) | |
4419 | cg_cpu_usage[i].online = false; | |
4420 | ||
4421 | curcpu = physcpu; | |
4422 | } | |
4423 | ||
4424 | cg_cpu_usage[curcpu].online = true; | |
4425 | ||
056adcef JS |
4426 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
4427 | &user, | |
4428 | &nice, | |
4429 | &system, | |
4430 | &idle, | |
4431 | &iowait, | |
4432 | &irq, | |
4433 | &softirq, | |
4434 | &steal, | |
4435 | &guest, | |
4436 | &guest_nice); | |
4437 | ||
4438 | if (ret != 10) | |
4439 | continue; | |
4440 | ||
4441 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
4442 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
4443 | ||
4444 | if (all_used >= cg_used) { | |
4445 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
4446 | ||
4447 | } else { | |
4448 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
4449 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
4450 | curcpu, cg, all_used, cg_used); | |
4451 | cg_cpu_usage[curcpu].idle = idle; | |
4452 | } | |
4453 | } | |
4454 | ||
4455 | /* Cannot use more CPUs than is available due to cpuset */ | |
4456 | if (max_cpus > cpu_cnt) | |
4457 | max_cpus = cpu_cnt; | |
4458 | ||
2f49b662 | 4459 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
056adcef JS |
4460 | |
4461 | if (!stat_node) { | |
2f49b662 | 4462 | lxcfs_error("unable to find/create stat node for %s\n", cg); |
54a6d46a | 4463 | return 0; |
056adcef JS |
4464 | } |
4465 | ||
4466 | diff = malloc(sizeof(struct cpuacct_usage) * nprocs); | |
4467 | if (!diff) { | |
54a6d46a | 4468 | return 0; |
056adcef JS |
4469 | } |
4470 | ||
4471 | /* | |
4472 | * If the new values are LOWER than values stored in memory, it means | |
4473 | * the cgroup has been reset/recreated and we should reset too. | |
4474 | */ | |
77005a6c JS |
4475 | for (curcpu = 0; curcpu < nprocs; curcpu++) { |
4476 | if (!cg_cpu_usage[curcpu].online) | |
4477 | continue; | |
4478 | ||
4479 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
4480 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
4481 | ||
4482 | break; | |
4483 | } | |
056adcef | 4484 | |
77005a6c JS |
4485 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); |
4486 | ||
4487 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
4488 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
4489 | ||
4490 | if (!stat_node->usage[curcpu].online) | |
4491 | continue; | |
4492 | ||
4493 | i++; | |
056adcef | 4494 | |
056adcef JS |
4495 | stat_node->usage[curcpu].user += diff[curcpu].user; |
4496 | stat_node->usage[curcpu].system += diff[curcpu].system; | |
4497 | stat_node->usage[curcpu].idle += diff[curcpu].idle; | |
4498 | ||
77005a6c | 4499 | if (max_cpus > 0 && i >= max_cpus) { |
056adcef JS |
4500 | user_surplus += diff[curcpu].user; |
4501 | system_surplus += diff[curcpu].system; | |
4502 | } | |
4503 | } | |
4504 | ||
4505 | /* Calculate usage counters of visible CPUs */ | |
4506 | if (max_cpus > 0) { | |
54a6d46a CB |
4507 | unsigned long diff_user = 0; |
4508 | unsigned long diff_system = 0; | |
4509 | unsigned long diff_idle = 0; | |
4510 | unsigned long max_diff_idle = 0; | |
4511 | unsigned long max_diff_idle_index = 0; | |
4512 | double exact_cpus; | |
4513 | ||
056adcef JS |
4514 | /* threshold = maximum usage per cpu, including idle */ |
4515 | threshold = total_sum / cpu_cnt * max_cpus; | |
4516 | ||
77005a6c | 4517 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
77005a6c JS |
4518 | if (!stat_node->usage[curcpu].online) |
4519 | continue; | |
4520 | ||
4521 | i++; | |
4522 | ||
db1b32f6 SX |
4523 | if (i == max_cpus) |
4524 | break; | |
4525 | ||
056adcef JS |
4526 | if (diff[curcpu].user + diff[curcpu].system >= threshold) |
4527 | continue; | |
4528 | ||
4529 | /* Add user */ | |
54a6d46a CB |
4530 | add_cpu_usage(&user_surplus, &diff[curcpu], |
4531 | &diff[curcpu].user, threshold); | |
056adcef JS |
4532 | |
4533 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
4534 | continue; | |
4535 | ||
4536 | /* If there is still room, add system */ | |
54a6d46a CB |
4537 | add_cpu_usage(&system_surplus, &diff[curcpu], |
4538 | &diff[curcpu].system, threshold); | |
056adcef JS |
4539 | } |
4540 | ||
4541 | if (user_surplus > 0) | |
4542 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
4543 | if (system_surplus > 0) | |
4544 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
4545 | ||
77005a6c | 4546 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
77005a6c JS |
4547 | if (!stat_node->usage[curcpu].online) |
4548 | continue; | |
4549 | ||
4550 | i++; | |
4551 | ||
db1b32f6 SX |
4552 | if (i == max_cpus) |
4553 | break; | |
4554 | ||
056adcef JS |
4555 | stat_node->view[curcpu].user += diff[curcpu].user; |
4556 | stat_node->view[curcpu].system += diff[curcpu].system; | |
4557 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
4558 | ||
4559 | user_sum += stat_node->view[curcpu].user; | |
4560 | system_sum += stat_node->view[curcpu].system; | |
4561 | idle_sum += stat_node->view[curcpu].idle; | |
056adcef | 4562 | |
db1b32f6 SX |
4563 | diff_user += diff[curcpu].user; |
4564 | diff_system += diff[curcpu].system; | |
4565 | diff_idle += diff[curcpu].idle; | |
4566 | if (diff[curcpu].idle > max_diff_idle) { | |
4567 | max_diff_idle = diff[curcpu].idle; | |
4568 | max_diff_idle_index = curcpu; | |
4569 | } | |
4570 | ||
4571 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
4572 | } | |
4573 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
4574 | ||
54a6d46a CB |
4575 | /* revise cpu usage view to support partial cpu case. */ |
4576 | exact_cpus = exact_cpu_count(cg); | |
db1b32f6 | 4577 | if (exact_cpus < (double)max_cpus){ |
db1b32f6 | 4578 | unsigned long delta = (unsigned long)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); |
54a6d46a CB |
4579 | |
4580 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
db1b32f6 SX |
4581 | lxcfs_v("delta: %lu\n", delta); |
4582 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
4583 | idle_sum = idle_sum > delta ? idle_sum - delta : 0; | |
4584 | lxcfs_v("idle_sum after: %lu\n", idle_sum); | |
1c4b4e38 | 4585 | |
db1b32f6 SX |
4586 | curcpu = max_diff_idle_index; |
4587 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
4588 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle > delta ? stat_node->view[curcpu].idle - delta : 0; | |
4589 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
4590 | } | |
056adcef | 4591 | } else { |
77005a6c JS |
4592 | for (curcpu = 0; curcpu < nprocs; curcpu++) { |
4593 | if (!stat_node->usage[curcpu].online) | |
4594 | continue; | |
4595 | ||
056adcef JS |
4596 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; |
4597 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
4598 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
4599 | ||
4600 | user_sum += stat_node->view[curcpu].user; | |
4601 | system_sum += stat_node->view[curcpu].system; | |
4602 | idle_sum += stat_node->view[curcpu].idle; | |
4603 | } | |
4604 | } | |
4605 | ||
4606 | /* Render the file */ | |
4607 | /* cpu-all */ | |
4608 | l = snprintf(buf, buf_size, "cpu %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
4609 | user_sum, | |
4610 | system_sum, | |
4611 | idle_sum); | |
db1b32f6 | 4612 | lxcfs_v("cpu-all: %s\n", buf); |
056adcef JS |
4613 | |
4614 | if (l < 0) { | |
4615 | perror("Error writing to cache"); | |
54a6d46a | 4616 | return 0; |
056adcef JS |
4617 | } |
4618 | if (l >= buf_size) { | |
4619 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4620 | return 0; |
056adcef JS |
4621 | } |
4622 | ||
4623 | buf += l; | |
4624 | buf_size -= l; | |
4625 | total_len += l; | |
4626 | ||
4627 | /* Render visible CPUs */ | |
77005a6c JS |
4628 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
4629 | if (!stat_node->usage[curcpu].online) | |
