]>
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 | ||
9 | #ifdef HAVE_FUSE3 | |
10 | #ifndef FUSE_USE_VERSION | |
11 | #define FUSE_USE_VERSION 30 | |
12 | #endif | |
13 | #else | |
1f5596dd CB |
14 | #ifndef FUSE_USE_VERSION |
15 | #define FUSE_USE_VERSION 26 | |
16 | #endif | |
f834b6bf | 17 | #endif |
1f5596dd CB |
18 | |
19 | #define _FILE_OFFSET_BITS 64 | |
20 | ||
21 | #define __STDC_FORMAT_MACROS | |
22 | #include <dirent.h> | |
23 | #include <errno.h> | |
24 | #include <fcntl.h> | |
25 | #include <fuse.h> | |
26 | #include <inttypes.h> | |
27 | #include <libgen.h> | |
28 | #include <pthread.h> | |
29 | #include <sched.h> | |
30 | #include <stdarg.h> | |
31 | #include <stdbool.h> | |
32 | #include <stdint.h> | |
33 | #include <stdio.h> | |
34 | #include <stdlib.h> | |
35 | #include <string.h> | |
36 | #include <time.h> | |
37 | #include <unistd.h> | |
38 | #include <wait.h> | |
39 | #include <linux/magic.h> | |
40 | #include <linux/sched.h> | |
41 | #include <sys/epoll.h> | |
42 | #include <sys/mman.h> | |
43 | #include <sys/mount.h> | |
44 | #include <sys/param.h> | |
45 | #include <sys/socket.h> | |
46 | #include <sys/syscall.h> | |
47 | #include <sys/sysinfo.h> | |
48 | #include <sys/vfs.h> | |
49 | ||
50 | #include "bindings.h" | |
1f5596dd CB |
51 | #include "cgroup_fuse.h" |
52 | #include "cpuset_parse.h" | |
53 | #include "cgroups/cgroup.h" | |
54 | #include "cgroups/cgroup_utils.h" | |
55 | #include "memory_utils.h" | |
4ec5c9da | 56 | #include "proc_loadavg.h" |
1f5596dd CB |
57 | #include "utils.h" |
58 | ||
1f5596dd CB |
59 | /* Data for CPU view */ |
60 | struct cg_proc_stat { | |
61 | char *cg; | |
ce617d73 CB |
62 | struct cpuacct_usage *usage; /* Real usage as read from the host's /proc/stat. */ |
63 | struct cpuacct_usage *view; /* Usage stats reported to the container. */ | |
1f5596dd | 64 | int cpu_count; |
ce617d73 | 65 | pthread_mutex_t lock; /* For node manipulation. */ |
1f5596dd CB |
66 | struct cg_proc_stat *next; |
67 | }; | |
68 | ||
69 | struct cg_proc_stat_head { | |
70 | struct cg_proc_stat *next; | |
71 | time_t lastcheck; | |
72 | ||
73 | /* | |
74 | * For access to the list. Reading can be parallel, pruning is exclusive. | |
75 | */ | |
76 | pthread_rwlock_t lock; | |
77 | }; | |
78 | ||
79 | #define CPUVIEW_HASH_SIZE 100 | |
80 | static struct cg_proc_stat_head *proc_stat_history[CPUVIEW_HASH_SIZE]; | |
81 | ||
b456d40d CB |
82 | static void reset_proc_stat_node(struct cg_proc_stat *node, |
83 | struct cpuacct_usage *usage, int cpu_count) | |
1f5596dd | 84 | { |
1f5596dd CB |
85 | lxcfs_debug("Resetting stat node for %s\n", node->cg); |
86 | memcpy(node->usage, usage, sizeof(struct cpuacct_usage) * cpu_count); | |
87 | ||
b456d40d | 88 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
89 | node->view[i].user = 0; |
90 | node->view[i].system = 0; | |
91 | node->view[i].idle = 0; | |
92 | } | |
93 | ||
94 | node->cpu_count = cpu_count; | |
95 | } | |
96 | ||
97 | static bool expand_proc_stat_node(struct cg_proc_stat *node, int cpu_count) | |
98 | { | |
99 | __do_free struct cpuacct_usage *new_usage = NULL, *new_view = NULL; | |
100 | ||
101 | /* Allocate new memory */ | |
82d74a95 | 102 | new_usage = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
103 | if (!new_usage) |
104 | return false; | |
105 | ||
82d74a95 | 106 | new_view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd CB |
107 | if (!new_view) |
108 | return false; | |
109 | ||
110 | /* Copy existing data & initialize new elements */ | |
111 | for (int i = 0; i < cpu_count; i++) { | |
112 | if (i < node->cpu_count) { | |
82d74a95 CB |
113 | new_usage[i].user = node->usage[i].user; |
114 | new_usage[i].system = node->usage[i].system; | |
115 | new_usage[i].idle = node->usage[i].idle; | |
116 | ||
117 | new_view[i].user = node->view[i].user; | |
118 | new_view[i].system = node->view[i].system; | |
119 | new_view[i].idle = node->view[i].idle; | |
1f5596dd CB |
120 | } |
121 | } | |
122 | ||
123 | free(node->usage); | |
124 | node->usage = move_ptr(new_usage); | |
125 | ||
126 | free(node->view); | |
127 | node->view = move_ptr(new_view); | |
128 | node->cpu_count = cpu_count; | |
129 | ||
130 | return true; | |
131 | } | |
132 | ||
4ec5c9da CB |
133 | static void free_proc_stat_node(struct cg_proc_stat *node) |
134 | { | |
6a4dceb1 CB |
135 | if (node) { |
136 | /* | |
137 | * We're abusing the usage pointer to indicate that | |
138 | * pthread_mutex_init() was successful. Don't judge me. | |
139 | */ | |
140 | if (node->usage) | |
141 | pthread_mutex_destroy(&node->lock); | |
142 | free_disarm(node->cg); | |
143 | free_disarm(node->usage); | |
144 | free_disarm(node->view); | |
145 | free_disarm(node); | |
146 | } | |
4ec5c9da CB |
147 | } |
148 | ||
6a4dceb1 CB |
149 | define_cleanup_function(struct cg_proc_stat *, free_proc_stat_node); |
150 | ||
1f5596dd CB |
151 | static struct cg_proc_stat *add_proc_stat_node(struct cg_proc_stat *new_node) |
152 | { | |
0d129671 CB |
153 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *new = new_node; |
154 | struct cg_proc_stat *rv = new_node; | |
155 | int hash = calc_hash(new->cg) % CPUVIEW_HASH_SIZE; | |
1f5596dd | 156 | struct cg_proc_stat_head *head = proc_stat_history[hash]; |
0d129671 | 157 | struct cg_proc_stat *cur; |
1f5596dd CB |
158 | |
159 | pthread_rwlock_wrlock(&head->lock); | |
160 | ||
161 | if (!head->next) { | |
0d129671 | 162 | head->next = move_ptr(new); |
164acda7 | 163 | goto out_rwlock_unlock; |
1f5596dd CB |
164 | } |
165 | ||
0d129671 | 166 | cur = head->next; |
1f5596dd CB |
167 | |
168 | for (;;) { | |
0d129671 CB |
169 | /* |
170 | * The node to be added is already present in the list, so | |
171 | * free the newly allocated one and return the one we found. | |
172 | */ | |
173 | if (strcmp(cur->cg, new->cg) == 0) { | |
174 | rv = cur; | |
164acda7 | 175 | goto out_rwlock_unlock; |
1f5596dd CB |
176 | } |
177 | ||
0d129671 CB |
178 | /* Keep walking. */ |
179 | if (cur->next) { | |
180 | cur = cur->next; | |
1f5596dd CB |
181 | continue; |
182 | } | |
183 | ||
0d129671 CB |
184 | /* Add new node to end of list. */ |
185 | cur->next = move_ptr(new); | |
164acda7 | 186 | goto out_rwlock_unlock; |
1f5596dd CB |
187 | } |
188 | ||
164acda7 | 189 | out_rwlock_unlock: |
1f5596dd | 190 | pthread_rwlock_unlock(&head->lock); |
0d129671 | 191 | return move_ptr(rv); |
1f5596dd CB |
192 | } |
193 | ||
6a4dceb1 CB |
194 | static struct cg_proc_stat *new_proc_stat_node(struct cpuacct_usage *usage, |
195 | int cpu_count, const char *cg) | |
1f5596dd | 196 | { |
6a4dceb1 CB |
197 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *node = NULL; |
198 | __do_free struct cpuacct_usage *new_usage = NULL; | |
1f5596dd | 199 | |
6a4dceb1 | 200 | node = zalloc(sizeof(struct cg_proc_stat)); |
1f5596dd | 201 | if (!node) |
6a4dceb1 | 202 | return NULL; |
1f5596dd | 203 | |
6a4dceb1 | 204 | node->cg = strdup(cg); |
1f5596dd | 205 | if (!node->cg) |
6a4dceb1 | 206 | return NULL; |
1f5596dd | 207 | |
6a4dceb1 CB |
208 | new_usage = memdup(usage, sizeof(struct cpuacct_usage) * cpu_count); |
209 | if (!new_usage) | |
210 | return NULL; | |
1f5596dd | 211 | |
6a4dceb1 | 212 | node->view = zalloc(sizeof(struct cpuacct_usage) * cpu_count); |
1f5596dd | 213 | if (!node->view) |
6a4dceb1 | 214 | return NULL; |
1f5596dd CB |
215 | |
216 | node->cpu_count = cpu_count; | |
1f5596dd | 217 | |
6a4dceb1 CB |
218 | if (pthread_mutex_init(&node->lock, NULL)) |
219 | return NULL; | |
220 | /* | |
221 | * We're abusing the usage pointer to indicate that | |
222 | * pthread_mutex_init() was successful. Don't judge me. | |
223 | */ | |
224 | node->usage = move_ptr(new_usage); | |
1f5596dd | 225 | |
6a4dceb1 | 226 | return move_ptr(node); |
1f5596dd CB |
227 | } |
228 | ||
2d00d04c CB |
229 | static bool cgroup_supports(const char *controller, const char *cgroup, |
230 | const char *file) | |
4ec5c9da | 231 | { |
2c990b1d CB |
232 | __do_free char *path = NULL; |
233 | int cfd; | |
4ec5c9da CB |
234 | |
235 | cfd = get_cgroup_fd(controller); | |
236 | if (cfd < 0) | |
237 | return false; | |
238 | ||
925d5849 | 239 | path = must_make_path_relative(cgroup, file, NULL); |
2d00d04c | 240 | return faccessat(cfd, path, F_OK, 0) == 0; |
4ec5c9da CB |
241 | } |
242 | ||
1f5596dd CB |
243 | static struct cg_proc_stat *prune_proc_stat_list(struct cg_proc_stat *node) |
244 | { | |
b456d40d | 245 | struct cg_proc_stat *first = NULL; |
1f5596dd | 246 | |
b456d40d | 247 | for (struct cg_proc_stat *prev = NULL; node; ) { |
2d00d04c CB |
248 | if (!cgroup_supports("cpu", node->cg, "cpu.shares")) { |
249 | call_cleaner(free_proc_stat_node) struct cg_proc_stat *cur = node; | |
1f5596dd CB |
250 | |
251 | if (prev) | |
252 | prev->next = node->next; | |
253 | else | |
254 | first = node->next; | |
255 | ||
256 | node = node->next; | |
2d00d04c | 257 | lxcfs_debug("Removing stat node for %s\n", cur->cg); |
1f5596dd CB |
258 | } else { |
259 | if (!first) | |
260 | first = node; | |
261 | prev = node; | |
262 | node = node->next; | |
263 | } | |
264 | } | |
265 | ||
266 | return first; | |
267 | } | |
268 | ||
269 | #define PROC_STAT_PRUNE_INTERVAL 10 | |
270 | static void prune_proc_stat_history(void) | |
271 | { | |
1f5596dd CB |
272 | time_t now = time(NULL); |
273 | ||
b456d40d | 274 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) { |
1f5596dd CB |
275 | pthread_rwlock_wrlock(&proc_stat_history[i]->lock); |
276 | ||
277 | if ((proc_stat_history[i]->lastcheck + PROC_STAT_PRUNE_INTERVAL) > now) { | |
278 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
279 | return; | |
280 | } | |
281 | ||
282 | if (proc_stat_history[i]->next) { | |
283 | proc_stat_history[i]->next = prune_proc_stat_list(proc_stat_history[i]->next); | |
284 | proc_stat_history[i]->lastcheck = now; | |
285 | } | |
286 | ||
287 | pthread_rwlock_unlock(&proc_stat_history[i]->lock); | |
288 | } | |
289 | } | |
290 | ||
291 | static struct cg_proc_stat *find_proc_stat_node(struct cg_proc_stat_head *head, | |
292 | const char *cg) | |
293 | { | |
294 | struct cg_proc_stat *node; | |
295 | ||
296 | pthread_rwlock_rdlock(&head->lock); | |
297 | ||
298 | if (!head->next) { | |
299 | pthread_rwlock_unlock(&head->lock); | |
300 | return NULL; | |
301 | } | |
302 | ||
303 | node = head->next; | |
