1 // SPDX-License-Identifier: GPL-2.0-or-later
2
3 #define _GNU_SOURCE
4 #include "../kselftest_harness.h"
5 #include <errno.h>
6 #include <setjmp.h>
7 #include <signal.h>
8 #include <stdbool.h>
9 #include <stdio.h>
10 #include <stdlib.h>
11 #include <string.h>
12 #include <linux/mman.h>
13 #include <sys/syscall.h>
14 #include <unistd.h>
15 #include <sched.h>
16 #include "vm_util.h"
17
18 #include "../pidfd/pidfd.h"
19
FIXTURE(process_madvise)20 FIXTURE(process_madvise)
21 {
22 unsigned long page_size;
23 pid_t child_pid;
24 int remote_pidfd;
25 int pidfd;
26 };
27
FIXTURE_SETUP(process_madvise)28 FIXTURE_SETUP(process_madvise)
29 {
30 self->page_size = (unsigned long)sysconf(_SC_PAGESIZE);
31 self->pidfd = PIDFD_SELF;
32 self->remote_pidfd = -1;
33 self->child_pid = -1;
34 };
35
FIXTURE_TEARDOWN_PARENT(process_madvise)36 FIXTURE_TEARDOWN_PARENT(process_madvise)
37 {
38 /* This teardown is guaranteed to run, even if tests SKIP or ASSERT */
39 if (self->child_pid > 0) {
40 kill(self->child_pid, SIGKILL);
41 waitpid(self->child_pid, NULL, 0);
42 }
43
44 if (self->remote_pidfd >= 0)
45 close(self->remote_pidfd);
46 }
47
sys_process_madvise(int pidfd,const struct iovec * iovec,size_t vlen,int advice,unsigned int flags)48 static ssize_t sys_process_madvise(int pidfd, const struct iovec *iovec,
49 size_t vlen, int advice, unsigned int flags)
50 {
51 return syscall(__NR_process_madvise, pidfd, iovec, vlen, advice, flags);
52 }
53
54 /*
55 * This test uses PIDFD_SELF to target the current process. The main
56 * goal is to verify the basic behavior of process_madvise() with
57 * a vector of non-contiguous memory ranges, not its cross-process
58 * capabilities.
59 */
TEST_F(process_madvise,basic)60 TEST_F(process_madvise, basic)
61 {
62 const unsigned long pagesize = self->page_size;
63 const int madvise_pages = 4;
64 struct iovec vec[madvise_pages];
65 int pidfd = self->pidfd;
66 ssize_t ret;
67 char *map;
68
69 /*
70 * Create a single large mapping. We will pick pages from this
71 * mapping to advise on. This ensures we test non-contiguous iovecs.
72 */
73 map = mmap(NULL, pagesize * 10, PROT_READ | PROT_WRITE,
74 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
75 if (map == MAP_FAILED)
76 SKIP(return, "mmap failed, not enough memory.\n");
77
78 /* Fill the entire region with a known pattern. */
79 memset(map, 'A', pagesize * 10);
80
81 /*
82 * Setup the iovec to point to 4 non-contiguous pages
83 * within the mapping.
84 */
85 vec[0].iov_base = &map[0 * pagesize];
86 vec[0].iov_len = pagesize;
87 vec[1].iov_base = &map[3 * pagesize];
88 vec[1].iov_len = pagesize;
89 vec[2].iov_base = &map[5 * pagesize];
90 vec[2].iov_len = pagesize;
91 vec[3].iov_base = &map[8 * pagesize];
92 vec[3].iov_len = pagesize;
93
94 ret = sys_process_madvise(pidfd, vec, madvise_pages, MADV_DONTNEED, 0);
95 if (ret == -1 && errno == EPERM)
96 SKIP(return,
97 "process_madvise() unsupported or permission denied, try running as root.\n");
98 else if (errno == EINVAL)
99 SKIP(return,
100 "process_madvise() unsupported or parameter invalid, please check arguments.\n");
101
102 /* The call should succeed and report the total bytes processed. */
103 ASSERT_EQ(ret, madvise_pages * pagesize);
104
105 /* Check that advised pages are now zero. */
106 for (int i = 0; i < madvise_pages; i++) {
107 char *advised_page = (char *)vec[i].iov_base;
108
109 /* Content must be 0, not 'A'. */
110 ASSERT_EQ(*advised_page, '\0');
111 }
112
113 /* Check that an un-advised page in between is still 'A'. */
114 char *unadvised_page = &map[1 * pagesize];
115
116 for (int i = 0; i < pagesize; i++)
117 ASSERT_EQ(unadvised_page[i], 'A');
118
119 /* Cleanup. */
120 ASSERT_EQ(munmap(map, pagesize * 10), 0);
121 }
122
123 /*
124 * This test deterministically validates process_madvise() with MADV_COLLAPSE
125 * on a remote process, other advices are difficult to verify reliably.
