Solaris statfs fix

[edgargabriel]

This is part 1 of the fix to make ompio compile on solaris correctly.

@rhc54 would you mind having a look especially at the modifications in opal_path_nfs that I made? Basically I added an additional argument, which is however only set if the user does not pass NULL as an input pointer.

Part 2 of the fix will have to occur based on this modification on the release branch only (that code section does not exist in master)

[ggouaillardet]

@edgargabriel
fwiw, i cherry-picked these two commits into the v2.x branch, and added the inlined one-off commit and was able to compile on Solaris

commit ec7fc0839d0d41fc0b582b0f60055cc5e05869ba
Author: Gilles Gouaillardet <gilles@rist.or.jp>
Date:   Wed Jan 27 16:53:35 2016 +0900

    ompio: use mca_fs_base_get_fstype to make it works even on Solaris

    Thanks Siegmar Gross for reporting this issue

diff --git a/ompi/mca/io/ompio/io_ompio_component.c b/ompi/mca/io/ompio/io_ompio_component.c
index ff00c6d..c143d33 100644
--- a/ompi/mca/io/ompio/io_ompio_component.c
+++ b/ompi/mca/io/ompio/io_ompio_component.c
@@ -281,22 +281,11 @@ file_query(struct ompi_file_t *file,
     rank = ompi_comm_rank ( file->f_comm);
     if (!tmp) {
         if ( 0 == rank) {
-            do {
-                err = statfs (file->f_filename, &fsbuf);
-            } while (err && (errno == ESTALE));
-            
-            if (err && (errno == ENOENT)) {
-                mca_fs_base_get_parent_dir (file->f_filename, &dir);
-                err = statfs (dir, &fsbuf);
-                free (dir);
-            }
-#ifndef LL_SUPER_MAGIC
-#define LL_SUPER_MAGIC 0x0BD00BD0
-#endif
-            if (fsbuf.f_type == LL_SUPER_MAGIC) {
+            if (LUSTRE == mca_fs_base_get_fstype(file->f_filename)) {
                 is_lustre = 1; //true
             }
        }
+        
        file->f_comm->c_coll.coll_bcast (&is_lustre,
                                          1,
                                          MPI_INT,

[edgargabriel]

@ggouaillardet sounds good, thanks! I just would like @rhc54 to give a thumbs up to the change, since I modified his code. After that, its probably a good idea to got through at least one night of MTT testing and coverty reports before filing the pr to the v2.x branch.

[rhc54]

👍

[edgargabriel]

@ggouaillardet I wanted to cherry pick your commit ec7fc0839d0d41fc0b582b0f60055cc5e05869ba, but can not find the repository that you have that in. Could you please point me to that?

[ggouaillardet]

this must be on my workstation only.
feel free to copy/paste it and commit it, since I will not be able to upload it before monday

[ggouaillardet]

@edgargabriel i made PR open-mpi/ompi-release#935 that includes the backports of your commits plus my one-off one