...

Source file src/html/template/escape.go

Documentation: html/template

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

View as plain text