Add more tests for userfaultfd #1280

ricardobranco777 · 2025-12-07T12:02:42Z

Add more tests for userfaultfd:

UFFDIO_MOVE
UFFDIO_WRITEPROTECT
UFFDIO_ZEROPAGE
UFFD_USER_MODE_ONLY
Use /dev/userfaultfd instead of system call

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c

metan-ucw

Looks good, but there is too much duplicated code (should be mostly put into a common header) and some minor errors as well.

metan-ucw · 2025-12-17T12:03:15Z

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c

+	SAFE_IOCTL(uffd, UFFDIO_REGISTER, &uffdio_register);
+
+	SAFE_PTHREAD_CREATE(&thr, NULL,
+			(void * (*)(void *)) handle_thread, NULL);


This can be just (void*) cast, but it would be better if the handle_thread function was defined as void * and returned NULL at the end.

metan-ucw · 2025-12-17T12:05:31Z

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c

+			tst_brk(TBROK | TTERRNO,
+				"Could not create userfault file descriptor");
+	}
+


The sys_userfaultfd() should be in the test setup() so that things work properly when test is executed with -1 1000 parameter.

Also this pattern is repeated in all tests, so this piece of code should be put into a header as a static inline function and called from all tests that needs it.

metan-ucw · 2025-12-17T12:10:08Z

testcases/kernel/syscalls/userfaultfd/userfaultfd03.c

+static struct tst_test test = {
+	.test_all = run,
+	.min_kver = "5.11",
+};


This is nearly the same as the userfaultfd01, it just adds the UFFD_USER_MODE_ONLY flag. It would be far better if we added .tcnt = 2 to the usefaultfd02.c and add the flag to the uffdio_register.mode only when n == 1 in order to avoid copy & paste of basically the same test.

The .min_kver is different though.

metan-ucw · 2025-12-17T12:14:45Z

testcases/kernel/syscalls/userfaultfd/userfaultfd04.c

+
+/*\
+ * Force a pagefault event and handle it using :man2:`userfaultfd`
+ * from a different thread using /dev/userfaultfd instead of syscall.


You should mention the details here, in this case we do UFFDIO_COPY and use file descriptor created by USERFAULTFD_IOC_NEW

metan-ucw · 2025-12-17T12:15:42Z

testcases/kernel/syscalls/userfaultfd/userfaultfd05.c

+	SAFE_PTHREAD_CREATE(&thr, NULL,
+			(void * (*)(void *)) handle_thread, NULL);
+
+	(void)page[0xf];


I do not think that we need this here, we are checking the page in the loop below, which will trigger read access anyways.

metan-ucw · 2025-12-17T12:19:19Z

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c

+	pollfd.events = POLLIN;
+	nready = poll(&pollfd, 1, -1);
+	if (nready == -1)
+		tst_brk(TBROK | TERRNO, "Error on poll");


Do we need the poll() here? We are waiting indefinitely anyways so as long as uffd is not O_NONBLOCK (which should be default) we can just remove this and call the SAFE_READ() which will wait for the data.

Do we need the poll() here? We are waiting indefinitely anyways so as long as uffd is not O_NONBLOCK (which should be default) we can just remove this and call the SAFE_READ() which will wait for the data.

Technically no but uffd's are universally opened in non-blocking mode to use with poll as it seems it's the only way to catch some errors:

https://docs.kernel.org/admin-guide/mm/userfaultfd.html

Be sure to test for all errors including (pollfd[0].revents & POLLERR). This can happen, e.g. when ranges supplied were incorrect.

metan-ucw · 2025-12-17T12:20:50Z

testcases/kernel/syscalls/userfaultfd/userfaultfd05.c

+	uffdio_zeropage.range.len	= page_size;
+	uffdio_zeropage.mode		= 0;
+
+	SAFE_IOCTL(uffd, UFFDIO_ZEROPAGE, &uffdio_zeropage);


