package fs import ( "crypto/md5" "crypto/sha1" "encoding/hex" "fmt" "hash" "io" "strings" "github.com/pkg/errors" ) // HashType indicates a standard hashing algorithm type HashType int // ErrHashUnsupported should be returned by filesystem, // if it is requested to deliver an unsupported hash type. var ErrHashUnsupported = errors.New("hash type not supported") const ( // HashMD5 indicates MD5 support HashMD5 HashType = 1 << iota // HashSHA1 indicates SHA-1 support HashSHA1 // HashNone indicates no hashes are supported HashNone HashType = 0 ) // SupportedHashes returns a set of all the supported hashes by // HashStream and MultiHasher. var SupportedHashes = NewHashSet(HashMD5, HashSHA1) // HashWidth returns the width in characters for any HashType var HashWidth = map[HashType]int{ HashMD5: 32, HashSHA1: 40, } // HashStream will calculate hashes of all supported hash types. func HashStream(r io.Reader) (map[HashType]string, error) { return HashStreamTypes(r, SupportedHashes) } // HashStreamTypes will calculate hashes of the requested hash types. func HashStreamTypes(r io.Reader, set HashSet) (map[HashType]string, error) { hashers, err := hashFromTypes(set) if err != nil { return nil, err } _, err = io.Copy(hashToMultiWriter(hashers), r) if err != nil { return nil, err } var ret = make(map[HashType]string) for k, v := range hashers { ret[k] = hex.EncodeToString(v.Sum(nil)) } return ret, nil } // String returns a string representation of the hash type. // The function will panic if the hash type is unknown. func (h HashType) String() string { switch h { case HashNone: return "None" case HashMD5: return "MD5" case HashSHA1: return "SHA-1" default: err := fmt.Sprintf("internal error: unknown hash type: 0x%x", int(h)) panic(err) } } // hashFromTypes will return hashers for all the requested types. // The types must be a subset of SupportedHashes, // and this function must support all types. func hashFromTypes(set HashSet) (map[HashType]hash.Hash, error) { if !set.SubsetOf(SupportedHashes) { return nil, errors.Errorf("requested set %08x contains unknown hash types", int(set)) } var hashers = make(map[HashType]hash.Hash) types := set.Array() for _, t := range types { switch t { case HashMD5: hashers[t] = md5.New() case HashSHA1: hashers[t] = sha1.New() default: err := fmt.Sprintf("internal error: Unsupported hash type %v", t) panic(err) } } return hashers, nil } // hashToMultiWriter will return a set of hashers into a // single multiwriter, where one write will update all // the hashers. func hashToMultiWriter(h map[HashType]hash.Hash) io.Writer { // Convert to to slice var w = make([]io.Writer, 0, len(h)) for _, v := range h { w = append(w, v) } return io.MultiWriter(w...) } // A MultiHasher will construct various hashes on // all incoming writes. type MultiHasher struct { w io.Writer size int64 h map[HashType]hash.Hash // Hashes } // NewMultiHasher will return a hash writer that will write all // supported hash types. func NewMultiHasher() *MultiHasher { h, err := NewMultiHasherTypes(SupportedHashes) if err != nil { panic("internal error: could not create multihasher") } return h } // NewMultiHasherTypes will return a hash writer that will write // the requested hash types. func NewMultiHasherTypes(set HashSet) (*MultiHasher, error) { hashers, err := hashFromTypes(set) if err != nil { return nil, err } m := MultiHasher{h: hashers, w: hashToMultiWriter(hashers)} return &m, nil } func (m *MultiHasher) Write(p []byte) (n int, err error) { n, err = m.w.Write(p) m.size += int64(n) return n, err } // Sums returns the sums of all accumulated hashes as hex encoded // strings. func (m *MultiHasher) Sums() map[HashType]string { dst := make(map[HashType]string) for k, v := range m.h { dst[k] = hex.EncodeToString(v.Sum(nil)) } return dst } // Size returns the number of bytes written func (m *MultiHasher) Size() int64 { return m.size } // A HashSet Indicates one or more hash types. type HashSet int // NewHashSet will create a new hash set with the hash types supplied func NewHashSet(t ...HashType) HashSet { h := HashSet(HashNone) return h.Add(t...) } // Add one or more hash types to the set. // Returns the modified hash set. func (h *HashSet) Add(t ...HashType) HashSet { for _, v := range t { *h |= HashSet(v) } return *h } // Contains returns true if the func (h HashSet) Contains(t HashType) bool { return int(h)&int(t) != 0 } // Overlap returns the overlapping hash types func (h HashSet) Overlap(t HashSet) HashSet { return HashSet(int(h) & int(t)) } // SubsetOf will return true if all types of h // is present in the set c func (h HashSet) SubsetOf(c HashSet) bool { return int(h)|int(c) == int(c) } // GetOne will return a hash type. // Currently the first is returned, but it could be // improved to return the strongest. func (h HashSet) GetOne() HashType { v := int(h) i := uint(0) for v != 0 { if v&1 != 0 { return HashType(1 << i) } i++ v >>= 1 } return HashType(HashNone) } // Array returns an array of all hash types in the set func (h HashSet) Array() (ht []HashType) { v := int(h) i := uint(0) for v != 0 { if v&1 != 0 { ht = append(ht, HashType(1<>= 1 } return ht } // Count returns the number of hash types in the set func (h HashSet) Count() int { if int(h) == 0 { return 0 } // credit: https://code.google.com/u/arnehormann/ x := uint64(h) x -= (x >> 1) & 0x5555555555555555 x = (x>>2)&0x3333333333333333 + x&0x3333333333333333 x += x >> 4 x &= 0x0f0f0f0f0f0f0f0f x *= 0x0101010101010101 return int(x >> 56) } // String returns a string representation of the hash set. // The function will panic if it contains an unknown type. func (h HashSet) String() string { a := h.Array() var r []string for _, v := range a { r = append(r, v.String()) } return "[" + strings.Join(r, ", ") + "]" }