|author||H. Peter Anvin <email@example.com>||2016-02-10 13:47:06 -0800|
|committer||H. Peter Anvin <firstname.lastname@example.org>||2016-02-10 13:50:18 -0800|
getfssec: update comment explaining next_extent() semanticsobsolete-20160210
Update comments which explains next_extent() semantics to be slightly clearer, and also correct the statement that next_extent can be NULL. Signed-off-by: H. Peter Anvin <email@example.com>
1 files changed, 8 insertions, 3 deletions
diff --git a/core/fs/getfssec.c b/core/fs/getfssec.c
index e099b64e..2ea122d9 100644
@@ -39,18 +39,23 @@
* will store the initial sector number into inode->next_extent.lstart
* on return.)
+ * If inode->next_extent.pstart is EXTENT_ZERO, then no disk I/O is
+ * performed, and the data in the extent is all zero.
* If inode->next_extent.len != 0 on entry then the routine is allowed
* to assume inode->next_extent contains valid data from the previous
* usage, which can be used for optimization purposes.
* If the filesystem can map the entire file as a single extent
* (e.g. iso9660), then the filesystem can simply insert the extent
- * information into inode->next_extent at searchdir/iget time, and leave
- * next_extent() as NULL.
+ * information into inode->next_extent at searchdir/iget time, and point
+ * next_extent() to the generic function no_next_extent().
* Note: the filesystem driver is not required to do extent coalescing,
* if that is difficult to do; this routine will perform extent lookahead
- * and coalescing.
+ * and coalescing. However, if the filesystem can do extent coalescing
+ * very cheaply by using filesystem-specific knowledge, then that is
+ * preferred (e.g. FAT).