]>
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" |
c9236032 HY |
41 | #include "memory_utils.h" |
42 | #include "config.h" | |
237e200e | 43 | |
29a73c2f CB |
44 | /* Define pivot_root() if missing from the C library */ |
45 | #ifndef HAVE_PIVOT_ROOT | |
46 | static int pivot_root(const char * new_root, const char * put_old) | |
47 | { | |
48 | #ifdef __NR_pivot_root | |
49 | return syscall(__NR_pivot_root, new_root, put_old); | |
50 | #else | |
51 | errno = ENOSYS; | |
52 | return -1; | |
53 | #endif | |
54 | } | |
55 | #else | |
56 | extern int pivot_root(const char * new_root, const char * put_old); | |
57 | #endif | |
58 | ||
8be92dd1 JS |
59 | struct cpuacct_usage { |
60 | uint64_t user; | |
61 | uint64_t system; | |
056adcef | 62 | uint64_t idle; |
77005a6c | 63 | bool online; |
8be92dd1 JS |
64 | }; |
65 | ||
0e47acaa | 66 | /* The function of hash table.*/ |
67 | #define LOAD_SIZE 100 /*the size of hash_table */ | |
6db4f7a3 | 68 | #define FLUSH_TIME 5 /*the flush rate */ |
69 | #define DEPTH_DIR 3 /*the depth of per cgroup */ | |
70 | /* The function of calculate loadavg .*/ | |
71 | #define FSHIFT 11 /* nr of bits of precision */ | |
72 | #define FIXED_1 (1<<FSHIFT) /* 1.0 as fixed-point */ | |
73 | #define EXP_1 1884 /* 1/exp(5sec/1min) as fixed-point */ | |
74 | #define EXP_5 2014 /* 1/exp(5sec/5min) */ | |
75 | #define EXP_15 2037 /* 1/exp(5sec/15min) */ | |
76 | #define LOAD_INT(x) ((x) >> FSHIFT) | |
77 | #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) | |
beb5024e | 78 | /* |
6db4f7a3 | 79 | * This parameter is used for proc_loadavg_read(). |
80 | * 1 means use loadavg, 0 means not use. | |
81 | */ | |
82 | static int loadavg = 0; | |
a83618e2 | 83 | static volatile sig_atomic_t loadavg_stop = 0; |
056adcef | 84 | static int calc_hash(const char *name) |
0e47acaa | 85 | { |
86 | unsigned int hash = 0; | |
87 | unsigned int x = 0; | |
88 | /* ELFHash algorithm. */ | |
89 | while (*name) { | |
90 | hash = (hash << 4) + *name++; | |
91 | x = hash & 0xf0000000; | |
92 | if (x != 0) | |
93 | hash ^= (x >> 24); | |
94 | hash &= ~x; | |
95 | } | |
b077527b | 96 | return (hash & 0x7fffffff); |
0e47acaa | 97 | } |
98 | ||
99 | struct load_node { | |
100 | char *cg; /*cg */ | |
101 | unsigned long avenrun[3]; /* Load averages */ | |
102 | unsigned int run_pid; | |
103 | unsigned int total_pid; | |
104 | unsigned int last_pid; | |
105 | int cfd; /* The file descriptor of the mounted cgroup */ | |
106 | struct load_node *next; | |
107 | struct load_node **pre; | |
108 | }; | |
109 | ||
110 | struct load_head { | |
111 | /* | |
112 | * The lock is about insert load_node and refresh load_node.To the first | |
113 | * load_node of each hash bucket, insert and refresh in this hash bucket is | |
114 | * mutually exclusive. | |
115 | */ | |
116 | pthread_mutex_t lock; | |
117 | /* | |
118 | * The rdlock is about read loadavg and delete load_node.To each hash | |
119 | * bucket, read and delete is mutually exclusive. But at the same time, we | |
120 | * allow paratactic read operation. This rdlock is at list level. | |
121 | */ | |
122 | pthread_rwlock_t rdlock; | |
123 | /* | |
124 | * The rilock is about read loadavg and insert load_node.To the first | |
125 | * load_node of each hash bucket, read and insert is mutually exclusive. | |
126 | * But at the same time, we allow paratactic read operation. | |
127 | */ | |
128 | pthread_rwlock_t rilock; | |
129 | struct load_node *next; | |
130 | }; | |
131 | ||
132 | static struct load_head load_hash[LOAD_SIZE]; /* hash table */ | |
133 | /* | |
134 | * init_load initialize the hash table. | |
135 | * Return 0 on success, return -1 on failure. | |
136 | */ | |
137 | static int init_load(void) | |
138 | { | |
139 | int i; | |
140 | int ret; | |
141 | ||
142 | for (i = 0; i < LOAD_SIZE; i++) { | |
143 | load_hash[i].next = NULL; | |
144 | ret = pthread_mutex_init(&load_hash[i].lock, NULL); | |
145 | if (ret != 0) { | |
146 | lxcfs_error("%s\n", "Failed to initialize lock"); | |
147 | goto out3; | |
148 | } | |
149 | ret = pthread_rwlock_init(&load_hash[i].rdlock, NULL); | |
150 | if (ret != 0) { | |
151 | lxcfs_error("%s\n", "Failed to initialize rdlock"); | |
152 | goto out2; | |
153 | } | |
154 | ret = pthread_rwlock_init(&load_hash[i].rilock, NULL); | |
155 | if (ret != 0) { | |
156 | lxcfs_error("%s\n", "Failed to initialize rilock"); | |
157 | goto out1; | |
158 | } | |
159 | } | |
160 | return 0; | |
161 | out1: | |
162 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
163 | out2: | |
164 | pthread_mutex_destroy(&load_hash[i].lock); | |
165 | out3: | |
166 | while (i > 0) { | |
167 | i--; | |
168 | pthread_mutex_destroy(&load_hash[i].lock); | |
169 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
170 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
171 | } | |
172 | return -1; | |
173 | } | |
174 | ||
175 | static void insert_node(struct load_node **n, int locate) | |
176 | { | |
177 | struct load_node *f; | |
178 | ||
179 | pthread_mutex_lock(&load_hash[locate].lock); | |
180 | pthread_rwlock_wrlock(&load_hash[locate].rilock); | |
181 | f = load_hash[locate].next; | |
182 | load_hash[locate].next = *n; | |
183 | ||
184 | (*n)->pre = &(load_hash[locate].next); | |
185 | if (f) | |
186 | f->pre = &((*n)->next); | |
187 | (*n)->next = f; | |
188 | pthread_mutex_unlock(&load_hash[locate].lock); | |
189 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
190 | } | |
191 | /* | |
192 | * locate_node() finds special node. Not return NULL means success. | |
193 | * It should be noted that rdlock isn't unlocked at the end of code | |
194 | * because this function is used to read special node. Delete is not | |
195 | * allowed before read has ended. | |
196 | * unlock rdlock only in proc_loadavg_read(). | |
197 | */ | |
198 | static struct load_node *locate_node(char *cg, int locate) | |
199 | { | |
200 | struct load_node *f = NULL; | |
201 | int i = 0; | |
202 | ||
203 | pthread_rwlock_rdlock(&load_hash[locate].rilock); | |
204 | pthread_rwlock_rdlock(&load_hash[locate].rdlock); | |
205 | if (load_hash[locate].next == NULL) { | |
206 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
207 | return f; | |
208 | } | |
209 | f = load_hash[locate].next; | |
210 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
211 | while (f && ((i = strcmp(f->cg, cg)) != 0)) | |
212 | f = f->next; | |
213 | return f; | |
214 | } | |
215 | /* Delete the load_node n and return the next node of it. */ | |
216 | static struct load_node *del_node(struct load_node *n, int locate) | |
217 | { | |
218 | struct load_node *g; | |
219 | ||
220 | pthread_rwlock_wrlock(&load_hash[locate].rdlock); | |
221 | if (n->next == NULL) { | |
222 | *(n->pre) = NULL; | |
223 | } else { | |
224 | *(n->pre) = n->next; | |
225 | n->next->pre = n->pre; | |
226 | } | |
227 | g = n->next; | |
228 | free(n->cg); | |
229 | free(n); | |
230 | pthread_rwlock_unlock(&load_hash[locate].rdlock); | |
231 | return g; | |
232 | } | |
233 | ||
a83618e2 | 234 | static void load_free(void) |
9c480eb7 | 235 | { |
236 | int i; | |
237 | struct load_node *f, *p; | |
238 | ||
239 | for (i = 0; i < LOAD_SIZE; i++) { | |
240 | pthread_mutex_lock(&load_hash[i].lock); | |
241 | pthread_rwlock_wrlock(&load_hash[i].rilock); | |
242 | pthread_rwlock_wrlock(&load_hash[i].rdlock); | |
243 | if (load_hash[i].next == NULL) { | |
244 | pthread_mutex_unlock(&load_hash[i].lock); | |
245 | pthread_mutex_destroy(&load_hash[i].lock); | |
246 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
247 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
248 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
249 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
250 | continue; | |
251 | } | |
252 | for (f = load_hash[i].next; f; ) { | |
253 | free(f->cg); | |
254 | p = f->next; | |
255 | free(f); | |
256 | f = p; | |
257 | } | |
258 | pthread_mutex_unlock(&load_hash[i].lock); | |
259 | pthread_mutex_destroy(&load_hash[i].lock); | |
260 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
261 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
262 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
263 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
264 | } | |
265 | } | |
056adcef JS |
266 | |
267 | /* Data for CPU view */ | |
268 | struct cg_proc_stat { | |
269 | char *cg; | |
270 | struct cpuacct_usage *usage; // Real usage as read from the host's /proc/stat | |
271 | struct cpuacct_usage *view; // Usage stats reported to the container | |
272 | int cpu_count; | |
2f49b662 | 273 | pthread_mutex_t lock; // For node manipulation |
056adcef JS |
274 | struct cg_proc_stat *next; |
275 | }; | |
276 | ||
277 | struct cg_proc_stat_head { | |
278 | struct cg_proc_stat *next; | |
951acc94 | 279 | time_t lastcheck; |
2f49b662 JS |
280 | |
281 | /* | |
282 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
283 | */ | |
284 | pthread_rwlock_t lock; | |
056adcef JS |
285 | }; |
286 | ||
287 | #define CPUVIEW_HASH_SIZE 100 | |
288 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
289 | ||
290 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
291 | { | |
292 | *head = malloc(sizeof(struct cg_proc_stat_head)); | |
293 | if (!(*head)) { | |
294 | lxcfs_error("%s\n", strerror(errno)); | |
295 | return false; | |
296 | } | |
297 | ||
951acc94 | 298 | (*head)->lastcheck = time(NULL); |
056adcef | 299 | (*head)->next = NULL; |
2f49b662 JS |
300 | |
301 | if (pthread_rwlock_init(&(*head)->lock, NULL) != 0) { | |
302 | lxcfs_error("%s\n", "Failed to initialize list lock"); | |
303 | free(*head); | |
304 | return false; | |
305 | } | |
306 | ||
056adcef JS |
307 | return true; |
308 | } | |
309 | ||
310 | static bool init_cpuview() | |
311 | { | |
312 | int i; | |
313 | ||
314 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
315 | proc_stat_history[i] = NULL; | |
316 | ||
317 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
318 | if (!cpuview_init_head(&proc_stat_history[i])) | |
319 | goto err; | |
320 | } | |
321 | ||
322 | return true; | |
323 | ||
324 | err: | |
325 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
326 | if (proc_stat_history[i]) { | |
327 | free(proc_stat_history[i]); | |
328 | proc_stat_history[i] = NULL; | |
329 | } | |
330 | } | |
331 | ||
332 | return false; | |
333 | } | |
334 | ||
951acc94 JS |
335 | static void free_proc_stat_node(struct cg_proc_stat *node) |
336 | { | |
2f49b662 | 337 | pthread_mutex_destroy(&node->lock); |
951acc94 JS |
338 | free(node->cg); |
339 | free(node->usage); | |
340 | free(node->view); | |
341 | free(node); | |
342 | } | |
343 | ||
056adcef JS |
344 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
345 | { | |
346 | struct cg_proc_stat *node, *tmp; | |
347 | ||
348 | if (head->next) { | |
349 | node = head->next; | |
350 | ||
351 | for (;;) { | |
352 | tmp = node; | |
353 | node = node->next; | |
951acc94 | 354 | free_proc_stat_node(tmp); |
056adcef JS |
355 | |
356 | if (!node) | |
357 | break; | |
358 | } | |
359 | } | |
360 | ||
2f49b662 | 361 | pthread_rwlock_destroy(&head->lock); |
056adcef JS |
362 | free(head); |
363 | } | |
364 | ||
365 | static void free_cpuview() | |
366 | { | |
367 | int i; | |
368 | ||
369 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
370 | if (proc_stat_history[i]) | |
371 | cpuview_free_head(proc_stat_history[i]); | |
372 | } | |
373 | } | |
374 | ||
237e200e SH |
375 | /* |
376 | * A table caching which pid is init for a pid namespace. | |
377 | * When looking up which pid is init for $qpid, we first | |
378 | * 1. Stat /proc/$qpid/ns/pid. | |
379 | * 2. Check whether the ino_t is in our store. | |
380 | * a. if not, fork a child in qpid's ns to send us | |
381 | * ucred.pid = 1, and read the initpid. Cache | |
382 | * initpid and creation time for /proc/initpid | |
383 | * in a new store entry. | |
384 | * b. if so, verify that /proc/initpid still matches | |
385 | * what we have saved. If not, clear the store | |
386 | * entry and go back to a. If so, return the | |
387 | * cached initpid. | |
388 | */ | |
389 | struct pidns_init_store { | |
390 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
391 | pid_t initpid; // the pid of nit in that ns | |
392 | long int ctime; // the time at which /proc/$initpid was created | |
393 | struct pidns_init_store *next; | |
394 | long int lastcheck; | |
395 | }; | |
396 | ||
397 | /* lol - look at how they are allocated in the kernel */ | |
398 | #define PIDNS_HASH_SIZE 4096 | |
399 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
400 | ||
401 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
402 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
403 | static void lock_mutex(pthread_mutex_t *l) | |
404 | { | |
405 | int ret; | |
406 | ||
407 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
b8defc3d | 408 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
409 | exit(1); |
410 | } | |
411 | } | |
412 | ||
29a73c2f CB |
413 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
414 | * Number of hierarchies mounted. */ | |
415 | static int num_hierarchies; | |
416 | ||
417 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
418 | * Hierachies mounted {cpuset, blkio, ...}: | |
419 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
420 | static char **hierarchies; | |
421 | ||
422 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
423 | * Open file descriptors: | |
424 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
425 | * private mount namespace. | |
426 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
427 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
428 | * mounts and respective files in the private namespace even when located in | |
429 | * another namespace using the *at() family of functions | |
430 | * {openat(), fchownat(), ...}. */ | |
431 | static int *fd_hierarchies; | |
a257a8ee | 432 | static int cgroup_mount_ns_fd = -1; |
29a73c2f | 433 | |
237e200e SH |
434 | static void unlock_mutex(pthread_mutex_t *l) |
435 | { | |
436 | int ret; | |
437 | ||
438 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
b8defc3d | 439 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
440 | exit(1); |
441 | } | |
442 | } | |
443 | ||
444 | static void store_lock(void) | |
445 | { | |
446 | lock_mutex(&pidns_store_mutex); | |
447 | } | |
448 | ||
449 | static void store_unlock(void) | |
450 | { | |
451 | unlock_mutex(&pidns_store_mutex); | |
452 | } | |
453 | ||
454 | /* Must be called under store_lock */ | |
455 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
456 | { | |
457 | struct stat initsb; | |
458 | char fnam[100]; | |
459 | ||
460 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
461 | if (stat(fnam, &initsb) < 0) | |
462 | return false; | |
7dd6560a CB |
463 | |
464 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
465 | initsb.st_ctime, e->initpid); | |
466 | ||
237e200e SH |
467 | if (e->ctime != initsb.st_ctime) |
468 | return false; | |
469 | return true; | |
470 | } | |
471 | ||
472 | /* Must be called under store_lock */ | |
473 | static void remove_initpid(struct pidns_init_store *e) | |
474 | { | |
475 | struct pidns_init_store *tmp; | |
476 | int h; | |
477 | ||
7dd6560a CB |
478 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
479 | ||
237e200e SH |
480 | h = HASH(e->ino); |
481 | if (pidns_hash_table[h] == e) { | |
482 | pidns_hash_table[h] = e->next; | |
483 | free(e); | |
484 | return; | |
485 | } | |
486 | ||
487 | tmp = pidns_hash_table[h]; | |
488 | while (tmp) { | |
489 | if (tmp->next == e) { | |
490 | tmp->next = e->next; | |
491 | free(e); | |
492 | return; | |
493 | } | |
494 | tmp = tmp->next; | |
495 | } | |
496 | } | |
497 | ||
498 | #define PURGE_SECS 5 | |
499 | /* Must be called under store_lock */ | |
500 | static void prune_initpid_store(void) | |
501 | { | |
502 | static long int last_prune = 0; | |
503 | struct pidns_init_store *e, *prev, *delme; | |
504 | long int now, threshold; | |
505 | int i; | |
506 | ||
507 | if (!last_prune) { | |
508 | last_prune = time(NULL); | |
509 | return; | |
510 | } | |
511 | now = time(NULL); | |
512 | if (now < last_prune + PURGE_SECS) | |
513 | return; | |
7dd6560a CB |
514 | |
515 | lxcfs_debug("%s\n", "Pruning."); | |
516 | ||
237e200e SH |
517 | last_prune = now; |
518 | threshold = now - 2 * PURGE_SECS; | |
519 | ||
520 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
521 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
522 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
523 | |
524 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
525 | ||
237e200e SH |
526 | delme = e; |
527 | if (prev) | |
528 | prev->next = e->next; | |
529 | else | |
530 | pidns_hash_table[i] = e->next; | |
531 | e = e->next; | |
532 | free(delme); | |
533 | } else { | |
534 | prev = e; | |
535 | e = e->next; | |
536 | } | |
537 | } | |
538 | } | |
539 | } | |
540 | ||
541 | /* Must be called under store_lock */ | |
542 | static void save_initpid(struct stat *sb, pid_t pid) | |
543 | { | |
544 | struct pidns_init_store *e; | |
545 | char fpath[100]; | |
546 | struct stat procsb; | |
547 | int h; | |
548 | ||
7dd6560a CB |
549 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
550 | ||
237e200e SH |
551 | snprintf(fpath, 100, "/proc/%d", pid); |
552 | if (stat(fpath, &procsb) < 0) | |
553 | return; | |
554 | do { | |
555 | e = malloc(sizeof(*e)); | |
556 | } while (!e); | |
557 | e->ino = sb->st_ino; | |
558 | e->initpid = pid; | |
559 | e->ctime = procsb.st_ctime; | |
560 | h = HASH(e->ino); | |
561 | e->next = pidns_hash_table[h]; | |
562 | e->lastcheck = time(NULL); | |
563 | pidns_hash_table[h] = e; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
568 | * entry for the inode number and creation time. Verify that the init pid | |
569 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
570 | * otherwise. | |
571 | * Must be called under store_lock | |
572 | */ | |
573 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
574 | { | |
575 | int h = HASH(sb->st_ino); | |
576 | struct pidns_init_store *e = pidns_hash_table[h]; | |
577 | ||
578 | while (e) { | |
579 | if (e->ino == sb->st_ino) { | |
580 | if (initpid_still_valid(e, sb)) { | |
581 | e->lastcheck = time(NULL); | |
582 | return e; | |
583 | } | |
584 | remove_initpid(e); | |
585 | return NULL; | |
586 | } | |
587 | e = e->next; | |
588 | } | |
589 | ||
590 | return NULL; | |
591 | } | |
592 | ||
0f657ce3 | 593 | static int is_dir(const char *path, int fd) |
237e200e SH |
594 | { |
595 | struct stat statbuf; | |
0f657ce3 | 596 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
597 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
598 | return 1; | |
599 | return 0; | |
600 | } | |
601 | ||
602 | static char *must_copy_string(const char *str) | |
603 | { | |
604 | char *dup = NULL; | |
605 | if (!str) | |
606 | return NULL; | |
607 | do { | |
608 | dup = strdup(str); | |
609 | } while (!dup); | |
610 | ||
611 | return dup; | |
612 | } | |
613 | ||
614 | static inline void drop_trailing_newlines(char *s) | |
615 | { | |
616 | int l; | |
617 | ||
618 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
619 | s[l-1] = '\0'; | |
620 | } | |
621 | ||
622 | #define BATCH_SIZE 50 | |
623 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
624 | { | |
625 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
626 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
627 | ||
628 | if (!*mem || newbatches > oldbatches) { | |
629 | char *tmp; | |
630 | do { | |
631 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
632 | } while (!tmp); | |
633 | *mem = tmp; | |
634 | } | |
635 | } | |
636 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
637 | { | |
638 | size_t newlen = *len + linelen; | |
639 | dorealloc(contents, *len, newlen + 1); | |
640 | memcpy(*contents + *len, line, linelen+1); | |
641 | *len = newlen; | |
642 | } | |
643 | ||
60f2ae53 | 644 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
645 | { |
646 | char *line = NULL; | |
647 | char *contents = NULL; | |
60f2ae53 | 648 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
649 | size_t len = 0, fulllen = 0; |
650 | ssize_t linelen; | |
651 | ||
652 | if (!f) | |
653 | return NULL; | |
654 | ||
655 | while ((linelen = getline(&line, &len, f)) != -1) { | |
656 | append_line(&contents, &fulllen, line, linelen); | |
657 | } | |
658 | fclose(f); | |
659 | ||
660 | if (contents) | |
661 | drop_trailing_newlines(contents); | |
662 | free(line); | |
663 | return contents; | |
664 | } | |
665 | ||
c9236032 HY |
666 | static int preserve_ns(const int pid, const char *ns) |
667 | { | |
668 | int ret; | |
669 | /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */ | |
670 | #define __NS_PATH_LEN 50 | |
671 | char path[__NS_PATH_LEN]; | |
672 | ||
673 | /* This way we can use this function to also check whether namespaces | |
674 | * are supported by the kernel by passing in the NULL or the empty | |
675 | * string. | |
676 | */ | |
677 | ret = snprintf(path, __NS_PATH_LEN, "/proc/%d/ns%s%s", pid, | |
678 | !ns || strcmp(ns, "") == 0 ? "" : "/", | |
679 | !ns || strcmp(ns, "") == 0 ? "" : ns); | |
680 | if (ret < 0 || (size_t)ret >= __NS_PATH_LEN) { | |
681 | errno = EFBIG; | |
682 | return -1; | |
683 | } | |
684 | ||
685 | return open(path, O_RDONLY | O_CLOEXEC); | |
686 | } | |
687 | ||
688 | /** | |
689 | * in_same_namespace - Check whether two processes are in the same namespace. | |
690 | * @pid1 - PID of the first process. | |
691 | * @pid2 - PID of the second process. | |
692 | * @ns - Name of the namespace to check. Must correspond to one of the names | |
693 | * for the namespaces as shown in /proc/<pid/ns/ | |
694 | * | |
695 | * If the two processes are not in the same namespace returns an fd to the | |
696 | * namespace of the second process identified by @pid2. If the two processes are | |
697 | * in the same namespace returns -EINVAL, -1 if an error occurred. | |
698 | */ | |
699 | static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns) | |
700 | { | |
701 | __do_close_prot_errno int ns_fd1 = -1, ns_fd2 = -1; | |
702 | int ret = -1; | |
703 | struct stat ns_st1, ns_st2; | |
704 | ||
705 | ns_fd1 = preserve_ns(pid1, ns); | |
706 | if (ns_fd1 < 0) { | |
707 | /* The kernel does not support this namespace. This is not an | |
708 | * error. | |
709 | */ | |
710 | if (errno == ENOENT) | |
711 | return -EINVAL; | |
712 | ||
713 | return -1; | |
714 | } | |
715 | ||
716 | ns_fd2 = preserve_ns(pid2, ns); | |
717 | if (ns_fd2 < 0) | |
718 | return -1; | |
719 | ||
720 | ret = fstat(ns_fd1, &ns_st1); | |
721 | if (ret < 0) | |
722 | return -1; | |
723 | ||
724 | ret = fstat(ns_fd2, &ns_st2); | |
725 | if (ret < 0) | |
726 | return -1; | |
727 | ||
728 | /* processes are in the same namespace */ | |
729 | if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino)) | |
730 | return -EINVAL; | |
731 | ||
732 | /* processes are in different namespaces */ | |
733 | return move_fd(ns_fd2); | |
734 | } | |
735 | ||
736 | static bool is_shared_pidns(pid_t pid) | |
737 | { | |
738 | if (pid != 1) | |
739 | return false; | |
740 | ||
6e3637bb | 741 | if (in_same_namespace(pid, getpid(), "pid") == -EINVAL) |
c9236032 HY |
742 | return true; |
743 | ||
744 | return false; | |
745 | } | |
746 | ||
ba59ea09 | 747 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
748 | { |
749 | FILE *f; | |
750 | size_t len, ret; | |
751 | ||
beb5024e CB |
752 | f = fdopen(fd, "w"); |
753 | if (!f) | |
237e200e | 754 | return false; |
beb5024e | 755 | |
237e200e SH |
756 | len = strlen(string); |
757 | ret = fwrite(string, 1, len, f); | |
758 | if (ret != len) { | |
beb5024e CB |
759 | lxcfs_error("%s - Error writing \"%s\" to \"%s\"\n", |
760 | strerror(errno), string, fnam); | |
237e200e SH |
761 | fclose(f); |
762 | return false; | |
763 | } | |
beb5024e | 764 | |
237e200e | 765 | if (fclose(f) < 0) { |
beb5024e | 766 | lxcfs_error("%s - Failed to close \"%s\"\n", strerror(errno), fnam); |
237e200e SH |
767 | return false; |
768 | } | |
beb5024e | 769 | |
237e200e SH |
770 | return true; |
771 | } | |
772 | ||
237e200e SH |
773 | struct cgfs_files { |
774 | char *name; | |
775 | uint32_t uid, gid; | |
776 | uint32_t mode; | |
777 | }; | |
778 | ||
0619767c | 779 | #define ALLOC_NUM 20 |
237e200e SH |
780 | static bool store_hierarchy(char *stridx, char *h) |
781 | { | |
0619767c SH |
782 | if (num_hierarchies % ALLOC_NUM == 0) { |
783 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
784 | n *= ALLOC_NUM; | |
785 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c | 786 | if (!tmp) { |
b8defc3d | 787 | lxcfs_error("%s\n", strerror(errno)); |
0619767c SH |
788 | exit(1); |
789 | } | |
237e200e | 790 | hierarchies = tmp; |
237e200e | 791 | } |
f676eb79 | 792 | |
0619767c | 793 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
794 | return true; |
795 | } | |
796 | ||
797 | static void print_subsystems(void) | |
798 | { | |
799 | int i; | |
800 | ||
a257a8ee | 801 | fprintf(stderr, "mount namespace: %d\n", cgroup_mount_ns_fd); |
cc97d34c | 802 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
803 | for (i = 0; i < num_hierarchies; i++) { |
804 | if (hierarchies[i]) | |
b8defc3d CB |
805 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, |
806 | fd_hierarchies[i], hierarchies[i]); | |
237e200e SH |
807 | } |
808 | } | |
809 | ||
810 | static bool in_comma_list(const char *needle, const char *haystack) | |
811 | { | |
812 | const char *s = haystack, *e; | |
813 | size_t nlen = strlen(needle); | |
814 | ||
06081b29 | 815 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
816 | if (nlen != e - s) { |
817 | s = e + 1; | |
818 | continue; | |
819 | } | |
820 | if (strncmp(needle, s, nlen) == 0) | |
821 | return true; | |
822 | s = e + 1; | |
823 | } | |
824 | if (strcmp(needle, s) == 0) | |
825 | return true; | |
826 | return false; | |
827 | } | |
828 | ||
829 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
830 | /* Return the mounted controller and store the corresponding open file descriptor |
831 | * referring to the controller mountpoint in the private lxcfs namespace in | |
832 | * @cfd. | |
833 | */ | |
834 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
835 | { |
836 | int i; | |
837 | ||
838 | for (i = 0; i < num_hierarchies; i++) { | |
839 | if (!hierarchies[i]) | |
840 | continue; | |
5dd3e6fd CB |
841 | if (strcmp(hierarchies[i], controller) == 0) { |
842 | *cfd = fd_hierarchies[i]; | |
237e200e | 843 | return hierarchies[i]; |
5dd3e6fd CB |
844 | } |
845 | if (in_comma_list(controller, hierarchies[i])) { | |
846 | *cfd = fd_hierarchies[i]; | |
237e200e | 847 | return hierarchies[i]; |
5dd3e6fd | 848 | } |
237e200e SH |
849 | } |
850 | ||
851 | return NULL; | |
