summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--libglusterfs/src/iatt.h18
1 files changed, 18 insertions, 0 deletions
diff --git a/libglusterfs/src/iatt.h b/libglusterfs/src/iatt.h
index eb916ad5edc..60ae5904703 100644
--- a/libglusterfs/src/iatt.h
+++ b/libglusterfs/src/iatt.h
@@ -265,6 +265,24 @@ iatt_from_stat (struct iatt *iatt, struct stat *stat)
iatt->ia_blksize = stat->st_blksize;
iatt->ia_blocks = stat->st_blocks;
+ /* There is a possibility that the backend FS (like XFS) can
+ allocate blocks beyond EOF for better performance reasons, which
+ results in 'st_blocks' with higher values than what is consumed by
+ the file descriptor. This would break few logic inside GlusterFS,
+ like quota behavior etc, thus we need the exact number of blocks
+ which are consumed by the file to the higher layers inside GlusterFS.
+ Currently, this logic won't work for sparse files (ie, file with
+ holes)
+ */
+ {
+ uint64_t maxblocks;
+
+ maxblocks = (iatt->ia_size + 511) / 512;
+
+ if (iatt->ia_blocks > maxblocks)
+ iatt->ia_blocks = maxblocks;
+ }
+
iatt->ia_atime = stat->st_atime;
iatt->ia_atime_nsec = ST_ATIM_NSEC (stat);