blob: a7b2fa420b7b99d8c57849723c32ef89f86edb3e [file] [log] [blame]
// Copyright 2013 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "components/policy/core/common/schema.h"
#include <limits.h>
#include <stddef.h>
#include <algorithm>
#include <climits>
#include <map>
#include <memory>
#include <ostream>
#include <set>
#include <sstream>
#include <string>
#include <utility>
#include "base/check_op.h"
#include "base/compiler_specific.h"
#include "base/containers/contains.h"
#include "base/containers/flat_set.h"
#include "base/json/json_reader.h"
#include "base/memory/ptr_util.h"
#include "base/notreached.h"
#include "base/strings/strcat.h"
#include "base/strings/string_number_conversions.h"
#include "base/strings/stringprintf.h"
#include "base/types/expected.h"
#include "base/types/expected_macros.h"
#include "base/values.h"
#include "components/policy/core/common/json_schema_constants.h"
#include "components/policy/core/common/schema_internal.h"
#include "third_party/re2/src/re2/re2.h"
namespace schema = json_schema_constants;
namespace policy {
using internal::PropertiesNode;
using internal::PropertyNode;
using internal::RestrictionNode;
using internal::SchemaData;
using internal::SchemaNode;
std::string ErrorPathToString(const std::string& policy_name,
PolicyErrorPath error_path) {
if (error_path.empty())
return std::string();
std::stringstream error_path_string{policy_name};
error_path_string << policy_name;
for (auto& entry : error_path) {
if (absl::holds_alternative<int>(entry)) {
error_path_string << "[" << absl::get<int>(entry) << "]";
} else if (absl::holds_alternative<std::string>(entry)) {
error_path_string << "." << absl::get<std::string>(entry);
}
}
return error_path_string.str();
}
const char kSensitiveValueMask[] = "********";
namespace {
struct ReferencesAndIDs {
// Maps schema "id" attributes to the corresponding SchemaNode index.
std::map<std::string, short> id_map;
// List of pairs of references to be assigned later. The string is the "id"
// whose corresponding index should be stored in the pointer, once all the IDs
// are available.
std::vector<std::pair<std::string, short*>> reference_list;
};
// Sizes for the storage arrays. These are calculated in advance so that the
// arrays don't have to be resized during parsing, which would invalidate
// pointers into their contents (i.e. string's c_str() and address of indices
// for "$ref" attributes).
struct StorageSizes {
size_t strings = 0;
size_t schema_nodes = 0;
size_t property_nodes = 0;
size_t properties_nodes = 0;
size_t restriction_nodes = 0;
size_t required_properties = 0;
size_t int_enums = 0;
size_t string_enums = 0;
};
// An invalid index, indicating that a node is not present; similar to a NULL
// pointer.
const short kInvalid = -1;
// Maps a schema key to the corresponding base::Value::Type
struct SchemaKeyToValueType {
const char* key;
base::Value::Type type;
};
// Allowed types and their base::Value::Type equivalent. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kSchemaTypesToValueTypes[] = {
{schema::kArray, base::Value::Type::LIST},
{schema::kBoolean, base::Value::Type::BOOLEAN},
{schema::kInteger, base::Value::Type::INTEGER},
{schema::kNumber, base::Value::Type::DOUBLE},
{schema::kObject, base::Value::Type::DICT},
{schema::kString, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kSchemaTypesToValueTypesEnd =
kSchemaTypesToValueTypes + std::size(kSchemaTypesToValueTypes);
// Allowed attributes and types for type 'array'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForArray[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kId, base::Value::Type::STRING},
{schema::kItems, base::Value::Type::DICT},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForArrayEnd =
kAttributesAndTypesForArray + std::size(kAttributesAndTypesForArray);
// Allowed attributes and types for type 'boolean'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForBoolean[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kId, base::Value::Type::STRING},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForBooleanEnd =
kAttributesAndTypesForBoolean + std::size(kAttributesAndTypesForBoolean);
// Allowed attributes and types for type 'integer'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForInteger[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kEnum, base::Value::Type::LIST},
{schema::kId, base::Value::Type::STRING},
{schema::kMaximum, base::Value::Type::DOUBLE},
{schema::kMinimum, base::Value::Type::DOUBLE},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForIntegerEnd =
kAttributesAndTypesForInteger + std::size(kAttributesAndTypesForInteger);
// Allowed attributes and types for type 'number'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForNumber[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kId, base::Value::Type::STRING},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForNumberEnd =
kAttributesAndTypesForNumber + std::size(kAttributesAndTypesForNumber);
// Allowed attributes and types for type 'object'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForObject[] = {
{schema::kAdditionalProperties, base::Value::Type::DICT},
{schema::kDescription, base::Value::Type::STRING},
{schema::kId, base::Value::Type::STRING},
{schema::kPatternProperties, base::Value::Type::DICT},
{schema::kProperties, base::Value::Type::DICT},
{schema::kRequired, base::Value::Type::LIST},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForObjectEnd =
kAttributesAndTypesForObject + std::size(kAttributesAndTypesForObject);
// Allowed attributes and types for $ref. These are ordered alphabetically to
// perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForRef[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kRef, base::Value::Type::STRING},
{schema::kTitle, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForRefEnd =
kAttributesAndTypesForRef + std::size(kAttributesAndTypesForRef);
// Allowed attributes and types for type 'string'. These are ordered
// alphabetically to perform binary search.
const SchemaKeyToValueType kAttributesAndTypesForString[] = {
{schema::kDescription, base::Value::Type::STRING},
{schema::kEnum, base::Value::Type::LIST},
{schema::kId, base::Value::Type::STRING},
{schema::kPattern, base::Value::Type::STRING},
{schema::kSensitiveValue, base::Value::Type::BOOLEAN},
{schema::kTitle, base::Value::Type::STRING},
{schema::kType, base::Value::Type::STRING},
};
const SchemaKeyToValueType* kAttributesAndTypesForStringEnd =
kAttributesAndTypesForString + std::size(kAttributesAndTypesForString);
// Helper for std::lower_bound.
bool CompareToString(const SchemaKeyToValueType& entry,
const std::string& key) {
return entry.key < key;
}
// Returns true if a SchemaKeyToValueType with key==|schema_key| can be found in
// the array represented by |begin| and |end|. If so, |value_type| will be set
// to the SchemaKeyToValueType value type.
bool MapSchemaKeyToValueType(const std::string& schema_key,
const SchemaKeyToValueType* begin,
const SchemaKeyToValueType* end,
base::Value::Type* value_type) {
const SchemaKeyToValueType* entry =
std::lower_bound(begin, end, schema_key, CompareToString);
if (entry == end || entry->key != schema_key)
return false;
if (value_type)
*value_type = entry->type;
return true;
}
// Shorthand method for |SchemaTypeToValueType()| with
// |kSchemaTypesToValueTypes|.
bool SchemaTypeToValueType(const std::string& schema_type,
base::Value::Type* value_type) {
return MapSchemaKeyToValueType(schema_type, kSchemaTypesToValueTypes,
kSchemaTypesToValueTypesEnd, value_type);
}
bool StrategyAllowUnknown(SchemaOnErrorStrategy strategy) {
return strategy != SCHEMA_STRICT;
}
bool StrategyAllowInvalidListEntry(SchemaOnErrorStrategy strategy) {
return strategy == SCHEMA_ALLOW_UNKNOWN_AND_INVALID_LIST_ENTRY;
}
bool StrategyAllowUnknownWithoutWarning(SchemaOnErrorStrategy strategy) {
return strategy == SCHEMA_ALLOW_UNKNOWN_WITHOUT_WARNING;
}
void SchemaErrorFound(PolicyErrorPath* out_error_path,
std::string* out_error,
const std::string& msg) {
if (out_error_path)
*out_error_path = {};
if (out_error)
*out_error = msg;
}
void AddListIndexPrefixToPath(int index, PolicyErrorPath* path) {
if (path) {
path->emplace(path->begin(), index);
}
}
void AddDictKeyPrefixToPath(const std::string& key, PolicyErrorPath* path) {
if (path) {
path->emplace(path->begin(), key);
}
}
bool IgnoreUnknownAttributes(int options) {
return (options & kSchemaOptionsIgnoreUnknownAttributes);
}
// Check that the value's type and the expected type are equal. We also allow
// integers when expecting doubles.
bool CheckType(const base::Value* value, base::Value::Type expected_type) {
return value->type() == expected_type ||
(value->is_int() && expected_type == base::Value::Type::DOUBLE);
}
// Returns true if |type| is supported as schema's 'type' value.
bool IsValidType(const std::string& type) {
return MapSchemaKeyToValueType(type, kSchemaTypesToValueTypes,
kSchemaTypesToValueTypesEnd, nullptr);
}
// Validate that |dict| only contains attributes that are allowed for the
// corresponding value of 'type'. Also ensure that all of those attributes are
// of the expected type. |options| can be used to ignore unknown attributes.
base::expected<void, std::string> ValidateAttributesAndTypes(
const base::Value::Dict& dict,
const std::string& type,
int options) {
const SchemaKeyToValueType* begin = nullptr;
const SchemaKeyToValueType* end = nullptr;
if (type == schema::kArray) {
begin = kAttributesAndTypesForArray;
end = kAttributesAndTypesForArrayEnd;
} else if (type == schema::kBoolean) {
begin = kAttributesAndTypesForBoolean;
end = kAttributesAndTypesForBooleanEnd;
} else if (type == schema::kInteger) {
begin = kAttributesAndTypesForInteger;
end = kAttributesAndTypesForIntegerEnd;
} else if (type == schema::kNumber) {
begin = kAttributesAndTypesForNumber;
end = kAttributesAndTypesForNumberEnd;
} else if (type == schema::kObject) {
begin = kAttributesAndTypesForObject;
end = kAttributesAndTypesForObjectEnd;
} else if (type == schema::kRef) {
begin = kAttributesAndTypesForRef;
end = kAttributesAndTypesForRefEnd;
} else if (type == schema::kString) {
begin = kAttributesAndTypesForString;
end = kAttributesAndTypesForStringEnd;
} else {
NOTREACHED_IN_MIGRATION()
<< "Type should be a valid schema type or '$ref'.";
}
base::Value::Type expected_type = base::Value::Type::NONE;
for (auto it : dict) {
if (MapSchemaKeyToValueType(it.first, begin, end, &expected_type)) {
if (!CheckType(&it.second, expected_type)) {
return base::unexpected(base::StringPrintf(
"Invalid type for attribute '%s'", it.first.c_str()));
}
} else if (!IgnoreUnknownAttributes(options)) {
return base::unexpected(
base::StringPrintf("Unknown attribute '%s'", it.first.c_str()));
}
}
return base::ok();
}
// Validates that |enum_list| is a list and its items are all of type |type|.
base::expected<void, std::string> ValidateEnum(const base::Value* enum_list,
const std::string& type) {
if (!enum_list->is_list() || enum_list->GetList().empty()) {
return base::unexpected("Attribute 'enum' must be a non-empty list.");
}
base::Value::Type expected_item_type = base::Value::Type::NONE;
MapSchemaKeyToValueType(type, kSchemaTypesToValueTypes,
kSchemaTypesToValueTypesEnd, &expected_item_type);
for (const base::Value& item : enum_list->GetList()) {
if (item.type() != expected_item_type) {
return base::unexpected(base::StringPrintf(
"Attribute 'enum' for type '%s' contains items with invalid types",
type.c_str()));
}
}
return base::ok();
}
// Forward declaration (used in ValidateProperties).
base::expected<void, std::string> IsValidSchema(const base::Value::Dict& dict,
int options);
// Validates that the values in the |properties| dict are valid schemas.
base::expected<void, std::string> ValidateProperties(
const base::Value::Dict& properties,
int options) {
for (auto dict_it : properties) {
if (!dict_it.second.is_dict()) {
return base::unexpected(base::StringPrintf(
"Schema for property '%s' must be a dict.", dict_it.first.c_str()));
}
RETURN_IF_ERROR(IsValidSchema(dict_it.second.GetDict(), options));
}
return base::ok();
}
base::expected<void, std::string> IsFieldTypeObject(
const base::Value& field,
const std::string& field_name) {
if (!field.is_dict()) {
return base::unexpected(base::StringPrintf("Field '%s' must be an object.",
field_name.c_str()));
}
return base::ok();
}
// Checks whether the passed dict is a valid schema. See
// |kAllowedAttributesAndTypes| for a list of supported types, supported
// attributes and their expected types. Values for 'minimum' and 'maximum' for
// type 'integer' can be of type int or double. Referenced IDs ($ref) are not
// checked for existence and IDs are not checked for duplicates. The 'pattern'
// attribute and keys for 'patternProperties' are not checked for valid regular
// expression syntax. Invalid regular expressions will cause a value validation
// error.
base::expected<void, std::string> IsValidSchema(const base::Value::Dict& dict,
int options) {
// Validate '$ref'.
if (dict.contains(schema::kRef)) {
return ValidateAttributesAndTypes(dict, schema::kRef, options);
}
// Validate 'type'.
if (!dict.contains(schema::kType)) {
return base::unexpected("Each schema must have a 'type' or '$ref'.");
}
const std::string* type = dict.FindString(schema::kType);
if (!type) {
return base::unexpected("Attribute 'type' must be a string.");
}
const std::string& type_string = *type;
if (!IsValidType(type_string)) {
return base::unexpected(
base::StringPrintf("Unknown type '%s'.", type_string.c_str()));
}
// Validate attributes and expected types.