852 | } | |
853 | ||
854 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
855 | const char *value) | |
856 | { | |
ba59ea09 | 857 | int ret, fd, cfd; |
237e200e | 858 | size_t len; |
f5a6d92e | 859 | char *fnam, *tmpc; |
237e200e | 860 | |
f5a6d92e | 861 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
862 | if (!tmpc) |
863 | return false; | |
f5a6d92e CB |
864 | |
865 | /* Make sure we pass a relative path to *at() family of functions. | |
866 | * . + /cgroup + / + file + \0 | |
867 | */ | |
ba59ea09 | 868 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 869 | fnam = alloca(len); |
ba59ea09 CB |
870 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
871 | if (ret < 0 || (size_t)ret >= len) | |
872 | return false; | |
873 | ||
874 | fd = openat(cfd, fnam, O_WRONLY); | |
875 | if (fd < 0) | |
876 | return false; | |
f676eb79 | 877 | |
ba59ea09 | 878 | return write_string(fnam, value, fd); |
237e200e SH |
879 | } |
880 | ||
881 | // Chown all the files in the cgroup directory. We do this when we create | |
882 | // a cgroup on behalf of a user. | |
f23fe717 | 883 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 884 | { |
f23fe717 | 885 | struct dirent *direntp; |
237e200e SH |
886 | char path[MAXPATHLEN]; |
887 | size_t len; | |
888 | DIR *d; | |
f23fe717 | 889 | int fd1, ret; |
237e200e SH |
890 | |
891 | len = strlen(dirname); | |
892 | if (len >= MAXPATHLEN) { | |
b8defc3d | 893 | lxcfs_error("Pathname too long: %s\n", dirname); |
237e200e SH |
894 | return; |
895 | } | |
896 | ||
f23fe717 CB |
897 | fd1 = openat(fd, dirname, O_DIRECTORY); |
898 | if (fd1 < 0) | |
899 | return; | |
900 | ||
901 | d = fdopendir(fd1); | |
237e200e | 902 | if (!d) { |
b8defc3d | 903 | lxcfs_error("Failed to open %s\n", dirname); |
237e200e SH |
904 | return; |
905 | } | |
906 | ||
f23fe717 | 907 | while ((direntp = readdir(d))) { |
237e200e SH |
908 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
909 | continue; | |
910 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
911 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 912 | lxcfs_error("Pathname too long under %s\n", dirname); |
237e200e SH |
913 | continue; |
914 | } | |
f23fe717 | 915 | if (fchownat(fd, path, uid, gid, 0) < 0) |
b8defc3d | 916 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); |
237e200e SH |
917 | } |
918 | closedir(d); | |
919 | } | |
920 | ||
921 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
922 | { | |
5dd3e6fd | 923 | int cfd; |
237e200e | 924 | size_t len; |
f5a6d92e | 925 | char *dirnam, *tmpc; |
237e200e | 926 | |
f5a6d92e | 927 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
928 | if (!tmpc) |
929 | return -EINVAL; | |
f5a6d92e CB |
930 | |
931 | /* Make sure we pass a relative path to *at() family of functions. | |
932 | * . + /cg + \0 | |
933 | */ | |
f23fe717 | 934 | len = strlen(cg) + 2; |
237e200e | 935 | dirnam = alloca(len); |
f23fe717 | 936 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 937 | |
f23fe717 | 938 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
939 | return -errno; |
940 | ||
941 | if (uid == 0 && gid == 0) | |
942 | return 0; | |
943 | ||
f23fe717 | 944 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
945 | return -errno; |
946 | ||
f23fe717 | 947 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
948 | |
949 | return 0; | |
950 | } | |
951 | ||
7213ec5c | 952 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) |
237e200e | 953 | { |
b7672ded | 954 | struct dirent *direntp; |
237e200e SH |
955 | DIR *dir; |
956 | bool ret = false; | |
957 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
958 | int dupfd; |
959 | ||
960 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
961 | if (dupfd < 0) | |
962 | return false; | |
237e200e | 963 | |
b7672ded | 964 | dir = fdopendir(dupfd); |
237e200e | 965 | if (!dir) { |
7dd6560a | 966 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
7213ec5c | 967 | close(dupfd); |
237e200e SH |
968 | return false; |
969 | } | |
970 | ||
b7672ded | 971 | while ((direntp = readdir(dir))) { |
237e200e SH |
972 | struct stat mystat; |
973 | int rc; | |
974 | ||
237e200e SH |
975 | if (!strcmp(direntp->d_name, ".") || |
976 | !strcmp(direntp->d_name, "..")) | |
977 | continue; | |
978 | ||
979 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
980 | if (rc < 0 || rc >= MAXPATHLEN) { | |
b8defc3d | 981 | lxcfs_error("%s\n", "Pathname too long."); |
237e200e SH |
982 | continue; |
983 | } | |
984 | ||
2e81a5e3 CB |
985 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
986 | if (rc) { | |
7dd6560a | 987 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
988 | continue; |
989 | } | |
7dd6560a | 990 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 991 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 992 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
993 | } |
994 | ||
995 | ret = true; | |
996 | if (closedir(dir) < 0) { | |
b8defc3d | 997 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); |
237e200e SH |
998 | ret = false; |
999 | } | |
1000 | ||
2e81a5e3 | 1001 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 1002 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
1003 | ret = false; |
1004 | } | |
7213ec5c CB |
1005 | |
1006 | close(dupfd); | |
237e200e SH |
1007 | |
1008 | return ret; | |
1009 | } | |
1010 | ||
1011 | bool cgfs_remove(const char *controller, const char *cg) | |
1012 | { | |
b7672ded | 1013 | int fd, cfd; |
237e200e | 1014 | size_t len; |
f5a6d92e | 1015 | char *dirnam, *tmpc; |
7213ec5c | 1016 | bool bret; |
237e200e | 1017 | |
f5a6d92e | 1018 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1019 | if (!tmpc) |
1020 | return false; | |
f5a6d92e CB |
1021 | |
1022 | /* Make sure we pass a relative path to *at() family of functions. | |
1023 | * . + /cg + \0 | |
1024 | */ | |
b7672ded | 1025 | len = strlen(cg) + 2; |
237e200e | 1026 | dirnam = alloca(len); |
b7672ded CB |
1027 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
1028 | ||
1029 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
1030 | if (fd < 0) | |
1031 | return false; | |
1032 | ||
7213ec5c CB |
1033 | bret = recursive_rmdir(dirnam, fd, cfd); |
1034 | close(fd); | |
1035 | return bret; | |
237e200e SH |
1036 | } |
1037 | ||
1038 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
1039 | { | |
5dd3e6fd | 1040 | int cfd; |
237e200e | 1041 | size_t len; |
f5a6d92e | 1042 | char *pathname, *tmpc; |
237e200e | 1043 | |
f5a6d92e | 1044 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1045 | if (!tmpc) |
1046 | return false; | |
f5a6d92e CB |
1047 | |
1048 | /* Make sure we pass a relative path to *at() family of functions. | |
1049 | * . + /file + \0 | |
1050 | */ | |
534690b4 | 1051 | len = strlen(file) + 2; |
237e200e | 1052 | pathname = alloca(len); |
534690b4 CB |
1053 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
1054 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
1055 | return false; |
1056 | return true; | |
1057 | } | |
1058 | ||
0f657ce3 | 1059 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
1060 | { |
1061 | size_t len; | |
1062 | char *fname; | |
1063 | ||
1064 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
1065 | fname = alloca(len); | |
1066 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 1067 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
1068 | return -errno; |
1069 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 1070 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
1071 | return -errno; |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
1076 | { | |
5dd3e6fd | 1077 | int cfd; |
237e200e | 1078 | size_t len; |
f5a6d92e | 1079 | char *pathname, *tmpc; |
237e200e | 1080 | |
f5a6d92e | 1081 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1082 | if (!tmpc) |
1083 | return -EINVAL; | |
f5a6d92e CB |
1084 | |
1085 | /* Make sure we pass a relative path to *at() family of functions. | |
1086 | * . + /file + \0 | |
1087 | */ | |
0f657ce3 | 1088 | len = strlen(file) + 2; |
237e200e | 1089 | pathname = alloca(len); |
0f657ce3 CB |
1090 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
1091 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
1092 | return -errno; |
1093 | ||
0f657ce3 | 1094 | if (is_dir(pathname, cfd)) |
237e200e | 1095 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 1096 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
1097 | |
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
1102 | { | |
3ffd08ee | 1103 | int fd, cfd; |
237e200e | 1104 | size_t len; |
f5a6d92e | 1105 | char *pathname, *tmpc; |
237e200e | 1106 | |
f5a6d92e | 1107 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1108 | if (!tmpc) |
1109 | return NULL; | |
f5a6d92e CB |
1110 | |
1111 | /* Make sure we pass a relative path to *at() family of functions. | |
1112 | * . + /cgroup + / "cgroup.procs" + \0 | |
1113 | */ | |
3ffd08ee | 1114 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 1115 | pathname = alloca(len); |
3ffd08ee CB |
1116 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
1117 | ||
1118 | fd = openat(cfd, pathname, O_WRONLY); | |
1119 | if (fd < 0) | |
1120 | return NULL; | |
1121 | ||
1122 | return fdopen(fd, "w"); | |
237e200e SH |
1123 | } |
1124 | ||
f366da65 WB |
1125 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
1126 | void ***list, size_t typesize, | |
1127 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 1128 | { |
4ea38a4c | 1129 | int cfd, fd, ret; |
237e200e | 1130 | size_t len; |
4ea38a4c | 1131 | char *cg, *tmpc; |
237e200e | 1132 | char pathname[MAXPATHLEN]; |
f366da65 | 1133 | size_t sz = 0, asz = 0; |
4ea38a4c | 1134 | struct dirent *dirent; |
237e200e | 1135 | DIR *dir; |
237e200e | 1136 | |
4ea38a4c | 1137 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 1138 | *list = NULL; |
237e200e | 1139 | if (!tmpc) |
e97c834b | 1140 | return false; |
237e200e | 1141 | |
f5a6d92e | 1142 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
1143 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
1144 | cg = alloca(len); | |
1145 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
1146 | if (ret < 0 || (size_t)ret >= len) { | |
b8defc3d | 1147 | lxcfs_error("Pathname too long under %s\n", cgroup); |
4ea38a4c CB |
1148 | return false; |
1149 | } | |
237e200e | 1150 | |
4ea38a4c CB |
1151 | fd = openat(cfd, cg, O_DIRECTORY); |
1152 | if (fd < 0) | |
1153 | return false; | |
1154 | ||
1155 | dir = fdopendir(fd); | |
237e200e SH |
1156 | if (!dir) |
1157 | return false; | |
1158 | ||
4ea38a4c | 1159 | while ((dirent = readdir(dir))) { |
237e200e | 1160 | struct stat mystat; |
237e200e | 1161 | |
4ea38a4c CB |
1162 | if (!strcmp(dirent->d_name, ".") || |
1163 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
1164 | continue; |
1165 | ||
4ea38a4c CB |
1166 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
1167 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 1168 | lxcfs_error("Pathname too long under %s\n", cg); |
237e200e SH |
1169 | continue; |
1170 | } | |
1171 | ||
4ea38a4c | 1172 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e | 1173 | if (ret) { |
b8defc3d | 1174 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); |
237e200e SH |
1175 | continue; |
1176 | } | |
f366da65 WB |
1177 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
1178 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
1179 | continue; |
1180 | ||
1181 | if (sz+2 >= asz) { | |
f366da65 | 1182 | void **tmp; |
237e200e SH |
1183 | asz += BATCH_SIZE; |
1184 | do { | |
f366da65 | 1185 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
1186 | } while (!tmp); |
1187 | *list = tmp; | |
1188 | } | |
4ea38a4c | 1189 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
1190 | (*list)[sz+1] = NULL; |
1191 | sz++; | |
1192 | } | |
1193 | if (closedir(dir) < 0) { | |
b8defc3d | 1194 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); |
237e200e SH |
1195 | return false; |
1196 | } | |
1197 | return true; | |
1198 | } | |
1199 | ||
f366da65 WB |
1200 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
1201 | { | |
1202 | char *dup; | |
1203 | do { | |
1204 | dup = strdup(dir_entry); | |
1205 | } while (!dup); | |
1206 | return dup; | |
1207 | } | |
1208 | ||
1209 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
1210 | { | |
1211 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
1212 | } | |
1213 | ||
237e200e SH |
1214 | void free_key(struct cgfs_files *k) |
1215 | { | |
1216 | if (!k) | |
1217 | return; | |
1218 | free(k->name); | |
1219 | free(k); | |
1220 | } | |
1221 | ||
1222 | void free_keys(struct cgfs_files **keys) | |
1223 | { | |
1224 | int i; | |
1225 | ||
1226 | if (!keys) | |
1227 | return; | |
1228 | for (i = 0; keys[i]; i++) { | |
1229 | free_key(keys[i]); | |
1230 | } | |
1231 | free(keys); | |
1232 | } | |
1233 | ||
1234 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
1235 | { | |
60f2ae53 | 1236 | int ret, fd, cfd; |
237e200e | 1237 | size_t len; |
f5a6d92e | 1238 | char *fnam, *tmpc; |
237e200e | 1239 | |
f5a6d92e | 1240 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1241 | if (!tmpc) |
1242 | return false; | |
f5a6d92e CB |
1243 | |
1244 | /* Make sure we pass a relative path to *at() family of functions. | |
1245 | * . + /cgroup + / + file + \0 | |
1246 | */ | |
60f2ae53 | 1247 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 1248 | fnam = alloca(len); |
60f2ae53 CB |
1249 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
1250 | if (ret < 0 || (size_t)ret >= len) | |
234a820c | 1251 | return false; |
60f2ae53 CB |
1252 | |
1253 | fd = openat(cfd, fnam, O_RDONLY); | |
1254 | if (fd < 0) | |
234a820c | 1255 | return false; |
237e200e | 1256 | |
60f2ae53 | 1257 | *value = slurp_file(fnam, fd); |
237e200e SH |
1258 | return *value != NULL; |
1259 | } | |
1260 | ||
951acc94 JS |
1261 | bool cgfs_param_exist(const char *controller, const char *cgroup, const char *file) |
1262 | { | |
1263 | int ret, cfd; | |
1264 | size_t len; | |
1265 | char *fnam, *tmpc; | |
1266 | ||
1267 | tmpc = find_mounted_controller(controller, &cfd); | |
1268 | if (!tmpc) | |
1269 | return false; | |
1270 | ||
1271 | /* Make sure we pass a relative path to *at() family of functions. | |
1272 | * . + /cgroup + / + file + \0 | |
1273 | */ | |
1274 | len = strlen(cgroup) + strlen(file) + 3; | |
1275 | fnam = alloca(len); | |
1276 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); | |
1277 | if (ret < 0 || (size_t)ret >= len) | |
1278 | return false; | |
1279 | ||
1280 | return (faccessat(cfd, fnam, F_OK, 0) == 0); | |
1281 | } | |
1282 | ||
237e200e SH |
1283 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) |
1284 | { | |
4ea38a4c | 1285 | int ret, cfd; |
237e200e | 1286 | size_t len; |
f5a6d92e | 1287 | char *fnam, *tmpc; |
237e200e SH |
1288 | struct stat sb; |
1289 | struct cgfs_files *newkey; | |
237e200e | 1290 | |
f5a6d92e | 1291 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1292 | if (!tmpc) |
1293 | return false; | |
1294 | ||
1295 | if (file && *file == '/') | |
1296 | file++; | |
1297 | ||
06081b29 | 1298 | if (file && strchr(file, '/')) |
237e200e SH |
1299 | return NULL; |
1300 | ||
f5a6d92e CB |
1301 | /* Make sure we pass a relative path to *at() family of functions. |
1302 | * . + /cgroup + / + file + \0 | |
1303 | */ | |
4ea38a4c | 1304 | len = strlen(cgroup) + 3; |
237e200e SH |
1305 | if (file) |
1306 | len += strlen(file) + 1; | |
1307 | fnam = alloca(len); | |
4ea38a4c CB |
1308 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
1309 | file ? "/" : "", file ? file : ""); | |
237e200e | 1310 | |
4ea38a4c | 1311 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1312 | if (ret < 0) |
1313 | return NULL; | |
1314 | ||
1315 | do { | |
1316 | newkey = malloc(sizeof(struct cgfs_files)); | |
1317 | } while (!newkey); | |
1318 | if (file) | |
1319 | newkey->name = must_copy_string(file); | |
06081b29 CB |
1320 | else if (strrchr(cgroup, '/')) |
1321 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
1322 | else |
1323 | newkey->name = must_copy_string(cgroup); | |
1324 | newkey->uid = sb.st_uid; | |
1325 | newkey->gid = sb.st_gid; | |
1326 | newkey->mode = sb.st_mode; | |
1327 | ||
1328 | return newkey; | |
1329 | } | |
1330 | ||
f366da65 | 1331 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 1332 | { |
f366da65 WB |
1333 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
1334 | if (!entry) { | |
b8defc3d CB |
1335 | lxcfs_error("Error getting files under %s:%s\n", controller, |
1336 | cgroup); | |
237e200e | 1337 | } |
f366da65 WB |
1338 | return entry; |
1339 | } | |
1340 | ||
1341 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
1342 | { | |
1343 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
1344 | } |
1345 | ||
1346 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
1347 | { |
1348 | int cfd; | |
1349 | size_t len; | |
f5a6d92e | 1350 | char *fnam, *tmpc; |
237e200e SH |
1351 | int ret; |
1352 | struct stat sb; | |
1353 | ||
f5a6d92e | 1354 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1355 | if (!tmpc) |
1356 | return false; | |
f5a6d92e CB |
1357 | |
1358 | /* Make sure we pass a relative path to *at() family of functions. | |
1359 | * . + /cgroup + / + f + \0 | |
1360 | */ | |
d04232f2 | 1361 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 1362 | fnam = alloca(len); |
d04232f2 CB |
1363 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
1364 | if (ret < 0 || (size_t)ret >= len) | |
1365 | return false; | |
237e200e | 1366 | |
d04232f2 | 1367 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1368 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
1369 | return false; | |
f5a6d92e | 1370 | |
237e200e SH |
1371 | return true; |
1372 | } | |
1373 | ||
1374 | #define SEND_CREDS_OK 0 | |
1375 | #define SEND_CREDS_NOTSK 1 | |
1376 | #define SEND_CREDS_FAIL 2 | |
1377 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
1378 | static int wait_for_pid(pid_t pid); | |
1379 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 1380 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
1381 | |
1382 | /* | |
b10bdd6c | 1383 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
1384 | * over a unix sock so we can read the task's reaper's pid in our |
1385 | * namespace | |
b10bdd6c FG |
1386 | * |
1387 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
1388 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
1389 | * the pidns and the parent pid outside are identical. Using clone prevents | |
1390 | * this issue. | |
237e200e SH |
1391 | */ |
1392 | static void write_task_init_pid_exit(int sock, pid_t target) | |
1393 | { | |
237e200e SH |
1394 | char fnam[100]; |
1395 | pid_t pid; | |
237e200e | 1396 | int fd, ret; |
b10bdd6c FG |
1397 | size_t stack_size = sysconf(_SC_PAGESIZE); |
1398 | void *stack = alloca(stack_size); | |
237e200e SH |
1399 | |
1400 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1401 | if (ret < 0 || ret >= sizeof(fnam)) | |
1402 | _exit(1); | |
1403 | ||
1404 | fd = open(fnam, O_RDONLY); | |
1405 | if (fd < 0) { | |
1406 | perror("write_task_init_pid_exit open of ns/pid"); | |
1407 | _exit(1); | |
1408 | } | |
1409 | if (setns(fd, 0)) { | |
1410 | perror("write_task_init_pid_exit setns 1"); | |
1411 | close(fd); | |
1412 | _exit(1); | |
1413 | } | |
b10bdd6c | 1414 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1415 | if (pid < 0) |
1416 | _exit(1); | |
1417 | if (pid != 0) { | |
1418 | if (!wait_for_pid(pid)) | |
1419 | _exit(1); | |
1420 | _exit(0); | |
1421 | } | |
b10bdd6c FG |
1422 | } |
1423 | ||
1424 | static int send_creds_clone_wrapper(void *arg) { | |
1425 | struct ucred cred; | |
1426 | char v; | |
1427 | int sock = *(int *)arg; | |
237e200e SH |
1428 | |
1429 | /* we are the child */ | |
1430 | cred.uid = 0; | |
1431 | cred.gid = 0; | |
1432 | cred.pid = 1; | |
1433 | v = '1'; | |
1434 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1435 | return 1; |
1436 | return 0; | |
237e200e SH |
1437 | } |
1438 | ||
1439 | static pid_t get_init_pid_for_task(pid_t task) | |
1440 | { | |
1441 | int sock[2]; | |
1442 | pid_t pid; | |
1443 | pid_t ret = -1; | |
1444 | char v = '0'; | |
1445 | struct ucred cred; | |
1446 | ||
1447 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1448 | perror("socketpair"); | |
1449 | return -1; | |
1450 | } | |
1451 | ||
1452 | pid = fork(); | |
1453 | if (pid < 0) | |
1454 | goto out; | |
1455 | if (!pid) { | |
1456 | close(sock[1]); | |
1457 | write_task_init_pid_exit(sock[0], task); | |
1458 | _exit(0); | |
1459 | } | |
1460 | ||
1461 | if (!recv_creds(sock[1], &cred, &v)) | |
1462 | goto out; | |
1463 | ret = cred.pid; | |
1464 | ||
1465 | out: | |
1466 | close(sock[0]); | |
1467 | close(sock[1]); | |
1468 | if (pid > 0) | |
1469 | wait_for_pid(pid); | |
1470 | return ret; | |
1471 | } | |
1472 | ||
71f17cd2 | 1473 | pid_t lookup_initpid_in_store(pid_t qpid) |
237e200e SH |
1474 | { |
1475 | pid_t answer = 0; | |
1476 | struct stat sb; | |
1477 | struct pidns_init_store *e; | |
1478 | char fnam[100]; | |
1479 | ||
1480 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1481 | store_lock(); | |
1482 | if (stat(fnam, &sb) < 0) | |
1483 | goto out; | |
1484 | e = lookup_verify_initpid(&sb); | |
1485 | if (e) { | |
1486 | answer = e->initpid; | |
1487 | goto out; | |
1488 | } | |
1489 | answer = get_init_pid_for_task(qpid); | |
1490 | if (answer > 0) | |
1491 | save_initpid(&sb, answer); | |
1492 | ||
1493 | out: | |
1494 | /* we prune at end in case we are returning | |
1495 | * the value we were about to return */ | |
1496 | prune_initpid_store(); | |
1497 | store_unlock(); | |
1498 | return answer; | |
1499 | } | |
1500 | ||
1501 | static int wait_for_pid(pid_t pid) | |
1502 | { | |
1503 | int status, ret; | |
1504 | ||
1505 | if (pid <= 0) | |
1506 | return -1; | |
1507 | ||
1508 | again: | |
1509 | ret = waitpid(pid, &status, 0); | |
1510 | if (ret == -1) { | |
1511 | if (errno == EINTR) | |
1512 | goto again; | |
1513 | return -1; | |
1514 | } | |
1515 | if (ret != pid) | |
1516 | goto again; | |
1517 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1518 | return -1; | |
1519 | return 0; | |
1520 | } | |
1521 | ||
237e200e | 1522 | /* |
db1b32f6 SX |
1523 | * append the given formatted string to *src. |
1524 | * src: a pointer to a char* in which to append the formatted string. | |
237e200e SH |
1525 | * sz: the number of characters printed so far, minus trailing \0. |
1526 | * asz: the allocated size so far | |
db1b32f6 SX |
1527 | * format: string format. See printf for details. |
1528 | * ...: varargs. See printf for details. | |
237e200e | 1529 | */ |
db1b32f6 | 1530 | static void must_strcat(char **src, size_t *sz, size_t *asz, const char *format, ...) |
237e200e | 1531 | { |
db1b32f6 SX |
1532 | char tmp[BUF_RESERVE_SIZE]; |
1533 | va_list args; | |
237e200e | 1534 | |
db1b32f6 SX |
1535 | va_start (args, format); |
1536 | int tmplen = vsnprintf(tmp, BUF_RESERVE_SIZE, format, args); | |
1537 | va_end(args); | |
237e200e SH |
1538 | |
1539 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1540 | char *tmp; | |
1541 | do { | |
1542 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1543 | } while (!tmp); | |
1544 | *src = tmp; | |
1545 | *asz += BUF_RESERVE_SIZE; | |
1546 | } | |
bbfd0e33 | 1547 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1548 | *sz += tmplen; |
237e200e SH |
1549 | } |
1550 | ||
db1b32f6 SX |
1551 | /* |
1552 | * append pid to *src. | |
1553 | * src: a pointer to a char* in which ot append the pid. | |
1554 | * sz: the number of characters printed so far, minus trailing \0. | |
1555 | * asz: the allocated size so far | |
1556 | * pid: the pid to append | |
1557 | */ | |
1558 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1559 | { | |
1560 | must_strcat(src, sz, asz, "%d\n", (int)pid); | |
1561 | } | |
1562 | ||
237e200e SH |
1563 | /* |
1564 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1565 | * valid in the caller's namespace, return the id mapped into | |
1566 | * pid's namespace. | |
1567 | * Returns the mapped id, or -1 on error. | |
1568 | */ | |
1569 | unsigned int | |
1570 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1571 | { | |
1572 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1573 | hostuid, // base id for a range in the caller's namespace | |
1574 | count; // number of ids in this range | |
1575 | char line[400]; | |
1576 | int ret; | |
1577 | ||
1578 | fseek(idfile, 0L, SEEK_SET); | |
1579 | while (fgets(line, 400, idfile)) { | |
1580 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1581 | if (ret != 3) | |
1582 | continue; | |
1583 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1584 | /* | |
1585 | * uids wrapped around - unexpected as this is a procfile, | |
1586 | * so just bail. | |
1587 | */ | |
b8defc3d | 1588 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", |
237e200e SH |
1589 | nsuid, hostuid, count, line); |
1590 | return -1; | |
1591 | } | |
1592 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1593 | /* | |
1594 | * now since hostuid <= in_id < hostuid+count, and | |
1595 | * hostuid+count and nsuid+count do not wrap around, | |
1596 | * we know that nsuid+(in_id-hostuid) which must be | |
1597 | * less that nsuid+(count) must not wrap around | |
1598 | */ | |
1599 | return (in_id - hostuid) + nsuid; | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | // no answer found | |
1604 | return -1; | |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * for is_privileged_over, | |
1609 | * specify whether we require the calling uid to be root in his | |
1610 | * namespace | |
1611 | */ | |
1612 | #define NS_ROOT_REQD true | |
1613 | #define NS_ROOT_OPT false | |
1614 | ||
1615 | #define PROCLEN 100 | |
1616 | ||
1617 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1618 | { | |
1619 | char fpath[PROCLEN]; | |
1620 | int ret; | |
1621 | bool answer = false; | |
1622 | uid_t nsuid; | |
1623 | ||
1624 | if (victim == -1 || uid == -1) | |
1625 | return false; | |
1626 | ||
1627 | /* | |
1628 | * If the request is one not requiring root in the namespace, | |
1629 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1630 | * access to files owned by uid 1000 | |
1631 | */ | |
1632 | if (!req_ns_root && uid == victim) | |
1633 | return true; | |
1634 | ||
1635 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1636 | if (ret < 0 || ret >= PROCLEN) | |
1637 | return false; | |
1638 | FILE *f = fopen(fpath, "r"); | |
1639 | if (!f) | |
1640 | return false; | |
1641 | ||
1642 | /* if caller's not root in his namespace, reject */ | |
1643 | nsuid = convert_id_to_ns(f, uid); | |
1644 | if (nsuid) | |
1645 | goto out; | |
1646 | ||
1647 | /* | |
1648 | * If victim is not mapped into caller's ns, reject. | |
