]>
Commit | Line | Data |
---|---|---|
db0463bf | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
1f5596dd CB |
2 | |
3 | #ifndef _GNU_SOURCE | |
4 | #define _GNU_SOURCE | |
5 | #endif | |
6 | ||
7 | #ifndef FUSE_USE_VERSION | |
8 | #define FUSE_USE_VERSION 26 | |
9 | #endif | |
10 | ||
11 | #define _FILE_OFFSET_BITS 64 | |
12 | ||
13 | #define __STDC_FORMAT_MACROS | |
14 | #include <dirent.h> | |
15 | #include <errno.h> | |
16 | #include <fcntl.h> | |
17 | #include <fuse.h> | |
18 | #include <inttypes.h> | |
19 | #include <libgen.h> | |
20 | #include <pthread.h> | |
21 | #include <sched.h> | |
22 | #include <stdarg.h> | |
23 | #include <stdbool.h> | |
24 | #include <stdint.h> | |
25 | #include <stdio.h> | |
26 | #include <stdlib.h> | |
27 | #include <string.h> | |
28 | #include <time.h> | |
29 | #include <unistd.h> | |
30 | #include <wait.h> | |
31 | #include <linux/magic.h> | |
32 | #include <linux/sched.h> | |
33 | #include <sys/epoll.h> | |
34 | #include <sys/mman.h> | |
35 | #include <sys/mount.h> | |
36 | #include <sys/param.h> | |
37 | #include <sys/socket.h> | |
38 | #include <sys/syscall.h> | |
39 | #include <sys/sysinfo.h> | |
40 | #include <sys/vfs.h> | |
41 | ||
42 | #include "bindings.h" | |
43 | #include "config.h" | |
44 | #include "cgroup_fuse.h" | |
45 | #include "cpuset_parse.h" | |
46 | #include "cgroups/cgroup.h" | |
47 | #include "cgroups/cgroup_utils.h" | |
48 | #include "memory_utils.h" | |
4ec5c9da | 49 | #include "proc_loadavg.h" |
1f5596dd CB |
50 | #include "utils.h" |
51 | ||
1f5596dd CB |
52 | /* Data for CPU view */ |
53 | struct cg_proc_stat { | |
54 | char *cg; | |
ce617d73 CB |
55 | struct cpuacct_usage *usage; /* Real usage as read from the host's /proc/stat. */ |
56 | struct cpuacct_usage *view; /* Usage stats reported to the container. */ | |
1f5596dd | 57 | int cpu_count; |
ce617d73 | 58 | pthread_mutex_t lock; /* For node manipulation. */ |
1f5596dd CB |
59 | struct cg_proc_stat *next; |
60 | }; | |
61 | ||
62 | struct cg_proc_stat_head { | |
63 | struct cg_proc_stat *next; | |
64 | time_t lastcheck; | |
65 | ||
66 | /* | |
67 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
68 | */ | |
69 | pthread_rwlock_t lock; | |
70 | }; | |
71 | ||
72 | #define CPUVIEW_HASH_SIZE 100 | |
73 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
74 | ||
b456d40d CB |
75 | static void reset_proc_stat_node(struct cg_proc_stat *node, |
76 | struct cpuacct_usage *usage, int cpu_count) | |
1f5596dd | 77 | { |
1f5596dd CB |
78 | lxcfs_debug("Resetting stat node for %s\n", node->cg); |
79 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
80 | ||
b456d40d | 81 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
82 | node->view[i].user = 0; |
83 | node->view[i].system = 0; | |
84 | node->view[i].idle = 0; | |
85 | } | |
86 | ||
87 | node->cpu_count = cpu_count; | |
88 | } | |
89 | ||
90 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) | |
91 | { | |
92 | __do_free struct cpuacct_usage *new_usage = NULL, *new_view = NULL; | |
93 | ||
94 | /* Allocate new memory */ | |
82d74a95 | 95 | new_usage = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
96 | if (!new_usage) |
97 | return false; | |
98 | ||
82d74a95 | 99 | new_view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
100 | if (!new_view) |
101 | return false; | |
102 | ||
103 | /* Copy existing data & initialize new elements */ | |
104 | for (int i = 0; i < cpu_count; i++) { | |
105 | if (i < node->cpu_count) { | |
82d74a95 CB |
106 | new_usage[i].user = node->usage[i].user; |
107 | new_usage[i].system = node->usage[i].system; | |
108 | new_usage[i].idle = node->usage[i].idle; | |
109 | ||
110 | new_view[i].user = node->view[i].user; | |
111 | new_view[i].system = node->view[i].system; | |
112 | new_view[i].idle = node->view[i].idle; | |
1f5596dd CB |
113 | } |
114 | } | |
115 | ||
116 | free(node->usage); | |
117 | node->usage = move_ptr(new_usage); | |
118 | ||
119 | free(node->view); | |
120 | node->view = move_ptr(new_view); | |
121 | node->cpu_count = cpu_count; | |
122 | ||
123 | return true; | |
124 | } | |
125 | ||
4ec5c9da CB |
126 | static void free_proc_stat_node(struct cg_proc_stat *node) |
127 | { | |
6a4dceb1 CB |
128 | if (node) { |
129 | /* | |
130 | * We're abusing the usage pointer to indicate that | |
131 | * pthread_mutex_init() was successful. Don't judge me. | |
132 | */ | |
133 | if (node->usage) | |
134 | pthread_mutex_destroy(&node->lock); | |
135 | free_disarm(node->cg); | |
136 | free_disarm(node->usage); | |
137 | free_disarm(node->view); | |
138 | free_disarm(node); | |
139 | } | |
4ec5c9da CB |
140 | } |
141 | ||
6a4dceb1 CB |
142 | define_cleanup_function(struct cg_proc_stat *, free_proc_stat_node); |
143 | ||
1f5596dd CB |
144 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
145 | { | |
0d129671 CB |
146 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *new = new_node; |
