]>
Commit | Line | Data |
---|---|---|
237e200e SH |
1 | /* lxcfs |
2 | * | |
3 | * Copyright © 2014-2016 Canonical, Inc | |
4 | * Author: Serge Hallyn <serge.hallyn@ubuntu.com> | |
5 | * | |
6 | * See COPYING file for details. | |
7 | */ | |
8 | ||
9 | #define FUSE_USE_VERSION 26 | |
10 | ||
0ecddf02 | 11 | #define __STDC_FORMAT_MACROS |
237e200e | 12 | #include <dirent.h> |
29a73c2f | 13 | #include <errno.h> |
237e200e SH |
14 | #include <fcntl.h> |
15 | #include <fuse.h> | |
0ecddf02 | 16 | #include <inttypes.h> |
237e200e | 17 | #include <libgen.h> |
237e200e | 18 | #include <pthread.h> |
29a73c2f CB |
19 | #include <sched.h> |
20 | #include <stdbool.h> | |
0ecddf02 | 21 | #include <stdint.h> |
29a73c2f CB |
22 | #include <stdio.h> |
23 | #include <stdlib.h> | |
24 | #include <string.h> | |
25 | #include <time.h> | |
26 | #include <unistd.h> | |
27 | #include <wait.h> | |
d89504c4 | 28 | #include <linux/magic.h> |
237e200e | 29 | #include <linux/sched.h> |
29a73c2f CB |
30 | #include <sys/epoll.h> |
31 | #include <sys/mman.h> | |
32 | #include <sys/mount.h> | |
237e200e SH |
33 | #include <sys/param.h> |
34 | #include <sys/socket.h> | |
29a73c2f | 35 | #include <sys/syscall.h> |
0ecddf02 | 36 | #include <sys/sysinfo.h> |
d89504c4 | 37 | #include <sys/vfs.h> |
237e200e | 38 | |
237e200e | 39 | #include "bindings.h" |
237e200e SH |
40 | #include "config.h" // for VERSION |
41 | ||
0ecddf02 CB |
42 | /* Maximum number for 64 bit integer is a string with 21 digits: 2^64 - 1 = 21 */ |
43 | #define LXCFS_NUMSTRLEN64 21 | |
44 | ||
29a73c2f CB |
45 | /* Define pivot_root() if missing from the C library */ |
46 | #ifndef HAVE_PIVOT_ROOT | |
47 | static int pivot_root(const char * new_root, const char * put_old) | |
48 | { | |
49 | #ifdef __NR_pivot_root | |
50 | return syscall(__NR_pivot_root, new_root, put_old); | |
51 | #else | |
52 | errno = ENOSYS; | |
53 | return -1; | |
54 | #endif | |
55 | } | |
56 | #else | |
57 | extern int pivot_root(const char * new_root, const char * put_old); | |
58 | #endif | |
59 | ||
237e200e SH |
60 | enum { |
61 | LXC_TYPE_CGDIR, | |
62 | LXC_TYPE_CGFILE, | |
63 | LXC_TYPE_PROC_MEMINFO, | |
64 | LXC_TYPE_PROC_CPUINFO, | |
65 | LXC_TYPE_PROC_UPTIME, | |
66 | LXC_TYPE_PROC_STAT, | |
67 | LXC_TYPE_PROC_DISKSTATS, | |
70dcc12e | 68 | LXC_TYPE_PROC_SWAPS, |
46be8eed | 69 | LXC_TYPE_PROC_LOADAVG, |
237e200e SH |
70 | }; |
71 | ||
72 | struct file_info { | |
73 | char *controller; | |
74 | char *cgroup; | |
75 | char *file; | |
76 | int type; | |
77 | char *buf; // unused as of yet | |
78 | int buflen; | |
79 | int size; //actual data size | |
80 | int cached; | |
81 | }; | |
82 | ||
8be92dd1 JS |
83 | struct cpuacct_usage { |
84 | uint64_t user; | |
85 | uint64_t system; | |
86 | }; | |
87 | ||
0e47acaa | 88 | /* The function of hash table.*/ |
89 | #define LOAD_SIZE 100 /*the size of hash_table */ | |
6db4f7a3 | 90 | #define FLUSH_TIME 5 /*the flush rate */ |
91 | #define DEPTH_DIR 3 /*the depth of per cgroup */ | |
92 | /* The function of calculate loadavg .*/ | |
93 | #define FSHIFT 11 /* nr of bits of precision */ | |
94 | #define FIXED_1 (1<<FSHIFT) /* 1.0 as fixed-point */ | |
95 | #define EXP_1 1884 /* 1/exp(5sec/1min) as fixed-point */ | |
96 | #define EXP_5 2014 /* 1/exp(5sec/5min) */ | |
97 | #define EXP_15 2037 /* 1/exp(5sec/15min) */ | |
98 | #define LOAD_INT(x) ((x) >> FSHIFT) | |
99 | #define LOAD_FRAC(x) LOAD_INT(((x) & (FIXED_1-1)) * 100) | |
beb5024e | 100 | /* |
6db4f7a3 | 101 | * This parameter is used for proc_loadavg_read(). |
102 | * 1 means use loadavg, 0 means not use. | |
103 | */ | |
104 | static int loadavg = 0; | |
a83618e2 | 105 | static volatile sig_atomic_t loadavg_stop = 0; |
0e47acaa | 106 | static int calc_hash(char *name) |
107 | { | |
108 | unsigned int hash = 0; | |
109 | unsigned int x = 0; | |
110 | /* ELFHash algorithm. */ | |
111 | while (*name) { | |
112 | hash = (hash << 4) + *name++; | |
113 | x = hash & 0xf0000000; | |
114 | if (x != 0) | |
115 | hash ^= (x >> 24); | |
116 | hash &= ~x; | |
117 | } | |
118 | return ((hash & 0x7fffffff) % LOAD_SIZE); | |
119 | } | |
120 | ||
121 | struct load_node { | |
122 | char *cg; /*cg */ | |
123 | unsigned long avenrun[3]; /* Load averages */ | |
124 | unsigned int run_pid; | |
125 | unsigned int total_pid; | |
126 | unsigned int last_pid; | |
127 | int cfd; /* The file descriptor of the mounted cgroup */ | |
128 | struct load_node *next; | |
129 | struct load_node **pre; | |
130 | }; | |
131 | ||
132 | struct load_head { | |
133 | /* | |
134 | * The lock is about insert load_node and refresh load_node.To the first | |
135 | * load_node of each hash bucket, insert and refresh in this hash bucket is | |
136 | * mutually exclusive. | |
137 | */ | |
138 | pthread_mutex_t lock; | |
139 | /* | |
140 | * The rdlock is about read loadavg and delete load_node.To each hash | |
141 | * bucket, read and delete is mutually exclusive. But at the same time, we | |
142 | * allow paratactic read operation. This rdlock is at list level. | |
143 | */ | |
144 | pthread_rwlock_t rdlock; | |
145 | /* | |
146 | * The rilock is about read loadavg and insert load_node.To the first | |
147 | * load_node of each hash bucket, read and insert is mutually exclusive. | |
148 | * But at the same time, we allow paratactic read operation. | |
149 | */ | |
150 | pthread_rwlock_t rilock; | |
151 | struct load_node *next; | |
152 | }; | |
153 | ||
154 | static struct load_head load_hash[LOAD_SIZE]; /* hash table */ | |
155 | /* | |
156 | * init_load initialize the hash table. | |
157 | * Return 0 on success, return -1 on failure. | |
158 | */ | |
159 | static int init_load(void) | |
160 | { | |
161 | int i; | |
162 | int ret; | |
163 | ||
164 | for (i = 0; i < LOAD_SIZE; i++) { | |
165 | load_hash[i].next = NULL; | |
166 | ret = pthread_mutex_init(&load_hash[i].lock, NULL); | |
167 | if (ret != 0) { | |
168 | lxcfs_error("%s\n", "Failed to initialize lock"); | |
169 | goto out3; | |
170 | } | |
171 | ret = pthread_rwlock_init(&load_hash[i].rdlock, NULL); | |
172 | if (ret != 0) { | |
173 | lxcfs_error("%s\n", "Failed to initialize rdlock"); | |
174 | goto out2; | |
175 | } | |
176 | ret = pthread_rwlock_init(&load_hash[i].rilock, NULL); | |
177 | if (ret != 0) { | |
178 | lxcfs_error("%s\n", "Failed to initialize rilock"); | |
179 | goto out1; | |
180 | } | |
181 | } | |
182 | return 0; | |
183 | out1: | |
184 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
185 | out2: | |
186 | pthread_mutex_destroy(&load_hash[i].lock); | |
187 | out3: | |
188 | while (i > 0) { | |
189 | i--; | |
190 | pthread_mutex_destroy(&load_hash[i].lock); | |
191 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
192 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
193 | } | |
194 | return -1; | |
195 | } | |
196 | ||
197 | static void insert_node(struct load_node **n, int locate) | |
198 | { | |
199 | struct load_node *f; | |
200 | ||
201 | pthread_mutex_lock(&load_hash[locate].lock); | |
202 | pthread_rwlock_wrlock(&load_hash[locate].rilock); | |
203 | f = load_hash[locate].next; | |
204 | load_hash[locate].next = *n; | |
205 | ||
206 | (*n)->pre = &(load_hash[locate].next); | |
207 | if (f) | |
208 | f->pre = &((*n)->next); | |
209 | (*n)->next = f; | |
210 | pthread_mutex_unlock(&load_hash[locate].lock); | |
211 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
212 | } | |
213 | /* | |
214 | * locate_node() finds special node. Not return NULL means success. | |
215 | * It should be noted that rdlock isn't unlocked at the end of code | |
216 | * because this function is used to read special node. Delete is not | |
217 | * allowed before read has ended. | |
218 | * unlock rdlock only in proc_loadavg_read(). | |
219 | */ | |
220 | static struct load_node *locate_node(char *cg, int locate) | |
221 | { | |
222 | struct load_node *f = NULL; | |
223 | int i = 0; | |
224 | ||
225 | pthread_rwlock_rdlock(&load_hash[locate].rilock); | |
226 | pthread_rwlock_rdlock(&load_hash[locate].rdlock); | |
227 | if (load_hash[locate].next == NULL) { | |
228 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
229 | return f; | |
230 | } | |
231 | f = load_hash[locate].next; | |
232 | pthread_rwlock_unlock(&load_hash[locate].rilock); | |
233 | while (f && ((i = strcmp(f->cg, cg)) != 0)) | |
234 | f = f->next; | |
235 | return f; | |
236 | } | |
237 | /* Delete the load_node n and return the next node of it. */ | |
238 | static struct load_node *del_node(struct load_node *n, int locate) | |
239 | { | |
240 | struct load_node *g; | |
241 | ||
242 | pthread_rwlock_wrlock(&load_hash[locate].rdlock); | |
243 | if (n->next == NULL) { | |
244 | *(n->pre) = NULL; | |
245 | } else { | |
246 | *(n->pre) = n->next; | |
247 | n->next->pre = n->pre; | |
248 | } | |
249 | g = n->next; | |
250 | free(n->cg); | |
251 | free(n); | |
252 | pthread_rwlock_unlock(&load_hash[locate].rdlock); | |
253 | return g; | |
254 | } | |
255 | ||
a83618e2 | 256 | static void load_free(void) |
9c480eb7 | 257 | { |
258 | int i; | |
259 | struct load_node *f, *p; | |
260 | ||
261 | for (i = 0; i < LOAD_SIZE; i++) { | |
262 | pthread_mutex_lock(&load_hash[i].lock); | |
263 | pthread_rwlock_wrlock(&load_hash[i].rilock); | |
264 | pthread_rwlock_wrlock(&load_hash[i].rdlock); | |
265 | if (load_hash[i].next == NULL) { | |
266 | pthread_mutex_unlock(&load_hash[i].lock); | |
267 | pthread_mutex_destroy(&load_hash[i].lock); | |
268 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
269 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
270 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
271 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
272 | continue; | |
273 | } | |
274 | for (f = load_hash[i].next; f; ) { | |
275 | free(f->cg); | |
276 | p = f->next; | |
277 | free(f); | |
278 | f = p; | |
279 | } | |
280 | pthread_mutex_unlock(&load_hash[i].lock); | |
281 | pthread_mutex_destroy(&load_hash[i].lock); | |
282 | pthread_rwlock_unlock(&load_hash[i].rilock); | |
283 | pthread_rwlock_destroy(&load_hash[i].rilock); | |
284 | pthread_rwlock_unlock(&load_hash[i].rdlock); | |
285 | pthread_rwlock_destroy(&load_hash[i].rdlock); | |
286 | } | |
287 | } | |
f34de69a CB |
288 | /* Reserve buffer size to account for file size changes. */ |
289 | #define BUF_RESERVE_SIZE 512 | |
237e200e SH |
290 | |
291 | /* | |
292 | * A table caching which pid is init for a pid namespace. | |
293 | * When looking up which pid is init for $qpid, we first | |
294 | * 1. Stat /proc/$qpid/ns/pid. | |
295 | * 2. Check whether the ino_t is in our store. | |
296 | * a. if not, fork a child in qpid's ns to send us | |
297 | * ucred.pid = 1, and read the initpid. Cache | |
298 | * initpid and creation time for /proc/initpid | |
299 | * in a new store entry. | |
300 | * b. if so, verify that /proc/initpid still matches | |
301 | * what we have saved. If not, clear the store | |
302 | * entry and go back to a. If so, return the | |
303 | * cached initpid. | |
304 | */ | |
305 | struct pidns_init_store { | |
306 | ino_t ino; // inode number for /proc/$pid/ns/pid | |
307 | pid_t initpid; // the pid of nit in that ns | |
308 | long int ctime; // the time at which /proc/$initpid was created | |
309 | struct pidns_init_store *next; | |
310 | long int lastcheck; | |
311 | }; | |
312 | ||
313 | /* lol - look at how they are allocated in the kernel */ | |
314 | #define PIDNS_HASH_SIZE 4096 | |
315 | #define HASH(x) ((x) % PIDNS_HASH_SIZE) | |
316 | ||
317 | static struct pidns_init_store *pidns_hash_table[PIDNS_HASH_SIZE]; | |
318 | static pthread_mutex_t pidns_store_mutex = PTHREAD_MUTEX_INITIALIZER; | |
319 | static void lock_mutex(pthread_mutex_t *l) | |
320 | { | |
321 | int ret; | |
322 | ||
323 | if ((ret = pthread_mutex_lock(l)) != 0) { | |
b8defc3d | 324 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
325 | exit(1); |
326 | } | |
327 | } | |
328 | ||
29a73c2f CB |
329 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. |
330 | * Number of hierarchies mounted. */ | |
331 | static int num_hierarchies; | |
332 | ||
333 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
334 | * Hierachies mounted {cpuset, blkio, ...}: | |
335 | * Initialized via __constructor__ collect_and_mount_subsystems(). */ | |
336 | static char **hierarchies; | |
337 | ||
338 | /* READ-ONLY after __constructor__ collect_and_mount_subsystems() has run. | |
339 | * Open file descriptors: | |
340 | * @fd_hierarchies[i] refers to cgroup @hierarchies[i]. They are mounted in a | |
341 | * private mount namespace. | |
342 | * Initialized via __constructor__ collect_and_mount_subsystems(). | |
343 | * @fd_hierarchies[i] can be used to perform file operations on the cgroup | |
344 | * mounts and respective files in the private namespace even when located in | |
345 | * another namespace using the *at() family of functions | |
346 | * {openat(), fchownat(), ...}. */ | |
347 | static int *fd_hierarchies; | |
a257a8ee | 348 | static int cgroup_mount_ns_fd = -1; |
29a73c2f | 349 | |
237e200e SH |
350 | static void unlock_mutex(pthread_mutex_t *l) |
351 | { | |
352 | int ret; | |
353 | ||
354 | if ((ret = pthread_mutex_unlock(l)) != 0) { | |
b8defc3d | 355 | lxcfs_error("returned:%d %s\n", ret, strerror(ret)); |
237e200e SH |
356 | exit(1); |
357 | } | |
358 | } | |
359 | ||
360 | static void store_lock(void) | |
361 | { | |
362 | lock_mutex(&pidns_store_mutex); | |
363 | } | |
364 | ||
365 | static void store_unlock(void) | |
366 | { | |
367 | unlock_mutex(&pidns_store_mutex); | |
368 | } | |
369 | ||
370 | /* Must be called under store_lock */ | |
371 | static bool initpid_still_valid(struct pidns_init_store *e, struct stat *nsfdsb) | |
372 | { | |
373 | struct stat initsb; | |
374 | char fnam[100]; | |
375 | ||
376 | snprintf(fnam, 100, "/proc/%d", e->initpid); | |
377 | if (stat(fnam, &initsb) < 0) | |
378 | return false; | |
7dd6560a CB |
379 | |
380 | lxcfs_debug("Comparing ctime %ld == %ld for pid %d.\n", e->ctime, | |
381 | initsb.st_ctime, e->initpid); | |
382 | ||
237e200e SH |
383 | if (e->ctime != initsb.st_ctime) |
384 | return false; | |
385 | return true; | |
386 | } | |
387 | ||
388 | /* Must be called under store_lock */ | |
389 | static void remove_initpid(struct pidns_init_store *e) | |
390 | { | |
391 | struct pidns_init_store *tmp; | |
392 | int h; | |
393 | ||
7dd6560a CB |
394 | lxcfs_debug("Remove_initpid: removing entry for %d.\n", e->initpid); |
395 | ||
237e200e SH |
396 | h = HASH(e->ino); |
397 | if (pidns_hash_table[h] == e) { | |
398 | pidns_hash_table[h] = e->next; | |
399 | free(e); | |
400 | return; | |
401 | } | |
402 | ||
403 | tmp = pidns_hash_table[h]; | |
404 | while (tmp) { | |
405 | if (tmp->next == e) { | |
406 | tmp->next = e->next; | |
407 | free(e); | |
408 | return; | |
409 | } | |
410 | tmp = tmp->next; | |
411 | } | |
412 | } | |
413 | ||
414 | #define PURGE_SECS 5 | |
415 | /* Must be called under store_lock */ | |
416 | static void prune_initpid_store(void) | |
417 | { | |
418 | static long int last_prune = 0; | |
419 | struct pidns_init_store *e, *prev, *delme; | |
420 | long int now, threshold; | |
421 | int i; | |
422 | ||
423 | if (!last_prune) { | |
424 | last_prune = time(NULL); | |
425 | return; | |
426 | } | |
427 | now = time(NULL); | |
428 | if (now < last_prune + PURGE_SECS) | |
429 | return; | |
7dd6560a CB |
430 | |
431 | lxcfs_debug("%s\n", "Pruning."); | |
432 | ||
237e200e SH |
433 | last_prune = now; |
434 | threshold = now - 2 * PURGE_SECS; | |
435 | ||
436 | for (i = 0; i < PIDNS_HASH_SIZE; i++) { | |
437 | for (prev = NULL, e = pidns_hash_table[i]; e; ) { | |
438 | if (e->lastcheck < threshold) { | |
7dd6560a CB |
439 | |
440 | lxcfs_debug("Removing cached entry for %d.\n", e->initpid); | |
441 | ||
237e200e SH |
442 | delme = e; |
443 | if (prev) | |
444 | prev->next = e->next; | |
445 | else | |
446 | pidns_hash_table[i] = e->next; | |
447 | e = e->next; | |
448 | free(delme); | |
449 | } else { | |
450 | prev = e; | |
451 | e = e->next; | |
452 | } | |
453 | } | |
454 | } | |
455 | } | |
456 | ||
457 | /* Must be called under store_lock */ | |
458 | static void save_initpid(struct stat *sb, pid_t pid) | |
459 | { | |
460 | struct pidns_init_store *e; | |
461 | char fpath[100]; | |
462 | struct stat procsb; | |
463 | int h; | |
464 | ||
7dd6560a CB |
465 | lxcfs_debug("Save_initpid: adding entry for %d.\n", pid); |
466 | ||
237e200e SH |
467 | snprintf(fpath, 100, "/proc/%d", pid); |
468 | if (stat(fpath, &procsb) < 0) | |
469 | return; | |
470 | do { | |
471 | e = malloc(sizeof(*e)); | |
472 | } while (!e); | |
473 | e->ino = sb->st_ino; | |
474 | e->initpid = pid; | |
475 | e->ctime = procsb.st_ctime; | |
476 | h = HASH(e->ino); | |
477 | e->next = pidns_hash_table[h]; | |
478 | e->lastcheck = time(NULL); | |
479 | pidns_hash_table[h] = e; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store | |
484 | * entry for the inode number and creation time. Verify that the init pid | |
485 | * is still valid. If not, remove it. Return the entry if valid, NULL | |
486 | * otherwise. | |
487 | * Must be called under store_lock | |
488 | */ | |
489 | static struct pidns_init_store *lookup_verify_initpid(struct stat *sb) | |
490 | { | |
491 | int h = HASH(sb->st_ino); | |
492 | struct pidns_init_store *e = pidns_hash_table[h]; | |
493 | ||
494 | while (e) { | |
495 | if (e->ino == sb->st_ino) { | |
496 | if (initpid_still_valid(e, sb)) { | |
497 | e->lastcheck = time(NULL); | |
498 | return e; | |
499 | } | |
500 | remove_initpid(e); | |
501 | return NULL; | |
502 | } | |
503 | e = e->next; | |
504 | } | |
505 | ||
506 | return NULL; | |
507 | } | |
508 | ||
0f657ce3 | 509 | static int is_dir(const char *path, int fd) |
237e200e SH |
510 | { |
511 | struct stat statbuf; | |
0f657ce3 | 512 | int ret = fstatat(fd, path, &statbuf, fd); |
237e200e SH |
513 | if (ret == 0 && S_ISDIR(statbuf.st_mode)) |
514 | return 1; | |
515 | return 0; | |
516 | } | |
517 | ||
518 | static char *must_copy_string(const char *str) | |
519 | { | |
520 | char *dup = NULL; | |
521 | if (!str) | |
522 | return NULL; | |
523 | do { | |
524 | dup = strdup(str); | |
525 | } while (!dup); | |
526 | ||
527 | return dup; | |
528 | } | |
529 | ||
530 | static inline void drop_trailing_newlines(char *s) | |
531 | { | |
532 | int l; | |
533 | ||
534 | for (l=strlen(s); l>0 && s[l-1] == '\n'; l--) | |
535 | s[l-1] = '\0'; | |
536 | } | |
537 | ||
538 | #define BATCH_SIZE 50 | |
539 | static void dorealloc(char **mem, size_t oldlen, size_t newlen) | |
540 | { | |
541 | int newbatches = (newlen / BATCH_SIZE) + 1; | |
542 | int oldbatches = (oldlen / BATCH_SIZE) + 1; | |
543 | ||
544 | if (!*mem || newbatches > oldbatches) { | |
545 | char *tmp; | |
546 | do { | |
547 | tmp = realloc(*mem, newbatches * BATCH_SIZE); | |
548 | } while (!tmp); | |
549 | *mem = tmp; | |
550 | } | |
551 | } | |
552 | static void append_line(char **contents, size_t *len, char *line, ssize_t linelen) | |
553 | { | |
554 | size_t newlen = *len + linelen; | |
555 | dorealloc(contents, *len, newlen + 1); | |
556 | memcpy(*contents + *len, line, linelen+1); | |
557 | *len = newlen; | |
558 | } | |
559 | ||
60f2ae53 | 560 | static char *slurp_file(const char *from, int fd) |
237e200e SH |
561 | { |
562 | char *line = NULL; | |
563 | char *contents = NULL; | |
60f2ae53 | 564 | FILE *f = fdopen(fd, "r"); |
237e200e SH |
565 | size_t len = 0, fulllen = 0; |
566 | ssize_t linelen; | |
567 | ||
568 | if (!f) | |
569 | return NULL; | |
570 | ||
571 | while ((linelen = getline(&line, &len, f)) != -1) { | |
572 | append_line(&contents, &fulllen, line, linelen); | |
573 | } | |
574 | fclose(f); | |
575 | ||
576 | if (contents) | |
577 | drop_trailing_newlines(contents); | |
578 | free(line); | |
579 | return contents; | |
580 | } | |
581 | ||
ba59ea09 | 582 | static bool write_string(const char *fnam, const char *string, int fd) |
237e200e SH |
583 | { |
584 | FILE *f; | |
585 | size_t len, ret; | |
586 | ||
beb5024e CB |
587 | f = fdopen(fd, "w"); |
588 | if (!f) | |
237e200e | 589 | return false; |
beb5024e | 590 | |
237e200e SH |
591 | len = strlen(string); |
592 | ret = fwrite(string, 1, len, f); | |
593 | if (ret != len) { | |
beb5024e CB |
594 | lxcfs_error("%s - Error writing \"%s\" to \"%s\"\n", |
595 | strerror(errno), string, fnam); | |
237e200e SH |
596 | fclose(f); |
597 | return false; | |
598 | } | |
beb5024e | 599 | |
237e200e | 600 | if (fclose(f) < 0) { |
beb5024e | 601 | lxcfs_error("%s - Failed to close \"%s\"\n", strerror(errno), fnam); |
237e200e SH |
602 | return false; |
603 | } | |
beb5024e | 604 | |
237e200e SH |
605 | return true; |
606 | } | |
607 | ||
237e200e SH |
608 | struct cgfs_files { |
609 | char *name; | |
610 | uint32_t uid, gid; | |
611 | uint32_t mode; | |
612 | }; | |
613 | ||
0619767c | 614 | #define ALLOC_NUM 20 |
237e200e SH |
615 | static bool store_hierarchy(char *stridx, char *h) |
616 | { | |
0619767c SH |
617 | if (num_hierarchies % ALLOC_NUM == 0) { |
618 | size_t n = (num_hierarchies / ALLOC_NUM) + 1; | |
619 | n *= ALLOC_NUM; | |
620 | char **tmp = realloc(hierarchies, n * sizeof(char *)); | |
0619767c | 621 | if (!tmp) { |
b8defc3d | 622 | lxcfs_error("%s\n", strerror(errno)); |
0619767c SH |
623 | exit(1); |
624 | } | |
237e200e | 625 | hierarchies = tmp; |
237e200e | 626 | } |
f676eb79 | 627 | |
0619767c | 628 | hierarchies[num_hierarchies++] = must_copy_string(h); |
237e200e SH |
629 | return true; |
630 | } | |
631 | ||
632 | static void print_subsystems(void) | |
633 | { | |
634 | int i; | |
635 | ||
a257a8ee | 636 | fprintf(stderr, "mount namespace: %d\n", cgroup_mount_ns_fd); |
cc97d34c | 637 | fprintf(stderr, "hierarchies:\n"); |
237e200e SH |
638 | for (i = 0; i < num_hierarchies; i++) { |
639 | if (hierarchies[i]) | |
b8defc3d CB |
640 | fprintf(stderr, " %2d: fd: %3d: %s\n", i, |
641 | fd_hierarchies[i], hierarchies[i]); | |
237e200e SH |
642 | } |
643 | } | |
644 | ||
645 | static bool in_comma_list(const char *needle, const char *haystack) | |
646 | { | |
647 | const char *s = haystack, *e; | |
648 | size_t nlen = strlen(needle); | |
649 | ||
06081b29 | 650 | while (*s && (e = strchr(s, ','))) { |
237e200e SH |
651 | if (nlen != e - s) { |
652 | s = e + 1; | |
653 | continue; | |
654 | } | |
655 | if (strncmp(needle, s, nlen) == 0) | |
656 | return true; | |
657 | s = e + 1; | |
658 | } | |
659 | if (strcmp(needle, s) == 0) | |
660 | return true; | |
661 | return false; | |
662 | } | |
663 | ||
664 | /* do we need to do any massaging here? I'm not sure... */ | |
5dd3e6fd CB |
665 | /* Return the mounted controller and store the corresponding open file descriptor |
666 | * referring to the controller mountpoint in the private lxcfs namespace in | |
667 | * @cfd. | |
668 | */ | |
669 | static char *find_mounted_controller(const char *controller, int *cfd) | |
237e200e SH |
670 | { |
671 | int i; | |
672 | ||
673 | for (i = 0; i < num_hierarchies; i++) { | |
674 | if (!hierarchies[i]) | |
675 | continue; | |
5dd3e6fd CB |
676 | if (strcmp(hierarchies[i], controller) == 0) { |
677 | *cfd = fd_hierarchies[i]; | |
237e200e | 678 | return hierarchies[i]; |
5dd3e6fd CB |
679 | } |
680 | if (in_comma_list(controller, hierarchies[i])) { | |
681 | *cfd = fd_hierarchies[i]; | |
237e200e | 682 | return hierarchies[i]; |
5dd3e6fd | 683 | } |
237e200e SH |
684 | } |
685 | ||
686 | return NULL; | |
687 | } | |
688 | ||
689 | bool cgfs_set_value(const char *controller, const char *cgroup, const char *file, | |
690 | const char *value) | |
691 | { | |
ba59ea09 | 692 | int ret, fd, cfd; |
237e200e | 693 | size_t len; |
f5a6d92e | 694 | char *fnam, *tmpc; |
237e200e | 695 | |
f5a6d92e | 696 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
697 | if (!tmpc) |
698 | return false; | |
f5a6d92e CB |
699 | |
700 | /* Make sure we pass a relative path to *at() family of functions. | |
701 | * . + /cgroup + / + file + \0 | |
702 | */ | |
ba59ea09 | 703 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 704 | fnam = alloca(len); |
ba59ea09 CB |
705 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
706 | if (ret < 0 || (size_t)ret >= len) | |
707 | return false; | |
708 | ||
709 | fd = openat(cfd, fnam, O_WRONLY); | |
710 | if (fd < 0) | |
711 | return false; | |
f676eb79 | 712 | |
ba59ea09 | 713 | return write_string(fnam, value, fd); |
237e200e SH |
714 | } |
715 | ||
716 | // Chown all the files in the cgroup directory. We do this when we create | |
717 | // a cgroup on behalf of a user. | |
f23fe717 | 718 | static void chown_all_cgroup_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e | 719 | { |
f23fe717 | 720 | struct dirent *direntp; |
237e200e SH |
721 | char path[MAXPATHLEN]; |
722 | size_t len; | |
723 | DIR *d; | |
f23fe717 | 724 | int fd1, ret; |
237e200e SH |
725 | |
726 | len = strlen(dirname); | |
727 | if (len >= MAXPATHLEN) { | |
b8defc3d | 728 | lxcfs_error("Pathname too long: %s\n", dirname); |
237e200e SH |
729 | return; |
730 | } | |
731 | ||
f23fe717 CB |
732 | fd1 = openat(fd, dirname, O_DIRECTORY); |
733 | if (fd1 < 0) | |
734 | return; | |
735 | ||
736 | d = fdopendir(fd1); | |
237e200e | 737 | if (!d) { |
b8defc3d | 738 | lxcfs_error("Failed to open %s\n", dirname); |
237e200e SH |
739 | return; |
740 | } | |
741 | ||
f23fe717 | 742 | while ((direntp = readdir(d))) { |
237e200e SH |
743 | if (!strcmp(direntp->d_name, ".") || !strcmp(direntp->d_name, "..")) |
744 | continue; | |
745 | ret = snprintf(path, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
746 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 747 | lxcfs_error("Pathname too long under %s\n", dirname); |
237e200e SH |
748 | continue; |
749 | } | |
f23fe717 | 750 | if (fchownat(fd, path, uid, gid, 0) < 0) |
b8defc3d | 751 | lxcfs_error("Failed to chown file %s to %u:%u", path, uid, gid); |
237e200e SH |
752 | } |
753 | closedir(d); | |
754 | } | |
755 | ||
756 | int cgfs_create(const char *controller, const char *cg, uid_t uid, gid_t gid) | |
757 | { | |
5dd3e6fd | 758 | int cfd; |
237e200e | 759 | size_t len; |
f5a6d92e | 760 | char *dirnam, *tmpc; |
237e200e | 761 | |
f5a6d92e | 762 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
763 | if (!tmpc) |
764 | return -EINVAL; | |
f5a6d92e CB |
765 | |
766 | /* Make sure we pass a relative path to *at() family of functions. | |
767 | * . + /cg + \0 | |
768 | */ | |
f23fe717 | 769 | len = strlen(cg) + 2; |
237e200e | 770 | dirnam = alloca(len); |
f23fe717 | 771 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
237e200e | 772 | |
f23fe717 | 773 | if (mkdirat(cfd, dirnam, 0755) < 0) |
237e200e SH |
774 | return -errno; |
775 | ||
776 | if (uid == 0 && gid == 0) | |
777 | return 0; | |
778 | ||
f23fe717 | 779 | if (fchownat(cfd, dirnam, uid, gid, 0) < 0) |
237e200e SH |
780 | return -errno; |
781 | ||
f23fe717 | 782 | chown_all_cgroup_files(dirnam, uid, gid, cfd); |
237e200e SH |
783 | |
784 | return 0; | |
785 | } | |
786 | ||
7213ec5c | 787 | static bool recursive_rmdir(const char *dirname, int fd, const int cfd) |
237e200e | 788 | { |
b7672ded | 789 | struct dirent *direntp; |
237e200e SH |
790 | DIR *dir; |
791 | bool ret = false; | |
792 | char pathname[MAXPATHLEN]; | |
b7672ded CB |
793 | int dupfd; |
794 | ||
795 | dupfd = dup(fd); // fdopendir() does bad things once it uses an fd. | |
796 | if (dupfd < 0) | |
797 | return false; | |
237e200e | 798 | |
b7672ded | 799 | dir = fdopendir(dupfd); |
237e200e | 800 | if (!dir) { |
7dd6560a | 801 | lxcfs_debug("Failed to open %s: %s.\n", dirname, strerror(errno)); |
7213ec5c | 802 | close(dupfd); |
237e200e SH |
803 | return false; |
804 | } | |
805 | ||
b7672ded | 806 | while ((direntp = readdir(dir))) { |
237e200e SH |
807 | struct stat mystat; |
808 | int rc; | |
809 | ||
237e200e SH |
810 | if (!strcmp(direntp->d_name, ".") || |
811 | !strcmp(direntp->d_name, "..")) | |
812 | continue; | |
813 | ||
814 | rc = snprintf(pathname, MAXPATHLEN, "%s/%s", dirname, direntp->d_name); | |
815 | if (rc < 0 || rc >= MAXPATHLEN) { | |
b8defc3d | 816 | lxcfs_error("%s\n", "Pathname too long."); |
237e200e SH |
817 | continue; |
818 | } | |
819 | ||
2e81a5e3 CB |
820 | rc = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
821 | if (rc) { | |
7dd6560a | 822 | lxcfs_debug("Failed to stat %s: %s.\n", pathname, strerror(errno)); |
237e200e SH |
823 | continue; |
824 | } | |
7dd6560a | 825 | if (S_ISDIR(mystat.st_mode)) |
2e81a5e3 | 826 | if (!recursive_rmdir(pathname, fd, cfd)) |
7dd6560a | 827 | lxcfs_debug("Error removing %s.\n", pathname); |
237e200e SH |
828 | } |
829 | ||
830 | ret = true; | |
831 | if (closedir(dir) < 0) { | |
b8defc3d | 832 | lxcfs_error("Failed to close directory %s: %s\n", dirname, strerror(errno)); |
237e200e SH |
833 | ret = false; |
834 | } | |
835 | ||
2e81a5e3 | 836 | if (unlinkat(cfd, dirname, AT_REMOVEDIR) < 0) { |
7dd6560a | 837 | lxcfs_debug("Failed to delete %s: %s.\n", dirname, strerror(errno)); |
237e200e SH |
838 | ret = false; |
839 | } | |
7213ec5c CB |
840 | |
841 | close(dupfd); | |
237e200e SH |
842 | |
843 | return ret; | |
844 | } | |
845 | ||
846 | bool cgfs_remove(const char *controller, const char *cg) | |
847 | { | |
b7672ded | 848 | int fd, cfd; |
237e200e | 849 | size_t len; |
f5a6d92e | 850 | char *dirnam, *tmpc; |
7213ec5c | 851 | bool bret; |
237e200e | 852 | |
f5a6d92e | 853 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
854 | if (!tmpc) |
855 | return false; | |
f5a6d92e CB |
856 | |
857 | /* Make sure we pass a relative path to *at() family of functions. | |
858 | * . + /cg + \0 | |
859 | */ | |
b7672ded | 860 | len = strlen(cg) + 2; |
237e200e | 861 | dirnam = alloca(len); |
b7672ded CB |
862 | snprintf(dirnam, len, "%s%s", *cg == '/' ? "." : "", cg); |
863 | ||
864 | fd = openat(cfd, dirnam, O_DIRECTORY); | |
865 | if (fd < 0) | |
866 | return false; | |
867 | ||
7213ec5c CB |
868 | bret = recursive_rmdir(dirnam, fd, cfd); |
869 | close(fd); | |
870 | return bret; | |
237e200e SH |
871 | } |
872 | ||
873 | bool cgfs_chmod_file(const char *controller, const char *file, mode_t mode) | |
874 | { | |
5dd3e6fd | 875 | int cfd; |
237e200e | 876 | size_t len; |
f5a6d92e | 877 | char *pathname, *tmpc; |
237e200e | 878 | |
f5a6d92e | 879 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
880 | if (!tmpc) |
881 | return false; | |
f5a6d92e CB |
882 | |
883 | /* Make sure we pass a relative path to *at() family of functions. | |
884 | * . + /file + \0 | |
885 | */ | |
534690b4 | 886 | len = strlen(file) + 2; |
237e200e | 887 | pathname = alloca(len); |
534690b4 CB |
888 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
889 | if (fchmodat(cfd, pathname, mode, 0) < 0) | |
237e200e SH |
890 | return false; |
891 | return true; | |
892 | } | |
893 | ||
0f657ce3 | 894 | static int chown_tasks_files(const char *dirname, uid_t uid, gid_t gid, int fd) |
237e200e SH |
895 | { |
896 | size_t len; | |
897 | char *fname; | |
898 | ||
899 | len = strlen(dirname) + strlen("/cgroup.procs") + 1; | |
900 | fname = alloca(len); | |
901 | snprintf(fname, len, "%s/tasks", dirname); | |
0f657ce3 | 902 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
903 | return -errno; |
904 | snprintf(fname, len, "%s/cgroup.procs", dirname); | |
0f657ce3 | 905 | if (fchownat(fd, fname, uid, gid, 0) != 0) |
237e200e SH |
906 | return -errno; |
907 | return 0; | |
908 | } | |
909 | ||
910 | int cgfs_chown_file(const char *controller, const char *file, uid_t uid, gid_t gid) | |
911 | { | |
5dd3e6fd | 912 | int cfd; |
237e200e | 913 | size_t len; |
f5a6d92e | 914 | char *pathname, *tmpc; |
237e200e | 915 | |
f5a6d92e | 916 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
917 | if (!tmpc) |
918 | return -EINVAL; | |
f5a6d92e CB |
919 | |
920 | /* Make sure we pass a relative path to *at() family of functions. | |
921 | * . + /file + \0 | |
922 | */ | |
0f657ce3 | 923 | len = strlen(file) + 2; |
237e200e | 924 | pathname = alloca(len); |
0f657ce3 CB |
925 | snprintf(pathname, len, "%s%s", *file == '/' ? "." : "", file); |
926 | if (fchownat(cfd, pathname, uid, gid, 0) < 0) | |
237e200e SH |
927 | return -errno; |
928 | ||
0f657ce3 | 929 | if (is_dir(pathname, cfd)) |
237e200e | 930 | // like cgmanager did, we want to chown the tasks file as well |
0f657ce3 | 931 | return chown_tasks_files(pathname, uid, gid, cfd); |
237e200e SH |
932 | |
933 | return 0; | |
934 | } | |
935 | ||
936 | FILE *open_pids_file(const char *controller, const char *cgroup) | |
937 | { | |
3ffd08ee | 938 | int fd, cfd; |
237e200e | 939 | size_t len; |
f5a6d92e | 940 | char *pathname, *tmpc; |
237e200e | 941 | |
f5a6d92e | 942 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
943 | if (!tmpc) |
944 | return NULL; | |
f5a6d92e CB |
945 | |
946 | /* Make sure we pass a relative path to *at() family of functions. | |
947 | * . + /cgroup + / "cgroup.procs" + \0 | |
948 | */ | |
3ffd08ee | 949 | len = strlen(cgroup) + strlen("cgroup.procs") + 3; |
237e200e | 950 | pathname = alloca(len); |
3ffd08ee CB |
951 | snprintf(pathname, len, "%s%s/cgroup.procs", *cgroup == '/' ? "." : "", cgroup); |
952 | ||
953 | fd = openat(cfd, pathname, O_WRONLY); | |
954 | if (fd < 0) | |
955 | return NULL; | |
956 | ||
957 | return fdopen(fd, "w"); | |
237e200e SH |
958 | } |
959 | ||
f366da65 WB |
960 | static bool cgfs_iterate_cgroup(const char *controller, const char *cgroup, bool directories, |
961 | void ***list, size_t typesize, | |
962 | void* (*iterator)(const char*, const char*, const char*)) | |
237e200e | 963 | { |
4ea38a4c | 964 | int cfd, fd, ret; |
237e200e | 965 | size_t len; |
4ea38a4c | 966 | char *cg, *tmpc; |
237e200e | 967 | char pathname[MAXPATHLEN]; |
f366da65 | 968 | size_t sz = 0, asz = 0; |
4ea38a4c | 969 | struct dirent *dirent; |
237e200e | 970 | DIR *dir; |
237e200e | 971 | |
4ea38a4c | 972 | tmpc = find_mounted_controller(controller, &cfd); |
f366da65 | 973 | *list = NULL; |
237e200e | 974 | if (!tmpc) |
e97c834b | 975 | return false; |
237e200e | 976 | |
f5a6d92e | 977 | /* Make sure we pass a relative path to *at() family of functions. */ |
4ea38a4c CB |
978 | len = strlen(cgroup) + 1 /* . */ + 1 /* \0 */; |
979 | cg = alloca(len); | |
980 | ret = snprintf(cg, len, "%s%s", *cgroup == '/' ? "." : "", cgroup); | |
981 | if (ret < 0 || (size_t)ret >= len) { | |
b8defc3d | 982 | lxcfs_error("Pathname too long under %s\n", cgroup); |
4ea38a4c CB |
983 | return false; |
984 | } | |
237e200e | 985 | |
4ea38a4c CB |
986 | fd = openat(cfd, cg, O_DIRECTORY); |
987 | if (fd < 0) | |
988 | return false; | |
989 | ||
990 | dir = fdopendir(fd); | |
237e200e SH |
991 | if (!dir) |
992 | return false; | |
993 | ||
4ea38a4c | 994 | while ((dirent = readdir(dir))) { |
237e200e | 995 | struct stat mystat; |
237e200e | 996 | |
4ea38a4c CB |
997 | if (!strcmp(dirent->d_name, ".") || |
998 | !strcmp(dirent->d_name, "..")) | |
237e200e SH |
999 | continue; |
1000 | ||
4ea38a4c CB |
1001 | ret = snprintf(pathname, MAXPATHLEN, "%s/%s", cg, dirent->d_name); |
1002 | if (ret < 0 || ret >= MAXPATHLEN) { | |
b8defc3d | 1003 | lxcfs_error("Pathname too long under %s\n", cg); |
237e200e SH |
1004 | continue; |
1005 | } | |
1006 | ||
4ea38a4c | 1007 | ret = fstatat(cfd, pathname, &mystat, AT_SYMLINK_NOFOLLOW); |
237e200e | 1008 | if (ret) { |
b8defc3d | 1009 | lxcfs_error("Failed to stat %s: %s\n", pathname, strerror(errno)); |
237e200e SH |
1010 | continue; |
1011 | } | |
f366da65 WB |
1012 | if ((!directories && !S_ISREG(mystat.st_mode)) || |
1013 | (directories && !S_ISDIR(mystat.st_mode))) | |
237e200e SH |
1014 | continue; |
1015 | ||
1016 | if (sz+2 >= asz) { | |
f366da65 | 1017 | void **tmp; |
237e200e SH |
1018 | asz += BATCH_SIZE; |
1019 | do { | |
f366da65 | 1020 | tmp = realloc(*list, asz * typesize); |
237e200e SH |
1021 | } while (!tmp); |
1022 | *list = tmp; | |
1023 | } | |
4ea38a4c | 1024 | (*list)[sz] = (*iterator)(controller, cg, dirent->d_name); |
237e200e SH |
1025 | (*list)[sz+1] = NULL; |
1026 | sz++; | |
1027 | } | |
1028 | if (closedir(dir) < 0) { | |
b8defc3d | 1029 | lxcfs_error("Failed closedir for %s: %s\n", cgroup, strerror(errno)); |
237e200e SH |
1030 | return false; |
1031 | } | |
1032 | return true; | |
1033 | } | |
1034 | ||
f366da65 WB |
1035 | static void *make_children_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
1036 | { | |
1037 | char *dup; | |
1038 | do { | |
1039 | dup = strdup(dir_entry); | |
1040 | } while (!dup); | |
1041 | return dup; | |
1042 | } | |
1043 | ||
1044 | bool cgfs_list_children(const char *controller, const char *cgroup, char ***list) | |
1045 | { | |
1046 | return cgfs_iterate_cgroup(controller, cgroup, true, (void***)list, sizeof(*list), &make_children_list_entry); | |
1047 | } | |
1048 | ||
237e200e SH |
1049 | void free_key(struct cgfs_files *k) |
1050 | { | |
1051 | if (!k) | |
1052 | return; | |
1053 | free(k->name); | |
1054 | free(k); | |
1055 | } | |
1056 | ||
1057 | void free_keys(struct cgfs_files **keys) | |
1058 | { | |
1059 | int i; | |
1060 | ||
1061 | if (!keys) | |
1062 | return; | |
1063 | for (i = 0; keys[i]; i++) { | |
1064 | free_key(keys[i]); | |
1065 | } | |
1066 | free(keys); | |
1067 | } | |
1068 | ||
1069 | bool cgfs_get_value(const char *controller, const char *cgroup, const char *file, char **value) | |
1070 | { | |
60f2ae53 | 1071 | int ret, fd, cfd; |
237e200e | 1072 | size_t len; |
f5a6d92e | 1073 | char *fnam, *tmpc; |
237e200e | 1074 | |
f5a6d92e | 1075 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1076 | if (!tmpc) |
1077 | return false; | |
f5a6d92e CB |
1078 | |
1079 | /* Make sure we pass a relative path to *at() family of functions. | |
1080 | * . + /cgroup + / + file + \0 | |
1081 | */ | |
60f2ae53 | 1082 | len = strlen(cgroup) + strlen(file) + 3; |
237e200e | 1083 | fnam = alloca(len); |
60f2ae53 CB |
1084 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, file); |
1085 | if (ret < 0 || (size_t)ret >= len) | |
234a820c | 1086 | return false; |
60f2ae53 CB |
1087 | |
1088 | fd = openat(cfd, fnam, O_RDONLY); | |
1089 | if (fd < 0) | |
234a820c | 1090 | return false; |
237e200e | 1091 | |
60f2ae53 | 1092 | *value = slurp_file(fnam, fd); |
237e200e SH |
1093 | return *value != NULL; |
1094 | } | |
1095 | ||
1096 | struct cgfs_files *cgfs_get_key(const char *controller, const char *cgroup, const char *file) | |
1097 | { | |
4ea38a4c | 1098 | int ret, cfd; |
237e200e | 1099 | size_t len; |
f5a6d92e | 1100 | char *fnam, *tmpc; |
237e200e SH |
1101 | struct stat sb; |
1102 | struct cgfs_files *newkey; | |
237e200e | 1103 | |
f5a6d92e | 1104 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1105 | if (!tmpc) |
1106 | return false; | |
1107 | ||
1108 | if (file && *file == '/') | |
1109 | file++; | |
1110 | ||
06081b29 | 1111 | if (file && strchr(file, '/')) |
237e200e SH |
1112 | return NULL; |
1113 | ||
f5a6d92e CB |
1114 | /* Make sure we pass a relative path to *at() family of functions. |
1115 | * . + /cgroup + / + file + \0 | |
1116 | */ | |
4ea38a4c | 1117 | len = strlen(cgroup) + 3; |
237e200e SH |
1118 | if (file) |
1119 | len += strlen(file) + 1; | |
1120 | fnam = alloca(len); | |
4ea38a4c CB |
1121 | snprintf(fnam, len, "%s%s%s%s", *cgroup == '/' ? "." : "", cgroup, |
1122 | file ? "/" : "", file ? file : ""); | |
237e200e | 1123 | |
4ea38a4c | 1124 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1125 | if (ret < 0) |
1126 | return NULL; | |
1127 | ||
1128 | do { | |
1129 | newkey = malloc(sizeof(struct cgfs_files)); | |
1130 | } while (!newkey); | |
1131 | if (file) | |
1132 | newkey->name = must_copy_string(file); | |
06081b29 CB |
1133 | else if (strrchr(cgroup, '/')) |
1134 | newkey->name = must_copy_string(strrchr(cgroup, '/')); | |
237e200e SH |
1135 | else |
1136 | newkey->name = must_copy_string(cgroup); | |
1137 | newkey->uid = sb.st_uid; | |
1138 | newkey->gid = sb.st_gid; | |
1139 | newkey->mode = sb.st_mode; | |
1140 | ||
1141 | return newkey; | |
1142 | } | |
1143 | ||
f366da65 | 1144 | static void *make_key_list_entry(const char *controller, const char *cgroup, const char *dir_entry) |
237e200e | 1145 | { |
f366da65 WB |
1146 | struct cgfs_files *entry = cgfs_get_key(controller, cgroup, dir_entry); |
1147 | if (!entry) { | |
b8defc3d CB |
1148 | lxcfs_error("Error getting files under %s:%s\n", controller, |
1149 | cgroup); | |
237e200e | 1150 | } |
f366da65 WB |
1151 | return entry; |
1152 | } | |
1153 | ||
1154 | bool cgfs_list_keys(const char *controller, const char *cgroup, struct cgfs_files ***keys) | |
1155 | { | |
1156 | return cgfs_iterate_cgroup(controller, cgroup, false, (void***)keys, sizeof(*keys), &make_key_list_entry); | |
237e200e SH |
1157 | } |
1158 | ||
1159 | bool is_child_cgroup(const char *controller, const char *cgroup, const char *f) | |
5dd3e6fd CB |
1160 | { |
1161 | int cfd; | |
1162 | size_t len; | |
f5a6d92e | 1163 | char *fnam, *tmpc; |
237e200e SH |
1164 | int ret; |
1165 | struct stat sb; | |
1166 | ||
f5a6d92e | 1167 | tmpc = find_mounted_controller(controller, &cfd); |
237e200e SH |
1168 | if (!tmpc) |
1169 | return false; | |
f5a6d92e CB |
1170 | |
1171 | /* Make sure we pass a relative path to *at() family of functions. | |
1172 | * . + /cgroup + / + f + \0 | |
1173 | */ | |
d04232f2 | 1174 | len = strlen(cgroup) + strlen(f) + 3; |
237e200e | 1175 | fnam = alloca(len); |
d04232f2 CB |
1176 | ret = snprintf(fnam, len, "%s%s/%s", *cgroup == '/' ? "." : "", cgroup, f); |
1177 | if (ret < 0 || (size_t)ret >= len) | |
1178 | return false; | |
237e200e | 1179 | |
d04232f2 | 1180 | ret = fstatat(cfd, fnam, &sb, 0); |
237e200e SH |
1181 | if (ret < 0 || !S_ISDIR(sb.st_mode)) |
1182 | return false; | |
f5a6d92e | 1183 | |
237e200e SH |
1184 | return true; |
1185 | } | |
1186 | ||
1187 | #define SEND_CREDS_OK 0 | |
1188 | #define SEND_CREDS_NOTSK 1 | |
1189 | #define SEND_CREDS_FAIL 2 | |
1190 | static bool recv_creds(int sock, struct ucred *cred, char *v); | |
1191 | static int wait_for_pid(pid_t pid); | |
1192 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst); | |
b10bdd6c | 1193 | static int send_creds_clone_wrapper(void *arg); |
237e200e SH |
1194 | |
1195 | /* | |
b10bdd6c | 1196 | * clone a task which switches to @task's namespace and writes '1'. |
237e200e SH |
1197 | * over a unix sock so we can read the task's reaper's pid in our |
1198 | * namespace | |
b10bdd6c FG |
1199 | * |
1200 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
1201 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
1202 | * the pidns and the parent pid outside are identical. Using clone prevents | |
1203 | * this issue. | |
237e200e SH |
1204 | */ |
1205 | static void write_task_init_pid_exit(int sock, pid_t target) | |
1206 | { | |
237e200e SH |
1207 | char fnam[100]; |
1208 | pid_t pid; | |
237e200e | 1209 | int fd, ret; |
b10bdd6c FG |
1210 | size_t stack_size = sysconf(_SC_PAGESIZE); |
1211 | void *stack = alloca(stack_size); | |
237e200e SH |
1212 | |
1213 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", (int)target); | |
1214 | if (ret < 0 || ret >= sizeof(fnam)) | |
1215 | _exit(1); | |
1216 | ||
1217 | fd = open(fnam, O_RDONLY); | |
1218 | if (fd < 0) { | |
1219 | perror("write_task_init_pid_exit open of ns/pid"); | |
1220 | _exit(1); | |
1221 | } | |
1222 | if (setns(fd, 0)) { | |
1223 | perror("write_task_init_pid_exit setns 1"); | |
1224 | close(fd); | |
1225 | _exit(1); | |
1226 | } | |
b10bdd6c | 1227 | pid = clone(send_creds_clone_wrapper, stack + stack_size, SIGCHLD, &sock); |
237e200e SH |
1228 | if (pid < 0) |
1229 | _exit(1); | |
1230 | if (pid != 0) { | |
1231 | if (!wait_for_pid(pid)) | |
1232 | _exit(1); | |
1233 | _exit(0); | |
1234 | } | |
b10bdd6c FG |
1235 | } |
1236 | ||
1237 | static int send_creds_clone_wrapper(void *arg) { | |
1238 | struct ucred cred; | |
1239 | char v; | |
1240 | int sock = *(int *)arg; | |
237e200e SH |
1241 | |
1242 | /* we are the child */ | |
1243 | cred.uid = 0; | |
1244 | cred.gid = 0; | |
1245 | cred.pid = 1; | |
1246 | v = '1'; | |
1247 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) | |
b10bdd6c FG |
1248 | return 1; |
1249 | return 0; | |
237e200e SH |
1250 | } |
1251 | ||
1252 | static pid_t get_init_pid_for_task(pid_t task) | |
1253 | { | |
1254 | int sock[2]; | |
1255 | pid_t pid; | |
1256 | pid_t ret = -1; | |
1257 | char v = '0'; | |
1258 | struct ucred cred; | |
1259 | ||
1260 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
1261 | perror("socketpair"); | |
1262 | return -1; | |
1263 | } | |
1264 | ||
1265 | pid = fork(); | |
1266 | if (pid < 0) | |
1267 | goto out; | |
1268 | if (!pid) { | |
1269 | close(sock[1]); | |
1270 | write_task_init_pid_exit(sock[0], task); | |
1271 | _exit(0); | |
1272 | } | |
1273 | ||
1274 | if (!recv_creds(sock[1], &cred, &v)) | |
1275 | goto out; | |
1276 | ret = cred.pid; | |
1277 | ||
1278 | out: | |
1279 | close(sock[0]); | |
1280 | close(sock[1]); | |
1281 | if (pid > 0) | |
1282 | wait_for_pid(pid); | |
1283 | return ret; | |
1284 | } | |
1285 | ||
1286 | static pid_t lookup_initpid_in_store(pid_t qpid) | |
1287 | { | |
1288 | pid_t answer = 0; | |
1289 | struct stat sb; | |
1290 | struct pidns_init_store *e; | |
1291 | char fnam[100]; | |
1292 | ||
1293 | snprintf(fnam, 100, "/proc/%d/ns/pid", qpid); | |
1294 | store_lock(); | |
1295 | if (stat(fnam, &sb) < 0) | |
1296 | goto out; | |
1297 | e = lookup_verify_initpid(&sb); | |
1298 | if (e) { | |
1299 | answer = e->initpid; | |
1300 | goto out; | |
1301 | } | |
1302 | answer = get_init_pid_for_task(qpid); | |
1303 | if (answer > 0) | |
1304 | save_initpid(&sb, answer); | |
1305 | ||
1306 | out: | |
1307 | /* we prune at end in case we are returning | |
1308 | * the value we were about to return */ | |
1309 | prune_initpid_store(); | |
1310 | store_unlock(); | |
1311 | return answer; | |
1312 | } | |
1313 | ||
1314 | static int wait_for_pid(pid_t pid) | |
1315 | { | |
1316 | int status, ret; | |
1317 | ||
1318 | if (pid <= 0) | |
1319 | return -1; | |
1320 | ||
1321 | again: | |
1322 | ret = waitpid(pid, &status, 0); | |
1323 | if (ret == -1) { | |
1324 | if (errno == EINTR) | |
1325 | goto again; | |
1326 | return -1; | |
1327 | } | |
1328 | if (ret != pid) | |
1329 | goto again; | |
1330 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) | |
1331 | return -1; | |
1332 | return 0; | |
1333 | } | |
1334 | ||
1335 | ||
1336 | /* | |
1337 | * append pid to *src. | |
1338 | * src: a pointer to a char* in which ot append the pid. | |
1339 | * sz: the number of characters printed so far, minus trailing \0. | |
1340 | * asz: the allocated size so far | |
1341 | * pid: the pid to append | |
1342 | */ | |
1343 | static void must_strcat_pid(char **src, size_t *sz, size_t *asz, pid_t pid) | |
1344 | { | |
1345 | char tmp[30]; | |
1346 | ||
1347 | int tmplen = sprintf(tmp, "%d\n", (int)pid); | |
1348 | ||
1349 | if (!*src || tmplen + *sz + 1 >= *asz) { | |
1350 | char *tmp; | |
1351 | do { | |
1352 | tmp = realloc(*src, *asz + BUF_RESERVE_SIZE); | |
1353 | } while (!tmp); | |
1354 | *src = tmp; | |
1355 | *asz += BUF_RESERVE_SIZE; | |
1356 | } | |
bbfd0e33 | 1357 | memcpy((*src) +*sz , tmp, tmplen+1); /* include the \0 */ |
237e200e | 1358 | *sz += tmplen; |
237e200e SH |
1359 | } |
1360 | ||
1361 | /* | |
1362 | * Given a open file * to /proc/pid/{u,g}id_map, and an id | |
1363 | * valid in the caller's namespace, return the id mapped into | |
1364 | * pid's namespace. | |
1365 | * Returns the mapped id, or -1 on error. | |
1366 | */ | |
1367 | unsigned int | |
1368 | convert_id_to_ns(FILE *idfile, unsigned int in_id) | |
1369 | { | |
1370 | unsigned int nsuid, // base id for a range in the idfile's namespace | |
1371 | hostuid, // base id for a range in the caller's namespace | |
1372 | count; // number of ids in this range | |
1373 | char line[400]; | |
1374 | int ret; | |
1375 | ||
1376 | fseek(idfile, 0L, SEEK_SET); | |
1377 | while (fgets(line, 400, idfile)) { | |
1378 | ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count); | |
1379 | if (ret != 3) | |
1380 | continue; | |
1381 | if (hostuid + count < hostuid || nsuid + count < nsuid) { | |
1382 | /* | |
1383 | * uids wrapped around - unexpected as this is a procfile, | |
1384 | * so just bail. | |
1385 | */ | |
b8defc3d | 1386 | lxcfs_error("pid wrapparound at entry %u %u %u in %s\n", |
237e200e SH |
1387 | nsuid, hostuid, count, line); |
1388 | return -1; | |
1389 | } | |
1390 | if (hostuid <= in_id && hostuid+count > in_id) { | |
1391 | /* | |
1392 | * now since hostuid <= in_id < hostuid+count, and | |
1393 | * hostuid+count and nsuid+count do not wrap around, | |
1394 | * we know that nsuid+(in_id-hostuid) which must be | |
1395 | * less that nsuid+(count) must not wrap around | |
1396 | */ | |
1397 | return (in_id - hostuid) + nsuid; | |
1398 | } | |
1399 | } | |
1400 | ||
1401 | // no answer found | |
1402 | return -1; | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * for is_privileged_over, | |
1407 | * specify whether we require the calling uid to be root in his | |
1408 | * namespace | |
1409 | */ | |
1410 | #define NS_ROOT_REQD true | |
1411 | #define NS_ROOT_OPT false | |
1412 | ||
1413 | #define PROCLEN 100 | |
1414 | ||
1415 | static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root) | |
1416 | { | |
1417 | char fpath[PROCLEN]; | |
1418 | int ret; | |
1419 | bool answer = false; | |
1420 | uid_t nsuid; | |
1421 | ||
1422 | if (victim == -1 || uid == -1) | |
1423 | return false; | |
1424 | ||
1425 | /* | |
1426 | * If the request is one not requiring root in the namespace, | |
1427 | * then having the same uid suffices. (i.e. uid 1000 has write | |
1428 | * access to files owned by uid 1000 | |
1429 | */ | |
1430 | if (!req_ns_root && uid == victim) | |
1431 | return true; | |
1432 | ||
1433 | ret = snprintf(fpath, PROCLEN, "/proc/%d/uid_map", pid); | |
1434 | if (ret < 0 || ret >= PROCLEN) | |
1435 | return false; | |
1436 | FILE *f = fopen(fpath, "r"); | |
1437 | if (!f) | |
1438 | return false; | |
1439 | ||
1440 | /* if caller's not root in his namespace, reject */ | |
1441 | nsuid = convert_id_to_ns(f, uid); | |
1442 | if (nsuid) | |
1443 | goto out; | |
1444 | ||
1445 | /* | |
1446 | * If victim is not mapped into caller's ns, reject. | |
1447 | * XXX I'm not sure this check is needed given that fuse | |
1448 | * will be sending requests where the vfs has converted | |
1449 | */ | |
1450 | nsuid = convert_id_to_ns(f, victim); | |
1451 | if (nsuid == -1) | |
1452 | goto out; | |
1453 | ||
1454 | answer = true; | |
1455 | ||
1456 | out: | |
1457 | fclose(f); | |
1458 | return answer; | |
1459 | } | |
1460 | ||
1461 | static bool perms_include(int fmode, mode_t req_mode) | |
1462 | { | |
1463 | mode_t r; | |
1464 | ||
1465 | switch (req_mode & O_ACCMODE) { | |
1466 | case O_RDONLY: | |
1467 | r = S_IROTH; | |
1468 | break; | |
1469 | case O_WRONLY: | |
1470 | r = S_IWOTH; | |
1471 | break; | |
1472 | case O_RDWR: | |
1473 | r = S_IROTH | S_IWOTH; | |
1474 | break; | |
1475 | default: | |
1476 | return false; | |
1477 | } | |
1478 | return ((fmode & r) == r); | |
1479 | } | |
1480 | ||
1481 | ||
1482 | /* | |
1483 | * taskcg is a/b/c | |
1484 | * querycg is /a/b/c/d/e | |
1485 | * we return 'd' | |
1486 | */ | |
1487 | static char *get_next_cgroup_dir(const char *taskcg, const char *querycg) | |
1488 | { | |
1489 | char *start, *end; | |
1490 | ||
1491 | if (strlen(taskcg) <= strlen(querycg)) { | |
b8defc3d | 1492 | lxcfs_error("%s\n", "I was fed bad input."); |
237e200e SH |
1493 | return NULL; |
1494 | } | |
1495 | ||
06081b29 | 1496 | if ((strcmp(querycg, "/") == 0) || (strcmp(querycg, "./") == 0)) |
237e200e SH |
1497 | start = strdup(taskcg + 1); |
1498 | else | |
1499 | start = strdup(taskcg + strlen(querycg) + 1); | |
1500 | if (!start) | |
1501 | return NULL; | |
1502 | end = strchr(start, '/'); | |
1503 | if (end) | |
1504 | *end = '\0'; | |
1505 | return start; | |
1506 | } | |
1507 | ||
1508 | static void stripnewline(char *x) | |
1509 | { | |
1510 | size_t l = strlen(x); | |
1511 | if (l && x[l-1] == '\n') | |
1512 | x[l-1] = '\0'; | |
1513 | } | |
1514 | ||
1515 | static char *get_pid_cgroup(pid_t pid, const char *contrl) | |
1516 | { | |
5dd3e6fd | 1517 | int cfd; |
237e200e SH |
1518 | char fnam[PROCLEN]; |
1519 | FILE *f; | |
1520 | char *answer = NULL; | |
1521 | char *line = NULL; | |
1522 | size_t len = 0; | |
1523 | int ret; | |
5dd3e6fd | 1524 | const char *h = find_mounted_controller(contrl, &cfd); |
237e200e SH |
1525 | if (!h) |
1526 | return NULL; | |
1527 | ||
1528 | ret = snprintf(fnam, PROCLEN, "/proc/%d/cgroup", pid); | |
1529 | if (ret < 0 || ret >= PROCLEN) | |
1530 | return NULL; | |
1531 | if (!(f = fopen(fnam, "r"))) | |
1532 | return NULL; | |
1533 | ||
1534 | while (getline(&line, &len, f) != -1) { | |
1535 | char *c1, *c2; | |
1536 | if (!line[0]) | |
1537 | continue; | |
1538 | c1 = strchr(line, ':'); | |
1539 | if (!c1) | |
1540 | goto out; | |
1541 | c1++; | |
1542 | c2 = strchr(c1, ':'); | |
1543 | if (!c2) | |
1544 | goto out; | |
1545 | *c2 = '\0'; | |
1546 | if (strcmp(c1, h) != 0) | |
1547 | continue; | |
1548 | c2++; | |
1549 | stripnewline(c2); | |
1550 | do { | |
1551 | answer = strdup(c2); | |
1552 | } while (!answer); | |
1553 | break; | |
1554 | } | |
1555 | ||
1556 | out: | |
1557 | fclose(f); | |
1558 | free(line); | |
1559 | return answer; | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * check whether a fuse context may access a cgroup dir or file | |
1564 | * | |
1565 | * If file is not null, it is a cgroup file to check under cg. | |
1566 | * If file is null, then we are checking perms on cg itself. | |
1567 | * | |
1568 | * For files we can check the mode of the list_keys result. | |
1569 | * For cgroups, we must make assumptions based on the files under the | |
1570 | * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups | |
1571 | * yet. | |
1572 | */ | |
1573 | static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode) | |
1574 | { | |
1575 | struct cgfs_files *k = NULL; | |
1576 | bool ret = false; | |
1577 | ||
1578 | k = cgfs_get_key(contrl, cg, file); | |
1579 | if (!k) | |
1580 | return false; | |
1581 | ||
1582 | if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
1583 | if (perms_include(k->mode >> 6, mode)) { | |
1584 | ret = true; | |
1585 | goto out; | |
1586 | } | |
1587 | } | |
1588 | if (fc->gid == k->gid) { | |
1589 | if (perms_include(k->mode >> 3, mode)) { | |
1590 | ret = true; | |
1591 | goto out; | |
1592 | } | |
1593 | } | |
1594 | ret = perms_include(k->mode, mode); | |
1595 | ||
1596 | out: | |
1597 | free_key(k); | |
1598 | return ret; | |
1599 | } | |
1600 | ||
1601 | #define INITSCOPE "/init.scope" | |
1602 | static void prune_init_slice(char *cg) | |
1603 | { | |
1604 | char *point; | |
1605 | size_t cg_len = strlen(cg), initscope_len = strlen(INITSCOPE); | |
1606 | ||
1607 | if (cg_len < initscope_len) | |
1608 | return; | |
1609 | ||
1610 | point = cg + cg_len - initscope_len; | |
1611 | if (strcmp(point, INITSCOPE) == 0) { | |
1612 | if (point == cg) | |
1613 | *(point+1) = '\0'; | |
1614 | else | |
1615 | *point = '\0'; | |
1616 | } | |
1617 | } | |
1618 | ||
1619 | /* | |
1620 | * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d. | |
1621 | * If pid is in /a, he may act on /a/b, but not on /b. | |
1622 | * if the answer is false and nextcg is not NULL, then *nextcg will point | |
1623 | * to a string containing the next cgroup directory under cg, which must be | |
1624 | * freed by the caller. | |
1625 | */ | |
1626 | static bool caller_is_in_ancestor(pid_t pid, const char *contrl, const char *cg, char **nextcg) | |
1627 | { | |
1628 | bool answer = false; | |
1629 | char *c2 = get_pid_cgroup(pid, contrl); | |
1630 | char *linecmp; | |
1631 | ||
1632 | if (!c2) | |
1633 | return false; | |
1634 | prune_init_slice(c2); | |
1635 | ||
1636 | /* | |
12c31268 CB |
1637 | * callers pass in '/' or './' (openat()) for root cgroup, otherwise |
1638 | * they pass in a cgroup without leading '/' | |
1639 | * | |
1640 | * The original line here was: | |
1641 | * linecmp = *cg == '/' ? c2 : c2+1; | |
1642 | * TODO: I'm not sure why you'd want to increment when *cg != '/'? | |
1643 | * Serge, do you know? | |
237e200e | 1644 | */ |
12c31268 CB |
1645 | if (*cg == '/' || !strncmp(cg, "./", 2)) |
1646 | linecmp = c2; | |
1647 | else | |
1648 | linecmp = c2 + 1; | |
237e200e SH |
1649 | if (strncmp(linecmp, cg, strlen(linecmp)) != 0) { |
1650 | if (nextcg) { | |
1651 | *nextcg = get_next_cgroup_dir(linecmp, cg); | |
1652 | } | |
1653 | goto out; | |
1654 | } | |
1655 | answer = true; | |
1656 | ||
1657 | out: | |
1658 | free(c2); | |
1659 | return answer; | |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c. | |
1664 | */ | |
1665 | static bool caller_may_see_dir(pid_t pid, const char *contrl, const char *cg) | |
1666 | { | |
1667 | bool answer = false; | |
1668 | char *c2, *task_cg; | |
1669 | size_t target_len, task_len; | |
1670 | ||
f7bff426 | 1671 | if (strcmp(cg, "/") == 0 || strcmp(cg, "./") == 0) |
237e200e SH |
1672 | return true; |
1673 | ||
1674 | c2 = get_pid_cgroup(pid, contrl); | |
1675 | if (!c2) | |
1676 | return false; | |
1677 | prune_init_slice(c2); | |
1678 | ||
1679 | task_cg = c2 + 1; | |
1680 | target_len = strlen(cg); | |
1681 | task_len = strlen(task_cg); | |
1682 | if (task_len == 0) { | |
1683 | /* Task is in the root cg, it can see everything. This case is | |
1684 | * not handled by the strmcps below, since they test for the | |
1685 | * last /, but that is the first / that we've chopped off | |
1686 | * above. | |
1687 | */ | |
1688 | answer = true; | |
1689 | goto out; | |
1690 | } | |
1691 | if (strcmp(cg, task_cg) == 0) { | |
1692 | answer = true; | |
1693 | goto out; | |
1694 | } | |
1695 | if (target_len < task_len) { | |
1696 | /* looking up a parent dir */ | |
1697 | if (strncmp(task_cg, cg, target_len) == 0 && task_cg[target_len] == '/') | |
1698 | answer = true; | |
1699 | goto out; | |
1700 | } | |
1701 | if (target_len > task_len) { | |
1702 | /* looking up a child dir */ | |
1703 | if (strncmp(task_cg, cg, task_len) == 0 && cg[task_len] == '/') | |
1704 | answer = true; | |
1705 | goto out; | |
1706 | } | |
1707 | ||
1708 | out: | |
1709 | free(c2); | |
1710 | return answer; | |
1711 | } | |
1712 | ||
1713 | /* | |
1714 | * given /cgroup/freezer/a/b, return "freezer". | |
1715 | * the returned char* should NOT be freed. | |
1716 | */ | |
1717 | static char *pick_controller_from_path(struct fuse_context *fc, const char *path) | |
1718 | { | |
1719 | const char *p1; | |
1720 | char *contr, *slash; | |
1721 | ||
99142521 | 1722 | if (strlen(path) < 9) { |
e254948f | 1723 | errno = EACCES; |
237e200e | 1724 | return NULL; |
99142521 CB |
1725 | } |
1726 | if (*(path + 7) != '/') { | |
1727 | errno = EINVAL; | |
237e200e | 1728 | return NULL; |
99142521 | 1729 | } |
3adc421c | 1730 | p1 = path + 8; |
237e200e | 1731 | contr = strdupa(p1); |
99142521 CB |
1732 | if (!contr) { |
1733 | errno = ENOMEM; | |
237e200e | 1734 | return NULL; |
99142521 | 1735 | } |
237e200e SH |
1736 | slash = strstr(contr, "/"); |
1737 | if (slash) | |
1738 | *slash = '\0'; | |
1739 | ||
1740 | int i; | |
3adc421c | 1741 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
1742 | if (hierarchies[i] && strcmp(hierarchies[i], contr) == 0) |
1743 | return hierarchies[i]; | |
1744 | } | |
99142521 | 1745 | errno = ENOENT; |
237e200e SH |
1746 | return NULL; |
1747 | } | |
1748 | ||
1749 | /* | |
1750 | * Find the start of cgroup in /cgroup/controller/the/cgroup/path | |
1751 | * Note that the returned value may include files (keynames) etc | |
1752 | */ | |
1753 | static const char *find_cgroup_in_path(const char *path) | |
1754 | { | |
1755 | const char *p1; | |
1756 | ||
bc70ba9b | 1757 | if (strlen(path) < 9) { |
e254948f | 1758 | errno = EACCES; |
237e200e | 1759 | return NULL; |
bc70ba9b CB |
1760 | } |
1761 | p1 = strstr(path + 8, "/"); | |
1762 | if (!p1) { | |
1763 | errno = EINVAL; | |
237e200e | 1764 | return NULL; |
bc70ba9b CB |
1765 | } |
1766 | errno = 0; | |
1767 | return p1 + 1; | |
237e200e SH |
1768 | } |
1769 | ||
1770 | /* | |
1771 | * split the last path element from the path in @cg. | |
1772 | * @dir is newly allocated and should be freed, @last not | |
1773 | */ | |
1774 | static void get_cgdir_and_path(const char *cg, char **dir, char **last) | |
1775 | { | |
1776 | char *p; | |
1777 | ||
1778 | do { | |
1779 | *dir = strdup(cg); | |
1780 | } while (!*dir); | |
1781 | *last = strrchr(cg, '/'); | |
1782 | if (!*last) { | |
1783 | *last = NULL; | |
1784 | return; | |
1785 | } | |
1786 | p = strrchr(*dir, '/'); | |
1787 | *p = '\0'; | |
1788 | } | |
1789 | ||
1790 | /* | |
1791 | * FUSE ops for /cgroup | |
1792 | */ | |
1793 | ||
1794 | int cg_getattr(const char *path, struct stat *sb) | |
1795 | { | |
1796 | struct timespec now; | |
1797 | struct fuse_context *fc = fuse_get_context(); | |
1798 | char * cgdir = NULL; | |
1799 | char *last = NULL, *path1, *path2; | |
1800 | struct cgfs_files *k = NULL; | |
1801 | const char *cgroup; | |
1802 | const char *controller = NULL; | |
1803 | int ret = -ENOENT; | |
1804 | ||
1805 | ||
1806 | if (!fc) | |
1807 | return -EIO; | |
1808 | ||
1809 | memset(sb, 0, sizeof(struct stat)); | |
1810 | ||
1811 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
1812 | return -EINVAL; | |
1813 | ||
1814 | sb->st_uid = sb->st_gid = 0; | |
1815 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
1816 | sb->st_size = 0; | |
1817 | ||
1818 | if (strcmp(path, "/cgroup") == 0) { | |
1819 | sb->st_mode = S_IFDIR | 00755; | |
1820 | sb->st_nlink = 2; | |
1821 | return 0; | |
1822 | } | |
1823 | ||
1824 | controller = pick_controller_from_path(fc, path); | |
1825 | if (!controller) | |
2f7036d0 | 1826 | return -errno; |
237e200e SH |
1827 | cgroup = find_cgroup_in_path(path); |
1828 | if (!cgroup) { | |
1829 | /* this is just /cgroup/controller, return it as a dir */ | |
1830 | sb->st_mode = S_IFDIR | 00755; | |
1831 | sb->st_nlink = 2; | |
1832 | return 0; | |
1833 | } | |
1834 | ||
1835 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
1836 | ||
1837 | if (!last) { | |
1838 | path1 = "/"; | |
1839 | path2 = cgdir; | |
1840 | } else { | |
1841 | path1 = cgdir; | |
1842 | path2 = last; | |
1843 | } | |
1844 | ||
1845 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1846 | if (initpid <= 0) | |
1847 | initpid = fc->pid; | |
1848 | /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys. | |
1849 | * Then check that caller's cgroup is under path if last is a child | |
1850 | * cgroup, or cgdir if last is a file */ | |
1851 | ||
1852 | if (is_child_cgroup(controller, path1, path2)) { | |
1853 | if (!caller_may_see_dir(initpid, controller, cgroup)) { | |
1854 | ret = -ENOENT; | |
1855 | goto out; | |
1856 | } | |
1857 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
1858 | /* this is just /cgroup/controller, return it as a dir */ | |
1859 | sb->st_mode = S_IFDIR | 00555; | |
1860 | sb->st_nlink = 2; | |
1861 | ret = 0; | |
1862 | goto out; | |
1863 | } | |
1864 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) { | |
1865 | ret = -EACCES; | |
1866 | goto out; | |
1867 | } | |
1868 | ||
1869 | // get uid, gid, from '/tasks' file and make up a mode | |
1870 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
1871 | sb->st_mode = S_IFDIR | 00755; | |
1872 | k = cgfs_get_key(controller, cgroup, NULL); | |
1873 | if (!k) { | |
1874 | sb->st_uid = sb->st_gid = 0; | |
1875 | } else { | |
1876 | sb->st_uid = k->uid; | |
1877 | sb->st_gid = k->gid; | |
1878 | } | |
1879 | free_key(k); | |
1880 | sb->st_nlink = 2; | |
1881 | ret = 0; | |
1882 | goto out; | |
1883 | } | |
1884 | ||
1885 | if ((k = cgfs_get_key(controller, path1, path2)) != NULL) { | |
1886 | sb->st_mode = S_IFREG | k->mode; | |
1887 | sb->st_nlink = 1; | |
1888 | sb->st_uid = k->uid; | |
1889 | sb->st_gid = k->gid; | |
1890 | sb->st_size = 0; | |
1891 | free_key(k); | |
1892 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
1893 | ret = -ENOENT; | |
1894 | goto out; | |
1895 | } | |
237e200e SH |
1896 | ret = 0; |
1897 | } | |
1898 | ||
1899 | out: | |
1900 | free(cgdir); | |
1901 | return ret; | |
1902 | } | |
1903 | ||
1904 | int cg_opendir(const char *path, struct fuse_file_info *fi) | |
1905 | { | |
1906 | struct fuse_context *fc = fuse_get_context(); | |
1907 | const char *cgroup; | |
1908 | struct file_info *dir_info; | |
1909 | char *controller = NULL; | |
1910 | ||
1911 | if (!fc) | |
1912 | return -EIO; | |
1913 | ||
1914 | if (strcmp(path, "/cgroup") == 0) { | |
1915 | cgroup = NULL; | |
1916 | controller = NULL; | |
1917 | } else { | |
1918 | // return list of keys for the controller, and list of child cgroups | |
1919 | controller = pick_controller_from_path(fc, path); | |
1920 | if (!controller) | |
2f7036d0 | 1921 | return -errno; |
237e200e SH |
1922 | |
1923 | cgroup = find_cgroup_in_path(path); | |
1924 | if (!cgroup) { | |
1925 | /* this is just /cgroup/controller, return its contents */ | |
1926 | cgroup = "/"; | |
1927 | } | |
1928 | } | |
1929 | ||
1930 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1931 | if (initpid <= 0) | |
1932 | initpid = fc->pid; | |
1933 | if (cgroup) { | |
1934 | if (!caller_may_see_dir(initpid, controller, cgroup)) | |
1935 | return -ENOENT; | |
1936 | if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY)) | |
1937 | return -EACCES; | |
1938 | } | |
1939 | ||
1940 | /* we'll free this at cg_releasedir */ | |
1941 | dir_info = malloc(sizeof(*dir_info)); | |
1942 | if (!dir_info) | |
1943 | return -ENOMEM; | |
1944 | dir_info->controller = must_copy_string(controller); | |
1945 | dir_info->cgroup = must_copy_string(cgroup); | |
1946 | dir_info->type = LXC_TYPE_CGDIR; | |
1947 | dir_info->buf = NULL; | |
1948 | dir_info->file = NULL; | |
1949 | dir_info->buflen = 0; | |
1950 | ||
1951 | fi->fh = (unsigned long)dir_info; | |
1952 | return 0; | |
1953 | } | |
1954 | ||
1955 | int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
1956 | struct fuse_file_info *fi) | |
1957 | { | |
1958 | struct file_info *d = (struct file_info *)fi->fh; | |
1959 | struct cgfs_files **list = NULL; | |
1960 | int i, ret; | |
1961 | char *nextcg = NULL; | |
1962 | struct fuse_context *fc = fuse_get_context(); | |
1963 | char **clist = NULL; | |
1964 | ||
d639f863 CB |
1965 | if (filler(buf, ".", NULL, 0) != 0 || filler(buf, "..", NULL, 0) != 0) |
1966 | return -EIO; | |
1967 | ||
237e200e | 1968 | if (d->type != LXC_TYPE_CGDIR) { |
b8defc3d | 1969 | lxcfs_error("%s\n", "Internal error: file cache info used in readdir."); |
237e200e SH |
1970 | return -EIO; |
1971 | } | |
1972 | if (!d->cgroup && !d->controller) { | |
1973 | // ls /var/lib/lxcfs/cgroup - just show list of controllers | |
1974 | int i; | |
1975 | ||
1976 | for (i = 0; i < num_hierarchies; i++) { | |
1977 | if (hierarchies[i] && filler(buf, hierarchies[i], NULL, 0) != 0) { | |
1978 | return -EIO; | |
1979 | } | |
1980 | } | |
1981 | return 0; | |
1982 | } | |
1983 | ||
1984 | if (!cgfs_list_keys(d->controller, d->cgroup, &list)) { | |
1985 | // not a valid cgroup | |
1986 | ret = -EINVAL; | |
1987 | goto out; | |
1988 | } | |
1989 | ||
1990 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
1991 | if (initpid <= 0) | |
1992 | initpid = fc->pid; | |
1993 | if (!caller_is_in_ancestor(initpid, d->controller, d->cgroup, &nextcg)) { | |
1994 | if (nextcg) { | |
1995 | ret = filler(buf, nextcg, NULL, 0); | |
1996 | free(nextcg); | |
1997 | if (ret != 0) { | |
1998 | ret = -EIO; | |
1999 | goto out; | |
2000 | } | |
2001 | } | |
2002 | ret = 0; | |
2003 | goto out; | |
2004 | } | |
2005 | ||
2006 | for (i = 0; list[i]; i++) { | |
2007 | if (filler(buf, list[i]->name, NULL, 0) != 0) { | |
2008 | ret = -EIO; | |
2009 | goto out; | |
2010 | } | |
2011 | } | |
2012 | ||
2013 | // now get the list of child cgroups | |
2014 | ||
2015 | if (!cgfs_list_children(d->controller, d->cgroup, &clist)) { | |
2016 | ret = 0; | |
2017 | goto out; | |
2018 | } | |
f366da65 WB |
2019 | if (clist) { |
2020 | for (i = 0; clist[i]; i++) { | |
2021 | if (filler(buf, clist[i], NULL, 0) != 0) { | |
2022 | ret = -EIO; | |
2023 | goto out; | |
2024 | } | |
237e200e SH |
2025 | } |
2026 | } | |
2027 | ret = 0; | |
2028 | ||
2029 | out: | |
2030 | free_keys(list); | |
2031 | if (clist) { | |
2032 | for (i = 0; clist[i]; i++) | |
2033 | free(clist[i]); | |
2034 | free(clist); | |
2035 | } | |
2036 | return ret; | |
2037 | } | |
2038 | ||
43215927 | 2039 | static void do_release_file_info(struct fuse_file_info *fi) |
237e200e | 2040 | { |
43215927 SH |
2041 | struct file_info *f = (struct file_info *)fi->fh; |
2042 | ||
237e200e SH |
2043 | if (!f) |
2044 | return; | |
43215927 SH |
2045 | |
2046 | fi->fh = 0; | |
2047 | ||
237e200e | 2048 | free(f->controller); |
43215927 | 2049 | f->controller = NULL; |
237e200e | 2050 | free(f->cgroup); |
43215927 | 2051 | f->cgroup = NULL; |
237e200e | 2052 | free(f->file); |
43215927 | 2053 | f->file = NULL; |
237e200e | 2054 | free(f->buf); |
43215927 | 2055 | f->buf = NULL; |
237e200e SH |
2056 | free(f); |
2057 | } | |
2058 | ||
2059 | int cg_releasedir(const char *path, struct fuse_file_info *fi) | |