1649 | * XXX I'm not sure this check is needed given that fuse | |
1650 | * will be sending requests where the vfs has converted | |
1651 | */ | |
1652 | nsuid = convert_id_to_ns(f, victim); | |
1653 | if (nsuid == -1) | |
1654 | goto out; | |
1655 | ||
1656 | answer = true; | |
1657 | ||
1658 | out: | |
1659 | fclose(f); | |
1660 | return answer; | |
1661 | } | |
1662 | ||
1663 | static bool perms_include(int fmode, mode_t req_mode) | |
1664 | { | |
1665 | mode_t r; | |
1666 | ||
1667 | switch (req_mode & O_ACCMODE) { | |
1668 | case O_RDONLY: | |
1669 | r = S_IROTH; | |
1670 | break; | |
1671 | case O_WRONLY: | |
1672 | r = S_IWOTH; | |
1673 | break; | |
1674 | case O_RDWR: | |
1675 | r = S_IROTH | S_IWOTH; | |
1676 | break; | |
1677 | default: | |
1678 | return false; | |
1679 | } | |
1680 | return ((fmode & r) == r); | |
1681 | } | |
1682 | ||
1683 | ||
1684 | /* | |
1685 | * taskcg is a/b/c | |
1686 | * querycg is /a/b/c/d/e | |
1687 | * we return 'd' | |
1688 | */ | |
1689 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1690 | { | |
1691 | char *start, *end; | |
1692 | ||
1693 | if (strlen(taskcg) <= strlen(querycg)) { | |
b8defc3d | 1694 | lxcfs_error("%s\n", "I was fed bad input."); |
237e200e SH |
1695 | return NULL; |
1696 | } | |
1697 | ||
06081b29 | 1698 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1699 | start = strdup(taskcg + 1); |
1700 | else | |
1701 | start = strdup(taskcg + strlen(querycg) + 1); | |
1702 | if (!start) | |
1703 | return NULL; | |
1704 | end = strchr(start, '/'); | |
1705 | if (end) | |
1706 | *end = '\0'; | |
1707 | return start; | |
1708 | } | |
1709 | ||
1710 | static void stripnewline(char *x) | |
1711 | { | |
1712 | size_t l = strlen(x); | |
1713 | if (l && x[l-1] == '\n') | |
1714 | x[l-1] = '\0'; | |
1715 | } | |
1716 | ||
71f17cd2 | 1717 | char *get_pid_cgroup(pid_t pid, const char *contrl) |
237e200e | 1718 | { |
5dd3e6fd | 1719 | int cfd; |
237e200e SH |
1720 | char fnam[PROCLEN]; |
1721 | FILE *f; | |
1722 | char *answer = NULL; | |
1723 | char *line = NULL; | |
1724 | size_t len = 0; | |
1725 | int ret; | |
5dd3e6fd | 1726 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1727 | if (!h) |
1728 | return NULL; | |
1729 | ||
1730 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1731 | if (ret < 0 || ret >= PROCLEN) | |
1732 | return NULL; | |
1733 | if (!(f = fopen(fnam, "r"))) | |
1734 | return NULL; | |
1735 | ||
1736 | while (getline(&line, &len, f) != -1) { | |
1737 | char *c1, *c2; | |
1738 | if (!line[0]) | |
1739 | continue; | |
1740 | c1 = strchr(line, ':'); | |
1741 | if (!c1) | |
1742 | goto out; | |
1743 | c1++; | |
1744 | c2 = strchr(c1, ':'); | |
1745 | if (!c2) | |
1746 | goto out; | |
1747 | *c2 = '\0'; | |
1748 | if (strcmp(c1, h) != 0) | |
1749 | continue; | |
1750 | c2++; | |
1751 | stripnewline(c2); | |
1752 | do { | |
1753 | answer = strdup(c2); | |
1754 | } while (!answer); | |
1755 | break; | |
1756 | } | |
1757 | ||
1758 | out: | |
1759 | fclose(f); | |
1760 | free(line); | |
1761 | return answer; | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * check whether a fuse context may access a cgroup dir or file | |
1766 | * | |
1767 | * If file is not null, it is a cgroup file to check under cg. | |
1768 | * If file is null, then we are checking perms on cg itself. | |
1769 | * | |
1770 | * For files we can check the mode of the list_keys result. | |
1771 | * For cgroups, we must make assumptions based on the files under the | |
1772 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1773 | * yet. | |
1774 | */ | |
1775 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1776 | { | |
1777 | struct cgfs_files *k = NULL; | |
1778 | bool ret = false; | |
1779 | ||
1780 | k = cgfs_get_key(contrl, cg, file); | |
1781 | if (!k) | |
1782 | return false; | |
1783 | ||
1784 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1785 | if (perms_include(k->mode >> 6, mode)) { | |
1786 | ret = true; | |
1787 | goto out; | |
1788 | } | |
1789 | } | |
1790 | if (fc->gid == k->gid) { | |
1791 | if (perms_include(k->mode >> 3, mode)) { | |
1792 | ret = true; | |
1793 | goto out; | |
1794 | } | |
1795 | } | |
1796 | ret = perms_include(k->mode, mode); | |
1797 | ||
1798 | out: | |
1799 | free_key(k); | |
1800 | return ret; | |
1801 | } | |
1802 | ||
1803 | #define INITSCOPE "/init.scope" | |
71f17cd2 | 1804 | void prune_init_slice(char *cg) |
237e200e SH |
1805 | { |
1806 | char *point; | |
1807 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1808 | ||
1809 | if (cg_len < initscope_len) | |
1810 | return; | |
1811 | ||
1812 | point = cg + cg_len - initscope_len; | |
1813 | if (strcmp(point, INITSCOPE) == 0) { | |
1814 | if (point == cg) | |
1815 | *(point+1) = '\0'; | |
1816 | else | |
1817 | *point = '\0'; | |
1818 | } | |
1819 | } | |
1820 | ||
1821 | /* | |
1822 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1823 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1824 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1825 | * to a string containing the next cgroup directory under cg, which must be | |
1826 | * freed by the caller. | |
1827 | */ | |
1828 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1829 | { | |
1830 | bool answer = false; | |
1831 | char *c2 = get_pid_cgroup(pid, contrl); | |
1832 | char *linecmp; | |
1833 | ||
1834 | if (!c2) | |
1835 | return false; | |
1836 | prune_init_slice(c2); | |
1837 | ||
1838 | /* | |
12c31268 CB |
1839 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1840 | * they pass in a cgroup without leading '/' | |
1841 | * | |
1842 | * The original line here was: | |
1843 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1844 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1845 | * Serge, do you know? | |
237e200e | 1846 | */ |
12c31268 CB |
1847 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1848 | linecmp = c2; | |
1849 | else | |
1850 | linecmp = c2 + 1; | |
237e200e SH |
1851 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1852 | if (nextcg) { | |
1853 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1854 | } | |
1855 | goto out; | |
1856 | } | |
1857 | answer = true; | |
1858 | ||
1859 | out: | |
1860 | free(c2); | |
1861 | return answer; | |
1862 | } | |
1863 | ||
1864 | /* | |
1865 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1866 | */ | |
1867 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1868 | { | |
1869 | bool answer = false; | |
1870 | char *c2, *task_cg; | |
1871 | size_t target_len, task_len; | |
1872 | ||
f7bff426 | 1873 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1874 | return true; |
1875 | ||
1876 | c2 = get_pid_cgroup(pid, contrl); | |
1877 | if (!c2) | |
1878 | return false; | |
1879 | prune_init_slice(c2); | |
1880 | ||
1881 | task_cg = c2 + 1; | |
1882 | target_len = strlen(cg); | |
1883 | task_len = strlen(task_cg); | |
1884 | if (task_len == 0) { | |
1885 | /* Task is in the root cg, it can see everything. This case is | |
1886 | * not handled by the strmcps below, since they test for the | |
1887 | * last /, but that is the first / that we've chopped off | |
1888 | * above. | |
1889 | */ | |
1890 | answer = true; | |
1891 | goto out; | |
1892 | } | |
1893 | if (strcmp(cg, task_cg) == 0) { | |
1894 | answer = true; | |
1895 | goto out; | |
1896 | } | |
1897 | if (target_len < task_len) { | |
1898 | /* looking up a parent dir */ | |
1899 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1900 | answer = true; | |
1901 | goto out; | |
1902 | } | |
1903 | if (target_len > task_len) { | |
1904 | /* looking up a child dir */ | |
1905 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1906 | answer = true; | |
1907 | goto out; | |
1908 | } | |
1909 | ||
1910 | out: | |
1911 | free(c2); | |
1912 | return answer; | |
1913 | } | |
1914 | ||
1915 | /* | |
1916 | * given /cgroup/freezer/a/b, return "freezer". | |
1917 | * the returned char* should NOT be freed. | |
1918 | */ | |
1919 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1920 | { | |
1921 | const char *p1; | |
1922 | char *contr, *slash; | |
1923 | ||
99142521 | 1924 | if (strlen(path) < 9) { |
e254948f | 1925 | errno = EACCES; |
237e200e | 1926 | return NULL; |
99142521 CB |
1927 | } |
1928 | if (*(path + 7) != '/') { | |
1929 | errno = EINVAL; | |
237e200e | 1930 | return NULL; |
99142521 | 1931 | } |
3adc421c | 1932 | p1 = path + 8; |
237e200e | 1933 | contr = strdupa(p1); |
99142521 CB |
1934 | if (!contr) { |
1935 | errno = ENOMEM; | |
237e200e | 1936 | return NULL; |
99142521 | 1937 | } |
237e200e SH |
1938 | slash = strstr(contr, "/"); |
1939 | if (slash) | |
1940 | *slash = '\0'; | |
1941 | ||
1942 | int i; | |
3adc421c | 1943 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
1944 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) |
1945 | return hierarchies[i]; | |
1946 | } | |
99142521 | 1947 | errno = ENOENT; |
237e200e SH |
1948 | return NULL; |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1953 | * Note that the returned value may include files (keynames) etc | |
1954 | */ | |
1955 | static const char *find_cgroup_in_path(const char *path) | |
1956 | { | |
1957 | const char *p1; | |
1958 | ||
bc70ba9b | 1959 | if (strlen(path) < 9) { |
e254948f | 1960 | errno = EACCES; |
237e200e | 1961 | return NULL; |
bc70ba9b CB |
1962 | } |
1963 | p1 = strstr(path + 8, "/"); | |
1964 | if (!p1) { | |
1965 | errno = EINVAL; | |
237e200e | 1966 | return NULL; |
bc70ba9b CB |
1967 | } |
1968 | errno = 0; | |
1969 | return p1 + 1; | |
237e200e SH |
1970 | } |
1971 | ||
1972 | /* | |
1973 | * split the last path element from the path in @cg. | |
1974 | * @dir is newly allocated and should be freed, @last not | |
1975 | */ | |
1976 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1977 | { | |
1978 | char *p; | |
1979 | ||
1980 | do { | |
1981 | *dir = strdup(cg); | |
1982 | } while (!*dir); | |
1983 | *last = strrchr(cg, '/'); | |
1984 | if (!*last) { | |
1985 | *last = NULL; | |
1986 | return; | |
1987 | } | |
1988 | p = strrchr(*dir, '/'); | |
1989 | *p = '\0'; | |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * FUSE ops for /cgroup | |
1994 | */ | |
1995 | ||
1996 | int cg_getattr(const char *path, struct stat *sb) | |
1997 | { | |
1998 | struct timespec now; | |
1999 | struct fuse_context *fc = fuse_get_context(); | |
2000 | char * cgdir = NULL; | |
2001 | char *last = NULL, *path1, *path2; | |
2002 | struct cgfs_files *k = NULL; | |
2003 | const char *cgroup; | |
2004 | const char *controller = NULL; | |
2005 | int ret = -ENOENT; | |
2006 | ||
2007 | ||
2008 | if (!fc) | |
2009 | return -EIO; | |
2010 | ||
2011 | memset(sb, 0, sizeof(struct stat)); | |
2012 | ||
2013 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
2014 | return -EINVAL; | |
2015 | ||
2016 | sb->st_uid = sb->st_gid = 0; | |
2017 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
2018 | sb->st_size = 0; | |
2019 | ||
2020 | if (strcmp(path, "/cgroup") == 0) { | |
2021 | sb->st_mode = S_IFDIR | 00755; | |
2022 | sb->st_nlink = 2; | |
2023 | return 0; | |
2024 | } | |
2025 | ||
2026 | controller = pick_controller_from_path(fc, path); | |
2027 | if (!controller) | |
2f7036d0 | 2028 | return -errno; |
237e200e SH |
2029 | cgroup = find_cgroup_in_path(path); |
2030 | if (!cgroup) { | |
2031 | /* this is just /cgroup/controller, return it as a dir */ | |
2032 | sb->st_mode = S_IFDIR | 00755; | |
2033 | sb->st_nlink = 2; | |
2034 | return 0; | |
2035 | } | |
2036 | ||
2037 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2038 | ||
2039 | if (!last) { | |
2040 | path1 = "/"; | |
2041 | path2 = cgdir; | |
2042 | } else { | |
2043 | path1 = cgdir; | |
2044 | path2 = last; | |
2045 | } | |
2046 | ||
2047 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2048 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
2049 | initpid = fc->pid; |
2050 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
2051 | * Then check that caller's cgroup is under path if last is a child | |
2052 | * cgroup, or cgdir if last is a file */ | |
2053 | ||
2054 | if (is_child_cgroup(controller, path1, path2)) { | |
2055 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
2056 | ret = -ENOENT; | |
2057 | goto out; | |
2058 | } | |
2059 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
2060 | /* this is just /cgroup/controller, return it as a dir */ | |
2061 | sb->st_mode = S_IFDIR | 00555; | |
2062 | sb->st_nlink = 2; | |
2063 | ret = 0; | |
2064 | goto out; | |
2065 | } | |
2066 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
2067 | ret = -EACCES; | |
2068 | goto out; | |
2069 | } | |
2070 | ||
2071 | // get uid, gid, from '/tasks' file and make up a mode | |
2072 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2073 | sb->st_mode = S_IFDIR | 00755; | |
2074 | k = cgfs_get_key(controller, cgroup, NULL); | |
2075 | if (!k) { | |
2076 | sb->st_uid = sb->st_gid = 0; | |
2077 | } else { | |
2078 | sb->st_uid = k->uid; | |
2079 | sb->st_gid = k->gid; | |
2080 | } | |
2081 | free_key(k); | |
2082 | sb->st_nlink = 2; | |
2083 | ret = 0; | |
2084 | goto out; | |
2085 | } | |
2086 | ||
2087 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
2088 | sb->st_mode = S_IFREG | k->mode; | |
2089 | sb->st_nlink = 1; | |
2090 | sb->st_uid = k->uid; | |
2091 | sb->st_gid = k->gid; | |
2092 | sb->st_size = 0; | |
2093 | free_key(k); | |
2094 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
2095 | ret = -ENOENT; | |
2096 | goto out; | |
2097 | } | |
237e200e SH |
2098 | ret = 0; |
2099 | } | |
2100 | ||
2101 | out: | |
2102 | free(cgdir); | |
2103 | return ret; | |
2104 | } | |
2105 | ||
2106 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
2107 | { | |
2108 | struct fuse_context *fc = fuse_get_context(); | |
2109 | const char *cgroup; | |
2110 | struct file_info *dir_info; | |
2111 | char *controller = NULL; | |
2112 | ||
2113 | if (!fc) | |
2114 | return -EIO; | |
2115 | ||
2116 | if (strcmp(path, "/cgroup") == 0) { | |
2117 | cgroup = NULL; | |
2118 | controller = NULL; | |
2119 | } else { | |
2120 | // return list of keys for the controller, and list of child cgroups | |
2121 | controller = pick_controller_from_path(fc, path); | |
2122 | if (!controller) | |
2f7036d0 | 2123 | return -errno; |
237e200e SH |
2124 | |
2125 | cgroup = find_cgroup_in_path(path); | |
2126 | if (!cgroup) { | |
2127 | /* this is just /cgroup/controller, return its contents */ | |
2128 | cgroup = "/"; | |
2129 | } | |
2130 | } | |
2131 | ||
2132 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2133 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
2134 | initpid = fc->pid; |
2135 | if (cgroup) { | |
2136 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
2137 | return -ENOENT; | |
2138 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
2139 | return -EACCES; | |
2140 | } | |
2141 | ||
2142 | /* we'll free this at cg_releasedir */ | |
2143 | dir_info = malloc(sizeof(*dir_info)); | |
2144 | if (!dir_info) | |
2145 | return -ENOMEM; | |
2146 | dir_info->controller = must_copy_string(controller); | |
2147 | dir_info->cgroup = must_copy_string(cgroup); | |
2148 | dir_info->type = LXC_TYPE_CGDIR; | |
2149 | dir_info->buf = NULL; | |
2150 | dir_info->file = NULL; | |
2151 | dir_info->buflen = 0; | |
2152 | ||
2153 | fi->fh = (unsigned long)dir_info; | |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
2158 | struct fuse_file_info *fi) | |
2159 | { | |
2160 | struct file_info *d = (struct file_info *)fi->fh; | |
2161 | struct cgfs_files **list = NULL; | |
2162 | int i, ret; | |
2163 | char *nextcg = NULL; | |
2164 | struct fuse_context *fc = fuse_get_context(); | |
2165 | char **clist = NULL; | |
2166 | ||
d639f863 CB |
2167 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) |
2168 | return -EIO; | |
2169 | ||
237e200e | 2170 | if (d->type != LXC_TYPE_CGDIR) { |
b8defc3d | 2171 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); |
237e200e SH |
2172 | return -EIO; |
2173 | } | |
2174 | if (!d->cgroup && !d->controller) { | |
2175 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
2176 | int i; | |
2177 | ||
2178 | for (i = 0; i < num_hierarchies; i++) { | |
2179 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
2180 | return -EIO; | |
2181 | } | |
2182 | } | |
2183 | return 0; | |
2184 | } | |
2185 | ||
2186 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
2187 | // not a valid cgroup | |
2188 | ret = -EINVAL; | |
2189 | goto out; | |
2190 | } | |
2191 | ||
2192 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2193 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
2194 | initpid = fc->pid; |
2195 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
2196 | if (nextcg) { | |
2197 | ret = filler(buf, nextcg, NULL, 0); | |
2198 | free(nextcg); | |
2199 | if (ret != 0) { | |
2200 | ret = -EIO; | |
2201 | goto out; | |
2202 | } | |
2203 | } | |
2204 | ret = 0; | |
2205 | goto out; | |
2206 | } | |
2207 | ||
b737a54a | 2208 | for (i = 0; list && list[i]; i++) { |
237e200e SH |
2209 | if (filler(buf, list[i]->name, NULL, 0) != 0) { |
2210 | ret = -EIO; | |
2211 | goto out; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | // now get the list of child cgroups | |
2216 | ||
2217 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
2218 | ret = 0; | |
2219 | goto out; | |
2220 | } | |
f366da65 WB |
2221 | if (clist) { |
2222 | for (i = 0; clist[i]; i++) { | |
2223 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
2224 | ret = -EIO; | |
2225 | goto out; | |
2226 | } | |
237e200e SH |
2227 | } |
2228 | } | |
2229 | ret = 0; | |
2230 | ||
2231 | out: | |
2232 | free_keys(list); | |
2233 | if (clist) { | |
2234 | for (i = 0; clist[i]; i++) | |
2235 | free(clist[i]); | |
2236 | free(clist); | |
2237 | } | |
2238 | return ret; | |
2239 | } | |
2240 | ||
71f17cd2 | 2241 | void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 2242 | { |
43215927 SH |
2243 | struct file_info *f = (struct file_info *)fi->fh; |
2244 | ||
237e200e SH |
2245 | if (!f) |
2246 | return; | |
43215927 SH |
2247 | |
2248 | fi->fh = 0; | |
2249 | ||
237e200e | 2250 | free(f->controller); |
43215927 | 2251 | f->controller = NULL; |
237e200e | 2252 | free(f->cgroup); |
43215927 | 2253 | f->cgroup = NULL; |
237e200e | 2254 | free(f->file); |
43215927 | 2255 | f->file = NULL; |
237e200e | 2256 | free(f->buf); |
43215927 | 2257 | f->buf = NULL; |
237e200e | 2258 | free(f); |
bbb508dd | 2259 | f = NULL; |
237e200e SH |
2260 | } |
2261 | ||
2262 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
2263 | { | |
43215927 | 2264 | do_release_file_info(fi); |
237e200e SH |
2265 | return 0; |
2266 | } | |
2267 | ||
2268 | int cg_open(const char *path, struct fuse_file_info *fi) | |
2269 | { | |
2270 | const char *cgroup; | |
2271 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
2272 | struct cgfs_files *k = NULL; | |
2273 | struct file_info *file_info; | |
2274 | struct fuse_context *fc = fuse_get_context(); | |
2275 | int ret; | |
2276 | ||
2277 | if (!fc) | |
2278 | return -EIO; | |
2279 | ||
2280 | controller = pick_controller_from_path(fc, path); | |
2281 | if (!controller) | |
2f7036d0 | 2282 | return -errno; |
237e200e SH |
2283 | cgroup = find_cgroup_in_path(path); |
2284 | if (!cgroup) | |
bc70ba9b | 2285 | return -errno; |
237e200e SH |
2286 | |
2287 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2288 | if (!last) { | |
2289 | path1 = "/"; | |
2290 | path2 = cgdir; | |
2291 | } else { | |
2292 | path1 = cgdir; | |
2293 | path2 = last; | |
2294 | } | |
2295 | ||
2296 | k = cgfs_get_key(controller, path1, path2); | |
2297 | if (!k) { | |
2298 | ret = -EINVAL; | |
2299 | goto out; | |
2300 | } | |
2301 | free_key(k); | |
2302 | ||
2303 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2304 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
2305 | initpid = fc->pid; |
2306 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2307 | ret = -ENOENT; | |
2308 | goto out; | |
2309 | } | |
2310 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
2311 | ret = -EACCES; |
2312 | goto out; | |
2313 | } | |
2314 | ||
2315 | /* we'll free this at cg_release */ | |
2316 | file_info = malloc(sizeof(*file_info)); | |
2317 | if (!file_info) { | |
2318 | ret = -ENOMEM; | |
2319 | goto out; | |
2320 | } | |
2321 | file_info->controller = must_copy_string(controller); | |
2322 | file_info->cgroup = must_copy_string(path1); | |
2323 | file_info->file = must_copy_string(path2); | |
2324 | file_info->type = LXC_TYPE_CGFILE; | |
2325 | file_info->buf = NULL; | |
2326 | file_info->buflen = 0; | |
2327 | ||
2328 | fi->fh = (unsigned long)file_info; | |
2329 | ret = 0; | |
2330 | ||
2331 | out: | |
2332 | free(cgdir); | |
2333 | return ret; | |
2334 | } | |
2335 | ||
bddbb106 SH |
2336 | int cg_access(const char *path, int mode) |
2337 | { | |
6f0f6b83 | 2338 | int ret; |
bddbb106 | 2339 | const char *cgroup; |
6f0f6b83 CB |
2340 | char *path1, *path2, *controller; |
2341 | char *last = NULL, *cgdir = NULL; | |
bddbb106 SH |
2342 | struct cgfs_files *k = NULL; |
2343 | struct fuse_context *fc = fuse_get_context(); | |
6f0f6b83 | 2344 | |
9873c5e8 | 2345 | if (strcmp(path, "/cgroup") == 0) |
6f0f6b83 | 2346 | return 0; |
bddbb106 SH |
2347 | |
2348 | if (!fc) | |
2349 | return -EIO; | |
2350 | ||
2351 | controller = pick_controller_from_path(fc, path); | |
2352 | if (!controller) | |
2f7036d0 | 2353 | return -errno; |
bddbb106 | 2354 | cgroup = find_cgroup_in_path(path); |
575316c4 SH |
2355 | if (!cgroup) { |
2356 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
2357 | if ((mode & W_OK) == 0) |
2358 | return 0; | |
2359 | return -EACCES; | |
575316c4 | 2360 | } |
bddbb106 SH |
2361 | |
2362 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2363 | if (!last) { | |
2364 | path1 = "/"; | |
2365 | path2 = cgdir; | |
2366 | } else { | |
2367 | path1 = cgdir; | |
2368 | path2 = last; | |
2369 | } | |
2370 | ||
2371 | k = cgfs_get_key(controller, path1, path2); | |
2372 | if (!k) { | |
3f441bc7 SH |
2373 | if ((mode & W_OK) == 0) |
2374 | ret = 0; | |
2375 | else | |
2376 | ret = -EACCES; | |
bddbb106 SH |
2377 | goto out; |
2378 | } | |
2379 | free_key(k); | |
2380 | ||
2381 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 2382 | if (initpid <= 1 || is_shared_pidns(initpid)) |
bddbb106 SH |
2383 | initpid = fc->pid; |
2384 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2385 | ret = -ENOENT; | |
2386 | goto out; | |
2387 | } | |
2388 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
2389 | ret = -EACCES; | |
2390 | goto out; | |
2391 | } | |
2392 | ||
2393 | ret = 0; | |
2394 | ||
2395 | out: | |
2396 | free(cgdir); | |
2397 | return ret; | |
2398 | } | |
2399 | ||
237e200e SH |
2400 | int cg_release(const char *path, struct fuse_file_info *fi) |
2401 | { | |
43215927 | 2402 | do_release_file_info(fi); |
237e200e SH |
2403 | return 0; |
2404 | } | |
2405 | ||
2406 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
2407 | ||
2408 | static bool wait_for_sock(int sock, int timeout) | |
2409 | { | |
2410 | struct epoll_event ev; | |
2411 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
2412 | ||
2413 | if ((starttime = time(NULL)) < 0) | |
2414 | return false; | |
2415 | ||
2416 | if ((epfd = epoll_create(1)) < 0) { | |
b8defc3d | 2417 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); |
237e200e SH |
2418 | return false; |
2419 | } | |
2420 | ||
2421 | ev.events = POLLIN_SET; | |
2422 | ev.data.fd = sock; | |
2423 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
b8defc3d | 2424 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); |
237e200e SH |
2425 | close(epfd); |
2426 | return false; | |
2427 | } | |
2428 | ||
2429 | again: | |
2430 | if ((now = time(NULL)) < 0) { | |
2431 | close(epfd); | |
2432 | return false; | |
2433 | } | |
2434 | ||
2435 | deltatime = (starttime + timeout) - now; | |
2436 | if (deltatime < 0) { // timeout | |
2437 | errno = 0; | |
2438 | close(epfd); | |
2439 | return false; | |
2440 | } | |
2441 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2442 | if (ret < 0 && errno == EINTR) | |
2443 | goto again; | |
2444 | saved_errno = errno; | |
2445 | close(epfd); | |
2446 | ||
2447 | if (ret <= 0) { | |
2448 | errno = saved_errno; | |
2449 | return false; | |
2450 | } | |
2451 | return true; | |
2452 | } | |
2453 | ||
2454 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2455 | { | |
2456 | if (!wait_for_sock(sockfd, 2)) | |
2457 | return -1; | |
2458 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2459 | } | |
2460 | ||
2461 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2462 | { | |
2463 | struct msghdr msg = { 0 }; | |
2464 | struct iovec iov; | |
2465 | struct cmsghdr *cmsg; | |
2466 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2467 | char buf[1]; | |
2468 | buf[0] = 'p'; | |
2469 | ||
2470 | if (pingfirst) { | |
2471 | if (msgrecv(sock, buf, 1) != 1) { | |
b8defc3d | 2472 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); |
237e200e SH |
2473 | return SEND_CREDS_FAIL; |
2474 | } | |
2475 | } | |
2476 | ||
2477 | msg.msg_control = cmsgbuf; | |
2478 | msg.msg_controllen = sizeof(cmsgbuf); | |
2479 | ||
2480 | cmsg = CMSG_FIRSTHDR(&msg); | |
2481 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2482 | cmsg->cmsg_level = SOL_SOCKET; | |
2483 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2484 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2485 | ||
2486 | msg.msg_name = NULL; | |
2487 | msg.msg_namelen = 0; | |
2488 | ||
2489 | buf[0] = v; | |
2490 | iov.iov_base = buf; | |
2491 | iov.iov_len = sizeof(buf); | |
2492 | msg.msg_iov = &iov; | |
2493 | msg.msg_iovlen = 1; | |
2494 | ||
2495 | if (sendmsg(sock, &msg, 0) < 0) { | |
b8defc3d | 2496 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); |
237e200e SH |
2497 | if (errno == 3) |
2498 | return SEND_CREDS_NOTSK; | |
2499 | return SEND_CREDS_FAIL; | |
2500 | } | |
2501 | ||
2502 | return SEND_CREDS_OK; | |
2503 | } | |
2504 | ||
2505 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2506 | { | |
2507 | struct msghdr msg = { 0 }; | |
2508 | struct iovec iov; | |
2509 | struct cmsghdr *cmsg; | |
2510 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2511 | char buf[1]; | |
2512 | int ret; | |
2513 | int optval = 1; | |
2514 | ||
2515 | *v = '1'; | |
2516 | ||
2517 | cred->pid = -1; | |
2518 | cred->uid = -1; | |
2519 | cred->gid = -1; | |
2520 | ||
2521 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
b8defc3d | 2522 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); |
237e200e SH |
2523 | return false; |
2524 | } | |
2525 | buf[0] = '1'; | |
2526 | if (write(sock, buf, 1) != 1) { | |
b8defc3d | 2527 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); |
237e200e SH |
2528 | return false; |
