solaris librados port #5909
solaris librados port #5909
Conversation
| @@ -2123,16 +2198,27 @@ int Pipe::do_sendmsg(struct msghdr *msg, int len, bool more) | |||
| assert(l == len); | |||
| } | |||
|
|
|||
| int r = ::sendmsg(sd, msg, MSG_NOSIGNAL | (more ? MSG_MORE : 0)); | |||
There was a problem hiding this comment.
Hmm, I guess we could pick up some ways to kick out MSG_NOSIGNAL? @liewegas
|
Maybe async messenger need to impl evport for event methods |
| -D_PTHREADS \ | ||
| -D_POSIX_C_SOURCE | ||
|
|
||
| endif |
There was a problem hiding this comment.
this may break compilation on OSX and my pending FreeBSD port. please avoid doing this
There was a problem hiding this comment.
Is -D_REENTRANT required for threading for OSX and FreeBSD? If so I'll add those.
There was a problem hiding this comment.
guess PTHREAD_CFLAGS offered by ACX_PTHREAD would suffice? and we already have it. and the interesting thing is what i found in m4/acx_pthread.m4 and
case "${host_cpu}-${host_os}" in
*-aix* | *-freebsd* | *-darwin*) flag="-D_THREAD_SAFE";;
*solaris* | *-osf* | *-hpux*) flag="-D_REENTRANT";;
esachttp://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m4/ax_pthread.m4 shows something different, but similar
AC_MSG_CHECKING([if more special flags are required for pthreads])
flag=no
case ${host_os} in
aix* | freebsd* | darwin*) flag="-D_THREAD_SAFE";;
osf* | hpux*) flag="-D_REENTRANT";;
solaris*)
if test "$GCC" = "yes"; then
flag="-D_REENTRANT"
else
# TODO: What about Clang on Solaris?
flag="-mt -D_REENTRANT"
fi
;;
esacThere was a problem hiding this comment.
It gets complicated: http://docs.oracle.com/cd/E19455-01/806-5257/6je9h033k/index.html
I chose the pure posix approach. I didn't choose the mixed mode because of compile problems in the boost headers.
Signed-off-by: Rohan Mars <code@rohanmars.com>
rohanmars
force-pushed
the
wip-solaris-port
branch
from
03e01ef
to
8680c4d
Compare
|
got following errors on OSX. msg/simple/Pipe.cc:2249:3: error: use of undeclared identifier 'restore_sigpipe' I apply your patch on top of my RP #5826 |
Signed-off-by: Rohan Mars <code@rohanmars.com>
|
@ukernel that needed a SO_NOSIGPIPE guard fix. I've just added that to the pull request. thankyou for testing this on OSX! |
| // converts from linux errno values to host values | ||
| __s32 translate_errno(CephContext *cct,__s32 r) | ||
| { | ||
| if (r < -34) { |
There was a problem hiding this comment.
I think a cleaner approach here would be to have a generic
translate_errno()
method that is a #define on Linux but does this conversoin on solaris. Then change all of the return sites in librados.cc to pass the return value through. For example,
https://github.com/ceph/ceph/blob/master/src/librados/librados.cc#L1149
would do
return translate_errno(io_ctx_impl->sparse_read(oid, m, bl, len, off));
There was a problem hiding this comment.
@liewegas I see pros/cons with both approaches. How will native errors from local systems calls be differentiated using this approach? Using the translation at the unmarshalling layer means that errors from local calls won't be mapped (which is what we want I'm thinking).
There was a problem hiding this comment.
On Tue, 27 Oct 2015, rohanmars wrote:
In src/common/solaris_errno.cc:
+#include "messages/MRemoveSnaps.h"
+#include "messages/MRoute.h"
+#include "messages/MStatfs.h"
+#include "messages/MStatfsReply.h"
+#include "messages/MTimeCheck.h"
+#include "messages/MWatchNotify.h"
+
+#define dout_subsys ceph_subsys_ms
+
+#include "solaris_errno.h"
+#include "solaris_message_errno.h"
+
+// converts from linux errno values to host values
+__s32 translate_errno(CephContext *cct,__s32 r)
+{
- if (r < -34) {
@liewegas I see pros/cons with both approaches. How will native errors from local systems calls be differentiated using
this approach? Using the translation at the unmarshalling layer means that errors from local calls won't be mapped (which
is what we want I'm thinking).
Hmm, yeah. Maybe a better approach would be to create a new set of types:
errorcode32_t
errorcode64_t
that do the translation in the encode/decode methods, and have the magic
conversion methods like
errorcode32_t(int i) : val(i) {}
and
operator int() {
return val;
}
Then we'd just change everything in messages/* that sends a result code
over the wire to use this type instead of int32_t or int64_t...
There was a problem hiding this comment.
There was a problem hiding this comment.
Okay, I think I get the approach. I'll try it out. Thanks.
|
It would help final review if you could break this into a few different patches. One for the errno, one for the socket stuff, one for the sockaddr_storage, etc. Generally looks okay, though! Thanks! |
|
Inadvertently closed this pull request. Created a new one: |
Signed-off-by: Rohan Mars code@rohanmars.com