| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Now that we always require a 64 bit off_t, we no longer need SMB_OFF_T.
Andrew Bartlett
Autobuild-User: Andrew Bartlett <abartlet@samba.org>
Autobuild-Date: Fri Apr 6 01:47:43 CEST 2012 on sn-devel-104
|
|
|
|
leaves data in the pipe on short write.
Fix default_sys_recvfile() to correctly cope with
short writes. Return the amount written. Return
-1 and set errno if no data could be written.
|
|
Linux leaves data in the pipe on short write.
Split out the functionality of drain_socket() into a separate
function from default_sys_recvfile().
|
|
the pipe on short write
Bug found and fix suggested by Andrew Bartlett.
Autobuild-User: Jeremy Allison <jra@samba.org>
Autobuild-Date: Sun Dec 25 07:46:38 CET 2011 on sn-devel-104
|
|
Guenther
|
|
The splice manpage explicitly requires loff_t as offset. Copy "offset" there.
Probably not required, but I wanted to make sure it's as required.
Splice blocks with large buffers. For me it worked with 16k and blocked with
32k and beyond. It would be nice to see a clarification in the manpage of
splice for this behaviour.
Splice if used with an offset increments the offset. From the manpage this was
at least not entirely obvious :-)
I haven't yet activated this (try_splice_call ist still false by default), it
needs more testing.
Volker
|
|
On a debian lenny installation with a kernel that calls itself 2.6.28, I could
repeatably block smbd using splice. According to the output of netstat, the
recvq of the TCP connection was full and at the same time according to strace
smbd was stuck in a splice syscall trying to read from that TCP connection. We
can not protect ourselves from this condition, as nobody will ever notify us
that something is broken. So sadly we can't use splice yet at all. Maybe in
2.6.29...
|
|
According to the splice(2) manpage, one of the file descriptors *must* be a
pipe. With this patch I could successfully run splice(2) on a debian lenny
installation.
Jeremy, please check!
|
|
Jeremy.
(This used to be commit 830337f054a6c0646d85df33d9958e99283e727a)
|
|
so we can detect broken Linux recvfile splice and correctly fall
back.
Jeremy.
(This used to be commit ec2d301a7aac173aba41dd2074037f27d05095ce)
|
|
capability for large UNIX write if not signing and
recvfile set. Cope with large UNIX write length on
incoming processing. Stevef - we can now test 1-16Mb
writes from CIFFS.
Jeremy.
(This used to be commit 8cf78776b0a44bd026cef3d74eb11cfb415f8303)
|
|
Jeremy.
(This used to be commit b7088bb9c2a00d4717b9a7efa4bddc0c005f4efb)
|
|
to zero). If non-zero, writeX calls greater than this
value will be left in the socket buffer for later handling
with recvfile (or userspace equivalent). Definition of
recvfile for your system is left as an exercise for
the reader (I'm working on getting splice working :-).
Jeremy.
(This used to be commit 11c03b75ddbcb6e36b231bb40a1773d1c550621c)
|
|
Jeremy.
(This used to be commit d5886218e53910b7015e6f13bc634e1bf338565a)
|
