ceph: fix readdir EOVERFLOW on 32-bit archs
authorSage Weil <sage@newdream.net>
Thu, 18 Nov 2010 17:15:07 +0000 (09:15 -0800)
committerSage Weil <sage@newdream.net>
Thu, 18 Nov 2010 17:15:07 +0000 (09:15 -0800)
One of the readdir filldir_t callers was passing the raw ceph 64-bit ino
instead of the hashed 32-bit one, producing an EOVERFLOW in the filler
callback.  Fix this by calling the ceph_vino_to_ino() helper to do the
conversion.

Reported-by: Jan Smets <jan.smets@alcatel-lucent.com>
Tested-by: Jan Smets <jan.smets@alcatel-lucent.com>
Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/dir.c

index 5f67728..7d447af 100644 (file)
@@ -358,18 +358,22 @@ more:
                u64 pos = ceph_make_fpos(frag, off);
                struct ceph_mds_reply_inode *in =
                        rinfo->dir_in[off - fi->offset].in;
+               struct ceph_vino vino;
+               ino_t ino;
+
                dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
                     off, off - fi->offset, rinfo->dir_nr, pos,
                     rinfo->dir_dname_len[off - fi->offset],
                     rinfo->dir_dname[off - fi->offset], in);
                BUG_ON(!in);
                ftype = le32_to_cpu(in->mode) >> 12;
+               vino.ino = le64_to_cpu(in->ino);
+               vino.snap = le64_to_cpu(in->snapid);
+               ino = ceph_vino_to_ino(vino);
                if (filldir(dirent,
                            rinfo->dir_dname[off - fi->offset],
                            rinfo->dir_dname_len[off - fi->offset],
-                           pos,
-                           le64_to_cpu(in->ino),
-                           ftype) < 0) {
+                           pos, ino, ftype) < 0) {
                        dout("filldir stopping us...\n");
                        return 0;
                }