1 /*
2 * HTT/1 mux-demux for connections
3 *
4 * Copyright 2018 Christopher Faulet <cfaulet@haproxy.com>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12 #include <import/ebistree.h>
13
14 #include <haproxy/api.h>
15 #include <haproxy/cfgparse.h>
16 #include <haproxy/connection.h>
17 #include <haproxy/h1.h>
18 #include <haproxy/h1_htx.h>
19 #include <haproxy/h2.h>
20 #include <haproxy/http_htx.h>
21 #include <haproxy/htx.h>
22 #include <haproxy/istbuf.h>
23 #include <haproxy/log.h>
24 #include <haproxy/pipe-t.h>
25 #include <haproxy/proxy-t.h>
26 #include <haproxy/session-t.h>
27 #include <haproxy/stream.h>
28 #include <haproxy/stream_interface.h>
29 #include <haproxy/trace.h>
30
31 /*
32 * H1 Connection flags (32 bits)
33 */
34 #define H1C_F_NONE 0x00000000
35
36 /* Flags indicating why writing output data are blocked */
37 #define H1C_F_OUT_ALLOC 0x00000001 /* mux is blocked on lack of output buffer */
38 #define H1C_F_OUT_FULL 0x00000002 /* mux is blocked on output buffer full */
39 /* 0x00000004 - 0x00000008 unused */
40
41 /* Flags indicating why reading input data are blocked. */
42 #define H1C_F_IN_ALLOC 0x00000010 /* mux is blocked on lack of input buffer */
43 #define H1C_F_IN_FULL 0x00000020 /* mux is blocked on input buffer full */
44 #define H1C_F_IN_BUSY 0x00000040 /* mux is blocked on input waiting the other side */
45 /* 0x00000080 unused */
46
47 #define H1C_F_ST_SILENT_SHUT 0x00000100 /* silent (or dirty) shutdown must be performed */
48 /* 0x000000200 - 0x000000800 unused */
49
50 /* Flags indicating the connection state */
51 #define H1C_F_CS_ERROR 0x00001000 /* connection must be closed ASAP because an error occurred */
52 #define H1C_F_CS_SHUTW_NOW 0x00002000 /* connection must be shut down for writes ASAP */
53 #define H1C_F_CS_SHUTDOWN 0x00004000 /* connection is shut down */
54 #define H1C_F_CS_IDLE 0x00008000 /* connection is idle and may be reused
55 * (exclusive to all H1C_F_CS flags and never set when an h1s is attached) */
56
57 #define H1C_F_WAIT_NEXT_REQ 0x00010000 /* waiting for the next request to start, use keep-alive timeout */
58 #define H1C_F_UPG_H2C 0x00020000 /* set if an upgrade to h2 should be done */
59
60 #define H1C_F_CO_MSG_MORE 0x00040000 /* set if CO_SFL_MSG_MORE must be set when calling xprt->snd_buf() */
61 #define H1C_F_CO_STREAMER 0x00080000 /* set if CO_SFL_STREAMER must be set when calling xprt->snd_buf() */
62
63 /*
64 * H1 Stream flags (32 bits)
65 */
66 #define H1S_F_NONE 0x00000000
67 #define H1S_F_ERROR 0x00000001 /* An error occurred on the H1 stream */
68 #define H1S_F_REQ_ERROR 0x00000002 /* An error occurred during the request parsing/xfer */
69 #define H1S_F_RES_ERROR 0x00000004 /* An error occurred during the response parsing/xfer */
70
71 #define H1S_F_REOS 0x00000008 /* End of input stream seen even if not delivered yet */
72 #define H1S_F_WANT_KAL 0x00000010
73 #define H1S_F_WANT_TUN 0x00000020
74 #define H1S_F_WANT_CLO 0x00000040
75 #define H1S_F_WANT_MSK 0x00000070
76 #define H1S_F_NOT_FIRST 0x00000080 /* The H1 stream is not the first one */
77 #define H1S_F_BUF_FLUSH 0x00000100 /* Flush input buffer and don't read more data */
78 #define H1S_F_SPLICED_DATA 0x00000200 /* Set when the kernel splicing is in used */
79 #define H1S_F_PARSING_DONE 0x00000400 /* Set when incoming message parsing is finished (EOM added) */
80 /* 0x00000800 unused */
81 #define H1S_F_RX_CONGESTED 0x00001000 /* Cannot process input data RX path is congested (waiting for more space in channel's buffer) */
82 #define H1S_F_HAVE_SRV_NAME 0x00002000 /* Set during output process if the server name header was added to the request */
83 #define H1S_F_HAVE_O_CONN 0x00004000 /* Set during output process to know connection mode was processed */
84
85 /* H1 connection descriptor */
86 struct h1c {
87 struct connection *conn;
88 struct proxy *px;
89 uint32_t flags; /* Connection flags: H1C_F_* */
90
91 struct buffer ibuf; /* Input buffer to store data before parsing */
92 struct buffer obuf; /* Output buffer to store data after reformatting */
93
94 struct buffer_wait buf_wait; /* Wait list for buffer allocation */
95 struct wait_event wait_event; /* To be used if we're waiting for I/Os */
96
97 struct h1s *h1s; /* H1 stream descriptor */
98 struct task *task; /* timeout management task */
99 int timeout; /* idle timeout duration in ticks */
100 int shut_timeout; /* idle timeout duration in ticks after stream shutdown */
101 };
102
103 /* H1 stream descriptor */
104 struct h1s {
105 struct h1c *h1c;
106 struct conn_stream *cs;
107 struct cs_info csinfo; /* CS info, only used for client connections */
108 uint32_t flags; /* Connection flags: H1S_F_* */
109
110 struct wait_event *subs; /* Address of the wait_event the conn_stream associated is waiting on */
111
112 struct session *sess; /* Associated session */
113 struct h1m req;
114 struct h1m res;
115
116 enum http_meth_t meth; /* HTTP request method */
117 uint16_t status; /* HTTP response status */
118 };
119
120 /* Map of headers used to convert outgoing headers */
121 struct h1_hdrs_map {
122 char *name;
123 struct eb_root map;
124 };
125
126 /* An entry in a headers map */
127 struct h1_hdr_entry {
128 struct ist name;
129 struct ebpt_node node;
130 };
131
132 /* Declare the headers map */
133 static struct h1_hdrs_map hdrs_map = { .name = NULL, .map = EB_ROOT };
134
135
136 /* trace source and events */
137 static void h1_trace(enum trace_level level, uint64_t mask,
138 const struct trace_source *src,
139 const struct ist where, const struct ist func,
140 const void *a1, const void *a2, const void *a3, const void *a4);
141
142 /* The event representation is split like this :
143 * h1c - internal H1 connection
144 * h1s - internal H1 stream
145 * strm - application layer
146 * rx - data receipt
147 * tx - data transmission
148 *
149 */
150 static const struct trace_event h1_trace_events[] = {
151 #define H1_EV_H1C_NEW (1ULL << 0)
152 { .mask = H1_EV_H1C_NEW, .name = "h1c_new", .desc = "new H1 connection" },
153 #define H1_EV_H1C_RECV (1ULL << 1)
154 { .mask = H1_EV_H1C_RECV, .name = "h1c_recv", .desc = "Rx on H1 connection" },
155 #define H1_EV_H1C_SEND (1ULL << 2)
156 { .mask = H1_EV_H1C_SEND, .name = "h1c_send", .desc = "Tx on H1 connection" },
157 #define H1_EV_H1C_BLK (1ULL << 3)
158 { .mask = H1_EV_H1C_BLK, .name = "h1c_blk", .desc = "H1 connection blocked" },
159 #define H1_EV_H1C_WAKE (1ULL << 4)
160 { .mask = H1_EV_H1C_WAKE, .name = "h1c_wake", .desc = "H1 connection woken up" },
161 #define H1_EV_H1C_END (1ULL << 5)
162 { .mask = H1_EV_H1C_END, .name = "h1c_end", .desc = "H1 connection terminated" },
163 #define H1_EV_H1C_ERR (1ULL << 6)
164 { .mask = H1_EV_H1C_ERR, .name = "h1c_err", .desc = "error on H1 connection" },
165
166 #define H1_EV_RX_DATA (1ULL << 7)
167 { .mask = H1_EV_RX_DATA, .name = "rx_data", .desc = "receipt of any H1 data" },
168 #define H1_EV_RX_EOI (1ULL << 8)
169 { .mask = H1_EV_RX_EOI, .name = "rx_eoi", .desc = "receipt of end of H1 input" },
170 #define H1_EV_RX_HDRS (1ULL << 9)
171 { .mask = H1_EV_RX_HDRS, .name = "rx_headers", .desc = "receipt of H1 headers" },
172 #define H1_EV_RX_BODY (1ULL << 10)
173 { .mask = H1_EV_RX_BODY, .name = "rx_body", .desc = "receipt of H1 body" },
174 #define H1_EV_RX_TLRS (1ULL << 11)
175 { .mask = H1_EV_RX_TLRS, .name = "rx_trailerus", .desc = "receipt of H1 trailers" },
176
177 #define H1_EV_TX_DATA (1ULL << 12)
178 { .mask = H1_EV_TX_DATA, .name = "tx_data", .desc = "transmission of any H1 data" },
179 #define H1_EV_TX_EOI (1ULL << 13)
180 { .mask = H1_EV_TX_EOI, .name = "tx_eoi", .desc = "transmission of end of H1 input" },
181 #define H1_EV_TX_HDRS (1ULL << 14)
182 { .mask = H1_EV_TX_HDRS, .name = "tx_headers", .desc = "transmission of all headers" },
183 #define H1_EV_TX_BODY (1ULL << 15)
184 { .mask = H1_EV_TX_BODY, .name = "tx_body", .desc = "transmission of H1 body" },
185 #define H1_EV_TX_TLRS (1ULL << 16)
186 { .mask = H1_EV_TX_TLRS, .name = "tx_trailerus", .desc = "transmission of H1 trailers" },
187
188 #define H1_EV_H1S_NEW (1ULL << 17)
189 { .mask = H1_EV_H1S_NEW, .name = "h1s_new", .desc = "new H1 stream" },
190 #define H1_EV_H1S_BLK (1ULL << 18)
191 { .mask = H1_EV_H1S_BLK, .name = "h1s_blk", .desc = "H1 stream blocked" },
192 #define H1_EV_H1S_END (1ULL << 19)
193 { .mask = H1_EV_H1S_END, .name = "h1s_end", .desc = "H1 stream terminated" },
194 #define H1_EV_H1S_ERR (1ULL << 20)
195 { .mask = H1_EV_H1S_ERR, .name = "h1s_err", .desc = "error on H1 stream" },
196
197 #define H1_EV_STRM_NEW (1ULL << 21)
198 { .mask = H1_EV_STRM_NEW, .name = "strm_new", .desc = "app-layer stream creation" },
199 #define H1_EV_STRM_RECV (1ULL << 22)
200 { .mask = H1_EV_STRM_RECV, .name = "strm_recv", .desc = "receiving data for stream" },
201 #define H1_EV_STRM_SEND (1ULL << 23)
202 { .mask = H1_EV_STRM_SEND, .name = "strm_send", .desc = "sending data for stream" },
203 #define H1_EV_STRM_WAKE (1ULL << 24)
204 { .mask = H1_EV_STRM_WAKE, .name = "strm_wake", .desc = "stream woken up" },
205 #define H1_EV_STRM_SHUT (1ULL << 25)
206 { .mask = H1_EV_STRM_SHUT, .name = "strm_shut", .desc = "stream shutdown" },
207 #define H1_EV_STRM_END (1ULL << 26)
208 { .mask = H1_EV_STRM_END, .name = "strm_end", .desc = "detaching app-layer stream" },
209 #define H1_EV_STRM_ERR (1ULL << 27)
210 { .mask = H1_EV_STRM_ERR, .name = "strm_err", .desc = "stream error" },
211
212 { }
213 };
214
215 static const struct name_desc h1_trace_lockon_args[4] = {
216 /* arg1 */ { /* already used by the connection */ },
217 /* arg2 */ { .name="h1s", .desc="H1 stream" },
218 /* arg3 */ { },
219 /* arg4 */ { }
220 };
221
222 static const struct name_desc h1_trace_decoding[] = {
223 #define H1_VERB_CLEAN 1
224 { .name="clean", .desc="only user-friendly stuff, generally suitable for level \"user\"" },
225 #define H1_VERB_MINIMAL 2
226 { .name="minimal", .desc="report only h1c/h1s state and flags, no real decoding" },
227 #define H1_VERB_SIMPLE 3
228 { .name="simple", .desc="add request/response status line or htx info when available" },
229 #define H1_VERB_ADVANCED 4
230 { .name="advanced", .desc="add header fields or frame decoding when available" },
231 #define H1_VERB_COMPLETE 5
232 { .name="complete", .desc="add full data dump when available" },
233 { /* end */ }
234 };
235
236 static struct trace_source trace_h1 = {
237 .name = IST("h1"),
238 .desc = "HTTP/1 multiplexer",
239 .arg_def = TRC_ARG1_CONN, // TRACE()'s first argument is always a connection
240 .default_cb = h1_trace,
241 .known_events = h1_trace_events,
242 .lockon_args = h1_trace_lockon_args,
243 .decoding = h1_trace_decoding,
244 .report_events = ~0, // report everything by default
245 };
246
247 #define TRACE_SOURCE &trace_h1
248 INITCALL1(STG_REGISTER, trace_register_source, TRACE_SOURCE);
249
250 /* the h1c and h1s pools */
251 DECLARE_STATIC_POOL(pool_head_h1c, "h1c", sizeof(struct h1c));
252 DECLARE_STATIC_POOL(pool_head_h1s, "h1s", sizeof(struct h1s));
253
254 static int h1_recv(struct h1c *h1c);
255 static int h1_send(struct h1c *h1c);
256 static int h1_process(struct h1c *h1c);
257 static struct task *h1_io_cb(struct task *t, void *ctx, unsigned short state);
258 static void h1_shutw_conn(struct connection *conn);
259 static struct task *h1_timeout_task(struct task *t, void *context, unsigned short state);
260 static void h1_wake_stream_for_recv(struct h1s *h1s);
261 static void h1_wake_stream_for_send(struct h1s *h1s);
262
263 /* the H1 traces always expect that arg1, if non-null, is of type connection
264 * (from which we can derive h1c), that arg2, if non-null, is of type h1s, and
265 * that arg3, if non-null, is a htx for rx/tx headers.
266 */
h1_trace(enum trace_level level,uint64_t mask,const struct trace_source * src,const struct ist where,const struct ist func,const void * a1,const void * a2,const void * a3,const void * a4)267 static void h1_trace(enum trace_level level, uint64_t mask, const struct trace_source *src,
268 const struct ist where, const struct ist func,
269 const void *a1, const void *a2, const void *a3, const void *a4)
270 {
271 const struct connection *conn = a1;
272 const struct h1c *h1c = conn ? conn->ctx : NULL;
273 const struct h1s *h1s = a2;
274 const struct htx *htx = a3;
275 const size_t *val = a4;
276
277 if (!h1c)
278 h1c = (h1s ? h1s->h1c : NULL);
279
280 if (!h1c || src->verbosity < H1_VERB_CLEAN)
281 return;
282
283 /* Display frontend/backend info by default */
284 chunk_appendf(&trace_buf, " : [%c]", (conn_is_back(h1c->conn) ? 'B' : 'F'));
285
286 /* Display request and response states if h1s is defined */
287 if (h1s)
288 chunk_appendf(&trace_buf, " [%s, %s]",
289 h1m_state_str(h1s->req.state), h1m_state_str(h1s->res.state));
290
291 if (src->verbosity == H1_VERB_CLEAN)
292 return;
293
294 /* Display the value to the 4th argument (level > STATE) */
295 if (src->level > TRACE_LEVEL_STATE && val)
296 chunk_appendf(&trace_buf, " - VAL=%lu", (long)*val);
297
298 /* Display status-line if possible (verbosity > MINIMAL) */
299 if (src->verbosity > H1_VERB_MINIMAL && htx && htx_nbblks(htx)) {
300 const struct htx_blk *blk = htx_get_head_blk(htx);
301 const struct htx_sl *sl = htx_get_blk_ptr(htx, blk);
302 enum htx_blk_type type = htx_get_blk_type(blk);
303
304 if (type == HTX_BLK_REQ_SL || type == HTX_BLK_RES_SL)
305 chunk_appendf(&trace_buf, " - \"%.*s %.*s %.*s\"",
306 HTX_SL_P1_LEN(sl), HTX_SL_P1_PTR(sl),
307 HTX_SL_P2_LEN(sl), HTX_SL_P2_PTR(sl),
308 HTX_SL_P3_LEN(sl), HTX_SL_P3_PTR(sl));
309 }
310
311 /* Display h1c info and, if defined, h1s info (pointer + flags) */
312 chunk_appendf(&trace_buf, " - h1c=%p(0x%08x)", h1c, h1c->flags);
313 if (h1s)
314 chunk_appendf(&trace_buf, " h1s=%p(0x%08x)", h1s, h1s->flags);
315
316 if (src->verbosity == H1_VERB_MINIMAL)
317 return;
318
319 /* Display input and output buffer info (level > USER & verbosity > SIMPLE) */
320 if (src->level > TRACE_LEVEL_USER) {
321 if (src->verbosity == H1_VERB_COMPLETE ||
322 (src->verbosity == H1_VERB_ADVANCED && (mask & (H1_EV_H1C_RECV|H1_EV_STRM_RECV))))
323 chunk_appendf(&trace_buf, " ibuf=%u@%p+%u/%u",
324 (unsigned int)b_data(&h1c->ibuf), b_orig(&h1c->ibuf),
325 (unsigned int)b_head_ofs(&h1c->ibuf), (unsigned int)b_size(&h1c->ibuf));
326 if (src->verbosity == H1_VERB_COMPLETE ||
327 (src->verbosity == H1_VERB_ADVANCED && (mask & (H1_EV_H1C_SEND|H1_EV_STRM_SEND))))
328 chunk_appendf(&trace_buf, " obuf=%u@%p+%u/%u",
329 (unsigned int)b_data(&h1c->obuf), b_orig(&h1c->obuf),
330 (unsigned int)b_head_ofs(&h1c->obuf), (unsigned int)b_size(&h1c->obuf));
331 }
332
333 /* Display htx info if defined (level > USER) */
334 if (src->level > TRACE_LEVEL_USER && htx) {
335 int full = 0;
336
337 /* Full htx info (level > STATE && verbosity > SIMPLE) */
338 if (src->level > TRACE_LEVEL_STATE) {
339 if (src->verbosity == H1_VERB_COMPLETE)
340 full = 1;
341 else if (src->verbosity == H1_VERB_ADVANCED && (mask & (H1_EV_RX_HDRS|H1_EV_TX_HDRS)))
342 full = 1;
343 }
344
345 chunk_memcat(&trace_buf, "\n\t", 2);
346 htx_dump(&trace_buf, htx, full);
347 }
348 }
349
350
351 /*****************************************************/
352 /* functions below are for dynamic buffer management */
353 /*****************************************************/
354 /*
355 * Indicates whether or not we may receive data. The rules are the following :
356 * - if an error or a shutdown for reads was detected on the connection we
357 must not attempt to receive
358 * - if the input buffer failed to be allocated or is full , we must not try
359 * to receive
360 * - if he input processing is busy waiting for the output side, we may
361 * attempt to receive
362 * - otherwise must may not attempt to receive
363 */
h1_recv_allowed(const struct h1c * h1c)364 static inline int h1_recv_allowed(const struct h1c *h1c)
365 {
366 if (h1c->flags & H1C_F_CS_ERROR) {
367 TRACE_DEVEL("recv not allowed because of error on h1c", H1_EV_H1C_RECV|H1_EV_H1C_BLK, h1c->conn);
368 return 0;
369 }
370
371 if (h1c->conn->flags & (CO_FL_ERROR|CO_FL_SOCK_RD_SH)) {
372 TRACE_DEVEL("recv not allowed because of (error|read0) on connection", H1_EV_H1C_RECV|H1_EV_H1C_BLK, h1c->conn);
373 return 0;
374 }
375
376 if (!(h1c->flags & (H1C_F_IN_ALLOC|H1C_F_IN_FULL)))
377 return 1;
378
379 TRACE_DEVEL("recv not allowed because input is blocked", H1_EV_H1C_RECV|H1_EV_H1C_BLK, h1c->conn);
380 return 0;
381 }
382
383 /*
384 * Tries to grab a buffer and to re-enables processing on mux <target>. The h1
385 * flags are used to figure what buffer was requested. It returns 1 if the
386 * allocation succeeds, in which case the connection is woken up, or 0 if it's
387 * impossible to wake up and we prefer to be woken up later.
