1 // SPDX-License-Identifier: GPL-2.0
2 #define _GNU_SOURCE
3 #include <sched.h>
4 #include <sys/mount.h>
5 #include <sys/stat.h>
6 #include <sys/types.h>
7 #include <linux/limits.h>
8 #include <stdio.h>
9 #include <stdlib.h>
10 #include <linux/sched.h>
11 #include <fcntl.h>
12 #include <unistd.h>
13 #include <ftw.h>
14
15 #include "cgroup_helpers.h"
16 #include "bpf_util.h"
17
18 /*
19 * To avoid relying on the system setup, when setup_cgroup_env is called
20 * we create a new mount namespace, and cgroup namespace. The cgroupv2
21 * root is mounted at CGROUP_MOUNT_PATH. Unfortunately, most people don't
22 * have cgroupv2 enabled at this point in time. It's easier to create our
23 * own mount namespace and manage it ourselves. We assume /mnt exists.
24 *
25 * Related cgroupv1 helpers are named *classid*(), since we only use the
26 * net_cls controller for tagging net_cls.classid. We assume the default
27 * mount under /sys/fs/cgroup/net_cls, which should be the case for the
28 * vast majority of users.
29 */
30
31 #define WALK_FD_LIMIT 16
32
33 #define CGROUP_MOUNT_PATH "/mnt"
34 #define CGROUP_MOUNT_DFLT "/sys/fs/cgroup"
35 #define NETCLS_MOUNT_PATH CGROUP_MOUNT_DFLT "/net_cls"
36 #define CGROUP_WORK_DIR "/cgroup-test-work-dir"
37
38 #define format_cgroup_path_pid(buf, path, pid) \
39 snprintf(buf, sizeof(buf), "%s%s%d%s", CGROUP_MOUNT_PATH, \
40 CGROUP_WORK_DIR, pid, path)
41
42 #define format_cgroup_path(buf, path) \
43 format_cgroup_path_pid(buf, path, getpid())
44
45 #define format_parent_cgroup_path(buf, path) \
46 format_cgroup_path_pid(buf, path, getppid())
47
48 #define format_classid_path(buf) \
49 snprintf(buf, sizeof(buf), "%s%s", NETCLS_MOUNT_PATH, \
50 CGROUP_WORK_DIR)
51
__enable_controllers(const char * cgroup_path,const char * controllers)52 static int __enable_controllers(const char *cgroup_path, const char *controllers)
53 {
54 char path[PATH_MAX + 1];
55 char enable[PATH_MAX + 1];
56 char *c, *c2;
57 int fd, cfd;
58 ssize_t len;
59
60 /* If not controllers are passed, enable all available controllers */
61 if (!controllers) {
62 snprintf(path, sizeof(path), "%s/cgroup.controllers",
63 cgroup_path);
64 fd = open(path, O_RDONLY);
65 if (fd < 0) {
66 log_err("Opening cgroup.controllers: %s", path);
67 return 1;
68 }
69 len = read(fd, enable, sizeof(enable) - 1);
70 if (len < 0) {
71 close(fd);
72 log_err("Reading cgroup.controllers: %s", path);
73 return 1;
74 } else if (len == 0) { /* No controllers to enable */
75 close(fd);
76 return 0;
77 }
78 enable[len] = 0;
79 close(fd);
80 } else {
81 bpf_strlcpy(enable, controllers, sizeof(enable));
82 }
83
84 snprintf(path, sizeof(path), "%s/cgroup.subtree_control", cgroup_path);
85 cfd = open(path, O_RDWR);
86 if (cfd < 0) {
87 log_err("Opening cgroup.subtree_control: %s", path);
88 return 1;
89 }
90
91 for (c = strtok_r(enable, " ", &c2); c; c = strtok_r(NULL, " ", &c2)) {
92 if (dprintf(cfd, "+%s\n", c) <= 0) {
93 log_err("Enabling controller %s: %s", c, path);
94 close(cfd);
95 return 1;
96 }
97 }
98 close(cfd);
99 return 0;
100 }
101
102 /**
103 * enable_controllers() - Enable cgroup v2 controllers
104 * @relative_path: The cgroup path, relative to the workdir
105 * @controllers: List of controllers to enable in cgroup.controllers format
106 *
107 *
108 * Enable given cgroup v2 controllers, if @controllers is NULL, enable all
109 * available controllers.
