]>
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> | |
1f5596dd CB |
13 | #include <inttypes.h> |
14 | #include <libgen.h> | |
15 | #include <pthread.h> | |
16 | #include <sched.h> | |
17 | #include <stdarg.h> | |
18 | #include <stdbool.h> | |
19 | #include <stdint.h> | |
20 | #include <stdio.h> | |
21 | #include <stdlib.h> | |
22 | #include <string.h> | |
23 | #include <time.h> | |
24 | #include <unistd.h> | |
25 | #include <wait.h> | |
26 | #include <linux/magic.h> | |
27 | #include <linux/sched.h> | |
28 | #include <sys/epoll.h> | |
29 | #include <sys/mman.h> | |
30 | #include <sys/mount.h> | |
31 | #include <sys/param.h> | |
32 | #include <sys/socket.h> | |
33 | #include <sys/syscall.h> | |
34 | #include <sys/sysinfo.h> | |
35 | #include <sys/vfs.h> | |
36 | ||
e01afbb7 CB |
37 | #include "proc_cpuview.h" |
38 | ||
1f5596dd | 39 | #include "bindings.h" |
1f5596dd CB |
40 | #include "cgroup_fuse.h" |
41 | #include "cpuset_parse.h" | |
42 | #include "cgroups/cgroup.h" | |
43 | #include "cgroups/cgroup_utils.h" | |
44 | #include "memory_utils.h" | |
4ec5c9da | 45 | #include "proc_loadavg.h" |
1f5596dd CB |
46 | #include "utils.h" |
47 | ||
1f5596dd CB |
48 | /* Data for CPU view */ |
49 | struct cg_proc_stat { | |
50 | char *cg; | |
ce617d73 CB |
51 | struct cpuacct_usage *usage; /* Real usage as read from the host's /proc/stat. */ |
52 | struct cpuacct_usage *view; /* Usage stats reported to the container. */ | |
1f5596dd | 53 | int cpu_count; |
ce617d73 | 54 | pthread_mutex_t lock; /* For node manipulation. */ |
1f5596dd CB |
55 | struct cg_proc_stat *next; |
56 | }; | |
57 | ||
58 | struct cg_proc_stat_head { | |
59 | struct cg_proc_stat *next; | |
60 | time_t lastcheck; | |
61 | ||
62 | /* | |
63 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
64 | */ | |
65 | pthread_rwlock_t lock; | |
66 | }; | |
67 | ||
68 | #define CPUVIEW_HASH_SIZE 100 | |
69 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
70 | ||
b456d40d CB |
71 | static void reset_proc_stat_node(struct cg_proc_stat *node, |
72 | struct cpuacct_usage *usage, int cpu_count) | |
1f5596dd | 73 | { |
1f5596dd CB |
74 | lxcfs_debug("Resetting stat node for %s\n", node->cg); |
75 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
76 | ||
b456d40d | 77 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
78 | node->view[i].user = 0; |
79 | node->view[i].system = 0; | |
80 | node->view[i].idle = 0; | |
81 | } | |
82 | ||
83 | node->cpu_count = cpu_count; | |
84 | } | |
85 | ||
86 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) | |
87 | { | |
88 | __do_free struct cpuacct_usage *new_usage = NULL, *new_view = NULL; | |
89 | ||
90 | /* Allocate new memory */ | |
82d74a95 | 91 | new_usage = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
92 | if (!new_usage) |
93 | return false; | |
94 | ||
82d74a95 | 95 | new_view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
96 | if (!new_view) |
97 | return false; | |
98 | ||
99 | /* Copy existing data & initialize new elements */ | |
100 | for (int i = 0; i < cpu_count; i++) { | |
101 | if (i < node->cpu_count) { | |
82d74a95 CB |
102 | new_usage[i].user = node->usage[i].user; |
103 | new_usage[i].system = node->usage[i].system; | |
104 | new_usage[i].idle = node->usage[i].idle; | |
105 | ||
106 | new_view[i].user = node->view[i].user; | |
107 | new_view[i].system = node->view[i].system; | |
108 | new_view[i].idle = node->view[i].idle; | |
1f5596dd CB |
109 | } |
110 | } | |
111 | ||
112 | free(node->usage); | |
113 | node->usage = move_ptr(new_usage); | |
114 | ||
115 | free(node->view); | |
116 | node->view = move_ptr(new_view); | |
117 | node->cpu_count = cpu_count; | |
118 | ||
119 | return true; | |
120 | } | |
121 | ||
4ec5c9da CB |
122 | static void free_proc_stat_node(struct cg_proc_stat *node) |
123 | { | |
6a4dceb1 CB |
124 | if (node) { |
125 | /* | |
126 | * We're abusing the usage pointer to indicate that | |
127 | * pthread_mutex_init() was successful. Don't judge me. | |
128 | */ | |
129 | if (node->usage) | |
130 | pthread_mutex_destroy(&node->lock); | |
131 | free_disarm(node->cg); | |
132 | free_disarm(node->usage); | |
133 | free_disarm(node->view); | |
134 | free_disarm(node); | |
135 | } | |
4ec5c9da CB |
136 | } |
137 | ||
6a4dceb1 CB |
138 | define_cleanup_function(struct cg_proc_stat *, free_proc_stat_node); |
139 | ||
1f5596dd CB |
140 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
141 | { | |
0d129671 CB |
142 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *new = new_node; |
143 | struct cg_proc_stat *rv = new_node; | |
144 | int hash = calc_hash(new->cg) % CPUVIEW_HASH_SIZE; | |
1f5596dd | 145 | struct cg_proc_stat_head *head = proc_stat_history[hash]; |
0d129671 | 146 | struct cg_proc_stat *cur; |
1f5596dd CB |
147 | |
148 | pthread_rwlock_wrlock(&head->lock); | |
149 | ||
150 | if (!head->next) { | |
0d129671 | 151 | head->next = move_ptr(new); |
164acda7 | 152 | goto out_rwlock_unlock; |
1f5596dd CB |
153 | } |
154 | ||
0d129671 | 155 | cur = head->next; |
1f5596dd CB |
156 | |
157 | for (;;) { | |
0d129671 CB |
158 | /* |