2529 | } | |
2530 | ||
2531 | msg.msg_name = NULL; | |
2532 | msg.msg_namelen = 0; | |
2533 | msg.msg_control = cmsgbuf; | |
2534 | msg.msg_controllen = sizeof(cmsgbuf); | |
2535 | ||
2536 | iov.iov_base = buf; | |
2537 | iov.iov_len = sizeof(buf); | |
2538 | msg.msg_iov = &iov; | |
2539 | msg.msg_iovlen = 1; | |
2540 | ||
2541 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2542 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2543 | return false; |
2544 | } | |
2545 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2546 | if (ret < 0) { | |
b8defc3d | 2547 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2548 | return false; |
2549 | } | |
2550 | ||
2551 | cmsg = CMSG_FIRSTHDR(&msg); | |
2552 | ||
2553 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2554 | cmsg->cmsg_level == SOL_SOCKET && | |
2555 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2556 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2557 | } | |
2558 | *v = buf[0]; | |
2559 | ||
2560 | return true; | |
2561 | } | |
2562 | ||
35174b0f FG |
2563 | struct pid_ns_clone_args { |
2564 | int *cpipe; | |
2565 | int sock; | |
2566 | pid_t tpid; | |
2567 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2568 | }; | |
2569 | ||
2570 | /* | |
2571 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2572 | * with clone(). This simply writes '1' as ACK back to the parent | |
2573 | * before calling the actual wrapped function. | |
2574 | */ | |
2575 | static int pid_ns_clone_wrapper(void *arg) { | |
2576 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2577 | char b = '1'; | |
2578 | ||
2579 | close(args->cpipe[0]); | |
b8defc3d CB |
2580 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) |
2581 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
35174b0f FG |
2582 | close(args->cpipe[1]); |
2583 | return args->wrapped(args->sock, args->tpid); | |
2584 | } | |
237e200e SH |
2585 | |
2586 | /* | |
2587 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2588 | * int value back over the socket. This shifts the pid from the | |
2589 | * sender's pidns into tpid's pidns. | |
2590 | */ | |
35174b0f | 2591 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2592 | { |
2593 | char v = '0'; | |
2594 | struct ucred cred; | |
2595 | ||
2596 | while (recv_creds(sock, &cred, &v)) { | |
2597 | if (v == '1') | |
35174b0f | 2598 | return 0; |
237e200e | 2599 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2600 | return 1; |
237e200e | 2601 | } |
35174b0f | 2602 | return 0; |
237e200e SH |
2603 | } |
2604 | ||
35174b0f | 2605 | |
237e200e SH |
2606 | /* |
2607 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2608 | * in your old pidns. Only children which you clone will be in the target |
2609 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2610 | * actually convert pids. | |
2611 | * | |
2612 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2613 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2614 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2615 | * this issue. | |
237e200e SH |
2616 | */ |
2617 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2618 | { | |
2619 | int newnsfd = -1, ret, cpipe[2]; | |
2620 | char fnam[100]; | |
2621 | pid_t cpid; | |
2622 | char v; | |
2623 | ||
2624 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2625 | if (ret < 0 || ret >= sizeof(fnam)) | |
2626 | _exit(1); | |
2627 | newnsfd = open(fnam, O_RDONLY); | |
2628 | if (newnsfd < 0) | |
2629 | _exit(1); | |
2630 | if (setns(newnsfd, 0) < 0) | |
2631 | _exit(1); | |
2632 | close(newnsfd); | |
2633 | ||
2634 | if (pipe(cpipe) < 0) | |
2635 | _exit(1); | |
2636 | ||
35174b0f FG |
2637 | struct pid_ns_clone_args args = { |
2638 | .cpipe = cpipe, | |
2639 | .sock = sock, | |
2640 | .tpid = tpid, | |
2641 | .wrapped = &pid_to_ns | |
2642 | }; | |
2643 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2644 | void *stack = alloca(stack_size); | |
2645 | ||
2646 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2647 | if (cpid < 0) |
2648 | _exit(1); | |
2649 | ||
237e200e SH |
2650 | // give the child 1 second to be done forking and |
2651 | // write its ack | |
2652 | if (!wait_for_sock(cpipe[0], 1)) | |
2653 | _exit(1); | |
2654 | ret = read(cpipe[0], &v, 1); | |
2655 | if (ret != sizeof(char) || v != '1') | |
2656 | _exit(1); | |
2657 | ||
2658 | if (!wait_for_pid(cpid)) | |
2659 | _exit(1); | |
2660 | _exit(0); | |
2661 | } | |
2662 | ||
2663 | /* | |
2664 | * To read cgroup files with a particular pid, we will setns into the child | |
2665 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2666 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2667 | */ | |
2668 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2669 | { | |
2670 | int sock[2] = {-1, -1}; | |
2671 | char *tmpdata = NULL; | |
2672 | int ret; | |
2673 | pid_t qpid, cpid = -1; | |
2674 | bool answer = false; | |
2675 | char v = '0'; | |
2676 | struct ucred cred; | |
2677 | size_t sz = 0, asz = 0; | |
2678 | ||
2679 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2680 | return false; | |
2681 | ||
2682 | /* | |
2683 | * Now we read the pids from returned data one by one, pass | |
2684 | * them into a child in the target namespace, read back the | |
2685 | * translated pids, and put them into our to-return data | |
2686 | */ | |
2687 | ||
2688 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2689 | perror("socketpair"); | |
2690 | free(tmpdata); | |
2691 | return false; | |
2692 | } | |
2693 | ||
2694 | cpid = fork(); | |
2695 | if (cpid == -1) | |
2696 | goto out; | |
2697 | ||
2698 | if (!cpid) // child - exits when done | |
2699 | pid_to_ns_wrapper(sock[1], tpid); | |
2700 | ||
2701 | char *ptr = tmpdata; | |
2702 | cred.uid = 0; | |
2703 | cred.gid = 0; | |
2704 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2705 | cred.pid = qpid; | |
2706 | ret = send_creds(sock[0], &cred, v, true); | |
2707 | ||
2708 | if (ret == SEND_CREDS_NOTSK) | |
2709 | goto next; | |
2710 | if (ret == SEND_CREDS_FAIL) | |
2711 | goto out; | |
2712 | ||
2713 | // read converted results | |
2714 | if (!wait_for_sock(sock[0], 2)) { | |
b8defc3d | 2715 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2716 | goto out; |
2717 | } | |
2718 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2719 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2720 | goto out; |
2721 | } | |
2722 | must_strcat_pid(d, &sz, &asz, qpid); | |
2723 | next: | |
2724 | ptr = strchr(ptr, '\n'); | |
2725 | if (!ptr) | |
2726 | break; | |
2727 | ptr++; | |
2728 | } | |
2729 | ||
2730 | cred.pid = getpid(); | |
2731 | v = '1'; | |
2732 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2733 | // failed to ask child to exit | |
b8defc3d | 2734 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); |
237e200e SH |
2735 | goto out; |
2736 | } | |
2737 | ||
2738 | answer = true; | |
2739 | ||
2740 | out: | |
2741 | free(tmpdata); | |
2742 | if (cpid != -1) | |
2743 | wait_for_pid(cpid); | |
2744 | if (sock[0] != -1) { | |
2745 | close(sock[0]); | |
2746 | close(sock[1]); | |
2747 | } | |
2748 | return answer; | |
2749 | } | |
2750 | ||
2751 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2752 | struct fuse_file_info *fi) | |
2753 | { | |
2754 | struct fuse_context *fc = fuse_get_context(); | |
2755 | struct file_info *f = (struct file_info *)fi->fh; | |
2756 | struct cgfs_files *k = NULL; | |
2757 | char *data = NULL; | |
2758 | int ret, s; | |
2759 | bool r; | |
2760 | ||
2761 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2762 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); |
237e200e SH |
2763 | return -EIO; |
2764 | } | |
2765 | ||
2766 | if (offset) | |
2767 | return 0; | |
2768 | ||
2769 | if (!fc) | |
2770 | return -EIO; | |
2771 | ||
2772 | if (!f->controller) | |
2773 | return -EINVAL; | |
2774 | ||
2775 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2776 | return -EINVAL; | |
2777 | } | |
2778 | free_key(k); | |
2779 | ||
2780 | ||
888f8f3c | 2781 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2782 | ret = -EACCES; |
2783 | goto out; | |
2784 | } | |
2785 | ||
2786 | if (strcmp(f->file, "tasks") == 0 || | |
2787 | strcmp(f->file, "/tasks") == 0 || | |
2788 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2789 | strcmp(f->file, "cgroup.procs") == 0) | |
2790 | // special case - we have to translate the pids | |
2791 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2792 | else | |
2793 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2794 | ||
2795 | if (!r) { | |
2796 | ret = -EINVAL; | |
2797 | goto out; | |
2798 | } | |
2799 | ||
2800 | if (!data) { | |
2801 | ret = 0; | |
2802 | goto out; | |
2803 | } | |
2804 | s = strlen(data); | |
2805 | if (s > size) | |
2806 | s = size; | |
2807 | memcpy(buf, data, s); | |
2808 | if (s > 0 && s < size && data[s-1] != '\n') | |
2809 | buf[s++] = '\n'; | |
2810 | ||
2811 | ret = s; | |
2812 | ||
2813 | out: | |
2814 | free(data); | |
2815 | return ret; | |
2816 | } | |
2817 | ||
35174b0f | 2818 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2819 | { |
2820 | pid_t vpid; | |
2821 | struct ucred cred; | |
2822 | char v; | |
2823 | int ret; | |
2824 | ||
2825 | cred.uid = 0; | |
2826 | cred.gid = 0; | |
2827 | while (1) { | |
2828 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2829 | lxcfs_error("%s\n", "Timeout reading from parent."); |
35174b0f | 2830 | return 1; |
237e200e SH |
2831 | } |
2832 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
b8defc3d | 2833 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); |
35174b0f | 2834 | return 1; |
237e200e SH |
2835 | } |
2836 | if (vpid == -1) // done | |
2837 | break; | |
2838 | v = '0'; | |
2839 | cred.pid = vpid; | |
2840 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2841 | v = '1'; | |
2842 | cred.pid = getpid(); | |
2843 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2844 | return 1; |
237e200e SH |
2845 | } |
2846 | } | |
35174b0f | 2847 | return 0; |
237e200e SH |
2848 | } |
2849 | ||
2850 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2851 | { | |
2852 | int newnsfd = -1, ret, cpipe[2]; | |
2853 | char fnam[100]; | |
2854 | pid_t cpid; | |
2855 | char v; | |
2856 | ||
2857 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2858 | if (ret < 0 || ret >= sizeof(fnam)) | |
2859 | _exit(1); | |
2860 | newnsfd = open(fnam, O_RDONLY); | |
2861 | if (newnsfd < 0) | |
2862 | _exit(1); | |
2863 | if (setns(newnsfd, 0) < 0) | |
2864 | _exit(1); | |
2865 | close(newnsfd); | |
2866 | ||
2867 | if (pipe(cpipe) < 0) | |
2868 | _exit(1); | |
2869 | ||
35174b0f FG |
2870 | struct pid_ns_clone_args args = { |
2871 | .cpipe = cpipe, | |
2872 | .sock = sock, | |
2873 | .tpid = tpid, | |
2874 | .wrapped = &pid_from_ns | |
2875 | }; | |
f0f8b851 SH |
2876 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2877 | void *stack = alloca(stack_size); | |
35174b0f FG |
2878 | |
2879 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2880 | if (cpid < 0) |
2881 | _exit(1); | |
2882 | ||
237e200e SH |
2883 | // give the child 1 second to be done forking and |
2884 | // write its ack | |
2885 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2886 | _exit(1); |
237e200e | 2887 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2888 | if (ret != sizeof(char) || v != '1') |
2889 | _exit(1); | |
237e200e SH |
2890 | |
2891 | if (!wait_for_pid(cpid)) | |
2892 | _exit(1); | |
2893 | _exit(0); | |
237e200e SH |
2894 | } |
2895 | ||
2896 | /* | |
2897 | * Given host @uid, return the uid to which it maps in | |
2898 | * @pid's user namespace, or -1 if none. | |
2899 | */ | |
2900 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2901 | { | |
2902 | FILE *f; | |
2903 | char line[400]; | |
2904 | ||
2905 | sprintf(line, "/proc/%d/uid_map", pid); | |
2906 | if ((f = fopen(line, "r")) == NULL) { | |
2907 | return false; | |
2908 | } | |
2909 | ||
2910 | *answer = convert_id_to_ns(f, uid); | |
2911 | fclose(f); | |
2912 | ||
2913 | if (*answer == -1) | |
2914 | return false; | |
2915 | return true; | |
2916 | } | |
2917 | ||
2918 | /* | |
2919 | * get_pid_creds: get the real uid and gid of @pid from | |
2920 | * /proc/$$/status | |
2921 | * (XXX should we use euid here?) | |
2922 | */ | |
2923 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2924 | { | |
2925 | char line[400]; | |
2926 | uid_t u; | |
2927 | gid_t g; | |
2928 | FILE *f; | |
2929 | ||
2930 | *uid = -1; | |
2931 | *gid = -1; | |
2932 | sprintf(line, "/proc/%d/status", pid); | |
2933 | if ((f = fopen(line, "r")) == NULL) { | |
b8defc3d | 2934 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); |
237e200e SH |
2935 | return; |
2936 | } | |
2937 | while (fgets(line, 400, f)) { | |
2938 | if (strncmp(line, "Uid:", 4) == 0) { | |
2939 | if (sscanf(line+4, "%u", &u) != 1) { | |
b8defc3d | 2940 | lxcfs_error("bad uid line for pid %u\n", pid); |
237e200e SH |
2941 | fclose(f); |
2942 | return; | |
2943 | } | |
2944 | *uid = u; | |
2945 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2946 | if (sscanf(line+4, "%u", &g) != 1) { | |
b8defc3d | 2947 | lxcfs_error("bad gid line for pid %u\n", pid); |
237e200e SH |
2948 | fclose(f); |
2949 | return; | |
2950 | } | |
2951 | *gid = g; | |
2952 | } | |
2953 | } | |
2954 | fclose(f); | |
2955 | } | |
2956 | ||
2957 | /* | |
2958 | * May the requestor @r move victim @v to a new cgroup? | |
2959 | * This is allowed if | |
2960 | * . they are the same task | |
2961 | * . they are ownedy by the same uid | |
2962 | * . @r is root on the host, or | |
2963 | * . @v's uid is mapped into @r's where @r is root. | |
2964 | */ | |
2965 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2966 | { | |
2967 | uid_t v_uid, tmpuid; | |
2968 | gid_t v_gid; | |
2969 | ||
2970 | if (r == v) | |
2971 | return true; | |
2972 | if (r_uid == 0) | |
2973 | return true; | |
2974 | get_pid_creds(v, &v_uid, &v_gid); | |
2975 | if (r_uid == v_uid) | |
2976 | return true; | |
2977 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2978 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2979 | return true; | |
2980 | return false; | |
2981 | } | |
2982 | ||
2983 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2984 | const char *file, const char *buf) | |
2985 | { | |
2986 | int sock[2] = {-1, -1}; | |
2987 | pid_t qpid, cpid = -1; | |
2988 | FILE *pids_file = NULL; | |
2989 | bool answer = false, fail = false; | |
2990 | ||
2991 | pids_file = open_pids_file(contrl, cg); | |
2992 | if (!pids_file) | |
2993 | return false; | |
2994 | ||
2995 | /* | |
2996 | * write the pids to a socket, have helper in writer's pidns | |
2997 | * call movepid for us | |
2998 | */ | |
2999 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
3000 | perror("socketpair"); | |
3001 | goto out; | |
3002 | } | |
3003 | ||
3004 | cpid = fork(); | |
3005 | if (cpid == -1) | |
3006 | goto out; | |
3007 | ||
3008 | if (!cpid) { // child | |
3009 | fclose(pids_file); | |
3010 | pid_from_ns_wrapper(sock[1], tpid); | |
3011 | } | |
3012 | ||
3013 | const char *ptr = buf; | |
3014 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
3015 | struct ucred cred; | |
3016 | char v; | |
3017 | ||
3018 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 3019 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); |
237e200e SH |
3020 | goto out; |
3021 | } | |
3022 | ||
3023 | if (recv_creds(sock[0], &cred, &v)) { | |
3024 | if (v == '0') { | |
3025 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
3026 | fail = true; | |
3027 | break; | |
3028 | } | |
3029 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
3030 | fail = true; | |
3031 | } | |
3032 | } | |
3033 | ||
3034 | ptr = strchr(ptr, '\n'); | |
3035 | if (!ptr) | |
3036 | break; | |
3037 | ptr++; | |
3038 | } | |
3039 | ||
3040 | /* All good, write the value */ | |
3041 | qpid = -1; | |
3042 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
b8defc3d | 3043 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); |
237e200e SH |
3044 | |
3045 | if (!fail) | |
3046 | answer = true; | |
3047 | ||
3048 | out: | |
3049 | if (cpid != -1) | |
3050 | wait_for_pid(cpid); | |
3051 | if (sock[0] != -1) { | |
3052 | close(sock[0]); | |
3053 | close(sock[1]); | |
3054 | } | |
3055 | if (pids_file) { | |
3056 | if (fclose(pids_file) != 0) | |
3057 | answer = false; | |
3058 | } | |
3059 | return answer; | |
3060 | } | |
3061 | ||
3062 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
3063 | struct fuse_file_info *fi) | |
3064 | { | |
3065 | struct fuse_context *fc = fuse_get_context(); | |
3066 | char *localbuf = NULL; | |
3067 | struct cgfs_files *k = NULL; | |
3068 | struct file_info *f = (struct file_info *)fi->fh; | |
3069 | bool r; | |
3070 | ||
3071 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 3072 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); |
237e200e SH |
3073 | return -EIO; |
3074 | } | |
3075 | ||
3076 | if (offset) | |
3077 | return 0; | |
3078 | ||
3079 | if (!fc) | |
3080 | return -EIO; | |
3081 | ||
3082 | localbuf = alloca(size+1); | |
3083 | localbuf[size] = '\0'; | |
3084 | memcpy(localbuf, buf, size); | |
3085 | ||
3086 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
3087 | size = -EINVAL; | |
3088 | goto out; | |
3089 | } | |
3090 | ||
3091 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
3092 | size = -EACCES; | |
3093 | goto out; | |
3094 | } | |
3095 | ||
3096 | if (strcmp(f->file, "tasks") == 0 || | |
3097 | strcmp(f->file, "/tasks") == 0 || | |
3098 | strcmp(f->file, "/cgroup.procs") == 0 || | |
3099 | strcmp(f->file, "cgroup.procs") == 0) | |
3100 | // special case - we have to translate the pids | |
3101 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
3102 | else | |
3103 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
3104 | ||
3105 | if (!r) | |
3106 | size = -EINVAL; | |
3107 | ||
3108 | out: | |
3109 | free_key(k); | |
3110 | return size; | |
3111 | } | |
3112 | ||
3113 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
3114 | { | |
3115 | struct fuse_context *fc = fuse_get_context(); | |
3116 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
3117 | struct cgfs_files *k = NULL; | |
3118 | const char *cgroup; | |
3119 | int ret; | |
3120 | ||
3121 | if (!fc) | |
3122 | return -EIO; | |
3123 | ||
3124 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 3125 | return -EPERM; |
237e200e SH |
3126 | |
3127 | controller = pick_controller_from_path(fc, path); | |
3128 | if (!controller) | |
bc70ba9b CB |
3129 | return errno == ENOENT ? -EPERM : -errno; |
3130 | ||
237e200e SH |
3131 | cgroup = find_cgroup_in_path(path); |
3132 | if (!cgroup) | |
3133 | /* this is just /cgroup/controller */ | |
bc70ba9b | 3134 | return -EPERM; |
237e200e SH |
3135 | |
3136 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3137 | ||
3138 | if (!last) { | |
3139 | path1 = "/"; | |
3140 | path2 = cgdir; | |
3141 | } else { | |
3142 | path1 = cgdir; | |
3143 | path2 = last; | |
3144 | } | |
3145 | ||
3146 | if (is_child_cgroup(controller, path1, path2)) { | |
3147 | // get uid, gid, from '/tasks' file and make up a mode | |
3148 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
3149 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
3150 | ||
3151 | } else | |
3152 | k = cgfs_get_key(controller, path1, path2); | |
3153 | ||
3154 | if (!k) { | |
3155 | ret = -EINVAL; | |
3156 | goto out; | |
3157 | } | |
3158 | ||
3159 | /* | |
3160 | * This being a fuse request, the uid and gid must be valid | |
3161 | * in the caller's namespace. So we can just check to make | |
3162 | * sure that the caller is root in his uid, and privileged | |
3163 | * over the file's current owner. | |
3164 | */ | |
3165 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
3166 | ret = -EACCES; | |
3167 | goto out; | |
3168 | } | |
3169 | ||
3170 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
3171 | ||
3172 | out: | |
3173 | free_key(k); | |
3174 | free(cgdir); | |
3175 | ||
3176 | return ret; | |
3177 | } | |
3178 | ||
3179 | int cg_chmod(const char *path, mode_t mode) | |
3180 | { | |
3181 | struct fuse_context *fc = fuse_get_context(); | |
3182 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
3183 | struct cgfs_files *k = NULL; | |
3184 | const char *cgroup; | |
3185 | int ret; | |
3186 | ||
3187 | if (!fc) | |
3188 | return -EIO; | |
3189 | ||
3190 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 3191 | return -EPERM; |
237e200e SH |
3192 | |
3193 | controller = pick_controller_from_path(fc, path); | |
3194 | if (!controller) | |
bc70ba9b CB |
3195 | return errno == ENOENT ? -EPERM : -errno; |
3196 | ||
237e200e SH |
3197 | cgroup = find_cgroup_in_path(path); |
3198 | if (!cgroup) | |
3199 | /* this is just /cgroup/controller */ | |
bc70ba9b | 3200 | return -EPERM; |
237e200e SH |
3201 | |
3202 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3203 | ||
3204 | if (!last) { | |
3205 | path1 = "/"; | |
3206 | path2 = cgdir; | |
3207 | } else { | |
3208 | path1 = cgdir; | |
3209 | path2 = last; | |
3210 | } | |
3211 | ||
3212 | if (is_child_cgroup(controller, path1, path2)) { | |
3213 | // get uid, gid, from '/tasks' file and make up a mode | |
3214 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
3215 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
3216 | ||
3217 | } else | |
3218 | k = cgfs_get_key(controller, path1, path2); | |
3219 | ||
3220 | if (!k) { | |
3221 | ret = -EINVAL; | |
3222 | goto out; | |
3223 | } | |
3224 | ||
3225 | /* | |
3226 | * This being a fuse request, the uid and gid must be valid | |
3227 | * in the caller's namespace. So we can just check to make | |
3228 | * sure that the caller is root in his uid, and privileged | |
3229 | * over the file's current owner. | |
3230 | */ | |
3231 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
3232 | ret = -EPERM; | |
3233 | goto out; | |
3234 | } | |
3235 | ||
3236 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
3237 | ret = -EINVAL; | |
3238 | goto out; | |
3239 | } | |
3240 | ||
3241 | ret = 0; | |
3242 | out: | |
3243 | free_key(k); | |
3244 | free(cgdir); | |
3245 | return ret; | |
3246 | } | |
3247 | ||
3248 | int cg_mkdir(const char *path, mode_t mode) | |
3249 | { | |
3250 | struct fuse_context *fc = fuse_get_context(); | |
3251 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
3252 | const char *cgroup; | |
3253 | int ret; | |
3254 | ||
3255 | if (!fc) | |
3256 | return -EIO; | |
3257 | ||
237e200e SH |
3258 | controller = pick_controller_from_path(fc, path); |
3259 | if (!controller) | |
2f7036d0 | 3260 | return errno == ENOENT ? -EPERM : -errno; |
237e200e SH |
3261 | |
3262 | cgroup = find_cgroup_in_path(path); | |
3263 | if (!cgroup) | |
bc70ba9b | 3264 | return -errno; |
237e200e SH |
3265 | |
3266 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3267 | if (!last) | |
3268 | path1 = "/"; | |
3269 | else | |
3270 | path1 = cgdir; | |
3271 | ||
3272 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3273 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3274 | initpid = fc->pid; |
3275 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
3276 | if (!next) | |
3277 | ret = -EINVAL; | |
3278 | else if (last && strcmp(next, last) == 0) | |
3279 | ret = -EEXIST; | |
3280 | else | |
2f7036d0 | 3281 | ret = -EPERM; |
237e200e SH |
3282 | goto out; |
3283 | } | |
3284 | ||
3285 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
3286 | ret = -EACCES; | |
3287 | goto out; | |
3288 | } | |
3289 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
3290 | ret = -EACCES; | |
3291 | goto out; | |
3292 | } | |
3293 | ||
3294 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
3295 | ||
3296 | out: | |
3297 | free(cgdir); | |
3298 | free(next); | |
3299 | return ret; | |
3300 | } | |
3301 | ||
3302 | int cg_rmdir(const char *path) | |
3303 | { | |
3304 | struct fuse_context *fc = fuse_get_context(); | |
3305 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
3306 | const char *cgroup; | |
3307 | int ret; | |
3308 | ||
3309 | if (!fc) | |
3310 | return -EIO; | |
3311 | ||
3312 | controller = pick_controller_from_path(fc, path); | |
e254948f CB |
3313 | if (!controller) /* Someone's trying to delete "/cgroup". */ |
3314 | return -EPERM; | |
237e200e SH |
3315 | |
3316 | cgroup = find_cgroup_in_path(path); | |
e254948f CB |
3317 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ |
3318 | return -EPERM; | |
237e200e SH |
3319 | |
3320 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3321 | if (!last) { | |
e254948f CB |
3322 | /* Someone's trying to delete a cgroup on the same level as the |
3323 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
3324 | * rmdir "/cgroup/blkio/init.slice". | |
3325 | */ | |
3326 | ret = -EPERM; | |
237e200e SH |
3327 | goto out; |
3328 | } | |
3329 | ||
3330 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3331 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3332 | initpid = fc->pid; |
3333 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
de77249b | 3334 | if (!last || (next && (strcmp(next, last) == 0))) |
237e200e SH |
3335 | ret = -EBUSY; |
3336 | else | |
3337 | ret = -ENOENT; | |
3338 | goto out; | |
3339 | } | |
3340 | ||
3341 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
3342 | ret = -EACCES; | |
3343 | goto out; | |
3344 | } | |
3345 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
3346 | ret = -EACCES; | |
3347 | goto out; | |
3348 | } | |
3349 | ||
3350 | if (!cgfs_remove(controller, cgroup)) { | |
3351 | ret = -EINVAL; | |
3352 | goto out; | |
3353 | } | |
3354 | ||
3355 | ret = 0; | |
3356 | ||
3357 | out: | |
3358 | free(cgdir); | |
3359 | free(next); | |
3360 | return ret; | |
3361 | } | |
3362 | ||
3363 | static bool startswith(const char *line, const char *pref) | |
3364 | { | |
3365 | if (strncmp(line, pref, strlen(pref)) == 0) | |
3366 | return true; | |
3367 | return false; | |
3368 | } | |
3369 | ||
c6095b08 SH |
3370 | static void parse_memstat(char *memstat, unsigned long *cached, |
3371 | unsigned long *active_anon, unsigned long *inactive_anon, | |
3372 | unsigned long *active_file, unsigned long *inactive_file, | |
559eaa8f | 3373 | unsigned long *unevictable, unsigned long *shmem) |
237e200e SH |
3374 | { |
3375 | char *eol; | |
3376 | ||
237e200e | 3377 | while (*memstat) { |
4accebfb AS |
3378 | if (startswith(memstat, "total_cache")) { |
3379 | sscanf(memstat + 11, "%lu", cached); | |
c6095b08 | 3380 | *cached /= 1024; |
4accebfb AS |
3381 | } else if (startswith(memstat, "total_active_anon")) { |
3382 | sscanf(memstat + 17, "%lu", active_anon); | |
c6095b08 | 3383 | *active_anon /= 1024; |
4accebfb AS |
3384 | } else if (startswith(memstat, "total_inactive_anon")) { |
3385 | sscanf(memstat + 19, "%lu", inactive_anon); | |