110 *
111 * If successful, 0 is returned.
112 */
enable_controllers(const char * relative_path,const char * controllers)113 int enable_controllers(const char *relative_path, const char *controllers)
114 {
115 char cgroup_path[PATH_MAX + 1];
116
117 format_cgroup_path(cgroup_path, relative_path);
118 return __enable_controllers(cgroup_path, controllers);
119 }
120
__write_cgroup_file(const char * cgroup_path,const char * file,const char * buf)121 static int __write_cgroup_file(const char *cgroup_path, const char *file,
122 const char *buf)
123 {
124 char file_path[PATH_MAX + 1];
125 int fd;
126
127 snprintf(file_path, sizeof(file_path), "%s/%s", cgroup_path, file);
128 fd = open(file_path, O_RDWR);
129 if (fd < 0) {
130 log_err("Opening %s", file_path);
131 return 1;
132 }
133
134 if (dprintf(fd, "%s", buf) <= 0) {
135 log_err("Writing to %s", file_path);
136 close(fd);
137 return 1;
138 }
139 close(fd);
140 return 0;
141 }
142
143 /**
144 * write_cgroup_file() - Write to a cgroup file
145 * @relative_path: The cgroup path, relative to the workdir
146 * @file: The name of the file in cgroupfs to write to
147 * @buf: Buffer to write to the file
148 *
149 * Write to a file in the given cgroup's directory.
150 *
151 * If successful, 0 is returned.
152 */
write_cgroup_file(const char * relative_path,const char * file,const char * buf)153 int write_cgroup_file(const char *relative_path, const char *file,
154 const char *buf)
155 {
156 char cgroup_path[PATH_MAX - 24];
157
158 format_cgroup_path(cgroup_path, relative_path);
159 return __write_cgroup_file(cgroup_path, file, buf);
160 }
161
162 /**
163 * write_cgroup_file_parent() - Write to a cgroup file in the parent process
164 * workdir
165 * @relative_path: The cgroup path, relative to the parent process workdir
166 * @file: The name of the file in cgroupfs to write to
167 * @buf: Buffer to write to the file
168 *
169 * Write to a file in the given cgroup's directory under the parent process
170 * workdir.
171 *
172 * If successful, 0 is returned.
173 */
write_cgroup_file_parent(const char * relative_path,const char * file,const char * buf)174 int write_cgroup_file_parent(const char *relative_path, const char *file,
175 const char *buf)
176 {
177 char cgroup_path[PATH_MAX - 24];
178
179 format_parent_cgroup_path(cgroup_path, relative_path);
180 return __write_cgroup_file(cgroup_path, file, buf);
181 }
182
183 /**
184 * setup_cgroup_environment() - Setup the cgroup environment
185 *
186 * After calling this function, cleanup_cgroup_environment should be called
187 * once testing is complete.
188 *
189 * This function will print an error to stderr and return 1 if it is unable
190 * to setup the cgroup environment. If setup is successful, 0 is returned.
