Source file src/html/template/escape.go

     1  // Copyright 2011 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package template
     6  
     7  import (
     8  	"bytes"
     9  	"fmt"
    10  	"html"
    11  	"internal/godebug"
    12  	"io"
    13  	"regexp"
    14  	"text/template"
    15  	"text/template/parse"
    16  )
    17  
    18  // escapeTemplate rewrites the named template, which must be
    19  // associated with t, to guarantee that the output of any of the named
    20  // templates is properly escaped. If no error is returned, then the named templates have
    21  // been modified. Otherwise the named templates have been rendered
    22  // unusable.
    23  func escapeTemplate(tmpl *Template, node parse.Node, name string) error {
    24  	c, _ := tmpl.esc.escapeTree(context{}, node, name, 0)
    25  	var err error
    26  	if c.err != nil {
    27  		err, c.err.Name = c.err, name
    28  	} else if c.state != stateText {
    29  		err = &Error{ErrEndContext, nil, name, 0, fmt.Sprintf("ends in a non-text context: %v", c)}
    30  	}
    31  	if err != nil {
    32  		// Prevent execution of unsafe templates.
    33  		if t := tmpl.set[name]; t != nil {
    34  			t.escapeErr = err
    35  			t.text.Tree = nil
    36  			t.Tree = nil
    37  		}
    38  		return err
    39  	}
    40  	tmpl.esc.commit()
    41  	if t := tmpl.set[name]; t != nil {
    42  		t.escapeErr = escapeOK
    43  		t.Tree = t.text.Tree
    44  	}
    45  	return nil
    46  }
    47  
    48  // evalArgs formats the list of arguments into a string. It is equivalent to
    49  // fmt.Sprint(args...), except that it dereferences all pointers.
    50  func evalArgs(args ...any) string {
    51  	// Optimization for simple common case of a single string argument.
    52  	if len(args) == 1 {
    53  		if s, ok := args[0].(string); ok {
    54  			return s
    55  		}
    56  	}
    57  	for i, arg := range args {
    58  		args[i] = indirectToStringerOrError(arg)
    59  	}
    60  	return fmt.Sprint(args...)
    61  }
    62  
    63  // funcMap maps command names to functions that render their inputs safe.
    64  var funcMap = template.FuncMap{
    65  	"_html_template_attrescaper":      attrEscaper,
    66  	"_html_template_commentescaper":   commentEscaper,
    67  	"_html_template_cssescaper":       cssEscaper,
    68  	"_html_template_cssvaluefilter":   cssValueFilter,
    69  	"_html_template_htmlnamefilter":   htmlNameFilter,
    70  	"_html_template_htmlescaper":      htmlEscaper,
    71  	"_html_template_jsregexpescaper":  jsRegexpEscaper,
    72  	"_html_template_jsstrescaper":     jsStrEscaper,
    73  	"_html_template_jstmpllitescaper": jsTmplLitEscaper,
    74  	"_html_template_jsvalescaper":     jsValEscaper,
    75  	"_html_template_nospaceescaper":   htmlNospaceEscaper,
    76  	"_html_template_rcdataescaper":    rcdataEscaper,
    77  	"_html_template_srcsetescaper":    srcsetFilterAndEscaper,
    78  	"_html_template_urlescaper":       urlEscaper,
    79  	"_html_template_urlfilter":        urlFilter,
    80  	"_html_template_urlnormalizer":    urlNormalizer,
    81  	"_eval_args_":                     evalArgs,
    82  }
    83  
    84  // escaper collects type inferences about templates and changes needed to make
    85  // templates injection safe.
    86  type escaper struct {
    87  	// ns is the nameSpace that this escaper is associated with.
    88  	ns *nameSpace
    89  	// output[templateName] is the output context for a templateName that
    90  	// has been mangled to include its input context.
    91  	output map[string]context
    92  	// derived[c.mangle(name)] maps to a template derived from the template
    93  	// named name templateName for the start context c.
    94  	derived map[string]*template.Template
    95  	// called[templateName] is a set of called mangled template names.
    96  	called map[string]bool
    97  	// xxxNodeEdits are the accumulated edits to apply during commit.
    98  	// Such edits are not applied immediately in case a template set
    99  	// executes a given template in different escaping contexts.
   100  	actionNodeEdits   map[*parse.ActionNode][]string
   101  	templateNodeEdits map[*parse.TemplateNode]string
   102  	textNodeEdits     map[*parse.TextNode][]byte
   103  	// rangeContext holds context about the current range loop.
   104  	rangeContext *rangeContext
   105  }
   106  
   107  // rangeContext holds information about the current range loop.
   108  type rangeContext struct {
   109  	outer     *rangeContext // outer loop
   110  	breaks    []context     // context at each break action
   111  	continues []context     // context at each continue action
   112  }
   113  
   114  // makeEscaper creates a blank escaper for the given set.
   115  func makeEscaper(n *nameSpace) escaper {
   116  	return escaper{
   117  		n,
   118  		map[string]context{},
   119  		map[string]*template.Template{},
   120  		map[string]bool{},
   121  		map[*parse.ActionNode][]string{},
   122  		map[*parse.TemplateNode]string{},
   123  		map[*parse.TextNode][]byte{},
   124  		nil,
   125  	}
   126  }
   127  
   128  // filterFailsafe is an innocuous word that is emitted in place of unsafe values
   129  // by sanitizer functions. It is not a keyword in any programming language,
   130  // contains no special characters, is not empty, and when it appears in output
   131  // it is distinct enough that a developer can find the source of the problem
   132  // via a search engine.
