summaryrefslogtreecommitdiffhomepage
path: root/pkg/merkletree/merkletree.go
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/merkletree/merkletree.go')
-rw-r--r--pkg/merkletree/merkletree.go259
1 files changed, 219 insertions, 40 deletions
diff --git a/pkg/merkletree/merkletree.go b/pkg/merkletree/merkletree.go
index 906f67943..955c9c473 100644
--- a/pkg/merkletree/merkletree.go
+++ b/pkg/merkletree/merkletree.go
@@ -16,7 +16,9 @@
package merkletree
import (
+ "bytes"
"crypto/sha256"
+ "fmt"
"io"
"gvisor.dev/gvisor/pkg/usermem"
@@ -27,50 +29,78 @@ const (
sha256DigestSize = 32
)
-// Size defines the scale of a Merkle tree.
-type Size struct {
+// Layout defines the scale of a Merkle tree.
+type Layout struct {
// blockSize is the size of a data block to be hashed.
blockSize int64
// digestSize is the size of a generated hash.
digestSize int64
- // hashesPerBlock is the number of hashes in a block. For example, if
- // blockSize is 4096 bytes, and digestSize is 32 bytes, there will be 128
- // hashesPerBlock. Therefore 128 hashes in a lower level will be put into a
- // block and generate a single hash in an upper level.
- hashesPerBlock int64
- // levelStart is the start block index of each level. The number of levels in
- // the tree is the length of the slice. The leafs (level 0) are hashes of
- // blocks in the input data. The levels above are hashes of lower level
- // hashes. The highest level is the root hash.
- levelStart []int64
+ // levelOffset contains the offset of the begnning of each level in
+ // bytes. The number of levels in the tree is the length of the slice.
+ // The leaf nodes (level 0) contain hashes of blocks of the input data.
+ // Each level N contains hashes of the blocks in level N-1. The highest
+ // level is the root hash.
+ levelOffset []int64
}
-// MakeSize initializes and returns a new Size object describing the structure
-// of a tree. dataSize specifies the number of the file system size in bytes.
-func MakeSize(dataSize int64) Size {
- size := Size{
+// InitLayout initializes and returns a new Layout object describing the structure
+// of a tree. dataSize specifies the size of input data in bytes.
+func InitLayout(dataSize int64) Layout {
+ layout := Layout{
blockSize: usermem.PageSize,
// TODO(b/156980949): Allow config other hash methods (SHA384/SHA512).
- digestSize: sha256DigestSize,
- hashesPerBlock: usermem.PageSize / sha256DigestSize,
+ digestSize: sha256DigestSize,
}
- numBlocks := (dataSize + size.blockSize - 1) / size.blockSize
- level := int64(0)
+ numBlocks := (dataSize + layout.blockSize - 1) / layout.blockSize
+ level := 0
offset := int64(0)
- // Calcuate the number of levels in the Merkle tree and the beginning offset
- // of each level. Level 0 is the level directly above the data blocks, while
- // level NumLevels - 1 is the root.
+ // Calculate the number of levels in the Merkle tree and the beginning
+ // offset of each level. Level 0 consists of the leaf nodes that
+ // contain the hashes of the data blocks, while level numLevels - 1 is
+ // the root.
for numBlocks > 1 {
- size.levelStart = append(size.levelStart, offset)
+ layout.levelOffset = append(layout.levelOffset, offset*layout.blockSize)
// Round numBlocks up to fill up a block.
- numBlocks += (size.hashesPerBlock - numBlocks%size.hashesPerBlock) % size.hashesPerBlock
- offset += numBlocks / size.hashesPerBlock
- numBlocks = numBlocks / size.hashesPerBlock
+ numBlocks += (layout.hashesPerBlock() - numBlocks%layout.hashesPerBlock()) % layout.hashesPerBlock()
+ offset += numBlocks / layout.hashesPerBlock()
+ numBlocks = numBlocks / layout.hashesPerBlock()
level++
}
- size.levelStart = append(size.levelStart, offset)
- return size
+ layout.levelOffset = append(layout.levelOffset, offset*layout.blockSize)
+ return layout
+}
+
+// hashesPerBlock() returns the number of digests in each block. For example,
+// if blockSize is 4096 bytes, and digestSize is 32 bytes, there will be 128
+// hashesPerBlock. Therefore 128 hashes in one level will be combined in one
+// hash in the level above.
+func (layout Layout) hashesPerBlock() int64 {
+ return layout.blockSize / layout.digestSize
+}
+
+// numLevels returns the total number of levels in the Merkle tree.
