1 // SPDX-License-Identifier: GPL-2.0
2 #define _GNU_SOURCE
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <limits.h>
6 #include <sched.h>
7 #include <stdio.h>
8 #include <stdlib.h>
9 #include <string.h>
10 #include <linux/nsfs.h>
11 #include <poll.h>
12 #include <sys/mount.h>
13 #include <sys/socket.h>
14 #include <sys/stat.h>
15 #include <sys/types.h>
16 #include <sys/wait.h>
17 #include <sys/syscall.h>
18 #include <unistd.h>
19 #include <pthread.h>
20 #include "../kselftest_harness.h"
21 #include "../filesystems/utils.h"
22 #include "wrappers.h"
23
24 #ifndef PIDFD_THREAD
25 #define PIDFD_THREAD O_EXCL
26 #endif
27
28 #ifndef FD_NSFS_ROOT
29 #define FD_NSFS_ROOT -10003 /* Root of the nsfs filesystem */
30 #endif
31
32 #ifndef FILEID_NSFS
33 #define FILEID_NSFS 0xf1
34 #endif
35
36 /*
37 * Test that initial namespaces can be reopened via file handle.
38 * Initial namespaces should have active ref count of 1 from boot.
39 */
TEST(init_ns_always_active)40 TEST(init_ns_always_active)
41 {
42 struct file_handle *handle;
43 int mount_id;
44 int ret;
45 int fd1, fd2;
46 struct stat st1, st2;
47
48 handle = malloc(sizeof(*handle) + MAX_HANDLE_SZ);
49 ASSERT_NE(handle, NULL);
50
51 /* Open initial network namespace */
52 fd1 = open("/proc/1/ns/net", O_RDONLY);
53 ASSERT_GE(fd1, 0);
54
55 /* Get file handle for initial namespace */
56 handle->handle_bytes = MAX_HANDLE_SZ;
57 ret = name_to_handle_at(fd1, "", handle, &mount_id, AT_EMPTY_PATH);
58 if (ret < 0 && errno == EOPNOTSUPP) {
59 SKIP(free(handle); close(fd1);
60 return, "nsfs doesn't support file handles");
61 }
62 ASSERT_EQ(ret, 0);
63
64 /* Close the namespace fd */
65 close(fd1);
66
67 /* Try to reopen via file handle - should succeed since init ns is always active */
68 fd2 = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
69 if (fd2 < 0 && (errno == EINVAL || errno == EOPNOTSUPP)) {
70 SKIP(free(handle);
71 return, "open_by_handle_at with FD_NSFS_ROOT not supported");
72 }
73 ASSERT_GE(fd2, 0);
74
75 /* Verify we opened the same namespace */
76 fd1 = open("/proc/1/ns/net", O_RDONLY);
77 ASSERT_GE(fd1, 0);
78 ASSERT_EQ(fstat(fd1, &st1), 0);
79 ASSERT_EQ(fstat(fd2, &st2), 0);
80 ASSERT_EQ(st1.st_ino, st2.st_ino);
81
82 close(fd1);
83 close(fd2);
84 free(handle);
85 }
86
87 /*
88 * Test namespace lifecycle: create a namespace in a child process,
89 * get a file handle while it's active, then try to reopen after
90 * the process exits (namespace becomes inactive).
91 */
TEST(ns_inactive_after_exit)92 TEST(ns_inactive_after_exit)
93 {
94 struct file_handle *handle;
95 int mount_id;
96 int ret;
97 int fd;
98 int pipefd[2];
99 pid_t pid;
100 int status;
101 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
102
103 /* Create pipe for passing file handle from child */
104 ASSERT_EQ(pipe(pipefd), 0);
105
106 pid = fork();
107 ASSERT_GE(pid, 0);
108
109 if (pid == 0) {
110 /* Child process */
111 close(pipefd[0]);
112
113 /* Create new network namespace */
114 ret = unshare(CLONE_NEWNET);
115 if (ret < 0) {
116 close(pipefd[1]);
117 exit(1);
118 }
119
120 /* Open our new namespace */
121 fd = open("/proc/self/ns/net", O_RDONLY);
122 if (fd < 0) {
123 close(pipefd[1]);
124 exit(1);
125 }
126
127 /* Get file handle for the namespace */
128 handle = (struct file_handle *)buf;
129 handle->handle_bytes = MAX_HANDLE_SZ;
130 ret = name_to_handle_at(fd, "", handle, &mount_id, AT_EMPTY_PATH);
131 close(fd);
132
133 if (ret < 0) {
134 close(pipefd[1]);
135 exit(1);
136 }
137
138 /* Send handle to parent */
139 write(pipefd[1], buf, sizeof(*handle) + handle->handle_bytes);
140 close(pipefd[1]);
141
142 /* Exit - namespace should become inactive */
143 exit(0);
144 }
145
146 /* Parent process */
147 close(pipefd[1]);
148
149 /* Read file handle from child */
150 ret = read(pipefd[0], buf, sizeof(buf));
151 close(pipefd[0]);
152
153 /* Wait for child to exit */
154 waitpid(pid, &status, 0);
155 ASSERT_TRUE(WIFEXITED(status));
156 ASSERT_EQ(WEXITSTATUS(status), 0);
157
158 ASSERT_GT(ret, 0);
159 handle = (struct file_handle *)buf;
160
161 /* Try to reopen namespace - should fail with ENOENT since it's inactive */
162 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
163 ASSERT_LT(fd, 0);
164 /* Should fail with ENOENT (namespace inactive) or ESTALE */
165 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
166 }
167
168 /*
169 * Test that a namespace remains active while a process is using it,
170 * even after the creating process exits.
171 */
TEST(ns_active_with_multiple_processes)172 TEST(ns_active_with_multiple_processes)
173 {
174 struct file_handle *handle;
175 int mount_id;
176 int ret;
177 int fd;
178 int pipefd[2];
179 int syncpipe[2];
180 pid_t pid1, pid2;
181 int status;
182 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
183 char sync_byte;
184
185 /* Create pipes for communication */
186 ASSERT_EQ(pipe(pipefd), 0);
187 ASSERT_EQ(pipe(syncpipe), 0);
188
189 pid1 = fork();
190 ASSERT_GE(pid1, 0);
191
192 if (pid1 == 0) {
193 /* First child - creates namespace */
194 close(pipefd[0]);
195 close(syncpipe[1]);
196
197 /* Create new network namespace */
198 ret = unshare(CLONE_NEWNET);
199 if (ret < 0) {
200 close(pipefd[1]);
201 close(syncpipe[0]);
202 exit(1);
203 }
204
205 /* Open and get handle */
206 fd = open("/proc/self/ns/net", O_RDONLY);
207 if (fd < 0) {
208 close(pipefd[1]);
209 close(syncpipe[0]);
210 exit(1);
211 }
212
213 handle = (struct file_handle *)buf;
214 handle->handle_bytes = MAX_HANDLE_SZ;
215 ret = name_to_handle_at(fd, "", handle, &mount_id, AT_EMPTY_PATH);
216 close(fd);
217
218 if (ret < 0) {
219 close(pipefd[1]);
220 close(syncpipe[0]);
221 exit(1);
222 }
223
224 /* Send handle to parent */
225 write(pipefd[1], buf, sizeof(*handle) + handle->handle_bytes);
226 close(pipefd[1]);
227
228 /* Wait for signal before exiting */
229 read(syncpipe[0], &sync_byte, 1);
230 close(syncpipe[0]);
231 exit(0);
232 }
233
234 /* Parent reads handle */
235 close(pipefd[1]);
236 ret = read(pipefd[0], buf, sizeof(buf));
237 close(pipefd[0]);
238 ASSERT_GT(ret, 0);
239
240 handle = (struct file_handle *)buf;
241
242 /* Create second child that will keep namespace active */
243 pid2 = fork();
244 ASSERT_GE(pid2, 0);
245
246 if (pid2 == 0) {
247 /* Second child - reopens the namespace */
248 close(syncpipe[0]);
249 close(syncpipe[1]);
250
251 /* Open the namespace via handle */
252 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
253 if (fd < 0) {
254 exit(1);
255 }
256
257 /* Join the namespace */
258 ret = setns(fd, CLONE_NEWNET);
259 close(fd);
260 if (ret < 0) {
261 exit(1);
262 }
263
264 /* Sleep to keep namespace active */
265 sleep(1);
266 exit(0);
267 }
268
269 /* Let second child enter the namespace */
270 usleep(100000); /* 100ms */
271
272 /* Signal first child to exit */
273 close(syncpipe[0]);
274 sync_byte = 'X';
275 write(syncpipe[1], &sync_byte, 1);
276 close(syncpipe[1]);
277
278 /* Wait for first child */
279 waitpid(pid1, &status, 0);
280 ASSERT_TRUE(WIFEXITED(status));
281
282 /* Namespace should still be active because second child is using it */
283 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
284 ASSERT_GE(fd, 0);
285 close(fd);
286
287 /* Wait for second child */
288 waitpid(pid2, &status, 0);
289 ASSERT_TRUE(WIFEXITED(status));
290 }
291
292 /*
293 * Test user namespace active ref tracking via credential lifecycle
294 */
TEST(userns_active_ref_lifecycle)295 TEST(userns_active_ref_lifecycle)
296 {
297 struct file_handle *handle;
298 int mount_id;
299 int ret;
300 int fd;
301 int pipefd[2];
302 pid_t pid;
303 int status;
304 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
305
306 ASSERT_EQ(pipe(pipefd), 0);
307
308 pid = fork();
309 ASSERT_GE(pid, 0);
310
311 if (pid == 0) {
312 /* Child process */
313 close(pipefd[0]);
314
315 /* Create new user namespace */
316 ret = unshare(CLONE_NEWUSER);
317 if (ret < 0) {
318 close(pipefd[1]);
319 exit(1);
320 }
321
322 /* Set up uid/gid mappings */
323 int uid_map_fd = open("/proc/self/uid_map", O_WRONLY);
324 int gid_map_fd = open("/proc/self/gid_map", O_WRONLY);
325 int setgroups_fd = open("/proc/self/setgroups", O_WRONLY);
326
327 if (uid_map_fd >= 0 && gid_map_fd >= 0 && setgroups_fd >= 0) {
328 write(setgroups_fd, "deny", 4);
329 close(setgroups_fd);
330
331 char mapping[64];
332 snprintf(mapping, sizeof(mapping), "0 %d 1", getuid());
333 write(uid_map_fd, mapping, strlen(mapping));
334 close(uid_map_fd);
335
336 snprintf(mapping, sizeof(mapping), "0 %d 1", getgid());
337 write(gid_map_fd, mapping, strlen(mapping));
338 close(gid_map_fd);
339 }
340
341 /* Get file handle */
342 fd = open("/proc/self/ns/user", O_RDONLY);
343 if (fd < 0) {
344 close(pipefd[1]);
345 exit(1);
346 }
347
348 handle = (struct file_handle *)buf;
349 handle->handle_bytes = MAX_HANDLE_SZ;
350 ret = name_to_handle_at(fd, "", handle, &mount_id, AT_EMPTY_PATH);
351 close(fd);
352
353 if (ret < 0) {
354 close(pipefd[1]);
355 exit(1);
356 }
357
358 /* Send handle to parent */
359 write(pipefd[1], buf, sizeof(*handle) + handle->handle_bytes);
360 close(pipefd[1]);
361 exit(0);
362 }
363
364 /* Parent */
365 close(pipefd[1]);
366 ret = read(pipefd[0], buf, sizeof(buf));
367 close(pipefd[0]);
368
369 waitpid(pid, &status, 0);
370 ASSERT_TRUE(WIFEXITED(status));
371 ASSERT_EQ(WEXITSTATUS(status), 0);
372
373 ASSERT_GT(ret, 0);
374 handle = (struct file_handle *)buf;
375
376 /* Namespace should be inactive after all tasks exit */
377 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
378 ASSERT_LT(fd, 0);
379 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
380 }
381
382 /*
383 * Test PID namespace active ref tracking
384 */
TEST(pidns_active_ref_lifecycle)385 TEST(pidns_active_ref_lifecycle)
386 {
387 struct file_handle *handle;
388 int mount_id;
389 int ret;
390 int fd;
391 int pipefd[2];
392 pid_t pid;
393 int status;
394 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
395
396 ASSERT_EQ(pipe(pipefd), 0);
397
398 pid = fork();
399 ASSERT_GE(pid, 0);
400
401 if (pid == 0) {
402 /* Child process */
403 close(pipefd[0]);
404
405 /* Create new PID namespace */
406 ret = unshare(CLONE_NEWPID);
407 if (ret < 0) {
408 close(pipefd[1]);
409 exit(1);
410 }
411
412 /* Fork to actually enter the PID namespace */
413 pid_t child = fork();
414 if (child < 0) {
415 close(pipefd[1]);
416 exit(1);
417 }
418
419 if (child == 0) {
420 /* Grandchild - in new PID namespace */
421 fd = open("/proc/self/ns/pid", O_RDONLY);
422 if (fd < 0) {
423 exit(1);
424 }
425
426 handle = (struct file_handle *)buf;
427 handle->handle_bytes = MAX_HANDLE_SZ;
428 ret = name_to_handle_at(fd, "", handle, &mount_id, AT_EMPTY_PATH);
429 close(fd);
430
431 if (ret < 0) {
432 exit(1);
433 }
434
435 /* Send handle to grandparent */
436 write(pipefd[1], buf, sizeof(*handle) + handle->handle_bytes);
437 close(pipefd[1]);
438 exit(0);
439 }
440
441 /* Wait for grandchild */
442 waitpid(child, NULL, 0);
443 exit(0);
444 }
445
446 /* Parent */
447 close(pipefd[1]);
448 ret = read(pipefd[0], buf, sizeof(buf));
449 close(pipefd[0]);
450
451 waitpid(pid, &status, 0);
452 ASSERT_TRUE(WIFEXITED(status));
453 ASSERT_EQ(WEXITSTATUS(status), 0);
454
455 ASSERT_GT(ret, 0);
456 handle = (struct file_handle *)buf;
457
458 /* Namespace should be inactive after all processes exit */
459 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
460 ASSERT_LT(fd, 0);
461 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
462 }
463
464 /*
465 * Test that an open file descriptor keeps a namespace active.