388 */
h1_buf_available(void * target)389 static int h1_buf_available(void *target)
390 {
391 struct h1c *h1c = target;
392
393 if ((h1c->flags & H1C_F_IN_ALLOC) && b_alloc_margin(&h1c->ibuf, 0)) {
394 TRACE_STATE("unblocking h1c, ibuf allocated", H1_EV_H1C_RECV|H1_EV_H1C_BLK|H1_EV_H1C_WAKE, h1c->conn);
395 h1c->flags &= ~H1C_F_IN_ALLOC;
396 if (h1_recv_allowed(h1c))
397 tasklet_wakeup(h1c->wait_event.tasklet);
398 return 1;
399 }
400
401 if ((h1c->flags & H1C_F_OUT_ALLOC) && b_alloc_margin(&h1c->obuf, 0)) {
402 TRACE_STATE("unblocking h1s, obuf allocated", H1_EV_TX_DATA|H1_EV_H1S_BLK|H1_EV_STRM_WAKE, h1c->conn, h1c->h1s);
403 h1c->flags &= ~H1C_F_OUT_ALLOC;
404 tasklet_wakeup(h1c->wait_event.tasklet);
405 if (h1c->h1s)
406 h1_wake_stream_for_send(h1c->h1s);
407 return 1;
408 }
409
410 return 0;
411 }
412
413 /*
414 * Allocate a buffer. If if fails, it adds the mux in buffer wait queue.
415 */
h1_get_buf(struct h1c * h1c,struct buffer * bptr)416 static inline struct buffer *h1_get_buf(struct h1c *h1c, struct buffer *bptr)
417 {
418 struct buffer *buf = NULL;
419
420 if (likely(!MT_LIST_ADDED(&h1c->buf_wait.list)) &&
421 unlikely((buf = b_alloc_margin(bptr, 0)) == NULL)) {
422 h1c->buf_wait.target = h1c;
423 h1c->buf_wait.wakeup_cb = h1_buf_available;
424 MT_LIST_ADDQ(&buffer_wq, &h1c->buf_wait.list);
425 }
426 return buf;
427 }
428
429 /*
430 * Release a buffer, if any, and try to wake up entities waiting in the buffer
431 * wait queue.
432 */
h1_release_buf(struct h1c * h1c,struct buffer * bptr)433 static inline void h1_release_buf(struct h1c *h1c, struct buffer *bptr)
434 {
435 if (bptr->size) {
436 b_free(bptr);
437 offer_buffers(h1c->buf_wait.target, tasks_run_queue);
438 }
439 }
440
441 /* returns the number of streams in use on a connection to figure if it's idle
442 * or not. We rely on H1C_F_CS_IDLE to know if the connection is in-use or
443 * not. This flag is only set when no H1S is attached and when the previous
444 * stream, if any, was fully terminated without any error and in K/A mode.
445 */
h1_used_streams(struct connection * conn)446 static int h1_used_streams(struct connection *conn)
447 {
448 struct h1c *h1c = conn->ctx;
449
450 return ((h1c->flags & H1C_F_CS_IDLE) ? 0 : 1);
451 }
452
453 /* returns the number of streams still available on a connection */
h1_avail_streams(struct connection * conn)454 static int h1_avail_streams(struct connection *conn)
455 {
456 return 1 - h1_used_streams(conn);
457 }
458
459 /* Refresh the h1c task timeout if necessary */
h1_refresh_timeout(struct h1c * h1c)460 static void h1_refresh_timeout(struct h1c *h1c)
461 {
462 if (h1c->task) {
463 h1c->task->expire = TICK_ETERNITY;
464 if (h1c->flags & H1C_F_CS_SHUTDOWN) {
465 /* half-closed connections switch to clientfin/serverfin
466 * timeouts so that we don't hang too long on clients
467 * that have gone away (especially in tunnel mode).
468 */
469 h1c->task->expire = tick_add(now_ms, h1c->shut_timeout);
470 task_queue(h1c->task);
471 TRACE_DEVEL("refreshing connection's timeout (half-closed)", H1_EV_H1C_SEND, h1c->conn);
472 } else if (b_data(&h1c->obuf)) {
473 /* any connection with pending data, need a timeout (server or client).
474 */
475 h1c->task->expire = tick_add(now_ms, ((h1c->flags & H1C_F_CS_SHUTW_NOW)
476 ? h1c->shut_timeout
477 : h1c->timeout));
478 task_queue(h1c->task);
479 TRACE_DEVEL("refreshing connection's timeout", H1_EV_H1C_SEND, h1c->conn);
480 } else if ((h1c->flags & (H1C_F_CS_IDLE|H1C_F_WAIT_NEXT_REQ)) && !conn_is_back(h1c->conn)) {
481 /* front connections waiting for a stream need a timeout. client timeout by
482 * default but http-keep-alive if defined
483 */
484 int timeout = h1c->timeout;
485
486 if (h1c->flags & H1C_F_WAIT_NEXT_REQ)
487 timeout = tick_first(timeout, h1c->px->timeout.httpka);
488
489 h1c->task->expire = tick_add(now_ms, ((h1c->flags & H1C_F_CS_SHUTW_NOW)
490 ? h1c->shut_timeout
491 : timeout));
492 task_queue(h1c->task);
493 TRACE_DEVEL("refreshing connection's timeout", H1_EV_H1C_SEND, h1c->conn);
494 }
495 }
496 }
497 /*****************************************************************/
498 /* functions below are dedicated to the mux setup and management */
499 /*****************************************************************/
500
501 /* returns non-zero if there are input data pending for stream h1s. */
h1s_data_pending(const struct h1s * h1s)502 static inline size_t h1s_data_pending(const struct h1s *h1s)
503 {
504 const struct h1m *h1m;
505
506 h1m = conn_is_back(h1s->h1c->conn) ? &h1s->res : &h1s->req;
507 if (h1m->state == H1_MSG_DONE)
508 return !(h1s->flags & H1S_F_PARSING_DONE);
509
510 return b_data(&h1s->h1c->ibuf);
511 }
512
h1s_new_cs(struct h1s * h1s)513 static struct conn_stream *h1s_new_cs(struct h1s *h1s)
514 {
515 struct conn_stream *cs;
516
517 TRACE_ENTER(H1_EV_STRM_NEW, h1s->h1c->conn, h1s);
518 cs = cs_new(h1s->h1c->conn);
519 if (!cs) {
520 TRACE_DEVEL("leaving on CS allocation failure", H1_EV_STRM_NEW|H1_EV_STRM_END|H1_EV_STRM_ERR, h1s->h1c->conn, h1s);
521 goto err;
522 }
523 h1s->cs = cs;
524 cs->ctx = h1s;
525
526 if (h1s->flags & H1S_F_NOT_FIRST)
527 cs->flags |= CS_FL_NOT_FIRST;
528
529 if (global.tune.options & GTUNE_USE_SPLICE) {
530 TRACE_STATE("notify the mux can use splicing", H1_EV_STRM_NEW, h1s->h1c->conn, h1s);
531 cs->flags |= CS_FL_MAY_SPLICE;
532 }
533
534 if (stream_create_from_cs(cs) < 0) {
535 TRACE_DEVEL("leaving on stream creation failure", H1_EV_STRM_NEW|H1_EV_STRM_END|H1_EV_STRM_ERR, h1s->h1c->conn, h1s);
536 goto err;
537 }
538
539 TRACE_LEAVE(H1_EV_STRM_NEW, h1s->h1c->conn, h1s);
540 return cs;
541
542 err:
543 cs_free(cs);
544 h1s->cs = NULL;
545 return NULL;
546 }
547
h1s_create(struct h1c * h1c,struct conn_stream * cs,struct session * sess)548 static struct h1s *h1s_create(struct h1c *h1c, struct conn_stream *cs, struct session *sess)
549 {
550 struct h1s *h1s;
551
552 TRACE_ENTER(H1_EV_H1S_NEW, h1c->conn);
553
554 h1s = pool_alloc(pool_head_h1s);
555 if (!h1s)
556 goto fail;
557
558 h1s->h1c = h1c;
559 h1c->h1s = h1s;
560
561 h1s->sess = sess;
562
563 h1s->cs = NULL;
564 h1s->flags = H1S_F_WANT_KAL;
565
566 h1s->subs = NULL;
567
568 h1m_init_req(&h1s->req);
569 h1s->req.flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR);
570
571 h1m_init_res(&h1s->res);
572 h1s->res.flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR);
573
574 h1s->status = 0;
575 h1s->meth = HTTP_METH_OTHER;
576
577 if (h1c->flags & H1C_F_WAIT_NEXT_REQ)
578 h1s->flags |= H1S_F_NOT_FIRST;
579 h1c->flags &= ~(H1C_F_CS_IDLE|H1C_F_WAIT_NEXT_REQ);
580
581 if (!conn_is_back(h1c->conn)) {
582 if (h1c->px->options2 & PR_O2_REQBUG_OK)
583 h1s->req.err_pos = -1;
584
585 /* For frontend connections we should always have a session */
586 if (!sess)
587 h1s->sess = sess = h1c->conn->owner;
588
589 /* Timers for subsequent sessions on the same HTTP 1.x connection
590 * measure from `now`, not from the connection accept time */
591 if (h1s->flags & H1S_F_NOT_FIRST) {
592 h1s->csinfo.create_date = date;
593 h1s->csinfo.tv_create = now;
594 h1s->csinfo.t_handshake = 0;
595 h1s->csinfo.t_idle = -1;
596 }
597 else {
598 h1s->csinfo.create_date = sess->accept_date;
599 h1s->csinfo.tv_create = sess->tv_accept;
600 h1s->csinfo.t_handshake = sess->t_handshake;
601 h1s->csinfo.t_idle = -1;
602 }
603 }
604 else {
605 if (h1c->px->options2 & PR_O2_RSPBUG_OK)
606 h1s->res.err_pos = -1;
607
608 h1s->csinfo.create_date = date;
609 h1s->csinfo.tv_create = now;
610 h1s->csinfo.t_handshake = 0;
611 h1s->csinfo.t_idle = -1;
612 }
613
614 /* If a conn_stream already exists, attach it to this H1S. Otherwise we
615 * create a new one.
616 */
617 if (cs) {
618 cs->ctx = h1s;
619 h1s->cs = cs;
620 }
621 else {
622 cs = h1s_new_cs(h1s);
623 if (!cs)
624 goto fail;
625 }
626 TRACE_LEAVE(H1_EV_H1S_NEW, h1c->conn, h1s);
627 return h1s;
628
629 fail:
630 pool_free(pool_head_h1s, h1s);
631 TRACE_DEVEL("leaving in error", H1_EV_H1S_NEW|H1_EV_H1S_END|H1_EV_H1S_ERR, h1c->conn);
632 return NULL;
633 }
634
h1s_destroy(struct h1s * h1s)635 static void h1s_destroy(struct h1s *h1s)
636 {
637 if (h1s) {
638 struct h1c *h1c = h1s->h1c;
639
640 TRACE_POINT(H1_EV_H1S_END, h1c->conn, h1s);
641 h1c->h1s = NULL;
642
643 if (h1s->subs)
644 h1s->subs->events = 0;
645
646 h1c->flags &= ~H1C_F_IN_BUSY;
647 if (h1s->flags & (H1S_F_REQ_ERROR|H1S_F_RES_ERROR)) {
648 h1c->flags |= H1C_F_CS_ERROR;
649 TRACE_STATE("h1s on error, set error on h1c", H1_EV_H1C_ERR, h1c->conn, h1s);
650 }
651
652 if (!(h1c->flags & (H1C_F_CS_ERROR|H1C_F_CS_SHUTW_NOW|H1C_F_CS_SHUTDOWN)) && /* No error/shutdown on h1c */
653 !(h1c->conn->flags & (CO_FL_ERROR|CO_FL_SOCK_RD_SH|CO_FL_SOCK_WR_SH)) && /* No error/shutdown on conn */
654 (h1s->flags & (H1S_F_WANT_KAL|H1S_F_PARSING_DONE)) == (H1S_F_WANT_KAL|H1S_F_PARSING_DONE) && /* K/A possible */
655 h1s->req.state == H1_MSG_DONE && h1s->res.state == H1_MSG_DONE) { /* req/res in DONE state */
656 h1c->flags |= (H1C_F_CS_IDLE|H1C_F_WAIT_NEXT_REQ);
657 TRACE_STATE("set idle mode on h1c, waiting for the next request", H1_EV_H1C_ERR, h1c->conn, h1s);
658 }
659 pool_free(pool_head_h1s, h1s);
660 }
661 }
662
h1_get_cs_info(struct conn_stream * cs)663 static const struct cs_info *h1_get_cs_info(struct conn_stream *cs)
664 {
665 struct h1s *h1s = cs->ctx;
666
667 if (h1s && !conn_is_back(cs->conn))
668 return &h1s->csinfo;
669 return NULL;
670 }
671
672 /*
673 * Initialize the mux once it's attached. It is expected that conn->ctx points
674 * to the existing conn_stream (for outgoing connections or for incoming ones
675 * during a mux upgrade) or NULL (for incoming ones during the connection
676 * establishment). <input> is always used as Input buffer and may contain
677 * data. It is the caller responsibility to not reuse it anymore. Returns < 0 on
678 * error.
679 */
h1_init(struct connection * conn,struct proxy * proxy,struct session * sess,struct buffer * input)680 static int h1_init(struct connection *conn, struct proxy *proxy, struct session *sess,
681 struct buffer *input)
682 {
683 struct h1c *h1c;
684 struct task *t = NULL;
685 void *conn_ctx = conn->ctx;
686
687 TRACE_ENTER(H1_EV_H1C_NEW);
688
689 h1c = pool_alloc(pool_head_h1c);
690 if (!h1c)
691 goto fail_h1c;
692 h1c->conn = conn;
693 h1c->px = proxy;
694
695 h1c->flags = H1C_F_CS_IDLE;
696 h1c->ibuf = *input;
697 h1c->obuf = BUF_NULL;
698 h1c->h1s = NULL;
699 h1c->task = NULL;
700
701 MT_LIST_INIT(&h1c->buf_wait.list);
702 h1c->wait_event.tasklet = tasklet_new();
703 if (!h1c->wait_event.tasklet)
704 goto fail;
705 h1c->wait_event.tasklet->process = h1_io_cb;
706 h1c->wait_event.tasklet->context = h1c;
707 h1c->wait_event.events = 0;
708
709 if (conn_is_back(conn)) {
710 h1c->shut_timeout = h1c->timeout = proxy->timeout.server;
711 if (tick_isset(proxy->timeout.serverfin))
712 h1c->shut_timeout = proxy->timeout.serverfin;
713 } else {
714 h1c->shut_timeout = h1c->timeout = proxy->timeout.client;
715 if (tick_isset(proxy->timeout.clientfin))
716 h1c->shut_timeout = proxy->timeout.clientfin;
717 }
718 if (tick_isset(h1c->timeout)) {
719 t = task_new(tid_bit);
720 if (!t)
721 goto fail;
722
723 h1c->task = t;
724 t->process = h1_timeout_task;
725 t->context = h1c;
726 t->expire = tick_add(now_ms, h1c->timeout);
727 }
728
729 conn->ctx = h1c;
730
731 /* Always Create a new H1S */
732 if (!h1s_create(h1c, conn_ctx, sess))
733 goto fail;
734
735 if (t)
736 task_queue(t);
737
738 /* Try to read, if nothing is available yet we'll just subscribe */
739 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
740
741 /* mux->wake will be called soon to complete the operation */
742 TRACE_LEAVE(H1_EV_H1C_NEW, conn, h1c->h1s);
743 return 0;
744
745 fail:
746 task_destroy(t);
747 if (h1c->wait_event.tasklet)
748 tasklet_free(h1c->wait_event.tasklet);
749 pool_free(pool_head_h1c, h1c);
750 fail_h1c:
751 conn->ctx = conn_ctx; // restore saved context
752 TRACE_DEVEL("leaving in error", H1_EV_H1C_NEW|H1_EV_H1C_END|H1_EV_H1C_ERR);
753 return -1;
754 }
755
756 /* release function. This one should be called to free all resources allocated
757 * to the mux.
758 */
h1_release(struct h1c * h1c)759 static void h1_release(struct h1c *h1c)
760 {
761 struct connection *conn = NULL;
762
763 TRACE_POINT(H1_EV_H1C_END);
764
765 if (h1c) {
766 /* The connection must be aattached to this mux to be released */
767 if (h1c->conn && h1c->conn->ctx == h1c)
768 conn = h1c->conn;
769
770 TRACE_DEVEL("freeing h1c", H1_EV_H1C_END, conn);
771
772 if (conn && h1c->flags & H1C_F_UPG_H2C) {
773 TRACE_DEVEL("upgrading H1 to H2", H1_EV_H1C_END, conn);
774 h1c->flags &= ~H1C_F_UPG_H2C;
775 /* Make sure we're no longer subscribed to anything */
776 if (h1c->wait_event.events)
777 conn->xprt->unsubscribe(conn, conn->xprt_ctx,
778 h1c->wait_event.events, &h1c->wait_event);
779 if (conn_upgrade_mux_fe(conn, NULL, &h1c->ibuf, ist("h2"), PROTO_MODE_HTTP) != -1) {
780 /* connection successfully upgraded to H2, this
781 * mux was already released */
782 return;
783 }
784 TRACE_DEVEL("h2 upgrade failed", H1_EV_H1C_END|H1_EV_H1C_ERR, conn);
785 sess_log(conn->owner); /* Log if the upgrade failed */
786 }
787
788
789 if (MT_LIST_ADDED(&h1c->buf_wait.list))
790 MT_LIST_DEL(&h1c->buf_wait.list);
791
792 h1_release_buf(h1c, &h1c->ibuf);
793 h1_release_buf(h1c, &h1c->obuf);
794
795 if (h1c->task) {
796 h1c->task->context = NULL;
797 task_wakeup(h1c->task, TASK_WOKEN_OTHER);
798 h1c->task = NULL;
799 }
800
801 if (h1c->wait_event.tasklet)
802 tasklet_free(h1c->wait_event.tasklet);
803
804 h1s_destroy(h1c->h1s);
805 if (conn) {
806 if (h1c->wait_event.events != 0)
807 conn->xprt->unsubscribe(conn, conn->xprt_ctx, h1c->wait_event.events,
808 &h1c->wait_event);
809 h1_shutw_conn(conn);
810 }
811 pool_free(pool_head_h1c, h1c);
812 }
813
814 if (conn) {
815 conn->mux = NULL;
816 conn->ctx = NULL;
817 TRACE_DEVEL("freeing conn", H1_EV_H1C_END, conn);
818
819 conn_stop_tracking(conn);
820 conn_full_close(conn);
821 if (conn->destroy_cb)
822 conn->destroy_cb(conn);
823 conn_free(conn);
824 }
825 }
826
827 /******************************************************/
828 /* functions below are for the H1 protocol processing */
829 /******************************************************/
830 /* Parse the request version and set H1_MF_VER_11 on <h1m> if the version is
831 * greater or equal to 1.1
832 */
h1_parse_req_vsn(struct h1m * h1m,const struct htx_sl * sl)833 static void h1_parse_req_vsn(struct h1m *h1m, const struct htx_sl *sl)
834 {
835 const char *p = HTX_SL_REQ_VPTR(sl);
836
837 if ((HTX_SL_REQ_VLEN(sl) == 8) &&
838 (*(p + 5) > '1' ||
839 (*(p + 5) == '1' && *(p + 7) >= '1')))
840 h1m->flags |= H1_MF_VER_11;
841 }
842
843 /* Parse the response version and set H1_MF_VER_11 on <h1m> if the version is
844 * greater or equal to 1.1
845 */
h1_parse_res_vsn(struct h1m * h1m,const struct htx_sl * sl)846 static void h1_parse_res_vsn(struct h1m *h1m, const struct htx_sl *sl)
847 {
848 const char *p = HTX_SL_RES_VPTR(sl);
849
850 if ((HTX_SL_RES_VLEN(sl) == 8) &&
851 (*(p + 5) > '1' ||
852 (*(p + 5) == '1' && *(p + 7) >= '1')))
853 h1m->flags |= H1_MF_VER_11;
854 }
855
856 /* Deduce the connection mode of the client connection, depending on the
857 * configuration and the H1 message flags. This function is called twice, the
858 * first time when the request is parsed and the second time when the response
859 * is parsed.
