// Protocol Buffers - Google's data interchange format // Copyright 2023 Google LLC. All rights reserved. // // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file or at // https://developers.google.com/open-source/licenses/bsd #include "upb/wire/decode.h" #include #include #include #include #include #include "upb/base/descriptor_constants.h" #include "upb/base/internal/endian.h" #include "upb/base/string_view.h" #include "upb/hash/common.h" #include "upb/mem/arena.h" #include "upb/message/array.h" #include "upb/message/internal/accessors.h" #include "upb/message/internal/array.h" #include "upb/message/internal/extension.h" #include "upb/message/internal/map.h" #include "upb/message/internal/map_entry.h" #include "upb/message/internal/message.h" #include "upb/message/internal/tagged_ptr.h" #include "upb/message/map.h" #include "upb/message/message.h" #include "upb/message/tagged_ptr.h" #include "upb/mini_table/enum.h" #include "upb/mini_table/extension.h" #include "upb/mini_table/extension_registry.h" #include "upb/mini_table/field.h" #include "upb/mini_table/internal/field.h" #include "upb/mini_table/internal/message.h" #include "upb/mini_table/internal/size_log2.h" #include "upb/mini_table/message.h" #include "upb/mini_table/sub.h" #include "upb/port/atomic.h" #include "upb/wire/encode.h" #include "upb/wire/eps_copy_input_stream.h" #include "upb/wire/internal/constants.h" #include "upb/wire/internal/decoder.h" #include "upb/wire/reader.h" // Must be last. #include "upb/port/def.inc" // A few fake field types for our tables. enum { kUpb_FakeFieldType_FieldNotFound = 0, kUpb_FakeFieldType_MessageSetItem = 19, }; // DecodeOp: an action to be performed for a wire-type/field-type combination. enum { // Special ops: we don't write data to regular fields for these. kUpb_DecodeOp_UnknownField = -1, kUpb_DecodeOp_MessageSetItem = -2, // Scalar-only ops. kUpb_DecodeOp_Scalar1Byte = 0, kUpb_DecodeOp_Scalar4Byte = 2, kUpb_DecodeOp_Scalar8Byte = 3, kUpb_DecodeOp_Enum = 1, // Scalar/repeated ops. kUpb_DecodeOp_String = 4, kUpb_DecodeOp_Bytes = 5, kUpb_DecodeOp_SubMessage = 6, // Repeated-only ops (also see macros below). kUpb_DecodeOp_PackedEnum = 13, }; // For packed fields it is helpful to be able to recover the lg2 of the data // size from the op. #define OP_FIXPCK_LG2(n) (n + 5) /* n in [2, 3] => op in [7, 8] */ #define OP_VARPCK_LG2(n) (n + 9) /* n in [0, 2, 3] => op in [9, 11, 12] */ typedef union { bool bool_val; uint32_t uint32_val; uint64_t uint64_val; uint32_t size; } wireval; // Ideally these two functions should take the owning MiniTable pointer as a // first argument, then we could just put them in mini_table/message.h as nice // clean getters. But we don't have that so instead we gotta write these // Frankenfunctions that take an array of subtables. // TODO: Move these to mini_table/ anyway since there are other places // that could use them. // Returns the MiniTable corresponding to a given MiniTableField // from an array of MiniTableSubs. static const upb_MiniTable* _upb_MiniTableSubs_MessageByField( const upb_MiniTableSub* subs, const upb_MiniTableField* field) { return upb_MiniTableSub_Message(subs[field->UPB_PRIVATE(submsg_index)]); } // Returns the MiniTableEnum corresponding to a given MiniTableField // from an array of MiniTableSub. static const upb_MiniTableEnum* _upb_MiniTableSubs_EnumByField( const upb_MiniTableSub* subs, const upb_MiniTableField* field) { return upb_MiniTableSub_Enum(subs[field->UPB_PRIVATE(submsg_index)]); } static const char* _upb_Decoder_DecodeMessage(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTable* layout); UPB_NORETURN static void* _upb_Decoder_ErrorJmp(upb_Decoder* d, upb_DecodeStatus status) { UPB_ASSERT(status != kUpb_DecodeStatus_Ok); d->status = status; UPB_LONGJMP(d->err, 1); } const char* _upb_FastDecoder_ErrorJmp(upb_Decoder* d, int status) { UPB_ASSERT(status != kUpb_DecodeStatus_Ok); d->status = status; UPB_LONGJMP(d->err, 1); return NULL; } static void _upb_Decoder_VerifyUtf8(upb_Decoder* d, const char* buf, int len) { if (!_upb_Decoder_VerifyUtf8Inline(buf, len)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_BadUtf8); } } static bool _upb_Decoder_Reserve(upb_Decoder* d, upb_Array* arr, size_t elem) { bool need_realloc = arr->UPB_PRIVATE(capacity) - arr->UPB_PRIVATE(size) < elem; if (need_realloc && !UPB_PRIVATE(_upb_Array_Realloc)( arr, arr->UPB_PRIVATE(size) + elem, &d->arena)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } return need_realloc; } typedef struct { const char* ptr; uint64_t val; } _upb_DecodeLongVarintReturn; UPB_NOINLINE static _upb_DecodeLongVarintReturn _upb_Decoder_DecodeLongVarint( const char* ptr, uint64_t val) { _upb_DecodeLongVarintReturn ret = {NULL, 0}; uint64_t byte; for (int i = 1; i < 10; i++) { byte = (uint8_t)ptr[i]; val += (byte - 1) << (i * 7); if (!