On 6/7/24 10:50 AM, Donet Tom wrote:
> 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
>
> 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>
> ---
> tools/testing/selftests/mm/Makefile | 1 +
> tools/testing/selftests/mm/hugetlb_dio.c | 118 +++++++++++++++++++++++
Missed my feedback on adding the test to vm_test.sh
Other than this, LGTM. Please add following tag after the above change:
Reviewed-by: Muhammad Usama Anjum <usama.anjum@xxxxxxxxxxxxx>
> 2 files changed, 119 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();
> +}
> +
--
BR,
Muhammad Usama Anjum
