Linux 6.5: Replace generic_file_splice_read with filemap_splice_read

[ckane]

The generic_file_splice_read function was removed in Linux 6.5 in favor of filemap_splice_read. Add an autoconf test for filemap_splice_read and use it if it is found as the handler for .splice_read in the file_operations struct. Additionally, ITER_PIPE was removed in 6.5. This change removes the ITER_* macros that OpenZFS doesn't use from being tested in config/kernel-vfs-iov_iter.m4. The removal of ITER_PIPE was causing the test to fail, which also affected the code responsible for setting the .splice_read handler, above. That behavior caused run-time panics on Linux 6.5.

Motivation and Context

When running ZFS on 6.5rc*, a run-time panic occurs with the following stack trace:

[ 2417.825843]  ? __die+0x23/0x70
[ 2417.825974]  ? page_fault_oops+0x171/0x4e0
[ 2417.826107]  ? exc_page_fault+0x7f/0x180
[ 2417.826239]  ? asm_exc_page_fault+0x26/0x30
[ 2417.826372]  ? memcpy_orig+0x78/0x140
[ 2417.826513]  zfs_uiomove_iov+0x58/0x220 [zfs 4a5a286022a33f632f1f726c7834b01d83a9bf66]
[ 2417.826759]  dmu_read_uio_dnode+0xd3/0x150 [zfs 4a5a286022a33f632f1f726c7834b01d83a9bf66]
[ 2417.826988]  dmu_read_uio_dbuf+0x46/0x60 [zfs 4a5a286022a33f632f1f726c7834b01d83a9bf66]
[ 2417.827226]  zfs_read+0x139/0x3e0 [zfs 4a5a286022a33f632f1f726c7834b01d83a9bf66]
[ 2417.827460]  zpl_iter_read+0x116/0x1d0 [zfs 4a5a286022a33f632f1f726c7834b01d83a9bf66]
[ 2417.827689]  vfs_read+0x201/0x350
[ 2417.827839]  __x64_sys_pread64+0x98/0xd0
[ 2417.827993]  do_syscall_64+0x60/0x90
[ 2417.828154]  ? exc_page_fault+0x7f/0x180
[ 2417.828307]  entry_SYSCALL_64_after_hwframe+0x6e/0xd8

It turns out that ITER_PIPE was removed in 6.5, and this in turn causes one of the tests setting HAVE_VFS_IOV_ITER to result in a false-negative during configure. This results in not assigning the .splice_read member of struct file_operations in module/os/linux/zfs/zpl_file.c, which was likely the cause of the memory access error above. Additionally, this struct member has been set to generic_file_splice_read up until now, but that function has also been removed in Linux 6.5. It has been superseded by filemap_splice_read. This additional regression was masked by the configure test failure due to missing ITER_PIPE.

Description

The only ITER_* macro used by OpenZFS is ITER_KVEC, so instead of trying to set an int with all possible bits set in the test, I changed it to only try setting the ITER_KVEC bit, and succeed if that works. In the future, if the remaining ones (or any future ones) are used by the code, they can be added to this test in config/kernel-vfs-iov_iter.m4. As far as I can tell, we shouldn't be failing the build if macros we don't use are removed in the kernel.

Add a new configure test for the presence of filemap_splice_read and ensure that it can be assigned to file_operations.splice_read without any type conflicts. This tests for the existence of the function, and also ensures that its type definition matches the handler in struct file_operations, the context in which we rely upon it.

Finally, add a #ifdef in the module/os/linux/zfs/zpl_file.c code that will assign file_operations.splice_read to filemap_splice_read if the latter is found by configure.

How Has This Been Tested?

Build-tested and Run-tested on Linux 6.5rc1

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Library ABI change (libzfs, libzfs_core, libnvpair, libuutil and libzfsbootenv)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[ckane]

Alternatively, this could be written to test for generic_file_splice_read() and prefer to use it where it exists. This would minimize the behavior drift across kernels where both of these functions exist. The filemap_splice_read() was added in Feb 2023, so this probably will affect 6.4 kernels, and maybe some 6.3 kernels, as well. I'll let the CI/CD tests run and if it looks like this is causing failures on older kernels, then I'll go ahead and change the logic to favor the removed function where it is found, rather than favor its replacement.

[bwatkinson]

Can you look at the comment in zfs_uiomove_iter() in module/os/linux/zfs/zfs_uio.c about returning the value of EFAULT as it will be converted to EAGAIN by generic_file_splice_read()?

When I was looking at filemap_splice_read() I noticed it just returned the error without converting the error of EFAULT. We may need to also update this part of the code with the check to just return EAGAIN? Not sure though without spending more time looking at what error should be returned from zfs_uiomove_iter() in this case.

[behlendorf]

It looks to me like filemap_splice_read() will handle this correctly.