[0/2] Fix fallocate error in hugetlbfs when fallocating again

Message

Ackerley Tng May 2, 2023, 11:56 p.m. UTC

When fallocate() is used twice on the same offset, it should succeed
and return 0 to userspace. The second fallocate() in
test_multiple_fallocates() in the following code will return -EEXIST
for hugetlbfs, but not tmpfs/shmem.

#define _GNU_SOURCE
#include <fcntl.h>
#include <linux/falloc.h>
#include <linux/memfd.h>
#include <sys/mman.h>
#include <stdio.h>
#include <unistd.h>

void test_multiple_fallocates(unsigned int flags)
{
	int fd = memfd_create("foo", flags);

	if (fallocate(fd, 0, 0, 0x1000))
		printf("error with fallocate %m\n");

	if (fallocate(fd, 0, 0, 0x1000))
		printf("error with second fallocate %m\n");
}

int main(void) {

	printf("Testing tmpfs:\n");
	test_multiple_fallocates(0);

	printf("Testing hugetlbfs:\n");
	test_multiple_fallocates(MFD_HUGETLB | MFD_HUGE_2MB);

}

This patch series replaces page_cache_next_miss(), used to determine a
page cache hit, with a more direct filemap_has_folio() (a new
function).

I hope that this is also the desired refactoring as mentioned in [1].

[1] https://lore.kernel.org/all/[e-mail]/

---

Ackerley Tng (2):
  mm: filemap: Add filemap_has_folio function
  fs: hugetlbfs: Fix logic to skip allocation on hit in page cache

 fs/hugetlbfs/inode.c    |  6 +-----
 include/linux/pagemap.h |  1 +
 mm/filemap.c            | 17 +++++++++++++++++
 3 files changed, 19 insertions(+), 5 deletions(-)

--
2.40.1.495.gc816e09b53d-goog