466 * Even after the creating process exits, the namespace should remain
467 * active as long as an fd is held open.
468 */
TEST(ns_fd_keeps_active)469 TEST(ns_fd_keeps_active)
470 {
471 struct file_handle *handle;
472 int mount_id;
473 int ret;
474 int nsfd;
475 int pipe_child_ready[2];
476 int pipe_parent_ready[2];
477 pid_t pid;
478 int status;
479 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
480 char sync_byte;
481 char proc_path[64];
482
483 ASSERT_EQ(pipe(pipe_child_ready), 0);
484 ASSERT_EQ(pipe(pipe_parent_ready), 0);
485
486 pid = fork();
487 ASSERT_GE(pid, 0);
488
489 if (pid == 0) {
490 /* Child process */
491 close(pipe_child_ready[0]);
492 close(pipe_parent_ready[1]);
493
494 TH_LOG("Child: creating new network namespace");
495
496 /* Create new network namespace */
497 ret = unshare(CLONE_NEWNET);
498 if (ret < 0) {
499 TH_LOG("Child: unshare(CLONE_NEWNET) failed: %s", strerror(errno));
500 close(pipe_child_ready[1]);
501 close(pipe_parent_ready[0]);
502 exit(1);
503 }
504
505 TH_LOG("Child: network namespace created successfully");
506
507 /* Get file handle for the namespace */
508 nsfd = open("/proc/self/ns/net", O_RDONLY);
509 if (nsfd < 0) {
510 TH_LOG("Child: failed to open /proc/self/ns/net: %s", strerror(errno));
511 close(pipe_child_ready[1]);
512 close(pipe_parent_ready[0]);
513 exit(1);
514 }
515
516 TH_LOG("Child: opened namespace fd %d", nsfd);
517
518 handle = (struct file_handle *)buf;
519 handle->handle_bytes = MAX_HANDLE_SZ;
520 ret = name_to_handle_at(nsfd, "", handle, &mount_id, AT_EMPTY_PATH);
521 close(nsfd);
522
523 if (ret < 0) {
524 TH_LOG("Child: name_to_handle_at failed: %s", strerror(errno));
525 close(pipe_child_ready[1]);
526 close(pipe_parent_ready[0]);
527 exit(1);
528 }
529
530 TH_LOG("Child: got file handle (bytes=%u)", handle->handle_bytes);
531
532 /* Send file handle to parent */
533 ret = write(pipe_child_ready[1], buf, sizeof(*handle) + handle->handle_bytes);
534 TH_LOG("Child: sent %d bytes of file handle to parent", ret);
535 close(pipe_child_ready[1]);
536
537 /* Wait for parent to open the fd */
538 TH_LOG("Child: waiting for parent to open fd");
539 ret = read(pipe_parent_ready[0], &sync_byte, 1);
540 close(pipe_parent_ready[0]);
541
542 TH_LOG("Child: parent signaled (read %d bytes), exiting now", ret);
543 /* Exit - namespace should stay active because parent holds fd */
544 exit(0);
545 }
546
547 /* Parent process */
548 close(pipe_child_ready[1]);
549 close(pipe_parent_ready[0]);
550
551 TH_LOG("Parent: reading file handle from child");
552
553 /* Read file handle from child */
554 ret = read(pipe_child_ready[0], buf, sizeof(buf));
555 close(pipe_child_ready[0]);
556 ASSERT_GT(ret, 0);
557 handle = (struct file_handle *)buf;
558
559 TH_LOG("Parent: received %d bytes, handle size=%u", ret, handle->handle_bytes);
560
561 /* Open the child's namespace while it's still alive */
562 snprintf(proc_path, sizeof(proc_path), "/proc/%d/ns/net", pid);
563 TH_LOG("Parent: opening child's namespace at %s", proc_path);
564 nsfd = open(proc_path, O_RDONLY);
565 if (nsfd < 0) {
566 TH_LOG("Parent: failed to open %s: %s", proc_path, strerror(errno));
567 close(pipe_parent_ready[1]);
568 kill(pid, SIGKILL);
569 waitpid(pid, NULL, 0);
570 SKIP(return, "Failed to open child's namespace");
571 }
572
573 TH_LOG("Parent: opened child's namespace, got fd %d", nsfd);
574
575 /* Signal child that we have the fd */
576 sync_byte = 'G';
577 write(pipe_parent_ready[1], &sync_byte, 1);
578 close(pipe_parent_ready[1]);
579 TH_LOG("Parent: signaled child that we have the fd");
580
581 /* Wait for child to exit */
582 waitpid(pid, &status, 0);
583 ASSERT_TRUE(WIFEXITED(status));
584 ASSERT_EQ(WEXITSTATUS(status), 0);
585
586 TH_LOG("Child exited, parent holds fd %d to namespace", nsfd);
587
588 /*
589 * Namespace should still be ACTIVE because we hold an fd.
590 * We should be able to reopen it via file handle.
591 */
592 TH_LOG("Attempting to reopen namespace via file handle (should succeed - fd held)");
593 int fd2 = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
594 ASSERT_GE(fd2, 0);
595
596 TH_LOG("Successfully reopened namespace via file handle, got fd %d", fd2);
597
598 /* Verify it's the same namespace */
599 struct stat st1, st2;
600 ASSERT_EQ(fstat(nsfd, &st1), 0);
601 ASSERT_EQ(fstat(fd2, &st2), 0);
602 TH_LOG("Namespace inodes: nsfd=%lu, fd2=%lu", st1.st_ino, st2.st_ino);
603 ASSERT_EQ(st1.st_ino, st2.st_ino);
604 close(fd2);
605
606 /* Now close the fd - namespace should become inactive */
607 TH_LOG("Closing fd %d - namespace should become inactive", nsfd);
608 close(nsfd);
609
610 /* Now reopening should fail - namespace is inactive */
611 TH_LOG("Attempting to reopen namespace via file handle (should fail - inactive)");
612 fd2 = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
613 ASSERT_LT(fd2, 0);
614 /* Should fail with ENOENT (inactive) or ESTALE (gone) */
615 TH_LOG("Reopen failed as expected: %s (errno=%d)", strerror(errno), errno);
616 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
617 }
618
619 /*
620 * Test hierarchical active reference propagation.
621 * When a child namespace is active, its owning user namespace should also
622 * be active automatically due to hierarchical active reference propagation.
623 * This ensures parents are always reachable when children are active.
624 */
TEST(ns_parent_always_reachable)625 TEST(ns_parent_always_reachable)
626 {
627 struct file_handle *parent_handle, *child_handle;
628 int ret;
629 int child_nsfd;
630 int pipefd[2];
631 pid_t pid;
632 int status;
633 __u64 parent_id, child_id;
634 char parent_buf[sizeof(*parent_handle) + MAX_HANDLE_SZ];
635 char child_buf[sizeof(*child_handle) + MAX_HANDLE_SZ];
636
637 ASSERT_EQ(pipe(pipefd), 0);
638
639 pid = fork();
640 ASSERT_GE(pid, 0);
641
642 if (pid == 0) {
643 /* Child process */
644 close(pipefd[0]);
645
646 TH_LOG("Child: creating parent user namespace and setting up mappings");
647
648 /* Create parent user namespace with mappings */
649 ret = setup_userns();
650 if (ret < 0) {
651 TH_LOG("Child: setup_userns() for parent failed: %s", strerror(errno));
652 close(pipefd[1]);
653 exit(1);
654 }
655
656 TH_LOG("Child: parent user namespace created, now uid=%d gid=%d", getuid(), getgid());
657
658 /* Get namespace ID for parent user namespace */
659 int parent_fd = open("/proc/self/ns/user", O_RDONLY);
660 if (parent_fd < 0) {
661 TH_LOG("Child: failed to open parent /proc/self/ns/user: %s", strerror(errno));
662 close(pipefd[1]);
663 exit(1);
664 }
665
666 TH_LOG("Child: opened parent userns fd %d", parent_fd);
667
668 if (ioctl(parent_fd, NS_GET_ID, &parent_id) < 0) {
669 TH_LOG("Child: NS_GET_ID for parent failed: %s", strerror(errno));
670 close(parent_fd);
671 close(pipefd[1]);
672 exit(1);
673 }
674 close(parent_fd);
675
676 TH_LOG("Child: got parent namespace ID %llu", (unsigned long long)parent_id);
677
678 /* Create child user namespace within parent */
679 TH_LOG("Child: creating nested child user namespace");
680 ret = setup_userns();
681 if (ret < 0) {
682 TH_LOG("Child: setup_userns() for child failed: %s", strerror(errno));
683 close(pipefd[1]);
684 exit(1);
685 }
686
687 TH_LOG("Child: nested child user namespace created, uid=%d gid=%d", getuid(), getgid());
688
689 /* Get namespace ID for child user namespace */
690 int child_fd = open("/proc/self/ns/user", O_RDONLY);
691 if (child_fd < 0) {
692 TH_LOG("Child: failed to open child /proc/self/ns/user: %s", strerror(errno));
693 close(pipefd[1]);
694 exit(1);
695 }
696
697 TH_LOG("Child: opened child userns fd %d", child_fd);
698
699 if (ioctl(child_fd, NS_GET_ID, &child_id) < 0) {
700 TH_LOG("Child: NS_GET_ID for child failed: %s", strerror(errno));
701 close(child_fd);
702 close(pipefd[1]);
703 exit(1);
704 }
705 close(child_fd);
706
707 TH_LOG("Child: got child namespace ID %llu", (unsigned long long)child_id);
708
709 /* Send both namespace IDs to parent */
710 TH_LOG("Child: sending both namespace IDs to parent");
711 write(pipefd[1], &parent_id, sizeof(parent_id));
712 write(pipefd[1], &child_id, sizeof(child_id));
713 close(pipefd[1]);
714
715 TH_LOG("Child: exiting - parent userns should become inactive");
716 /* Exit - parent user namespace should become inactive */
717 exit(0);
718 }
719
720 /* Parent process */
721 close(pipefd[1]);
722
723 TH_LOG("Parent: reading both namespace IDs from child");
724
725 /* Read both namespace IDs - fixed size, no parsing needed */
726 ret = read(pipefd[0], &parent_id, sizeof(parent_id));
727 if (ret != sizeof(parent_id)) {
728 close(pipefd[0]);
729 waitpid(pid, NULL, 0);
730 SKIP(return, "Failed to read parent namespace ID from child");
731 }
732
733 ret = read(pipefd[0], &child_id, sizeof(child_id));
734 close(pipefd[0]);
735 if (ret != sizeof(child_id)) {
736 waitpid(pid, NULL, 0);
737 SKIP(return, "Failed to read child namespace ID from child");
738 }
739
740 TH_LOG("Parent: received parent_id=%llu, child_id=%llu",
741 (unsigned long long)parent_id, (unsigned long long)child_id);
742
743 /* Construct file handles from namespace IDs */
744 parent_handle = (struct file_handle *)parent_buf;
745 parent_handle->handle_bytes = sizeof(struct nsfs_file_handle);
746 parent_handle->handle_type = FILEID_NSFS;
747 struct nsfs_file_handle *parent_fh = (struct nsfs_file_handle *)parent_handle->f_handle;
748 parent_fh->ns_id = parent_id;
749 parent_fh->ns_type = 0;
750 parent_fh->ns_inum = 0;
751
752 child_handle = (struct file_handle *)child_buf;
753 child_handle->handle_bytes = sizeof(struct nsfs_file_handle);
754 child_handle->handle_type = FILEID_NSFS;
755 struct nsfs_file_handle *child_fh = (struct nsfs_file_handle *)child_handle->f_handle;
756 child_fh->ns_id = child_id;
757 child_fh->ns_type = 0;
758 child_fh->ns_inum = 0;
759
760 TH_LOG("Parent: opening child namespace BEFORE child exits");
761
762 /* Open child namespace while child is still alive to keep it active */
763 child_nsfd = open_by_handle_at(FD_NSFS_ROOT, child_handle, O_RDONLY);
764 if (child_nsfd < 0) {
765 TH_LOG("Failed to open child namespace: %s (errno=%d)", strerror(errno), errno);
766 waitpid(pid, NULL, 0);
767 SKIP(return, "Failed to open child namespace");
768 }
769
770 TH_LOG("Opened child namespace fd %d", child_nsfd);
771
772 /* Now wait for child to exit */
773 TH_LOG("Parent: waiting for child to exit");
774 waitpid(pid, &status, 0);
775 ASSERT_TRUE(WIFEXITED(status));
776 ASSERT_EQ(WEXITSTATUS(status), 0);
777
778 TH_LOG("Child process exited, parent holds fd to child namespace");
779
780 /*
781 * With hierarchical active reference propagation:
782 * Since the child namespace is active (parent process holds fd),
783 * the parent user namespace should ALSO be active automatically.
