1 /* SPDX-License-Identifier: GPL-2.0 */
2 #define _GNU_SOURCE
3
4 #include <linux/limits.h>
5 #include <linux/oom.h>
6 #include <fcntl.h>
7 #include <stdio.h>
8 #include <stdlib.h>
9 #include <string.h>
10 #include <sys/stat.h>
11 #include <sys/types.h>
12 #include <unistd.h>
13 #include <sys/inotify.h>
14 #include <sys/socket.h>
15 #include <sys/wait.h>
16 #include <arpa/inet.h>
17 #include <netinet/in.h>
18 #include <netdb.h>
19 #include <errno.h>
20 #include <sys/mman.h>
21
22 #include "kselftest.h"
23 #include "cgroup_util.h"
24
25 #define MEMCG_SOCKSTAT_WAIT_RETRIES 30
26
27 static bool has_localevents;
28 static bool has_recursiveprot;
29 static int page_size;
30
get_temp_fd(void)31 int get_temp_fd(void)
32 {
33 return open(".", O_TMPFILE | O_RDWR | O_EXCL);
34 }
35
alloc_pagecache(int fd,size_t size)36 int alloc_pagecache(int fd, size_t size)
37 {
38 char buf[BUF_SIZE];
39 struct stat st;
40 int i;
41
42 if (fstat(fd, &st))
43 goto cleanup;
44
45 size += st.st_size;
46
47 if (ftruncate(fd, size))
48 goto cleanup;
49
50 for (i = 0; i < size; i += sizeof(buf))
51 read(fd, buf, sizeof(buf));
52
53 return 0;
54
55 cleanup:
56 return -1;
57 }
58
alloc_and_populate_anon(size_t size)59 static char *alloc_and_populate_anon(size_t size)
60 {
61 char *buf, *ptr;
62
63 buf = malloc(size);
64 if (buf == NULL) {
65 fprintf(stderr, "malloc() failed\n");
66 return NULL;
67 }
68
69 for (ptr = buf; ptr < buf + size; ptr += page_size)
70 *ptr = 0;
71
72 return buf;
73 }
74
alloc_anon(const char * cgroup,void * arg)75 int alloc_anon(const char *cgroup, void *arg)
76 {
77 size_t size = (unsigned long)arg;
78 char *buf;
79
80 buf = alloc_and_populate_anon(size);
81 if (!buf)
82 return -1;
83
84 free(buf);
85 return 0;
86 }
87
is_swap_enabled(void)88 int is_swap_enabled(void)
89 {
90 char buf[BUF_SIZE];
91 const char delim[] = "\n";
92 int cnt = 0;
93 char *line;
94
95 if (read_text("/proc/swaps", buf, sizeof(buf)) <= 0)
96 return -1;
97
98 for (line = strtok(buf, delim); line; line = strtok(NULL, delim))
99 cnt++;
100
101 return cnt > 1;
102 }
103
set_oom_adj_score(int pid,int score)104 int set_oom_adj_score(int pid, int score)
105 {
106 char path[PATH_MAX];
107 int fd, len;
108
109 sprintf(path, "/proc/%d/oom_score_adj", pid);
110
111 fd = open(path, O_WRONLY | O_APPEND);
112 if (fd < 0)
113 return fd;
114
115 len = dprintf(fd, "%d", score);
116 if (len < 0) {
117 close(fd);
118 return len;
119 }
120
121 close(fd);
122 return 0;
123 }
124
125 /*
126 * This test creates two nested cgroups with and without enabling
127 * the memory controller.
128 */
test_memcg_subtree_control(const char * root)129 static int test_memcg_subtree_control(const char *root)
130 {
131 char *parent, *child, *parent2 = NULL, *child2 = NULL;
132 int ret = KSFT_FAIL;
133 char buf[BUF_SIZE];
134
135 /* Create two nested cgroups with the memory controller enabled */
136 parent = cg_name(root, "memcg_test_0");
137 child = cg_name(root, "memcg_test_0/memcg_test_1");
138 if (!parent || !child)
139 goto cleanup_free;
140
141 if (cg_create(parent))
142 goto cleanup_free;
143
144 if (cg_write(parent, "cgroup.subtree_control", "+memory"))
145 goto cleanup_parent;
146
147 if (cg_create(child))
148 goto cleanup_parent;
149
150 if (cg_read_strstr(child, "cgroup.controllers", "memory"))
151 goto cleanup_child;
152
153 /* Create two nested cgroups without enabling memory controller */
154 parent2 = cg_name(root, "memcg_test_1");
155 child2 = cg_name(root, "memcg_test_1/memcg_test_1");
156 if (!parent2 || !child2)
157 goto cleanup_free2;
158
159 if (cg_create(parent2))
160 goto cleanup_free2;
161
162 if (cg_create(child2))
163 goto cleanup_parent2;
164
165 if (cg_read(child2, "cgroup.controllers", buf, sizeof(buf)))
166 goto cleanup_all;
167
168 if (!cg_read_strstr(child2, "cgroup.controllers", "memory"))
169 goto cleanup_all;
170
171 ret = KSFT_PASS;
172
173 cleanup_all:
174 cg_destroy(child2);
175 cleanup_parent2:
176 cg_destroy(parent2);
177 cleanup_free2:
178 free(parent2);
179 free(child2);
180 cleanup_child:
181 cg_destroy(child);
182 cleanup_parent:
183 cg_destroy(parent);
184 cleanup_free:
185 free(parent);
186 free(child);
187
188 return ret;
189 }
190
alloc_anon_50M_check(const char * cgroup,void * arg)191 static int alloc_anon_50M_check(const char *cgroup, void *arg)
192 {
193 size_t size = MB(50);
194 char *buf;
195 long anon, current;
196 int ret = -1;
197
198 buf = alloc_and_populate_anon(size);
199 if (!buf)
200 return -1;
201
202 current = cg_read_long(cgroup, "memory.current");
203 if (current < size)
204 goto cleanup;
205
206 if (!values_close(size, current, 3))
207 goto cleanup;
208
209 anon = cg_read_key_long(cgroup, "memory.stat", "anon ");
210 if (anon < 0)
211 goto cleanup;
212
213 if (!values_close(anon, current, 3))
214 goto cleanup;
215
216 ret = 0;
217 cleanup:
218 free(buf);
219 return ret;
220 }
221
alloc_pagecache_50M_check(const char * cgroup,void * arg)222 static int alloc_pagecache_50M_check(const char *cgroup, void *arg)
223 {
224 size_t size = MB(50);
225 int ret = -1;
226 long current, file;
227 int fd;
228
229 fd = get_temp_fd();
230 if (fd < 0)
231 return -1;
232
233 if (alloc_pagecache(fd, size))
234 goto cleanup;
235
236 current = cg_read_long(cgroup, "memory.current");
237 if (current < size)
238 goto cleanup;
239
240 file = cg_read_key_long(cgroup, "memory.stat", "file ");
241 if (file < 0)
242 goto cleanup;
243
244 if (!values_close(file, current, 10))
245 goto cleanup;
246
247 ret = 0;
248
249 cleanup:
250 close(fd);
251 return ret;
252 }
253
254 /*
255 * This test create a memory cgroup, allocates
256 * some anonymous memory and some pagecache
257 * and checks memory.current, memory.peak, and some memory.stat values.
