fs-verity: implement readahead of Merkle tree pages

When fs-verity verifies data pages, currently it reads each Merkle tree
page synchronously using read_mapping_page().

Therefore, when the Merkle tree pages aren't already cached, fs-verity
causes an extra 4 KiB I/O request for every 512 KiB of data (assuming
that the Merkle tree uses SHA-256 and 4 KiB blocks). This results in
more I/O requests and performance loss than is strictly necessary.

Therefore, implement readahead of the Merkle tree pages.

For simplicity, we take advantage of the fact that the kernel already
does readahead of the file's *data*, just like it does for any other
file. Due to this, we don't really need a separate readahead state
(struct file_ra_state) just for the Merkle tree, but rather we just need
to piggy-back on the existing data readahead requests.

We also only really need to bother with the first level of the Merkle
tree, since the usual fan-out factor is 128, so normally over 99% of
Merkle tree I/O requests are for the first level.

Therefore, make fsverity_verify_bio() enable readahead of the first
Merkle tree level, for up to 1/4 the number of pages in the bio, when it
sees that the REQ_RAHEAD flag is set on the bio. The readahead size is
then passed down to ->read_merkle_tree_page() for the filesystem to
(optionally) implement if it sees that the requested page is uncached.

While we're at it, also make build_merkle_tree_level() set the Merkle
tree readahead size, since it's easy to do there.

However, for now don't set the readahead size in fsverity_verify_page(),
since currently it's only used to verify holes on ext4 and f2fs, and it
would need parameters added to know how much to read ahead.

This patch significantly improves fs-verity sequential read performance.
Some quick benchmarks with 'cat'-ing a 250MB file after dropping caches:

On an ARM64 phone (using sha256-ce):
Before: 217 MB/s
After: 263 MB/s
(compare to sha256sum of non-verity file: 357 MB/s)

In an x86_64 VM (using sha256-avx2):
Before: 173 MB/s
After: 215 MB/s
(compare to sha256sum of non-verity file: 223 MB/s)

Link: https://lore.kernel.org/r/[e-mail]
Reviewed-by: Theodore Ts'o <[e-mail]>
Signed-off-by: Eric Biggers <[e-mail]>

fs/ext4/verity.c

···
 	return desc_size;
 }
 
-static struct page *ext4_read_merkle_tree_page(struct inode *inode,
-					       pgoff_t index)
+/*
+ * Prefetch some pages from the file's Merkle tree.
+ *
+ * This is basically a stripped-down version of __do_page_cache_readahead()
+ * which works on pages past i_size.
+ */
+static void ext4_merkle_tree_readahead(struct address_space *mapping,
+				       pgoff_t start_index, unsigned long count)
 {
+	LIST_HEAD(pages);
+	unsigned int nr_pages = 0;
+	struct page *page;
+	pgoff_t index;
+	struct blk_plug plug;
+
+	for (index = start_index; index < start_index + count; index++) {
+		page = xa_load(&mapping->i_pages, index);
+		if (!page || xa_is_value(page)) {
+			page = __page_cache_alloc(readahead_gfp_mask(mapping));
+			if (!page)
+				break;
+			page->index = index;
+			list_add(&page->lru, &pages);
+			nr_pages++;
+		}
+	}
+	blk_start_plug(&plug);
+	ext4_mpage_readpages(mapping, &pages, NULL, nr_pages, true);
+	blk_finish_plug(&plug);
+}
+
+static struct page *ext4_read_merkle_tree_page(struct inode *inode,
+					       pgoff_t index,
+					       unsigned long num_ra_pages)
+{
+	struct page *page;
+
 	index += ext4_verity_metadata_pos(inode) >> PAGE_SHIFT;
 