784 * This is because when we took an active reference on the child,
785 * it propagated up to the owning user namespace.
786 */
787 TH_LOG("Attempting to reopen parent namespace (should SUCCEED - hierarchical propagation)");
788 int parent_fd = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
789 ASSERT_GE(parent_fd, 0);
790
791 TH_LOG("SUCCESS: Parent namespace is active (fd=%d) due to active child", parent_fd);
792
793 /* Verify we can also get parent via NS_GET_USERNS */
794 TH_LOG("Verifying NS_GET_USERNS also works");
795 int parent_fd2 = ioctl(child_nsfd, NS_GET_USERNS);
796 if (parent_fd2 < 0) {
797 close(parent_fd);
798 close(child_nsfd);
799 TH_LOG("NS_GET_USERNS failed: %s (errno=%d)", strerror(errno), errno);
800 SKIP(return, "NS_GET_USERNS not supported or failed");
801 }
802
803 TH_LOG("NS_GET_USERNS succeeded, got parent fd %d", parent_fd2);
804
805 /* Verify both methods give us the same namespace */
806 struct stat st1, st2;
807 ASSERT_EQ(fstat(parent_fd, &st1), 0);
808 ASSERT_EQ(fstat(parent_fd2, &st2), 0);
809 TH_LOG("Parent namespace inodes: parent_fd=%lu, parent_fd2=%lu", st1.st_ino, st2.st_ino);
810 ASSERT_EQ(st1.st_ino, st2.st_ino);
811
812 /*
813 * Close child fd - parent should remain active because we still
814 * hold direct references to it (parent_fd and parent_fd2).
815 */
816 TH_LOG("Closing child fd - parent should remain active (direct refs held)");
817 close(child_nsfd);
818
819 /* Parent should still be openable */
820 TH_LOG("Verifying parent still active via file handle");
821 int parent_fd3 = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
822 ASSERT_GE(parent_fd3, 0);
823 close(parent_fd3);
824
825 TH_LOG("Closing all fds to parent namespace");
826 close(parent_fd);
827 close(parent_fd2);
828
829 /* Both should now be inactive */
830 TH_LOG("Attempting to reopen parent (should fail - inactive, no refs)");
831 parent_fd = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
832 ASSERT_LT(parent_fd, 0);
833 TH_LOG("Parent inactive as expected: %s (errno=%d)", strerror(errno), errno);
834 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
835 }
836
837 /*
838 * Test that bind mounts keep namespaces in the tree even when inactive
839 */
TEST(ns_bind_mount_keeps_in_tree)840 TEST(ns_bind_mount_keeps_in_tree)
841 {
842 struct file_handle *handle;
843 int mount_id;
844 int ret;
845 int fd;
846 int pipefd[2];
847 pid_t pid;
848 int status;
849 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
850 char tmpfile[] = "/tmp/ns-test-XXXXXX";
851 int tmpfd;
852
853 /* Create temporary file for bind mount */
854 tmpfd = mkstemp(tmpfile);
855 if (tmpfd < 0) {
856 SKIP(return, "Cannot create temporary file");
857 }
858 close(tmpfd);
859
860 ASSERT_EQ(pipe(pipefd), 0);
861
862 pid = fork();
863 ASSERT_GE(pid, 0);
864
865 if (pid == 0) {
866 /* Child process */
867 close(pipefd[0]);
868
869 /* Unshare mount namespace and make mounts private to avoid propagation */
870 ret = unshare(CLONE_NEWNS);
871 if (ret < 0) {
872 close(pipefd[1]);
873 unlink(tmpfile);
874 exit(1);
875 }
876 ret = mount(NULL, "/", NULL, MS_PRIVATE | MS_REC, NULL);
877 if (ret < 0) {
878 close(pipefd[1]);
879 unlink(tmpfile);
880 exit(1);
881 }
882
883 /* Create new network namespace */
884 ret = unshare(CLONE_NEWNET);
885 if (ret < 0) {
886 close(pipefd[1]);
887 unlink(tmpfile);
888 exit(1);
889 }
890
891 /* Bind mount the namespace */
892 ret = mount("/proc/self/ns/net", tmpfile, NULL, MS_BIND, NULL);
893 if (ret < 0) {
894 close(pipefd[1]);
895 unlink(tmpfile);
896 exit(1);
897 }
898
899 /* Get file handle */
900 fd = open("/proc/self/ns/net", O_RDONLY);
901 if (fd < 0) {
902 umount(tmpfile);
903 close(pipefd[1]);
904 unlink(tmpfile);
905 exit(1);
906 }
907
908 handle = (struct file_handle *)buf;
909 handle->handle_bytes = MAX_HANDLE_SZ;
910 ret = name_to_handle_at(fd, "", handle, &mount_id, AT_EMPTY_PATH);
911 close(fd);
912
913 if (ret < 0) {
914 umount(tmpfile);
915 close(pipefd[1]);
916 unlink(tmpfile);
917 exit(1);
918 }
919
920 /* Send handle to parent */
921 write(pipefd[1], buf, sizeof(*handle) + handle->handle_bytes);
922 close(pipefd[1]);
923 exit(0);
924 }
925
926 /* Parent */
927 close(pipefd[1]);
928 ret = read(pipefd[0], buf, sizeof(buf));
929 close(pipefd[0]);
930
931 waitpid(pid, &status, 0);
932 ASSERT_TRUE(WIFEXITED(status));
933 ASSERT_EQ(WEXITSTATUS(status), 0);
934
935 ASSERT_GT(ret, 0);
936 handle = (struct file_handle *)buf;
937
938 /*
939 * Namespace should be inactive but still in tree due to bind mount.
940 * Reopening should fail with ENOENT (inactive) not ESTALE (not in tree).
941 */
942 fd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
943 ASSERT_LT(fd, 0);
944 /* Should be ENOENT (inactive) since bind mount keeps it in tree */
945 if (errno != ENOENT && errno != ESTALE) {
946 TH_LOG("Unexpected error: %d", errno);
947 }
948
949 /* Cleanup */
950 umount(tmpfile);
951 unlink(tmpfile);
952 }
953
954 /*
955 * Test multi-level hierarchy (3+ levels deep).
956 * Grandparent → Parent → Child
957 * When child is active, both parent AND grandparent should be active.
958 */
TEST(ns_multilevel_hierarchy)959 TEST(ns_multilevel_hierarchy)
960 {
961 struct file_handle *gp_handle, *p_handle, *c_handle;
962 int ret, pipefd[2];
963 pid_t pid;
964 int status;
965 __u64 gp_id, p_id, c_id;
966 char gp_buf[sizeof(*gp_handle) + MAX_HANDLE_SZ];
967 char p_buf[sizeof(*p_handle) + MAX_HANDLE_SZ];
968 char c_buf[sizeof(*c_handle) + MAX_HANDLE_SZ];
969
970 ASSERT_EQ(pipe(pipefd), 0);
971 pid = fork();
972 ASSERT_GE(pid, 0);
973
974 if (pid == 0) {
975 close(pipefd[0]);
976
977 /* Create grandparent user namespace */
978 if (setup_userns() < 0) {
979 close(pipefd[1]);
980 exit(1);
981 }
982
983 int gp_fd = open("/proc/self/ns/user", O_RDONLY);
984 if (gp_fd < 0) {
985 close(pipefd[1]);
986 exit(1);
987 }
988 if (ioctl(gp_fd, NS_GET_ID, &gp_id) < 0) {
989 close(gp_fd);
990 close(pipefd[1]);
991 exit(1);
992 }
993 close(gp_fd);
994
995 /* Create parent user namespace */
996 if (setup_userns() < 0) {
997 close(pipefd[1]);
998 exit(1);
999 }
1000
1001 int p_fd = open("/proc/self/ns/user", O_RDONLY);
1002 if (p_fd < 0) {
1003 close(pipefd[1]);
1004 exit(1);
1005 }
1006 if (ioctl(p_fd, NS_GET_ID, &p_id) < 0) {
1007 close(p_fd);
1008 close(pipefd[1]);
1009 exit(1);
1010 }
1011 close(p_fd);
1012
1013 /* Create child user namespace */
1014 if (setup_userns() < 0) {
1015 close(pipefd[1]);
1016 exit(1);
1017 }
1018
1019 int c_fd = open("/proc/self/ns/user", O_RDONLY);
1020 if (c_fd < 0) {
1021 close(pipefd[1]);
1022 exit(1);
1023 }
1024 if (ioctl(c_fd, NS_GET_ID, &c_id) < 0) {
1025 close(c_fd);
1026 close(pipefd[1]);
1027 exit(1);
1028 }
1029 close(c_fd);
1030
1031 /* Send all three namespace IDs */
1032 write(pipefd[1], &gp_id, sizeof(gp_id));
1033 write(pipefd[1], &p_id, sizeof(p_id));
1034 write(pipefd[1], &c_id, sizeof(c_id));
1035 close(pipefd[1]);
1036 exit(0);
1037 }
1038
1039 close(pipefd[1]);
1040
1041 /* Read all three namespace IDs - fixed size, no parsing needed */
1042 ret = read(pipefd[0], &gp_id, sizeof(gp_id));
1043 if (ret != sizeof(gp_id)) {
1044 close(pipefd[0]);
1045 waitpid(pid, NULL, 0);
1046 SKIP(return, "Failed to read grandparent namespace ID from child");
1047 }
1048
1049 ret = read(pipefd[0], &p_id, sizeof(p_id));
1050 if (ret != sizeof(p_id)) {
1051 close(pipefd[0]);
1052 waitpid(pid, NULL, 0);
1053 SKIP(return, "Failed to read parent namespace ID from child");
1054 }
1055
1056 ret = read(pipefd[0], &c_id, sizeof(c_id));
1057 close(pipefd[0]);
1058 if (ret != sizeof(c_id)) {
1059 waitpid(pid, NULL, 0);
1060 SKIP(return, "Failed to read child namespace ID from child");
1061 }
1062
1063 /* Construct file handles from namespace IDs */
1064 gp_handle = (struct file_handle *)gp_buf;
1065 gp_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1066 gp_handle->handle_type = FILEID_NSFS;
1067 struct nsfs_file_handle *gp_fh = (struct nsfs_file_handle *)gp_handle->f_handle;
1068 gp_fh->ns_id = gp_id;
1069 gp_fh->ns_type = 0;
1070 gp_fh->ns_inum = 0;
1071
1072 p_handle = (struct file_handle *)p_buf;
1073 p_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1074 p_handle->handle_type = FILEID_NSFS;
1075 struct nsfs_file_handle *p_fh = (struct nsfs_file_handle *)p_handle->f_handle;
1076 p_fh->ns_id = p_id;
1077 p_fh->ns_type = 0;
1078 p_fh->ns_inum = 0;
1079
1080 c_handle = (struct file_handle *)c_buf;
1081 c_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1082 c_handle->handle_type = FILEID_NSFS;
1083 struct nsfs_file_handle *c_fh = (struct nsfs_file_handle *)c_handle->f_handle;
1084 c_fh->ns_id = c_id;
1085 c_fh->ns_type = 0;
1086 c_fh->ns_inum = 0;
1087
1088 /* Open child before process exits */
1089 int c_fd = open_by_handle_at(FD_NSFS_ROOT, c_handle, O_RDONLY);
1090 if (c_fd < 0) {
1091 waitpid(pid, NULL, 0);
1092 SKIP(return, "Failed to open child namespace");
1093 }
1094
1095 waitpid(pid, &status, 0);
1096 ASSERT_TRUE(WIFEXITED(status));
1097 ASSERT_EQ(WEXITSTATUS(status), 0);
1098
1099 /*
1100 * With 3-level hierarchy and child active:
1101 * - Child is active (we hold fd)
1102 * - Parent should be active (propagated from child)
1103 * - Grandparent should be active (propagated from parent)
1104 */
1105 TH_LOG("Testing parent active when child is active");
1106 int p_fd = open_by_handle_at(FD_NSFS_ROOT, p_handle, O_RDONLY);
1107 ASSERT_GE(p_fd, 0);
1108
1109 TH_LOG("Testing grandparent active when child is active");
1110 int gp_fd = open_by_handle_at(FD_NSFS_ROOT, gp_handle, O_RDONLY);
1111 ASSERT_GE(gp_fd, 0);
1112
1113 close(c_fd);
1114 close(p_fd);
1115 close(gp_fd);
1116 }
1117
1118 /*
1119 * Test multiple children sharing same parent.