258 */
test_memcg_current_peak(const char * root)259 static int test_memcg_current_peak(const char *root)
260 {
261 int ret = KSFT_FAIL;
262 long current, peak, peak_reset;
263 char *memcg;
264 bool fd2_closed = false, fd3_closed = false, fd4_closed = false;
265 int peak_fd = -1, peak_fd2 = -1, peak_fd3 = -1, peak_fd4 = -1;
266 struct stat ss;
267
268 memcg = cg_name(root, "memcg_test");
269 if (!memcg)
270 goto cleanup;
271
272 if (cg_create(memcg))
273 goto cleanup;
274
275 current = cg_read_long(memcg, "memory.current");
276 if (current != 0)
277 goto cleanup;
278
279 peak = cg_read_long(memcg, "memory.peak");
280 if (peak != 0)
281 goto cleanup;
282
283 if (cg_run(memcg, alloc_anon_50M_check, NULL))
284 goto cleanup;
285
286 peak = cg_read_long(memcg, "memory.peak");
287 if (peak < MB(50))
288 goto cleanup;
289
290 /*
291 * We'll open a few FDs for the same memory.peak file to exercise the free-path
292 * We need at least three to be closed in a different order than writes occurred to test
293 * the linked-list handling.
294 */
295 peak_fd = cg_open(memcg, "memory.peak", O_RDWR | O_APPEND | O_CLOEXEC);
296
297 if (peak_fd == -1) {
298 if (errno == ENOENT)
299 ret = KSFT_SKIP;
300 goto cleanup;
301 }
302
303 /*
304 * Before we try to use memory.peak's fd, try to figure out whether
305 * this kernel supports writing to that file in the first place. (by
306 * checking the writable bit on the file's st_mode)
307 */
308 if (fstat(peak_fd, &ss))
309 goto cleanup;
310
311 if ((ss.st_mode & S_IWUSR) == 0) {
312 ret = KSFT_SKIP;
313 goto cleanup;
314 }
315
316 peak_fd2 = cg_open(memcg, "memory.peak", O_RDWR | O_APPEND | O_CLOEXEC);
317
318 if (peak_fd2 == -1)
319 goto cleanup;
320
321 peak_fd3 = cg_open(memcg, "memory.peak", O_RDWR | O_APPEND | O_CLOEXEC);
322
323 if (peak_fd3 == -1)
324 goto cleanup;
325
326 /* any non-empty string resets, but make it clear */
327 static const char reset_string[] = "reset\n";
328
329 peak_reset = write(peak_fd, reset_string, sizeof(reset_string));
330 if (peak_reset != sizeof(reset_string))
331 goto cleanup;
332
333 peak_reset = write(peak_fd2, reset_string, sizeof(reset_string));
334 if (peak_reset != sizeof(reset_string))
335 goto cleanup;
336
337 peak_reset = write(peak_fd3, reset_string, sizeof(reset_string));
338 if (peak_reset != sizeof(reset_string))
339 goto cleanup;
340
341 /* Make sure a completely independent read isn't affected by our FD-local reset above*/
342 peak = cg_read_long(memcg, "memory.peak");
343 if (peak < MB(50))
344 goto cleanup;
345
346 fd2_closed = true;
347 if (close(peak_fd2))
348 goto cleanup;
349
350 peak_fd4 = cg_open(memcg, "memory.peak", O_RDWR | O_APPEND | O_CLOEXEC);
351
352 if (peak_fd4 == -1)
353 goto cleanup;
354
355 peak_reset = write(peak_fd4, reset_string, sizeof(reset_string));
356 if (peak_reset != sizeof(reset_string))
357 goto cleanup;
358
359 peak = cg_read_long_fd(peak_fd);
360 if (peak > MB(30) || peak < 0)
361 goto cleanup;
362
363 if (cg_run(memcg, alloc_pagecache_50M_check, NULL))
364 goto cleanup;
365
366 peak = cg_read_long(memcg, "memory.peak");
367 if (peak < MB(50))
368 goto cleanup;
369
370 /* Make sure everything is back to normal */
371 peak = cg_read_long_fd(peak_fd);
372 if (peak < MB(50))
373 goto cleanup;
374
375 peak = cg_read_long_fd(peak_fd4);
376 if (peak < MB(50))
377 goto cleanup;
378
379 fd3_closed = true;
380 if (close(peak_fd3))
381 goto cleanup;
382
383 fd4_closed = true;
384 if (close(peak_fd4))
385 goto cleanup;
386
387 ret = KSFT_PASS;
388
389 cleanup:
390 close(peak_fd);
391 if (!fd2_closed)
392 close(peak_fd2);
393 if (!fd3_closed)
394 close(peak_fd3);
395 if (!fd4_closed)
396 close(peak_fd4);
397 cg_destroy(memcg);
398 free(memcg);
399
400 return ret;
401 }
402
alloc_pagecache_50M_noexit(const char * cgroup,void * arg)403 static int alloc_pagecache_50M_noexit(const char *cgroup, void *arg)
404 {
405 int fd = (long)arg;
406 int ppid = getppid();
407
408 if (alloc_pagecache(fd, MB(50)))
409 return -1;
410
411 while (getppid() == ppid)
412 sleep(1);
413
414 return 0;
415 }
416
alloc_anon_noexit(const char * cgroup,void * arg)417 static int alloc_anon_noexit(const char *cgroup, void *arg)
418 {
419 int ppid = getppid();
420 size_t size = (unsigned long)arg;
421 char *buf;
422
423 buf = alloc_and_populate_anon(size);
424 if (!buf)
425 return -1;
426
427 while (getppid() == ppid)
428 sleep(1);
429
430 free(buf);
431 return 0;
432 }
433
434 /*
435 * Wait until processes are killed asynchronously by the OOM killer
436 * If we exceed a timeout, fail.
437 */
cg_test_proc_killed(const char * cgroup)438 static int cg_test_proc_killed(const char *cgroup)
439 {
440 int limit;
441
442 for (limit = 10; limit > 0; limit--) {
443 if (cg_read_strcmp(cgroup, "cgroup.procs", "") == 0)
444 return 0;
445
446 usleep(100000);
447 }
448 return -1;
449 }
450
451 static bool reclaim_until(const char *memcg, long goal);
452
453 /*
454 * First, this test creates the following hierarchy:
455 * A memory.min = 0, memory.max = 200M
456 * A/B memory.min = 50M
457 * A/B/C memory.min = 75M, memory.current = 50M
458 * A/B/D memory.min = 25M, memory.current = 50M
459 * A/B/E memory.min = 0, memory.current = 50M
460 * A/B/F memory.min = 500M, memory.current = 0
461 *
462 * (or memory.low if we test soft protection)
463 *
464 * Usages are pagecache and the test keeps a running
465 * process in every leaf cgroup.
466 * Then it creates A/G and creates a significant
467 * memory pressure in A.
468 *
469 * Then it checks actual memory usages and expects that:
470 * A/B memory.current ~= 50M
471 * A/B/C memory.current ~= 29M [memory.events:low > 0]
472 * A/B/D memory.current ~= 21M [memory.events:low > 0]
473 * A/B/E memory.current ~= 0 [memory.events:low == 0 if !memory_recursiveprot,
474 * undefined otherwise]
475 * A/B/F memory.current = 0 [memory.events:low == 0]
476 * (for origin of the numbers, see model in memcg_protection.m.)
477 *
478 * After that it tries to allocate more than there is
479 * unprotected memory in A available, and checks that:
480 * a) memory.min protects pagecache even in this case,
481 * b) memory.low allows reclaiming page cache with low events.
482 *
483 * Then we try to reclaim from A/B/C using memory.reclaim until its
484 * usage reaches 10M.
485 * This makes sure that:
486 * (a) We ignore the protection of the reclaim target memcg.
487 * (b) The previously calculated emin value (~29M) should be dismissed.
488 */
test_memcg_protection(const char * root,bool min)489 static int test_memcg_protection(const char *root, bool min)
490 {
491 int ret = KSFT_FAIL, rc;
492 char *parent[3] = {NULL};
493 char *children[4] = {NULL};
494 const char *attribute = min ? "memory.min" : "memory.low";
495 long c[4];
496 long current;
497 int i, attempts;
498 int fd;
499
500 fd = get_temp_fd();
501 if (fd < 0)
502 goto cleanup;
503
504 parent[0] = cg_name(root, "memcg_test_0");
505 if (!parent[0])
506 goto cleanup;
507
508 parent[1] = cg_name(parent[0], "memcg_test_1");
509 if (!parent[1])
510 goto cleanup;
511
512 parent[2] = cg_name(parent[0], "memcg_test_2");
513 if (!parent[2])
514 goto cleanup;
515
516 if (cg_create(parent[0]))
517 goto cleanup;
518
519 if (cg_read_long(parent[0], attribute)) {
520 /* No memory.min on older kernels is fine */
521 if (min)
522 ret = KSFT_SKIP;
523 goto cleanup;
524 }
525
526 if (cg_write(parent[0], "cgroup.subtree_control", "+memory"))
527 goto cleanup;
528
529 if (cg_write(parent[0], "memory.max", "200M"))
530 goto cleanup;
531
532 if (cg_write(parent[0], "memory.swap.max", "0"))
533 goto cleanup;
534
535 if (cg_create(parent[1]))
536 goto cleanup;
537
538 if (cg_write(parent[1], "cgroup.subtree_control", "+memory"))
539 goto cleanup;
540
541 if (cg_create(parent[2]))
542 goto cleanup;
543
544 for (i = 0; i < ARRAY_SIZE(children); i++) {
545 children[i] = cg_name_indexed(parent[1], "child_memcg", i);
546 if (!children[i])
547 goto cleanup;
548
549 if (cg_create(children[i]))
550 goto cleanup;
551
552 if (i > 2)
553 continue;
554
555 cg_run_nowait(children[i], alloc_pagecache_50M_noexit,
556 (void *)(long)fd);
557 }
558
559 if (cg_write(parent[1], attribute, "50M"))
560 goto cleanup;
561 if (cg_write(children[0], attribute, "75M"))
562 goto cleanup;
563 if (cg_write(children[1], attribute, "25M"))
564 goto cleanup;
565 if (cg_write(children[2], attribute, "0"))
566 goto cleanup;
567 if (cg_write(children[3], attribute, "500M"))
568 goto cleanup;
569
570 attempts = 0;
571 while (!values_close(cg_read_long(parent[1], "memory.current"),
572 MB(150), 3)) {
573 if (attempts++ > 5)
574 break;
575 sleep(1);
576 }
577
578 if (cg_run(parent[2], alloc_anon, (void *)MB(148)))
579 goto cleanup;
580
581 if (!values_close(cg_read_long(parent[1], "memory.current"), MB(50), 3))
582 goto cleanup;
583
584 for (i = 0; i < ARRAY_SIZE(children); i++)
585 c[i] = cg_read_long(children[i], "memory.current");
586
587 if (!values_close(c[0], MB(29), 15))
588 goto cleanup;
589
590 if (!values_close(c[1], MB(21), 20))
591 goto cleanup;
592
593 if (c[3] != 0)
594 goto cleanup;
595
596 rc = cg_run(parent[2], alloc_anon, (void *)MB(170));
597 if (min && !rc)
598 goto cleanup;
599 else if (!min && rc) {
600 fprintf(stderr,
601 "memory.low prevents from allocating anon memory\n");
602 goto cleanup;
603 }
604
605 current = min ? MB(50) : MB(30);
606 if (!values_close(cg_read_long(parent[1], "memory.current"), current, 3))
607 goto cleanup;
608
609 if (!reclaim_until(children[0], MB(10)))
610 goto cleanup;
611
612 if (min) {
613 ret = KSFT_PASS;
614 goto cleanup;
615 }
616
617 /*
618 * Child 2 has memory.low=0, but some low protection may still be
619 * distributed down from its parent with memory.low=50M if cgroup2
620 * memory_recursiveprot mount option is enabled. Ignore the low
621 * event count in this case.
622 */
623 for (i = 0; i < ARRAY_SIZE(children); i++) {
624 int ignore_low_events_index = has_recursiveprot ? 2 : -1;
625 int no_low_events_index = 1;
626 long low, oom;
627
628 oom = cg_read_key_long(children[i], "memory.events", "oom ");
629 low = cg_read_key_long(children[i], "memory.events", "low ");
630
631 if (oom)
632 goto cleanup;
633 if (i == ignore_low_events_index)
634 continue;
635 if (i <= no_low_events_index && low <= 0)
636 goto cleanup;
637 if (i > no_low_events_index && low)
638 goto cleanup;
639
640 }
641
642 ret = KSFT_PASS;
643
644 cleanup:
645 for (i = ARRAY_SIZE(children) - 1; i >= 0; i--) {
646 if (!children[i])
647 continue;
648
649 cg_destroy(children[i]);
650 free(children[i]);
651 }
652
653 for (i = ARRAY_SIZE(parent) - 1; i >= 0; i--) {
654 if (!parent[i])
655 continue;
656
657 cg_destroy(parent[i]);
658 free(parent[i]);
659 }
660 close(fd);
661 return ret;
662 }
663
test_memcg_min(const char * root)664 static int test_memcg_min(const char *root)
665 {
666 return test_memcg_protection(root, true);
667 }
668
test_memcg_low(const char * root)669 static int test_memcg_low(const char *root)
670 {
671 return test_memcg_protection(root, false);
672 }
673
alloc_pagecache_max_30M(const char * cgroup,void * arg)674 static int alloc_pagecache_max_30M(const char *cgroup, void *arg)
675 {
676 size_t size = MB(50);
677 int ret = -1;
678 long current, high, max;
679 int fd;
680
681 high = cg_read_long(cgroup, "memory.high");
682 max = cg_read_long(cgroup, "memory.max");
683 if (high != MB(30) && max != MB(30))
684 return -1;
685
686 fd = get_temp_fd();
687 if (fd < 0)
688 return -1;
689
690 if (alloc_pagecache(fd, size))
691 goto cleanup;
692
693 current = cg_read_long(cgroup, "memory.current");
694 if (!values_close(current, MB(30), 5))
695 goto cleanup;
696
697 ret = 0;
698
699 cleanup:
700 close(fd);
701 return ret;
702
703 }
704
705 /*
706 * This test checks that memory.high limits the amount of
707 * memory which can be consumed by either anonymous memory
708 * or pagecache.