147 | struct cg_proc_stat *rv = new_node; | |
148 | int hash = calc_hash(new->cg) % CPUVIEW_HASH_SIZE; | |
1f5596dd | 149 | struct cg_proc_stat_head *head = proc_stat_history[hash]; |
0d129671 | 150 | struct cg_proc_stat *cur; |
1f5596dd CB |
151 | |
152 | pthread_rwlock_wrlock(&head->lock); | |
153 | ||
154 | if (!head->next) { | |
0d129671 | 155 | head->next = move_ptr(new); |
164acda7 | 156 | goto out_rwlock_unlock; |
1f5596dd CB |
157 | } |
158 | ||
0d129671 | 159 | cur = head->next; |
1f5596dd CB |
160 | |
161 | for (;;) { | |
0d129671 CB |
162 | /* |
163 | * The node to be added is already present in the list, so | |
164 | * free the newly allocated one and return the one we found. | |
165 | */ | |
166 | if (strcmp(cur->cg, new->cg) == 0) { | |
167 | rv = cur; | |
164acda7 | 168 | goto out_rwlock_unlock; |
1f5596dd CB |
169 | } |
170 | ||
0d129671 CB |
171 | /* Keep walking. */ |
172 | if (cur->next) { | |
173 | cur = cur->next; | |
1f5596dd CB |
174 | continue; |
175 | } | |
176 | ||
0d129671 CB |
177 | /* Add new node to end of list. */ |
178 | cur->next = move_ptr(new); | |
164acda7 | 179 | goto out_rwlock_unlock; |
1f5596dd CB |
180 | } |
181 | ||
164acda7 | 182 | out_rwlock_unlock: |
1f5596dd | 183 | pthread_rwlock_unlock(&head->lock); |
0d129671 | 184 | return move_ptr(rv); |
1f5596dd CB |
185 | } |
186 | ||
6a4dceb1 CB |
187 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, |
188 | int cpu_count, const char *cg) | |
1f5596dd | 189 | { |
6a4dceb1 CB |
190 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *node = NULL; |
191 | __do_free struct cpuacct_usage *new_usage = NULL; | |
1f5596dd | 192 | |
6a4dceb1 | 193 | node = zalloc(sizeof(struct cg_proc_stat)); |
1f5596dd | 194 | if (!node) |
6a4dceb1 | 195 | return NULL; |
1f5596dd | 196 | |
6a4dceb1 | 197 | node->cg = strdup(cg); |
1f5596dd | 198 | if (!node->cg) |
6a4dceb1 | 199 | return NULL; |
1f5596dd | 200 | |
6a4dceb1 CB |
201 | new_usage = memdup(usage, sizeof(struct cpuacct_usage) * cpu_count); |
202 | if (!new_usage) | |
203 | return NULL; | |
1f5596dd | 204 | |
6a4dceb1 | 205 | node->view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd | 206 | if (!node->view) |
6a4dceb1 | 207 | return NULL; |
1f5596dd CB |
208 | |
209 | node->cpu_count = cpu_count; | |
1f5596dd | 210 | |
6a4dceb1 CB |
211 | if (pthread_mutex_init(&node->lock, NULL)) |
212 | return NULL; | |
213 | /* | |
214 | * We're abusing the usage pointer to indicate that | |
215 | * pthread_mutex_init() was successful. Don't judge me. | |
216 | */ | |
217 | node->usage = move_ptr(new_usage); | |
1f5596dd | 218 | |
6a4dceb1 | 219 | return move_ptr(node); |
1f5596dd CB |
220 | } |
221 | ||
2d00d04c CB |
222 | static bool cgroup_supports(const char *controller, const char *cgroup, |
223 | const char *file) | |
4ec5c9da | 224 | { |
2c990b1d CB |
225 | __do_free char *path = NULL; |
226 | int cfd; | |
4ec5c9da CB |
227 | |
228 | cfd = get_cgroup_fd(controller); | |
229 | if (cfd < 0) | |
230 | return false; | |
231 | ||
925d5849 | 232 | path = must_make_path_relative(cgroup, file, NULL); |
2d00d04c | 233 | return faccessat(cfd, path, F_OK, 0) == 0; |
4ec5c9da CB |
234 | } |
235 | ||
1f5596dd CB |
236 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
237 | { | |
b456d40d | 238 | struct cg_proc_stat *first = NULL; |
1f5596dd | 239 | |
b456d40d | 240 | for (struct cg_proc_stat *prev = NULL; node; ) { |
2d00d04c CB |
241 | if (!cgroup_supports("cpu", node->cg, "cpu.shares")) { |
242 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *cur = node; | |
1f5596dd CB |
243 | |
244 | if (prev) | |
245 | prev->next = node->next; | |
246 | else | |
247 | first = node->next; | |
248 | ||
249 | node = node->next; | |
2d00d04c | 250 | lxcfs_debug("Removing stat node for %s\n", cur->cg); |
1f5596dd CB |
251 | } else { |
252 | if (!first) | |
253 | first = node; | |
254 | prev = node; | |
255 | node = node->next; | |
256 | } | |
257 | } | |
258 | ||
259 | return first; | |
260 | } | |
261 | ||
262 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
263 | static void prune_proc_stat_history(void) | |
264 | { | |
1f5596dd CB |
265 | time_t now = time(NULL); |
266 | ||
b456d40d | 267 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) { |
1f5596dd CB |
268 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
269 | ||
270 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
271 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
272 | return; | |
273 | } | |
274 | ||
275 | if (proc_stat_history[i]->next) { | |
276 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
277 | proc_stat_history[i]->lastcheck = now; | |
278 | } | |
279 | ||
280 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
281 | } | |
282 | } | |
283 | ||
284 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, | |
285 | const char *cg) | |
286 | { | |
287 | struct cg_proc_stat *node; | |
288 | ||
289 | pthread_rwlock_rdlock(&head->lock); | |
290 | ||