4630 | continue; | |
4631 | ||
4632 | i++; | |
4633 | ||
4634 | if (max_cpus > 0 && i == max_cpus) | |
056adcef JS |
4635 | break; |
4636 | ||
4637 | l = snprintf(buf, buf_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
77005a6c | 4638 | i, |
056adcef JS |
4639 | stat_node->view[curcpu].user, |
4640 | stat_node->view[curcpu].system, | |
4641 | stat_node->view[curcpu].idle); | |
db1b32f6 | 4642 | lxcfs_v("cpu: %s\n", buf); |
056adcef JS |
4643 | |
4644 | if (l < 0) { | |
4645 | perror("Error writing to cache"); | |
54a6d46a | 4646 | return 0; |
056adcef JS |
4647 | |
4648 | } | |
4649 | if (l >= buf_size) { | |
4650 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4651 | return 0; |
056adcef JS |
4652 | } |
4653 | ||
4654 | buf += l; | |
4655 | buf_size -= l; | |
4656 | total_len += l; | |
4657 | } | |
4658 | ||
4659 | /* Pass the rest of /proc/stat, start with the last line read */ | |
4660 | l = snprintf(buf, buf_size, "%s", line); | |
4661 | ||
4662 | if (l < 0) { | |
4663 | perror("Error writing to cache"); | |
54a6d46a | 4664 | return 0; |
056adcef JS |
4665 | |
4666 | } | |
4667 | if (l >= buf_size) { | |
4668 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4669 | return 0; |
056adcef JS |
4670 | } |
4671 | ||
4672 | buf += l; | |
4673 | buf_size -= l; | |
4674 | total_len += l; | |
4675 | ||
4676 | /* Pass the rest of the host's /proc/stat */ | |
4677 | while (getline(&line, &linelen, f) != -1) { | |
4678 | l = snprintf(buf, buf_size, "%s", line); | |
4679 | if (l < 0) { | |
4680 | perror("Error writing to cache"); | |
54a6d46a | 4681 | return 0; |
056adcef JS |
4682 | } |
4683 | if (l >= buf_size) { | |
4684 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4685 | return 0; |
056adcef JS |
4686 | } |
4687 | buf += l; | |
4688 | buf_size -= l; | |
4689 | total_len += l; | |
4690 | } | |
4691 | ||
2f49b662 JS |
4692 | if (stat_node) |
4693 | pthread_mutex_unlock(&stat_node->lock); | |
54a6d46a | 4694 | return total_len; |
056adcef JS |
4695 | } |
4696 | ||
f34de69a | 4697 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) |
237e200e | 4698 | static int proc_stat_read(char *buf, size_t size, off_t offset, |
54a6d46a | 4699 | struct fuse_file_info *fi) |
237e200e | 4700 | { |
54a6d46a CB |
4701 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; |
4702 | __do_free struct cpuacct_usage *cg_cpu_usage = NULL; | |
4703 | __do_fclose FILE *f = NULL; | |
237e200e SH |
4704 | struct fuse_context *fc = fuse_get_context(); |
4705 | struct file_info *d = (struct file_info *)fi->fh; | |
54a6d46a | 4706 | size_t linelen = 0, total_len = 0; |
237e200e | 4707 | int curcpu = -1; /* cpu numbering starts at 0 */ |
77005a6c | 4708 | int physcpu = 0; |
54a6d46a CB |
4709 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, |
4710 | irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
4711 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, | |
4712 | iowait_sum = 0, irq_sum = 0, softirq_sum = 0, | |
4713 | steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; | |
237e200e SH |
4714 | char cpuall[CPUALL_MAX_SIZE]; |
4715 | /* reserve for cpu all */ | |
4716 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
4717 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
79612c8b | 4718 | int cg_cpu_usage_size = 0; |
237e200e SH |
4719 | |
4720 | if (offset){ | |
4721 | if (offset > d->size) | |
4722 | return -EINVAL; | |
4723 | if (!d->cached) | |
4724 | return 0; | |
4725 | int left = d->size - offset; | |
4726 | total_len = left > size ? size: left; | |
4727 | memcpy(buf, d->buf + offset, total_len); | |
4728 | return total_len; | |
4729 | } | |
4730 | ||
4731 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
db1b32f6 | 4732 | lxcfs_v("initpid: %d\n", initpid); |
237e200e SH |
4733 | if (initpid <= 0) |
4734 | initpid = fc->pid; | |
a4bd57ae HY |
4735 | |
4736 | /* | |
4737 | * when container run with host pid namespace initpid == 1, cgroup will "/" | |
4738 | * we should return host os's /proc contents. | |
4739 | * in some case cpuacct_usage.all in "/" will larger then /proc/stat | |
4740 | */ | |
4741 | if (initpid == 1) { | |
5fbea8a6 | 4742 | return read_file_fuse("/proc/stat", buf, size, d); |
a4bd57ae HY |
4743 | } |
4744 | ||
237e200e | 4745 | cg = get_pid_cgroup(initpid, "cpuset"); |
db1b32f6 | 4746 | lxcfs_v("cg: %s\n", cg); |
237e200e | 4747 | if (!cg) |
5fbea8a6 | 4748 | return read_file_fuse("/proc/stat", buf, size, d); |
6d2f6996 | 4749 | prune_init_slice(cg); |
237e200e SH |
4750 | |
4751 | cpuset = get_cpuset(cg); | |
4752 | if (!cpuset) | |
54a6d46a | 4753 | return 0; |
237e200e | 4754 | |
8be92dd1 JS |
4755 | /* |
4756 | * Read cpuacct.usage_all for all CPUs. | |
4757 | * If the cpuacct cgroup is present, it is used to calculate the container's | |
4758 | * CPU usage. If not, values from the host's /proc/stat are used. | |
4759 | */ | |
79612c8b | 4760 | if (read_cpuacct_usage_all(cg, cpuset, &cg_cpu_usage, &cg_cpu_usage_size) != 0) { |
db1b32f6 | 4761 | lxcfs_v("%s\n", "proc_stat_read failed to read from cpuacct, " |
8be92dd1 JS |
4762 | "falling back to the host's /proc/stat"); |
4763 | } | |
4764 | ||
237e200e SH |
4765 | f = fopen("/proc/stat", "r"); |
4766 | if (!f) | |
54a6d46a | 4767 | return 0; |
237e200e SH |
4768 | |
4769 | //skip first line | |
4770 | if (getline(&line, &linelen, f) < 0) { | |
b8defc3d | 4771 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); |
54a6d46a | 4772 | return 0; |
237e200e SH |
4773 | } |
4774 | ||
77f4399a | 4775 | if (cgroup_ops->can_use_cpuview(cgroup_ops) && cg_cpu_usage) { |
79612c8b JS |
4776 | total_len = cpuview_proc_stat(cg, cpuset, cg_cpu_usage, cg_cpu_usage_size, |
4777 | f, d->buf, d->buflen); | |
056adcef JS |
4778 | goto out; |
4779 | } | |
4780 | ||
237e200e | 4781 | while (getline(&line, &linelen, f) != -1) { |
a262ddb7 | 4782 | ssize_t l; |
237e200e SH |
4783 | char cpu_char[10]; /* That's a lot of cores */ |
4784 | char *c; | |
8be92dd1 JS |
4785 | uint64_t all_used, cg_used, new_idle; |
4786 | int ret; | |
237e200e | 4787 | |
b4665ce0 SH |
4788 | if (strlen(line) == 0) |
4789 | continue; | |
237e200e SH |
4790 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
4791 | /* not a ^cpuN line containing a number N, just print it */ | |
9502bae2 | 4792 | l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
4793 | if (l < 0) { |
4794 | perror("Error writing to cache"); | |
54a6d46a | 4795 | return 0; |
237e200e SH |
4796 | } |
4797 | if (l >= cache_size) { | |
b8defc3d | 4798 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
54a6d46a | 4799 | return 0; |
237e200e SH |
4800 | } |
4801 | cache += l; | |
4802 | cache_size -= l; | |
4803 | total_len += l; | |
4804 | continue; | |
4805 | } | |
4806 | ||
77005a6c | 4807 | if (sscanf(cpu_char, "%d", &physcpu) != 1) |
237e200e | 4808 | continue; |
77005a6c | 4809 | if (!cpu_in_cpuset(physcpu, cpuset)) |
237e200e SH |
4810 | continue; |
4811 | curcpu ++; | |
4812 | ||
8be92dd1 | 4813 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
7144f069 CB |
4814 | &user, |
4815 | &nice, | |
4816 | &system, | |
4817 | &idle, | |
4818 | &iowait, | |
4819 | &irq, | |
4820 | &softirq, | |
4821 | &steal, | |
4822 | &guest, | |
8be92dd1 JS |
4823 | &guest_nice); |
4824 | ||
4825 | if (ret != 10 || !cg_cpu_usage) { | |
4826 | c = strchr(line, ' '); | |
4827 | if (!c) | |
4828 | continue; | |
4829 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
4830 | if (l < 0) { | |
4831 | perror("Error writing to cache"); | |
54a6d46a | 4832 | return 0; |
8be92dd1 JS |
4833 | |
4834 | } | |
4835 | if (l >= cache_size) { | |
4836 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4837 | return 0; |
8be92dd1 JS |
4838 | } |
4839 | ||
