| | varnish-cache/bin/varnishd/http1/cache_http1_line.c |
0 |
|
/*- |
1 |
|
* Copyright (c) 2006 Verdens Gang AS |
2 |
|
* Copyright (c) 2006-2011 Varnish Software AS |
3 |
|
* All rights reserved. |
4 |
|
* |
5 |
|
* Author: Poul-Henning Kamp <phk@phk.freebsd.dk> |
6 |
|
* |
7 |
|
* SPDX-License-Identifier: BSD-2-Clause |
8 |
|
* |
9 |
|
* Redistribution and use in source and binary forms, with or without |
10 |
|
* modification, are permitted provided that the following conditions |
11 |
|
* are met: |
12 |
|
* 1. Redistributions of source code must retain the above copyright |
13 |
|
* notice, this list of conditions and the following disclaimer. |
14 |
|
* 2. Redistributions in binary form must reproduce the above copyright |
15 |
|
* notice, this list of conditions and the following disclaimer in the |
16 |
|
* documentation and/or other materials provided with the distribution. |
17 |
|
* |
18 |
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
19 |
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
20 |
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
21 |
|
* ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE |
22 |
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
23 |
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
24 |
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
25 |
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
26 |
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
27 |
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
28 |
|
* SUCH DAMAGE. |
29 |
|
* |
30 |
|
* Write data to fd |
31 |
|
* We try to use writev() if possible in order to minimize number of |
32 |
|
* syscalls made and packets sent. It also just might allow the worker |
33 |
|
* thread to complete the request without holding stuff locked. |
34 |
|
* |
35 |
|
* XXX: chunked header (generated in Flush) and Tail (EndChunk) |
36 |
|
* are not accounted by means of the size_t returned. Obvious ideas: |
37 |
|
* - add size_t return value to Flush and EndChunk |
38 |
|
* - base accounting on (struct v1l).cnt |
39 |
|
*/ |
40 |
|
|
41 |
|
#include "config.h" |
42 |
|
|
43 |
|
#include <sys/uio.h> |
44 |
|
#include "cache/cache_varnishd.h" |
45 |
|
#include "cache/cache_filter.h" |
46 |
|
|
47 |
|
#include <stdio.h> |
48 |
|
|
49 |
|
#include "cache_http1.h" |
50 |
|
#include "vtim.h" |
51 |
|
|
52 |
|
/*--------------------------------------------------------------------*/ |
53 |
|
|
54 |
|
struct v1l { |
55 |
|
unsigned magic; |
56 |
|
#define V1L_MAGIC 0x2f2142e5 |
57 |
|
int *wfd; |
58 |
|
stream_close_t werr; /* valid after V1L_Flush() */ |
59 |
|
struct iovec *iov; |
60 |
|
int siov; |
61 |
|
int niov; |
62 |
|
size_t liov; |
63 |
|
size_t cliov; |