291 | if (!head->next) { | |
292 | pthread_rwlock_unlock(&head->lock); | |
293 | return NULL; | |
294 | } | |
295 | ||
296 | node = head->next; | |
297 | ||
298 | do { | |
299 | if (strcmp(cg, node->cg) == 0) | |
300 | goto out; | |
301 | } while ((node = node->next)); | |
302 | ||
303 | node = NULL; | |
304 | ||
305 | out: | |
306 | pthread_rwlock_unlock(&head->lock); | |
307 | prune_proc_stat_history(); | |
308 | return node; | |
309 | } | |
310 | ||
311 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) | |
312 | { | |
313 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
314 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
315 | struct cg_proc_stat *node; | |
316 | ||
317 | node = find_proc_stat_node(head, cg); | |
1f5596dd CB |
318 | if (!node) { |
319 | node = new_proc_stat_node(usage, cpu_count, cg); | |
320 | if (!node) | |
321 | return NULL; | |
322 | ||
323 | node = add_proc_stat_node(node); | |
324 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
325 | } | |
326 | ||
327 | pthread_mutex_lock(&node->lock); | |
328 | ||
ce089f10 CB |
329 | /* |
330 | * If additional CPUs on the host have been enabled, CPU usage counter | |
331 | * arrays have to be expanded. | |
332 | */ | |
1f5596dd CB |
333 | if (node->cpu_count < cpu_count) { |
334 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
ce089f10 | 335 | node->cpu_count, cpu_count, cg); |
1f5596dd CB |
336 | |
337 | if (!expand_proc_stat_node(node, cpu_count)) { | |
338 | pthread_mutex_unlock(&node->lock); | |
b456d40d | 339 | return log_debug(NULL, "Unable to expand stat node %d->%d for %s", node->cpu_count, cpu_count, cg); |
1f5596dd CB |
340 | } |
341 | } | |
342 | ||
343 | return node; | |
344 | } | |
345 | ||
2b8eff1d CB |
346 | static void add_cpu_usage(uint64_t *surplus, struct cpuacct_usage *usage, |
347 | uint64_t *counter, uint64_t threshold) | |
1f5596dd | 348 | { |
1ba088ae | 349 | uint64_t free_space, to_add; |
1f5596dd CB |
350 | |
351 | free_space = threshold - usage->user - usage->system; | |
352 | ||
353 | if (free_space > usage->idle) | |
354 | free_space = usage->idle; | |
355 | ||
8206874a CB |
356 | if (free_space > *surplus) |
357 | to_add = *surplus; | |
358 | else | |
359 | to_add = free_space; | |
1f5596dd CB |
360 | |
361 | *counter += to_add; | |
362 | usage->idle -= to_add; | |
363 | *surplus -= to_add; | |
364 | } | |
365 | ||
1ba088ae CB |
366 | static uint64_t diff_cpu_usage(struct cpuacct_usage *older, |
367 | struct cpuacct_usage *newer, | |
368 | struct cpuacct_usage *diff, int cpu_count) | |
1f5596dd | 369 | { |
1ba088ae | 370 | uint64_t sum = 0; |
1f5596dd | 371 | |
b456d40d | 372 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
373 | if (!newer[i].online) |
374 | continue; | |
375 | ||
b456d40d CB |
376 | /* |
377 | * When cpuset is changed on the fly, the CPUs might get | |
378 | * reordered. We could either reset all counters, or check | |
379 | * that the substractions below will return expected results. | |
1f5596dd CB |
380 | */ |
381 | if (newer[i].user > older[i].user) | |
382 | diff[i].user = newer[i].user - older[i].user; | |
383 | else | |
384 | diff[i].user = 0; | |
385 | ||
386 | if (newer[i].system > older[i].system) | |
387 | diff[i].system = newer[i].system - older[i].system; | |
388 | else | |
389 | diff[i].system = 0; | |
390 | ||
391 | if (newer[i].idle > older[i].idle) | |
392 | diff[i].idle = newer[i].idle - older[i].idle; | |
393 | else | |
394 | diff[i].idle = 0; | |
395 | ||
396 | sum += diff[i].user; | |
397 | sum += diff[i].system; | |
398 | sum += diff[i].idle; | |
399 | } | |
400 | ||
401 | return sum; | |
402 | } | |
403 | ||
404 | /* | |
b456d40d CB |
405 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or |
406 | * `cpu.cfs_period_us`, depending on `param`. Parameter value is returned | |
407 | * throuh `value`. | |
1f5596dd CB |
408 | */ |
409 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
410 | { | |
411 | __do_free char *str = NULL; | |
48f6862e | 412 | char file[STRLITERALLEN("cpu.cfs_period_us") + 1]; |
9844eea7 | 413 | bool first = true; |
48f6862e | 414 | int ret; |
1f5596dd | 415 | |
48f6862e | 416 | if (pure_unified_layout(cgroup_ops)) { |
9844eea7 | 417 | first = !strcmp(param, "quota"); |
48f6862e CB |
418 | ret = snprintf(file, sizeof(file), "cpu.max"); |
419 | } else { | |
420 | ret = snprintf(file, sizeof(file), "cpu.cfs_%s_us", param); | |
9844eea7 | 421 | } |
48f6862e | 422 | if (ret < 0 || (size_t)ret >= sizeof(file)) |
1f5596dd CB |
423 | return false; |
424 | ||
48f6862e | 425 | if (!cgroup_ops->get(cgroup_ops, "cpu", cg, file, &str)) |
1f5596dd CB |
426 | return false; |
427 | ||
48f6862e | 428 | return sscanf(str, first ? "%" PRId64 : "%*d %" PRId64, value) == 1; |
1f5596dd CB |
429 | } |
430 | ||
431 | /* | |
432 | * Return the exact number of visible CPUs based on CPU quotas. | |
433 | * If there is no quota set, zero is returned. | |
434 | */ | |
