fsverity: Use 2-way interleaved SHA-256 hashing when supported

When the crypto library provides an optimized implementation of
sha256_finup_2x(), use it to interleave the hashing of pairs of data
blocks.  On some CPUs this nearly doubles hashing performance.  The
increase in overall throughput of cold-cache fsverity reads that I'm
seeing on arm64 and x86_64 is roughly 35% (though this metric is hard to
measure as it jumps around a lot).

For now this is only done on the verification path, and only for data
blocks, not Merkle tree blocks.  We could use sha256_finup_2x() on
Merkle tree blocks too, but that is less important as there aren't as
many Merkle tree blocks as data blocks, and that would require some
additional code restructuring.  We could also use sha256_finup_2x() to
accelerate building the Merkle tree, but verification performance is
more important.

Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250915160819.140019-7-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>

Author: Eric Biggers

fs/verity/verify.c

@@ -10,6 +10,31 @@
 #include <linux/bio.h>
 #include <linux/export.h>
 
+#define FS_VERITY_MAX_PENDING_BLOCKS 2
+
+struct fsverity_pending_block {
+	const void *data;
+	u64 pos;
+	u8 real_hash[FS_VERITY_MAX_DIGEST_SIZE];
+};
+
+struct fsverity_verification_context {
+	struct inode *inode;
+	struct fsverity_info *vi;
+	unsigned long max_ra_pages;
+
+	/*
+	 * This is the queue of data blocks that are pending verification.  When
+	 * the crypto layer supports interleaved hashing, we allow multiple
+	 * blocks to be queued up in order to utilize it.  This can improve
+	 * performance significantly vs. sequential hashing of each block.
+	 */
+	int num_pending;
+	int max_pending;
+	struct fsverity_pending_block
+		pending_blocks[FS_VERITY_MAX_PENDING_BLOCKS];
+};
+
 static struct workqueue_struct *fsverity_read_workqueue;
 
 /*
@@ -79,7 +104,7 @@ static bool is_hash_block_verified(struct fsverity_info *vi, struct page *hpage,
 }
 
 /*
- * Verify a single data block against the file's Merkle tree.
+ * Verify the hash of a single data block against the file's Merkle tree.
  *
  * In principle, we need to verify the entire path to the root node.  However,
  * for efficiency the filesystem may cache the hash blocks.  Therefore we need
@@ -88,10 +113,11 @@ static bool is_hash_block_verified(struct fsverity_info *vi, struct page *hpage,
  *
  * Return: %true if the data block is valid, else %false.
  */
-static bool
-verify_data_block(struct inode *inode, struct fsverity_info *vi,
-		  const void *data, u64 data_pos, unsigned long max_ra_pages)
+static bool verify_data_block(struct inode *inode, struct fsverity_info *vi,
+			      const struct fsverity_pending_block *dblock,
+			      unsigned long max_ra_pages)
 {
+	const u64 data_pos = dblock->pos;
 	const struct merkle_tree_params *params = &vi->tree_params;
 	const unsigned int hsize = params->digest_size;
 	int level;
@@ -115,8 +141,12 @@ verify_data_block(struct inode *inode, struct fsverity_info *vi,
 	 */
 	u64 hidx = data_pos >> params->log_blocksize;
 
-	/* Up to 1 + FS_VERITY_MAX_LEVELS pages may be mapped at once */
-	BUILD_BUG_ON(1 + FS_VERITY_MAX_LEVELS > KM_MAX_IDX);
+	/*
+	 * Up to FS_VERITY_MAX_PENDING_BLOCKS + FS_VERITY_MAX_LEVELS pages may
+	 * be mapped at once.
+	 */
+	static_assert(FS_VERITY_MAX_PENDING_BLOCKS + FS_VERITY_MAX_LEVELS <=
+		      KM_MAX_IDX);
 
 	if (unlikely(data_pos >= inode->i_size)) {
 		/*
@@ -127,7 +157,7 @@ verify_data_block(struct inode *inode, struct fsverity_info *vi,
 		 * any part past EOF should be all zeroes.  Therefore, we need
 		 * to verify that any data blocks fully past EOF are all zeroes.
 		 */
-		if (memchr_inv(data, 0, params->block_size)) {
+		if (memchr_inv(dblock->data, 0, params->block_size)) {
 			fsverity_err(inode,
 				     "FILE CORRUPTED!  Data past EOF is not zeroed");
 			return false;
@@ -220,18 +250,18 @@ verify_data_block(struct inode *inode, struct fsverity_info *vi,
 		put_page(hpage);
 	}
 