2060 | { | |
43215927 | 2061 | do_release_file_info(fi); |
237e200e SH |
2062 | return 0; |
2063 | } | |
2064 | ||
2065 | int cg_open(const char *path, struct fuse_file_info *fi) | |
2066 | { | |
2067 | const char *cgroup; | |
2068 | char *last = NULL, *path1, *path2, * cgdir = NULL, *controller; | |
2069 | struct cgfs_files *k = NULL; | |
2070 | struct file_info *file_info; | |
2071 | struct fuse_context *fc = fuse_get_context(); | |
2072 | int ret; | |
2073 | ||
2074 | if (!fc) | |
2075 | return -EIO; | |
2076 | ||
2077 | controller = pick_controller_from_path(fc, path); | |
2078 | if (!controller) | |
2f7036d0 | 2079 | return -errno; |
237e200e SH |
2080 | cgroup = find_cgroup_in_path(path); |
2081 | if (!cgroup) | |
bc70ba9b | 2082 | return -errno; |
237e200e SH |
2083 | |
2084 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2085 | if (!last) { | |
2086 | path1 = "/"; | |
2087 | path2 = cgdir; | |
2088 | } else { | |
2089 | path1 = cgdir; | |
2090 | path2 = last; | |
2091 | } | |
2092 | ||
2093 | k = cgfs_get_key(controller, path1, path2); | |
2094 | if (!k) { | |
2095 | ret = -EINVAL; | |
2096 | goto out; | |
2097 | } | |
2098 | free_key(k); | |
2099 | ||
2100 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2101 | if (initpid <= 0) | |
2102 | initpid = fc->pid; | |
2103 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2104 | ret = -ENOENT; | |
2105 | goto out; | |
2106 | } | |
2107 | if (!fc_may_access(fc, controller, path1, path2, fi->flags)) { | |
237e200e SH |
2108 | ret = -EACCES; |
2109 | goto out; | |
2110 | } | |
2111 | ||
2112 | /* we'll free this at cg_release */ | |
2113 | file_info = malloc(sizeof(*file_info)); | |
2114 | if (!file_info) { | |
2115 | ret = -ENOMEM; | |
2116 | goto out; | |
2117 | } | |
2118 | file_info->controller = must_copy_string(controller); | |
2119 | file_info->cgroup = must_copy_string(path1); | |
2120 | file_info->file = must_copy_string(path2); | |
2121 | file_info->type = LXC_TYPE_CGFILE; | |
2122 | file_info->buf = NULL; | |
2123 | file_info->buflen = 0; | |
2124 | ||
2125 | fi->fh = (unsigned long)file_info; | |
2126 | ret = 0; | |
2127 | ||
2128 | out: | |
2129 | free(cgdir); | |
2130 | return ret; | |
2131 | } | |
2132 | ||
bddbb106 SH |
2133 | int cg_access(const char *path, int mode) |
2134 | { | |
6f0f6b83 | 2135 | int ret; |
bddbb106 | 2136 | const char *cgroup; |
6f0f6b83 CB |
2137 | char *path1, *path2, *controller; |
2138 | char *last = NULL, *cgdir = NULL; | |
bddbb106 SH |
2139 | struct cgfs_files *k = NULL; |
2140 | struct fuse_context *fc = fuse_get_context(); | |
6f0f6b83 | 2141 | |
9873c5e8 | 2142 | if (strcmp(path, "/cgroup") == 0) |
6f0f6b83 | 2143 | return 0; |
bddbb106 SH |
2144 | |
2145 | if (!fc) | |
2146 | return -EIO; | |
2147 | ||
2148 | controller = pick_controller_from_path(fc, path); | |
2149 | if (!controller) | |
2f7036d0 | 2150 | return -errno; |
bddbb106 | 2151 | cgroup = find_cgroup_in_path(path); |
575316c4 SH |
2152 | if (!cgroup) { |
2153 | // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not | |
3f441bc7 SH |
2154 | if ((mode & W_OK) == 0) |
2155 | return 0; | |
2156 | return -EACCES; | |
575316c4 | 2157 | } |
bddbb106 SH |
2158 | |
2159 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2160 | if (!last) { | |
2161 | path1 = "/"; | |
2162 | path2 = cgdir; | |
2163 | } else { | |
2164 | path1 = cgdir; | |
2165 | path2 = last; | |
2166 | } | |
2167 | ||
2168 | k = cgfs_get_key(controller, path1, path2); | |
2169 | if (!k) { | |
3f441bc7 SH |
2170 | if ((mode & W_OK) == 0) |
2171 | ret = 0; | |
2172 | else | |
2173 | ret = -EACCES; | |
bddbb106 SH |
2174 | goto out; |
2175 | } | |
2176 | free_key(k); | |
2177 | ||
2178 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
2179 | if (initpid <= 0) | |
2180 | initpid = fc->pid; | |
2181 | if (!caller_may_see_dir(initpid, controller, path1)) { | |
2182 | ret = -ENOENT; | |
2183 | goto out; | |
2184 | } | |
2185 | if (!fc_may_access(fc, controller, path1, path2, mode)) { | |
2186 | ret = -EACCES; | |
2187 | goto out; | |
2188 | } | |
2189 | ||
2190 | ret = 0; | |
2191 | ||
2192 | out: | |
2193 | free(cgdir); | |
2194 | return ret; | |
2195 | } | |
2196 | ||
237e200e SH |
2197 | int cg_release(const char *path, struct fuse_file_info *fi) |
2198 | { | |
43215927 | 2199 | do_release_file_info(fi); |
237e200e SH |
2200 | return 0; |
2201 | } | |
2202 | ||
2203 | #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP ) | |
2204 | ||
2205 | static bool wait_for_sock(int sock, int timeout) | |
2206 | { | |
2207 | struct epoll_event ev; | |
2208 | int epfd, ret, now, starttime, deltatime, saved_errno; | |
2209 | ||
2210 | if ((starttime = time(NULL)) < 0) | |
2211 | return false; | |
2212 | ||
2213 | if ((epfd = epoll_create(1)) < 0) { | |
b8defc3d | 2214 | lxcfs_error("%s\n", "Failed to create epoll socket: %m."); |
237e200e SH |
2215 | return false; |
2216 | } | |
2217 | ||
2218 | ev.events = POLLIN_SET; | |
2219 | ev.data.fd = sock; | |
2220 | if (epoll_ctl(epfd, EPOLL_CTL_ADD, sock, &ev) < 0) { | |
b8defc3d | 2221 | lxcfs_error("%s\n", "Failed adding socket to epoll: %m."); |
237e200e SH |
2222 | close(epfd); |
2223 | return false; | |
2224 | } | |
2225 | ||
2226 | again: | |
2227 | if ((now = time(NULL)) < 0) { | |
2228 | close(epfd); | |
2229 | return false; | |
2230 | } | |
2231 | ||
2232 | deltatime = (starttime + timeout) - now; | |
2233 | if (deltatime < 0) { // timeout | |
2234 | errno = 0; | |
2235 | close(epfd); | |
2236 | return false; | |
2237 | } | |
2238 | ret = epoll_wait(epfd, &ev, 1, 1000*deltatime + 1); | |
2239 | if (ret < 0 && errno == EINTR) | |
2240 | goto again; | |
2241 | saved_errno = errno; | |
2242 | close(epfd); | |
2243 | ||
2244 | if (ret <= 0) { | |
2245 | errno = saved_errno; | |
2246 | return false; | |
2247 | } | |
2248 | return true; | |
2249 | } | |
2250 | ||
2251 | static int msgrecv(int sockfd, void *buf, size_t len) | |
2252 | { | |
2253 | if (!wait_for_sock(sockfd, 2)) | |
2254 | return -1; | |
2255 | return recv(sockfd, buf, len, MSG_DONTWAIT); | |
2256 | } | |
2257 | ||
2258 | static int send_creds(int sock, struct ucred *cred, char v, bool pingfirst) | |
2259 | { | |
2260 | struct msghdr msg = { 0 }; | |
2261 | struct iovec iov; | |
2262 | struct cmsghdr *cmsg; | |
2263 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2264 | char buf[1]; | |
2265 | buf[0] = 'p'; | |
2266 | ||
2267 | if (pingfirst) { | |
2268 | if (msgrecv(sock, buf, 1) != 1) { | |
b8defc3d | 2269 | lxcfs_error("%s\n", "Error getting reply from server over socketpair."); |
237e200e SH |
2270 | return SEND_CREDS_FAIL; |
2271 | } | |
2272 | } | |
2273 | ||
2274 | msg.msg_control = cmsgbuf; | |
2275 | msg.msg_controllen = sizeof(cmsgbuf); | |
2276 | ||
2277 | cmsg = CMSG_FIRSTHDR(&msg); | |
2278 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred)); | |
2279 | cmsg->cmsg_level = SOL_SOCKET; | |
2280 | cmsg->cmsg_type = SCM_CREDENTIALS; | |
2281 | memcpy(CMSG_DATA(cmsg), cred, sizeof(*cred)); | |
2282 | ||
2283 | msg.msg_name = NULL; | |
2284 | msg.msg_namelen = 0; | |
2285 | ||
2286 | buf[0] = v; | |
2287 | iov.iov_base = buf; | |
2288 | iov.iov_len = sizeof(buf); | |
2289 | msg.msg_iov = &iov; | |
2290 | msg.msg_iovlen = 1; | |
2291 | ||
2292 | if (sendmsg(sock, &msg, 0) < 0) { | |
b8defc3d | 2293 | lxcfs_error("Failed at sendmsg: %s.\n",strerror(errno)); |
237e200e SH |
2294 | if (errno == 3) |
2295 | return SEND_CREDS_NOTSK; | |
2296 | return SEND_CREDS_FAIL; | |
2297 | } | |
2298 | ||
2299 | return SEND_CREDS_OK; | |
2300 | } | |
2301 | ||
2302 | static bool recv_creds(int sock, struct ucred *cred, char *v) | |
2303 | { | |
2304 | struct msghdr msg = { 0 }; | |
2305 | struct iovec iov; | |
2306 | struct cmsghdr *cmsg; | |
2307 | char cmsgbuf[CMSG_SPACE(sizeof(*cred))]; | |
2308 | char buf[1]; | |
2309 | int ret; | |
2310 | int optval = 1; | |
2311 | ||
2312 | *v = '1'; | |
2313 | ||
2314 | cred->pid = -1; | |
2315 | cred->uid = -1; | |
2316 | cred->gid = -1; | |
2317 | ||
2318 | if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED, &optval, sizeof(optval)) == -1) { | |
b8defc3d | 2319 | lxcfs_error("Failed to set passcred: %s\n", strerror(errno)); |
237e200e SH |
2320 | return false; |
2321 | } | |
2322 | buf[0] = '1'; | |
2323 | if (write(sock, buf, 1) != 1) { | |
b8defc3d | 2324 | lxcfs_error("Failed to start write on scm fd: %s\n", strerror(errno)); |
237e200e SH |
2325 | return false; |
2326 | } | |
2327 | ||
2328 | msg.msg_name = NULL; | |
2329 | msg.msg_namelen = 0; | |
2330 | msg.msg_control = cmsgbuf; | |
2331 | msg.msg_controllen = sizeof(cmsgbuf); | |
2332 | ||
2333 | iov.iov_base = buf; | |
2334 | iov.iov_len = sizeof(buf); | |
2335 | msg.msg_iov = &iov; | |
2336 | msg.msg_iovlen = 1; | |
2337 | ||
2338 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2339 | lxcfs_error("Timed out waiting for scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2340 | return false; |
2341 | } | |
2342 | ret = recvmsg(sock, &msg, MSG_DONTWAIT); | |
2343 | if (ret < 0) { | |
b8defc3d | 2344 | lxcfs_error("Failed to receive scm_cred: %s\n", strerror(errno)); |
237e200e SH |
2345 | return false; |
2346 | } | |
2347 | ||
2348 | cmsg = CMSG_FIRSTHDR(&msg); | |
2349 | ||
2350 | if (cmsg && cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)) && | |
2351 | cmsg->cmsg_level == SOL_SOCKET && | |
2352 | cmsg->cmsg_type == SCM_CREDENTIALS) { | |
2353 | memcpy(cred, CMSG_DATA(cmsg), sizeof(*cred)); | |
2354 | } | |
2355 | *v = buf[0]; | |
2356 | ||
2357 | return true; | |
2358 | } | |
2359 | ||
35174b0f FG |
2360 | struct pid_ns_clone_args { |
2361 | int *cpipe; | |
2362 | int sock; | |
2363 | pid_t tpid; | |
2364 | int (*wrapped) (int, pid_t); // pid_from_ns or pid_to_ns | |
2365 | }; | |
2366 | ||
2367 | /* | |
2368 | * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage | |
2369 | * with clone(). This simply writes '1' as ACK back to the parent | |
2370 | * before calling the actual wrapped function. | |
2371 | */ | |
2372 | static int pid_ns_clone_wrapper(void *arg) { | |
2373 | struct pid_ns_clone_args* args = (struct pid_ns_clone_args *) arg; | |
2374 | char b = '1'; | |
2375 | ||
2376 | close(args->cpipe[0]); | |
b8defc3d CB |
2377 | if (write(args->cpipe[1], &b, sizeof(char)) < 0) |
2378 | lxcfs_error("(child): error on write: %s.\n", strerror(errno)); | |
35174b0f FG |
2379 | close(args->cpipe[1]); |
2380 | return args->wrapped(args->sock, args->tpid); | |
2381 | } | |
237e200e SH |
2382 | |
2383 | /* | |
2384 | * pid_to_ns - reads pids from a ucred over a socket, then writes the | |
2385 | * int value back over the socket. This shifts the pid from the | |
2386 | * sender's pidns into tpid's pidns. | |
2387 | */ | |
35174b0f | 2388 | static int pid_to_ns(int sock, pid_t tpid) |
237e200e SH |
2389 | { |
2390 | char v = '0'; | |
2391 | struct ucred cred; | |
2392 | ||
2393 | while (recv_creds(sock, &cred, &v)) { | |
2394 | if (v == '1') | |
35174b0f | 2395 | return 0; |
237e200e | 2396 | if (write(sock, &cred.pid, sizeof(pid_t)) != sizeof(pid_t)) |
35174b0f | 2397 | return 1; |
237e200e | 2398 | } |
35174b0f | 2399 | return 0; |
237e200e SH |
2400 | } |
2401 | ||
35174b0f | 2402 | |
237e200e SH |
2403 | /* |
2404 | * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain | |
35174b0f FG |
2405 | * in your old pidns. Only children which you clone will be in the target |
2406 | * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to | |
2407 | * actually convert pids. | |
2408 | * | |
2409 | * Note: glibc's fork() does not respect pidns, which can lead to failed | |
2410 | * assertions inside glibc (and thus failed forks) if the child's pid in | |
2411 | * the pidns and the parent pid outside are identical. Using clone prevents | |
2412 | * this issue. | |
237e200e SH |
2413 | */ |
2414 | static void pid_to_ns_wrapper(int sock, pid_t tpid) | |
2415 | { | |
2416 | int newnsfd = -1, ret, cpipe[2]; | |
2417 | char fnam[100]; | |
2418 | pid_t cpid; | |
2419 | char v; | |
2420 | ||
2421 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2422 | if (ret < 0 || ret >= sizeof(fnam)) | |
2423 | _exit(1); | |
2424 | newnsfd = open(fnam, O_RDONLY); | |
2425 | if (newnsfd < 0) | |
2426 | _exit(1); | |
2427 | if (setns(newnsfd, 0) < 0) | |
2428 | _exit(1); | |
2429 | close(newnsfd); | |
2430 | ||
2431 | if (pipe(cpipe) < 0) | |
2432 | _exit(1); | |
2433 | ||
35174b0f FG |
2434 | struct pid_ns_clone_args args = { |
2435 | .cpipe = cpipe, | |
2436 | .sock = sock, | |
2437 | .tpid = tpid, | |
2438 | .wrapped = &pid_to_ns | |
2439 | }; | |
2440 | size_t stack_size = sysconf(_SC_PAGESIZE); | |
2441 | void *stack = alloca(stack_size); | |
2442 | ||
2443 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2444 | if (cpid < 0) |
2445 | _exit(1); | |
2446 | ||
237e200e SH |
2447 | // give the child 1 second to be done forking and |
2448 | // write its ack | |
2449 | if (!wait_for_sock(cpipe[0], 1)) | |
2450 | _exit(1); | |
2451 | ret = read(cpipe[0], &v, 1); | |
2452 | if (ret != sizeof(char) || v != '1') | |
2453 | _exit(1); | |
2454 | ||
2455 | if (!wait_for_pid(cpid)) | |
2456 | _exit(1); | |
2457 | _exit(0); | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * To read cgroup files with a particular pid, we will setns into the child | |
2462 | * pidns, open a pipe, fork a child - which will be the first to really be in | |
2463 | * the child ns - which does the cgfs_get_value and writes the data to the pipe. | |
2464 | */ | |
2465 | bool do_read_pids(pid_t tpid, const char *contrl, const char *cg, const char *file, char **d) | |
2466 | { | |
2467 | int sock[2] = {-1, -1}; | |
2468 | char *tmpdata = NULL; | |
2469 | int ret; | |
2470 | pid_t qpid, cpid = -1; | |
2471 | bool answer = false; | |
2472 | char v = '0'; | |
2473 | struct ucred cred; | |
2474 | size_t sz = 0, asz = 0; | |
2475 | ||
2476 | if (!cgfs_get_value(contrl, cg, file, &tmpdata)) | |
2477 | return false; | |
2478 | ||
2479 | /* | |
2480 | * Now we read the pids from returned data one by one, pass | |
2481 | * them into a child in the target namespace, read back the | |
2482 | * translated pids, and put them into our to-return data | |
2483 | */ | |
2484 | ||
2485 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2486 | perror("socketpair"); | |
2487 | free(tmpdata); | |
2488 | return false; | |
2489 | } | |
2490 | ||
2491 | cpid = fork(); | |
2492 | if (cpid == -1) | |
2493 | goto out; | |
2494 | ||
2495 | if (!cpid) // child - exits when done | |
2496 | pid_to_ns_wrapper(sock[1], tpid); | |
2497 | ||
2498 | char *ptr = tmpdata; | |
2499 | cred.uid = 0; | |
2500 | cred.gid = 0; | |
2501 | while (sscanf(ptr, "%d\n", &qpid) == 1) { | |
2502 | cred.pid = qpid; | |
2503 | ret = send_creds(sock[0], &cred, v, true); | |
2504 | ||
2505 | if (ret == SEND_CREDS_NOTSK) | |
2506 | goto next; | |
2507 | if (ret == SEND_CREDS_FAIL) | |
2508 | goto out; | |
2509 | ||
2510 | // read converted results | |
2511 | if (!wait_for_sock(sock[0], 2)) { | |
b8defc3d | 2512 | lxcfs_error("Timed out waiting for pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2513 | goto out; |
2514 | } | |
2515 | if (read(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2516 | lxcfs_error("Error reading pid from child: %s.\n", strerror(errno)); |
237e200e SH |
2517 | goto out; |
2518 | } | |
2519 | must_strcat_pid(d, &sz, &asz, qpid); | |
2520 | next: | |
2521 | ptr = strchr(ptr, '\n'); | |
2522 | if (!ptr) | |
2523 | break; | |
2524 | ptr++; | |
2525 | } | |
2526 | ||
2527 | cred.pid = getpid(); | |
2528 | v = '1'; | |
2529 | if (send_creds(sock[0], &cred, v, true) != SEND_CREDS_OK) { | |
2530 | // failed to ask child to exit | |
b8defc3d | 2531 | lxcfs_error("Failed to ask child to exit: %s.\n", strerror(errno)); |
237e200e SH |
2532 | goto out; |
2533 | } | |
2534 | ||
2535 | answer = true; | |
2536 | ||
2537 | out: | |
2538 | free(tmpdata); | |
2539 | if (cpid != -1) | |
2540 | wait_for_pid(cpid); | |
2541 | if (sock[0] != -1) { | |
2542 | close(sock[0]); | |
2543 | close(sock[1]); | |
2544 | } | |
2545 | return answer; | |
2546 | } | |
2547 | ||
2548 | int cg_read(const char *path, char *buf, size_t size, off_t offset, | |
2549 | struct fuse_file_info *fi) | |
2550 | { | |
2551 | struct fuse_context *fc = fuse_get_context(); | |
2552 | struct file_info *f = (struct file_info *)fi->fh; | |
2553 | struct cgfs_files *k = NULL; | |
2554 | char *data = NULL; | |
2555 | int ret, s; | |
2556 | bool r; | |
2557 | ||
2558 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2559 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_read."); |
237e200e SH |
2560 | return -EIO; |
2561 | } | |
2562 | ||
2563 | if (offset) | |
2564 | return 0; | |
2565 | ||
2566 | if (!fc) | |
2567 | return -EIO; | |
2568 | ||
2569 | if (!f->controller) | |
2570 | return -EINVAL; | |
2571 | ||
2572 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2573 | return -EINVAL; | |
2574 | } | |
2575 | free_key(k); | |
2576 | ||
2577 | ||
888f8f3c | 2578 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_RDONLY)) { |
237e200e SH |
2579 | ret = -EACCES; |
2580 | goto out; | |
2581 | } | |
2582 | ||
2583 | if (strcmp(f->file, "tasks") == 0 || | |
2584 | strcmp(f->file, "/tasks") == 0 || | |
2585 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2586 | strcmp(f->file, "cgroup.procs") == 0) | |
2587 | // special case - we have to translate the pids | |
2588 | r = do_read_pids(fc->pid, f->controller, f->cgroup, f->file, &data); | |
2589 | else | |
2590 | r = cgfs_get_value(f->controller, f->cgroup, f->file, &data); | |
2591 | ||
2592 | if (!r) { | |
2593 | ret = -EINVAL; | |
2594 | goto out; | |
2595 | } | |
2596 | ||
2597 | if (!data) { | |
2598 | ret = 0; | |
2599 | goto out; | |
2600 | } | |
2601 | s = strlen(data); | |
2602 | if (s > size) | |
2603 | s = size; | |
2604 | memcpy(buf, data, s); | |
2605 | if (s > 0 && s < size && data[s-1] != '\n') | |
2606 | buf[s++] = '\n'; | |
2607 | ||
2608 | ret = s; | |
2609 | ||
2610 | out: | |
2611 | free(data); | |
2612 | return ret; | |
2613 | } | |
2614 | ||
35174b0f | 2615 | static int pid_from_ns(int sock, pid_t tpid) |
237e200e SH |
2616 | { |
2617 | pid_t vpid; | |
2618 | struct ucred cred; | |
2619 | char v; | |
2620 | int ret; | |
2621 | ||
2622 | cred.uid = 0; | |
2623 | cred.gid = 0; | |
2624 | while (1) { | |
2625 | if (!wait_for_sock(sock, 2)) { | |
b8defc3d | 2626 | lxcfs_error("%s\n", "Timeout reading from parent."); |
35174b0f | 2627 | return 1; |
237e200e SH |
2628 | } |
2629 | if ((ret = read(sock, &vpid, sizeof(pid_t))) != sizeof(pid_t)) { | |
b8defc3d | 2630 | lxcfs_error("Bad read from parent: %s.\n", strerror(errno)); |
35174b0f | 2631 | return 1; |
237e200e SH |
2632 | } |
2633 | if (vpid == -1) // done | |
2634 | break; | |
2635 | v = '0'; | |
2636 | cred.pid = vpid; | |
2637 | if (send_creds(sock, &cred, v, true) != SEND_CREDS_OK) { | |
2638 | v = '1'; | |
2639 | cred.pid = getpid(); | |
2640 | if (send_creds(sock, &cred, v, false) != SEND_CREDS_OK) | |
35174b0f | 2641 | return 1; |
237e200e SH |
2642 | } |
2643 | } | |
35174b0f | 2644 | return 0; |
237e200e SH |
2645 | } |
2646 | ||
2647 | static void pid_from_ns_wrapper(int sock, pid_t tpid) | |
2648 | { | |
2649 | int newnsfd = -1, ret, cpipe[2]; | |
2650 | char fnam[100]; | |
2651 | pid_t cpid; | |
2652 | char v; | |
2653 | ||
2654 | ret = snprintf(fnam, sizeof(fnam), "/proc/%d/ns/pid", tpid); | |
2655 | if (ret < 0 || ret >= sizeof(fnam)) | |
2656 | _exit(1); | |
2657 | newnsfd = open(fnam, O_RDONLY); | |
2658 | if (newnsfd < 0) | |
2659 | _exit(1); | |
2660 | if (setns(newnsfd, 0) < 0) | |
2661 | _exit(1); | |
2662 | close(newnsfd); | |
2663 | ||
2664 | if (pipe(cpipe) < 0) | |
2665 | _exit(1); | |
2666 | ||
35174b0f FG |
2667 | struct pid_ns_clone_args args = { |
2668 | .cpipe = cpipe, | |
2669 | .sock = sock, | |
2670 | .tpid = tpid, | |
2671 | .wrapped = &pid_from_ns | |
2672 | }; | |
f0f8b851 SH |
2673 | size_t stack_size = sysconf(_SC_PAGESIZE); |
2674 | void *stack = alloca(stack_size); | |
35174b0f FG |
2675 | |
2676 | cpid = clone(pid_ns_clone_wrapper, stack + stack_size, SIGCHLD, &args); | |
237e200e SH |
2677 | if (cpid < 0) |
2678 | _exit(1); | |
2679 | ||
237e200e SH |
2680 | // give the child 1 second to be done forking and |
2681 | // write its ack | |
2682 | if (!wait_for_sock(cpipe[0], 1)) | |
f0f8b851 | 2683 | _exit(1); |
237e200e | 2684 | ret = read(cpipe[0], &v, 1); |
f0f8b851 SH |
2685 | if (ret != sizeof(char) || v != '1') |
2686 | _exit(1); | |
237e200e SH |
2687 | |
2688 | if (!wait_for_pid(cpid)) | |
2689 | _exit(1); | |
2690 | _exit(0); | |
237e200e SH |
2691 | } |
2692 | ||
2693 | /* | |
2694 | * Given host @uid, return the uid to which it maps in | |
2695 | * @pid's user namespace, or -1 if none. | |
2696 | */ | |
2697 | bool hostuid_to_ns(uid_t uid, pid_t pid, uid_t *answer) | |
2698 | { | |
2699 | FILE *f; | |
2700 | char line[400]; | |
2701 | ||
2702 | sprintf(line, "/proc/%d/uid_map", pid); | |
2703 | if ((f = fopen(line, "r")) == NULL) { | |
2704 | return false; | |
2705 | } | |
2706 | ||
2707 | *answer = convert_id_to_ns(f, uid); | |
2708 | fclose(f); | |
2709 | ||
2710 | if (*answer == -1) | |
2711 | return false; | |
2712 | return true; | |
2713 | } | |
2714 | ||
2715 | /* | |
2716 | * get_pid_creds: get the real uid and gid of @pid from | |
2717 | * /proc/$$/status | |
2718 | * (XXX should we use euid here?) | |
2719 | */ | |
2720 | void get_pid_creds(pid_t pid, uid_t *uid, gid_t *gid) | |
2721 | { | |
2722 | char line[400]; | |
2723 | uid_t u; | |
2724 | gid_t g; | |
2725 | FILE *f; | |
2726 | ||
2727 | *uid = -1; | |
2728 | *gid = -1; | |
2729 | sprintf(line, "/proc/%d/status", pid); | |
2730 | if ((f = fopen(line, "r")) == NULL) { | |
b8defc3d | 2731 | lxcfs_error("Error opening %s: %s\n", line, strerror(errno)); |
237e200e SH |
2732 | return; |
2733 | } | |
2734 | while (fgets(line, 400, f)) { | |
2735 | if (strncmp(line, "Uid:", 4) == 0) { | |
2736 | if (sscanf(line+4, "%u", &u) != 1) { | |
b8defc3d | 2737 | lxcfs_error("bad uid line for pid %u\n", pid); |
237e200e SH |
2738 | fclose(f); |
2739 | return; | |
2740 | } | |
2741 | *uid = u; | |
2742 | } else if (strncmp(line, "Gid:", 4) == 0) { | |
2743 | if (sscanf(line+4, "%u", &g) != 1) { | |
b8defc3d | 2744 | lxcfs_error("bad gid line for pid %u\n", pid); |
237e200e SH |
2745 | fclose(f); |
2746 | return; | |
2747 | } | |
2748 | *gid = g; | |
2749 | } | |
2750 | } | |
2751 | fclose(f); | |
2752 | } | |
2753 | ||
2754 | /* | |
2755 | * May the requestor @r move victim @v to a new cgroup? | |
2756 | * This is allowed if | |
2757 | * . they are the same task | |
2758 | * . they are ownedy by the same uid | |
2759 | * . @r is root on the host, or | |
2760 | * . @v's uid is mapped into @r's where @r is root. | |
2761 | */ | |
2762 | bool may_move_pid(pid_t r, uid_t r_uid, pid_t v) | |
2763 | { | |
2764 | uid_t v_uid, tmpuid; | |
2765 | gid_t v_gid; | |
2766 | ||
2767 | if (r == v) | |
2768 | return true; | |
2769 | if (r_uid == 0) | |
2770 | return true; | |