   133  const filterFailsafe = "ZgotmplZ"
   134  
   135  // escape escapes a template node.
   136  func (e *escaper) escape(c context, n parse.Node) context {
   137  	switch n := n.(type) {
   138  	case *parse.ActionNode:
   139  		return e.escapeAction(c, n)
   140  	case *parse.BreakNode:
   141  		c.n = n
   142  		e.rangeContext.breaks = append(e.rangeContext.breaks, c)
   143  		return context{state: stateDead}
   144  	case *parse.CommentNode:
   145  		return c
   146  	case *parse.ContinueNode:
   147  		c.n = n
   148  		e.rangeContext.continues = append(e.rangeContext.breaks, c)
   149  		return context{state: stateDead}
   150  	case *parse.IfNode:
   151  		return e.escapeBranch(c, &n.BranchNode, "if")
   152  	case *parse.ListNode:
   153  		return e.escapeList(c, n)
   154  	case *parse.RangeNode:
   155  		return e.escapeBranch(c, &n.BranchNode, "range")
   156  	case *parse.TemplateNode:
   157  		return e.escapeTemplate(c, n)
   158  	case *parse.TextNode:
   159  		return e.escapeText(c, n)
   160  	case *parse.WithNode:
   161  		return e.escapeBranch(c, &n.BranchNode, "with")
   162  	}
   163  	panic("escaping " + n.String() + " is unimplemented")
   164  }
   165  
   166  var debugAllowActionJSTmpl = godebug.New("jstmpllitinterp")
   167  
   168  // escapeAction escapes an action template node.
   169  func (e *escaper) escapeAction(c context, n *parse.ActionNode) context {
   170  	if len(n.Pipe.Decl) != 0 {
   171  		// A local variable assignment, not an interpolation.
   172  		return c
   173  	}
   174  	c = nudge(c)
   175  	// Check for disallowed use of predefined escapers in the pipeline.
   176  	for pos, idNode := range n.Pipe.Cmds {
   177  		node, ok := idNode.Args[0].(*parse.IdentifierNode)
   178  		if !ok {
   179  			// A predefined escaper "esc" will never be found as an identifier in a
   180  			// Chain or Field node, since:
   181  			// - "esc.x ..." is invalid, since predefined escapers return strings, and
   182  			//   strings do not have methods, keys or fields.
   183  			// - "... .esc" is invalid, since predefined escapers are global functions,
   184  			//   not methods or fields of any types.
   185  			// Therefore, it is safe to ignore these two node types.
   186  			continue
   187  		}
   188  		ident := node.Ident
   189  		if _, ok := predefinedEscapers[ident]; ok {
   190  			if pos < len(n.Pipe.Cmds)-1 ||
   191  				c.state == stateAttr && c.delim == delimSpaceOrTagEnd && ident == "html" {
   192  				return context{
   193  					state: stateError,
   194  					err:   errorf(ErrPredefinedEscaper, n, n.Line, "predefined escaper %q disallowed in template", ident),
   195  				}
   196  			}
   197  		}
   198  	}
   199  	s := make([]string, 0, 3)
   200  	switch c.state {
   201  	case stateError:
   202  		return c
   203  	case stateURL, stateCSSDqStr, stateCSSSqStr, stateCSSDqURL, stateCSSSqURL, stateCSSURL:
   204  		switch c.urlPart {
   205  		case urlPartNone:
   206  			s = append(s, "_html_template_urlfilter")
   207  			fallthrough
   208  		case urlPartPreQuery:
   209  			switch c.state {
   210  			case stateCSSDqStr, stateCSSSqStr:
   211  				s = append(s, "_html_template_cssescaper")
   212  			default:
   213  				s = append(s, "_html_template_urlnormalizer")
   214  			}
   215  		case urlPartQueryOrFrag:
   216  			s = append(s, "_html_template_urlescaper")
   217  		case urlPartUnknown:
   218  			return context{
   219  				state: stateError,
   220  				err:   errorf(ErrAmbigContext, n, n.Line, "%s appears in an ambiguous context within a URL", n),
   221  			}
   222  		default:
   223  			panic(c.urlPart.String())
   224  		}
   225  	case stateJS:
   226  		s = append(s, "_html_template_jsvalescaper")
   227  		// A slash after a value starts a div operator.
   228  		c.jsCtx = jsCtxDivOp
   229  	case stateJSDqStr, stateJSSqStr:
   230  		s = append(s, "_html_template_jsstrescaper")
   231  	case stateJSTmplLit:
   232  		s = append(s, "_html_template_jstmpllitescaper")
   233  	case stateJSRegexp:
   234  		s = append(s, "_html_template_jsregexpescaper")
   235  	case stateCSS:
   236  		s = append(s, "_html_template_cssvaluefilter")
   237  	case stateText:
   238  		s = append(s, "_html_template_htmlescaper")
   239  	case stateRCDATA:
   240  		s = append(s, "_html_template_rcdataescaper")
   241  	case stateAttr:
   242  		// Handled below in delim check.
   243  	case stateAttrName, stateTag:
   244  		c.state = stateAttrName
   245  		s = append(s, "_html_template_htmlnamefilter")
   246  	case stateSrcset:
   247  		s = append(s, "_html_template_srcsetescaper")
   248  	default:
   249  		if isComment(c.state) {
   250  			s = append(s, "_html_template_commentescaper")
   251  		} else {
   252  			panic("unexpected state " + c.state.String())
   253  		}
   254  	}
   255  	switch c.delim {
   256  	case delimNone:
   257  		// No extra-escaping needed for raw text content.
