xref: /unit/src/nxt_http_parse.c (revision 576:7567b2a5ebfb)
1 
2 /*
3  * Copyright (C) NGINX, Inc.
4  * Copyright (C) Valentin V. Bartenev
5  */
6 
7 #include <nxt_main.h>
8 
9 
10 static nxt_int_t nxt_http_parse_unusual_target(nxt_http_request_parse_t *rp,
11     u_char **pos, u_char *end);
12 static nxt_int_t nxt_http_parse_request_line(nxt_http_request_parse_t *rp,
13     u_char **pos, u_char *end);
14 static nxt_int_t nxt_http_parse_field_name(nxt_http_request_parse_t *rp,
15     u_char **pos, u_char *end);
16 static nxt_int_t nxt_http_parse_field_value(nxt_http_request_parse_t *rp,
17     u_char **pos, u_char *end);
18 static u_char *nxt_http_lookup_field_end(u_char *p, u_char *end);
19 static nxt_int_t nxt_http_parse_field_end(nxt_http_request_parse_t *rp,
20     u_char **pos, u_char *end);
21 
22 static nxt_int_t nxt_http_parse_complex_target(nxt_http_request_parse_t *rp);
23 
24 static nxt_int_t nxt_http_field_hash_test(nxt_lvlhsh_query_t *lhq, void *data);
25 static void *nxt_http_field_hash_alloc(void *pool, size_t size);
26 static void nxt_http_field_hash_free(void *pool, void *p);
27 
28 static nxt_int_t nxt_http_field_hash_collision(nxt_lvlhsh_query_t *lhq,
29     void *data);
30 
31 
32 #define NXT_HTTP_MAX_FIELD_NAME         0xff
33 #define NXT_HTTP_MAX_FIELD_VALUE        NXT_INT32_T_MAX
34 
35 #define NXT_HTTP_FIELD_LVLHSH_SHIFT     5
36 
37 #define NXT_HTTP_FIELD_HASH_INIT        159406
38 #define nxt_http_field_hash_char(h, c)  (((h) << 4) + (h) + (c))
39 #define nxt_http_field_hash_end(h)      (((h) >> 16) ^ (h))
40 
41 
42 typedef enum {
43     NXT_HTTP_TARGET_SPACE = 1,   /* \s  */
44     NXT_HTTP_TARGET_HASH,        /*  #  */
45     NXT_HTTP_TARGET_AGAIN,
46     NXT_HTTP_TARGET_BAD,         /* \0\r\n */
47 
48     /* traps below are used for extended check only */
49 
50     NXT_HTTP_TARGET_SLASH = 5,   /*  /  */
51     NXT_HTTP_TARGET_DOT,         /*  .  */
52     NXT_HTTP_TARGET_ARGS_MARK,   /*  ?  */
53     NXT_HTTP_TARGET_QUOTE_MARK,  /*  %  */
54     NXT_HTTP_TARGET_PLUS,        /*  +  */
55 } nxt_http_target_traps_e;
56 
57 
58 static const uint8_t  nxt_http_target_chars[256] nxt_aligned(64) = {
59     /* \0                               \n        \r       */
60         4, 0, 0, 0,  0, 0, 0, 0,   0, 0, 4, 0,  0, 4, 0, 0,
61         0, 0, 0, 0,  0, 0, 0, 0,   0, 0, 0, 0,  0, 0, 0, 0,
62 
63     /* \s  !  "  #   $  %  &  '    (  )  *  +   ,  -  .  / */
64         1, 0, 0, 2,  0, 8, 0, 0,   0, 0, 0, 9,  0, 0, 6, 5,
65 
66     /*  0  1  2  3   4  5  6  7    8  9  :  ;   <  =  >  ? */
67         0, 0, 0, 0,  0, 0, 0, 0,   0, 0, 0, 0,  0, 0, 0, 7,
68 };
69 
70 
71 nxt_inline nxt_http_target_traps_e
72 nxt_http_parse_target(u_char **pos, u_char *end)
73 {
74     u_char      *p;
75     nxt_uint_t  trap;
76 
77     p = *pos;
78 
79     while (nxt_fast_path(end - p >= 10)) {
80 
81 #define nxt_target_test_char(ch)                                              \
82                                                                               \
83         trap = nxt_http_target_chars[ch];                                     \
84                                                                               \
85         if (nxt_slow_path(trap != 0)) {                                       \
86             *pos = &(ch);                                                     \
87             return trap;                                                      \
88         }
89 
90 /* enddef */
91 
92         nxt_target_test_char(p[0]);
93         nxt_target_test_char(p[1]);
94         nxt_target_test_char(p[2]);
95         nxt_target_test_char(p[3]);
96 
97         nxt_target_test_char(p[4]);
98         nxt_target_test_char(p[5]);
99         nxt_target_test_char(p[6]);
100         nxt_target_test_char(p[7]);
101 
102         nxt_target_test_char(p[8]);
103         nxt_target_test_char(p[9]);
104 
105         p += 10;
106     }
107 
108     while (p != end) {
109         nxt_target_test_char(*p); p++;
110     }
111 
112     return NXT_HTTP_TARGET_AGAIN;
113 }
114 
115 
116 nxt_int_t
117 nxt_http_parse_request_init(nxt_http_request_parse_t *rp, nxt_mp_t *mp)
118 {
119     rp->mem_pool = mp;
120 
121     rp->fields = nxt_list_create(mp, 8, sizeof(nxt_http_field_t));
122     if (nxt_slow_path(rp->fields == NULL)){
123         return NXT_ERROR;
124     }
125 
126     rp->field_hash = NXT_HTTP_FIELD_HASH_INIT;
127 
128     return NXT_OK;
129 }
130 
131 
132 nxt_int_t
133 nxt_http_parse_request(nxt_http_request_parse_t *rp, nxt_buf_mem_t *b)
134 {
135     nxt_int_t  rc;
136 
137     if (rp->handler == NULL) {
138         rp->handler = &nxt_http_parse_request_line;
139     }
140 
141     do {
142         rc = rp->handler(rp, &b->pos, b->free);
143     } while (rc == NXT_OK);
144 
145     return rc;
146 }
147 
148 
149 nxt_int_t
150 nxt_http_parse_fields(nxt_http_request_parse_t *rp, nxt_buf_mem_t *b)
151 {
152     nxt_int_t  rc;
153 
154     if (rp->handler == NULL) {
155         rp->handler = &nxt_http_parse_field_name;
156     }
157 
158     do {
159         rc = rp->handler(rp, &b->pos, b->free);
160     } while (rc == NXT_OK);
161 
162     return rc;
163 }
164 
165 
166 static nxt_int_t
167 nxt_http_parse_request_line(nxt_http_request_parse_t *rp, u_char **pos,
168     u_char *end)
169 {
170     u_char                   *p, ch, *after_slash;
171     nxt_int_t                rc;
172     nxt_http_ver_t           ver;
173     nxt_http_target_traps_e  trap;
174 
175     static const nxt_http_ver_t  http11 = { "HTTP/1.1" };
176     static const nxt_http_ver_t  http10 = { "HTTP/1.0" };
177 
178     p = *pos;
179 
180     rp->method.start = p;
181 
182     for ( ;; ) {
183 
184         while (nxt_fast_path(end - p >= 8)) {
185 
186 #define nxt_method_test_char(ch)                                              \
187                                                                               \
188             if (nxt_slow_path((ch) < 'A' || (ch) > 'Z')) {                    \
189                 p = &(ch);                                                    \
190                 goto method_unusual_char;                                     \
191             }
192 
193 /* enddef */
194 
195             nxt_method_test_char(p[0]);
196             nxt_method_test_char(p[1]);
197             nxt_method_test_char(p[2]);
198             nxt_method_test_char(p[3]);
199 
200             nxt_method_test_char(p[4]);
201             nxt_method_test_char(p[5]);
202             nxt_method_test_char(p[6]);
203             nxt_method_test_char(p[7]);
204 
205             p += 8;
206         }
207 
208         while (p != end) {
209             nxt_method_test_char(*p); p++;
210         }
211 
212         return NXT_AGAIN;
213 
214     method_unusual_char:
215 
216         ch = *p;
217 
218         if (nxt_fast_path(ch == ' ')) {
219             rp->method.length = p - rp->method.start;
220             break;
221         }
222 
223         if (ch == '_' || ch == '-') {
224             p++;
225             continue;
226         }
227 
228         if (rp->method.start == p && (ch == NXT_CR || ch == NXT_LF)) {
229             rp->method.start++;
230             p++;
231             continue;
232         }
233 
234         return NXT_HTTP_PARSE_INVALID;
235     }
236 
237     p++;
238 
239     if (nxt_slow_path(p == end)) {
240         return NXT_AGAIN;
241     }
242 
243     /* target */
244 
245     ch = *p;
246 
247     if (nxt_slow_path(ch != '/')) {
248         rc = nxt_http_parse_unusual_target(rp, &p, end);
249 
250         if (nxt_slow_path(rc != NXT_OK)) {
251             return rc;
252         }
253     }
254 
255     rp->target_start = p;
256 
257     after_slash = p + 1;
258 
259     for ( ;; ) {
260         p++;
261 
262         trap = nxt_http_parse_target(&p, end);
263 
264         switch (trap) {
265         case NXT_HTTP_TARGET_SLASH:
266             if (nxt_slow_path(after_slash == p)) {
267                 rp->complex_target = 1;
268                 goto rest_of_target;
269             }
270 
271             after_slash = p + 1;
272 
273             rp->exten_start = NULL;
274             continue;
275 
276         case NXT_HTTP_TARGET_DOT:
277             if (nxt_slow_path(after_slash == p)) {
278                 rp->complex_target = 1;
279                 goto rest_of_target;
280             }
281 
282             rp->exten_start = p + 1;
283             continue;
284 
285         case NXT_HTTP_TARGET_ARGS_MARK:
286             rp->args_start = p + 1;
287             goto rest_of_target;
288 
289         case NXT_HTTP_TARGET_SPACE:
290             rp->target_end = p;
291             goto space_after_target;
292 
293         case NXT_HTTP_TARGET_QUOTE_MARK:
294             rp->quoted_target = 1;
295             goto rest_of_target;
296 
297         case NXT_HTTP_TARGET_PLUS:
298             rp->plus_in_target = 1;
299             continue;
300 
301         case NXT_HTTP_TARGET_HASH:
302             rp->complex_target = 1;
303             goto rest_of_target;
304 
305         case NXT_HTTP_TARGET_AGAIN:
306             return NXT_AGAIN;
307 
308         case NXT_HTTP_TARGET_BAD:
309             return NXT_HTTP_PARSE_INVALID;
310         }
311 
312         nxt_unreachable();
313     }
314 
315 rest_of_target:
316 
317     for ( ;; ) {
318         p++;
319 
320         trap = nxt_http_parse_target(&p, end);
321 
322         switch (trap) {
323         case NXT_HTTP_TARGET_SPACE:
324             rp->target_end = p;
325             goto space_after_target;
326 
327         case NXT_HTTP_TARGET_HASH:
328             rp->complex_target = 1;
329             continue;
330 
331         case NXT_HTTP_TARGET_AGAIN:
332             return NXT_AGAIN;
333 
334         case NXT_HTTP_TARGET_BAD:
335             return NXT_HTTP_PARSE_INVALID;
336 
337         default:
338             continue;
339         }
340 
341         nxt_unreachable();
342     }
343 
344 space_after_target:
345 
346     if (nxt_slow_path(end - p < 10)) {
347 
348         do {
349             p++;
350 
351             if (p == end) {
352                 return NXT_AGAIN;
353             }
354 
355         } while (*p == ' ');
356 
357         if (nxt_memcmp(p, "HTTP/", nxt_min(end - p, 5)) == 0) {
358 
359             switch (end - p) {
360             case 8:
361                 if (p[7] < '0' || p[7] > '9') {
362                     break;
363                 }
364                 /* Fall through. */
365             case 7:
366                 if (p[6] != '.') {
367                     break;
368                 }
369                 /* Fall through. */
370             case 6:
371                 if (p[5] < '0' || p[5] > '9') {
372                     break;
373                 }
374                 /* Fall through. */
375             default:
376                 return NXT_AGAIN;
377             }
378         }
379 
380         rp->space_in_target = 1;
381         goto rest_of_target;
382     }
383 
384     /* " HTTP/1.1\r\n" or " HTTP/1.1\n" */
385 
386     if (nxt_slow_path(p[9] != '\r' && p[9] != '\n')) {
387 
388         if (p[1] == ' ') {
389             /* surplus space after tartet */
390             p++;
391             goto space_after_target;
392         }
393 
394         rp->space_in_target = 1;
395         goto rest_of_target;
396     }
397 
398     nxt_memcpy(ver.str, &p[1], 8);
399 
400     if (nxt_fast_path(ver.ui64 == http11.ui64
401                       || ver.ui64 == http10.ui64
402                       || (nxt_memcmp(ver.str, "HTTP/1.", 7) == 0
403                           && ver.s.minor >= '0' && ver.s.minor <= '9')))
404     {
405         rp->version.ui64 = ver.ui64;
406 
407         if (nxt_fast_path(p[9] == '\r')) {
408             p += 10;
409 
410             if (nxt_slow_path(p == end)) {
411                 return NXT_AGAIN;
412             }
413 
414             if (nxt_slow_path(*p != '\n')) {
415                 return NXT_HTTP_PARSE_INVALID;
416             }
417 
418             *pos = p + 1;
419 
420         } else {
421             *pos = p + 10;
422         }
423 
424         if (rp->complex_target != 0 || rp->quoted_target != 0) {
425             rc = nxt_http_parse_complex_target(rp);
426 
427             if (nxt_slow_path(rc != NXT_OK)) {
428                 return rc;
429             }
430 
431             return nxt_http_parse_field_name(rp, pos, end);
432         }
433 
434         rp->path.start = rp->target_start;
435 
436         if (rp->args_start != NULL) {
437             rp->path.length = rp->args_start - rp->target_start - 1;
438 
439             rp->args.start = rp->args_start;
440             rp->args.length = rp->target_end - rp->args_start;
441 
442         } else {
443             rp->path.length = rp->target_end - rp->target_start;
444         }
445 
446         if (rp->exten_start) {
447             rp->exten.length = rp->path.start + rp->path.length -
448                                rp->exten_start;
449             rp->exten.start = rp->exten_start;
450         }
451 
452         return nxt_http_parse_field_name(rp, pos, end);
453     }
454 
455     if (nxt_memcmp(ver.s.prefix, "HTTP/", 5) == 0
456         && ver.s.major >= '0' && ver.s.major <= '9'
457         && ver.s.point == '.'