1120 * Parent should stay active as long as ANY child is active.
1121 */
TEST(ns_multiple_children_same_parent)1122 TEST(ns_multiple_children_same_parent)
1123 {
1124 struct file_handle *p_handle, *c1_handle, *c2_handle;
1125 int ret, pipefd[2];
1126 pid_t pid;
1127 int status;
1128 __u64 p_id, c1_id, c2_id;
1129 char p_buf[sizeof(*p_handle) + MAX_HANDLE_SZ];
1130 char c1_buf[sizeof(*c1_handle) + MAX_HANDLE_SZ];
1131 char c2_buf[sizeof(*c2_handle) + MAX_HANDLE_SZ];
1132
1133 ASSERT_EQ(pipe(pipefd), 0);
1134 pid = fork();
1135 ASSERT_GE(pid, 0);
1136
1137 if (pid == 0) {
1138 close(pipefd[0]);
1139
1140 /* Create parent user namespace */
1141 if (setup_userns() < 0) {
1142 close(pipefd[1]);
1143 exit(1);
1144 }
1145
1146 int p_fd = open("/proc/self/ns/user", O_RDONLY);
1147 if (p_fd < 0) {
1148 close(pipefd[1]);
1149 exit(1);
1150 }
1151 if (ioctl(p_fd, NS_GET_ID, &p_id) < 0) {
1152 close(p_fd);
1153 close(pipefd[1]);
1154 exit(1);
1155 }
1156 close(p_fd);
1157
1158 /* Create first child user namespace */
1159 if (setup_userns() < 0) {
1160 close(pipefd[1]);
1161 exit(1);
1162 }
1163
1164 int c1_fd = open("/proc/self/ns/user", O_RDONLY);
1165 if (c1_fd < 0) {
1166 close(pipefd[1]);
1167 exit(1);
1168 }
1169 if (ioctl(c1_fd, NS_GET_ID, &c1_id) < 0) {
1170 close(c1_fd);
1171 close(pipefd[1]);
1172 exit(1);
1173 }
1174 close(c1_fd);
1175
1176 /* Return to parent user namespace and create second child */
1177 /* We can't actually do this easily, so let's create a sibling namespace
1178 * by creating a network namespace instead */
1179 if (unshare(CLONE_NEWNET) < 0) {
1180 close(pipefd[1]);
1181 exit(1);
1182 }
1183
1184 int c2_fd = open("/proc/self/ns/net", O_RDONLY);
1185 if (c2_fd < 0) {
1186 close(pipefd[1]);
1187 exit(1);
1188 }
1189 if (ioctl(c2_fd, NS_GET_ID, &c2_id) < 0) {
1190 close(c2_fd);
1191 close(pipefd[1]);
1192 exit(1);
1193 }
1194 close(c2_fd);
1195
1196 /* Send all namespace IDs */
1197 write(pipefd[1], &p_id, sizeof(p_id));
1198 write(pipefd[1], &c1_id, sizeof(c1_id));
1199 write(pipefd[1], &c2_id, sizeof(c2_id));
1200 close(pipefd[1]);
1201 exit(0);
1202 }
1203
1204 close(pipefd[1]);
1205
1206 /* Read all three namespace IDs - fixed size, no parsing needed */
1207 ret = read(pipefd[0], &p_id, sizeof(p_id));
1208 if (ret != sizeof(p_id)) {
1209 close(pipefd[0]);
1210 waitpid(pid, NULL, 0);
1211 SKIP(return, "Failed to read parent namespace ID");
1212 }
1213
1214 ret = read(pipefd[0], &c1_id, sizeof(c1_id));
1215 if (ret != sizeof(c1_id)) {
1216 close(pipefd[0]);
1217 waitpid(pid, NULL, 0);
1218 SKIP(return, "Failed to read first child namespace ID");
1219 }
1220
1221 ret = read(pipefd[0], &c2_id, sizeof(c2_id));
1222 close(pipefd[0]);
1223 if (ret != sizeof(c2_id)) {
1224 waitpid(pid, NULL, 0);
1225 SKIP(return, "Failed to read second child namespace ID");
1226 }
1227
1228 /* Construct file handles from namespace IDs */
1229 p_handle = (struct file_handle *)p_buf;
1230 p_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1231 p_handle->handle_type = FILEID_NSFS;
1232 struct nsfs_file_handle *p_fh = (struct nsfs_file_handle *)p_handle->f_handle;
1233 p_fh->ns_id = p_id;
1234 p_fh->ns_type = 0;
1235 p_fh->ns_inum = 0;
1236
1237 c1_handle = (struct file_handle *)c1_buf;
1238 c1_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1239 c1_handle->handle_type = FILEID_NSFS;
1240 struct nsfs_file_handle *c1_fh = (struct nsfs_file_handle *)c1_handle->f_handle;
1241 c1_fh->ns_id = c1_id;
1242 c1_fh->ns_type = 0;
1243 c1_fh->ns_inum = 0;
1244
1245 c2_handle = (struct file_handle *)c2_buf;
1246 c2_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1247 c2_handle->handle_type = FILEID_NSFS;
1248 struct nsfs_file_handle *c2_fh = (struct nsfs_file_handle *)c2_handle->f_handle;
1249 c2_fh->ns_id = c2_id;
1250 c2_fh->ns_type = 0;
1251 c2_fh->ns_inum = 0;
1252
1253 /* Open both children before process exits */
1254 int c1_fd = open_by_handle_at(FD_NSFS_ROOT, c1_handle, O_RDONLY);
1255 int c2_fd = open_by_handle_at(FD_NSFS_ROOT, c2_handle, O_RDONLY);
1256
1257 if (c1_fd < 0 || c2_fd < 0) {
1258 if (c1_fd >= 0) close(c1_fd);
1259 if (c2_fd >= 0) close(c2_fd);
1260 waitpid(pid, NULL, 0);
1261 SKIP(return, "Failed to open child namespaces");
1262 }
1263
1264 waitpid(pid, &status, 0);
1265 ASSERT_TRUE(WIFEXITED(status));
1266 ASSERT_EQ(WEXITSTATUS(status), 0);
1267
1268 /* Parent should be active (both children active) */
1269 TH_LOG("Both children active - parent should be active");
1270 int p_fd = open_by_handle_at(FD_NSFS_ROOT, p_handle, O_RDONLY);
1271 ASSERT_GE(p_fd, 0);
1272 close(p_fd);
1273
1274 /* Close first child - parent should STILL be active */
1275 TH_LOG("Closing first child - parent should still be active");
1276 close(c1_fd);
1277 p_fd = open_by_handle_at(FD_NSFS_ROOT, p_handle, O_RDONLY);
1278 ASSERT_GE(p_fd, 0);
1279 close(p_fd);
1280
1281 /* Close second child - NOW parent should become inactive */
1282 TH_LOG("Closing second child - parent should become inactive");
1283 close(c2_fd);
1284 p_fd = open_by_handle_at(FD_NSFS_ROOT, p_handle, O_RDONLY);
1285 ASSERT_LT(p_fd, 0);
1286 }
1287
1288 /*
1289 * Test that different namespace types with same owner all contribute
1290 * active references to the owning user namespace.
1291 */
TEST(ns_different_types_same_owner)1292 TEST(ns_different_types_same_owner)
1293 {
1294 struct file_handle *u_handle, *n_handle, *ut_handle;
1295 int ret, pipefd[2];
1296 pid_t pid;
1297 int status;
1298 __u64 u_id, n_id, ut_id;
1299 char u_buf[sizeof(*u_handle) + MAX_HANDLE_SZ];
1300 char n_buf[sizeof(*n_handle) + MAX_HANDLE_SZ];
1301 char ut_buf[sizeof(*ut_handle) + MAX_HANDLE_SZ];
1302
1303 ASSERT_EQ(pipe(pipefd), 0);
1304 pid = fork();
1305 ASSERT_GE(pid, 0);
1306
1307 if (pid == 0) {
1308 close(pipefd[0]);
1309
1310 /* Create user namespace */
1311 if (setup_userns() < 0) {
1312 close(pipefd[1]);
1313 exit(1);
1314 }
1315
1316 int u_fd = open("/proc/self/ns/user", O_RDONLY);
1317 if (u_fd < 0) {
1318 close(pipefd[1]);
1319 exit(1);
1320 }
1321 if (ioctl(u_fd, NS_GET_ID, &u_id) < 0) {
1322 close(u_fd);
1323 close(pipefd[1]);
1324 exit(1);
1325 }
1326 close(u_fd);
1327
1328 /* Create network namespace (owned by user namespace) */
1329 if (unshare(CLONE_NEWNET) < 0) {
1330 close(pipefd[1]);
1331 exit(1);
1332 }
1333
1334 int n_fd = open("/proc/self/ns/net", O_RDONLY);
1335 if (n_fd < 0) {
1336 close(pipefd[1]);
1337 exit(1);
1338 }
1339 if (ioctl(n_fd, NS_GET_ID, &n_id) < 0) {
1340 close(n_fd);
1341 close(pipefd[1]);
1342 exit(1);
1343 }
1344 close(n_fd);
1345
1346 /* Create UTS namespace (also owned by user namespace) */
1347 if (unshare(CLONE_NEWUTS) < 0) {
1348 close(pipefd[1]);
1349 exit(1);
1350 }
1351
1352 int ut_fd = open("/proc/self/ns/uts", O_RDONLY);
1353 if (ut_fd < 0) {
1354 close(pipefd[1]);
1355 exit(1);
1356 }
1357 if (ioctl(ut_fd, NS_GET_ID, &ut_id) < 0) {
1358 close(ut_fd);
1359 close(pipefd[1]);
1360 exit(1);
1361 }
1362 close(ut_fd);
1363
1364 /* Send all namespace IDs */
1365 write(pipefd[1], &u_id, sizeof(u_id));
1366 write(pipefd[1], &n_id, sizeof(n_id));
1367 write(pipefd[1], &ut_id, sizeof(ut_id));
1368 close(pipefd[1]);
1369 exit(0);
1370 }
1371
1372 close(pipefd[1]);
1373
1374 /* Read all three namespace IDs - fixed size, no parsing needed */
1375 ret = read(pipefd[0], &u_id, sizeof(u_id));
1376 if (ret != sizeof(u_id)) {
1377 close(pipefd[0]);
1378 waitpid(pid, NULL, 0);
1379 SKIP(return, "Failed to read user namespace ID");
1380 }
1381
1382 ret = read(pipefd[0], &n_id, sizeof(n_id));
1383 if (ret != sizeof(n_id)) {
1384 close(pipefd[0]);
1385 waitpid(pid, NULL, 0);
1386 SKIP(return, "Failed to read network namespace ID");
1387 }
1388
1389 ret = read(pipefd[0], &ut_id, sizeof(ut_id));
1390 close(pipefd[0]);
1391 if (ret != sizeof(ut_id)) {
1392 waitpid(pid, NULL, 0);
1393 SKIP(return, "Failed to read UTS namespace ID");
1394 }
1395
1396 /* Construct file handles from namespace IDs */
1397 u_handle = (struct file_handle *)u_buf;
1398 u_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1399 u_handle->handle_type = FILEID_NSFS;
1400 struct nsfs_file_handle *u_fh = (struct nsfs_file_handle *)u_handle->f_handle;
1401 u_fh->ns_id = u_id;
1402 u_fh->ns_type = 0;
1403 u_fh->ns_inum = 0;
1404
1405 n_handle = (struct file_handle *)n_buf;
1406 n_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1407 n_handle->handle_type = FILEID_NSFS;
1408 struct nsfs_file_handle *n_fh = (struct nsfs_file_handle *)n_handle->f_handle;
1409 n_fh->ns_id = n_id;
1410 n_fh->ns_type = 0;
1411 n_fh->ns_inum = 0;
1412
1413 ut_handle = (struct file_handle *)ut_buf;
1414 ut_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1415 ut_handle->handle_type = FILEID_NSFS;
1416 struct nsfs_file_handle *ut_fh = (struct nsfs_file_handle *)ut_handle->f_handle;
1417 ut_fh->ns_id = ut_id;
1418 ut_fh->ns_type = 0;
1419 ut_fh->ns_inum = 0;
1420
1421 /* Open both non-user namespaces before process exits */
1422 int n_fd = open_by_handle_at(FD_NSFS_ROOT, n_handle, O_RDONLY);
1423 int ut_fd = open_by_handle_at(FD_NSFS_ROOT, ut_handle, O_RDONLY);
1424
1425 if (n_fd < 0 || ut_fd < 0) {
1426 if (n_fd >= 0) close(n_fd);
1427 if (ut_fd >= 0) close(ut_fd);
1428 waitpid(pid, NULL, 0);
1429 SKIP(return, "Failed to open namespaces");
1430 }
1431
1432 waitpid(pid, &status, 0);
1433 ASSERT_TRUE(WIFEXITED(status));
1434 ASSERT_EQ(WEXITSTATUS(status), 0);
1435
1436 /*
1437 * Both network and UTS namespaces are active.