709 */
test_memcg_high(const char * root)710 static int test_memcg_high(const char *root)
711 {
712 int ret = KSFT_FAIL;
713 char *memcg;
714 long high;
715
716 memcg = cg_name(root, "memcg_test");
717 if (!memcg)
718 goto cleanup;
719
720 if (cg_create(memcg))
721 goto cleanup;
722
723 if (cg_read_strcmp(memcg, "memory.high", "max\n"))
724 goto cleanup;
725
726 if (cg_write(memcg, "memory.swap.max", "0"))
727 goto cleanup;
728
729 if (cg_write(memcg, "memory.high", "30M"))
730 goto cleanup;
731
732 if (cg_run(memcg, alloc_anon, (void *)MB(31)))
733 goto cleanup;
734
735 if (!cg_run(memcg, alloc_pagecache_50M_check, NULL))
736 goto cleanup;
737
738 if (cg_run(memcg, alloc_pagecache_max_30M, NULL))
739 goto cleanup;
740
741 high = cg_read_key_long(memcg, "memory.events", "high ");
742 if (high <= 0)
743 goto cleanup;
744
745 ret = KSFT_PASS;
746
747 cleanup:
748 cg_destroy(memcg);
749 free(memcg);
750
751 return ret;
752 }
753
alloc_anon_mlock(const char * cgroup,void * arg)754 static int alloc_anon_mlock(const char *cgroup, void *arg)
755 {
756 size_t size = (size_t)arg;
757 void *buf;
758
759 buf = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON,
760 0, 0);
761 if (buf == MAP_FAILED)
762 return -1;
763
764 mlock(buf, size);
765 munmap(buf, size);
766 return 0;
767 }
768
769 /*
770 * This test checks that memory.high is able to throttle big single shot
771 * allocation i.e. large allocation within one kernel entry.
772 */
test_memcg_high_sync(const char * root)773 static int test_memcg_high_sync(const char *root)
774 {
775 int ret = KSFT_FAIL, pid, fd = -1;
776 char *memcg;
777 long pre_high, pre_max;
778 long post_high, post_max;
779
780 memcg = cg_name(root, "memcg_test");
781 if (!memcg)
782 goto cleanup;
783
784 if (cg_create(memcg))
785 goto cleanup;
786
787 pre_high = cg_read_key_long(memcg, "memory.events", "high ");
788 pre_max = cg_read_key_long(memcg, "memory.events", "max ");
789 if (pre_high < 0 || pre_max < 0)
790 goto cleanup;
791
792 if (cg_write(memcg, "memory.swap.max", "0"))
793 goto cleanup;
794
795 if (cg_write(memcg, "memory.high", "30M"))
796 goto cleanup;
797
798 if (cg_write(memcg, "memory.max", "140M"))
799 goto cleanup;
800
801 fd = memcg_prepare_for_wait(memcg);
802 if (fd < 0)
803 goto cleanup;
804
805 pid = cg_run_nowait(memcg, alloc_anon_mlock, (void *)MB(200));
806 if (pid < 0)
807 goto cleanup;
808
809 cg_wait_for(fd);
810
811 post_high = cg_read_key_long(memcg, "memory.events", "high ");
812 post_max = cg_read_key_long(memcg, "memory.events", "max ");
813 if (post_high < 0 || post_max < 0)
814 goto cleanup;
815
816 if (pre_high == post_high || pre_max != post_max)
817 goto cleanup;
818
819 ret = KSFT_PASS;
820
821 cleanup:
822 if (fd >= 0)
823 close(fd);
824 cg_destroy(memcg);
825 free(memcg);
826
827 return ret;
828 }
829
830 /*
831 * This test checks that memory.max limits the amount of
832 * memory which can be consumed by either anonymous memory
833 * or pagecache.
834 */
test_memcg_max(const char * root)835 static int test_memcg_max(const char *root)
836 {
837 int ret = KSFT_FAIL;
838 char *memcg;
839 long current, max;
840
841 memcg = cg_name(root, "memcg_test");
842 if (!memcg)
843 goto cleanup;
844
845 if (cg_create(memcg))
846 goto cleanup;
847
848 if (cg_read_strcmp(memcg, "memory.max", "max\n"))
849 goto cleanup;
850
851 if (cg_write(memcg, "memory.swap.max", "0"))
852 goto cleanup;
853
854 if (cg_write(memcg, "memory.max", "30M"))
855 goto cleanup;
856
857 /* Should be killed by OOM killer */
858 if (!cg_run(memcg, alloc_anon, (void *)MB(100)))
859 goto cleanup;
860
861 if (cg_run(memcg, alloc_pagecache_max_30M, NULL))
862 goto cleanup;
863
864 current = cg_read_long(memcg, "memory.current");
865 if (current > MB(30) || !current)
866 goto cleanup;
867
868 max = cg_read_key_long(memcg, "memory.events", "max ");
869 if (max <= 0)
870 goto cleanup;
871
872 ret = KSFT_PASS;
873
874 cleanup:
875 cg_destroy(memcg);
876 free(memcg);
877
878 return ret;
879 }
880
881 /*
882 * Reclaim from @memcg until usage reaches @goal by writing to
883 * memory.reclaim.
884 *
885 * This function will return false if the usage is already below the
886 * goal.
887 *
888 * This function assumes that writing to memory.reclaim is the only
889 * source of change in memory.current (no concurrent allocations or
890 * reclaim).
891 *
892 * This function makes sure memory.reclaim is sane. It will return
893 * false if memory.reclaim's error codes do not make sense, even if
894 * the usage goal was satisfied.
895 */
reclaim_until(const char * memcg,long goal)896 static bool reclaim_until(const char *memcg, long goal)
897 {
898 char buf[64];
899 int retries, err;
900 long current, to_reclaim;
901 bool reclaimed = false;
902
903 for (retries = 5; retries > 0; retries--) {
904 current = cg_read_long(memcg, "memory.current");
905
906 if (current < goal || values_close(current, goal, 3))
907 break;
908 /* Did memory.reclaim return 0 incorrectly? */
909 else if (reclaimed)
910 return false;
911
912 to_reclaim = current - goal;
913 snprintf(buf, sizeof(buf), "%ld", to_reclaim);
914 err = cg_write(memcg, "memory.reclaim", buf);
915 if (!err)
916 reclaimed = true;
917 else if (err != -EAGAIN)
918 return false;
919 }
920 return reclaimed;
921 }
922
923 /*
924 * This test checks that memory.reclaim reclaims the given
925 * amount of memory (from both anon and file, if possible).
926 */
test_memcg_reclaim(const char * root)927 static int test_memcg_reclaim(const char *root)
928 {
929 int ret = KSFT_FAIL;
930 int fd = -1;
931 int retries;
932 char *memcg;
933 long current, expected_usage;
934
935 memcg = cg_name(root, "memcg_test");
936 if (!memcg)
937 goto cleanup;
938
939 if (cg_create(memcg))
940 goto cleanup;
941
942 current = cg_read_long(memcg, "memory.current");
943 if (current != 0)
944 goto cleanup;
945
946 fd = get_temp_fd();
947 if (fd < 0)
948 goto cleanup;
949
950 cg_run_nowait(memcg, alloc_pagecache_50M_noexit, (void *)(long)fd);
951
952 /*
953 * If swap is enabled, try to reclaim from both anon and file, else try
954 * to reclaim from file only.
955 */
956 if (is_swap_enabled()) {
957 cg_run_nowait(memcg, alloc_anon_noexit, (void *) MB(50));
958 expected_usage = MB(100);
959 } else
960 expected_usage = MB(50);
961
962 /*
963 * Wait until current usage reaches the expected usage (or we run out of
964 * retries).
965 */
966 retries = 5;
967 while (!values_close(cg_read_long(memcg, "memory.current"),
968 expected_usage, 10)) {
969 if (retries--) {
970 sleep(1);
971 continue;
972 } else {
973 fprintf(stderr,
974 "failed to allocate %ld for memcg reclaim test\n",
975 expected_usage);
976 goto cleanup;
977 }
978 }
979
980 /*
981 * Reclaim until current reaches 30M, this makes sure we hit both anon
982 * and file if swap is enabled.
983 */
984 if (!reclaim_until(memcg, MB(30)))
985 goto cleanup;
986
987 ret = KSFT_PASS;
988 cleanup:
989 cg_destroy(memcg);
990 free(memcg);
991 close(fd);
992
993 return ret;
994 }
995
alloc_anon_50M_check_swap(const char * cgroup,void * arg)996 static int alloc_anon_50M_check_swap(const char *cgroup, void *arg)
997 {
998 long mem_max = (long)arg;
999 size_t size = MB(50);
1000 char *buf;
1001 long mem_current, swap_current;
1002 int ret = -1;
1003
1004 buf = alloc_and_populate_anon(size);
1005 if (!buf)
1006 return -1;
1007
1008 mem_current = cg_read_long(cgroup, "memory.current");
1009 if (!mem_current || !values_close(mem_current, mem_max, 3))
1010 goto cleanup;
1011
1012 swap_current = cg_read_long(cgroup, "memory.swap.current");
1013 if (!swap_current ||
1014 !values_close(mem_current + swap_current, size, 3))
1015 goto cleanup;
1016
1017 ret = 0;
1018 cleanup:
1019 free(buf);
1020 return ret;
1021 }
1022
1023 /*
1024 * This test checks that memory.swap.max limits the amount of
1025 * anonymous memory which can be swapped out. Additionally, it verifies that
1026 * memory.swap.peak reflects the high watermark and can be reset.
1027 */
test_memcg_swap_max_peak(const char * root)1028 static int test_memcg_swap_max_peak(const char *root)
1029 {
1030 int ret = KSFT_FAIL;
1031 char *memcg;
1032 long max, peak;
1033 struct stat ss;
1034 int swap_peak_fd = -1, mem_peak_fd = -1;
1035
1036 /* any non-empty string resets */
1037 static const char reset_string[] = "foobarbaz";
1038
1039 if (!is_swap_enabled())
1040 return KSFT_SKIP;
1041
1042 memcg = cg_name(root, "memcg_test");
1043 if (!memcg)
1044 goto cleanup;
1045
1046 if (cg_create(memcg))
1047 goto cleanup;
1048
1049 if (cg_read_long(memcg, "memory.swap.current")) {
1050 ret = KSFT_SKIP;
1051 goto cleanup;
1052 }
1053
1054 swap_peak_fd = cg_open(memcg, "memory.swap.peak",
1055 O_RDWR | O_APPEND | O_CLOEXEC);
1056
1057 if (swap_peak_fd == -1) {
1058 if (errno == ENOENT)
1059 ret = KSFT_SKIP;
1060 goto cleanup;
1061 }
1062
1063 /*
1064 * Before we try to use memory.swap.peak's fd, try to figure out
1065 * whether this kernel supports writing to that file in the first
1066 * place. (by checking the writable bit on the file's st_mode)
1067 */
1068 if (fstat(swap_peak_fd, &ss))
1069 goto cleanup;
1070
1071 if ((ss.st_mode & S_IWUSR) == 0) {
1072 ret = KSFT_SKIP;
1073 goto cleanup;
1074 }
1075
1076 mem_peak_fd = cg_open(memcg, "memory.peak", O_RDWR | O_APPEND | O_CLOEXEC);
1077
1078 if (mem_peak_fd == -1)
1079 goto cleanup;
1080
1081 if (cg_read_long(memcg, "memory.swap.peak"))
1082 goto cleanup;
1083
1084 if (cg_read_long_fd(swap_peak_fd))
1085 goto cleanup;
1086
1087 /* switch the swap and mem fds into local-peak tracking mode*/
1088 int peak_reset = write(swap_peak_fd, reset_string, sizeof(reset_string));
1089
1090 if (peak_reset != sizeof(reset_string))
1091 goto cleanup;
1092
1093 if (cg_read_long_fd(swap_peak_fd))
1094 goto cleanup;
1095
1096 if (cg_read_long(memcg, "memory.peak"))
1097 goto cleanup;
1098
1099 if (cg_read_long_fd(mem_peak_fd))
1100 goto cleanup;
1101
1102 peak_reset = write(mem_peak_fd, reset_string, sizeof(reset_string));
1103 if (peak_reset != sizeof(reset_string))
1104 goto cleanup;
1105
1106 if (cg_read_long_fd(mem_peak_fd))
1107 goto cleanup;
1108
1109 if (cg_read_strcmp(memcg, "memory.max", "max\n"))
1110 goto cleanup;
1111
1112 if (cg_read_strcmp(memcg, "memory.swap.max", "max\n"))
1113 goto cleanup;
1114
1115 if (cg_write(memcg, "memory.swap.max", "30M"))
1116 goto cleanup;
1117
1118 if (cg_write(memcg, "memory.max", "30M"))
1119 goto cleanup;
1120
1121 /* Should be killed by OOM killer */
1122 if (!cg_run(memcg, alloc_anon, (void *)MB(100)))
1123 goto cleanup;
1124
1125 if (cg_read_key_long(memcg, "memory.events", "oom ") != 1)
1126 goto cleanup;
1127
1128 if (cg_read_key_long(memcg, "memory.events", "oom_kill ") != 1)
1129 goto cleanup;
1130
1131 peak = cg_read_long(memcg, "memory.peak");
1132 if (peak < MB(29))
1133 goto cleanup;
1134
1135 peak = cg_read_long(memcg, "memory.swap.peak");
1136 if (peak < MB(29))
1137 goto cleanup;
1138
1139 peak = cg_read_long_fd(mem_peak_fd);
1140 if (peak < MB(29))
1141 goto cleanup;
1142
1143 peak = cg_read_long_fd(swap_peak_fd);
1144 if (peak < MB(29))
1145 goto cleanup;
1146
1147 /*
1148 * open, reset and close the peak swap on another FD to make sure
1149 * multiple extant fds don't corrupt the linked-list
1150 */
1151 peak_reset = cg_write(memcg, "memory.swap.peak", (char *)reset_string);
1152 if (peak_reset)
1153 goto cleanup;
1154
1155 peak_reset = cg_write(memcg, "memory.peak", (char *)reset_string);
1156 if (peak_reset)
1157 goto cleanup;
1158
1159 /* actually reset on the fds */
1160 peak_reset = write(swap_peak_fd, reset_string, sizeof(reset_string));
1161 if (peak_reset != sizeof(reset_string))
1162 goto cleanup;
1163
1164 peak_reset = write(mem_peak_fd, reset_string, sizeof(reset_string));
1165 if (peak_reset != sizeof(reset_string))
1166 goto cleanup;
1167
1168 peak = cg_read_long_fd(swap_peak_fd);
1169 if (peak > MB(10))
1170 goto cleanup;
1171
1172 /*
1173 * The cgroup is now empty, but there may be a page or two associated
1174 * with the open FD accounted to it.