4840 | cache += l; | |
4841 | cache_size -= l; | |
4842 | total_len += l; | |
4843 | ||
4844 | if (ret != 10) | |
4845 | continue; | |
4846 | } | |
4847 | ||
4848 | if (cg_cpu_usage) { | |
79612c8b JS |
4849 | if (physcpu >= cg_cpu_usage_size) |
4850 | break; | |
4851 | ||
8be92dd1 | 4852 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; |
77005a6c | 4853 | cg_used = cg_cpu_usage[physcpu].user + cg_cpu_usage[physcpu].system; |
8be92dd1 JS |
4854 | |
4855 | if (all_used >= cg_used) { | |
4856 | new_idle = idle + (all_used - cg_used); | |
4857 | ||
4858 | } else { | |
4859 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
4860 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
4861 | curcpu, cg, all_used, cg_used); | |
4862 | new_idle = idle; | |
4863 | } | |
4864 | ||
4865 | l = snprintf(cache, cache_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
77005a6c | 4866 | curcpu, cg_cpu_usage[physcpu].user, cg_cpu_usage[physcpu].system, |
8be92dd1 JS |
4867 | new_idle); |
4868 | ||
4869 | if (l < 0) { | |
4870 | perror("Error writing to cache"); | |
54a6d46a | 4871 | return 0; |
8be92dd1 JS |
4872 | |
4873 | } | |
4874 | if (l >= cache_size) { | |
4875 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
54a6d46a | 4876 | return 0; |
8be92dd1 JS |
4877 | } |
4878 | ||
4879 | cache += l; | |
4880 | cache_size -= l; | |
4881 | total_len += l; | |
4882 | ||
77005a6c JS |
4883 | user_sum += cg_cpu_usage[physcpu].user; |
4884 | system_sum += cg_cpu_usage[physcpu].system; | |
8be92dd1 JS |
4885 | idle_sum += new_idle; |
4886 | ||
4887 | } else { | |
4888 | user_sum += user; | |
4889 | nice_sum += nice; | |
4890 | system_sum += system; | |
4891 | idle_sum += idle; | |
4892 | iowait_sum += iowait; | |
4893 | irq_sum += irq; | |
4894 | softirq_sum += softirq; | |
4895 | steal_sum += steal; | |
4896 | guest_sum += guest; | |
4897 | guest_nice_sum += guest_nice; | |
4898 | } | |
237e200e SH |
4899 | } |
4900 | ||
4901 | cache = d->buf; | |
4902 | ||
7144f069 CB |
4903 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
4904 | user_sum, | |
4905 | nice_sum, | |
4906 | system_sum, | |
4907 | idle_sum, | |
4908 | iowait_sum, | |
4909 | irq_sum, | |
4910 | softirq_sum, | |
4911 | steal_sum, | |
4912 | guest_sum, | |
4913 | guest_nice_sum); | |
4914 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
237e200e SH |
4915 | memcpy(cache, cpuall, cpuall_len); |
4916 | cache += cpuall_len; | |
7144f069 | 4917 | } else { |
237e200e | 4918 | /* shouldn't happen */ |
b8defc3d | 4919 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); |
237e200e SH |
4920 | cpuall_len = 0; |
4921 | } | |
4922 | ||
4923 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
4924 | total_len += cpuall_len; | |
056adcef JS |
4925 | |
4926 | out: | |
237e200e SH |
4927 | d->cached = 1; |
4928 | d->size = total_len; | |
7144f069 CB |
4929 | if (total_len > size) |
4930 | total_len = size; | |
237e200e SH |
4931 | |
4932 | memcpy(buf, d->buf, total_len); | |
54a6d46a | 4933 | return total_len; |
237e200e SH |
4934 | } |
4935 | ||
0ecddf02 CB |
4936 | /* This function retrieves the busy time of a group of tasks by looking at |
4937 | * cpuacct.usage. Unfortunately, this only makes sense when the container has | |
4938 | * been given it's own cpuacct cgroup. If not, this function will take the busy | |
4939 | * time of all other taks that do not actually belong to the container into | |
4940 | * account as well. If someone has a clever solution for this please send a | |
4941 | * patch! | |
4942 | */ | |
1c4b4e38 | 4943 | static double get_reaper_busy(pid_t task) |
237e200e | 4944 | { |
54a6d46a | 4945 | __do_free char *cgroup = NULL, *usage_str = NULL; |
237e200e | 4946 | unsigned long usage = 0; |
54a6d46a | 4947 | pid_t initpid; |
237e200e | 4948 | |
54a6d46a | 4949 | initpid = lookup_initpid_in_store(task); |
237e200e SH |
4950 | if (initpid <= 0) |
4951 | return 0; | |
4952 | ||
4953 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
4954 | if (!cgroup) | |
54a6d46a | 4955 | return 0; |
6d2f6996 | 4956 | prune_init_slice(cgroup); |
54a6d46a CB |
4957 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cgroup, "cpuacct.usage", |
4958 | &usage_str)) | |
4959 | return 0; | |
237e200e | 4960 | |
54a6d46a CB |
4961 | usage = strtoul(usage_str, NULL, 10); |
4962 | return ((double)usage / 1000000000); | |
237e200e SH |
4963 | } |
4964 | ||
4965 | #if RELOADTEST | |
4966 | void iwashere(void) | |
4967 | { | |
237e200e SH |
4968 | int fd; |
4969 | ||
ec2b5e7c | 4970 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
4971 | if (fd >= 0) |
4972 | close(fd); | |
4973 | } | |
4974 | #endif | |
4975 | ||
4976 | /* | |
4977 | * We read /proc/uptime and reuse its second field. | |
4978 | * For the first field, we use the mtime for the reaper for | |
4979 | * the calling pid as returned by getreaperage | |
4980 | */ | |
4981 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
4982 | struct fuse_file_info *fi) | |
4983 | { | |
4984 | struct fuse_context *fc = fuse_get_context(); | |
4985 | struct file_info *d = (struct file_info *)fi->fh; | |
1c4b4e38 | 4986 | double busytime = get_reaper_busy(fc->pid); |
237e200e | 4987 | char *cache = d->buf; |
a262ddb7 | 4988 | ssize_t total_len = 0; |
1c4b4e38 | 4989 | double idletime, reaperage; |
237e200e SH |
4990 | |
4991 | #if RELOADTEST | |
4992 | iwashere(); | |
4993 | #endif | |
4994 | ||
4995 | if (offset){ | |
237e200e SH |
4996 | if (!d->cached) |
4997 | return 0; | |
bbdf646b BM |
4998 | if (offset > d->size) |
4999 | return -EINVAL; | |
237e200e SH |
5000 | int left = d->size - offset; |
5001 | total_len = left > size ? size: left; | |
5002 | memcpy(buf, cache + offset, total_len); | |
5003 | return total_len; | |
5004 | } | |
5005 | ||
0ecddf02 CB |
5006 | reaperage = get_reaper_age(fc->pid); |
5007 | /* To understand why this is done, please read the comment to the | |
5008 | * get_reaper_busy() function. | |
5009 | */ | |
5010 | idletime = reaperage; | |
5011 | if (reaperage >= busytime) | |
5012 | idletime = reaperage - busytime; | |
237e200e | 5013 | |
1c4b4e38 | 5014 | total_len = snprintf(d->buf, d->buflen, "%.2lf %.2lf\n", reaperage, idletime); |
bbdf646b | 5015 | if (total_len < 0 || total_len >= d->buflen){ |
0ecddf02 | 5016 | lxcfs_error("%s\n", "failed to write to cache"); |
237e200e SH |
5017 | return 0; |
5018 | } | |
5019 | ||
5020 | d->size = (int)total_len; | |
5021 | d->cached = 1; | |
5022 | ||
5023 | if (total_len > size) total_len = size; | |
5024 | ||
5025 | memcpy(buf, d->buf, total_len); | |
5026 | return total_len; | |
5027 | } | |
5028 | ||
5029 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
9a9484ab | 5030 | struct fuse_file_info *fi) |
237e200e | 5031 | { |
9a9484ab CB |
5032 | __do_free char *cg = NULL, *io_serviced_str = NULL, |
5033 | *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
5034 | *io_wait_time_str = NULL, *io_service_time_str = NULL, | |
5035 | *line = NULL; | |
5036 | __do_fclose FILE *f = NULL; | |
237e200e SH |
5037 | struct fuse_context *fc = fuse_get_context(); |
5038 | struct file_info *d = (struct file_info *)fi->fh; | |
237e200e SH |
5039 | unsigned long read = 0, write = 0; |
5040 | unsigned long read_merged = 0, write_merged = 0; | |
5041 | unsigned long read_sectors = 0, write_sectors = 0; | |
5042 | unsigned long read_ticks = 0, write_ticks = 0; | |
5043 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
5044 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
5045 | char *cache = d->buf; | |
5046 | size_t cache_size = d->buflen; | |
9a9484ab | 5047 | size_t linelen = 0, total_len = 0; |
237e200e SH |
5048 | unsigned int major = 0, minor = 0; |
5049 | int i = 0; | |
9a9484ab CB |
5050 | int ret; |