   258  	case delimSpaceOrTagEnd:
   259  		s = append(s, "_html_template_nospaceescaper")
   260  	default:
   261  		s = append(s, "_html_template_attrescaper")
   262  	}
   263  	e.editActionNode(n, s)
   264  	return c
   265  }
   266  
   267  // ensurePipelineContains ensures that the pipeline ends with the commands with
   268  // the identifiers in s in order. If the pipeline ends with a predefined escaper
   269  // (i.e. "html" or "urlquery"), merge it with the identifiers in s.
   270  func ensurePipelineContains(p *parse.PipeNode, s []string) {
   271  	if len(s) == 0 {
   272  		// Do not rewrite pipeline if we have no escapers to insert.
   273  		return
   274  	}
   275  	// Precondition: p.Cmds contains at most one predefined escaper and the
   276  	// escaper will be present at p.Cmds[len(p.Cmds)-1]. This precondition is
   277  	// always true because of the checks in escapeAction.
   278  	pipelineLen := len(p.Cmds)
   279  	if pipelineLen > 0 {
   280  		lastCmd := p.Cmds[pipelineLen-1]
   281  		if idNode, ok := lastCmd.Args[0].(*parse.IdentifierNode); ok {
   282  			if esc := idNode.Ident; predefinedEscapers[esc] {
   283  				// Pipeline ends with a predefined escaper.
   284  				if len(p.Cmds) == 1 && len(lastCmd.Args) > 1 {
   285  					// Special case: pipeline is of the form {{ esc arg1 arg2 ... argN }},
   286  					// where esc is the predefined escaper, and arg1...argN are its arguments.
   287  					// Convert this into the equivalent form
   288  					// {{ _eval_args_ arg1 arg2 ... argN | esc }}, so that esc can be easily
   289  					// merged with the escapers in s.
   290  					lastCmd.Args[0] = parse.NewIdentifier("_eval_args_").SetTree(nil).SetPos(lastCmd.Args[0].Position())
   291  					p.Cmds = appendCmd(p.Cmds, newIdentCmd(esc, p.Position()))
   292  					pipelineLen++
   293  				}
   294  				// If any of the commands in s that we are about to insert is equivalent
   295  				// to the predefined escaper, use the predefined escaper instead.
   296  				dup := false
   297  				for i, escaper := range s {
   298  					if escFnsEq(esc, escaper) {
   299  						s[i] = idNode.Ident
   300  						dup = true
   301  					}
   302  				}
   303  				if dup {
   304  					// The predefined escaper will already be inserted along with the
   305  					// escapers in s, so do not copy it to the rewritten pipeline.
   306  					pipelineLen--
   307  				}
   308  			}
   309  		}
   310  	}
   311  	// Rewrite the pipeline, creating the escapers in s at the end of the pipeline.
   312  	newCmds := make([]*parse.CommandNode, pipelineLen, pipelineLen+len(s))
   313  	insertedIdents := make(map[string]bool)
   314  	for i := 0; i < pipelineLen; i++ {
   315  		cmd := p.Cmds[i]
   316  		newCmds[i] = cmd
   317  		if idNode, ok := cmd.Args[0].(*parse.IdentifierNode); ok {
   318  			insertedIdents[normalizeEscFn(idNode.Ident)] = true
   319  		}
   320  	}
   321  	for _, name := range s {
   322  		if !insertedIdents[normalizeEscFn(name)] {
   323  			// When two templates share an underlying parse tree via the use of
   324  			// AddParseTree and one template is executed after the other, this check
   325  			// ensures that escapers that were already inserted into the pipeline on
   326  			// the first escaping pass do not get inserted again.
   327  			newCmds = appendCmd(newCmds, newIdentCmd(name, p.Position()))
   328  		}
   329  	}
   330  	p.Cmds = newCmds
   331  }
   332  
   333  // predefinedEscapers contains template predefined escapers that are equivalent
   334  // to some contextual escapers. Keep in sync with equivEscapers.
   335  var predefinedEscapers = map[string]bool{
   336  	"html":     true,
   337  	"urlquery": true,
   338  }
   339  
   340  // equivEscapers matches contextual escapers to equivalent predefined
   341  // template escapers.
   342  var equivEscapers = map[string]string{
   343  	// The following pairs of HTML escapers provide equivalent security
   344  	// guarantees, since they all escape '\000', '\'', '"', '&', '<', and '>'.
   345  	"_html_template_attrescaper":   "html",
   346  	"_html_template_htmlescaper":   "html",
   347  	"_html_template_rcdataescaper": "html",
   348  	// These two URL escapers produce URLs safe for embedding in a URL query by
   349  	// percent-encoding all the reserved characters specified in RFC 3986 Section
   350  	// 2.2
   351  	"_html_template_urlescaper": "urlquery",
   352  	// These two functions are not actually equivalent; urlquery is stricter as it
   353  	// escapes reserved characters (e.g. '#'), while _html_template_urlnormalizer
   354  	// does not. It is therefore only safe to replace _html_template_urlnormalizer
   355  	// with urlquery (this happens in ensurePipelineContains), but not the otherI've
   356  	// way around. We keep this entry around to preserve the behavior of templates
   357  	// written before Go 1.9, which might depend on this substitution taking place.