304 | ||
305 | do { | |
306 | if (strcmp(cg, node->cg) == 0) | |
307 | goto out; | |
308 | } while ((node = node->next)); | |
309 | ||
310 | node = NULL; | |
311 | ||
312 | out: | |
313 | pthread_rwlock_unlock(&head->lock); | |
314 | prune_proc_stat_history(); | |
315 | return node; | |
316 | } | |
317 | ||
318 | static struct cg_proc_stat *find_or_create_proc_stat_node(struct cpuacct_usage *usage, int cpu_count, const char *cg) | |
319 | { | |
320 | int hash = calc_hash(cg) % CPUVIEW_HASH_SIZE; | |
321 | struct cg_proc_stat_head *head = proc_stat_history[hash]; | |
322 | struct cg_proc_stat *node; | |
323 | ||
324 | node = find_proc_stat_node(head, cg); | |
1f5596dd CB |
325 | if (!node) { |
326 | node = new_proc_stat_node(usage, cpu_count, cg); | |
327 | if (!node) | |
328 | return NULL; | |
329 | ||
330 | node = add_proc_stat_node(node); | |
331 | lxcfs_debug("New stat node (%d) for %s\n", cpu_count, cg); | |
332 | } | |
333 | ||
334 | pthread_mutex_lock(&node->lock); | |
335 | ||
ce089f10 CB |
336 | /* |
337 | * If additional CPUs on the host have been enabled, CPU usage counter | |
338 | * arrays have to be expanded. | |
339 | */ | |
1f5596dd CB |
340 | if (node->cpu_count < cpu_count) { |
341 | lxcfs_debug("Expanding stat node %d->%d for %s\n", | |
ce089f10 | 342 | node->cpu_count, cpu_count, cg); |
1f5596dd CB |
343 | |
344 | if (!expand_proc_stat_node(node, cpu_count)) { | |
345 | pthread_mutex_unlock(&node->lock); | |
b456d40d | 346 | return log_debug(NULL, "Unable to expand stat node %d->%d for %s", node->cpu_count, cpu_count, cg); |
1f5596dd CB |
347 | } |
348 | } | |
349 | ||
350 | return node; | |
351 | } | |
352 | ||
2b8eff1d CB |
353 | static void add_cpu_usage(uint64_t *surplus, struct cpuacct_usage *usage, |
354 | uint64_t *counter, uint64_t threshold) | |
1f5596dd | 355 | { |
1ba088ae | 356 | uint64_t free_space, to_add; |
1f5596dd CB |
357 | |
358 | free_space = threshold - usage->user - usage->system; | |
359 | ||
360 | if (free_space > usage->idle) | |
361 | free_space = usage->idle; | |
362 | ||
8206874a CB |
363 | if (free_space > *surplus) |
364 | to_add = *surplus; | |
365 | else | |
366 | to_add = free_space; | |
1f5596dd CB |
367 | |
368 | *counter += to_add; | |
369 | usage->idle -= to_add; | |
370 | *surplus -= to_add; | |
371 | } | |
372 | ||
1ba088ae CB |
373 | static uint64_t diff_cpu_usage(struct cpuacct_usage *older, |
374 | struct cpuacct_usage *newer, | |
375 | struct cpuacct_usage *diff, int cpu_count) | |
1f5596dd | 376 | { |
1ba088ae | 377 | uint64_t sum = 0; |
1f5596dd | 378 | |
b456d40d | 379 | for (int i = 0; i < cpu_count; i++) { |
1f5596dd CB |
380 | if (!newer[i].online) |
381 | continue; | |
382 | ||
b456d40d CB |
383 | /* |
384 | * When cpuset is changed on the fly, the CPUs might get | |
385 | * reordered. We could either reset all counters, or check | |
386 | * that the substractions below will return expected results. | |
1f5596dd CB |
387 | */ |
388 | if (newer[i].user > older[i].user) | |
389 | diff[i].user = newer[i].user - older[i].user; | |
390 | else | |
391 | diff[i].user = 0; | |
392 | ||
393 | if (newer[i].system > older[i].system) | |
394 | diff[i].system = newer[i].system - older[i].system; | |
395 | else | |
396 | diff[i].system = 0; | |
397 | ||
398 | if (newer[i].idle > older[i].idle) | |
399 | diff[i].idle = newer[i].idle - older[i].idle; | |
400 | else | |
401 | diff[i].idle = 0; | |
402 | ||
403 | sum += diff[i].user; | |
404 | sum += diff[i].system; | |
405 | sum += diff[i].idle; | |
406 | } | |
407 | ||
408 | return sum; | |
409 | } | |
410 | ||
411 | /* | |
b456d40d CB |
412 | * Read cgroup CPU quota parameters from `cpu.cfs_quota_us` or |
413 | * `cpu.cfs_period_us`, depending on `param`. Parameter value is returned | |
92264841 | 414 | * through `value`. |
1f5596dd CB |
415 | */ |
416 | static bool read_cpu_cfs_param(const char *cg, const char *param, int64_t *value) | |
417 | { | |
418 | __do_free char *str = NULL; | |
48f6862e | 419 | char file[STRLITERALLEN("cpu.cfs_period_us") + 1]; |
9844eea7 | 420 | bool first = true; |
48f6862e | 421 | int ret; |
1f5596dd | 422 | |
48f6862e | 423 | if (pure_unified_layout(cgroup_ops)) { |
9844eea7 | 424 | first = !strcmp(param, "quota"); |
48f6862e CB |
425 | ret = snprintf(file, sizeof(file), "cpu.max"); |
426 | } else { | |
427 | ret = snprintf(file, sizeof(file), "cpu.cfs_%s_us", param); | |
9844eea7 | 428 | } |
48f6862e | 429 | if (ret < 0 || (size_t)ret >= sizeof(file)) |
1f5596dd CB |
430 | return false; |
431 | ||
48f6862e | 432 | if (!cgroup_ops->get(cgroup_ops, "cpu", cg, file, &str)) |
1f5596dd CB |
433 | return false; |
434 | ||
48f6862e | 435 | return sscanf(str, first ? "%" PRId64 : "%*d %" PRId64, value) == 1; |
1f5596dd CB |
436 | } |
437 | ||
438 | /* | |
439 | * Return the exact number of visible CPUs based on CPU quotas. | |
440 | * If there is no quota set, zero is returned. | |
441 | */ | |
442 | static double exact_cpu_count(const char *cg) | |
443 | { | |
444 | double rv; | |
445 | int nprocs; | |
446 | int64_t cfs_quota, cfs_period; | |
447 | ||
c602a0d0 CB |
448 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