(byte & 0x80)) { ret.ptr = ptr + i + 1; ret.val = val; return ret; } } return ret; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeVarint(upb_Decoder* d, const char* ptr, uint64_t* val) { uint64_t byte = (uint8_t)*ptr; if (UPB_LIKELY((byte & 0x80) == 0)) { *val = byte; return ptr + 1; } else { _upb_DecodeLongVarintReturn res = _upb_Decoder_DecodeLongVarint(ptr, byte); if (!res.ptr) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); *val = res.val; return res.ptr; } } UPB_FORCEINLINE const char* _upb_Decoder_DecodeTag(upb_Decoder* d, const char* ptr, uint32_t* val) { uint64_t byte = (uint8_t)*ptr; if (UPB_LIKELY((byte & 0x80) == 0)) { *val = byte; return ptr + 1; } else { const char* start = ptr; _upb_DecodeLongVarintReturn res = _upb_Decoder_DecodeLongVarint(ptr, byte); if (!res.ptr || res.ptr - start > 5 || res.val > UINT32_MAX) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } *val = res.val; return res.ptr; } } UPB_FORCEINLINE const char* upb_Decoder_DecodeSize(upb_Decoder* d, const char* ptr, uint32_t* size) { uint64_t size64; ptr = _upb_Decoder_DecodeVarint(d, ptr, &size64); if (size64 >= INT32_MAX || !upb_EpsCopyInputStream_CheckSize(&d->input, ptr, (int)size64)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } *size = size64; return ptr; } static void _upb_Decoder_MungeInt32(wireval* val) { if (!upb_IsLittleEndian()) { /* The next stage will memcpy(dst, &val, 4) */ val->uint32_val = val->uint64_val; } } static void _upb_Decoder_Munge(int type, wireval* val) { switch (type) { case kUpb_FieldType_Bool: val->bool_val = val->uint64_val != 0; break; case kUpb_FieldType_SInt32: { uint32_t n = val->uint64_val; val->uint32_val = (n >> 1) ^ -(int32_t)(n & 1); break; } case kUpb_FieldType_SInt64: { uint64_t n = val->uint64_val; val->uint64_val = (n >> 1) ^ -(int64_t)(n & 1); break; } case kUpb_FieldType_Int32: case kUpb_FieldType_UInt32: case kUpb_FieldType_Enum: _upb_Decoder_MungeInt32(val); break; } } static upb_Message* _upb_Decoder_NewSubMessage(upb_Decoder* d, const upb_MiniTableSub* subs, const upb_MiniTableField* field, upb_TaggedMessagePtr* target) { const upb_MiniTable* subl = _upb_MiniTableSubs_MessageByField(subs, field); UPB_ASSERT(subl); upb_Message* msg = _upb_Message_New(subl, &d->arena); if (!msg) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); // Extensions should not be unlinked. A message extension should not be // registered until its sub-message type is available to be linked. bool is_empty = UPB_PRIVATE(_upb_MiniTable_IsEmpty)(subl); bool is_extension = field->UPB_PRIVATE(mode) & kUpb_LabelFlags_IsExtension; UPB_ASSERT(!(is_empty && is_extension)); if (is_empty && !(d->options & kUpb_DecodeOption_ExperimentalAllowUnlinked)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_UnlinkedSubMessage); } upb_TaggedMessagePtr tagged = UPB_PRIVATE(_upb_TaggedMessagePtr_Pack)(msg, is_empty); memcpy(target, &tagged, sizeof(tagged)); return msg; } static upb_Message* _upb_Decoder_ReuseSubMessage( upb_Decoder* d, const upb_MiniTableSub* subs, const upb_MiniTableField* field, upb_TaggedMessagePtr* target) { upb_TaggedMessagePtr tagged = *target; const upb_MiniTable* subl = _upb_MiniTableSubs_MessageByField(subs, field); UPB_ASSERT(subl); if (!upb_TaggedMessagePtr_IsEmpty(tagged) || UPB_PRIVATE(_upb_MiniTable_IsEmpty)(subl)) { return UPB_PRIVATE(_upb_TaggedMessagePtr_GetMessage)(tagged); } // We found an empty message from a previous parse that was performed before // this field was linked. But it is linked now, so we want to allocate a new // message of the correct type and promote data into it before continuing. upb_Message* existing = UPB_PRIVATE(_upb_TaggedMessagePtr_GetEmptyMessage)(tagged); upb_Message* promoted = _upb_Decoder_NewSubMessage(d, subs, field, target); size_t size; const char* unknown = upb_Message_GetUnknown(existing, &size); upb_DecodeStatus status = upb_Decode(unknown, size, promoted, subl, d->extreg, d->options, &d->arena); if (status != kUpb_DecodeStatus_Ok) _upb_Decoder_ErrorJmp(d, status); return promoted; } static const char* _upb_Decoder_ReadString(upb_Decoder* d, const char* ptr, int size, upb_StringView* str) { const char* str_ptr = ptr; ptr = upb_EpsCopyInputStream_ReadString(&d->input, &str_ptr, size, &d->arena); if (!ptr) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); str->data = str_ptr; str->size = size; return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_RecurseSubMessage(upb_Decoder* d, const char* ptr, upb_Message* submsg, const upb_MiniTable* subl, uint32_t expected_end_group) { if (--d->depth < 0) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_MaxDepthExceeded); } ptr = _upb_Decoder_DecodeMessage(d, ptr, submsg, subl); d->depth++; if (d->end_group != expected_end_group) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeSubMessage(upb_Decoder* d, const char* ptr, upb_Message* submsg, const upb_MiniTableSub* subs, const upb_MiniTableField* field, int size) { int saved_delta = upb_EpsCopyInputStream_PushLimit(&d->input, ptr, size); const upb_MiniTable* subl = _upb_MiniTableSubs_MessageByField(subs, field); UPB_ASSERT(subl); ptr = _upb_Decoder_RecurseSubMessage(d, ptr, submsg, subl, DECODE_NOGROUP); upb_EpsCopyInputStream_PopLimit(&d->input, ptr, saved_delta); return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeGroup(upb_Decoder* d, const char* ptr, upb_Message* submsg, const upb_MiniTable* subl, uint32_t number) { if (_upb_Decoder_IsDone(d, &ptr)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } ptr = _upb_Decoder_RecurseSubMessage(d, ptr, submsg, subl, number); d->end_group = DECODE_NOGROUP; return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeUnknownGroup(upb_Decoder* d, const char* ptr, uint32_t number) { return _upb_Decoder_DecodeGroup(d, ptr, NULL, NULL, number); } UPB_FORCEINLINE const char* _upb_Decoder_DecodeKnownGroup(upb_Decoder* d, const char* ptr, upb_Message* submsg, const upb_MiniTableSub* subs, const upb_MiniTableField* field) { const upb_MiniTable* subl = _upb_MiniTableSubs_MessageByField(subs, field); UPB_ASSERT(subl); return _upb_Decoder_DecodeGroup(d, ptr, submsg, subl, field->UPB_PRIVATE(number)); } static char* upb_Decoder_EncodeVarint32(uint32_t val, char* ptr) { do { uint8_t byte = val & 0x7fU; val >>= 7; if (val) byte |= 0x80U; *(ptr++) = byte; } while (val); return ptr; } static void _upb_Decoder_AddUnknownVarints(upb_Decoder* d, upb_Message* msg, uint32_t val1, uint32_t val2) { char buf[20]; char* end = buf; end = upb_Decoder_EncodeVarint32(val1, end); end = upb_Decoder_EncodeVarint32(val2, end); if (!UPB_PRIVATE(_upb_Message_AddUnknown)(msg, buf, end - buf, &d->arena)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } } UPB_FORCEINLINE bool _upb_Decoder_CheckEnum(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTableEnum* e, const upb_MiniTableField* field, wireval* val) { const uint32_t v = val->uint32_val; if (UPB_LIKELY(upb_MiniTableEnum_CheckValue(e, v))) return true; // Unrecognized enum goes into unknown fields. // For packed fields the tag could be arbitrarily far in the past, // so we just re-encode the tag and value here. const uint32_t tag = ((uint32_t)field->UPB_PRIVATE(number) << 3) | kUpb_WireType_Varint; upb_Message* unknown_msg = field->UPB_PRIVATE(mode) & kUpb_LabelFlags_IsExtension ? d->unknown_msg : msg; _upb_Decoder_AddUnknownVarints(d, unknown_msg, tag, v); return false; } UPB_NOINLINE static const char* _upb_Decoder_DecodeEnumArray(upb_Decoder* d, const char* ptr, upb_Message* msg, upb_Array* arr, const upb_MiniTableSub* subs, const upb_MiniTableField* field, wireval* val) { const upb_MiniTableEnum* e = _upb_MiniTableSubs_EnumByField(subs, field); if (!_upb_Decoder_CheckEnum(d, ptr, msg, e, field, val)) return ptr; void* mem = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) * 4, void); arr->UPB_PRIVATE(size)++; memcpy(mem, val, 4); return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeFixedPacked(upb_Decoder* d, const char* ptr, upb_Array* arr, wireval* val, const upb_MiniTableField* field, int lg2) { int mask = (1 << lg2) - 1; size_t count = val->size >> lg2; if ((val->size & mask) != 0) { // Length isn't a round multiple of elem size. _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } _upb_Decoder_Reserve(d, arr, count); void* mem = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) << lg2, void); arr->UPB_PRIVATE(size) += count; // Note: if/when the decoder supports multi-buffer input, we will need to // handle buffer seams here. if (upb_IsLittleEndian()) { ptr = upb_EpsCopyInputStream_Copy(&d->input, ptr, mem, val->size); } else { int delta = upb_EpsCopyInputStream_PushLimit(&d->input, ptr, val->size); char* dst = mem; while (!_upb_Decoder_IsDone(d, &ptr)) { if (lg2 == 2) { ptr = upb_WireReader_ReadFixed32(ptr, dst); dst += 4; } else { UPB_ASSERT(lg2 == 3); ptr = upb_WireReader_ReadFixed64(ptr, dst); dst += 8; } } upb_EpsCopyInputStream_PopLimit(&d->input, ptr, delta); } return ptr; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeVarintPacked(upb_Decoder* d, const char* ptr, upb_Array* arr, wireval* val, const upb_MiniTableField* field, int lg2) { int scale = 1 << lg2; int saved_limit = upb_EpsCopyInputStream_PushLimit(&d->input, ptr, val->size); char* out = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) << lg2, void); while (!_upb_Decoder_IsDone(d, &ptr)) { wireval elem; ptr = _upb_Decoder_DecodeVarint(d, ptr, &elem.uint64_val); _upb_Decoder_Munge(field->UPB_PRIVATE(descriptortype), &elem); if (_upb_Decoder_Reserve(d, arr, 1)) { out = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) << lg2, void); } arr->UPB_PRIVATE(size)++; memcpy(out, &elem, scale); out += scale; } upb_EpsCopyInputStream_PopLimit(&d->input, ptr, saved_limit); return ptr; } UPB_NOINLINE static const char* _upb_Decoder_DecodeEnumPacked( upb_Decoder* d, const char* ptr, upb_Message* msg, upb_Array* arr, const upb_MiniTableSub* subs, const upb_MiniTableField* field, wireval* val) { const upb_MiniTableEnum* e = _upb_MiniTableSubs_EnumByField(subs, field); int saved_limit = upb_EpsCopyInputStream_PushLimit(&d->input, ptr, val->size); char* out = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) * 4, void); while (!_upb_Decoder_IsDone(d, &ptr)) { wireval elem; ptr = _upb_Decoder_DecodeVarint(d, ptr, &elem.uint64_val); _upb_Decoder_MungeInt32(&elem); if (!_upb_Decoder_CheckEnum(d, ptr, msg, e, field, &elem)) { continue; } if (_upb_Decoder_Reserve(d, arr, 1)) { out = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) * 4, void); } arr->UPB_PRIVATE(size)++; memcpy(out, &elem, 4); out += 4; } upb_EpsCopyInputStream_PopLimit(&d->input, ptr, saved_limit); return ptr; } static upb_Array* _upb_Decoder_CreateArray(upb_Decoder* d, const upb_MiniTableField* field) { const upb_FieldType field_type = field->UPB_PRIVATE(descriptortype); const size_t lg2 = UPB_PRIVATE(_upb_FieldType_SizeLg2)(field_type); upb_Array* ret = UPB_PRIVATE(_upb_Array_New)(&d->arena, 4, lg2); if (!ret) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); return ret; } static const char* _upb_Decoder_DecodeToArray(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTableSub* subs, const upb_MiniTableField* field, wireval* val, int op) { upb_Array** arrp = UPB_PTR_AT(msg, field->UPB_PRIVATE(offset), void); upb_Array* arr = *arrp; void* mem; if (arr) { _upb_Decoder_Reserve(d, arr, 1); } else { arr = _upb_Decoder_CreateArray(d, field); *arrp = arr; } switch (op) { case kUpb_DecodeOp_Scalar1Byte: case kUpb_DecodeOp_Scalar4Byte: case kUpb_DecodeOp_Scalar8Byte: /* Append scalar value. */ mem = UPB_PTR_AT(upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) << op, void); arr->UPB_PRIVATE(size)++; memcpy(mem, val, 1 << op); return ptr; case kUpb_DecodeOp_String: _upb_Decoder_VerifyUtf8(d, ptr, val->size); /* Fallthrough. */ case kUpb_DecodeOp_Bytes: { /* Append bytes. */ upb_StringView* str = (upb_StringView*)upb_Array_MutableDataPtr(arr) + arr->UPB_PRIVATE(size); arr->UPB_PRIVATE(size)++; return _upb_Decoder_ReadString(d, ptr, val->size, str); } case kUpb_DecodeOp_SubMessage: { /* Append submessage / group. */ upb_TaggedMessagePtr* target = UPB_PTR_AT( upb_Array_MutableDataPtr(arr), arr->UPB_PRIVATE(size) * sizeof(void*), upb_TaggedMessagePtr); upb_Message* submsg = _upb_Decoder_NewSubMessage(d, subs, field, target); arr->UPB_PRIVATE(size)++; if (UPB_UNLIKELY(field->UPB_PRIVATE(descriptortype) == kUpb_FieldType_Group)) { return _upb_Decoder_DecodeKnownGroup(d, ptr, submsg, subs, field); } else { return _upb_Decoder_DecodeSubMessage(d, ptr, submsg, subs, field, val->size); } } case OP_FIXPCK_LG2(2): case OP_FIXPCK_LG2(3): return _upb_Decoder_DecodeFixedPacked(d, ptr, arr, val, field, op - OP_FIXPCK_LG2(0)); case OP_VARPCK_LG2(0): case OP_VARPCK_LG2(2): case OP_VARPCK_LG2(3): return _upb_Decoder_DecodeVarintPacked(d, ptr, arr, val, field, op - OP_VARPCK_LG2(0)); case kUpb_DecodeOp_Enum: return _upb_Decoder_DecodeEnumArray(d, ptr, msg, arr, subs, field, val); case kUpb_DecodeOp_PackedEnum: return _upb_Decoder_DecodeEnumPacked(d, ptr, msg, arr, subs, field, val); default: UPB_UNREACHABLE(); } } static upb_Map* _upb_Decoder_CreateMap(upb_Decoder* d, const upb_MiniTable* entry) { // Maps descriptor type -> upb map size static const uint8_t kSizeInMap[] = { [0] = -1, // invalid descriptor type [kUpb_FieldType_Double] = 8, [kUpb_FieldType_Float] = 4, [kUpb_FieldType_Int64] = 8, [kUpb_FieldType_UInt64] = 8, [kUpb_FieldType_Int32] = 4, [kUpb_FieldType_Fixed64] = 8, [kUpb_FieldType_Fixed32] = 4, [kUpb_FieldType_Bool] = 1, [kUpb_FieldType_String] = UPB_MAPTYPE_STRING, [kUpb_FieldType_Group] = sizeof(void*), [kUpb_FieldType_Message] = sizeof(void*), [kUpb_FieldType_Bytes] = UPB_MAPTYPE_STRING, [kUpb_FieldType_UInt32] = 4, [kUpb_FieldType_Enum] = 4, [kUpb_FieldType_SFixed32] = 4, [kUpb_FieldType_SFixed64] = 8, [kUpb_FieldType_SInt32] = 4, [kUpb_FieldType_SInt64] = 8, }; const upb_MiniTableField* key_field = &entry->UPB_PRIVATE(fields)[0]; const upb_MiniTableField* val_field = &entry->UPB_PRIVATE(fields)[1]; char key_size = kSizeInMap[key_field->UPB_PRIVATE(descriptortype)]; char val_size = kSizeInMap[val_field->UPB_PRIVATE(descriptortype)]; UPB_ASSERT(key_field->UPB_PRIVATE(offset) == offsetof(upb_MapEntry, k)); UPB_ASSERT(val_field->UPB_PRIVATE(offset) == offsetof(upb_MapEntry, v)); upb_Map* ret = _upb_Map_New(&d->arena, key_size, val_size); if (!ret) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); return ret; } static const char* _upb_Decoder_DecodeToMap(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTableSub* subs, const upb_MiniTableField* field, wireval* val) { upb_Map** map_p = UPB_PTR_AT(msg, field->UPB_PRIVATE(offset), upb_Map*); upb_Map* map = *map_p; upb_MapEntry ent; UPB_ASSERT(upb_MiniTableField_Type(field) == kUpb_FieldType_Message); const upb_MiniTable* entry = _upb_MiniTableSubs_MessageByField(subs, field); UPB_ASSERT(entry); UPB_ASSERT(entry->UPB_PRIVATE(field_count) == 2); UPB_ASSERT(upb_MiniTableField_IsScalar(&entry->UPB_PRIVATE(fields)[0])); UPB_ASSERT(upb_MiniTableField_IsScalar(&entry->UPB_PRIVATE(fields)[1])); if (!map) { map = _upb_Decoder_CreateMap(d, entry); *map_p = map; } // Parse map entry. memset(&ent, 0, sizeof(ent)); if (entry->UPB_PRIVATE(fields)[1].UPB_PRIVATE(descriptortype) == kUpb_FieldType_Message || entry->UPB_PRIVATE(fields)[1].UPB_PRIVATE(descriptortype) == kUpb_FieldType_Group) { // Create proactively to handle the case where it doesn't appear. upb_TaggedMessagePtr msg; _upb_Decoder_NewSubMessage(d, entry->UPB_PRIVATE(subs), &entry->UPB_PRIVATE(fields)[1], &msg); ent.v.val = upb_value_uintptr(msg); } ptr = _upb_Decoder_DecodeSubMessage(d, ptr, &ent.message, subs, field, val->size); // check if ent had any unknown fields size_t size; upb_Message_GetUnknown(&ent.message, &size); if (size != 0) { char* buf; size_t size; uint32_t tag = ((uint32_t)field->UPB_PRIVATE(number) << 3) | kUpb_WireType_Delimited; upb_EncodeStatus status = upb_Encode(&ent.message, entry, 0, &d->arena, &buf, &size); if (status != kUpb_EncodeStatus_Ok) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } _upb_Decoder_AddUnknownVarints(d, msg, tag, size); if (!UPB_PRIVATE(_upb_Message_AddUnknown)(msg, buf, size, &d->arena)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } } else { if (_upb_Map_Insert(map, &ent.k, map->key_size, &ent.v, map->val_size, &d->arena) == kUpb_MapInsertStatus_OutOfMemory) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } } return ptr; } static const char* _upb_Decoder_DecodeToSubMessage( upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTableSub* subs, const upb_MiniTableField* field, wireval* val, int op) { void* mem = UPB_PTR_AT(msg, field->UPB_PRIVATE(offset), void); int type = field->UPB_PRIVATE(descriptortype); if (UPB_UNLIKELY(op == kUpb_DecodeOp_Enum) && !