   358  	"_html_template_urlnormalizer": "urlquery",
   359  }
   360  
   361  // escFnsEq reports whether the two escaping functions are equivalent.
   362  func escFnsEq(a, b string) bool {
   363  	return normalizeEscFn(a) == normalizeEscFn(b)
   364  }
   365  
   366  // normalizeEscFn(a) is equal to normalizeEscFn(b) for any pair of names of
   367  // escaper functions a and b that are equivalent.
   368  func normalizeEscFn(e string) string {
   369  	if norm := equivEscapers[e]; norm != "" {
   370  		return norm
   371  	}
   372  	return e
   373  }
   374  
   375  // redundantFuncs[a][b] implies that funcMap[b](funcMap[a](x)) == funcMap[a](x)
   376  // for all x.
   377  var redundantFuncs = map[string]map[string]bool{
   378  	"_html_template_commentescaper": {
   379  		"_html_template_attrescaper": true,
   380  		"_html_template_htmlescaper": true,
   381  	},
   382  	"_html_template_cssescaper": {
   383  		"_html_template_attrescaper": true,
   384  	},
   385  	"_html_template_jsregexpescaper": {
   386  		"_html_template_attrescaper": true,
   387  	},
   388  	"_html_template_jsstrescaper": {
   389  		"_html_template_attrescaper": true,
   390  	},
   391  	"_html_template_jstmpllitescaper": {
   392  		"_html_template_attrescaper": true,
   393  	},
   394  	"_html_template_urlescaper": {
   395  		"_html_template_urlnormalizer": true,
   396  	},
   397  }
   398  
   399  // appendCmd appends the given command to the end of the command pipeline
   400  // unless it is redundant with the last command.
   401  func appendCmd(cmds []*parse.CommandNode, cmd *parse.CommandNode) []*parse.CommandNode {
   402  	if n := len(cmds); n != 0 {
   403  		last, okLast := cmds[n-1].Args[0].(*parse.IdentifierNode)
   404  		next, okNext := cmd.Args[0].(*parse.IdentifierNode)
   405  		if okLast && okNext && redundantFuncs[last.Ident][next.Ident] {
   406  			return cmds
   407  		}
   408  	}
   409  	return append(cmds, cmd)
   410  }
   411  
   412  // newIdentCmd produces a command containing a single identifier node.
   413  func newIdentCmd(identifier string, pos parse.Pos) *parse.CommandNode {
   414  	return &parse.CommandNode{
   415  		NodeType: parse.NodeCommand,
   416  		Args:     []parse.Node{parse.NewIdentifier(identifier).SetTree(nil).SetPos(pos)}, // TODO: SetTree.
   417  	}
   418  }
   419  
   420  // nudge returns the context that would result from following empty string
   421  // transitions from the input context.
   422  // For example, parsing:
   423  //
   424  //	`<a href=`
   425  //
   426  // will end in context{stateBeforeValue, attrURL}, but parsing one extra rune:
   427  //
   428  //	`<a href=x`
   429  //
   430  // will end in context{stateURL, delimSpaceOrTagEnd, ...}.
   431  // There are two transitions that happen when the 'x' is seen:
   432  // (1) Transition from a before-value state to a start-of-value state without
   433  //
   434  //	consuming any character.
   435  //
   436  // (2) Consume 'x' and transition past the first value character.
   437  // In this case, nudging produces the context after (1) happens.
   438  func nudge(c context) context {
   439  	switch c.state {
   440  	case stateTag:
   441  		// In `<foo {{.}}`, the action should emit an attribute.
   442  		c.state = stateAttrName
   443  	case stateBeforeValue:
   444  		// In `<foo bar={{.}}`, the action is an undelimited value.
   445  		c.state, c.delim, c.attr = attrStartStates[c.attr], delimSpaceOrTagEnd, attrNone
   446  	case stateAfterName:
   447  		// In `<foo bar {{.}}`, the action is an attribute name.
   448  		c.state, c.attr = stateAttrName, attrNone
   449  	}
   450  	return c
   451  }
   452  
   453  // join joins the two contexts of a branch template node. The result is an
   454  // error context if either of the input contexts are error contexts, or if the
   455  // input contexts differ.
   456  func join(a, b context, node parse.Node, nodeName string) context {
   457  	if a.state == stateError {
   458  		return a
   459  	}
   460  	if b.state == stateError {
   461  		return b
   462  	}
   463  	if a.state == stateDead {
   464  		return b
   465  	}
   466  	if b.state == stateDead {
   467  		return a
   468  	}
   469  	if a.eq(b) {
   470  		return a
   471  	}
   472  
   473  	c := a
   474  	c.urlPart = b.urlPart
   475  	if c.eq(b) {
   476  		// The contexts differ only by urlPart.
   477  		c.urlPart = urlPartUnknown
   478  		return c
   479  	}
   480  
   481  	c = a
   482  	c.jsCtx = b.jsCtx
   483  	if c.eq(b) {
   484  		// The contexts differ only by jsCtx.
   485  		c.jsCtx = jsCtxUnknown
   486  		return c
   487  	}
   488  
   489  	// Allow a nudged context to join with an unnudged one.
   490  	// This means that
   491  	//   <p title={{if .C}}{{.}}{{end}}
   492  	// ends in an unquoted value state even though the else branch
   493  	// ends in stateBeforeValue.