159 | * The node to be added is already present in the list, so | |
160 | * free the newly allocated one and return the one we found. | |
161 | */ | |
162 | if (strcmp(cur->cg, new->cg) == 0) { | |
163 | rv = cur; | |
164acda7 | 164 | goto out_rwlock_unlock; |
1f5596dd CB |
165 | } |
166 | ||
0d129671 CB |
167 | /* Keep walking. */ |
168 | if (cur->next) { | |
169 | cur = cur->next; | |
1f5596dd CB |
170 | continue; |
171 | } | |
172 | ||
0d129671 CB |
173 | /* Add new node to end of list. */ |
174 | cur->next = move_ptr(new); | |
164acda7 | 175 | goto out_rwlock_unlock; |
1f5596dd CB |
176 | } |
177 | ||
164acda7 | 178 | out_rwlock_unlock: |
1f5596dd | 179 | pthread_rwlock_unlock(&head->lock); |
0d129671 | 180 | return move_ptr(rv); |
1f5596dd CB |
181 | } |
182 | ||
6a4dceb1 CB |
183 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, |
184 | int cpu_count, const char *cg) | |
1f5596dd | 185 | { |
6a4dceb1 CB |
186 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *node = NULL; |
187 | __do_free struct cpuacct_usage *new_usage = NULL; | |
1f5596dd | 188 | |
6a4dceb1 | 189 | node = zalloc(sizeof(struct cg_proc_stat)); |
1f5596dd | 190 | if (!node) |
6a4dceb1 | 191 | return NULL; |
1f5596dd | 192 | |
6a4dceb1 | 193 | node->cg = strdup(cg); |
1f5596dd | 194 | if (!node->cg) |
6a4dceb1 | 195 | return NULL; |
1f5596dd | 196 | |
6a4dceb1 CB |
197 | new_usage = memdup(usage, sizeof(struct cpuacct_usage) * cpu_count); |
198 | if (!new_usage) | |
199 | return NULL; | |
1f5596dd | 200 | |
6a4dceb1 | 201 | node->view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd | 202 | if (!node->view) |
6a4dceb1 | 203 | return NULL; |
1f5596dd CB |
204 | |
205 | node->cpu_count = cpu_count; | |
1f5596dd | 206 | |
6a4dceb1 CB |
207 | if (pthread_mutex_init(&node->lock, NULL)) |
208 | return NULL; | |
209 | /* | |
210 | * We're abusing the usage pointer to indicate that | |
211 | * pthread_mutex_init() was successful. Don't judge me. | |
212 | */ | |
213 | node->usage = move_ptr(new_usage); | |
1f5596dd | 214 | |
6a4dceb1 | 215 | return move_ptr(node); |
1f5596dd CB |
216 | } |
217 | ||
2d00d04c CB |
218 | static bool cgroup_supports(const char *controller, const char *cgroup, |
219 | const char *file) | |
4ec5c9da | 220 | { |
2c990b1d CB |
221 | __do_free char *path = NULL; |
222 | int cfd; | |
4ec5c9da CB |
223 | |
224 | cfd = get_cgroup_fd(controller); | |
225 | if (cfd < 0) | |
226 | return false; | |
227 | ||
925d5849 | 228 | path = must_make_path_relative(cgroup, file, NULL); |
2d00d04c | 229 | return faccessat(cfd, path, F_OK, 0) == 0; |
4ec5c9da CB |
230 | } |
231 | ||
1f5596dd CB |
232 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
233 | { | |
b456d40d | 234 | struct cg_proc_stat *first = NULL; |
1f5596dd | 235 | |
b456d40d | 236 | for (struct cg_proc_stat *prev = NULL; node; ) { |
2d00d04c | 237 | if (!cgroup_supports("cpu", node->cg, "cpu.shares")) { |
d5e34313 | 238 | struct cg_proc_stat *cur = node; |
1f5596dd CB |
239 | |
240 | if (prev) | |
241 | prev->next = node->next; | |
242 | else | |
243 | first = node->next; | |
244 | ||
245 | node = node->next; | |
d5e34313 CB |
246 | lxcfs_debug("Removing stat node for %s\n", cur); |
247 | ||
248 | free_proc_stat_node(cur); | |
1f5596dd CB |
249 | } else { |
250 | if (!first) | |
251 | first = node; | |
252 | prev = node; | |
253 | node = node->next; | |
254 | } | |
255 | } | |
256 | ||
257 | return first; | |
258 | } | |
259 | ||
260 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
261 | static void prune_proc_stat_history(void) | |
262 | { | |
1f5596dd CB |
263 | time_t now = time(NULL); |
264 | ||
b456d40d | 265 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) { |
1f5596dd CB |
266 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
267 | ||
268 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
269 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
270 | return; | |
271 | } | |
272 | ||
273 | if (proc_stat_history[i]->next) { | |
274 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
275 | proc_stat_history[i]->lastcheck = now; | |
276 | } | |
277 | ||
278 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
279 | } | |
280 | } | |
281 | ||
282 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, | |
283 | const char *cg) | |
284 | { | |
285 | struct cg_proc_stat *node; | |
286 | ||
287 | pthread_rwlock_rdlock(&head->lock); | |
288 | ||
289 | if (!head->next) { | |
290 | pthread_rwlock_unlock(&head->lock); | |
291 | return NULL; | |
292 | } | |
293 | ||
294 | node = head->next; | |
295 | ||
296 | do { | |
297 | if (strcmp(cg, node->cg) == 0) | |
298 | goto out; | |
299 | } while ((node = node->next)); | |
300 | ||
301 | node = NULL; | |
302 | ||
303 | out: | |
304 | pthread_rwlock_unlock(&head->lock); | |
305 | prune_proc_stat_history(); | |
306 | return node; | |
307 | } | |
308 | ||
692f48eb CB |
309 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, |
310 | int cpu_count, const char *cg) | |
1f5596dd CB |
311 | { |
312 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
313 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
314 | struct cg_proc_stat *node; | |
315 | ||
316 | node = find_proc_stat_node(head, cg); | |
1f5596dd CB |
317 | if (!node) { |
318 | node = new_proc_stat_node(usage, cpu_count, cg); | |
319 | if (!node) | |
320 | return NULL; | |
321 | ||
322 | node = add_proc_stat_node(node); | |
323 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
324 | } | |
325 | ||
326 | pthread_mutex_lock(&node->lock); | |
327 | ||
ce089f10 CB |
328 | /* |
329 | * If additional CPUs on the host have been enabled, CPU usage counter | |
330 | * arrays have to be expanded. | |
331 | */ | |
1f5596dd CB |
332 | if (node->cpu_count < cpu_count) { |
333 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
ce089f10 | 334 | node->cpu_count, cpu_count, cg); |
1f5596dd CB |
335 | |
336 | if (!expand_proc_stat_node(node, cpu_count)) { | |
337 | pthread_mutex_unlock(&node->lock); | |
b456d40d | 338 | return log_debug(NULL, "Unable to expand stat node %d->%d for %s", node->cpu_count, cpu_count, cg); |
1f5596dd CB |
339 | } |
340 | } | |
341 | ||
342 | return node; | |
343 | } | |
344 | ||
2b8eff1d CB |
345 | static void add_cpu_usage(uint64_t *surplus, struct cpuacct_usage *usage, |
346 | uint64_t *counter, uint64_t threshold) | |
1f5596dd | 347 | { |
1ba088ae | 348 | uint64_t free_space, to_add; |
1f5596dd CB |
349 | |
350 | free_space = threshold - usage->user - usage->system; | |
351 | ||
352 | if (free_space > usage->idle) | |
353 | free_space = usage->idle; | |
354 | ||
8206874a CB |
355 | if (free_space > *surplus) |
356 | to_add = *surplus; | |
357 | else | |
358 | to_add = free_space; | |
1f5596dd CB |
359 | |
360 | *counter += to_add; | |
361 | usage->idle -= to_add; | |
362 | *surplus -= to_add; | |
363 | } | |
364 | ||
1ba088ae CB |
365 | static uint64_t diff_cpu_usage(struct cpuacct_usage *older, |
366 | struct cpuacct_usage *newer, | |
367 | struct cpuacct_usage *diff, int cpu_count) | |
1f5596dd | 368 | { |
1ba088ae | 369 | uint64_t sum = 0; |
1f5596dd | 370 | |
b456d40d | 371 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
372 | if (!newer[i].online) |
373 | continue; | |
374 | ||
b456d40d CB |
375 | /* |
376 | * When cpuset is changed on the fly, the CPUs might get | |
377 | * reordered. We could either reset all counters, or check | |
378 | * that the substractions below will return expected results. | |
1f5596dd CB |
379 | */ |
380 | if (newer[i].user > older[i].user) | |
381 | diff[i].user = newer[i].user - older[i].user; | |
382 | else | |
383 | diff[i].user = 0; | |
384 | ||
385 | if (newer[i].system > older[i].system) | |
386 | diff[i].system = newer[i].system - older[i].system; | |
387 | else | |
388 | diff[i].system = 0; | |
389 | ||
390 | if (newer[i].idle > older[i].idle) | |
391 | diff[i].idle = newer[i].idle - older[i].idle; | |
392 | else | |
393 | diff[i].idle = 0; | |
394 | ||
395 | sum += diff[i].user; | |
396 | sum += diff[i].system; | |
397 | sum += diff[i].idle; | |
398 | } | |
399 | ||
400 | return sum; | |
401 | } | |
402 | ||
403 | /* | |
b456d40d CB |
404 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or |
405 | * `cpu.cfs_period_us`, depending on `param`. Parameter value is returned | |
92264841 | 406 | * through `value`. |
1f5596dd CB |
407 | */ |
408 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
409 | { | |
410 | __do_free char *str = NULL; | |
48f6862e | 411 | char file[STRLITERALLEN("cpu.cfs_period_us") + 1]; |
9844eea7 | 412 | bool first = true; |
48f6862e | 413 | int ret; |
1f5596dd | 414 | |
48f6862e | 415 | if (pure_unified_layout(cgroup_ops)) { |
9844eea7 | 416 | first = !strcmp(param, "quota"); |
48f6862e CB |
417 | ret = snprintf(file, sizeof(file), "cpu.max"); |
418 | } else { | |
419 | ret = snprintf(file, sizeof(file), "cpu.cfs_%s_us", param); | |
9844eea7 | 420 | } |
48f6862e | 421 | if (ret < 0 || (size_t)ret >= sizeof(file)) |
1f5596dd CB |
422 | return false; |
423 | ||
48f6862e | 424 | if (!cgroup_ops->get(cgroup_ops, "cpu", cg, file, &str)) |
1f5596dd CB |
425 | return false; |
426 | ||
48f6862e | 427 | return sscanf(str, first ? "%" PRId64 : "%*d %" PRId64, value) == 1; |
1f5596dd CB |
428 | } |
429 | ||
430 | /* | |
431 | * Return the exact number of visible CPUs based on CPU quotas. | |
432 | * If there is no quota set, zero is returned. | |
433 | */ | |
434 | static double exact_cpu_count(const char *cg) | |
435 | { | |
436 | double rv; | |
437 | int nprocs; | |
438 | int64_t cfs_quota, cfs_period; | |
439 | ||
c602a0d0 CB |
440 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
441 | return 0; | |
442 | ||
443 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
444 | return 0; | |
1f5596dd CB |
445 | |
446 | if (cfs_quota <= 0 || cfs_period <= 0) | |
447 | return 0; | |
448 | ||
449 | rv = (double)cfs_quota / (double)cfs_period; | |
450 | ||
451 | nprocs = get_nprocs(); | |
452 | ||
453 | if (rv > nprocs) | |
454 | rv = nprocs; | |
455 | ||
456 | return rv; | |
457 | } | |
458 | ||
459 | /* | |
460 | * Return the maximum number of visible CPUs based on CPU quotas. | |
461 | * If there is no quota set, zero is returned. | |