64 |
|
int ciov; /* Chunked header marker */ |
65 |
|
vtim_real deadline; |
66 |
|
struct vsl_log *vsl; |
67 |
|
uint64_t cnt; /* Flushed byte count */ |
68 |
|
struct ws *ws; |
69 |
|
uintptr_t ws_snap; |
70 |
|
void **vdp_priv; |
71 |
|
}; |
72 |
|
|
73 |
|
/*-------------------------------------------------------------------- |
74 |
|
* for niov == 0, reserve the ws for max number of iovs |
75 |
|
* otherwise, up to niov |
76 |
|
*/ |
77 |
|
|
78 |
|
struct v1l * |
79 |
25612 |
V1L_Open(struct ws *ws, int *fd, struct vsl_log *vsl, |
80 |
|
vtim_real deadline, unsigned niov) |
81 |
|
{ |
82 |
|
struct v1l *v1l; |
83 |
|
unsigned u; |
84 |
|
uintptr_t ws_snap; |
85 |
|
size_t sz; |
86 |
|
|
87 |
25612 |
if (WS_Overflowed(ws)) |
88 |
0 |
return (NULL); |
89 |
|
|
90 |
25612 |
if (niov != 0) |
91 |
15112 |
assert(niov >= 3); |
92 |
|
|
93 |
25612 |
ws_snap = WS_Snapshot(ws); |
94 |
|
|
95 |
25612 |
v1l = WS_Alloc(ws, sizeof *v1l); |
96 |
25612 |
if (v1l == NULL) |
97 |
5 |
return (NULL); |
98 |
25607 |
INIT_OBJ(v1l, V1L_MAGIC); |
99 |
|
|
100 |
25607 |
v1l->ws = ws; |
101 |
25607 |
v1l->ws_snap = ws_snap; |
102 |
|
|
103 |
25607 |
u = WS_ReserveLumps(ws, sizeof(struct iovec)); |
104 |
25607 |
if (u < 3) { |
105 |
|
/* Must have at least 3 in case of chunked encoding */ |
106 |
0 |
WS_Release(ws, 0); |
107 |
0 |
WS_MarkOverflow(ws); |
108 |
0 |
return (NULL); |
109 |
|
} |
110 |
25607 |
if (u > IOV_MAX) |
111 |
160 |
u = IOV_MAX; |
112 |
25607 |
if (niov != 0 && u > niov) |
113 |
14733 |
u = niov; |
114 |
25607 |
v1l->iov = WS_Reservation(ws); |
115 |
25607 |
v1l->siov = (int)u; |
116 |
25607 |
v1l->ciov = (int)u; |
117 |
25607 |
v1l->wfd = fd; |
118 |
25607 |
v1l->deadline = deadline; |
119 |
25607 |
v1l->vsl = vsl; |
120 |
25607 |
v1l->werr = SC_NULL; |
121 |
|
|
122 |
25607 |
sz = u * sizeof(struct iovec); |
123 |
25607 |
assert(sz < UINT_MAX); |
124 |
25607 |
WS_Release(ws, (unsigned)sz); |
125 |
25607 |
return (v1l); |
126 |
25612 |
} |
127 |
|
|
128 |
|
void |
129 |
160 |
V1L_NoRollback(struct v1l *v1l) |
130 |
|
{ |
131 |
|
|
132 |
160 |
CHECK_OBJ_NOTNULL(v1l, V1L_MAGIC); |
133 |
160 |
v1l->ws_snap = 0; |
134 |
160 |
} |
135 |
|
|
136 |
|
stream_close_t |
137 |
25608 |
V1L_Close(struct v1l **v1lp, uint64_t *cnt) |
138 |
|
{ |
139 |
|
struct v1l *v1l; |
140 |
|
struct ws *ws; |
141 |
|
uintptr_t ws_snap; |
142 |
|
stream_close_t sc; |
143 |
|
|
144 |
25608 |
AN(cnt); |
145 |
25608 |
TAKE_OBJ_NOTNULL(v1l, v1lp, V1L_MAGIC); |
146 |
25608 |
if (v1l->vdp_priv != NULL) { |
147 |
20104 |
assert(*v1l->vdp_priv == v1l); |
148 |
20104 |
*v1l->vdp_priv = NULL; |
149 |
20104 |
} |
150 |
25608 |
sc = V1L_Flush(v1l); |
151 |
25608 |
*cnt = v1l->cnt; |
152 |
25608 |
ws = v1l->ws; |
153 |
25608 |
ws_snap = v1l->ws_snap; |
154 |
25608 |
ZERO_OBJ(v1l, sizeof *v1l); |
155 |
25608 |
if (ws_snap != 0) |
156 |
25449 |
WS_Rollback(ws, ws_snap); |
157 |
25608 |
return (sc); |
158 |
|
} |
159 |
|
|
160 |
|
static void |
161 |
55 |
v1l_prune(struct v1l *v1l, ssize_t abytes) |
162 |
|
{ |
163 |
55 |
size_t used = 0; |
164 |
|
size_t sz, bytes, used_here; |
165 |
|
int j; |
166 |
|
|
167 |
55 |
assert(abytes > 0); |
168 |
55 |
bytes = (size_t)abytes; |
169 |
|
|
170 |
305 |
for (j = 0; j < v1l->niov; j++) { |
171 |
305 |
if (used + v1l->iov[j].iov_len > bytes) { |
172 |
|
/* Cutoff is in this iov */ |
173 |
55 |
used_here = bytes - used; |
174 |
55 |
v1l->iov[j].iov_len -= used_here; |
175 |
55 |
v1l->iov[j].iov_base = |
176 |
55 |
(char*)v1l->iov[j].iov_base + used_here; |
177 |
55 |
sz = (unsigned)v1l->niov - (unsigned)j; |
178 |
55 |
sz *= sizeof(struct iovec); |
179 |
55 |
memmove(v1l->iov, &v1l->iov[j], sz); |
180 |
55 |
v1l->niov -= j; |
181 |
55 |
assert(v1l->liov >= bytes); |
182 |
55 |
v1l->liov -= bytes; |
183 |
55 |
return; |
184 |
|
} |
185 |
250 |
used += v1l->iov[j].iov_len; |
186 |
250 |
} |
187 |
0 |
AZ(v1l->liov); |
188 |
55 |
} |
189 |
|
|
190 |
|
stream_close_t |
191 |
51383 |
V1L_Flush(struct v1l *v1l) |
192 |
|
{ |
193 |
|
ssize_t i; |
194 |
|
size_t sz; |
195 |
|
int err; |
196 |
|
char cbuf[32]; |
197 |
|
|
198 |
51383 |
CHECK_OBJ_NOTNULL(v1l, V1L_MAGIC); |
199 |
51383 |
CHECK_OBJ_NOTNULL(v1l->werr, STREAM_CLOSE_MAGIC); |
200 |
51383 |
AN(v1l->wfd); |
201 |
|
|
202 |
51383 |
assert(v1l->niov <= v1l->siov); |
203 |
|
|
204 |
51383 |
if (*v1l->wfd >= 0 && v1l->liov > 0 && v1l->werr == SC_NULL) { |
205 |
34637 |
if (v1l->ciov < v1l->siov && v1l->cliov > 0) { |
206 |
|
/* Add chunk head & tail */ |
207 |
5339 |
bprintf(cbuf, "00%zx\r\n", v1l->cliov); |
208 |
5339 |
sz = strlen(cbuf); |
209 |
5339 |
v1l->iov[v1l->ciov].iov_base = cbuf; |
210 |
5339 |
v1l->iov[v1l->ciov].iov_len = sz; |
211 |
5339 |
v1l->liov += sz; |
212 |
|
|
213 |
|
/* This is OK, because siov was --'ed */ |
214 |
5339 |
v1l->iov[v1l->niov].iov_base = cbuf + sz - 2; |
215 |
5339 |
v1l->iov[v1l->niov++].iov_len = 2; |
216 |
5339 |
v1l->liov += 2; |
217 |
34637 |
} else if (v1l->ciov < v1l->siov) { |
218 |
314 |
v1l->iov[v1l->ciov].iov_base = cbuf; |
219 |
314 |
v1l->iov[v1l->ciov].iov_len = 0; |
220 |
314 |
} |
221 |
|
|
222 |
34637 |
i = 0; |
223 |
34637 |
err = 0; |
224 |
34637 |
do { |
225 |
34829 |
if (VTIM_real() > v1l->deadline) { |
226 |
40 |
VSLb(v1l->vsl, SLT_Debug, |
227 |
|
"Hit total send timeout, " |
228 |
|
"wrote = %zd/%zd; not retrying", |
229 |
20 |
i, v1l->liov); |
230 |
20 |
i = -1; |
231 |
20 |
break; |
232 |
|
} |
233 |
|
|
234 |
34809 |