1175 */
1176 peak = cg_read_long_fd(mem_peak_fd);
1177 if (peak > MB(1))
1178 goto cleanup;
1179
1180 if (cg_read_long(memcg, "memory.peak") < MB(29))
1181 goto cleanup;
1182
1183 if (cg_read_long(memcg, "memory.swap.peak") < MB(29))
1184 goto cleanup;
1185
1186 if (cg_run(memcg, alloc_anon_50M_check_swap, (void *)MB(30)))
1187 goto cleanup;
1188
1189 max = cg_read_key_long(memcg, "memory.events", "max ");
1190 if (max <= 0)
1191 goto cleanup;
1192
1193 peak = cg_read_long(memcg, "memory.peak");
1194 if (peak < MB(29))
1195 goto cleanup;
1196
1197 peak = cg_read_long(memcg, "memory.swap.peak");
1198 if (peak < MB(29))
1199 goto cleanup;
1200
1201 peak = cg_read_long_fd(mem_peak_fd);
1202 if (peak < MB(29))
1203 goto cleanup;
1204
1205 peak = cg_read_long_fd(swap_peak_fd);
1206 if (peak < MB(19))
1207 goto cleanup;
1208
1209 ret = KSFT_PASS;
1210
1211 cleanup:
1212 if (mem_peak_fd != -1 && close(mem_peak_fd))
1213 ret = KSFT_FAIL;
1214 if (swap_peak_fd != -1 && close(swap_peak_fd))
1215 ret = KSFT_FAIL;
1216 cg_destroy(memcg);
1217 free(memcg);
1218
1219 return ret;
1220 }
1221
1222 /*
1223 * This test disables swapping and tries to allocate anonymous memory
1224 * up to OOM. Then it checks for oom and oom_kill events in
1225 * memory.events.
1226 */
test_memcg_oom_events(const char * root)1227 static int test_memcg_oom_events(const char *root)
1228 {
1229 int ret = KSFT_FAIL;
1230 char *memcg;
1231
1232 memcg = cg_name(root, "memcg_test");
1233 if (!memcg)
1234 goto cleanup;
1235
1236 if (cg_create(memcg))
1237 goto cleanup;
1238
1239 if (cg_write(memcg, "memory.max", "30M"))
1240 goto cleanup;
1241
1242 if (cg_write(memcg, "memory.swap.max", "0"))
1243 goto cleanup;
1244
1245 if (!cg_run(memcg, alloc_anon, (void *)MB(100)))
1246 goto cleanup;
1247
1248 if (cg_read_strcmp(memcg, "cgroup.procs", ""))
1249 goto cleanup;
1250
1251 if (cg_read_key_long(memcg, "memory.events", "oom ") != 1)
1252 goto cleanup;
1253
1254 if (cg_read_key_long(memcg, "memory.events", "oom_kill ") != 1)
1255 goto cleanup;
1256
1257 ret = KSFT_PASS;
1258
1259 cleanup:
1260 cg_destroy(memcg);
1261 free(memcg);
1262
1263 return ret;
1264 }
1265
1266 struct tcp_server_args {
1267 unsigned short port;
1268 int ctl[2];
1269 };
1270
tcp_server(const char * cgroup,void * arg)1271 static int tcp_server(const char *cgroup, void *arg)
1272 {
1273 struct tcp_server_args *srv_args = arg;
1274 struct sockaddr_in6 saddr = { 0 };
1275 socklen_t slen = sizeof(saddr);
1276 int sk, client_sk, ctl_fd, yes = 1, ret = -1;
1277
1278 close(srv_args->ctl[0]);
1279 ctl_fd = srv_args->ctl[1];
1280
1281 saddr.sin6_family = AF_INET6;
1282 saddr.sin6_addr = in6addr_any;
1283 saddr.sin6_port = htons(srv_args->port);
1284
1285 sk = socket(AF_INET6, SOCK_STREAM, 0);
1286 if (sk < 0) {
1287 /* Pass back errno to the ctl_fd */
1288 write(ctl_fd, &errno, sizeof(errno));
1289 return ret;
1290 }
1291
1292 if (setsockopt(sk, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes)) < 0)
1293 goto cleanup;
1294
1295 if (bind(sk, (struct sockaddr *)&saddr, slen)) {
1296 write(ctl_fd, &errno, sizeof(errno));
1297 goto cleanup;
1298 }
1299
1300 if (listen(sk, 1))
1301 goto cleanup;
1302
1303 ret = 0;
1304 if (write(ctl_fd, &ret, sizeof(ret)) != sizeof(ret)) {
1305 ret = -1;
1306 goto cleanup;
1307 }
1308
1309 client_sk = accept(sk, NULL, NULL);
1310 if (client_sk < 0)
1311 goto cleanup;
1312
1313 ret = -1;
1314 for (;;) {
1315 uint8_t buf[0x100000];
1316
1317 if (write(client_sk, buf, sizeof(buf)) <= 0) {
1318 if (errno == ECONNRESET)
1319 ret = 0;
1320 break;
1321 }
1322 }
1323
1324 close(client_sk);
1325
1326 cleanup:
1327 close(sk);
1328 return ret;
1329 }
1330
tcp_client(const char * cgroup,unsigned short port)1331 static int tcp_client(const char *cgroup, unsigned short port)
1332 {
1333 const char server[] = "localhost";
1334 struct addrinfo *ai;
1335 char servport[6];
1336 int retries = 0x10; /* nice round number */
1337 int sk, ret;
1338 long allocated;
1339
1340 allocated = cg_read_long(cgroup, "memory.current");
1341 snprintf(servport, sizeof(servport), "%hd", port);
1342 ret = getaddrinfo(server, servport, NULL, &ai);
1343 if (ret)
1344 return ret;
1345
1346 sk = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol);
1347 if (sk < 0)
1348 goto free_ainfo;
1349
1350 ret = connect(sk, ai->ai_addr, ai->ai_addrlen);
1351 if (ret < 0)
1352 goto close_sk;
1353
1354 ret = KSFT_FAIL;
1355 while (retries--) {
1356 uint8_t buf[0x100000];
1357 long current, sock;
1358
1359 if (read(sk, buf, sizeof(buf)) <= 0)
1360 goto close_sk;
1361
1362 current = cg_read_long(cgroup, "memory.current");
1363 sock = cg_read_key_long(cgroup, "memory.stat", "sock ");
1364
1365 if (current < 0 || sock < 0)
1366 goto close_sk;
1367
1368 /* exclude the memory not related to socket connection */
1369 if (values_close(current - allocated, sock, 10)) {
1370 ret = KSFT_PASS;
1371 break;
1372 }
1373 }
1374
1375 close_sk:
1376 close(sk);
1377 free_ainfo:
1378 freeaddrinfo(ai);
1379 return ret;
1380 }
1381
1382 /*
1383 * This test checks socket memory accounting.
1384 * The test forks a TCP server listens on a random port between 1000
1385 * and 61000. Once it gets a client connection, it starts writing to
1386 * its socket.
1387 * The TCP client interleaves reads from the socket with check whether
1388 * memory.current and memory.stat.sock are similar.
