]>
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 | ||
f834b6bf SP |
7 | #include "config.h" |
8 | ||
1f5596dd CB |
9 | #define __STDC_FORMAT_MACROS |
10 | #include <dirent.h> | |
11 | #include <errno.h> | |
12 | #include <fcntl.h> | |
13 | #include <fuse.h> | |
14 | #include <inttypes.h> | |
15 | #include <libgen.h> | |
16 | #include <pthread.h> | |
17 | #include <sched.h> | |
18 | #include <stdarg.h> | |
19 | #include <stdbool.h> | |
20 | #include <stdint.h> | |
21 | #include <stdio.h> | |
22 | #include <stdlib.h> | |
23 | #include <string.h> | |
24 | #include <time.h> | |
25 | #include <unistd.h> | |
26 | #include <wait.h> | |
27 | #include <linux/magic.h> | |
28 | #include <linux/sched.h> | |
29 | #include <sys/epoll.h> | |
30 | #include <sys/mman.h> | |
31 | #include <sys/mount.h> | |
32 | #include <sys/param.h> | |
33 | #include <sys/socket.h> | |
34 | #include <sys/syscall.h> | |
35 | #include <sys/sysinfo.h> | |
36 | #include <sys/vfs.h> | |
37 | ||
38 | #include "bindings.h" | |
1f5596dd CB |
39 | #include "cgroup_fuse.h" |
40 | #include "cpuset_parse.h" | |
41 | #include "cgroups/cgroup.h" | |
42 | #include "cgroups/cgroup_utils.h" | |
43 | #include "memory_utils.h" | |
4ec5c9da | 44 | #include "proc_loadavg.h" |
1f5596dd CB |
45 | #include "utils.h" |
46 | ||
1f5596dd CB |
47 | /* Data for CPU view */ |
48 | struct cg_proc_stat { | |
49 | char *cg; | |
ce617d73 CB |
50 | struct cpuacct_usage *usage; /* Real usage as read from the host's /proc/stat. */ |
51 | struct cpuacct_usage *view; /* Usage stats reported to the container. */ | |
1f5596dd | 52 | int cpu_count; |
ce617d73 | 53 | pthread_mutex_t lock; /* For node manipulation. */ |
1f5596dd CB |
54 | struct cg_proc_stat *next; |
55 | }; | |
56 | ||
57 | struct cg_proc_stat_head { | |
58 | struct cg_proc_stat *next; | |
59 | time_t lastcheck; | |
60 | ||
61 | /* | |
62 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
63 | */ | |
64 | pthread_rwlock_t lock; | |
65 | }; | |
66 | ||
67 | #define CPUVIEW_HASH_SIZE 100 | |
68 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
69 | ||
b456d40d CB |
70 | static void reset_proc_stat_node(struct cg_proc_stat *node, |
71 | struct cpuacct_usage *usage, int cpu_count) | |
1f5596dd | 72 | { |
1f5596dd CB |
73 | lxcfs_debug("Resetting stat node for %s\n", node->cg); |
74 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
75 | ||
b456d40d | 76 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
77 | node->view[i].user = 0; |
78 | node->view[i].system = 0; | |
79 | node->view[i].idle = 0; | |
80 | } | |
81 | ||
82 | node->cpu_count = cpu_count; | |
83 | } | |
84 | ||
85 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) | |
86 | { | |
87 | __do_free struct cpuacct_usage *new_usage = NULL, *new_view = NULL; | |
88 | ||
89 | /* Allocate new memory */ | |
82d74a95 | 90 | new_usage = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
91 | if (!new_usage) |
92 | return false; | |
93 | ||
82d74a95 | 94 | new_view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
95 | if (!new_view) |
96 | return false; | |
97 | ||
98 | /* Copy existing data & initialize new elements */ | |
99 | for (int i = 0; i < cpu_count; i++) { | |
100 | if (i < node->cpu_count) { | |
82d74a95 CB |
101 | new_usage[i].user = node->usage[i].user; |
102 | new_usage[i].system = node->usage[i].system; | |
103 | new_usage[i].idle = node->usage[i].idle; | |
104 | ||
105 | new_view[i].user = node->view[i].user; | |
106 | new_view[i].system = node->view[i].system; | |
107 | new_view[i].idle = node->view[i].idle; | |
1f5596dd CB |
108 | } |
109 | } | |
110 | ||
111 | free(node->usage); | |
112 | node->usage = move_ptr(new_usage); | |
113 | ||
114 | free(node->view); | |
115 | node->view = move_ptr(new_view); | |
116 | node->cpu_count = cpu_count; | |
117 | ||
118 | return true; | |
119 | } | |
120 | ||
4ec5c9da CB |
121 | static void free_proc_stat_node(struct cg_proc_stat *node) |
122 | { | |
6a4dceb1 CB |
123 | if (node) { |
124 | /* | |
125 | * We're abusing the usage pointer to indicate that | |
126 | * pthread_mutex_init() was successful. Don't judge me. | |
127 | */ | |
128 | if (node->usage) | |
129 | pthread_mutex_destroy(&node->lock); | |
130 | free_disarm(node->cg); | |
131 | free_disarm(node->usage); | |
132 | free_disarm(node->view); | |
133 | free_disarm(node); | |
134 | } | |
4ec5c9da CB |
135 | } |
136 | ||
6a4dceb1 CB |
137 | define_cleanup_function(struct cg_proc_stat *, free_proc_stat_node); |
138 | ||
1f5596dd CB |
139 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
140 | { | |
0d129671 CB |
141 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *new = new_node; |
142 | struct cg_proc_stat *rv = new_node; | |
143 | int hash = calc_hash(new->cg) % CPUVIEW_HASH_SIZE; | |
1f5596dd | 144 | struct cg_proc_stat_head *head = proc_stat_history[hash]; |
0d129671 | 145 | struct cg_proc_stat *cur; |
1f5596dd CB |
146 | |
147 | pthread_rwlock_wrlock(&head->lock); | |
148 | ||
149 | if (!head->next) { | |
0d129671 | 150 | head->next = move_ptr(new); |
164acda7 | 151 | goto out_rwlock_unlock; |
1f5596dd CB |
152 | } |
153 | ||
0d129671 | 154 | cur = head->next; |
1f5596dd CB |
155 | |
156 | for (;;) { | |
0d129671 CB |
157 | /* |
158 | * The node to be added is already present in the list, so | |
159 | * free the newly allocated one and return the one we found. | |
160 | */ | |
161 | if (strcmp(cur->cg, new->cg) == 0) { | |
162 | rv = cur; | |
164acda7 | 163 | goto out_rwlock_unlock; |
1f5596dd CB |
164 | } |
165 | ||
0d129671 CB |
166 | /* Keep walking. */ |
167 | if (cur->next) { | |
168 | cur = cur->next; | |
1f5596dd CB |
169 | continue; |
170 | } | |
171 | ||
0d129671 CB |
172 | /* Add new node to end of list. */ |
173 | cur->next = move_ptr(new); | |
164acda7 | 174 | goto out_rwlock_unlock; |
1f5596dd CB |
175 | } |
176 | ||
164acda7 | 177 | out_rwlock_unlock: |
1f5596dd | 178 | pthread_rwlock_unlock(&head->lock); |
0d129671 | 179 | return move_ptr(rv); |
1f5596dd CB |
180 | } |
181 | ||
6a4dceb1 CB |
182 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, |
183 | int cpu_count, const char *cg) | |
1f5596dd | 184 | { |
6a4dceb1 CB |
185 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *node = NULL; |
186 | __do_free struct cpuacct_usage *new_usage = NULL; | |
1f5596dd | 187 | |
6a4dceb1 | 188 | node = zalloc(sizeof(struct cg_proc_stat)); |
1f5596dd | 189 | if (!node) |
6a4dceb1 | 190 | return NULL; |
1f5596dd | 191 | |
6a4dceb1 | 192 | node->cg = strdup(cg); |
1f5596dd | 193 | if (!node->cg) |
6a4dceb1 | 194 | return NULL; |
1f5596dd | 195 | |
6a4dceb1 CB |
196 | new_usage = memdup(usage, sizeof(struct cpuacct_usage) * cpu_count); |
197 | if (!new_usage) | |
198 | return NULL; | |
1f5596dd | 199 | |
6a4dceb1 | 200 | node->view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd | 201 | if (!node->view) |
6a4dceb1 | 202 | return NULL; |
1f5596dd CB |
203 | |
204 | node->cpu_count = cpu_count; | |
1f5596dd | 205 | |
6a4dceb1 CB |
206 | if (pthread_mutex_init(&node->lock, NULL)) |
207 | return NULL; | |
208 | /* | |
209 | * We're abusing the usage pointer to indicate that | |
210 | * pthread_mutex_init() was successful. Don't judge me. | |
211 | */ | |
212 | node->usage = move_ptr(new_usage); | |
1f5596dd | 213 | |
6a4dceb1 | 214 | return move_ptr(node); |
1f5596dd CB |
215 | } |
216 | ||
2d00d04c CB |
217 | static bool cgroup_supports(const char *controller, const char *cgroup, |
218 | const char *file) | |
4ec5c9da | 219 | { |
2c990b1d CB |
220 | __do_free char *path = NULL; |
221 | int cfd; | |
4ec5c9da CB |
222 | |
223 | cfd = get_cgroup_fd(controller); | |
224 | if (cfd < 0) | |
225 | return false; | |
226 | ||
925d5849 | 227 | path = must_make_path_relative(cgroup, file, NULL); |
2d00d04c | 228 | return faccessat(cfd, path, F_OK, 0) == 0; |
4ec5c9da CB |
229 | } |
230 | ||
1f5596dd CB |
231 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
232 | { | |
b456d40d | 233 | struct cg_proc_stat *first = NULL; |
1f5596dd | 234 | |
b456d40d | 235 | for (struct cg_proc_stat *prev = NULL; node; ) { |
2d00d04c CB |
236 | if (!cgroup_supports("cpu", node->cg, "cpu.shares")) { |
237 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *cur = node; | |
1f5596dd CB |
238 | |
239 | if (prev) | |
240 | prev->next = node->next; | |
241 | else | |
242 | first = node->next; | |
243 | ||
244 | node = node->next; | |
2d00d04c | 245 | lxcfs_debug("Removing stat node for %s\n", cur->cg); |
1f5596dd CB |
246 | } else { |
247 | if (!first) | |
248 | first = node; | |
249 | prev = node; | |
250 | node = node->next; | |
251 | } | |
252 | } | |
253 | ||
254 | return first; | |
255 | } | |
256 | ||
257 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
258 | static void prune_proc_stat_history(void) | |
259 | { | |
1f5596dd CB |
260 | time_t now = time(NULL); |
261 | ||
b456d40d | 262 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) { |
1f5596dd CB |
263 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
264 | ||
265 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
266 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
267 | return; | |
268 | } | |
269 | ||
270 | if (proc_stat_history[i]->next) { | |
271 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
272 | proc_stat_history[i]->lastcheck = now; | |
273 | } | |
274 | ||
275 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
276 | } | |
277 | } | |
278 | ||
279 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, | |
280 | const char *cg) | |
281 | { | |
282 | struct cg_proc_stat *node; | |
283 | ||
284 | pthread_rwlock_rdlock(&head->lock); | |
285 | ||
286 | if (!head->next) { | |
287 | pthread_rwlock_unlock(&head->lock); | |
288 | return NULL; | |
289 | } | |
290 | ||
291 | node = head->next; | |
292 | ||
293 | do { | |
294 | if (strcmp(cg, node->cg) == 0) | |
295 | goto out; | |
296 | } while ((node = node->next)); | |
297 | ||
298 | node = NULL; | |
299 | ||
300 | out: | |
301 | pthread_rwlock_unlock(&head->lock); | |
302 | prune_proc_stat_history(); | |
303 | return node; | |
304 | } | |
305 | ||
692f48eb CB |
306 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, |