860 */
h1_set_cli_conn_mode(struct h1s * h1s,struct h1m * h1m)861 static void h1_set_cli_conn_mode(struct h1s *h1s, struct h1m *h1m)
862 {
863 struct proxy *fe = h1s->h1c->px;
864
865 if (h1m->flags & H1_MF_RESP) {
866 /* Output direction: second pass */
867 if ((h1s->meth == HTTP_METH_CONNECT && h1s->status == 200) ||
868 h1s->status == 101) {
869 /* Either we've established an explicit tunnel, or we're
870 * switching the protocol. In both cases, we're very unlikely to
871 * understand the next protocols. We have to switch to tunnel
872 * mode, so that we transfer the request and responses then let
873 * this protocol pass unmodified. When we later implement
874 * specific parsers for such protocols, we'll want to check the
875 * Upgrade header which contains information about that protocol
876 * for responses with status 101 (eg: see RFC2817 about TLS).
877 */
878 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_TUN;
879 TRACE_STATE("set tunnel mode (resp)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
880 }
881 else if (h1s->flags & H1S_F_WANT_KAL) {
882 /* By default the client is in KAL mode. CLOSE mode mean
883 * it is imposed by the client itself. So only change
884 * KAL mode here. */
885 if (!(h1m->flags & H1_MF_XFER_LEN) || (h1m->flags & H1_MF_CONN_CLO)) {
886 /* no length known or explicit close => close */
887 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
888 TRACE_STATE("detect close mode (resp)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
889 }
890 else if (!(h1m->flags & H1_MF_CONN_KAL) &&
891 (fe->options & PR_O_HTTP_MODE) == PR_O_HTTP_CLO) {
892 /* no explicit keep-alive and option httpclose => close */
893 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
894 TRACE_STATE("force close mode (resp)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
895 }
896 }
897 }
898 else {
899 /* Input direction: first pass */
900 if (!(h1m->flags & (H1_MF_VER_11|H1_MF_CONN_KAL)) || h1m->flags & H1_MF_CONN_CLO) {
901 /* no explicit keep-alive in HTTP/1.0 or explicit close => close*/
902 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
903 TRACE_STATE("detect close mode (req)", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s);
904 }
905 }
906
907 /* If KAL, check if the frontend is stopping. If yes, switch in CLO mode */
908 if (h1s->flags & H1S_F_WANT_KAL && fe->state == PR_STSTOPPED) {
909 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
910 TRACE_STATE("stopping, set close mode", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
911 }
912 }
913
914 /* Deduce the connection mode of the client connection, depending on the
915 * configuration and the H1 message flags. This function is called twice, the
916 * first time when the request is parsed and the second time when the response
917 * is parsed.
918 */
h1_set_srv_conn_mode(struct h1s * h1s,struct h1m * h1m)919 static void h1_set_srv_conn_mode(struct h1s *h1s, struct h1m *h1m)
920 {
921 struct session *sess = h1s->sess;
922 struct proxy *be = h1s->h1c->px;
923 int fe_flags = sess ? sess->fe->options : 0;
924
925 if (h1m->flags & H1_MF_RESP) {
926 /* Input direction: second pass */
927 if ((h1s->meth == HTTP_METH_CONNECT && h1s->status == 200) ||
928 h1s->status == 101) {
929 /* Either we've established an explicit tunnel, or we're
930 * switching the protocol. In both cases, we're very unlikely to
931 * understand the next protocols. We have to switch to tunnel
932 * mode, so that we transfer the request and responses then let
933 * this protocol pass unmodified. When we later implement
934 * specific parsers for such protocols, we'll want to check the
935 * Upgrade header which contains information about that protocol
936 * for responses with status 101 (eg: see RFC2817 about TLS).
937 */
938 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_TUN;
939 TRACE_STATE("set tunnel mode (resp)", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s);
940 }
941 else if (h1s->flags & H1S_F_WANT_KAL) {
942 /* By default the server is in KAL mode. CLOSE mode mean
943 * it is imposed by haproxy itself. So only change KAL
944 * mode here. */
945 if (!(h1m->flags & H1_MF_XFER_LEN) || h1m->flags & H1_MF_CONN_CLO ||
946 !(h1m->flags & (H1_MF_VER_11|H1_MF_CONN_KAL))){
947 /* no length known or explicit close or no explicit keep-alive in HTTP/1.0 => close */
948 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
949 TRACE_STATE("detect close mode (resp)", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s);
950 }
951 }
952 }
953 else {
954 /* Output direction: first pass */
955 if (h1m->flags & H1_MF_CONN_CLO) {
956 /* explicit close => close */
957 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
958 TRACE_STATE("detect close mode (req)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
959 }
960 else if (!(h1m->flags & H1_MF_CONN_KAL) &&
961 ((fe_flags & PR_O_HTTP_MODE) == PR_O_HTTP_SCL ||
962 (be->options & PR_O_HTTP_MODE) == PR_O_HTTP_SCL ||
963 (fe_flags & PR_O_HTTP_MODE) == PR_O_HTTP_CLO ||
964 (be->options & PR_O_HTTP_MODE) == PR_O_HTTP_CLO)) {
965 /* no explicit keep-alive option httpclose/server-close => close */
966 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
967 TRACE_STATE("force close mode (req)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
968 }
969 }
970
971 /* If KAL, check if the backend is stopping. If yes, switch in CLO mode */
972 if (h1s->flags & H1S_F_WANT_KAL && be->state == PR_STSTOPPED) {
973 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
974 TRACE_STATE("stopping, set close mode", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
975 }
976 }
977
h1_update_req_conn_value(struct h1s * h1s,struct h1m * h1m,struct ist * conn_val)978 static void h1_update_req_conn_value(struct h1s *h1s, struct h1m *h1m, struct ist *conn_val)
979 {
980 struct proxy *px = h1s->h1c->px;
981
982 /* Don't update "Connection:" header in TUNNEL mode or if "Upgrage"
983 * token is found
984 */
985 if (h1s->flags & H1S_F_WANT_TUN || h1m->flags & H1_MF_CONN_UPG)
986 return;
987
988 if (h1s->flags & H1S_F_WANT_KAL || px->options2 & PR_O2_FAKE_KA) {
989 if (!(h1m->flags & H1_MF_VER_11)) {
990 TRACE_STATE("add \"Connection: keep-alive\"", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
991 *conn_val = ist("keep-alive");
992 }
993 }
994 else { /* H1S_F_WANT_CLO && !PR_O2_FAKE_KA */
995 if (h1m->flags & H1_MF_VER_11) {
996 TRACE_STATE("add \"Connection: close\"", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
997 *conn_val = ist("close");
998 }
999 }
1000 }
1001
h1_update_res_conn_value(struct h1s * h1s,struct h1m * h1m,struct ist * conn_val)1002 static void h1_update_res_conn_value(struct h1s *h1s, struct h1m *h1m, struct ist *conn_val)
1003 {
1004 /* Don't update "Connection:" header in TUNNEL mode or if "Upgrage"
1005 * token is found
1006 */
1007 if (h1s->flags & H1S_F_WANT_TUN || h1m->flags & H1_MF_CONN_UPG)
1008 return;
1009
1010 if (h1s->flags & H1S_F_WANT_KAL) {
1011 if (!(h1m->flags & H1_MF_VER_11) ||
1012 !((h1m->flags & h1s->req.flags) & H1_MF_VER_11)) {
1013 TRACE_STATE("add \"Connection: keep-alive\"", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1014 *conn_val = ist("keep-alive");
1015 }
1016 }
1017 else { /* H1S_F_WANT_CLO */
1018 if (h1m->flags & H1_MF_VER_11) {
1019 TRACE_STATE("add \"Connection: close\"", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1020 *conn_val = ist("close");
1021 }
1022 }
1023 }
1024
h1_process_input_conn_mode(struct h1s * h1s,struct h1m * h1m,struct htx * htx)1025 static void h1_process_input_conn_mode(struct h1s *h1s, struct h1m *h1m, struct htx *htx)
1026 {
1027 if (!conn_is_back(h1s->h1c->conn))
1028 h1_set_cli_conn_mode(h1s, h1m);
1029 else
1030 h1_set_srv_conn_mode(h1s, h1m);
1031 }
1032
h1_process_output_conn_mode(struct h1s * h1s,struct h1m * h1m,struct ist * conn_val)1033 static void h1_process_output_conn_mode(struct h1s *h1s, struct h1m *h1m, struct ist *conn_val)
1034 {
1035 if (!conn_is_back(h1s->h1c->conn))
1036 h1_set_cli_conn_mode(h1s, h1m);
1037 else
1038 h1_set_srv_conn_mode(h1s, h1m);
1039
1040 if (!(h1m->flags & H1_MF_RESP))
1041 h1_update_req_conn_value(h1s, h1m, conn_val);
1042 else
1043 h1_update_res_conn_value(h1s, h1m, conn_val);
1044 }
1045
1046 /* Try to adjust the case of the message header name using the global map
1047 * <hdrs_map>.
1048 */
h1_adjust_case_outgoing_hdr(struct h1s * h1s,struct h1m * h1m,struct ist * name)1049 static void h1_adjust_case_outgoing_hdr(struct h1s *h1s, struct h1m *h1m, struct ist *name)
1050 {
1051 struct ebpt_node *node;
1052 struct h1_hdr_entry *entry;
1053
1054 /* No entry in the map, do nothing */
1055 if (eb_is_empty(&hdrs_map.map))
1056 return;
1057
1058 /* No conversion fo the request headers */
1059 if (!(h1m->flags & H1_MF_RESP) && !(h1s->h1c->px->options2 & PR_O2_H1_ADJ_BUGSRV))
1060 return;
1061
1062 /* No conversion fo the response headers */
1063 if ((h1m->flags & H1_MF_RESP) && !(h1s->h1c->px->options2 & PR_O2_H1_ADJ_BUGCLI))
1064 return;
1065
1066 node = ebis_lookup_len(&hdrs_map.map, name->ptr, name->len);
1067 if (!node)
1068 return;
1069 entry = container_of(node, struct h1_hdr_entry, node);
1070 name->ptr = entry->name.ptr;
1071 name->len = entry->name.len;
1072 }
1073
1074 /* Append the description of what is present in error snapshot <es> into <out>.
1075 * The description must be small enough to always fit in a buffer. The output
1076 * buffer may be the trash so the trash must not be used inside this function.
1077 */
h1_show_error_snapshot(struct buffer * out,const struct error_snapshot * es)1078 static void h1_show_error_snapshot(struct buffer *out, const struct error_snapshot *es)
1079 {
1080 chunk_appendf(out,
1081 " H1 connection flags 0x%08x, H1 stream flags 0x%08x\n"
1082 " H1 msg state %s(%d), H1 msg flags 0x%08x\n"
1083 " H1 chunk len %lld bytes, H1 body len %lld bytes :\n",
1084 es->ctx.h1.c_flags, es->ctx.h1.s_flags,
1085 h1m_state_str(es->ctx.h1.state), es->ctx.h1.state,
1086 es->ctx.h1.m_flags, es->ctx.h1.m_clen, es->ctx.h1.m_blen);
1087 }
1088 /*
1089 * Capture a bad request or response and archive it in the proxy's structure.
1090 * By default it tries to report the error position as h1m->err_pos. However if
1091 * this one is not set, it will then report h1m->next, which is the last known
1092 * parsing point. The function is able to deal with wrapping buffers. It always
1093 * displays buffers as a contiguous area starting at buf->p. The direction is
1094 * determined thanks to the h1m's flags.
1095 */
h1_capture_bad_message(struct h1c * h1c,struct h1s * h1s,struct h1m * h1m,struct buffer * buf)1096 static void h1_capture_bad_message(struct h1c *h1c, struct h1s *h1s,
1097 struct h1m *h1m, struct buffer *buf)
1098 {
1099 struct session *sess = h1s->sess;
1100 struct proxy *proxy = h1c->px;
1101 struct proxy *other_end;
1102 union error_snapshot_ctx ctx;
1103
1104 if (h1s->cs && h1s->cs->data) {
1105 if (sess == NULL)
1106 sess = si_strm(h1s->cs->data)->sess;
1107 if (!(h1m->flags & H1_MF_RESP))
1108 other_end = si_strm(h1s->cs->data)->be;
1109 else
1110 other_end = sess->fe;
1111 } else
1112 other_end = NULL;
1113
1114 /* http-specific part now */
1115 ctx.h1.state = h1m->state;
1116 ctx.h1.c_flags = h1c->flags;
1117 ctx.h1.s_flags = h1s->flags;
1118 ctx.h1.m_flags = h1m->flags;
1119 ctx.h1.m_clen = h1m->curr_len;
1120 ctx.h1.m_blen = h1m->body_len;
1121
1122 proxy_capture_error(proxy, !!(h1m->flags & H1_MF_RESP), other_end,
1123 h1c->conn->target, sess, buf, 0, 0,
1124 (h1m->err_pos >= 0) ? h1m->err_pos : h1m->next,
1125 &ctx, h1_show_error_snapshot);
1126 }
1127
1128 /* Emit the chunksize followed by a CRLF in front of data of the buffer
1129 * <buf>. It goes backwards and starts with the byte before the buffer's
1130 * head. The caller is responsible for ensuring there is enough room left before
1131 * the buffer's head for the string.
1132 */
h1_emit_chunk_size(struct buffer * buf,size_t chksz)1133 static void h1_emit_chunk_size(struct buffer *buf, size_t chksz)
1134 {
1135 char *beg, *end;
1136
1137 beg = end = b_head(buf);
1138 *--beg = '\n';
1139 *--beg = '\r';
1140 do {
1141 *--beg = hextab[chksz & 0xF];
1142 } while (chksz >>= 4);
1143 buf->head -= (end - beg);
1144 b_add(buf, end - beg);
1145 }
1146
1147 /* Emit a CRLF after the data of the buffer <buf>. The caller is responsible for
1148 * ensuring there is enough room left in the buffer for the string. */
h1_emit_chunk_crlf(struct buffer * buf)1149 static void h1_emit_chunk_crlf(struct buffer *buf)
1150 {
1151 *(b_peek(buf, b_data(buf))) = '\r';
1152 *(b_peek(buf, b_data(buf) + 1)) = '\n';
1153 b_add(buf, 2);
1154 }
1155
1156 /*
1157 * Switch the request to tunnel mode. This function must only be called for
1158 * CONNECT requests. On the client side, if the response is not finished, the
1159 * mux is mark as busy on input.
1160 */
h1_set_req_tunnel_mode(struct h1s * h1s)1161 static void h1_set_req_tunnel_mode(struct h1s *h1s)
1162 {
1163 h1s->req.flags &= ~(H1_MF_XFER_LEN|H1_MF_CLEN|H1_MF_CHNK);
1164 h1s->req.state = H1_MSG_TUNNEL;
1165 TRACE_STATE("switch H1 request in tunnel mode", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1166
1167 if (!conn_is_back(h1s->h1c->conn)) {
1168 h1s->flags &= ~H1S_F_PARSING_DONE;
1169 if (h1s->res.state < H1_MSG_DONE) {
1170 h1s->h1c->flags |= H1C_F_IN_BUSY;
1171 TRACE_STATE("switch h1c in busy mode", H1_EV_RX_DATA|H1_EV_H1C_BLK, h1s->h1c->conn, h1s);
1172 }
1173 }
1174 else if (h1s->h1c->flags & H1C_F_IN_BUSY) {
1175 h1s->h1c->flags &= ~H1C_F_IN_BUSY;
1176 tasklet_wakeup(h1s->h1c->wait_event.tasklet);
1177 TRACE_STATE("h1c no more busy", H1_EV_RX_DATA|H1_EV_H1C_BLK|H1_EV_H1C_WAKE, h1s->h1c->conn, h1s);
1178 }
1179 }
1180
1181 /*
1182 * Switch the response to tunnel mode. This function must only be called on
1183 * successful replies to CONNECT requests or on protocol switching. In this
1184 * last case, this function takes care to switch the request to tunnel mode if
1185 * possible. On the server side, if the request is not finished, the mux is mark
1186 * as busy on input.
1187 */
h1_set_res_tunnel_mode(struct h1s * h1s)1188 static void h1_set_res_tunnel_mode(struct h1s *h1s)
1189 {
1190
1191 h1s->res.flags &= ~(H1_MF_XFER_LEN|H1_MF_CLEN|H1_MF_CHNK);
1192 h1s->res.state = H1_MSG_TUNNEL;
1193 TRACE_STATE("switch H1 response in tunnel mode", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1194
1195 if (conn_is_back(h1s->h1c->conn)) {
1196 h1s->flags &= ~H1S_F_PARSING_DONE;
1197 /* On protocol switching, switch the request to tunnel mode if it is in
1198 * DONE state. Otherwise we will wait the end of the request to switch
1199 * it in tunnel mode.
1200 */
1201 if (h1s->req.state < H1_MSG_DONE) {
1202 h1s->h1c->flags |= H1C_F_IN_BUSY;
1203 TRACE_STATE("switch h1c in busy mode", H1_EV_RX_DATA|H1_EV_H1C_BLK, h1s->h1c->conn, h1s);
1204 }
1205 else if (h1s->status == 101 && h1s->req.state == H1_MSG_DONE) {
1206 h1s->req.flags &= ~(H1_MF_XFER_LEN|H1_MF_CLEN|H1_MF_CHNK);
1207 h1s->req.state = H1_MSG_TUNNEL;
1208 TRACE_STATE("switch H1 request in tunnel mode", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1209 }
1210 }
1211 else if (h1s->h1c->flags & H1C_F_IN_BUSY) {
1212 h1s->h1c->flags &= ~H1C_F_IN_BUSY;
1213 tasklet_wakeup(h1s->h1c->wait_event.tasklet);
1214 TRACE_STATE("h1c no more busy", H1_EV_RX_DATA|H1_EV_H1C_BLK|H1_EV_H1C_WAKE, h1s->h1c->conn, h1s);
1215 }
1216 }
1217
1218 /*
1219 * Parse HTTP/1 headers. It returns the number of bytes parsed if > 0, or 0 if
1220 * it couldn't proceed. Parsing errors are reported by setting H1S_F_*_ERROR
1221 * flag. If more room is requested, H1S_F_RX_CONGESTED flag is set. If relies on
1222 * the function http_parse_msg_hdrs() to do the parsing.