RETURN_IF_ERROR(ValidateAttributesAndTypes(dict, type_string, options));
// Validate 'enum' attribute.
if (type_string == schema::kString || type_string == schema::kInteger) {
if (const base::Value* enum_list = dict.Find(schema::kEnum)) {
RETURN_IF_ERROR(ValidateEnum(enum_list, type_string));
}
}
// TODO(b/341873894): Refactor type validation to helper functions.
if (type_string == schema::kInteger) {
// Validate 'minimum' > 'maximum'.
const std::optional<double> minimum_value =
dict.FindDouble(schema::kMinimum);
const std::optional<double> maximum_value =
dict.FindDouble(schema::kMaximum);
if (minimum_value && maximum_value) {
if (minimum_value.value() > maximum_value.value()) {
return base::unexpected(
base::StringPrintf("Invalid range specified [%f;%f].",
minimum_value.value(), maximum_value.value()));
}
}
} else if (type_string == schema::kArray) {
// Validate type 'array'.
const base::Value* items = dict.Find(schema::kItems);
if (!items || !items->is_dict()) {
return base::unexpected(
"Schema of type 'array' must have a schema in 'items' of type "
"dictionary.");
}
RETURN_IF_ERROR(IsValidSchema(items->GetDict(), options));
} else if (type_string == schema::kObject) {
// Validate type 'object'.
const base::Value* properties = dict.Find(schema::kProperties);
if (properties) {
RETURN_IF_ERROR(IsFieldTypeObject(*properties, schema::kProperties));
RETURN_IF_ERROR(ValidateProperties(properties->GetDict(), options));
}
if (const base::Value* pattern_properties =
dict.Find(schema::kPatternProperties)) {
RETURN_IF_ERROR(
IsFieldTypeObject(*pattern_properties, schema::kPatternProperties));
RETURN_IF_ERROR(
ValidateProperties(pattern_properties->GetDict(), options));
}
if (const base::Value* additional_properties =
dict.Find(schema::kAdditionalProperties)) {
RETURN_IF_ERROR(IsFieldTypeObject(*additional_properties,
schema::kAdditionalProperties));
RETURN_IF_ERROR(IsValidSchema(additional_properties->GetDict(), options));
}
if (const base::Value::List* required = dict.FindList(schema::kRequired)) {
for (const base::Value& item : *required) {
if (!item.is_string()) {
return base::unexpected(
"Attribute 'required' may only contain strings.");
}
const std::string property_name = item.GetString();
if (!properties || !properties->GetDict().contains(property_name)) {
return base::unexpected(base::StringPrintf(
"Attribute 'required' contains unknown property '%s'.",
property_name.c_str()));
}
}
}
}
return base::ok();
}
} // namespace
// Contains the internal data representation of a Schema. This can either wrap
// a SchemaData owned elsewhere (currently used to wrap the Chrome schema, which
// is generated at compile time), or it can own its own SchemaData.
class Schema::InternalStorage
: public base::RefCountedThreadSafe<InternalStorage> {
public:
InternalStorage(const InternalStorage&) = delete;
InternalStorage& operator=(const InternalStorage&) = delete;
static scoped_refptr<const InternalStorage> Wrap(const SchemaData* data);
static base::expected<scoped_refptr<const InternalStorage>, std::string>
ParseSchema(const base::Value::Dict& schema);
const SchemaData* data() const { return &schema_data_; }
const SchemaNode* root_node() const { return schema(0); }
// Returns the validation_schema root node if one was generated, or nullptr.
const SchemaNode* validation_schema_root_node() const {
return schema_data_.validation_schema_root_index >= 0
? schema(schema_data_.validation_schema_root_index)
: nullptr;
}
const SchemaNode* schema(int index) const {
DCHECK_GE(index, 0);
return schema_data_.schema_nodes + index;
}
const PropertiesNode* properties(int index) const {
DCHECK_GE(index, 0);
return schema_data_.properties_nodes + index;
}
const PropertyNode* property(int index) const {
DCHECK_GE(index, 0);
return schema_data_.property_nodes + index;
}
const RestrictionNode* restriction(int index) const {
DCHECK_GE(index, 0);
return schema_data_.restriction_nodes + index;
}
const char* const* required_property(int index) const {
DCHECK_GE(index, 0);
return schema_data_.required_properties + index;
}
const int* int_enums(int index) const {
DCHECK_GE(index, 0);
return schema_data_.int_enums + index;
}
const char* const* string_enums(int index) const {
DCHECK_GE(index, 0);
return schema_data_.string_enums + index;
}
// Compiles regular expression |pattern|. The result is cached and will be
// returned directly next time.
re2::RE2* CompileRegex(const std::string& pattern) const;
private:
friend class base::RefCountedThreadSafe<InternalStorage>;
InternalStorage();
~InternalStorage();
// Determines the expected |sizes| of the storage for the representation
// of |schema|.
static void DetermineStorageSizes(const base::Value::Dict& schema,
StorageSizes* sizes);
// Parses the JSON schema in |schema|.
//
// If |schema| has a "$ref" attribute then a pending reference is appended
// to the |reference_list|, and nothing else is done.
//
// Otherwise, |index| gets assigned the index of the corresponding SchemaNode
// in |schema_nodes_|. If the |schema| contains an "id" then that ID is mapped
// to the |index| in the |id_map|.
//
// If |schema| is invalid, it returns an error reason.
base::expected<void, std::string> Parse(const base::Value::Dict& schema,
short* index,
ReferencesAndIDs* references_and_ids);
// Helper for Parse() that gets an already assigned |schema_node| instead of
// an |index| pointer.
base::expected<void, std::string> ParseDictionary(
const base::Value::Dict& schema,
SchemaNode* schema_node,
ReferencesAndIDs* references_and_ids);
// Helper for Parse() that gets an already assigned |schema_node| instead of
// an |index| pointer.
base::expected<void, std::string> ParseList(
const base::Value::Dict& schema,
SchemaNode* schema_node,
ReferencesAndIDs* references_and_ids);
base::expected<void, std::string> ParseEnum(const base::Value::Dict& schema,
base::Value::Type type,
SchemaNode* schema_node);
base::expected<void, std::string> ParseRangedInt(
const base::Value::Dict& schema,
SchemaNode* schema_node);
base::expected<void, std::string> ParseStringPattern(
const base::Value::Dict& schema,
SchemaNode* schema_node);
// Assigns the IDs in |id_map| to the pending references in the
// |reference_list|. If an ID is missing then |error| is set and false is
// returned; otherwise returns true.
static bool ResolveReferences(const ReferencesAndIDs& references_and_ids,
std::string* error);
// Sets |has_sensitive_children| for all |SchemaNode|s in |schema_nodes_|.
void FindSensitiveChildren();
// Returns true iff the node at |index| has sensitive child elements or
// contains a sensitive value itself.
bool FindSensitiveChildrenRecursive(int index,
std::set<int>* handled_schema_nodes);
// Cache for CompileRegex(), will memorize return value of every call to
// CompileRegex() and return results directly next time.