c6095b08 | 3386 | *inactive_anon /= 1024; |
4accebfb AS |
3387 | } else if (startswith(memstat, "total_active_file")) { |
3388 | sscanf(memstat + 17, "%lu", active_file); | |
c6095b08 | 3389 | *active_file /= 1024; |
4accebfb AS |
3390 | } else if (startswith(memstat, "total_inactive_file")) { |
3391 | sscanf(memstat + 19, "%lu", inactive_file); | |
c6095b08 | 3392 | *inactive_file /= 1024; |
4accebfb AS |
3393 | } else if (startswith(memstat, "total_unevictable")) { |
3394 | sscanf(memstat + 17, "%lu", unevictable); | |
c6095b08 | 3395 | *unevictable /= 1024; |
559eaa8f JS |
3396 | } else if (startswith(memstat, "total_shmem")) { |
3397 | sscanf(memstat + 11, "%lu", shmem); | |
3398 | *shmem /= 1024; | |
237e200e SH |
3399 | } |
3400 | eol = strchr(memstat, '\n'); | |
3401 | if (!eol) | |
3402 | return; | |
3403 | memstat = eol+1; | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
3408 | { | |
3409 | char *eol; | |
3410 | char key[32]; | |
3411 | ||
3412 | memset(key, 0, 32); | |
3413 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
3414 | ||
3415 | size_t len = strlen(key); | |
3416 | *v = 0; | |
3417 | ||
3418 | while (*str) { | |
3419 | if (startswith(str, key)) { | |
3420 | sscanf(str + len, "%lu", v); | |
3421 | return; | |
3422 | } | |
3423 | eol = strchr(str, '\n'); | |
3424 | if (!eol) | |
3425 | return; | |
3426 | str = eol+1; | |
3427 | } | |
3428 | } | |
3429 | ||
71f17cd2 | 3430 | int read_file(const char *path, char *buf, size_t size, struct file_info *d) |
237e200e SH |
3431 | { |
3432 | size_t linelen = 0, total_len = 0, rv = 0; | |
3433 | char *line = NULL; | |
3434 | char *cache = d->buf; | |
3435 | size_t cache_size = d->buflen; | |
3436 | FILE *f = fopen(path, "r"); | |
3437 | if (!f) | |
3438 | return 0; | |
3439 | ||
3440 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3441 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3442 | if (l < 0) { |
3443 | perror("Error writing to cache"); | |
3444 | rv = 0; | |
3445 | goto err; | |
3446 | } | |
3447 | if (l >= cache_size) { | |
b8defc3d | 3448 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3449 | rv = 0; |
3450 | goto err; | |
3451 | } | |
3452 | cache += l; | |
3453 | cache_size -= l; | |
3454 | total_len += l; | |
3455 | } | |
3456 | ||
3457 | d->size = total_len; | |
a262ddb7 CB |
3458 | if (total_len > size) |
3459 | total_len = size; | |
237e200e SH |
3460 | |
3461 | /* read from off 0 */ | |
3462 | memcpy(buf, d->buf, total_len); | |
3463 | rv = total_len; | |
3464 | err: | |
3465 | fclose(f); | |
3466 | free(line); | |
030d022c HY |
3467 | if (d->size > rv) |
3468 | d->cached = d->size - rv; | |
237e200e SH |
3469 | return rv; |
3470 | } | |
3471 | ||
3472 | /* | |
3473 | * FUSE ops for /proc | |
3474 | */ | |
3475 | ||
018246ff | 3476 | static unsigned long get_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3477 | { |
3478 | char *memlimit_str = NULL; | |
3479 | unsigned long memlimit = -1; | |
3480 | ||
018246ff | 3481 | if (cgfs_get_value("memory", cgroup, file, &memlimit_str)) |
237e200e SH |
3482 | memlimit = strtoul(memlimit_str, NULL, 10); |
3483 | ||
3484 | free(memlimit_str); | |
3485 | ||
3486 | return memlimit; | |
3487 | } | |
3488 | ||
018246ff | 3489 | static unsigned long get_min_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3490 | { |
3491 | char *copy = strdupa(cgroup); | |
3492 | unsigned long memlimit = 0, retlimit; | |
3493 | ||
018246ff | 3494 | retlimit = get_memlimit(copy, file); |
237e200e SH |
3495 | |
3496 | while (strcmp(copy, "/") != 0) { | |
3497 | copy = dirname(copy); | |
018246ff | 3498 | memlimit = get_memlimit(copy, file); |
237e200e SH |
3499 | if (memlimit != -1 && memlimit < retlimit) |
3500 | retlimit = memlimit; | |
3501 | }; | |
3502 | ||
3503 | return retlimit; | |
3504 | } | |
3505 | ||
3506 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3507 | struct fuse_file_info *fi) | |
3508 | { | |
3509 | struct fuse_context *fc = fuse_get_context(); | |
7e60aa1b | 3510 | struct lxcfs_opts *opts = (struct lxcfs_opts *) fuse_get_context()->private_data; |
237e200e SH |
3511 | struct file_info *d = (struct file_info *)fi->fh; |
3512 | char *cg; | |
3513 | char *memusage_str = NULL, *memstat_str = NULL, | |
018246ff | 3514 | *memswlimit_str = NULL, *memswusage_str = NULL; |
237e200e | 3515 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, |
c6095b08 | 3516 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
559eaa8f | 3517 | active_file = 0, inactive_file = 0, unevictable = 0, shmem = 0, |
594a10e6 | 3518 | hostswtotal = 0; |
237e200e SH |
3519 | char *line = NULL; |
3520 | size_t linelen = 0, total_len = 0, rv = 0; | |
3521 | char *cache = d->buf; | |
3522 | size_t cache_size = d->buflen; | |
3523 | FILE *f = NULL; | |
3524 | ||
3525 | if (offset){ | |
3526 | if (offset > d->size) | |
3527 | return -EINVAL; | |
3528 | if (!d->cached) | |
3529 | return 0; | |
3530 | int left = d->size - offset; | |
3531 | total_len = left > size ? size: left; | |
3532 | memcpy(buf, cache + offset, total_len); | |
3533 | return total_len; | |
3534 | } | |
3535 | ||
3536 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3537 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3538 | initpid = fc->pid; |
3539 | cg = get_pid_cgroup(initpid, "memory"); | |
3540 | if (!cg) | |
3541 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3542 | prune_init_slice(cg); |
237e200e | 3543 | |
018246ff | 3544 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
237e200e SH |
3545 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) |
3546 | goto err; | |
3547 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3548 | goto err; | |
3549 | ||
3550 | // Following values are allowed to fail, because swapaccount might be turned | |
3551 | // off for current kernel | |
3552 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3553 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3554 | { | |
018246ff | 3555 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
237e200e SH |
3556 | memswusage = strtoul(memswusage_str, NULL, 10); |
3557 | ||
237e200e SH |
3558 | memswlimit = memswlimit / 1024; |
3559 | memswusage = memswusage / 1024; | |
3560 | } | |
3561 | ||
3562 | memusage = strtoul(memusage_str, NULL, 10); | |
3563 | memlimit /= 1024; | |
3564 | memusage /= 1024; | |
3565 | ||
c6095b08 SH |
3566 | parse_memstat(memstat_str, &cached, &active_anon, |
3567 | &inactive_anon, &active_file, &inactive_file, | |
559eaa8f | 3568 | &unevictable, &shmem); |
237e200e SH |
3569 | |
3570 | f = fopen("/proc/meminfo", "r"); | |
3571 | if (!f) | |
3572 | goto err; | |
3573 | ||
3574 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3575 | ssize_t l; |
237e200e SH |
3576 | char *printme, lbuf[100]; |
3577 | ||
3578 | memset(lbuf, 0, 100); | |
3579 | if (startswith(line, "MemTotal:")) { | |
594a10e6 | 3580 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); |
237e200e SH |
3581 | if (hosttotal < memlimit) |
3582 | memlimit = hosttotal; | |
3583 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3584 | printme = lbuf; | |
3585 | } else if (startswith(line, "MemFree:")) { | |
3586 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3587 | printme = lbuf; | |
3588 | } else if (startswith(line, "MemAvailable:")) { | |
ad19b86d | 3589 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage + cached); |
237e200e | 3590 | printme = lbuf; |
b3aa7f1a | 3591 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0 && opts && opts->swap_off == false) { |
594a10e6 | 3592 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); |
4127e51b | 3593 | if (hostswtotal < memswlimit) |
3594 | memswlimit = hostswtotal; | |
3595 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit); | |
237e200e | 3596 | printme = lbuf; |
b3aa7f1a | 3597 | } else if (startswith(line, "SwapTotal:") && opts && opts->swap_off == true) { |
7e60aa1b | 3598 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", 0UL); |
3599 | printme = lbuf; | |
b3aa7f1a | 3600 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0 && opts && opts->swap_off == false) { |
4127e51b | 3601 | unsigned long swaptotal = memswlimit, |
e715dc60 | 3602 | swapusage = memusage > memswusage ? 0 : memswusage - memusage, |
b4665ce0 SH |
3603 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; |
3604 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3605 | printme = lbuf; |
b3aa7f1a | 3606 | } else if (startswith(line, "SwapFree:") && opts && opts->swap_off == true) { |
7e60aa1b | 3607 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", 0UL); |
3608 | printme = lbuf; | |
da35d72a SH |
3609 | } else if (startswith(line, "Slab:")) { |
3610 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3611 | printme = lbuf; | |
237e200e SH |
3612 | } else if (startswith(line, "Buffers:")) { |
3613 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3614 | printme = lbuf; | |
3615 | } else if (startswith(line, "Cached:")) { | |
3616 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3617 | printme = lbuf; | |
3618 | } else if (startswith(line, "SwapCached:")) { | |
3619 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3620 | printme = lbuf; | |
2f306ad3 | 3621 | } else if (startswith(line, "Active:")) { |
c6095b08 SH |
3622 | snprintf(lbuf, 100, "Active: %8lu kB\n", |
3623 | active_anon + active_file); | |
3624 | printme = lbuf; | |
2f306ad3 | 3625 | } else if (startswith(line, "Inactive:")) { |
c6095b08 SH |
3626 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", |
3627 | inactive_anon + inactive_file); | |
3628 | printme = lbuf; | |
3629 | } else if (startswith(line, "Active(anon)")) { | |
3630 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3631 | printme = lbuf; | |
3632 | } else if (startswith(line, "Inactive(anon)")) { | |
3633 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3634 | printme = lbuf; | |
3635 | } else if (startswith(line, "Active(file)")) { | |
3636 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3637 | printme = lbuf; | |
3638 | } else if (startswith(line, "Inactive(file)")) { | |
3639 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3640 | printme = lbuf; | |
3641 | } else if (startswith(line, "Unevictable")) { | |
3642 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3643 | printme = lbuf; | |
3644 | } else if (startswith(line, "SReclaimable")) { | |
3645 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3646 | printme = lbuf; | |
3647 | } else if (startswith(line, "SUnreclaim")) { | |
3648 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3649 | printme = lbuf; | |
559eaa8f JS |
3650 | } else if (startswith(line, "Shmem:")) { |
3651 | snprintf(lbuf, 100, "Shmem: %8lu kB\n", shmem); | |
3652 | printme = lbuf; | |
28cdea9b JS |
3653 | } else if (startswith(line, "ShmemHugePages")) { |
3654 | snprintf(lbuf, 100, "ShmemHugePages: %8lu kB\n", 0UL); | |
3655 | printme = lbuf; | |
3656 | } else if (startswith(line, "ShmemPmdMapped")) { | |
3657 | snprintf(lbuf, 100, "ShmemPmdMapped: %8lu kB\n", 0UL); | |
3658 | printme = lbuf; | |
237e200e SH |
3659 | } else |
3660 | printme = line; | |
3661 | ||
3662 | l = snprintf(cache, cache_size, "%s", printme); | |
3663 | if (l < 0) { | |
3664 | perror("Error writing to cache"); | |
3665 | rv = 0; | |
3666 | goto err; | |
3667 | ||
3668 | } | |
3669 | if (l >= cache_size) { | |
b8defc3d | 3670 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3671 | rv = 0; |
3672 | goto err; | |
3673 | } | |
3674 | ||
3675 | cache += l; | |
3676 | cache_size -= l; | |
3677 | total_len += l; | |
3678 | } | |
3679 | ||
3680 | d->cached = 1; | |
3681 | d->size = total_len; | |
3682 | if (total_len > size ) total_len = size; | |
3683 | memcpy(buf, d->buf, total_len); | |
3684 | ||
3685 | rv = total_len; | |
3686 | err: | |
3687 | if (f) | |
3688 | fclose(f); | |
3689 | free(line); | |
3690 | free(cg); | |
3691 | free(memusage_str); | |
3692 | free(memswlimit_str); | |
3693 | free(memswusage_str); | |
3694 | free(memstat_str); | |
237e200e SH |
3695 | return rv; |
3696 | } | |
3697 | ||
3698 | /* | |
3699 | * Read the cpuset.cpus for cg | |
3700 | * Return the answer in a newly allocated string which must be freed | |
3701 | */ | |
71f17cd2 | 3702 | char *get_cpuset(const char *cg) |
237e200e SH |
3703 | { |
3704 | char *answer; | |
3705 | ||
3706 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3707 | return NULL; | |
3708 | return answer; | |
3709 | } | |
3710 | ||
3711 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3712 | ||
3713 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3714 | { | |
3715 | int cpu; | |
3716 | ||
3717 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3718 | return false; | |
3719 | return cpu_in_cpuset(cpu, cpuset); | |
3720 | } | |
3721 | ||
c59d6a55 JS |
3722 | /* |
3723 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or `cpu.cfs_period_us`, | |
3724 | * depending on `param`. Parameter value is returned throuh `value`. | |
3725 | */ | |
3726 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
3727 | { | |
3728 | bool rv = false; | |
3729 | char file[11 + 6 + 1]; // cpu.cfs__us + quota/period + \0 | |
3730 | char *str = NULL; | |
3731 | ||
3732 | sprintf(file, "cpu.cfs_%s_us", param); | |
3733 | ||
3734 | if (!cgfs_get_value("cpu", cg, file, &str)) | |
3735 | goto err; | |
3736 | ||
3737 | if (sscanf(str, "%ld", value) != 1) | |
3738 | goto err; | |
3739 | ||
3740 | rv = true; | |
3741 | ||
3742 | err: | |
3743 | if (str) | |
3744 | free(str); | |
3745 | return rv; | |
3746 | } | |
3747 | ||
3748 | /* | |
3749 | * Return the maximum number of visible CPUs based on CPU quotas. | |
3750 | * If there is no quota set, zero is returned. | |
3751 | */ | |
3752 | int max_cpu_count(const char *cg) | |
3753 | { | |
3754 | int rv, nprocs; | |
3755 | int64_t cfs_quota, cfs_period; | |
8c23db36 HY |
3756 | int nr_cpus_in_cpuset = 0; |
3757 | char *cpuset = NULL; | |
c59d6a55 JS |
3758 | |
3759 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) | |
3760 | return 0; | |
3761 | ||
3762 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
3763 | return 0; | |
3764 | ||
8c23db36 HY |
3765 | cpuset = get_cpuset(cg); |
3766 | if (cpuset) | |
3767 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
3768 | ||
3769 | if (cfs_quota <= 0 || cfs_period <= 0){ | |
3770 | if (nr_cpus_in_cpuset > 0) | |
3771 | return nr_cpus_in_cpuset; | |
3772 | ||
c59d6a55 | 3773 | return 0; |
8c23db36 | 3774 | } |
c59d6a55 JS |
3775 | |
3776 | rv = cfs_quota / cfs_period; | |
3777 | ||
3778 | /* In case quota/period does not yield a whole number, add one CPU for | |
3779 | * the remainder. | |
3780 | */ | |
3781 | if ((cfs_quota % cfs_period) > 0) | |
3782 | rv += 1; | |
3783 | ||
3784 | nprocs = get_nprocs(); | |
3785 | ||
3786 | if (rv > nprocs) | |
3787 | rv = nprocs; | |
3788 | ||
8c23db36 HY |
3789 | /* use min value in cpu quota and cpuset */ |
3790 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) | |
3791 | rv = nr_cpus_in_cpuset; | |
3792 | ||
c59d6a55 JS |
3793 | return rv; |
3794 | } | |
3795 | ||
db1b32f6 SX |
3796 | /* |
3797 | * Return the exact number of visible CPUs based on CPU quotas. | |
3798 | * If there is no quota set, zero is returned. | |
3799 | */ | |
3800 | static double exact_cpu_count(const char *cg) | |
3801 | { | |
3802 | double rv; | |
3803 | int nprocs; | |
3804 | int64_t cfs_quota, cfs_period; | |
3805 | ||
3806 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) | |
3807 | return 0; | |
3808 | ||
3809 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
3810 | return 0; | |
3811 | ||
3812 | if (cfs_quota <= 0 || cfs_period <= 0) | |
3813 | return 0; | |
3814 | ||
3815 | rv = (double)cfs_quota / (double)cfs_period; | |
3816 | ||
3817 | nprocs = get_nprocs(); | |
3818 | ||
3819 | if (rv > nprocs) | |
3820 | rv = nprocs; | |
3821 | ||
3822 | return rv; | |
3823 | } | |
3824 | ||
c59d6a55 JS |
3825 | /* |
3826 | * Determine whether CPU views should be used or not. | |
3827 | */ | |
3828 | bool use_cpuview(const char *cg) | |
3829 | { | |
3830 | int cfd; | |
3831 | char *tmpc; | |
3832 | ||
3833 | tmpc = find_mounted_controller("cpu", &cfd); | |
3834 | if (!tmpc) | |
3835 | return false; | |
3836 | ||
3837 | tmpc = find_mounted_controller("cpuacct", &cfd); | |
3838 | if (!tmpc) | |
3839 | return false; | |
3840 | ||
3841 | return true; | |
3842 | } | |
3843 | ||
237e200e SH |
3844 | /* |
3845 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3846 | */ | |
3847 | static bool is_processor_line(const char *line) | |
3848 | { | |
3849 | int cpu; | |
3850 | ||
3851 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3852 | return true; | |
3853 | return false; | |
3854 | } | |
3855 | ||
3856 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3857 | struct fuse_file_info *fi) | |
3858 | { | |
3859 | struct fuse_context *fc = fuse_get_context(); | |
3860 | struct file_info *d = (struct file_info *)fi->fh; | |
3861 | char *cg; | |
3862 | char *cpuset = NULL; | |
3863 | char *line = NULL; | |
3864 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 | 3865 | bool am_printing = false, firstline = true, is_s390x = false; |
c59d6a55 JS |
3866 | int curcpu = -1, cpu, max_cpus = 0; |
3867 | bool use_view; | |
237e200e SH |
3868 | char *cache = d->buf; |
3869 | size_t cache_size = d->buflen; | |
3870 | FILE *f = NULL; | |
3871 | ||
3872 | if (offset){ | |
3873 | if (offset > d->size) | |
3874 | return -EINVAL; | |
3875 | if (!d->cached) | |
3876 | return 0; | |
3877 | int left = d->size - offset; | |
3878 | total_len = left > size ? size: left; | |
3879 | memcpy(buf, cache + offset, total_len); | |
3880 | return total_len; | |
3881 | } | |
3882 | ||
3883 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 3884 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
3885 | initpid = fc->pid; |
3886 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3887 | if (!cg) | |
3888 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3889 | prune_init_slice(cg); |
237e200e SH |
3890 | |
3891 | cpuset = get_cpuset(cg); | |
3892 | if (!cpuset) | |
3893 | goto err; | |
3894 | ||
c59d6a55 JS |
3895 | use_view = use_cpuview(cg); |
3896 | ||
3897 | if (use_view) | |
3898 | max_cpus = max_cpu_count(cg); | |
3899 | ||
237e200e SH |
3900 | f = fopen("/proc/cpuinfo", "r"); |
3901 | if (!f) | |
3902 | goto err; | |
3903 | ||
3904 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3905 | ssize_t l; |
f676eb79 SH |
3906 | if (firstline) { |
3907 | firstline = false; | |
3908 | if (strstr(line, "IBM/S390") != NULL) { | |
3909 | is_s390x = true; | |
3910 | am_printing = true; | |
5ed9d4e2 | 3911 | continue; |
f676eb79 SH |
3912 | } |
3913 | } | |
5ed9d4e2 SH |
3914 | if (strncmp(line, "# processors:", 12) == 0) |
3915 | continue; | |
237e200e | 3916 | if (is_processor_line(line)) { |
c59d6a55 JS |
3917 | if (use_view && max_cpus > 0 && (curcpu+1) == max_cpus) |
3918 | break; | |
237e200e SH |
3919 | am_printing = cpuline_in_cpuset(line, cpuset); |
3920 | if (am_printing) { | |
3921 | curcpu ++; | |
3922 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3923 | if (l < 0) { | |
3924 | perror("Error writing to cache"); | |
3925 | rv = 0; | |
3926 | goto err; | |
3927 | } | |
3928 | if (l >= cache_size) { | |
b8defc3d | 3929 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3930 | rv = 0; |
3931 | goto err; | |
3932 | } | |
3933 | cache += l; | |
3934 | cache_size -= l; | |
3935 | total_len += l; | |
3936 | } | |
3937 | continue; | |
f676eb79 SH |
3938 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3939 | char *p; | |
c59d6a55 JS |
3940 | if (use_view && max_cpus > 0 && (curcpu+1) == max_cpus) |
3941 | break; | |
f676eb79 SH |
3942 | if (!cpu_in_cpuset(cpu, cpuset)) |
3943 | continue; | |
3944 | curcpu ++; | |
3945 | p = strchr(line, ':'); | |
3946 | if (!p || !*p) | |
3947 | goto err; | |
3948 | p++; | |
5ed9d4e2 | 3949 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3950 | if (l < 0) { |
3951 | perror("Error writing to cache"); | |
3952 | rv = 0; | |
3953 | goto err; | |
3954 | } | |
3955 | if (l >= cache_size) { | |
b8defc3d | 3956 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
f676eb79 SH |
3957 | rv = 0; |
3958 | goto err; | |
3959 | } | |
3960 | cache += l; | |
3961 | cache_size -= l; | |
3962 | total_len += l; | |
3963 | continue; | |
3964 | ||
237e200e SH |
3965 | } |
3966 | if (am_printing) { | |
3967 | l = snprintf(cache, cache_size, "%s", line); | |
3968 | if (l < 0) { | |
3969 | perror("Error writing to cache"); | |
3970 | rv = 0; | |
3971 | goto err; | |
3972 | } | |
3973 | if (l >= cache_size) { | |
b8defc3d | 3974 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3975 | rv = 0; |
3976 | goto err; | |
3977 | } | |
3978 | cache += l; | |
3979 | cache_size -= l; | |
3980 | total_len += l; | |
3981 | } | |
3982 | } | |
3983 | ||
5ed9d4e2 SH |
3984 | if (is_s390x) { |
3985 | char *origcache = d->buf; | |
a262ddb7 | 3986 | ssize_t l; |
5ed9d4e2 SH |
3987 | do { |
3988 | d->buf = malloc(d->buflen); | |
3989 | } while (!d->buf); | |
3990 | cache = d->buf; | |
3991 | cache_size = d->buflen; | |
3992 | total_len = 0; | |
3993 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3994 | if (l < 0 || l >= cache_size) { | |
3995 | free(origcache); | |
3996 | goto err; | |
3997 | } | |
3998 | cache_size -= l; | |
3999 | cache += l; | |
4000 | total_len += l; | |
4001 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
4002 | if (l < 0 || l >= cache_size) { | |
4003 | free(origcache); | |
4004 | goto err; | |
4005 | } | |
4006 | cache_size -= l; | |
4007 | cache += l; | |
4008 | total_len += l; | |
4009 | l = snprintf(cache, cache_size, "%s", origcache); | |
4010 | free(origcache); | |
4011 | if (l < 0 || l >= cache_size) | |
4012 | goto err; | |
4013 | total_len += l; | |
4014 | } | |
4015 | ||
237e200e SH |
4016 | d->cached = 1; |
4017 | d->size = total_len; | |
4018 | if (total_len > size ) total_len = size; | |
4019 | ||
4020 | /* read from off 0 */ | |
4021 | memcpy(buf, d->buf, total_len); | |
4022 | rv = total_len; | |
4023 | err: | |
4024 | if (f) | |
4025 | fclose(f); | |
4026 | free(line); | |
4027 | free(cpuset); | |
4028 | free(cg); | |
4029 | return rv; | |
4030 | } | |
4031 | ||
0ecddf02 | 4032 | static uint64_t get_reaper_start_time(pid_t pid) |
9ac264cf | 4033 | { |
9ac264cf | 4034 | int ret; |
0ecddf02 CB |
4035 | FILE *f; |
4036 | uint64_t starttime; | |
4037 | /* strlen("/proc/") = 6 | |
4038 | * + | |
4039 | * LXCFS_NUMSTRLEN64 | |
4040 | * + | |
4041 | * strlen("/stat") = 5 | |
4042 | * + | |
4043 | * \0 = 1 | |
4044 | * */ | |
4045 | #define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1) | |
4046 | char path[__PROC_PID_STAT_LEN]; | |
9ac264cf JB |
4047 | pid_t qpid; |
4048 | ||
4049 | qpid = lookup_initpid_in_store(pid); | |
0ecddf02 CB |
4050 | if (qpid <= 0) { |
4051 | /* Caller can check for EINVAL on 0. */ | |
4052 | errno = EINVAL; | |
9ac264cf | 4053 | return 0; |
0ecddf02 | 4054 | } |
9ac264cf | 4055 | |
0ecddf02 CB |
4056 | ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid); |
4057 | if (ret < 0 || ret >= __PROC_PID_STAT_LEN) { | |
4058 | /* Caller can check for EINVAL on 0. */ | |
4059 | errno = EINVAL; | |
9ac264cf | 4060 | return 0; |
0ecddf02 | 4061 | } |
9ac264cf | 4062 | |
0ecddf02 CB |
4063 | f = fopen(path, "r"); |
4064 | if (!f) { | |
4065 | /* Caller can check for EINVAL on 0. */ | |
4066 | errno = EINVAL; | |
9ac264cf | 4067 | return 0; |
0ecddf02 | 4068 | } |
9ac264cf | 4069 | |
0ecddf02 CB |
4070 | /* Note that the *scanf() argument supression requires that length |
4071 | * modifiers such as "l" are omitted. Otherwise some compilers will yell | |
4072 | * at us. It's like telling someone you're not married and then asking | |
4073 | * if you can bring your wife to the party. | |
4074 | */ | |
4075 | ret = fscanf(f, "%*d " /* (1) pid %d */ | |
4076 | "%*s " /* (2) comm %s */ | |
4077 | "%*c " /* (3) state %c */ | |
4078 | "%*d " /* (4) ppid %d */ | |
4079 | "%*d " /* (5) pgrp %d */ | |
4080 | "%*d " /* (6) session %d */ | |
4081 | "%*d " /* (7) tty_nr %d */ | |
4082 | "%*d " /* (8) tpgid %d */ | |
4083 | "%*u " /* (9) flags %u */ | |
4084 | "%*u " /* (10) minflt %lu */ | |
4085 | "%*u " /* (11) cminflt %lu */ | |
4086 | "%*u " /* (12) majflt %lu */ | |
4087 | "%*u " /* (13) cmajflt %lu */ | |
4088 | "%*u " /* (14) utime %lu */ | |
4089 | "%*u " /* (15) stime %lu */ | |
4090 | "%*d " /* (16) cutime %ld */ | |
4091 | "%*d " /* (17) cstime %ld */ | |
4092 | "%*d " /* (18) priority %ld */ | |
4093 | "%*d " /* (19) nice %ld */ | |
4094 | "%*d " /* (20) num_threads %ld */ | |
4095 | "%*d " /* (21) itrealvalue %ld */ | |
4096 | "%" PRIu64, /* (22) starttime %llu */ | |
4097 | &starttime); | |
4098 | if (ret != 1) { | |
4099 | fclose(f); | |
4100 | /* Caller can check for EINVAL on 0. */ | |
4101 | errno = EINVAL; | |
4102 | return 0; | |
4103 | } | |
4104 | ||
4105 | fclose(f); | |
4106 | ||
4107 | errno = 0; | |
4108 | return starttime; | |
4109 | } | |
4110 | ||
1c4b4e38 | 4111 | static double get_reaper_start_time_in_sec(pid_t pid) |
0ecddf02 | 4112 | { |
1c4b4e38 CB |
4113 | uint64_t clockticks, ticks_per_sec; |
4114 | int64_t ret; | |
4115 | double res = 0; | |
0ecddf02 CB |
4116 | |
4117 | clockticks = get_reaper_start_time(pid); | |
4118 | if (clockticks == 0 && errno == EINVAL) { | |
4119 | lxcfs_debug("failed to retrieve start time of pid %d\n", pid); | |
4120 | return 0; | |
4121 | } | |
4122 | ||
1c4b4e38 CB |
4123 | ret = sysconf(_SC_CLK_TCK); |
4124 | if (ret < 0 && errno == EINVAL) { | |
0ecddf02 CB |
4125 | lxcfs_debug( |
4126 | "%s\n", | |
4127 | "failed to determine number of clock ticks in a second"); | |
4128 | return 0; | |
4129 | } | |
4130 | ||
1c4b4e38 CB |
4131 | ticks_per_sec = (uint64_t)ret; |
4132 | res = (double)clockticks / ticks_per_sec; | |
4133 | return res; | |
0ecddf02 CB |
4134 | } |
4135 | ||
1c4b4e38 | 4136 | static double get_reaper_age(pid_t pid) |
0ecddf02 | 4137 | { |
1c4b4e38 CB |
4138 | uint64_t uptime_ms; |
4139 | double procstart, procage; | |
0ecddf02 CB |
4140 | |
4141 | /* We need to substract the time the process has started since system | |
4142 | * boot minus the time when the system has started to get the actual | |
4143 | * reaper age. | |
4144 | */ | |
4145 | procstart = get_reaper_start_time_in_sec(pid); | |
4146 | procage = procstart; | |
4147 | if (procstart > 0) { | |
4148 | int ret; | |
4149 | struct timespec spec; | |
4150 | ||
4151 | ret = clock_gettime(CLOCK_BOOTTIME, &spec); | |
4152 | if (ret < 0) | |
4153 | return 0; | |
1c4b4e38 | 4154 | |
0ecddf02 CB |
4155 | /* We could make this more precise here by using the tv_nsec |
4156 | * field in the timespec struct and convert it to milliseconds | |
