DAOS-15682 dfuse: Perform reads in larger chunks. (#14212)

When dfuse sees I/O as well-aligned 128k reads then read MB at
a time and cache the result allowing for faster read bandwidth
for well behaved applicaions and large files.

Create a new in-memory descriptor for file contents, pull in a
whole descriptor on first read and perform all other reads from
the same result.

This should give much higher bandwidth for well behaved applications
and should be easy to extend to proper readahead in future.

Signed-off-by: Ashley Pittman <ashley.m.pittman@intel.com>

src/client/dfuse/dfuse.h

@@ -405,6 +405,7 @@ struct dfuse_event {
 	union {
 		struct dfuse_obj_hdl     *de_oh;
 		struct dfuse_inode_entry *de_ie;
+		struct read_chunk_data   *de_cd;
 	};
 	off_t  de_req_position; /**< The file position requested by fuse */
 	union {
@@ -1011,6 +1012,8 @@ struct dfuse_inode_entry {
 
 	/* Entry on the evict list */
 	d_list_t                  ie_evict_entry;
+
+	struct read_chunk_core   *ie_chunk;
 };
 
 /* Flush write-back cache writes to a inode.  It does this by waiting for and then releasing an
@@ -1108,6 +1111,13 @@ dfuse_compute_inode(struct dfuse_cont *dfs,
 void
 dfuse_cache_evict_dir(struct dfuse_info *dfuse_info, struct dfuse_inode_entry *ie);
 
+/* Free any read chunk data for an inode.
+ *
+ * Returns true if feature was used.
+ */
+bool
+read_chunk_close(struct dfuse_inode_entry *ie);
+
 /* Metadata caching functions. */
 
 /* Mark the cache as up-to-date from now */

src/client/dfuse/ops/open.c

@@ -139,6 +139,7 @@ dfuse_cb_release(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi)
 	struct dfuse_obj_hdl     *oh         = (struct dfuse_obj_hdl *)fi->fh;
 	struct dfuse_inode_entry *ie         = NULL;
 	int                       rc;
+	uint32_t                  oc;
 	uint32_t                  il_calls;
 
 	/* Perform the opposite of what the ioctl call does, always change the open handle count
@@ -206,7 +207,11 @@ dfuse_cb_release(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi)
 	if (il_calls != 0) {
 		atomic_fetch_sub_relaxed(&oh->doh_ie->ie_il_count, 1);
 	}
-	atomic_fetch_sub_relaxed(&oh->doh_ie->ie_open_count, 1);
+	oc = atomic_fetch_sub_relaxed(&oh->doh_ie->ie_open_count, 1);
+	if (oc == 1) {
+		if (read_chunk_close(oh->doh_ie))
+			oh->doh_linear_read = true;
+	}
 
 	if (oh->doh_evict_on_close) {
 		ie = oh->doh_ie;