458         && ver.s.minor >= '0' && ver.s.minor <= '9')
459     {
460         return NXT_HTTP_PARSE_UNSUPPORTED_VERSION;
461     }
462 
463     return NXT_HTTP_PARSE_INVALID;
464 }
465 
466 
467 static nxt_int_t
468 nxt_http_parse_unusual_target(nxt_http_request_parse_t *rp, u_char **pos,
469     u_char *end)
470 {
471     u_char  *p, ch;
472 
473     p = *pos;
474 
475     ch = *p;
476 
477     if (ch == ' ') {
478         /* skip surplus spaces before target */
479 
480         do {
481             p++;
482 
483             if (nxt_slow_path(p == end)) {
484                 return NXT_AGAIN;
485             }
486 
487             ch = *p;
488 
489         } while (ch == ' ');
490 
491         if (ch == '/') {
492             *pos = p;
493             return NXT_OK;
494         }
495     }
496 
497     /* absolute path or '*' */
498 
499     /* TODO */
500 
501     return NXT_HTTP_PARSE_INVALID;
502 }
503 
504 
505 static nxt_int_t
506 nxt_http_parse_field_name(nxt_http_request_parse_t *rp, u_char **pos,
507     u_char *end)
508 {
509     u_char    *p, c;
510     size_t    len;
511     uint32_t  hash;
512 
513     static const u_char  normal[256]  nxt_aligned(64) =
514         "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"
515         "\0\0\0\0\0\0\0\0\0\0\0\0\0-\0\0" "0123456789\0\0\0\0\0\0"
516 
517         /* These 64 bytes should reside in one cache line. */
518         "\0abcdefghijklmnopqrstuvwxyz\0\0\0\0_"
519         "\0abcdefghijklmnopqrstuvwxyz\0\0\0\0\0"
520 
521         "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"
522         "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"
523         "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"
524         "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0";
525 
526     p = *pos + rp->field_name.length;
527     hash = rp->field_hash;
528 
529     while (nxt_fast_path(end - p >= 8)) {
530 
531 #define nxt_field_name_test_char(ch)                                          \
532                                                                               \
533         c = normal[ch];                                                       \
534                                                                               \
535         if (nxt_slow_path(c == '\0')) {                                       \
536             p = &(ch);                                                        \
537             goto name_end;                                                    \
538         }                                                                     \
539                                                                               \
540         hash = nxt_http_field_hash_char(hash, c);
541 
542 /* enddef */
543 
544         nxt_field_name_test_char(p[0]);
545         nxt_field_name_test_char(p[1]);
546         nxt_field_name_test_char(p[2]);
547         nxt_field_name_test_char(p[3]);
548 
549         nxt_field_name_test_char(p[4]);
550         nxt_field_name_test_char(p[5]);
551         nxt_field_name_test_char(p[6]);
552         nxt_field_name_test_char(p[7]);
553 
554         p += 8;
555     }
556 
557     while (nxt_fast_path(p != end)) {
558         nxt_field_name_test_char(*p); p++;
559     }
560 
561     len = p - *pos;
562 
563     if (nxt_slow_path(len > NXT_HTTP_MAX_FIELD_NAME)) {
564         return NXT_HTTP_PARSE_TOO_LARGE_FIELD;
565     }
566 
567     rp->field_hash = hash;
568     rp->field_name.length = len;
569 
570     rp->handler = &nxt_http_parse_field_name;
571 
572     return NXT_AGAIN;
573 
574 name_end:
575 
576     if (nxt_fast_path(*p == ':')) {
577         if (nxt_slow_path(p == *pos)) {
578             return NXT_HTTP_PARSE_INVALID;
579         }
580 
581         len = p - *pos;
582 
583         if (nxt_slow_path(len > NXT_HTTP_MAX_FIELD_NAME)) {
584             return NXT_HTTP_PARSE_TOO_LARGE_FIELD;
585         }
586 
587         rp->field_hash = hash;
588 
589         rp->field_name.length = len;
590         rp->field_name.start = *pos;
591 
592         *pos = p + 1;
593 
594         return nxt_http_parse_field_value(rp, pos, end);
595     }
596 
597     if (nxt_slow_path(p != *pos)) {
598         return NXT_HTTP_PARSE_INVALID;
599     }
600 
601     return nxt_http_parse_field_end(rp, pos, end);
602 }
603 
604 
605 static nxt_int_t
606 nxt_http_parse_field_value(nxt_http_request_parse_t *rp, u_char **pos,
607     u_char *end)
608 {
609     u_char  *p, *start, ch;
610     size_t  len;
611 
612     p = *pos;
613 
614     for ( ;; ) {
615         if (nxt_slow_path(p == end)) {