_upb_Decoder_CheckEnum(d, ptr, msg, _upb_MiniTableSubs_EnumByField(subs, field), field, val)) { return ptr; } // Set presence if necessary. if (UPB_PRIVATE(_upb_MiniTableField_HasHasbit)(field)) { UPB_PRIVATE(_upb_Message_SetHasbit)(msg, field); } else if (upb_MiniTableField_IsInOneof(field)) { // Oneof case uint32_t* oneof_case = UPB_PRIVATE(_upb_Message_OneofCasePtr)(msg, field); if (op == kUpb_DecodeOp_SubMessage && *oneof_case != field->UPB_PRIVATE(number)) { memset(mem, 0, sizeof(void*)); } *oneof_case = field->UPB_PRIVATE(number); } // Store into message. switch (op) { case kUpb_DecodeOp_SubMessage: { upb_TaggedMessagePtr* submsgp = mem; upb_Message* submsg; if (*submsgp) { submsg = _upb_Decoder_ReuseSubMessage(d, subs, field, submsgp); } else { submsg = _upb_Decoder_NewSubMessage(d, subs, field, submsgp); } if (UPB_UNLIKELY(type == kUpb_FieldType_Group)) { ptr = _upb_Decoder_DecodeKnownGroup(d, ptr, submsg, subs, field); } else { ptr = _upb_Decoder_DecodeSubMessage(d, ptr, submsg, subs, field, val->size); } break; } case kUpb_DecodeOp_String: _upb_Decoder_VerifyUtf8(d, ptr, val->size); /* Fallthrough. */ case kUpb_DecodeOp_Bytes: return _upb_Decoder_ReadString(d, ptr, val->size, mem); case kUpb_DecodeOp_Scalar8Byte: memcpy(mem, val, 8); break; case kUpb_DecodeOp_Enum: case kUpb_DecodeOp_Scalar4Byte: memcpy(mem, val, 4); break; case kUpb_DecodeOp_Scalar1Byte: memcpy(mem, val, 1); break; default: UPB_UNREACHABLE(); } return ptr; } UPB_NOINLINE const char* _upb_Decoder_CheckRequired(upb_Decoder* d, const char* ptr, const upb_Message* msg, const upb_MiniTable* m) { UPB_ASSERT(m->UPB_PRIVATE(required_count)); if (UPB_UNLIKELY(d->options & kUpb_DecodeOption_CheckRequired)) { d->missing_required = !UPB_PRIVATE(_upb_Message_IsInitializedShallow)(msg, m); } return ptr; } UPB_FORCEINLINE bool _upb_Decoder_TryFastDispatch(upb_Decoder* d, const char** ptr, upb_Message* msg, const upb_MiniTable* m) { #if UPB_FASTTABLE if (m && m->UPB_PRIVATE(table_mask) != (unsigned char)-1) { uint16_t tag = _upb_FastDecoder_LoadTag(*ptr); intptr_t table = decode_totable(m); *ptr = _upb_FastDecoder_TagDispatch(d, *ptr, msg, table, 0, tag); return true; } #endif return false; } static const char* upb_Decoder_SkipField(upb_Decoder* d, const char* ptr, uint32_t tag) { int field_number = tag >> 3; int wire_type = tag & 7; switch (wire_type) { case kUpb_WireType_Varint: { uint64_t val; return _upb_Decoder_DecodeVarint(d, ptr, &val); } case kUpb_WireType_64Bit: return ptr + 8; case kUpb_WireType_32Bit: return ptr + 4; case kUpb_WireType_Delimited: { uint32_t size; ptr = upb_Decoder_DecodeSize(d, ptr, &size); return ptr + size; } case kUpb_WireType_StartGroup: return _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number); default: _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } } enum { kStartItemTag = ((kUpb_MsgSet_Item << 3) | kUpb_WireType_StartGroup), kEndItemTag = ((kUpb_MsgSet_Item << 3) | kUpb_WireType_EndGroup), kTypeIdTag = ((kUpb_MsgSet_TypeId << 3) | kUpb_WireType_Varint), kMessageTag = ((kUpb_MsgSet_Message << 3) | kUpb_WireType_Delimited), }; static void upb_Decoder_AddKnownMessageSetItem( upb_Decoder* d, upb_Message* msg, const upb_MiniTableExtension* item_mt, const char* data, uint32_t size) { upb_Extension* ext = UPB_PRIVATE(_upb_Message_GetOrCreateExtension)(msg, item_mt, &d->arena); if (UPB_UNLIKELY(!ext)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } upb_Message* submsg = _upb_Decoder_NewSubMessage( d, &ext->ext->UPB_PRIVATE(sub), &ext->ext->UPB_PRIVATE(field), (upb_TaggedMessagePtr*)&ext->data); upb_DecodeStatus status = upb_Decode( data, size, submsg, upb_MiniTableExtension_GetSubMessage(item_mt), d->extreg, d->options, &d->arena); if (status != kUpb_DecodeStatus_Ok) _upb_Decoder_ErrorJmp(d, status); } static void upb_Decoder_AddUnknownMessageSetItem(upb_Decoder* d, upb_Message* msg, uint32_t type_id, const char* message_data, uint32_t message_size) { char buf[60]; char* ptr = buf; ptr = upb_Decoder_EncodeVarint32(kStartItemTag, ptr); ptr = upb_Decoder_EncodeVarint32(kTypeIdTag, ptr); ptr = upb_Decoder_EncodeVarint32(type_id, ptr); ptr = upb_Decoder_EncodeVarint32(kMessageTag, ptr); ptr = upb_Decoder_EncodeVarint32(message_size, ptr); char* split = ptr; ptr = upb_Decoder_EncodeVarint32(kEndItemTag, ptr); char* end = ptr; if (!UPB_PRIVATE(_upb_Message_AddUnknown)(msg, buf, split - buf, &d->arena) || !UPB_PRIVATE(_upb_Message_AddUnknown)(msg, message_data, message_size, &d->arena) || !UPB_PRIVATE(_upb_Message_AddUnknown)(msg, split, end - split, &d->arena)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } } static void upb_Decoder_AddMessageSetItem(upb_Decoder* d, upb_Message* msg, const upb_MiniTable* t, uint32_t type_id, const char* data, uint32_t size) { const upb_MiniTableExtension* item_mt = upb_ExtensionRegistry_Lookup(d->extreg, t, type_id); if (item_mt) { upb_Decoder_AddKnownMessageSetItem(d, msg, item_mt, data, size); } else { upb_Decoder_AddUnknownMessageSetItem(d, msg, type_id, data, size); } } static const char* upb_Decoder_DecodeMessageSetItem( upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTable* layout) { uint32_t type_id = 0; upb_StringView preserved = {NULL, 0}; typedef enum { kUpb_HaveId = 1 << 0, kUpb_HavePayload = 1 << 1, } StateMask; StateMask state_mask = 0; while (!_upb_Decoder_IsDone(d, &ptr)) { uint32_t tag; ptr = _upb_Decoder_DecodeTag(d, ptr, &tag); switch (tag) { case kEndItemTag: return ptr; case kTypeIdTag: { uint64_t tmp; ptr = _upb_Decoder_DecodeVarint(d, ptr, &tmp); if (state_mask & kUpb_HaveId) break; // Ignore dup. state_mask |= kUpb_HaveId; type_id = tmp; if (state_mask & kUpb_HavePayload) { upb_Decoder_AddMessageSetItem(d, msg, layout, type_id, preserved.data, preserved.size); } break; } case kMessageTag: { uint32_t size; ptr = upb_Decoder_DecodeSize(d, ptr, &size); const char* data = ptr; ptr += size; if (state_mask & kUpb_HavePayload) break; // Ignore dup. state_mask |= kUpb_HavePayload; if (state_mask & kUpb_HaveId) { upb_Decoder_AddMessageSetItem(d, msg, layout, type_id, data, size); } else { // Out of order, we must preserve the payload. preserved.data = data; preserved.size = size; } break; } default: // We do not preserve unexpected fields inside a message set item. ptr = upb_Decoder_SkipField(d, ptr, tag); break; } } _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } static const upb_MiniTableField* _upb_Decoder_FindField(upb_Decoder* d, const upb_MiniTable* t, uint32_t field_number, int* last_field_index) { static upb_MiniTableField none = { 0, 0, 0, 0, kUpb_FakeFieldType_FieldNotFound, 0}; if (t == NULL) return &none; size_t idx = ((size_t)field_number) - 1; // 0 wraps to SIZE_MAX if (idx < t->UPB_PRIVATE(dense_below)) { // Fastest case: index into dense fields. goto found; } if (t->UPB_PRIVATE(dense_below) < t->UPB_PRIVATE(field_count)) { // Linear search non-dense fields. Resume scanning from last_field_index // since fields are usually in order. size_t last = *last_field_index; for (idx = last; idx < t->UPB_PRIVATE(field_count); idx++) { if (t->UPB_PRIVATE(fields)[idx].UPB_PRIVATE(number) == field_number) { goto found; } } for (idx = t->UPB_PRIVATE(dense_below); idx < last; idx++) { if (t->UPB_PRIVATE(fields)[idx].UPB_PRIVATE(number) == field_number) { goto found; } } } if (d->extreg) { switch (t->UPB_PRIVATE(ext)) { case kUpb_ExtMode_Extendable: { const upb_MiniTableExtension* ext = upb_ExtensionRegistry_Lookup(d->extreg, t, field_number); if (ext) return &ext->UPB_PRIVATE(field); break; } case kUpb_ExtMode_IsMessageSet: if (field_number == kUpb_MsgSet_Item) { static upb_MiniTableField item = { 0, 0, 0, 0, kUpb_FakeFieldType_MessageSetItem, 0}; return &item; } break; } } return &none; // Unknown field. found: UPB_ASSERT(t->UPB_PRIVATE(fields)[idx].UPB_PRIVATE(number) == field_number); *last_field_index = idx; return &t->UPB_PRIVATE(fields)[idx]; } static int _upb_Decoder_GetVarintOp(const upb_MiniTableField* field) { static const int8_t kVarintOps[] = { [kUpb_FakeFieldType_FieldNotFound] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Double] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Float] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Int64] = kUpb_DecodeOp_Scalar8Byte, [kUpb_FieldType_UInt64] = kUpb_DecodeOp_Scalar8Byte, [kUpb_FieldType_Int32] = kUpb_DecodeOp_Scalar4Byte, [kUpb_FieldType_Fixed64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Fixed32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Bool] = kUpb_DecodeOp_Scalar1Byte, [kUpb_FieldType_String] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Group] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Message] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Bytes] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_UInt32] = kUpb_DecodeOp_Scalar4Byte, [kUpb_FieldType_Enum] = kUpb_DecodeOp_Enum, [kUpb_FieldType_SFixed32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SFixed64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SInt32] = kUpb_DecodeOp_Scalar4Byte, [kUpb_FieldType_SInt64] = kUpb_DecodeOp_Scalar8Byte, [kUpb_FakeFieldType_MessageSetItem] = kUpb_DecodeOp_UnknownField, }; return kVarintOps[field->UPB_PRIVATE(descriptortype)]; } UPB_FORCEINLINE void _upb_Decoder_CheckUnlinked(upb_Decoder* d, const upb_MiniTable* mt, const upb_MiniTableField* field, int* op) { // If sub-message is not linked, treat as unknown. if (field->UPB_PRIVATE(mode) & kUpb_LabelFlags_IsExtension) return; const upb_MiniTable* mt_sub = _upb_MiniTableSubs_MessageByField(mt->UPB_PRIVATE(subs), field); if ((d->options & kUpb_DecodeOption_ExperimentalAllowUnlinked) || !UPB_PRIVATE(_upb_MiniTable_IsEmpty)(mt_sub)) { return; } #ifndef NDEBUG const upb_MiniTableField* oneof = upb_MiniTable_GetOneof(mt, field); if (oneof) { // All other members of the oneof must be message fields that are also // unlinked. do { UPB_ASSERT(upb_MiniTableField_CType(oneof) == kUpb_CType_Message); const upb_MiniTableSub* oneof_sub = &mt->UPB_PRIVATE(subs)[oneof->UPB_PRIVATE(submsg_index)]; UPB_ASSERT(!oneof_sub); } while (upb_MiniTable_NextOneofField(mt, &oneof)); } #endif // NDEBUG *op = kUpb_DecodeOp_UnknownField; } UPB_FORCEINLINE void _upb_Decoder_MaybeVerifyUtf8(upb_Decoder* d, const upb_MiniTableField* field, int* op) { if ((field->UPB_ONLYBITS(mode) & kUpb_LabelFlags_IsAlternate) && UPB_UNLIKELY(d->options & kUpb_DecodeOption_AlwaysValidateUtf8)) *op = kUpb_DecodeOp_String; } static int _upb_Decoder_GetDelimitedOp(upb_Decoder* d, const upb_MiniTable* mt, const upb_MiniTableField* field) { enum { kRepeatedBase = 19 }; static const int8_t kDelimitedOps[] = { // For non-repeated field type. [kUpb_FakeFieldType_FieldNotFound] = kUpb_DecodeOp_UnknownField, // Field not found. [kUpb_FieldType_Double] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Float] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Int64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_UInt64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Int32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Fixed64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Fixed32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Bool] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_String] = kUpb_DecodeOp_String, [kUpb_FieldType_Group] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Message] = kUpb_DecodeOp_SubMessage, [kUpb_FieldType_Bytes] = kUpb_DecodeOp_Bytes, [kUpb_FieldType_UInt32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_Enum] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SFixed32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SFixed64] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SInt32] = kUpb_DecodeOp_UnknownField, [kUpb_FieldType_SInt64] = kUpb_DecodeOp_UnknownField, [kUpb_FakeFieldType_MessageSetItem] = kUpb_DecodeOp_UnknownField, // For