126 *
127 * The test verifies that a memory region in a child process,
128 * focus on process_madv remote result, only check addresses and lengths.
129 * The correctness of the MADV_COLLAPSE can be found in the relevant test examples in khugepaged.
130 */
TEST_F(process_madvise,remote_collapse)131 TEST_F(process_madvise, remote_collapse)
132 {
133 const unsigned long pagesize = self->page_size;
134 long huge_page_size;
135 int pipe_info[2];
136 ssize_t ret;
137 struct iovec vec;
138
139 struct child_info {
140 pid_t pid;
141 void *map_addr;
142 } info;
143
144 huge_page_size = read_pmd_pagesize();
145 if (huge_page_size <= 0)
146 SKIP(return, "Could not determine a valid huge page size.\n");
147
148 ASSERT_EQ(pipe(pipe_info), 0);
149
150 self->child_pid = fork();
151 ASSERT_NE(self->child_pid, -1);
152
153 if (self->child_pid == 0) {
154 char *map;
155 size_t map_size = 2 * huge_page_size;
156
157 close(pipe_info[0]);
158
159 map = mmap(NULL, map_size, PROT_READ | PROT_WRITE,
160 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
161 ASSERT_NE(map, MAP_FAILED);
162
163 /* Fault in as small pages */
164 for (size_t i = 0; i < map_size; i += pagesize)
165 map[i] = 'A';
166
167 /* Send info and pause */
168 info.pid = getpid();
169 info.map_addr = map;
170 ret = write(pipe_info[1], &info, sizeof(info));
171 ASSERT_EQ(ret, sizeof(info));
172 close(pipe_info[1]);
173
174 pause();
175 exit(0);
176 }
177
178 close(pipe_info[1]);
179
180 /* Receive child info */
181 ret = read(pipe_info[0], &info, sizeof(info));
182 if (ret <= 0) {
183 waitpid(self->child_pid, NULL, 0);
184 SKIP(return, "Failed to read child info from pipe.\n");
185 }
186 ASSERT_EQ(ret, sizeof(info));
187 close(pipe_info[0]);
188 self->child_pid = info.pid;
189
190 self->remote_pidfd = syscall(__NR_pidfd_open, self->child_pid, 0);
191 ASSERT_GE(self->remote_pidfd, 0);
192
193 vec.iov_base = info.map_addr;
194 vec.iov_len = huge_page_size;
195
196 ret = sys_process_madvise(self->remote_pidfd, &vec, 1, MADV_COLLAPSE,
197 0);
198 if (ret == -1) {
199 if (errno == EINVAL)
200 SKIP(return, "PROCESS_MADV_ADVISE is not supported.\n");
201 else if (errno == EPERM)
202 SKIP(return,
203 "No process_madvise() permissions, try running as root.\n");
204 return;
205 }
206
207 ASSERT_EQ(ret, huge_page_size);
208 }
209
210 /*
211 * Test process_madvise() with a pidfd for a process that has already
212 * exited to ensure correct error handling.
213 */
TEST_F(process_madvise,exited_process_pidfd)214 TEST_F(process_madvise, exited_process_pidfd)
215 {
216 const unsigned long pagesize = self->page_size;
217 struct iovec vec;
218 char *map;
219 ssize_t ret;
220
221 map = mmap(NULL, pagesize, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1,
222 0);
223 if (map == MAP_FAILED)
224 SKIP(return, "mmap failed, not enough memory.\n");
225
226 vec.iov_base = map;
227 vec.iov_len = pagesize;
228
229 /*
230 * Using a pidfd for a process that has already exited should fail
231 * with ESRCH.