   494  	if c, d := nudge(a), nudge(b); !(c.eq(a) && d.eq(b)) {
   495  		if e := join(c, d, node, nodeName); e.state != stateError {
   496  			return e
   497  		}
   498  	}
   499  
   500  	return context{
   501  		state: stateError,
   502  		err:   errorf(ErrBranchEnd, node, 0, "{{%s}} branches end in different contexts: %v, %v", nodeName, a, b),
   503  	}
   504  }
   505  
   506  // escapeBranch escapes a branch template node: "if", "range" and "with".
   507  func (e *escaper) escapeBranch(c context, n *parse.BranchNode, nodeName string) context {
   508  	if nodeName == "range" {
   509  		e.rangeContext = &rangeContext{outer: e.rangeContext}
   510  	}
   511  	c0 := e.escapeList(c, n.List)
   512  	if nodeName == "range" {
   513  		if c0.state != stateError {
   514  			c0 = joinRange(c0, e.rangeContext)
   515  		}
   516  		e.rangeContext = e.rangeContext.outer
   517  		if c0.state == stateError {
   518  			return c0
   519  		}
   520  
   521  		// The "true" branch of a "range" node can execute multiple times.
   522  		// We check that executing n.List once results in the same context
   523  		// as executing n.List twice.
   524  		e.rangeContext = &rangeContext{outer: e.rangeContext}
   525  		c1, _ := e.escapeListConditionally(c0, n.List, nil)
   526  		c0 = join(c0, c1, n, nodeName)
   527  		if c0.state == stateError {
   528  			e.rangeContext = e.rangeContext.outer
   529  			// Make clear that this is a problem on loop re-entry
   530  			// since developers tend to overlook that branch when
   531  			// debugging templates.
   532  			c0.err.Line = n.Line
   533  			c0.err.Description = "on range loop re-entry: " + c0.err.Description
   534  			return c0
   535  		}
   536  		c0 = joinRange(c0, e.rangeContext)
   537  		e.rangeContext = e.rangeContext.outer
   538  		if c0.state == stateError {
   539  			return c0
   540  		}
   541  	}
   542  	c1 := e.escapeList(c, n.ElseList)
   543  	return join(c0, c1, n, nodeName)
   544  }
   545  
   546  func joinRange(c0 context, rc *rangeContext) context {
   547  	// Merge contexts at break and continue statements into overall body context.
   548  	// In theory we could treat breaks differently from continues, but for now it is
   549  	// enough to treat them both as going back to the start of the loop (which may then stop).
   550  	for _, c := range rc.breaks {
   551  		c0 = join(c0, c, c.n, "range")
   552  		if c0.state == stateError {
   553  			c0.err.Line = c.n.(*parse.BreakNode).Line
   554  			c0.err.Description = "at range loop break: " + c0.err.Description
   555  			return c0
   556  		}
   557  	}
   558  	for _, c := range rc.continues {
   559  		c0 = join(c0, c, c.n, "range")
   560  		if c0.state == stateError {
   561  			c0.err.Line = c.n.(*parse.ContinueNode).Line
   562  			c0.err.Description = "at range loop continue: " + c0.err.Description
   563  			return c0
   564  		}
   565  	}
   566  	return c0
   567  }
   568  
   569  // escapeList escapes a list template node.
   570  func (e *escaper) escapeList(c context, n *parse.ListNode) context {
   571  	if n == nil {
   572  		return c
   573  	}
   574  	for _, m := range n.Nodes {
   575  		c = e.escape(c, m)
   576  		if c.state == stateDead {
   577  			break
   578  		}
   579  	}
   580  	return c
   581  }
   582  
   583  // escapeListConditionally escapes a list node but only preserves edits and
   584  // inferences in e if the inferences and output context satisfy filter.
   585  // It returns the best guess at an output context, and the result of the filter
   586  // which is the same as whether e was updated.
   587  func (e *escaper) escapeListConditionally(c context, n *parse.ListNode, filter func(*escaper, context) bool) (context, bool) {
   588  	e1 := makeEscaper(e.ns)
   589  	e1.rangeContext = e.rangeContext
   590  	// Make type inferences available to f.
   591  	for k, v := range e.output {
   592  		e1.output[k] = v
   593  	}
   594  	c = e1.escapeList(c, n)
   595  	ok := filter != nil && filter(&e1, c)
   596  	if ok {
   597  		// Copy inferences and edits from e1 back into e.
   598  		for k, v := range e1.output {
   599  			e.output[k] = v
   600  		}
   601  		for k, v := range e1.derived {
   602  			e.derived[k] = v
   603  		}
   604  		for k, v := range e1.called {
   605  			e.called[k] = v
   606  		}
   607  		for k, v := range e1.actionNodeEdits {
   608  			e.editActionNode(k, v)
   609  		}
   610  		for k, v := range e1.templateNodeEdits {
   611  			e.editTemplateNode(k, v)
   612  		}
   613  		for k, v := range e1.textNodeEdits {
   614  			e.editTextNode(k, v)
   615  		}
   616  	}
   617  	return c, ok
   618  }
   619  
   620  // escapeTemplate escapes a {{template}} call node.
   621  func (e *escaper) escapeTemplate(c context, n *parse.TemplateNode) context {
   622  	c, name := e.escapeTree(c, n, n.Name, n.Line)
   623  	if name != n.Name {
   624  		e.editTemplateNode(n, name)
   625  	}
   626  	return c
   627  }
   628  
   629  // escapeTree escapes the named template starting in the given context as
   630  // necessary and returns its output context.