616             *pos = p;
617             rp->handler = &nxt_http_parse_field_value;
618             return NXT_AGAIN;
619         }
620 
621         if (*p != ' ') {
622             break;
623         }
624 
625         p++;
626     }
627 
628     start = p;
629 
630     p += rp->field_value.length;
631 
632     for ( ;; ) {
633         p = nxt_http_lookup_field_end(p, end);
634 
635         if (nxt_slow_path(p == end)) {
636             *pos = start;
637 
638             len = p - start;
639 
640             if (nxt_slow_path(len > NXT_HTTP_MAX_FIELD_VALUE)) {
641                 return NXT_HTTP_PARSE_TOO_LARGE_FIELD;
642             }
643 
644             rp->field_value.length = len;
645             rp->handler = &nxt_http_parse_field_value;
646             return NXT_AGAIN;
647         }
648 
649         ch = *p;
650 
651         if (nxt_fast_path(ch == '\r' || ch == '\n')) {
652             break;
653         }
654 
655         if (ch != '\t') {
656             return NXT_HTTP_PARSE_INVALID;
657         }
658 
659         p++;
660     }
661 
662     *pos = p;
663 
664     if (nxt_fast_path(p != start)) {
665         while (p[-1] == ' ') {
666             p--;
667         }
668     }
669 
670     len = p - start;
671 
672     if (nxt_slow_path(len > NXT_HTTP_MAX_FIELD_VALUE)) {
673         return NXT_HTTP_PARSE_TOO_LARGE_FIELD;
674     }
675 
676     rp->field_value.length = len;
677     rp->field_value.start = start;
678 
679     return nxt_http_parse_field_end(rp, pos, end);
680 }
681 
682 
683 static u_char *
684 nxt_http_lookup_field_end(u_char *p, u_char *end)
685 {
686     while (nxt_fast_path(end - p >= 16)) {
687 
688 #define nxt_field_end_test_char(ch)                                           \
689                                                                               \
690         /* Values below 0x20 become more than 0xdf. */                        \
691         if (nxt_slow_path((u_char) ((ch) - 0x20) > 0x5e)) {                   \
692             return &(ch);                                                     \
693         }
694 
695 /* enddef */
696 
697         nxt_field_end_test_char(p[0]);
698         nxt_field_end_test_char(p[1]);
699         nxt_field_end_test_char(p[2]);
700         nxt_field_end_test_char(p[3]);
701 
702         nxt_field_end_test_char(p[4]);
703         nxt_field_end_test_char(p[5]);
704         nxt_field_end_test_char(p[6]);
705         nxt_field_end_test_char(p[7]);
706 
707         nxt_field_end_test_char(p[8]);
708         nxt_field_end_test_char(p[9]);
709         nxt_field_end_test_char(p[10]);
710         nxt_field_end_test_char(p[11]);
711 
712         nxt_field_end_test_char(p[12]);
713         nxt_field_end_test_char(p[13]);
714         nxt_field_end_test_char(p[14]);
715         nxt_field_end_test_char(p[15]);
716 
717         p += 16;
718     }
719 
720     while (nxt_fast_path(end - p >= 4)) {
721 
722         nxt_field_end_test_char(p[0]);
723         nxt_field_end_test_char(p[1]);
724         nxt_field_end_test_char(p[2]);
725         nxt_field_end_test_char(p[3]);
726 
727         p += 4;
728     }
729 
730     switch (end - p) {
731     case 3:
732         nxt_field_end_test_char(*p); p++;
733         /* Fall through. */
734     case 2:
735         nxt_field_end_test_char(*p); p++;
736         /* Fall through. */
737     case 1:
738         nxt_field_end_test_char(*p); p++;
739         /* Fall through. */
740     case 0:
741         break;
742     default:
743         nxt_unreachable();
744     }
745 
746     return p;
747 }
748 
749 
750 static nxt_int_t
751 nxt_http_parse_field_end(nxt_http_request_parse_t *rp, u_char **pos,
752     u_char *end)
753 {
754     u_char            *p;
755     nxt_http_field_t  *field;
756 
757     p = *pos;
758 
759     if (nxt_fast_path(*p == '\r')) {
760         p++;
761 
762         if (nxt_slow_path(p == end)) {
763             rp->handler = &nxt_http_parse_field_end;
764             return NXT_AGAIN;
765         }
766     }
767 
768     if (nxt_fast_path(*p == '\n')) {
769         *pos = p + 1;
770 
771         if (rp->field_name.length != 0) {
772             field = nxt_list_add(rp->fields);
773 
774             if (nxt_slow_path(field == NULL)) {
775                 return NXT_ERROR;
776             }
777 
778             field->hash = nxt_http_field_hash_end(rp->field_hash);
779             field->skip = 0;
780 
781             field->name_length = rp->field_name.length;
782             field->value_length = rp->field_value.length;
