[PATCH 5/5] Squashfs: optimize reading uncompressed data

From: Daniel Rosenberg
Date: Fri Sep 22 2017 - 17:56:18 EST


From: Adrien Schildknecht <adriens@xxxxxxxxxx>

When dealing with uncompressed data, there is no need to read a whole
block (default 128K) to get the desired page: the pages are
independent from each others.

This patch change the readpages logic so that reading uncompressed
data only read the number of pages advised by the readahead algorithm.

Moreover, if the page actor contains holes (i.e. pages that are already
up-to-date), squashfs skips the buffer_head associated to those pages.

This patch greatly improve the performance of random reads for
uncompressed files because squashfs only read what is needed. It also
reduces the number of unnecessary reads.

Signed-off-by: Adrien Schildknecht <adriens@xxxxxxxxxx>
Signed-off-by: Daniel Rosenberg <drosen@xxxxxxxxxx>
---
fs/squashfs/block.c | 25 +++++++++++++++++++++++++
fs/squashfs/file_direct.c | 37 ++++++++++++++++++++++++++++++-------
2 files changed, 55 insertions(+), 7 deletions(-)

diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c
index 252dfc82ae72..37658b6e83ee 100644
--- a/fs/squashfs/block.c
+++ b/fs/squashfs/block.c
@@ -207,6 +207,22 @@ static void squashfs_bio_end_io(struct bio *bio)
kfree(bio_req);
}

+static int bh_is_optional(struct squashfs_read_request *req, int idx)
+{
+ int start_idx, end_idx;
+ struct squashfs_sb_info *msblk = req->sb->s_fs_info;
+
+ start_idx = (idx * msblk->devblksize - req->offset) >> PAGE_SHIFT;
+ end_idx = ((idx + 1) * msblk->devblksize - req->offset + 1) >> PAGE_SHIFT;
+ if (start_idx >= req->output->pages)
+ return 1;
+ if (start_idx < 0)
+ start_idx = end_idx;
+ if (end_idx >= req->output->pages)
+ end_idx = start_idx;
+ return !req->output->page[start_idx] && !req->output->page[end_idx];
+}
+
static int actor_getblks(struct squashfs_read_request *req, u64 block)
{
int i;
@@ -216,6 +232,15 @@ static int actor_getblks(struct squashfs_read_request *req, u64 block)
return -ENOMEM;

for (i = 0; i < req->nr_buffers; ++i) {
+ /*
+ * When dealing with an uncompressed block, the actor may
+ * contains NULL pages. There's no need to read the buffers
+ * associated with these pages.
+ */
+ if (!req->compressed && bh_is_optional(req, i)) {
+ req->bh[i] = NULL;
+ continue;
+ }
req->bh[i] = sb_getblk(req->sb, block + i);
if (!req->bh[i]) {
while (--i) {
diff --git a/fs/squashfs/file_direct.c b/fs/squashfs/file_direct.c
index a978811de327..dc87f77ce11e 100644
--- a/fs/squashfs/file_direct.c
+++ b/fs/squashfs/file_direct.c
@@ -111,15 +111,38 @@ int squashfs_readpages_block(struct page *target_page,
struct squashfs_page_actor *actor;
struct inode *inode = mapping->host;
struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
- int file_end = (i_size_read(inode) - 1) >> PAGE_SHIFT;
+ int start_index, end_index, file_end, actor_pages, res;
int mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1;
- int start_index = page_index & ~mask;
- int end_index = start_index | mask;
- int actor_pages, res;

- if (end_index > file_end)
- end_index = file_end;
- actor_pages = end_index - start_index + 1;
+ /*
+ * If readpage() is called on an uncompressed datablock, we can just
+ * read the pages instead of fetching the whole block.
+ * This greatly improves the performance when a process keep doing
+ * random reads because we only fetch the necessary data.
+ * The readahead algorithm will take care of doing speculative reads
+ * if necessary.
+ * We can't read more than 1 block even if readahead provides use more
+ * pages because we don't know yet if the next block is compressed or
+ * not.
+ */
+ if (bsize && !SQUASHFS_COMPRESSED_BLOCK(bsize)) {
+ u64 block_end = block + msblk->block_size;
+
+ block += (page_index & mask) * PAGE_SIZE;
+ actor_pages = (block_end - block) / PAGE_SIZE;
+ if (*nr_pages < actor_pages)
+ actor_pages = *nr_pages;
+ start_index = page_index;
+ bsize = min_t(int, bsize, (PAGE_SIZE * actor_pages)
+ | SQUASHFS_COMPRESSED_BIT_BLOCK);
+ } else {
+ file_end = (i_size_read(inode) - 1) >> PAGE_SHIFT;
+ start_index = page_index & ~mask;
+ end_index = start_index | mask;
+ if (end_index > file_end)
+ end_index = file_end;
+ actor_pages = end_index - start_index + 1;
+ }

actor = actor_from_page_cache(actor_pages, target_page,
readahead_pages, nr_pages, start_index,
--
2.14.1.821.g8fa685d3b7-goog