1438 * User namespace should be active (gets 2 active refs).
1439 */
1440 TH_LOG("Both net and uts active - user namespace should be active");
1441 int u_fd = open_by_handle_at(FD_NSFS_ROOT, u_handle, O_RDONLY);
1442 ASSERT_GE(u_fd, 0);
1443 close(u_fd);
1444
1445 /* Close network namespace - user namespace should STILL be active */
1446 TH_LOG("Closing network ns - user ns should still be active (uts still active)");
1447 close(n_fd);
1448 u_fd = open_by_handle_at(FD_NSFS_ROOT, u_handle, O_RDONLY);
1449 ASSERT_GE(u_fd, 0);
1450 close(u_fd);
1451
1452 /* Close UTS namespace - user namespace should become inactive */
1453 TH_LOG("Closing uts ns - user ns should become inactive");
1454 close(ut_fd);
1455 u_fd = open_by_handle_at(FD_NSFS_ROOT, u_handle, O_RDONLY);
1456 ASSERT_LT(u_fd, 0);
1457 }
1458
1459 /*
1460 * Test hierarchical propagation with deep namespace hierarchy.
1461 * Create: init_user_ns -> user_A -> user_B -> net_ns
1462 * When net_ns is active, both user_A and user_B should be active.
1463 * This verifies the conditional recursion in __ns_ref_active_put() works.
1464 */
TEST(ns_deep_hierarchy_propagation)1465 TEST(ns_deep_hierarchy_propagation)
1466 {
1467 struct file_handle *ua_handle, *ub_handle, *net_handle;
1468 int ret, pipefd[2];
1469 pid_t pid;
1470 int status;
1471 __u64 ua_id, ub_id, net_id;
1472 char ua_buf[sizeof(*ua_handle) + MAX_HANDLE_SZ];
1473 char ub_buf[sizeof(*ub_handle) + MAX_HANDLE_SZ];
1474 char net_buf[sizeof(*net_handle) + MAX_HANDLE_SZ];
1475
1476 ASSERT_EQ(pipe(pipefd), 0);
1477 pid = fork();
1478 ASSERT_GE(pid, 0);
1479
1480 if (pid == 0) {
1481 close(pipefd[0]);
1482
1483 /* Create user_A -> user_B -> net hierarchy */
1484 if (setup_userns() < 0) {
1485 close(pipefd[1]);
1486 exit(1);
1487 }
1488
1489 int ua_fd = open("/proc/self/ns/user", O_RDONLY);
1490 if (ua_fd < 0) {
1491 close(pipefd[1]);
1492 exit(1);
1493 }
1494 if (ioctl(ua_fd, NS_GET_ID, &ua_id) < 0) {
1495 close(ua_fd);
1496 close(pipefd[1]);
1497 exit(1);
1498 }
1499 close(ua_fd);
1500
1501 if (setup_userns() < 0) {
1502 close(pipefd[1]);
1503 exit(1);
1504 }
1505
1506 int ub_fd = open("/proc/self/ns/user", O_RDONLY);
1507 if (ub_fd < 0) {
1508 close(pipefd[1]);
1509 exit(1);
1510 }
1511 if (ioctl(ub_fd, NS_GET_ID, &ub_id) < 0) {
1512 close(ub_fd);
1513 close(pipefd[1]);
1514 exit(1);
1515 }
1516 close(ub_fd);
1517
1518 if (unshare(CLONE_NEWNET) < 0) {
1519 close(pipefd[1]);
1520 exit(1);
1521 }
1522
1523 int net_fd = open("/proc/self/ns/net", O_RDONLY);
1524 if (net_fd < 0) {
1525 close(pipefd[1]);
1526 exit(1);
1527 }
1528 if (ioctl(net_fd, NS_GET_ID, &net_id) < 0) {
1529 close(net_fd);
1530 close(pipefd[1]);
1531 exit(1);
1532 }
1533 close(net_fd);
1534
1535 /* Send all three namespace IDs */
1536 write(pipefd[1], &ua_id, sizeof(ua_id));
1537 write(pipefd[1], &ub_id, sizeof(ub_id));
1538 write(pipefd[1], &net_id, sizeof(net_id));
1539 close(pipefd[1]);
1540 exit(0);
1541 }
1542
1543 close(pipefd[1]);
1544
1545 /* Read all three namespace IDs - fixed size, no parsing needed */
1546 ret = read(pipefd[0], &ua_id, sizeof(ua_id));
1547 if (ret != sizeof(ua_id)) {
1548 close(pipefd[0]);
1549 waitpid(pid, NULL, 0);
1550 SKIP(return, "Failed to read user_A namespace ID");
1551 }
1552
1553 ret = read(pipefd[0], &ub_id, sizeof(ub_id));
1554 if (ret != sizeof(ub_id)) {
1555 close(pipefd[0]);
1556 waitpid(pid, NULL, 0);
1557 SKIP(return, "Failed to read user_B namespace ID");
1558 }
1559
1560 ret = read(pipefd[0], &net_id, sizeof(net_id));
1561 close(pipefd[0]);
1562 if (ret != sizeof(net_id)) {
1563 waitpid(pid, NULL, 0);
1564 SKIP(return, "Failed to read network namespace ID");
1565 }
1566
1567 /* Construct file handles from namespace IDs */
1568 ua_handle = (struct file_handle *)ua_buf;
1569 ua_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1570 ua_handle->handle_type = FILEID_NSFS;
1571 struct nsfs_file_handle *ua_fh = (struct nsfs_file_handle *)ua_handle->f_handle;
1572 ua_fh->ns_id = ua_id;
1573 ua_fh->ns_type = 0;
1574 ua_fh->ns_inum = 0;
1575
1576 ub_handle = (struct file_handle *)ub_buf;
1577 ub_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1578 ub_handle->handle_type = FILEID_NSFS;
1579 struct nsfs_file_handle *ub_fh = (struct nsfs_file_handle *)ub_handle->f_handle;
1580 ub_fh->ns_id = ub_id;
1581 ub_fh->ns_type = 0;
1582 ub_fh->ns_inum = 0;
1583
1584 net_handle = (struct file_handle *)net_buf;
1585 net_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1586 net_handle->handle_type = FILEID_NSFS;
1587 struct nsfs_file_handle *net_fh = (struct nsfs_file_handle *)net_handle->f_handle;
1588 net_fh->ns_id = net_id;
1589 net_fh->ns_type = 0;
1590 net_fh->ns_inum = 0;
1591
1592 /* Open net_ns before child exits to keep it active */
1593 int net_fd = open_by_handle_at(FD_NSFS_ROOT, net_handle, O_RDONLY);
1594 if (net_fd < 0) {
1595 waitpid(pid, NULL, 0);
1596 SKIP(return, "Failed to open network namespace");
1597 }
1598
1599 waitpid(pid, &status, 0);
1600 ASSERT_TRUE(WIFEXITED(status));
1601 ASSERT_EQ(WEXITSTATUS(status), 0);
1602
1603 /* With net_ns active, both user_A and user_B should be active */
1604 TH_LOG("Testing user_B active (net_ns active causes propagation)");
1605 int ub_fd = open_by_handle_at(FD_NSFS_ROOT, ub_handle, O_RDONLY);
1606 ASSERT_GE(ub_fd, 0);
1607
1608 TH_LOG("Testing user_A active (propagated through user_B)");
1609 int ua_fd = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1610 ASSERT_GE(ua_fd, 0);
1611
1612 /* Close net_ns - user_B should stay active (we hold direct ref) */
1613 TH_LOG("Closing net_ns, user_B should remain active (direct ref held)");
1614 close(net_fd);
1615 int ub_fd2 = open_by_handle_at(FD_NSFS_ROOT, ub_handle, O_RDONLY);
1616 ASSERT_GE(ub_fd2, 0);
1617 close(ub_fd2);
1618
1619 /* Close user_B - user_A should stay active (we hold direct ref) */
1620 TH_LOG("Closing user_B, user_A should remain active (direct ref held)");
1621 close(ub_fd);
1622 int ua_fd2 = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1623 ASSERT_GE(ua_fd2, 0);
1624 close(ua_fd2);
1625
1626 /* Close user_A - everything should become inactive */
1627 TH_LOG("Closing user_A, all should become inactive");
1628 close(ua_fd);
1629
1630 /* All should now be inactive */
1631 ua_fd = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1632 ASSERT_LT(ua_fd, 0);
1633 }
1634
1635 /*
1636 * Test that parent stays active as long as ANY child is active.
1637 * Create parent user namespace with two child net namespaces.
1638 * Parent should remain active until BOTH children are inactive.
1639 */
TEST(ns_parent_multiple_children_refcount)1640 TEST(ns_parent_multiple_children_refcount)
1641 {
1642 struct file_handle *parent_handle, *net1_handle, *net2_handle;
1643 int ret, pipefd[2], syncpipe[2];
1644 pid_t pid;
1645 int status;
1646 __u64 p_id, n1_id, n2_id;
1647 char p_buf[sizeof(*parent_handle) + MAX_HANDLE_SZ];
1648 char n1_buf[sizeof(*net1_handle) + MAX_HANDLE_SZ];
1649 char n2_buf[sizeof(*net2_handle) + MAX_HANDLE_SZ];
1650 char sync_byte;
1651
1652 ASSERT_EQ(pipe(pipefd), 0);
1653 ASSERT_EQ(pipe(syncpipe), 0);
1654 pid = fork();
1655 ASSERT_GE(pid, 0);
1656
1657 if (pid == 0) {
1658 close(pipefd[0]);
1659 close(syncpipe[1]);
1660
1661 /* Create parent user namespace */
1662 if (setup_userns() < 0) {
1663 close(pipefd[1]);
1664 exit(1);
1665 }
1666
1667 int p_fd = open("/proc/self/ns/user", O_RDONLY);
1668 if (p_fd < 0) {
1669 close(pipefd[1]);
1670 exit(1);
1671 }
1672 if (ioctl(p_fd, NS_GET_ID, &p_id) < 0) {
1673 close(p_fd);
1674 close(pipefd[1]);
1675 exit(1);
1676 }
1677 close(p_fd);
1678
1679 /* Create first network namespace */
1680 if (unshare(CLONE_NEWNET) < 0) {
1681 close(pipefd[1]);
1682 close(syncpipe[0]);
1683 exit(1);
1684 }
1685
1686 int n1_fd = open("/proc/self/ns/net", O_RDONLY);
1687 if (n1_fd < 0) {
1688 close(pipefd[1]);
1689 close(syncpipe[0]);
1690 exit(1);
1691 }
1692 if (ioctl(n1_fd, NS_GET_ID, &n1_id) < 0) {
1693 close(n1_fd);
1694 close(pipefd[1]);
1695 close(syncpipe[0]);
1696 exit(1);
1697 }
1698 /* Keep n1_fd open so first namespace stays active */
1699
1700 /* Create second network namespace */
1701 if (unshare(CLONE_NEWNET) < 0) {
1702 close(n1_fd);
1703 close(pipefd[1]);
1704 close(syncpipe[0]);
1705 exit(1);
1706 }
1707
1708 int n2_fd = open("/proc/self/ns/net", O_RDONLY);
1709 if (n2_fd < 0) {
1710 close(n1_fd);
1711 close(pipefd[1]);
1712 close(syncpipe[0]);
1713 exit(1);
1714 }
1715 if (ioctl(n2_fd, NS_GET_ID, &n2_id) < 0) {
1716 close(n1_fd);
1717 close(n2_fd);
1718 close(pipefd[1]);
1719 close(syncpipe[0]);
1720 exit(1);
1721 }
1722 /* Keep both n1_fd and n2_fd open */
1723
1724 /* Send all namespace IDs */
1725 write(pipefd[1], &p_id, sizeof(p_id));
1726 write(pipefd[1], &n1_id, sizeof(n1_id));
1727 write(pipefd[1], &n2_id, sizeof(n2_id));
1728 close(pipefd[1]);
1729
1730 /* Wait for parent to signal before exiting */
1731 read(syncpipe[0], &sync_byte, 1);
1732 close(syncpipe[0]);
1733 exit(0);
1734 }
1735
1736 close(pipefd[1]);
1737 close(syncpipe[0]);
1738
1739 /* Read all three namespace IDs - fixed size, no parsing needed */
1740 ret = read(pipefd[0], &p_id, sizeof(p_id));
1741 if (ret != sizeof(p_id)) {
1742 close(pipefd[0]);
1743 waitpid(pid, NULL, 0);
1744 SKIP(return, "Failed to read parent namespace ID");
1745 }
1746
1747 ret = read(pipefd[0], &n1_id, sizeof(n1_id));
1748 if (ret != sizeof(n1_id)) {
1749 close(pipefd[0]);
1750 waitpid(pid, NULL, 0);
1751 SKIP(return, "Failed to read first network namespace ID");
1752 }
1753
1754 ret = read(pipefd[0], &n2_id, sizeof(n2_id));
1755 close(pipefd[0]);
1756 if (ret != sizeof(n2_id)) {
1757 waitpid(pid, NULL, 0);
1758 SKIP(return, "Failed to read second network namespace ID");
1759 }
1760
1761 /* Construct file handles from namespace IDs */
1762 parent_handle = (struct file_handle *)p_buf;
1763 parent_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1764 parent_handle->handle_type = FILEID_NSFS;
1765 struct nsfs_file_handle *p_fh = (struct nsfs_file_handle *)parent_handle->f_handle;
1766 p_fh->ns_id = p_id;
1767 p_fh->ns_type = 0;
1768 p_fh->ns_inum = 0;
1769
1770 net1_handle = (struct file_handle *)n1_buf;
1771 net1_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1772 net1_handle->handle_type = FILEID_NSFS;
1773 struct nsfs_file_handle *n1_fh = (struct nsfs_file_handle *)net1_handle->f_handle;
1774 n1_fh->ns_id = n1_id;
1775 n1_fh->ns_type = 0;
1776 n1_fh->ns_inum = 0;
1777
1778 net2_handle = (struct file_handle *)n2_buf;
1779 net2_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1780 net2_handle->handle_type = FILEID_NSFS;
1781 struct nsfs_file_handle *n2_fh = (struct nsfs_file_handle *)net2_handle->f_handle;
1782 n2_fh->ns_id = n2_id;
1783 n2_fh->ns_type = 0;
1784 n2_fh->ns_inum = 0;
1785
1786 /* Open both net namespaces while child is still alive */
1787 int n1_fd = open_by_handle_at(FD_NSFS_ROOT, net1_handle, O_RDONLY);
1788 int n2_fd = open_by_handle_at(FD_NSFS_ROOT, net2_handle, O_RDONLY);
1789 if (n1_fd < 0 || n2_fd < 0) {
1790 if (n1_fd >= 0) close(n1_fd);
1791 if (n2_fd >= 0) close(n2_fd);
1792 sync_byte = 'G';
1793 write(syncpipe[1], &sync_byte, 1);
1794 close(syncpipe[1]);
1795 waitpid(pid, NULL, 0);
1796 SKIP(return, "Failed to open net namespaces");
1797 }
1798
1799 /* Signal child that we have opened the namespaces */
1800 sync_byte = 'G';
1801 write(syncpipe[1], &sync_byte, 1);
1802 close(syncpipe[1]);
1803
1804 /* Wait for child to exit */
1805 waitpid(pid, &status, 0);
1806 ASSERT_TRUE(WIFEXITED(status));
1807 ASSERT_EQ(WEXITSTATUS(status), 0);
1808
1809 /* Parent should be active (has 2 active children) */
1810 TH_LOG("Both net namespaces active - parent should be active");
1811 int p_fd = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
1812 ASSERT_GE(p_fd, 0);
1813 close(p_fd);
1814
1815 /* Close first net namespace - parent should STILL be active */
1816 TH_LOG("Closing first net ns - parent should still be active");
1817 close(n1_fd);
1818 p_fd = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
1819 ASSERT_GE(p_fd, 0);
1820 close(p_fd);
1821
1822 /* Close second net namespace - parent should become inactive */
1823 TH_LOG("Closing second net ns - parent should become inactive");
1824 close(n2_fd);
1825 p_fd = open_by_handle_at(FD_NSFS_ROOT, parent_handle, O_RDONLY);
1826 ASSERT_LT(p_fd, 0);
1827 }
1828
1829 /*
1830 * Test that user namespace as a child also propagates correctly.
