caddy/context.go
Matt Holt 3c90e370a4
v2: Module documentation; refactor LoadModule(); new caddy struct tags (#2924)
This commit goes a long way toward making automated documentation of
Caddy config and Caddy modules possible. It's a broad, sweeping change,
but mostly internal. It allows us to automatically generate docs for all
Caddy modules (including future third-party ones) and make them viewable
on a web page; it also doubles as godoc comments.

As such, this commit makes significant progress in migrating the docs
from our temporary wiki page toward our new website which is still under
construction.

With this change, all host modules will use ctx.LoadModule() and pass in
both the struct pointer and the field name as a string. This allows the
reflect package to read the struct tag from that field so that it can
get the necessary information like the module namespace and the inline
key.

This has the nice side-effect of unifying the code and documentation. It
also simplifies module loading, and handles several variations on field
types for raw module fields (i.e. variations on json.RawMessage, such as
arrays and maps).

I also renamed ModuleInfo.Name -> ModuleInfo.ID, to make it clear that
the ID is the "full name" which includes both the module namespace and
the name. This clarity is helpful when describing module hierarchy.

As of this change, Caddy modules are no longer an experimental design.
I think the architecture is good enough to go forward.
2019-12-10 13:36:46 -07:00

402 lines
14 KiB
Go

// Copyright 2015 Matthew Holt and The Caddy Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package caddy
import (
"context"
"encoding/json"
"fmt"
"log"
"reflect"
"github.com/mholt/certmagic"
"go.uber.org/zap"
)
// Context is a type which defines the lifetime of modules that
// are loaded and provides access to the parent configuration
// that spawned the modules which are loaded. It should be used
// with care and wrapped with derivation functions from the
// standard context package only if you don't need the Caddy
// specific features. These contexts are cancelled when the
// lifetime of the modules loaded from it is over.
//
// Use NewContext() to get a valid value (but most modules will
// not actually need to do this).
type Context struct {
context.Context
moduleInstances map[string][]interface{}
cfg *Config
cleanupFuncs []func()
}
// NewContext provides a new context derived from the given
// context ctx. Normally, you will not need to call this
// function unless you are loading modules which have a
// different lifespan than the ones for the context the
// module was provisioned with. Be sure to call the cancel
// func when the context is to be cleaned up so that
// modules which are loaded will be properly unloaded.
// See standard library context package's documentation.
func NewContext(ctx Context) (Context, context.CancelFunc) {
newCtx := Context{moduleInstances: make(map[string][]interface{}), cfg: ctx.cfg}
c, cancel := context.WithCancel(ctx.Context)
wrappedCancel := func() {
cancel()
for _, f := range ctx.cleanupFuncs {
f()
}
for modName, modInstances := range newCtx.moduleInstances {
for _, inst := range modInstances {
if cu, ok := inst.(CleanerUpper); ok {
err := cu.Cleanup()
if err != nil {
log.Printf("[ERROR] %s (%p): cleanup: %v", modName, inst, err)
}
}
}
}
}
newCtx.Context = c
return newCtx, wrappedCancel
}
// OnCancel executes f when ctx is cancelled.
func (ctx *Context) OnCancel(f func()) {
ctx.cleanupFuncs = append(ctx.cleanupFuncs, f)
}
// LoadModule loads the Caddy module(s) from the specified field of the parent struct
// pointer and returns the loaded module(s). The struct pointer and its field name as
// a string are necessary so that reflection can be used to read the struct tag on the
// field to get the module namespace and inline module name key (if specified).
//
// The field can be any one of the supported raw module types: json.RawMessage,
// []json.RawMessage, map[string]json.RawMessage, or []map[string]json.RawMessage.
// ModuleMap may be used in place of map[string]json.RawMessage. The return value's
// underlying type mirrors the input field's type:
//
// json.RawMessage => interface{}
// []json.RawMessage => []interface{}
// map[string]json.RawMessage => map[string]interface{}
// []map[string]json.RawMessage => []map[string]interface{}
//
// The field must have a "caddy" struct tag in this format:
//
// caddy:"key1=val1 key2=val2"
//
// To load modules, a "namespace" key is required. For example, to load modules
// in the "http.handlers" namespace, you'd put: `namespace=http.handlers` in the
// Caddy struct tag.
//
// The module name must also be available. If the field type is a map or slice of maps,
// then key is assumed to be the module name if an "inline_key" is NOT specified in the
// caddy struct tag. In this case, the module name does NOT need to be specified in-line
// with the module itself.
//
// If not a map, or if inline_key is non-empty, then the module name must be embedded