mutable std::map<std::string, std::unique_ptr<re2::RE2>> regex_cache_;
SchemaData schema_data_;
std::vector<std::string> strings_;
std::vector<SchemaNode> schema_nodes_;
std::vector<PropertyNode> property_nodes_;
std::vector<PropertiesNode> properties_nodes_;
std::vector<RestrictionNode> restriction_nodes_;
std::vector<const char*> required_properties_;
std::vector<int> int_enums_;
std::vector<const char*> string_enums_;
};
Schema::InternalStorage::InternalStorage() = default;
Schema::InternalStorage::~InternalStorage() = default;
// static
scoped_refptr<const Schema::InternalStorage> Schema::InternalStorage::Wrap(
const SchemaData* data) {
InternalStorage* storage = new InternalStorage();
storage->schema_data_ = *data;
return storage;
}
// static
base::expected<scoped_refptr<const Schema::InternalStorage>, std::string>
Schema::InternalStorage::ParseSchema(const base::Value::Dict& schema) {
// Determine the sizes of the storage arrays and reserve the capacity before
// starting to append nodes and strings. This is important to prevent the
// arrays from being reallocated, which would invalidate the c_str() pointers
// and the addresses of indices to fix.
StorageSizes sizes;
DetermineStorageSizes(schema, &sizes);
scoped_refptr<InternalStorage> storage = new InternalStorage();
storage->strings_.reserve(sizes.strings);
storage->schema_nodes_.reserve(sizes.schema_nodes);
storage->property_nodes_.reserve(sizes.property_nodes);
storage->properties_nodes_.reserve(sizes.properties_nodes);
storage->restriction_nodes_.reserve(sizes.restriction_nodes);
storage->required_properties_.reserve(sizes.required_properties);
storage->int_enums_.reserve(sizes.int_enums);
storage->string_enums_.reserve(sizes.string_enums);
short root_index = kInvalid;
ReferencesAndIDs references_and_ids;
RETURN_IF_ERROR(storage->Parse(schema, &root_index, &references_and_ids));
if (root_index == kInvalid) {
return base::unexpected("The main schema can't have a $ref");
}
// None of this should ever happen without having been already detected.
// But, if it does happen, then it will lead to corrupted memory; drop
// everything in that case.
if (root_index != 0 || sizes.strings != storage->strings_.size() ||
sizes.schema_nodes != storage->schema_nodes_.size() ||
sizes.property_nodes != storage->property_nodes_.size() ||
sizes.properties_nodes != storage->properties_nodes_.size() ||
sizes.restriction_nodes != storage->restriction_nodes_.size() ||
sizes.required_properties != storage->required_properties_.size() ||
sizes.int_enums != storage->int_enums_.size() ||
sizes.string_enums != storage->string_enums_.size()) {
return base::unexpected(
"Failed to parse the schema due to a Chrome bug. Please file a "
"new issue at http://crbug.com");
}
std::string error;
if (!ResolveReferences(references_and_ids, &error)) {
return base::unexpected(error);
}
storage->FindSensitiveChildren();
SchemaData* data = &storage->schema_data_;
data->schema_nodes = storage->schema_nodes_.data();
data->property_nodes = storage->property_nodes_.data();
data->properties_nodes = storage->properties_nodes_.data();
data->restriction_nodes = storage->restriction_nodes_.data();
data->required_properties = storage->required_properties_.data();
data->int_enums = storage->int_enums_.data();
data->string_enums = storage->string_enums_.data();
data->validation_schema_root_index = -1;
return base::ok(std::move(storage));
}
re2::RE2* Schema::InternalStorage::CompileRegex(
const std::string& pattern) const {
auto it = regex_cache_.find(pattern);
if (it == regex_cache_.end()) {
std::unique_ptr<re2::RE2> compiled(new re2::RE2(pattern));
re2::RE2* compiled_ptr = compiled.get();
regex_cache_.insert(std::make_pair(pattern, std::move(compiled)));
return compiled_ptr;
}
return it->second.get();
}
// static
void Schema::InternalStorage::DetermineStorageSizes(
const base::Value::Dict& schema,
StorageSizes* sizes) {
if (schema.FindString(schema::kRef)) {
// Schemas with a "$ref" attribute don't take additional storage.
return;
}
base::Value::Type type = base::Value::Type::NONE;
const std::string* type_string = schema.FindString(schema::kType);
if (!type_string || !SchemaTypeToValueType(*type_string, &type)) {
// This schema is invalid.
return;
}
sizes->schema_nodes++;
if (type == base::Value::Type::LIST) {
const base::Value* items = schema.Find(schema::kItems);
if (items && items->is_dict()) {
DetermineStorageSizes(items->GetDict(), sizes);
}
} else if (type == base::Value::Type::DICT) {
sizes->properties_nodes++;
const base::Value* additional_properties =
schema.Find(schema::kAdditionalProperties);
if (additional_properties && additional_properties->is_dict()) {
DetermineStorageSizes(additional_properties->GetDict(), sizes);
}
const base::Value::Dict* properties = schema.FindDict(schema::kProperties);
if (properties) {
for (auto property : *properties) {
if (property.second.is_dict()) {
DetermineStorageSizes(property.second.GetDict(), sizes);
}
sizes->strings++;
sizes->property_nodes++;
}
}
const base::Value::Dict* pattern_properties =
schema.FindDict(schema::kPatternProperties);
if (pattern_properties) {
for (auto pattern_property : *pattern_properties) {
if (pattern_property.second.is_dict()) {
DetermineStorageSizes(pattern_property.second.GetDict(), sizes);
}
sizes->strings++;
sizes->property_nodes++;
}
}
const base::Value::List* required_properties =
schema.FindList(schema::kRequired);
if (required_properties) {
sizes->strings += required_properties->size();
sizes->required_properties += required_properties->size();
}
} else if (schema.FindList(schema::kEnum)) {
const base::Value::List* possible_values = schema.FindList(schema::kEnum);
if (possible_values) {
size_t num_possible_values = possible_values->size();
if (type == base::Value::Type::INTEGER) {
sizes->int_enums += num_possible_values;
} else if (type == base::Value::Type::STRING) {
sizes->string_enums += num_possible_values;
sizes->strings += num_possible_values;
}
sizes->restriction_nodes++;
}
} else if (type == base::Value::Type::INTEGER) {
if (schema.contains(schema::kMinimum) ||
schema.contains(schema::kMaximum)) {
sizes->restriction_nodes++;
}
} else if (type == base::Value::Type::STRING) {
if (schema.contains(schema::kPattern)) {
sizes->strings++;
sizes->string_enums++;
sizes->restriction_nodes++;
}
}
}
base::expected<void, std::string> Schema::InternalStorage::Parse(
const base::Value::Dict& schema,
short* index,
ReferencesAndIDs* references_and_ids) {
