def write(self, *vals):
ctx = context.force_epsilon_transition(self.ctx_)
underlying = self.underlying_
for val in vals:
ctx_after, esc_modes, problem = (escaping.esc_mode_for_hole(ctx))
if problem is not None:
raise AutoescapeError(problem)
ctx = ctx_after
for esc_mode in esc_modes:
escaper = escaping.SANITIZER_FOR_ESC_MODE[esc_mode]
val = escaper(val)
underlying.write(val)
self.ctx_ = ctx
def write(self, *vals):
ctx = context.force_epsilon_transition(self.ctx_)
underlying = self.underlying_
for val in vals:
ctx_after, esc_modes, problem = (
escaping.esc_mode_for_hole(ctx)
)
if problem is not None:
raise AutoescapeError(problem)
ctx = ctx_after
for esc_mode in esc_modes:
escaper = escaping.SANITIZER_FOR_ESC_MODE[esc_mode]
val = escaper(val)
underlying.write(val)
self.ctx_ = ctx
def esc_mode_for_hole(context_before):
"""
Given a context in which an untrusted value hole appears, computes the
escaping modes needed to render that untrusted value safe for interpolation
and the context after the hole.
context_before - The input context before the substitution.
Returns (context after, (escaping_modes...,))
"""
ctx = context.force_epsilon_transition(context_before)
state, url_part = context.state_of(ctx), context.url_part_of(ctx)
esc_modes = [ESC_MODE_FOR_STATE[state]]
problem = None
if url_part == context.URL_PART_NONE:
# Make sure that at the start of a URL, we filter out dangerous
# protocols.
if state in (
context.STATE_URL, context.STATE_CSS_URL, context.STATE_CSSDQ_URL,
context.STATE_CSSSQ_URL):
esc_modes = [ESC_MODE_FILTER_URL, ESC_MODE_NORMALIZE_URL]
ctx = (ctx & ~context.URL_PART_ALL) | context.URL_PART_PRE_QUERY
elif state in (context.STATE_CSSDQ_STR, context.STATE_CSSSQ_STR):
esc_modes[:0] = [ESC_MODE_FILTER_URL]
ctx = (ctx & ~context.URL_PART_ALL) | context.URL_PART_PRE_QUERY
elif url_part == context.URL_PART_PRE_QUERY:
if state not in (context.STATE_CSSDQ_STR, context.STATE_CSSSQ_STR):
esc_modes[0] = ESC_MODE_NORMALIZE_URL
elif url_part == context.URL_PART_QUERY_OR_FRAG:
esc_modes[0] = ESC_MODE_ESCAPE_URL
elif url_part == context.URL_PART_UNKNOWN:
ctx = context.STATE_ERROR
problem = 'hole appears in an ambiguous URL context'
if state == context.STATE_JS:
ctx = (ctx & ~context.JS_CTX_ALL) | context.JS_CTX_DIV_OP
elif (state == context.STATE_ATTR_NAME
and context.attr_type_of(ctx) != context.ATTR_NONE):
esc_modes[0] = ESC_MODE_FILTER_HTML_ATTR_SUFFIX
if esc_modes[0] is None:
ctx = context.STATE_ERROR
esc_mode = esc_modes[-1]
delim_type = context.delim_type_of(ctx)
if delim_type != context.DELIM_NONE:
# Figure out how to escape the attribute value.
if esc_mode != ESC_MODE_ESCAPE_HTML_ATTRIBUTE:
esc_modes.append(ESC_MODE_ESCAPE_HTML_ATTRIBUTE)
if (context.delim_type_of(context_before) == context.DELIM_NONE
and delim_type == context.DELIM_SPACE_OR_TAG_END):
esc_modes.append(ESC_MODE_OPEN_QUOTE)
last, i = esc_modes[0], 1
while i < len(esc_modes):
curr = esc_modes[i]
# If, for all x, f(g(x)) == g(x), we can skip f.
if (last, curr) in REDUNDANT_ESC_MODES:
esc_modes[i:i+1] = []
else:
last = curr
i += 1
return ctx, tuple(esc_modes), problem
def esc_mode_for_hole(context_before):
"""
Given a context in which an untrusted value hole appears, computes the
escaping modes needed to render that untrusted value safe for interpolation
and the context after the hole.
context_before - The input context before the substitution.
Returns (context after, (escaping_modes...,))
"""
ctx = context.force_epsilon_transition(context_before)
state, url_part = context.state_of(ctx), context.url_part_of(ctx)
esc_modes = [ESC_MODE_FOR_STATE[state]]
problem = None
if url_part == context.URL_PART_NONE:
# Make sure that at the start of a URL, we filter out dangerous
# protocols.
if state in (context.STATE_URL, context.STATE_CSS_URL,
context.STATE_CSSDQ_URL, context.STATE_CSSSQ_URL):
esc_modes = [ESC_MODE_FILTER_URL, ESC_MODE_NORMALIZE_URL]
ctx = (ctx & ~context.URL_PART_ALL) | context.URL_PART_PRE_QUERY
elif state in (context.STATE_CSSDQ_STR, context.STATE_CSSSQ_STR):
esc_modes[:0] = [ESC_MODE_FILTER_URL]
ctx = (ctx & ~context.URL_PART_ALL) | context.URL_PART_PRE_QUERY
elif url_part == context.URL_PART_PRE_QUERY:
if state not in (context.STATE_CSSDQ_STR, context.STATE_CSSSQ_STR):
esc_modes[0] = ESC_MODE_NORMALIZE_URL
elif url_part == context.URL_PART_QUERY_OR_FRAG:
esc_modes[0] = ESC_MODE_ESCAPE_URL
elif url_part == context.URL_PART_UNKNOWN:
ctx = context.STATE_ERROR
problem = 'hole appears in an ambiguous URL context'
if state == context.STATE_JS:
ctx = (ctx & ~context.JS_CTX_ALL) | context.JS_CTX_DIV_OP
elif (state == context.STATE_ATTR_NAME
and context.attr_type_of(ctx) != context.ATTR_NONE):
esc_modes[0] = ESC_MODE_FILTER_HTML_ATTR_SUFFIX
if esc_modes[0] is None:
ctx = context.STATE_ERROR
esc_mode = esc_modes[-1]
delim_type = context.delim_type_of(ctx)
if delim_type != context.DELIM_NONE:
# Figure out how to escape the attribute value.
if esc_mode != ESC_MODE_ESCAPE_HTML_ATTRIBUTE:
esc_modes.append(ESC_MODE_ESCAPE_HTML_ATTRIBUTE)
if (context.delim_type_of(context_before) == context.DELIM_NONE
and delim_type == context.DELIM_SPACE_OR_TAG_END):
esc_modes.append(ESC_MODE_OPEN_QUOTE)
last, i = esc_modes[0], 1
while i < len(esc_modes):
curr = esc_modes[i]
# If, for all x, f(g(x)) == g(x), we can skip f.
if (last, curr) in REDUNDANT_ESC_MODES:
esc_modes[i:i + 1] = []
else:
last = curr
i += 1
return ctx, tuple(esc_modes), problem