5051 | char dev_name[72]; | |
237e200e SH |
5052 | |
5053 | if (offset){ | |
9a9484ab CB |
5054 | int left; |
5055 | ||
237e200e SH |
5056 | if (offset > d->size) |
5057 | return -EINVAL; | |
9a9484ab | 5058 | |
237e200e SH |
5059 | if (!d->cached) |
5060 | return 0; | |
9a9484ab CB |
5061 | |
5062 | left = d->size - offset; | |
237e200e SH |
5063 | total_len = left > size ? size: left; |
5064 | memcpy(buf, cache + offset, total_len); | |
9a9484ab | 5065 | |
237e200e SH |
5066 | return total_len; |
5067 | } | |
5068 | ||
5069 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5070 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
5071 | initpid = fc->pid; |
5072 | cg = get_pid_cgroup(initpid, "blkio"); | |
5073 | if (!cg) | |
5fbea8a6 | 5074 | return read_file_fuse("/proc/diskstats", buf, size, d); |
6d2f6996 | 5075 | prune_init_slice(cg); |
237e200e | 5076 | |
9a9484ab CB |
5077 | ret = cgroup_ops->get_io_serviced(cgroup_ops, cg, &io_serviced_str); |
5078 | if (ret < 0) { | |
5079 | if (ret == -EOPNOTSUPP) | |
5080 | return read_file_fuse("/proc/diskstats", buf, size, d); | |
5081 | } | |
237e200e | 5082 | |
9a9484ab CB |
5083 | ret = cgroup_ops->get_io_merged(cgroup_ops, cg, &io_merged_str); |
5084 | if (ret < 0) { | |
5085 | if (ret == -EOPNOTSUPP) | |
5086 | return read_file_fuse("/proc/diskstats", buf, size, d); | |
5087 | } | |
5088 | ||
5089 | ret = cgroup_ops->get_io_service_bytes(cgroup_ops, cg, &io_service_bytes_str); | |
5090 | if (ret < 0) { | |
5091 | if (ret == -EOPNOTSUPP) | |
5092 | return read_file_fuse("/proc/diskstats", buf, size, d); | |
5093 | } | |
5094 | ||
5095 | ret = cgroup_ops->get_io_wait_time(cgroup_ops, cg, &io_wait_time_str); | |
5096 | if (ret < 0) { | |
5097 | if (ret == -EOPNOTSUPP) | |
5098 | return read_file_fuse("/proc/diskstats", buf, size, d); | |
5099 | } | |
5100 | ||
5101 | ret = cgroup_ops->get_io_service_time(cgroup_ops, cg, &io_service_time_str); | |
5102 | if (ret < 0) { | |
5103 | if (ret == -EOPNOTSUPP) | |
5104 | return read_file_fuse("/proc/diskstats", buf, size, d); | |
5105 | } | |
237e200e SH |
5106 | |
5107 | f = fopen("/proc/diskstats", "r"); | |
5108 | if (!f) | |
9a9484ab | 5109 | return 0; |
237e200e SH |
5110 | |
5111 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 5112 | ssize_t l; |
2209fe50 | 5113 | char lbuf[256]; |
237e200e SH |
5114 | |
5115 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 5116 | if (i != 3) |
237e200e | 5117 | continue; |
2209fe50 SH |
5118 | |
5119 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
5120 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
5121 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
5122 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
5123 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
5124 | read_sectors = read_sectors/512; | |
5125 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
5126 | write_sectors = write_sectors/512; | |
5127 | ||
5128 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
5129 | rd_svctm = rd_svctm/1000000; | |
5130 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
5131 | rd_wait = rd_wait/1000000; | |
5132 | read_ticks = rd_svctm + rd_wait; | |
5133 | ||
5134 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
5135 | wr_svctm = wr_svctm/1000000; | |
5136 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
5137 | wr_wait = wr_wait/1000000; | |
5138 | write_ticks = wr_svctm + wr_wait; | |
5139 | ||
5140 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
5141 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
5142 | |
5143 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
5144 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
5145 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
5146 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
5147 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
5148 | else | |
5149 | continue; | |
237e200e | 5150 | |
2209fe50 | 5151 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
5152 | if (l < 0) { |
5153 | perror("Error writing to fuse buf"); | |
9a9484ab | 5154 | return 0; |
237e200e SH |
5155 | } |
5156 | if (l >= cache_size) { | |
b8defc3d | 5157 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
9a9484ab | 5158 | return 0; |
237e200e SH |
5159 | } |
5160 | cache += l; | |
5161 | cache_size -= l; | |
5162 | total_len += l; | |
5163 | } | |
5164 | ||
5165 | d->cached = 1; | |
5166 | d->size = total_len; | |
5167 | if (total_len > size ) total_len = size; | |
5168 | memcpy(buf, d->buf, total_len); | |
5169 | ||
9a9484ab | 5170 | return total_len; |
237e200e SH |
5171 | } |
5172 | ||
70dcc12e | 5173 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
66c5e848 | 5174 | struct fuse_file_info *fi) |
70dcc12e | 5175 | { |
66c5e848 CB |
5176 | __do_free char *cg = NULL, *memswlimit_str = NULL, *memusage_str = NULL, |
5177 | *memswusage_str = NULL; | |
70dcc12e SH |
5178 | struct fuse_context *fc = fuse_get_context(); |
5179 | struct file_info *d = (struct file_info *)fi->fh; | |
66c5e848 CB |
5180 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, |
5181 | memswusage = 0, swap_total = 0, swap_free = 0; | |
5182 | ssize_t total_len = 0; | |
a262ddb7 | 5183 | ssize_t l = 0; |
70dcc12e | 5184 | char *cache = d->buf; |
66c5e848 | 5185 | int ret; |
70dcc12e SH |
5186 | |
5187 | if (offset) { | |
66c5e848 CB |
5188 | int left; |
5189 | ||
70dcc12e SH |
5190 | if (offset > d->size) |
5191 | return -EINVAL; | |
66c5e848 | 5192 | |
70dcc12e SH |
5193 | if (!d->cached) |
5194 | return 0; | |
66c5e848 CB |
5195 | |
5196 | left = d->size - offset; | |
70dcc12e SH |
5197 | total_len = left > size ? size: left; |
5198 | memcpy(buf, cache + offset, total_len); | |
66c5e848 | 5199 | |
70dcc12e SH |
5200 | return total_len; |
5201 | } | |
5202 | ||
5203 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5204 | if (initpid <= 1 || is_shared_pidns(initpid)) |
70dcc12e SH |
5205 | initpid = fc->pid; |
5206 | cg = get_pid_cgroup(initpid, "memory"); | |
5207 | if (!cg) | |
5fbea8a6 | 5208 | return read_file_fuse("/proc/swaps", buf, size, d); |
6d2f6996 | 5209 | prune_init_slice(cg); |
70dcc12e | 5210 | |
66c5e848 | 5211 | memlimit = get_min_memlimit(cg, false); |
70dcc12e | 5212 | |
66c5e848 CB |
5213 | ret = cgroup_ops->get_memory_current(cgroup_ops, cg, &memusage_str); |
5214 | if (ret < 0) | |
5215 | return 0; | |
70dcc12e | 5216 | |
70dcc12e SH |
5217 | memusage = strtoul(memusage_str, NULL, 10); |
5218 | ||
66c5e848 CB |
5219 | ret = cgroup_ops->get_memory_swap_max(cgroup_ops, cg, &memswlimit_str); |
5220 | if (ret >= 0) | |
5221 | ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cg, &memswusage_str); | |
5222 | if (ret >= 0) { | |
5223 | memswlimit = get_min_memlimit(cg, true); | |
70dcc12e | 5224 | memswusage = strtoul(memswusage_str, NULL, 10); |
70dcc12e SH |
5225 | swap_total = (memswlimit - memlimit) / 1024; |
5226 | swap_free = (memswusage - memusage) / 1024; | |
5227 | } | |
5228 | ||
5229 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
5230 | ||
5231 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
5232 | if (!memswlimit) { | |
66c5e848 CB |
5233 | __do_free char *line = NULL; |
5234 | __do_fclose FILE *f = NULL; | |
70dcc12e | 5235 | size_t linelen = 0; |
70dcc12e | 5236 | |
66c5e848 | 5237 | f = fopen("/proc/meminfo", "r"); |
70dcc12e | 5238 | if (!f) |
66c5e848 | 5239 | return 0; |
70dcc12e SH |
5240 | |
5241 | while (getline(&line, &linelen, f) != -1) { | |
66c5e848 | 5242 | if (startswith(line, "SwapTotal:")) |
70dcc12e | 5243 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); |
66c5e848 | 5244 | else if (startswith(line, "SwapFree:")) |
70dcc12e | 5245 | sscanf(line, "SwapFree: %8lu kB", &swap_free); |
70dcc12e | 5246 | } |
