[Linux 5.18+] Fix detection of filemap_range_has_page

[rrevans]

Motivation and Context

While testing #16031, I noted that HAVE_FILEMAP_RANGE_HAS_PAGE is not being defined for Linux 6.5 or 6.6.

Turns out the symbol moved in v5.18 from fs.h to pagemap.h, and the fallback path has been in use for all versions of Linux since then.

For the history of the symbol, run from linux git repo:

$ git grep -e filemap_range_has_page $(git rev-parse --tags=v*) -- include/linux/pagemap.h | git name-rev --annotate-stdin --tags

I also verified this by compiling the test src at all major releases from v5.18 ... v6.8 and v6.9-rc1

$ cat /tmp/x.c
#include <linux/module.h>
#include <linux/fs.h>

int main(void) {
                struct address_space *mapping = NULL;
                loff_t lstart = 0;
                loff_t lend = 0;
                bool ret __attribute__ ((unused));

                ret = filemap_range_has_page(mapping, lstart, lend);
                return 0;
}
$ gcc -Wp,-MMD,arch/x86/realmode/.init.o.d -nostdinc -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/compiler-version.h -include ./include/linux/kconfig.h -include ./include/linux/compiler_types.h -D__KERNEL__ -fmacro-prefix-map=./= -Wall -Wundef -Werror=strict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -fshort-wchar -fno-PIE -Werror=implicit-function-declaration -std=gnu11 -m64 -c -o /tmp/x.o /tmp/x.c
... pass for v5.17, fail for all versions after ...

In the fallback path, ZFS calls mappedread and update_pages for every read or write (respectively) after the first mmap call to the file. This is unnecessary for unmapped ranges and has a slight performance cost.

Edit: The cost maybe isn't so slight. The fallback code within mappedread only reads in page-sized increments which is somewhat unfortunate. See #16031 (comment)

Description

In v5.18 filemap_range_has_page moved to pagemap.h

pagemap.h has been around since 3.10 so just include both

How Has This Been Tested?

Local ZTS passes (linux.run, common.run, and sanity.run) on 6.6.8-200.fc39.x86_64 w/ debug build

This is not without risk since the non-fallback path has been disabled for a long while

I don't have a test setup that can build against older kernels.

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.

[robn]

Well that stinks. Good find.

[behlendorf]

Well that's unfortunate, thanks for the fix.

[ssergiienko]

Given the noticeable performance improvement and visual safety of the change, wouldn't it be nice to cherry-pick it to v2.2.4?

[rrevans]

Seems reasonable to backport with the caveat that this code path hasn't been exercised in recent kernels.