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 
16 #include "cgroup_helpers.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 cgroup2
21  * root is mounted at CGROUP_MOUNT_PATH
22  *
23  * Unfortunately, most people don't have cgroupv2 enabled at this point in time.
24  * It's easier to create our own mount namespace and manage it ourselves.
25  *
26  * We assume /mnt exists.
27  */
28 
29 #define WALK_FD_LIMIT			16
30 #define CGROUP_MOUNT_PATH		"/mnt"
31 #define CGROUP_WORK_DIR			"/cgroup-test-work-dir"
32 #define format_cgroup_path(buf, path) \
33 	snprintf(buf, sizeof(buf), "%s%s%s", CGROUP_MOUNT_PATH, \
34 		 CGROUP_WORK_DIR, path)
35 
36 /**
37  * enable_all_controllers() - Enable all available cgroup v2 controllers
38  *
39  * Enable all available cgroup v2 controllers in order to increase
40  * the code coverage.
41  *
42  * If successful, 0 is returned.
43  */
44 int enable_all_controllers(char *cgroup_path)
45 {
46 	char path[PATH_MAX + 1];
47 	char buf[PATH_MAX];
48 	char *c, *c2;
49 	int fd, cfd;
50 	ssize_t len;
51 
52 	snprintf(path, sizeof(path), "%s/cgroup.controllers", cgroup_path);
53 	fd = open(path, O_RDONLY);
54 	if (fd < 0) {
55 		log_err("Opening cgroup.controllers: %s", path);
56 		return 1;
57 	}
58 
59 	len = read(fd, buf, sizeof(buf) - 1);
60 	if (len < 0) {
61 		close(fd);
62 		log_err("Reading cgroup.controllers: %s", path);
63 		return 1;
64 	}
65 	buf[len] = 0;
66 	close(fd);
67 
68 	/* No controllers available? We're probably on cgroup v1. */
69 	if (len == 0)
70 		return 0;
71 
72 	snprintf(path, sizeof(path), "%s/cgroup.subtree_control", cgroup_path);
73 	cfd = open(path, O_RDWR);
74 	if (cfd < 0) {
75 		log_err("Opening cgroup.subtree_control: %s", path);
76 		return 1;
77 	}
78 
79 	for (c = strtok_r(buf, " ", &c2); c; c = strtok_r(NULL, " ", &c2)) {
80 		if (dprintf(cfd, "+%s\n", c) <= 0) {
81 			log_err("Enabling controller %s: %s", c, path);
82 			close(cfd);
83 			return 1;
84 		}
85 	}
86 	close(cfd);
87 	return 0;
88 }
89 
90 /**
91  * setup_cgroup_environment() - Setup the cgroup environment
92  *
93  * After calling this function, cleanup_cgroup_environment should be called
94  * once testing is complete.
95  *
96  * This function will print an error to stderr and return 1 if it is unable
97  * to setup the cgroup environment. If setup is successful, 0 is returned.
98  */
99 int setup_cgroup_environment(void)
100 {
101 	char cgroup_workdir[PATH_MAX + 1];
102 
103 	format_cgroup_path(cgroup_workdir, "");
104 
105 	if (unshare(CLONE_NEWNS)) {
106 		log_err("unshare");
107 		return 1;
108 	}
109 
110 	if (mount("none", "/", NULL, MS_REC | MS_PRIVATE, NULL)) {
111 		log_err("mount fakeroot");
112 		return 1;
113 	}
114 
115 	if (mount("none", CGROUP_MOUNT_PATH, "cgroup2", 0, NULL) && errno != EBUSY) {
116 		log_err("mount cgroup2");
117 		return 1;
118 	}
119 
120 	/* Cleanup existing failed runs, now that the environment is setup */
121 	cleanup_cgroup_environment();
122 
123 	if (mkdir(cgroup_workdir, 0777) && errno != EEXIST) {
124 		log_err("mkdir cgroup work dir");
125 		return 1;
126 	}
127 
128 	if (enable_all_controllers(cgroup_workdir))
129 		return 1;
130 
131 	return 0;
132 }
133 
134 static int nftwfunc(const char *filename, const struct stat *statptr,
135 		    int fileflags, struct FTW *pfwt)
136 {
137 	if ((fileflags & FTW_D) && rmdir(filename))
138 		log_err("Removing cgroup: %s", filename);
139 	return 0;
140 }
141 
142 
143 static int join_cgroup_from_top(char *cgroup_path)
144 {
145 	char cgroup_procs_path[PATH_MAX + 1];
146 	pid_t pid = getpid();
147 	int fd, rc = 0;
148 
149 	snprintf(cgroup_procs_path, sizeof(cgroup_procs_path),
150 		 "%s/cgroup.procs", cgroup_path);
151 
152 	fd = open(cgroup_procs_path, O_WRONLY);
153 	if (fd < 0) {
154 		log_err("Opening Cgroup Procs: %s", cgroup_procs_path);
155 		return 1;
156 	}
157 
158 	if (dprintf(fd, "%d\n", pid) < 0) {
159 		log_err("Joining Cgroup");
160 		rc = 1;
161 	}
162 
163 	close(fd);
164 	return rc;
165 }
166 
167 /**
168  * join_cgroup() - Join a cgroup
169  * @path: The cgroup path, relative to the workdir, to join
170  *
171  * This function expects a cgroup to already be created, relative to the cgroup
172  * work dir, and it joins it. For example, passing "/my-cgroup" as the path
173  * would actually put the calling process into the cgroup
174  * "/cgroup-test-work-dir/my-cgroup"
175  *
176  * On success, it returns 0, otherwise on failure it returns 1.