4157 | * and then create a double for the seconds and milliseconds but | |
4158 | * that seems more work than it is worth. | |
4159 | */ | |
1c4b4e38 CB |
4160 | uptime_ms = (spec.tv_sec * 1000) + (spec.tv_nsec * 1e-6); |
4161 | procage = (uptime_ms - (procstart * 1000)) / 1000; | |
0ecddf02 CB |
4162 | } |
4163 | ||
4164 | return procage; | |
4165 | } | |
4166 | ||
8be92dd1 JS |
4167 | /* |
4168 | * Returns 0 on success. | |
4169 | * It is the caller's responsibility to free `return_usage`, unless this | |
4170 | * function returns an error. | |
4171 | */ | |
79612c8b | 4172 | static int read_cpuacct_usage_all(char *cg, char *cpuset, struct cpuacct_usage **return_usage, int *size) |
8be92dd1 | 4173 | { |
77005a6c | 4174 | int cpucount = get_nprocs_conf(); |
8be92dd1 | 4175 | struct cpuacct_usage *cpu_usage; |
db1b32f6 | 4176 | int rv = 0, i, j, ret; |
8be92dd1 JS |
4177 | int cg_cpu; |
4178 | uint64_t cg_user, cg_system; | |
4179 | int64_t ticks_per_sec; | |
4180 | char *usage_str = NULL; | |
4181 | ||
4182 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
4183 | ||
4184 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
db1b32f6 | 4185 | lxcfs_v( |
8be92dd1 JS |
4186 | "%s\n", |
4187 | "read_cpuacct_usage_all failed to determine number of clock ticks " | |
4188 | "in a second"); | |
4189 | return -1; | |
4190 | } | |
4191 | ||
4192 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); | |
4193 | if (!cpu_usage) | |
4194 | return -ENOMEM; | |
4195 | ||
db1b32f6 | 4196 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); |
8be92dd1 | 4197 | if (!cgfs_get_value("cpuacct", cg, "cpuacct.usage_all", &usage_str)) { |
db1b32f6 SX |
4198 | // read cpuacct.usage_percpu instead |
4199 | lxcfs_v("failed to read cpuacct.usage_all. reading cpuacct.usage_percpu instead\n%s", ""); | |
4200 | if (!cgfs_get_value("cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) { | |
4201 | rv = -1; | |
4202 | goto err; | |
4203 | } | |
4204 | lxcfs_v("usage_str: %s\n", usage_str); | |
4205 | ||
4206 | // convert cpuacct.usage_percpu into cpuacct.usage_all | |
4207 | lxcfs_v("converting cpuacct.usage_percpu into cpuacct.usage_all\n%s", ""); | |
1c4b4e38 | 4208 | |
db1b32f6 SX |
4209 | char *data = NULL; |
4210 | size_t sz = 0, asz = 0; | |
4211 | ||
4212 | must_strcat(&data, &sz, &asz, "cpu user system\n"); | |
4213 | ||
4214 | int i = 0, read_pos = 0, read_cnt=0; | |
4215 | while (sscanf(usage_str + read_pos, "%lu %n", &cg_user, &read_cnt) > 0) { | |
4216 | lxcfs_debug("i: %d, cg_user: %lu, read_pos: %d, read_cnt: %d\n", i, cg_user, read_pos, read_cnt); | |
4217 | must_strcat(&data, &sz, &asz, "%d %lu 0\n", i, cg_user); | |
4218 | i++; | |
4219 | read_pos += read_cnt; | |
4220 | } | |
4221 | ||
4222 | free(usage_str); | |
4223 | usage_str = data; | |
4224 | ||
4225 | lxcfs_v("usage_str: %s\n", usage_str); | |
8be92dd1 JS |
4226 | } |
4227 | ||
db1b32f6 | 4228 | int read_pos = 0, read_cnt=0; |
8be92dd1 JS |
4229 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) { |
4230 | lxcfs_error("read_cpuacct_usage_all reading first line from " | |
4231 | "%s/cpuacct.usage_all failed.\n", cg); | |
4232 | rv = -1; | |
4233 | goto err; | |
4234 | } | |
4235 | ||
4236 | read_pos += read_cnt; | |
4237 | ||
4238 | for (i = 0, j = 0; i < cpucount; i++) { | |
4239 | ret = sscanf(usage_str + read_pos, "%d %lu %lu\n%n", &cg_cpu, &cg_user, | |
4240 | &cg_system, &read_cnt); | |
4241 | ||
4242 | if (ret == EOF) | |
4243 | break; | |
4244 | ||
4245 | if (ret != 3) { | |
4246 | lxcfs_error("read_cpuacct_usage_all reading from %s/cpuacct.usage_all " | |
4247 | "failed.\n", cg); | |
4248 | rv = -1; | |
4249 | goto err; | |
4250 | } | |
4251 | ||
4252 | read_pos += read_cnt; | |
4253 | ||
8be92dd1 JS |
4254 | /* Convert the time from nanoseconds to USER_HZ */ |
4255 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
4256 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
4257 | j++; | |
4258 | } | |
4259 | ||
4260 | rv = 0; | |
4261 | *return_usage = cpu_usage; | |
79612c8b | 4262 | *size = cpucount; |
8be92dd1 JS |
4263 | |
4264 | err: | |
4265 | if (usage_str) | |
4266 | free(usage_str); | |
4267 | ||
4268 | if (rv != 0) { | |
4269 | free(cpu_usage); | |
4270 | *return_usage = NULL; | |
4271 | } | |
4272 | ||
4273 | return rv; | |
4274 | } | |
4275 | ||
056adcef JS |
4276 | static unsigned long diff_cpu_usage(struct cpuacct_usage *older, struct cpuacct_usage *newer, struct cpuacct_usage *diff, int cpu_count) |
4277 | { | |
4278 | int i; | |
4279 | unsigned long sum = 0; | |
4280 | ||
4281 | for (i = 0; i < cpu_count; i++) { | |
77005a6c JS |
4282 | if (!newer[i].online) |
4283 | continue; | |
4284 | ||
056adcef JS |
4285 | /* When cpuset is changed on the fly, the CPUs might get reordered. |
4286 | * We could either reset all counters, or check that the substractions | |
4287 | * below will return expected results. | |
4288 | */ | |
4289 | if (newer[i].user > older[i].user) | |
4290 | diff[i].user = newer[i].user - older[i].user; | |
4291 | else | |
4292 | diff[i].user = 0; | |
4293 | ||
4294 | if (newer[i].system > older[i].system) | |
4295 | diff[i].system = newer[i].system - older[i].system; | |
4296 | else | |
4297 | diff[i].system = 0; | |
4298 | ||
4299 | if (newer[i].idle > older[i].idle) | |
4300 | diff[i].idle = newer[i].idle - older[i].idle; | |
4301 | else | |
4302 | diff[i].idle = 0; | |
4303 | ||
4304 | sum += diff[i].user; | |
4305 | sum += diff[i].system; | |
4306 | sum += diff[i].idle; | |
4307 | } | |
4308 | ||
4309 | return sum; | |
4310 | } | |
4311 | ||
4312 | static void add_cpu_usage(unsigned long *surplus, struct cpuacct_usage *usage, unsigned long *counter, unsigned long threshold) | |
4313 | { | |
4314 | unsigned long free_space, to_add; | |
4315 | ||
4316 | free_space = threshold - usage->user - usage->system; | |
4317 | ||
4318 | if (free_space > usage->idle) | |
4319 | free_space = usage->idle; | |
4320 | ||
4321 | to_add = free_space > *surplus ? *surplus : free_space; | |
4322 | ||
4323 | *counter += to_add; | |
4324 | usage->idle -= to_add; | |
4325 | *surplus -= to_add; | |
4326 | } | |
4327 | ||
951acc94 JS |
4328 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
4329 | { | |
4330 | struct cg_proc_stat *first = NULL, *prev, *tmp; | |
4331 | ||
4332 | for (prev = NULL; node; ) { | |
4333 | if (!cgfs_param_exist("cpu", node->cg, "cpu.shares")) { | |
4334 | tmp = node; | |
4335 | lxcfs_debug("Removing stat node for %s\n", node->cg); | |
4336 | ||
4337 | if (prev) | |
4338 | prev->next = node->next; | |
4339 | else | |
4340 | first = node->next; | |
4341 | ||
4342 | node = node->next; | |
4343 | free_proc_stat_node(tmp); | |
4344 | } else { | |
4345 | if (!first) | |
4346 | first = node; | |
4347 | prev = node; | |
4348 | node = node->next; | |
4349 | } | |
4350 | } | |
4351 | ||
4352 | return first; | |
4353 | } | |
4354 | ||
4355 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
4356 | static void prune_proc_stat_history(void) | |
4357 | { | |
4358 | int i; | |
4359 | time_t now = time(NULL); | |
4360 | ||
4361 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
2f49b662 JS |
4362 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
4363 | ||
4364 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
4365 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
951acc94 | 4366 | return; |
2f49b662 | 4367 | } |
951acc94 | 4368 | |
2f49b662 JS |
4369 | if (proc_stat_history[i]->next) { |
4370 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
4371 | proc_stat_history[i]->lastcheck = now; | |
4372 | } | |
951acc94 | 4373 | |
2f49b662 | 4374 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); |
951acc94 JS |
4375 | } |
4376 | } | |
4377 | ||
2f49b662 | 4378 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, const char *cg) |
056adcef | 4379 | { |
056adcef JS |
4380 | struct cg_proc_stat *node; |
4381 | ||
2f49b662 JS |
4382 | pthread_rwlock_rdlock(&head->lock); |
4383 | ||
4384 | if (!head->next) { | |
4385 | pthread_rwlock_unlock(&head->lock); | |
056adcef | 4386 | return NULL; |
2f49b662 | 4387 | } |
056adcef JS |
4388 | |
4389 | node = head->next; | |
4390 | ||
4391 | do { | |
4392 | if (strcmp(cg, node->cg) == 0) | |
951acc94 | 4393 | goto out; |
056adcef JS |
4394 | } while ((node = node->next)); |
4395 | ||
951acc94 JS |
4396 | node = NULL; |
4397 | ||
4398 | out: | |
2f49b662 | 4399 | pthread_rwlock_unlock(&head->lock); |
951acc94 JS |
4400 | prune_proc_stat_history(); |
4401 | return node; | |
056adcef JS |
4402 | } |
4403 | ||
4404 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) | |
4405 | { | |
4406 | struct cg_proc_stat *node; | |
4407 | int i; | |
4408 | ||
4409 | node = malloc(sizeof(struct cg_proc_stat)); | |
4410 | if (!node) | |
4411 | goto err; | |
4412 | ||
4413 | node->cg = NULL; | |
4414 | node->usage = NULL; | |
4415 | node->view = NULL; | |
4416 | ||
4417 | node->cg = malloc(strlen(cg) + 1); | |
4418 | if (!node->cg) | |
4419 | goto err; | |
4420 | ||
4421 | strcpy(node->cg, cg); | |
4422 | ||
4423 | node->usage = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4424 | if (!node->usage) | |
4425 | goto err; | |
4426 | ||
4427 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
4428 | ||
4429 | node->view = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4430 | if (!node->view) | |
4431 | goto err; | |
4432 | ||
4433 | node->cpu_count = cpu_count; | |
4434 | node->next = NULL; | |
4435 | ||
2f49b662 JS |
4436 | if (pthread_mutex_init(&node->lock, NULL) != 0) { |
4437 | lxcfs_error("%s\n", "Failed to initialize node lock"); | |
4438 | goto err; | |
4439 | } | |
4440 | ||
056adcef JS |
4441 | for (i = 0; i < cpu_count; i++) { |
4442 | node->view[i].user = 0; | |
4443 | node->view[i].system = 0; | |
4444 | node->view[i].idle = 0; | |
4445 | } | |
4446 | ||
4447 | return node; | |
4448 | ||
4449 | err: | |
4450 | if (node && node->cg) | |
4451 | free(node->cg); | |
4452 | if (node && node->usage) | |
4453 | free(node->usage); | |
4454 | if (node && node->view) | |
4455 | free(node->view); | |
4456 | if (node) | |
4457 | free(node); | |
4458 | ||
4459 | return NULL; | |
4460 | } | |
4461 | ||
2f49b662 | 4462 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
056adcef JS |
4463 | { |
4464 | int hash = calc_hash(new_node->cg) % CPUVIEW_HASH_SIZE; | |
4465 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
2f49b662 JS |
4466 | struct cg_proc_stat *node, *rv = new_node; |
4467 | ||
4468 | pthread_rwlock_wrlock(&head->lock); | |
056adcef JS |
4469 | |
4470 | if (!head->next) { | |
4471 | head->next = new_node; | |
2f49b662 | 4472 | goto out; |
056adcef JS |
4473 | } |
4474 | ||
2f49b662 JS |
4475 | node = head->next; |
4476 | ||
056adcef | 4477 | for (;;) { |
2f49b662 JS |
4478 | if (strcmp(node->cg, new_node->cg) == 0) { |
4479 | /* The node is already present, return it */ | |
4480 | free_proc_stat_node(new_node); | |
4481 | rv = node; | |
4482 | goto out; | |
4483 | } | |
056adcef JS |
4484 | |
4485 | if (node->next) { | |
4486 | node = node->next; | |
4487 | continue; | |
4488 | } | |
4489 | ||
4490 | node->next = new_node; | |
2f49b662 JS |
4491 | goto out; |
4492 | } | |
4493 | ||
4494 | out: | |
4495 | pthread_rwlock_unlock(&head->lock); | |
4496 | return rv; | |
4497 | } | |
4498 | ||
895f28e5 JS |
4499 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) |
4500 | { | |
4501 | struct cpuacct_usage *new_usage, *new_view; | |
4502 | int i; | |
4503 | ||
4504 | /* Allocate new memory */ | |
4505 | new_usage = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4506 | if (!new_usage) | |
4507 | return false; | |
4508 | ||
4509 | new_view = malloc(sizeof(struct cpuacct_usage) * cpu_count); | |
4510 | if (!new_view) { | |
4511 | free(new_usage); | |
4512 | return false; | |
4513 | } | |
4514 | ||
4515 | /* Copy existing data & initialize new elements */ | |
4516 | for (i = 0; i < cpu_count; i++) { | |
4517 | if (i < node->cpu_count) { | |
4518 | new_usage[i].user = node->usage[i].user; | |
4519 | new_usage[i].system = node->usage[i].system; | |
4520 | new_usage[i].idle = node->usage[i].idle; | |
4521 | ||
4522 | new_view[i].user = node->view[i].user; | |
4523 | new_view[i].system = node->view[i].system; | |
4524 | new_view[i].idle = node->view[i].idle; | |
4525 | } else { | |
4526 | new_usage[i].user = 0; | |
4527 | new_usage[i].system = 0; | |
4528 | new_usage[i].idle = 0; | |
4529 | ||
4530 | new_view[i].user = 0; | |
4531 | new_view[i].system = 0; | |
4532 | new_view[i].idle = 0; | |
4533 | } | |
4534 | } | |
4535 | ||
4536 | free(node->usage); | |
4537 | free(node->view); | |
4538 | ||
4539 | node->usage = new_usage; | |
4540 | node->view = new_view; | |
4541 | node->cpu_count = cpu_count; | |
4542 | ||
4543 | return true; | |
4544 | } | |
4545 | ||
2f49b662 JS |
4546 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) |
4547 | { | |
4548 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
4549 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
4550 | struct cg_proc_stat *node; | |
4551 | ||
4552 | node = find_proc_stat_node(head, cg); | |
4553 | ||
4554 | if (!node) { | |
4555 | node = new_proc_stat_node(usage, cpu_count, cg); | |
4556 | if (!node) | |
4557 | return NULL; | |
4558 | ||
4559 | node = add_proc_stat_node(node); | |
4560 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
056adcef | 4561 | } |
2f49b662 JS |
4562 | |
4563 | pthread_mutex_lock(&node->lock); | |
895f28e5 JS |
4564 | |
4565 | /* If additional CPUs on the host have been enabled, CPU usage counter | |
4566 | * arrays have to be expanded */ | |
4567 | if (node->cpu_count < cpu_count) { | |
4568 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
4569 | node->cpu_count, cpu_count, cg); | |
4570 | ||
4571 | if (!expand_proc_stat_node(node, cpu_count)) { | |
4572 | pthread_mutex_unlock(&node->lock); | |
4573 | lxcfs_debug("Unable to expand stat node %d->%d for %s\n", | |
4574 | node->cpu_count, cpu_count, cg); | |
4575 | return NULL; | |
4576 | } | |
4577 | } | |
4578 | ||
2f49b662 | 4579 | return node; |
056adcef JS |
4580 | } |
4581 | ||
4582 | static void reset_proc_stat_node(struct cg_proc_stat *node, struct cpuacct_usage *usage, int cpu_count) | |
4583 | { | |
4584 | int i; | |
4585 | ||
4586 | lxcfs_debug("Resetting stat node for %s\n", node->cg); | |
4587 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
4588 | ||
4589 | for (i = 0; i < cpu_count; i++) { | |
4590 | node->view[i].user = 0; | |
4591 | node->view[i].system = 0; | |
4592 | node->view[i].idle = 0; | |
4593 | } | |
4594 | ||
4595 | node->cpu_count = cpu_count; | |
4596 | } | |
4597 | ||
79612c8b | 4598 | static int cpuview_proc_stat(const char *cg, const char *cpuset, struct cpuacct_usage *cg_cpu_usage, int cg_cpu_usage_size, FILE *f, char *buf, size_t buf_size) |
056adcef JS |
4599 | { |
4600 | char *line = NULL; | |
4601 | size_t linelen = 0, total_len = 0, rv = 0, l; | |
4602 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
77005a6c | 4603 | int physcpu, i; |
056adcef JS |
4604 | int max_cpus = max_cpu_count(cg), cpu_cnt = 0; |
4605 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
4606 | unsigned long user_sum = 0, system_sum = 0, idle_sum = 0; | |
4607 | unsigned long user_surplus = 0, system_surplus = 0; | |
4608 | unsigned long total_sum, threshold; | |
4609 | struct cg_proc_stat *stat_node; | |
4610 | struct cpuacct_usage *diff = NULL; | |
77005a6c | 4611 | int nprocs = get_nprocs_conf(); |
056adcef | 4612 | |
79612c8b JS |
4613 | if (cg_cpu_usage_size < nprocs) |
4614 | nprocs = cg_cpu_usage_size; | |
4615 | ||
056adcef JS |
4616 | /* Read all CPU stats and stop when we've encountered other lines */ |
4617 | while (getline(&line, &linelen, f) != -1) { | |
77005a6c | 4618 | int ret; |
056adcef JS |
4619 | char cpu_char[10]; /* That's a lot of cores */ |
4620 | uint64_t all_used, cg_used; | |
4621 | ||
4622 | if (strlen(line) == 0) | |
4623 | continue; | |
4624 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { | |
4625 | /* not a ^cpuN line containing a number N */ | |
4626 | break; | |
4627 | } | |
4628 | ||
77005a6c | 4629 | if (sscanf(cpu_char, "%d", &physcpu) != 1) |
056adcef | 4630 | continue; |
77005a6c | 4631 | |
79612c8b JS |
4632 | if (physcpu >= cg_cpu_usage_size) |
4633 | continue; | |
4634 | ||
056adcef JS |
4635 | curcpu ++; |
4636 | cpu_cnt ++; | |
4637 | ||
77005a6c JS |
4638 | if (!cpu_in_cpuset(physcpu, cpuset)) { |
4639 | for (i = curcpu; i <= physcpu; i++) { | |
4640 | cg_cpu_usage[i].online = false; | |
4641 | } | |
4642 | continue; | |
4643 | } | |
4644 | ||
4645 | if (curcpu < physcpu) { | |
4646 | /* Some CPUs may be disabled */ | |
4647 | for (i = curcpu; i < physcpu; i++) | |
4648 | cg_cpu_usage[i].online = false; | |
4649 | ||
4650 | curcpu = physcpu; | |
4651 | } | |
4652 | ||
4653 | cg_cpu_usage[curcpu].online = true; | |
4654 | ||
056adcef JS |
4655 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
4656 | &user, | |
4657 | &nice, | |
4658 | &system, | |
4659 | &idle, | |
4660 | &iowait, | |
4661 | &irq, | |
4662 | &softirq, | |
4663 | &steal, | |
4664 | &guest, | |
4665 | &guest_nice); | |
4666 | ||
4667 | if (ret != 10) | |
4668 | continue; | |
4669 | ||
4670 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
4671 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
4672 | ||
4673 | if (all_used >= cg_used) { | |
4674 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
4675 | ||
4676 | } else { | |
4677 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
4678 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
4679 | curcpu, cg, all_used, cg_used); | |
4680 | cg_cpu_usage[curcpu].idle = idle; | |
4681 | } | |
4682 | } | |
4683 | ||
4684 | /* Cannot use more CPUs than is available due to cpuset */ | |
4685 | if (max_cpus > cpu_cnt) | |
4686 | max_cpus = cpu_cnt; | |
4687 | ||
2f49b662 | 4688 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
056adcef JS |
4689 | |
4690 | if (!stat_node) { | |
2f49b662 JS |
4691 | lxcfs_error("unable to find/create stat node for %s\n", cg); |
4692 | rv = 0; | |
4693 | goto err; | |
056adcef JS |
4694 | } |
4695 | ||
4696 | diff = malloc(sizeof(struct cpuacct_usage) * nprocs); | |
4697 | if (!diff) { | |
4698 | rv = 0; | |
4699 | goto err; | |
4700 | } | |
4701 | ||
4702 | /* | |
4703 | * If the new values are LOWER than values stored in memory, it means | |
4704 | * the cgroup has been reset/recreated and we should reset too. | |
4705 | */ | |
77005a6c JS |
4706 | for (curcpu = 0; curcpu < nprocs; curcpu++) { |
4707 | if (!cg_cpu_usage[curcpu].online) | |
4708 | continue; | |
4709 | ||
4710 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
4711 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
4712 | ||
4713 | break; | |
4714 | } | |
056adcef | 4715 | |
77005a6c JS |
4716 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); |
4717 | ||
4718 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
4719 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
4720 | ||
4721 | if (!stat_node->usage[curcpu].online) | |
4722 | continue; | |
4723 | ||
4724 | i++; | |
056adcef | 4725 | |
056adcef JS |
4726 | stat_node->usage[curcpu].user += diff[curcpu].user; |
4727 | stat_node->usage[curcpu].system += diff[curcpu].system; | |
4728 | stat_node->usage[curcpu].idle += diff[curcpu].idle; | |
4729 | ||
77005a6c | 4730 | if (max_cpus > 0 && i >= max_cpus) { |
056adcef JS |
4731 | user_surplus += diff[curcpu].user; |
4732 | system_surplus += diff[curcpu].system; | |
4733 | } | |
4734 | } | |
4735 | ||
4736 | /* Calculate usage counters of visible CPUs */ | |
4737 | if (max_cpus > 0) { | |
4738 | /* threshold = maximum usage per cpu, including idle */ | |
4739 | threshold = total_sum / cpu_cnt * max_cpus; | |
4740 | ||
77005a6c | 4741 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
77005a6c JS |
4742 | if (!stat_node->usage[curcpu].online) |
4743 | continue; | |
4744 | ||
4745 | i++; | |
4746 | ||
db1b32f6 SX |
4747 | if (i == max_cpus) |
4748 | break; | |
4749 | ||
056adcef JS |
4750 | if (diff[curcpu].user + diff[curcpu].system >= threshold) |
4751 | continue; | |
4752 | ||
4753 | /* Add user */ | |
4754 | add_cpu_usage( | |
4755 | &user_surplus, | |
4756 | &diff[curcpu], | |
4757 | &diff[curcpu].user, | |
4758 | threshold); | |
4759 | ||
4760 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
4761 | continue; | |
4762 | ||
4763 | /* If there is still room, add system */ | |
4764 | add_cpu_usage( | |
4765 | &system_surplus, | |
4766 | &diff[curcpu], | |
4767 | &diff[curcpu].system, | |
4768 | threshold); | |
4769 | } | |
4770 | ||
4771 | if (user_surplus > 0) | |
4772 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
4773 | if (system_surplus > 0) | |
4774 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
4775 | ||
db1b32f6 SX |
4776 | unsigned long diff_user = 0; |
4777 | unsigned long diff_system = 0; | |
4778 | unsigned long diff_idle = 0; | |
4779 | unsigned long max_diff_idle = 0; | |
4780 | unsigned long max_diff_idle_index = 0; | |
77005a6c | 4781 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
77005a6c JS |
4782 | if (!stat_node->usage[curcpu].online) |
4783 | continue; | |
4784 | ||
4785 | i++; | |
4786 | ||
db1b32f6 SX |
4787 | if (i == max_cpus) |
4788 | break; | |
4789 | ||
056adcef JS |
4790 | stat_node->view[curcpu].user += diff[curcpu].user; |
4791 | stat_node->view[curcpu].system += diff[curcpu].system; | |
4792 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
4793 | ||
4794 | user_sum += stat_node->view[curcpu].user; | |
4795 | system_sum += stat_node->view[curcpu].system; | |
4796 | idle_sum += stat_node->view[curcpu].idle; | |
056adcef | 4797 | |
db1b32f6 SX |
4798 | diff_user += diff[curcpu].user; |
4799 | diff_system += diff[curcpu].system; | |
4800 | diff_idle += diff[curcpu].idle; | |
4801 | if (diff[curcpu].idle > max_diff_idle) { | |
4802 | max_diff_idle = diff[curcpu].idle; | |
4803 | max_diff_idle_index = curcpu; | |
4804 | } | |
4805 | ||
4806 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
4807 | } | |
4808 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
4809 | ||
4810 | // revise cpu usage view to support partial cpu case | |
4811 | double exact_cpus = exact_cpu_count(cg); | |
4812 | if (exact_cpus < (double)max_cpus){ | |
4813 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
4814 | unsigned long delta = (unsigned long)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); | |
4815 | lxcfs_v("delta: %lu\n", delta); | |
4816 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
4817 | idle_sum = idle_sum > delta ? idle_sum - delta : 0; | |
4818 | lxcfs_v("idle_sum after: %lu\n", idle_sum); | |
1c4b4e38 | 4819 | |
db1b32f6 SX |
4820 | curcpu = max_diff_idle_index; |
4821 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
4822 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle > delta ? stat_node->view[curcpu].idle - delta : 0; | |
4823 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
4824 | } | |
056adcef | 4825 | } else { |
77005a6c JS |
4826 | for (curcpu = 0; curcpu < nprocs; curcpu++) { |
4827 | if (!stat_node->usage[curcpu].online) | |
4828 | continue; | |
4829 | ||
056adcef JS |
4830 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; |
4831 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
4832 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
4833 | ||
4834 | user_sum += stat_node->view[curcpu].user; | |
4835 | system_sum += stat_node->view[curcpu].system; | |
4836 | idle_sum += stat_node->view[curcpu].idle; | |
4837 | } | |
4838 | } | |
4839 | ||
4840 | /* Render the file */ | |
4841 | /* cpu-all */ | |
4842 | l = snprintf(buf, buf_size, "cpu %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
4843 | user_sum, | |
4844 | system_sum, | |
4845 | idle_sum); | |
db1b32f6 | 4846 | lxcfs_v("cpu-all: %s\n", buf); |
056adcef JS |
4847 | |
4848 | if (l < 0) { | |
4849 | perror("Error writing to cache"); | |
4850 | rv = 0; | |
4851 | goto err; | |
056adcef JS |
4852 | } |
4853 | if (l >= buf_size) { | |
4854 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4855 | rv = 0; | |
4856 | goto err; | |
4857 | } | |
4858 | ||
4859 | buf += l; | |
4860 | buf_size -= l; | |
4861 | total_len += l; | |
4862 | ||
4863 | /* Render visible CPUs */ | |
77005a6c JS |
4864 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { |
4865 | if (!stat_node->usage[curcpu].online) | |
4866 | continue; | |
4867 | ||
4868 | i++; | |
4869 | ||
4870 | if (max_cpus > 0 && i == max_cpus) | |
056adcef JS |
4871 | break; |
4872 | ||
4873 | l = snprintf(buf, buf_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
77005a6c | 4874 | i, |
056adcef JS |
4875 | stat_node->view[curcpu].user, |
4876 | stat_node->view[curcpu].system, | |
4877 | stat_node->view[curcpu].idle); | |
db1b32f6 | 4878 | lxcfs_v("cpu: %s\n", buf); |
056adcef JS |
4879 | |
4880 | if (l < 0) { | |
4881 | perror("Error writing to cache"); | |
4882 | rv = 0; | |
4883 | goto err; | |
4884 | ||
4885 | } | |
4886 | if (l >= buf_size) { | |
4887 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4888 | rv = 0; | |
4889 | goto err; | |
4890 | } | |
4891 | ||
4892 | buf += l; | |
4893 | buf_size -= l; | |
4894 | total_len += l; | |
4895 | } | |
4896 | ||
4897 | /* Pass the rest of /proc/stat, start with the last line read */ | |
4898 | l = snprintf(buf, buf_size, "%s", line); | |
4899 | ||
4900 | if (l < 0) { | |
4901 | perror("Error writing to cache"); | |
4902 | rv = 0; | |
4903 | goto err; | |
4904 | ||
4905 | } | |
4906 | if (l >= buf_size) { | |
4907 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4908 | rv = 0; | |
4909 | goto err; | |
4910 | } | |
4911 | ||
4912 | buf += l; | |
4913 | buf_size -= l; | |
4914 | total_len += l; | |
4915 | ||
4916 | /* Pass the rest of the host's /proc/stat */ | |
4917 | while (getline(&line, &linelen, f) != -1) { | |
4918 | l = snprintf(buf, buf_size, "%s", line); | |
4919 | if (l < 0) { | |
4920 | perror("Error writing to cache"); | |
4921 | rv = 0; | |
4922 | goto err; | |
4923 | } | |
4924 | if (l >= buf_size) { | |
4925 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4926 | rv = 0; | |
4927 | goto err; | |
4928 | } | |
4929 | buf += l; | |
4930 | buf_size -= l; | |
4931 | total_len += l; | |
4932 | } | |
4933 | ||
4934 | rv = total_len; | |
4935 | ||
4936 | err: | |
2f49b662 JS |
4937 | if (stat_node) |
4938 | pthread_mutex_unlock(&stat_node->lock); | |
056adcef JS |
4939 | if (line) |
4940 | free(line); | |
4941 | if (diff) | |
4942 | free(diff); | |
4943 | return rv; | |
4944 | } | |
4945 | ||
f34de69a | 4946 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) |
237e200e SH |
4947 | static int proc_stat_read(char *buf, size_t size, off_t offset, |
4948 | struct fuse_file_info *fi) | |
4949 | { | |
4950 | struct fuse_context *fc = fuse_get_context(); | |