-	/* Finally, verify the data block. */
-	fsverity_hash_block(params, data, real_hash);
-	if (memcmp(want_hash, real_hash, hsize) != 0)
+	/* Finally, verify the hash of the data block. */
+	if (memcmp(want_hash, dblock->real_hash, hsize) != 0)
 		goto corrupted;
 	return true;
 
 corrupted:
-	fsverity_err(inode,
-		     "FILE CORRUPTED! pos=%llu, level=%d, want_hash=%s:%*phN, real_hash=%s:%*phN",
-		     data_pos, level - 1,
-		     params->hash_alg->name, hsize, want_hash,
-		     params->hash_alg->name, hsize, real_hash);
+	fsverity_err(
+		inode,
+		"FILE CORRUPTED! pos=%llu, level=%d, want_hash=%s:%*phN, real_hash=%s:%*phN",
+		data_pos, level - 1, params->hash_alg->name, hsize, want_hash,
+		params->hash_alg->name, hsize,
+		level == 0 ? dblock->real_hash : real_hash);
 error:
 	for (; level > 0; level--) {
 		kunmap_local(hblocks[level - 1].addr);
@@ -240,13 +270,73 @@ verify_data_block(struct inode *inode, struct fsverity_info *vi,
 	return false;
 }
 
-static bool
-verify_data_blocks(struct folio *data_folio, size_t len, size_t offset,
-		   unsigned long max_ra_pages)
+static void
+fsverity_init_verification_context(struct fsverity_verification_context *ctx,
+				   struct inode *inode,
+				   unsigned long max_ra_pages)
 {
-	struct inode *inode = data_folio->mapping->host;
 	struct fsverity_info *vi = *fsverity_info_addr(inode);
-	const unsigned int block_size = vi->tree_params.block_size;
+
+	ctx->inode = inode;
+	ctx->vi = vi;
+	ctx->max_ra_pages = max_ra_pages;
+	ctx->num_pending = 0;
+	if (vi->tree_params.hash_alg->algo_id == HASH_ALGO_SHA256 &&
+	    sha256_finup_2x_is_optimized())
+		ctx->max_pending = 2;
+	else
+		ctx->max_pending = 1;
+}
+
+static void
+fsverity_clear_pending_blocks(struct fsverity_verification_context *ctx)
+{
+	int i;
+
+	for (i = ctx->num_pending - 1; i >= 0; i--) {
+		kunmap_local(ctx->pending_blocks[i].data);
+		ctx->pending_blocks[i].data = NULL;
+	}
+	ctx->num_pending = 0;
+}
+
+static bool
+fsverity_verify_pending_blocks(struct fsverity_verification_context *ctx)
+{
+	struct fsverity_info *vi = ctx->vi;
+	const struct merkle_tree_params *params = &vi->tree_params;
+	int i;
+
+	if (ctx->num_pending == 2) {
+		/* num_pending == 2 implies that the algorithm is SHA-256 */
+		sha256_finup_2x(params->hashstate ? &params->hashstate->sha256 :
+						    NULL,
+				ctx->pending_blocks[0].data,
+				ctx->pending_blocks[1].data, params->block_size,
+				ctx->pending_blocks[0].real_hash,
+				ctx->pending_blocks[1].real_hash);
+	} else {
+		for (i = 0; i < ctx->num_pending; i++)
+			fsverity_hash_block(params, ctx->pending_blocks[i].data,
+					    ctx->pending_blocks[i].real_hash);
+	}
+
+	for (i = 0; i < ctx->num_pending; i++) {
+		if (!verify_data_block(ctx->inode, vi, &ctx->pending_blocks[i],
+				       ctx->max_ra_pages))
+			return false;
+	}
+	fsverity_clear_pending_blocks(ctx);
+	return true;
+}
+
+static bool fsverity_add_data_blocks(struct fsverity_verification_context *ctx,
+				     struct folio *data_folio, size_t len,
+				     size_t offset)
+{
+	struct fsverity_info *vi = ctx->vi;
+	const struct merkle_tree_params *params = &vi->tree_params;
+	const unsigned int block_size = params->block_size;
 	u64 pos = (u64)data_folio->index << PAGE_SHIFT;
 