435 | static double exact_cpu_count(const char *cg) | |
436 | { | |
437 | double rv; | |
438 | int nprocs; | |
439 | int64_t cfs_quota, cfs_period; | |
440 | ||
9844eea7 JC |
441 | read_cpu_cfs_param(cg, "quota", &cfs_quota); |
442 | read_cpu_cfs_param(cg, "period", &cfs_period); | |
1f5596dd CB |
443 | |
444 | if (cfs_quota <= 0 || cfs_period <= 0) | |
445 | return 0; | |
446 | ||
447 | rv = (double)cfs_quota / (double)cfs_period; | |
448 | ||
449 | nprocs = get_nprocs(); | |
450 | ||
451 | if (rv > nprocs) | |
452 | rv = nprocs; | |
453 | ||
454 | return rv; | |
455 | } | |
456 | ||
457 | /* | |
458 | * Return the maximum number of visible CPUs based on CPU quotas. | |
459 | * If there is no quota set, zero is returned. | |
460 | */ | |
4ec5c9da | 461 | int max_cpu_count(const char *cg) |
1f5596dd | 462 | { |
700dd417 | 463 | __do_free char *cpuset = NULL; |
1f5596dd CB |
464 | int rv, nprocs; |
465 | int64_t cfs_quota, cfs_period; | |
466 | int nr_cpus_in_cpuset = 0; | |
1f5596dd | 467 | |
9844eea7 JC |
468 | read_cpu_cfs_param(cg, "quota", &cfs_quota); |
469 | read_cpu_cfs_param(cg, "period", &cfs_period); | |
1f5596dd CB |
470 | |
471 | cpuset = get_cpuset(cg); | |
472 | if (cpuset) | |
473 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
474 | ||
475 | if (cfs_quota <= 0 || cfs_period <= 0){ | |
476 | if (nr_cpus_in_cpuset > 0) | |
477 | return nr_cpus_in_cpuset; | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | rv = cfs_quota / cfs_period; | |
483 | ||
484 | /* In case quota/period does not yield a whole number, add one CPU for | |
485 | * the remainder. | |
486 | */ | |
487 | if ((cfs_quota % cfs_period) > 0) | |
488 | rv += 1; | |
489 | ||
490 | nprocs = get_nprocs(); | |
1f5596dd CB |
491 | if (rv > nprocs) |
492 | rv = nprocs; | |
493 | ||
494 | /* use min value in cpu quota and cpuset */ | |
495 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) | |
496 | rv = nr_cpus_in_cpuset; | |
497 | ||
498 | return rv; | |
499 | } | |
500 | ||
501 | int cpuview_proc_stat(const char *cg, const char *cpuset, | |
502 | struct cpuacct_usage *cg_cpu_usage, int cg_cpu_usage_size, | |
503 | FILE *f, char *buf, size_t buf_size) | |
504 | { | |
505 | __do_free char *line = NULL; | |
506 | __do_free struct cpuacct_usage *diff = NULL; | |
4f18a602 | 507 | size_t linelen = 0, total_len = 0; |
1f5596dd CB |
508 | int curcpu = -1; /* cpu numbering starts at 0 */ |
509 | int physcpu, i; | |
39f231da | 510 | int cpu_cnt = 0; |
2b8eff1d CB |
511 | uint64_t user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, |
512 | softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
513 | uint64_t user_sum = 0, system_sum = 0, idle_sum = 0; | |
514 | uint64_t user_surplus = 0, system_surplus = 0; | |
39f231da | 515 | int nprocs, max_cpus; |
4f18a602 | 516 | ssize_t l; |
2b8eff1d | 517 | uint64_t total_sum, threshold; |
1f5596dd | 518 | struct cg_proc_stat *stat_node; |
1f5596dd | 519 | |
39f231da | 520 | nprocs = get_nprocs_conf(); |
1f5596dd CB |
521 | if (cg_cpu_usage_size < nprocs) |
522 | nprocs = cg_cpu_usage_size; | |
523 | ||
524 | /* Read all CPU stats and stop when we've encountered other lines */ | |
525 | while (getline(&line, &linelen, f) != -1) { | |
526 | int ret; | |
527 | char cpu_char[10]; /* That's a lot of cores */ | |
528 | uint64_t all_used, cg_used; | |
529 | ||
530 | if (strlen(line) == 0) | |
531 | continue; | |
532 | ||
533 | /* not a ^cpuN line containing a number N */ | |
534 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) | |
535 | break; | |
536 | ||
537 | if (sscanf(cpu_char, "%d", &physcpu) != 1) | |
538 | continue; | |
539 | ||
540 | if (physcpu >= cg_cpu_usage_size) | |
541 | continue; | |
542 | ||
fd65c77c CB |
543 | curcpu++; |
544 | cpu_cnt++; | |
1f5596dd CB |
545 | |
546 | if (!cpu_in_cpuset(physcpu, cpuset)) { | |
547 | for (i = curcpu; i <= physcpu; i++) | |
548 | cg_cpu_usage[i].online = false; | |
549 | continue; | |
550 | } | |
551 | ||
552 | if (curcpu < physcpu) { | |
553 | /* Some CPUs may be disabled */ | |
554 | for (i = curcpu; i < physcpu; i++) | |
555 | cg_cpu_usage[i].online = false; | |
556 | ||
557 | curcpu = physcpu; | |
558 | } | |
559 | ||
560 | cg_cpu_usage[curcpu].online = true; | |
561 | ||
2b8eff1d | 562 | ret = sscanf(line, "%*s %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 "lu", |
1f5596dd CB |
563 | &user, |
564 | &nice, | |
565 | &system, | |
566 | &idle, | |
567 | &iowait, | |
568 | &irq, | |
569 | &softirq, | |
570 | &steal, | |
571 | &guest, | |
572 | &guest_nice); | |
1f5596dd CB |
573 | if (ret != 10) |
574 | continue; | |
575 | ||
576 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
577 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
578 | ||
579 | if (all_used >= cg_used) { | |
580 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
581 | ||
582 | } else { | |
2b8eff1d CB |