307 | int cpu_count, const char *cg) | |
1f5596dd CB |
308 | { |
309 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
310 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
311 | struct cg_proc_stat *node; | |
312 | ||
313 | node = find_proc_stat_node(head, cg); | |
1f5596dd CB |
314 | if (!node) { |
315 | node = new_proc_stat_node(usage, cpu_count, cg); | |
316 | if (!node) | |
317 | return NULL; | |
318 | ||
319 | node = add_proc_stat_node(node); | |
320 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
321 | } | |
322 | ||
323 | pthread_mutex_lock(&node->lock); | |
324 | ||
ce089f10 CB |
325 | /* |
326 | * If additional CPUs on the host have been enabled, CPU usage counter | |
327 | * arrays have to be expanded. | |
328 | */ | |
1f5596dd CB |
329 | if (node->cpu_count < cpu_count) { |
330 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
ce089f10 | 331 | node->cpu_count, cpu_count, cg); |
1f5596dd CB |
332 | |
333 | if (!expand_proc_stat_node(node, cpu_count)) { | |
334 | pthread_mutex_unlock(&node->lock); | |
b456d40d | 335 | return log_debug(NULL, "Unable to expand stat node %d->%d for %s", node->cpu_count, cpu_count, cg); |
1f5596dd CB |
336 | } |
337 | } | |
338 | ||
339 | return node; | |
340 | } | |
341 | ||
2b8eff1d CB |
342 | static void add_cpu_usage(uint64_t *surplus, struct cpuacct_usage *usage, |
343 | uint64_t *counter, uint64_t threshold) | |
1f5596dd | 344 | { |
1ba088ae | 345 | uint64_t free_space, to_add; |
1f5596dd CB |
346 | |
347 | free_space = threshold - usage->user - usage->system; | |
348 | ||
349 | if (free_space > usage->idle) | |
350 | free_space = usage->idle; | |
351 | ||
8206874a CB |
352 | if (free_space > *surplus) |
353 | to_add = *surplus; | |
354 | else | |
355 | to_add = free_space; | |
1f5596dd CB |
356 | |
357 | *counter += to_add; | |
358 | usage->idle -= to_add; | |
359 | *surplus -= to_add; | |
360 | } | |
361 | ||
1ba088ae CB |
362 | static uint64_t diff_cpu_usage(struct cpuacct_usage *older, |
363 | struct cpuacct_usage *newer, | |
364 | struct cpuacct_usage *diff, int cpu_count) | |
1f5596dd | 365 | { |
1ba088ae | 366 | uint64_t sum = 0; |
1f5596dd | 367 | |
b456d40d | 368 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
369 | if (!newer[i].online) |
370 | continue; | |
371 | ||
b456d40d CB |
372 | /* |
373 | * When cpuset is changed on the fly, the CPUs might get | |
374 | * reordered. We could either reset all counters, or check | |
375 | * that the substractions below will return expected results. | |
1f5596dd CB |
376 | */ |
377 | if (newer[i].user > older[i].user) | |
378 | diff[i].user = newer[i].user - older[i].user; | |
379 | else | |
380 | diff[i].user = 0; | |
381 | ||
382 | if (newer[i].system > older[i].system) | |
383 | diff[i].system = newer[i].system - older[i].system; | |
384 | else | |
385 | diff[i].system = 0; | |
386 | ||
387 | if (newer[i].idle > older[i].idle) | |
388 | diff[i].idle = newer[i].idle - older[i].idle; | |
389 | else | |
390 | diff[i].idle = 0; | |
391 | ||
392 | sum += diff[i].user; | |
393 | sum += diff[i].system; | |
394 | sum += diff[i].idle; | |
395 | } | |
396 | ||
397 | return sum; | |
398 | } | |
399 | ||
400 | /* | |
b456d40d CB |
401 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or |
402 | * `cpu.cfs_period_us`, depending on `param`. Parameter value is returned | |
92264841 | 403 | * through `value`. |
1f5596dd CB |
404 | */ |
405 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
406 | { | |
407 | __do_free char *str = NULL; | |
48f6862e | 408 | char file[STRLITERALLEN("cpu.cfs_period_us") + 1]; |
9844eea7 | 409 | bool first = true; |
48f6862e | 410 | int ret; |
1f5596dd | 411 | |
48f6862e | 412 | if (pure_unified_layout(cgroup_ops)) { |
9844eea7 | 413 | first = !strcmp(param, "quota"); |
48f6862e CB |
414 | ret = snprintf(file, sizeof(file), "cpu.max"); |
415 | } else { | |
416 | ret = snprintf(file, sizeof(file), "cpu.cfs_%s_us", param); | |
9844eea7 | 417 | } |
48f6862e | 418 | if (ret < 0 || (size_t)ret >= sizeof(file)) |
1f5596dd CB |
419 | return false; |
420 | ||
48f6862e | 421 | if (!cgroup_ops->get(cgroup_ops, "cpu", cg, file, &str)) |
1f5596dd CB |
422 | return false; |
423 | ||
48f6862e | 424 | return sscanf(str, first ? "%" PRId64 : "%*d %" PRId64, value) == 1; |
1f5596dd CB |
425 | } |
426 | ||
427 | /* | |
428 | * Return the exact number of visible CPUs based on CPU quotas. | |
429 | * If there is no quota set, zero is returned. | |
430 | */ | |
431 | static double exact_cpu_count(const char *cg) | |
432 | { | |
433 | double rv; | |
434 | int nprocs; | |
435 | int64_t cfs_quota, cfs_period; | |
436 | ||
c602a0d0 CB |
437 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
438 | return 0; | |
439 | ||
440 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
441 | return 0; | |
1f5596dd CB |
442 | |
443 | if (cfs_quota <= 0 || cfs_period <= 0) | |
444 | return 0; | |
445 | ||
446 | rv = (double)cfs_quota / (double)cfs_period; | |
447 | ||
448 | nprocs = get_nprocs(); | |
449 | ||
450 | if (rv > nprocs) | |
451 | rv = nprocs; | |