1223 */
h1_process_headers(struct h1s * h1s,struct h1m * h1m,struct htx * htx,struct buffer * buf,size_t * ofs,size_t max)1224 static size_t h1_process_headers(struct h1s *h1s, struct h1m *h1m, struct htx *htx,
1225 struct buffer *buf, size_t *ofs, size_t max)
1226 {
1227 union h1_sl h1sl;
1228 int ret = 0;
1229
1230 TRACE_ENTER(H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s,, (size_t[]){max});
1231
1232 if (!(h1s->h1c->px->options2 & PR_O2_NO_H2_UPGRADE) && /* H2 upgrade supported by the proxy */
1233 !(h1s->flags & H1S_F_NOT_FIRST) && /* It is the first transaction */
1234 !(h1m->flags & H1_MF_RESP)) { /* It is a request */
1235 /* Try to match H2 preface before parsing the request headers. */
1236 ret = b_isteq(buf, 0, b_data(buf), ist(H2_CONN_PREFACE));
1237 if (ret > 0) {
1238 goto h2c_upgrade;
1239 }
1240 }
1241 else {
1242 if (h1s->meth == HTTP_METH_CONNECT)
1243 h1m->flags |= H1_MF_METH_CONNECT;
1244 if (h1s->meth == HTTP_METH_HEAD)
1245 h1m->flags |= H1_MF_METH_HEAD;
1246 }
1247
1248 ret = h1_parse_msg_hdrs(h1m, &h1sl, htx, buf, *ofs, max);
1249 if (ret <= 0) {
1250 TRACE_DEVEL("leaving on missing data or error", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s);
1251 if (ret == -1) {
1252 if (!(h1m->flags & H1_MF_RESP)) {
1253 h1s->flags |= H1S_F_REQ_ERROR;
1254 TRACE_USER("rejected H1 request", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1255 }
1256 else {
1257 h1s->flags |= H1S_F_RES_ERROR;
1258 TRACE_USER("rejected H1 response", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1259 }
1260 h1s->cs->flags |= CS_FL_EOI;
1261 TRACE_STATE("parsing error", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1262 h1_capture_bad_message(h1s->h1c, h1s, h1m, buf);
1263 }
1264 else if (ret == -2) {
1265 TRACE_STATE("RX path congested, waiting for more space", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_H1S_BLK, h1s->h1c->conn, h1s);
1266 h1s->flags |= H1S_F_RX_CONGESTED;
1267 }
1268 ret = 0;
1269 goto end;
1270 }
1271
1272 if (h1m->err_pos >= 0) {
1273 /* Maybe we found an error during the parsing while we were
1274 * configured not to block on that, so we have to capture it
1275 * now.
1276 */
1277 TRACE_STATE("Ignored parsing error", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s);
1278 h1_capture_bad_message(h1s->h1c, h1s, h1m, buf);
1279 }
1280
1281 if (!(h1m->flags & H1_MF_RESP)) {
1282 h1s->meth = h1sl.rq.meth;
1283 if (h1m->state == H1_MSG_TUNNEL)
1284 h1_set_req_tunnel_mode(h1s);
1285 }
1286 else {
1287 h1s->status = h1sl.st.status;
1288 if (h1m->state == H1_MSG_TUNNEL)
1289 h1_set_res_tunnel_mode(h1s);
1290 }
1291 h1_process_input_conn_mode(h1s, h1m, htx);
1292 *ofs += ret;
1293
1294 end:
1295 TRACE_LEAVE(H1_EV_RX_DATA|H1_EV_RX_HDRS, h1s->h1c->conn, h1s,, (size_t[]){ret});
1296 return ret;
1297
1298 h2c_upgrade:
1299 h1s->h1c->flags |= H1C_F_UPG_H2C;
1300 h1s->cs->flags |= CS_FL_EOI;
1301 htx->flags |= HTX_FL_UPGRADE;
1302 TRACE_DEVEL("leaving on H2 update", H1_EV_RX_DATA|H1_EV_RX_HDRS|H1_EV_RX_EOI, h1s->h1c->conn, h1s);
1303 return 0;
1304 }
1305
1306 /*
1307 * Parse HTTP/1 body. It returns the number of bytes parsed if > 0, or 0 if it
1308 * couldn't proceed. Parsing errors are reported by setting H1S_F_*_ERROR flag.
1309 * If relies on the function http_parse_msg_data() to do the parsing.
1310 */
h1_process_data(struct h1s * h1s,struct h1m * h1m,struct htx ** htx,struct buffer * buf,size_t * ofs,size_t max,struct buffer * htxbuf)1311 static size_t h1_process_data(struct h1s *h1s, struct h1m *h1m, struct htx **htx,
1312 struct buffer *buf, size_t *ofs, size_t max,
1313 struct buffer *htxbuf)
1314 {
1315 int ret;
1316
1317 TRACE_ENTER(H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s,, (size_t[]){max});
1318 ret = h1_parse_msg_data(h1m, htx, buf, *ofs, max, htxbuf);
1319 if (!ret) {
1320 TRACE_DEVEL("leaving on missing data or error", H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s);
1321 if ((*htx)->flags & HTX_FL_PARSING_ERROR) {
1322 if (!(h1m->flags & H1_MF_RESP)) {
1323 h1s->flags |= H1S_F_REQ_ERROR;
1324 TRACE_USER("rejected H1 request", H1_EV_RX_DATA|H1_EV_RX_BODY|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1325 }
1326 else {
1327 h1s->flags |= H1S_F_RES_ERROR;
1328 TRACE_USER("rejected H1 response", H1_EV_RX_DATA|H1_EV_RX_BODY|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1329 }
1330 h1s->cs->flags |= CS_FL_EOI;
1331 TRACE_STATE("parsing error", H1_EV_RX_DATA|H1_EV_RX_BODY|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1332 h1_capture_bad_message(h1s->h1c, h1s, h1m, buf);
1333 }
1334 goto end;
1335 }
1336
1337 if (h1s->cs && !(h1m->flags & H1_MF_CHNK) &&
1338 ((h1m->state == H1_MSG_DATA && h1m->curr_len) || (h1m->state == H1_MSG_TUNNEL))) {
1339 TRACE_STATE("notify the mux can use splicing", H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s);
1340 h1s->cs->flags |= CS_FL_MAY_SPLICE;
1341 }
1342 else if (h1s->cs) {
1343 TRACE_STATE("notify the mux can't use splicing anymore", H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s);
1344 h1s->cs->flags &= ~CS_FL_MAY_SPLICE;
1345 }
1346
1347 *ofs += ret;
1348
1349 end:
1350 if (b_data(buf) != *ofs && (h1m->state == H1_MSG_DATA || h1m->state == H1_MSG_TUNNEL)) {
1351 TRACE_STATE("RX path congested, waiting for more space", H1_EV_RX_DATA|H1_EV_RX_BODY|H1_EV_H1S_BLK, h1s->h1c->conn, h1s);
1352 h1s->flags |= H1S_F_RX_CONGESTED;
1353 }
1354
1355 TRACE_LEAVE(H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s,, (size_t[]){ret});
1356 return ret;
1357 }
1358
1359 /*
1360 * Parse HTTP/1 trailers. It returns the number of bytes parsed if > 0, or 0 if
1361 * it couldn't proceed. Parsing errors are reported by setting H1S_F_*_ERROR
1362 * flag and filling h1s->err_pos and h1s->err_state fields. This functions is
1363 * responsible to update the parser state <h1m>. If more room is requested,
1364 * H1S_F_RX_CONGESTED flag is set.
1365 */
h1_process_trailers(struct h1s * h1s,struct h1m * h1m,struct htx * htx,struct buffer * buf,size_t * ofs,size_t max)1366 static size_t h1_process_trailers(struct h1s *h1s, struct h1m *h1m, struct htx *htx,
1367 struct buffer *buf, size_t *ofs, size_t max)
1368 {
1369 int ret;
1370
1371 TRACE_ENTER(H1_EV_RX_DATA|H1_EV_RX_TLRS, h1s->h1c->conn, h1s,, (size_t[]){max});
1372 ret = h1_parse_msg_tlrs(h1m, htx, buf, *ofs, max);
1373 if (ret <= 0) {
1374 TRACE_DEVEL("leaving on missing data or error", H1_EV_RX_DATA|H1_EV_RX_BODY, h1s->h1c->conn, h1s);
1375 if (ret == -1) {
1376 if (!(h1m->flags & H1_MF_RESP)) {
1377 h1s->flags |= H1S_F_REQ_ERROR;
1378 TRACE_USER("rejected H1 request", H1_EV_RX_DATA|H1_EV_RX_TLRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1379 }
1380 else {
1381 h1s->flags |= H1S_F_RES_ERROR;
1382 TRACE_USER("rejected H1 response", H1_EV_RX_DATA|H1_EV_RX_TLRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1383 }
1384 h1s->cs->flags |= CS_FL_EOI;
1385 TRACE_STATE("parsing error", H1_EV_RX_DATA|H1_EV_RX_TLRS|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1386 h1_capture_bad_message(h1s->h1c, h1s, h1m, buf);
1387 }
1388 else if (ret == -2) {
1389 TRACE_STATE("RX path congested, waiting for more space", H1_EV_RX_DATA|H1_EV_RX_TLRS|H1_EV_H1S_BLK, h1s->h1c->conn, h1s);
1390 h1s->flags |= H1S_F_RX_CONGESTED;
1391 }
1392 ret = 0;
1393 goto end;
1394 }
1395
1396 *ofs += ret;
1397
1398 end:
1399 TRACE_LEAVE(H1_EV_RX_DATA|H1_EV_RX_TLRS, h1s->h1c->conn, h1s,, (size_t[]){ret});
1400 return ret;
1401 }
1402
1403 /*
1404 * Add the EOM in the HTX message. It returns 1 on success or 0 if it couldn't
1405 * proceed. This functions is responsible to update the parser state <h1m>. It
1406 * also add the flag CS_FL_EOI on the CS.
1407 */
h1_process_eom(struct h1s * h1s,struct h1m * h1m,struct htx * htx,struct buffer * buf,size_t * ofs,size_t max)1408 static size_t h1_process_eom(struct h1s *h1s, struct h1m *h1m, struct htx *htx,
1409 struct buffer *buf, size_t *ofs, size_t max)
1410 {
1411 int ret;
1412
1413 TRACE_ENTER(H1_EV_RX_DATA|H1_EV_RX_EOI, h1s->h1c->conn, h1s,, (size_t[]){max});
1414 ret = h1_parse_msg_eom(h1m, htx, max);
1415 if (!ret) {
1416 TRACE_DEVEL("leaving on missing data or error", H1_EV_RX_DATA|H1_EV_RX_EOI, h1s->h1c->conn, h1s);
1417 if (htx->flags & HTX_FL_PARSING_ERROR) {
1418 if (!(h1m->flags & H1_MF_RESP)) {
1419 h1s->flags |= H1S_F_REQ_ERROR;
1420 TRACE_USER("rejected H1 request", H1_EV_RX_DATA|H1_EV_RX_EOI|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1421 }
1422 else {
1423 h1s->flags |= H1S_F_RES_ERROR;
1424 TRACE_USER("rejected H1 response", H1_EV_RX_DATA|H1_EV_RX_EOI|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1425 }
1426 h1s->cs->flags |= CS_FL_EOI;
1427 TRACE_STATE("parsing error", H1_EV_RX_DATA|H1_EV_RX_EOI|H1_EV_H1S_ERR, h1s->h1c->conn, h1s);
1428 h1_capture_bad_message(h1s->h1c, h1s, h1m, buf);
1429 }
1430 goto end;
1431 }
1432
1433 h1s->flags |= H1S_F_PARSING_DONE;
1434 /* Set EOI on conn-stream in DONE state iff:
1435 * - it is a response
1436 * - it is a request but no a protocol upgrade nor a CONNECT
1437 *
1438 * If not set, Wait the response to do so or not depending on the status
1439 * code.
1440 */
1441 if ((h1m->flags & H1_MF_RESP) || ((h1s->meth != HTTP_METH_CONNECT) && !(h1m->flags & H1_MF_CONN_UPG)))
1442 h1s->cs->flags |= CS_FL_EOI;
1443 end:
1444 TRACE_LEAVE(H1_EV_RX_DATA|H1_EV_RX_EOI, h1s->h1c->conn, h1s,, (size_t[]){ret});
1445 return ret;
1446 }
1447
1448 /*
1449 * Process incoming data. It parses data and transfer them from h1c->ibuf into
1450 * <buf>. It returns the number of bytes parsed and transferred if > 0, or 0 if
1451 * it couldn't proceed.
1452 *
1453 * WARNING: H1S_F_RX_CONGESTED flag must be removed before processing input data.
1454 */
h1_process_input(struct h1c * h1c,struct buffer * buf,size_t count)1455 static size_t h1_process_input(struct h1c *h1c, struct buffer *buf, size_t count)
1456 {
1457 struct h1s *h1s = h1c->h1s;
1458 struct h1m *h1m;
1459 struct htx *htx;
1460 size_t ret, data;
1461 size_t total = 0;
1462 int errflag;
1463
1464 htx = htx_from_buf(buf);
1465 TRACE_ENTER(H1_EV_RX_DATA, h1c->conn, h1s, htx, (size_t[]){count});
1466
1467 if (!conn_is_back(h1c->conn)) {
1468 h1m = &h1s->req;
1469 errflag = H1S_F_REQ_ERROR;
1470 }
1471 else {
1472 h1m = &h1s->res;
1473 errflag = H1S_F_RES_ERROR;
1474 }
1475
1476 data = htx->data;
1477 if (h1s->flags & errflag)
1478 goto end;
1479
1480 if (h1c->flags & H1C_F_IN_BUSY)
1481 goto end;
1482
1483 /* Always remove congestion flags and try to process more input data */
1484 h1s->flags &= ~H1S_F_RX_CONGESTED;
1485
1486 do {
1487 size_t used = htx_used_space(htx);
1488
1489 if (h1m->state <= H1_MSG_LAST_LF) {
1490 TRACE_PROTO("parsing message headers", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1c->conn, h1s);
1491 ret = h1_process_headers(h1s, h1m, htx, &h1c->ibuf, &total, count);
1492 if (!ret)
1493 break;
1494
1495 TRACE_USER((!(h1m->flags & H1_MF_RESP) ? "rcvd H1 request headers" : "rcvd H1 response headers"),
1496 H1_EV_RX_DATA|H1_EV_RX_HDRS, h1c->conn, h1s, htx, (size_t[]){ret});
1497
1498 if ((h1m->flags & H1_MF_RESP) &&
1499 h1s->status < 200 && (h1s->status == 100 || h1s->status >= 102)) {
1500 h1m_init_res(&h1s->res);
1501 h1m->flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR);
1502 TRACE_STATE("1xx response rcvd", H1_EV_RX_DATA|H1_EV_RX_HDRS, h1c->conn, h1s);
1503 }
1504 }
1505 else if (h1m->state < H1_MSG_TRAILERS) {
1506 TRACE_PROTO("parsing message payload", H1_EV_RX_DATA|H1_EV_RX_BODY, h1c->conn, h1s);
1507 ret = h1_process_data(h1s, h1m, &htx, &h1c->ibuf, &total, count, buf);
1508 if (!ret && h1m->state != H1_MSG_DONE)
1509 break;
1510
1511 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "rcvd H1 request payload data" : "rcvd H1 response payload data"),
1512 H1_EV_RX_DATA|H1_EV_RX_BODY, h1c->conn, h1s, htx, (size_t[]){ret});
1513 }
1514 else if (h1m->state == H1_MSG_TRAILERS) {
1515 TRACE_PROTO("parsing message trailers", H1_EV_RX_DATA|H1_EV_RX_TLRS, h1c->conn, h1s);
1516 ret = h1_process_trailers(h1s, h1m, htx, &h1c->ibuf, &total, count);
1517 if (!ret && h1m->state != H1_MSG_DONE)
1518 break;
1519
1520 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "rcvd H1 request trailers" : "rcvd H1 response trailers"),
1521 H1_EV_RX_DATA|H1_EV_RX_TLRS, h1c->conn, h1s, htx, (size_t[]){ret});
1522 }
1523 else if (h1m->state == H1_MSG_DONE) {
1524 if (!(h1s->flags & H1S_F_PARSING_DONE)) {
1525 if (!h1_process_eom(h1s, h1m, htx, &h1c->ibuf, &total, count))
1526 break;
1527
1528 TRACE_USER((!(h1m->flags & H1_MF_RESP) ? "H1 request fully rcvd" : "H1 response fully rcvd"),
1529 H1_EV_RX_DATA|H1_EV_RX_EOI, h1c->conn, h1s, htx);
1530 }
1531
1532 if (!(h1m->flags & H1_MF_RESP) && h1s->status == 101)
1533 h1_set_req_tunnel_mode(h1s);
1534 else if ((h1m->flags & H1_MF_RESP) && h1s->req.state == H1_MSG_TUNNEL) {
1535 TRACE_STATE("switch back H1 request from tunnel mode", H1_EV_RX_DATA|H1_EV_H1C_BLK, h1c->conn, h1s);
1536 h1s->req.state = H1_MSG_DONE;
1537 }
1538 else if (h1s->req.state < H1_MSG_DONE || h1s->res.state < H1_MSG_DONE) {
1539 h1c->flags |= H1C_F_IN_BUSY;
1540 TRACE_STATE("switch h1c in busy mode", H1_EV_RX_DATA|H1_EV_H1C_BLK, h1c->conn, h1s);
1541 break;
1542 }
1543 else
1544 break;
1545 }
1546 else if (h1m->state == H1_MSG_TUNNEL) {
1547 TRACE_PROTO("parsing tunneled data", H1_EV_RX_DATA, h1c->conn, h1s);
1548 ret = h1_process_data(h1s, h1m, &htx, &h1c->ibuf, &total, count, buf);
1549 if (!ret)
1550 break;
1551
1552 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "rcvd H1 request tunneled data" : "rcvd H1 response tunneled data"),
1553 H1_EV_RX_DATA|H1_EV_RX_EOI, h1c->conn, h1s, htx, (size_t[]){ret});
1554 }
1555 else {
1556 h1s->flags |= errflag;
1557 break;
1558 }
1559
1560 count -= htx_used_space(htx) - used;
1561 } while (!(h1s->flags & (errflag|H1S_F_RX_CONGESTED)));
1562
1563 if (h1s->flags & errflag) {
1564 TRACE_PROTO("parsing error", H1_EV_RX_DATA, h1c->conn, h1s);
1565 goto parsing_err;
1566 }
1567
1568 b_del(&h1c->ibuf, total);
1569
1570 end:
1571 htx_to_buf(htx, buf);
1572 ret = htx->data - data;
1573 if ((h1c->flags & H1C_F_IN_FULL) && buf_room_for_htx_data(&h1c->ibuf)) {
1574 h1c->flags &= ~H1C_F_IN_FULL;
1575 TRACE_STATE("h1c ibuf not full anymore", H1_EV_RX_DATA|H1_EV_H1C_BLK|H1_EV_H1C_WAKE);
1576 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
1577 }
1578
1579 h1s->cs->flags &= ~(CS_FL_RCV_MORE | CS_FL_WANT_ROOM);
1580
1581 if (!b_data(&h1c->ibuf))
1582 h1_release_buf(h1c, &h1c->ibuf);
1583
1584 /* When Input data are pending for this message, notify upper layer that
1585 * the mux need more space in the HTX buffer to continue if :
1586 *
1587 * - The parser is blocked in MSG_DATA or MSG_TUNNEL state
1588 * - Headers or trailers are pending to be copied.
1589 */
1590 if (h1s->flags & (H1S_F_RX_CONGESTED)) {
1591 h1s->cs->flags |= CS_FL_RCV_MORE | CS_FL_WANT_ROOM;
1592 TRACE_STATE("waiting for more room", H1_EV_RX_DATA|H1_EV_H1S_BLK, h1c->conn, h1s);
1593 }
1594 else if (h1s->flags & H1S_F_REOS) {
1595 h1s->cs->flags |= CS_FL_EOS;
1596 if (h1m->state >= H1_MSG_DONE)
1597 h1s->cs->flags |= CS_FL_EOI;
1598 else if (h1m->state > H1_MSG_LAST_LF && h1m->state < H1_MSG_DONE)
1599 h1s->cs->flags |= CS_FL_ERROR;
1600 }
1601
1602 TRACE_LEAVE(H1_EV_RX_DATA, h1c->conn, h1s, htx, (size_t[]){ret});
1603 return ret;
1604
1605 parsing_err:
1606 b_reset(&h1c->ibuf);
1607 htx_to_buf(htx, buf);
1608 TRACE_DEVEL("leaving on error", H1_EV_RX_DATA|H1_EV_STRM_ERR, h1c->conn, h1s);
1609 return 0;
1610 }
1611
1612 /*
1613 * Process outgoing data. It parses data and transfer them from the channel buffer into
1614 * h1c->obuf. It returns the number of bytes parsed and transferred if > 0, or
1615 * 0 if it couldn't proceed.