449 | return 0; | |
450 | ||
451 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
452 | return 0; | |
1f5596dd CB |
453 | |
454 | if (cfs_quota <= 0 || cfs_period <= 0) | |
455 | return 0; | |
456 | ||
457 | rv = (double)cfs_quota / (double)cfs_period; | |
458 | ||
459 | nprocs = get_nprocs(); | |
460 | ||
461 | if (rv > nprocs) | |
462 | rv = nprocs; | |
463 | ||
464 | return rv; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Return the maximum number of visible CPUs based on CPU quotas. | |
469 | * If there is no quota set, zero is returned. | |
470 | */ | |
4ec5c9da | 471 | int max_cpu_count(const char *cg) |
1f5596dd | 472 | { |
700dd417 | 473 | __do_free char *cpuset = NULL; |
1f5596dd CB |
474 | int rv, nprocs; |
475 | int64_t cfs_quota, cfs_period; | |
476 | int nr_cpus_in_cpuset = 0; | |
1f5596dd | 477 | |
921bdfdb CB |
478 | if (!read_cpu_cfs_param(cg, "quota", &cfs_quota)) |
479 | return 0; | |
480 | ||
481 | if (!read_cpu_cfs_param(cg, "period", &cfs_period)) | |
482 | return 0; | |
1f5596dd CB |
483 | |
484 | cpuset = get_cpuset(cg); | |
485 | if (cpuset) | |
486 | nr_cpus_in_cpuset = cpu_number_in_cpuset(cpuset); | |
487 | ||
921bdfdb | 488 | if (cfs_quota <= 0 || cfs_period <= 0) { |
1f5596dd CB |
489 | if (nr_cpus_in_cpuset > 0) |
490 | return nr_cpus_in_cpuset; | |
491 | ||
492 | return 0; | |
493 | } | |
494 | ||
495 | rv = cfs_quota / cfs_period; | |
496 | ||
921bdfdb CB |
497 | /* |
498 | * In case quota/period does not yield a whole number, add one CPU for | |
1f5596dd CB |
499 | * the remainder. |
500 | */ | |
501 | if ((cfs_quota % cfs_period) > 0) | |
502 | rv += 1; | |
503 | ||
504 | nprocs = get_nprocs(); | |
1f5596dd CB |
505 | if (rv > nprocs) |
506 | rv = nprocs; | |
507 | ||
921bdfdb | 508 | /* Use min value in cpu quota and cpuset. */ |
1f5596dd CB |
509 | if (nr_cpus_in_cpuset > 0 && nr_cpus_in_cpuset < rv) |
510 | rv = nr_cpus_in_cpuset; | |
511 | ||
512 | return rv; | |
513 | } | |
514 | ||
515 | int cpuview_proc_stat(const char *cg, const char *cpuset, | |
516 | struct cpuacct_usage *cg_cpu_usage, int cg_cpu_usage_size, | |
517 | FILE *f, char *buf, size_t buf_size) | |
518 | { | |
519 | __do_free char *line = NULL; | |
520 | __do_free struct cpuacct_usage *diff = NULL; | |
4f18a602 | 521 | size_t linelen = 0, total_len = 0; |
1f5596dd CB |
522 | int curcpu = -1; /* cpu numbering starts at 0 */ |
523 | int physcpu, i; | |
39f231da | 524 | int cpu_cnt = 0; |
2b8eff1d CB |
525 | uint64_t user = 0, nice = 0, system = 0, idle = 0, iowait = 0, irq = 0, |
526 | softirq = 0, steal = 0, guest = 0, guest_nice = 0; | |
527 | uint64_t user_sum = 0, system_sum = 0, idle_sum = 0; | |
528 | uint64_t user_surplus = 0, system_surplus = 0; | |
39f231da | 529 | int nprocs, max_cpus; |
4f18a602 | 530 | ssize_t l; |
2b8eff1d | 531 | uint64_t total_sum, threshold; |
1f5596dd | 532 | struct cg_proc_stat *stat_node; |
1f5596dd | 533 | |
39f231da | 534 | nprocs = get_nprocs_conf(); |
1f5596dd CB |
535 | if (cg_cpu_usage_size < nprocs) |
536 | nprocs = cg_cpu_usage_size; | |
537 | ||
538 | /* Read all CPU stats and stop when we've encountered other lines */ | |
539 | while (getline(&line, &linelen, f) != -1) { | |
540 | int ret; | |
541 | char cpu_char[10]; /* That's a lot of cores */ | |
542 | uint64_t all_used, cg_used; | |
543 | ||
544 | if (strlen(line) == 0) | |
545 | continue; | |
546 | ||
547 | /* not a ^cpuN line containing a number N */ | |
548 | if (sscanf(line, "cpu%9[^ ]", cpu_char) != 1) | |
549 | break; | |
550 | ||
551 | if (sscanf(cpu_char, "%d", &physcpu) != 1) | |
552 | continue; | |
553 | ||
554 | if (physcpu >= cg_cpu_usage_size) | |
555 | continue; | |
556 | ||
fd65c77c CB |
557 | curcpu++; |
558 | cpu_cnt++; | |
1f5596dd CB |
559 | |
560 | if (!cpu_in_cpuset(physcpu, cpuset)) { | |
561 | for (i = curcpu; i <= physcpu; i++) | |
562 | cg_cpu_usage[i].online = false; | |
563 | continue; | |
564 | } | |
565 | ||
566 | if (curcpu < physcpu) { | |
567 | /* Some CPUs may be disabled */ | |
568 | for (i = curcpu; i < physcpu; i++) | |
569 | cg_cpu_usage[i].online = false; | |
570 | ||
571 | curcpu = physcpu; | |
572 | } | |
573 | ||
574 | cg_cpu_usage[curcpu].online = true; | |
575 | ||
2b8eff1d | 576 | ret = sscanf(line, "%*s %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 " %" PRIu64 "lu", |
1f5596dd CB |
577 | &user, |
578 | &nice, | |
579 | &system, | |
580 | &idle, | |
581 | &iowait, | |
582 | &irq, | |
583 | &softirq, | |
584 | &steal, | |
585 | &guest, | |
586 | &guest_nice); | |
1f5596dd CB |
587 | if (ret != 10) |
588 | continue; | |
589 | ||
590 | all_used = user + nice + system + iowait + irq + softirq + steal + guest + guest_nice; | |
591 | cg_used = cg_cpu_usage[curcpu].user + cg_cpu_usage[curcpu].system; | |
592 | ||
593 | if (all_used >= cg_used) { | |
594 | cg_cpu_usage[curcpu].idle = idle + (all_used - cg_used); | |
595 | ||
596 | } else { | |
2b8eff1d CB |
597 | lxcfs_error("cpu%d from %s has unexpected cpu time: %" PRIu64 " in /proc/stat, %" PRIu64 " in cpuacct.usage_all; unable to determine idle time", |
598 | curcpu, cg, all_used, cg_used); | |
1f5596dd CB |
599 | cg_cpu_usage[curcpu].idle = idle; |
600 | } | |
601 | } | |
602 | ||
f9434b9a CB |
603 | /* Cannot use more CPUs than is available in cpuset. */ |
604 | max_cpus = max_cpu_count(cg); | |
605 | if (max_cpus > cpu_cnt || !max_cpus) | |
606 | max_cpus = cpu_cnt; | |
607 | ||
1f5596dd | 608 | stat_node = find_or_create_proc_stat_node(cg_cpu_usage, nprocs, cg); |
b456d40d CB |
609 | if (!stat_node) |
610 | return log_error(0, "Failed to find/create stat node for %s", cg); | |
1f5596dd | 611 | |
b4572722 | 612 | diff = zalloc(sizeof(struct cpuacct_usage) * nprocs); |
700dd417 | 613 | if (!diff) |
1f5596dd | 614 | return 0; |
1f5596dd CB |
615 | |
616 | /* | |
617 | * If the new values are LOWER than values stored in memory, it means | |
618 | * the cgroup has been reset/recreated and we should reset too. | |
619 | */ | |
620 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
621 | if (!cg_cpu_usage[curcpu].online) | |
622 | continue; | |
623 | ||
624 | if (cg_cpu_usage[curcpu].user < stat_node->usage[curcpu].user) | |
625 | reset_proc_stat_node(stat_node, cg_cpu_usage, nprocs); | |
626 | ||
627 | break; | |
628 | } | |
629 | ||
630 | total_sum = diff_cpu_usage(stat_node->usage, cg_cpu_usage, diff, nprocs); | |
631 | ||
632 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
633 | stat_node->usage[curcpu].online = cg_cpu_usage[curcpu].online; | |
634 | ||
635 | if (!stat_node->usage[curcpu].online) | |
636 | continue; | |
637 | ||
638 | i++; | |
639 | ||
b4572722 | 640 | stat_node->usage[curcpu].user += diff[curcpu].user; |
1f5596dd | 641 | stat_node->usage[curcpu].system += diff[curcpu].system; |
b4572722 | 642 | stat_node->usage[curcpu].idle += diff[curcpu].idle; |
1f5596dd CB |
643 | |
644 | if (max_cpus > 0 && i >= max_cpus) { | |
b4572722 CB |
645 | user_surplus += diff[curcpu].user; |
646 | system_surplus += diff[curcpu].system; | |
1f5596dd CB |
647 | } |
648 | } | |
649 | ||
650 | /* Calculate usage counters of visible CPUs */ | |
651 | if (max_cpus > 0) { | |
2b8eff1d CB |
652 | uint64_t diff_user = 0; |
653 | uint64_t diff_system = 0; | |
654 | uint64_t diff_idle = 0; | |
655 | uint64_t max_diff_idle = 0; | |
656 | uint64_t max_diff_idle_index = 0; | |
1f5596dd CB |
657 | double exact_cpus; |
658 | ||
659 | /* threshold = maximum usage per cpu, including idle */ | |
660 | threshold = total_sum / cpu_cnt * max_cpus; | |
661 | ||
662 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
663 | if (!stat_node->usage[curcpu].online) | |
664 | continue; | |
665 | ||
666 | i++; | |
667 | ||
668 | if (i == max_cpus) | |
669 | break; | |
670 | ||
671 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
672 | continue; | |
673 | ||
674 | /* Add user */ | |
675 | add_cpu_usage(&user_surplus, &diff[curcpu], | |
676 | &diff[curcpu].user, threshold); | |
677 | ||
678 | if (diff[curcpu].user + diff[curcpu].system >= threshold) | |
679 | continue; | |
680 | ||
681 | /* If there is still room, add system */ | |
682 | add_cpu_usage(&system_surplus, &diff[curcpu], | |
683 | &diff[curcpu].system, threshold); | |
684 | } | |
685 | ||
686 | if (user_surplus > 0) | |
687 | lxcfs_debug("leftover user: %lu for %s\n", user_surplus, cg); | |
688 | if (system_surplus > 0) | |
689 | lxcfs_debug("leftover system: %lu for %s\n", system_surplus, cg); | |
690 | ||
691 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
692 | if (!stat_node->usage[curcpu].online) | |
693 | continue; | |
694 | ||
695 | i++; | |
696 | ||
697 | if (i == max_cpus) | |
698 | break; | |
699 | ||
b4572722 CB |
700 | stat_node->view[curcpu].user += diff[curcpu].user; |
701 | stat_node->view[curcpu].system += diff[curcpu].system; | |
702 | stat_node->view[curcpu].idle += diff[curcpu].idle; | |
1f5596dd | 703 | |
b4572722 CB |
704 | user_sum += stat_node->view[curcpu].user; |
705 | system_sum += stat_node->view[curcpu].system; | |
706 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd | 707 | |
b4572722 CB |
708 | diff_user += diff[curcpu].user; |
709 | diff_system += diff[curcpu].system; | |
710 | diff_idle += diff[curcpu].idle; | |
1f5596dd | 711 | if (diff[curcpu].idle > max_diff_idle) { |
b4572722 CB |
712 | max_diff_idle = diff[curcpu].idle; |
713 | max_diff_idle_index = curcpu; | |
1f5596dd CB |
714 | } |
715 | ||
716 | lxcfs_v("curcpu: %d, diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", curcpu, diff[curcpu].user, diff[curcpu].system, diff[curcpu].idle); | |
717 | } | |
718 | lxcfs_v("total. diff_user: %lu, diff_system: %lu, diff_idle: %lu\n", diff_user, diff_system, diff_idle); | |
719 | ||
720 | /* revise cpu usage view to support partial cpu case. */ | |
721 | exact_cpus = exact_cpu_count(cg); | |
722 | if (exact_cpus < (double)max_cpus){ | |
1ba088ae | 723 | uint64_t delta = (uint64_t)((double)(diff_user + diff_system + diff_idle) * (1 - exact_cpus / (double)max_cpus)); |
1f5596dd CB |
724 | |
725 | lxcfs_v("revising cpu usage view to match the exact cpu count [%f]\n", exact_cpus); | |
726 | lxcfs_v("delta: %lu\n", delta); | |
727 | lxcfs_v("idle_sum before: %lu\n", idle_sum); | |
b4572722 CB |
728 | if (idle_sum > delta) |
729 | idle_sum = idle_sum - delta; | |
730 | else | |
