orangefs: invalidate stored directory on seek

If an application seeks to a position before the point which has been
read, it must want updates which have been made to the directory.  So
delete the copy stored in the kernel so it will be fetched again.

Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
This commit is contained in:
Martin Brandenburg 2017-05-02 12:15:11 -04:00 коммит произвёл Mike Marshall
Родитель bf15ba7c1f
Коммит 942835d68f
1 изменённых файлов: 23 добавлений и 1 удалений

Просмотреть файл

@ -275,6 +275,28 @@ static int orangefs_dir_fill(struct orangefs_inode_s *oi,
return 0;
}
static loff_t orangefs_dir_llseek(struct file *file, loff_t offset,
int whence)
{
struct orangefs_dir *od = file->private_data;
/*
* Delete the stored data so userspace sees new directory
* entries.
*/
if (!whence && offset < od->end) {
struct orangefs_dir_part *part = od->part;
while (part) {
struct orangefs_dir_part *next = part->next;
vfree(part);
part = next;
}
od->token = ORANGEFS_ITERATE_START;
od->part = NULL;
od->end = 1 << PART_SHIFT;
}
return default_llseek(file, offset, whence);
}
static int orangefs_dir_iterate(struct file *file,
struct dir_context *ctx)
{
@ -371,7 +393,7 @@ static int orangefs_dir_release(struct inode *inode, struct file *file)
}
const struct file_operations orangefs_dir_operations = {
.llseek = default_llseek,
.llseek = orangefs_dir_llseek,
.read = generic_read_dir,
.iterate = orangefs_dir_iterate,
.open = orangefs_dir_open,