1389 */
test_memcg_sock(const char * root)1390 static int test_memcg_sock(const char *root)
1391 {
1392 int bind_retries = 5, ret = KSFT_FAIL, pid, err;
1393 unsigned short port;
1394 char *memcg;
1395 long sock_post = -1;
1396
1397 memcg = cg_name(root, "memcg_test");
1398 if (!memcg)
1399 goto cleanup;
1400
1401 if (cg_create(memcg))
1402 goto cleanup;
1403
1404 while (bind_retries--) {
1405 struct tcp_server_args args;
1406
1407 if (pipe(args.ctl))
1408 goto cleanup;
1409
1410 port = args.port = 1000 + rand() % 60000;
1411
1412 pid = cg_run_nowait(memcg, tcp_server, &args);
1413 if (pid < 0)
1414 goto cleanup;
1415
1416 close(args.ctl[1]);
1417 if (read(args.ctl[0], &err, sizeof(err)) != sizeof(err))
1418 goto cleanup;
1419 close(args.ctl[0]);
1420
1421 /* Skip if address family not supported by protocol */
1422 if (err == EAFNOSUPPORT) {
1423 ret = KSFT_SKIP;
1424 goto cleanup;
1425 }
1426
1427 if (!err)
1428 break;
1429 if (err != EADDRINUSE)
1430 goto cleanup;
1431
1432 waitpid(pid, NULL, 0);
1433 }
1434
1435 if (err == EADDRINUSE) {
1436 ret = KSFT_SKIP;
1437 goto cleanup;
1438 }
1439
1440 if (tcp_client(memcg, port) != KSFT_PASS)
1441 goto cleanup;
1442
1443 waitpid(pid, &err, 0);
1444 if (WEXITSTATUS(err))
1445 goto cleanup;
1446
1447 if (cg_read_long(memcg, "memory.current") < 0)
1448 goto cleanup;
1449
1450 /*
1451 * memory.stat is updated asynchronously via the memcg rstat
1452 * flushing worker, which runs periodically (every 2 seconds,
1453 * see FLUSH_TIME). On a busy system, the "sock " counter may
1454 * stay non-zero for a short period of time after the TCP
1455 * connection is closed and all socket memory has been
1456 * uncharged.
1457 *
1458 * Poll memory.stat for up to 3 seconds (~FLUSH_TIME plus some
1459 * scheduling slack) and require that the "sock " counter
1460 * eventually drops to zero.
1461 */
1462 sock_post = cg_read_key_long_poll(memcg, "memory.stat", "sock ", 0,
1463 MEMCG_SOCKSTAT_WAIT_RETRIES,
1464 DEFAULT_WAIT_INTERVAL_US);
1465 if (sock_post)
1466 goto cleanup;
1467
1468 ret = KSFT_PASS;
1469
1470 cleanup:
1471 cg_destroy(memcg);
1472 free(memcg);
1473
1474 return ret;
1475 }
1476
1477 /*
1478 * This test disables swapping and tries to allocate anonymous memory
1479 * up to OOM with memory.group.oom set. Then it checks that all
1480 * processes in the leaf were killed. It also checks that oom_events
1481 * were propagated to the parent level.
1482 */
test_memcg_oom_group_leaf_events(const char * root)1483 static int test_memcg_oom_group_leaf_events(const char *root)
1484 {
1485 int ret = KSFT_FAIL;
1486 char *parent, *child;
1487 long parent_oom_events;
1488
1489 parent = cg_name(root, "memcg_test_0");
1490 child = cg_name(root, "memcg_test_0/memcg_test_1");
1491
1492 if (!parent || !child)
1493 goto cleanup;
1494
1495 if (cg_create(parent))
1496 goto cleanup;
1497
1498 if (cg_create(child))
1499 goto cleanup;
1500
1501 if (cg_write(parent, "cgroup.subtree_control", "+memory"))
1502 goto cleanup;
1503
1504 if (cg_write(child, "memory.max", "50M"))
1505 goto cleanup;
1506
1507 if (cg_write(child, "memory.swap.max", "0"))
1508 goto cleanup;
1509
1510 if (cg_write(child, "memory.oom.group", "1"))
1511 goto cleanup;
1512
1513 cg_run_nowait(parent, alloc_anon_noexit, (void *) MB(60));
1514 cg_run_nowait(child, alloc_anon_noexit, (void *) MB(1));
1515 cg_run_nowait(child, alloc_anon_noexit, (void *) MB(1));
1516 if (!cg_run(child, alloc_anon, (void *)MB(100)))
1517 goto cleanup;
1518
1519 if (cg_test_proc_killed(child))
1520 goto cleanup;
1521
1522 if (cg_read_key_long(child, "memory.events", "oom_kill ") <= 0)
1523 goto cleanup;
1524
1525 parent_oom_events = cg_read_key_long(
1526 parent, "memory.events", "oom_kill ");
1527 /*
1528 * If memory_localevents is not enabled (the default), the parent should
1529 * count OOM events in its children groups. Otherwise, it should not
1530 * have observed any events.
1531 */
1532 if (has_localevents && parent_oom_events != 0)
1533 goto cleanup;
1534 else if (!has_localevents && parent_oom_events <= 0)
1535 goto cleanup;
1536
1537 ret = KSFT_PASS;
1538
1539 cleanup:
1540 if (child)
1541 cg_destroy(child);
1542 if (parent)
1543 cg_destroy(parent);
1544 free(child);
1545 free(parent);
1546
1547 return ret;
1548 }
1549
1550 /*
1551 * This test disables swapping and tries to allocate anonymous memory
1552 * up to OOM with memory.group.oom set. Then it checks that all
1553 * processes in the parent and leaf were killed.
1554 */
test_memcg_oom_group_parent_events(const char * root)1555 static int test_memcg_oom_group_parent_events(const char *root)
1556 {
1557 int ret = KSFT_FAIL;
1558 char *parent, *child;
1559
1560 parent = cg_name(root, "memcg_test_0");
1561 child = cg_name(root, "memcg_test_0/memcg_test_1");
1562
1563 if (!parent || !child)
1564 goto cleanup;
1565
1566 if (cg_create(parent))
1567 goto cleanup;
1568
1569 if (cg_create(child))
1570 goto cleanup;
1571
1572 if (cg_write(parent, "memory.max", "80M"))
1573 goto cleanup;
1574
1575 if (cg_write(parent, "memory.swap.max", "0"))
1576 goto cleanup;
1577
1578 if (cg_write(parent, "memory.oom.group", "1"))
1579 goto cleanup;
1580
1581 cg_run_nowait(parent, alloc_anon_noexit, (void *) MB(60));
1582 cg_run_nowait(child, alloc_anon_noexit, (void *) MB(1));
1583 cg_run_nowait(child, alloc_anon_noexit, (void *) MB(1));
1584
1585 if (!cg_run(child, alloc_anon, (void *)MB(100)))
1586 goto cleanup;
1587
1588 if (cg_test_proc_killed(child))
1589 goto cleanup;
1590 if (cg_test_proc_killed(parent))
1591 goto cleanup;
1592
1593 ret = KSFT_PASS;
1594
1595 cleanup:
1596 if (child)
1597 cg_destroy(child);
1598 if (parent)
1599 cg_destroy(parent);
1600 free(child);
1601 free(parent);
1602
1603 return ret;
1604 }
1605
1606 /*
1607 * This test disables swapping and tries to allocate anonymous memory
1608 * up to OOM with memory.group.oom set. Then it checks that all