const std::string* ref = schema.FindString(schema::kRef);
if (ref) {
if (schema.FindString(schema::kId)) {
return base::unexpected("Schemas with a $ref can't have an id");
}
references_and_ids->reference_list.emplace_back(*ref, index);
return base::ok();
}
const std::string* type_string = schema.FindString(schema::kType);
if (!type_string) {
return base::unexpected("The schema type must be declared.");
}
base::Value::Type type = base::Value::Type::NONE;
if (!SchemaTypeToValueType(*type_string, &type)) {
return base::unexpected("Type not supported: " + *type_string);
}
if (schema_nodes_.size() > std::numeric_limits<short>::max()) {
return base::unexpected(
"Can't have more than " +
base::NumberToString(std::numeric_limits<short>::max()) +
" schema nodes.");
}
*index = static_cast<short>(schema_nodes_.size());
schema_nodes_.push_back(
{.type = type,
.extra = kInvalid,
.is_sensitive_value =
schema.FindBool(schema::kSensitiveValue).value_or(false)});
SchemaNode* schema_node = &schema_nodes_.back();
if (type == base::Value::Type::DICT) {
RETURN_IF_ERROR(ParseDictionary(schema, schema_node, references_and_ids));
} else if (type == base::Value::Type::LIST) {
RETURN_IF_ERROR(ParseList(schema, schema_node, references_and_ids));
} else if (schema.contains(schema::kEnum)) {
RETURN_IF_ERROR(ParseEnum(schema, type, schema_node));
} else if (schema.contains(schema::kPattern)) {
RETURN_IF_ERROR(ParseStringPattern(schema, schema_node));
} else if (schema.contains(schema::kMinimum) ||
schema.contains(schema::kMaximum)) {
if (type != base::Value::Type::INTEGER) {
return base::unexpected("Only integers can have minimum and maximum");
}
RETURN_IF_ERROR(ParseRangedInt(schema, schema_node));
}
const std::string* id = schema.FindString(schema::kId);
if (id) {
auto& id_map = references_and_ids->id_map;
if (base::Contains(id_map, *id)) {
return base::unexpected("Duplicated id: " + *id);
}
id_map[*id] = *index;
}
return base::ok();
}
base::expected<void, std::string> Schema::InternalStorage::ParseDictionary(
const base::Value::Dict& schema,
SchemaNode* schema_node,
ReferencesAndIDs* references_and_ids) {
int extra = static_cast<int>(properties_nodes_.size());
properties_nodes_.push_back({.additional = kInvalid});
schema_node->extra = extra;
const base::Value::Dict* additional_properties =
schema.FindDict(schema::kAdditionalProperties);
if (additional_properties) {
RETURN_IF_ERROR(Parse(*additional_properties,
&properties_nodes_[extra].additional,
references_and_ids));
}
properties_nodes_[extra].begin = static_cast<int>(property_nodes_.size());
const base::Value::Dict* properties = schema.FindDict(schema::kProperties);
if (properties) {
// This and below reserves nodes for all of the |properties|, and makes sure
// they are contiguous. Recursive calls to Parse() will append after these
// elements.
property_nodes_.resize(property_nodes_.size() + properties->size());
}
properties_nodes_[extra].end = static_cast<int>(property_nodes_.size());
const base::Value::Dict* pattern_properties =
schema.FindDict(schema::kPatternProperties);
if (pattern_properties) {
property_nodes_.resize(property_nodes_.size() + pattern_properties->size());
}
properties_nodes_[extra].pattern_end =
static_cast<int>(property_nodes_.size());
if (properties != nullptr) {
int base_index = properties_nodes_[extra].begin;
int index = base_index;
for (auto property : *properties) {
strings_.push_back(property.first);
property_nodes_[index].key = strings_.back().c_str();
if (!property.second.is_dict()) {
return base::unexpected(std::string());
}
RETURN_IF_ERROR(Parse(property.second.GetDict(),
&property_nodes_[index].schema,
references_and_ids));
++index;
}
CHECK_EQ(static_cast<int>(properties->size()), index - base_index);
}
if (pattern_properties != nullptr) {
int base_index = properties_nodes_[extra].end;
int index = base_index;
for (auto pattern_property : *pattern_properties) {
re2::RE2* compiled_regex = CompileRegex(pattern_property.first);
if (!compiled_regex->ok()) {
return base::unexpected(
"/" + pattern_property.first +
"/ is a invalid regex: " + compiled_regex->error());
}
strings_.push_back(pattern_property.first);
property_nodes_[index].key = strings_.back().c_str();
if (!pattern_property.second.is_dict()) {
return base::unexpected(std::string());
}
RETURN_IF_ERROR(Parse(pattern_property.second.GetDict(),
&property_nodes_[index].schema,
references_and_ids));
++index;
}
CHECK_EQ(static_cast<int>(pattern_properties->size()), index - base_index);
}
properties_nodes_[extra].required_begin = required_properties_.size();
const base::Value::List* required_properties =
schema.FindList(schema::kRequired);
if (required_properties) {
for (const base::Value& val : *required_properties) {
strings_.push_back(val.GetString());
required_properties_.push_back(strings_.back().c_str());
}
}
properties_nodes_[extra].required_end = required_properties_.size();
if (properties_nodes_[extra].begin == properties_nodes_[extra].pattern_end) {
properties_nodes_[extra].begin = kInvalid;
properties_nodes_[extra].end = kInvalid;
properties_nodes_[extra].pattern_end = kInvalid;
properties_nodes_[extra].required_begin = kInvalid;
properties_nodes_[extra].required_end = kInvalid;
}
return base::ok();
}
base::expected<void, std::string> Schema::InternalStorage::ParseList(
const base::Value::Dict& schema,
SchemaNode* schema_node,
ReferencesAndIDs* references_and_ids) {
const base::Value::Dict* items = schema.FindDict(schema::kItems);
if (!items) {
return base::unexpected(
"Arrays must declare a single schema for their items.");
}
return Parse(*items, &schema_node->extra, references_and_ids);
}
base::expected<void, std::string> Schema::InternalStorage::ParseEnum(
const base::Value::Dict& schema,
base::Value::Type type,
SchemaNode* schema_node) {
const base::Value::List* possible_values = schema.FindList(schema::kEnum);
if (!possible_values) {
return base::unexpected("Enum attribute must be a list value");
}
if (possible_values->empty()) {
return base::unexpected("Enum attribute must be non-empty");
}
int offset_begin;
int offset_end;
if (type == base::Value::Type::INTEGER) {
offset_begin = static_cast<int>(int_enums_.size());
for (const auto& possible_value : *possible_values) {
if (!possible_value.is_int()) {
return base::unexpected("Invalid enumeration member type");
}
int_enums_.push_back(possible_value.GetInt());
}
offset_end = static_cast<int>(int_enums_.size());
} else if (type == base::Value::Type::STRING) {
offset_begin = static_cast<int>(string_enums_.size());
