Commit 1b151e2435fc ("block: Remove special-casing of compound
pages") caused a change in behaviour when releasing the pages
if the buffer does not start at the beginning of the page. This
was because the calculation of the number of pages to release
was incorrect.
This was fixed by commit 38b43539d64b ("block: Fix page refcounts
for unaligned buffers in __bio_release_pages()").
We pin the user buffer during direct I/O writes. If this buffer is a
hugepage, bio_release_page() will unpin it and decrement all references
and pin counts at ->bi_end_io. However, if any references to the hugepage
remain post-I/O, the hugepage will not be freed upon unmap, leading
to a memory leak.
This patch verifies that a hugepage, used as a user buffer for DIO
operations, is correctly freed upon unmapping, regardless of whether
the offsets are aligned or unaligned w.r.t page boundary.
Test Result Fail Scenario (Without the fix)
--------------------------------------------------------
[]# ./hugetlb_dio
TAP version 13
1..4
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 1 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 2 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 3 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 6
not ok 4 : Huge pages not freed!
Totals: pass:3 fail:1 xfail:0 xpass:0 skip:0 error:0
Test Result PASS Scenario (With the fix)
---------------------------------------------------------
[]#./hugetlb_dio
TAP version 13
1..4
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 1 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 2 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 3 : Huge pages freed successfully !
No. Free pages before allocation : 7
No. Free pages after munmap : 7
ok 4 : Huge pages freed successfully !
Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0
V4:
- Added this test to run_vmtests.sh.
V3:
- Fixed the build error when it is compiled with _FORTIFY_SOURCE.
V2:
- Addressed all review commets from Muhammad Usama Anjum
https://lore.kernel.org/all/20240604132801.23377-1-donettom@xxxxxxxxxxxxx/
V1:
https://lore.kernel.org/all/20240523063905.3173-1-donettom@xxxxxxxxxxxxx/#t
Signed-off-by: Donet Tom <donettom@xxxxxxxxxxxxx>
Co-developed-by: Ritesh Harjani (IBM) <ritesh.list@xxxxxxxxx>
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@xxxxxxxxx>
Reviewed-by: Muhammad Usama Anjum <usama.anjum@xxxxxxxxxxxxx>
---
tools/testing/selftests/mm/Makefile | 1 +
tools/testing/selftests/mm/hugetlb_dio.c | 118 ++++++++++++++++++++++
tools/testing/selftests/mm/run_vmtests.sh | 1 +
3 files changed, 120 insertions(+)
create mode 100644 tools/testing/selftests/mm/hugetlb_dio.c
diff --git a/tools/testing/selftests/mm/Makefile b/tools/testing/selftests/mm/Makefile
index 3b49bc3d0a3b..a1748a4c7df1 100644
--- a/tools/testing/selftests/mm/Makefile
+++ b/tools/testing/selftests/mm/Makefile
@@ -73,6 +73,7 @@ TEST_GEN_FILES += ksm_functional_tests
TEST_GEN_FILES += mdwe_test
TEST_GEN_FILES += hugetlb_fault_after_madv
TEST_GEN_FILES += hugetlb_madv_vs_map
+TEST_GEN_FILES += hugetlb_dio
ifneq ($(ARCH),arm64)
TEST_GEN_FILES += soft-dirty
diff --git a/tools/testing/selftests/mm/hugetlb_dio.c b/tools/testing/selftests/mm/hugetlb_dio.c
new file mode 100644
index 000000000000..986f3b6c7f7b
--- /dev/null
+++ b/tools/testing/selftests/mm/hugetlb_dio.c
@@ -0,0 +1,118 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This program tests for hugepage leaks after DIO writes to a file using a
+ * hugepage as the user buffer. During DIO, the user buffer is pinned and
+ * should be properly unpinned upon completion. This patch verifies that the
+ * kernel correctly unpins the buffer at DIO completion for both aligned and
+ * unaligned user buffer offsets (w.r.t page boundary), ensuring the hugepage
+ * is freed upon unmapping.