i = writev(*v1l->wfd, v1l->iov, v1l->niov); |
235 |
34809 |
if (i > 0) { |
236 |
34634 |
v1l->cnt += (size_t)i; |
237 |
34634 |
if ((size_t)i == v1l->liov) |
238 |
34579 |
break; |
239 |
55 |
} |
240 |
|
|
241 |
|
/* we hit a timeout, and some data may have been sent: |
242 |
|
* Remove sent data from start of I/O vector, then retry |
243 |
|
* |
244 |
|
* XXX: Add a "minimum sent data per timeout counter to |
245 |
|
* prevent slowloris attacks |
246 |
|
*/ |
247 |
|
|
248 |
230 |
err = errno; |
249 |
|
|
250 |
230 |
if (err == EWOULDBLOCK) { |
251 |
270 |
VSLb(v1l->vsl, SLT_Debug, |
252 |
|
"Hit idle send timeout, " |
253 |
|
"wrote = %zd/%zd; retrying", |
254 |
135 |
i, v1l->liov); |
255 |
135 |
} |
256 |
|
|
257 |
230 |
if (i > 0) |
258 |
55 |
v1l_prune(v1l, i); |
259 |
230 |
} while (i > 0 || err == EWOULDBLOCK); |
260 |
|
|
261 |
34637 |
if (i <= 0) { |
262 |
120 |
VSLb(v1l->vsl, SLT_Debug, |
263 |
|
"Write error, retval = %zd, len = %zd, errno = %s", |
264 |
60 |
i, v1l->liov, VAS_errtxt(err)); |
265 |
60 |
assert(v1l->werr == SC_NULL); |
266 |
60 |
if (err == EPIPE) |
267 |
39 |
v1l->werr = SC_REM_CLOSE; |
268 |
|
else |
269 |
21 |
v1l->werr = SC_TX_ERROR; |
270 |
60 |
errno = err; |
271 |
60 |
} |
272 |
34637 |
} |
273 |
51389 |
v1l->liov = 0; |
274 |
51389 |
v1l->cliov = 0; |
275 |
51389 |
v1l->niov = 0; |
276 |
51389 |
if (v1l->ciov < v1l->siov) |
277 |
10371 |
v1l->ciov = v1l->niov++; |
278 |
51377 |
CHECK_OBJ_NOTNULL(v1l->werr, STREAM_CLOSE_MAGIC); |
279 |
51377 |
return (v1l->werr); |
280 |
|
} |
281 |
|
|
282 |
|
size_t |
283 |
572836 |
V1L_Write(struct v1l *v1l, const void *ptr, ssize_t alen) |
284 |
|
{ |
285 |
572836 |
size_t len = 0; |
286 |
|
|
287 |
572836 |
CHECK_OBJ_NOTNULL(v1l, V1L_MAGIC); |
288 |
572836 |
AN(v1l->wfd); |
289 |
572836 |
if (alen == 0 || *v1l->wfd < 0) |
290 |
158 |
return (0); |
291 |
572836 |
if (alen > 0) |
292 |
282176 |
len = (size_t)alen; |
293 |
290660 |
else if (alen == -1) |
294 |
290660 |
len = strlen(ptr); |
295 |
|
else |
296 |
0 |
WRONG("alen"); |
297 |
|
|
298 |
572836 |
assert(v1l->niov < v1l->siov); |
299 |
572836 |
v1l->iov[v1l->niov].iov_base = TRUST_ME(ptr); |
300 |
572836 |
v1l->iov[v1l->niov].iov_len = len; |
301 |
572836 |
v1l->liov += len; |
302 |
572836 |
v1l->niov++; |
303 |
572836 |
v1l->cliov += len; |
304 |
572836 |
if (v1l->niov >= v1l->siov) { |
305 |
550 |
(void)V1L_Flush(v1l); |
306 |
550 |
VSC_C_main->http1_iovs_flush++; |
307 |
550 |
} |
308 |
572836 |
return (len); |
309 |
572836 |
} |
310 |
|
|
311 |
|
void |
312 |
1385 |
V1L_Chunked(struct v1l *v1l) |
313 |
|
{ |
314 |
|
|
315 |
1385 |
CHECK_OBJ_NOTNULL(v1l, V1L_MAGIC); |
316 |
|
|
317 |
1385 |
assert(v1l->ciov == v1l->siov); |
318 |
1385 |
assert(v1l->siov >= 3); |