2771 | get_pid_creds(v, &v_uid, &v_gid); | |
2772 | if (r_uid == v_uid) | |
2773 | return true; | |
2774 | if (hostuid_to_ns(r_uid, r, &tmpuid) && tmpuid == 0 | |
2775 | && hostuid_to_ns(v_uid, r, &tmpuid)) | |
2776 | return true; | |
2777 | return false; | |
2778 | } | |
2779 | ||
2780 | static bool do_write_pids(pid_t tpid, uid_t tuid, const char *contrl, const char *cg, | |
2781 | const char *file, const char *buf) | |
2782 | { | |
2783 | int sock[2] = {-1, -1}; | |
2784 | pid_t qpid, cpid = -1; | |
2785 | FILE *pids_file = NULL; | |
2786 | bool answer = false, fail = false; | |
2787 | ||
2788 | pids_file = open_pids_file(contrl, cg); | |
2789 | if (!pids_file) | |
2790 | return false; | |
2791 | ||
2792 | /* | |
2793 | * write the pids to a socket, have helper in writer's pidns | |
2794 | * call movepid for us | |
2795 | */ | |
2796 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, sock) < 0) { | |
2797 | perror("socketpair"); | |
2798 | goto out; | |
2799 | } | |
2800 | ||
2801 | cpid = fork(); | |
2802 | if (cpid == -1) | |
2803 | goto out; | |
2804 | ||
2805 | if (!cpid) { // child | |
2806 | fclose(pids_file); | |
2807 | pid_from_ns_wrapper(sock[1], tpid); | |
2808 | } | |
2809 | ||
2810 | const char *ptr = buf; | |
2811 | while (sscanf(ptr, "%d", &qpid) == 1) { | |
2812 | struct ucred cred; | |
2813 | char v; | |
2814 | ||
2815 | if (write(sock[0], &qpid, sizeof(qpid)) != sizeof(qpid)) { | |
b8defc3d | 2816 | lxcfs_error("Error writing pid to child: %s.\n", strerror(errno)); |
237e200e SH |
2817 | goto out; |
2818 | } | |
2819 | ||
2820 | if (recv_creds(sock[0], &cred, &v)) { | |
2821 | if (v == '0') { | |
2822 | if (!may_move_pid(tpid, tuid, cred.pid)) { | |
2823 | fail = true; | |
2824 | break; | |
2825 | } | |
2826 | if (fprintf(pids_file, "%d", (int) cred.pid) < 0) | |
2827 | fail = true; | |
2828 | } | |
2829 | } | |
2830 | ||
2831 | ptr = strchr(ptr, '\n'); | |
2832 | if (!ptr) | |
2833 | break; | |
2834 | ptr++; | |
2835 | } | |
2836 | ||
2837 | /* All good, write the value */ | |
2838 | qpid = -1; | |
2839 | if (write(sock[0], &qpid ,sizeof(qpid)) != sizeof(qpid)) | |
b8defc3d | 2840 | lxcfs_error("%s\n", "Warning: failed to ask child to exit."); |
237e200e SH |
2841 | |
2842 | if (!fail) | |
2843 | answer = true; | |
2844 | ||
2845 | out: | |
2846 | if (cpid != -1) | |
2847 | wait_for_pid(cpid); | |
2848 | if (sock[0] != -1) { | |
2849 | close(sock[0]); | |
2850 | close(sock[1]); | |
2851 | } | |
2852 | if (pids_file) { | |
2853 | if (fclose(pids_file) != 0) | |
2854 | answer = false; | |
2855 | } | |
2856 | return answer; | |
2857 | } | |
2858 | ||
2859 | int cg_write(const char *path, const char *buf, size_t size, off_t offset, | |
2860 | struct fuse_file_info *fi) | |
2861 | { | |
2862 | struct fuse_context *fc = fuse_get_context(); | |
2863 | char *localbuf = NULL; | |
2864 | struct cgfs_files *k = NULL; | |
2865 | struct file_info *f = (struct file_info *)fi->fh; | |
2866 | bool r; | |
2867 | ||
2868 | if (f->type != LXC_TYPE_CGFILE) { | |
b8defc3d | 2869 | lxcfs_error("%s\n", "Internal error: directory cache info used in cg_write."); |
237e200e SH |
2870 | return -EIO; |
2871 | } | |
2872 | ||
2873 | if (offset) | |
2874 | return 0; | |
2875 | ||
2876 | if (!fc) | |
2877 | return -EIO; | |
2878 | ||
2879 | localbuf = alloca(size+1); | |
2880 | localbuf[size] = '\0'; | |
2881 | memcpy(localbuf, buf, size); | |
2882 | ||
2883 | if ((k = cgfs_get_key(f->controller, f->cgroup, f->file)) == NULL) { | |
2884 | size = -EINVAL; | |
2885 | goto out; | |
2886 | } | |
2887 | ||
2888 | if (!fc_may_access(fc, f->controller, f->cgroup, f->file, O_WRONLY)) { | |
2889 | size = -EACCES; | |
2890 | goto out; | |
2891 | } | |
2892 | ||
2893 | if (strcmp(f->file, "tasks") == 0 || | |
2894 | strcmp(f->file, "/tasks") == 0 || | |
2895 | strcmp(f->file, "/cgroup.procs") == 0 || | |
2896 | strcmp(f->file, "cgroup.procs") == 0) | |
2897 | // special case - we have to translate the pids | |
2898 | r = do_write_pids(fc->pid, fc->uid, f->controller, f->cgroup, f->file, localbuf); | |
2899 | else | |
2900 | r = cgfs_set_value(f->controller, f->cgroup, f->file, localbuf); | |
2901 | ||
2902 | if (!r) | |
2903 | size = -EINVAL; | |
2904 | ||
2905 | out: | |
2906 | free_key(k); | |
2907 | return size; | |
2908 | } | |
2909 | ||
2910 | int cg_chown(const char *path, uid_t uid, gid_t gid) | |
2911 | { | |
2912 | struct fuse_context *fc = fuse_get_context(); | |
2913 | char *cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2914 | struct cgfs_files *k = NULL; | |
2915 | const char *cgroup; | |
2916 | int ret; | |
2917 | ||
2918 | if (!fc) | |
2919 | return -EIO; | |
2920 | ||
2921 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2922 | return -EPERM; |
237e200e SH |
2923 | |
2924 | controller = pick_controller_from_path(fc, path); | |
2925 | if (!controller) | |
bc70ba9b CB |
2926 | return errno == ENOENT ? -EPERM : -errno; |
2927 | ||
237e200e SH |
2928 | cgroup = find_cgroup_in_path(path); |
2929 | if (!cgroup) | |
2930 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2931 | return -EPERM; |
237e200e SH |
2932 | |
2933 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
2934 | ||
2935 | if (!last) { | |
2936 | path1 = "/"; | |
2937 | path2 = cgdir; | |
2938 | } else { | |
2939 | path1 = cgdir; | |
2940 | path2 = last; | |
2941 | } | |
2942 | ||
2943 | if (is_child_cgroup(controller, path1, path2)) { | |
2944 | // get uid, gid, from '/tasks' file and make up a mode | |
2945 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
2946 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
2947 | ||
2948 | } else | |
2949 | k = cgfs_get_key(controller, path1, path2); | |
2950 | ||
2951 | if (!k) { | |
2952 | ret = -EINVAL; | |
2953 | goto out; | |
2954 | } | |
2955 | ||
2956 | /* | |
2957 | * This being a fuse request, the uid and gid must be valid | |
2958 | * in the caller's namespace. So we can just check to make | |
2959 | * sure that the caller is root in his uid, and privileged | |
2960 | * over the file's current owner. | |
2961 | */ | |
2962 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD)) { | |
2963 | ret = -EACCES; | |
2964 | goto out; | |
2965 | } | |
2966 | ||
2967 | ret = cgfs_chown_file(controller, cgroup, uid, gid); | |
2968 | ||
2969 | out: | |
2970 | free_key(k); | |
2971 | free(cgdir); | |
2972 | ||
2973 | return ret; | |
2974 | } | |
2975 | ||
2976 | int cg_chmod(const char *path, mode_t mode) | |
2977 | { | |
2978 | struct fuse_context *fc = fuse_get_context(); | |
2979 | char * cgdir = NULL, *last = NULL, *path1, *path2, *controller; | |
2980 | struct cgfs_files *k = NULL; | |
2981 | const char *cgroup; | |
2982 | int ret; | |
2983 | ||
2984 | if (!fc) | |
2985 | return -EIO; | |
2986 | ||
2987 | if (strcmp(path, "/cgroup") == 0) | |
bc70ba9b | 2988 | return -EPERM; |
237e200e SH |
2989 | |
2990 | controller = pick_controller_from_path(fc, path); | |
2991 | if (!controller) | |
bc70ba9b CB |
2992 | return errno == ENOENT ? -EPERM : -errno; |
2993 | ||
237e200e SH |
2994 | cgroup = find_cgroup_in_path(path); |
2995 | if (!cgroup) | |
2996 | /* this is just /cgroup/controller */ | |
bc70ba9b | 2997 | return -EPERM; |
237e200e SH |
2998 | |
2999 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3000 | ||
3001 | if (!last) { | |
3002 | path1 = "/"; | |
3003 | path2 = cgdir; | |
3004 | } else { | |
3005 | path1 = cgdir; | |
3006 | path2 = last; | |
3007 | } | |
3008 | ||
3009 | if (is_child_cgroup(controller, path1, path2)) { | |
3010 | // get uid, gid, from '/tasks' file and make up a mode | |
3011 | // That is a hack, until cgmanager gains a GetCgroupPerms fn. | |
3012 | k = cgfs_get_key(controller, cgroup, "tasks"); | |
3013 | ||
3014 | } else | |
3015 | k = cgfs_get_key(controller, path1, path2); | |
3016 | ||
3017 | if (!k) { | |
3018 | ret = -EINVAL; | |
3019 | goto out; | |
3020 | } | |
3021 | ||
3022 | /* | |
3023 | * This being a fuse request, the uid and gid must be valid | |
3024 | * in the caller's namespace. So we can just check to make | |
3025 | * sure that the caller is root in his uid, and privileged | |
3026 | * over the file's current owner. | |
3027 | */ | |
3028 | if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) { | |
3029 | ret = -EPERM; | |
3030 | goto out; | |
3031 | } | |
3032 | ||
3033 | if (!cgfs_chmod_file(controller, cgroup, mode)) { | |
3034 | ret = -EINVAL; | |
3035 | goto out; | |
3036 | } | |
3037 | ||
3038 | ret = 0; | |
3039 | out: | |
3040 | free_key(k); | |
3041 | free(cgdir); | |
3042 | return ret; | |
3043 | } | |
3044 | ||
3045 | int cg_mkdir(const char *path, mode_t mode) | |
3046 | { | |
3047 | struct fuse_context *fc = fuse_get_context(); | |
3048 | char *last = NULL, *path1, *cgdir = NULL, *controller, *next = NULL; | |
3049 | const char *cgroup; | |
3050 | int ret; | |
3051 | ||
3052 | if (!fc) | |
3053 | return -EIO; | |
3054 | ||
237e200e SH |
3055 | controller = pick_controller_from_path(fc, path); |
3056 | if (!controller) | |
2f7036d0 | 3057 | return errno == ENOENT ? -EPERM : -errno; |
237e200e SH |
3058 | |
3059 | cgroup = find_cgroup_in_path(path); | |
3060 | if (!cgroup) | |
bc70ba9b | 3061 | return -errno; |
237e200e SH |
3062 | |
3063 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3064 | if (!last) | |
3065 | path1 = "/"; | |
3066 | else | |
3067 | path1 = cgdir; | |
3068 | ||
3069 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3070 | if (initpid <= 0) | |
3071 | initpid = fc->pid; | |
3072 | if (!caller_is_in_ancestor(initpid, controller, path1, &next)) { | |
3073 | if (!next) | |
3074 | ret = -EINVAL; | |
3075 | else if (last && strcmp(next, last) == 0) | |
3076 | ret = -EEXIST; | |
3077 | else | |
2f7036d0 | 3078 | ret = -EPERM; |
237e200e SH |
3079 | goto out; |
3080 | } | |
3081 | ||
3082 | if (!fc_may_access(fc, controller, path1, NULL, O_RDWR)) { | |
3083 | ret = -EACCES; | |
3084 | goto out; | |
3085 | } | |
3086 | if (!caller_is_in_ancestor(initpid, controller, path1, NULL)) { | |
3087 | ret = -EACCES; | |
3088 | goto out; | |
3089 | } | |
3090 | ||
3091 | ret = cgfs_create(controller, cgroup, fc->uid, fc->gid); | |
3092 | ||
3093 | out: | |
3094 | free(cgdir); | |
3095 | free(next); | |
3096 | return ret; | |
3097 | } | |
3098 | ||
3099 | int cg_rmdir(const char *path) | |
3100 | { | |
3101 | struct fuse_context *fc = fuse_get_context(); | |
3102 | char *last = NULL, *cgdir = NULL, *controller, *next = NULL; | |
3103 | const char *cgroup; | |
3104 | int ret; | |
3105 | ||
3106 | if (!fc) | |
3107 | return -EIO; | |
3108 | ||
3109 | controller = pick_controller_from_path(fc, path); | |
e254948f CB |
3110 | if (!controller) /* Someone's trying to delete "/cgroup". */ |
3111 | return -EPERM; | |
237e200e SH |
3112 | |
3113 | cgroup = find_cgroup_in_path(path); | |
e254948f CB |
3114 | if (!cgroup) /* Someone's trying to delete a controller e.g. "/blkio". */ |
3115 | return -EPERM; | |
237e200e SH |
3116 | |
3117 | get_cgdir_and_path(cgroup, &cgdir, &last); | |
3118 | if (!last) { | |
e254948f CB |
3119 | /* Someone's trying to delete a cgroup on the same level as the |
3120 | * "/lxc" cgroup e.g. rmdir "/cgroup/blkio/lxc" or | |
3121 | * rmdir "/cgroup/blkio/init.slice". | |
3122 | */ | |
3123 | ret = -EPERM; | |
237e200e SH |
3124 | goto out; |
3125 | } | |
3126 | ||
3127 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3128 | if (initpid <= 0) | |
3129 | initpid = fc->pid; | |
3130 | if (!caller_is_in_ancestor(initpid, controller, cgroup, &next)) { | |
de77249b | 3131 | if (!last || (next && (strcmp(next, last) == 0))) |
237e200e SH |
3132 | ret = -EBUSY; |
3133 | else | |
3134 | ret = -ENOENT; | |
3135 | goto out; | |
3136 | } | |
3137 | ||
3138 | if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY)) { | |
3139 | ret = -EACCES; | |
3140 | goto out; | |
3141 | } | |
3142 | if (!caller_is_in_ancestor(initpid, controller, cgroup, NULL)) { | |
3143 | ret = -EACCES; | |
3144 | goto out; | |
3145 | } | |
3146 | ||
3147 | if (!cgfs_remove(controller, cgroup)) { | |
3148 | ret = -EINVAL; | |
3149 | goto out; | |
3150 | } | |
3151 | ||
3152 | ret = 0; | |
3153 | ||
3154 | out: | |
3155 | free(cgdir); | |
3156 | free(next); | |
3157 | return ret; | |
3158 | } | |
3159 | ||
3160 | static bool startswith(const char *line, const char *pref) | |
3161 | { | |
3162 | if (strncmp(line, pref, strlen(pref)) == 0) | |
3163 | return true; | |
3164 | return false; | |
3165 | } | |
3166 | ||
c6095b08 SH |
3167 | static void parse_memstat(char *memstat, unsigned long *cached, |
3168 | unsigned long *active_anon, unsigned long *inactive_anon, | |
3169 | unsigned long *active_file, unsigned long *inactive_file, | |
3170 | unsigned long *unevictable) | |
237e200e SH |
3171 | { |
3172 | char *eol; | |
3173 | ||
237e200e | 3174 | while (*memstat) { |
4accebfb AS |
3175 | if (startswith(memstat, "total_cache")) { |
3176 | sscanf(memstat + 11, "%lu", cached); | |
c6095b08 | 3177 | *cached /= 1024; |
4accebfb AS |
3178 | } else if (startswith(memstat, "total_active_anon")) { |
3179 | sscanf(memstat + 17, "%lu", active_anon); | |
c6095b08 | 3180 | *active_anon /= 1024; |
4accebfb AS |
3181 | } else if (startswith(memstat, "total_inactive_anon")) { |
3182 | sscanf(memstat + 19, "%lu", inactive_anon); | |
c6095b08 | 3183 | *inactive_anon /= 1024; |
4accebfb AS |
3184 | } else if (startswith(memstat, "total_active_file")) { |
3185 | sscanf(memstat + 17, "%lu", active_file); | |
c6095b08 | 3186 | *active_file /= 1024; |
4accebfb AS |
3187 | } else if (startswith(memstat, "total_inactive_file")) { |
3188 | sscanf(memstat + 19, "%lu", inactive_file); | |
c6095b08 | 3189 | *inactive_file /= 1024; |
4accebfb AS |
3190 | } else if (startswith(memstat, "total_unevictable")) { |
3191 | sscanf(memstat + 17, "%lu", unevictable); | |
c6095b08 | 3192 | *unevictable /= 1024; |
237e200e SH |
3193 | } |
3194 | eol = strchr(memstat, '\n'); | |
3195 | if (!eol) | |
3196 | return; | |
3197 | memstat = eol+1; | |
3198 | } | |
3199 | } | |
3200 | ||
3201 | static void get_blkio_io_value(char *str, unsigned major, unsigned minor, char *iotype, unsigned long *v) | |
3202 | { | |
3203 | char *eol; | |
3204 | char key[32]; | |
3205 | ||
3206 | memset(key, 0, 32); | |
3207 | snprintf(key, 32, "%u:%u %s", major, minor, iotype); | |
3208 | ||
3209 | size_t len = strlen(key); | |
3210 | *v = 0; | |
3211 | ||
3212 | while (*str) { | |
3213 | if (startswith(str, key)) { | |
3214 | sscanf(str + len, "%lu", v); | |
3215 | return; | |
3216 | } | |
3217 | eol = strchr(str, '\n'); | |
3218 | if (!eol) | |
3219 | return; | |
3220 | str = eol+1; | |
3221 | } | |
3222 | } | |
3223 | ||
3224 | static int read_file(const char *path, char *buf, size_t size, | |
3225 | struct file_info *d) | |
3226 | { | |
3227 | size_t linelen = 0, total_len = 0, rv = 0; | |
3228 | char *line = NULL; | |
3229 | char *cache = d->buf; | |
3230 | size_t cache_size = d->buflen; | |
3231 | FILE *f = fopen(path, "r"); | |
3232 | if (!f) | |
3233 | return 0; | |
3234 | ||
3235 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3236 | ssize_t l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3237 | if (l < 0) { |
3238 | perror("Error writing to cache"); | |
3239 | rv = 0; | |
3240 | goto err; | |
3241 | } | |
3242 | if (l >= cache_size) { | |
b8defc3d | 3243 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3244 | rv = 0; |
3245 | goto err; | |
3246 | } | |
3247 | cache += l; | |
3248 | cache_size -= l; | |
3249 | total_len += l; | |
3250 | } | |
3251 | ||
3252 | d->size = total_len; | |
a262ddb7 CB |
3253 | if (total_len > size) |
3254 | total_len = size; | |
237e200e SH |
3255 | |
3256 | /* read from off 0 */ | |
3257 | memcpy(buf, d->buf, total_len); | |
3258 | rv = total_len; | |
3259 | err: | |
3260 | fclose(f); | |
3261 | free(line); | |
3262 | return rv; | |
3263 | } | |
3264 | ||
3265 | /* | |
3266 | * FUSE ops for /proc | |
3267 | */ | |
3268 | ||
018246ff | 3269 | static unsigned long get_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3270 | { |
3271 | char *memlimit_str = NULL; | |
3272 | unsigned long memlimit = -1; | |
3273 | ||
018246ff | 3274 | if (cgfs_get_value("memory", cgroup, file, &memlimit_str)) |
237e200e SH |
3275 | memlimit = strtoul(memlimit_str, NULL, 10); |
3276 | ||
3277 | free(memlimit_str); | |
3278 | ||
3279 | return memlimit; | |
3280 | } | |
3281 | ||
018246ff | 3282 | static unsigned long get_min_memlimit(const char *cgroup, const char *file) |
237e200e SH |
3283 | { |
3284 | char *copy = strdupa(cgroup); | |
3285 | unsigned long memlimit = 0, retlimit; | |
3286 | ||
018246ff | 3287 | retlimit = get_memlimit(copy, file); |
237e200e SH |
3288 | |
3289 | while (strcmp(copy, "/") != 0) { | |
3290 | copy = dirname(copy); | |
018246ff | 3291 | memlimit = get_memlimit(copy, file); |
237e200e SH |
3292 | if (memlimit != -1 && memlimit < retlimit) |
3293 | retlimit = memlimit; | |
3294 | }; | |
3295 | ||
3296 | return retlimit; | |
3297 | } | |
3298 | ||
3299 | static int proc_meminfo_read(char *buf, size_t size, off_t offset, | |
3300 | struct fuse_file_info *fi) | |
3301 | { | |
3302 | struct fuse_context *fc = fuse_get_context(); | |
3303 | struct file_info *d = (struct file_info *)fi->fh; | |
3304 | char *cg; | |
3305 | char *memusage_str = NULL, *memstat_str = NULL, | |
018246ff | 3306 | *memswlimit_str = NULL, *memswusage_str = NULL; |
237e200e | 3307 | unsigned long memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0, |
c6095b08 | 3308 | cached = 0, hosttotal = 0, active_anon = 0, inactive_anon = 0, |
594a10e6 WB |
3309 | active_file = 0, inactive_file = 0, unevictable = 0, |
3310 | hostswtotal = 0; | |
237e200e SH |
3311 | char *line = NULL; |
3312 | size_t linelen = 0, total_len = 0, rv = 0; | |
3313 | char *cache = d->buf; | |
3314 | size_t cache_size = d->buflen; | |
3315 | FILE *f = NULL; | |
3316 | ||
3317 | if (offset){ | |
3318 | if (offset > d->size) | |
3319 | return -EINVAL; | |
3320 | if (!d->cached) | |
3321 | return 0; | |
3322 | int left = d->size - offset; | |
3323 | total_len = left > size ? size: left; | |
3324 | memcpy(buf, cache + offset, total_len); | |
3325 | return total_len; | |
3326 | } | |
3327 | ||
3328 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3329 | if (initpid <= 0) | |
3330 | initpid = fc->pid; | |
3331 | cg = get_pid_cgroup(initpid, "memory"); | |
3332 | if (!cg) | |
3333 | return read_file("/proc/meminfo", buf, size, d); | |
6d2f6996 | 3334 | prune_init_slice(cg); |
237e200e | 3335 | |
018246ff | 3336 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
237e200e SH |
3337 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) |
3338 | goto err; | |
3339 | if (!cgfs_get_value("memory", cg, "memory.stat", &memstat_str)) | |
3340 | goto err; | |
3341 | ||
3342 | // Following values are allowed to fail, because swapaccount might be turned | |
3343 | // off for current kernel | |
3344 | if(cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str) && | |
3345 | cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str)) | |
3346 | { | |
018246ff | 3347 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
237e200e SH |
3348 | memswusage = strtoul(memswusage_str, NULL, 10); |
3349 | ||
237e200e SH |
3350 | memswlimit = memswlimit / 1024; |
3351 | memswusage = memswusage / 1024; | |
3352 | } | |
3353 | ||
3354 | memusage = strtoul(memusage_str, NULL, 10); | |
3355 | memlimit /= 1024; | |
3356 | memusage /= 1024; | |
3357 | ||
c6095b08 SH |
3358 | parse_memstat(memstat_str, &cached, &active_anon, |
3359 | &inactive_anon, &active_file, &inactive_file, | |
3360 | &unevictable); | |
237e200e SH |
3361 | |
3362 | f = fopen("/proc/meminfo", "r"); | |
3363 | if (!f) | |
3364 | goto err; | |
3365 | ||
3366 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3367 | ssize_t l; |
237e200e SH |
3368 | char *printme, lbuf[100]; |
3369 | ||
3370 | memset(lbuf, 0, 100); | |
3371 | if (startswith(line, "MemTotal:")) { | |
594a10e6 | 3372 | sscanf(line+sizeof("MemTotal:")-1, "%lu", &hosttotal); |
237e200e SH |
3373 | if (hosttotal < memlimit) |
3374 | memlimit = hosttotal; | |
3375 | snprintf(lbuf, 100, "MemTotal: %8lu kB\n", memlimit); | |
3376 | printme = lbuf; | |
3377 | } else if (startswith(line, "MemFree:")) { | |
3378 | snprintf(lbuf, 100, "MemFree: %8lu kB\n", memlimit - memusage); | |
3379 | printme = lbuf; | |
3380 | } else if (startswith(line, "MemAvailable:")) { | |
ad19b86d | 3381 | snprintf(lbuf, 100, "MemAvailable: %8lu kB\n", memlimit - memusage + cached); |
237e200e SH |
3382 | printme = lbuf; |
3383 | } else if (startswith(line, "SwapTotal:") && memswlimit > 0) { | |
594a10e6 | 3384 | sscanf(line+sizeof("SwapTotal:")-1, "%lu", &hostswtotal); |
4127e51b | 3385 | if (hostswtotal < memswlimit) |
3386 | memswlimit = hostswtotal; | |
3387 | snprintf(lbuf, 100, "SwapTotal: %8lu kB\n", memswlimit); | |
237e200e SH |
3388 | printme = lbuf; |
3389 | } else if (startswith(line, "SwapFree:") && memswlimit > 0 && memswusage > 0) { | |
4127e51b | 3390 | unsigned long swaptotal = memswlimit, |
b4665ce0 SH |
3391 | swapusage = memswusage - memusage, |
3392 | swapfree = swapusage < swaptotal ? swaptotal - swapusage : 0; | |
3393 | snprintf(lbuf, 100, "SwapFree: %8lu kB\n", swapfree); | |
237e200e | 3394 | printme = lbuf; |
da35d72a SH |
3395 | } else if (startswith(line, "Slab:")) { |
3396 | snprintf(lbuf, 100, "Slab: %8lu kB\n", 0UL); | |
3397 | printme = lbuf; | |
237e200e SH |
3398 | } else if (startswith(line, "Buffers:")) { |
3399 | snprintf(lbuf, 100, "Buffers: %8lu kB\n", 0UL); | |
3400 | printme = lbuf; | |
3401 | } else if (startswith(line, "Cached:")) { | |
3402 | snprintf(lbuf, 100, "Cached: %8lu kB\n", cached); | |
3403 | printme = lbuf; | |
3404 | } else if (startswith(line, "SwapCached:")) { | |
3405 | snprintf(lbuf, 100, "SwapCached: %8lu kB\n", 0UL); | |
3406 | printme = lbuf; | |
2f306ad3 | 3407 | } else if (startswith(line, "Active:")) { |
c6095b08 SH |
3408 | snprintf(lbuf, 100, "Active: %8lu kB\n", |
3409 | active_anon + active_file); | |
3410 | printme = lbuf; | |
2f306ad3 | 3411 | } else if (startswith(line, "Inactive:")) { |
c6095b08 SH |
3412 | snprintf(lbuf, 100, "Inactive: %8lu kB\n", |
3413 | inactive_anon + inactive_file); | |
3414 | printme = lbuf; | |
3415 | } else if (startswith(line, "Active(anon)")) { | |
3416 | snprintf(lbuf, 100, "Active(anon): %8lu kB\n", active_anon); | |
3417 | printme = lbuf; | |
3418 | } else if (startswith(line, "Inactive(anon)")) { | |
3419 | snprintf(lbuf, 100, "Inactive(anon): %8lu kB\n", inactive_anon); | |
3420 | printme = lbuf; | |
3421 | } else if (startswith(line, "Active(file)")) { | |
3422 | snprintf(lbuf, 100, "Active(file): %8lu kB\n", active_file); | |
3423 | printme = lbuf; | |
3424 | } else if (startswith(line, "Inactive(file)")) { | |
3425 | snprintf(lbuf, 100, "Inactive(file): %8lu kB\n", inactive_file); | |
3426 | printme = lbuf; | |
3427 | } else if (startswith(line, "Unevictable")) { | |
3428 | snprintf(lbuf, 100, "Unevictable: %8lu kB\n", unevictable); | |
3429 | printme = lbuf; | |
