As of kernel version 4.18-rc4, Linux has an issue with ext4+DAX where DMA and direct I/O operations aren't synchronized with respect to operations which can change the block mappings of an inode. This can potentially lead to data corruption, and is fixed in the following kernel series: https://lists.01.org/pipermail/linux-nvdimm/2018-July/016842.html These patches add a regression test for that series. --- Changes since v1: * Added copyright and license information to files I've previously authored in src/. * Added copyright, license information and info about the test's purpose to src/t_mmap_collision.c. (Dave) * Added comments to tests/generic/999 explaining why we unset the mount flags for SCRATCH_MNT. (Dave) * Subtests in src/t_mmap_collision.c are now conditional based on the existence of the necessary FALLOC_FL_* flags. (Eryu) * Added the test to the punch, collapse and zero groups. (Eryu) Ross Zwisler (2): src/: add license and copyright info to files generic/999: test DAX DMA vs truncate/hole-punch .gitignore | 1 + src/Makefile | 2 +- src/t_ext4_dax_inline_corruption.c | 2 + src/t_ext4_dax_journal_corruption.c | 2 + src/t_mmap_collision.c | 235 ++++++++++++++++++++++++++++++++++++ src/t_mmap_cow_race.c | 2 + src/t_mmap_stale_pmd.c | 2 + tests/generic/999 | 53 ++++++++ tests/generic/999.out | 2 + tests/generic/group | 1 + 10 files changed, 301 insertions(+), 1 deletion(-) create mode 100644 src/t_mmap_collision.c create mode 100755 tests/generic/999 create mode 100644 tests/generic/999.out -- 2.14.4