731 | idle_sum = 0; | |
1f5596dd CB |
732 | lxcfs_v("idle_sum after: %lu\n", idle_sum); |
733 | ||
734 | curcpu = max_diff_idle_index; | |
735 | lxcfs_v("curcpu: %d, idle before: %lu\n", curcpu, stat_node->view[curcpu].idle); | |
b4572722 CB |
736 | if (stat_node->view[curcpu].idle > delta) |
737 | stat_node->view[curcpu].idle = stat_node->view[curcpu].idle - delta; | |
738 | else | |
739 | stat_node->view[curcpu].idle = 0; | |
1f5596dd CB |
740 | lxcfs_v("curcpu: %d, idle after: %lu\n", curcpu, stat_node->view[curcpu].idle); |
741 | } | |
742 | } else { | |
743 | for (curcpu = 0; curcpu < nprocs; curcpu++) { | |
744 | if (!stat_node->usage[curcpu].online) | |
745 | continue; | |
746 | ||
b4572722 CB |
747 | stat_node->view[curcpu].user = stat_node->usage[curcpu].user; |
748 | stat_node->view[curcpu].system = stat_node->usage[curcpu].system; | |
749 | stat_node->view[curcpu].idle = stat_node->usage[curcpu].idle; | |
1f5596dd | 750 | |
b4572722 CB |
751 | user_sum += stat_node->view[curcpu].user; |
752 | system_sum += stat_node->view[curcpu].system; | |
753 | idle_sum += stat_node->view[curcpu].idle; | |
1f5596dd CB |
754 | } |
755 | } | |
756 | ||
757 | /* Render the file */ | |
758 | /* cpu-all */ | |
2b8eff1d CB |
759 | l = snprintf(buf, buf_size, |
760 | "cpu %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", | |
761 | user_sum, system_sum, idle_sum); | |
1f5596dd | 762 | lxcfs_v("cpu-all: %s\n", buf); |
b456d40d CB |
763 | if (l < 0) |
764 | return log_error(0, "Failed to write cache"); | |
765 | if (l >= buf_size) | |
766 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
767 | |
768 | buf += l; | |
769 | buf_size -= l; | |
770 | total_len += l; | |
771 | ||
772 | /* Render visible CPUs */ | |
773 | for (curcpu = 0, i = -1; curcpu < nprocs; curcpu++) { | |
774 | if (!stat_node->usage[curcpu].online) | |
775 | continue; | |
776 | ||
777 | i++; | |
778 | ||
779 | if (max_cpus > 0 && i == max_cpus) | |
780 | break; | |
781 | ||
2b8eff1d CB |
782 | l = snprintf(buf, buf_size, "cpu%d %" PRIu64 " 0 %" PRIu64 " %" PRIu64 " 0 0 0 0 0 0\n", |
783 | i, | |
784 | stat_node->view[curcpu].user, | |
785 | stat_node->view[curcpu].system, | |
786 | stat_node->view[curcpu].idle); | |
1f5596dd | 787 | lxcfs_v("cpu: %s\n", buf); |
b456d40d CB |
788 | if (l < 0) |
789 | return log_error(0, "Failed to write cache"); | |
790 | if (l >= buf_size) | |
791 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
792 | |
793 | buf += l; | |
794 | buf_size -= l; | |
795 | total_len += l; | |
796 | } | |
797 | ||
798 | /* Pass the rest of /proc/stat, start with the last line read */ | |
799 | l = snprintf(buf, buf_size, "%s", line); | |
b456d40d CB |
800 | if (l < 0) |
801 | return log_error(0, "Failed to write cache"); | |
802 | if (l >= buf_size) | |
803 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
804 | |
805 | buf += l; | |
806 | buf_size -= l; | |
807 | total_len += l; | |
808 | ||
809 | /* Pass the rest of the host's /proc/stat */ | |
810 | while (getline(&line, &linelen, f) != -1) { | |
811 | l = snprintf(buf, buf_size, "%s", line); | |
b456d40d CB |
812 | if (l < 0) |
813 | return log_error(0, "Failed to write cache"); | |
814 | if (l >= buf_size) | |
815 | return log_error(0, "Write to cache was truncated"); | |
816 | ||
1f5596dd CB |
817 | buf += l; |
818 | buf_size -= l; | |
819 | total_len += l; | |
820 | } | |
821 | ||
822 | if (stat_node) | |
823 | pthread_mutex_unlock(&stat_node->lock); | |
b456d40d | 824 | |
1f5596dd CB |
825 | return total_len; |
826 | } | |
827 | ||
828 | /* | |
829 | * check whether this is a '^processor" line in /proc/cpuinfo | |
830 | */ | |
b456d40d | 831 | static inline bool is_processor_line(const char *line) |
1f5596dd CB |
832 | { |
833 | int cpu; | |
b456d40d | 834 | return sscanf(line, "processor : %d", &cpu) == 1; |
1f5596dd CB |
835 | } |
836 | ||
b456d40d | 837 | static inline bool cpuline_in_cpuset(const char *line, const char *cpuset) |
1f5596dd CB |
838 | { |
839 | int cpu; | |
c539526c CB |
840 | |
841 | if (sscanf(line, "processor : %d", &cpu) == 1) | |
842 | return cpu_in_cpuset(cpu, cpuset); | |
843 | ||
844 | return false; | |
1f5596dd CB |
845 | } |
846 | ||
847 | int proc_cpuinfo_read(char *buf, size_t size, off_t offset, | |
848 | struct fuse_file_info *fi) | |
849 | { | |
850 | __do_free char *cg = NULL, *cpuset = NULL, *line = NULL; | |
757a63e7 | 851 | __do_free void *fopen_cache = NULL; |
1f5596dd CB |
852 | __do_fclose FILE *f = NULL; |
853 | struct fuse_context *fc = fuse_get_context(); | |
0274438c | 854 | struct lxcfs_opts *opts = (struct lxcfs_opts *)fc->private_data; |
99b183fb | 855 | struct file_info *d = INTTYPE_TO_PTR(fi->fh); |
1f5596dd CB |
856 | size_t linelen = 0, total_len = 0; |
857 | bool am_printing = false, firstline = true, is_s390x = false; | |
858 | int curcpu = -1, cpu, max_cpus = 0; | |
859 | bool use_view; | |
860 | char *cache = d->buf; | |
861 | size_t cache_size = d->buflen; | |
862 | ||
f9434b9a | 863 | if (offset) { |
1f5596dd CB |
864 | int left; |
865 | ||
866 | if (offset > d->size) | |
867 | return -EINVAL; | |
868 | ||
869 | if (!d->cached) | |
870 | return 0; | |