583 | lxcfs_error("cpu%d from %s has unexpected cpu time: %" PRIu64 " in /proc/stat, %" PRIu64 " in cpuacct.usage_all; unable to determine idle time", |
584 | curcpu, cg, all_used, cg_used); | |
1f5596dd CB |
585 | cg_cpu_usage[curcpu].idle = idle; |
586 | } | |
587 | } | |
588 | ||
f9434b9a CB |
589 | /* Cannot use more CPUs than is available in cpuset. */ |
590 | max_cpus = max_cpu_count(cg); | |
591 | if (max_cpus > cpu_cnt || !max_cpus) | |
592 | max_cpus = cpu_cnt; | |
593 | ||
1f5596dd | 594 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
b456d40d CB |
595 | if (!stat_node) |
596 | return log_error(0, "Failed to find/create stat node for %s", cg); | |
1f5596dd CB |
597 | |
598 | diff = malloc(sizeof(struct cpuacct_usage) * nprocs); | |
700dd417 | 599 | if (!diff) |
1f5596dd | 600 | return 0; |
1f5596dd CB |
601 | |
602 | /* | |
603 | * If the new values are LOWER than values stored in memory, it means | |
604 | * the cgroup has been reset/recreated and we should reset too. | |
605 | */ | |
606 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
607 | if (!cg_cpu_usage[curcpu].online) | |
608 | continue; | |
609 | ||
610 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
611 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
612 | ||
613 | break; | |
614 | } | |
615 | ||
616 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); | |
617 | ||
618 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
619 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
620 | ||
621 | if (!stat_node->usage[curcpu].online) | |
622 | continue; | |
623 | ||
624 | i++; | |
625 | ||
626 | stat_node->usage[curcpu].user += diff[curcpu].user; | |
627 | stat_node->usage[curcpu].system += diff[curcpu].system; | |
628 | stat_node->usage[curcpu].idle += diff[curcpu].idle; | |
629 | ||
630 | if (max_cpus > 0 && i >= max_cpus) { | |
631 | user_surplus += diff[curcpu].user; | |
632 | system_surplus += diff[curcpu].system; | |
633 | } | |
634 | } | |
635 | ||
636 | /* Calculate usage counters of visible CPUs */ | |
637 | if (max_cpus > 0) { | |
2b8eff1d CB |
638 | uint64_t diff_user = 0; |
639 | uint64_t diff_system = 0; | |
640 | uint64_t diff_idle = 0; | |
641 | uint64_t max_diff_idle = 0; | |
642 | uint64_t max_diff_idle_index = 0; | |
1f5596dd CB |
643 | double exact_cpus; |
644 | ||
645 | /* threshold = maximum usage per cpu, including idle */ | |
646 | threshold = total_sum / cpu_cnt * max_cpus; | |
647 | ||
648 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
649 | if (!stat_node->usage[curcpu].online) | |
650 | continue; | |
651 | ||
652 | i++; | |
653 | ||
654 | if (i == max_cpus) | |
655 | break; | |
656 | ||
657 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
658 | continue; | |
659 | ||
660 | /* Add user */ | |
661 | add_cpu_usage(&user_surplus, &diff[curcpu], | |
662 | &diff[curcpu].user, threshold); | |
663 | ||
664 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
665 | continue; | |
666 | ||
667 | /* If there is still room, add system */ | |
668 | add_cpu_usage(&system_surplus, &diff[curcpu], | |
669 | &diff[curcpu].system, threshold); | |
670 | } | |
671 | ||
672 | if (user_surplus > 0) | |
673 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
674 | if (system_surplus > 0) | |
675 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
676 | ||
677 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
678 | if (!stat_node->usage[curcpu].online) | |
679 | continue; | |
680 | ||
681 | i++; | |
682 | ||
683 | if (i == max_cpus) | |
684 | break; | |
685 | ||
686 | stat_node->view[curcpu].user += diff[curcpu].user; | |
687 | stat_node->view[curcpu].system += diff[curcpu].system; | |
688 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
689 | ||
690 | user_sum += stat_node->view[curcpu].user; | |
691 | system_sum += stat_node->view[curcpu].system; | |
692 | idle_sum += stat_node->view[curcpu].idle; | |
693 | ||
694 | diff_user += diff[curcpu].user; | |
695 | diff_system += diff[curcpu].system; | |
696 | diff_idle += diff[curcpu].idle; | |
697 | if (diff[curcpu].idle > max_diff_idle) { | |
698 | max_diff_idle = diff[curcpu].idle; | |
699 | max_diff_idle_index = curcpu; | |
700 | } | |
701 | ||
702 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
703 | } | |
704 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
705 | ||
706 | /* revise cpu usage view to support partial cpu case. */ | |
707 | exact_cpus = exact_cpu_count(cg); | |
708 | if (exact_cpus < (double)max_cpus){ | |
1ba088ae | 709 | uint64_t delta = (uint64_t)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); |
1f5596dd CB |
710 | |
711 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
712 | lxcfs_v("delta: %lu\n", delta); | |
713 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
714 | idle_sum = idle_sum > delta ? idle_sum - delta : 0; | |
715 | lxcfs_v("idle_sum after: %lu\n", idle_sum); | |
716 | ||
717 | curcpu = max_diff_idle_index; | |
718 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
