Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Yue Hu <zbestahu@gmail.com>
To:     xiang@kernel.org, chao@kernel.org
Cc:     linux-erofs@lists.ozlabs.org, linux-kernel@vger.kernel.org,
        zhangwen@coolpad.com, Yue Hu <huyue2@coolpad.com>
Subject: [RFC PATCH 1/2] erofs: support on-disk offset for shifted decompression
Date:   Mon, 22 Aug 2022 13:53:00 +0800
Message-Id: <d424334dc32483e93baf0a215dab88c90d14e5c2.1661146058.git.huyue2@coolpad.com>
In-Reply-To: <cover.1661146058.git.huyue2@coolpad.com>
References: <cover.1661146058.git.huyue2@coolpad.com>
In-Reply-To: <cover.1661146058.git.huyue2@coolpad.com>
References: <cover.1661146058.git.huyue2@coolpad.com>
Precedence: bulk

From: Yue Hu <huyue2@coolpad.com>

Currently, there is no start offset when writing uncompressed data to
disk blocks for compressed files. However, we are using in-place I/O
which will decrease the number of memory copies a lot if we write it
just from an offset of 'pageofs_out'. So, let's support it.

Signed-off-by: Yue Hu <huyue2@coolpad.com>
---
 fs/erofs/decompressor.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/fs/erofs/decompressor.c b/fs/erofs/decompressor.c
index 2d55569f96ac..dc02d95b52d7 100644
--- a/fs/erofs/decompressor.c
+++ b/fs/erofs/decompressor.c
@@ -326,6 +326,7 @@ static int z_erofs_shifted_transform(struct z_erofs_decompress_req *rq,
 					     PAGE_SIZE - rq->pageofs_out);
 	const unsigned int lefthalf = rq->outputsize - righthalf;
 	unsigned char *src, *dst;
+	unsigned int headofs_in;
 
 	if (nrpages_out > 2) {
 		DBG_BUGON(1);
@@ -337,21 +338,25 @@ static int z_erofs_shifted_transform(struct z_erofs_decompress_req *rq,
 		return 0;
 	}
 
+	/* set it to pageofs_out if fragments feature is used */
+	headofs_in = 0;
+
 	src = kmap_atomic(*rq->in) + rq->pageofs_in;
 	if (rq->out[0]) {
 		dst = kmap_atomic(rq->out[0]);
-		memcpy(dst + rq->pageofs_out, src, righthalf);
+		memcpy(dst + rq->pageofs_out, src + headofs_in, righthalf);
 		kunmap_atomic(dst);
 	}
 
 	if (nrpages_out == 2) {
 		DBG_BUGON(!rq->out[1]);
-		if (rq->out[1] == *rq->in) {
-			memmove(src, src + righthalf, lefthalf);
-		} else {
+		if (rq->out[1] != *rq->in) {
 			dst = kmap_atomic(rq->out[1]);
-			memcpy(dst, src + righthalf, lefthalf);
+			memcpy(dst, src + (headofs_in ? 0 : righthalf),
+			       lefthalf);
 			kunmap_atomic(dst);
+		} else if (!headofs_in) {
+			memmove(src, src + righthalf, lefthalf);
 		}
 	}
 	kunmap_atomic(src);
-- 
2.17.1