1616 */
h1_process_output(struct h1c * h1c,struct buffer * buf,size_t count)1617 static size_t h1_process_output(struct h1c *h1c, struct buffer *buf, size_t count)
1618 {
1619 struct h1s *h1s = h1c->h1s;
1620 struct h1m *h1m;
1621 struct htx *chn_htx = NULL;
1622 struct htx_blk *blk;
1623 struct buffer tmp;
1624 size_t total = 0;
1625 int errflag;
1626
1627 if (!count)
1628 goto end;
1629
1630 chn_htx = htxbuf(buf);
1631 TRACE_ENTER(H1_EV_TX_DATA, h1c->conn, h1s, chn_htx, (size_t[]){count});
1632
1633 if (htx_is_empty(chn_htx))
1634 goto end;
1635
1636 if (!h1_get_buf(h1c, &h1c->obuf)) {
1637 h1c->flags |= H1C_F_OUT_ALLOC;
1638 TRACE_STATE("waiting for h1c obuf allocation", H1_EV_TX_DATA|H1_EV_H1S_BLK, h1c->conn, h1s);
1639 goto end;
1640 }
1641
1642 if (!conn_is_back(h1c->conn)) {
1643 h1m = &h1s->res;
1644 errflag = H1S_F_RES_ERROR;
1645 }
1646 else {
1647 h1m = &h1s->req;
1648 errflag = H1S_F_REQ_ERROR;
1649 }
1650
1651 if (h1s->flags & errflag)
1652 goto end;
1653
1654 /* the htx is non-empty thus has at least one block */
1655 blk = htx_get_head_blk(chn_htx);
1656
1657 /* Perform some optimizations to reduce the number of buffer copies.
1658 * First, if the mux's buffer is empty and the htx area contains
1659 * exactly one data block of the same size as the requested count,
1660 * then it's possible to simply swap the caller's buffer with the
1661 * mux's output buffer and adjust offsets and length to match the
1662 * entire DATA HTX block in the middle. In this case we perform a
1663 * true zero-copy operation from end-to-end. This is the situation
1664 * that happens all the time with large files. Second, if this is not
1665 * possible, but the mux's output buffer is empty, we still have an
1666 * opportunity to avoid the copy to the intermediary buffer, by making
1667 * the intermediary buffer's area point to the output buffer's area.
1668 * In this case we want to skip the HTX header to make sure that copies
1669 * remain aligned and that this operation remains possible all the
1670 * time. This goes for headers, data blocks and any data extracted from
1671 * the HTX blocks.
1672 */
1673 if (!b_data(&h1c->obuf)) {
1674 if (htx_nbblks(chn_htx) == 1 &&
1675 htx_get_blk_type(blk) == HTX_BLK_DATA &&
1676 htx_get_blk_value(chn_htx, blk).len == count) {
1677 void *old_area = h1c->obuf.area;
1678
1679 TRACE_PROTO("sending message data (zero-copy)", H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s, chn_htx, (size_t[]){count});
1680 h1c->obuf.area = buf->area;
1681 h1c->obuf.head = sizeof(struct htx) + blk->addr;
1682 h1c->obuf.data = count;
1683
1684 buf->area = old_area;
1685 buf->data = buf->head = 0;
1686
1687 chn_htx = (struct htx *)buf->area;
1688 htx_reset(chn_htx);
1689
1690 /* The message is chunked. We need to emit the chunk
1691 * size. We have at least the size of the struct htx to
1692 * write the chunk envelope. It should be enough.
1693 */
1694 if (h1m->flags & H1_MF_CHNK) {
1695 h1_emit_chunk_size(&h1c->obuf, count);
1696 h1_emit_chunk_crlf(&h1c->obuf);
1697 }
1698
1699 total += count;
1700 if (h1m->state == H1_MSG_DATA)
1701 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request payload data xferred" : "H1 response payload data xferred"),
1702 H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s,, (size_t[]){count});
1703 else
1704 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request tunneled data xferred" : "H1 response tunneled data xferred"),
1705 H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s,, (size_t[]){count});
1706 goto out;
1707 }
1708 tmp.area = h1c->obuf.area + h1c->obuf.head;
1709 }
1710 else
1711 tmp.area = trash.area;
1712
1713 tmp.data = 0;
1714 tmp.size = b_room(&h1c->obuf);
1715 while (count && !(h1s->flags & errflag) && blk) {
1716 struct htx_sl *sl;
1717 struct ist n, v;
1718 enum htx_blk_type type = htx_get_blk_type(blk);
1719 uint32_t sz = htx_get_blksz(blk);
1720 uint32_t vlen, chklen;
1721
1722 vlen = sz;
1723 if (type != HTX_BLK_DATA && vlen > count)
1724 goto full;
1725
1726 if (type == HTX_BLK_UNUSED)
1727 goto nextblk;
1728
1729 switch (h1m->state) {
1730 case H1_MSG_RQBEFORE:
1731 if (type != HTX_BLK_REQ_SL)
1732 goto error;
1733 TRACE_USER("sending request headers", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s, chn_htx);
1734 sl = htx_get_blk_ptr(chn_htx, blk);
1735 h1s->meth = sl->info.req.meth;
1736 h1_parse_req_vsn(h1m, sl);
1737 if (!h1_format_htx_reqline(sl, &tmp))
1738 goto full;
1739 h1m->flags |= H1_MF_XFER_LEN;
1740 if (sl->flags & HTX_SL_F_BODYLESS)
1741 h1m->flags |= H1_MF_CLEN;
1742 h1m->state = H1_MSG_HDR_FIRST;
1743 break;
1744
1745 case H1_MSG_RPBEFORE:
1746 if (type != HTX_BLK_RES_SL)
1747 goto error;
1748 TRACE_USER("sending response headers", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s, chn_htx);
1749 sl = htx_get_blk_ptr(chn_htx, blk);
1750 h1s->status = sl->info.res.status;
1751 h1_parse_res_vsn(h1m, sl);
1752 if (!h1_format_htx_stline(sl, &tmp))
1753 goto full;
1754 if (sl->flags & HTX_SL_F_XFER_LEN)
1755 h1m->flags |= H1_MF_XFER_LEN;
1756 if (sl->info.res.status < 200 &&
1757 (sl->info.res.status == 100 || sl->info.res.status >= 102))
1758 h1s->flags |= H1S_F_HAVE_O_CONN;
1759 h1m->state = H1_MSG_HDR_FIRST;
1760 break;
1761
1762 case H1_MSG_HDR_FIRST:
1763 case H1_MSG_HDR_NAME:
1764 case H1_MSG_HDR_L2_LWS:
1765 if (type == HTX_BLK_EOH)
1766 goto last_lf;
1767 if (type != HTX_BLK_HDR)
1768 goto error;
1769
1770 h1m->state = H1_MSG_HDR_NAME;
1771 n = htx_get_blk_name(chn_htx, blk);
1772 v = htx_get_blk_value(chn_htx, blk);
1773
1774 /* Skip all pseudo-headers */
1775 if (*(n.ptr) == ':')
1776 goto skip_hdr;
1777
1778 if (isteq(n, ist("transfer-encoding")))
1779 h1_parse_xfer_enc_header(h1m, v);
1780 else if (isteq(n, ist("content-length"))) {
1781 /* Only skip C-L header with invalid value. */
1782 if (h1_parse_cont_len_header(h1m, &v) < 0)
1783 goto skip_hdr;
1784 }
1785 else if (isteq(n, ist("connection"))) {
1786 h1_parse_connection_header(h1m, &v);
1787 if (!v.len)
1788 goto skip_hdr;
1789 }
1790 else if (isteq(n, ist("te"))) {
1791 /* "te" may only be sent with "trailers" if this value
1792 * is present, otherwise it must be deleted.
1793 */
1794 v = istist(v, ist("trailers"));
1795 if (!isttest(v) || (v.len > 8 && v.ptr[8] != ','))
1796 goto skip_hdr;
1797 v = ist("trailers");
1798 }
1799
1800 /* Skip header if same name is used to add the server name */
1801 if (!(h1m->flags & H1_MF_RESP) && h1c->px->server_id_hdr_name &&
1802 isteqi(n, ist2(h1c->px->server_id_hdr_name, h1c->px->server_id_hdr_len)))
1803 goto skip_hdr;
1804
1805 /* Try to adjust the case of the header name */
1806 if (h1c->px->options2 & (PR_O2_H1_ADJ_BUGCLI|PR_O2_H1_ADJ_BUGSRV))
1807 h1_adjust_case_outgoing_hdr(h1s, h1m, &n);
1808 if (!h1_format_htx_hdr(n, v, &tmp))
1809 goto full;
1810 skip_hdr:
1811 h1m->state = H1_MSG_HDR_L2_LWS;
1812 break;
1813
1814 case H1_MSG_LAST_LF:
1815 if (type != HTX_BLK_EOH)
1816 goto error;
1817 last_lf:
1818 h1m->state = H1_MSG_LAST_LF;
1819 if (!(h1s->flags & H1S_F_HAVE_O_CONN)) {
1820 /* If the reply comes from haproxy while the request is
1821 * not finished, we force the connection close. */
1822 if ((chn_htx->flags & HTX_FL_PROXY_RESP) && h1s->req.state != H1_MSG_DONE) {
1823 h1s->flags = (h1s->flags & ~H1S_F_WANT_MSK) | H1S_F_WANT_CLO;
1824 TRACE_STATE("force close mode (resp)", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1s->h1c->conn, h1s);
1825 }
1826
1827 /* the conn_mode must be processed. So do it */
1828 n = ist("connection");
1829 v = ist("");
1830 h1_process_output_conn_mode(h1s, h1m, &v);
1831 if (v.len) {
1832 /* Try to adjust the case of the header name */
1833 if (h1c->px->options2 & (PR_O2_H1_ADJ_BUGCLI|PR_O2_H1_ADJ_BUGSRV))
1834 h1_adjust_case_outgoing_hdr(h1s, h1m, &n);
1835 if (!h1_format_htx_hdr(n, v, &tmp))
1836 goto full;
1837 }
1838 h1s->flags |= H1S_F_HAVE_O_CONN;
1839 }
1840
1841 if ((h1s->meth != HTTP_METH_CONNECT &&
1842 (h1m->flags & (H1_MF_VER_11|H1_MF_RESP|H1_MF_CLEN|H1_MF_CHNK|H1_MF_XFER_LEN)) ==
1843 (H1_MF_VER_11|H1_MF_XFER_LEN)) ||
1844 (h1s->status >= 200 && h1s->status != 204 && h1s->status != 304 &&
1845 h1s->meth != HTTP_METH_HEAD && !(h1s->meth == HTTP_METH_CONNECT && h1s->status == 200) &&
1846 (h1m->flags & (H1_MF_VER_11|H1_MF_RESP|H1_MF_CLEN|H1_MF_CHNK|H1_MF_XFER_LEN)) ==
1847 (H1_MF_VER_11|H1_MF_RESP|H1_MF_XFER_LEN))) {
1848 /* chunking needed but header not seen */
1849 n = ist("transfer-encoding");
1850 v = ist("chunked");
1851 if (h1c->px->options2 & (PR_O2_H1_ADJ_BUGCLI|PR_O2_H1_ADJ_BUGSRV))
1852 h1_adjust_case_outgoing_hdr(h1s, h1m, &n);
1853 if (!h1_format_htx_hdr(n, v, &tmp))
1854 goto full;
1855 TRACE_STATE("add \"Transfer-Encoding: chunked\"", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1856 h1m->flags |= H1_MF_CHNK;
1857 }
1858
1859 /* Now add the server name to a header (if requested) */
1860 if (!(h1s->flags & H1S_F_HAVE_SRV_NAME) &&
1861 !(h1m->flags & H1_MF_RESP) && h1c->px->server_id_hdr_name) {
1862 struct server *srv = objt_server(h1c->conn->target);
1863
1864 if (srv) {
1865 n = ist2(h1c->px->server_id_hdr_name, h1c->px->server_id_hdr_len);
1866 v = ist(srv->id);
1867
1868 /* Try to adjust the case of the header name */
1869 if (h1c->px->options2 & (PR_O2_H1_ADJ_BUGCLI|PR_O2_H1_ADJ_BUGSRV))
1870 h1_adjust_case_outgoing_hdr(h1s, h1m, &n);
1871 if (!h1_format_htx_hdr(n, v, &tmp))
1872 goto full;
1873 }
1874 TRACE_STATE("add server name header", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1875 h1s->flags |= H1S_F_HAVE_SRV_NAME;
1876 }
1877
1878 if (!chunk_memcat(&tmp, "\r\n", 2))
1879 goto full;
1880
1881 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request headers xferred" : "H1 response headers xferred"),
1882 H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1883
1884 if (!(h1m->flags & H1_MF_RESP) && h1s->meth == HTTP_METH_CONNECT) {
1885 /* a CONNECT request is sent to the server. Switch it to tunnel mode. */
1886 h1_set_req_tunnel_mode(h1s);
1887 }
1888 else if ((h1m->flags & H1_MF_RESP) &&
1889 ((h1s->meth == HTTP_METH_CONNECT && h1s->status == 200) || h1s->status == 101)) {
1890 /* a successful reply to a CONNECT or a protocol switching is sent
1891 * to the client. Switch the response to tunnel mode.
1892 */
1893 h1_set_res_tunnel_mode(h1s);
1894 TRACE_STATE("switch H1 response in tunnel mode", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1895 }
1896 else if ((h1m->flags & H1_MF_RESP) &&
1897 h1s->status < 200 && (h1s->status == 100 || h1s->status >= 102)) {
1898 h1m_init_res(&h1s->res);
1899 h1m->flags |= (H1_MF_NO_PHDR|H1_MF_CLEAN_CONN_HDR);
1900 h1s->flags &= ~H1S_F_HAVE_O_CONN;
1901 TRACE_STATE("1xx response xferred", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1902 }
1903 else if ((h1m->flags & H1_MF_RESP) && h1s->meth == HTTP_METH_HEAD) {
1904 h1m->state = H1_MSG_DONE;
1905 TRACE_STATE("HEAD response processed", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
1906 }
1907 else
1908 h1m->state = H1_MSG_DATA;
1909 break;
1910
1911 case H1_MSG_DATA:
1912 case H1_MSG_TUNNEL:
1913 if (type == HTX_BLK_EOM) {
1914 /* Chunked message without explicit trailers */
1915 if (h1m->flags & H1_MF_CHNK) {
1916 if (!chunk_memcat(&tmp, "0\r\n\r\n", 5))
1917 goto full;
1918 }
1919 goto done;
1920 }
1921 else if (type == HTX_BLK_EOT || type == HTX_BLK_TLR) {
1922 /* If the message is not chunked, never
1923 * add the last chunk. */
1924 if ((h1m->flags & H1_MF_CHNK) && !chunk_memcat(&tmp, "0\r\n", 3))
1925 goto full;
1926 TRACE_PROTO("sending message trailers", H1_EV_TX_DATA|H1_EV_TX_TLRS, h1c->conn, h1s, chn_htx);
1927 goto trailers;
1928 }
1929 else if (type != HTX_BLK_DATA)
1930 goto error;
1931
1932 TRACE_PROTO("sending message data", H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s, chn_htx, (size_t[]){sz});
1933
1934
1935 if (vlen > count) {
1936 /* Get the maximum amount of data we can xferred */
1937 vlen = count;
1938 }
1939
1940 chklen = 0;
1941 if (h1m->flags & H1_MF_CHNK) {
1942 chklen = b_room(&tmp);
1943 chklen = ((chklen < 16) ? 1 : (chklen < 256) ? 2 :
1944 (chklen < 4096) ? 3 : (chklen < 65536) ? 4 :
1945 (chklen < 1048576) ? 5 : 8);
1946 chklen += 4; /* 2 x CRLF */
1947 }
1948
1949 if (vlen + chklen > b_room(&tmp)) {
1950 /* too large for the buffer */
1951 if (chklen >= b_room(&tmp))
1952 goto full;
1953 vlen = b_room(&tmp) - chklen;
1954 }
1955 v = htx_get_blk_value(chn_htx, blk);
1956 v.len = vlen;
1957 if (!h1_format_htx_data(v, &tmp, !!(h1m->flags & H1_MF_CHNK)))
1958 goto full;
1959
1960 if (h1m->state == H1_MSG_DATA)
1961 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request payload data xferred" : "H1 response payload data xferred"),
1962 H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s,, (size_t[]){v.len});
1963 else
1964 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request tunneled data xferred" : "H1 response tunneled data xferred"),
1965 H1_EV_TX_DATA|H1_EV_TX_BODY, h1c->conn, h1s,, (size_t[]){v.len});
1966 break;
1967
1968 case H1_MSG_TRAILERS:
1969 if (type == HTX_BLK_EOM)
1970 goto done;
1971 else if (type != HTX_BLK_TLR && type != HTX_BLK_EOT)
1972 goto error;
1973 trailers:
1974 h1m->state = H1_MSG_TRAILERS;
1975 /* If the message is not chunked, ignore
1976 * trailers. It may happen with H2 messages. */
1977 if (!(h1m->flags & H1_MF_CHNK))
1978 break;
1979
1980 if (type == HTX_BLK_EOT) {
1981 if (!chunk_memcat(&tmp, "\r\n", 2))
1982 goto full;
1983 TRACE_PROTO((!(h1m->flags & H1_MF_RESP) ? "H1 request trailers xferred" : "H1 response trailers xferred"),
1984 H1_EV_TX_DATA|H1_EV_TX_TLRS, h1c->conn, h1s);
1985 }
1986 else { // HTX_BLK_TLR
1987 n = htx_get_blk_name(chn_htx, blk);
1988 v = htx_get_blk_value(chn_htx, blk);
1989
1990 /* Try to adjust the case of the header name */
1991 if (h1c->px->options2 & (PR_O2_H1_ADJ_BUGCLI|PR_O2_H1_ADJ_BUGSRV))
1992 h1_adjust_case_outgoing_hdr(h1s, h1m, &n);
1993 if (!h1_format_htx_hdr(n, v, &tmp))
1994 goto full;
1995 }
1996 break;
1997
1998 case H1_MSG_DONE:
1999 if (type != HTX_BLK_EOM)
2000 goto error;
2001 done:
2002 h1m->state = H1_MSG_DONE;
2003 if (!(h1m->flags & H1_MF_RESP) && h1s->status == 101) {
2004 h1_set_req_tunnel_mode(h1s);
2005 TRACE_STATE("switch H1 request in tunnel mode", H1_EV_TX_DATA|H1_EV_TX_HDRS, h1c->conn, h1s);
2006 }
2007 else if ((h1m->flags & H1_MF_RESP) && h1s->req.state == H1_MSG_TUNNEL) {
2008 TRACE_STATE("switch back H1 request from tunnel mode", H1_EV_RX_DATA|H1_EV_H1C_BLK, h1c->conn, h1s);
2009 h1s->req.state = H1_MSG_DONE;
2010 h1s->flags |= H1S_F_PARSING_DONE;
2011 }
2012
2013 if (h1s->h1c->flags & H1C_F_IN_BUSY) {
2014 h1s->h1c->flags &= ~H1C_F_IN_BUSY;
2015 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
2016 TRACE_STATE("h1c no more busy", H1_EV_TX_DATA|H1_EV_H1C_BLK|H1_EV_H1C_WAKE, h1c->conn, h1s);
2017 }
2018
2019 TRACE_USER((!(h1m->flags & H1_MF_RESP) ? "H1 request fully xferred" : "H1 response fully xferred"),
2020 H1_EV_TX_DATA, h1c->conn, h1s);
2021 break;
2022
2023 default:
2024 error:
2025 TRACE_PROTO("formatting error", H1_EV_TX_DATA, h1c->conn, h1s);
2026 /* Unexpected error during output processing */
2027 chn_htx->flags |= HTX_FL_PROCESSING_ERROR;
2028 h1s->flags |= errflag;
2029 h1c->flags |= H1C_F_CS_ERROR;
2030 TRACE_STATE("processing error, set error on h1c/h1s", H1_EV_H1C_ERR|H1_EV_H1S_ERR, h1c->conn, h1s);
2031 TRACE_DEVEL("unexpected error", H1_EV_TX_DATA|H1_EV_STRM_ERR, h1c->conn, h1s);
2032 break;
2033 }
2034
2035 nextblk:
2036 total += vlen;
2037 count -= vlen;
2038 if (sz == vlen)
2039 blk = htx_remove_blk(chn_htx, blk);
2040 else {
2041 htx_cut_data_blk(chn_htx, blk, vlen);
2042 break;
2043 }
2044 }
2045
2046 copy:
2047 /* when the output buffer is empty, tmp shares the same area so that we
2048 * only have to update pointers and lengths.