452 | ||
453 | return rv; | |
454 | } | |
455 | ||
456 | /* | |
457 | * Return the maximum number of visible CPUs based on CPU quotas. | |
458 | * If there is no quota set, zero is returned. | |
459 | */ | |
4ec5c9da | 460 | int max_cpu_count(const char *cg) |
1f5596dd | 461 | { |
700dd417 | 462 | __do_free char *cpuset = NULL; |
1f5596dd CB |
463 | int rv, nprocs; |
464 | int64_t cfs_quota, cfs_period; | |
465 | int nr_cpus_in_cpuset = 0; | |
1f5596dd | 466 | |
921bdfdb CB |
467 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
468 | return 0; | |
469 | ||
470 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
471 | return 0; | |
1f5596dd CB |
472 | |
473 | cpuset = get_cpuset(cg); | |
474 | if (cpuset) | |
475 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
476 | ||
921bdfdb | 477 | if (cfs_quota <= 0 || cfs_period <= 0) { |
1f5596dd CB |
478 | if (nr_cpus_in_cpuset > 0) |
479 | return nr_cpus_in_cpuset; | |
480 | ||
481 | return 0; | |
482 | } | |
483 | ||
484 | rv = cfs_quota / cfs_period; | |
485 | ||
921bdfdb CB |
486 | /* |
487 | * In case quota/period does not yield a whole number, add one CPU for | |
1f5596dd CB |
488 | * the remainder. |
489 | */ | |
490 | if ((cfs_quota % cfs_period) > 0) | |
491 | rv += 1; | |
492 | ||
493 | nprocs = get_nprocs(); | |
1f5596dd CB |
494 | if (rv > nprocs) |
495 | rv = nprocs; | |
496 | ||
921bdfdb | 497 | /* Use min value in cpu quota and cpuset. */ |
1f5596dd CB |
498 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) |
499 | rv = nr_cpus_in_cpuset; | |
500 | ||
501 | return rv; | |
502 | } | |
503 | ||
504 | int cpuview_proc_stat(const char *cg, const char *cpuset, | |
505 | struct cpuacct_usage *cg_cpu_usage, int cg_cpu_usage_size, | |
506 | FILE *f, char *buf, size_t buf_size) | |
507 | { | |
508 | __do_free char *line = NULL; | |
509 | __do_free struct cpuacct_usage *diff = NULL; | |
4f18a602 | 510 | size_t linelen = 0, total_len = 0; |
1f5596dd CB |
511 | int curcpu = -1; /* cpu numbering starts at 0 */ |
512 | int physcpu, i; | |
39f231da | 513 | int cpu_cnt = 0; |
2b8eff1d CB |
514 | uint64_t user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, |
515 | softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
516 | uint64_t user_sum = 0, system_sum = 0, idle_sum = 0; | |
517 | uint64_t user_surplus = 0, system_surplus = 0; | |
39f231da | 518 | int nprocs, max_cpus; |
4f18a602 | 519 | ssize_t l; |
2b8eff1d | 520 | uint64_t total_sum, threshold; |
1f5596dd | 521 | struct cg_proc_stat *stat_node; |
1f5596dd | 522 | |
39f231da | 523 | nprocs = get_nprocs_conf(); |
1f5596dd CB |
524 | if (cg_cpu_usage_size < nprocs) |
525 | nprocs = cg_cpu_usage_size; | |
526 | ||
527 | /* Read all CPU stats and stop when we've encountered other lines */ | |
528 | while (getline(&line, &linelen, f) != -1) { | |
529 | int ret; | |
530 | char cpu_char[10]; /* That's a lot of cores */ | |
531 | uint64_t all_used, cg_used; | |
532 | ||
533 | if (strlen(line) == 0) | |
534 | continue; | |
535 | ||
536 | /* not a ^cpuN line containing a number N */ | |
537 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) | |
538 | break; | |
539 | ||
540 | if (sscanf(cpu_char, "%d", &physcpu) != 1) | |
541 | continue; | |
542 | ||
543 | if (physcpu >= cg_cpu_usage_size) | |
544 | continue; | |
545 | ||
fd65c77c CB |
546 | curcpu++; |
547 | cpu_cnt++; | |
1f5596dd CB |
548 | |
549 | if (!cpu_in_cpuset(physcpu, cpuset)) { | |
550 | for (i = curcpu; i <= physcpu; i++) | |
551 | cg_cpu_usage[i].online = false; | |
552 | continue; | |
553 | } | |
554 | ||
555 | if (curcpu < physcpu) { | |
556 | /* Some CPUs may be disabled */ | |
557 | for (i = curcpu; i < physcpu; i++) | |
558 | cg_cpu_usage[i].online = false; | |
559 | ||
560 | curcpu = physcpu; | |
561 | } | |
562 | ||
563 | cg_cpu_usage[curcpu].online = true; | |
564 | ||
2b8eff1d | 565 | ret = sscanf(line, "%*s %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 "lu", |
1f5596dd CB |
566 | &user, |
567 | &nice, | |
568 | &system, | |
569 | &idle, | |
570 | &iowait, | |
571 | &irq, | |
572 | &softirq, | |
573 | &steal, | |
574 | &guest, | |
575 | &guest_nice); | |
1f5596dd CB |
576 | if (ret != 10) |
577 | continue; | |
578 | ||
579 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
580 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
581 | ||
582 | if (all_used >= cg_used) { | |
583 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
584 | ||
585 | } else { | |
2b8eff1d CB |
586 | lxcfs_error("cpu%d from %s has unexpected cpu time: %" PRIu64 " in /proc/stat, %" PRIu64 " in cpuacct.usage_all; unable to determine idle time", |
587 | curcpu, cg, all_used, cg_used); | |
1f5596dd CB |
588 | cg_cpu_usage[curcpu].idle = idle; |
589 | } | |
590 | } | |
591 | ||
f9434b9a CB |
592 | /* Cannot use more CPUs than is available in cpuset. */ |
593 | max_cpus = max_cpu_count(cg); | |
594 | if (max_cpus > cpu_cnt || !max_cpus) | |
595 | max_cpus = cpu_cnt; | |
596 | ||
692f48eb | 597 | /* takes lock pthread_mutex_lock(&node->lock) */ |
1f5596dd | 598 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