+ */
+
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <sys/stat.h>
+#include <stdlib.h>
+#include <fcntl.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/mman.h>
+#include "vm_util.h"
+#include "../kselftest.h"
+
+void run_dio_using_hugetlb(unsigned int start_off, unsigned int end_off)
+{
+ int fd;
+ char *buffer = NULL;
+ char *orig_buffer = NULL;
+ size_t h_pagesize = 0;
+ size_t writesize;
+ int free_hpage_b = 0;
+ int free_hpage_a = 0;
+ const int mmap_flags = MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB;
+ const int mmap_prot = PROT_READ | PROT_WRITE;
+
+ writesize = end_off - start_off;
+
+ /* Get the default huge page size */
+ h_pagesize = default_huge_page_size();
+ if (!h_pagesize)
+ ksft_exit_fail_msg("Unable to determine huge page size\n");
+
+ /* Open the file to DIO */
+ fd = open("/tmp", O_TMPFILE | O_RDWR | O_DIRECT, 0664);
+ if (fd < 0)
+ ksft_exit_fail_perror("Error opening file\n");
+
+ /* Get the free huge pages before allocation */
+ free_hpage_b = get_free_hugepages();
+ if (free_hpage_b == 0) {
+ close(fd);
+ ksft_exit_skip("No free hugepage, exiting!\n");
+ }
+
+ /* Allocate a hugetlb page */
+ orig_buffer = mmap(NULL, h_pagesize, mmap_prot, mmap_flags, -1, 0);
+ if (orig_buffer == MAP_FAILED) {
+ close(fd);
+ ksft_exit_fail_perror("Error mapping memory\n");
+ }
+ buffer = orig_buffer;
+ buffer += start_off;
+
+ memset(buffer, 'A', writesize);
+
+ /* Write the buffer to the file */
+ if (write(fd, buffer, writesize) != (writesize)) {
+ munmap(orig_buffer, h_pagesize);
+ close(fd);
+ ksft_exit_fail_perror("Error writing to file\n");
+ }
+
+ /* unmap the huge page */
+ munmap(orig_buffer, h_pagesize);
+ close(fd);
+
+ /* Get the free huge pages after unmap*/
+ free_hpage_a = get_free_hugepages();
+
+ /*
+ * If the no. of free hugepages before allocation and after unmap does
+ * not match - that means there could still be a page which is pinned.
+ */
+ if (free_hpage_a != free_hpage_b) {
+ ksft_print_msg("No. Free pages before allocation : %d\n", free_hpage_b);
+ ksft_print_msg("No. Free pages after munmap : %d\n", free_hpage_a);
+ ksft_test_result_fail(": Huge pages not freed!\n");
+ } else {
+ ksft_print_msg("No. Free pages before allocation : %d\n", free_hpage_b);
+ ksft_print_msg("No. Free pages after munmap : %d\n", free_hpage_a);
+ ksft_test_result_pass(": Huge pages freed successfully !\n");
+ }
+}
+
+int main(void)
+{
+ size_t pagesize = 0;
+
+ ksft_print_header();
+ ksft_set_plan(4);
+
+ /* Get base page size */
+ pagesize = psize();
+
+ /* start and end is aligned to pagesize */
+ run_dio_using_hugetlb(0, (pagesize * 3));
+
+ /* start is aligned but end is not aligned */
+ run_dio_using_hugetlb(0, (pagesize * 3) - (pagesize / 2));
+
+ /* start is unaligned and end is aligned */
+ run_dio_using_hugetlb(pagesize / 2, (pagesize * 3));
+
+ /* both start and end are unaligned */
+ run_dio_using_hugetlb(pagesize / 2, (pagesize * 3) + (pagesize / 2));
+
+ ksft_finished();
+}
+
diff --git a/tools/testing/selftests/mm/run_vmtests.sh b/tools/testing/selftests/mm/run_vmtests.sh
index 3157204b9047..5698d519170d 100755
--- a/tools/testing/selftests/mm/run_vmtests.sh
+++ b/tools/testing/selftests/mm/run_vmtests.sh
@@ -265,6 +265,7 @@ CATEGORY="hugetlb" run_test ./map_hugetlb
CATEGORY="hugetlb" run_test ./hugepage-mremap
CATEGORY="hugetlb" run_test ./hugepage-vmemmap
CATEGORY="hugetlb" run_test ./hugetlb-madvise
+CATEGORY="hugetlb" run_test ./hugetlb_dio
nr_hugepages_tmp=$(cat /proc/sys/vm/nr_hugepages)
# For this test, we need one and just one huge page
--
2.43.0