+func (layout Layout) numLevels() int {
+ return len(layout.levelOffset)
+}
+
+// rootLevel returns the level of the root hash.
+func (layout Layout) rootLevel() int {
+ return layout.numLevels() - 1
+}
+
+// digestOffset finds the offset of a digest from the beginning of the tree.
+// The target digest is at level of the tree, with index from the beginning of
+// the current level.
+func (layout Layout) digestOffset(level int, index int64) int64 {
+ return layout.levelOffset[level] + index*layout.digestSize
+}
+
+// blockOffset finds the offset of a block from the beginning of the tree. The
+// target block is at level of the tree, with index from the beginning of the
+// current level.
+func (layout Layout) blockOffset(level int, index int64) int64 {
+ return layout.levelOffset[level] + index*layout.blockSize
}
// Generate constructs a Merkle tree for the contents of data. The output is
@@ -78,21 +108,21 @@ func MakeSize(dataSize int64) Size {
// it has been written. That is, treeWriter and treeReader should point to the
// same underlying data but have separate cursors.
func Generate(data io.Reader, dataSize int64, treeReader io.Reader, treeWriter io.Writer) ([]byte, error) {
- size := MakeSize(dataSize)
+ layout := InitLayout(dataSize)
- numBlocks := (dataSize + size.blockSize - 1) / size.blockSize
+ numBlocks := (dataSize + layout.blockSize - 1) / layout.blockSize
var root []byte
- for level := 0; level < len(size.levelStart); level++ {
+ for level := 0; level < layout.numLevels(); level++ {
for i := int64(0); i < numBlocks; i++ {
- buf := make([]byte, size.blockSize)
+ buf := make([]byte, layout.blockSize)
var (
n int
err error
)
if level == 0 {
- // Read data block from the target file since level 0 is directly above
- // the raw data block.
+ // Read data block from the target file since level 0 includes hashes
+ // of blocks in the input data.
n, err = data.Read(buf)
} else {
// Read data block from the tree file since levels higher than 0 are
@@ -112,7 +142,7 @@ func Generate(data io.Reader, dataSize int64, treeReader io.Reader, treeWriter i
// Hash the bytes in buf.
digest := sha256.Sum256(buf)
- if level == len(size.levelStart)-1 {
+ if level == layout.rootLevel() {
root = digest[:]
}
@@ -121,15 +151,164 @@ func Generate(data io.Reader, dataSize int64, treeReader io.Reader, treeWriter i
return nil, err
}
}
- // If the genereated digests do not round up to a block, zero-padding the
+ // If the generated digests do not round up to a block, zero-padding the
// remaining of the last block. But no need to do so for root.
- if level != len(size.levelStart)-1 && numBlocks%size.hashesPerBlock != 0 {
- zeroBuf := make([]byte, size.blockSize-(numBlocks%size.hashesPerBlock)*size.digestSize)
+ if level != layout.rootLevel() && numBlocks%layout.hashesPerBlock() != 0 {
+ zeroBuf := make([]byte, layout.blockSize-(numBlocks%layout.hashesPerBlock())*layout.digestSize)
if _, err := treeWriter.Write(zeroBuf[:]); err != nil {
return nil, err
}
}
- numBlocks = (numBlocks + size.hashesPerBlock - 1) / size.hashesPerBlock
+ numBlocks = (numBlocks + layout.hashesPerBlock() - 1) / layout.hashesPerBlock()
}
return root, nil
}
+
+// Verify verifies the content read from data with offset. The content is
+// verified against tree. If content spans across multiple blocks, each block is
+// verified. Verification fails if the hash of the data does not match the tree
+// at any level, or if the final root hash does not match expectedRoot.
+// Once the data is verified, it will be written using w.
+// Verify will modify the cursor for data, but always restores it to its
+// original position upon exit. The cursor for tree is modified and not
+// restored.
+func Verify(w io.Writer, data, tree io.ReadSeeker, dataSize int64, readOffset int64, readSize int64, expectedRoot []byte) error {
+ if readSize <= 0 {
+ return fmt.Errorf("Unexpected read size: %d", readSize)
+ }
+ layout := InitLayout(int64(dataSize))
+
+ // Calculate the index of blocks that includes the target range in input
+ // data.
+ firstDataBlock := readOffset / layout.blockSize
+ lastDataBlock := (readOffset + readSize - 1) / layout.blockSize
+
+ // Store the current offset, so we can set it back once verification
+ // finishes.
+ origOffset, err := data.Seek(0, io.SeekCurrent)
+ if err != nil {
+ return fmt.Errorf("Find current data offset failed: %v", err)
+ }
+ defer data.Seek(origOffset, io.SeekStart)
+
+ // Move to the first block that contains target data.