1609 * processes were killed except those set with OOM_SCORE_ADJ_MIN
1610 */
test_memcg_oom_group_score_events(const char * root)1611 static int test_memcg_oom_group_score_events(const char *root)
1612 {
1613 int ret = KSFT_FAIL;
1614 char *memcg;
1615 int safe_pid;
1616
1617 memcg = cg_name(root, "memcg_test_0");
1618
1619 if (!memcg)
1620 goto cleanup;
1621
1622 if (cg_create(memcg))
1623 goto cleanup;
1624
1625 if (cg_write(memcg, "memory.max", "50M"))
1626 goto cleanup;
1627
1628 if (cg_write(memcg, "memory.swap.max", "0"))
1629 goto cleanup;
1630
1631 if (cg_write(memcg, "memory.oom.group", "1"))
1632 goto cleanup;
1633
1634 safe_pid = cg_run_nowait(memcg, alloc_anon_noexit, (void *) MB(1));
1635 if (set_oom_adj_score(safe_pid, OOM_SCORE_ADJ_MIN))
1636 goto cleanup;
1637
1638 cg_run_nowait(memcg, alloc_anon_noexit, (void *) MB(1));
1639 if (!cg_run(memcg, alloc_anon, (void *)MB(100)))
1640 goto cleanup;
1641
1642 if (cg_read_key_long(memcg, "memory.events", "oom_kill ") != 3)
1643 goto cleanup;
1644
1645 if (kill(safe_pid, SIGKILL))
1646 goto cleanup;
1647
1648 ret = KSFT_PASS;
1649
1650 cleanup:
1651 if (memcg)
1652 cg_destroy(memcg);
1653 free(memcg);
1654
1655 return ret;
1656 }
1657
read_event(int inotify_fd,int expected_event,int expected_wd)1658 static int read_event(int inotify_fd, int expected_event, int expected_wd)
1659 {
1660 struct inotify_event event;
1661 ssize_t len = 0;
1662
1663 len = read(inotify_fd, &event, sizeof(event));
1664 if (len < (ssize_t)sizeof(event))
1665 return -1;
1666
1667 if (event.mask != expected_event || event.wd != expected_wd) {
1668 fprintf(stderr,
1669 "event does not match expected values: mask %d (expected %d) wd %d (expected %d)\n",
1670 event.mask, expected_event, event.wd, expected_wd);
1671 return -1;
1672 }
1673
1674 return 0;
1675 }
1676
test_memcg_inotify_delete_file(const char * root)1677 static int test_memcg_inotify_delete_file(const char *root)
1678 {
1679 int ret = KSFT_FAIL;
1680 char *memcg = NULL;
1681 int fd, wd;
1682
1683 memcg = cg_name(root, "memcg_test_0");
1684
1685 if (!memcg)
1686 goto cleanup;
1687
1688 if (cg_create(memcg))
1689 goto cleanup;
1690
1691 fd = inotify_init1(0);
1692 if (fd == -1)
1693 goto cleanup;
1694
1695 wd = inotify_add_watch(fd, cg_control(memcg, "memory.events"), IN_DELETE_SELF);
1696 if (wd == -1)
1697 goto cleanup;
1698
1699 if (cg_destroy(memcg))
1700 goto cleanup;
1701 free(memcg);
1702 memcg = NULL;
1703
1704 if (read_event(fd, IN_DELETE_SELF, wd))
1705 goto cleanup;
1706
1707 if (read_event(fd, IN_IGNORED, wd))
1708 goto cleanup;
1709
1710 ret = KSFT_PASS;
1711
1712 cleanup:
1713 if (fd >= 0)
1714 close(fd);
1715 if (memcg)
1716 cg_destroy(memcg);
1717 free(memcg);
1718
1719 return ret;
1720 }
1721
test_memcg_inotify_delete_dir(const char * root)1722 static int test_memcg_inotify_delete_dir(const char *root)
1723 {
1724 int ret = KSFT_FAIL;
1725 char *memcg = NULL;
1726 int fd, wd;
1727
1728 memcg = cg_name(root, "memcg_test_0");
1729
1730 if (!memcg)
1731 goto cleanup;
1732
1733 if (cg_create(memcg))
1734 goto cleanup;
1735
1736 fd = inotify_init1(0);
1737 if (fd == -1)
1738 goto cleanup;
1739
1740 wd = inotify_add_watch(fd, memcg, IN_DELETE_SELF);
1741 if (wd == -1)
1742 goto cleanup;
1743
1744 if (cg_destroy(memcg))
1745 goto cleanup;
1746 free(memcg);
1747 memcg = NULL;
1748
1749 if (read_event(fd, IN_DELETE_SELF, wd))
1750 goto cleanup;
1751
1752 if (read_event(fd, IN_IGNORED, wd))
1753 goto cleanup;
1754
1755 ret = KSFT_PASS;
1756
1757 cleanup:
1758 if (fd >= 0)
1759 close(fd);
1760 if (memcg)
1761 cg_destroy(memcg);
1762 free(memcg);
1763
1764 return ret;
1765 }
1766
1767 #define T(x) { x, #x }
1768 struct memcg_test {
1769 int (*fn)(const char *root);
1770 const char *name;
1771 } tests[] = {
1772 T(test_memcg_subtree_control),
1773 T(test_memcg_current_peak),
1774 T(test_memcg_min),
1775 T(test_memcg_low),
1776 T(test_memcg_high),
1777 T(test_memcg_high_sync),
1778 T(test_memcg_max),
1779 T(test_memcg_reclaim),
1780 T(test_memcg_oom_events),
1781 T(test_memcg_swap_max_peak),
1782 T(test_memcg_sock),
1783 T(test_memcg_oom_group_leaf_events),
1784 T(test_memcg_oom_group_parent_events),
1785 T(test_memcg_oom_group_score_events),
1786 T(test_memcg_inotify_delete_file),
1787 T(test_memcg_inotify_delete_dir),
1788 };
1789 #undef T
1790
main(int argc,char ** argv)1791 int main(int argc, char **argv)
1792 {
1793 char root[PATH_MAX];
1794 int i, proc_status;
1795
1796 page_size = sysconf(_SC_PAGE_SIZE);
1797 if (page_size <= 0)
1798 page_size = BUF_SIZE;
1799
1800 ksft_print_header();
1801 if (cg_find_unified_root(root, sizeof(root), NULL))
1802 ksft_exit_skip("cgroup v2 isn't mounted\n");
1803
1804 /*
1805 * Check that memory controller is available:
1806 * memory is listed in cgroup.controllers
1807 */
1808 if (cg_read_strstr(root, "cgroup.controllers", "memory"))
1809 ksft_exit_skip("memory controller isn't available\n");
1810
1811 if (cg_read_strstr(root, "cgroup.subtree_control", "memory"))
1812 if (cg_write(root, "cgroup.subtree_control", "+memory"))
1813 ksft_exit_skip("Failed to set memory controller\n");
1814
1815 proc_status = proc_mount_contains("memory_recursiveprot");
1816 if (proc_status < 0)
1817 ksft_exit_skip("Failed to query cgroup mount option\n");
1818 has_recursiveprot = proc_status;
1819
1820 proc_status = proc_mount_contains("memory_localevents");
1821 if (proc_status < 0)
1822 ksft_exit_skip("Failed to query cgroup mount option\n");
1823 has_localevents = proc_status;
1824
1825 ksft_set_plan(ARRAY_SIZE(tests));
1826 for (i = 0; i < ARRAY_SIZE(tests); i++) {
1827 switch (tests[i].fn(root)) {
1828 case KSFT_PASS:
1829 ksft_test_result_pass("%s\n", tests[i].name);
1830 break;
1831 case KSFT_SKIP:
1832 ksft_test_result_skip("%s\n", tests[i].name);
1833 break;
1834 default:
1835 ksft_test_result_fail("%s\n", tests[i].name);
1836 break;
1837 }
1838 }
1839
1840 ksft_finished();
1841 }
1842