783             field->name = rp->field_name.start;
784             field->value = rp->field_value.start;
785 
786             rp->field_hash = NXT_HTTP_FIELD_HASH_INIT;
787 
788             rp->field_name.length = 0;
789             rp->field_value.length = 0;
790 
791             rp->handler = &nxt_http_parse_field_name;
792             return NXT_OK;
793         }
794 
795         return NXT_DONE;
796     }
797 
798     return NXT_HTTP_PARSE_INVALID;
799 }
800 
801 
802 #define                                                                       \
803 nxt_http_is_normal(c)                                                         \
804     (nxt_fast_path((nxt_http_normal[c / 8] & (1 << (c & 7))) != 0))
805 
806 
807 static const uint8_t  nxt_http_normal[32]  nxt_aligned(32) = {
808 
809                              /*        \0   \r  \n                         */
810     0xfe, 0xdb, 0xff, 0xff,  /* 1111 1110  1101 1011  1111 1111  1111 1111 */
811 
812                              /* '&%$ #"!   /.-, |*)(  7654 3210  ?>=< ;:98 */
813     0xd6, 0x37, 0xff, 0x7f,  /* 1101 0110  0011 0111  1111 1111  0111 1111 */
814 
815                              /* GFED CBA@  ONML KJIH  WVUT SRQP  _^]\ [ZYX */
816     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
817 
818                              /* gfed cba`  onml kjih  wvut srqp   ~}| {zyx */
819     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
820 
821     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
822     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
823     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
824     0xff, 0xff, 0xff, 0xff,  /* 1111 1111  1111 1111  1111 1111  1111 1111 */
825 };
826 
827 
828 static nxt_int_t
829 nxt_http_parse_complex_target(nxt_http_request_parse_t *rp)
830 {
831     u_char  *p, *u, c, ch, high;
832     enum {
833         sw_normal = 0,
834         sw_slash,
835         sw_dot,
836         sw_dot_dot,
837         sw_quoted,
838         sw_quoted_second,
839     } state, saved_state;
840 
841     nxt_prefetch(nxt_http_normal);
842 
843     state = sw_normal;
844     saved_state = sw_normal;
845     p = rp->target_start;
846 
847     u = nxt_mp_alloc(rp->mem_pool, rp->target_end - p + 1);
848 
849     if (nxt_slow_path(u == NULL)) {
850         return NXT_ERROR;
851     }
852 
853     rp->path.length = 0;
854     rp->path.start = u;
855 
856     high = '\0';
857     rp->exten_start = NULL;
858     rp->args_start = NULL;
859 
860     while (p < rp->target_end) {
861 
862         ch = *p++;
863 
864     again:
865 
866         switch (state) {
867 
868         case sw_normal:
869 
870             if (nxt_http_is_normal(ch)) {
871                 *u++ = ch;
872                 continue;
873             }
874 
875             switch (ch) {
876             case '/':
877                 rp->exten_start = NULL;
878                 state = sw_slash;
879                 *u++ = ch;
880                 continue;
881             case '%':
882                 saved_state = state;
883                 state = sw_quoted;
884                 continue;
885             case '?':
886                 rp->args_start = p;
887                 goto args;
888             case '#':
889                 goto done;
890             case '.':
891                 rp->exten_start = u + 1;
892                 *u++ = ch;
893                 continue;
894             case '+':
895                 rp->plus_in_target = 1;
896                 /* Fall through. */
897             default:
898                 *u++ = ch;
899                 continue;
900             }
901 
902             break;
903 
904         case sw_slash:
905 
906             if (nxt_http_is_normal(ch)) {
907                 state = sw_normal;
908                 *u++ = ch;
909                 continue;
910             }
911 
912             switch (ch) {
913             case '/':
914                 continue;
915             case '.':
916                 state = sw_dot;
917                 *u++ = ch;
918                 continue;
919             case '%':
920                 saved_state = state;
921                 state = sw_quoted;
922                 continue;
923             case '?':
924                 rp->args_start = p;
925                 goto args;
926             case '#':
927                 goto done;
928             case '+':
929                 rp->plus_in_target = 1;
930                 /* Fall through. */
931             default:
932                 state = sw_normal;
933                 *u++ = ch;
934                 continue;
935             }