for (const auto& possible_value : *possible_values) {
if (!possible_value.is_string()) {
return base::unexpected("Invalid enumeration member type");
}
strings_.push_back(possible_value.GetString());
string_enums_.push_back(strings_.back().c_str());
}
offset_end = static_cast<int>(string_enums_.size());
} else {
return base::unexpected(
"Enumeration is only supported for integer and string.");
}
schema_node->extra = static_cast<int>(restriction_nodes_.size());
restriction_nodes_.push_back(RestrictionNode{
.enumeration_restriction = RestrictionNode::EnumerationRestriction{
.offset_begin = offset_begin, .offset_end = offset_end}});
return base::ok();
}
base::expected<void, std::string> Schema::InternalStorage::ParseRangedInt(
const base::Value::Dict& schema,
SchemaNode* schema_node) {
int min_value = schema.FindInt(schema::kMinimum).value_or(INT_MIN);
int max_value = schema.FindInt(schema::kMaximum).value_or(INT_MAX);
if (min_value > max_value) {
return base::unexpected("Invalid range restriction for int type.");
}
schema_node->extra = static_cast<int>(restriction_nodes_.size());
restriction_nodes_.push_back(
RestrictionNode{.ranged_restriction = RestrictionNode::RangedRestriction{
.max_value = max_value, .min_value = min_value}});
return base::ok();
}
base::expected<void, std::string> Schema::InternalStorage::ParseStringPattern(
const base::Value::Dict& schema,
SchemaNode* schema_node) {
const std::string* pattern = schema.FindString(schema::kPattern);
if (!pattern) {
return base::unexpected("Schema pattern must be a string.");
}
re2::RE2* compiled_regex = CompileRegex(*pattern);
if (!compiled_regex->ok()) {
return base::unexpected("/" + *pattern +
"/ is invalid regex: " + compiled_regex->error());
}
int index = static_cast<int>(string_enums_.size());
strings_.push_back(*pattern);
string_enums_.push_back(strings_.back().c_str());
schema_node->extra = static_cast<int>(restriction_nodes_.size());
restriction_nodes_.push_back(RestrictionNode{
.string_pattern_restriction = RestrictionNode::StringPatternRestriction{
.pattern_index = index, .pattern_index_backup = index}});
return base::ok();
}
// static
bool Schema::InternalStorage::ResolveReferences(
const ReferencesAndIDs& references_and_ids,
std::string* error) {
const auto& reference_list = references_and_ids.reference_list;
const auto& id_map = references_and_ids.id_map;
for (auto& ref : reference_list) {
auto id = id_map.find(ref.first);
if (id == id_map.end()) {
*error = "Invalid $ref: " + ref.first;
return false;
}
*ref.second = id->second;
}
return true;
}
void Schema::InternalStorage::FindSensitiveChildren() {
if (schema_nodes_.empty())
return;
std::set<int> handled_schema_nodes;
FindSensitiveChildrenRecursive(0, &handled_schema_nodes);
}
bool Schema::InternalStorage::FindSensitiveChildrenRecursive(
int index,
std::set<int>* handled_schema_nodes) {
DCHECK(static_cast<unsigned long>(index) < schema_nodes_.size());
SchemaNode& schema_node = schema_nodes_[index];
if (handled_schema_nodes->find(index) != handled_schema_nodes->end())
return schema_node.has_sensitive_children || schema_node.is_sensitive_value;
handled_schema_nodes->insert(index);
bool has_sensitive_children = false;
if (schema_node.type == base::Value::Type::DICT) {
const PropertiesNode& properties_node =
properties_nodes_[schema_node.extra];
// Iterate through properties and patternProperties.
for (int i = properties_node.begin; i < properties_node.pattern_end; ++i) {
int sub_index = property_nodes_[i].schema;
has_sensitive_children |=
FindSensitiveChildrenRecursive(sub_index, handled_schema_nodes);
}
if (properties_node.additional != kInvalid) {
has_sensitive_children |= FindSensitiveChildrenRecursive(
properties_node.additional, handled_schema_nodes);
}
} else if (schema_node.type == base::Value::Type::LIST) {
int sub_index = schema_node.extra;
has_sensitive_children |=
FindSensitiveChildrenRecursive(sub_index, handled_schema_nodes);
}
schema_node.has_sensitive_children = has_sensitive_children;
return schema_node.has_sensitive_children || schema_node.is_sensitive_value;
}
Schema::Iterator::Iterator(const scoped_refptr<const InternalStorage>& storage,
const PropertiesNode* node) {
if (node->begin == kInvalid || node->end == kInvalid) {
it_ = nullptr;
end_ = nullptr;
} else {
storage_ = storage;
it_ = storage->property(node->begin);
end_ = storage->property(node->end);
}
}
Schema::Iterator::Iterator(const Iterator& iterator)
: storage_(iterator.storage_), it_(iterator.it_), end_(iterator.end_) {}
Schema::Iterator::~Iterator() = default;
Schema::Iterator& Schema::Iterator::operator=(const Iterator& iterator) {
storage_ = iterator.storage_;
it_ = iterator.it_;
end_ = iterator.end_;
return *this;
}
bool Schema::Iterator::IsAtEnd() const {
return it_ == end_;
}
void Schema::Iterator::Advance() {
DCHECK(it_);
++it_;
}
const char* Schema::Iterator::key() const {
return it_->key;
}
Schema Schema::Iterator::schema() const {
return Schema(storage_, storage_->schema(it_->schema));
}
Schema::Schema() : node_(nullptr) {}
Schema::Schema(const scoped_refptr<const InternalStorage>& storage,
const SchemaNode* node)
: storage_(storage), node_(node) {}
Schema::Schema(const Schema& schema)
: storage_(schema.storage_), node_(schema.node_) {}
Schema::~Schema() = default;
Schema& Schema::operator=(const Schema& schema) {
storage_ = schema.storage_;
node_ = schema.node_;
return *this;
}
// static
Schema Schema::Wrap(const SchemaData* data) {
scoped_refptr<const InternalStorage> storage = InternalStorage::Wrap(data);
return Schema(storage, storage->root_node());
}
bool Schema::Validate(const base::Value& value,
SchemaOnErrorStrategy strategy,
PolicyErrorPath* out_error_path,
std::string* out_error) const {
if (!valid()) {
SchemaErrorFound(out_error_path, out_error, "The schema is invalid.");
return false;
}
if (value.type() != type()) {
// Allow the integer to double promotion. Note that range restriction on
// double is not supported now.
if (value.is_int() && type() == base::Value::Type::DOUBLE) {
return true;
}
SchemaErrorFound(
out_error_path, out_error,
base::StringPrintf(
"Policy type mismatch: expected: \"%s\", actual: \"%s\".",
base::Value::GetTypeName(type()),
base::Value::GetTypeName(value.type())));
return false;
}
if (value.is_dict()) {
base::flat_set<std::string> present_properties;
for (auto dict_item : value.GetDict()) {
SchemaList schema_list = GetMatchingProperties(dict_item.first);
if (schema_list.empty()) {
// Unknown property was detected.