4951 | struct file_info *d = (struct file_info *)fi->fh; | |
4952 | char *cg; | |
4953 | char *cpuset = NULL; | |
4954 | char *line = NULL; | |
4955 | size_t linelen = 0, total_len = 0, rv = 0; | |
4956 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
77005a6c | 4957 | int physcpu = 0; |
7144f069 | 4958 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; |
237e200e | 4959 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, |
7144f069 | 4960 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; |
237e200e SH |
4961 | char cpuall[CPUALL_MAX_SIZE]; |
4962 | /* reserve for cpu all */ | |
4963 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
4964 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
4965 | FILE *f = NULL; | |
8be92dd1 | 4966 | struct cpuacct_usage *cg_cpu_usage = NULL; |
79612c8b | 4967 | int cg_cpu_usage_size = 0; |
237e200e SH |
4968 | |
4969 | if (offset){ | |
4970 | if (offset > d->size) | |
4971 | return -EINVAL; | |
4972 | if (!d->cached) | |
4973 | return 0; | |
4974 | int left = d->size - offset; | |
4975 | total_len = left > size ? size: left; | |
4976 | memcpy(buf, d->buf + offset, total_len); | |
4977 | return total_len; | |
4978 | } | |
4979 | ||
4980 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
db1b32f6 | 4981 | lxcfs_v("initpid: %d\n", initpid); |
237e200e SH |
4982 | if (initpid <= 0) |
4983 | initpid = fc->pid; | |
a4bd57ae HY |
4984 | |
4985 | /* | |
4986 | * when container run with host pid namespace initpid == 1, cgroup will "/" | |
4987 | * we should return host os's /proc contents. | |
4988 | * in some case cpuacct_usage.all in "/" will larger then /proc/stat | |
4989 | */ | |
4990 | if (initpid == 1) { | |
4991 | return read_file("/proc/stat", buf, size, d); | |
4992 | } | |
4993 | ||
237e200e | 4994 | cg = get_pid_cgroup(initpid, "cpuset"); |
db1b32f6 | 4995 | lxcfs_v("cg: %s\n", cg); |
237e200e SH |
4996 | if (!cg) |
4997 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 4998 | prune_init_slice(cg); |
237e200e SH |
4999 | |
5000 | cpuset = get_cpuset(cg); | |
5001 | if (!cpuset) | |
5002 | goto err; | |
5003 | ||
8be92dd1 JS |
5004 | /* |
5005 | * Read cpuacct.usage_all for all CPUs. | |
5006 | * If the cpuacct cgroup is present, it is used to calculate the container's | |
5007 | * CPU usage. If not, values from the host's /proc/stat are used. | |
5008 | */ | |
79612c8b | 5009 | if (read_cpuacct_usage_all(cg, cpuset, &cg_cpu_usage, &cg_cpu_usage_size) != 0) { |
db1b32f6 | 5010 | lxcfs_v("%s\n", "proc_stat_read failed to read from cpuacct, " |
8be92dd1 JS |
5011 | "falling back to the host's /proc/stat"); |
5012 | } | |
5013 | ||
237e200e SH |
5014 | f = fopen("/proc/stat", "r"); |
5015 | if (!f) | |
5016 | goto err; | |
5017 | ||
5018 | //skip first line | |
5019 | if (getline(&line, &linelen, f) < 0) { | |
b8defc3d | 5020 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); |
237e200e SH |
5021 | goto err; |
5022 | } | |
5023 | ||
056adcef | 5024 | if (use_cpuview(cg) && cg_cpu_usage) { |
79612c8b JS |
5025 | total_len = cpuview_proc_stat(cg, cpuset, cg_cpu_usage, cg_cpu_usage_size, |
5026 | f, d->buf, d->buflen); | |
056adcef JS |
5027 | goto out; |
5028 | } | |
5029 | ||
237e200e | 5030 | while (getline(&line, &linelen, f) != -1) { |
a262ddb7 | 5031 | ssize_t l; |
237e200e SH |
5032 | char cpu_char[10]; /* That's a lot of cores */ |
5033 | char *c; | |
8be92dd1 JS |
5034 | uint64_t all_used, cg_used, new_idle; |
5035 | int ret; | |
237e200e | 5036 | |
b4665ce0 SH |
5037 | if (strlen(line) == 0) |
5038 | continue; | |
237e200e SH |
5039 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
5040 | /* not a ^cpuN line containing a number N, just print it */ | |
9502bae2 | 5041 | l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
5042 | if (l < 0) { |
5043 | perror("Error writing to cache"); | |
5044 | rv = 0; | |
5045 | goto err; | |
5046 | } | |
5047 | if (l >= cache_size) { | |
b8defc3d | 5048 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
5049 | rv = 0; |
5050 | goto err; | |
5051 | } | |
5052 | cache += l; | |
5053 | cache_size -= l; | |
5054 | total_len += l; | |
5055 | continue; | |
5056 | } | |
5057 | ||
77005a6c | 5058 | if (sscanf(cpu_char, "%d", &physcpu) != 1) |
237e200e | 5059 | continue; |
77005a6c | 5060 | if (!cpu_in_cpuset(physcpu, cpuset)) |
237e200e SH |
5061 | continue; |
5062 | curcpu ++; | |
5063 | ||
8be92dd1 | 5064 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
7144f069 CB |
5065 | &user, |
5066 | &nice, | |
5067 | &system, | |
5068 | &idle, | |
5069 | &iowait, | |
5070 | &irq, | |
5071 | &softirq, | |
5072 | &steal, | |
5073 | &guest, | |
8be92dd1 JS |
5074 | &guest_nice); |
5075 | ||
5076 | if (ret != 10 || !cg_cpu_usage) { | |
5077 | c = strchr(line, ' '); | |
5078 | if (!c) | |
5079 | continue; | |
5080 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
5081 | if (l < 0) { | |
5082 | perror("Error writing to cache"); | |
5083 | rv = 0; | |
5084 | goto err; | |
5085 | ||
5086 | } | |
5087 | if (l >= cache_size) { | |
5088 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
5089 | rv = 0; | |
5090 | goto err; | |
5091 | } | |
5092 | ||
5093 | cache += l; | |
5094 | cache_size -= l; | |
5095 | total_len += l; | |
5096 | ||
5097 | if (ret != 10) | |
5098 | continue; | |
5099 | } | |
5100 | ||
5101 | if (cg_cpu_usage) { | |
79612c8b JS |
5102 | if (physcpu >= cg_cpu_usage_size) |
5103 | break; | |
5104 | ||
8be92dd1 | 5105 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; |
77005a6c | 5106 | cg_used = cg_cpu_usage[physcpu].user + cg_cpu_usage[physcpu].system; |
8be92dd1 JS |
5107 | |
5108 | if (all_used >= cg_used) { | |
5109 | new_idle = idle + (all_used - cg_used); | |
5110 | ||
5111 | } else { | |
5112 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
5113 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
5114 | curcpu, cg, all_used, cg_used); | |
5115 | new_idle = idle; | |
5116 | } | |
5117 | ||
5118 | l = snprintf(cache, cache_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
77005a6c | 5119 | curcpu, cg_cpu_usage[physcpu].user, cg_cpu_usage[physcpu].system, |
8be92dd1 JS |
5120 | new_idle); |
5121 | ||
5122 | if (l < 0) { | |
5123 | perror("Error writing to cache"); | |
5124 | rv = 0; | |
5125 | goto err; | |
5126 | ||
5127 | } | |
5128 | if (l >= cache_size) { | |
5129 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
5130 | rv = 0; | |
5131 | goto err; | |
5132 | } | |
5133 | ||
5134 | cache += l; | |
5135 | cache_size -= l; | |
5136 | total_len += l; | |
5137 | ||
77005a6c JS |
5138 | user_sum += cg_cpu_usage[physcpu].user; |
5139 | system_sum += cg_cpu_usage[physcpu].system; | |
8be92dd1 JS |
5140 | idle_sum += new_idle; |
5141 | ||
5142 | } else { | |
5143 | user_sum += user; | |
5144 | nice_sum += nice; | |
5145 | system_sum += system; | |
5146 | idle_sum += idle; | |
5147 | iowait_sum += iowait; | |
5148 | irq_sum += irq; | |
5149 | softirq_sum += softirq; | |
5150 | steal_sum += steal; | |
5151 | guest_sum += guest; | |
5152 | guest_nice_sum += guest_nice; | |
5153 | } | |
237e200e SH |
5154 | } |
5155 | ||
5156 | cache = d->buf; | |
5157 | ||
7144f069 CB |
5158 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
5159 | user_sum, | |
5160 | nice_sum, | |
5161 | system_sum, | |
5162 | idle_sum, | |
5163 | iowait_sum, | |
5164 | irq_sum, | |
5165 | softirq_sum, | |
5166 | steal_sum, | |
5167 | guest_sum, | |
5168 | guest_nice_sum); | |
5169 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
237e200e SH |
5170 | memcpy(cache, cpuall, cpuall_len); |
5171 | cache += cpuall_len; | |
7144f069 | 5172 | } else { |
237e200e | 5173 | /* shouldn't happen */ |
b8defc3d | 5174 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); |
237e200e SH |
5175 | cpuall_len = 0; |
5176 | } | |
5177 | ||
5178 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
5179 | total_len += cpuall_len; | |
056adcef JS |
5180 | |
5181 | out: | |
237e200e SH |
5182 | d->cached = 1; |
5183 | d->size = total_len; | |
7144f069 CB |
5184 | if (total_len > size) |
5185 | total_len = size; | |
237e200e SH |
5186 | |
5187 | memcpy(buf, d->buf, total_len); | |
5188 | rv = total_len; | |
5189 | ||
5190 | err: | |
5191 | if (f) | |
5192 | fclose(f); | |
8be92dd1 JS |
5193 | if (cg_cpu_usage) |
5194 | free(cg_cpu_usage); | |
237e200e SH |
5195 | free(line); |
5196 | free(cpuset); | |
5197 | free(cg); | |
5198 | return rv; | |
5199 | } | |
5200 | ||
0ecddf02 CB |
5201 | /* This function retrieves the busy time of a group of tasks by looking at |
5202 | * cpuacct.usage. Unfortunately, this only makes sense when the container has | |
5203 | * been given it's own cpuacct cgroup. If not, this function will take the busy | |
5204 | * time of all other taks that do not actually belong to the container into | |
5205 | * account as well. If someone has a clever solution for this please send a | |
5206 | * patch! | |
5207 | */ | |
1c4b4e38 | 5208 | static double get_reaper_busy(pid_t task) |
237e200e SH |
5209 | { |
5210 | pid_t initpid = lookup_initpid_in_store(task); | |
5211 | char *cgroup = NULL, *usage_str = NULL; | |
5212 | unsigned long usage = 0; | |
1c4b4e38 | 5213 | double res = 0; |
237e200e SH |
5214 | |
5215 | if (initpid <= 0) | |
5216 | return 0; | |
5217 | ||
5218 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
5219 | if (!cgroup) | |
5220 | goto out; | |
6d2f6996 | 5221 | prune_init_slice(cgroup); |
237e200e SH |
5222 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
5223 | goto out; | |
5224 | usage = strtoul(usage_str, NULL, 10); | |
1c4b4e38 | 5225 | res = (double)usage / 1000000000; |
237e200e SH |
5226 | |
5227 | out: | |
5228 | free(cgroup); | |
5229 | free(usage_str); | |
1c4b4e38 | 5230 | return res; |
237e200e SH |
5231 | } |
5232 | ||
5233 | #if RELOADTEST | |
5234 | void iwashere(void) | |
5235 | { | |
237e200e SH |
5236 | int fd; |
5237 | ||
ec2b5e7c | 5238 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
5239 | if (fd >= 0) |
5240 | close(fd); | |
5241 | } | |
5242 | #endif | |
5243 | ||
5244 | /* | |
5245 | * We read /proc/uptime and reuse its second field. | |
5246 | * For the first field, we use the mtime for the reaper for | |
5247 | * the calling pid as returned by getreaperage | |
5248 | */ | |
5249 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
5250 | struct fuse_file_info *fi) | |
5251 | { | |
5252 | struct fuse_context *fc = fuse_get_context(); | |
5253 | struct file_info *d = (struct file_info *)fi->fh; | |
1c4b4e38 | 5254 | double busytime = get_reaper_busy(fc->pid); |
237e200e | 5255 | char *cache = d->buf; |
a262ddb7 | 5256 | ssize_t total_len = 0; |
1c4b4e38 | 5257 | double idletime, reaperage; |
237e200e SH |
5258 | |
5259 | #if RELOADTEST | |
5260 | iwashere(); | |
5261 | #endif | |
5262 | ||
5263 | if (offset){ | |
237e200e SH |
5264 | if (!d->cached) |
5265 | return 0; | |
bbdf646b BM |
5266 | if (offset > d->size) |
5267 | return -EINVAL; | |
237e200e SH |
5268 | int left = d->size - offset; |
5269 | total_len = left > size ? size: left; | |
5270 | memcpy(buf, cache + offset, total_len); | |
5271 | return total_len; | |
5272 | } | |
5273 | ||
0ecddf02 CB |
5274 | reaperage = get_reaper_age(fc->pid); |
5275 | /* To understand why this is done, please read the comment to the | |
5276 | * get_reaper_busy() function. | |
5277 | */ | |
5278 | idletime = reaperage; | |
5279 | if (reaperage >= busytime) | |
5280 | idletime = reaperage - busytime; | |
237e200e | 5281 | |
1c4b4e38 | 5282 | total_len = snprintf(d->buf, d->buflen, "%.2lf %.2lf\n", reaperage, idletime); |
bbdf646b | 5283 | if (total_len < 0 || total_len >= d->buflen){ |
0ecddf02 | 5284 | lxcfs_error("%s\n", "failed to write to cache"); |
237e200e SH |
5285 | return 0; |
5286 | } | |
5287 | ||
5288 | d->size = (int)total_len; | |
5289 | d->cached = 1; | |
5290 | ||
5291 | if (total_len > size) total_len = size; | |
5292 | ||
5293 | memcpy(buf, d->buf, total_len); | |
5294 | return total_len; | |
5295 | } | |
5296 | ||
5297 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
5298 | struct fuse_file_info *fi) | |
5299 | { | |
5300 | char dev_name[72]; | |
5301 | struct fuse_context *fc = fuse_get_context(); | |
5302 | struct file_info *d = (struct file_info *)fi->fh; | |
5303 | char *cg; | |
5304 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
5305 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
5306 | unsigned long read = 0, write = 0; | |
5307 | unsigned long read_merged = 0, write_merged = 0; | |
5308 | unsigned long read_sectors = 0, write_sectors = 0; | |
5309 | unsigned long read_ticks = 0, write_ticks = 0; | |
5310 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
5311 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
5312 | char *cache = d->buf; | |
5313 | size_t cache_size = d->buflen; | |
5314 | char *line = NULL; | |
5315 | size_t linelen = 0, total_len = 0, rv = 0; | |
5316 | unsigned int major = 0, minor = 0; | |
5317 | int i = 0; | |
5318 | FILE *f = NULL; | |
5319 | ||
5320 | if (offset){ | |
5321 | if (offset > d->size) | |
5322 | return -EINVAL; | |
5323 | if (!d->cached) | |
5324 | return 0; | |
5325 | int left = d->size - offset; | |
5326 | total_len = left > size ? size: left; | |
5327 | memcpy(buf, cache + offset, total_len); | |
5328 | return total_len; | |
5329 | } | |
5330 | ||
5331 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5332 | if (initpid <= 1 || is_shared_pidns(initpid)) |
237e200e SH |
5333 | initpid = fc->pid; |
5334 | cg = get_pid_cgroup(initpid, "blkio"); | |
5335 | if (!cg) | |
5336 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 5337 | prune_init_slice(cg); |
237e200e | 5338 | |
2209fe50 | 5339 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 5340 | goto err; |
2209fe50 | 5341 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 5342 | goto err; |
2209fe50 | 5343 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 5344 | goto err; |
2209fe50 | 5345 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 5346 | goto err; |
2209fe50 | 5347 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
5348 | goto err; |
5349 | ||
5350 | ||
5351 | f = fopen("/proc/diskstats", "r"); | |
5352 | if (!f) | |
5353 | goto err; | |
5354 | ||
5355 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 5356 | ssize_t l; |
2209fe50 | 5357 | char lbuf[256]; |
237e200e SH |
5358 | |
5359 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 5360 | if (i != 3) |
237e200e | 5361 | continue; |
2209fe50 SH |
5362 | |
5363 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
5364 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
5365 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
5366 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
5367 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
5368 | read_sectors = read_sectors/512; | |
5369 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
5370 | write_sectors = write_sectors/512; | |
5371 | ||
5372 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
5373 | rd_svctm = rd_svctm/1000000; | |
5374 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
5375 | rd_wait = rd_wait/1000000; | |
5376 | read_ticks = rd_svctm + rd_wait; | |
5377 | ||
5378 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
5379 | wr_svctm = wr_svctm/1000000; | |
5380 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
5381 | wr_wait = wr_wait/1000000; | |
5382 | write_ticks = wr_svctm + wr_wait; | |
5383 | ||
5384 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
5385 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
5386 | |
5387 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
5388 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
5389 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
5390 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
5391 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
5392 | else | |
5393 | continue; | |
237e200e | 5394 | |
2209fe50 | 5395 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
5396 | if (l < 0) { |
5397 | perror("Error writing to fuse buf"); | |
5398 | rv = 0; | |
5399 | goto err; | |
5400 | } | |
5401 | if (l >= cache_size) { | |
b8defc3d | 5402 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
5403 | rv = 0; |
5404 | goto err; | |
5405 | } | |
5406 | cache += l; | |
5407 | cache_size -= l; | |
5408 | total_len += l; | |
5409 | } | |
5410 | ||
5411 | d->cached = 1; | |
5412 | d->size = total_len; | |
5413 | if (total_len > size ) total_len = size; | |
5414 | memcpy(buf, d->buf, total_len); | |
5415 | ||
5416 | rv = total_len; | |
5417 | err: | |
5418 | free(cg); | |
5419 | if (f) | |
5420 | fclose(f); | |
5421 | free(line); | |
5422 | free(io_serviced_str); | |
5423 | free(io_merged_str); | |
5424 | free(io_service_bytes_str); | |
5425 | free(io_wait_time_str); | |
5426 | free(io_service_time_str); | |
5427 | return rv; | |
5428 | } | |
5429 | ||
70dcc12e SH |
5430 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
5431 | struct fuse_file_info *fi) | |
5432 | { | |
5433 | struct fuse_context *fc = fuse_get_context(); | |
5434 | struct file_info *d = (struct file_info *)fi->fh; | |
5435 | char *cg = NULL; | |
018246ff | 5436 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL; |
70dcc12e | 5437 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; |
a262ddb7 CB |
5438 | ssize_t total_len = 0, rv = 0; |
5439 | ssize_t l = 0; | |
70dcc12e SH |
5440 | char *cache = d->buf; |
5441 | ||
5442 | if (offset) { | |
5443 | if (offset > d->size) | |
5444 | return -EINVAL; | |
5445 | if (!d->cached) | |
5446 | return 0; | |
5447 | int left = d->size - offset; | |
5448 | total_len = left > size ? size: left; | |
5449 | memcpy(buf, cache + offset, total_len); | |
5450 | return total_len; | |
5451 | } | |
5452 | ||
5453 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5454 | if (initpid <= 1 || is_shared_pidns(initpid)) |
70dcc12e SH |
5455 | initpid = fc->pid; |
5456 | cg = get_pid_cgroup(initpid, "memory"); | |
5457 | if (!cg) | |
5458 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 5459 | prune_init_slice(cg); |
70dcc12e | 5460 | |
018246ff | 5461 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
70dcc12e SH |
5462 | |
5463 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
5464 | goto err; | |
5465 | ||
70dcc12e SH |
5466 | memusage = strtoul(memusage_str, NULL, 10); |
5467 | ||
5468 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
5469 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
5470 | ||
018246ff | 5471 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
70dcc12e SH |
5472 | memswusage = strtoul(memswusage_str, NULL, 10); |
5473 | ||
70dcc12e SH |
5474 | swap_total = (memswlimit - memlimit) / 1024; |
5475 | swap_free = (memswusage - memusage) / 1024; | |
5476 | } | |
5477 | ||
5478 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
5479 | ||
5480 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
5481 | if (!memswlimit) { | |
5482 | char *line = NULL; | |
5483 | size_t linelen = 0; | |
5484 | FILE *f = fopen("/proc/meminfo", "r"); | |
5485 | ||
5486 | if (!f) | |
5487 | goto err; | |
5488 | ||
5489 | while (getline(&line, &linelen, f) != -1) { | |
5490 | if (startswith(line, "SwapTotal:")) { | |
5491 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
5492 | } else if (startswith(line, "SwapFree:")) { | |
5493 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
5494 | } | |
5495 | } | |
5496 | ||
5497 | free(line); | |
5498 | fclose(f); | |
5499 | } | |
5500 | ||
5501 | if (swap_total > 0) { | |
a262ddb7 CB |
5502 | l = snprintf(d->buf + total_len, d->size - total_len, |
5503 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
5504 | swap_total, swap_free); | |
5505 | total_len += l; | |
70dcc12e SH |
5506 | } |
5507 | ||
a262ddb7 | 5508 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
5509 | perror("Error writing to cache"); |
5510 | rv = 0; | |
5511 | goto err; | |
5512 | } | |
5513 | ||
5514 | d->cached = 1; | |
5515 | d->size = (int)total_len; | |
5516 | ||
5517 | if (total_len > size) total_len = size; | |
5518 | memcpy(buf, d->buf, total_len); | |
5519 | rv = total_len; | |
5520 | ||
5521 | err: | |
5522 | free(cg); | |
5523 | free(memswlimit_str); | |
5524 | free(memlimit_str); | |
5525 | free(memusage_str); | |
5526 | free(memswusage_str); | |
70dcc12e SH |
5527 | return rv; |
5528 | } | |
6db4f7a3 | 5529 | /* |
5530 | * Find the process pid from cgroup path. | |
5531 | * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid. | |
5532 | * @pid_buf : put pid to pid_buf. | |
5533 | * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ... | |
5534 | * @depth : the depth of cgroup in container. | |
5535 | * @sum : return the number of pid. | |
5536 | * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu | |
5537 | */ | |
5538 | static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd) | |
5539 | { | |
5540 | DIR *dir; | |
5541 | int fd; | |
5542 | struct dirent *file; | |
5543 | FILE *f = NULL; | |
5544 | size_t linelen = 0; | |
5545 | char *line = NULL; | |
5546 | int pd; | |
5547 | char *path_dir, *path; | |
5548 | char **pid; | |
5549 | ||
5550 | /* path = dpath + "/cgroup.procs" + /0 */ | |
5551 | do { | |
5552 | path = malloc(strlen(dpath) + 20); | |
5553 | } while (!path); | |
5554 | ||
5555 | strcpy(path, dpath); | |
5556 | fd = openat(cfd, path, O_RDONLY); | |
5557 | if (fd < 0) | |
5558 | goto out; | |
5559 | ||
5560 | dir = fdopendir(fd); | |
5561 | if (dir == NULL) { | |
5562 | close(fd); | |
5563 | goto out; | |
5564 | } | |
5565 | ||
5566 | while (((file = readdir(dir)) != NULL) && depth > 0) { | |
5567 | if (strncmp(file->d_name, ".", 1) == 0) | |
5568 | continue; | |
5569 | if (strncmp(file->d_name, "..", 1) == 0) | |
5570 | continue; | |
5571 | if (file->d_type == DT_DIR) { | |
5572 | /* path + '/' + d_name +/0 */ | |
5573 | do { | |
5574 | path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name)); | |
5575 | } while (!path_dir); | |
5576 | strcpy(path_dir, path); | |
5577 | strcat(path_dir, "/"); | |
5578 | strcat(path_dir, file->d_name); | |
5579 | pd = depth - 1; | |
5580 | sum = calc_pid(pid_buf, path_dir, pd, sum, cfd); | |
5581 | free(path_dir); | |
5582 | } | |
5583 | } | |
5584 | closedir(dir); | |
5585 | ||
5586 | strcat(path, "/cgroup.procs"); | |
5587 | fd = openat(cfd, path, O_RDONLY); | |
5588 | if (fd < 0) | |
5589 | goto out; | |
5590 | ||
5591 | f = fdopen(fd, "r"); | |
5592 | if (!f) { | |
5593 | close(fd); | |
5594 | goto out; | |
5595 | } | |
5596 | ||
5597 | while (getline(&line, &linelen, f) != -1) { | |
5598 | do { | |
5599 | pid = realloc(*pid_buf, sizeof(char *) * (sum + 1)); | |
5600 | } while (!pid); | |
5601 | *pid_buf = pid; | |
5602 | do { | |
5603 | *(*pid_buf + sum) = malloc(strlen(line) + 1); | |
5604 | } while (*(*pid_buf + sum) == NULL); | |
5605 | strcpy(*(*pid_buf + sum), line); | |
5606 | sum++; | |
5607 | } | |
5608 | fclose(f); | |
5609 | out: | |
832904c1 JS |
5610 | if (line) |
5611 | free(line); | |
6db4f7a3 | 5612 | free(path); |
5613 | return sum; | |
5614 | } | |
5615 | /* | |
5616 | * calc_load calculates the load according to the following formula: | |
5617 | * load1 = load0 * exp + active * (1 - exp) | |
5618 | * | |
5619 | * @load1: the new loadavg. | |
5620 | * @load0: the former loadavg. | |
5621 | * @active: the total number of running pid at this moment. | |
5622 | * @exp: the fixed-point defined in the beginning. | |
5623 | */ | |
5624 | static unsigned long | |
5625 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
5626 | { | |
5627 | unsigned long newload; | |
5628 | ||
5629 | active = active > 0 ? active * FIXED_1 : 0; | |
5630 | newload = load * exp + active * (FIXED_1 - exp); | |
5631 | if (active >= load) | |
5632 | newload += FIXED_1 - 1; | |
5633 | ||
5634 | return newload / FIXED_1; | |
5635 | } | |
5636 | ||
5637 | /* | |
5638 | * Return 0 means that container p->cg is closed. | |
5639 | * Return -1 means that error occurred in refresh. | |
5640 | * Positive num equals the total number of pid. | |
5641 | */ | |
5642 | static int refresh_load(struct load_node *p, char *path) | |
5643 | { | |
5644 | FILE *f = NULL; | |
5645 | char **idbuf; | |
5646 | char proc_path[256]; | |
5647 | int i, ret, run_pid = 0, total_pid = 0, last_pid = 0; | |
5648 | char *line = NULL; | |
5649 | size_t linelen = 0; | |
5650 | int sum, length; | |
5651 | DIR *dp; | |
5652 | struct dirent *file; | |
5653 | ||
5654 | do { | |
5655 | idbuf = malloc(sizeof(char *)); | |
5656 | } while (!idbuf); | |
5657 | sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd); | |
5658 | /* normal exit */ | |
5659 | if (sum == 0) | |
5660 | goto out; | |
5661 | ||
5662 | for (i = 0; i < sum; i++) { | |
5663 | /*clean up '\n' */ | |
5664 | length = strlen(idbuf[i])-1; | |
5665 | idbuf[i][length] = '\0'; | |
5666 | ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]); | |
5667 | if (ret < 0 || ret > 255) { | |
5668 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
5669 | i = sum; | |
5670 | sum = -1; | |
5671 | goto err_out; | |
5672 | } | |
5673 | ||
5674 | dp = opendir(proc_path); | |
5675 | if (!dp) { | |
5676 | lxcfs_error("%s\n", "Open proc_path failed in refresh_load."); | |
5677 | continue; | |
5678 | } | |
5679 | while ((file = readdir(dp)) != NULL) { | |
5680 | if (strncmp(file->d_name, ".", 1) == 0) | |
5681 | continue; | |
5682 | if (strncmp(file->d_name, "..", 1) == 0) | |
5683 | continue; | |
5684 | total_pid++; | |
5685 | /* We make the biggest pid become last_pid.*/ | |
5686 | ret = atof(file->d_name); | |
5687 | last_pid = (ret > last_pid) ? ret : last_pid; | |
5688 | ||
5689 | ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name); | |
5690 | if (ret < 0 || ret > 255) { | |
5691 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
5692 | i = sum; | |
5693 | sum = -1; | |
5694 | closedir(dp); | |
5695 | goto err_out; | |
5696 | } | |
5697 | f = fopen(proc_path, "r"); | |
5698 | if (f != NULL) { | |
5699 | while (getline(&line, &linelen, f) != -1) { | |
5700 | /* Find State */ | |
5701 | if ((line[0] == 'S') && (line[1] == 't')) | |
5702 | break; | |
5703 | } | |
5704 | if ((line[7] == 'R') || (line[7] == 'D')) | |
5705 | run_pid++; | |
5706 | fclose(f); | |
5707 | } | |
5708 | } | |
5709 | closedir(dp); | |
5710 | } | |
5711 | /*Calculate the loadavg.*/ | |
5712 | p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid); | |
5713 | p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid); | |
5714 | p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid); | |
5715 | p->run_pid = run_pid; | |
5716 | p->total_pid = total_pid; | |
5717 | p->last_pid = last_pid; | |
5718 | ||
5719 | free(line); | |
beb5024e | 5720 | err_out: |
6db4f7a3 | 5721 | for (; i > 0; i--) |
