NFSD/blocklayout: Support multiple extents per LAYOUTGET

Allow the pNFS server to respond with multiple extents to a LAYOUTGET
request, thereby avoiding unnecessary load on the server and improving
performance for the client. The number of LAYOUTGET requests is
significantly reduced for various file access patterns, including
random and parallel writes.

Additionally, this change allows the client to request layouts with the
loga_minlength value greater than the minimum possible length of a single
extent in XFS. We use this functionality to fix a livelock in the client.

Signed-off-by: Sergey Bashirov <sergeybashirov@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>

Author: sbashiro

fs/nfsd/blocklayout.c

@@ -89,9 +89,9 @@ nfsd4_block_proc_layoutget(struct svc_rqst *rqstp, struct inode *inode,
 {
 	struct nfsd4_layout_seg *seg = &args->lg_seg;
 	struct pnfs_block_layout *bl;
-	struct pnfs_block_extent *bex;
-	u64 length;
-	u32 nr_extents_max = 1, block_size = i_blocksize(inode);
+	struct pnfs_block_extent *first_bex, *last_bex;
+	u64 offset = seg->offset, length = seg->length;
+	u32 i, nr_extents_max, block_size = i_blocksize(inode);
 	__be32 nfserr;
 
 	if (locks_in_grace(SVC_NET(rqstp)))
@@ -118,6 +118,13 @@ nfsd4_block_proc_layoutget(struct svc_rqst *rqstp, struct inode *inode,
 				PNFS_BLOCK_EXTENT_SIZE)
 		goto out_error;
 
+	/*
+	 * Limit the maximum layout size to avoid allocating
+	 * a large buffer on the server for each layout request.
+	 */
+	nr_extents_max = (min(args->lg_maxcount, PAGE_SIZE) -
+			  PNFS_BLOCK_LAYOUT4_SIZE) / PNFS_BLOCK_EXTENT_SIZE;
+
 	/*
 	 * Some clients barf on non-zero block numbers for NONE or INVALID
 	 * layouts, so make sure to zero the whole structure.
@@ -129,23 +136,37 @@ nfsd4_block_proc_layoutget(struct svc_rqst *rqstp, struct inode *inode,
 	bl->nr_extents = nr_extents_max;
 	args->lg_content = bl;
 
-	bex = &bl->extents[0];
-	nfserr = nfsd4_block_map_extent(inode, fhp, seg->offset, seg->length,
-			seg->iomode, args->lg_minlength, bex);
-	if (nfserr != nfs_ok)
-		goto out_error;
+	for (i = 0; i < bl->nr_extents; i++) {
+		struct pnfs_block_extent *bex = bl->extents + i;
+		u64 bex_length;
+
+		nfserr = nfsd4_block_map_extent(inode, fhp, offset, length,
+				seg->iomode, args->lg_minlength, bex);
+		if (nfserr != nfs_ok)
+			goto out_error;
+
+		bex_length = bex->len - (offset - bex->foff);
+		if (bex_length >= length) {
+			bl->nr_extents = i + 1;
+			break;
+		}
+
+		offset = bex->foff + bex->len;
+		length -= bex_length;
+	}
+
+	first_bex = bl->extents;
+	last_bex = bl->extents + bl->nr_extents - 1;
 
 	nfserr = nfserr_layoutunavailable;
-	length = bex->foff + bex->len - seg->offset;
+	length = last_bex->foff + last_bex->len - seg->offset;
 	if (length < args->lg_minlength) {
 		dprintk("pnfsd: extent smaller than minlength\n");
 		goto out_error;
 	}
 
-	seg->offset = bex->foff;
-	seg->length = bex->len;
-
-	dprintk("GET: 0x%llx:0x%llx %d\n", bex->foff, bex->len, bex->es);
+	seg->offset = first_bex->foff;
+	seg->length = last_bex->foff - first_bex->foff + last_bex->len;
 	return nfs_ok;
 
 out_error: