1//! Operations related to UTF-8 validation.
2
3use super::Utf8Error;
4use crate::intrinsics::const_eval_select;
5
6/// Returns the initial codepoint accumulator for the first byte.
7/// The first byte is special, only want bottom 5 bits for width 2, 4 bits
8/// for width 3, and 3 bits for width 4.
9#[inline]
10const fn utf8_first_byte(byte: u8, width: u32) -> u32 {
11 (byte & (0x7F >> width)) as u32
12}
13
14/// Returns the value of `ch` updated with continuation byte `byte`.
15#[inline]
16const fn utf8_acc_cont_byte(ch: u32, byte: u8) -> u32 {
17 (ch << 6) | (byte & CONT_MASK) as u32
18}
19
20/// Checks whether the byte is a UTF-8 continuation byte (i.e., starts with the
21/// bits `10`).
22#[inline]
23pub(super) const fn utf8_is_cont_byte(byte: u8) -> bool {
24 (byte as i8) < -64
25}
26
27/// Reads the next code point out of a byte iterator (assuming a
28/// UTF-8-like encoding).
29///
30/// # Safety
31///
32/// `bytes` must produce a valid UTF-8-like (UTF-8 or WTF-8) string
33#[unstable(feature = "str_internals", issue = "none")]
34#[inline]
35pub unsafe fn next_code_point<'a, I: Iterator<Item = &'a u8>>(bytes: &mut I) -> Option<u32> {
36 // Decode UTF-8
37 let x = *bytes.next()?;
38 if x < 128 {
39 return Some(x as u32);
40 }
41
42 // Multibyte case follows
43 // Decode from a byte combination out of: [[[x y] z] w]
44 // NOTE: Performance is sensitive to the exact formulation here
45 let init = utf8_first_byte(x, 2);
46 // SAFETY: `bytes` produces an UTF-8-like string,
47 // so the iterator must produce a value here.
48 let y = unsafe { *bytes.next().unwrap_unchecked() };
49 let mut ch = utf8_acc_cont_byte(init, y);
50 if x >= 0xE0 {
51 // [[x y z] w] case
52 // 5th bit in 0xE0 .. 0xEF is always clear, so `init` is still valid
53 // SAFETY: `bytes` produces an UTF-8-like string,
54 // so the iterator must produce a value here.
55 let z = unsafe { *bytes.next().unwrap_unchecked() };
56 let y_z = utf8_acc_cont_byte((y & CONT_MASK) as u32, z);
57 ch = init << 12 | y_z;
58 if x >= 0xF0 {
59 // [x y z w] case
60 // use only the lower 3 bits of `init`
61 // SAFETY: `bytes` produces an UTF-8-like string,
62 // so the iterator must produce a value here.
63 let w = unsafe { *bytes.next().unwrap_unchecked() };
64 ch = (init & 7) << 18 | utf8_acc_cont_byte(y_z, w);
65 }
66 }
67
68 Some(ch)
69}
70
71/// Reads the last code point out of a byte iterator (assuming a
72/// UTF-8-like encoding).
73///
74/// # Safety
75///
76/// `bytes` must produce a valid UTF-8-like (UTF-8 or WTF-8) string
77#[inline]
78pub(super) unsafe fn next_code_point_reverse<'a, I>(bytes: &mut I) -> Option<u32>
79where
80 I: DoubleEndedIterator<Item = &'a u8>,
81{
82 // Decode UTF-8
83 let w = match *bytes.next_back()? {
84 next_byte if next_byte < 128 => return Some(next_byte as u32),
85 back_byte => back_byte,
86 };
87
88 // Multibyte case follows
89 // Decode from a byte combination out of: [x [y [z w]]]
90 let mut ch;
91 // SAFETY: `bytes` produces an UTF-8-like string,
92 // so the iterator must produce a value here.
93 let z = unsafe { *bytes.next_back().unwrap_unchecked() };
94 ch = utf8_first_byte(z, 2);
95 if utf8_is_cont_byte(z) {
96 // SAFETY: `bytes` produces an UTF-8-like string,
97 // so the iterator must produce a value here.
98 let y = unsafe { *bytes.next_back().unwrap_unchecked() };
99 ch = utf8_first_byte(y, 3);
100 if utf8_is_cont_byte(y) {
101 // SAFETY: `bytes` produces an UTF-8-like string,
102 // so the iterator must produce a value here.
103 let x = unsafe { *bytes.next_back().unwrap_unchecked() };
104 ch = utf8_first_byte(x, 4);
105 ch = utf8_acc_cont_byte(ch, y);
106 }
107 ch = utf8_acc_cont_byte(ch, z);
108 }
109 ch = utf8_acc_cont_byte(ch, w);
110
111 Some(ch)
112}
113
114const NONASCII_MASK: usize = usize::repeat_u8(0x80);
115
116/// Returns `true` if any byte in the word `x` is nonascii (>= 128).
117#[inline]
118const fn contains_nonascii(x: usize) -> bool {
119 (x & NONASCII_MASK) != 0
120}
121
122/// Walks through `v` checking that it's a valid UTF-8 sequence,
123/// returning `Ok(())` in that case, or, if it is invalid, `Err(err)`.