 	if (WARN_ON_ONCE(len <= 0 || !IS_ALIGNED(len | offset, block_size)))
@@ -255,14 +345,11 @@ verify_data_blocks(struct folio *data_folio, size_t len, size_t offset,
 			 folio_test_uptodate(data_folio)))
 		return false;
 	do {
-		void *data;
-		bool valid;
-
-		data = kmap_local_folio(data_folio, offset);
-		valid = verify_data_block(inode, vi, data, pos + offset,
-					  max_ra_pages);
-		kunmap_local(data);
-		if (!valid)
+		ctx->pending_blocks[ctx->num_pending].data =
+			kmap_local_folio(data_folio, offset);
+		ctx->pending_blocks[ctx->num_pending].pos = pos + offset;
+		if (++ctx->num_pending == ctx->max_pending &&
+		    !fsverity_verify_pending_blocks(ctx))
 			return false;
 		offset += block_size;
 		len -= block_size;
@@ -284,7 +371,15 @@ verify_data_blocks(struct folio *data_folio, size_t len, size_t offset,
  */
 bool fsverity_verify_blocks(struct folio *folio, size_t len, size_t offset)
 {
-	return verify_data_blocks(folio, len, offset, 0);
+	struct fsverity_verification_context ctx;
+
+	fsverity_init_verification_context(&ctx, folio->mapping->host, 0);
+
+	if (fsverity_add_data_blocks(&ctx, folio, len, offset) &&
+	    fsverity_verify_pending_blocks(&ctx))
+		return true;
+	fsverity_clear_pending_blocks(&ctx);
+	return false;
 }
 EXPORT_SYMBOL_GPL(fsverity_verify_blocks);
 
@@ -305,6 +400,8 @@ EXPORT_SYMBOL_GPL(fsverity_verify_blocks);
  */
 void fsverity_verify_bio(struct bio *bio)
 {
+	struct inode *inode = bio_first_folio_all(bio)->mapping->host;
+	struct fsverity_verification_context ctx;
 	struct folio_iter fi;
 	unsigned long max_ra_pages = 0;
 
@@ -321,13 +418,21 @@ void fsverity_verify_bio(struct bio *bio)
 		max_ra_pages = bio->bi_iter.bi_size >> (PAGE_SHIFT + 2);
 	}
 
+	fsverity_init_verification_context(&ctx, inode, max_ra_pages);
+
 	bio_for_each_folio_all(fi, bio) {
-		if (!verify_data_blocks(fi.folio, fi.length, fi.offset,
-					max_ra_pages)) {
-			bio->bi_status = BLK_STS_IOERR;
-			break;
-		}
+		if (!fsverity_add_data_blocks(&ctx, fi.folio, fi.length,
+					      fi.offset))
+			goto ioerr;
 	}
+
+	if (!fsverity_verify_pending_blocks(&ctx))
+		goto ioerr;
+	return;
+
+ioerr:
+	fsverity_clear_pending_blocks(&ctx);
+	bio->bi_status = BLK_STS_IOERR;
 }
 EXPORT_SYMBOL_GPL(fsverity_verify_bio);
 #endif /* CONFIG_BLOCK */