70dcc12e SH |
5247 | } |
5248 | ||
5249 | if (swap_total > 0) { | |
a262ddb7 CB |
5250 | l = snprintf(d->buf + total_len, d->size - total_len, |
5251 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
5252 | swap_total, swap_free); | |
5253 | total_len += l; | |
70dcc12e SH |
5254 | } |
5255 | ||
a262ddb7 | 5256 | if (total_len < 0 || l < 0) { |
70dcc12e | 5257 | perror("Error writing to cache"); |
66c5e848 | 5258 | return 0; |
70dcc12e SH |
5259 | } |
5260 | ||
5261 | d->cached = 1; | |
5262 | d->size = (int)total_len; | |
5263 | ||
5264 | if (total_len > size) total_len = size; | |
5265 | memcpy(buf, d->buf, total_len); | |
66c5e848 | 5266 | return total_len; |
70dcc12e | 5267 | } |
66c5e848 | 5268 | |
6db4f7a3 | 5269 | /* |
5270 | * Find the process pid from cgroup path. | |
5271 | * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid. | |
5272 | * @pid_buf : put pid to pid_buf. | |
5273 | * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ... | |
5274 | * @depth : the depth of cgroup in container. | |
5275 | * @sum : return the number of pid. | |
5276 | * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu | |
5277 | */ | |
5278 | static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd) | |
5279 | { | |
54a6d46a CB |
5280 | __do_free char *path = NULL; |
5281 | __do_close_prot_errno int fd = -EBADF; | |
5282 | __do_fclose FILE *f = NULL; | |
5283 | __do_closedir DIR *dir = NULL; | |
6db4f7a3 | 5284 | struct dirent *file; |
6db4f7a3 | 5285 | size_t linelen = 0; |
5286 | char *line = NULL; | |
5287 | int pd; | |
6db4f7a3 | 5288 | char **pid; |
5289 | ||
5290 | /* path = dpath + "/cgroup.procs" + /0 */ | |
54a6d46a CB |
5291 | path = malloc(strlen(dpath) + 20); |
5292 | if (!path) | |
5293 | return sum; | |
6db4f7a3 | 5294 | |
5295 | strcpy(path, dpath); | |
54a6d46a | 5296 | fd = openat(cfd, path, O_RDONLY | O_CLOEXEC | O_NOFOLLOW); |
6db4f7a3 | 5297 | if (fd < 0) |
54a6d46a | 5298 | return sum; |
6db4f7a3 | 5299 | |
54a6d46a CB |
5300 | dir = fdopendir(move_fd(fd)); |
5301 | if (!dir) | |
5302 | return sum; | |
6db4f7a3 | 5303 | |
5304 | while (((file = readdir(dir)) != NULL) && depth > 0) { | |
54a6d46a | 5305 | if (strcmp(file->d_name, ".") == 0) |
6db4f7a3 | 5306 | continue; |
54a6d46a CB |
5307 | |
5308 | if (strcmp(file->d_name, "..") == 0) | |
6db4f7a3 | 5309 | continue; |
54a6d46a | 5310 | |
6db4f7a3 | 5311 | if (file->d_type == DT_DIR) { |
54a6d46a CB |
5312 | __do_free char *path_dir = NULL; |
5313 | ||
6db4f7a3 | 5314 | /* path + '/' + d_name +/0 */ |
54a6d46a CB |
5315 | path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name)); |
5316 | if (!path_dir) | |
5317 | return sum; | |
5318 | ||
6db4f7a3 | 5319 | strcpy(path_dir, path); |
5320 | strcat(path_dir, "/"); | |
5321 | strcat(path_dir, file->d_name); | |
5322 | pd = depth - 1; | |
5323 | sum = calc_pid(pid_buf, path_dir, pd, sum, cfd); | |
6db4f7a3 | 5324 | } |
5325 | } | |
6db4f7a3 | 5326 | |
5327 | strcat(path, "/cgroup.procs"); | |
5328 | fd = openat(cfd, path, O_RDONLY); | |
5329 | if (fd < 0) | |
54a6d46a | 5330 | return sum; |
6db4f7a3 | 5331 | |
54a6d46a CB |
5332 | f = fdopen(move_fd(fd), "r"); |
5333 | if (!f) | |
5334 | return sum; | |
6db4f7a3 | 5335 | |
5336 | while (getline(&line, &linelen, f) != -1) { | |
54a6d46a CB |
5337 | pid = realloc(*pid_buf, sizeof(char *) * (sum + 1)); |
5338 | if (!pid) | |
5339 | return sum; | |
6db4f7a3 | 5340 | *pid_buf = pid; |
54a6d46a CB |
5341 | |
5342 | *(*pid_buf + sum) = malloc(strlen(line) + 1); | |
5343 | if (!*(*pid_buf + sum)) | |
5344 | return sum; | |
5345 | ||
6db4f7a3 | 5346 | strcpy(*(*pid_buf + sum), line); |
5347 | sum++; | |
5348 | } | |
54a6d46a | 5349 | |
6db4f7a3 | 5350 | return sum; |
5351 | } | |
54a6d46a | 5352 | |
6db4f7a3 | 5353 | /* |
5354 | * calc_load calculates the load according to the following formula: | |
5355 | * load1 = load0 * exp + active * (1 - exp) | |
5356 | * | |
5357 | * @load1: the new loadavg. | |
5358 | * @load0: the former loadavg. | |
5359 | * @active: the total number of running pid at this moment. | |
5360 | * @exp: the fixed-point defined in the beginning. | |
5361 | */ | |
5362 | static unsigned long | |
5363 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
5364 | { | |
5365 | unsigned long newload; | |
5366 | ||
5367 | active = active > 0 ? active * FIXED_1 : 0; | |
5368 | newload = load * exp + active * (FIXED_1 - exp); | |
5369 | if (active >= load) | |
5370 | newload += FIXED_1 - 1; | |
5371 | ||
5372 | return newload / FIXED_1; | |
5373 | } | |
5374 | ||
5375 | /* | |
5376 | * Return 0 means that container p->cg is closed. | |
5377 | * Return -1 means that error occurred in refresh. | |
5378 | * Positive num equals the total number of pid. | |
5379 | */ | |
5380 | static int refresh_load(struct load_node *p, char *path) | |
5381 | { | |
54a6d46a | 5382 | __do_free char *line = NULL; |
6db4f7a3 | 5383 | char **idbuf; |
5384 | char proc_path[256]; | |
5385 | int i, ret, run_pid = 0, total_pid = 0, last_pid = 0; | |
6db4f7a3 | 5386 | size_t linelen = 0; |
5387 | int sum, length; | |
6db4f7a3 | 5388 | struct dirent *file; |
5389 | ||
54a6d46a CB |
5390 | idbuf = malloc(sizeof(char *)); |
5391 | if (!idbuf) | |
5392 | return -1; | |
5393 | ||
6db4f7a3 | 5394 | sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd); |
5395 | /* normal exit */ | |
5396 | if (sum == 0) | |
5397 | goto out; | |
5398 | ||
5399 | for (i = 0; i < sum; i++) { | |
54a6d46a CB |
5400 | __do_closedir DIR *dp = NULL; |
5401 | ||
6db4f7a3 | 5402 | /*clean up '\n' */ |
5403 | length = strlen(idbuf[i])-1; | |
5404 | idbuf[i][length] = '\0'; | |
5405 | ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]); | |
5406 | if (ret < 0 || ret > 255) { | |
5407 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
5408 | i = sum; | |
5409 | sum = -1; | |
5410 | goto err_out; | |
5411 | } | |
5412 | ||
5413 | dp = opendir(proc_path); | |
5414 | if (!dp) { | |
5415 | lxcfs_error("%s\n", "Open proc_path failed in refresh_load."); | |
5416 | continue; | |
5417 | } | |
5418 | while ((file = readdir(dp)) != NULL) { | |
54a6d46a CB |
5419 | __do_fclose FILE *f = NULL; |
5420 | ||
6db4f7a3 | 5421 | if (strncmp(file->d_name, ".", 1) == 0) |
5422 | continue; | |
5423 | if (strncmp(file->d_name, "..", 1) == 0) | |
5424 | continue; | |
5425 | total_pid++; | |
5426 | /* We make the biggest pid become last_pid.*/ | |
5427 | ret = atof(file->d_name); | |
5428 | last_pid = (ret > last_pid) ? ret : last_pid; | |
5429 | ||
5430 | ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name); | |
5431 | if (ret < 0 || ret > 255) { | |
5432 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
5433 | i = sum; | |
5434 | sum = -1; | |
6db4f7a3 | 5435 | goto err_out; |
5436 | } | |
54a6d46a | 5437 | |
6db4f7a3 | 5438 | f = fopen(proc_path, "r"); |
5439 | if (f != NULL) { | |
5440 | while (getline(&line, &linelen, f) != -1) { | |
5441 | /* Find State */ | |
5442 | if ((line[0] == 'S') && (line[1] == 't')) | |
5443 | break; | |
5444 | } | |
54a6d46a | 5445 | |
6db4f7a3 | 5446 | if ((line[7] == 'R') || (line[7] == 'D')) |
5447 | run_pid++; | |
6db4f7a3 | 5448 | } |
5449 | } | |
6db4f7a3 | 5450 | } |
5451 | /*Calculate the loadavg.*/ | |
5452 | p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid); | |
5453 | p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid); | |
5454 | p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid); | |
5455 | p->run_pid = run_pid; | |
5456 | p->total_pid = total_pid; | |
5457 | p->last_pid = last_pid; | |
5458 | ||
beb5024e | 5459 | err_out: |
6db4f7a3 | 5460 | for (; i > 0; i--) |
5461 | free(idbuf[i-1]); | |
5462 | out: | |
5463 | free(idbuf); | |
5464 | return sum; | |
5465 | } | |
54a6d46a | 5466 | |
6db4f7a3 | 5467 | /* |
5468 | * Traverse the hash table and update it. | |
5469 | */ | |
5470 | void *load_begin(void *arg) | |
5471 | { | |
5472 | ||
6db4f7a3 | 5473 | int i, sum, length, ret; |
5474 | struct load_node *f; | |
5475 | int first_node; | |
5476 | clock_t time1, time2; | |
5477 | ||
5478 | while (1) { | |
a83618e2 JS |
5479 | if (loadavg_stop == 1) |
5480 | return NULL; | |
5481 | ||
6db4f7a3 | 5482 | time1 = clock(); |
5483 | for (i = 0; i < LOAD_SIZE; i++) { | |
5484 | pthread_mutex_lock(&load_hash[i].lock); | |
5485 | if (load_hash[i].next == NULL) { | |
5486 | pthread_mutex_unlock(&load_hash[i].lock); | |
5487 | continue; | |
5488 | } | |
5489 | f = load_hash[i].next; | |
5490 | first_node = 1; | |
5491 | while (f) { | |
54a6d46a CB |
5492 | __do_free char *path = NULL; |
5493 | ||
6db4f7a3 | 5494 | length = strlen(f->cg) + 2; |
6db4f7a3 | 5495 | /* strlen(f->cg) + '.' or '' + \0 */ |
54a6d46a CB |
5496 | path = malloc(length); |
5497 | if (!path) | |
5498 | goto out; | |
6db4f7a3 | 5499 | |
075387cd | 5500 | ret = snprintf(path, length, "%s%s", dot_or_empty(f->cg), f->cg); |
6db4f7a3 | 5501 | if (ret < 0 || ret > length - 1) { |
5502 | /* snprintf failed, ignore the node.*/ | |
5503 | lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg); | |
5504 | goto out; | |
5505 | } | |
54a6d46a | 5506 | |
6db4f7a3 | 5507 | sum = refresh_load(f, path); |
54a6d46a | 5508 | if (sum == 0) |
6db4f7a3 | 5509 | f = del_node(f, i); |
54a6d46a | 5510 | else |
6db4f7a3 | 5511 | out: f = f->next; |
6db4f7a3 | 5512 | /* load_hash[i].lock locks only on the first node.*/ |
5513 | if (first_node == 1) { | |
5514 | first_node = 0; | |
5515 | pthread_mutex_unlock(&load_hash[i].lock); | |
5516 | } | |
5517 | } | |
5518 | } | |
a83618e2 JS |
5519 | |
5520 | if (loadavg_stop == 1) | |
5521 | return NULL; | |
5522 | ||
6db4f7a3 | 5523 | time2 = clock(); |
5524 | usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC)); | |
5525 | } | |
5526 | } | |
5527 | ||
5528 | static int proc_loadavg_read(char *buf, size_t size, off_t offset, | |
5529 | struct fuse_file_info *fi) | |
5530 | { | |
5531 | struct fuse_context *fc = fuse_get_context(); | |
5532 | struct file_info *d = (struct file_info *)fi->fh; | |
5533 | pid_t initpid; | |
5534 | char *cg; | |
5535 | size_t total_len = 0; | |
5536 | char *cache = d->buf; | |
5537 | struct load_node *n; | |
5538 | int hash; | |
01d88ede | 5539 | int cfd, rv = 0; |
6db4f7a3 | 5540 | unsigned long a, b, c; |
5541 | ||
5542 | if (offset) { | |
5543 | if (offset > d->size) | |
5544 | return -EINVAL; | |
5545 | if (!d->cached) | |
5546 | return 0; | |
5547 | int left = d->size - offset; | |
5548 | total_len = left > size ? size : left; | |
5549 | memcpy(buf, cache + offset, total_len); | |
5550 | return total_len; | |
5551 | } | |
5552 | if (!loadavg) | |
5fbea8a6 | 5553 | return read_file_fuse("/proc/loadavg", buf, size, d); |
6db4f7a3 | 5554 | |
5555 | initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5556 | if (initpid <= 1 || is_shared_pidns(initpid)) |
6db4f7a3 | 5557 | initpid = fc->pid; |
5558 | cg = get_pid_cgroup(initpid, "cpu"); | |
5559 | if (!cg) | |
5fbea8a6 | 5560 | return read_file_fuse("/proc/loadavg", buf, size, d); |
6db4f7a3 | 5561 | |
5562 | prune_init_slice(cg); | |
b077527b | 5563 | hash = calc_hash(cg) % LOAD_SIZE; |
6db4f7a3 | 5564 | n = locate_node(cg, hash); |
5565 | ||
5566 | /* First time */ | |
5567 | if (n == NULL) { | |
d298bba1 | 5568 | cfd = get_cgroup_fd("cpu"); |
5fbea8a6 | 5569 | if (cfd >= 0) { |
6db4f7a3 | 5570 | /* |
5571 | * In locate_node() above, pthread_rwlock_unlock() isn't used | |
5572 | * because delete is not allowed before read has ended. | |
5573 | */ | |
5574 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
01d88ede JS |
5575 | rv = 0; |
5576 | goto err; | |
6db4f7a3 | 5577 | } |
5578 | do { | |
5579 | n = malloc(sizeof(struct load_node)); | |
5580 | } while (!n); | |
5581 | ||
5582 | do { | |
5583 | n->cg = malloc(strlen(cg)+1); | |
5584 | } while (!n->cg); | |
5585 | strcpy(n->cg, cg); | |
5586 | n->avenrun[0] = 0; | |
5587 | n->avenrun[1] = 0; | |
5588 | n->avenrun[2] = 0; | |
5589 | n->run_pid = 0; | |
5590 | n->total_pid = 1; | |
5591 | n->last_pid = initpid; | |
5592 | n->cfd = cfd; | |
5593 | insert_node(&n, hash); | |
5594 | } | |
5595 | a = n->avenrun[0] + (FIXED_1/200); | |
5596 | b = n->avenrun[1] + (FIXED_1/200); | |
5597 | c = n->avenrun[2] + (FIXED_1/200); | |
5598 | total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n", | |
5599 | LOAD_INT(a), LOAD_FRAC(a), | |
5600 | LOAD_INT(b), LOAD_FRAC(b), | |
5601 | LOAD_INT(c), LOAD_FRAC(c), | |
5602 | n->run_pid, n->total_pid, n->last_pid); | |
5603 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
5604 | if (total_len < 0 || total_len >= d->buflen) { | |
5605 | lxcfs_error("%s\n", "Failed to write to cache"); | |
01d88ede JS |
5606 | rv = 0; |
5607 | goto err; | |
6db4f7a3 | 5608 | } |
5609 | d->size = (int)total_len; | |
5610 | d->cached = 1; | |
5611 | ||
5612 | if (total_len > size) | |
5613 | total_len = size; | |
5614 | memcpy(buf, d->buf, total_len); | |
01d88ede JS |
5615 | rv = total_len; |
5616 | ||
5617 | err: | |
5618 | free(cg); | |
5619 | return rv; | |
6db4f7a3 | 5620 | } |
5621 | /* Return a positive number on success, return 0 on failure.*/ | |
5622 | pthread_t load_daemon(int load_use) | |
5623 | { | |
5624 | int ret; | |
5625 | pthread_t pid; | |
5626 | ||
5627 | ret = init_load(); | |
5628 | if (ret == -1) { | |
5629 | lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!"); | |
5630 | return 0; | |
5631 | } | |
5632 | ret = pthread_create(&pid, NULL, load_begin, NULL); | |
5633 | if (ret != 0) { | |
5634 | lxcfs_error("%s\n", "Create pthread fails in load_daemon!"); | |
5635 | load_free(); | |
5636 | return 0; | |
5637 | } | |
5638 | /* use loadavg, here loadavg = 1*/ | |
5639 | loadavg = load_use; | |
5640 | return pid; | |
5641 | } | |
70dcc12e | 5642 | |
a83618e2 JS |
5643 | /* Returns 0 on success. */ |
5644 | int stop_load_daemon(pthread_t pid) | |
5645 | { | |
5646 | int s; | |
5647 | ||
5648 | /* Signal the thread to gracefully stop */ | |
5649 | loadavg_stop = 1; | |
5650 | ||
5651 | s = pthread_join(pid, NULL); /* Make sure sub thread has been canceled. */ | |
5652 | if (s != 0) { | |
5653 | lxcfs_error("%s\n", "stop_load_daemon error: failed to join"); | |
5654 | return -1; | |
5655 | } | |
5656 | ||
5657 | load_free(); | |
5658 | loadavg_stop = 0; | |
5659 | ||
5660 | return 0; | |
5661 | } | |
5662 | ||
237e200e SH |
5663 | static off_t get_procfile_size(const char *which) |
5664 | { | |
5665 | FILE *f = fopen(which, "r"); | |
5666 | char *line = NULL; | |
5667 | size_t len = 0; | |
5668 | ssize_t sz, answer = 0; | |
5669 | if (!f) | |
5670 | return 0; | |
5671 | ||
5672 | while ((sz = getline(&line, &len, f)) != -1) | |
5673 | answer += sz; | |
5674 | fclose (f); | |
5675 | free(line); | |
5676 | ||
5677 | return answer; | |
5678 | } | |
5679 | ||
5680 | int proc_getattr(const char *path, struct stat *sb) | |
5681 | { | |
5682 | struct timespec now; | |
5683 | ||
5684 | memset(sb, 0, sizeof(struct stat)); | |
5685 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
5686 | return -EINVAL; | |
5687 | sb->st_uid = sb->st_gid = 0; | |
5688 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
5689 | if (strcmp(path, "/proc") == 0) { | |
5690 | sb->st_mode = S_IFDIR | 00555; | |
5691 | sb->st_nlink = 2; | |
5692 | return 0; | |
5693 | } | |
5694 | if (strcmp(path, "/proc/meminfo") == 0 || | |
5695 | strcmp(path, "/proc/cpuinfo") == 0 || | |
5696 | strcmp(path, "/proc/uptime") == 0 || | |
5697 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e | 5698 | strcmp(path, "/proc/diskstats") == 0 || |