   631  func (e *escaper) escapeTree(c context, node parse.Node, name string, line int) (context, string) {
   632  	// Mangle the template name with the input context to produce a reliable
   633  	// identifier.
   634  	dname := c.mangle(name)
   635  	e.called[dname] = true
   636  	if out, ok := e.output[dname]; ok {
   637  		// Already escaped.
   638  		return out, dname
   639  	}
   640  	t := e.template(name)
   641  	if t == nil {
   642  		// Two cases: The template exists but is empty, or has never been mentioned at
   643  		// all. Distinguish the cases in the error messages.
   644  		if e.ns.set[name] != nil {
   645  			return context{
   646  				state: stateError,
   647  				err:   errorf(ErrNoSuchTemplate, node, line, "%q is an incomplete or empty template", name),
   648  			}, dname
   649  		}
   650  		return context{
   651  			state: stateError,
   652  			err:   errorf(ErrNoSuchTemplate, node, line, "no such template %q", name),
   653  		}, dname
   654  	}
   655  	if dname != name {
   656  		// Use any template derived during an earlier call to escapeTemplate
   657  		// with different top level templates, or clone if necessary.
   658  		dt := e.template(dname)
   659  		if dt == nil {
   660  			dt = template.New(dname)
   661  			dt.Tree = &parse.Tree{Name: dname, Root: t.Root.CopyList()}
   662  			e.derived[dname] = dt
   663  		}
   664  		t = dt
   665  	}
   666  	return e.computeOutCtx(c, t), dname
   667  }
   668  
   669  // computeOutCtx takes a template and its start context and computes the output
   670  // context while storing any inferences in e.
   671  func (e *escaper) computeOutCtx(c context, t *template.Template) context {
   672  	// Propagate context over the body.
   673  	c1, ok := e.escapeTemplateBody(c, t)
   674  	if !ok {
   675  		// Look for a fixed point by assuming c1 as the output context.
   676  		if c2, ok2 := e.escapeTemplateBody(c1, t); ok2 {
   677  			c1, ok = c2, true
   678  		}
   679  		// Use c1 as the error context if neither assumption worked.
   680  	}
   681  	if !ok && c1.state != stateError {
   682  		return context{
   683  			state: stateError,
   684  			err:   errorf(ErrOutputContext, t.Tree.Root, 0, "cannot compute output context for template %s", t.Name()),
   685  		}
   686  	}
   687  	return c1
   688  }
   689  
   690  // escapeTemplateBody escapes the given template assuming the given output
   691  // context, and returns the best guess at the output context and whether the
   692  // assumption was correct.
   693  func (e *escaper) escapeTemplateBody(c context, t *template.Template) (context, bool) {
   694  	filter := func(e1 *escaper, c1 context) bool {
   695  		if c1.state == stateError {
   696  			// Do not update the input escaper, e.
   697  			return false
   698  		}
   699  		if !e1.called[t.Name()] {
   700  			// If t is not recursively called, then c1 is an
   701  			// accurate output context.
   702  			return true
   703  		}
   704  		// c1 is accurate if it matches our assumed output context.
   705  		return c.eq(c1)
   706  	}
   707  	// We need to assume an output context so that recursive template calls
   708  	// take the fast path out of escapeTree instead of infinitely recurring.
   709  	// Naively assuming that the input context is the same as the output
   710  	// works >90% of the time.
   711  	e.output[t.Name()] = c
   712  	return e.escapeListConditionally(c, t.Tree.Root, filter)
   713  }
   714  
   715  // delimEnds maps each delim to a string of characters that terminate it.
   716  var delimEnds = [...]string{
   717  	delimDoubleQuote: `"`,
   718  	delimSingleQuote: "'",
   719  	// Determined empirically by running the below in various browsers.
   720  	// var div = document.createElement("DIV");
   721  	// for (var i = 0; i < 0x10000; ++i) {
   722  	//   div.innerHTML = "<span title=x" + String.fromCharCode(i) + "-bar>";
   723  	//   if (div.getElementsByTagName("SPAN")[0].title.indexOf("bar") < 0)
   724  	//     document.write("<p>U+" + i.toString(16));
   725  	// }
   726  	delimSpaceOrTagEnd: " \t\n\f\r>",
   727  }
   728  
   729  var (
   730  	// Per WHATWG HTML specification, section 4.12.1.3, there are extremely
   731  	// complicated rules for how to handle the set of opening tags <!--,
   732  	// <script, and </script when they appear in JS literals (i.e. strings,
   733  	// regexs, and comments). The specification suggests a simple solution,
   734  	// rather than implementing the arcane ABNF, which involves simply escaping
   735  	// the opening bracket with \x3C. We use the below regex for this, since it
   736  	// makes doing the case-insensitive find-replace much simpler.
   737  	specialScriptTagRE          = regexp.MustCompile("(?i)<(script|/script|!--)")
   738  	specialScriptTagReplacement = []byte("\\x3C$1")
   739  )
   740  
   741  func containsSpecialScriptTag(s []byte) bool {
   742  	return specialScriptTagRE.Match(s)
   743  }
   744  
   745  func escapeSpecialScriptTags(s []byte) []byte {
   746  	return specialScriptTagRE.ReplaceAll(s, specialScriptTagReplacement)
   747  }
   748  
   749  var doctypeBytes = []byte("<!DOCTYPE")
   750  
   751  // escapeText escapes a text template node.