repeated field type. [kRepeatedBase + kUpb_FieldType_Double] = OP_FIXPCK_LG2(3), [kRepeatedBase + kUpb_FieldType_Float] = OP_FIXPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_Int64] = OP_VARPCK_LG2(3), [kRepeatedBase + kUpb_FieldType_UInt64] = OP_VARPCK_LG2(3), [kRepeatedBase + kUpb_FieldType_Int32] = OP_VARPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_Fixed64] = OP_FIXPCK_LG2(3), [kRepeatedBase + kUpb_FieldType_Fixed32] = OP_FIXPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_Bool] = OP_VARPCK_LG2(0), [kRepeatedBase + kUpb_FieldType_String] = kUpb_DecodeOp_String, [kRepeatedBase + kUpb_FieldType_Group] = kUpb_DecodeOp_SubMessage, [kRepeatedBase + kUpb_FieldType_Message] = kUpb_DecodeOp_SubMessage, [kRepeatedBase + kUpb_FieldType_Bytes] = kUpb_DecodeOp_Bytes, [kRepeatedBase + kUpb_FieldType_UInt32] = OP_VARPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_Enum] = kUpb_DecodeOp_PackedEnum, [kRepeatedBase + kUpb_FieldType_SFixed32] = OP_FIXPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_SFixed64] = OP_FIXPCK_LG2(3), [kRepeatedBase + kUpb_FieldType_SInt32] = OP_VARPCK_LG2(2), [kRepeatedBase + kUpb_FieldType_SInt64] = OP_VARPCK_LG2(3), // Omitting kUpb_FakeFieldType_MessageSetItem, because we never emit a // repeated msgset type }; int ndx = field->UPB_PRIVATE(descriptortype); if (upb_MiniTableField_IsArray(field)) ndx += kRepeatedBase; int op = kDelimitedOps[ndx]; if (op == kUpb_DecodeOp_SubMessage) { _upb_Decoder_CheckUnlinked(d, mt, field, &op); } else if (op == kUpb_DecodeOp_Bytes) { _upb_Decoder_MaybeVerifyUtf8(d, field, &op); } return op; } UPB_FORCEINLINE const char* _upb_Decoder_DecodeWireValue(upb_Decoder* d, const char* ptr, const upb_MiniTable* mt, const upb_MiniTableField* field, int wire_type, wireval* val, int* op) { static const unsigned kFixed32OkMask = (1 << kUpb_FieldType_Float) | (1 << kUpb_FieldType_Fixed32) | (1 << kUpb_FieldType_SFixed32); static const unsigned kFixed64OkMask = (1 << kUpb_FieldType_Double) | (1 << kUpb_FieldType_Fixed64) | (1 << kUpb_FieldType_SFixed64); switch (wire_type) { case kUpb_WireType_Varint: ptr = _upb_Decoder_DecodeVarint(d, ptr, &val->uint64_val); *op = _upb_Decoder_GetVarintOp(field); _upb_Decoder_Munge(field->UPB_PRIVATE(descriptortype), val); return ptr; case kUpb_WireType_32Bit: *op = kUpb_DecodeOp_Scalar4Byte; if (((1 << field->UPB_PRIVATE(descriptortype)) & kFixed32OkMask) == 0) { *op = kUpb_DecodeOp_UnknownField; } return upb_WireReader_ReadFixed32(ptr, &val->uint32_val); case kUpb_WireType_64Bit: *op = kUpb_DecodeOp_Scalar8Byte; if (((1 << field->UPB_PRIVATE(descriptortype)) & kFixed64OkMask) == 0) { *op = kUpb_DecodeOp_UnknownField; } return upb_WireReader_ReadFixed64(ptr, &val->uint64_val); case kUpb_WireType_Delimited: ptr = upb_Decoder_DecodeSize(d, ptr, &val->size); *op = _upb_Decoder_GetDelimitedOp(d, mt, field); return ptr; case kUpb_WireType_StartGroup: val->uint32_val = field->UPB_PRIVATE(number); if (field->UPB_PRIVATE(descriptortype) == kUpb_FieldType_Group) { *op = kUpb_DecodeOp_SubMessage; _upb_Decoder_CheckUnlinked(d, mt, field, op); } else if (field->UPB_PRIVATE(descriptortype) == kUpb_FakeFieldType_MessageSetItem) { *op = kUpb_DecodeOp_MessageSetItem; } else { *op = kUpb_DecodeOp_UnknownField; } return ptr; default: break; } _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); } UPB_FORCEINLINE const char* _upb_Decoder_DecodeKnownField(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTable* layout, const upb_MiniTableField* field, int op, wireval* val) { const upb_MiniTableSub* subs = layout->UPB_PRIVATE(subs); uint8_t mode = field->UPB_PRIVATE(mode); if (UPB_UNLIKELY(mode & kUpb_LabelFlags_IsExtension)) { const upb_MiniTableExtension* ext_layout = (const upb_MiniTableExtension*)field; upb_Extension* ext = UPB_PRIVATE(_upb_Message_GetOrCreateExtension)( msg, ext_layout, &d->arena); if (UPB_UNLIKELY(!ext)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } d->unknown_msg = msg; msg = (upb_Message*)&ext->data; subs = &ext->ext->UPB_PRIVATE(sub); } switch (mode & kUpb_FieldMode_Mask) { case kUpb_FieldMode_Array: return _upb_Decoder_DecodeToArray(d, ptr, msg, subs, field, val, op); case kUpb_FieldMode_Map: return _upb_Decoder_DecodeToMap(d, ptr, msg, subs, field, val); case kUpb_FieldMode_Scalar: return _upb_Decoder_DecodeToSubMessage(d, ptr, msg, subs, field, val, op); default: UPB_UNREACHABLE(); } } static const char* _upb_Decoder_ReverseSkipVarint(const char* ptr, uint32_t val) { uint32_t seen = 0; do { ptr--; seen <<= 7; seen |= *ptr & 0x7f; } while (seen != val); return ptr; } static const char* _upb_Decoder_DecodeUnknownField(upb_Decoder* d, const char* ptr, upb_Message* msg, int field_number, int wire_type, wireval val) { if (field_number == 0) _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_Malformed); // Since unknown fields are the uncommon case, we do a little extra work here // to walk backwards through the buffer to find the field start. This frees // up a register in the fast paths (when the field is known), which leads to // significant speedups in benchmarks. const char* start = ptr; if (wire_type == kUpb_WireType_Delimited) ptr += val.size; if (msg) { switch (wire_type) { case kUpb_WireType_Varint: case kUpb_WireType_Delimited: start--; while (start[-1] & 0x80) start--; break; case kUpb_WireType_32Bit: start -= 4; break; case kUpb_WireType_64Bit: start -= 8; break; default: break; } assert(start == d->debug_valstart); uint32_t tag = ((uint32_t)field_number << 3) | wire_type; start = _upb_Decoder_ReverseSkipVarint(start, tag); assert(start == d->debug_tagstart); if (wire_type == kUpb_WireType_StartGroup) { d->unknown = start; d->unknown_msg = msg; ptr = _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number); start = d->unknown; d->unknown = NULL; } if (!UPB_PRIVATE(_upb_Message_AddUnknown)(msg, start, ptr - start, &d->arena)) { _upb_Decoder_ErrorJmp(d, kUpb_DecodeStatus_OutOfMemory); } } else if (wire_type == kUpb_WireType_StartGroup) { ptr = _upb_Decoder_DecodeUnknownGroup(d, ptr, field_number); } return ptr; } UPB_NOINLINE static const char* _upb_Decoder_DecodeMessage(upb_Decoder* d, const char* ptr, upb_Message* msg, const upb_MiniTable* layout) { int last_field_index = 0; #if UPB_FASTTABLE // The first time we want to skip fast dispatch, because we may have just been // invoked by the fast parser to handle a case that it bailed on. if (!_upb_Decoder_IsDone(d, &ptr)) goto nofast; #endif while (!_upb_Decoder_IsDone(d, &ptr)) { uint32_t tag; const upb_MiniTableField* field; int field_number; int wire_type; wireval val; int op; if (_upb_Decoder_TryFastDispatch(d, &ptr, msg, layout)) break; #if UPB_FASTTABLE nofast: #endif #ifndef NDEBUG d->debug_tagstart = ptr; #endif UPB_ASSERT(ptr < d->input.limit_ptr); ptr = _upb_Decoder_DecodeTag(d, ptr, &tag); field_number = tag >> 3; wire_type = tag & 7; #ifndef NDEBUG d->debug_valstart = ptr; #endif if (wire_type == kUpb_WireType_EndGroup) { d->end_group = field_number; return ptr; } field = _upb_Decoder_FindField(d, layout, field_number, &last_field_index); ptr = _upb_Decoder_DecodeWireValue(d, ptr, layout, field, wire_type, &val, &op); if (op >= 0) { ptr = _upb_Decoder_DecodeKnownField(d, ptr, msg, layout, field, op, &val); } else { switch (op) { case kUpb_DecodeOp_UnknownField: ptr = _upb_Decoder_DecodeUnknownField(d, ptr, msg, field_number, wire_type, val); break; case kUpb_DecodeOp_MessageSetItem: ptr = upb_Decoder_DecodeMessageSetItem(d, ptr, msg, layout); break; } } } return UPB_UNLIKELY(layout && layout->UPB_PRIVATE(required_count)) ? _upb_Decoder_CheckRequired(d, ptr, msg, layout) : ptr; } const char* _upb_FastDecoder_DecodeGeneric(struct upb_Decoder* d, const char* ptr, upb_Message* msg, intptr_t table, uint64_t hasbits, uint64_t data) { (void)data; *(uint32_t*)msg |= hasbits; return _upb_Decoder_DecodeMessage(d, ptr, msg, decode_totablep(table)); } static upb_DecodeStatus _upb_Decoder_DecodeTop(struct upb_Decoder* d, const char* buf, upb_Message* msg, const upb_MiniTable* m) { if (!_upb_Decoder_TryFastDispatch(d, &buf, msg, m)) { _upb_Decoder_DecodeMessage(d, buf, msg, m); } if (d->end_group != DECODE_NOGROUP) return kUpb_DecodeStatus_Malformed; if (d->missing_required) return kUpb_DecodeStatus_MissingRequired; return kUpb_DecodeStatus_Ok; } UPB_NOINLINE const char* _upb_Decoder_IsDoneFallback(upb_EpsCopyInputStream* e, const char* ptr, int overrun) { return _upb_EpsCopyInputStream_IsDoneFallbackInline( e, ptr, overrun, _upb_Decoder_BufferFlipCallback); } static upb_DecodeStatus upb_Decoder_Decode(upb_Decoder* const decoder, const char* const buf, upb_Message* const msg, const upb_MiniTable* const m, upb_Arena* const arena) { if (UPB_SETJMP(decoder->err) == 0) { decoder->status = _upb_Decoder_DecodeTop(decoder, buf, msg, m); } else { UPB_ASSERT(decoder->status != kUpb_DecodeStatus_Ok); } UPB_PRIVATE(_upb_Arena_SwapOut)(arena, &decoder->arena); return decoder->status; } upb_DecodeStatus upb_Decode(const char* buf, size_t size, upb_Message* msg, const upb_MiniTable* mt, const upb_ExtensionRegistry* extreg, int options, upb_Arena* arena) { UPB_ASSERT(!upb_Message_IsFrozen(msg)); upb_Decoder decoder; unsigned depth = (unsigned)options >> 16; upb_EpsCopyInputStream_Init(&decoder.input, &buf, size, options & kUpb_DecodeOption_AliasString); decoder.extreg = extreg; decoder.unknown = NULL; decoder.depth = depth ? depth : kUpb_WireFormat_DefaultDepthLimit; decoder.end_group = DECODE_NOGROUP; decoder.options = (uint16_t)options; decoder.missing_required = false; decoder.status = kUpb_DecodeStatus_Ok; // Violating the encapsulation of the arena for performance reasons. // This is a temporary arena that we swap into and swap out of when we are // done. The temporary arena only needs to be able to handle allocation, // not fuse or free, so it does not need many of the members to be initialized // (particularly parent_or_count). UPB_PRIVATE(_upb_Arena_SwapIn)(&decoder.arena, arena); return upb_Decoder_Decode(&decoder, buf, msg, mt, arena); } upb_DecodeStatus upb_DecodeLengthPrefixed(const char* buf, size_t size, upb_Message* msg, size_t* num_bytes_read, const upb_MiniTable* mt, const upb_ExtensionRegistry* extreg, int options, upb_Arena* arena) { // To avoid needing to make a Decoder just to decode the initial length, // hand-decode the leading varint for the message length here. uint64_t msg_len = 0; for (size_t i = 0;; ++i) { if (i >= size || i > 9) { return kUpb_DecodeStatus_Malformed; } uint64_t b = *buf; buf++; msg_len += (b & 0x7f) << (i * 7); if ((b & 0x80) == 0) { *num_bytes_read = i + 1 + msg_len; break; } } // If the total number of bytes we would read (= the bytes from the varint // plus however many bytes that varint says we should read) is larger then the // input buffer then error as malformed. if (*num_bytes_read > size) { return kUpb_DecodeStatus_Malformed; } if (msg_len > INT32_MAX) { return kUpb_DecodeStatus_Malformed; } return upb_Decode(buf, msg_len, msg, mt, extreg, options, arena); } #undef OP_FIXPCK_LG2 #undef OP_VARPCK_LG2