1831 * Create user_A -> user_B, verify when user_B is active that user_A
1832 * is also active. This is different from non-user namespace children.
1833 */
TEST(ns_userns_child_propagation)1834 TEST(ns_userns_child_propagation)
1835 {
1836 struct file_handle *ua_handle, *ub_handle;
1837 int ret, pipefd[2];
1838 pid_t pid;
1839 int status;
1840 __u64 ua_id, ub_id;
1841 char ua_buf[sizeof(*ua_handle) + MAX_HANDLE_SZ];
1842 char ub_buf[sizeof(*ub_handle) + MAX_HANDLE_SZ];
1843
1844 ASSERT_EQ(pipe(pipefd), 0);
1845 pid = fork();
1846 ASSERT_GE(pid, 0);
1847
1848 if (pid == 0) {
1849 close(pipefd[0]);
1850
1851 /* Create user_A */
1852 if (setup_userns() < 0) {
1853 close(pipefd[1]);
1854 exit(1);
1855 }
1856
1857 int ua_fd = open("/proc/self/ns/user", O_RDONLY);
1858 if (ua_fd < 0) {
1859 close(pipefd[1]);
1860 exit(1);
1861 }
1862 if (ioctl(ua_fd, NS_GET_ID, &ua_id) < 0) {
1863 close(ua_fd);
1864 close(pipefd[1]);
1865 exit(1);
1866 }
1867 close(ua_fd);
1868
1869 /* Create user_B (child of user_A) */
1870 if (setup_userns() < 0) {
1871 close(pipefd[1]);
1872 exit(1);
1873 }
1874
1875 int ub_fd = open("/proc/self/ns/user", O_RDONLY);
1876 if (ub_fd < 0) {
1877 close(pipefd[1]);
1878 exit(1);
1879 }
1880 if (ioctl(ub_fd, NS_GET_ID, &ub_id) < 0) {
1881 close(ub_fd);
1882 close(pipefd[1]);
1883 exit(1);
1884 }
1885 close(ub_fd);
1886
1887 /* Send both namespace IDs */
1888 write(pipefd[1], &ua_id, sizeof(ua_id));
1889 write(pipefd[1], &ub_id, sizeof(ub_id));
1890 close(pipefd[1]);
1891 exit(0);
1892 }
1893
1894 close(pipefd[1]);
1895
1896 /* Read both namespace IDs - fixed size, no parsing needed */
1897 ret = read(pipefd[0], &ua_id, sizeof(ua_id));
1898 if (ret != sizeof(ua_id)) {
1899 close(pipefd[0]);
1900 waitpid(pid, NULL, 0);
1901 SKIP(return, "Failed to read user_A namespace ID");
1902 }
1903
1904 ret = read(pipefd[0], &ub_id, sizeof(ub_id));
1905 close(pipefd[0]);
1906 if (ret != sizeof(ub_id)) {
1907 waitpid(pid, NULL, 0);
1908 SKIP(return, "Failed to read user_B namespace ID");
1909 }
1910
1911 /* Construct file handles from namespace IDs */
1912 ua_handle = (struct file_handle *)ua_buf;
1913 ua_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1914 ua_handle->handle_type = FILEID_NSFS;
1915 struct nsfs_file_handle *ua_fh = (struct nsfs_file_handle *)ua_handle->f_handle;
1916 ua_fh->ns_id = ua_id;
1917 ua_fh->ns_type = 0;
1918 ua_fh->ns_inum = 0;
1919
1920 ub_handle = (struct file_handle *)ub_buf;
1921 ub_handle->handle_bytes = sizeof(struct nsfs_file_handle);
1922 ub_handle->handle_type = FILEID_NSFS;
1923 struct nsfs_file_handle *ub_fh = (struct nsfs_file_handle *)ub_handle->f_handle;
1924 ub_fh->ns_id = ub_id;
1925 ub_fh->ns_type = 0;
1926 ub_fh->ns_inum = 0;
1927
1928 /* Open user_B before child exits */
1929 int ub_fd = open_by_handle_at(FD_NSFS_ROOT, ub_handle, O_RDONLY);
1930 if (ub_fd < 0) {
1931 waitpid(pid, NULL, 0);
1932 SKIP(return, "Failed to open user_B");
1933 }
1934
1935 waitpid(pid, &status, 0);
1936 ASSERT_TRUE(WIFEXITED(status));
1937 ASSERT_EQ(WEXITSTATUS(status), 0);
1938
1939 /* With user_B active, user_A should also be active */
1940 TH_LOG("Testing user_A active when child user_B is active");
1941 int ua_fd = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1942 ASSERT_GE(ua_fd, 0);
1943
1944 /* Close user_B */
1945 TH_LOG("Closing user_B");
1946 close(ub_fd);
1947
1948 /* user_A should remain active (we hold direct ref) */
1949 int ua_fd2 = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1950 ASSERT_GE(ua_fd2, 0);
1951 close(ua_fd2);
1952
1953 /* Close user_A - should become inactive */
1954 TH_LOG("Closing user_A - should become inactive");
1955 close(ua_fd);
1956
1957 ua_fd = open_by_handle_at(FD_NSFS_ROOT, ua_handle, O_RDONLY);
1958 ASSERT_LT(ua_fd, 0);
1959 }
1960
1961 /*
1962 * Test different namespace types (net, uts, ipc) all contributing
1963 * active references to the same owning user namespace.
1964 */
TEST(ns_mixed_types_same_owner)1965 TEST(ns_mixed_types_same_owner)
1966 {
1967 struct file_handle *user_handle, *net_handle, *uts_handle;
1968 int ret, pipefd[2];
1969 pid_t pid;
1970 int status;
1971 __u64 u_id, n_id, ut_id;
1972 char u_buf[sizeof(*user_handle) + MAX_HANDLE_SZ];
1973 char n_buf[sizeof(*net_handle) + MAX_HANDLE_SZ];
1974 char ut_buf[sizeof(*uts_handle) + MAX_HANDLE_SZ];
1975
1976 ASSERT_EQ(pipe(pipefd), 0);
1977 pid = fork();
1978 ASSERT_GE(pid, 0);
1979
1980 if (pid == 0) {
1981 close(pipefd[0]);
1982
1983 if (setup_userns() < 0) {
1984 close(pipefd[1]);
1985 exit(1);
1986 }
1987
1988 int u_fd = open("/proc/self/ns/user", O_RDONLY);
1989 if (u_fd < 0) {
1990 close(pipefd[1]);
1991 exit(1);
1992 }
1993 if (ioctl(u_fd, NS_GET_ID, &u_id) < 0) {
1994 close(u_fd);
1995 close(pipefd[1]);
1996 exit(1);
1997 }
1998 close(u_fd);
1999
2000 if (unshare(CLONE_NEWNET) < 0) {
2001 close(pipefd[1]);
2002 exit(1);
2003 }
2004
2005 int n_fd = open("/proc/self/ns/net", O_RDONLY);
2006 if (n_fd < 0) {
2007 close(pipefd[1]);
2008 exit(1);
2009 }
2010 if (ioctl(n_fd, NS_GET_ID, &n_id) < 0) {
2011 close(n_fd);
2012 close(pipefd[1]);
2013 exit(1);
2014 }
2015 close(n_fd);
2016
2017 if (unshare(CLONE_NEWUTS) < 0) {
2018 close(pipefd[1]);
2019 exit(1);
2020 }
2021
2022 int ut_fd = open("/proc/self/ns/uts", O_RDONLY);
2023 if (ut_fd < 0) {
2024 close(pipefd[1]);
2025 exit(1);
2026 }
2027 if (ioctl(ut_fd, NS_GET_ID, &ut_id) < 0) {
2028 close(ut_fd);
2029 close(pipefd[1]);
2030 exit(1);
2031 }
2032 close(ut_fd);
2033
2034 /* Send all namespace IDs */
2035 write(pipefd[1], &u_id, sizeof(u_id));
2036 write(pipefd[1], &n_id, sizeof(n_id));
2037 write(pipefd[1], &ut_id, sizeof(ut_id));
2038 close(pipefd[1]);
2039 exit(0);
2040 }
2041
2042 close(pipefd[1]);
2043
2044 /* Read all three namespace IDs - fixed size, no parsing needed */
2045 ret = read(pipefd[0], &u_id, sizeof(u_id));
2046 if (ret != sizeof(u_id)) {
2047 close(pipefd[0]);
2048 waitpid(pid, NULL, 0);
2049 SKIP(return, "Failed to read user namespace ID");
2050 }
2051
2052 ret = read(pipefd[0], &n_id, sizeof(n_id));
2053 if (ret != sizeof(n_id)) {
2054 close(pipefd[0]);
2055 waitpid(pid, NULL, 0);
2056 SKIP(return, "Failed to read network namespace ID");
2057 }
2058
2059 ret = read(pipefd[0], &ut_id, sizeof(ut_id));
2060 close(pipefd[0]);
2061 if (ret != sizeof(ut_id)) {
2062 waitpid(pid, NULL, 0);
2063 SKIP(return, "Failed to read UTS namespace ID");
2064 }
2065
2066 /* Construct file handles from namespace IDs */
2067 user_handle = (struct file_handle *)u_buf;
2068 user_handle->handle_bytes = sizeof(struct nsfs_file_handle);
2069 user_handle->handle_type = FILEID_NSFS;
2070 struct nsfs_file_handle *u_fh = (struct nsfs_file_handle *)user_handle->f_handle;
2071 u_fh->ns_id = u_id;
2072 u_fh->ns_type = 0;
2073 u_fh->ns_inum = 0;
2074
2075 net_handle = (struct file_handle *)n_buf;
2076 net_handle->handle_bytes = sizeof(struct nsfs_file_handle);
2077 net_handle->handle_type = FILEID_NSFS;
2078 struct nsfs_file_handle *n_fh = (struct nsfs_file_handle *)net_handle->f_handle;
2079 n_fh->ns_id = n_id;
2080 n_fh->ns_type = 0;
2081 n_fh->ns_inum = 0;
2082
2083 uts_handle = (struct file_handle *)ut_buf;
2084 uts_handle->handle_bytes = sizeof(struct nsfs_file_handle);
2085 uts_handle->handle_type = FILEID_NSFS;
2086 struct nsfs_file_handle *ut_fh = (struct nsfs_file_handle *)uts_handle->f_handle;
2087 ut_fh->ns_id = ut_id;
2088 ut_fh->ns_type = 0;
2089 ut_fh->ns_inum = 0;
2090
2091 /* Open both non-user namespaces */
2092 int n_fd = open_by_handle_at(FD_NSFS_ROOT, net_handle, O_RDONLY);
2093 int ut_fd = open_by_handle_at(FD_NSFS_ROOT, uts_handle, O_RDONLY);
2094 if (n_fd < 0 || ut_fd < 0) {
2095 if (n_fd >= 0) close(n_fd);
2096 if (ut_fd >= 0) close(ut_fd);
2097 waitpid(pid, NULL, 0);
2098 SKIP(return, "Failed to open namespaces");
2099 }
2100
2101 waitpid(pid, &status, 0);
2102 ASSERT_TRUE(WIFEXITED(status));
2103 ASSERT_EQ(WEXITSTATUS(status), 0);
2104
2105 /* User namespace should be active (2 active children) */
2106 TH_LOG("Both net and uts active - user ns should be active");
2107 int u_fd = open_by_handle_at(FD_NSFS_ROOT, user_handle, O_RDONLY);
2108 ASSERT_GE(u_fd, 0);
2109 close(u_fd);
2110
2111 /* Close net - user ns should STILL be active (uts still active) */
2112 TH_LOG("Closing net - user ns should still be active");
2113 close(n_fd);
2114 u_fd = open_by_handle_at(FD_NSFS_ROOT, user_handle, O_RDONLY);
2115 ASSERT_GE(u_fd, 0);
2116 close(u_fd);
2117
2118 /* Close uts - user ns should become inactive */
2119 TH_LOG("Closing uts - user ns should become inactive");
2120 close(ut_fd);
2121 u_fd = open_by_handle_at(FD_NSFS_ROOT, user_handle, O_RDONLY);
2122 ASSERT_LT(u_fd, 0);
2123 }
2124
2125 /* Thread test helpers and structures */
2126 struct thread_ns_info {
2127 __u64 ns_id;
2128 int pidfd;
2129 int pipefd;