319 |
|
/* |
320 |
|
* If there is no space for chunked header, a chunk of data and |
321 |
|
* a chunk tail, we might as well flush right away. |
322 |
|
*/ |
323 |
1385 |
if (v1l->niov + 3 >= v1l->siov) { |
324 |
0 |
(void)V1L_Flush(v1l); |
325 |
0 |
VSC_C_main->http1_iovs_flush++; |
326 |
0 |
} |
327 |
1385 |
v1l->siov--; |
328 |
1385 |
v1l->ciov = v1l->niov++; |
329 |
1385 |
v1l->cliov = 0; |
330 |
1385 |
assert(v1l->ciov < v1l->siov); |
331 |
1385 |
assert(v1l->niov < v1l->siov); |
332 |
1385 |
} |
333 |
|
|
334 |
|
/* |
335 |
|
* XXX: It is not worth the complexity to attempt to get the |
336 |
|
* XXX: end of chunk into the V1L_Flush(), because most of the time |
337 |
|
* XXX: if not always, that is a no-op anyway, because the calling |
338 |
|
* XXX: code already called V1L_Flush() to release local storage. |
339 |
|
*/ |
340 |
|
|
341 |
|
void |
342 |
1279 |
V1L_EndChunk(struct v1l *v1l) |
343 |
|
{ |
344 |
|
|
345 |
1279 |
CHECK_OBJ_NOTNULL(v1l, V1L_MAGIC); |
346 |
|
|
347 |
1279 |
assert(v1l->ciov < v1l->siov); |
348 |
1279 |
(void)V1L_Flush(v1l); |
349 |
1279 |
v1l->siov++; |
350 |
1279 |
v1l->ciov = v1l->siov; |
351 |
1279 |
v1l->niov = 0; |
352 |
1279 |
v1l->cliov = 0; |
353 |
1279 |
(void)V1L_Write(v1l, "0\r\n\r\n", -1); |
354 |
1279 |
} |
355 |
|
|
356 |
|
/*-------------------------------------------------------------------- |
357 |
|
* VDP using V1L |
358 |
|
*/ |
359 |
|
|
360 |
|
/* remember priv pointer for V1L_Close() to clear */ |
361 |
|
static int v_matchproto_(vdp_init_f) |
362 |
20105 |
v1l_init(VRT_CTX, struct vdp_ctx *vdc, void **priv) |
363 |
|
{ |
364 |
|
struct v1l *v1l; |
365 |
|
|
366 |
20105 |
(void) ctx; |
367 |
20105 |
(void) vdc; |
368 |
20105 |
AN(priv); |
369 |
20105 |
CAST_OBJ_NOTNULL(v1l, *priv, V1L_MAGIC); |
370 |
|
|
371 |
20105 |
v1l->vdp_priv = priv; |
372 |
20105 |
return (0); |
373 |
|
} |
374 |
|
|
375 |
|
static int v_matchproto_(vdp_bytes_f) |
376 |
24520 |
v1l_bytes(struct vdp_ctx *vdc, enum vdp_action act, void **priv, |
377 |
|
const void *ptr, ssize_t len) |
378 |
|
{ |
379 |
24520 |
size_t wl = 0; |
380 |
|
|
381 |
24520 |
CHECK_OBJ_NOTNULL(vdc, VDP_CTX_MAGIC); |
382 |
24520 |
AN(priv); |
383 |
|
|
384 |
24520 |
AZ(vdc->nxt); /* always at the bottom of the pile */ |
385 |
|
|
386 |
24520 |
if (len > 0) |
387 |
17779 |
wl = V1L_Write(*priv, ptr, len); |
388 |
24520 |
if (act > VDP_NULL && V1L_Flush(*priv) != SC_NULL) |
389 |
59 |
return (-1); |
390 |
24461 |
if ((size_t)len != wl) |
391 |
0 |
return (-1); |
392 |
24461 |
return (0); |
393 |
24520 |
} |
394 |
|
|
395 |
|
const struct vdp * const VDP_v1l = &(struct vdp){ |
396 |
|
.name = "V1B", |
397 |
|
.init = v1l_init, |
398 |
|
.bytes = v1l_bytes, |
399 |
|
}; |