if (!StrategyAllowUnknownWithoutWarning(strategy)) {
SchemaErrorFound(out_error_path, out_error,
"Unknown property: " + dict_item.first);
}
if (!StrategyAllowUnknown(strategy))
return false;
} else {
for (const auto& subschema : schema_list) {
std::string new_error;
const bool validation_result = subschema.Validate(
dict_item.second, strategy, out_error_path, &new_error);
if (!new_error.empty()) {
AddDictKeyPrefixToPath(dict_item.first, out_error_path);
if (out_error)
*out_error = std::move(new_error);
}
if (!validation_result) {
// Invalid property was detected.
return false;
}
}
present_properties.insert(dict_item.first);
}
}
for (const auto& required_property : GetRequiredProperties()) {
if (base::Contains(present_properties, required_property))
continue;
SchemaErrorFound(
out_error_path, out_error,
"Missing or invalid required property: " + required_property);
return false;
}
} else if (value.is_list()) {
for (size_t index = 0; index < value.GetList().size(); ++index) {
const base::Value& list_item = value.GetList()[index];
std::string new_error;
const bool validation_result =
GetItems().Validate(list_item, strategy, out_error_path, &new_error);
if (!new_error.empty()) {
AddListIndexPrefixToPath(index, out_error_path);
if (out_error)
*out_error = std::move(new_error);
}
if (!validation_result && !StrategyAllowInvalidListEntry(strategy))
return false; // Invalid list item was detected.
}
} else if (value.is_int()) {
if (node_->extra != kInvalid &&
!ValidateIntegerRestriction(node_->extra, value.GetInt())) {
SchemaErrorFound(out_error_path, out_error, "Invalid value for integer");
return false;
}
} else if (value.is_string()) {
if (node_->extra != kInvalid &&
!ValidateStringRestriction(node_->extra, value.GetString().c_str())) {
SchemaErrorFound(out_error_path, out_error, "Invalid value for string");
return false;
}
}
return true;
}
bool Schema::Normalize(base::Value* value,
SchemaOnErrorStrategy strategy,
PolicyErrorPath* out_error_path,
std::string* out_error,
bool* out_changed) const {
if (!valid()) {
SchemaErrorFound(out_error_path, out_error, "The schema is invalid.");
return false;
}
if (value->type() != type()) {
// Allow the integer to double promotion. Note that range restriction on
// double is not supported now.
if (value->is_int() && type() == base::Value::Type::DOUBLE) {
return true;
}
SchemaErrorFound(
out_error_path, out_error,
base::StringPrintf(
"Policy type mismatch: expected: \"%s\", actual: \"%s\".",
base::Value::GetTypeName(type()),
base::Value::GetTypeName(value->type())));
return false;
}
if (value->is_dict()) {
base::flat_set<std::string> present_properties;
std::vector<std::string> drop_list; // Contains the keys to drop.
for (auto dict_item : value->GetDict()) {
SchemaList schema_list = GetMatchingProperties(dict_item.first);
if (schema_list.empty()) {
// Unknown property was detected.
if (!StrategyAllowUnknownWithoutWarning(strategy)) {
SchemaErrorFound(out_error_path, out_error,
"Unknown property: " + dict_item.first);
}
if (!StrategyAllowUnknown(strategy))
return false;
if (!StrategyAllowUnknownWithoutWarning(strategy)) {
drop_list.push_back(dict_item.first);
}
} else {
for (const auto& subschema : schema_list) {
std::string new_error;
const bool normalization_result =
subschema.Normalize(&dict_item.second, strategy, out_error_path,
&new_error, out_changed);
if (!new_error.empty()) {
AddDictKeyPrefixToPath(dict_item.first, out_error_path);
if (out_error)
*out_error = std::move(new_error);
}
if (!normalization_result) {
// Invalid property was detected.
return false;
}
}
present_properties.insert(dict_item.first);
}
}
for (const auto& required_property : GetRequiredProperties()) {
if (base::Contains(present_properties, required_property))
continue;
SchemaErrorFound(
out_error_path, out_error,
"Missing or invalid required property: " + required_property);
return false;
}
if (out_changed && !drop_list.empty())
*out_changed = true;
for (const auto& drop_key : drop_list)
value->GetDict().Remove(drop_key);
return true;
} else if (value->is_list()) {
base::Value::List& list = value->GetList();
// Instead of removing invalid list items afterwards, we push valid items
// forward in the list by overriding invalid items. The next free position
// is indicated by |write_index|, which gets increased for every valid item.
// At the end |list| is resized to |write_index|'s size.
size_t write_index = 0;
for (size_t index = 0; index < list.size(); ++index) {
base::Value& list_item = list[index];
std::string new_error;
const bool normalization_result = GetItems().Normalize(
&list_item, strategy, out_error_path, &new_error, out_changed);
if (!new_error.empty()) {
AddListIndexPrefixToPath(index, out_error_path);
if (out_error)
*out_error = new_error;
}
if (!normalization_result) {
// Invalid list item was detected.
if (!StrategyAllowInvalidListEntry(strategy))
return false;
} else {
if (write_index != index)
list[write_index] = std::move(list_item);
++write_index;
}
}
if (out_changed && write_index < list.size())
*out_changed = true;
while (write_index < list.size()) {
list.erase(list.end() - 1);
}
return true;
}
return Validate(*value, strategy, out_error_path, out_error);
}
void Schema::MaskSensitiveValues(base::Value* value) const {
if (!valid())
return;
MaskSensitiveValuesRecursive(value);
}
// static
base::expected<Schema, std::string> Schema::Parse(const std::string& content) {
// Validate as a generic JSON schema, and ignore unknown attributes; they
// may become used in a future version of the schema format.
ASSIGN_OR_RETURN(auto dict,
Schema::ParseToDictAndValidate(
content, kSchemaOptionsIgnoreUnknownAttributes));
// Validate the main type.
const std::string* type = dict.FindString(schema::kType);
if (!type || *type != schema::kObject) {
return base::unexpected(
"The main schema must have a type attribute with \"object\" value.");
}
// Checks for invalid attributes at the top-level.