b456d40d CB |
599 | if (!stat_node) |
600 | return log_error(0, "Failed to find/create stat node for %s", cg); | |
1f5596dd | 601 | |
b4572722 | 602 | diff = zalloc(sizeof(struct cpuacct_usage) * nprocs); |
700dd417 | 603 | if (!diff) |
08d61303 | 604 | goto out_pthread_mutex_unlock; |
1f5596dd CB |
605 | |
606 | /* | |
607 | * If the new values are LOWER than values stored in memory, it means | |
608 | * the cgroup has been reset/recreated and we should reset too. | |
609 | */ | |
610 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
611 | if (!cg_cpu_usage[curcpu].online) | |
612 | continue; | |
613 | ||
614 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
615 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
616 | ||
617 | break; | |
618 | } | |
619 | ||
620 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); | |
621 | ||
622 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
623 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
624 | ||
625 | if (!stat_node->usage[curcpu].online) | |
626 | continue; | |
627 | ||
628 | i++; | |
629 | ||
b4572722 | 630 | stat_node->usage[curcpu].user += diff[curcpu].user; |
1f5596dd | 631 | stat_node->usage[curcpu].system += diff[curcpu].system; |
b4572722 | 632 | stat_node->usage[curcpu].idle += diff[curcpu].idle; |
1f5596dd CB |
633 | |
634 | if (max_cpus > 0 && i >= max_cpus) { | |
b4572722 CB |
635 | user_surplus += diff[curcpu].user; |
636 | system_surplus += diff[curcpu].system; | |
1f5596dd CB |
637 | } |
638 | } | |
639 | ||
640 | /* Calculate usage counters of visible CPUs */ | |
641 | if (max_cpus > 0) { | |
2b8eff1d CB |
642 | uint64_t diff_user = 0; |
643 | uint64_t diff_system = 0; | |
644 | uint64_t diff_idle = 0; | |
645 | uint64_t max_diff_idle = 0; | |
646 | uint64_t max_diff_idle_index = 0; | |
1f5596dd | 647 | double exact_cpus; |
1f5596dd CB |
648 | /* threshold = maximum usage per cpu, including idle */ |
649 | threshold = total_sum / cpu_cnt * max_cpus; | |
650 | ||
651 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
652 | if (!stat_node->usage[curcpu].online) | |
653 | continue; | |
654 | ||
655 | i++; | |
656 | ||
657 | if (i == max_cpus) | |
658 | break; | |
659 | ||
660 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
661 | continue; | |
662 | ||
663 | /* Add user */ | |
664 | add_cpu_usage(&user_surplus, &diff[curcpu], | |
665 | &diff[curcpu].user, threshold); | |
666 | ||
667 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
668 | continue; | |
669 | ||
670 | /* If there is still room, add system */ | |
671 | add_cpu_usage(&system_surplus, &diff[curcpu], | |
672 | &diff[curcpu].system, threshold); | |
673 | } | |
674 | ||
675 | if (user_surplus > 0) | |
676 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
677 | if (system_surplus > 0) | |
678 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
679 | ||
680 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
681 | if (!stat_node->usage[curcpu].online) | |
682 | continue; | |
683 | ||
684 | i++; | |
685 | ||
686 | if (i == max_cpus) | |
687 | break; | |
688 | ||
b4572722 CB |
689 | stat_node->view[curcpu].user += diff[curcpu].user; |
690 | stat_node->view[curcpu].system += diff[curcpu].system; | |
691 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
1f5596dd | 692 | |
b4572722 CB |
693 | user_sum += stat_node->view[curcpu].user; |
694 | system_sum += stat_node->view[curcpu].system; | |
695 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd | 696 | |
b4572722 CB |
697 | diff_user += diff[curcpu].user; |
698 | diff_system += diff[curcpu].system; | |
699 | diff_idle += diff[curcpu].idle; | |
1f5596dd | 700 | if (diff[curcpu].idle > max_diff_idle) { |
b4572722 CB |
701 | max_diff_idle = diff[curcpu].idle; |
702 | max_diff_idle_index = curcpu; | |
1f5596dd CB |
703 | } |
704 | ||
705 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
706 | } | |
707 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
708 | ||
709 | /* revise cpu usage view to support partial cpu case. */ | |
710 | exact_cpus = exact_cpu_count(cg); | |
711 | if (exact_cpus < (double)max_cpus){ | |
1ba088ae | 712 | uint64_t delta = (uint64_t)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); |
1f5596dd CB |
713 | |
714 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
715 | lxcfs_v("delta: %lu\n", delta); | |
716 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
b4572722 CB |
717 | if (idle_sum > delta) |
718 | idle_sum = idle_sum - delta; | |
719 | else | |
720 | idle_sum = 0; | |
1f5596dd CB |
721 | lxcfs_v("idle_sum after: %lu\n", idle_sum); |
722 | ||
723 | curcpu = max_diff_idle_index; | |
724 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
b4572722 CB |
725 | if (stat_node->view[curcpu].idle > delta) |
726 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle - delta; | |
727 | else | |
728 | stat_node->view[curcpu].idle = 0; | |
1f5596dd CB |
729 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); |
730 | } | |
731 | } else { | |
732 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
733 | if (!stat_node->usage[curcpu].online) | |
734 | continue; | |
735 | ||
b4572722 CB |
736 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; |
737 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
738 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
1f5596dd | 739 | |
b4572722 CB |
740 | user_sum += stat_node->view[curcpu].user; |
741 | system_sum += stat_node->view[curcpu].system; | |
742 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd CB |
743 | } |
744 | } | |
745 | ||
746 | /* Render the file */ | |
747 | /* cpu-all */ | |
2b8eff1d CB |
748 | l = snprintf(buf, buf_size, |
749 | "cpu %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", | |
750 | user_sum, system_sum, idle_sum); | |
1f5596dd | 751 | lxcfs_v("cpu-all: %s\n", buf); |
692f48eb CB |
752 | if (l < 0) { |
753 | lxcfs_error("Failed to write cache"); | |
754 | total_len = 0; | |
755 | goto out_pthread_mutex_unlock; | |
756 | } | |
08d61303 CB |
757 | if (l >= buf_size) { |
758 | lxcfs_error("Write to cache was truncated"); | |
759 | total_len = 0; | |
760 | goto out_pthread_mutex_unlock; | |
761 | } | |
1f5596dd CB |
762 | |
763 | buf += l; | |
764 | buf_size -= l; | |
765 | total_len += l; | |
766 | ||
767 | /* Render visible CPUs */ | |
768 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
769 | if (!stat_node->usage[curcpu].online) | |
770 | continue; | |
771 | ||
772 | i++; | |
773 | ||
774 | if (max_cpus > 0 && i == max_cpus) | |
775 | break; | |
776 | ||
2b8eff1d CB |
777 | l = snprintf(buf, buf_size, "cpu%d %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", |
778 | i, | |
779 | stat_node->view[curcpu].user, | |
780 | stat_node->view[curcpu].system, | |
781 | stat_node->view[curcpu].idle); | |
1f5596dd | 782 | lxcfs_v("cpu: %s\n", buf); |
692f48eb CB |
783 | if (l < 0) { |
784 | lxcfs_error("Failed to write cache"); | |
785 | total_len = 0; | |
786 | goto out_pthread_mutex_unlock; | |
787 | } | |
788 | if (l >= buf_size) { | |
789 | lxcfs_error("Write to cache was truncated"); | |
790 | total_len = 0; | |
791 | goto out_pthread_mutex_unlock; | |
792 | } | |
1f5596dd CB |
793 | |
794 | buf += l; | |
795 | buf_size -= l; | |
796 | total_len += l; | |
797 | } | |
798 | ||
799 | /* Pass the rest of /proc/stat, start with the last line read */ | |
800 | l = snprintf(buf, buf_size, "%s", line); | |
692f48eb CB |
801 | if (l < 0) { |
802 | lxcfs_error("Failed to write cache"); | |
803 | total_len = 0; | |
804 | goto out_pthread_mutex_unlock; | |
805 | } | |
806 | if (l >= buf_size) { | |
807 | lxcfs_error("Write to cache was truncated"); | |
808 | total_len = 0; | |
809 | goto out_pthread_mutex_unlock; | |
810 | } | |
1f5596dd CB |
811 | |
812 | buf += l; | |
813 | buf_size -= l; | |
814 | total_len += l; | |
815 | ||
816 | /* Pass the rest of the host's /proc/stat */ | |
817 | while (getline(&line, &linelen, f) != -1) { | |
818 | l = snprintf(buf, buf_size, "%s", line); | |
692f48eb CB |
819 | if (l < 0) { |
820 | lxcfs_error("Failed to write cache"); | |
821 | total_len = 0; | |
822 | goto out_pthread_mutex_unlock; | |
823 | } | |
824 | if (l >= buf_size) { | |
825 | lxcfs_error("Write to cache was truncated"); | |
826 | total_len = 0; | |
827 | goto out_pthread_mutex_unlock; | |
828 | } | |
b456d40d | 829 | |
1f5596dd CB |
830 | buf += l; |
831 | buf_size -= l; | |
832 | total_len += l; | |
833 | } | |
834 | ||
692f48eb | 835 | out_pthread_mutex_unlock: |
1f5596dd CB |
836 | if (stat_node) |
837 | pthread_mutex_unlock(&stat_node->lock); | |
b456d40d | 838 | |
1f5596dd CB |
839 | return total_len; |
840 | } | |
841 | ||
842 | /* | |
843 | * check whether this is a '^processor" line in /proc/cpuinfo | |
844 | */ | |
b456d40d | 845 | static inline bool is_processor_line(const char *line) |
1f5596dd CB |
846 | { |
847 | int cpu; | |
b456d40d | 848 | return sscanf(line, "processor : %d", &cpu) == 1; |
1f5596dd CB |
849 | } |
850 | ||
b456d40d | 851 | static inline bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1f5596dd CB |
852 | { |
853 | int cpu; | |
c539526c CB |
854 | |
855 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
856 | return cpu_in_cpuset(cpu, cpuset); | |
857 | ||
858 | return false; | |
1f5596dd CB |
859 | } |
860 | ||
861 | int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
862 | struct fuse_file_info *fi) | |
863 | { | |
864 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; | |
757a63e7 | 865 | __do_free void *fopen_cache = NULL; |
1f5596dd CB |
866 | __do_fclose FILE *f = NULL; |
867 | struct fuse_context *fc = fuse_get_context(); | |
0274438c | 868 | struct lxcfs_opts *opts = (struct lxcfs_opts *)fc->private_data; |
99b183fb | 869 | struct file_info *d = INTTYPE_TO_PTR(fi->fh); |
1f5596dd CB |
870 | size_t linelen = 0, total_len = 0; |
871 | bool am_printing = false, firstline = true, is_s390x = false; | |
872 | int curcpu = -1, cpu, max_cpus = 0; | |
873 | bool use_view; | |
874 | char *cache = d->buf; | |
875 | size_t cache_size = d->buflen; | |
876 | ||
f9434b9a | 877 | if (offset) { |
1f5596dd CB |
878 | int left; |