191 */
setup_cgroup_environment(void)192 int setup_cgroup_environment(void)
193 {
194 char cgroup_workdir[PATH_MAX - 24];
195
196 format_cgroup_path(cgroup_workdir, "");
197
198 if (unshare(CLONE_NEWNS)) {
199 log_err("unshare");
200 return 1;
201 }
202
203 if (mount("none", "/", NULL, MS_REC | MS_PRIVATE, NULL)) {
204 log_err("mount fakeroot");
205 return 1;
206 }
207
208 if (mount("none", CGROUP_MOUNT_PATH, "cgroup2", 0, NULL) && errno != EBUSY) {
209 log_err("mount cgroup2");
210 return 1;
211 }
212
213 /* Cleanup existing failed runs, now that the environment is setup */
214 cleanup_cgroup_environment();
215
216 if (mkdir(cgroup_workdir, 0777) && errno != EEXIST) {
217 log_err("mkdir cgroup work dir");
218 return 1;
219 }
220
221 /* Enable all available controllers to increase test coverage */
222 if (__enable_controllers(CGROUP_MOUNT_PATH, NULL) ||
223 __enable_controllers(cgroup_workdir, NULL))
224 return 1;
225
226 return 0;
227 }
228
nftwfunc(const char * filename,const struct stat * statptr,int fileflags,struct FTW * pfwt)229 static int nftwfunc(const char *filename, const struct stat *statptr,
230 int fileflags, struct FTW *pfwt)
231 {
232 if ((fileflags & FTW_D) && rmdir(filename))
233 log_err("Removing cgroup: %s", filename);
234 return 0;
235 }
236
join_cgroup_from_top(const char * cgroup_path)237 static int join_cgroup_from_top(const char *cgroup_path)
238 {
239 char cgroup_procs_path[PATH_MAX + 1];
240 pid_t pid = getpid();
241 int fd, rc = 0;
242
243 snprintf(cgroup_procs_path, sizeof(cgroup_procs_path),
244 "%s/cgroup.procs", cgroup_path);
245
246 fd = open(cgroup_procs_path, O_WRONLY);
247 if (fd < 0) {
248 log_err("Opening Cgroup Procs: %s", cgroup_procs_path);
249 return 1;
250 }
251
252 if (dprintf(fd, "%d\n", pid) < 0) {
253 log_err("Joining Cgroup");
254 rc = 1;
255 }
256
257 close(fd);
258 return rc;
259 }
260
261 /**
262 * join_cgroup() - Join a cgroup
263 * @relative_path: The cgroup path, relative to the workdir, to join
264 *
265 * This function expects a cgroup to already be created, relative to the cgroup
266 * work dir, and it joins it. For example, passing "/my-cgroup" as the path
267 * would actually put the calling process into the cgroup
268 * "/cgroup-test-work-dir/my-cgroup"
269 *
270 * On success, it returns 0, otherwise on failure it returns 1.
271 */
join_cgroup(const char * relative_path)272 int join_cgroup(const char *relative_path)
273 {
274 char cgroup_path[PATH_MAX + 1];
275
276 format_cgroup_path(cgroup_path, relative_path);
277 return join_cgroup_from_top(cgroup_path);
278 }
279
280 /**
281 * join_parent_cgroup() - Join a cgroup in the parent process workdir
282 * @relative_path: The cgroup path, relative to parent process workdir, to join
283 *
284 * See join_cgroup().
285 *
286 * On success, it returns 0, otherwise on failure it returns 1.
287 */
join_parent_cgroup(const char * relative_path)288 int join_parent_cgroup(const char *relative_path)
289 {
290 char cgroup_path[PATH_MAX + 1];
291
292 format_parent_cgroup_path(cgroup_path, relative_path);
293 return join_cgroup_from_top(cgroup_path);
294 }
295
296 /**
297 * cleanup_cgroup_environment() - Cleanup Cgroup Testing Environment
298 *
299 * This is an idempotent function to delete all temporary cgroups that
300 * have been created during the test, including the cgroup testing work
301 * directory.
302 *
303 * At call time, it moves the calling process to the root cgroup, and then
304 * runs the deletion process. It is idempotent, and should not fail, unless
305 * a process is lingering.
306 *
307 * On failure, it will print an error to stderr, and try to continue.
308 */
cleanup_cgroup_environment(void)309 void cleanup_cgroup_environment(void)
310 {
311 char cgroup_workdir[PATH_MAX + 1];
312
313 format_cgroup_path(cgroup_workdir, "");
314 join_cgroup_from_top(CGROUP_MOUNT_PATH);
315 nftw(cgroup_workdir, nftwfunc, WALK_FD_LIMIT, FTW_DEPTH | FTW_MOUNT);
316 }
317
318 /**
319 * get_root_cgroup() - Get the FD of the root cgroup
320 *
321 * On success, it returns the file descriptor. On failure, it returns -1.
322 * If there is a failure, it prints the error to stderr.
323 */
get_root_cgroup(void)324 int get_root_cgroup(void)
325 {
326 int fd;
327
328 fd = open(CGROUP_MOUNT_PATH, O_RDONLY);
329 if (fd < 0) {
330 log_err("Opening root cgroup");
331 return -1;
332 }
333 return fd;
334 }
335
336 /*
337 * remove_cgroup() - Remove a cgroup
338 * @relative_path: The cgroup path, relative to the workdir, to remove
339 *
340 * This function expects a cgroup to already be created, relative to the cgroup
341 * work dir. It also expects the cgroup doesn't have any children or live
342 * processes and it removes the cgroup.