2130 int syncfd_read;
2131 int syncfd_write;
2132 int exit_code;
2133 };
2134
thread_create_namespace(void * arg)2135 static void *thread_create_namespace(void *arg)
2136 {
2137 struct thread_ns_info *info = (struct thread_ns_info *)arg;
2138 int ret;
2139
2140 /* Create new network namespace */
2141 ret = unshare(CLONE_NEWNET);
2142 if (ret < 0) {
2143 info->exit_code = 1;
2144 return NULL;
2145 }
2146
2147 /* Get namespace ID */
2148 int fd = open("/proc/thread-self/ns/net", O_RDONLY);
2149 if (fd < 0) {
2150 info->exit_code = 2;
2151 return NULL;
2152 }
2153
2154 ret = ioctl(fd, NS_GET_ID, &info->ns_id);
2155 close(fd);
2156 if (ret < 0) {
2157 info->exit_code = 3;
2158 return NULL;
2159 }
2160
2161 /* Open a pidfd so the main thread can later poll it to detect actual thread exit. */
2162 info->pidfd = syscall(__NR_pidfd_open, gettid(), PIDFD_THREAD);
2163 if (info->pidfd < 0) {
2164 info->exit_code = 6;
2165 return NULL;
2166 }
2167
2168 /* Send namespace ID to main thread */
2169 if (write(info->pipefd, &info->ns_id, sizeof(info->ns_id)) != sizeof(info->ns_id)) {
2170 info->exit_code = 4;
2171 return NULL;
2172 }
2173
2174 /* Wait for signal to exit */
2175 char sync_byte;
2176 if (read(info->syncfd_read, &sync_byte, 1) != 1) {
2177 info->exit_code = 5;
2178 return NULL;
2179 }
2180
2181 info->exit_code = 0;
2182 return NULL;
2183 }
2184
2185 /*
2186 * pthread_join() returns once exit_mm() wakes the CLEARTID futex, before
2187 * exit_nsproxy_namespaces() drops the namespace reference. A pidfd becomes
2188 * readable only from exit_notify(), after that point.
2189 */
wait_thread_exited(int pidfd)2190 static int wait_thread_exited(int pidfd)
2191 {
2192 struct pollfd pfd = { .fd = pidfd, .events = POLLIN };
2193 int ret = poll(&pfd, 1, -1);
2194
2195 close(pidfd);
2196 return ret;
2197 }
2198
2199 /*
2200 * Test that namespace becomes inactive after thread exits.
2201 * This verifies active reference counting works with threads, not just processes.
2202 */
TEST(thread_ns_inactive_after_exit)2203 TEST(thread_ns_inactive_after_exit)
2204 {
2205 pthread_t thread;
2206 struct thread_ns_info info;
2207 struct file_handle *handle;
2208 int pipefd[2];
2209 int syncpipe[2];
2210 int ret;
2211 char sync_byte;
2212 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
2213
2214 ASSERT_EQ(pipe(pipefd), 0);
2215 ASSERT_EQ(pipe(syncpipe), 0);
2216
2217 info.pipefd = pipefd[1];
2218 info.syncfd_read = syncpipe[0];
2219 info.syncfd_write = -1;
2220 info.exit_code = -1;
2221
2222 /* Create thread that will create a namespace */
2223 ret = pthread_create(&thread, NULL, thread_create_namespace, &info);
2224 ASSERT_EQ(ret, 0);
2225
2226 /* Read namespace ID from thread */
2227 __u64 ns_id;
2228 ret = read(pipefd[0], &ns_id, sizeof(ns_id));
2229 if (ret != sizeof(ns_id)) {
2230 sync_byte = 'X';
2231 write(syncpipe[1], &sync_byte, 1);
2232 pthread_join(thread, NULL);
2233 close(pipefd[0]);
2234 close(pipefd[1]);
2235 close(syncpipe[0]);
2236 close(syncpipe[1]);
2237 SKIP(return, "Failed to read namespace ID from thread");
2238 }
2239
2240 TH_LOG("Thread created namespace with ID %llu", (unsigned long long)ns_id);
2241
2242 /* Construct file handle */
2243 handle = (struct file_handle *)buf;
2244 handle->handle_bytes = sizeof(struct nsfs_file_handle);
2245 handle->handle_type = FILEID_NSFS;
2246 struct nsfs_file_handle *fh = (struct nsfs_file_handle *)handle->f_handle;
2247 fh->ns_id = ns_id;
2248 fh->ns_type = 0;
2249 fh->ns_inum = 0;
2250
2251 /* Namespace should be active while thread is alive */
2252 TH_LOG("Attempting to open namespace while thread is alive (should succeed)");
2253 int nsfd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
2254 ASSERT_GE(nsfd, 0);
2255 close(nsfd);
2256
2257 /* Signal thread to exit */
2258 TH_LOG("Signaling thread to exit");
2259 sync_byte = 'X';
2260 ASSERT_EQ(write(syncpipe[1], &sync_byte, 1), 1);
2261 close(syncpipe[1]);
2262
2263 /* Wait for thread to exit */
2264 ASSERT_EQ(wait_thread_exited(info.pidfd), 1);
2265 ASSERT_EQ(pthread_join(thread, NULL), 0);
2266 close(pipefd[0]);
2267 close(pipefd[1]);
2268 close(syncpipe[0]);
2269
2270 if (info.exit_code != 0)
2271 SKIP(return, "Thread failed to create namespace");
2272
2273 TH_LOG("Thread exited, namespace should be inactive");
2274
2275 /* Namespace should now be inactive */
2276 nsfd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
2277 ASSERT_LT(nsfd, 0);
2278 /* Should fail with ENOENT (inactive) or ESTALE (gone) */
2279 TH_LOG("Namespace inactive as expected: %s (errno=%d)", strerror(errno), errno);
2280 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
2281 }
2282
2283 /*
2284 * Test that a namespace remains active while a thread holds an fd to it.
2285 * Even after the thread exits, the namespace should remain active as long as
2286 * another thread holds a file descriptor to it.
2287 */
TEST(thread_ns_fd_keeps_active)2288 TEST(thread_ns_fd_keeps_active)
2289 {
2290 pthread_t thread;
2291 struct thread_ns_info info;
2292 struct file_handle *handle;
2293 int pipefd[2];
2294 int syncpipe[2];
2295 int ret;
2296 char sync_byte;
2297 char buf[sizeof(*handle) + MAX_HANDLE_SZ];
2298
2299 ASSERT_EQ(pipe(pipefd), 0);
2300 ASSERT_EQ(pipe(syncpipe), 0);
2301
2302 info.pipefd = pipefd[1];
2303 info.syncfd_read = syncpipe[0];
2304 info.syncfd_write = -1;
2305 info.exit_code = -1;
2306
2307 /* Create thread that will create a namespace */
2308 ret = pthread_create(&thread, NULL, thread_create_namespace, &info);
2309 ASSERT_EQ(ret, 0);
2310
2311 /* Read namespace ID from thread */
2312 __u64 ns_id;
2313 ret = read(pipefd[0], &ns_id, sizeof(ns_id));
2314 if (ret != sizeof(ns_id)) {
2315 sync_byte = 'X';
2316 write(syncpipe[1], &sync_byte, 1);
2317 pthread_join(thread, NULL);
2318 close(pipefd[0]);
2319 close(pipefd[1]);
2320 close(syncpipe[0]);
2321 close(syncpipe[1]);
2322 SKIP(return, "Failed to read namespace ID from thread");
2323 }
2324
2325 TH_LOG("Thread created namespace with ID %llu", (unsigned long long)ns_id);
2326
2327 /* Construct file handle */
2328 handle = (struct file_handle *)buf;
2329 handle->handle_bytes = sizeof(struct nsfs_file_handle);
2330 handle->handle_type = FILEID_NSFS;
2331 struct nsfs_file_handle *fh = (struct nsfs_file_handle *)handle->f_handle;
2332 fh->ns_id = ns_id;
2333 fh->ns_type = 0;
2334 fh->ns_inum = 0;
2335
2336 /* Open namespace while thread is alive */
2337 TH_LOG("Opening namespace while thread is alive");
2338 int nsfd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
2339 ASSERT_GE(nsfd, 0);
2340
2341 /* Signal thread to exit */
2342 TH_LOG("Signaling thread to exit");
2343 sync_byte = 'X';
2344 write(syncpipe[1], &sync_byte, 1);
2345 close(syncpipe[1]);
2346
2347 /* Wait for thread to exit */
2348 ASSERT_EQ(wait_thread_exited(info.pidfd), 1);
2349 pthread_join(thread, NULL);
2350 close(pipefd[0]);
2351 close(pipefd[1]);
2352 close(syncpipe[0]);
2353
2354 if (info.exit_code != 0) {
2355 close(nsfd);
2356 SKIP(return, "Thread failed to create namespace");
2357 }
2358
2359 TH_LOG("Thread exited, but main thread holds fd - namespace should remain active");
2360
2361 /* Namespace should still be active because we hold an fd */
2362 int nsfd2 = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
2363 ASSERT_GE(nsfd2, 0);
2364
2365 /* Verify it's the same namespace */
2366 struct stat st1, st2;
2367 ASSERT_EQ(fstat(nsfd, &st1), 0);
2368 ASSERT_EQ(fstat(nsfd2, &st2), 0);
2369 ASSERT_EQ(st1.st_ino, st2.st_ino);
2370 close(nsfd2);
2371
2372 TH_LOG("Closing fd - namespace should become inactive");
2373 close(nsfd);
2374
2375 /* Now namespace should be inactive */
2376 nsfd = open_by_handle_at(FD_NSFS_ROOT, handle, O_RDONLY);
2377 ASSERT_LT(nsfd, 0);
2378 /* Should fail with ENOENT (inactive) or ESTALE (gone) */
2379 TH_LOG("Namespace inactive as expected: %s (errno=%d)", strerror(errno), errno);
2380 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
2381 }
2382
2383 /* Structure for thread data in subprocess */
2384 struct thread_sleep_data {
2385 int syncfd_read;
2386 };
2387
thread_sleep_and_wait(void * arg)2388 static void *thread_sleep_and_wait(void *arg)
2389 {
2390 struct thread_sleep_data *data = (struct thread_sleep_data *)arg;
2391 char sync_byte;
2392
2393 /* Wait for signal to exit - read will unblock when pipe is closed */
2394 (void)read(data->syncfd_read, &sync_byte, 1);
2395 return NULL;
2396 }
2397
2398 /*
2399 * Test that namespaces become inactive after subprocess with multiple threads exits.
2400 * Create a subprocess that unshares user and network namespaces, then creates two
2401 * threads that share those namespaces. Verify that after all threads and subprocess
2402 * exit, the namespaces are no longer listed by listns() and cannot be opened by
2403 * open_by_handle_at().