I guess that we need a test where we will attempt to map a ZEROPAGE for a write access, I suppose that we will get error from this ioctl()

metan-ucw · 2025-12-17T12:21:12Z

testcases/kernel/syscalls/userfaultfd/userfaultfd05.c

+	(void)page[0xf];
+
+	for (int i = 0; i < page_size; i++) {
+		if (page[i] != '\0') {


just use 0 instead of '\0'

metan-ucw · 2025-12-17T12:38:01Z

testcases/kernel/syscalls/userfaultfd/userfaultfd06.c

+	SAFE_READ(1, uffd, &msg, sizeof(msg));
+
+	if (msg.event != UFFD_EVENT_PAGEFAULT)
+		tst_brk(TBROK | TERRNO, "Received unexpected UFFD_EVENT");


That's TFAIL

metan-ucw · 2025-12-17T12:38:11Z

testcases/kernel/syscalls/userfaultfd/userfaultfd06.c

+
+	if (!(msg.arg.pagefault.flags & UFFD_PAGEFAULT_FLAG_WP) ||
+	    !(msg.arg.pagefault.flags & UFFD_PAGEFAULT_FLAG_WRITE)) {
+		tst_brk(TBROK,


And here as well TFAIL

Signed-off-by: Ricardo Branco <rbranco@suse.de>

ricardobranco777 · 2025-12-19T22:20:27Z

Looks good, but there is too much duplicated code (should be mostly put into a common header) and some minor errors as well.

Thanks for reviewing. I addressed most of your suggestions. Left some questions regarding further refactoring.

ricardobranco777 force-pushed the userfaultfd branch 8 times, most recently from 45b04e6 to bffe686 Compare December 8, 2025 11:38

pevik reviewed Dec 8, 2025

View reviewed changes

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c Outdated Show resolved Hide resolved

pevik reviewed Dec 8, 2025

View reviewed changes

testcases/kernel/syscalls/userfaultfd/userfaultfd02.c Outdated Show resolved Hide resolved

pevik mentioned this pull request Dec 8, 2025

Write test for userfaultfd() #296

Open

ricardobranco777 force-pushed the userfaultfd branch from bffe686 to 9da6afe Compare December 8, 2025 19:27

metan-ucw requested changes Dec 17, 2025

View reviewed changes

userfaultfd: Add test for UFFDIO_MOVE

3b6189b

Signed-off-by: Ricardo Branco <rbranco@suse.de>

ricardobranco777 force-pushed the userfaultfd branch from 9da6afe to 60ef374 Compare December 19, 2025 10:35

ricardobranco777 added 6 commits December 19, 2025 15:32

userfaultfd: Minor fixes for userfaultfd01.c

dce50b9

Signed-off-by: Ricardo Branco <rbranco@suse.de>

userfaultfd: Put sys_userfaultfd in separate header

784a17f

Signed-off-by: Ricardo Branco <rbranco@suse.de>

userfaultfd: Add test using UFFD_USER_MODE_ONLY

3a4ce37

Signed-off-by: Ricardo Branco <rbranco@suse.de>

userfaultfd: Add test using /dev/userfaultfd instead of syscall

7f08651

Signed-off-by: Ricardo Branco <rbranco@suse.de>

userfaultfd: Add test using UFFDIO_ZEROPAGE

cee17c0

Signed-off-by: Ricardo Branco <rbranco@suse.de>

userfaultfd: Add test using UFFDIO_WRITEPROTECT

05b2ba2

Signed-off-by: Ricardo Branco <rbranco@suse.de>

ricardobranco777 force-pushed the userfaultfd branch from 60ef374 to 05b2ba2 Compare December 19, 2025 14:32

wangli5665 force-pushed the master branch from 1785a68 to 5aa4a4f Compare December 20, 2025 02:29

Add more tests for userfaultfd #1280

Are you sure you want to change the base?

Add more tests for userfaultfd #1280

Conversation

ricardobranco777 commented Dec 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

metan-ucw left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ricardobranco777 commented Dec 19, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ricardobranco777 commented Dec 7, 2025 •

edited

Loading