3430 | } else if (startswith(line, "SReclaimable")) { | |
3431 | snprintf(lbuf, 100, "SReclaimable: %8lu kB\n", 0UL); | |
3432 | printme = lbuf; | |
3433 | } else if (startswith(line, "SUnreclaim")) { | |
3434 | snprintf(lbuf, 100, "SUnreclaim: %8lu kB\n", 0UL); | |
3435 | printme = lbuf; | |
237e200e SH |
3436 | } else |
3437 | printme = line; | |
3438 | ||
3439 | l = snprintf(cache, cache_size, "%s", printme); | |
3440 | if (l < 0) { | |
3441 | perror("Error writing to cache"); | |
3442 | rv = 0; | |
3443 | goto err; | |
3444 | ||
3445 | } | |
3446 | if (l >= cache_size) { | |
b8defc3d | 3447 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3448 | rv = 0; |
3449 | goto err; | |
3450 | } | |
3451 | ||
3452 | cache += l; | |
3453 | cache_size -= l; | |
3454 | total_len += l; | |
3455 | } | |
3456 | ||
3457 | d->cached = 1; | |
3458 | d->size = total_len; | |
3459 | if (total_len > size ) total_len = size; | |
3460 | memcpy(buf, d->buf, total_len); | |
3461 | ||
3462 | rv = total_len; | |
3463 | err: | |
3464 | if (f) | |
3465 | fclose(f); | |
3466 | free(line); | |
3467 | free(cg); | |
3468 | free(memusage_str); | |
3469 | free(memswlimit_str); | |
3470 | free(memswusage_str); | |
3471 | free(memstat_str); | |
237e200e SH |
3472 | return rv; |
3473 | } | |
3474 | ||
3475 | /* | |
3476 | * Read the cpuset.cpus for cg | |
3477 | * Return the answer in a newly allocated string which must be freed | |
3478 | */ | |
3479 | static char *get_cpuset(const char *cg) | |
3480 | { | |
3481 | char *answer; | |
3482 | ||
3483 | if (!cgfs_get_value("cpuset", cg, "cpuset.cpus", &answer)) | |
3484 | return NULL; | |
3485 | return answer; | |
3486 | } | |
3487 | ||
3488 | bool cpu_in_cpuset(int cpu, const char *cpuset); | |
3489 | ||
3490 | static bool cpuline_in_cpuset(const char *line, const char *cpuset) | |
3491 | { | |
3492 | int cpu; | |
3493 | ||
3494 | if (sscanf(line, "processor : %d", &cpu) != 1) | |
3495 | return false; | |
3496 | return cpu_in_cpuset(cpu, cpuset); | |
3497 | } | |
3498 | ||
3499 | /* | |
3500 | * check whether this is a '^processor" line in /proc/cpuinfo | |
3501 | */ | |
3502 | static bool is_processor_line(const char *line) | |
3503 | { | |
3504 | int cpu; | |
3505 | ||
3506 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
3507 | return true; | |
3508 | return false; | |
3509 | } | |
3510 | ||
3511 | static int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
3512 | struct fuse_file_info *fi) | |
3513 | { | |
3514 | struct fuse_context *fc = fuse_get_context(); | |
3515 | struct file_info *d = (struct file_info *)fi->fh; | |
3516 | char *cg; | |
3517 | char *cpuset = NULL; | |
3518 | char *line = NULL; | |
3519 | size_t linelen = 0, total_len = 0, rv = 0; | |
f676eb79 SH |
3520 | bool am_printing = false, firstline = true, is_s390x = false; |
3521 | int curcpu = -1, cpu; | |
237e200e SH |
3522 | char *cache = d->buf; |
3523 | size_t cache_size = d->buflen; | |
3524 | FILE *f = NULL; | |
3525 | ||
3526 | if (offset){ | |
3527 | if (offset > d->size) | |
3528 | return -EINVAL; | |
3529 | if (!d->cached) | |
3530 | return 0; | |
3531 | int left = d->size - offset; | |
3532 | total_len = left > size ? size: left; | |
3533 | memcpy(buf, cache + offset, total_len); | |
3534 | return total_len; | |
3535 | } | |
3536 | ||
3537 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3538 | if (initpid <= 0) | |
3539 | initpid = fc->pid; | |
3540 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3541 | if (!cg) | |
3542 | return read_file("proc/cpuinfo", buf, size, d); | |
6d2f6996 | 3543 | prune_init_slice(cg); |
237e200e SH |
3544 | |
3545 | cpuset = get_cpuset(cg); | |
3546 | if (!cpuset) | |
3547 | goto err; | |
3548 | ||
3549 | f = fopen("/proc/cpuinfo", "r"); | |
3550 | if (!f) | |
3551 | goto err; | |
3552 | ||
3553 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3554 | ssize_t l; |
f676eb79 SH |
3555 | if (firstline) { |
3556 | firstline = false; | |
3557 | if (strstr(line, "IBM/S390") != NULL) { | |
3558 | is_s390x = true; | |
3559 | am_printing = true; | |
5ed9d4e2 | 3560 | continue; |
f676eb79 SH |
3561 | } |
3562 | } | |
5ed9d4e2 SH |
3563 | if (strncmp(line, "# processors:", 12) == 0) |
3564 | continue; | |
237e200e SH |
3565 | if (is_processor_line(line)) { |
3566 | am_printing = cpuline_in_cpuset(line, cpuset); | |
3567 | if (am_printing) { | |
3568 | curcpu ++; | |
3569 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); | |
3570 | if (l < 0) { | |
3571 | perror("Error writing to cache"); | |
3572 | rv = 0; | |
3573 | goto err; | |
3574 | } | |
3575 | if (l >= cache_size) { | |
b8defc3d | 3576 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3577 | rv = 0; |
3578 | goto err; | |
3579 | } | |
3580 | cache += l; | |
3581 | cache_size -= l; | |
3582 | total_len += l; | |
3583 | } | |
3584 | continue; | |
f676eb79 SH |
3585 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { |
3586 | char *p; | |
3587 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3588 | continue; | |
3589 | curcpu ++; | |
3590 | p = strchr(line, ':'); | |
3591 | if (!p || !*p) | |
3592 | goto err; | |
3593 | p++; | |
5ed9d4e2 | 3594 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
f676eb79 SH |
3595 | if (l < 0) { |
3596 | perror("Error writing to cache"); | |
3597 | rv = 0; | |
3598 | goto err; | |
3599 | } | |
3600 | if (l >= cache_size) { | |
b8defc3d | 3601 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
f676eb79 SH |
3602 | rv = 0; |
3603 | goto err; | |
3604 | } | |
3605 | cache += l; | |
3606 | cache_size -= l; | |
3607 | total_len += l; | |
3608 | continue; | |
3609 | ||
237e200e SH |
3610 | } |
3611 | if (am_printing) { | |
3612 | l = snprintf(cache, cache_size, "%s", line); | |
3613 | if (l < 0) { | |
3614 | perror("Error writing to cache"); | |
3615 | rv = 0; | |
3616 | goto err; | |
3617 | } | |
3618 | if (l >= cache_size) { | |
b8defc3d | 3619 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3620 | rv = 0; |
3621 | goto err; | |
3622 | } | |
3623 | cache += l; | |
3624 | cache_size -= l; | |
3625 | total_len += l; | |
3626 | } | |
3627 | } | |
3628 | ||
5ed9d4e2 SH |
3629 | if (is_s390x) { |
3630 | char *origcache = d->buf; | |
a262ddb7 | 3631 | ssize_t l; |
5ed9d4e2 SH |
3632 | do { |
3633 | d->buf = malloc(d->buflen); | |
3634 | } while (!d->buf); | |
3635 | cache = d->buf; | |
3636 | cache_size = d->buflen; | |
3637 | total_len = 0; | |
3638 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3639 | if (l < 0 || l >= cache_size) { | |
3640 | free(origcache); | |
3641 | goto err; | |
3642 | } | |
3643 | cache_size -= l; | |
3644 | cache += l; | |
3645 | total_len += l; | |
3646 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3647 | if (l < 0 || l >= cache_size) { | |
3648 | free(origcache); | |
3649 | goto err; | |
3650 | } | |
3651 | cache_size -= l; | |
3652 | cache += l; | |
3653 | total_len += l; | |
3654 | l = snprintf(cache, cache_size, "%s", origcache); | |
3655 | free(origcache); | |
3656 | if (l < 0 || l >= cache_size) | |
3657 | goto err; | |
3658 | total_len += l; | |
3659 | } | |
3660 | ||
237e200e SH |
3661 | d->cached = 1; |
3662 | d->size = total_len; | |
3663 | if (total_len > size ) total_len = size; | |
3664 | ||
3665 | /* read from off 0 */ | |
3666 | memcpy(buf, d->buf, total_len); | |
3667 | rv = total_len; | |
3668 | err: | |
3669 | if (f) | |
3670 | fclose(f); | |
3671 | free(line); | |
3672 | free(cpuset); | |
3673 | free(cg); | |
3674 | return rv; | |
3675 | } | |
3676 | ||
0ecddf02 | 3677 | static uint64_t get_reaper_start_time(pid_t pid) |
9ac264cf | 3678 | { |
9ac264cf | 3679 | int ret; |
0ecddf02 CB |
3680 | FILE *f; |
3681 | uint64_t starttime; | |
3682 | /* strlen("/proc/") = 6 | |
3683 | * + | |
3684 | * LXCFS_NUMSTRLEN64 | |
3685 | * + | |
3686 | * strlen("/stat") = 5 | |
3687 | * + | |
3688 | * \0 = 1 | |
3689 | * */ | |
3690 | #define __PROC_PID_STAT_LEN (6 + LXCFS_NUMSTRLEN64 + 5 + 1) | |
3691 | char path[__PROC_PID_STAT_LEN]; | |
9ac264cf JB |
3692 | pid_t qpid; |
3693 | ||
3694 | qpid = lookup_initpid_in_store(pid); | |
0ecddf02 CB |
3695 | if (qpid <= 0) { |
3696 | /* Caller can check for EINVAL on 0. */ | |
3697 | errno = EINVAL; | |
9ac264cf | 3698 | return 0; |
0ecddf02 | 3699 | } |
9ac264cf | 3700 | |
0ecddf02 CB |
3701 | ret = snprintf(path, __PROC_PID_STAT_LEN, "/proc/%d/stat", qpid); |
3702 | if (ret < 0 || ret >= __PROC_PID_STAT_LEN) { | |
3703 | /* Caller can check for EINVAL on 0. */ | |
3704 | errno = EINVAL; | |
9ac264cf | 3705 | return 0; |
0ecddf02 | 3706 | } |
9ac264cf | 3707 | |
0ecddf02 CB |
3708 | f = fopen(path, "r"); |
3709 | if (!f) { | |
3710 | /* Caller can check for EINVAL on 0. */ | |
3711 | errno = EINVAL; | |
9ac264cf | 3712 | return 0; |
0ecddf02 | 3713 | } |
9ac264cf | 3714 | |
0ecddf02 CB |
3715 | /* Note that the *scanf() argument supression requires that length |
3716 | * modifiers such as "l" are omitted. Otherwise some compilers will yell | |
3717 | * at us. It's like telling someone you're not married and then asking | |
3718 | * if you can bring your wife to the party. | |
3719 | */ | |
3720 | ret = fscanf(f, "%*d " /* (1) pid %d */ | |
3721 | "%*s " /* (2) comm %s */ | |
3722 | "%*c " /* (3) state %c */ | |
3723 | "%*d " /* (4) ppid %d */ | |
3724 | "%*d " /* (5) pgrp %d */ | |
3725 | "%*d " /* (6) session %d */ | |
3726 | "%*d " /* (7) tty_nr %d */ | |
3727 | "%*d " /* (8) tpgid %d */ | |
3728 | "%*u " /* (9) flags %u */ | |
3729 | "%*u " /* (10) minflt %lu */ | |
3730 | "%*u " /* (11) cminflt %lu */ | |
3731 | "%*u " /* (12) majflt %lu */ | |
3732 | "%*u " /* (13) cmajflt %lu */ | |
3733 | "%*u " /* (14) utime %lu */ | |
3734 | "%*u " /* (15) stime %lu */ | |
3735 | "%*d " /* (16) cutime %ld */ | |
3736 | "%*d " /* (17) cstime %ld */ | |
3737 | "%*d " /* (18) priority %ld */ | |
3738 | "%*d " /* (19) nice %ld */ | |
3739 | "%*d " /* (20) num_threads %ld */ | |
3740 | "%*d " /* (21) itrealvalue %ld */ | |
3741 | "%" PRIu64, /* (22) starttime %llu */ | |
3742 | &starttime); | |
3743 | if (ret != 1) { | |
3744 | fclose(f); | |
3745 | /* Caller can check for EINVAL on 0. */ | |
3746 | errno = EINVAL; | |
3747 | return 0; | |
3748 | } | |
3749 | ||
3750 | fclose(f); | |
3751 | ||
3752 | errno = 0; | |
3753 | return starttime; | |
3754 | } | |
3755 | ||
3756 | static uint64_t get_reaper_start_time_in_sec(pid_t pid) | |
3757 | { | |
3758 | uint64_t clockticks; | |
3759 | int64_t ticks_per_sec; | |
3760 | ||
3761 | clockticks = get_reaper_start_time(pid); | |
3762 | if (clockticks == 0 && errno == EINVAL) { | |
3763 | lxcfs_debug("failed to retrieve start time of pid %d\n", pid); | |
3764 | return 0; | |
3765 | } | |
3766 | ||
3767 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3768 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
3769 | lxcfs_debug( | |
3770 | "%s\n", | |
3771 | "failed to determine number of clock ticks in a second"); | |
3772 | return 0; | |
3773 | } | |
3774 | ||
3775 | return (clockticks /= ticks_per_sec); | |
3776 | } | |
3777 | ||
3778 | static uint64_t get_reaper_age(pid_t pid) | |
3779 | { | |
3780 | uint64_t procstart, uptime, procage; | |
3781 | ||
3782 | /* We need to substract the time the process has started since system | |
3783 | * boot minus the time when the system has started to get the actual | |
3784 | * reaper age. | |
3785 | */ | |
3786 | procstart = get_reaper_start_time_in_sec(pid); | |
3787 | procage = procstart; | |
3788 | if (procstart > 0) { | |
3789 | int ret; | |
3790 | struct timespec spec; | |
3791 | ||
3792 | ret = clock_gettime(CLOCK_BOOTTIME, &spec); | |
3793 | if (ret < 0) | |
3794 | return 0; | |
3795 | /* We could make this more precise here by using the tv_nsec | |
3796 | * field in the timespec struct and convert it to milliseconds | |
3797 | * and then create a double for the seconds and milliseconds but | |
3798 | * that seems more work than it is worth. | |
3799 | */ | |
3800 | uptime = spec.tv_sec; | |
3801 | procage = uptime - procstart; | |
3802 | } | |
3803 | ||
3804 | return procage; | |
3805 | } | |
3806 | ||
8be92dd1 JS |
3807 | /* |
3808 | * Returns 0 on success. | |
3809 | * It is the caller's responsibility to free `return_usage`, unless this | |
3810 | * function returns an error. | |
3811 | */ | |
3812 | static int read_cpuacct_usage_all(char *cg, char *cpuset, struct cpuacct_usage **return_usage) | |
3813 | { | |
3814 | int cpucount = get_nprocs(); | |
3815 | struct cpuacct_usage *cpu_usage; | |
3816 | int rv = 0, i, j, ret, read_pos = 0, read_cnt; | |
3817 | int cg_cpu; | |
3818 | uint64_t cg_user, cg_system; | |
3819 | int64_t ticks_per_sec; | |
3820 | char *usage_str = NULL; | |
3821 | ||
3822 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
3823 | ||
3824 | if (ticks_per_sec < 0 && errno == EINVAL) { | |
3825 | lxcfs_debug( | |
3826 | "%s\n", | |
3827 | "read_cpuacct_usage_all failed to determine number of clock ticks " | |
3828 | "in a second"); | |
3829 | return -1; | |
3830 | } | |
3831 | ||
3832 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); | |
3833 | if (!cpu_usage) | |
3834 | return -ENOMEM; | |
3835 | ||
3836 | if (!cgfs_get_value("cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
3837 | rv = -1; | |
3838 | goto err; | |
3839 | } | |
3840 | ||
3841 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) { | |
3842 | lxcfs_error("read_cpuacct_usage_all reading first line from " | |
3843 | "%s/cpuacct.usage_all failed.\n", cg); | |
3844 | rv = -1; | |
3845 | goto err; | |
3846 | } | |
3847 | ||
3848 | read_pos += read_cnt; | |
3849 | ||
3850 | for (i = 0, j = 0; i < cpucount; i++) { | |
3851 | ret = sscanf(usage_str + read_pos, "%d %lu %lu\n%n", &cg_cpu, &cg_user, | |
3852 | &cg_system, &read_cnt); | |
3853 | ||
3854 | if (ret == EOF) | |
3855 | break; | |
3856 | ||
3857 | if (ret != 3) { | |
3858 | lxcfs_error("read_cpuacct_usage_all reading from %s/cpuacct.usage_all " | |
3859 | "failed.\n", cg); | |
3860 | rv = -1; | |
3861 | goto err; | |
3862 | } | |
3863 | ||
3864 | read_pos += read_cnt; | |
3865 | ||
3866 | if (!cpu_in_cpuset(i, cpuset)) | |
3867 | continue; | |
3868 | ||
3869 | /* Convert the time from nanoseconds to USER_HZ */ | |
3870 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
3871 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
3872 | j++; | |
3873 | } | |
3874 | ||
3875 | rv = 0; | |
3876 | *return_usage = cpu_usage; | |
3877 | ||
3878 | err: | |
3879 | if (usage_str) | |
3880 | free(usage_str); | |
3881 | ||
3882 | if (rv != 0) { | |
3883 | free(cpu_usage); | |
3884 | *return_usage = NULL; | |
3885 | } | |
3886 | ||
3887 | return rv; | |
3888 | } | |
3889 | ||
f34de69a | 3890 | #define CPUALL_MAX_SIZE (BUF_RESERVE_SIZE / 2) |
237e200e SH |
3891 | static int proc_stat_read(char *buf, size_t size, off_t offset, |
3892 | struct fuse_file_info *fi) | |
3893 | { | |
3894 | struct fuse_context *fc = fuse_get_context(); | |
3895 | struct file_info *d = (struct file_info *)fi->fh; | |
3896 | char *cg; | |
3897 | char *cpuset = NULL; | |
3898 | char *line = NULL; | |
3899 | size_t linelen = 0, total_len = 0, rv = 0; | |
3900 | int curcpu = -1; /* cpu numbering starts at 0 */ | |
7144f069 | 3901 | unsigned long user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, softirq = 0, steal = 0, guest = 0, guest_nice = 0; |
237e200e | 3902 | unsigned long user_sum = 0, nice_sum = 0, system_sum = 0, idle_sum = 0, iowait_sum = 0, |
7144f069 | 3903 | irq_sum = 0, softirq_sum = 0, steal_sum = 0, guest_sum = 0, guest_nice_sum = 0; |
237e200e SH |
3904 | char cpuall[CPUALL_MAX_SIZE]; |
3905 | /* reserve for cpu all */ | |
3906 | char *cache = d->buf + CPUALL_MAX_SIZE; | |
3907 | size_t cache_size = d->buflen - CPUALL_MAX_SIZE; | |
3908 | FILE *f = NULL; | |
8be92dd1 | 3909 | struct cpuacct_usage *cg_cpu_usage = NULL; |
237e200e SH |
3910 | |
3911 | if (offset){ | |
3912 | if (offset > d->size) | |
3913 | return -EINVAL; | |
3914 | if (!d->cached) | |
3915 | return 0; | |
3916 | int left = d->size - offset; | |
3917 | total_len = left > size ? size: left; | |
3918 | memcpy(buf, d->buf + offset, total_len); | |
3919 | return total_len; | |
3920 | } | |
3921 | ||
3922 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
3923 | if (initpid <= 0) | |
3924 | initpid = fc->pid; | |
3925 | cg = get_pid_cgroup(initpid, "cpuset"); | |
3926 | if (!cg) | |
3927 | return read_file("/proc/stat", buf, size, d); | |
6d2f6996 | 3928 | prune_init_slice(cg); |
237e200e SH |
3929 | |
3930 | cpuset = get_cpuset(cg); | |
3931 | if (!cpuset) | |
3932 | goto err; | |
3933 | ||
8be92dd1 JS |
3934 | /* |
3935 | * Read cpuacct.usage_all for all CPUs. | |
3936 | * If the cpuacct cgroup is present, it is used to calculate the container's | |
3937 | * CPU usage. If not, values from the host's /proc/stat are used. | |
3938 | */ | |
3939 | if (read_cpuacct_usage_all(cg, cpuset, &cg_cpu_usage) != 0) { | |
3940 | lxcfs_debug("%s\n", "proc_stat_read failed to read from cpuacct, " | |
3941 | "falling back to the host's /proc/stat"); | |
3942 | } | |
3943 | ||
237e200e SH |
3944 | f = fopen("/proc/stat", "r"); |
3945 | if (!f) | |
3946 | goto err; | |
3947 | ||
3948 | //skip first line | |
3949 | if (getline(&line, &linelen, f) < 0) { | |
b8defc3d | 3950 | lxcfs_error("%s\n", "proc_stat_read read first line failed."); |
237e200e SH |
3951 | goto err; |
3952 | } | |
3953 | ||
3954 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 3955 | ssize_t l; |
237e200e SH |
3956 | int cpu; |
3957 | char cpu_char[10]; /* That's a lot of cores */ | |
3958 | char *c; | |
8be92dd1 JS |
3959 | uint64_t all_used, cg_used, new_idle; |
3960 | int ret; | |
237e200e | 3961 | |
b4665ce0 SH |
3962 | if (strlen(line) == 0) |
3963 | continue; | |
237e200e SH |
3964 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) { |
3965 | /* not a ^cpuN line containing a number N, just print it */ | |
9502bae2 | 3966 | l = snprintf(cache, cache_size, "%s", line); |
237e200e SH |
3967 | if (l < 0) { |
3968 | perror("Error writing to cache"); | |
3969 | rv = 0; | |
3970 | goto err; | |
3971 | } | |
3972 | if (l >= cache_size) { | |
b8defc3d | 3973 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
3974 | rv = 0; |
3975 | goto err; | |
3976 | } | |
3977 | cache += l; | |
3978 | cache_size -= l; | |
3979 | total_len += l; | |
3980 | continue; | |
3981 | } | |
3982 | ||
3983 | if (sscanf(cpu_char, "%d", &cpu) != 1) | |
3984 | continue; | |
3985 | if (!cpu_in_cpuset(cpu, cpuset)) | |
3986 | continue; | |
3987 | curcpu ++; | |
3988 | ||
8be92dd1 | 3989 | ret = sscanf(line, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", |
7144f069 CB |
3990 | &user, |
3991 | &nice, | |
3992 | &system, | |
3993 | &idle, | |
3994 | &iowait, | |
3995 | &irq, | |
3996 | &softirq, | |
3997 | &steal, | |
3998 | &guest, | |
8be92dd1 JS |
3999 | &guest_nice); |
4000 | ||
4001 | if (ret != 10 || !cg_cpu_usage) { | |
4002 | c = strchr(line, ' '); | |
4003 | if (!c) | |
4004 | continue; | |
4005 | l = snprintf(cache, cache_size, "cpu%d%s", curcpu, c); | |
4006 | if (l < 0) { | |
4007 | perror("Error writing to cache"); | |
4008 | rv = 0; | |
4009 | goto err; | |
4010 | ||
4011 | } | |
4012 | if (l >= cache_size) { | |
4013 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4014 | rv = 0; | |
4015 | goto err; | |
4016 | } | |
4017 | ||
4018 | cache += l; | |
4019 | cache_size -= l; | |
4020 | total_len += l; | |
4021 | ||
4022 | if (ret != 10) | |
4023 | continue; | |
4024 | } | |
4025 | ||
4026 | if (cg_cpu_usage) { | |
4027 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
4028 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
4029 | ||
4030 | if (all_used >= cg_used) { | |
4031 | new_idle = idle + (all_used - cg_used); | |
4032 | ||
4033 | } else { | |
4034 | lxcfs_error("cpu%d from %s has unexpected cpu time: %lu in /proc/stat, " | |
4035 | "%lu in cpuacct.usage_all; unable to determine idle time\n", | |
4036 | curcpu, cg, all_used, cg_used); | |
4037 | new_idle = idle; | |
4038 | } | |
4039 | ||
4040 | l = snprintf(cache, cache_size, "cpu%d %lu 0 %lu %lu 0 0 0 0 0 0\n", | |
4041 | curcpu, cg_cpu_usage[curcpu].user, cg_cpu_usage[curcpu].system, | |
4042 | new_idle); | |
4043 | ||
4044 | if (l < 0) { | |
4045 | perror("Error writing to cache"); | |
4046 | rv = 0; | |
4047 | goto err; | |
4048 | ||
4049 | } | |
4050 | if (l >= cache_size) { | |
4051 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); | |
4052 | rv = 0; | |
4053 | goto err; | |
4054 | } | |
4055 | ||
4056 | cache += l; | |
4057 | cache_size -= l; | |
4058 | total_len += l; | |
4059 | ||
4060 | user_sum += cg_cpu_usage[curcpu].user; | |
4061 | system_sum += cg_cpu_usage[curcpu].system; | |
4062 | idle_sum += new_idle; | |
4063 | ||
4064 | } else { | |
4065 | user_sum += user; | |
4066 | nice_sum += nice; | |
4067 | system_sum += system; | |
4068 | idle_sum += idle; | |
4069 | iowait_sum += iowait; | |
4070 | irq_sum += irq; | |
4071 | softirq_sum += softirq; | |
4072 | steal_sum += steal; | |
4073 | guest_sum += guest; | |
4074 | guest_nice_sum += guest_nice; | |
4075 | } | |
237e200e SH |
4076 | } |
4077 | ||
4078 | cache = d->buf; | |
4079 | ||
7144f069 CB |
4080 | int cpuall_len = snprintf(cpuall, CPUALL_MAX_SIZE, "cpu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", |
4081 | user_sum, | |
4082 | nice_sum, | |
4083 | system_sum, | |
4084 | idle_sum, | |
4085 | iowait_sum, | |
4086 | irq_sum, | |
4087 | softirq_sum, | |
4088 | steal_sum, | |
4089 | guest_sum, | |
4090 | guest_nice_sum); | |
4091 | if (cpuall_len > 0 && cpuall_len < CPUALL_MAX_SIZE) { | |
237e200e SH |
4092 | memcpy(cache, cpuall, cpuall_len); |
4093 | cache += cpuall_len; | |
7144f069 | 4094 | } else { |
237e200e | 4095 | /* shouldn't happen */ |
b8defc3d | 4096 | lxcfs_error("proc_stat_read copy cpuall failed, cpuall_len=%d.", cpuall_len); |
237e200e SH |
4097 | cpuall_len = 0; |
4098 | } | |
4099 | ||
4100 | memmove(cache, d->buf + CPUALL_MAX_SIZE, total_len); | |
4101 | total_len += cpuall_len; | |
4102 | d->cached = 1; | |
4103 | d->size = total_len; | |
7144f069 CB |
4104 | if (total_len > size) |
4105 | total_len = size; | |
237e200e SH |
4106 | |
4107 | memcpy(buf, d->buf, total_len); | |
4108 | rv = total_len; | |
4109 | ||
4110 | err: | |
4111 | if (f) | |
4112 | fclose(f); | |
8be92dd1 JS |
4113 | if (cg_cpu_usage) |
4114 | free(cg_cpu_usage); | |
237e200e SH |
4115 | free(line); |
4116 | free(cpuset); | |
4117 | free(cg); | |
4118 | return rv; | |
4119 | } | |
4120 | ||
0ecddf02 CB |
4121 | /* This function retrieves the busy time of a group of tasks by looking at |
4122 | * cpuacct.usage. Unfortunately, this only makes sense when the container has | |
4123 | * been given it's own cpuacct cgroup. If not, this function will take the busy | |
4124 | * time of all other taks that do not actually belong to the container into | |
4125 | * account as well. If someone has a clever solution for this please send a | |
4126 | * patch! | |
4127 | */ | |
237e200e SH |
4128 | static unsigned long get_reaper_busy(pid_t task) |
4129 | { | |
4130 | pid_t initpid = lookup_initpid_in_store(task); | |