   752  func (e *escaper) escapeText(c context, n *parse.TextNode) context {
   753  	s, written, i, b := n.Text, 0, 0, new(bytes.Buffer)
   754  	for i != len(s) {
   755  		c1, nread := contextAfterText(c, s[i:])
   756  		i1 := i + nread
   757  		if c.state == stateText || c.state == stateRCDATA {
   758  			end := i1
   759  			if c1.state != c.state {
   760  				for j := end - 1; j >= i; j-- {
   761  					if s[j] == '<' {
   762  						end = j
   763  						break
   764  					}
   765  				}
   766  			}
   767  			for j := i; j < end; j++ {
   768  				if s[j] == '<' && !bytes.HasPrefix(bytes.ToUpper(s[j:]), doctypeBytes) {
   769  					b.Write(s[written:j])
   770  					b.WriteString("&lt;")
   771  					written = j + 1
   772  				}
   773  			}
   774  		} else if isComment(c.state) && c.delim == delimNone {
   775  			switch c.state {
   776  			case stateJSBlockCmt:
   777  				// https://es5.github.io/#x7.4:
   778  				// "Comments behave like white space and are
   779  				// discarded except that, if a MultiLineComment
   780  				// contains a line terminator character, then
   781  				// the entire comment is considered to be a
   782  				// LineTerminator for purposes of parsing by
   783  				// the syntactic grammar."
   784  				if bytes.ContainsAny(s[written:i1], "\n\r\u2028\u2029") {
   785  					b.WriteByte('\n')
   786  				} else {
   787  					b.WriteByte(' ')
   788  				}
   789  			case stateCSSBlockCmt:
   790  				b.WriteByte(' ')
   791  			}
   792  			written = i1
   793  		}
   794  		if c.state != c1.state && isComment(c1.state) && c1.delim == delimNone {
   795  			// Preserve the portion between written and the comment start.
   796  			cs := i1 - 2
   797  			if c1.state == stateHTMLCmt || c1.state == stateJSHTMLOpenCmt {
   798  				// "<!--" instead of "/*" or "//"
   799  				cs -= 2
   800  			} else if c1.state == stateJSHTMLCloseCmt {
   801  				// "-->" instead of "/*" or "//"
   802  				cs -= 1
   803  			}
   804  			b.Write(s[written:cs])
   805  			written = i1
   806  		}
   807  		if isInScriptLiteral(c.state) && containsSpecialScriptTag(s[i:i1]) {
   808  			b.Write(s[written:i])
   809  			b.Write(escapeSpecialScriptTags(s[i:i1]))
   810  			written = i1
   811  		}
   812  		if i == i1 && c.state == c1.state {
   813  			panic(fmt.Sprintf("infinite loop from %v to %v on %q..%q", c, c1, s[:i], s[i:]))
   814  		}
   815  		c, i = c1, i1
   816  	}
   817  
   818  	if written != 0 && c.state != stateError {
   819  		if !isComment(c.state) || c.delim != delimNone {
   820  			b.Write(n.Text[written:])
   821  		}
   822  		e.editTextNode(n, b.Bytes())
   823  	}
   824  	return c
   825  }
   826  
   827  // contextAfterText starts in context c, consumes some tokens from the front of
   828  // s, then returns the context after those tokens and the unprocessed suffix.
   829  func contextAfterText(c context, s []byte) (context, int) {
   830  	if c.delim == delimNone {
   831  		c1, i := tSpecialTagEnd(c, s)
   832  		if i == 0 {
   833  			// A special end tag (`</script>`) has been seen and
   834  			// all content preceding it has been consumed.
   835  			return c1, 0
   836  		}
   837  		// Consider all content up to any end tag.
   838  		return transitionFunc[c.state](c, s[:i])
   839  	}
   840  
   841  	// We are at the beginning of an attribute value.
   842  
   843  	i := bytes.IndexAny(s, delimEnds[c.delim])
   844  	if i == -1 {
   845  		i = len(s)
   846  	}
   847  	if c.delim == delimSpaceOrTagEnd {
   848  		// https://www.w3.org/TR/html5/syntax.html#attribute-value-(unquoted)-state
   849  		// lists the runes below as error characters.
   850  		// Error out because HTML parsers may differ on whether
   851  		// "<a id= onclick=f("     ends inside id's or onclick's value,
   852  		// "<a class=`foo "        ends inside a value,
   853  		// "<a style=font:'Arial'" needs open-quote fixup.
   854  		// IE treats '`' as a quotation character.
   855  		if j := bytes.IndexAny(s[:i], "\"'<=`"); j >= 0 {
   856  			return context{
   857  				state: stateError,
   858  				err:   errorf(ErrBadHTML, nil, 0, "%q in unquoted attr: %q", s[j:j+1], s[:i]),
   859  			}, len(s)
   860  		}
   861  	}
   862  	if i == len(s) {
   863  		// Remain inside the attribute.
   864  		// Decode the value so non-HTML rules can easily handle
   865  		//     <button onclick="alert(&quot;Hi!&quot;)">
   866  		// without having to entity decode token boundaries.
   867  		for u := []byte(html.UnescapeString(string(s))); len(u) != 0; {
   868  			c1, i1 := transitionFunc[c.state](c, u)
   869  			c, u = c1, u[i1:]
   870  		}
   871  		return c, len(s)
   872  	}
   873  
   874  	element := c.element
   875  
   876  	// If this is a non-JS "type" attribute inside "script" tag, do not treat the contents as JS.