871 | ||
872 | left = d->size - offset; | |
873 | total_len = left > size ? size: left; | |
874 | memcpy(buf, cache + offset, total_len); | |
875 | ||
876 | return total_len; | |
877 | } | |
878 | ||
879 | pid_t initpid = lookup_initpid_in_store(fc->pid); | |
880 | if (initpid <= 1 || is_shared_pidns(initpid)) | |
881 | initpid = fc->pid; | |
b456d40d | 882 | |
1f5596dd CB |
883 | cg = get_pid_cgroup(initpid, "cpuset"); |
884 | if (!cg) | |
885 | return read_file_fuse("proc/cpuinfo", buf, size, d); | |
886 | prune_init_slice(cg); | |
887 | ||
888 | cpuset = get_cpuset(cg); | |
889 | if (!cpuset) | |
890 | return 0; | |
891 | ||
8044f626 | 892 | if (cgroup_ops->can_use_cpuview(cgroup_ops) && opts && opts->use_cfs) |
0274438c | 893 | use_view = true; |
8044f626 CB |
894 | else |
895 | use_view = false; | |
1f5596dd CB |
896 | if (use_view) |
897 | max_cpus = max_cpu_count(cg); | |
898 | ||
757a63e7 | 899 | f = fopen_cached("/proc/cpuinfo", "re", &fopen_cache); |
1f5596dd CB |
900 | if (!f) |
901 | return 0; | |
902 | ||
903 | while (getline(&line, &linelen, f) != -1) { | |
904 | ssize_t l; | |
905 | if (firstline) { | |
906 | firstline = false; | |
907 | if (strstr(line, "IBM/S390") != NULL) { | |
908 | is_s390x = true; | |
909 | am_printing = true; | |
910 | continue; | |
911 | } | |
912 | } | |
b456d40d | 913 | |
1f5596dd CB |
914 | if (strncmp(line, "# processors:", 12) == 0) |
915 | continue; | |
b456d40d | 916 | |
1f5596dd | 917 | if (is_processor_line(line)) { |
d0031abf | 918 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 919 | break; |
b456d40d | 920 | |
1f5596dd CB |
921 | am_printing = cpuline_in_cpuset(line, cpuset); |
922 | if (am_printing) { | |
d0031abf | 923 | curcpu++; |
1f5596dd | 924 | l = snprintf(cache, cache_size, "processor : %d\n", curcpu); |
b456d40d CB |
925 | if (l < 0) |
926 | return log_error(0, "Failed to write cache"); | |
927 | if (l >= cache_size) | |
928 | return log_error(0, "Write to cache was truncated"); | |
1f5596dd CB |
929 | cache += l; |
930 | cache_size -= l; | |
931 | total_len += l; | |
932 | } | |
933 | continue; | |
934 | } else if (is_s390x && sscanf(line, "processor %d:", &cpu) == 1) { | |
935 | char *p; | |
b456d40d | 936 | |
d0031abf | 937 | if (use_view && max_cpus > 0 && (curcpu + 1) == max_cpus) |
1f5596dd | 938 | break; |
b456d40d | 939 | |
1f5596dd CB |
940 | if (!cpu_in_cpuset(cpu, cpuset)) |
941 | continue; | |
b456d40d | 942 | |
1f5596dd CB |
943 | curcpu ++; |
944 | p = strchr(line, ':'); | |
945 | if (!p || !*p) | |
946 | return 0; | |
947 | p++; | |
b456d40d | 948 | |
1f5596dd | 949 | l = snprintf(cache, cache_size, "processor %d:%s", curcpu, p); |
b456d40d CB |
950 | if (l < 0) |
951 | return log_error(0, "Failed to write cache"); | |
952 | if (l >= cache_size) | |
953 | return log_error(0, "Write to cache was truncated"); | |
954 | ||
1f5596dd CB |
955 | cache += l; |
956 | cache_size -= l; | |
957 | total_len += l; | |
958 | continue; | |
959 | ||
960 | } | |
961 | if (am_printing) { | |
962 | l = snprintf(cache, cache_size, "%s", line); | |
b456d40d CB |
963 | if (l < 0) |
964 | return log_error(0, "Failed to write cache"); | |
965 | if (l >= cache_size) | |
966 | return log_error(0, "Write to cache was truncated"); | |
967 | ||
1f5596dd CB |
968 | cache += l; |
969 | cache_size -= l; | |
970 | total_len += l; | |
971 | } | |
972 | } | |
973 | ||
974 | if (is_s390x) { | |
975 | __do_free char *origcache = d->buf; | |
976 | ssize_t l; | |
977 | ||
978 | d->buf = malloc(d->buflen); | |
979 | if (!d->buf) { | |
980 | d->buf = move_ptr(origcache); | |
981 | return 0; | |
982 | } | |
983 | ||
984 | cache = d->buf; | |
985 | cache_size = d->buflen; | |
986 | total_len = 0; | |
987 | l = snprintf(cache, cache_size, "vendor_id : IBM/S390\n"); | |
988 | if (l < 0 || l >= cache_size) | |
989 | return 0; | |
990 | ||
991 | cache_size -= l; | |
992 | cache += l; | |
993 | total_len += l; | |
994 | l = snprintf(cache, cache_size, "# processors : %d\n", curcpu + 1); | |
995 | if (l < 0 || l >= cache_size) | |
996 | return 0; | |
997 | ||
998 | cache_size -= l; | |
999 | cache += l; | |
1000 | total_len += l; | |
1001 | l = snprintf(cache, cache_size, "%s", origcache); | |
1002 | if (l < 0 || l >= cache_size) | |
1003 | return 0; | |
1004 | total_len += l; | |
1005 | } | |
1006 | ||
1007 | d->cached = 1; | |
1008 | d->size = total_len; | |
d0031abf CB |
1009 | if (total_len > size) |
1010 | total_len = size; | |
1f5596dd CB |
1011 | |
1012 | /* read from off 0 */ | |
1013 | memcpy(buf, d->buf, total_len); | |
d0031abf | 1014 | |
1f5596dd CB |
1015 | return total_len; |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Returns 0 on success. | |
1020 | * It is the caller's responsibility to free `return_usage`, unless this | |
1021 | * function returns an error. | |
1022 | */ | |
1023 | int read_cpuacct_usage_all(char *cg, char *cpuset, | |
1024 | struct cpuacct_usage **return_usage, int *size) | |
1025 | { | |
1026 | __do_free char *usage_str = NULL; | |
1027 | __do_free struct cpuacct_usage *cpu_usage = NULL; | |
9ce186dc | 1028 | int i = 0, j = 0, read_pos = 0, read_cnt = 0; |