232 */
233 self->child_pid = fork();
234 ASSERT_NE(self->child_pid, -1);
235
236 if (self->child_pid == 0)
237 exit(0);
238
239 self->remote_pidfd = syscall(__NR_pidfd_open, self->child_pid, 0);
240 ASSERT_GE(self->remote_pidfd, 0);
241
242 /* Wait for the child to ensure it has terminated. */
243 waitpid(self->child_pid, NULL, 0);
244
245 ret = sys_process_madvise(self->remote_pidfd, &vec, 1, MADV_DONTNEED,
246 0);
247 ASSERT_EQ(ret, -1);
248 ASSERT_EQ(errno, ESRCH);
249 }
250
251 /*
252 * Test process_madvise() with bad pidfds to ensure correct error
253 * handling.
254 */
TEST_F(process_madvise,bad_pidfd)255 TEST_F(process_madvise, bad_pidfd)
256 {
257 const unsigned long pagesize = self->page_size;
258 struct iovec vec;
259 char *map;
260 ssize_t ret;
261
262 map = mmap(NULL, pagesize, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1,
263 0);
264 if (map == MAP_FAILED)
265 SKIP(return, "mmap failed, not enough memory.\n");
266
267 vec.iov_base = map;
268 vec.iov_len = pagesize;
269
270 /* Using an invalid fd number (-1) should fail with EBADF. */
271 ret = sys_process_madvise(-1, &vec, 1, MADV_DONTNEED, 0);
272 ASSERT_EQ(ret, -1);
273 ASSERT_EQ(errno, EBADF);
274
275 /*
276 * Using a valid fd that is not a pidfd (e.g. stdin) should fail
277 * with EBADF.
278 */
279 ret = sys_process_madvise(STDIN_FILENO, &vec, 1, MADV_DONTNEED, 0);
280 ASSERT_EQ(ret, -1);
281 ASSERT_EQ(errno, EBADF);
282 }
283
284 /*
285 * Test that process_madvise() rejects vlen > UIO_MAXIOV.
286 * The kernel should return -EINVAL when the number of iovecs exceeds 1024.
287 */
TEST_F(process_madvise,invalid_vlen)288 TEST_F(process_madvise, invalid_vlen)
289 {
290 const unsigned long pagesize = self->page_size;
291 int pidfd = self->pidfd;
292 struct iovec vec;
293 char *map;
294 ssize_t ret;
295
296 map = mmap(NULL, pagesize, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1,
297 0);
298 if (map == MAP_FAILED)
299 SKIP(return, "mmap failed, not enough memory.\n");
300
301 vec.iov_base = map;
302 vec.iov_len = pagesize;
303
304 ret = sys_process_madvise(pidfd, &vec, 1025, MADV_DONTNEED, 0);
305 ASSERT_EQ(ret, -1);
306 ASSERT_EQ(errno, EINVAL);
307
308 /* Cleanup. */
309 ASSERT_EQ(munmap(map, pagesize), 0);
310 }
311
312 /*
313 * Test process_madvise() with an invalid flag value. Currently, only a flag
314 * value of 0 is supported. This test is reserved for the future, e.g., if
315 * synchronous flags are added.
316 */
TEST_F(process_madvise,flag)317 TEST_F(process_madvise, flag)
318 {
319 const unsigned long pagesize = self->page_size;
320 unsigned int invalid_flag;
321 int pidfd = self->pidfd;
322 struct iovec vec;
323 char *map;
324 ssize_t ret;
325
326 map = mmap(NULL, pagesize, PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1,
327 0);
328 if (map == MAP_FAILED)
329 SKIP(return, "mmap failed, not enough memory.\n");
330
331 vec.iov_base = map;
332 vec.iov_len = pagesize;
333
334 invalid_flag = 0x80000000;
335
336 ret = sys_process_madvise(pidfd, &vec, 1, MADV_DONTNEED, invalid_flag);
337 ASSERT_EQ(ret, -1);
338 ASSERT_EQ(errno, EINVAL);
339
340 /* Cleanup. */
341 ASSERT_EQ(munmap(map, pagesize), 0);
342 }
343
344 TEST_HARNESS_MAIN
345