46be8eed | 5699 | strcmp(path, "/proc/swaps") == 0 || |
5700 | strcmp(path, "/proc/loadavg") == 0) { | |
237e200e SH |
5701 | sb->st_size = 0; |
5702 | sb->st_mode = S_IFREG | 00444; | |
5703 | sb->st_nlink = 1; | |
5704 | return 0; | |
5705 | } | |
5706 | ||
5707 | return -ENOENT; | |
5708 | } | |
5709 | ||
5710 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
5711 | struct fuse_file_info *fi) | |
5712 | { | |
d639f863 CB |
5713 | if (filler(buf, ".", NULL, 0) != 0 || |
5714 | filler(buf, "..", NULL, 0) != 0 || | |
5715 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
5716 | filler(buf, "meminfo", NULL, 0) != 0 || | |
5717 | filler(buf, "stat", NULL, 0) != 0 || | |
5718 | filler(buf, "uptime", NULL, 0) != 0 || | |
5719 | filler(buf, "diskstats", NULL, 0) != 0 || | |
46be8eed | 5720 | filler(buf, "swaps", NULL, 0) != 0 || |
5721 | filler(buf, "loadavg", NULL, 0) != 0) | |
237e200e SH |
5722 | return -EINVAL; |
5723 | return 0; | |
5724 | } | |
5725 | ||
5726 | int proc_open(const char *path, struct fuse_file_info *fi) | |
5727 | { | |
5728 | int type = -1; | |
5729 | struct file_info *info; | |
5730 | ||
5731 | if (strcmp(path, "/proc/meminfo") == 0) | |
5732 | type = LXC_TYPE_PROC_MEMINFO; | |
5733 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
5734 | type = LXC_TYPE_PROC_CPUINFO; | |
5735 | else if (strcmp(path, "/proc/uptime") == 0) | |
5736 | type = LXC_TYPE_PROC_UPTIME; | |
5737 | else if (strcmp(path, "/proc/stat") == 0) | |
5738 | type = LXC_TYPE_PROC_STAT; | |
5739 | else if (strcmp(path, "/proc/diskstats") == 0) | |
5740 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
5741 | else if (strcmp(path, "/proc/swaps") == 0) |
5742 | type = LXC_TYPE_PROC_SWAPS; | |
46be8eed | 5743 | else if (strcmp(path, "/proc/loadavg") == 0) |
5744 | type = LXC_TYPE_PROC_LOADAVG; | |
237e200e SH |
5745 | if (type == -1) |
5746 | return -ENOENT; | |
5747 | ||
5748 | info = malloc(sizeof(*info)); | |
5749 | if (!info) | |
5750 | return -ENOMEM; | |
5751 | ||
5752 | memset(info, 0, sizeof(*info)); | |
5753 | info->type = type; | |
5754 | ||
5755 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
5756 | do { | |
5757 | info->buf = malloc(info->buflen); | |
5758 | } while (!info->buf); | |
5759 | memset(info->buf, 0, info->buflen); | |
5760 | /* set actual size to buffer size */ | |
5761 | info->size = info->buflen; | |
5762 | ||
5763 | fi->fh = (unsigned long)info; | |
5764 | return 0; | |
5765 | } | |
5766 | ||
bddbb106 SH |
5767 | int proc_access(const char *path, int mask) |
5768 | { | |
e7849aa3 CB |
5769 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) |
5770 | return 0; | |
5771 | ||
bddbb106 SH |
5772 | /* these are all read-only */ |
5773 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 5774 | return -EACCES; |
bddbb106 SH |
5775 | return 0; |
5776 | } | |
5777 | ||
237e200e SH |
5778 | int proc_release(const char *path, struct fuse_file_info *fi) |
5779 | { | |
43215927 | 5780 | do_release_file_info(fi); |
237e200e SH |
5781 | return 0; |
5782 | } | |
5783 | ||
5784 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
5785 | struct fuse_file_info *fi) | |
5786 | { | |
5787 | struct file_info *f = (struct file_info *) fi->fh; | |
5788 | ||
5789 | switch (f->type) { | |
5790 | case LXC_TYPE_PROC_MEMINFO: | |
5791 | return proc_meminfo_read(buf, size, offset, fi); | |
5792 | case LXC_TYPE_PROC_CPUINFO: | |
5793 | return proc_cpuinfo_read(buf, size, offset, fi); | |
5794 | case LXC_TYPE_PROC_UPTIME: | |
5795 | return proc_uptime_read(buf, size, offset, fi); | |
5796 | case LXC_TYPE_PROC_STAT: | |
5797 | return proc_stat_read(buf, size, offset, fi); | |
5798 | case LXC_TYPE_PROC_DISKSTATS: | |
5799 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
5800 | case LXC_TYPE_PROC_SWAPS: |
5801 | return proc_swaps_read(buf, size, offset, fi); | |
46be8eed | 5802 | case LXC_TYPE_PROC_LOADAVG: |
5803 | return proc_loadavg_read(buf, size, offset, fi); | |
237e200e SH |
5804 | default: |
5805 | return -EINVAL; | |
5806 | } | |
5807 | } | |
5808 | ||
29a73c2f CB |
5809 | /* |
5810 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
5811 | */ |
5812 | ||
29a73c2f CB |
5813 | static bool umount_if_mounted(void) |
5814 | { | |
5815 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
b8defc3d | 5816 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
5817 | return false; |
5818 | } | |
5819 | return true; | |
5820 | } | |
5821 | ||
2283e240 CB |
5822 | /* __typeof__ should be safe to use with all compilers. */ |
5823 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
5824 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
5825 | { | |
5826 | return (fs->f_type == (fs_type_magic)magic_val); | |
5827 | } | |
5828 | ||
0a4dea41 CB |
5829 | /* |
5830 | * looking at fs/proc_namespace.c, it appears we can | |
5831 | * actually expect the rootfs entry to very specifically contain | |
5832 | * " - rootfs rootfs " | |
5833 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
5834 | * the rootfs entry should always be skipped in mountinfo contents. | |
5835 | */ | |
5836 | static bool is_on_ramfs(void) | |
5837 | { | |
5838 | FILE *f; | |
5839 | char *p, *p2; | |
5840 | char *line = NULL; | |
5841 | size_t len = 0; | |
5842 | int i; | |
5843 | ||
5844 | f = fopen("/proc/self/mountinfo", "r"); | |
5845 | if (!f) | |
5846 | return false; | |
5847 | ||
5848 | while (getline(&line, &len, f) != -1) { | |
5849 | for (p = line, i = 0; p && i < 4; i++) | |
5850 | p = strchr(p + 1, ' '); | |
5851 | if (!p) | |
5852 | continue; | |
5853 | p2 = strchr(p + 1, ' '); | |
5854 | if (!p2) | |
5855 | continue; | |
5856 | *p2 = '\0'; | |
5857 | if (strcmp(p + 1, "/") == 0) { | |
5858 | // this is '/'. is it the ramfs? | |
5859 | p = strchr(p2 + 1, '-'); | |
5860 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
5861 | free(line); | |
5862 | fclose(f); | |
5863 | return true; | |
5864 | } | |
5865 | } | |
5866 | } | |
5867 | free(line); | |
5868 | fclose(f); | |
5869 | return false; | |
5870 | } | |
5871 | ||
cc309f33 | 5872 | static int pivot_enter() |
0a4dea41 | 5873 | { |
cc309f33 CB |
5874 | int ret = -1, oldroot = -1, newroot = -1; |
5875 | ||
5876 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
5877 | if (oldroot < 0) { | |
5878 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
5879 | return ret; | |
5880 | } | |
5881 | ||
5882 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
5883 | if (newroot < 0) { | |
5884 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
5885 | goto err; | |
5886 | } | |
5887 | ||
5888 | /* change into new root fs */ | |
5889 | if (fchdir(newroot) < 0) { | |
5890 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
5891 | goto err; | |
5892 | } | |
5893 | ||
0a4dea41 CB |
5894 | /* pivot_root into our new root fs */ |
5895 | if (pivot_root(".", ".") < 0) { | |
5896 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
cc309f33 | 5897 | goto err; |
0a4dea41 CB |
5898 | } |
5899 | ||
5900 | /* | |
5901 | * At this point the old-root is mounted on top of our new-root. | |
5902 | * To unmounted it we must not be chdir'd into it, so escape back | |
5903 | * to the old-root. | |
5904 | */ | |
5905 | if (fchdir(oldroot) < 0) { | |
5906 | lxcfs_error("%s\n", "Failed to enter old root."); | |
cc309f33 | 5907 | goto err; |
0a4dea41 CB |
5908 | } |
5909 | ||
5910 | if (umount2(".", MNT_DETACH) < 0) { | |
5911 | lxcfs_error("%s\n", "Failed to detach old root."); | |
cc309f33 | 5912 | goto err; |
0a4dea41 CB |
5913 | } |
5914 | ||
5915 | if (fchdir(newroot) < 0) { | |
5916 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
cc309f33 | 5917 | goto err; |
0a4dea41 CB |
5918 | } |
5919 | ||
cc309f33 CB |
5920 | ret = 0; |
5921 | ||
5922 | err: | |
5923 | if (oldroot > 0) | |