8b6987a2 | 1029 | int cpucount; |
9ce186dc | 1030 | int ret; |
1f5596dd CB |
1031 | int cg_cpu; |
1032 | uint64_t cg_user, cg_system; | |
1033 | int64_t ticks_per_sec; | |
1034 | ||
1035 | ticks_per_sec = sysconf(_SC_CLK_TCK); | |
1f5596dd | 1036 | if (ticks_per_sec < 0 && errno == EINVAL) { |
8b6987a2 | 1037 | lxcfs_debug("%m - Failed to determine number of ticks per second"); |
1f5596dd CB |
1038 | return -1; |
1039 | } | |
1040 | ||
f9434b9a | 1041 | cpucount = get_nprocs_conf(); |
1f5596dd CB |
1042 | cpu_usage = malloc(sizeof(struct cpuacct_usage) * cpucount); |
1043 | if (!cpu_usage) | |
1044 | return -ENOMEM; | |
1045 | ||
1046 | memset(cpu_usage, 0, sizeof(struct cpuacct_usage) * cpucount); | |
1047 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_all", &usage_str)) { | |
8b6987a2 CB |
1048 | char *sep = " \t\n"; |
1049 | char *tok; | |
1f5596dd | 1050 | |
8b6987a2 CB |
1051 | /* Read cpuacct.usage_percpu instead. */ |
1052 | lxcfs_debug("Falling back to cpuacct.usage_percpu"); | |
1f5596dd CB |
1053 | if (!cgroup_ops->get(cgroup_ops, "cpuacct", cg, "cpuacct.usage_percpu", &usage_str)) |
1054 | return -1; | |
1f5596dd | 1055 | |
8b6987a2 CB |
1056 | lxc_iterate_parts(tok, usage_str, sep) { |
1057 | uint64_t percpu_user; | |
1058 | ||
1059 | if (i >= cpucount) | |
1060 | break; | |
1f5596dd | 1061 | |
8b6987a2 CB |
1062 | tok = trim_whitespace_in_place(tok); |
1063 | ret = safe_uint64(tok, &percpu_user, 10); | |
1064 | if (ret) | |
1065 | return -1; | |
1f5596dd | 1066 | |
8b6987a2 CB |
1067 | /* Convert the time from nanoseconds to USER_HZ */ |
1068 | cpu_usage[i].user = percpu_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1069 | cpu_usage[i].system = cpu_usage[i].user; | |
1f5596dd | 1070 | i++; |
8b6987a2 | 1071 | lxcfs_debug("cpu%d with time %s", i, tok); |
1f5596dd | 1072 | } |
8b6987a2 CB |
1073 | } else { |
1074 | if (sscanf(usage_str, "cpu user system\n%n", &read_cnt) != 0) | |
1075 | return log_error(-1, "read_cpuacct_usage_all reading first line from %s/cpuacct.usage_all failed", cg); | |
1f5596dd | 1076 | |
8b6987a2 | 1077 | read_pos += read_cnt; |
1f5596dd | 1078 | |
8b6987a2 CB |
1079 | for (i = 0, j = 0; i < cpucount; i++) { |
1080 | ret = sscanf(usage_str + read_pos, | |
1081 | "%d %" PRIu64 " %" PRIu64 "\n%n", &cg_cpu, | |
1082 | &cg_user, &cg_system, &read_cnt); | |
1f5596dd | 1083 | |
8b6987a2 CB |
1084 | if (ret == EOF) |
1085 | break; | |
1f5596dd | 1086 | |
8b6987a2 CB |
1087 | if (ret != 3) |
1088 | return log_error(-EINVAL, "Failed to parse cpuacct.usage_all line %s from cgroup %s", | |
1089 | usage_str + read_pos, cg); | |
1f5596dd | 1090 | |
8b6987a2 | 1091 | read_pos += read_cnt; |
1f5596dd | 1092 | |
8b6987a2 CB |
1093 | /* Convert the time from nanoseconds to USER_HZ */ |
1094 | cpu_usage[j].user = cg_user / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1095 | cpu_usage[j].system = cg_system / 1000.0 / 1000 / 1000 * ticks_per_sec; | |
1096 | j++; | |
1097 | } | |
1f5596dd CB |
1098 | } |
1099 | ||
1100 | *return_usage = move_ptr(cpu_usage); | |
1101 | *size = cpucount; | |
1102 | return 0; | |
1103 | } | |
1104 | ||
1105 | static bool cpuview_init_head(struct cg_proc_stat_head **head) | |
1106 | { | |
9d7fc1a3 | 1107 | __do_free struct cg_proc_stat_head *h; |
1f5596dd | 1108 | |
9d7fc1a3 CB |
1109 | h = zalloc(sizeof(struct cg_proc_stat_head)); |
1110 | if (!h) | |
1111 | return false; | |
1f5596dd | 1112 | |
9d7fc1a3 CB |
1113 | if (pthread_rwlock_init(&h->lock, NULL)) |
1114 | return false; | |
1115 | ||
1116 | h->lastcheck = time(NULL); | |
1f5596dd | 1117 | |
9d7fc1a3 | 1118 | *head = move_ptr(h); |
1f5596dd CB |
1119 | return true; |
1120 | } | |
1121 | ||
4ec5c9da | 1122 | bool init_cpuview(void) |
1f5596dd CB |
1123 | { |
1124 | int i; | |
1125 | ||
1126 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) | |
1127 | proc_stat_history[i] = NULL; | |
1128 | ||
1129 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1130 | if (!cpuview_init_head(&proc_stat_history[i])) | |
1131 | goto err; | |
1132 | } | |
1133 | ||
1134 | return true; | |
1135 | ||
1136 | err: | |
1137 | for (i = 0; i < CPUVIEW_HASH_SIZE; i++) { | |
1138 | if (proc_stat_history[i]) | |
1139 | free_disarm(proc_stat_history[i]); | |
1140 | } | |
1141 | ||
1142 | return false; | |
1143 | } | |
1144 | ||
1f5596dd CB |
1145 | static void cpuview_free_head(struct cg_proc_stat_head *head) |
1146 | { | |
905769cd | 1147 | struct cg_proc_stat *node; |
1f5596dd CB |
1148 | |
1149 | if (head->next) { | |
1150 | node = head->next; | |
1151 | ||
1152 | for (;;) { | |
905769cd | 1153 | struct cg_proc_stat *cur = node; |
1f5596dd | 1154 | node = node->next; |
905769cd | 1155 | free_proc_stat_node(cur); |
1f5596dd CB |
1156 | if (!node) |
1157 | break; | |
1158 | } | |
1159 | } | |
1160 | ||
1161 | pthread_rwlock_destroy(&head->lock); | |
1162 | free_disarm(head); | |
1163 | } | |
1164 | ||
4ec5c9da | 1165 | void free_cpuview(void) |
1f5596dd | 1166 | { |
4ec5c9da | 1167 | for (int i = 0; i < CPUVIEW_HASH_SIZE; i++) |
1f5596dd CB |
1168 | if (proc_stat_history[i]) |
1169 | cpuview_free_head(proc_stat_history[i]); | |
1f5596dd | 1170 | } |