343 *
344 * On failure, it will print an error to stderr.
345 */
remove_cgroup(const char * relative_path)346 void remove_cgroup(const char *relative_path)
347 {
348 char cgroup_path[PATH_MAX + 1];
349
350 format_cgroup_path(cgroup_path, relative_path);
351 if (rmdir(cgroup_path))
352 log_err("rmdiring cgroup %s .. %s", relative_path, cgroup_path);
353 }
354
355 /**
356 * create_and_get_cgroup() - Create a cgroup, relative to workdir, and get the FD
357 * @relative_path: The cgroup path, relative to the workdir, to join
358 *
359 * This function creates a cgroup under the top level workdir and returns the
360 * file descriptor. It is idempotent.
361 *
362 * On success, it returns the file descriptor. On failure it returns -1.
363 * If there is a failure, it prints the error to stderr.
364 */
create_and_get_cgroup(const char * relative_path)365 int create_and_get_cgroup(const char *relative_path)
366 {
367 char cgroup_path[PATH_MAX + 1];
368 int fd;
369
370 format_cgroup_path(cgroup_path, relative_path);
371 if (mkdir(cgroup_path, 0777) && errno != EEXIST) {
372 log_err("mkdiring cgroup %s .. %s", relative_path, cgroup_path);
373 return -1;
374 }
375
376 fd = open(cgroup_path, O_RDONLY);
377 if (fd < 0) {
378 log_err("Opening Cgroup");
379 return -1;
380 }
381
382 return fd;
383 }
384
385 /**
386 * get_cgroup_id() - Get cgroup id for a particular cgroup path
387 * @relative_path: The cgroup path, relative to the workdir, to join
388 *
389 * On success, it returns the cgroup id. On failure it returns 0,
390 * which is an invalid cgroup id.
391 * If there is a failure, it prints the error to stderr.
392 */
get_cgroup_id(const char * relative_path)393 unsigned long long get_cgroup_id(const char *relative_path)
394 {
395 int dirfd, err, flags, mount_id, fhsize;
396 union {
397 unsigned long long cgid;
398 unsigned char raw_bytes[8];
399 } id;
400 char cgroup_workdir[PATH_MAX + 1];
401 struct file_handle *fhp, *fhp2;
402 unsigned long long ret = 0;
403
404 format_cgroup_path(cgroup_workdir, relative_path);
405
406 dirfd = AT_FDCWD;
407 flags = 0;
408 fhsize = sizeof(*fhp);
409 fhp = calloc(1, fhsize);
410 if (!fhp) {
411 log_err("calloc");
412 return 0;
413 }
414 err = name_to_handle_at(dirfd, cgroup_workdir, fhp, &mount_id, flags);
415 if (err >= 0 || fhp->handle_bytes != 8) {
416 log_err("name_to_handle_at");
417 goto free_mem;
418 }
419
420 fhsize = sizeof(struct file_handle) + fhp->handle_bytes;
421 fhp2 = realloc(fhp, fhsize);
422 if (!fhp2) {
423 log_err("realloc");
424 goto free_mem;
425 }
426 err = name_to_handle_at(dirfd, cgroup_workdir, fhp2, &mount_id, flags);
427 fhp = fhp2;
428 if (err < 0) {
429 log_err("name_to_handle_at");
430 goto free_mem;
431 }
432
433 memcpy(id.raw_bytes, fhp->f_handle, 8);
434 ret = id.cgid;
435
436 free_mem:
437 free(fhp);
438 return ret;
439 }
440
cgroup_setup_and_join(const char * path)441 int cgroup_setup_and_join(const char *path) {
442 int cg_fd;
443
444 if (setup_cgroup_environment()) {
445 fprintf(stderr, "Failed to setup cgroup environment\n");
446 return -EINVAL;
447 }
448
449 cg_fd = create_and_get_cgroup(path);
450 if (cg_fd < 0) {
451 fprintf(stderr, "Failed to create test cgroup\n");
452 cleanup_cgroup_environment();
453 return cg_fd;
454 }
455
456 if (join_cgroup(path)) {
457 fprintf(stderr, "Failed to join cgroup\n");
458 cleanup_cgroup_environment();
459 return -EINVAL;
460 }
461 return cg_fd;
462 }
463
464 /**
465 * setup_classid_environment() - Setup the cgroupv1 net_cls environment
466 *
467 * After calling this function, cleanup_classid_environment should be called
468 * once testing is complete.