462 | */ | |
4ec5c9da | 463 | int max_cpu_count(const char *cg) |
1f5596dd | 464 | { |
700dd417 | 465 | __do_free char *cpuset = NULL; |
1f5596dd CB |
466 | int rv, nprocs; |
467 | int64_t cfs_quota, cfs_period; | |
468 | int nr_cpus_in_cpuset = 0; | |
1f5596dd | 469 | |
921bdfdb CB |
470 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
471 | return 0; | |
472 | ||
473 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
474 | return 0; | |
1f5596dd CB |
475 | |
476 | cpuset = get_cpuset(cg); | |
477 | if (cpuset) | |
478 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
479 | ||
921bdfdb | 480 | if (cfs_quota <= 0 || cfs_period <= 0) { |
1f5596dd CB |
481 | if (nr_cpus_in_cpuset > 0) |
482 | return nr_cpus_in_cpuset; | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | rv = cfs_quota / cfs_period; | |
488 | ||
921bdfdb CB |
489 | /* |
490 | * In case quota/period does not yield a whole number, add one CPU for | |
1f5596dd CB |
491 | * the remainder. |
492 | */ | |
493 | if ((cfs_quota % cfs_period) > 0) | |
494 | rv += 1; | |
495 | ||
496 | nprocs = get_nprocs(); | |
1f5596dd CB |
497 | if (rv > nprocs) |
498 | rv = nprocs; | |
499 | ||
921bdfdb | 500 | /* Use min value in cpu quota and cpuset. */ |
1f5596dd CB |
501 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) |
502 | rv = nr_cpus_in_cpuset; | |
503 | ||
504 | return rv; | |
505 | } | |
506 | ||
507 | int cpuview_proc_stat(const char *cg, const char *cpuset, | |
508 | struct cpuacct_usage *cg_cpu_usage, int cg_cpu_usage_size, | |
509 | FILE *f, char *buf, size_t buf_size) | |
510 | { | |
511 | __do_free char *line = NULL; | |
512 | __do_free struct cpuacct_usage *diff = NULL; | |
4f18a602 | 513 | size_t linelen = 0, total_len = 0; |
1f5596dd CB |
514 | int curcpu = -1; /* cpu numbering starts at 0 */ |
515 | int physcpu, i; | |
39f231da | 516 | int cpu_cnt = 0; |
2b8eff1d CB |
517 | uint64_t user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, |
518 | softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
519 | uint64_t user_sum = 0, system_sum = 0, idle_sum = 0; | |
520 | uint64_t user_surplus = 0, system_surplus = 0; | |
39f231da | 521 | int nprocs, max_cpus; |
4f18a602 | 522 | ssize_t l; |
2b8eff1d | 523 | uint64_t total_sum, threshold; |
1f5596dd | 524 | struct cg_proc_stat *stat_node; |
1f5596dd | 525 | |
39f231da | 526 | nprocs = get_nprocs_conf(); |
1f5596dd CB |
527 | if (cg_cpu_usage_size < nprocs) |
528 | nprocs = cg_cpu_usage_size; | |
529 | ||
530 | /* Read all CPU stats and stop when we've encountered other lines */ | |
531 | while (getline(&line, &linelen, f) != -1) { | |
532 | int ret; | |
533 | char cpu_char[10]; /* That's a lot of cores */ | |
534 | uint64_t all_used, cg_used; | |
535 | ||
536 | if (strlen(line) == 0) | |
537 | continue; | |
538 | ||
539 | /* not a ^cpuN line containing a number N */ | |
540 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) | |
541 | break; | |
542 | ||
543 | if (sscanf(cpu_char, "%d", &physcpu) != 1) | |
544 | continue; | |
545 | ||
546 | if (physcpu >= cg_cpu_usage_size) | |
547 | continue; | |
548 | ||
fd65c77c CB |
549 | curcpu++; |
550 | cpu_cnt++; | |
1f5596dd CB |
551 | |
552 | if (!cpu_in_cpuset(physcpu, cpuset)) { | |
553 | for (i = curcpu; i <= physcpu; i++) | |
554 | cg_cpu_usage[i].online = false; | |
555 | continue; | |
556 | } | |
557 | ||
558 | if (curcpu < physcpu) { | |
559 | /* Some CPUs may be disabled */ | |
560 | for (i = curcpu; i < physcpu; i++) | |
561 | cg_cpu_usage[i].online = false; | |
562 | ||
563 | curcpu = physcpu; | |
564 | } | |
565 | ||
566 | cg_cpu_usage[curcpu].online = true; | |
567 | ||
2b8eff1d | 568 | ret = sscanf(line, "%*s %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 "lu", |
1f5596dd CB |
569 | &user, |
570 | &nice, | |
571 | &system, | |
572 | &idle, | |
573 | &iowait, | |
574 | &irq, | |
575 | &softirq, | |
576 | &steal, | |
577 | &guest, | |
578 | &guest_nice); | |
1f5596dd CB |
579 | if (ret != 10) |
580 | continue; | |
581 | ||
582 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
583 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
584 | ||
585 | if (all_used >= cg_used) { | |
586 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
587 | ||
588 | } else { | |
2b8eff1d CB |
589 | lxcfs_error("cpu%d from %s has unexpected cpu time: %" PRIu64 " in /proc/stat, %" PRIu64 " in cpuacct.usage_all; unable to determine idle time", |
590 | curcpu, cg, all_used, cg_used); | |
1f5596dd CB |
591 | cg_cpu_usage[curcpu].idle = idle; |
592 | } | |
593 | } | |
594 | ||
f9434b9a CB |
595 | /* Cannot use more CPUs than is available in cpuset. */ |
596 | max_cpus = max_cpu_count(cg); | |
597 | if (max_cpus > cpu_cnt || !max_cpus) | |
598 | max_cpus = cpu_cnt; | |
599 | ||
692f48eb | 600 | /* takes lock pthread_mutex_lock(&node->lock) */ |
1f5596dd | 601 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
b456d40d CB |
602 | if (!stat_node) |
603 | return log_error(0, "Failed to find/create stat node for %s", cg); | |
1f5596dd | 604 | |
b4572722 | 605 | diff = zalloc(sizeof(struct cpuacct_usage) * nprocs); |
700dd417 | 606 | if (!diff) |
08d61303 | 607 | goto out_pthread_mutex_unlock; |
1f5596dd CB |
608 | |
609 | /* | |
610 | * If the new values are LOWER than values stored in memory, it means | |
611 | * the cgroup has been reset/recreated and we should reset too. | |
612 | */ | |
613 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
614 | if (!cg_cpu_usage[curcpu].online) | |
615 | continue; | |
616 | ||
617 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
618 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
619 | ||
620 | break; | |
621 | } | |
622 | ||
623 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); | |
624 | ||
625 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
626 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
627 | ||
628 | if (!stat_node->usage[curcpu].online) | |
629 | continue; | |
630 | ||
631 | i++; | |
632 | ||
b4572722 | 633 | stat_node->usage[curcpu].user += diff[curcpu].user; |
1f5596dd | 634 | stat_node->usage[curcpu].system += diff[curcpu].system; |
b4572722 | 635 | stat_node->usage[curcpu].idle += diff[curcpu].idle; |
1f5596dd CB |
636 | |
637 | if (max_cpus > 0 && i >= max_cpus) { | |
b4572722 CB |
638 | user_surplus += diff[curcpu].user; |
639 | system_surplus += diff[curcpu].system; | |
1f5596dd CB |
640 | } |
641 | } | |
642 | ||
643 | /* Calculate usage counters of visible CPUs */ | |
644 | if (max_cpus > 0) { | |
2b8eff1d CB |
645 | uint64_t diff_user = 0; |
646 | uint64_t diff_system = 0; | |
647 | uint64_t diff_idle = 0; | |
648 | uint64_t max_diff_idle = 0; | |
649 | uint64_t max_diff_idle_index = 0; | |
1f5596dd | 650 | double exact_cpus; |
1f5596dd CB |
651 | /* threshold = maximum usage per cpu, including idle */ |
652 | threshold = total_sum / cpu_cnt * max_cpus; | |
653 | ||
654 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
655 | if (!stat_node->usage[curcpu].online) | |
656 | continue; | |
657 | ||
658 | i++; | |
659 | ||
660 | if (i == max_cpus) | |
661 | break; | |
662 | ||
663 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
664 | continue; | |
665 | ||
666 | /* Add user */ | |
667 | add_cpu_usage(&user_surplus, &diff[curcpu], | |
668 | &diff[curcpu].user, threshold); | |
669 | ||
670 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
671 | continue; | |
672 | ||
673 | /* If there is still room, add system */ | |
674 | add_cpu_usage(&system_surplus, &diff[curcpu], | |
675 | &diff[curcpu].system, threshold); | |
676 | } | |
677 | ||
678 | if (user_surplus > 0) | |
679 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
680 | if (system_surplus > 0) | |
681 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
682 | ||
683 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
684 | if (!stat_node->usage[curcpu].online) | |
685 | continue; | |
686 | ||
687 | i++; | |
688 | ||
689 | if (i == max_cpus) | |
690 | break; | |
691 | ||
b4572722 CB |
692 | stat_node->view[curcpu].user += diff[curcpu].user; |
693 | stat_node->view[curcpu].system += diff[curcpu].system; | |
694 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
1f5596dd | 695 | |
b4572722 CB |
696 | user_sum += stat_node->view[curcpu].user; |
697 | system_sum += stat_node->view[curcpu].system; | |
698 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd | 699 | |
b4572722 CB |
700 | diff_user += diff[curcpu].user; |
701 | diff_system += diff[curcpu].system; | |
702 | diff_idle += diff[curcpu].idle; | |
1f5596dd | 703 | if (diff[curcpu].idle > max_diff_idle) { |
b4572722 CB |
704 | max_diff_idle = diff[curcpu].idle; |
705 | max_diff_idle_index = curcpu; | |
1f5596dd CB |
706 | } |
707 | ||
708 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
709 | } | |
710 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
711 | ||
712 | /* revise cpu usage view to support partial cpu case. */ | |
713 | exact_cpus = exact_cpu_count(cg); | |
714 | if (exact_cpus < (double)max_cpus){ | |
1ba088ae | 715 | uint64_t delta = (uint64_t)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); |
1f5596dd CB |
716 | |
717 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
718 | lxcfs_v("delta: %lu\n", delta); | |
719 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
b4572722 CB |
720 | if (idle_sum > delta) |
721 | idle_sum = idle_sum - delta; | |
722 | else | |
723 | idle_sum = 0; | |
1f5596dd CB |
724 | lxcfs_v("idle_sum after: %lu\n", idle_sum); |
725 | ||
726 | curcpu = max_diff_idle_index; | |
727 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
b4572722 CB |
728 | if (stat_node->view[curcpu].idle > delta) |
729 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle - delta; | |
730 | else | |
731 | stat_node->view[curcpu].idle = 0; | |
1f5596dd CB |
732 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); |
733 | } | |
734 | } else { | |
735 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
736 | if (!stat_node->usage[curcpu].online) | |
737 | continue; | |
738 | ||
b4572722 CB |
739 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; |
740 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
741 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