936 
937             break;
938 
939         case sw_dot:
940 
941             if (nxt_http_is_normal(ch)) {
942                 state = sw_normal;
943                 *u++ = ch;
944                 continue;
945             }
946 
947             switch (ch) {
948             case '/':
949                 state = sw_slash;
950                 u--;
951                 continue;
952             case '.':
953                 state = sw_dot_dot;
954                 *u++ = ch;
955                 continue;
956             case '%':
957                 saved_state = state;
958                 state = sw_quoted;
959                 continue;
960             case '?':
961                 rp->args_start = p;
962                 goto args;
963             case '#':
964                 goto done;
965             case '+':
966                 rp->plus_in_target = 1;
967                 /* Fall through. */
968             default:
969                 state = sw_normal;
970                 *u++ = ch;
971                 continue;
972             }
973 
974             break;
975 
976         case sw_dot_dot:
977 
978             if (nxt_http_is_normal(ch)) {
979                 state = sw_normal;
980                 *u++ = ch;
981                 continue;
982             }
983 
984             switch (ch) {
985             case '/':
986                 state = sw_slash;
987                 u -= 5;
988                 for ( ;; ) {
989                     if (u < rp->path.start) {
990                         return NXT_HTTP_PARSE_INVALID;
991                     }
992                     if (*u == '/') {
993                         u++;
994                         break;
995                     }
996                     u--;
997                 }
998                 break;
999 
1000             case '%':
1001                 saved_state = state;
1002                 state = sw_quoted;
1003                 continue;
1004             case '?':
1005                 rp->args_start = p;
1006                 goto args;
1007             case '#':
1008                 goto done;
1009             case '+':
1010                 rp->plus_in_target = 1;
1011                 /* Fall through. */
1012             default:
1013                 state = sw_normal;
1014                 *u++ = ch;
1015                 continue;
1016             }
1017 
1018             break;
1019 
1020         case sw_quoted:
1021             rp->quoted_target = 1;
1022 
1023             if (ch >= '0' && ch <= '9') {
1024                 high = (u_char) (ch - '0');
1025                 state = sw_quoted_second;
1026                 continue;
1027             }
1028 
1029             c = (u_char) (ch | 0x20);
1030             if (c >= 'a' && c <= 'f') {
1031                 high = (u_char) (c - 'a' + 10);
1032                 state = sw_quoted_second;
1033                 continue;
1034             }
1035 
1036             return NXT_HTTP_PARSE_INVALID;
1037 
1038         case sw_quoted_second:
1039             if (ch >= '0' && ch <= '9') {
1040                 ch = (u_char) ((high << 4) + ch - '0');
1041 
1042                 if (ch == '%' || ch == '#') {
1043                     state = sw_normal;
1044                     *u++ = ch;
1045                     continue;
1046 
1047                 } else if (ch == '\0') {
1048                     return NXT_HTTP_PARSE_INVALID;
1049                 }
1050 
1051                 state = saved_state;
1052                 goto again;
1053             }
1054 
1055             c = (u_char) (ch | 0x20);
1056             if (c >= 'a' && c <= 'f') {
1057                 ch = (u_char) ((high << 4) + c - 'a' + 10);
1058 
1059                 if (ch == '?') {
1060                     state = sw_normal;
1061                     *u++ = ch;
1062                     continue;
1063 
1064                 } else if (ch == '+') {
1065                     rp->plus_in_target = 1;
1066                 }
1067 
1068                 state = saved_state;
1069                 goto again;
1070             }
1071 
1072             return NXT_HTTP_PARSE_INVALID;
1073         }
1074     }
1075 
1076     if (state >= sw_quoted) {
1077         return NXT_HTTP_PARSE_INVALID;
1078     }
1079 
1080 args:
1081 
1082     for (/* void */; p < rp->target_end; p++) {
1083         if (*p == '#') {
1084             break;
1085         }
1086     }
1087 
1088     if (rp->args_start != NULL) {
1089         rp->args.length = p - rp->args_start;
1090         rp->args.start = rp->args_start;
1091     }
1092 
1093 done:
1094 