5722 | free(idbuf[i-1]); | |
5723 | out: | |
5724 | free(idbuf); | |
5725 | return sum; | |
5726 | } | |
5727 | /* | |
5728 | * Traverse the hash table and update it. | |
5729 | */ | |
5730 | void *load_begin(void *arg) | |
5731 | { | |
5732 | ||
5733 | char *path = NULL; | |
5734 | int i, sum, length, ret; | |
5735 | struct load_node *f; | |
5736 | int first_node; | |
5737 | clock_t time1, time2; | |
5738 | ||
5739 | while (1) { | |
a83618e2 JS |
5740 | if (loadavg_stop == 1) |
5741 | return NULL; | |
5742 | ||
6db4f7a3 | 5743 | time1 = clock(); |
5744 | for (i = 0; i < LOAD_SIZE; i++) { | |
5745 | pthread_mutex_lock(&load_hash[i].lock); | |
5746 | if (load_hash[i].next == NULL) { | |
5747 | pthread_mutex_unlock(&load_hash[i].lock); | |
5748 | continue; | |
5749 | } | |
5750 | f = load_hash[i].next; | |
5751 | first_node = 1; | |
5752 | while (f) { | |
5753 | length = strlen(f->cg) + 2; | |
5754 | do { | |
5755 | /* strlen(f->cg) + '.' or '' + \0 */ | |
5756 | path = malloc(length); | |
5757 | } while (!path); | |
5758 | ||
5759 | ret = snprintf(path, length, "%s%s", *(f->cg) == '/' ? "." : "", f->cg); | |
5760 | if (ret < 0 || ret > length - 1) { | |
5761 | /* snprintf failed, ignore the node.*/ | |
5762 | lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg); | |
5763 | goto out; | |
5764 | } | |
5765 | sum = refresh_load(f, path); | |
5766 | if (sum == 0) { | |
5767 | f = del_node(f, i); | |
5768 | } else { | |
5769 | out: f = f->next; | |
5770 | } | |
5771 | free(path); | |
5772 | /* load_hash[i].lock locks only on the first node.*/ | |
5773 | if (first_node == 1) { | |
5774 | first_node = 0; | |
5775 | pthread_mutex_unlock(&load_hash[i].lock); | |
5776 | } | |
5777 | } | |
5778 | } | |
a83618e2 JS |
5779 | |
5780 | if (loadavg_stop == 1) | |
5781 | return NULL; | |
5782 | ||
6db4f7a3 | 5783 | time2 = clock(); |
5784 | usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC)); | |
5785 | } | |
5786 | } | |
5787 | ||
5788 | static int proc_loadavg_read(char *buf, size_t size, off_t offset, | |
5789 | struct fuse_file_info *fi) | |
5790 | { | |
5791 | struct fuse_context *fc = fuse_get_context(); | |
5792 | struct file_info *d = (struct file_info *)fi->fh; | |
5793 | pid_t initpid; | |
5794 | char *cg; | |
5795 | size_t total_len = 0; | |
5796 | char *cache = d->buf; | |
5797 | struct load_node *n; | |
5798 | int hash; | |
01d88ede | 5799 | int cfd, rv = 0; |
6db4f7a3 | 5800 | unsigned long a, b, c; |
5801 | ||
5802 | if (offset) { | |
5803 | if (offset > d->size) | |
5804 | return -EINVAL; | |
5805 | if (!d->cached) | |
5806 | return 0; | |
5807 | int left = d->size - offset; | |
5808 | total_len = left > size ? size : left; | |
5809 | memcpy(buf, cache + offset, total_len); | |
5810 | return total_len; | |
5811 | } | |
5812 | if (!loadavg) | |
5813 | return read_file("/proc/loadavg", buf, size, d); | |
5814 | ||
5815 | initpid = lookup_initpid_in_store(fc->pid); | |
6e3637bb | 5816 | if (initpid <= 1 || is_shared_pidns(initpid)) |
6db4f7a3 | 5817 | initpid = fc->pid; |
5818 | cg = get_pid_cgroup(initpid, "cpu"); | |
5819 | if (!cg) | |
5820 | return read_file("/proc/loadavg", buf, size, d); | |
5821 | ||
5822 | prune_init_slice(cg); | |
b077527b | 5823 | hash = calc_hash(cg) % LOAD_SIZE; |
6db4f7a3 | 5824 | n = locate_node(cg, hash); |
5825 | ||
5826 | /* First time */ | |
5827 | if (n == NULL) { | |
5828 | if (!find_mounted_controller("cpu", &cfd)) { | |
5829 | /* | |
5830 | * In locate_node() above, pthread_rwlock_unlock() isn't used | |
5831 | * because delete is not allowed before read has ended. | |
5832 | */ | |
5833 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
01d88ede JS |
5834 | rv = 0; |
5835 | goto err; | |
6db4f7a3 | 5836 | } |
5837 | do { | |
5838 | n = malloc(sizeof(struct load_node)); | |
5839 | } while (!n); | |
5840 | ||
5841 | do { | |
5842 | n->cg = malloc(strlen(cg)+1); | |
5843 | } while (!n->cg); | |
5844 | strcpy(n->cg, cg); | |
5845 | n->avenrun[0] = 0; | |
5846 | n->avenrun[1] = 0; | |
5847 | n->avenrun[2] = 0; | |
5848 | n->run_pid = 0; | |
5849 | n->total_pid = 1; | |
5850 | n->last_pid = initpid; | |
5851 | n->cfd = cfd; | |
5852 | insert_node(&n, hash); | |
5853 | } | |
5854 | a = n->avenrun[0] + (FIXED_1/200); | |
5855 | b = n->avenrun[1] + (FIXED_1/200); | |
5856 | c = n->avenrun[2] + (FIXED_1/200); | |
5857 | total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n", | |
5858 | LOAD_INT(a), LOAD_FRAC(a), | |
5859 | LOAD_INT(b), LOAD_FRAC(b), | |
5860 | LOAD_INT(c), LOAD_FRAC(c), | |
5861 | n->run_pid, n->total_pid, n->last_pid); | |
5862 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
5863 | if (total_len < 0 || total_len >= d->buflen) { | |
5864 | lxcfs_error("%s\n", "Failed to write to cache"); | |
01d88ede JS |
5865 | rv = 0; |
5866 | goto err; | |
6db4f7a3 | 5867 | } |
5868 | d->size = (int)total_len; | |
5869 | d->cached = 1; | |
5870 | ||
5871 | if (total_len > size) | |
5872 | total_len = size; | |
5873 | memcpy(buf, d->buf, total_len); | |
01d88ede JS |
5874 | rv = total_len; |
5875 | ||
5876 | err: | |
5877 | free(cg); | |
5878 | return rv; | |
6db4f7a3 | 5879 | } |
5880 | /* Return a positive number on success, return 0 on failure.*/ | |
5881 | pthread_t load_daemon(int load_use) | |
5882 | { | |
5883 | int ret; | |
5884 | pthread_t pid; | |
5885 | ||
5886 | ret = init_load(); | |
5887 | if (ret == -1) { | |
5888 | lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!"); | |
5889 | return 0; | |
5890 | } | |
5891 | ret = pthread_create(&pid, NULL, load_begin, NULL); | |
5892 | if (ret != 0) { | |
5893 | lxcfs_error("%s\n", "Create pthread fails in load_daemon!"); | |
5894 | load_free(); | |
5895 | return 0; | |
5896 | } | |
5897 | /* use loadavg, here loadavg = 1*/ | |
5898 | loadavg = load_use; | |
5899 | return pid; | |
5900 | } | |
70dcc12e | 5901 | |
a83618e2 JS |
5902 | /* Returns 0 on success. */ |
5903 | int stop_load_daemon(pthread_t pid) | |
5904 | { | |
5905 | int s; | |
5906 | ||
5907 | /* Signal the thread to gracefully stop */ | |
5908 | loadavg_stop = 1; | |
5909 | ||
5910 | s = pthread_join(pid, NULL); /* Make sure sub thread has been canceled. */ | |
5911 | if (s != 0) { | |
5912 | lxcfs_error("%s\n", "stop_load_daemon error: failed to join"); | |
5913 | return -1; | |
5914 | } | |
5915 | ||
5916 | load_free(); | |
5917 | loadavg_stop = 0; | |
5918 | ||
5919 | return 0; | |
5920 | } | |
5921 | ||
237e200e SH |
5922 | static off_t get_procfile_size(const char *which) |
5923 | { | |
5924 | FILE *f = fopen(which, "r"); | |
5925 | char *line = NULL; | |
5926 | size_t len = 0; | |
5927 | ssize_t sz, answer = 0; | |
5928 | if (!f) | |
5929 | return 0; | |
5930 | ||
5931 | while ((sz = getline(&line, &len, f)) != -1) | |
5932 | answer += sz; | |
5933 | fclose (f); | |
5934 | free(line); | |
5935 | ||
5936 | return answer; | |
5937 | } | |
5938 | ||
5939 | int proc_getattr(const char *path, struct stat *sb) | |
5940 | { | |
5941 | struct timespec now; | |
5942 | ||
5943 | memset(sb, 0, sizeof(struct stat)); | |
5944 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
5945 | return -EINVAL; | |
5946 | sb->st_uid = sb->st_gid = 0; | |
5947 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
5948 | if (strcmp(path, "/proc") == 0) { | |
5949 | sb->st_mode = S_IFDIR | 00555; | |
5950 | sb->st_nlink = 2; | |
5951 | return 0; | |
5952 | } | |
5953 | if (strcmp(path, "/proc/meminfo") == 0 || | |
5954 | strcmp(path, "/proc/cpuinfo") == 0 || | |
5955 | strcmp(path, "/proc/uptime") == 0 || | |
5956 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e | 5957 | strcmp(path, "/proc/diskstats") == 0 || |
46be8eed | 5958 | strcmp(path, "/proc/swaps") == 0 || |
5959 | strcmp(path, "/proc/loadavg") == 0) { | |
237e200e SH |
5960 | sb->st_size = 0; |
5961 | sb->st_mode = S_IFREG | 00444; | |
5962 | sb->st_nlink = 1; | |
5963 | return 0; | |
5964 | } | |
5965 | ||
5966 | return -ENOENT; | |
5967 | } | |
5968 | ||
5969 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
5970 | struct fuse_file_info *fi) | |
5971 | { | |
d639f863 CB |
5972 | if (filler(buf, ".", NULL, 0) != 0 || |
5973 | filler(buf, "..", NULL, 0) != 0 || | |
5974 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
5975 | filler(buf, "meminfo", NULL, 0) != 0 || | |
5976 | filler(buf, "stat", NULL, 0) != 0 || | |
5977 | filler(buf, "uptime", NULL, 0) != 0 || | |
5978 | filler(buf, "diskstats", NULL, 0) != 0 || | |
46be8eed | 5979 | filler(buf, "swaps", NULL, 0) != 0 || |
5980 | filler(buf, "loadavg", NULL, 0) != 0) | |
237e200e SH |
5981 | return -EINVAL; |
5982 | return 0; | |
5983 | } | |
5984 | ||
5985 | int proc_open(const char *path, struct fuse_file_info *fi) | |
5986 | { | |
5987 | int type = -1; | |
5988 | struct file_info *info; | |
5989 | ||
5990 | if (strcmp(path, "/proc/meminfo") == 0) | |
5991 | type = LXC_TYPE_PROC_MEMINFO; | |
5992 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
5993 | type = LXC_TYPE_PROC_CPUINFO; | |
5994 | else if (strcmp(path, "/proc/uptime") == 0) | |
5995 | type = LXC_TYPE_PROC_UPTIME; | |
5996 | else if (strcmp(path, "/proc/stat") == 0) | |
5997 | type = LXC_TYPE_PROC_STAT; | |
5998 | else if (strcmp(path, "/proc/diskstats") == 0) | |
5999 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
6000 | else if (strcmp(path, "/proc/swaps") == 0) |
6001 | type = LXC_TYPE_PROC_SWAPS; | |
46be8eed | 6002 | else if (strcmp(path, "/proc/loadavg") == 0) |
6003 | type = LXC_TYPE_PROC_LOADAVG; | |
237e200e SH |
6004 | if (type == -1) |
6005 | return -ENOENT; | |
6006 | ||
6007 | info = malloc(sizeof(*info)); | |
6008 | if (!info) | |
6009 | return -ENOMEM; | |
6010 | ||
6011 | memset(info, 0, sizeof(*info)); | |
6012 | info->type = type; | |
6013 | ||
6014 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
6015 | do { | |
6016 | info->buf = malloc(info->buflen); | |
6017 | } while (!info->buf); | |
6018 | memset(info->buf, 0, info->buflen); | |
6019 | /* set actual size to buffer size */ | |
6020 | info->size = info->buflen; | |
6021 | ||
6022 | fi->fh = (unsigned long)info; | |
6023 | return 0; | |
6024 | } | |
6025 | ||
bddbb106 SH |
6026 | int proc_access(const char *path, int mask) |
6027 | { | |
e7849aa3 CB |
6028 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) |
6029 | return 0; | |
6030 | ||
bddbb106 SH |
6031 | /* these are all read-only */ |
6032 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 6033 | return -EACCES; |
bddbb106 SH |
6034 | return 0; |
6035 | } | |
6036 | ||
237e200e SH |
6037 | int proc_release(const char *path, struct fuse_file_info *fi) |
6038 | { | |
43215927 | 6039 | do_release_file_info(fi); |
237e200e SH |
6040 | return 0; |
6041 | } | |
6042 | ||
6043 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
6044 | struct fuse_file_info *fi) | |
6045 | { | |
6046 | struct file_info *f = (struct file_info *) fi->fh; | |
6047 | ||
6048 | switch (f->type) { | |
6049 | case LXC_TYPE_PROC_MEMINFO: | |
6050 | return proc_meminfo_read(buf, size, offset, fi); | |
6051 | case LXC_TYPE_PROC_CPUINFO: | |
6052 | return proc_cpuinfo_read(buf, size, offset, fi); | |
6053 | case LXC_TYPE_PROC_UPTIME: | |
6054 | return proc_uptime_read(buf, size, offset, fi); | |
6055 | case LXC_TYPE_PROC_STAT: | |
6056 | return proc_stat_read(buf, size, offset, fi); | |
6057 | case LXC_TYPE_PROC_DISKSTATS: | |
6058 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
6059 | case LXC_TYPE_PROC_SWAPS: |
6060 | return proc_swaps_read(buf, size, offset, fi); | |
46be8eed | 6061 | case LXC_TYPE_PROC_LOADAVG: |
6062 | return proc_loadavg_read(buf, size, offset, fi); | |
237e200e SH |
6063 | default: |
6064 | return -EINVAL; | |
6065 | } | |
6066 | } | |
6067 | ||
29a73c2f CB |
6068 | /* |
6069 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
6070 | */ |
6071 | ||
6072 | static bool mkdir_p(const char *dir, mode_t mode) | |
6073 | { | |
6074 | const char *tmp = dir; | |
6075 | const char *orig = dir; | |
6076 | char *makeme; | |
6077 | ||
6078 | do { | |
6079 | dir = tmp + strspn(tmp, "/"); | |
6080 | tmp = dir + strcspn(dir, "/"); | |
6081 | makeme = strndup(orig, dir - orig); | |
6082 | if (!makeme) | |
6083 | return false; | |
6084 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
b8defc3d | 6085 | lxcfs_error("Failed to create directory '%s': %s.\n", |
29a73c2f CB |
6086 | makeme, strerror(errno)); |
6087 | free(makeme); | |
6088 | return false; | |
6089 | } | |
6090 | free(makeme); | |
6091 | } while(tmp != dir); | |
6092 | ||
6093 | return true; | |
6094 | } | |
6095 | ||
6096 | static bool umount_if_mounted(void) | |
6097 | { | |
6098 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
b8defc3d | 6099 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
6100 | return false; |
6101 | } | |
6102 | return true; | |
6103 | } | |
6104 | ||
2283e240 CB |
6105 | /* __typeof__ should be safe to use with all compilers. */ |
6106 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
6107 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
6108 | { | |
6109 | return (fs->f_type == (fs_type_magic)magic_val); | |
6110 | } | |
6111 | ||
0a4dea41 CB |
6112 | /* |
6113 | * looking at fs/proc_namespace.c, it appears we can | |
6114 | * actually expect the rootfs entry to very specifically contain | |
6115 | * " - rootfs rootfs " | |
6116 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
6117 | * the rootfs entry should always be skipped in mountinfo contents. | |
6118 | */ | |
6119 | static bool is_on_ramfs(void) | |
6120 | { | |
6121 | FILE *f; | |
6122 | char *p, *p2; | |
6123 | char *line = NULL; | |
6124 | size_t len = 0; | |
6125 | int i; | |
6126 | ||
6127 | f = fopen("/proc/self/mountinfo", "r"); | |
6128 | if (!f) | |
6129 | return false; | |
6130 | ||
6131 | while (getline(&line, &len, f) != -1) { | |
6132 | for (p = line, i = 0; p && i < 4; i++) | |
6133 | p = strchr(p + 1, ' '); | |
6134 | if (!p) | |
6135 | continue; | |
6136 | p2 = strchr(p + 1, ' '); | |
6137 | if (!p2) | |
6138 | continue; | |
6139 | *p2 = '\0'; | |
6140 | if (strcmp(p + 1, "/") == 0) { | |
6141 | // this is '/'. is it the ramfs? | |
6142 | p = strchr(p2 + 1, '-'); | |
6143 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
6144 | free(line); | |
6145 | fclose(f); | |
6146 | return true; | |
6147 | } | |
6148 | } | |
6149 | } | |
6150 | free(line); | |
6151 | fclose(f); | |
6152 | return false; | |
6153 | } | |
6154 | ||
cc309f33 | 6155 | static int pivot_enter() |
0a4dea41 | 6156 | { |
cc309f33 CB |
6157 | int ret = -1, oldroot = -1, newroot = -1; |
6158 | ||
6159 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
6160 | if (oldroot < 0) { | |
6161 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
6162 | return ret; | |
6163 | } | |
6164 | ||
6165 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
6166 | if (newroot < 0) { | |
6167 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
6168 | goto err; | |
6169 | } | |
6170 | ||
6171 | /* change into new root fs */ | |
6172 | if (fchdir(newroot) < 0) { | |
6173 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
6174 | goto err; | |
6175 | } | |
6176 | ||
0a4dea41 CB |
6177 | /* pivot_root into our new root fs */ |
6178 | if (pivot_root(".", ".") < 0) { | |
6179 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
cc309f33 | 6180 | goto err; |
0a4dea41 CB |
6181 | } |
6182 | ||
6183 | /* | |
6184 | * At this point the old-root is mounted on top of our new-root. | |
6185 | * To unmounted it we must not be chdir'd into it, so escape back | |
6186 | * to the old-root. | |
6187 | */ | |
6188 | if (fchdir(oldroot) < 0) { | |
6189 | lxcfs_error("%s\n", "Failed to enter old root."); | |
cc309f33 | 6190 | goto err; |
0a4dea41 CB |
6191 | } |
6192 | ||
6193 | if (umount2(".", MNT_DETACH) < 0) { | |
6194 | lxcfs_error("%s\n", "Failed to detach old root."); | |
cc309f33 | 6195 | goto err; |
0a4dea41 CB |
6196 | } |
6197 | ||
6198 | if (fchdir(newroot) < 0) { | |
6199 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
cc309f33 | 6200 | goto err; |
0a4dea41 CB |
6201 | } |
6202 | ||
cc309f33 CB |
6203 | ret = 0; |
6204 | ||
6205 | err: | |
6206 | if (oldroot > 0) | |
6207 | close(oldroot); | |
6208 | if (newroot > 0) | |
6209 | close(newroot); | |
6210 | ||
6211 | return ret; | |
0a4dea41 CB |
6212 | } |
6213 | ||
6214 | static int chroot_enter() | |
6215 | { | |
6216 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
6217 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
6218 | return -1; | |
6219 | } | |
6220 | ||
6221 | if (chroot(".") < 0) { | |
6222 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
6223 | return -1; | |
6224 | } | |
6225 | ||
6226 | if (chdir("/") < 0) { | |
6227 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
6228 | return -1; | |
6229 | } | |
6230 | ||
6231 | return 0; | |
6232 | } | |
6233 | ||
0232cbac | 6234 | static int permute_and_enter(void) |
29a73c2f | 6235 | { |
0a4dea41 CB |
6236 | struct statfs sb; |
6237 | ||
6238 | if (statfs("/", &sb) < 0) { | |
6239 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
cc309f33 | 6240 | return -1; |
0a4dea41 CB |
6241 | } |
6242 | ||
6243 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
6244 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
6245 | * /proc/1/mountinfo. */ | |
6246 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
6247 | return chroot_enter(); | |
29a73c2f | 6248 | |
cc309f33 | 6249 | if (pivot_enter() < 0) { |
0a4dea41 | 6250 | lxcfs_error("%s\n", "Could not perform pivot root."); |
cc309f33 | 6251 | return -1; |
29a73c2f CB |
6252 | } |
6253 | ||
cc309f33 | 6254 | return 0; |
29a73c2f CB |
6255 | } |
6256 | ||
6257 | /* Prepare our new clean root. */ | |
0232cbac | 6258 | static int permute_prepare(void) |
29a73c2f CB |
6259 | { |
6260 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
b8defc3d | 6261 | lxcfs_error("%s\n", "Failed to create directory for new root."); |
29a73c2f CB |
6262 | return -1; |
6263 | } | |
6264 | ||
6265 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 6266 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); |
29a73c2f CB |
6267 | return -1; |
6268 | } | |
6269 | ||
6270 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 6271 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
6272 | return -1; |
6273 | } | |
6274 | ||
6275 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
b8defc3d | 6276 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
6277 | return -1; |
6278 | } | |
6279 | ||
6280 | return 0; | |
6281 | } | |
6282 | ||
0232cbac CB |
6283 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ |
6284 | static bool permute_root(void) | |
29a73c2f CB |
6285 | { |
6286 | /* Prepare new root. */ | |
0232cbac | 6287 | if (permute_prepare() < 0) |
29a73c2f CB |
6288 | return false; |
6289 | ||
6290 | /* Pivot into new root. */ | |
0232cbac | 6291 | if (permute_and_enter() < 0) |
29a73c2f CB |
6292 | return false; |
6293 | ||
6294 | return true; | |
6295 | } | |
6296 | ||
a257a8ee CB |
6297 | static int preserve_mnt_ns(int pid) |
6298 | { | |
6299 | int ret; | |
6300 | size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt"); | |
6301 | char path[len]; | |
6302 | ||
6303 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
6304 | if (ret < 0 || (size_t)ret >= len) | |
6305 | return -1; | |
6306 | ||
6307 | return open(path, O_RDONLY | O_CLOEXEC); | |
6308 | } | |
6309 | ||
0a4dea41 | 6310 | static bool cgfs_prepare_mounts(void) |
29a73c2f CB |
6311 | { |
6312 | if (!mkdir_p(BASEDIR, 0700)) { | |
b8defc3d | 6313 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); |
29a73c2f CB |
6314 | return false; |
6315 | } | |
480262c9 | 6316 | |
29a73c2f | 6317 | if (!umount_if_mounted()) { |
b8defc3d | 6318 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); |
480262c9 CB |
6319 | return false; |
6320 | } | |
6321 | ||
6322 | if (unshare(CLONE_NEWNS) < 0) { | |
b8defc3d | 6323 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); |
480262c9 CB |
6324 | return false; |
6325 | } | |
6326 | ||
a257a8ee CB |
6327 | cgroup_mount_ns_fd = preserve_mnt_ns(getpid()); |
6328 | if (cgroup_mount_ns_fd < 0) { | |
6329 | lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno)); | |
6330 | return false; | |
6331 | } | |
6332 | ||
480262c9 | 6333 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { |
b8defc3d | 6334 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); |
29a73c2f CB |
6335 | return false; |
6336 | } | |
480262c9 | 6337 | |
29a73c2f | 6338 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
b8defc3d | 6339 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); |
29a73c2f CB |
6340 | return false; |
6341 | } | |
480262c9 | 6342 | |
29a73c2f CB |
6343 | return true; |
6344 | } | |
6345 | ||
0a4dea41 | 6346 | static bool cgfs_mount_hierarchies(void) |
29a73c2f CB |
6347 | { |
6348 | char *target; | |
6349 | size_t clen, len; | |
6350 | int i, ret; | |
6351 | ||
6352 | for (i = 0; i < num_hierarchies; i++) { | |
6353 | char *controller = hierarchies[i]; | |
51c7ca35 | 6354 | |
29a73c2f CB |
6355 | clen = strlen(controller); |
6356 | len = strlen(BASEDIR) + clen + 2; | |
6357 | target = malloc(len); | |
6358 | if (!target) | |
6359 | return false; | |
51c7ca35 | 6360 | |
29a73c2f CB |
6361 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); |
6362 | if (ret < 0 || ret >= len) { | |
6363 | free(target); | |
6364 | return false; | |
6365 | } | |
6366 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
6367 | free(target); | |
6368 | return false; | |
6369 | } | |
51c7ca35 CB |
6370 | if (!strcmp(controller, "unified")) |
6371 | ret = mount("none", target, "cgroup2", 0, NULL); | |
6372 | else | |
6373 | ret = mount(controller, target, "cgroup", 0, controller); | |
6374 | if (ret < 0) { | |
6375 | lxcfs_error("Failed mounting cgroup %s: %s\n", controller, strerror(errno)); | |
29a73c2f CB |
6376 | free(target); |
6377 | return false; | |
6378 | } | |
6379 | ||
6380 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
6381 | if (fd_hierarchies[i] < 0) { | |
6382 | free(target); | |
6383 | return false; | |
6384 | } | |
6385 | free(target); | |
6386 | } | |
6387 | return true; | |
6388 | } | |
6389 | ||
480262c9 | 6390 | static bool cgfs_setup_controllers(void) |
29a73c2f | 6391 | { |
0a4dea41 | 6392 | if (!cgfs_prepare_mounts()) |
29a73c2f | 6393 | return false; |
29a73c2f | 6394 | |
0a4dea41 | 6395 | if (!cgfs_mount_hierarchies()) { |
b8defc3d | 6396 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); |
29a73c2f CB |
6397 | return false; |
6398 | } | |
6399 | ||
0232cbac | 6400 | if (!permute_root()) |
29a73c2f CB |
6401 | return false; |
6402 | ||
6403 | return true; | |
6404 | } | |
6405 | ||
6406 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) | |
237e200e SH |
6407 | { |
6408 | FILE *f; | |
e58dab00 CB |
6409 | char *cret, *line = NULL; |
6410 | char cwd[MAXPATHLEN]; | |
237e200e | 6411 | size_t len = 0; |
480262c9 | 6412 | int i, init_ns = -1; |
51c7ca35 | 6413 | bool found_unified = false; |
237e200e SH |
6414 | |
6415 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
b8defc3d | 6416 | lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno)); |
237e200e SH |
6417 | return; |
6418 | } | |
e58dab00 | 6419 | |
237e200e | 6420 | while (getline(&line, &len, f) != -1) { |
51c7ca35 | 6421 | char *idx, *p, *p2; |
237e200e SH |
6422 | |
6423 | p = strchr(line, ':'); | |
6424 | if (!p) | |
6425 | goto out; | |
51c7ca35 | 6426 | idx = line; |
237e200e SH |
6427 | *(p++) = '\0'; |
6428 | ||
6429 | p2 = strrchr(p, ':'); | |
6430 | if (!p2) | |
6431 | goto out; | |
6432 | *p2 = '\0'; | |
6433 | ||
a67719f6 CB |
6434 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
6435 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
6436 | * because it parses out the empty string "" and later on passes | |
6437 | * it to mount(). Let's skip such entries. | |
6438 | */ | |
51c7ca35 CB |
6439 | if (!strcmp(p, "") && !strcmp(idx, "0") && !found_unified) { |
6440 | found_unified = true; | |
6441 | p = "unified"; | |
6442 | } | |
a67719f6 | 6443 | |
237e200e SH |
6444 | if (!store_hierarchy(line, p)) |
6445 | goto out; | |
6446 | } | |
6447 | ||
480262c9 | 6448 | /* Preserve initial namespace. */ |
a257a8ee | 6449 | init_ns = preserve_mnt_ns(getpid()); |
b8defc3d CB |
6450 | if (init_ns < 0) { |
6451 | lxcfs_error("%s\n", "Failed to preserve initial mount namespace."); | |
480262c9 | 6452 | goto out; |
b8defc3d | 6453 | } |
480262c9 | 6454 | |
92c3ee11 | 6455 | fd_hierarchies = malloc(sizeof(int) * num_hierarchies); |
b8defc3d CB |
6456 | if (!fd_hierarchies) { |
6457 | lxcfs_error("%s\n", strerror(errno)); | |
29a73c2f | 6458 | goto out; |
b8defc3d | 6459 | } |
29a73c2f | 6460 | |
480262c9 CB |
6461 | for (i = 0; i < num_hierarchies; i++) |
6462 | fd_hierarchies[i] = -1; | |
6463 | ||
e58dab00 CB |
6464 | cret = getcwd(cwd, MAXPATHLEN); |
6465 | if (!cret) | |
6466 | lxcfs_debug("Could not retrieve current working directory: %s.\n", strerror(errno)); | |
6467 | ||
480262c9 CB |
6468 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace |
6469 | * to privately mount lxcfs cgroups. */ | |
b8defc3d CB |
6470 | if (!cgfs_setup_controllers()) { |
6471 | lxcfs_error("%s\n", "Failed to setup private cgroup mounts for lxcfs."); | |
29a73c2f | 6472 | goto out; |
b8defc3d | 6473 | } |
480262c9 | 6474 | |
b8defc3d CB |
6475 | if (setns(init_ns, 0) < 0) { |
6476 | lxcfs_error("Failed to switch back to initial mount namespace: %s.\n", strerror(errno)); | |
29a73c2f | 6477 | goto out; |
b8defc3d | 6478 | } |
29a73c2f | 6479 | |
e58dab00 CB |
6480 | if (!cret || chdir(cwd) < 0) |
6481 | lxcfs_debug("Could not change back to original working directory: %s.\n", strerror(errno)); | |
6482 | ||
056adcef JS |
6483 | if (!init_cpuview()) { |
6484 | lxcfs_error("%s\n", "failed to init CPU view"); | |
6485 | goto out; | |
6486 | } | |
6487 | ||
237e200e SH |
6488 | print_subsystems(); |
6489 | ||
6490 | out: | |
6491 | free(line); | |
6492 | fclose(f); | |
480262c9 CB |
6493 | if (init_ns >= 0) |
6494 | close(init_ns); | |
237e200e SH |
6495 | } |
6496 | ||
6497 | static void __attribute__((destructor)) free_subsystems(void) | |
6498 | { | |
6499 | int i; | |
6500 | ||
b8defc3d CB |
6501 | lxcfs_debug("%s\n", "Running destructor for liblxcfs."); |
6502 | ||
29a73c2f | 6503 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
6504 | if (hierarchies[i]) |
6505 | free(hierarchies[i]); | |
480262c9 | 6506 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
6507 | close(fd_hierarchies[i]); |
6508 | } | |
237e200e | 6509 | free(hierarchies); |
480262c9 | 6510 | free(fd_hierarchies); |
056adcef | 6511 | free_cpuview(); |
a257a8ee CB |
6512 | |
6513 | if (cgroup_mount_ns_fd >= 0) | |
6514 | close(cgroup_mount_ns_fd); | |
1c4b4e38 | 6515 | } |