2049 */
2050 if (tmp.area == h1c->obuf.area + h1c->obuf.head)
2051 h1c->obuf.data = tmp.data;
2052 else
2053 b_putblk(&h1c->obuf, tmp.area, tmp.data);
2054
2055 htx_to_buf(chn_htx, buf);
2056 out:
2057 /* Both the request and the response reached the DONE state. So set EOI
2058 * flag on the conn-stream. Most of time, the flag will already be set,
2059 * except for protocol upgrades.
2060 */
2061 if (h1s->cs && h1s->req.state == H1_MSG_DONE && h1s->res.state == H1_MSG_DONE)
2062 h1s->cs->flags |= CS_FL_EOI;
2063
2064 if (!buf_room_for_htx_data(&h1c->obuf)) {
2065 TRACE_STATE("h1c obuf full", H1_EV_TX_DATA|H1_EV_H1S_BLK, h1c->conn, h1s);
2066 h1c->flags |= H1C_F_OUT_FULL;
2067 }
2068 end:
2069 TRACE_LEAVE(H1_EV_TX_DATA, h1c->conn, h1s, chn_htx, (size_t[]){total});
2070 return total;
2071
2072 full:
2073 TRACE_STATE("h1c obuf full", H1_EV_TX_DATA|H1_EV_H1S_BLK, h1c->conn, h1s);
2074 h1c->flags |= H1C_F_OUT_FULL;
2075 goto copy;
2076 }
2077
2078 /*********************************************************/
2079 /* functions below are I/O callbacks from the connection */
2080 /*********************************************************/
h1_wake_stream_for_recv(struct h1s * h1s)2081 static void h1_wake_stream_for_recv(struct h1s *h1s)
2082 {
2083 if (h1s && h1s->subs && h1s->subs->events & SUB_RETRY_RECV) {
2084 TRACE_POINT(H1_EV_STRM_WAKE, h1s->h1c->conn, h1s);
2085 tasklet_wakeup(h1s->subs->tasklet);
2086 h1s->subs->events &= ~SUB_RETRY_RECV;
2087 if (!h1s->subs->events)
2088 h1s->subs = NULL;
2089 }
2090 }
h1_wake_stream_for_send(struct h1s * h1s)2091 static void h1_wake_stream_for_send(struct h1s *h1s)
2092 {
2093 if (h1s && h1s->subs && h1s->subs->events & SUB_RETRY_SEND) {
2094 TRACE_POINT(H1_EV_STRM_WAKE, h1s->h1c->conn, h1s);
2095 tasklet_wakeup(h1s->subs->tasklet);
2096 h1s->subs->events &= ~SUB_RETRY_SEND;
2097 if (!h1s->subs->events)
2098 h1s->subs = NULL;
2099 }
2100 }
2101
2102 /*
2103 * Attempt to read data, and subscribe if none available
2104 */
h1_recv(struct h1c * h1c)2105 static int h1_recv(struct h1c *h1c)
2106 {
2107 struct connection *conn = h1c->conn;
2108 struct h1s *h1s = h1c->h1s;
2109 size_t ret = 0, max;
2110 int rcvd = 0;
2111 int flags = 0;
2112
2113 TRACE_ENTER(H1_EV_H1C_RECV, h1c->conn);
2114
2115 if (h1c->wait_event.events & SUB_RETRY_RECV) {
2116 TRACE_DEVEL("leaving on sub_recv", H1_EV_H1C_RECV, h1c->conn);
2117 return (b_data(&h1c->ibuf));
2118 }
2119
2120 if (!h1_recv_allowed(h1c)) {
2121 TRACE_DEVEL("leaving on !recv_allowed", H1_EV_H1C_RECV, h1c->conn);
2122 rcvd = 1;
2123 goto end;
2124 }
2125
2126 if (!h1_get_buf(h1c, &h1c->ibuf)) {
2127 h1c->flags |= H1C_F_IN_ALLOC;
2128 TRACE_STATE("waiting for h1c ibuf allocation", H1_EV_H1C_RECV|H1_EV_H1C_BLK, h1c->conn);
2129 goto end;
2130 }
2131
2132 if (h1s && (h1s->flags & (H1S_F_BUF_FLUSH|H1S_F_SPLICED_DATA))) {
2133 if (!h1s_data_pending(h1s))
2134 h1_wake_stream_for_recv(h1s);
2135 rcvd = 1;
2136 TRACE_DEVEL("leaving on (buf_flush|spliced_data)", H1_EV_H1C_RECV, h1c->conn);
2137 goto end;
2138 }
2139
2140 /*
2141 * If we only have a small amount of data, realign it,
2142 * it's probably cheaper than doing 2 recv() calls.
2143 */
2144 if (b_data(&h1c->ibuf) > 0 && b_data(&h1c->ibuf) < 128)
2145 b_slow_realign(&h1c->ibuf, trash.area, 0);
2146
2147 /* avoid useless reads after first responses */
2148 if (h1s && ((!conn_is_back(conn) && h1s->req.state == H1_MSG_RQBEFORE) ||
2149 (conn_is_back(conn) && h1s->res.state == H1_MSG_RPBEFORE)))
2150 flags |= CO_RFL_READ_ONCE;
2151
2152 max = buf_room_for_htx_data(&h1c->ibuf);
2153 if (max) {
2154 if (h1c->flags & H1C_F_IN_FULL) {
2155 h1c->flags &= ~H1C_F_IN_FULL;
2156 TRACE_STATE("h1c ibuf not full anymore", H1_EV_H1C_RECV|H1_EV_H1C_BLK);
2157 }
2158
2159 b_realign_if_empty(&h1c->ibuf);
2160 if (!b_data(&h1c->ibuf)) {
2161 /* try to pre-align the buffer like the rxbufs will be
2162 * to optimize memory copies.
2163 */
2164 h1c->ibuf.head = sizeof(struct htx);
2165 }
2166 ret = conn->xprt->rcv_buf(conn, conn->xprt_ctx, &h1c->ibuf, max, flags);
2167 }
2168 if (ret > 0) {
2169 TRACE_DATA("data received", H1_EV_H1C_RECV, h1c->conn,,, (size_t[]){ret});
2170 rcvd = 1;
2171 if (h1s && h1s->cs) {
2172 h1s->cs->flags |= (CS_FL_READ_PARTIAL|CS_FL_RCV_MORE);
2173 if (h1s->csinfo.t_idle == -1)
2174 h1s->csinfo.t_idle = tv_ms_elapsed(&h1s->csinfo.tv_create, &now) - h1s->csinfo.t_handshake;
2175 }
2176 }
2177
2178 if (ret > 0 || !h1_recv_allowed(h1c) || !buf_room_for_htx_data(&h1c->ibuf)) {
2179 rcvd = 1;
2180 goto end;
2181 }
2182
2183 TRACE_STATE("failed to receive data, subscribing", H1_EV_H1C_RECV, h1c->conn);
2184 conn->xprt->subscribe(conn, conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
2185
2186 end:
2187 if (ret > 0 || (conn->flags & CO_FL_ERROR) || conn_xprt_read0_pending(conn))
2188 h1_wake_stream_for_recv(h1s);
2189
2190 if (conn_xprt_read0_pending(conn) && h1s) {
2191 h1s->flags |= H1S_F_REOS;
2192 TRACE_STATE("read0 on connection", H1_EV_H1C_RECV, conn, h1s);
2193 rcvd = 1;
2194 }
2195
2196 if (!b_data(&h1c->ibuf))
2197 h1_release_buf(h1c, &h1c->ibuf);
2198 else if (!buf_room_for_htx_data(&h1c->ibuf)) {
2199 h1c->flags |= H1C_F_IN_FULL;
2200 TRACE_STATE("h1c ibuf full", H1_EV_H1C_RECV|H1_EV_H1C_BLK);
2201 }
2202
2203 TRACE_LEAVE(H1_EV_H1C_RECV, h1c->conn);
2204 return rcvd;
2205 }
2206
2207
2208 /*
2209 * Try to send data if possible
2210 */
h1_send(struct h1c * h1c)2211 static int h1_send(struct h1c *h1c)
2212 {
2213 struct connection *conn = h1c->conn;
2214 unsigned int flags = 0;
2215 size_t ret;
2216 int sent = 0;
2217
2218 TRACE_ENTER(H1_EV_H1C_SEND, h1c->conn);
2219
2220 if (conn->flags & CO_FL_ERROR) {
2221 TRACE_DEVEL("leaving on connection error", H1_EV_H1C_SEND, h1c->conn);
2222 return 0;
2223 }
2224
2225 if (!b_data(&h1c->obuf))
2226 goto end;
2227
2228 if (h1c->flags & H1C_F_CO_MSG_MORE)
2229 flags |= CO_SFL_MSG_MORE;
2230 if (h1c->flags & H1C_F_CO_STREAMER)
2231 flags |= CO_SFL_STREAMER;
2232
2233 ret = conn->xprt->snd_buf(conn, conn->xprt_ctx, &h1c->obuf, b_data(&h1c->obuf), flags);
2234 if (ret > 0) {
2235 TRACE_DATA("data sent", H1_EV_H1C_SEND, h1c->conn,,, (size_t[]){ret});
2236 if (h1c->flags & H1C_F_OUT_FULL) {
2237 h1c->flags &= ~H1C_F_OUT_FULL;
2238 TRACE_STATE("h1c obuf not full anymore", H1_EV_STRM_SEND|H1_EV_H1S_BLK, h1c->conn);
2239 }
2240 b_del(&h1c->obuf, ret);
2241 sent = 1;
2242 }
2243
2244 if (conn->flags & (CO_FL_ERROR|CO_FL_SOCK_WR_SH)) {
2245 TRACE_DEVEL("connection error or output closed", H1_EV_H1C_SEND, h1c->conn);
2246 /* error or output closed, nothing to send, clear the buffer to release it */
2247 b_reset(&h1c->obuf);
2248 }
2249
2250 end:
2251 if (!(h1c->flags & H1C_F_OUT_FULL))
2252 h1_wake_stream_for_send(h1c->h1s);
2253
2254 /* We're done, no more to send */
2255 if (!b_data(&h1c->obuf)) {
2256 TRACE_DEVEL("leaving with everything sent", H1_EV_H1C_SEND, h1c->conn);
2257 h1_release_buf(h1c, &h1c->obuf);
2258 if (h1c->flags & H1C_F_CS_SHUTW_NOW) {
2259 TRACE_STATE("process pending shutdown for writes", H1_EV_H1C_SEND, h1c->conn);
2260 h1_shutw_conn(conn);
2261 }
2262 }
2263 else if (!(h1c->wait_event.events & SUB_RETRY_SEND)) {
2264 TRACE_STATE("more data to send, subscribing", H1_EV_H1C_SEND, h1c->conn);
2265 conn->xprt->subscribe(conn, conn->xprt_ctx, SUB_RETRY_SEND, &h1c->wait_event);
2266 }
2267
2268 TRACE_LEAVE(H1_EV_H1C_SEND, h1c->conn);
2269 return sent;
2270 }
2271
2272
2273 /* callback called on any event by the connection handler.
2274 * It applies changes and returns zero, or < 0 if it wants immediate
2275 * destruction of the connection.
2276 */
h1_process(struct h1c * h1c)2277 static int h1_process(struct h1c * h1c)
2278 {
2279 struct connection *conn = h1c->conn;
2280 struct h1s *h1s = h1c->h1s;
2281
2282 TRACE_ENTER(H1_EV_H1C_WAKE, conn);
2283
2284 if (!conn->ctx)
2285 return -1;
2286
2287 if (!h1s) {
2288 if (h1c->flags & (H1C_F_CS_ERROR|H1C_F_CS_SHUTDOWN) ||
2289 conn->flags & (CO_FL_ERROR|CO_FL_SOCK_RD_SH|CO_FL_SOCK_WR_SH))
2290 goto release;
2291 if (!conn_is_back(conn) && (h1c->flags & H1C_F_CS_IDLE)) {
2292 TRACE_STATE("K/A incoming connection, create new H1 stream", H1_EV_H1C_WAKE, conn);
2293 if (!h1s_create(h1c, NULL, NULL))
2294 goto release;
2295 }
2296 else if (conn_is_back(conn) && (h1c->flags & H1C_F_CS_IDLE) && b_data(&h1c->ibuf))
2297 goto release;
2298 else
2299 goto end;
2300 h1s = h1c->h1s;
2301 }
2302
2303 if (b_data(&h1c->ibuf) && h1s->csinfo.t_idle == -1)
2304 h1s->csinfo.t_idle = tv_ms_elapsed(&h1s->csinfo.tv_create, &now) - h1s->csinfo.t_handshake;
2305
2306 if (conn_xprt_read0_pending(conn)) {
2307 h1s->flags |= H1S_F_REOS;
2308 TRACE_STATE("read0 on connection", H1_EV_H1C_RECV, conn, h1s);
2309 }
2310
2311 if (!h1s_data_pending(h1s) && h1s && h1s->cs && h1s->cs->data_cb->wake &&
2312 (h1s->flags & H1S_F_REOS || h1c->flags & H1C_F_CS_ERROR ||
2313 conn->flags & (CO_FL_ERROR | CO_FL_SOCK_WR_SH))) {
2314 if (h1c->flags & H1C_F_CS_ERROR || conn->flags & CO_FL_ERROR)
2315 h1s->cs->flags |= CS_FL_ERROR;
2316 TRACE_POINT(H1_EV_STRM_WAKE, h1c->conn, h1s);
2317 h1s->cs->data_cb->wake(h1s->cs);
2318 }
2319 end:
2320 h1_refresh_timeout(h1c);
2321 TRACE_LEAVE(H1_EV_H1C_WAKE, conn);
2322 return 0;
2323
2324 release:
2325 h1_release(h1c);
2326 TRACE_DEVEL("leaving after releasing the connection", H1_EV_H1C_WAKE);
2327 return -1;
2328 }
2329
h1_io_cb(struct task * t,void * ctx,unsigned short status)2330 struct task *h1_io_cb(struct task *t, void *ctx, unsigned short status)
2331 {
2332 struct connection *conn;
2333 struct tasklet *tl = (struct tasklet *)t;
2334 int conn_in_list;
2335 struct h1c *h1c;
2336 int ret = 0;
2337
2338
2339 HA_SPIN_LOCK(OTHER_LOCK, &idle_conns[tid].takeover_lock);
2340 if (tl->context == NULL) {
2341 /* The connection has been taken over by another thread,
2342 * we're no longer responsible for it, so just free the
2343 * tasklet, and do nothing.
2344 */
2345 HA_SPIN_UNLOCK(OTHER_LOCK, &idle_conns[tid].takeover_lock);
2346 tasklet_free(tl);
2347 return NULL;
2348 }
2349 h1c = ctx;
2350 conn = h1c->conn;
2351
2352 TRACE_POINT(H1_EV_H1C_WAKE, conn);
2353
2354 /* Remove the connection from the list, to be sure nobody attempts
2355 * to use it while we handle the I/O events
2356 */
2357 conn_in_list = conn->flags & CO_FL_LIST_MASK;
2358 if (conn_in_list)
2359 MT_LIST_DEL(&conn->list);
2360
2361 HA_SPIN_UNLOCK(OTHER_LOCK, &idle_conns[tid].takeover_lock);
2362
2363 if (!(h1c->wait_event.events & SUB_RETRY_SEND))
2364 ret = h1_send(h1c);
2365 if (!(h1c->wait_event.events & SUB_RETRY_RECV))
2366 ret |= h1_recv(h1c);
2367 if (ret || !h1c->h1s)
2368 ret = h1_process(h1c);
2369 /* If we were in an idle list, we want to add it back into it,
2370 * unless h1_process() returned -1, which mean it has destroyed
2371 * the connection (testing !ret is enough, if h1_process() wasn't
2372 * called then ret will be 0 anyway.
2373 */
2374 if (!ret && conn_in_list) {
2375 struct server *srv = objt_server(conn->target);
2376
2377 if (conn_in_list == CO_FL_SAFE_LIST)
2378 MT_LIST_ADDQ(&srv->safe_conns[tid], &conn->list);
2379 else
2380 MT_LIST_ADDQ(&srv->idle_conns[tid], &conn->list);
2381 }
2382 return NULL;
2383 }
2384
h1_reset(struct connection * conn)2385 static void h1_reset(struct connection *conn)
2386 {
2387
2388 }
2389
h1_wake(struct connection * conn)2390 static int h1_wake(struct connection *conn)
2391 {
2392 struct h1c *h1c = conn->ctx;
2393 int ret;
2394
2395 TRACE_POINT(H1_EV_H1C_WAKE, conn);
2396
2397 h1_send(h1c);
2398 ret = h1_process(h1c);
2399 if (ret == 0) {
2400 struct h1s *h1s = h1c->h1s;
2401
2402 if (h1s && h1s->cs && h1s->cs->data_cb->wake) {
2403 TRACE_POINT(H1_EV_STRM_WAKE, h1c->conn, h1s);
2404 ret = h1s->cs->data_cb->wake(h1s->cs);
2405 }
2406 }
2407 return ret;
2408 }
2409
2410 /* Connection timeout management. The principle is that if there's no receipt
2411 * nor sending for a certain amount of time, the connection is closed.
2412 */
h1_timeout_task(struct task * t,void * context,unsigned short state)2413 static struct task *h1_timeout_task(struct task *t, void *context, unsigned short state)
2414 {
2415 struct h1c *h1c = context;
2416 int expired = tick_is_expired(t->expire, now_ms);
2417
2418 TRACE_POINT(H1_EV_H1C_WAKE, h1c ? h1c->conn : NULL);
2419
2420 if (h1c) {
2421 if (!expired) {
2422 TRACE_DEVEL("leaving (not expired)", H1_EV_H1C_WAKE, h1c->conn);
2423 return t;
2424 }
2425
2426 /* We're about to destroy the connection, so make sure nobody attempts
2427 * to steal it from us.
2428 */
2429 HA_SPIN_LOCK(OTHER_LOCK, &idle_conns[tid].takeover_lock);
2430
2431 /* Somebody already stole the connection from us, so we should not
2432 * free it, we just have to free the task.
2433 */
2434 if (!t->context)
2435 h1c = NULL;
2436 else if (h1c->conn->flags & CO_FL_LIST_MASK)
2437 MT_LIST_DEL(&h1c->conn->list);
2438
2439 HA_SPIN_UNLOCK(OTHER_LOCK, &idle_conns[tid].takeover_lock);
2440 }
2441
2442 task_destroy(t);
2443
2444 if (!h1c) {
2445 /* resources were already deleted */
2446 TRACE_DEVEL("leaving (not more h1c)", H1_EV_H1C_WAKE);
2447 return NULL;
2448 }
2449
2450 h1c->task = NULL;
2451 /* If a stream is still attached to the mux, just set an error and wait
2452 * for the stream's timeout. Otherwise, release the mux. This is only ok
2453 * because same timeouts are used.