+ if _, err := data.Seek(firstDataBlock*layout.blockSize, io.SeekStart); err != nil {
+ return fmt.Errorf("Seek to datablock start failed: %v", err)
+ }
+
+ buf := make([]byte, layout.blockSize)
+ var readErr error
+ bytesRead := 0
+ for i := firstDataBlock; i <= lastDataBlock; i++ {
+ // Read a block that includes all or part of target range in
+ // input data.
+ bytesRead, readErr = data.Read(buf)
+ // If at the end of input data and all previous blocks are
+ // verified, return the verified input data and EOF.
+ if readErr == io.EOF && bytesRead == 0 {
+ break
+ }
+ if readErr != nil && readErr != io.EOF {
+ return fmt.Errorf("Read from data failed: %v", err)
+ }
+ // If this is the end of file, zero the remaining bytes in buf,
+ // otherwise they are still from the previous block.
+ // TODO(b/162908070): Investigate possible issues with zero
+ // padding the data.
+ if bytesRead < len(buf) {
+ for j := bytesRead; j < len(buf); j++ {
+ buf[j] = 0
+ }
+ }
+ if err := verifyBlock(tree, layout, buf, i, expectedRoot); err != nil {
+ return err
+ }
+ // startOff is the beginning of the read range within the
+ // current data block. Note that for all blocks other than the
+ // first, startOff should be 0.
+ startOff := int64(0)
+ if i == firstDataBlock {
+ startOff = readOffset % layout.blockSize
+ }
+ // endOff is the end of the read range within the current data
+ // block. Note that for all blocks other than the last, endOff
+ // should be the block size.
+ endOff := layout.blockSize
+ if i == lastDataBlock {
+ endOff = (readOffset+readSize-1)%layout.blockSize + 1
+ }
+ // If the provided size exceeds the end of input data, we should
+ // only copy the parts in buf that's part of input data.
+ if startOff > int64(bytesRead) {
+ startOff = int64(bytesRead)
+ }
+ if endOff > int64(bytesRead) {
+ endOff = int64(bytesRead)
+ }
+ w.Write(buf[startOff:endOff])
+
+ }
+ return readErr
+}
+
+// verifyBlock verifies a block against tree. index is the number of block in
+// original data. The block is verified through each level of the tree. It
+// fails if the calculated hash from block is different from any level of
+// hashes stored in tree. And the final root hash is compared with
+// expectedRoot. verifyBlock modifies the cursor for tree. Users needs to
+// maintain the cursor if intended.
+func verifyBlock(tree io.ReadSeeker, layout Layout, dataBlock []byte, blockIndex int64, expectedRoot []byte) error {
+ if len(dataBlock) != int(layout.blockSize) {
+ return fmt.Errorf("incorrect block size")
+ }
+
+ expectedDigest := make([]byte, layout.digestSize)
+ treeBlock := make([]byte, layout.blockSize)
+ var digest []byte
+ for level := 0; level < layout.numLevels(); level++ {
+ // Calculate hash.
+ if level == 0 {
+ digestArray := sha256.Sum256(dataBlock)
+ digest = digestArray[:]
+ } else {
+ // Read a block in previous level that contains the
+ // hash we just generated, and generate a next level
+ // hash from it.
+ if _, err := tree.Seek(layout.blockOffset(level-1, blockIndex), io.SeekStart); err != nil {
+ return err
+ }
+ if _, err := tree.Read(treeBlock); err != nil {
+ return err
+ }
+ digestArray := sha256.Sum256(treeBlock)
+ digest = digestArray[:]
+ }
+
+ // Move to stored hash for the current block, read the digest
+ // and store in expectedDigest.
+ if _, err := tree.Seek(layout.digestOffset(level, blockIndex), io.SeekStart); err != nil {
+ return err
+ }
+ if _, err := tree.Read(expectedDigest); err != nil {
+ return err
+ }
+
+ if !bytes.Equal(digest, expectedDigest) {
+ return fmt.Errorf("Verification failed")
+ }
+
+ // If this is the root layer, no need to generate next level
+ // hash.
+ if level == layout.rootLevel() {
+ break
+ }
+ blockIndex = blockIndex / layout.hashesPerBlock()
+ }
+
+ // Verification for the tree succeeded. Now compare the root hash in the
+ // tree with expectedRoot.
+ if !bytes.Equal(digest[:], expectedRoot) {
+ return fmt.Errorf("Verification failed")
+ }
+ return nil
+}