// into the values, which must be objects; then there must be a key in those objects
// where its associated value is the module name. This is called the "inline key",
// meaning the key containing the module's name that is defined inline with the module
// itself. You must specify the inline key in a struct tag, along with the namespace:
//
// caddy:"namespace=http.handlers inline_key=handler"
//
// This will look for a key/value pair like `"handler": "..."` in the json.RawMessage
// in order to know the module name.
//
// To make use of the loaded module(s) (the return value), you will probably want
// to type-assert each interface{} value(s) to the types that are useful to you
// and store them on the same struct. Storing them on the same struct makes for
// easy garbage collection when your host module is no longer needed.
//
// Loaded modules have already been provisioned and validated.
func (ctx Context) LoadModule(structPointer interface{}, fieldName string) (interface{}, error) {
val := reflect.ValueOf(structPointer).Elem().FieldByName(fieldName)
typ := val.Type()
field, ok := reflect.TypeOf(structPointer).Elem().FieldByName(fieldName)
if !ok {
panic(fmt.Sprintf("field %s does not exist in %#v", fieldName, structPointer))
}
opts, err := ParseStructTag(field.Tag.Get("caddy"))
if err != nil {
panic(fmt.Sprintf("malformed tag on field %s: %v", fieldName, err))
}
moduleNamespace, ok := opts["namespace"]
if !ok {
panic(fmt.Sprintf("missing 'namespace' key in struct tag on field %s", fieldName))
}
inlineModuleKey := opts["inline_key"]
var result interface{}
switch val.Kind() {
case reflect.Slice:
if isJSONRawMessage(typ) {
// val is `json.RawMessage` ([]uint8 under the hood)
if inlineModuleKey == "" {
panic("unable to determine module name without inline_key when type is not a ModuleMap")
}
val, err := ctx.loadModuleInline(inlineModuleKey, moduleNamespace, val.Interface().(json.RawMessage))
if err != nil {
return nil, err
}
result = val
} else if isJSONRawMessage(typ.Elem()) {
// val is `[]json.RawMessage`
if inlineModuleKey == "" {
panic("unable to determine module name without inline_key because type is not a ModuleMap")
}
var all []interface{}
for i := 0; i < val.Len(); i++ {
val, err := ctx.loadModuleInline(inlineModuleKey, moduleNamespace, val.Index(i).Interface().(json.RawMessage))
if err != nil {
return nil, fmt.Errorf("position %d: %v", i, err)
}
all = append(all, val)
}
result = all
} else if isModuleMapType(typ.Elem()) {
// val is `[]map[string]json.RawMessage`
var all []map[string]interface{}
for i := 0; i < val.Len(); i++ {
thisSet, err := ctx.loadModulesFromSomeMap(moduleNamespace, inlineModuleKey, val.Index(i))
if err != nil {
return nil, err
}
all = append(all, thisSet)
}
result = all
}
case reflect.Map:
// val is a ModuleMap or some other kind of map
result, err = ctx.loadModulesFromSomeMap(moduleNamespace, inlineModuleKey, val)
if err != nil {
return nil, err
}
default:
return nil, fmt.Errorf("unrecognized type for module: %s", typ)
}
// we're done with the raw bytes; allow GC to deallocate
val.Set(reflect.Zero(typ))
return result, nil
}
// loadModulesFromSomeMap loads modules from val, which must be a type of map[string]interface{}.
// Depending on inlineModuleKey, it will be interpeted as either a ModuleMap (key is the module
// name) or as a regular map (key is not the module name, and module name is defined inline).
func (ctx Context) loadModulesFromSomeMap(namespace, inlineModuleKey string, val reflect.Value) (map[string]interface{}, error) {
// if no inline_key is specified, then val must be a ModuleMap,
// where the key is the module name
if inlineModuleKey == "" {
if !isModuleMapType(val.Type()) {
panic(fmt.Sprintf("expected ModuleMap because inline_key is empty; but we do not recognize this type: %s", val.Type()))
}
return ctx.loadModuleMap(namespace, val)
}
// otherwise, val is a map with modules, but the module name is
// inline with each value (the key means something else)
return ctx.loadModulesFromRegularMap(namespace, inlineModuleKey, val)
}
// loadModulesFromRegularMap loads modules from val, where val is a map[string]json.RawMessage.
// Map keys are NOT interpreted as module names, so module names are still expected to appear
// inline with the objects.
func (ctx Context) loadModulesFromRegularMap(namespace, inlineModuleKey string, val reflect.Value) (map[string]interface{}, error) {
mods := make(map[string]interface{})
iter := val.MapRange()
for iter.Next() {
k := iter.Key()
v := iter.Value()
mod, err := ctx.loadModuleInline(inlineModuleKey, namespace, v.Interface().(json.RawMessage))
if err != nil {
return nil, fmt.Errorf("key %s: %v", k, err)
}
mods[k.String()] = mod
}
return mods, nil
}
// loadModuleMap loads modules from a ModuleMap, i.e. map[string]interface{}, where the key is the
// module name. With a module map, module names do not need to be defined inline with their values.
func (ctx Context) loadModuleMap(namespace string, val reflect.Value) (map[string]interface{}, error) {
all := make(map[string]interface{})
iter := val.MapRange()