   877  	if c.state == stateAttr && c.element == elementScript && c.attr == attrScriptType && !isJSType(string(s[:i])) {
   878  		element = elementNone
   879  	}
   880  
   881  	if c.delim != delimSpaceOrTagEnd {
   882  		// Consume any quote.
   883  		i++
   884  	}
   885  	// On exiting an attribute, we discard all state information
   886  	// except the state and element.
   887  	return context{state: stateTag, element: element}, i
   888  }
   889  
   890  // editActionNode records a change to an action pipeline for later commit.
   891  func (e *escaper) editActionNode(n *parse.ActionNode, cmds []string) {
   892  	if _, ok := e.actionNodeEdits[n]; ok {
   893  		panic(fmt.Sprintf("node %s shared between templates", n))
   894  	}
   895  	e.actionNodeEdits[n] = cmds
   896  }
   897  
   898  // editTemplateNode records a change to a {{template}} callee for later commit.
   899  func (e *escaper) editTemplateNode(n *parse.TemplateNode, callee string) {
   900  	if _, ok := e.templateNodeEdits[n]; ok {
   901  		panic(fmt.Sprintf("node %s shared between templates", n))
   902  	}
   903  	e.templateNodeEdits[n] = callee
   904  }
   905  
   906  // editTextNode records a change to a text node for later commit.
   907  func (e *escaper) editTextNode(n *parse.TextNode, text []byte) {
   908  	if _, ok := e.textNodeEdits[n]; ok {
   909  		panic(fmt.Sprintf("node %s shared between templates", n))
   910  	}
   911  	e.textNodeEdits[n] = text
   912  }
   913  
   914  // commit applies changes to actions and template calls needed to contextually
   915  // autoescape content and adds any derived templates to the set.
   916  func (e *escaper) commit() {
   917  	for name := range e.output {
   918  		e.template(name).Funcs(funcMap)
   919  	}
   920  	// Any template from the name space associated with this escaper can be used
   921  	// to add derived templates to the underlying text/template name space.
   922  	tmpl := e.arbitraryTemplate()
   923  	for _, t := range e.derived {
   924  		if _, err := tmpl.text.AddParseTree(t.Name(), t.Tree); err != nil {
   925  			panic("error adding derived template")
   926  		}
   927  	}
   928  	for n, s := range e.actionNodeEdits {
   929  		ensurePipelineContains(n.Pipe, s)
   930  	}
   931  	for n, name := range e.templateNodeEdits {
   932  		n.Name = name
   933  	}
   934  	for n, s := range e.textNodeEdits {
   935  		n.Text = s
   936  	}
   937  	// Reset state that is specific to this commit so that the same changes are
   938  	// not re-applied to the template on subsequent calls to commit.
   939  	e.called = make(map[string]bool)
   940  	e.actionNodeEdits = make(map[*parse.ActionNode][]string)
   941  	e.templateNodeEdits = make(map[*parse.TemplateNode]string)
   942  	e.textNodeEdits = make(map[*parse.TextNode][]byte)
   943  }
   944  
   945  // template returns the named template given a mangled template name.
   946  func (e *escaper) template(name string) *template.Template {
   947  	// Any template from the name space associated with this escaper can be used
   948  	// to look up templates in the underlying text/template name space.
   949  	t := e.arbitraryTemplate().text.Lookup(name)
   950  	if t == nil {
   951  		t = e.derived[name]
   952  	}
   953  	return t
   954  }
   955  
   956  // arbitraryTemplate returns an arbitrary template from the name space
   957  // associated with e and panics if no templates are found.
   958  func (e *escaper) arbitraryTemplate() *Template {
   959  	for _, t := range e.ns.set {
   960  		return t
   961  	}
   962  	panic("no templates in name space")
   963  }
   964  
   965  // Forwarding functions so that clients need only import this package
   966  // to reach the general escaping functions of text/template.
   967  
   968  // HTMLEscape writes to w the escaped HTML equivalent of the plain text data b.
   969  func HTMLEscape(w io.Writer, b []byte) {
   970  	template.HTMLEscape(w, b)
   971  }
   972  
   973  // HTMLEscapeString returns the escaped HTML equivalent of the plain text data s.
   974  func HTMLEscapeString(s string) string {
   975  	return template.HTMLEscapeString(s)
   976  }
   977  
   978  // HTMLEscaper returns the escaped HTML equivalent of the textual
   979  // representation of its arguments.
   980  func HTMLEscaper(args ...any) string {
   981  	return template.HTMLEscaper(args...)
   982  }
   983  
   984  // JSEscape writes to w the escaped JavaScript equivalent of the plain text data b.
   985  func JSEscape(w io.Writer, b []byte) {
   986  	template.JSEscape(w, b)
   987  }
   988  
   989  // JSEscapeString returns the escaped JavaScript equivalent of the plain text data s.
   990  func JSEscapeString(s string) string {
   991  	return template.JSEscapeString(s)
   992  }
   993  
   994  // JSEscaper returns the escaped JavaScript equivalent of the textual
   995  // representation of its arguments.
   996  func JSEscaper(args ...any) string {
   997  	return template.JSEscaper(args...)
   998  }
   999  
  1000  // URLQueryEscaper returns the escaped value of the textual representation of
  1001  // its arguments in a form suitable for embedding in a URL query.
  1002  func URLQueryEscaper(args ...any) string {
  1003  	return template.URLQueryEscaper(args...)
  1004  }
  1005  

View as plain text