719 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle > delta ? stat_node->view[curcpu].idle - delta : 0; | |
720 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
721 | } | |
722 | } else { | |
723 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
724 | if (!stat_node->usage[curcpu].online) | |
725 | continue; | |
726 | ||
727 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; | |
728 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
729 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
730 | ||
731 | user_sum += stat_node->view[curcpu].user; | |
732 | system_sum += stat_node->view[curcpu].system; | |
733 | idle_sum += stat_node->view[curcpu].idle; | |
734 | } | |
735 | } | |
736 | ||
737 | /* Render the file */ | |
738 | /* cpu-all */ | |
2b8eff1d CB |
739 | l = snprintf(buf, buf_size, |
740 | "cpu %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", | |
741 | user_sum, system_sum, idle_sum); | |
1f5596dd | 742 | lxcfs_v("cpu-all: %s\n", buf); |
b456d40d CB |
743 | if (l < 0) |
744 | return log_error(0, "Failed to write cache"); | |
745 | if (l >= buf_size) | |
746 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
747 | |
748 | buf += l; | |
749 | buf_size -= l; | |
750 | total_len += l; | |
751 | ||
752 | /* Render visible CPUs */ | |
753 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
754 | if (!stat_node->usage[curcpu].online) | |
755 | continue; | |
756 | ||
757 | i++; | |
758 | ||
759 | if (max_cpus > 0 && i == max_cpus) | |
760 | break; | |
761 | ||
2b8eff1d CB |
762 | l = snprintf(buf, buf_size, "cpu%d %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", |
763 | i, | |
764 | stat_node->view[curcpu].user, | |
765 | stat_node->view[curcpu].system, | |
766 | stat_node->view[curcpu].idle); | |
1f5596dd | 767 | lxcfs_v("cpu: %s\n", buf); |
b456d40d CB |
768 | if (l < 0) |
769 | return log_error(0, "Failed to write cache"); | |
770 | if (l >= buf_size) | |
771 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
772 | |
773 | buf += l; | |
774 | buf_size -= l; | |
775 | total_len += l; | |
776 | } | |
777 | ||
778 | /* Pass the rest of /proc/stat, start with the last line read */ | |
779 | l = snprintf(buf, buf_size, "%s", line); | |
b456d40d CB |
780 | if (l < 0) |
781 | return log_error(0, "Failed to write cache"); | |
782 | if (l >= buf_size) | |
783 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
784 | |
785 | buf += l; | |
786 | buf_size -= l; | |
787 | total_len += l; | |
788 | ||
789 | /* Pass the rest of the host's /proc/stat */ | |
790 | while (getline(&line, &linelen, f) != -1) { | |
791 | l = snprintf(buf, buf_size, "%s", line); | |
b456d40d CB |
792 | if (l < 0) |
793 | return log_error(0, "Failed to write cache"); | |
794 | if (l >= buf_size) | |
795 | return log_error(0, "Write to cache was truncated"); | |
796 | ||
1f5596dd CB |
797 | buf += l; |
798 | buf_size -= l; | |
799 | total_len += l; | |
800 | } | |
801 | ||
802 | if (stat_node) | |
803 | pthread_mutex_unlock(&stat_node->lock); | |
b456d40d | 804 | |
1f5596dd CB |
805 | return total_len; |
806 | } | |
807 | ||
808 | /* | |
809 | * check whether this is a '^processor" line in /proc/cpuinfo | |
810 | */ | |
b456d40d | 811 | static inline bool is_processor_line(const char *line) |
1f5596dd CB |
812 | { |
813 | int cpu; | |
b456d40d | 814 | return sscanf(line, "processor : %d", &cpu) == 1; |
1f5596dd CB |
815 | } |
816 | ||
b456d40d | 817 | static inline bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1f5596dd CB |
818 | { |
819 | int cpu; | |
c539526c CB |
820 | |
821 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
822 | return cpu_in_cpuset(cpu, cpuset); | |
823 | ||
824 | return false; | |
1f5596dd CB |
825 | } |
826 | ||
827 | int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
828 | struct fuse_file_info *fi) | |
829 | { | |
830 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; | |
757a63e7 | 831 | __do_free void *fopen_cache = NULL; |
1f5596dd CB |
832 | __do_fclose FILE *f = NULL; |
833 | struct fuse_context *fc = fuse_get_context(); | |
0274438c | 834 | struct lxcfs_opts *opts = (struct lxcfs_opts *)fc->private_data; |
99b183fb | 835 | struct file_info *d = INTTYPE_TO_PTR(fi->fh); |
1f5596dd CB |
836 | size_t linelen = 0, total_len = 0; |
837 | bool am_printing = false, firstline = true, is_s390x = false; | |
838 | int curcpu = -1, cpu, max_cpus = 0; | |
839 | bool use_view; | |
840 | char *cache = d->buf; | |
841 | size_t cache_size = d->buflen; | |
842 | ||
f9434b9a | 843 | if (offset) { |
1f5596dd CB |
844 | int left; |
845 | ||
846 | if (offset > d->size) | |
847 | return -EINVAL; | |
848 | ||
849 | if (!d->cached) | |
850 | return 0; | |
851 | ||
852 | left = d->size - offset; | |
853 | total_len = left > size ? size: left; | |
854 | memcpy(buf, cache + offset, total_len); | |
855 | ||
856 | return total_len; | |
857 | } | |
858 | ||
859 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