5924 | close(oldroot); | |
5925 | if (newroot > 0) | |
5926 | close(newroot); | |
5927 | ||
5928 | return ret; | |
0a4dea41 CB |
5929 | } |
5930 | ||
5931 | static int chroot_enter() | |
5932 | { | |
5933 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
5934 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
5935 | return -1; | |
5936 | } | |
5937 | ||
5938 | if (chroot(".") < 0) { | |
5939 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
5940 | return -1; | |
5941 | } | |
5942 | ||
5943 | if (chdir("/") < 0) { | |
5944 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
5945 | return -1; | |
5946 | } | |
5947 | ||
5948 | return 0; | |
5949 | } | |
5950 | ||
0232cbac | 5951 | static int permute_and_enter(void) |
29a73c2f | 5952 | { |
0a4dea41 CB |
5953 | struct statfs sb; |
5954 | ||
5955 | if (statfs("/", &sb) < 0) { | |
5956 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
cc309f33 | 5957 | return -1; |
0a4dea41 CB |
5958 | } |
5959 | ||
5960 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
5961 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
5962 | * /proc/1/mountinfo. */ | |
5963 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
5964 | return chroot_enter(); | |
29a73c2f | 5965 | |
cc309f33 | 5966 | if (pivot_enter() < 0) { |
0a4dea41 | 5967 | lxcfs_error("%s\n", "Could not perform pivot root."); |
cc309f33 | 5968 | return -1; |
29a73c2f CB |
5969 | } |
5970 | ||
cc309f33 | 5971 | return 0; |
29a73c2f CB |
5972 | } |
5973 | ||
5974 | /* Prepare our new clean root. */ | |
0232cbac | 5975 | static int permute_prepare(void) |
29a73c2f CB |
5976 | { |
5977 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
b8defc3d | 5978 | lxcfs_error("%s\n", "Failed to create directory for new root."); |
29a73c2f CB |
5979 | return -1; |
5980 | } | |
5981 | ||
5982 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5983 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5984 | return -1; |
5985 | } | |
5986 | ||
5987 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5988 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5989 | return -1; |
5990 | } | |
5991 | ||
5992 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
b8defc3d | 5993 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5994 | return -1; |
5995 | } | |
5996 | ||
5997 | return 0; | |
5998 | } | |
5999 | ||
0232cbac CB |
6000 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ |
6001 | static bool permute_root(void) | |
29a73c2f CB |
6002 | { |
6003 | /* Prepare new root. */ | |
0232cbac | 6004 | if (permute_prepare() < 0) |
29a73c2f CB |
6005 | return false; |
6006 | ||
6007 | /* Pivot into new root. */ | |
0232cbac | 6008 | if (permute_and_enter() < 0) |
29a73c2f CB |
6009 | return false; |
6010 | ||
6011 | return true; | |
6012 | } | |
6013 | ||
a257a8ee CB |
6014 | static int preserve_mnt_ns(int pid) |
6015 | { | |
6016 | int ret; | |
6017 | size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt"); | |
6018 | char path[len]; | |
6019 | ||
6020 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
6021 | if (ret < 0 || (size_t)ret >= len) | |
6022 | return -1; | |
6023 | ||
6024 | return open(path, O_RDONLY | O_CLOEXEC); | |
6025 | } | |
6026 | ||
0a4dea41 | 6027 | static bool cgfs_prepare_mounts(void) |
29a73c2f CB |
6028 | { |
6029 | if (!mkdir_p(BASEDIR, 0700)) { | |
b8defc3d | 6030 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); |
29a73c2f CB |
6031 | return false; |
6032 | } | |
480262c9 | 6033 | |
29a73c2f | 6034 | if (!umount_if_mounted()) { |
b8defc3d | 6035 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); |
480262c9 CB |
6036 | return false; |
6037 | } | |
6038 | ||
6039 | if (unshare(CLONE_NEWNS) < 0) { | |
b8defc3d | 6040 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); |
480262c9 CB |
6041 | return false; |
6042 | } | |
6043 | ||
0646f250 CB |
6044 | cgroup_ops->mntns_fd = preserve_mnt_ns(getpid()); |
6045 | if (cgroup_ops->mntns_fd < 0) { | |
a257a8ee CB |
6046 | lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno)); |
6047 | return false; | |
6048 | } | |
6049 | ||
480262c9 | 6050 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { |
b8defc3d | 6051 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); |
29a73c2f CB |
6052 | return false; |
6053 | } | |
480262c9 | 6054 | |
29a73c2f | 6055 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
b8defc3d | 6056 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); |
29a73c2f CB |
6057 | return false; |
6058 | } | |
480262c9 | 6059 | |
29a73c2f CB |
6060 | return true; |
6061 | } | |
6062 | ||
0a4dea41 | 6063 | static bool cgfs_mount_hierarchies(void) |
29a73c2f | 6064 | { |
5fbea8a6 CB |
6065 | if (!mkdir_p(BASEDIR DEFAULT_CGROUP_MOUNTPOINT, 0755)) |
6066 | return false; | |
51c7ca35 | 6067 | |
5fbea8a6 CB |
6068 | if (!cgroup_ops->mount(cgroup_ops, BASEDIR)) |
6069 | return false; | |
29a73c2f | 6070 | |
5fbea8a6 CB |
6071 | for (struct hierarchy **h = cgroup_ops->hierarchies; h && *h; h++) { |
6072 | __do_free char *path = must_make_path(BASEDIR, (*h)->mountpoint, NULL); | |
6073 | (*h)->fd = open(path, O_DIRECTORY | O_CLOEXEC | O_NOFOLLOW); | |
6074 | if ((*h)->fd < 0) | |
29a73c2f | 6075 | return false; |
29a73c2f | 6076 | } |
5fbea8a6 | 6077 | |
29a73c2f CB |
6078 | return true; |
6079 | } | |
6080 | ||
480262c9 | 6081 | static bool cgfs_setup_controllers(void) |
29a73c2f | 6082 | { |
0a4dea41 | 6083 | if (!cgfs_prepare_mounts()) |
29a73c2f | 6084 | return false; |
29a73c2f | 6085 | |
0a4dea41 | 6086 | if (!cgfs_mount_hierarchies()) { |
b8defc3d | 6087 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); |
29a73c2f CB |
6088 | return false; |
6089 | } | |
6090 | ||
0232cbac | 6091 | if (!permute_root()) |
29a73c2f CB |
6092 | return false; |
6093 | ||
6094 | return true; | |
6095 | } | |
6096 | ||
2243c5a9 | 6097 | static void __attribute__((constructor)) lxcfs_init(void) |
237e200e | 6098 | { |
2243c5a9 | 6099 | __do_close_prot_errno int init_ns = -EBADF; |
5fbea8a6 | 6100 | char *cret; |
e58dab00 | 6101 | char cwd[MAXPATHLEN]; |
237e200e | 6102 | |
5fbea8a6 CB |
6103 | cgroup_ops = cgroup_init(); |
6104 | if (!cgroup_ops) | |
2243c5a9 | 6105 | log_exit("Failed to initialize cgroup support"); |
237e200e | 6106 | |
480262c9 | 6107 | /* Preserve initial namespace. */ |
a257a8ee | 6108 | init_ns = preserve_mnt_ns(getpid()); |
2243c5a9 CB |
6109 | if (init_ns < 0) |
6110 | log_exit("Failed to preserve initial mount namespace"); | |
480262c9 | 6111 | |
e58dab00 | 6112 | cret = getcwd(cwd, MAXPATHLEN); |
2243c5a9 | 6113 | log_exit("%s - Could not retrieve current working directory", strerror(errno)); |
e58dab00 | 6114 | |
480262c9 CB |
6115 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace |
6116 | * to privately mount lxcfs cgroups. */ | |
2243c5a9 CB |
6117 | if (!cgfs_setup_controllers()) |
6118 | log_exit("Failed to setup private cgroup mounts for lxcfs"); | |
480262c9 | 6119 | |
2243c5a9 CB |
6120 | if (setns(init_ns, 0) < 0) |
6121 | log_exit("%s - Failed to switch back to initial mount namespace", strerror(errno)); | |
29a73c2f | 6122 | |
e58dab00 | 6123 | if (!cret || chdir(cwd) < 0) |
2243c5a9 | 6124 | log_exit("%s - Could not change back to original working directory", strerror(errno)); |
e58dab00 | 6125 | |
2243c5a9 CB |
6126 | if (!init_cpuview()) |
6127 | log_exit("Failed to init CPU view"); | |
056adcef | 6128 | |
237e200e | 6129 | print_subsystems(); |
237e200e SH |
6130 | } |
6131 | ||
2243c5a9 | 6132 | static void __attribute__((destructor)) lxcfs_exit(void) |
237e200e | 6133 | { |
0646f250 | 6134 | lxcfs_debug("%s\n", "Running destructor for liblxcfs"); |
056adcef | 6135 | free_cpuview(); |
2243c5a9 | 6136 | cgroup_exit(cgroup_ops); |
1c4b4e38 | 6137 | } |