1f5596dd | 742 | |
b4572722 CB |
743 | user_sum += stat_node->view[curcpu].user; |
744 | system_sum += stat_node->view[curcpu].system; | |
745 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd CB |
746 | } |
747 | } | |
748 | ||
749 | /* Render the file */ | |
750 | /* cpu-all */ | |
2b8eff1d CB |
751 | l = snprintf(buf, buf_size, |
752 | "cpu %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", | |
753 | user_sum, system_sum, idle_sum); | |
1f5596dd | 754 | lxcfs_v("cpu-all: %s\n", buf); |
692f48eb CB |
755 | if (l < 0) { |
756 | lxcfs_error("Failed to write cache"); | |
757 | total_len = 0; | |
758 | goto out_pthread_mutex_unlock; | |
759 | } | |
3cf1e562 | 760 | if ((size_t)l >= buf_size) { |
08d61303 CB |
761 | lxcfs_error("Write to cache was truncated"); |
762 | total_len = 0; | |
763 | goto out_pthread_mutex_unlock; | |
764 | } | |
1f5596dd CB |
765 | |
766 | buf += l; | |
767 | buf_size -= l; | |
768 | total_len += l; | |
769 | ||
770 | /* Render visible CPUs */ | |
771 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
772 | if (!stat_node->usage[curcpu].online) | |
773 | continue; | |
774 | ||
775 | i++; | |
776 | ||
777 | if (max_cpus > 0 && i == max_cpus) | |
778 | break; | |
779 | ||
2b8eff1d CB |
780 | l = snprintf(buf, buf_size, "cpu%d %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", |
781 | i, | |
782 | stat_node->view[curcpu].user, | |
783 | stat_node->view[curcpu].system, | |
784 | stat_node->view[curcpu].idle); | |
1f5596dd | 785 | lxcfs_v("cpu: %s\n", buf); |
692f48eb CB |
786 | if (l < 0) { |
787 | lxcfs_error("Failed to write cache"); | |
788 | total_len = 0; | |
789 | goto out_pthread_mutex_unlock; | |
790 | } | |
3cf1e562 | 791 | if ((size_t)l >= buf_size) { |
692f48eb CB |
792 | lxcfs_error("Write to cache was truncated"); |
793 | total_len = 0; | |
794 | goto out_pthread_mutex_unlock; | |
795 | } | |
1f5596dd CB |
796 | |
797 | buf += l; | |
798 | buf_size -= l; | |
799 | total_len += l; | |
800 | } | |
801 | ||
802 | /* Pass the rest of /proc/stat, start with the last line read */ | |
803 | l = snprintf(buf, buf_size, "%s", line); | |
692f48eb CB |
804 | if (l < 0) { |
805 | lxcfs_error("Failed to write cache"); | |
806 | total_len = 0; | |
807 | goto out_pthread_mutex_unlock; | |
808 | } | |
3cf1e562 | 809 | if ((size_t)l >= buf_size) { |
692f48eb CB |
810 | lxcfs_error("Write to cache was truncated"); |
811 | total_len = 0; | |
812 | goto out_pthread_mutex_unlock; | |
813 | } | |
1f5596dd CB |
814 | |
815 | buf += l; | |
816 | buf_size -= l; | |
817 | total_len += l; | |
818 | ||
819 | /* Pass the rest of the host's /proc/stat */ | |
820 | while (getline(&line, &linelen, f) != -1) { | |
821 | l = snprintf(buf, buf_size, "%s", line); | |
692f48eb CB |
822 | if (l < 0) { |
823 | lxcfs_error("Failed to write cache"); | |
824 | total_len = 0; | |
825 | goto out_pthread_mutex_unlock; | |
826 | } | |
3cf1e562 | 827 | if ((size_t)l >= buf_size) { |
692f48eb CB |
828 | lxcfs_error("Write to cache was truncated"); |
829 | total_len = 0; | |
830 | goto out_pthread_mutex_unlock; | |
831 | } | |
b456d40d | 832 | |
1f5596dd CB |
833 | buf += l; |
834 | buf_size -= l; | |
835 | total_len += l; | |
836 | } | |
837 | ||
692f48eb | 838 | out_pthread_mutex_unlock: |
1f5596dd CB |
839 | if (stat_node) |
840 | pthread_mutex_unlock(&stat_node->lock); | |
b456d40d | 841 | |
1f5596dd CB |
842 | return total_len; |
843 | } | |
844 | ||
845 | /* | |
846 | * check whether this is a '^processor" line in /proc/cpuinfo | |
847 | */ | |
b456d40d | 848 | static inline bool is_processor_line(const char *line) |
1f5596dd CB |
849 | { |
850 | int cpu; | |
b456d40d | 851 | return sscanf(line, "processor : %d", &cpu) == 1; |
1f5596dd CB |
852 | } |
853 | ||
b456d40d | 854 | static inline bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1f5596dd CB |
855 | { |
856 | int cpu; | |
c539526c CB |
857 | |
858 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
859 | return cpu_in_cpuset(cpu, cpuset); | |
860 | ||
861 | return false; | |
1f5596dd CB |
862 | } |
863 | ||
864 | int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
865 | struct fuse_file_info *fi) | |
866 | { | |
867 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; | |
757a63e7 | 868 | __do_free void *fopen_cache = NULL; |
1f5596dd CB |
869 | __do_fclose FILE *f = NULL; |
870 | struct fuse_context *fc = fuse_get_context(); | |
0274438c | 871 | struct lxcfs_opts *opts = (struct lxcfs_opts *)fc->private_data; |
99b183fb | 872 | struct file_info *d = INTTYPE_TO_PTR(fi->fh); |
1f5596dd CB |
873 | size_t linelen = 0, total_len = 0; |
874 | bool am_printing = false, firstline = true, is_s390x = false; | |
875 | int curcpu = -1, cpu, max_cpus = 0; | |
876 | bool use_view; | |
877 | char *cache = d->buf; | |
878 | size_t cache_size = d->buflen; | |
879 | ||
f9434b9a | 880 | if (offset) { |
3cf1e562 | 881 | size_t left; |
1f5596dd CB |
882 | |
883 | if (offset > d->size) | |
884 | return -EINVAL; | |
885 | ||
886 | if (!d->cached) | |
887 | return 0; | |
888 | ||
889 | left = d->size - offset; | |
890 | total_len = left > size ? size: left; | |
891 | memcpy(buf, cache + offset, total_len); | |
892 | ||
893 | return total_len; | |
894 | } | |
895 | ||
896 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
897 | if (initpid <= 1 || is_shared_pidns(initpid)) | |
898 | initpid = fc->pid; | |
b456d40d | 899 | |
1f5596dd CB |
900 | cg = get_pid_cgroup(initpid, "cpuset"); |
901 | if (!cg) | |
902 | return read_file_fuse("proc/cpuinfo", buf, size, d); | |
903 | prune_init_slice(cg); | |
904 | ||
905 | cpuset = get_cpuset(cg); | |
906 | if (!cpuset) | |
907 | return 0; | |
908 | ||
8044f626 | 909 | if (cgroup_ops->can_use_cpuview(cgroup_ops) && opts && opts->use_cfs) |
0274438c | 910 | use_view = true; |
8044f626 CB |
911 | else |
912 | use_view = false; | |
1f5596dd CB |
913 | if (use_view) |
914 | max_cpus = max_cpu_count(cg); | |
915 | ||
757a63e7 | 916 | f = fopen_cached("/proc/cpuinfo", "re", &fopen_cache); |
1f5596dd CB |
917 | if (!f) |
918 | return 0; | |
919 | ||
920 | while (getline(&line, &linelen, f) != -1) { | |
921 | ssize_t l; | |
922 | if (firstline) { | |
923 | firstline = false; | |
924 | if (strstr(line, "IBM/S390") != NULL) { | |
925 | is_s390x = true; | |
926 | am_printing = true; | |
927 | continue; | |
928 | } | |
929 | } | |
b456d40d | 930 | |
1f5596dd CB |
931 | if (strncmp(line, "# processors:", 12) == 0) |
932 | continue; | |
b456d40d | 933 | |
1f5596dd | 934 | if (is_processor_line(line)) { |
d0031abf | 935 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 936 | break; |
b456d40d | 937 | |
1f5596dd CB |
938 | am_printing = cpuline_in_cpuset(line, cpuset); |
939 | if (am_printing) { | |
d0031abf | 940 | curcpu++; |
1f5596dd | 941 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); |
b456d40d CB |
942 | if (l < 0) |
943 | return log_error(0, "Failed to write cache"); | |
3cf1e562 | 944 | if ((size_t)l >= cache_size) |
b456d40d | 945 | return log_error(0, "Write to cache was truncated"); |
1f5596dd CB |
946 | cache += l; |
947 | cache_size -= l; | |
948 | total_len += l; | |
949 | } | |
950 | continue; | |
951 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { | |
952 | char *p; | |
b456d40d | 953 | |
d0031abf | 954 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 955 | break; |
b456d40d | 956 | |
1f5596dd CB |
957 | if (!cpu_in_cpuset(cpu, cpuset)) |
958 | continue; | |
b456d40d | 959 | |
1f5596dd CB |
960 | curcpu ++; |
961 | p = strchr(line, ':'); | |
962 | if (!p || !*p) | |
963 | return 0; | |
964 | p++; | |
b456d40d | 965 | |
1f5596dd | 966 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
b456d40d CB |
967 | if (l < 0) |
968 | return log_error(0, "Failed to write cache"); | |
3cf1e562 | 969 | if ((size_t)l >= cache_size) |
b456d40d CB |
970 | return log_error(0, "Write to cache was truncated"); |
971 | ||
1f5596dd CB |
972 | cache += l; |
973 | cache_size -= l; | |
974 | total_len += l; | |
975 | continue; | |
976 | ||
977 | } | |
978 | if (am_printing) { | |
979 | l = snprintf(cache, cache_size, "%s", line); | |
b456d40d CB |
980 | if (l < 0) |
981 | return log_error(0, "Failed to write cache"); | |
3cf1e562 | 982 | if ((size_t)l >= cache_size) |
b456d40d CB |
983 | return log_error(0, "Write to cache was truncated"); |
984 | ||
1f5596dd CB |
985 | cache += l; |
986 | cache_size -= l; | |
987 | total_len += l; | |
988 | } | |
989 | } | |
990 | ||
991 | if (is_s390x) { | |
992 | __do_free char *origcache = d->buf; | |
993 | ssize_t l; | |
994 | ||
995 | d->buf = malloc(d->buflen); | |
996 | if (!d->buf) { | |
997 | d->buf = move_ptr(origcache); | |
998 | return 0; | |
999 | } | |
1000 | ||
1001 | cache = d->buf; | |
1002 | cache_size = d->buflen; | |
1003 | total_len = 0; | |
1004 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
3cf1e562 | 1005 | if (l < 0 || (size_t)l >= cache_size) |
1f5596dd CB |
1006 | return 0; |
1007 | ||
1008 | cache_size -= l; | |
1009 | cache += l; | |
1010 | total_len += l; | |
1011 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
3cf1e562 | 1012 | if (l < 0 || (size_t)l >= cache_size) |
1f5596dd CB |
1013 | return 0; |
1014 | ||
1015 | cache_size -= l; | |
1016 | cache += l; | |
1017 | total_len += l; | |
1018 | l = snprintf(cache, cache_size, "%s", origcache); | |
3cf1e562 | 1019 | if (l < 0 || (size_t)l >= cache_size) |
1f5596dd CB |
1020 | return 0; |
1021 | total_len += l; | |
1022 | } | |
1023 | ||
1024 | d->cached = 1; | |
1025 | d->size = total_len; | |
d0031abf CB |
1026 | if (total_len > size) |
1027 | total_len = size; | |
1f5596dd CB |
1028 | |
1029 | /* read from off 0 */ | |
1030 | memcpy(buf, d->buf, total_len); | |
d0031abf | 1031 | |
1f5596dd CB |
1032 | return total_len; |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Returns 0 on success. | |
1037 | * It is the caller's responsibility to free `return_usage`, unless this | |
1038 | * function returns an error. | |
1039 | */ | |
1040 | int read_cpuacct_usage_all(char *cg, char *cpuset, | |
1041 | struct cpuacct_usage **return_usage, int *size) | |
1042 | { | |
1043 | __do_free char *usage_str = NULL; | |
1044 | __do_free struct cpuacct_usage *cpu_usage = NULL; | |