if (dict.contains(schema::kAdditionalProperties) ||
dict.contains(schema::kPatternProperties)) {
return base::unexpected(
"\"additionalProperties\" and \"patternProperties\" are not "
"supported at the main schema.");
}
ASSIGN_OR_RETURN(auto storage, InternalStorage::ParseSchema(dict));
return base::ok(Schema(storage, storage->root_node()));
}
// static
base::expected<base::Value::Dict, std::string> Schema::ParseToDictAndValidate(
const std::string& schema,
int validator_options) {
ASSIGN_OR_RETURN(
auto json,
base::JSONReader::ReadAndReturnValueWithError(
schema, base::JSONParserOptions::JSON_ALLOW_TRAILING_COMMAS |
base::JSONParserOptions::JSON_PARSE_CHROMIUM_EXTENSIONS),
[](auto e) { return std::move(e.ToString()); });
if (!json.is_dict()) {
return base::unexpected("Schema must be a JSON object");
}
RETURN_IF_ERROR(IsValidSchema(json.GetDict(), validator_options));
return base::ok(std::move(json).TakeDict());
}
base::Value::Type Schema::type() const {
CHECK(valid());
return node_->type;
}
Schema::Iterator Schema::GetPropertiesIterator() const {
CHECK(valid());
CHECK_EQ(base::Value::Type::DICT, type());
return Iterator(storage_, storage_->properties(node_->extra));
}
namespace {
bool CompareKeys(const PropertyNode& node, const std::string& key) {
return node.key < key;
}
} // namespace
Schema Schema::GetKnownProperty(const std::string& key) const {
CHECK(valid());
CHECK_EQ(base::Value::Type::DICT, type());
const PropertiesNode* node = storage_->properties(node_->extra);
if (node->begin == kInvalid || node->end == kInvalid)
return Schema();
const PropertyNode* begin = storage_->property(node->begin);
const PropertyNode* end = storage_->property(node->end);
const PropertyNode* it = std::lower_bound(begin, end, key, CompareKeys);
if (it != end && it->key == key)
return Schema(storage_, storage_->schema(it->schema));
return Schema();
}
Schema Schema::GetAdditionalProperties() const {
CHECK(valid());
CHECK_EQ(base::Value::Type::DICT, type());
const PropertiesNode* node = storage_->properties(node_->extra);
if (node->additional == kInvalid)
return Schema();
return Schema(storage_, storage_->schema(node->additional));
}
SchemaList Schema::GetPatternProperties(const std::string& key) const {
CHECK(valid());
CHECK_EQ(base::Value::Type::DICT, type());
const PropertiesNode* node = storage_->properties(node_->extra);
if (node->end == kInvalid || node->pattern_end == kInvalid)
return {};
const PropertyNode* begin = storage_->property(node->end);
const PropertyNode* end = storage_->property(node->pattern_end);
SchemaList matching_properties;
for (const PropertyNode* it = begin; it != end; ++it) {
if (re2::RE2::PartialMatch(key, *storage_->CompileRegex(it->key))) {
matching_properties.push_back(
Schema(storage_, storage_->schema(it->schema)));
}
}
return matching_properties;
}
std::vector<std::string> Schema::GetRequiredProperties() const {
CHECK(valid());
CHECK_EQ(base::Value::Type::DICT, type());
const PropertiesNode* node = storage_->properties(node_->extra);
if (node->required_begin == kInvalid || node->required_end == kInvalid)
return {};
const size_t begin = node->required_begin;
const size_t end = node->required_end;
return std::vector<std::string>(storage_->required_property(begin),
storage_->required_property(end));
}
Schema Schema::GetProperty(const std::string& key) const {
Schema schema = GetKnownProperty(key);
if (schema.valid())
return schema;
return GetAdditionalProperties();
}
SchemaList Schema::GetMatchingProperties(const std::string& key) const {
SchemaList schema_list;
Schema known_property = GetKnownProperty(key);
if (known_property.valid())
schema_list.push_back(known_property);
SchemaList pattern_properties = GetPatternProperties(key);
schema_list.insert(schema_list.end(), pattern_properties.begin(),
pattern_properties.end());
if (schema_list.empty()) {
Schema additional_property = GetAdditionalProperties();
if (additional_property.valid())
schema_list.push_back(additional_property);
}
return schema_list;
}
Schema Schema::GetItems() const {
CHECK(valid());
CHECK_EQ(base::Value::Type::LIST, type());
if (node_->extra == kInvalid)
return Schema();
return Schema(storage_, storage_->schema(node_->extra));
}
bool Schema::ValidateIntegerRestriction(int index, int value) const {
const RestrictionNode* rnode = storage_->restriction(index);
if (rnode->ranged_restriction.min_value <=
rnode->ranged_restriction.max_value) {
return rnode->ranged_restriction.min_value <= value &&
rnode->ranged_restriction.max_value >= value;
} else {
for (int i = rnode->enumeration_restriction.offset_begin;
i < rnode->enumeration_restriction.offset_end; ++i) {
if (*storage_->int_enums(i) == value)
return true;
}
return false;
}
}
bool Schema::ValidateStringRestriction(int index, const char* str) const {
const RestrictionNode* rnode = storage_->restriction(index);
if (rnode->enumeration_restriction.offset_begin <
rnode->enumeration_restriction.offset_end) {
for (int i = rnode->enumeration_restriction.offset_begin;
i < rnode->enumeration_restriction.offset_end; ++i) {
if (strcmp(*storage_->string_enums(i), str) == 0)
return true;
}
return false;
} else {
int pattern_index = rnode->string_pattern_restriction.pattern_index;
DCHECK(pattern_index ==
rnode->string_pattern_restriction.pattern_index_backup);
re2::RE2* regex =
storage_->CompileRegex(*storage_->string_enums(pattern_index));
return re2::RE2::PartialMatch(str, *regex);
}
}
void Schema::MaskSensitiveValuesRecursive(base::Value* value) const {
if (IsSensitiveValue()) {
*value = base::Value(kSensitiveValueMask);
return;
}
if (!HasSensitiveChildren())
return;
if (value->type() != type())
return;
if (value->is_dict()) {
for (auto [key, sub_value] : value->GetDict()) {
SchemaList schema_list = GetMatchingProperties(key);
for (const auto& schema_item : schema_list)
schema_item.MaskSensitiveValuesRecursive(&sub_value);
}
} else if (value->is_list()) {
for (auto& list_elem : value->GetList())
GetItems().MaskSensitiveValuesRecursive(&list_elem);
}
}
Schema Schema::GetValidationSchema() const {
CHECK(valid());
const SchemaNode* validation_schema_root_node =
storage_->validation_schema_root_node();
if (!validation_schema_root_node)
return Schema();
return Schema(storage_, validation_schema_root_node);
}
bool Schema::IsSensitiveValue() const {
CHECK(valid());
// This is safe because |node_| is guaranteed to have been returned from
// |storage_| and |storage_->root_node()| always returns to the |SchemaNode|
// with index 0.
int index = node_ - storage_->root_node();
const SchemaNode* schema_node = storage_->schema(index);
if (!schema_node)
return false;
return schema_node->is_sensitive_value;
}
bool Schema::HasSensitiveChildren() const {
CHECK(valid());
// This is safe because |node_| is guaranteed to have been returned from
// |storage_| and |storage_->root_node()| always returns to the |SchemaNode|
// with index 0.
int index = node_ - storage_->root_node();
const SchemaNode* schema_node = storage_->schema(index);
if (!schema_node)
return false;
return schema_node->has_sensitive_children;
}
} // namespace policy