177  */
178 int join_cgroup(const char *path)
179 {
180 	char cgroup_path[PATH_MAX + 1];
181 
182 	format_cgroup_path(cgroup_path, path);
183 	return join_cgroup_from_top(cgroup_path);
184 }
185 
186 /**
187  * cleanup_cgroup_environment() - Cleanup Cgroup Testing Environment
188  *
189  * This is an idempotent function to delete all temporary cgroups that
190  * have been created during the test, including the cgroup testing work
191  * directory.
192  *
193  * At call time, it moves the calling process to the root cgroup, and then
194  * runs the deletion process. It is idempotent, and should not fail, unless
195  * a process is lingering.
196  *
197  * On failure, it will print an error to stderr, and try to continue.
198  */
199 void cleanup_cgroup_environment(void)
200 {
201 	char cgroup_workdir[PATH_MAX + 1];
202 
203 	format_cgroup_path(cgroup_workdir, "");
204 	join_cgroup_from_top(CGROUP_MOUNT_PATH);
205 	nftw(cgroup_workdir, nftwfunc, WALK_FD_LIMIT, FTW_DEPTH | FTW_MOUNT);
206 }
207 
208 /**
209  * create_and_get_cgroup() - Create a cgroup, relative to workdir, and get the FD
210  * @path: The cgroup path, relative to the workdir, to join
211  *
212  * This function creates a cgroup under the top level workdir and returns the
213  * file descriptor. It is idempotent.
214  *
215  * On success, it returns the file descriptor. On failure it returns -1.
216  * If there is a failure, it prints the error to stderr.
217  */
218 int create_and_get_cgroup(const char *path)
219 {
220 	char cgroup_path[PATH_MAX + 1];
221 	int fd;
222 
223 	format_cgroup_path(cgroup_path, path);
224 	if (mkdir(cgroup_path, 0777) && errno != EEXIST) {
225 		log_err("mkdiring cgroup %s .. %s", path, cgroup_path);
226 		return -1;
227 	}
228 
229 	fd = open(cgroup_path, O_RDONLY);
230 	if (fd < 0) {
231 		log_err("Opening Cgroup");
232 		return -1;
233 	}
234 
235 	return fd;
236 }
237 
238 /**
239  * get_cgroup_id() - Get cgroup id for a particular cgroup path
240  * @path: The cgroup path, relative to the workdir, to join
241  *
242  * On success, it returns the cgroup id. On failure it returns 0,
243  * which is an invalid cgroup id.
244  * If there is a failure, it prints the error to stderr.
245  */
246 unsigned long long get_cgroup_id(const char *path)
247 {
248 	int dirfd, err, flags, mount_id, fhsize;
249 	union {
250 		unsigned long long cgid;
251 		unsigned char raw_bytes[8];
252 	} id;
253 	char cgroup_workdir[PATH_MAX + 1];
254 	struct file_handle *fhp, *fhp2;
255 	unsigned long long ret = 0;
256 
257 	format_cgroup_path(cgroup_workdir, path);
258 
259 	dirfd = AT_FDCWD;
260 	flags = 0;
261 	fhsize = sizeof(*fhp);
262 	fhp = calloc(1, fhsize);
263 	if (!fhp) {
264 		log_err("calloc");
265 		return 0;
266 	}
267 	err = name_to_handle_at(dirfd, cgroup_workdir, fhp, &mount_id, flags);
268 	if (err >= 0 || fhp->handle_bytes != 8) {
269 		log_err("name_to_handle_at");
270 		goto free_mem;
271 	}
272 
273 	fhsize = sizeof(struct file_handle) + fhp->handle_bytes;
274 	fhp2 = realloc(fhp, fhsize);
275 	if (!fhp2) {
276 		log_err("realloc");
277 		goto free_mem;
278 	}
279 	err = name_to_handle_at(dirfd, cgroup_workdir, fhp2, &mount_id, flags);
280 	fhp = fhp2;
281 	if (err < 0) {
282 		log_err("name_to_handle_at");
283 		goto free_mem;
284 	}
285 
286 	memcpy(id.raw_bytes, fhp->f_handle, 8);
287 	ret = id.cgid;
288 
289 free_mem:
290 	free(fhp);
291 	return ret;
292 }
293