1095     rp->path.length = u - rp->path.start;
1096 
1097     if (rp->exten_start) {
1098         rp->exten.length = u - rp->exten_start;
1099         rp->exten.start = rp->exten_start;
1100     }
1101 
1102     return NXT_OK;
1103 }
1104 
1105 
1106 static const nxt_lvlhsh_proto_t  nxt_http_fields_hash_proto  nxt_aligned(64) = {
1107     NXT_LVLHSH_BUCKET_SIZE(64),
1108     { NXT_HTTP_FIELD_LVLHSH_SHIFT, 0, 0, 0, 0, 0, 0, 0 },
1109     nxt_http_field_hash_test,
1110     nxt_http_field_hash_alloc,
1111     nxt_http_field_hash_free,
1112 };
1113 
1114 
1115 static nxt_int_t
1116 nxt_http_field_hash_test(nxt_lvlhsh_query_t *lhq, void *data)
1117 {
1118     nxt_http_field_proc_t  *field;
1119 
1120     field = data;
1121 
1122     if (nxt_strcasestr_eq(&lhq->key, &field->name)) {
1123         return NXT_OK;
1124     }
1125 
1126     return NXT_DECLINED;
1127 }
1128 
1129 
1130 static void *
1131 nxt_http_field_hash_alloc(void *pool, size_t size)
1132 {
1133     return nxt_mp_align(pool, size, size);
1134 }
1135 
1136 
1137 static void
1138 nxt_http_field_hash_free(void *pool, void *p)
1139 {
1140     nxt_mp_free(pool, p);
1141 }
1142 
1143 
1144 static nxt_int_t
1145 nxt_http_field_hash_collision(nxt_lvlhsh_query_t *lhq, void *data)
1146 {
1147     return NXT_OK;
1148 }
1149 
1150 
1151 nxt_int_t
1152 nxt_http_fields_hash(nxt_lvlhsh_t *hash, nxt_mp_t *mp,
1153     nxt_http_field_proc_t items[], nxt_uint_t count)
1154 {
1155     u_char              ch;
1156     uint32_t            key;
1157     nxt_str_t           *name;
1158     nxt_int_t           ret;
1159     nxt_uint_t          i, j;
1160     nxt_lvlhsh_query_t  lhq;
1161 
1162     lhq.replace = 0;
1163     lhq.proto = &nxt_http_fields_hash_proto;
1164     lhq.pool = mp;
1165 
1166     for (i = 0; i < count; i++) {
1167         key = NXT_HTTP_FIELD_HASH_INIT;
1168         name = &items[i].name;
1169 
1170         for (j = 0; j < name->length; j++) {
1171             ch = nxt_lowcase(name->start[j]);
1172             key = nxt_http_field_hash_char(key, ch);
1173         }
1174 
1175         lhq.key_hash = nxt_http_field_hash_end(key) & 0xffff;
1176         lhq.key = *name;
1177         lhq.value = &items[i];
1178 
1179         ret = nxt_lvlhsh_insert(hash, &lhq);
1180 
1181         if (nxt_slow_path(ret != NXT_OK)) {
1182             return NXT_ERROR;
1183         }
1184     }
1185 
1186     return NXT_OK;
1187 }
1188 
1189 
1190 nxt_uint_t
1191 nxt_http_fields_hash_collisions(nxt_lvlhsh_t *hash, nxt_mp_t *mp,
1192     nxt_http_field_proc_t items[], nxt_uint_t count, nxt_bool_t level)
1193 {
1194     u_char              ch;
1195     uint32_t            key, mask;
1196     nxt_str_t           *name;
1197     nxt_uint_t          colls, i, j;
1198     nxt_lvlhsh_proto_t  proto;
1199     nxt_lvlhsh_query_t  lhq;
1200 
1201     proto = nxt_http_fields_hash_proto;
1202     proto.test = nxt_http_field_hash_collision;
1203 
1204     lhq.replace = 0;
1205     lhq.proto = &proto;
1206     lhq.pool = mp;
1207 
1208     mask = level ? (1 << NXT_HTTP_FIELD_LVLHSH_SHIFT) - 1 : 0xffff;
1209 
1210     colls = 0;
1211 
1212     for (i = 0; i < count; i++) {
1213         key = NXT_HTTP_FIELD_HASH_INIT;
1214         name = &items[i].name;
1215 
1216         for (j = 0; j < name->length; j++) {
1217             ch = nxt_lowcase(name->start[j]);
1218             key = nxt_http_field_hash_char(key, ch);
1219         }
1220 
1221         lhq.key_hash = nxt_http_field_hash_end(key) & mask;
1222         lhq.value = &items[i];
1223 
1224         if (nxt_lvlhsh_insert(hash, &lhq) == NXT_DECLINED) {
1225             colls++;
1226         }
1227     }
1228 
1229     return colls;
1230 }
1231 
1232 
1233 nxt_int_t
1234 nxt_http_fields_process(nxt_list_t *fields, nxt_lvlhsh_t *hash, void *ctx)
1235 {
1236     nxt_int_t              ret;
1237     nxt_http_field_t       *field;
1238     nxt_lvlhsh_query_t     lhq;
1239     nxt_http_field_proc_t  *proc;
1240 
1241     lhq.proto = &nxt_http_fields_hash_proto;
1242 
1243     nxt_list_each(field, fields) {
1244 
1245         lhq.key_hash = field->hash;
1246         lhq.key.length = field->name_length;
1247         lhq.key.start = field->name;
1248 
1249         if (nxt_lvlhsh_find(hash, &lhq) != NXT_OK) {
1250             continue;
1251         }
1252 
1253         proc = lhq.value;
1254 
1255         ret = proc->handler(ctx, field, proc->data);
1256 
1257         if (nxt_slow_path(ret != NXT_OK)) {
1258             return ret;
1259         }
1260 
1261     } nxt_list_loop;
1262 
1263     return NXT_OK;
1264 }
1265