for iter.Next() {
k := iter.Key().Interface().(string)
v := iter.Value().Interface().(json.RawMessage)
moduleName := namespace + "." + k
if namespace == "" {
moduleName = k
}
val, err := ctx.LoadModuleByID(moduleName, v)
if err != nil {
return nil, fmt.Errorf("module name '%s': %v", k, err)
}
all[k] = val
}
return all, nil
}
// LoadModuleByID decodes rawMsg into a new instance of mod and
// returns the value. If mod.New is nil, an error is returned.
// If the module implements Validator or Provisioner interfaces,
// those methods are invoked to ensure the module is fully
// configured and valid before being used.
//
// This is a lower-level method and will usually not be called
// directly by most modules. However, this method is useful when
// dynamically loading/unloading modules in their own context,
// like from embedded scripts, etc.
func (ctx Context) LoadModuleByID(id string, rawMsg json.RawMessage) (interface{}, error) {
modulesMu.RLock()
mod, ok := modules[id]
modulesMu.RUnlock()
if !ok {
return nil, fmt.Errorf("unknown module: %s", id)
}
if mod.New == nil {
return nil, fmt.Errorf("module '%s' has no constructor", mod.ID)
}
val := mod.New().(interface{})
// value must be a pointer for unmarshaling into concrete type, even if
// the module's concrete type is a slice or map; New() *should* return
// a pointer, otherwise unmarshaling errors or panics will occur
if rv := reflect.ValueOf(val); rv.Kind() != reflect.Ptr {
log.Printf("[WARNING] ModuleInfo.New() for module '%s' did not return a pointer,"+
" so we are using reflection to make a pointer instead; please fix this by"+
" using new(Type) or &Type notation in your module's New() function.", id)
val = reflect.New(rv.Type()).Elem().Addr().Interface().(Module)
}
// fill in its config only if there is a config to fill in
if len(rawMsg) > 0 {
err := strictUnmarshalJSON(rawMsg, &val)
if err != nil {
return nil, fmt.Errorf("decoding module config: %s: %v", mod, err)
}
}
if val == nil {
// returned module values are almost always type-asserted
// before being used, so a nil value would panic; and there
// is no good reason to explicitly declare null modules in
// a config; it might be because the user is trying to achieve
// a result the developer isn't expecting, which is a smell
return nil, fmt.Errorf("module value cannot be null")
}
if prov, ok := val.(Provisioner); ok {
err := prov.Provision(ctx)
if err != nil {
// incomplete provisioning could have left state
// dangling, so make sure it gets cleaned up
if cleanerUpper, ok := val.(CleanerUpper); ok {
err2 := cleanerUpper.Cleanup()
if err2 != nil {
err = fmt.Errorf("%v; additionally, cleanup: %v", err, err2)
}
}
return nil, fmt.Errorf("provision %s: %v", mod, err)
}
}
if validator, ok := val.(Validator); ok {
err := validator.Validate()
if err != nil {
// since the module was already provisioned, make sure we clean up
if cleanerUpper, ok := val.(CleanerUpper); ok {
err2 := cleanerUpper.Cleanup()
if err2 != nil {
err = fmt.Errorf("%v; additionally, cleanup: %v", err, err2)
}
}
return nil, fmt.Errorf("%s: invalid configuration: %v", mod, err)
}
}
ctx.moduleInstances[id] = append(ctx.moduleInstances[id], val)
return val, nil
}
// loadModuleInline loads a module from a JSON raw message which decodes to
// a map[string]interface{}, where one of the object keys is moduleNameKey
// and the corresponding value is the module name (as a string) which can
// be found in the given scope. In other words, the module name is declared
// in-line with the module itself.
//
// This allows modules to be decoded into their concrete types and used when
// their names cannot be the unique key in a map, such as when there are
// multiple instances in the map or it appears in an array (where there are
// no custom keys). In other words, the key containing the module name is
// treated special/separate from all the other keys in the object.
func (ctx Context) loadModuleInline(moduleNameKey, moduleScope string, raw json.RawMessage) (interface{}, error) {
moduleName, raw, err := getModuleNameInline(moduleNameKey, raw)
if err != nil {
return nil, err
}
val, err := ctx.LoadModuleByID(moduleScope+"."+moduleName, raw)
if err != nil {
return nil, fmt.Errorf("loading module '%s': %v", moduleName, err)
}
return val, nil
}
// App returns the configured app named name. If no app with
// that name is currently configured, a new empty one will be
// instantiated. (The app module must still be registered.)
func (ctx Context) App(name string) (interface{}, error) {
if app, ok := ctx.cfg.apps[name]; ok {
return app, nil
}
modVal, err := ctx.LoadModuleByID(name, nil)
if err != nil {
return nil, fmt.Errorf("instantiating new module %s: %v", name, err)
}
ctx.cfg.apps[name] = modVal.(App)
return modVal, nil
}
// Storage returns the configured Caddy storage implementation.
func (ctx Context) Storage() certmagic.Storage {
return ctx.cfg.storage
}
// Logger returns a logger that can be used by mod.
func (ctx Context) Logger(mod Module) *zap.Logger {
return ctx.cfg.Logging.Logger(mod)
}