2454 */
2455 if (h1c->h1s && h1c->h1s->cs) {
2456 h1c->flags |= H1C_F_CS_ERROR;
2457 TRACE_STATE("error on h1c, h1s still attached (expired)", H1_EV_H1C_WAKE|H1_EV_H1C_ERR, h1c->conn, h1c->h1s);
2458 }
2459 else
2460 h1_release(h1c);
2461
2462 return NULL;
2463 }
2464
2465 /*******************************************/
2466 /* functions below are used by the streams */
2467 /*******************************************/
2468
2469 /*
2470 * Attach a new stream to a connection
2471 * (Used for outgoing connections)
2472 */
h1_attach(struct connection * conn,struct session * sess)2473 static struct conn_stream *h1_attach(struct connection *conn, struct session *sess)
2474 {
2475 struct h1c *h1c = conn->ctx;
2476 struct conn_stream *cs = NULL;
2477 struct h1s *h1s;
2478
2479 TRACE_ENTER(H1_EV_STRM_NEW, conn);
2480 if (h1c->flags & H1C_F_CS_ERROR) {
2481 TRACE_DEVEL("leaving on h1c error", H1_EV_STRM_NEW|H1_EV_STRM_END|H1_EV_STRM_ERR, conn);
2482 goto end;
2483 }
2484
2485 cs = cs_new(h1c->conn);
2486 if (!cs) {
2487 TRACE_DEVEL("leaving on CS allocation failure", H1_EV_STRM_NEW|H1_EV_STRM_END|H1_EV_STRM_ERR, conn);
2488 goto end;
2489 }
2490
2491 h1s = h1s_create(h1c, cs, sess);
2492 if (h1s == NULL) {
2493 TRACE_DEVEL("leaving on h1s creation failure", H1_EV_STRM_NEW|H1_EV_STRM_END|H1_EV_STRM_ERR, conn);
2494 goto end;
2495 }
2496
2497 TRACE_LEAVE(H1_EV_STRM_NEW, conn, h1s);
2498 return cs;
2499 end:
2500 cs_free(cs);
2501 return NULL;
2502 }
2503
2504 /* Retrieves a valid conn_stream from this connection, or returns NULL. For
2505 * this mux, it's easy as we can only store a single conn_stream.
2506 */
h1_get_first_cs(const struct connection * conn)2507 static const struct conn_stream *h1_get_first_cs(const struct connection *conn)
2508 {
2509 struct h1c *h1c = conn->ctx;
2510 struct h1s *h1s = h1c->h1s;
2511
2512 if (h1s)
2513 return h1s->cs;
2514
2515 return NULL;
2516 }
2517
h1_destroy(void * ctx)2518 static void h1_destroy(void *ctx)
2519 {
2520 struct h1c *h1c = ctx;
2521
2522 TRACE_POINT(H1_EV_H1C_END, h1c->conn);
2523 if (!h1c->h1s || !h1c->conn || h1c->conn->ctx != h1c)
2524 h1_release(h1c);
2525 }
2526
2527 /*
2528 * Detach the stream from the connection and possibly release the connection.
2529 */
h1_detach(struct conn_stream * cs)2530 static void h1_detach(struct conn_stream *cs)
2531 {
2532 struct h1s *h1s = cs->ctx;
2533 struct h1c *h1c;
2534 struct session *sess;
2535 int is_not_first;
2536
2537 TRACE_ENTER(H1_EV_STRM_END, h1s ? h1s->h1c->conn : NULL, h1s);
2538
2539 cs->ctx = NULL;
2540 if (!h1s) {
2541 TRACE_LEAVE(H1_EV_STRM_END);
2542 return;
2543 }
2544
2545 sess = h1s->sess;
2546 h1c = h1s->h1c;
2547 h1s->cs = NULL;
2548
2549 is_not_first = h1s->flags & H1S_F_NOT_FIRST;
2550 h1s_destroy(h1s);
2551
2552 if (conn_is_back(h1c->conn) && (h1c->flags & H1C_F_CS_IDLE)) {
2553 /* If there are any excess server data in the input buffer,
2554 * release it and close the connection ASAP (some data may
2555 * remain in the output buffer). This happens if a server sends
2556 * invalid responses. So in such case, we don't want to reuse
2557 * the connection
2558 */
2559 if (b_data(&h1c->ibuf)) {
2560 h1_release_buf(h1c, &h1c->ibuf);
2561 h1c->flags = (h1c->flags & ~H1C_F_CS_IDLE) | H1C_F_CS_SHUTW_NOW;
2562 TRACE_DEVEL("remaining data on detach, kill connection", H1_EV_STRM_END|H1_EV_H1C_END);
2563 goto release;
2564 }
2565
2566 /* Never ever allow to reuse a connection from a non-reuse backend */
2567 if ((h1c->px->options & PR_O_REUSE_MASK) == PR_O_REUSE_NEVR)
2568 h1c->conn->flags |= CO_FL_PRIVATE;
2569
2570 if (!(h1c->conn->owner) && (h1c->conn->flags & CO_FL_PRIVATE)) {
2571 h1c->conn->owner = sess;
2572 if (!session_add_conn(sess, h1c->conn, h1c->conn->target)) {
2573 h1c->conn->owner = NULL;
2574 h1c->conn->mux->destroy(h1c);
2575 goto end;
2576 }
2577 if (session_check_idle_conn(sess, h1c->conn)) {
2578 /* The connection got destroyed, let's leave */
2579 TRACE_DEVEL("outgoing connection killed", H1_EV_STRM_END|H1_EV_H1C_END);
2580 goto end;
2581 }
2582 }
2583 if (!(h1c->conn->flags & CO_FL_PRIVATE)) {
2584 if (h1c->conn->owner == sess)
2585 h1c->conn->owner = NULL;
2586 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
2587 if (!srv_add_to_idle_list(objt_server(h1c->conn->target), h1c->conn, is_not_first)) {
2588 /* The server doesn't want it, let's kill the connection right away */
2589 h1c->conn->mux->destroy(h1c);
2590 TRACE_DEVEL("outgoing connection killed", H1_EV_STRM_END|H1_EV_H1C_END);
2591 goto end;
2592 }
2593 /* At this point, the connection has been added to the
2594 * server idle list, so another thread may already have
2595 * hijacked it, so we can't do anything with it.
2596 */
2597 return;
2598 }
2599 }
2600
2601 release:
2602 /* We don't want to close right now unless the connection is in error or shut down for writes */
2603 if ((h1c->flags & (H1C_F_CS_ERROR|H1C_F_CS_SHUTDOWN|H1C_F_UPG_H2C)) ||
2604 (h1c->conn->flags & (CO_FL_ERROR|CO_FL_SOCK_WR_SH)) ||
2605 ((h1c->flags & H1C_F_CS_SHUTW_NOW) && !b_data(&h1c->obuf)) ||
2606 !h1c->conn->owner) {
2607 TRACE_DEVEL("killing dead connection", H1_EV_STRM_END, h1c->conn);
2608 h1_release(h1c);
2609 }
2610 else {
2611 /* If we have a new request, process it immediately */
2612 if (unlikely(b_data(&h1c->ibuf))) {
2613 if (h1_process(h1c) == -1)
2614 goto end;
2615 }
2616 else
2617 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
2618 h1_refresh_timeout(h1c);
2619 }
2620 end:
2621 TRACE_LEAVE(H1_EV_STRM_END);
2622 }
2623
2624
h1_shutr(struct conn_stream * cs,enum cs_shr_mode mode)2625 static void h1_shutr(struct conn_stream *cs, enum cs_shr_mode mode)
2626 {
2627 struct h1s *h1s = cs->ctx;
2628 struct h1c *h1c;
2629
2630 if (!h1s)
2631 return;
2632 h1c = h1s->h1c;
2633
2634 TRACE_ENTER(H1_EV_STRM_SHUT, h1c->conn, h1s);
2635
2636 if (cs->flags & CS_FL_KILL_CONN) {
2637 TRACE_STATE("stream wants to kill the connection", H1_EV_STRM_SHUT, h1c->conn, h1s);
2638 goto do_shutr;
2639 }
2640 if (h1c->conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH)) {
2641 TRACE_STATE("shutdown on connection (error|rd_sh|wr_sh)", H1_EV_STRM_SHUT, h1c->conn, h1s);
2642 goto do_shutr;
2643 }
2644
2645 if ((h1c->flags & H1C_F_UPG_H2C) || (h1s->flags & H1S_F_WANT_KAL)) {
2646 TRACE_STATE("keep connection alive (upg_h2c|want_kal)", H1_EV_STRM_SHUT, h1c->conn, h1s);
2647 goto end;
2648 }
2649
2650 do_shutr:
2651 /* NOTE: Be sure to handle abort (cf. h2_shutr) */
2652 if (cs->flags & CS_FL_SHR)
2653 goto end;
2654 if (conn_xprt_ready(cs->conn) && cs->conn->xprt->shutr)
2655 cs->conn->xprt->shutr(cs->conn, cs->conn->xprt_ctx,
2656 (mode == CS_SHR_DRAIN));
2657 end:
2658 TRACE_LEAVE(H1_EV_STRM_SHUT, h1c->conn, h1s);
2659 }
2660
h1_shutw(struct conn_stream * cs,enum cs_shw_mode mode)2661 static void h1_shutw(struct conn_stream *cs, enum cs_shw_mode mode)
2662 {
2663 struct h1s *h1s = cs->ctx;
2664 struct h1c *h1c;
2665
2666 if (!h1s)
2667 return;
2668 h1c = h1s->h1c;
2669
2670 TRACE_ENTER(H1_EV_STRM_SHUT, h1c->conn, h1s);
2671
2672 if (cs->flags & CS_FL_KILL_CONN) {
2673 TRACE_STATE("stream wants to kill the connection", H1_EV_STRM_SHUT, h1c->conn, h1s);
2674 goto do_shutw;
2675 }
2676 if (h1c->conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH)) {
2677 TRACE_STATE("shutdown on connection (error|rd_sh|wr_sh)", H1_EV_STRM_SHUT, h1c->conn, h1s);
2678 goto do_shutw;
2679 }
2680
2681 if ((h1c->flags & H1C_F_UPG_H2C) ||
2682 ((h1s->flags & H1S_F_WANT_KAL) && h1s->req.state == H1_MSG_DONE && h1s->res.state == H1_MSG_DONE)) {
2683 TRACE_STATE("keep connection alive (upg_h2c|want_kal)", H1_EV_STRM_SHUT, h1c->conn, h1s);
2684 goto end;
2685 }
2686
2687 do_shutw:
2688 h1c->flags |= H1C_F_CS_SHUTW_NOW;
2689 if (mode != CS_SHW_NORMAL)
2690 h1c->flags |= H1C_F_ST_SILENT_SHUT;
2691 if ((cs->flags & CS_FL_SHW) || b_data(&h1c->obuf))
2692 goto end;
2693 h1_shutw_conn(cs->conn);
2694 end:
2695 TRACE_LEAVE(H1_EV_STRM_SHUT, h1c->conn, h1s);
2696 }
2697
h1_shutw_conn(struct connection * conn)2698 static void h1_shutw_conn(struct connection *conn)
2699 {
2700 struct h1c *h1c = conn->ctx;
2701
2702 if (conn->flags & CO_FL_SOCK_WR_SH)
2703 return;
2704
2705 TRACE_ENTER(H1_EV_H1C_END, conn);
2706 conn_xprt_shutw(conn);
2707 conn_sock_shutw(conn, (h1c && !(h1c->flags & H1C_F_ST_SILENT_SHUT)));
2708 h1c->flags = (h1c->flags & ~H1C_F_CS_SHUTW_NOW) | H1C_F_CS_SHUTDOWN;
2709 TRACE_LEAVE(H1_EV_STRM_SHUT, conn);
2710 }
2711
2712 /* Called from the upper layer, to unsubscribe <es> from events <event_type>
2713 * The <es> pointer is not allowed to differ from the one passed to the
2714 * subscribe() call. It always returns zero.
2715 */
h1_unsubscribe(struct conn_stream * cs,int event_type,struct wait_event * es)2716 static int h1_unsubscribe(struct conn_stream *cs, int event_type, struct wait_event *es)
2717 {
2718 struct h1s *h1s = cs->ctx;
2719
2720 if (!h1s)
2721 return 0;
2722
2723 BUG_ON(event_type & ~(SUB_RETRY_SEND|SUB_RETRY_RECV));
2724 BUG_ON(h1s->subs && h1s->subs != es);
2725
2726 es->events &= ~event_type;
2727 if (!es->events)
2728 h1s->subs = NULL;
2729
2730 if (event_type & SUB_RETRY_RECV)
2731 TRACE_DEVEL("unsubscribe(recv)", H1_EV_STRM_RECV, h1s->h1c->conn, h1s);
2732
2733 if (event_type & SUB_RETRY_SEND)
2734 TRACE_DEVEL("unsubscribe(send)", H1_EV_STRM_SEND, h1s->h1c->conn, h1s);
2735
2736 return 0;
2737 }
2738
2739 /* Called from the upper layer, to subscribe <es> to events <event_type>. The
2740 * event subscriber <es> is not allowed to change from a previous call as long
2741 * as at least one event is still subscribed. The <event_type> must only be a
2742 * combination of SUB_RETRY_RECV and SUB_RETRY_SEND. It always returns 0, unless
2743 * the conn_stream <cs> was already detached, in which case it will return -1.
2744 */
h1_subscribe(struct conn_stream * cs,int event_type,struct wait_event * es)2745 static int h1_subscribe(struct conn_stream *cs, int event_type, struct wait_event *es)
2746 {
2747 struct h1s *h1s = cs->ctx;
2748 struct h1c *h1c;
2749
2750 if (!h1s)
2751 return -1;
2752
2753 BUG_ON(event_type & ~(SUB_RETRY_SEND|SUB_RETRY_RECV));
2754 BUG_ON(h1s->subs && h1s->subs != es);
2755
2756 es->events |= event_type;
2757 h1s->subs = es;
2758
2759 if (event_type & SUB_RETRY_RECV)
2760 TRACE_DEVEL("subscribe(recv)", H1_EV_STRM_RECV, h1s->h1c->conn, h1s);
2761
2762
2763 if (event_type & SUB_RETRY_SEND) {
2764 TRACE_DEVEL("subscribe(send)", H1_EV_STRM_SEND, h1s->h1c->conn, h1s);
2765 /*
2766 * If the conn_stream attempt to subscribe, and the
2767 * mux isn't subscribed to the connection, then it
2768 * probably means the connection wasn't established
2769 * yet, so we have to subscribe.
2770 */
2771 h1c = h1s->h1c;
2772 if (!(h1c->wait_event.events & SUB_RETRY_SEND))
2773 h1c->conn->xprt->subscribe(h1c->conn,
2774 h1c->conn->xprt_ctx,
2775 SUB_RETRY_SEND,
2776 &h1c->wait_event);
2777 }
2778 return 0;
2779 }
2780
2781 /* Called from the upper layer, to receive data */
h1_rcv_buf(struct conn_stream * cs,struct buffer * buf,size_t count,int flags)2782 static size_t h1_rcv_buf(struct conn_stream *cs, struct buffer *buf, size_t count, int flags)
2783 {
2784 struct h1s *h1s = cs->ctx;
2785 struct h1c *h1c = h1s->h1c;
2786 struct h1m *h1m = (!conn_is_back(cs->conn) ? &h1s->req : &h1s->res);
2787 size_t ret = 0;
2788
2789 TRACE_ENTER(H1_EV_STRM_RECV, h1c->conn, h1s,, (size_t[]){count});
2790 if (!(h1c->flags & H1C_F_IN_ALLOC))
2791 ret = h1_process_input(h1c, buf, count);
2792 else
2793 TRACE_DEVEL("h1c ibuf not allocated", H1_EV_H1C_RECV|H1_EV_H1C_BLK, h1c->conn);
2794
2795 if (flags & CO_RFL_BUF_FLUSH) {
2796 if (h1m->state == H1_MSG_TUNNEL || (h1m->state == H1_MSG_DATA && h1m->curr_len)) {
2797 h1s->flags |= H1S_F_BUF_FLUSH;
2798 TRACE_STATE("flush stream's buffer", H1_EV_STRM_RECV, h1c->conn, h1s);
2799 }
2800 }
2801 else {
2802 if (ret && h1s->flags & H1S_F_SPLICED_DATA) {
2803 h1s->flags &= ~H1S_F_SPLICED_DATA;
2804 TRACE_STATE("disable splicing", H1_EV_STRM_RECV, h1c->conn, h1s);
2805 }
2806 if (((flags & CO_RFL_KEEP_RECV) || (h1m->state != H1_MSG_DONE)) && !(h1c->wait_event.events & SUB_RETRY_RECV))
2807 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx, SUB_RETRY_RECV, &h1c->wait_event);
2808 }
2809 TRACE_LEAVE(H1_EV_STRM_RECV, h1c->conn, h1s,, (size_t[]){ret});
2810 return ret;
2811 }
2812
2813
2814 /* Called from the upper layer, to send data */
h1_snd_buf(struct conn_stream * cs,struct buffer * buf,size_t count,int flags)2815 static size_t h1_snd_buf(struct conn_stream *cs, struct buffer *buf, size_t count, int flags)
2816 {
2817 struct h1s *h1s = cs->ctx;
2818 struct h1c *h1c;
2819 size_t total = 0;
2820
2821 if (!h1s)
2822 return 0;
2823 h1c = h1s->h1c;
2824
2825 TRACE_ENTER(H1_EV_STRM_SEND, h1c->conn, h1s,, (size_t[]){count});
2826
2827 /* If we're not connected yet, or we're waiting for a handshake, stop
2828 * now, as we don't want to remove everything from the channel buffer
2829 * before we're sure we can send it.
