Re: [RFC PATCH 00/31] netfs: [WIP] Allow the use of MSG_SPLICE_PAGES and use netmem allocator

[David Howells <dhowells@redhat.com>]

Enzo Matsumiya <ematsumiya@suse.de> wrote:

> Anyway, if you want me to test, just send me the patches.
> I have your linux-fs remote as well, if that's easier.

If you look at:

https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=cifs-experimental

You can see a patch with the subject "cifs: [!] FIX transport compression".
Grab that and everything up to it.

I'm pretty certain it won't work, but I can't test it.  Well, maybe I can
force it on and look at the packet trace if wireshark can handle it.

Some things to note:

 (1) In smb_compress(), netfs_alloc_bvecq_buffer() is used to allocate the
     destination buffer and attach it to a bvecq, which it also allocates.  An
     iterator can be set on this to define part of the buffer to operate on.

 (2) vmap_bvecq() is used to map the source and the destination buffers.  It
     extracts the pages and then calls vmap() on them.

 (3) Space for the compression header is allocated by smb_send_rqst() in the
     first bvecq slot along with the rfc1002 header and transform header (if
     sealing).  A pointer is passed down to smb_compress().

 (4) It attempts to adjust the values such that the compression header is
     included in the encrypted section if also sealing.  However, it might be
     better to have smb_compress() place it in the output buffer.

 (5) If compression is successful, smb_compress() switches the original bvecq
     and the output bvecq and moves the header segment from the original to
     the output.

 (6) If the compression algo returns -EMSGSIZE, then the compression header is
     excluded from the header segment.

David