860 | if (initpid <= 1 || is_shared_pidns(initpid)) | |
861 | initpid = fc->pid; | |
b456d40d | 862 | |
1f5596dd CB |
863 | cg = get_pid_cgroup(initpid, "cpuset"); |
864 | if (!cg) | |
865 | return read_file_fuse("proc/cpuinfo", buf, size, d); | |
866 | prune_init_slice(cg); | |
867 | ||
868 | cpuset = get_cpuset(cg); | |
869 | if (!cpuset) | |
870 | return 0; | |
871 | ||
8044f626 | 872 | if (cgroup_ops->can_use_cpuview(cgroup_ops) && opts && opts->use_cfs) |
0274438c | 873 | use_view = true; |
8044f626 CB |
874 | else |
875 | use_view = false; | |
1f5596dd CB |
876 | if (use_view) |
877 | max_cpus = max_cpu_count(cg); | |
878 | ||
757a63e7 | 879 | f = fopen_cached("/proc/cpuinfo", "re", &fopen_cache); |
1f5596dd CB |
880 | if (!f) |
881 | return 0; | |
882 | ||
883 | while (getline(&line, &linelen, f) != -1) { | |
884 | ssize_t l; | |
885 | if (firstline) { | |
886 | firstline = false; | |
887 | if (strstr(line, "IBM/S390") != NULL) { | |
888 | is_s390x = true; | |
889 | am_printing = true; | |
890 | continue; | |
891 | } | |
892 | } | |
b456d40d | 893 | |
1f5596dd CB |
894 | if (strncmp(line, "# processors:", 12) == 0) |
895 | continue; | |
b456d40d | 896 | |
1f5596dd | 897 | if (is_processor_line(line)) { |
d0031abf | 898 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 899 | break; |
b456d40d | 900 | |
1f5596dd CB |
901 | am_printing = cpuline_in_cpuset(line, cpuset); |
902 | if (am_printing) { | |
d0031abf | 903 | curcpu++; |
1f5596dd | 904 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); |
b456d40d CB |
905 | if (l < 0) |
906 | return log_error(0, "Failed to write cache"); | |
907 | if (l >= cache_size) | |
908 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
909 | cache += l; |
910 | cache_size -= l; | |
911 | total_len += l; | |
912 | } | |
913 | continue; | |
914 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { | |
915 | char *p; | |
b456d40d | 916 | |
d0031abf | 917 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 918 | break; |
b456d40d | 919 | |
1f5596dd CB |
920 | if (!cpu_in_cpuset(cpu, cpuset)) |
921 | continue; | |
b456d40d | 922 | |
1f5596dd CB |
923 | curcpu ++; |
924 | p = strchr(line, ':'); | |
925 | if (!p || !*p) | |
926 | return 0; | |
927 | p++; | |
b456d40d | 928 | |
1f5596dd | 929 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
b456d40d CB |
930 | if (l < 0) |
931 | return log_error(0, "Failed to write cache"); | |
932 | if (l >= cache_size) | |
933 | return log_error(0, "Write to cache was truncated"); | |
934 | ||
1f5596dd CB |
935 | cache += l; |
936 | cache_size -= l; | |
937 | total_len += l; | |
938 | continue; | |
939 | ||
940 | } | |
941 | if (am_printing) { | |
942 | l = snprintf(cache, cache_size, "%s", line); | |
b456d40d CB |
943 | if (l < 0) |
944 | return log_error(0, "Failed to write cache"); | |
945 | if (l >= cache_size) | |
946 | return log_error(0, "Write to cache was truncated"); | |
947 | ||
1f5596dd CB |
948 | cache += l; |
949 | cache_size -= l; | |
950 | total_len += l; | |
951 | } | |
952 | } | |
953 | ||
954 | if (is_s390x) { | |
955 | __do_free char *origcache = d->buf; | |
956 | ssize_t l; | |
957 | ||
958 | d->buf = malloc(d->buflen); | |
959 | if (!d->buf) { | |
960 | d->buf = move_ptr(origcache); | |
961 | return 0; | |
962 | } | |
963 | ||
964 | cache = d->buf; | |
965 | cache_size = d->buflen; | |
966 | total_len = 0; | |
967 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
968 | if (l < 0 || l >= cache_size) | |
969 | return 0; | |
970 | ||
971 | cache_size -= l; | |
972 | cache += l; | |
973 | total_len += l; | |
974 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
975 | if (l < 0 || l >= cache_size) | |
976 | return 0; | |
977 | ||
978 | cache_size -= l; | |
979 | cache += l; | |
980 | total_len += l; | |
981 | l = snprintf(cache, cache_size, "%s", origcache); | |
982 | if (l < 0 || l >= cache_size) | |
983 | return 0; | |
984 | total_len += l; | |
985 | } | |
986 | ||
987 | d->cached = 1; | |
988 | d->size = total_len; | |
d0031abf CB |
989 | if (total_len > size) |
990 | total_len = size; | |
1f5596dd CB |
991 | |
992 | /* read from off 0 */ | |
993 | memcpy(buf, d->buf, total_len); | |
d0031abf | 994 | |
1f5596dd CB |
995 | return total_len; |
996 | } | |
997 | ||
998 | /* | |
999 | * Returns 0 on success. | |
1000 | * It is the caller's responsibility to free `return_usage`, unless this | |
1001 | * function returns an error. | |
1002 | */ | |
1003 | int read_cpuacct_usage_all(char *cg, char *cpuset, | |
1004 | struct cpuacct_usage **return_usage, int *size) | |
1005 | { | |
1006 | __do_free char *usage_str = NULL; | |
1007 | __do_free struct cpuacct_usage *cpu_usage = NULL; | |
9ce186dc | 1008 | int i = 0, j = 0, read_pos = 0, read_cnt = 0; |
8b6987a2 | 1009 | int cpucount; |
9ce186dc | 1010 | int ret; |
1f5596dd CB |
1011 | int cg_cpu; |
1012 | uint64_t cg_user, cg_system; | |