469 *
470 * This function will print an error to stderr and return 1 if it is unable
471 * to setup the cgroup environment. If setup is successful, 0 is returned.
472 */
setup_classid_environment(void)473 int setup_classid_environment(void)
474 {
475 char cgroup_workdir[PATH_MAX + 1];
476
477 format_classid_path(cgroup_workdir);
478
479 if (mount("tmpfs", CGROUP_MOUNT_DFLT, "tmpfs", 0, NULL) &&
480 errno != EBUSY) {
481 log_err("mount cgroup base");
482 return 1;
483 }
484
485 if (mkdir(NETCLS_MOUNT_PATH, 0777) && errno != EEXIST) {
486 log_err("mkdir cgroup net_cls");
487 return 1;
488 }
489
490 if (mount("net_cls", NETCLS_MOUNT_PATH, "cgroup", 0, "net_cls") &&
491 errno != EBUSY) {
492 log_err("mount cgroup net_cls");
493 return 1;
494 }
495
496 cleanup_classid_environment();
497
498 if (mkdir(cgroup_workdir, 0777) && errno != EEXIST) {
499 log_err("mkdir cgroup work dir");
500 return 1;
501 }
502
503 return 0;
504 }
505
506 /**
507 * set_classid() - Set a cgroupv1 net_cls classid
508 * @id: the numeric classid
509 *
510 * Writes the passed classid into the cgroup work dir's net_cls.classid
511 * file in order to later on trigger socket tagging.
512 *
513 * On success, it returns 0, otherwise on failure it returns 1. If there
514 * is a failure, it prints the error to stderr.
515 */
set_classid(unsigned int id)516 int set_classid(unsigned int id)
517 {
518 char cgroup_workdir[PATH_MAX - 42];
519 char cgroup_classid_path[PATH_MAX + 1];
520 int fd, rc = 0;
521
522 format_classid_path(cgroup_workdir);
523 snprintf(cgroup_classid_path, sizeof(cgroup_classid_path),
524 "%s/net_cls.classid", cgroup_workdir);
525
526 fd = open(cgroup_classid_path, O_WRONLY);
527 if (fd < 0) {
528 log_err("Opening cgroup classid: %s", cgroup_classid_path);
529 return 1;
530 }
531
532 if (dprintf(fd, "%u\n", id) < 0) {
533 log_err("Setting cgroup classid");
534 rc = 1;
535 }
536
537 close(fd);
538 return rc;
539 }
540
541 /**
542 * join_classid() - Join a cgroupv1 net_cls classid
543 *
544 * This function expects the cgroup work dir to be already created, as we
545 * join it here. This causes the process sockets to be tagged with the given
546 * net_cls classid.
547 *
548 * On success, it returns 0, otherwise on failure it returns 1.
549 */
join_classid(void)550 int join_classid(void)
551 {
552 char cgroup_workdir[PATH_MAX + 1];
553
554 format_classid_path(cgroup_workdir);
555 return join_cgroup_from_top(cgroup_workdir);
556 }
557
558 /**
559 * cleanup_classid_environment() - Cleanup the cgroupv1 net_cls environment
560 *
561 * At call time, it moves the calling process to the root cgroup, and then
562 * runs the deletion process.
563 *
564 * On failure, it will print an error to stderr, and try to continue.
565 */
cleanup_classid_environment(void)566 void cleanup_classid_environment(void)
567 {
568 char cgroup_workdir[PATH_MAX + 1];
569
570 format_classid_path(cgroup_workdir);
571 join_cgroup_from_top(NETCLS_MOUNT_PATH);
572 nftw(cgroup_workdir, nftwfunc, WALK_FD_LIMIT, FTW_DEPTH | FTW_MOUNT);
573 }
574