LU-14423 osd: recognize holes in osd_is_mapped()

ldiskfs_fiemap() can return {0,0,0} for last non-allocated
region.  osd_is_mapped() should be able to recognize and
cache this state.

Fixes: 144b5a65c1 ("LU-7132 osd-ldiskfs: speedup rewrites")
Signed-off-by: Alex Zhuravlev <bzzz@whamcloud.com>
Change-Id: I03883038c2c0ec84754377a442c4947c7e3021a9
Reviewed-on: https://review.whamcloud.com/41481
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Andrew Perepechko <andrew.perepechko@hpe.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c
index d2a9c88..f99f625 100644 (file)
--- a/lustre/osd-ldiskfs/osd_io.c
+++ b/lustre/osd-ldiskfs/osd_io.c
@@ -1335,6 +1335,16 @@ static int osd_is_mapped(struct dt_object *dt, __u64 offset,
                return 0;
 
        start = fe.fe_logical >> inode->i_blkbits;
+       if (fei.fi_extents_mapped == 0) {
+               /* a special case - no extent found at this offset and forward.
+                * we can consider this as a hole to EOF. it's safe to cache
+                * as other threads can not allocate/punch blocks this thread
+                * is working on (LDLM). */
+               cached_extent->start = block;
+               cached_extent->end = i_size_read(inode) >> inode->i_blkbits;
+               cached_extent->mapped = 0;
+               return 0;
+       }
 
        if (start > block) {
                cached_extent->start = block;