-	return read_mapping_page(inode->i_mapping, index, NULL);
+	page = find_get_page_flags(inode->i_mapping, index, FGP_ACCESSED);
+	if (!page || !PageUptodate(page)) {
+		if (page)
+			put_page(page);
+		else if (num_ra_pages > 1)
+			ext4_merkle_tree_readahead(inode->i_mapping, index,
+						   num_ra_pages);
+		page = read_mapping_page(inode->i_mapping, index, NULL);
+	}
+	return page;
 }
 
 static int ext4_write_merkle_tree_block(struct inode *inode, const void *buf,

fs/f2fs/data.c

···
  * use ->readpage() or do the necessary surgery to decouple ->readpages()
  * from read-ahead.
  */
-static int f2fs_mpage_readpages(struct address_space *mapping,
+int f2fs_mpage_readpages(struct address_space *mapping,
 			struct list_head *pages, struct page *page,
 			unsigned nr_pages, bool is_readahead)
 {

fs/f2fs/f2fs.h

···
 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index);
 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from);
 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index);
+int f2fs_mpage_readpages(struct address_space *mapping,
+			struct list_head *pages, struct page *page,
+			unsigned nr_pages, bool is_readahead);
 struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
 			int op_flags, bool for_write);
 struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index);

fs/f2fs/verity.c

···
 	return size;
 }
 
-static struct page *f2fs_read_merkle_tree_page(struct inode *inode,
-					       pgoff_t index)
+/*
+ * Prefetch some pages from the file's Merkle tree.
+ *
+ * This is basically a stripped-down version of __do_page_cache_readahead()
+ * which works on pages past i_size.
+ */
+static void f2fs_merkle_tree_readahead(struct address_space *mapping,
+				       pgoff_t start_index, unsigned long count)
 {
+	LIST_HEAD(pages);
+	unsigned int nr_pages = 0;
+	struct page *page;
+	pgoff_t index;
+	struct blk_plug plug;
+
+	for (index = start_index; index < start_index + count; index++) {
+		page = xa_load(&mapping->i_pages, index);
+		if (!page || xa_is_value(page)) {
+			page = __page_cache_alloc(readahead_gfp_mask(mapping));
+			if (!page)
+				break;
+			page->index = index;
+			list_add(&page->lru, &pages);
+			nr_pages++;
+		}
+	}
+	blk_start_plug(&plug);
+	f2fs_mpage_readpages(mapping, &pages, NULL, nr_pages, true);
+	blk_finish_plug(&plug);
+}
+
+static struct page *f2fs_read_merkle_tree_page(struct inode *inode,
+					       pgoff_t index,
+					       unsigned long num_ra_pages)
+{
+	struct page *page;
+
 	index += f2fs_verity_metadata_pos(inode) >> PAGE_SHIFT;
 
-	return read_mapping_page(inode->i_mapping, index, NULL);
+	page = find_get_page_flags(inode->i_mapping, index, FGP_ACCESSED);
+	if (!page || !PageUptodate(page)) {
+		if (page)
+			put_page(page);
+		else if (num_ra_pages > 1)
+			f2fs_merkle_tree_readahead(inode->i_mapping, index,
+						   num_ra_pages);
+		page = read_mapping_page(inode->i_mapping, index, NULL);
+	}
+	return page;
 }
 
 static int f2fs_write_merkle_tree_block(struct inode *inode, const void *buf,

fs/verity/enable.c

···
 #include "fsverity_private.h"
 
 #include <crypto/hash.h>
+#include <linux/backing-dev.h>
 #include <linux/mount.h>
 #include <linux/pagemap.h>
 #include <linux/sched/signal.h>
···
 				return err;
 			}
 		} else {
+			unsigned long num_ra_pages =
+				min_t(unsigned long, num_blocks_to_hash - i,
+				      inode->i_sb->s_bdi->io_pages);
+
 			/* Non-leaf: hashing hash block from level below */
 			src_page = vops->read_merkle_tree_page(inode,
-					params->level_start[level - 1] + i);
+					params->level_start[level - 1] + i,
+					num_ra_pages);
 			if (IS_ERR(src_page)) {
 				err = PTR_ERR(src_page);
 				fsverity_err(inode,

fs/verity/fsverity_private.h

···
 	unsigned int log_arity;		/* log2(hashes_per_block) */
 	unsigned int num_levels;	/* number of levels in Merkle tree */
 	u64 tree_size;			/* Merkle tree size in bytes */
+	unsigned long level0_blocks;	/* number of blocks in tree level 0 */
 
 	/*
 	 * Starting block index for each tree level, ordered from leaf level (0)

fs/verity/open.c

···
 		/* temporarily using level_start[] to store blocks in level */
 		params->level_start[params->num_levels++] = blocks;
 	}
+	params->level0_blocks = params->level_start[0];
 
 	/* Compute the starting block of each level */
 	offset = 0;

fs/verity/verify.c

···
  * Return: true if the page is valid, else false.
  */
 static bool verify_page(struct inode *inode, const struct fsverity_info *vi,
-			struct ahash_request *req, struct page *data_page)
+			struct ahash_request *req, struct page *data_page,
+			unsigned long level0_ra_pages)
 {
 	const struct merkle_tree_params *params = &vi->tree_params;
 	const unsigned int hsize = params->digest_size;
···
 		pr_debug_ratelimited("Level %d: hindex=%lu, hoffset=%u\n",
 				     level, hindex, hoffset);
 
-		hpage = inode->i_sb->s_vop->read_merkle_tree_page(inode,
-								  hindex);
+		hpage = inode->i_sb->s_vop->read_merkle_tree_page(inode, hindex,
+				level == 0 ? level0_ra_pages : 0);
 		if (IS_ERR(hpage)) {
 			err = PTR_ERR(hpage);
 			fsverity_err(inode,
···
 	if (unlikely(!req))
 		return false;
 
-	valid = verify_page(inode, vi, req, page);
+	valid = verify_page(inode, vi, req, page, 0);
 
 	ahash_request_free(req);
 
···
 {
 	struct inode *inode = bio_first_page_all(bio)->mapping->host;
 	const struct fsverity_info *vi = inode->i_verity_info;
+	const struct merkle_tree_params *params = &vi->tree_params;
 	struct ahash_request *req;
 	struct bio_vec *bv;
 	struct bvec_iter_all iter_all;
+	unsigned long max_ra_pages = 0;
 
-	req = ahash_request_alloc(vi->tree_params.hash_alg->tfm, GFP_NOFS);
+	req = ahash_request_alloc(params->hash_alg->tfm, GFP_NOFS);
 	if (unlikely(!req)) {
 		bio_for_each_segment_all(bv, bio, iter_all)
 			SetPageError(bv->bv_page);
 		return;
 	}
 
+	if (bio->bi_opf & REQ_RAHEAD) {
+		/*
+		 * If this bio is for data readahead, then we also do readahead
+		 * of the first (largest) level of the Merkle tree.  Namely,
+		 * when a Merkle tree page is read, we also try to piggy-back on
+		 * some additional pages -- up to 1/4 the number of data pages.
+		 *
+		 * This improves sequential read performance, as it greatly
+		 * reduces the number of I/O requests made to the Merkle tree.
+		 */
+		bio_for_each_segment_all(bv, bio, iter_all)
+			max_ra_pages++;
+		max_ra_pages /= 4;
+	}
+
 	bio_for_each_segment_all(bv, bio, iter_all) {
 		struct page *page = bv->bv_page;
+		unsigned long level0_index = page->index >> params->log_arity;
+		unsigned long level0_ra_pages =
+			min(max_ra_pages, params->level0_blocks - level0_index);
 
-		if (!PageError(page) && !verify_page(inode, vi, req, page))
+		if (!PageError(page) &&
+		    !verify_page(inode, vi, req, page, level0_ra_pages))
 			SetPageError(page);
 	}
 

include/linux/fsverity.h

···
 	 *
 	 * @inode: the inode
 	 * @index: 0-based index of the page within the Merkle tree
+	 * @num_ra_pages: The number of Merkle tree pages that should be
+	 *		  prefetched starting at @index if the page at @index
+	 *		  isn't already cached.  Implementations may ignore this
+	 *		  argument; it's only a performance optimization.
 	 *
 	 * This can be called at any time on an open verity file, as well as
 	 * between ->begin_enable_verity() and ->end_enable_verity().  It may be
···
 	 * Return: the page on success, ERR_PTR() on failure
 	 */
 	struct page *(*read_merkle_tree_page)(struct inode *inode,
-					      pgoff_t index);
+					      pgoff_t index,
+					      unsigned long num_ra_pages);
 
 	/**
 	 * Write a Merkle tree block to the given inode.