4131 | char *cgroup = NULL, *usage_str = NULL; | |
4132 | unsigned long usage = 0; | |
4133 | ||
4134 | if (initpid <= 0) | |
4135 | return 0; | |
4136 | ||
4137 | cgroup = get_pid_cgroup(initpid, "cpuacct"); | |
4138 | if (!cgroup) | |
4139 | goto out; | |
6d2f6996 | 4140 | prune_init_slice(cgroup); |
237e200e SH |
4141 | if (!cgfs_get_value("cpuacct", cgroup, "cpuacct.usage", &usage_str)) |
4142 | goto out; | |
4143 | usage = strtoul(usage_str, NULL, 10); | |
4144 | usage /= 1000000000; | |
4145 | ||
4146 | out: | |
4147 | free(cgroup); | |
4148 | free(usage_str); | |
4149 | return usage; | |
4150 | } | |
4151 | ||
4152 | #if RELOADTEST | |
4153 | void iwashere(void) | |
4154 | { | |
237e200e SH |
4155 | int fd; |
4156 | ||
ec2b5e7c | 4157 | fd = creat("/tmp/lxcfs-iwashere", 0644); |
237e200e SH |
4158 | if (fd >= 0) |
4159 | close(fd); | |
4160 | } | |
4161 | #endif | |
4162 | ||
4163 | /* | |
4164 | * We read /proc/uptime and reuse its second field. | |
4165 | * For the first field, we use the mtime for the reaper for | |
4166 | * the calling pid as returned by getreaperage | |
4167 | */ | |
4168 | static int proc_uptime_read(char *buf, size_t size, off_t offset, | |
4169 | struct fuse_file_info *fi) | |
4170 | { | |
4171 | struct fuse_context *fc = fuse_get_context(); | |
4172 | struct file_info *d = (struct file_info *)fi->fh; | |
0ecddf02 | 4173 | unsigned long int busytime = get_reaper_busy(fc->pid); |
237e200e | 4174 | char *cache = d->buf; |
a262ddb7 | 4175 | ssize_t total_len = 0; |
0ecddf02 | 4176 | uint64_t idletime, reaperage; |
237e200e SH |
4177 | |
4178 | #if RELOADTEST | |
4179 | iwashere(); | |
4180 | #endif | |
4181 | ||
4182 | if (offset){ | |
237e200e SH |
4183 | if (!d->cached) |
4184 | return 0; | |
bbdf646b BM |
4185 | if (offset > d->size) |
4186 | return -EINVAL; | |
237e200e SH |
4187 | int left = d->size - offset; |
4188 | total_len = left > size ? size: left; | |
4189 | memcpy(buf, cache + offset, total_len); | |
4190 | return total_len; | |
4191 | } | |
4192 | ||
0ecddf02 CB |
4193 | reaperage = get_reaper_age(fc->pid); |
4194 | /* To understand why this is done, please read the comment to the | |
4195 | * get_reaper_busy() function. | |
4196 | */ | |
4197 | idletime = reaperage; | |
4198 | if (reaperage >= busytime) | |
4199 | idletime = reaperage - busytime; | |
237e200e | 4200 | |
bbdf646b BM |
4201 | total_len = snprintf(d->buf, d->buflen, "%"PRIu64".00 %"PRIu64".00\n", reaperage, idletime); |
4202 | if (total_len < 0 || total_len >= d->buflen){ | |
0ecddf02 | 4203 | lxcfs_error("%s\n", "failed to write to cache"); |
237e200e SH |
4204 | return 0; |
4205 | } | |
4206 | ||
4207 | d->size = (int)total_len; | |
4208 | d->cached = 1; | |
4209 | ||
4210 | if (total_len > size) total_len = size; | |
4211 | ||
4212 | memcpy(buf, d->buf, total_len); | |
4213 | return total_len; | |
4214 | } | |
4215 | ||
4216 | static int proc_diskstats_read(char *buf, size_t size, off_t offset, | |
4217 | struct fuse_file_info *fi) | |
4218 | { | |
4219 | char dev_name[72]; | |
4220 | struct fuse_context *fc = fuse_get_context(); | |
4221 | struct file_info *d = (struct file_info *)fi->fh; | |
4222 | char *cg; | |
4223 | char *io_serviced_str = NULL, *io_merged_str = NULL, *io_service_bytes_str = NULL, | |
4224 | *io_wait_time_str = NULL, *io_service_time_str = NULL; | |
4225 | unsigned long read = 0, write = 0; | |
4226 | unsigned long read_merged = 0, write_merged = 0; | |
4227 | unsigned long read_sectors = 0, write_sectors = 0; | |
4228 | unsigned long read_ticks = 0, write_ticks = 0; | |
4229 | unsigned long ios_pgr = 0, tot_ticks = 0, rq_ticks = 0; | |
4230 | unsigned long rd_svctm = 0, wr_svctm = 0, rd_wait = 0, wr_wait = 0; | |
4231 | char *cache = d->buf; | |
4232 | size_t cache_size = d->buflen; | |
4233 | char *line = NULL; | |
4234 | size_t linelen = 0, total_len = 0, rv = 0; | |
4235 | unsigned int major = 0, minor = 0; | |
4236 | int i = 0; | |
4237 | FILE *f = NULL; | |
4238 | ||
4239 | if (offset){ | |
4240 | if (offset > d->size) | |
4241 | return -EINVAL; | |
4242 | if (!d->cached) | |
4243 | return 0; | |
4244 | int left = d->size - offset; | |
4245 | total_len = left > size ? size: left; | |
4246 | memcpy(buf, cache + offset, total_len); | |
4247 | return total_len; | |
4248 | } | |
4249 | ||
4250 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
4251 | if (initpid <= 0) | |
4252 | initpid = fc->pid; | |
4253 | cg = get_pid_cgroup(initpid, "blkio"); | |
4254 | if (!cg) | |
4255 | return read_file("/proc/diskstats", buf, size, d); | |
6d2f6996 | 4256 | prune_init_slice(cg); |
237e200e | 4257 | |
2209fe50 | 4258 | if (!cgfs_get_value("blkio", cg, "blkio.io_serviced_recursive", &io_serviced_str)) |
237e200e | 4259 | goto err; |
2209fe50 | 4260 | if (!cgfs_get_value("blkio", cg, "blkio.io_merged_recursive", &io_merged_str)) |
237e200e | 4261 | goto err; |
2209fe50 | 4262 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_bytes_recursive", &io_service_bytes_str)) |
237e200e | 4263 | goto err; |
2209fe50 | 4264 | if (!cgfs_get_value("blkio", cg, "blkio.io_wait_time_recursive", &io_wait_time_str)) |
237e200e | 4265 | goto err; |
2209fe50 | 4266 | if (!cgfs_get_value("blkio", cg, "blkio.io_service_time_recursive", &io_service_time_str)) |
237e200e SH |
4267 | goto err; |
4268 | ||
4269 | ||
4270 | f = fopen("/proc/diskstats", "r"); | |
4271 | if (!f) | |
4272 | goto err; | |
4273 | ||
4274 | while (getline(&line, &linelen, f) != -1) { | |
a262ddb7 | 4275 | ssize_t l; |
2209fe50 | 4276 | char lbuf[256]; |
237e200e SH |
4277 | |
4278 | i = sscanf(line, "%u %u %71s", &major, &minor, dev_name); | |
2209fe50 | 4279 | if (i != 3) |
237e200e | 4280 | continue; |
2209fe50 SH |
4281 | |
4282 | get_blkio_io_value(io_serviced_str, major, minor, "Read", &read); | |
4283 | get_blkio_io_value(io_serviced_str, major, minor, "Write", &write); | |
4284 | get_blkio_io_value(io_merged_str, major, minor, "Read", &read_merged); | |
4285 | get_blkio_io_value(io_merged_str, major, minor, "Write", &write_merged); | |
4286 | get_blkio_io_value(io_service_bytes_str, major, minor, "Read", &read_sectors); | |
4287 | read_sectors = read_sectors/512; | |
4288 | get_blkio_io_value(io_service_bytes_str, major, minor, "Write", &write_sectors); | |
4289 | write_sectors = write_sectors/512; | |
4290 | ||
4291 | get_blkio_io_value(io_service_time_str, major, minor, "Read", &rd_svctm); | |
4292 | rd_svctm = rd_svctm/1000000; | |
4293 | get_blkio_io_value(io_wait_time_str, major, minor, "Read", &rd_wait); | |
4294 | rd_wait = rd_wait/1000000; | |
4295 | read_ticks = rd_svctm + rd_wait; | |
4296 | ||
4297 | get_blkio_io_value(io_service_time_str, major, minor, "Write", &wr_svctm); | |
4298 | wr_svctm = wr_svctm/1000000; | |
4299 | get_blkio_io_value(io_wait_time_str, major, minor, "Write", &wr_wait); | |
4300 | wr_wait = wr_wait/1000000; | |
4301 | write_ticks = wr_svctm + wr_wait; | |
4302 | ||
4303 | get_blkio_io_value(io_service_time_str, major, minor, "Total", &tot_ticks); | |
4304 | tot_ticks = tot_ticks/1000000; | |
237e200e SH |
4305 | |
4306 | memset(lbuf, 0, 256); | |
2db31eb6 SH |
4307 | if (read || write || read_merged || write_merged || read_sectors || write_sectors || read_ticks || write_ticks) |
4308 | snprintf(lbuf, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n", | |
4309 | major, minor, dev_name, read, read_merged, read_sectors, read_ticks, | |
4310 | write, write_merged, write_sectors, write_ticks, ios_pgr, tot_ticks, rq_ticks); | |
4311 | else | |
4312 | continue; | |
237e200e | 4313 | |
2209fe50 | 4314 | l = snprintf(cache, cache_size, "%s", lbuf); |
237e200e SH |
4315 | if (l < 0) { |
4316 | perror("Error writing to fuse buf"); | |
4317 | rv = 0; | |
4318 | goto err; | |
4319 | } | |
4320 | if (l >= cache_size) { | |
b8defc3d | 4321 | lxcfs_error("%s\n", "Internal error: truncated write to cache."); |
237e200e SH |
4322 | rv = 0; |
4323 | goto err; | |
4324 | } | |
4325 | cache += l; | |
4326 | cache_size -= l; | |
4327 | total_len += l; | |
4328 | } | |
4329 | ||
4330 | d->cached = 1; | |
4331 | d->size = total_len; | |
4332 | if (total_len > size ) total_len = size; | |
4333 | memcpy(buf, d->buf, total_len); | |
4334 | ||
4335 | rv = total_len; | |
4336 | err: | |
4337 | free(cg); | |
4338 | if (f) | |
4339 | fclose(f); | |
4340 | free(line); | |
4341 | free(io_serviced_str); | |
4342 | free(io_merged_str); | |
4343 | free(io_service_bytes_str); | |
4344 | free(io_wait_time_str); | |
4345 | free(io_service_time_str); | |
4346 | return rv; | |
4347 | } | |
4348 | ||
70dcc12e SH |
4349 | static int proc_swaps_read(char *buf, size_t size, off_t offset, |
4350 | struct fuse_file_info *fi) | |
4351 | { | |
4352 | struct fuse_context *fc = fuse_get_context(); | |
4353 | struct file_info *d = (struct file_info *)fi->fh; | |
4354 | char *cg = NULL; | |
018246ff | 4355 | char *memswlimit_str = NULL, *memlimit_str = NULL, *memusage_str = NULL, *memswusage_str = NULL; |
70dcc12e | 4356 | unsigned long memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0, swap_total = 0, swap_free = 0; |
a262ddb7 CB |
4357 | ssize_t total_len = 0, rv = 0; |
4358 | ssize_t l = 0; | |
70dcc12e SH |
4359 | char *cache = d->buf; |
4360 | ||
4361 | if (offset) { | |
4362 | if (offset > d->size) | |
4363 | return -EINVAL; | |
4364 | if (!d->cached) | |
4365 | return 0; | |
4366 | int left = d->size - offset; | |
4367 | total_len = left > size ? size: left; | |
4368 | memcpy(buf, cache + offset, total_len); | |
4369 | return total_len; | |
4370 | } | |
4371 | ||
4372 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
4373 | if (initpid <= 0) | |
4374 | initpid = fc->pid; | |
4375 | cg = get_pid_cgroup(initpid, "memory"); | |
4376 | if (!cg) | |
4377 | return read_file("/proc/swaps", buf, size, d); | |
6d2f6996 | 4378 | prune_init_slice(cg); |
70dcc12e | 4379 | |
018246ff | 4380 | memlimit = get_min_memlimit(cg, "memory.limit_in_bytes"); |
70dcc12e SH |
4381 | |
4382 | if (!cgfs_get_value("memory", cg, "memory.usage_in_bytes", &memusage_str)) | |
4383 | goto err; | |
4384 | ||
70dcc12e SH |
4385 | memusage = strtoul(memusage_str, NULL, 10); |
4386 | ||
4387 | if (cgfs_get_value("memory", cg, "memory.memsw.usage_in_bytes", &memswusage_str) && | |
4388 | cgfs_get_value("memory", cg, "memory.memsw.limit_in_bytes", &memswlimit_str)) { | |
4389 | ||
018246ff | 4390 | memswlimit = get_min_memlimit(cg, "memory.memsw.limit_in_bytes"); |
70dcc12e SH |
4391 | memswusage = strtoul(memswusage_str, NULL, 10); |
4392 | ||
70dcc12e SH |
4393 | swap_total = (memswlimit - memlimit) / 1024; |
4394 | swap_free = (memswusage - memusage) / 1024; | |
4395 | } | |
4396 | ||
4397 | total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n"); | |
4398 | ||
4399 | /* When no mem + swap limit is specified or swapaccount=0*/ | |
4400 | if (!memswlimit) { | |
4401 | char *line = NULL; | |
4402 | size_t linelen = 0; | |
4403 | FILE *f = fopen("/proc/meminfo", "r"); | |
4404 | ||
4405 | if (!f) | |
4406 | goto err; | |
4407 | ||
4408 | while (getline(&line, &linelen, f) != -1) { | |
4409 | if (startswith(line, "SwapTotal:")) { | |
4410 | sscanf(line, "SwapTotal: %8lu kB", &swap_total); | |
4411 | } else if (startswith(line, "SwapFree:")) { | |
4412 | sscanf(line, "SwapFree: %8lu kB", &swap_free); | |
4413 | } | |
4414 | } | |
4415 | ||
4416 | free(line); | |
4417 | fclose(f); | |
4418 | } | |
4419 | ||
4420 | if (swap_total > 0) { | |
a262ddb7 CB |
4421 | l = snprintf(d->buf + total_len, d->size - total_len, |
4422 | "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ", | |
4423 | swap_total, swap_free); | |
4424 | total_len += l; | |
70dcc12e SH |
4425 | } |
4426 | ||
a262ddb7 | 4427 | if (total_len < 0 || l < 0) { |
70dcc12e SH |
4428 | perror("Error writing to cache"); |
4429 | rv = 0; | |
4430 | goto err; | |
4431 | } | |
4432 | ||
4433 | d->cached = 1; | |
4434 | d->size = (int)total_len; | |
4435 | ||
4436 | if (total_len > size) total_len = size; | |
4437 | memcpy(buf, d->buf, total_len); | |
4438 | rv = total_len; | |
4439 | ||
4440 | err: | |
4441 | free(cg); | |
4442 | free(memswlimit_str); | |
4443 | free(memlimit_str); | |
4444 | free(memusage_str); | |
4445 | free(memswusage_str); | |
70dcc12e SH |
4446 | return rv; |
4447 | } | |
6db4f7a3 | 4448 | /* |
4449 | * Find the process pid from cgroup path. | |
4450 | * eg:from /sys/fs/cgroup/cpu/docker/containerid/cgroup.procs to find the process pid. | |
4451 | * @pid_buf : put pid to pid_buf. | |
4452 | * @dpath : the path of cgroup. eg: /docker/containerid or /docker/containerid/child-cgroup ... | |
4453 | * @depth : the depth of cgroup in container. | |
4454 | * @sum : return the number of pid. | |
4455 | * @cfd : the file descriptor of the mounted cgroup. eg: /sys/fs/cgroup/cpu | |
4456 | */ | |
4457 | static int calc_pid(char ***pid_buf, char *dpath, int depth, int sum, int cfd) | |
4458 | { | |
4459 | DIR *dir; | |
4460 | int fd; | |
4461 | struct dirent *file; | |
4462 | FILE *f = NULL; | |
4463 | size_t linelen = 0; | |
4464 | char *line = NULL; | |
4465 | int pd; | |
4466 | char *path_dir, *path; | |
4467 | char **pid; | |
4468 | ||
4469 | /* path = dpath + "/cgroup.procs" + /0 */ | |
4470 | do { | |
4471 | path = malloc(strlen(dpath) + 20); | |
4472 | } while (!path); | |
4473 | ||
4474 | strcpy(path, dpath); | |
4475 | fd = openat(cfd, path, O_RDONLY); | |
4476 | if (fd < 0) | |
4477 | goto out; | |
4478 | ||
4479 | dir = fdopendir(fd); | |
4480 | if (dir == NULL) { | |
4481 | close(fd); | |
4482 | goto out; | |
4483 | } | |
4484 | ||
4485 | while (((file = readdir(dir)) != NULL) && depth > 0) { | |
4486 | if (strncmp(file->d_name, ".", 1) == 0) | |
4487 | continue; | |
4488 | if (strncmp(file->d_name, "..", 1) == 0) | |
4489 | continue; | |
4490 | if (file->d_type == DT_DIR) { | |
4491 | /* path + '/' + d_name +/0 */ | |
4492 | do { | |
4493 | path_dir = malloc(strlen(path) + 2 + sizeof(file->d_name)); | |
4494 | } while (!path_dir); | |
4495 | strcpy(path_dir, path); | |
4496 | strcat(path_dir, "/"); | |
4497 | strcat(path_dir, file->d_name); | |
4498 | pd = depth - 1; | |
4499 | sum = calc_pid(pid_buf, path_dir, pd, sum, cfd); | |
4500 | free(path_dir); | |
4501 | } | |
4502 | } | |
4503 | closedir(dir); | |
4504 | ||
4505 | strcat(path, "/cgroup.procs"); | |
4506 | fd = openat(cfd, path, O_RDONLY); | |
4507 | if (fd < 0) | |
4508 | goto out; | |
4509 | ||
4510 | f = fdopen(fd, "r"); | |
4511 | if (!f) { | |
4512 | close(fd); | |
4513 | goto out; | |
4514 | } | |
4515 | ||
4516 | while (getline(&line, &linelen, f) != -1) { | |
4517 | do { | |
4518 | pid = realloc(*pid_buf, sizeof(char *) * (sum + 1)); | |
4519 | } while (!pid); | |
4520 | *pid_buf = pid; | |
4521 | do { | |
4522 | *(*pid_buf + sum) = malloc(strlen(line) + 1); | |
4523 | } while (*(*pid_buf + sum) == NULL); | |
4524 | strcpy(*(*pid_buf + sum), line); | |
4525 | sum++; | |
4526 | } | |
4527 | fclose(f); | |
4528 | out: | |
4529 | free(path); | |
4530 | return sum; | |
4531 | } | |
4532 | /* | |
4533 | * calc_load calculates the load according to the following formula: | |
4534 | * load1 = load0 * exp + active * (1 - exp) | |
4535 | * | |
4536 | * @load1: the new loadavg. | |
4537 | * @load0: the former loadavg. | |
4538 | * @active: the total number of running pid at this moment. | |
4539 | * @exp: the fixed-point defined in the beginning. | |
4540 | */ | |
4541 | static unsigned long | |
4542 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
4543 | { | |
4544 | unsigned long newload; | |
4545 | ||
4546 | active = active > 0 ? active * FIXED_1 : 0; | |
4547 | newload = load * exp + active * (FIXED_1 - exp); | |
4548 | if (active >= load) | |
4549 | newload += FIXED_1 - 1; | |
4550 | ||
4551 | return newload / FIXED_1; | |
4552 | } | |
4553 | ||
4554 | /* | |
4555 | * Return 0 means that container p->cg is closed. | |
4556 | * Return -1 means that error occurred in refresh. | |
4557 | * Positive num equals the total number of pid. | |
4558 | */ | |
4559 | static int refresh_load(struct load_node *p, char *path) | |
4560 | { | |
4561 | FILE *f = NULL; | |
4562 | char **idbuf; | |
4563 | char proc_path[256]; | |
4564 | int i, ret, run_pid = 0, total_pid = 0, last_pid = 0; | |
4565 | char *line = NULL; | |
4566 | size_t linelen = 0; | |
4567 | int sum, length; | |
4568 | DIR *dp; | |
4569 | struct dirent *file; | |
4570 | ||
4571 | do { | |
4572 | idbuf = malloc(sizeof(char *)); | |
4573 | } while (!idbuf); | |
4574 | sum = calc_pid(&idbuf, path, DEPTH_DIR, 0, p->cfd); | |
4575 | /* normal exit */ | |
4576 | if (sum == 0) | |
4577 | goto out; | |
4578 | ||
4579 | for (i = 0; i < sum; i++) { | |
4580 | /*clean up '\n' */ | |
4581 | length = strlen(idbuf[i])-1; | |
4582 | idbuf[i][length] = '\0'; | |
4583 | ret = snprintf(proc_path, 256, "/proc/%s/task", idbuf[i]); | |
4584 | if (ret < 0 || ret > 255) { | |
4585 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
4586 | i = sum; | |
4587 | sum = -1; | |
4588 | goto err_out; | |
4589 | } | |
4590 | ||
4591 | dp = opendir(proc_path); | |
4592 | if (!dp) { | |
4593 | lxcfs_error("%s\n", "Open proc_path failed in refresh_load."); | |
4594 | continue; | |
4595 | } | |
4596 | while ((file = readdir(dp)) != NULL) { | |
4597 | if (strncmp(file->d_name, ".", 1) == 0) | |
4598 | continue; | |
4599 | if (strncmp(file->d_name, "..", 1) == 0) | |
4600 | continue; | |
4601 | total_pid++; | |
4602 | /* We make the biggest pid become last_pid.*/ | |
4603 | ret = atof(file->d_name); | |
4604 | last_pid = (ret > last_pid) ? ret : last_pid; | |
4605 | ||
4606 | ret = snprintf(proc_path, 256, "/proc/%s/task/%s/status", idbuf[i], file->d_name); | |
4607 | if (ret < 0 || ret > 255) { | |
4608 | lxcfs_error("%s\n", "snprintf() failed in refresh_load."); | |
4609 | i = sum; | |
4610 | sum = -1; | |
4611 | closedir(dp); | |
4612 | goto err_out; | |
4613 | } | |
4614 | f = fopen(proc_path, "r"); | |
4615 | if (f != NULL) { | |
4616 | while (getline(&line, &linelen, f) != -1) { | |
4617 | /* Find State */ | |
4618 | if ((line[0] == 'S') && (line[1] == 't')) | |
4619 | break; | |
4620 | } | |
4621 | if ((line[7] == 'R') || (line[7] == 'D')) | |
4622 | run_pid++; | |
4623 | fclose(f); | |
4624 | } | |
4625 | } | |
4626 | closedir(dp); | |
4627 | } | |
4628 | /*Calculate the loadavg.*/ | |
4629 | p->avenrun[0] = calc_load(p->avenrun[0], EXP_1, run_pid); | |
4630 | p->avenrun[1] = calc_load(p->avenrun[1], EXP_5, run_pid); | |
4631 | p->avenrun[2] = calc_load(p->avenrun[2], EXP_15, run_pid); | |
4632 | p->run_pid = run_pid; | |
4633 | p->total_pid = total_pid; | |
4634 | p->last_pid = last_pid; | |
4635 | ||
4636 | free(line); | |
beb5024e | 4637 | err_out: |
6db4f7a3 | 4638 | for (; i > 0; i--) |
4639 | free(idbuf[i-1]); | |
4640 | out: | |
4641 | free(idbuf); | |
4642 | return sum; | |
4643 | } | |
4644 | /* | |
4645 | * Traverse the hash table and update it. | |
4646 | */ | |
4647 | void *load_begin(void *arg) | |
4648 | { | |
4649 | ||
4650 | char *path = NULL; | |
4651 | int i, sum, length, ret; | |
4652 | struct load_node *f; | |
4653 | int first_node; | |
4654 | clock_t time1, time2; | |
4655 | ||
4656 | while (1) { | |
a83618e2 JS |
4657 | if (loadavg_stop == 1) |
4658 | return NULL; | |
4659 | ||
6db4f7a3 | 4660 | time1 = clock(); |
4661 | for (i = 0; i < LOAD_SIZE; i++) { | |
4662 | pthread_mutex_lock(&load_hash[i].lock); | |
4663 | if (load_hash[i].next == NULL) { | |
4664 | pthread_mutex_unlock(&load_hash[i].lock); | |
4665 | continue; | |
4666 | } | |
4667 | f = load_hash[i].next; | |
4668 | first_node = 1; | |
4669 | while (f) { | |
4670 | length = strlen(f->cg) + 2; | |
4671 | do { | |
4672 | /* strlen(f->cg) + '.' or '' + \0 */ | |
4673 | path = malloc(length); | |
4674 | } while (!path); | |
4675 | ||
4676 | ret = snprintf(path, length, "%s%s", *(f->cg) == '/' ? "." : "", f->cg); | |
4677 | if (ret < 0 || ret > length - 1) { | |
4678 | /* snprintf failed, ignore the node.*/ | |
4679 | lxcfs_error("Refresh node %s failed for snprintf().\n", f->cg); | |
4680 | goto out; | |
4681 | } | |
4682 | sum = refresh_load(f, path); | |
4683 | if (sum == 0) { | |
4684 | f = del_node(f, i); | |
4685 | } else { | |
4686 | out: f = f->next; | |
4687 | } | |
4688 | free(path); | |
4689 | /* load_hash[i].lock locks only on the first node.*/ | |
4690 | if (first_node == 1) { | |
4691 | first_node = 0; | |
4692 | pthread_mutex_unlock(&load_hash[i].lock); | |
4693 | } | |
4694 | } | |
4695 | } | |
a83618e2 JS |
4696 | |
4697 | if (loadavg_stop == 1) | |
4698 | return NULL; | |
4699 | ||
6db4f7a3 | 4700 | time2 = clock(); |
4701 | usleep(FLUSH_TIME * 1000000 - (int)((time2 - time1) * 1000000 / CLOCKS_PER_SEC)); | |
4702 | } | |
4703 | } | |
4704 | ||
4705 | static int proc_loadavg_read(char *buf, size_t size, off_t offset, | |
4706 | struct fuse_file_info *fi) | |
4707 | { | |
4708 | struct fuse_context *fc = fuse_get_context(); | |
4709 | struct file_info *d = (struct file_info *)fi->fh; | |
4710 | pid_t initpid; | |
4711 | char *cg; | |
4712 | size_t total_len = 0; | |
4713 | char *cache = d->buf; | |
4714 | struct load_node *n; | |
4715 | int hash; | |
01d88ede | 4716 | int cfd, rv = 0; |
6db4f7a3 | 4717 | unsigned long a, b, c; |
4718 | ||
4719 | if (offset) { | |
4720 | if (offset > d->size) | |
4721 | return -EINVAL; | |
4722 | if (!d->cached) | |
4723 | return 0; | |
4724 | int left = d->size - offset; | |
4725 | total_len = left > size ? size : left; | |
4726 | memcpy(buf, cache + offset, total_len); | |
4727 | return total_len; | |
4728 | } | |
4729 | if (!loadavg) | |
4730 | return read_file("/proc/loadavg", buf, size, d); | |
4731 | ||
4732 | initpid = lookup_initpid_in_store(fc->pid); | |
4733 | if (initpid <= 0) | |
4734 | initpid = fc->pid; | |
4735 | cg = get_pid_cgroup(initpid, "cpu"); | |
4736 | if (!cg) | |
4737 | return read_file("/proc/loadavg", buf, size, d); | |
4738 | ||
4739 | prune_init_slice(cg); | |
4740 | hash = calc_hash(cg); | |
4741 | n = locate_node(cg, hash); | |
4742 | ||
4743 | /* First time */ | |
4744 | if (n == NULL) { | |
4745 | if (!find_mounted_controller("cpu", &cfd)) { | |
4746 | /* | |
4747 | * In locate_node() above, pthread_rwlock_unlock() isn't used | |
4748 | * because delete is not allowed before read has ended. | |
4749 | */ | |
4750 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
01d88ede JS |
4751 | rv = 0; |
4752 | goto err; | |
6db4f7a3 | 4753 | } |
4754 | do { | |
4755 | n = malloc(sizeof(struct load_node)); | |
4756 | } while (!n); | |
4757 | ||
4758 | do { | |
4759 | n->cg = malloc(strlen(cg)+1); | |
4760 | } while (!n->cg); | |
4761 | strcpy(n->cg, cg); | |
4762 | n->avenrun[0] = 0; | |
4763 | n->avenrun[1] = 0; | |
4764 | n->avenrun[2] = 0; | |
4765 | n->run_pid = 0; | |
4766 | n->total_pid = 1; | |
4767 | n->last_pid = initpid; | |
4768 | n->cfd = cfd; | |
4769 | insert_node(&n, hash); | |
4770 | } | |
4771 | a = n->avenrun[0] + (FIXED_1/200); | |
4772 | b = n->avenrun[1] + (FIXED_1/200); | |
4773 | c = n->avenrun[2] + (FIXED_1/200); | |
4774 | total_len = snprintf(d->buf, d->buflen, "%lu.%02lu %lu.%02lu %lu.%02lu %d/%d %d\n", | |
4775 | LOAD_INT(a), LOAD_FRAC(a), | |
4776 | LOAD_INT(b), LOAD_FRAC(b), | |
4777 | LOAD_INT(c), LOAD_FRAC(c), | |
4778 | n->run_pid, n->total_pid, n->last_pid); | |
4779 | pthread_rwlock_unlock(&load_hash[hash].rdlock); | |
4780 | if (total_len < 0 || total_len >= d->buflen) { | |
4781 | lxcfs_error("%s\n", "Failed to write to cache"); | |
01d88ede JS |
4782 | rv = 0; |
4783 | goto err; | |
6db4f7a3 | 4784 | } |
4785 | d->size = (int)total_len; | |
4786 | d->cached = 1; | |
4787 | ||
4788 | if (total_len > size) | |
4789 | total_len = size; | |
4790 | memcpy(buf, d->buf, total_len); | |
01d88ede JS |
4791 | rv = total_len; |
4792 | ||
4793 | err: | |
4794 | free(cg); | |
4795 | return rv; | |
6db4f7a3 | 4796 | } |
4797 | /* Return a positive number on success, return 0 on failure.*/ | |
4798 | pthread_t load_daemon(int load_use) | |
4799 | { | |
4800 | int ret; | |
4801 | pthread_t pid; | |
4802 | ||
4803 | ret = init_load(); | |
4804 | if (ret == -1) { | |
4805 | lxcfs_error("%s\n", "Initialize hash_table fails in load_daemon!"); | |
4806 | return 0; | |
4807 | } | |
4808 | ret = pthread_create(&pid, NULL, load_begin, NULL); | |
4809 | if (ret != 0) { | |
4810 | lxcfs_error("%s\n", "Create pthread fails in load_daemon!"); | |
4811 | load_free(); | |
4812 | return 0; | |
4813 | } | |
4814 | /* use loadavg, here loadavg = 1*/ | |
4815 | loadavg = load_use; | |
4816 | return pid; | |
4817 | } | |
70dcc12e | 4818 | |
a83618e2 JS |
4819 | /* Returns 0 on success. */ |
4820 | int stop_load_daemon(pthread_t pid) | |
4821 | { | |
4822 | int s; | |
4823 | ||
4824 | /* Signal the thread to gracefully stop */ | |
4825 | loadavg_stop = 1; | |
4826 | ||
4827 | s = pthread_join(pid, NULL); /* Make sure sub thread has been canceled. */ | |
4828 | if (s != 0) { | |
4829 | lxcfs_error("%s\n", "stop_load_daemon error: failed to join"); | |
4830 | return -1; | |
4831 | } | |
4832 | ||
4833 | load_free(); | |
4834 | loadavg_stop = 0; | |
4835 | ||
4836 | return 0; | |
4837 | } | |
4838 | ||
237e200e SH |
4839 | static off_t get_procfile_size(const char *which) |
4840 | { | |
4841 | FILE *f = fopen(which, "r"); | |
4842 | char *line = NULL; | |
4843 | size_t len = 0; | |
4844 | ssize_t sz, answer = 0; | |
4845 | if (!f) | |
4846 | return 0; | |
4847 | ||
4848 | while ((sz = getline(&line, &len, f)) != -1) | |
4849 | answer += sz; | |
4850 | fclose (f); | |
4851 | free(line); | |
4852 | ||
4853 | return answer; | |
4854 | } | |
4855 | ||
4856 | int proc_getattr(const char *path, struct stat *sb) | |
4857 | { | |
4858 | struct timespec now; | |
4859 | ||
4860 | memset(sb, 0, sizeof(struct stat)); | |
4861 | if (clock_gettime(CLOCK_REALTIME, &now) < 0) | |
4862 | return -EINVAL; | |
4863 | sb->st_uid = sb->st_gid = 0; | |
4864 | sb->st_atim = sb->st_mtim = sb->st_ctim = now; | |
4865 | if (strcmp(path, "/proc") == 0) { | |
4866 | sb->st_mode = S_IFDIR | 00555; | |
4867 | sb->st_nlink = 2; | |
4868 | return 0; | |
4869 | } | |
4870 | if (strcmp(path, "/proc/meminfo") == 0 || | |
4871 | strcmp(path, "/proc/cpuinfo") == 0 || | |
4872 | strcmp(path, "/proc/uptime") == 0 || | |
4873 | strcmp(path, "/proc/stat") == 0 || | |
70dcc12e | 4874 | strcmp(path, "/proc/diskstats") == 0 || |
46be8eed | 4875 | strcmp(path, "/proc/swaps") == 0 || |
4876 | strcmp(path, "/proc/loadavg") == 0) { | |
237e200e SH |
4877 | sb->st_size = 0; |
4878 | sb->st_mode = S_IFREG | 00444; | |
4879 | sb->st_nlink = 1; | |
4880 | return 0; | |
4881 | } | |
4882 | ||
4883 | return -ENOENT; | |
4884 | } | |
4885 | ||
4886 | int proc_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset, | |
4887 | struct fuse_file_info *fi) | |
4888 | { | |
d639f863 CB |
4889 | if (filler(buf, ".", NULL, 0) != 0 || |
4890 | filler(buf, "..", NULL, 0) != 0 || | |
4891 | filler(buf, "cpuinfo", NULL, 0) != 0 || | |
4892 | filler(buf, "meminfo", NULL, 0) != 0 || | |
4893 | filler(buf, "stat", NULL, 0) != 0 || | |
4894 | filler(buf, "uptime", NULL, 0) != 0 || | |
4895 | filler(buf, "diskstats", NULL, 0) != 0 || | |
46be8eed | 4896 | filler(buf, "swaps", NULL, 0) != 0 || |
4897 | filler(buf, "loadavg", NULL, 0) != 0) | |
237e200e SH |
4898 | return -EINVAL; |
4899 | return 0; | |
4900 | } | |
4901 | ||
4902 | int proc_open(const char *path, struct fuse_file_info *fi) | |
4903 | { | |
4904 | int type = -1; | |
4905 | struct file_info *info; | |
4906 | ||
4907 | if (strcmp(path, "/proc/meminfo") == 0) | |
4908 | type = LXC_TYPE_PROC_MEMINFO; | |
4909 | else if (strcmp(path, "/proc/cpuinfo") == 0) | |
4910 | type = LXC_TYPE_PROC_CPUINFO; | |
4911 | else if (strcmp(path, "/proc/uptime") == 0) | |
4912 | type = LXC_TYPE_PROC_UPTIME; | |
4913 | else if (strcmp(path, "/proc/stat") == 0) | |
4914 | type = LXC_TYPE_PROC_STAT; | |
4915 | else if (strcmp(path, "/proc/diskstats") == 0) | |
4916 | type = LXC_TYPE_PROC_DISKSTATS; | |
70dcc12e SH |
4917 | else if (strcmp(path, "/proc/swaps") == 0) |
4918 | type = LXC_TYPE_PROC_SWAPS; | |
46be8eed | 4919 | else if (strcmp(path, "/proc/loadavg") == 0) |
4920 | type = LXC_TYPE_PROC_LOADAVG; | |
237e200e SH |
4921 | if (type == -1) |
4922 | return -ENOENT; | |
4923 | ||
4924 | info = malloc(sizeof(*info)); | |
4925 | if (!info) | |
4926 | return -ENOMEM; | |
4927 | ||
4928 | memset(info, 0, sizeof(*info)); | |
4929 | info->type = type; | |
4930 | ||
4931 | info->buflen = get_procfile_size(path) + BUF_RESERVE_SIZE; | |
4932 | do { | |
4933 | info->buf = malloc(info->buflen); | |
4934 | } while (!info->buf); | |
4935 | memset(info->buf, 0, info->buflen); | |
4936 | /* set actual size to buffer size */ | |
4937 | info->size = info->buflen; | |
4938 | ||
4939 | fi->fh = (unsigned long)info; | |
4940 | return 0; | |
4941 | } | |
4942 | ||
bddbb106 SH |
4943 | int proc_access(const char *path, int mask) |
4944 | { | |
e7849aa3 CB |
4945 | if (strcmp(path, "/proc") == 0 && access(path, R_OK) == 0) |
4946 | return 0; | |
4947 | ||
bddbb106 SH |
4948 | /* these are all read-only */ |
4949 | if ((mask & ~R_OK) != 0) | |
1b060d0a | 4950 | return -EACCES; |
bddbb106 SH |
4951 | return 0; |
4952 | } | |
4953 | ||
237e200e SH |
4954 | int proc_release(const char *path, struct fuse_file_info *fi) |
4955 | { | |
43215927 | 4956 | do_release_file_info(fi); |
237e200e SH |
4957 | return 0; |
4958 | } | |
4959 | ||
4960 | int proc_read(const char *path, char *buf, size_t size, off_t offset, | |
4961 | struct fuse_file_info *fi) | |
4962 | { | |
4963 | struct file_info *f = (struct file_info *) fi->fh; | |
4964 | ||
4965 | switch (f->type) { | |
4966 | case LXC_TYPE_PROC_MEMINFO: | |
4967 | return proc_meminfo_read(buf, size, offset, fi); | |
4968 | case LXC_TYPE_PROC_CPUINFO: | |
4969 | return proc_cpuinfo_read(buf, size, offset, fi); | |
4970 | case LXC_TYPE_PROC_UPTIME: | |
4971 | return proc_uptime_read(buf, size, offset, fi); | |
4972 | case LXC_TYPE_PROC_STAT: | |
4973 | return proc_stat_read(buf, size, offset, fi); | |
4974 | case LXC_TYPE_PROC_DISKSTATS: | |
4975 | return proc_diskstats_read(buf, size, offset, fi); | |
70dcc12e SH |
4976 | case LXC_TYPE_PROC_SWAPS: |
4977 | return proc_swaps_read(buf, size, offset, fi); | |
46be8eed | 4978 | case LXC_TYPE_PROC_LOADAVG: |
4979 | return proc_loadavg_read(buf, size, offset, fi); | |
237e200e SH |
4980 | default: |
4981 | return -EINVAL; | |
4982 | } | |
4983 | } | |
4984 | ||
29a73c2f CB |
4985 | /* |
4986 | * Functions needed to setup cgroups in the __constructor__. | |
29a73c2f CB |
4987 | */ |
4988 | ||
4989 | static bool mkdir_p(const char *dir, mode_t mode) | |
4990 | { | |
4991 | const char *tmp = dir; | |
4992 | const char *orig = dir; | |
4993 | char *makeme; | |
4994 | ||
4995 | do { | |
4996 | dir = tmp + strspn(tmp, "/"); | |
4997 | tmp = dir + strcspn(dir, "/"); | |
4998 | makeme = strndup(orig, dir - orig); | |
4999 | if (!makeme) | |
5000 | return false; | |
5001 | if (mkdir(makeme, mode) && errno != EEXIST) { | |
b8defc3d | 5002 | lxcfs_error("Failed to create directory '%s': %s.\n", |
29a73c2f CB |
5003 | makeme, strerror(errno)); |
5004 | free(makeme); | |
5005 | return false; | |
5006 | } | |
5007 | free(makeme); | |
5008 | } while(tmp != dir); | |
5009 | ||
5010 | return true; | |
5011 | } | |
5012 | ||
5013 | static bool umount_if_mounted(void) | |
5014 | { | |
5015 | if (umount2(BASEDIR, MNT_DETACH) < 0 && errno != EINVAL) { | |
b8defc3d | 5016 | lxcfs_error("Failed to unmount %s: %s.\n", BASEDIR, strerror(errno)); |
29a73c2f CB |
5017 | return false; |
5018 | } | |
5019 | return true; | |
5020 | } | |
5021 | ||
2283e240 CB |
5022 | /* __typeof__ should be safe to use with all compilers. */ |
5023 | typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic; | |
5024 | static bool has_fs_type(const struct statfs *fs, fs_type_magic magic_val) | |
5025 | { | |
5026 | return (fs->f_type == (fs_type_magic)magic_val); | |
5027 | } | |
5028 | ||
0a4dea41 CB |
5029 | /* |
5030 | * looking at fs/proc_namespace.c, it appears we can | |
5031 | * actually expect the rootfs entry to very specifically contain | |
5032 | * " - rootfs rootfs " | |
5033 | * IIUC, so long as we've chrooted so that rootfs is not our root, | |
5034 | * the rootfs entry should always be skipped in mountinfo contents. | |
5035 | */ | |
5036 | static bool is_on_ramfs(void) | |
5037 | { | |
5038 | FILE *f; | |
5039 | char *p, *p2; | |
5040 | char *line = NULL; | |
5041 | size_t len = 0; | |
5042 | int i; | |
5043 | ||
5044 | f = fopen("/proc/self/mountinfo", "r"); | |
5045 | if (!f) | |
5046 | return false; | |
5047 | ||
5048 | while (getline(&line, &len, f) != -1) { | |
5049 | for (p = line, i = 0; p && i < 4; i++) | |
5050 | p = strchr(p + 1, ' '); | |
5051 | if (!p) | |
5052 | continue; | |
5053 | p2 = strchr(p + 1, ' '); | |
5054 | if (!p2) | |
5055 | continue; | |
5056 | *p2 = '\0'; | |
5057 | if (strcmp(p + 1, "/") == 0) { | |
5058 | // this is '/'. is it the ramfs? | |
5059 | p = strchr(p2 + 1, '-'); | |
5060 | if (p && strncmp(p, "- rootfs rootfs ", 16) == 0) { | |
5061 | free(line); | |
5062 | fclose(f); | |
5063 | return true; | |
5064 | } | |
5065 | } | |
5066 | } | |
5067 | free(line); | |
5068 | fclose(f); | |
5069 | return false; | |
5070 | } | |
5071 | ||
cc309f33 | 5072 | static int pivot_enter() |
0a4dea41 | 5073 | { |
cc309f33 CB |
5074 | int ret = -1, oldroot = -1, newroot = -1; |
5075 | ||
5076 | oldroot = open("/", O_DIRECTORY | O_RDONLY); | |
5077 | if (oldroot < 0) { | |
5078 | lxcfs_error("%s\n", "Failed to open old root for fchdir."); | |
5079 | return ret; | |
5080 | } | |
5081 | ||
5082 | newroot = open(ROOTDIR, O_DIRECTORY | O_RDONLY); | |
5083 | if (newroot < 0) { | |
5084 | lxcfs_error("%s\n", "Failed to open new root for fchdir."); | |
5085 | goto err; | |
5086 | } | |
5087 | ||
5088 | /* change into new root fs */ | |
5089 | if (fchdir(newroot) < 0) { | |
5090 | lxcfs_error("Failed to change directory to new rootfs: %s.\n", ROOTDIR); | |
5091 | goto err; | |
5092 | } | |
5093 | ||
0a4dea41 CB |
5094 | /* pivot_root into our new root fs */ |
5095 | if (pivot_root(".", ".") < 0) { | |
5096 | lxcfs_error("pivot_root() syscall failed: %s.\n", strerror(errno)); | |
cc309f33 | 5097 | goto err; |
0a4dea41 CB |
5098 | } |
5099 | ||
5100 | /* | |
5101 | * At this point the old-root is mounted on top of our new-root. | |
5102 | * To unmounted it we must not be chdir'd into it, so escape back | |
5103 | * to the old-root. | |
5104 | */ | |
5105 | if (fchdir(oldroot) < 0) { | |
5106 | lxcfs_error("%s\n", "Failed to enter old root."); | |
cc309f33 | 5107 | goto err; |
0a4dea41 CB |
5108 | } |
5109 | ||
5110 | if (umount2(".", MNT_DETACH) < 0) { | |
5111 | lxcfs_error("%s\n", "Failed to detach old root."); | |
cc309f33 | 5112 | goto err; |
0a4dea41 CB |
5113 | } |
5114 | ||
5115 | if (fchdir(newroot) < 0) { | |
5116 | lxcfs_error("%s\n", "Failed to re-enter new root."); | |
cc309f33 | 5117 | goto err; |
0a4dea41 CB |
5118 | } |
5119 | ||
cc309f33 CB |
5120 | ret = 0; |
5121 | ||
5122 | err: | |
5123 | if (oldroot > 0) | |
5124 | close(oldroot); | |
5125 | if (newroot > 0) | |
5126 | close(newroot); | |
5127 | ||
5128 | return ret; | |
0a4dea41 CB |
5129 | } |
5130 | ||
5131 | static int chroot_enter() | |
5132 | { | |
5133 | if (mount(ROOTDIR, "/", NULL, MS_REC | MS_BIND, NULL)) { | |
5134 | lxcfs_error("Failed to recursively bind-mount %s into /.", ROOTDIR); | |
5135 | return -1; | |
5136 | } | |
5137 | ||
5138 | if (chroot(".") < 0) { | |
5139 | lxcfs_error("Call to chroot() failed: %s.\n", strerror(errno)); | |
5140 | return -1; | |
5141 | } | |
5142 | ||
5143 | if (chdir("/") < 0) { | |
5144 | lxcfs_error("Failed to change directory: %s.\n", strerror(errno)); | |
5145 | return -1; | |
5146 | } | |
5147 | ||
5148 | return 0; | |
5149 | } | |
5150 | ||
0232cbac | 5151 | static int permute_and_enter(void) |
29a73c2f | 5152 | { |
0a4dea41 CB |
5153 | struct statfs sb; |
5154 | ||
5155 | if (statfs("/", &sb) < 0) { | |
5156 | lxcfs_error("%s\n", "Could not stat / mountpoint."); | |
cc309f33 | 5157 | return -1; |
0a4dea41 CB |
5158 | } |
5159 | ||
5160 | /* has_fs_type() is not reliable. When the ramfs is a tmpfs it will | |
5161 | * likely report TMPFS_MAGIC. Hence, when it reports no we still check | |
5162 | * /proc/1/mountinfo. */ | |
5163 | if (has_fs_type(&sb, RAMFS_MAGIC) || is_on_ramfs()) | |
5164 | return chroot_enter(); | |
29a73c2f | 5165 | |
cc309f33 | 5166 | if (pivot_enter() < 0) { |
0a4dea41 | 5167 | lxcfs_error("%s\n", "Could not perform pivot root."); |
cc309f33 | 5168 | return -1; |
29a73c2f CB |
5169 | } |
5170 | ||
cc309f33 | 5171 | return 0; |
29a73c2f CB |
5172 | } |
5173 | ||
5174 | /* Prepare our new clean root. */ | |
0232cbac | 5175 | static int permute_prepare(void) |
29a73c2f CB |
5176 | { |
5177 | if (mkdir(ROOTDIR, 0700) < 0 && errno != EEXIST) { | |
b8defc3d | 5178 | lxcfs_error("%s\n", "Failed to create directory for new root."); |
29a73c2f CB |
5179 | return -1; |
5180 | } | |
5181 | ||
5182 | if (mount("/", ROOTDIR, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5183 | lxcfs_error("Failed to bind-mount / for new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5184 | return -1; |
5185 | } | |
5186 | ||
5187 | if (mount(RUNTIME_PATH, ROOTDIR RUNTIME_PATH, NULL, MS_BIND, 0) < 0) { | |
b8defc3d | 5188 | lxcfs_error("Failed to bind-mount /run into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5189 | return -1; |
5190 | } | |
5191 | ||
5192 | if (mount(BASEDIR, ROOTDIR BASEDIR, NULL, MS_REC | MS_MOVE, 0) < 0) { | |
b8defc3d | 5193 | printf("Failed to move " BASEDIR " into new root: %s.\n", strerror(errno)); |
29a73c2f CB |
5194 | return -1; |
5195 | } | |
5196 | ||
5197 | return 0; | |
5198 | } | |
5199 | ||
0232cbac CB |
5200 | /* Calls chroot() on ramfs, pivot_root() in all other cases. */ |
5201 | static bool permute_root(void) | |
29a73c2f CB |
5202 | { |
5203 | /* Prepare new root. */ | |
0232cbac | 5204 | if (permute_prepare() < 0) |
29a73c2f CB |
5205 | return false; |
5206 | ||
5207 | /* Pivot into new root. */ | |
0232cbac | 5208 | if (permute_and_enter() < 0) |
29a73c2f CB |
5209 | return false; |
5210 | ||
5211 | return true; | |
5212 | } | |
5213 | ||
a257a8ee CB |
5214 | static int preserve_mnt_ns(int pid) |
5215 | { | |
5216 | int ret; | |
5217 | size_t len = sizeof("/proc/") + 21 + sizeof("/ns/mnt"); | |
5218 | char path[len]; | |
5219 | ||
5220 | ret = snprintf(path, len, "/proc/%d/ns/mnt", pid); | |
5221 | if (ret < 0 || (size_t)ret >= len) | |
5222 | return -1; | |
5223 | ||
5224 | return open(path, O_RDONLY | O_CLOEXEC); | |
5225 | } | |
5226 | ||
0a4dea41 | 5227 | static bool cgfs_prepare_mounts(void) |
29a73c2f CB |
5228 | { |
5229 | if (!mkdir_p(BASEDIR, 0700)) { | |
b8defc3d | 5230 | lxcfs_error("%s\n", "Failed to create lxcfs cgroup mountpoint."); |
29a73c2f CB |
5231 | return false; |
5232 | } | |
480262c9 | 5233 | |
29a73c2f | 5234 | if (!umount_if_mounted()) { |
b8defc3d | 5235 | lxcfs_error("%s\n", "Failed to clean up old lxcfs cgroup mountpoint."); |
480262c9 CB |
5236 | return false; |
5237 | } | |
5238 | ||
5239 | if (unshare(CLONE_NEWNS) < 0) { | |
b8defc3d | 5240 | lxcfs_error("Failed to unshare mount namespace: %s.\n", strerror(errno)); |
480262c9 CB |
5241 | return false; |
5242 | } | |
5243 | ||
a257a8ee CB |
5244 | cgroup_mount_ns_fd = preserve_mnt_ns(getpid()); |
5245 | if (cgroup_mount_ns_fd < 0) { | |
5246 | lxcfs_error("Failed to preserve mount namespace: %s.\n", strerror(errno)); | |
5247 | return false; | |
5248 | } | |
5249 | ||
480262c9 | 5250 | if (mount(NULL, "/", NULL, MS_REC | MS_PRIVATE, 0) < 0) { |
b8defc3d | 5251 | lxcfs_error("Failed to remount / private: %s.\n", strerror(errno)); |
29a73c2f CB |
5252 | return false; |
5253 | } | |
480262c9 | 5254 | |
29a73c2f | 5255 | if (mount("tmpfs", BASEDIR, "tmpfs", 0, "size=100000,mode=700") < 0) { |
b8defc3d | 5256 | lxcfs_error("%s\n", "Failed to mount tmpfs over lxcfs cgroup mountpoint."); |
29a73c2f CB |
5257 | return false; |
5258 | } | |
480262c9 | 5259 | |
29a73c2f CB |
5260 | return true; |
5261 | } | |
5262 | ||
0a4dea41 | 5263 | static bool cgfs_mount_hierarchies(void) |
29a73c2f CB |
5264 | { |
5265 | char *target; | |
5266 | size_t clen, len; | |
5267 | int i, ret; | |
5268 | ||
5269 | for (i = 0; i < num_hierarchies; i++) { | |
5270 | char *controller = hierarchies[i]; | |
51c7ca35 | 5271 | |
29a73c2f CB |
5272 | clen = strlen(controller); |
5273 | len = strlen(BASEDIR) + clen + 2; | |
5274 | target = malloc(len); | |
5275 | if (!target) | |
5276 | return false; | |
51c7ca35 | 5277 | |
29a73c2f CB |
5278 | ret = snprintf(target, len, "%s/%s", BASEDIR, controller); |
5279 | if (ret < 0 || ret >= len) { | |
5280 | free(target); | |
5281 | return false; | |
5282 | } | |
5283 | if (mkdir(target, 0755) < 0 && errno != EEXIST) { | |
5284 | free(target); | |
5285 | return false; | |
5286 | } | |
51c7ca35 CB |
5287 | if (!strcmp(controller, "unified")) |
5288 | ret = mount("none", target, "cgroup2", 0, NULL); | |
5289 | else | |
5290 | ret = mount(controller, target, "cgroup", 0, controller); | |
5291 | if (ret < 0) { | |
5292 | lxcfs_error("Failed mounting cgroup %s: %s\n", controller, strerror(errno)); | |
29a73c2f CB |
5293 | free(target); |
5294 | return false; | |
5295 | } | |
5296 | ||
5297 | fd_hierarchies[i] = open(target, O_DIRECTORY); | |
5298 | if (fd_hierarchies[i] < 0) { | |
5299 | free(target); | |
5300 | return false; | |
5301 | } | |
5302 | free(target); | |
5303 | } | |
5304 | return true; | |
5305 | } | |
5306 | ||
480262c9 | 5307 | static bool cgfs_setup_controllers(void) |
29a73c2f | 5308 | { |
0a4dea41 | 5309 | if (!cgfs_prepare_mounts()) |
29a73c2f | 5310 | return false; |
29a73c2f | 5311 | |
0a4dea41 | 5312 | if (!cgfs_mount_hierarchies()) { |
b8defc3d | 5313 | lxcfs_error("%s\n", "Failed to set up private lxcfs cgroup mounts."); |
29a73c2f CB |
5314 | return false; |
5315 | } | |
5316 | ||
0232cbac | 5317 | if (!permute_root()) |
29a73c2f CB |
5318 | return false; |
5319 | ||
5320 | return true; | |
5321 | } | |
5322 | ||
5323 | static void __attribute__((constructor)) collect_and_mount_subsystems(void) | |
237e200e SH |
5324 | { |
5325 | FILE *f; | |
e58dab00 CB |
5326 | char *cret, *line = NULL; |
5327 | char cwd[MAXPATHLEN]; | |
237e200e | 5328 | size_t len = 0; |
480262c9 | 5329 | int i, init_ns = -1; |
51c7ca35 | 5330 | bool found_unified = false; |
237e200e SH |
5331 | |
5332 | if ((f = fopen("/proc/self/cgroup", "r")) == NULL) { | |
b8defc3d | 5333 | lxcfs_error("Error opening /proc/self/cgroup: %s\n", strerror(errno)); |
237e200e SH |
5334 | return; |
5335 | } | |
e58dab00 | 5336 | |
237e200e | 5337 | while (getline(&line, &len, f) != -1) { |
51c7ca35 | 5338 | char *idx, *p, *p2; |
237e200e SH |
5339 | |
5340 | p = strchr(line, ':'); | |
5341 | if (!p) | |
5342 | goto out; | |
51c7ca35 | 5343 | idx = line; |
237e200e SH |
5344 | *(p++) = '\0'; |
5345 | ||
5346 | p2 = strrchr(p, ':'); | |
5347 | if (!p2) | |
5348 | goto out; | |
5349 | *p2 = '\0'; | |
5350 | ||
a67719f6 CB |
5351 | /* With cgroupv2 /proc/self/cgroup can contain entries of the |
5352 | * form: 0::/ This will cause lxcfs to fail the cgroup mounts | |
5353 | * because it parses out the empty string "" and later on passes | |
5354 | * it to mount(). Let's skip such entries. | |
5355 | */ | |
51c7ca35 CB |
5356 | if (!strcmp(p, "") && !strcmp(idx, "0") && !found_unified) { |
5357 | found_unified = true; | |
5358 | p = "unified"; | |
5359 | } | |
a67719f6 | 5360 | |
237e200e SH |
5361 | if (!store_hierarchy(line, p)) |
5362 | goto out; | |
5363 | } | |
5364 | ||
480262c9 | 5365 | /* Preserve initial namespace. */ |
a257a8ee | 5366 | init_ns = preserve_mnt_ns(getpid()); |
b8defc3d CB |
5367 | if (init_ns < 0) { |
5368 | lxcfs_error("%s\n", "Failed to preserve initial mount namespace."); | |
480262c9 | 5369 | goto out; |
b8defc3d | 5370 | } |
480262c9 | 5371 | |
92c3ee11 | 5372 | fd_hierarchies = malloc(sizeof(int) * num_hierarchies); |
b8defc3d CB |
5373 | if (!fd_hierarchies) { |
5374 | lxcfs_error("%s\n", strerror(errno)); | |
29a73c2f | 5375 | goto out; |
b8defc3d | 5376 | } |
29a73c2f | 5377 | |
480262c9 CB |
5378 | for (i = 0; i < num_hierarchies; i++) |
5379 | fd_hierarchies[i] = -1; | |
5380 | ||
e58dab00 CB |
5381 | cret = getcwd(cwd, MAXPATHLEN); |
5382 | if (!cret) | |
5383 | lxcfs_debug("Could not retrieve current working directory: %s.\n", strerror(errno)); | |
5384 | ||
480262c9 CB |
5385 | /* This function calls unshare(CLONE_NEWNS) our initial mount namespace |
5386 | * to privately mount lxcfs cgroups. */ | |
b8defc3d CB |
5387 | if (!cgfs_setup_controllers()) { |
5388 | lxcfs_error("%s\n", "Failed to setup private cgroup mounts for lxcfs."); | |
29a73c2f | 5389 | goto out; |
b8defc3d | 5390 | } |
480262c9 | 5391 | |
b8defc3d CB |
5392 | if (setns(init_ns, 0) < 0) { |
5393 | lxcfs_error("Failed to switch back to initial mount namespace: %s.\n", strerror(errno)); | |
29a73c2f | 5394 | goto out; |
b8defc3d | 5395 | } |
29a73c2f | 5396 | |
e58dab00 CB |
5397 | if (!cret || chdir(cwd) < 0) |
5398 | lxcfs_debug("Could not change back to original working directory: %s.\n", strerror(errno)); | |
5399 | ||
237e200e SH |
5400 | print_subsystems(); |
5401 | ||
5402 | out: | |
5403 | free(line); | |
5404 | fclose(f); | |
480262c9 CB |
5405 | if (init_ns >= 0) |
5406 | close(init_ns); | |
237e200e SH |
5407 | } |
5408 | ||
5409 | static void __attribute__((destructor)) free_subsystems(void) | |
5410 | { | |
5411 | int i; | |
5412 | ||
b8defc3d CB |
5413 | lxcfs_debug("%s\n", "Running destructor for liblxcfs."); |
5414 | ||
29a73c2f | 5415 | for (i = 0; i < num_hierarchies; i++) { |
237e200e SH |
5416 | if (hierarchies[i]) |
5417 | free(hierarchies[i]); | |
480262c9 | 5418 | if (fd_hierarchies && fd_hierarchies[i] >= 0) |
29a73c2f CB |
5419 | close(fd_hierarchies[i]); |
5420 | } | |
237e200e | 5421 | free(hierarchies); |
480262c9 | 5422 | free(fd_hierarchies); |
a257a8ee CB |
5423 | |
5424 | if (cgroup_mount_ns_fd >= 0) | |
5425 | close(cgroup_mount_ns_fd); | |
237e200e | 5426 | } |