879 | ||
880 | if (offset > d->size) | |
881 | return -EINVAL; | |
882 | ||
883 | if (!d->cached) | |
884 | return 0; | |
885 | ||
886 | left = d->size - offset; | |
887 | total_len = left > size ? size: left; | |
888 | memcpy(buf, cache + offset, total_len); | |
889 | ||
890 | return total_len; | |
891 | } | |
892 | ||
893 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
894 | if (initpid <= 1 || is_shared_pidns(initpid)) | |
895 | initpid = fc->pid; | |
b456d40d | 896 | |
1f5596dd CB |
897 | cg = get_pid_cgroup(initpid, "cpuset"); |
898 | if (!cg) | |
899 | return read_file_fuse("proc/cpuinfo", buf, size, d); | |
900 | prune_init_slice(cg); | |
901 | ||
902 | cpuset = get_cpuset(cg); | |
903 | if (!cpuset) | |
904 | return 0; | |
905 | ||
8044f626 | 906 | if (cgroup_ops->can_use_cpuview(cgroup_ops) && opts && opts->use_cfs) |
0274438c | 907 | use_view = true; |
8044f626 CB |
908 | else |
909 | use_view = false; | |
1f5596dd CB |
910 | if (use_view) |
911 | max_cpus = max_cpu_count(cg); | |
912 | ||
757a63e7 | 913 | f = fopen_cached("/proc/cpuinfo", "re", &fopen_cache); |
1f5596dd CB |
914 | if (!f) |
915 | return 0; | |
916 | ||
917 | while (getline(&line, &linelen, f) != -1) { | |
918 | ssize_t l; | |
919 | if (firstline) { | |
920 | firstline = false; | |
921 | if (strstr(line, "IBM/S390") != NULL) { | |
922 | is_s390x = true; | |
923 | am_printing = true; | |
924 | continue; | |
925 | } | |
926 | } | |
b456d40d | 927 | |
1f5596dd CB |
928 | if (strncmp(line, "# processors:", 12) == 0) |
929 | continue; | |
b456d40d | 930 | |
1f5596dd | 931 | if (is_processor_line(line)) { |
d0031abf | 932 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 933 | break; |
b456d40d | 934 | |
1f5596dd CB |
935 | am_printing = cpuline_in_cpuset(line, cpuset); |
936 | if (am_printing) { | |
d0031abf | 937 | curcpu++; |
1f5596dd | 938 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); |
b456d40d CB |
939 | if (l < 0) |
940 | return log_error(0, "Failed to write cache"); | |
941 | if (l >= cache_size) | |
942 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
943 | cache += l; |
944 | cache_size -= l; | |
945 | total_len += l; | |
946 | } | |
947 | continue; | |
948 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { | |
949 | char *p; | |
b456d40d | 950 | |
d0031abf | 951 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 952 | break; |
b456d40d | 953 | |
1f5596dd CB |
954 | if (!cpu_in_cpuset(cpu, cpuset)) |
955 | continue; | |
b456d40d | 956 | |
1f5596dd CB |
957 | curcpu ++; |
958 | p = strchr(line, ':'); | |
959 | if (!p || !*p) | |
960 | return 0; | |
961 | p++; | |
b456d40d | 962 | |
1f5596dd | 963 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
b456d40d CB |
964 | if (l < 0) |
965 | return log_error(0, "Failed to write cache"); | |
966 | if (l >= cache_size) | |
967 | return log_error(0, "Write to cache was truncated"); | |
968 | ||
1f5596dd CB |
969 | cache += l; |
970 | cache_size -= l; | |
971 | total_len += l; | |
972 | continue; | |
973 | ||
974 | } | |
975 | if (am_printing) { | |
976 | l = snprintf(cache, cache_size, "%s", line); | |
b456d40d CB |
977 | if (l < 0) |
978 | return log_error(0, "Failed to write cache"); | |
979 | if (l >= cache_size) | |
980 | return log_error(0, "Write to cache was truncated"); | |
981 | ||
1f5596dd CB |
982 | cache += l; |
983 | cache_size -= l; | |
984 | total_len += l; | |
985 | } | |
986 | } | |
987 | ||
988 | if (is_s390x) { | |
989 | __do_free char *origcache = d->buf; | |
990 | ssize_t l; | |
991 | ||
992 | d->buf = malloc(d->buflen); | |
993 | if (!d->buf) { | |
994 | d->buf = move_ptr(origcache); | |
995 | return 0; | |
996 | } | |
997 | ||
998 | cache = d->buf; | |
999 | cache_size = d->buflen; | |
1000 | total_len = 0; | |
1001 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
1002 | if (l < 0 || l >= cache_size) | |
1003 | return 0; | |
1004 | ||
1005 | cache_size -= l; | |
1006 | cache += l; | |
1007 | total_len += l; | |
1008 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
1009 | if (l < 0 || l >= cache_size) | |
1010 | return 0; | |
1011 | ||
1012 | cache_size -= l; | |
1013 | cache += l; | |
1014 | total_len += l; | |
1015 | l = snprintf(cache, cache_size, "%s", origcache); | |
1016 | if (l < 0 || l >= cache_size) | |
1017 | return 0; | |
1018 | total_len += l; | |
1019 | } | |
1020 | ||
1021 | d->cached = 1; | |
1022 | d->size = total_len; | |
d0031abf CB |
1023 | if (total_len > size) |
1024 | total_len = size; | |
1f5596dd CB |
1025 | |
1026 | /* read from off 0 */ | |
1027 | memcpy(buf, d->buf, total_len); | |
d0031abf | 1028 | |
1f5596dd CB |
1029 | return total_len; |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Returns 0 on success. | |
1034 | * It is the caller's responsibility to free `return_usage`, unless this | |
1035 | * function returns an error. | |
1036 | */ | |
1037 | int read_cpuacct_usage_all(char *cg, char *cpuset, | |
1038 | struct cpuacct_usage **return_usage, int *size) | |
1039 | { | |
1040 | __do_free char *usage_str = NULL; | |
