Re: [RFC v2] copy_file_range.2: Update cross-filesystem support for 5.12

[Amir Goldstein]

On Sat, Feb 27, 2021 at 3:59 PM Alejandro Colomar
<alx.manpages@xxxxxxxxx> wrote:
>
> Linux 5.12 fixes a regression.
>
> Cross-filesystem copies (introduced in 5.3) were buggy.
>
> Move the statements documenting cross-fs to BUGS.
> Kernels 5.3..5.11 should be patched soon.
>
> State version information for some errors related to this.
>
> Reported-by: Luis Henriques <lhenriques@xxxxxxx>
> Reported-by: Amir Goldstein <amir73il@xxxxxxxxx>
> Related: <https://lwn.net/Articles/846403/>
> Cc: Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx>
> Cc: Michael Kerrisk <mtk.manpages@xxxxxxxxx>
> Cc: Anna Schumaker <anna.schumaker@xxxxxxxxxx>
> Cc: Jeff Layton <jlayton@xxxxxxxxxx>
> Cc: Steve French <sfrench@xxxxxxxxx>
> Cc: Miklos Szeredi <miklos@xxxxxxxxxx>
> Cc: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
> Cc: "Darrick J. Wong" <darrick.wong@xxxxxxxxxx>
> Cc: Dave Chinner <dchinner@xxxxxxxxxx>
> Cc: Nicolas Boichat <drinkcat@xxxxxxxxxxxx>
> Cc: Ian Lance Taylor <iant@xxxxxxxxxx>
> Cc: Luis Lozano <llozano@xxxxxxxxxxxx>
> Cc: Andreas Dilger <adilger@xxxxxxxxx>
> Cc: Olga Kornievskaia <aglo@xxxxxxxxx>
> Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>
> Cc: ceph-devel <ceph-devel@xxxxxxxxxxxxxxx>
> Cc: linux-kernel <linux-kernel@xxxxxxxxxxxxxxx>
> Cc: CIFS <linux-cifs@xxxxxxxxxxxxxxx>
> Cc: samba-technical <samba-technical@xxxxxxxxxxxxxxx>
> Cc: linux-fsdevel <linux-fsdevel@xxxxxxxxxxxxxxx>
> Cc: Linux NFS Mailing List <linux-nfs@xxxxxxxxxxxxxxx>
> Cc: Walter Harms <wharms@xxxxxx>
> Signed-off-by: Alejandro Colomar <alx.manpages@xxxxxxxxx>
> ---
>
> Hi all,
>
> Please check that this is correct.
> I wrote it as I understood copy_file_range() from the LWN article,
> and the conversation on this thread,
> but maybe someone with more experience on this syscall find bugs in my patch.
>
> When kernels 5.3..5.11 fix this, some info could be compacted a bit more,
> and maybe the BUGS section could be removed.
>
> Also, I'd like to know which filesystems support cross-fs, and since when.
>
> Amir, you said that it was only cifs and nfs (since when? 5.3? 5.12?).
>
> Also, I'm a bit surprised that <5.3 could fail with EOPNOTSUPP
> and it wasn't documented.  Is that for sure, Amir?

No. You are right. EOPNOTSUPP is new.
Kernel always fell back to sendfile(2) if the filesystem did not support
copy_file_range().

>
> Thanks,
>
> Alex
>
> ---
>  man2/copy_file_range.2 | 29 ++++++++++++++++++++---------
>  1 file changed, 20 insertions(+), 9 deletions(-)
>
> diff --git a/man2/copy_file_range.2 b/man2/copy_file_range.2
> index 611a39b80..93f54889d 100644
> --- a/man2/copy_file_range.2
> +++ b/man2/copy_file_range.2
> @@ -169,6 +169,9 @@ Out of memory.
>  .B ENOSPC
>  There is not enough space on the target filesystem to complete the copy.
>  .TP
> +.BR EOPNOTSUPP " (before Linux 5.3; or since Linux 5.12)"
> +The filesystem does not support this operation.
> +.TP

so not before 5.3

>  .B EOVERFLOW
>  The requested source or destination range is too large to represent in the
>  specified data types.
> @@ -184,10 +187,17 @@ or
>  .I fd_out
>  refers to an active swap file.
>  .TP
> -.B EXDEV
> +.BR EXDEV " (before Linux 5.3)"
>  The files referred to by
>  .IR fd_in " and " fd_out
> -are not on the same mounted filesystem (pre Linux 5.3).
> +are not on the same filesystem.
> +.TP
> +.BR EXDEV " (or since Linux 5.12)"
> +The files referred to by
> +.IR fd_in " and " fd_out
> +are not on the same filesystem,
> +and the source and target filesystems are not of the same type,
> +or do not support cross-filesystem copy.

ok.

>  .SH VERSIONS
>  The
>  .BR copy_file_range ()
> @@ -195,13 +205,10 @@ system call first appeared in Linux 4.5, but glibc 2.27 provides a user-space
>  emulation when it is not available.
>  .\" https://sourceware.org/git/?p=glibc.git;a=commit;f=posix/unistd.h;h=bad7a0c81f501fbbcc79af9eaa4b8254441c4a1f
>  .PP
> -A major rework of the kernel implementation occurred in 5.3.
> -Areas of the API that weren't clearly defined were clarified and the API bounds
> -are much more strictly checked than on earlier kernels.
> -Applications should target the behaviour and requirements of 5.3 kernels.
> -.PP

That information is useful. Why remove it?
FYI, the LTP tests written to velidate the copy_file_range() API are not running
on kernel < 5.3 at all.

> -First support for cross-filesystem copies was introduced in Linux 5.3.
> -Older kernels will return -EXDEV when cross-filesystem copies are attempted.
> +Since 5.12,
> +cross-filesystem copies can be achieved
> +when both filesystems are of the same type,
> +and that filesystem implements support for it.
>  .SH CONFORMING TO
>  The
>  .BR copy_file_range ()
> @@ -226,6 +233,10 @@ gives filesystems an opportunity to implement "copy acceleration" techniques,
>  such as the use of reflinks (i.e., two or more inodes that share
>  pointers to the same copy-on-write disk blocks)
>  or server-side-copy (in the case of NFS).
> +.SH BUGS
> +In Linux kernels 5.3 to 5.11, cross-filesystem copies were supported.

I think it is a bit confusing to say "were supported", because how come
support went away from kernel 5.12? maybe something along the lines
that kernel implementation of copy was used if there was no filesystem
support for the operation...

> +However, on some virtual filesystems, the call failed to copy,
> +eventhough it may have reported success.
>  .SH EXAMPLES
>  .EX
>  #define _GNU_SOURCE
> --
> 2.30.1.721.g45526154a5
>