tpl/internal: Sync go_templates

Closes #10411
This commit is contained in:
Bjørn Erik Pedersen
2022-11-14 19:13:09 +01:00
parent 58a98c7758
commit f6ab9553f4
34 changed files with 739 additions and 514 deletions

View File

@@ -12,14 +12,14 @@ The documentation here focuses on the security features of the package.
For information about how to program the templates themselves, see the
documentation for text/template.
Introduction
# Introduction
This package wraps package text/template so you can share its template API
to parse and execute HTML templates safely.
tmpl, err := template.New("name").Parse(...)
// Error checking elided
err = tmpl.Execute(out, data)
tmpl, err := template.New("name").Parse(...)
// Error checking elided
err = tmpl.Execute(out, data)
If successful, tmpl will now be injection-safe. Otherwise, err is an error
defined in the docs for ErrorCode.
@@ -34,38 +34,37 @@ provided below.
Example
import template "github.com/gohugoio/hugo/tpl/internal/go_templates/texttemplate"
...
t, err := template.New("foo").Parse(`{{define "T"}}Hello, {{.}}!{{end}}`)
err = t.ExecuteTemplate(out, "T", "<script>alert('you have been pwned')</script>")
import template "github.com/gohugoio/hugo/tpl/internal/go_templates/texttemplate"
...
t, err := template.New("foo").Parse(`{{define "T"}}Hello, {{.}}!{{end}}`)
err = t.ExecuteTemplate(out, "T", "<script>alert('you have been pwned')</script>")
produces
Hello, <script>alert('you have been pwned')</script>!
Hello, <script>alert('you have been pwned')</script>!
but the contextual autoescaping in html/template
import template "github.com/gohugoio/hugo/tpl/internal/go_templates/htmltemplate"
...
t, err := template.New("foo").Parse(`{{define "T"}}Hello, {{.}}!{{end}}`)
err = t.ExecuteTemplate(out, "T", "<script>alert('you have been pwned')</script>")
import template "github.com/gohugoio/hugo/tpl/internal/go_templates/htmltemplate"
...
t, err := template.New("foo").Parse(`{{define "T"}}Hello, {{.}}!{{end}}`)
err = t.ExecuteTemplate(out, "T", "<script>alert('you have been pwned')</script>")
produces safe, escaped HTML output
Hello, &lt;script&gt;alert(&#39;you have been pwned&#39;)&lt;/script&gt;!
Hello, &lt;script&gt;alert(&#39;you have been pwned&#39;)&lt;/script&gt;!
Contexts
# Contexts
This package understands HTML, CSS, JavaScript, and URIs. It adds sanitizing
functions to each simple action pipeline, so given the excerpt
<a href="/search?q={{.}}">{{.}}</a>
<a href="/search?q={{.}}">{{.}}</a>
At parse time each {{.}} is overwritten to add escaping functions as necessary.
In this case it becomes
<a href="/search?q={{. | urlescaper | attrescaper}}">{{. | htmlescaper}}</a>
<a href="/search?q={{. | urlescaper | attrescaper}}">{{. | htmlescaper}}</a>
where urlescaper, attrescaper, and htmlescaper are aliases for internal escaping
functions.
@@ -73,117 +72,113 @@ functions.
For these internal escaping functions, if an action pipeline evaluates to
a nil interface value, it is treated as though it were an empty string.
Namespaced and data- attributes
# Namespaced and data- attributes
Attributes with a namespace are treated as if they had no namespace.
Given the excerpt
<a my:href="{{.}}"></a>
<a my:href="{{.}}"></a>
At parse time the attribute will be treated as if it were just "href".
So at parse time the template becomes:
<a my:href="{{. | urlescaper | attrescaper}}"></a>
<a my:href="{{. | urlescaper | attrescaper}}"></a>
Similarly to attributes with namespaces, attributes with a "data-" prefix are
treated as if they had no "data-" prefix. So given
<a data-href="{{.}}"></a>
<a data-href="{{.}}"></a>
At parse time this becomes
<a data-href="{{. | urlescaper | attrescaper}}"></a>
<a data-href="{{. | urlescaper | attrescaper}}"></a>
If an attribute has both a namespace and a "data-" prefix, only the namespace
will be removed when determining the context. For example
<a my:data-href="{{.}}"></a>
<a my:data-href="{{.}}"></a>
This is handled as if "my:data-href" was just "data-href" and not "href" as
it would be if the "data-" prefix were to be ignored too. Thus at parse
time this becomes just
<a my:data-href="{{. | attrescaper}}"></a>