124#[inline(always)]
125#[rustc_allow_const_fn_unstable(const_eval_select)] // fallback impl has same behavior
126pub(super) const fn run_utf8_validation(v: &[u8]) -> Result<(), Utf8Error> {
127 let mut index = 0;
128 let len = v.len();
129
130 const USIZE_BYTES: usize = size_of::<usize>();
131
132 let ascii_block_size = 2 * USIZE_BYTES;
133 let blocks_end = if len >= ascii_block_size { len - ascii_block_size + 1 } else { 0 };
134 // Below, we safely fall back to a slower codepath if the offset is `usize::MAX`,
135 // so the end-to-end behavior is the same at compiletime and runtime.
136 let align = const_eval_select!(
137 @capture { v: &[u8] } -> usize:
138 if const {
139 usize::MAX
140 } else {
141 v.as_ptr().align_offset(USIZE_BYTES)
142 }
143 );
144
145 while index < len {
146 let old_offset = index;
147 macro_rules! err {
148 ($error_len: expr) => {
149 return Err(Utf8Error { valid_up_to: old_offset, error_len: $error_len })
150 };
151 }
152
153 macro_rules! next {
154 () => {{
155 index += 1;
156 // we needed data, but there was none: error!
157 if index >= len {
158 err!(None)
159 }
160 v[index]
161 }};
162 }
163
164 let first = v[index];
165 if first >= 128 {
166 let w = utf8_char_width(first);
167 // 2-byte encoding is for codepoints \u{0080} to \u{07ff}
168 // first C2 80 last DF BF
169 // 3-byte encoding is for codepoints \u{0800} to \u{ffff}
170 // first E0 A0 80 last EF BF BF
171 // excluding surrogates codepoints \u{d800} to \u{dfff}
172 // ED A0 80 to ED BF BF
173 // 4-byte encoding is for codepoints \u{10000} to \u{10ffff}
174 // first F0 90 80 80 last F4 8F BF BF
175 //
176 // Use the UTF-8 syntax from the RFC
177 //
178 // https://tools.ietf.org/html/rfc3629
179 // UTF8-1 = %x00-7F
180 // UTF8-2 = %xC2-DF UTF8-tail
181 // UTF8-3 = %xE0 %xA0-BF UTF8-tail / %xE1-EC 2( UTF8-tail ) /
182 // %xED %x80-9F UTF8-tail / %xEE-EF 2( UTF8-tail )
183 // UTF8-4 = %xF0 %x90-BF 2( UTF8-tail ) / %xF1-F3 3( UTF8-tail ) /
184 // %xF4 %x80-8F 2( UTF8-tail )
185 match w {
186 2 => {
187 if next!() as i8 >= -64 {
188 err!(Some(1))
189 }
190 }
191 3 => {
192 match (first, next!()) {
193 (0xE0, 0xA0..=0xBF)
194 | (0xE1..=0xEC, 0x80..=0xBF)
195 | (0xED, 0x80..=0x9F)
196 | (0xEE..=0xEF, 0x80..=0xBF) => {}
197 _ => err!(Some(1)),
198 }
199 if next!() as i8 >= -64 {
200 err!(Some(2))
201 }
202 }
203 4 => {
204 match (first, next!()) {
205 (0xF0, 0x90..=0xBF) | (0xF1..=0xF3, 0x80..=0xBF) | (0xF4, 0x80..=0x8F) => {}
206 _ => err!(Some(1)),
207 }
208 if next!() as i8 >= -64 {
209 err!(Some(2))
210 }
211 if next!() as i8 >= -64 {
212 err!(Some(3))
213 }
214 }
215 _ => err!(Some(1)),
216 }
217 index += 1;
218 } else {
219 // Ascii case, try to skip forward quickly.
220 // When the pointer is aligned, read 2 words of data per iteration
221 // until we find a word containing a non-ascii byte.
222 if align != usize::MAX && align.wrapping_sub(index) % USIZE_BYTES == 0 {
223 let ptr = v.as_ptr();
224 while index < blocks_end {
225 // SAFETY: since `align - index` and `ascii_block_size` are
226 // multiples of `USIZE_BYTES`, `block = ptr.add(index)` is
227 // always aligned with a `usize` so it's safe to dereference
228 // both `block` and `block.add(1)`.
229 unsafe {
230 let block = ptr.add(index) as *const usize;
231 // break if there is a nonascii byte
232 let zu = contains_nonascii(*block);
233 let zv = contains_nonascii(*block.add(1));
234 if zu || zv {
235 break;
236 }
237 }
238 index += ascii_block_size;
239 }
240 // step from the point where the wordwise loop stopped
241 while index < len && v[index] < 128 {
242 index += 1;
243 }
244 } else {
245 index += 1;
246 }
247 }
248 }
249
250 Ok(())
251}
252
253// https://tools.ietf.org/html/rfc3629
254const UTF8_CHAR_WIDTH: &[u8; 256] = &[
255 // 1 2 3 4 5 6 7 8 9 A B C D E F
256 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 0
257 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 1
258 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 2
259 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 3
260 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 4
261 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 5
262 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 6
263 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, // 7
264 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 8
265 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 9
266 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // A
267 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // B
268 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, // C
269 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, // D
270 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, // E
271 4, 4, 4, 4, 4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // F
272];
273
274/// Given a first byte, determines how many bytes are in this UTF-8 character.
275#[unstable(feature = "str_internals", issue = "none")]
276#[must_use]
277#[inline]
278pub const fn utf8_char_width(b: u8) -> usize {
279 UTF8_CHAR_WIDTH[b as usize] as usize
280}
281
282/// Mask of the value bits of a continuation byte.
283const CONT_MASK: u8 = 0b0011_1111;
284