1013 | int64_t ticks_per_sec; | |
1014 | ||
1015 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
1f5596dd | 1016 | if (ticks_per_sec < 0 && errno == EINVAL) { |
8b6987a2 | 1017 | lxcfs_debug("%m - Failed to determine number of ticks per second"); |
1f5596dd CB |
1018 | return -1; |
1019 | } | |
1020 | ||
f9434b9a | 1021 | cpucount = get_nprocs_conf(); |
1f5596dd CB |
1022 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); |
1023 | if (!cpu_usage) | |
1024 | return -ENOMEM; | |
1025 | ||
1026 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); | |
1027 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
8b6987a2 CB |
1028 | char *sep = " \t\n"; |
1029 | char *tok; | |
1f5596dd | 1030 | |
8b6987a2 CB |
1031 | /* Read cpuacct.usage_percpu instead. */ |
1032 | lxcfs_debug("Falling back to cpuacct.usage_percpu"); | |
1f5596dd CB |
1033 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) |
1034 | return -1; | |
1f5596dd | 1035 | |
8b6987a2 CB |
1036 | lxc_iterate_parts(tok, usage_str, sep) { |
1037 | uint64_t percpu_user; | |
1038 | ||
1039 | if (i >= cpucount) | |
1040 | break; | |
1f5596dd | 1041 | |
8b6987a2 CB |
1042 | tok = trim_whitespace_in_place(tok); |
1043 | ret = safe_uint64(tok, &percpu_user, 10); | |
1044 | if (ret) | |
1045 | return -1; | |
1f5596dd | 1046 | |
8b6987a2 CB |
1047 | /* Convert the time from nanoseconds to USER_HZ */ |
1048 | cpu_usage[i].user = percpu_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1049 | cpu_usage[i].system = cpu_usage[i].user; | |
1f5596dd | 1050 | i++; |
8b6987a2 | 1051 | lxcfs_debug("cpu%d with time %s", i, tok); |
1f5596dd | 1052 | } |
8b6987a2 CB |
1053 | } else { |
1054 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) | |
1055 | return log_error(-1, "read_cpuacct_usage_all reading first line from %s/cpuacct.usage_all failed", cg); | |
1f5596dd | 1056 | |
8b6987a2 | 1057 | read_pos += read_cnt; |
1f5596dd | 1058 | |
8b6987a2 CB |
1059 | for (i = 0, j = 0; i < cpucount; i++) { |
1060 | ret = sscanf(usage_str + read_pos, | |
1061 | "%d %" PRIu64 " %" PRIu64 "\n%n", &cg_cpu, | |
1062 | &cg_user, &cg_system, &read_cnt); | |
1f5596dd | 1063 | |
8b6987a2 CB |
1064 | if (ret == EOF) |
1065 | break; | |
1f5596dd | 1066 | |
8b6987a2 CB |
1067 | if (ret != 3) |
1068 | return log_error(-EINVAL, "Failed to parse cpuacct.usage_all line %s from cgroup %s", | |
1069 | usage_str + read_pos, cg); | |
1f5596dd | 1070 | |
8b6987a2 | 1071 | read_pos += read_cnt; |
1f5596dd | 1072 | |
8b6987a2 CB |
1073 | /* Convert the time from nanoseconds to USER_HZ */ |
1074 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1075 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1076 | j++; | |
1077 | } | |
1f5596dd CB |
1078 | } |
1079 | ||
1080 | *return_usage = move_ptr(cpu_usage); | |
1081 | *size = cpucount; | |
1082 | return 0; | |
1083 | } | |
1084 | ||
1085 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
1086 | { | |
1087 | *head = malloc(sizeof(struct cg_proc_stat_head)); | |
b456d40d CB |
1088 | if (!(*head)) |
1089 | return log_error(false, "%s", strerror(errno)); | |
1f5596dd CB |
1090 | |
1091 | (*head)->lastcheck = time(NULL); | |
1092 | (*head)->next = NULL; | |
1093 | ||
1094 | if (pthread_rwlock_init(&(*head)->lock, NULL) != 0) { | |
1f5596dd | 1095 | free_disarm(*head); |
b456d40d | 1096 | return log_error(false, "Failed to initialize list lock"); |
1f5596dd CB |
1097 | } |
1098 | ||
1099 | return true; | |
1100 | } | |
1101 | ||
4ec5c9da | 1102 | bool init_cpuview(void) |
1f5596dd CB |
1103 | { |
1104 | int i; | |
1105 | ||
1106 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
1107 | proc_stat_history[i] = NULL; | |
1108 | ||
1109 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1110 | if (!cpuview_init_head(&proc_stat_history[i])) | |
1111 | goto err; | |
1112 | } | |
1113 | ||
1114 | return true; | |
1115 | ||
1116 | err: | |
1117 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1118 | if (proc_stat_history[i]) | |
1119 | free_disarm(proc_stat_history[i]); | |
1120 | } | |
1121 | ||
1122 | return false; | |
1123 | } | |
1124 | ||
1f5596dd CB |
1125 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
1126 | { | |
905769cd | 1127 | struct cg_proc_stat *node; |
1f5596dd CB |
1128 | |
1129 | if (head->next) { | |
1130 | node = head->next; | |
1131 | ||
1132 | for (;;) { | |
905769cd | 1133 | struct cg_proc_stat *cur = node; |
1f5596dd | 1134 | node = node->next; |
905769cd | 1135 | free_proc_stat_node(cur); |
1f5596dd CB |
1136 | if (!node) |
1137 | break; | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | pthread_rwlock_destroy(&head->lock); | |
1142 | free_disarm(head); | |
1143 | } | |
1144 | ||
4ec5c9da | 1145 | void free_cpuview(void) |
1f5596dd | 1146 | { |
4ec5c9da | 1147 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) |
1f5596dd CB |
1148 | if (proc_stat_history[i]) |
1149 | cpuview_free_head(proc_stat_history[i]); | |
1f5596dd | 1150 | } |