<a my:data-href="{{. | attrescaper}}"></a>
As a special case, attributes with the namespace "xmlns" are always treated
as containing URLs. Given the excerpts
<a xmlns:title="{{.}}"></a>
<a xmlns:href="{{.}}"></a>
<a xmlns:onclick="{{.}}"></a>
<a xmlns:title="{{.}}"></a>
<a xmlns:href="{{.}}"></a>
<a xmlns:onclick="{{.}}"></a>
At parse time they become:
<a xmlns:title="{{. | urlescaper | attrescaper}}"></a>
<a xmlns:href="{{. | urlescaper | attrescaper}}"></a>
<a xmlns:onclick="{{. | urlescaper | attrescaper}}"></a>
<a xmlns:title="{{. | urlescaper | attrescaper}}"></a>
<a xmlns:href="{{. | urlescaper | attrescaper}}"></a>
<a xmlns:onclick="{{. | urlescaper | attrescaper}}"></a>
Errors
# Errors
See the documentation of ErrorCode for details.
A fuller picture
# A fuller picture
The rest of this package comment may be skipped on first reading; it includes
details necessary to understand escaping contexts and error messages. Most users
will not need to understand these details.
Contexts
# Contexts
Assuming {{.}} is `O'Reilly: How are <i>you</i>?`, the table below shows
how {{.}} appears when used in the context to the left.
Context {{.}} After
{{.}} O'Reilly: How are &lt;i&gt;you&lt;/i&gt;?
<a title='{{.}}'> O&#39;Reilly: How are you?
<a href="/{{.}}"> O&#39;Reilly: How are %3ci%3eyou%3c/i%3e?
<a href="?q={{.}}"> O&#39;Reilly%3a%20How%20are%3ci%3e...%3f
<a onx='f("{{.}}")'> O\x27Reilly: How are \x3ci\x3eyou...?
<a onx='f({{.}})'> "O\x27Reilly: How are \x3ci\x3eyou...?"
<a onx='pattern = /{{.}}/;'> O\x27Reilly: How are \x3ci\x3eyou...\x3f
Context {{.}} After
{{.}} O'Reilly: How are &lt;i&gt;you&lt;/i&gt;?
<a title='{{.}}'> O&#39;Reilly: How are you?
<a href="/{{.}}"> O&#39;Reilly: How are %3ci%3eyou%3c/i%3e?
<a href="?q={{.}}"> O&#39;Reilly%3a%20How%20are%3ci%3e...%3f
<a onx='f("{{.}}")'> O\x27Reilly: How are \x3ci\x3eyou...?
<a onx='f({{.}})'> "O\x27Reilly: How are \x3ci\x3eyou...?"
<a onx='pattern = /{{.}}/;'> O\x27Reilly: How are \x3ci\x3eyou...\x3f
If used in an unsafe context, then the value might be filtered out:
Context {{.}} After
<a href="{{.}}"> #ZgotmplZ
Context {{.}} After
<a href="{{.}}"> #ZgotmplZ
since "O'Reilly:" is not an allowed protocol like "http:".
If {{.}} is the innocuous word, `left`, then it can appear more widely,
Context {{.}} After
{{.}} left
<a title='{{.}}'> left
<a href='{{.}}'> left
<a href='/{{.}}'> left
<a href='?dir={{.}}'> left
<a style="border-{{.}}: 4px"> left
<a style="align: {{.}}"> left
<a style="background: '{{.}}'> left
<a style="background: url('{{.}}')> left
<style>p.{{.}} {color:red}</style> left
Context {{.}} After
{{.}} left
<a title='{{.}}'> left
<a href='{{.}}'> left
<a href='/{{.}}'> left
<a href='?dir={{.}}'> left
<a style="border-{{.}}: 4px"> left
<a style="align: {{.}}"> left
<a style="background: '{{.}}'> left
<a style="background: url('{{.}}')> left
<style>p.{{.}} {color:red}</style> left
Non-string values can be used in JavaScript contexts.
If {{.}} is
struct{A,B string}{ "foo", "bar" }
struct{A,B string}{ "foo", "bar" }
in the escaped template
<script>var pair = {{.}};</script>
<script>var pair = {{.}};</script>
then the template output is
<script>var pair = {"A": "foo", "B": "bar"};</script>
<script>var pair = {"A": "foo", "B": "bar"};</script>
See package json to understand how non-string content is marshaled for
embedding in JavaScript contexts.
Typed Strings
# Typed Strings
By default, this package assumes that all pipelines produce a plain text string.
It adds escaping pipeline stages necessary to correctly and safely embed that
@@ -197,24 +192,23 @@ exempted from escaping.
The template
Hello, {{.}}!
Hello, {{.}}!
can be invoked with
tmpl.Execute(out, template.HTML(`<b>World</b>`))
tmpl.Execute(out, template.HTML(`<b>World</b>`))
to produce
Hello, <b>World</b>!
Hello, <b>World</b>!
instead of the
Hello, &lt;b&gt;World&lt;b&gt;!
Hello, &lt;b&gt;World&lt;b&gt;!
that would have been produced if {{.}} was a regular string.
Security Model
# Security Model
https://rawgit.com/mikesamuel/sanitized-jquery-templates/trunk/safetemplate.html#problem_definition defines "safe" as used by this package.