2830 */
2831 if (h1c->conn->flags & CO_FL_WAIT_XPRT) {
2832 TRACE_LEAVE(H1_EV_STRM_SEND, h1c->conn, h1s);
2833 return 0;
2834 }
2835
2836 if (h1c->flags & H1C_F_CS_ERROR) {
2837 cs->flags |= CS_FL_ERROR;
2838 TRACE_DEVEL("H1 connection is in error, leaving in error", H1_EV_STRM_SEND|H1_EV_H1C_ERR|H1_EV_H1S_ERR|H1_EV_STRM_ERR, h1c->conn, h1s);
2839 return 0;
2840 }
2841
2842 /* Inherit some flags from the upper layer */
2843 h1c->flags &= ~(H1C_F_CO_MSG_MORE|H1C_F_CO_STREAMER);
2844 if (flags & CO_SFL_MSG_MORE)
2845 h1c->flags |= H1C_F_CO_MSG_MORE;
2846 if (flags & CO_SFL_STREAMER)
2847 h1c->flags |= H1C_F_CO_STREAMER;
2848
2849 while (count) {
2850 size_t ret = 0;
2851
2852 if (!(h1c->flags & (H1C_F_OUT_FULL|H1C_F_OUT_ALLOC)))
2853 ret = h1_process_output(h1c, buf, count);
2854 else
2855 TRACE_DEVEL("h1c obuf not allocated", H1_EV_STRM_SEND|H1_EV_H1S_BLK, h1c->conn, h1s);
2856
2857 if ((count - ret) > 0)
2858 h1c->flags |= H1C_F_CO_MSG_MORE;
2859
2860 if (!ret)
2861 break;
2862 total += ret;
2863 count -= ret;
2864 if ((h1c->wait_event.events & SUB_RETRY_SEND) || !h1_send(h1c))
2865 break;
2866 }
2867
2868 if (h1c->flags & H1C_F_CS_ERROR) {
2869 TRACE_DEVEL("reporting error to the app-layer stream", H1_EV_STRM_SEND|H1_EV_H1S_ERR|H1_EV_STRM_ERR, h1c->conn, h1s);
2870 cs->flags |= CS_FL_ERROR;
2871 }
2872
2873 h1_refresh_timeout(h1c);
2874 TRACE_LEAVE(H1_EV_STRM_SEND, h1c->conn, h1s,, (size_t[]){total});
2875 return total;
2876 }
2877
2878 #if defined(USE_LINUX_SPLICE)
2879 /* Send and get, using splicing */
h1_rcv_pipe(struct conn_stream * cs,struct pipe * pipe,unsigned int count)2880 static int h1_rcv_pipe(struct conn_stream *cs, struct pipe *pipe, unsigned int count)
2881 {
2882 struct h1s *h1s = cs->ctx;
2883 struct h1m *h1m = (!conn_is_back(cs->conn) ? &h1s->req : &h1s->res);
2884 int ret = 0;
2885
2886 TRACE_ENTER(H1_EV_STRM_RECV, cs->conn, h1s,, (size_t[]){count});
2887
2888 if ((h1m->flags & H1_MF_CHNK) || (h1m->state != H1_MSG_DATA && h1m->state != H1_MSG_TUNNEL)) {
2889 h1s->flags &= ~(H1S_F_BUF_FLUSH|H1S_F_SPLICED_DATA);
2890 TRACE_STATE("disable splicing on !(msg_data|msg_tunnel)", H1_EV_STRM_RECV, cs->conn, h1s);
2891 if (!(h1s->h1c->wait_event.events & SUB_RETRY_RECV)) {
2892 TRACE_STATE("restart receiving data, subscribing", H1_EV_STRM_RECV, cs->conn, h1s);
2893 cs->conn->xprt->subscribe(cs->conn, cs->conn->xprt_ctx, SUB_RETRY_RECV, &h1s->h1c->wait_event);
2894 }
2895 goto end;
2896 }
2897
2898 if (h1s_data_pending(h1s)) {
2899 h1s->flags |= H1S_F_BUF_FLUSH;
2900 TRACE_STATE("flush input buffer before splicing", H1_EV_STRM_RECV, cs->conn, h1s);
2901 goto end;
2902 }
2903
2904 if (!(h1s->flags & H1S_F_SPLICED_DATA)) {
2905 h1s->flags &= ~H1S_F_BUF_FLUSH;
2906 h1s->flags |= H1S_F_SPLICED_DATA;
2907 TRACE_STATE("enable splicing", H1_EV_STRM_RECV, cs->conn, h1s);
2908 }
2909
2910 if (!h1_recv_allowed(h1s->h1c)) {
2911 TRACE_DEVEL("leaving on !recv_allowed", H1_EV_STRM_RECV, cs->conn, h1s);
2912 goto end;
2913 }
2914
2915 if (h1m->state == H1_MSG_DATA && count > h1m->curr_len)
2916 count = h1m->curr_len;
2917 ret = cs->conn->xprt->rcv_pipe(cs->conn, cs->conn->xprt_ctx, pipe, count);
2918 if (h1m->state == H1_MSG_DATA && ret >= 0) {
2919 h1m->curr_len -= ret;
2920 if (!h1m->curr_len) {
2921 h1s->flags &= ~(H1S_F_BUF_FLUSH|H1S_F_SPLICED_DATA);
2922 TRACE_STATE("disable splicing on !curr_len", H1_EV_STRM_RECV, cs->conn, h1s);
2923 }
2924 }
2925
2926 end:
2927 if (conn_xprt_read0_pending(cs->conn)) {
2928 h1s->flags |= H1S_F_REOS;
2929 h1s->flags &= ~(H1S_F_BUF_FLUSH|H1S_F_SPLICED_DATA);
2930 TRACE_STATE("read0 on connection", H1_EV_STRM_RECV, cs->conn, h1s);
2931 }
2932
2933 if ((h1s->flags & H1S_F_REOS) ||
2934 (h1m->state != H1_MSG_TUNNEL && h1m->state != H1_MSG_DATA) ||
2935 (h1m->state == H1_MSG_DATA && !h1m->curr_len)) {
2936 TRACE_STATE("notify the mux can't use splicing anymore", H1_EV_STRM_RECV, h1s->h1c->conn, h1s);
2937 cs->flags &= ~CS_FL_MAY_SPLICE;
2938 }
2939
2940 TRACE_LEAVE(H1_EV_STRM_RECV, cs->conn, h1s,, (size_t[]){ret});
2941 return ret;
2942 }
2943
h1_snd_pipe(struct conn_stream * cs,struct pipe * pipe)2944 static int h1_snd_pipe(struct conn_stream *cs, struct pipe *pipe)
2945 {
2946 struct h1s *h1s = cs->ctx;
2947 int ret = 0;
2948
2949 TRACE_ENTER(H1_EV_STRM_SEND, cs->conn, h1s,, (size_t[]){pipe->data});
2950
2951 if (b_data(&h1s->h1c->obuf))
2952 goto end;
2953
2954 ret = cs->conn->xprt->snd_pipe(cs->conn, cs->conn->xprt_ctx, pipe);
2955 end:
2956 if (pipe->data) {
2957 if (!(h1s->h1c->wait_event.events & SUB_RETRY_SEND)) {
2958 TRACE_STATE("more data to send, subscribing", H1_EV_STRM_SEND, cs->conn, h1s);
2959 cs->conn->xprt->subscribe(cs->conn, cs->conn->xprt_ctx, SUB_RETRY_SEND, &h1s->h1c->wait_event);
2960 }
2961 }
2962
2963 TRACE_LEAVE(H1_EV_STRM_SEND, cs->conn, h1s,, (size_t[]){ret});
2964 return ret;
2965 }
2966 #endif
2967
h1_ctl(struct connection * conn,enum mux_ctl_type mux_ctl,void * output)2968 static int h1_ctl(struct connection *conn, enum mux_ctl_type mux_ctl, void *output)
2969 {
2970 int ret = 0;
2971 switch (mux_ctl) {
2972 case MUX_STATUS:
2973 if (!(conn->flags & CO_FL_WAIT_XPRT))
2974 ret |= MUX_STATUS_READY;
2975 return ret;
2976 default:
2977 return -1;
2978 }
2979 }
2980
2981 /* for debugging with CLI's "show fd" command */
h1_show_fd(struct buffer * msg,struct connection * conn)2982 static int h1_show_fd(struct buffer *msg, struct connection *conn)
2983 {
2984 struct h1c *h1c = conn->ctx;
2985 struct h1s *h1s = h1c->h1s;
2986 int ret = 0;
2987
2988 chunk_appendf(msg, " h1c.flg=0x%x .sub=%d .ibuf=%u@%p+%u/%u .obuf=%u@%p+%u/%u",
2989 h1c->flags, h1c->wait_event.events,
2990 (unsigned int)b_data(&h1c->ibuf), b_orig(&h1c->ibuf),
2991 (unsigned int)b_head_ofs(&h1c->ibuf), (unsigned int)b_size(&h1c->ibuf),
2992 (unsigned int)b_data(&h1c->obuf), b_orig(&h1c->obuf),
2993 (unsigned int)b_head_ofs(&h1c->obuf), (unsigned int)b_size(&h1c->obuf));
2994
2995 if (h1s) {
2996 char *method;
2997
2998 if (h1s->meth < HTTP_METH_OTHER)
2999 method = http_known_methods[h1s->meth].ptr;
3000 else
3001 method = "UNKNOWN";
3002 chunk_appendf(msg, " h1s=%p h1s.flg=0x%x .req.state=%s .res.state=%s"
3003 " .meth=%s status=%d",
3004 h1s, h1s->flags,
3005 h1m_state_str(h1s->req.state),
3006 h1m_state_str(h1s->res.state), method, h1s->status);
3007 if (h1s->cs)
3008 chunk_appendf(msg, " .cs.flg=0x%08x .cs.data=%p",
3009 h1s->cs->flags, h1s->cs->data);
3010
3011 chunk_appendf(&trash, " .subs=%p", h1s->subs);
3012 if (h1s->subs) {
3013 chunk_appendf(&trash, "(ev=%d tl=%p", h1s->subs->events, h1s->subs->tasklet);
3014 chunk_appendf(&trash, " tl.calls=%d tl.ctx=%p tl.fct=",
3015 h1s->subs->tasklet->calls,
3016 h1s->subs->tasklet->context);
3017 if (h1s->subs->tasklet->calls >= 1000000)
3018 ret = 1;
3019 resolve_sym_name(&trash, NULL, h1s->subs->tasklet->process);
3020 chunk_appendf(&trash, ")");
3021 }
3022 }
3023 return ret;
3024 }
3025
3026
3027 /* Add an entry in the headers map. Returns -1 on error and 0 on success. */
add_hdr_case_adjust(const char * from,const char * to,char ** err)3028 static int add_hdr_case_adjust(const char *from, const char *to, char **err)
3029 {
3030 struct h1_hdr_entry *entry;
3031
3032 /* Be sure there is a non-empty <to> */
3033 if (!strlen(to)) {
3034 memprintf(err, "expect <to>");
3035 return -1;
3036 }
3037
3038 /* Be sure only the case differs between <from> and <to> */
3039 if (strcasecmp(from, to)) {
3040 memprintf(err, "<from> and <to> must not differ execpt the case");
3041 return -1;
3042 }
3043
3044 /* Be sure <from> does not already existsin the tree */
3045 if (ebis_lookup(&hdrs_map.map, from)) {
3046 memprintf(err, "duplicate entry '%s'", from);
3047 return -1;
3048 }
3049
3050 /* Create the entry and insert it in the tree */
3051 entry = malloc(sizeof(*entry));
3052 if (!entry) {
3053 memprintf(err, "out of memory");
3054 return -1;
3055 }
3056
3057 entry->node.key = strdup(from);
3058 entry->name.ptr = strdup(to);
3059 entry->name.len = strlen(to);
3060 if (!entry->node.key || !entry->name.ptr) {
3061 free(entry->node.key);
3062 istfree(&entry->name);
3063 free(entry);
3064 memprintf(err, "out of memory");
3065 return -1;
3066 }
3067 ebis_insert(&hdrs_map.map, &entry->node);
3068 return 0;
3069 }
3070
3071 /* Migrate the the connection to the current thread.
3072 * Return 0 if successful, non-zero otherwise.
3073 * Expected to be called with the old thread lock held.
3074 */
h1_takeover(struct connection * conn,int orig_tid)3075 static int h1_takeover(struct connection *conn, int orig_tid)
3076 {
3077 struct h1c *h1c = conn->ctx;
3078 struct task *task;
3079
3080 if (fd_takeover(conn->handle.fd, conn) != 0)
3081 return -1;
3082
3083 if (conn->xprt->takeover && conn->xprt->takeover(conn, conn->xprt_ctx, orig_tid) != 0) {
3084 /* We failed to takeover the xprt, even if the connection may
3085 * still be valid, flag it as error'd, as we have already
3086 * taken over the fd, and wake the tasklet, so that it will
3087 * destroy it.
3088 */
3089 conn->flags |= CO_FL_ERROR;
3090 tasklet_wakeup_on(h1c->wait_event.tasklet, orig_tid);
3091 return -1;
3092 }
3093
3094 if (h1c->wait_event.events)
3095 h1c->conn->xprt->unsubscribe(h1c->conn, h1c->conn->xprt_ctx,
3096 h1c->wait_event.events, &h1c->wait_event);
3097 /* To let the tasklet know it should free itself, and do nothing else,
3098 * set its context to NULL.
3099 */
3100 h1c->wait_event.tasklet->context = NULL;
3101 tasklet_wakeup_on(h1c->wait_event.tasklet, orig_tid);
3102
3103 task = h1c->task;
3104 if (task) {
3105 task->context = NULL;
3106 h1c->task = NULL;
3107 __ha_barrier_store();
3108 task_kill(task);
3109
3110 h1c->task = task_new(tid_bit);
3111 if (!h1c->task) {
3112 h1_release(h1c);
3113 return -1;
3114 }
3115 h1c->task->process = h1_timeout_task;
3116 h1c->task->context = h1c;
3117 }
3118 h1c->wait_event.tasklet = tasklet_new();
3119 if (!h1c->wait_event.tasklet) {
3120 h1_release(h1c);
3121 return -1;
3122 }
3123 h1c->wait_event.tasklet->process = h1_io_cb;
3124 h1c->wait_event.tasklet->context = h1c;
3125 h1c->conn->xprt->subscribe(h1c->conn, h1c->conn->xprt_ctx,
3126 SUB_RETRY_RECV, &h1c->wait_event);
3127
3128 return 0;
3129 }
3130
3131
h1_hdeaders_case_adjust_deinit()3132 static void h1_hdeaders_case_adjust_deinit()
3133 {
3134 struct ebpt_node *node, *next;
3135 struct h1_hdr_entry *entry;
3136
3137 node = ebpt_first(&hdrs_map.map);
3138 while (node) {
3139 next = ebpt_next(node);
3140 ebpt_delete(node);
3141 entry = container_of(node, struct h1_hdr_entry, node);
3142 free(entry->node.key);
3143 istfree(&entry->name);
3144 free(entry);
3145 node = next;
3146 }
3147 free(hdrs_map.name);
3148 }
3149
cfg_h1_headers_case_adjust_postparser()3150 static int cfg_h1_headers_case_adjust_postparser()
3151 {
3152 FILE *file = NULL;
3153 char *c, *key_beg, *key_end, *value_beg, *value_end;
3154 char *err;
3155 int rc, line = 0, err_code = 0;
3156
3157 if (!hdrs_map.name)
3158 goto end;
3159
3160 file = fopen(hdrs_map.name, "r");
3161 if (!file) {
3162 ha_alert("config : h1-outgoing-headers-case-adjust-file '%s': failed to open file.\n",
3163 hdrs_map.name);
3164 err_code |= ERR_ALERT | ERR_FATAL;
3165 goto end;
3166 }
3167
3168 /* now parse all lines. The file may contain only two header name per
3169 * line, separated by spaces. All heading and trailing spaces will be
3170 * ignored. Lines starting with a # are ignored.
3171 */
3172 while (fgets(trash.area, trash.size, file) != NULL) {
3173 line++;
3174 c = trash.area;
3175
3176 /* strip leading spaces and tabs */
3177 while (*c == ' ' || *c == '\t')
3178 c++;
3179
3180 /* ignore emptu lines, or lines beginning with a dash */
3181 if (*c == '#' || *c == '\0' || *c == '\r' || *c == '\n')
3182 continue;
3183
3184 /* look for the end of the key */
3185 key_beg = c;
3186 while (*c != '\0' && *c != ' ' && *c != '\t' && *c != '\n' && *c != '\r')
3187 c++;
3188 key_end = c;
3189
3190 /* strip middle spaces and tabs */
3191 while (*c == ' ' || *c == '\t')
3192 c++;
3193
3194 /* look for the end of the value, it is the end of the line */
3195 value_beg = c;
3196 while (*c && *c != '\n' && *c != '\r')
3197 c++;
3198 value_end = c;
3199
3200 /* trim possibly trailing spaces and tabs */
3201 while (value_end > value_beg && (value_end[-1] == ' ' || value_end[-1] == '\t'))
3202 value_end--;
3203
3204 /* set final \0 and check entries */
3205 *key_end = '\0';
3206 *value_end = '\0';
3207
3208 err = NULL;
3209 rc = add_hdr_case_adjust(key_beg, value_beg, &err);
3210 if (rc < 0) {
3211 ha_alert("config : h1-outgoing-headers-case-adjust-file '%s' : %s at line %d.\n",
3212 hdrs_map.name, err, line);
3213 err_code |= ERR_ALERT | ERR_FATAL;
3214 free(err);
3215 goto end;
3216 }
3217 if (rc > 0) {
3218 ha_warning("config : h1-outgoing-headers-case-adjust-file '%s' : %s at line %d.\n",
3219 hdrs_map.name, err, line);
3220 err_code |= ERR_WARN;
3221 free(err);
3222 }
3223 }
3224
3225 end:
3226 if (file)
3227 fclose(file);
3228 hap_register_post_deinit(h1_hdeaders_case_adjust_deinit);
3229 return err_code;
3230 }
3231
3232
3233 /* config parser for global "h1-outgoing-header-case-adjust" */
cfg_parse_h1_header_case_adjust(char ** args,int section_type,struct proxy * curpx,struct proxy * defpx,const char * file,int line,char ** err)3234 static int cfg_parse_h1_header_case_adjust(char **args, int section_type, struct proxy *curpx,
3235 struct proxy *defpx, const char *file, int line,
3236 char **err)
3237 {
3238 if (too_many_args(2, args, err, NULL))
3239 return -1;
3240 if (!*(args[1]) || !*(args[2])) {
3241 memprintf(err, "'%s' expects <from> and <to> as argument.", args[0]);
3242 return -1;
3243 }
3244 return add_hdr_case_adjust(args[1], args[2], err);
3245 }
3246
3247 /* config parser for global "h1-outgoing-headers-case-adjust-file" */
cfg_parse_h1_headers_case_adjust_file(char ** args,int section_type,struct proxy * curpx,struct proxy * defpx,const char * file,int line,char ** err)3248 static int cfg_parse_h1_headers_case_adjust_file(char **args, int section_type, struct proxy *curpx,
3249 struct proxy *defpx, const char *file, int line,
3250 char **err)
3251 {
3252 if (too_many_args(1, args, err, NULL))
3253 return -1;
3254 if (!*(args[1])) {
3255 memprintf(err, "'%s' expects <file> as argument.", args[0]);
3256 return -1;
3257 }
3258 free(hdrs_map.name);
3259 hdrs_map.name = strdup(args[1]);
3260 return 0;
3261 }
3262
3263
3264 /* config keyword parsers */
3265 static struct cfg_kw_list cfg_kws = {{ }, {
3266 { CFG_GLOBAL, "h1-case-adjust", cfg_parse_h1_header_case_adjust },
3267 { CFG_GLOBAL, "h1-case-adjust-file", cfg_parse_h1_headers_case_adjust_file },
3268 { 0, NULL, NULL },
3269 }
3270 };
3271
3272 INITCALL1(STG_REGISTER, cfg_register_keywords, &cfg_kws);
3273 REGISTER_CONFIG_POSTPARSER("h1-headers-map", cfg_h1_headers_case_adjust_postparser);
3274
3275
3276 /****************************************/
3277 /* MUX initialization and instantiation */
3278 /****************************************/
3279
3280 /* The mux operations */
3281 static const struct mux_ops mux_h1_ops = {
3282 .init = h1_init,
3283 .wake = h1_wake,
3284 .attach = h1_attach,
3285 .get_first_cs = h1_get_first_cs,
3286 .get_cs_info = h1_get_cs_info,
3287 .detach = h1_detach,
3288 .destroy = h1_destroy,
3289 .avail_streams = h1_avail_streams,
3290 .used_streams = h1_used_streams,
3291 .rcv_buf = h1_rcv_buf,
3292 .snd_buf = h1_snd_buf,
3293 #if defined(USE_LINUX_SPLICE)
3294 .rcv_pipe = h1_rcv_pipe,
3295 .snd_pipe = h1_snd_pipe,
3296 #endif
3297 .subscribe = h1_subscribe,
3298 .unsubscribe = h1_unsubscribe,
3299 .shutr = h1_shutr,
3300 .shutw = h1_shutw,
3301 .show_fd = h1_show_fd,
3302 .reset = h1_reset,
3303 .ctl = h1_ctl,
3304 .takeover = h1_takeover,
3305 .flags = MX_FL_HTX,
3306 .name = "H1",
3307 };
3308
3309
3310 /* this mux registers default HTX proto */
3311 static struct mux_proto_list mux_proto_htx =
3312 { .token = IST(""), .mode = PROTO_MODE_HTTP, .side = PROTO_SIDE_BOTH, .mux = &mux_h1_ops };
3313
3314 INITCALL1(STG_REGISTER, register_mux_proto, &mux_proto_htx);
3315
3316 /*
3317 * Local variables:
3318 * c-indent-level: 8
3319 * c-basic-offset: 8
3320 * End:
3321 */
3322