2404 */
TEST(thread_subprocess_ns_inactive_after_all_exit)2405 TEST(thread_subprocess_ns_inactive_after_all_exit)
2406 {
2407 int pipefd[2];
2408 int sv[2];
2409 pid_t pid;
2410 int status;
2411 __u64 user_id, net_id;
2412 struct file_handle *user_handle, *net_handle;
2413 char user_buf[sizeof(*user_handle) + MAX_HANDLE_SZ];
2414 char net_buf[sizeof(*net_handle) + MAX_HANDLE_SZ];
2415 char sync_byte;
2416 int ret;
2417
2418 ASSERT_EQ(pipe(pipefd), 0);
2419 ASSERT_EQ(socketpair(AF_UNIX, SOCK_STREAM, 0, sv), 0);
2420
2421 pid = fork();
2422 ASSERT_GE(pid, 0);
2423
2424 if (pid == 0) {
2425 /* Child process */
2426 close(pipefd[0]);
2427 close(sv[0]);
2428
2429 /* Create user namespace with mappings */
2430 if (setup_userns() < 0) {
2431 fprintf(stderr, "Child: setup_userns() failed: %s\n", strerror(errno));
2432 close(pipefd[1]);
2433 close(sv[1]);
2434 exit(1);
2435 }
2436 fprintf(stderr, "Child: setup_userns() succeeded\n");
2437
2438 /* Get user namespace ID */
2439 int user_fd = open("/proc/self/ns/user", O_RDONLY);
2440 if (user_fd < 0) {
2441 fprintf(stderr, "Child: open(/proc/self/ns/user) failed: %s\n", strerror(errno));
2442 close(pipefd[1]);
2443 close(sv[1]);
2444 exit(1);
2445 }
2446
2447 if (ioctl(user_fd, NS_GET_ID, &user_id) < 0) {
2448 fprintf(stderr, "Child: ioctl(NS_GET_ID) for user ns failed: %s\n", strerror(errno));
2449 close(user_fd);
2450 close(pipefd[1]);
2451 close(sv[1]);
2452 exit(1);
2453 }
2454 close(user_fd);
2455 fprintf(stderr, "Child: user ns ID = %llu\n", (unsigned long long)user_id);
2456
2457 /* Unshare network namespace */
2458 if (unshare(CLONE_NEWNET) < 0) {
2459 fprintf(stderr, "Child: unshare(CLONE_NEWNET) failed: %s\n", strerror(errno));
2460 close(pipefd[1]);
2461 close(sv[1]);
2462 exit(1);
2463 }
2464 fprintf(stderr, "Child: unshare(CLONE_NEWNET) succeeded\n");
2465
2466 /* Get network namespace ID */
2467 int net_fd = open("/proc/self/ns/net", O_RDONLY);
2468 if (net_fd < 0) {
2469 fprintf(stderr, "Child: open(/proc/self/ns/net) failed: %s\n", strerror(errno));
2470 close(pipefd[1]);
2471 close(sv[1]);
2472 exit(1);
2473 }
2474
2475 if (ioctl(net_fd, NS_GET_ID, &net_id) < 0) {
2476 fprintf(stderr, "Child: ioctl(NS_GET_ID) for net ns failed: %s\n", strerror(errno));
2477 close(net_fd);
2478 close(pipefd[1]);
2479 close(sv[1]);
2480 exit(1);
2481 }
2482 close(net_fd);
2483 fprintf(stderr, "Child: net ns ID = %llu\n", (unsigned long long)net_id);
2484
2485 /* Send namespace IDs to parent */
2486 if (write(pipefd[1], &user_id, sizeof(user_id)) != sizeof(user_id)) {
2487 fprintf(stderr, "Child: write(user_id) failed: %s\n", strerror(errno));
2488 exit(1);
2489 }
2490 if (write(pipefd[1], &net_id, sizeof(net_id)) != sizeof(net_id)) {
2491 fprintf(stderr, "Child: write(net_id) failed: %s\n", strerror(errno));
2492 exit(1);
2493 }
2494 close(pipefd[1]);
2495 fprintf(stderr, "Child: sent namespace IDs to parent\n");
2496
2497 /* Create two threads that share the namespaces */
2498 pthread_t thread1, thread2;
2499 struct thread_sleep_data data;
2500 data.syncfd_read = sv[1];
2501
2502 int ret_thread = pthread_create(&thread1, NULL, thread_sleep_and_wait, &data);
2503 if (ret_thread != 0) {
2504 fprintf(stderr, "Child: pthread_create(thread1) failed: %s\n", strerror(ret_thread));
2505 close(sv[1]);
2506 exit(1);
2507 }
2508 fprintf(stderr, "Child: created thread1\n");
2509
2510 ret_thread = pthread_create(&thread2, NULL, thread_sleep_and_wait, &data);
2511 if (ret_thread != 0) {
2512 fprintf(stderr, "Child: pthread_create(thread2) failed: %s\n", strerror(ret_thread));
2513 close(sv[1]);
2514 pthread_cancel(thread1);
2515 exit(1);
2516 }
2517 fprintf(stderr, "Child: created thread2\n");
2518
2519 /* Wait for threads to complete - they will unblock when parent writes */
2520 fprintf(stderr, "Child: waiting for threads to exit\n");
2521 pthread_join(thread1, NULL);
2522 fprintf(stderr, "Child: thread1 exited\n");
2523 pthread_join(thread2, NULL);
2524 fprintf(stderr, "Child: thread2 exited\n");
2525
2526 close(sv[1]);
2527
2528 /* Exit - namespaces should become inactive */
2529 fprintf(stderr, "Child: all threads joined, exiting with success\n");
2530 exit(0);
2531 }
2532
2533 /* Parent process */
2534 close(pipefd[1]);
2535 close(sv[1]);
2536
2537 TH_LOG("Parent: waiting to read namespace IDs from child");
2538
2539 /* Read namespace IDs from child */
2540 ret = read(pipefd[0], &user_id, sizeof(user_id));
2541 if (ret != sizeof(user_id)) {
2542 TH_LOG("Parent: failed to read user_id, ret=%d, errno=%s", ret, strerror(errno));
2543 close(pipefd[0]);
2544 sync_byte = 'X';
2545 (void)write(sv[0], &sync_byte, 1);
2546 close(sv[0]);
2547 waitpid(pid, NULL, 0);
2548 SKIP(return, "Failed to read user namespace ID from child");
2549 }
2550
2551 ret = read(pipefd[0], &net_id, sizeof(net_id));
2552 close(pipefd[0]);
2553 if (ret != sizeof(net_id)) {
2554 TH_LOG("Parent: failed to read net_id, ret=%d, errno=%s", ret, strerror(errno));
2555 sync_byte = 'X';
2556 (void)write(sv[0], &sync_byte, 1);
2557 close(sv[0]);
2558 waitpid(pid, NULL, 0);
2559 SKIP(return, "Failed to read network namespace ID from child");
2560 }
2561
2562 TH_LOG("Child created user ns %llu and net ns %llu with 2 threads",
2563 (unsigned long long)user_id, (unsigned long long)net_id);
2564
2565 /* Construct file handles */
2566 user_handle = (struct file_handle *)user_buf;
2567 user_handle->handle_bytes = sizeof(struct nsfs_file_handle);
2568 user_handle->handle_type = FILEID_NSFS;
2569 struct nsfs_file_handle *user_fh = (struct nsfs_file_handle *)user_handle->f_handle;
2570 user_fh->ns_id = user_id;
2571 user_fh->ns_type = 0;
2572 user_fh->ns_inum = 0;
2573
2574 net_handle = (struct file_handle *)net_buf;
2575 net_handle->handle_bytes = sizeof(struct nsfs_file_handle);
2576 net_handle->handle_type = FILEID_NSFS;
2577 struct nsfs_file_handle *net_fh = (struct nsfs_file_handle *)net_handle->f_handle;
2578 net_fh->ns_id = net_id;
2579 net_fh->ns_type = 0;
2580 net_fh->ns_inum = 0;
2581
2582 /* Verify namespaces are active while subprocess and threads are alive */
2583 TH_LOG("Verifying namespaces are active while subprocess with threads is running");
2584 int user_fd = open_by_handle_at(FD_NSFS_ROOT, user_handle, O_RDONLY);
2585 ASSERT_GE(user_fd, 0);
2586
2587 int net_fd = open_by_handle_at(FD_NSFS_ROOT, net_handle, O_RDONLY);
2588 ASSERT_GE(net_fd, 0);
2589
2590 close(user_fd);
2591 close(net_fd);
2592
2593 /* Also verify they appear in listns() */
2594 TH_LOG("Verifying namespaces appear in listns() while active");
2595 struct ns_id_req req = {
2596 .size = sizeof(struct ns_id_req),
2597 .spare = 0,
2598 .ns_id = 0,
2599 .ns_type = CLONE_NEWUSER,
2600 .spare2 = 0,
2601 .user_ns_id = 0,
2602 };
2603 __u64 ns_ids[256];
2604 int nr_ids = sys_listns(&req, ns_ids, 256, 0);
2605 if (nr_ids < 0) {
2606 TH_LOG("listns() not available, skipping listns verification");
2607 } else {
2608 /* Check if user_id is in the list */
2609 int found_user = 0;
2610 for (int i = 0; i < nr_ids; i++) {
2611 if (ns_ids[i] == user_id) {
2612 found_user = 1;
2613 break;
2614 }
2615 }
2616 ASSERT_TRUE(found_user);
2617 TH_LOG("User namespace found in listns() as expected");
2618
2619 /* Check network namespace */
2620 req.ns_type = CLONE_NEWNET;
2621 nr_ids = sys_listns(&req, ns_ids, 256, 0);
2622 if (nr_ids >= 0) {
2623 int found_net = 0;
2624 for (int i = 0; i < nr_ids; i++) {
2625 if (ns_ids[i] == net_id) {
2626 found_net = 1;
2627 break;
2628 }
2629 }
2630 ASSERT_TRUE(found_net);
2631 TH_LOG("Network namespace found in listns() as expected");
2632 }
2633 }
2634
2635 /* Signal threads to exit */
2636 TH_LOG("Signaling threads to exit");
2637 sync_byte = 'X';
2638 /* Write two bytes - one for each thread */
2639 ASSERT_EQ(write(sv[0], &sync_byte, 1), 1);
2640 ASSERT_EQ(write(sv[0], &sync_byte, 1), 1);
2641 close(sv[0]);
2642
2643 /* Wait for child process to exit */
2644 waitpid(pid, &status, 0);
2645 ASSERT_TRUE(WIFEXITED(status));
2646 if (WEXITSTATUS(status) != 0) {
2647 TH_LOG("Child process failed with exit code %d", WEXITSTATUS(status));
2648 SKIP(return, "Child process failed");
2649 }
2650
2651 TH_LOG("Subprocess and all threads have exited successfully");
2652
2653 /* Verify namespaces are now inactive - open_by_handle_at should fail */
2654 TH_LOG("Verifying namespaces are inactive after subprocess and threads exit");
2655 user_fd = open_by_handle_at(FD_NSFS_ROOT, user_handle, O_RDONLY);
2656 ASSERT_LT(user_fd, 0);
2657 TH_LOG("User namespace inactive as expected: %s (errno=%d)",
2658 strerror(errno), errno);
2659 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
2660
2661 net_fd = open_by_handle_at(FD_NSFS_ROOT, net_handle, O_RDONLY);
2662 ASSERT_LT(net_fd, 0);
2663 TH_LOG("Network namespace inactive as expected: %s (errno=%d)",
2664 strerror(errno), errno);
2665 ASSERT_TRUE(errno == ENOENT || errno == ESTALE);
2666
2667 /* Verify namespaces do NOT appear in listns() */
2668 TH_LOG("Verifying namespaces do NOT appear in listns() when inactive");
2669 memset(&req, 0, sizeof(req));
2670 req.size = sizeof(struct ns_id_req);
2671 req.ns_type = CLONE_NEWUSER;
2672 nr_ids = sys_listns(&req, ns_ids, 256, 0);
2673 if (nr_ids >= 0) {
2674 int found_user = 0;
2675 for (int i = 0; i < nr_ids; i++) {
2676 if (ns_ids[i] == user_id) {
2677 found_user = 1;
2678 break;
2679 }
2680 }
2681 ASSERT_FALSE(found_user);
2682 TH_LOG("User namespace correctly not listed in listns()");
2683
2684 /* Check network namespace */
2685 req.ns_type = CLONE_NEWNET;
2686 nr_ids = sys_listns(&req, ns_ids, 256, 0);
2687 if (nr_ids >= 0) {
2688 int found_net = 0;
2689 for (int i = 0; i < nr_ids; i++) {
2690 if (ns_ids[i] == net_id) {
2691 found_net = 1;
2692 break;
2693 }
2694 }
2695 ASSERT_FALSE(found_net);
2696 TH_LOG("Network namespace correctly not listed in listns()");
2697 }
2698 }
2699 }
2700
2701 TEST_HARNESS_MAIN
2702