1041 | __do_free struct cpuacct_usage *cpu_usage = NULL; | |
9ce186dc | 1042 | int i = 0, j = 0, read_pos = 0, read_cnt = 0; |
8b6987a2 | 1043 | int cpucount; |
9ce186dc | 1044 | int ret; |
1f5596dd CB |
1045 | int cg_cpu; |
1046 | uint64_t cg_user, cg_system; | |
1047 | int64_t ticks_per_sec; | |
1048 | ||
1049 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
1f5596dd | 1050 | if (ticks_per_sec < 0 && errno == EINVAL) { |
8b6987a2 | 1051 | lxcfs_debug("%m - Failed to determine number of ticks per second"); |
1f5596dd CB |
1052 | return -1; |
1053 | } | |
1054 | ||
f9434b9a | 1055 | cpucount = get_nprocs_conf(); |
1f5596dd CB |
1056 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); |
1057 | if (!cpu_usage) | |
1058 | return -ENOMEM; | |
1059 | ||
1060 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); | |
1061 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
8b6987a2 CB |
1062 | char *sep = " \t\n"; |
1063 | char *tok; | |
1f5596dd | 1064 | |
8b6987a2 CB |
1065 | /* Read cpuacct.usage_percpu instead. */ |
1066 | lxcfs_debug("Falling back to cpuacct.usage_percpu"); | |
1f5596dd CB |
1067 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) |
1068 | return -1; | |
1f5596dd | 1069 | |
8b6987a2 CB |
1070 | lxc_iterate_parts(tok, usage_str, sep) { |
1071 | uint64_t percpu_user; | |
1072 | ||
1073 | if (i >= cpucount) | |
1074 | break; | |
1f5596dd | 1075 | |
8b6987a2 CB |
1076 | tok = trim_whitespace_in_place(tok); |
1077 | ret = safe_uint64(tok, &percpu_user, 10); | |
1078 | if (ret) | |
1079 | return -1; | |
1f5596dd | 1080 | |
8b6987a2 CB |
1081 | /* Convert the time from nanoseconds to USER_HZ */ |
1082 | cpu_usage[i].user = percpu_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1083 | cpu_usage[i].system = cpu_usage[i].user; | |
1f5596dd | 1084 | i++; |
8b6987a2 | 1085 | lxcfs_debug("cpu%d with time %s", i, tok); |
1f5596dd | 1086 | } |
8b6987a2 CB |
1087 | } else { |
1088 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) | |
1089 | return log_error(-1, "read_cpuacct_usage_all reading first line from %s/cpuacct.usage_all failed", cg); | |
1f5596dd | 1090 | |
8b6987a2 | 1091 | read_pos += read_cnt; |
1f5596dd | 1092 | |
8b6987a2 CB |
1093 | for (i = 0, j = 0; i < cpucount; i++) { |
1094 | ret = sscanf(usage_str + read_pos, | |
1095 | "%d %" PRIu64 " %" PRIu64 "\n%n", &cg_cpu, | |
1096 | &cg_user, &cg_system, &read_cnt); | |
1f5596dd | 1097 | |
8b6987a2 CB |
1098 | if (ret == EOF) |
1099 | break; | |
1f5596dd | 1100 | |
8b6987a2 CB |
1101 | if (ret != 3) |
1102 | return log_error(-EINVAL, "Failed to parse cpuacct.usage_all line %s from cgroup %s", | |
1103 | usage_str + read_pos, cg); | |
1f5596dd | 1104 | |
8b6987a2 | 1105 | read_pos += read_cnt; |
1f5596dd | 1106 | |
8b6987a2 CB |
1107 | /* Convert the time from nanoseconds to USER_HZ */ |
1108 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1109 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1110 | j++; | |
1111 | } | |
1f5596dd CB |
1112 | } |
1113 | ||
1114 | *return_usage = move_ptr(cpu_usage); | |
1115 | *size = cpucount; | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
1120 | { | |
9d7fc1a3 | 1121 | __do_free struct cg_proc_stat_head *h; |
1f5596dd | 1122 | |
9d7fc1a3 CB |
1123 | h = zalloc(sizeof(struct cg_proc_stat_head)); |
1124 | if (!h) | |
1125 | return false; | |
1f5596dd | 1126 | |
9d7fc1a3 CB |
1127 | if (pthread_rwlock_init(&h->lock, NULL)) |
1128 | return false; | |
1129 | ||
1130 | h->lastcheck = time(NULL); | |
1f5596dd | 1131 | |
9d7fc1a3 | 1132 | *head = move_ptr(h); |
1f5596dd CB |
1133 | return true; |
1134 | } | |
1135 | ||
4ec5c9da | 1136 | bool init_cpuview(void) |
1f5596dd CB |
1137 | { |
1138 | int i; | |
1139 | ||
1140 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
1141 | proc_stat_history[i] = NULL; | |
1142 | ||
1143 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1144 | if (!cpuview_init_head(&proc_stat_history[i])) | |
1145 | goto err; | |
1146 | } | |
1147 | ||
1148 | return true; | |
1149 | ||
1150 | err: | |
1151 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1152 | if (proc_stat_history[i]) | |
1153 | free_disarm(proc_stat_history[i]); | |
1154 | } | |
1155 | ||
1156 | return false; | |
1157 | } | |
1158 | ||
1f5596dd CB |
1159 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
1160 | { | |
905769cd | 1161 | struct cg_proc_stat *node; |
1f5596dd CB |
1162 | |
1163 | if (head->next) { | |
1164 | node = head->next; | |
1165 | ||
1166 | for (;;) { | |
905769cd | 1167 | struct cg_proc_stat *cur = node; |
1f5596dd | 1168 | node = node->next; |
905769cd | 1169 | free_proc_stat_node(cur); |
1f5596dd CB |
1170 | if (!node) |
1171 | break; | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | pthread_rwlock_destroy(&head->lock); | |
1176 | free_disarm(head); | |
1177 | } | |
1178 | ||
4ec5c9da | 1179 | void free_cpuview(void) |
1f5596dd | 1180 | { |
4ec5c9da | 1181 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) |
1f5596dd CB |
1182 | if (proc_stat_history[i]) |
1183 | cpuview_free_head(proc_stat_history[i]); | |
1f5596dd | 1184 | } |