9ce186dc | 1045 | int i = 0, j = 0, read_pos = 0, read_cnt = 0; |
8b6987a2 | 1046 | int cpucount; |
9ce186dc | 1047 | int ret; |
1f5596dd CB |
1048 | int cg_cpu; |
1049 | uint64_t cg_user, cg_system; | |
1050 | int64_t ticks_per_sec; | |
1051 | ||
1052 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
1f5596dd | 1053 | if (ticks_per_sec < 0 && errno == EINVAL) { |
8b6987a2 | 1054 | lxcfs_debug("%m - Failed to determine number of ticks per second"); |
1f5596dd CB |
1055 | return -1; |
1056 | } | |
1057 | ||
f9434b9a | 1058 | cpucount = get_nprocs_conf(); |
1f5596dd CB |
1059 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); |
1060 | if (!cpu_usage) | |
1061 | return -ENOMEM; | |
1062 | ||
1063 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); | |
1064 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
8b6987a2 CB |
1065 | char *sep = " \t\n"; |
1066 | char *tok; | |
1f5596dd | 1067 | |
8b6987a2 CB |
1068 | /* Read cpuacct.usage_percpu instead. */ |
1069 | lxcfs_debug("Falling back to cpuacct.usage_percpu"); | |
1f5596dd CB |
1070 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) |
1071 | return -1; | |
1f5596dd | 1072 | |
8b6987a2 CB |
1073 | lxc_iterate_parts(tok, usage_str, sep) { |
1074 | uint64_t percpu_user; | |
1075 | ||
1076 | if (i >= cpucount) | |
1077 | break; | |
1f5596dd | 1078 | |
8b6987a2 CB |
1079 | tok = trim_whitespace_in_place(tok); |
1080 | ret = safe_uint64(tok, &percpu_user, 10); | |
1081 | if (ret) | |
1082 | return -1; | |
1f5596dd | 1083 | |
8b6987a2 CB |
1084 | /* Convert the time from nanoseconds to USER_HZ */ |
1085 | cpu_usage[i].user = percpu_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1086 | cpu_usage[i].system = cpu_usage[i].user; | |
1f5596dd | 1087 | i++; |
8b6987a2 | 1088 | lxcfs_debug("cpu%d with time %s", i, tok); |
1f5596dd | 1089 | } |
8b6987a2 CB |
1090 | } else { |
1091 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) | |
1092 | return log_error(-1, "read_cpuacct_usage_all reading first line from %s/cpuacct.usage_all failed", cg); | |
1f5596dd | 1093 | |
8b6987a2 | 1094 | read_pos += read_cnt; |
1f5596dd | 1095 | |
8b6987a2 CB |
1096 | for (i = 0, j = 0; i < cpucount; i++) { |
1097 | ret = sscanf(usage_str + read_pos, | |
1098 | "%d %" PRIu64 " %" PRIu64 "\n%n", &cg_cpu, | |
1099 | &cg_user, &cg_system, &read_cnt); | |
1f5596dd | 1100 | |
8b6987a2 CB |
1101 | if (ret == EOF) |
1102 | break; | |
1f5596dd | 1103 | |
8b6987a2 CB |
1104 | if (ret != 3) |
1105 | return log_error(-EINVAL, "Failed to parse cpuacct.usage_all line %s from cgroup %s", | |
1106 | usage_str + read_pos, cg); | |
1f5596dd | 1107 | |
8b6987a2 | 1108 | read_pos += read_cnt; |
1f5596dd | 1109 | |
8b6987a2 CB |
1110 | /* Convert the time from nanoseconds to USER_HZ */ |
1111 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1112 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1113 | j++; | |
1114 | } | |
1f5596dd CB |
1115 | } |
1116 | ||
1117 | *return_usage = move_ptr(cpu_usage); | |
1118 | *size = cpucount; | |
1119 | return 0; | |
1120 | } | |
1121 | ||
1122 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
1123 | { | |
9d7fc1a3 | 1124 | __do_free struct cg_proc_stat_head *h; |
1f5596dd | 1125 | |
9d7fc1a3 CB |
1126 | h = zalloc(sizeof(struct cg_proc_stat_head)); |
1127 | if (!h) | |
1128 | return false; | |
1f5596dd | 1129 | |
9d7fc1a3 CB |
1130 | if (pthread_rwlock_init(&h->lock, NULL)) |
1131 | return false; | |
1132 | ||
1133 | h->lastcheck = time(NULL); | |
1f5596dd | 1134 | |
9d7fc1a3 | 1135 | *head = move_ptr(h); |
1f5596dd CB |
1136 | return true; |
1137 | } | |
1138 | ||
4ec5c9da | 1139 | bool init_cpuview(void) |
1f5596dd CB |
1140 | { |
1141 | int i; | |
1142 | ||
1143 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
1144 | proc_stat_history[i] = NULL; | |
1145 | ||
1146 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1147 | if (!cpuview_init_head(&proc_stat_history[i])) | |
1148 | goto err; | |
1149 | } | |
1150 | ||
1151 | return true; | |
1152 | ||
1153 | err: | |
1154 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1155 | if (proc_stat_history[i]) | |
1156 | free_disarm(proc_stat_history[i]); | |
1157 | } | |
1158 | ||
1159 | return false; | |
1160 | } | |
1161 | ||
1f5596dd CB |
1162 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
1163 | { | |
905769cd | 1164 | struct cg_proc_stat *node; |
1f5596dd CB |
1165 | |
1166 | if (head->next) { | |
1167 | node = head->next; | |
1168 | ||
1169 | for (;;) { | |
905769cd | 1170 | struct cg_proc_stat *cur = node; |
1f5596dd | 1171 | node = node->next; |
905769cd | 1172 | free_proc_stat_node(cur); |
1f5596dd CB |
1173 | if (!node) |
1174 | break; | |
1175 | } | |
1176 | } | |
1177 | ||
1178 | pthread_rwlock_destroy(&head->lock); | |
1179 | free_disarm(head); | |
1180 | } | |
1181 | ||
4ec5c9da | 1182 | void free_cpuview(void) |
1f5596dd | 1183 | { |
4ec5c9da | 1184 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) |
1f5596dd CB |
1185 | if (proc_stat_history[i]) |
1186 | cpuview_free_head(proc_stat_history[i]); | |
1f5596dd | 1187 | } |