summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--.gitignore3
-rw-r--r--DESIGN99
-rw-r--r--Makefile15
-rw-r--r--README12
-rw-r--r--man4/opentracker.conf.411
-rw-r--r--opentracker.c131
-rw-r--r--opentracker.xcodeproj/project.pbxproj320
-rw-r--r--ot_config.c26
-rw-r--r--ot_event.c9
-rw-r--r--ot_event.h5
-rw-r--r--ot_fdstate.c7
-rw-r--r--ot_fdstate.h3
-rw-r--r--ot_http.c63
-rw-r--r--ot_http.h12
-rw-r--r--ot_socket.c21
-rw-r--r--ot_socket.h1
-rw-r--r--ot_tcp.c157
-rw-r--r--ot_tcp.h22
-rw-r--r--tests/accept_filter_probe.py110
-rw-r--r--tests/integration/test_config.py29
-rw-r--r--tests/integration/test_expiry.py7
-rw-r--r--tests/syscount.c144
-rw-r--r--tests/t_event.c69
-rw-r--r--tests/t_socket.c43
24 files changed, 824 insertions, 495 deletions
diff --git a/.gitignore b/.gitignore
index b5c40f0..3e469ff 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,5 +1,5 @@
1*.o
2*.dSYM 1*.dSYM
2*.o
3opentracker 3opentracker
4opentracker.debug 4opentracker.debug
5opentracker.integration 5opentracker.integration
@@ -7,4 +7,5 @@ opentracker.tsan
7proxy 7proxy
8proxy.debug 8proxy.debug
9tests/runner 9tests/runner
10tests/runner.poll
10tests/runner.sanitize 11tests/runner.sanitize
diff --git a/DESIGN b/DESIGN
new file mode 100644
index 0000000..ff37706
--- /dev/null
+++ b/DESIGN
@@ -0,0 +1,99 @@
1opentracker design notes
2========================
3
4Decisions that cost measurement to reach, and the alternatives that were
5tried and rejected.
6
7What the tracker is optimised for
8---------------------------------
9
10UDP carries most of the traffic. An HTTP request costs about ten times a UDP
11announce, 26 microseconds of tracker CPU against 2.5. Nearly all of that is
12the kernel setting up and tearing down a connection rather than tracker
13code.
14
15A connection carries one request and closes. Keep-alive works and is rarely
16asked for.
17
18Statistics are approximate on purpose. Counters are per thread and summed
19when read, so a /stats read can under-report once. Nothing drifts.
20
21
22Decisions
23---------
24
25One UDP socket per worker, where the kernel spreads datagrams over a group
26bound to one address. Cost per datagram then stays flat as workers are added,
270.71 to 0.72 microseconds from one worker to eight. Sharing one socket grows
28it eightfold over the same range, 0.71 to 6.07.
29
30listen.udp.workers defaults to 1. One worker beats none by 19 to 37 per cent
31everywhere, because a blocking read costs one system call where the event
32loop costs two. Beyond one the gain depends on the kernel balancing, so
33raising it is the operator's decision.
34
35A thread parked in a blocking read is woken by a signal, not by closing or
36shutting down its socket:
37
38 mechanism Linux FreeBSD macOS
39 close() blocked blocked 0 ms
40 shutdown(SHUT_RD) 1 ms 1 ms blocked
41 SO_RCVTIMEO works works works
42 pthread_kill 1 ms 1 ms 1 ms
43
44Only the signal works everywhere. A receive timeout also works, but wakes
45every thread on every tick for the life of the process to catch something
46that happens once: at eight workers it doubled idle context switches, 255 to
47495 over thirty seconds.
48
49Each binary is built in one compiler invocation. Without object files a build
50cannot link objects compiled with different FEATURES, which was a real
51failure mode. It costs incremental rebuilds, 0.075 seconds becoming 1.56.
52
53A socket is read or written, never both. http_interest_set replaces an
54add-and-drop pair so that the invariant cannot be broken by accident, and the
55event layer hears about a change once instead of twice.
56
57Accept waits until the client's request has arrived: TCP_DEFER_ACCEPT on
58Linux, the accf_data filter on FreeBSD. Measured against a flood of silent
59connections, the filter serves more requests than no filter, not fewer, so it
60is not the denial of service vector that FreeBSD-SA-02:26 once described.
61
62
63Paths not taken
64---------------
65
66accf_http, the HTTP-aware accept filter. It duplicates a parser the tracker
67has anyway, puts more kernel code in the path, and holds a connection that
68sends "GET /" and stops, which accf_data hands straight over.
69
70recvmmsg and sendmmsg. No measurable gain on either FreeBSD machine, between
71-4 and +10 per cent, and they do not exist on Darwin.
72
73Link time optimisation. It cuts .text by 13 per cent, 93,497 bytes to 81,630,
74and changes throughput not at all. A tracker waits on memory and system
75calls, not on instruction fetch: the announce path is 6.6 KB of code against
76a 32 KB instruction cache at worst.
77
78A directory for object files. bmake has no pattern rules and the Makefile
79builds under FreeBSD's make today. Building each binary in one invocation
80clears the objects out of the tree without giving that up.
81
82A src/ directory. 52 files sharing an ot_ prefix are a module list rather
83than clutter, and build output was most of what made the tree look crowded.
84
85
86Platform differences that cost us code
87--------------------------------------
88
89SO_REUSEPORT means three different things. Linux spreads datagrams over the
90group; FreeBSD permits the bind and needs SO_REUSEPORT_LB to spread. Darwin
91hands every datagram to whichever socket bound last, so a group there starves
92all but one reader.
93
94Waking a blocked reader differs per kernel. see the table above.
95
96Deferring accept is TCP_DEFER_ACCEPT, or accf_data, or nothing.
97
98make is bmake on FreeBSD and GNU make elsewhere, and the Makefile has to
99build under both.
diff --git a/Makefile b/Makefile
index 3ce127e..8f4dcf6 100644
--- a/Makefile
+++ b/Makefile
@@ -69,8 +69,8 @@ CFLAGS_libowfat=-I$(LIBOWFAT_HEADERS)
69LDFLAGS_libowfat=-L$(LIBOWFAT_LIBRARY) -lowfat 69LDFLAGS_libowfat=-L$(LIBOWFAT_LIBRARY) -lowfat
70 70
71BINARY =opentracker 71BINARY =opentracker
72HEADERS=trackerlogic.h ot_urlencoded.h ot_batch.h ot_buf.h ot_event.h ot_fdstate.h ot_fmt.h ot_hex.h ot_ip.h ot_config.h ot_scan.h ot_socket.h ot_mutex.h ot_random.h ot_stats.h ot_vector.h ot_clean.h ot_udp.h ot_iovec.h ot_fullscrape.h ot_accesslist.h ot_http.h ot_livesync.h ot_rijndael.h 72HEADERS=trackerlogic.h ot_urlencoded.h ot_batch.h ot_buf.h ot_event.h ot_fdstate.h ot_fmt.h ot_hex.h ot_ip.h ot_config.h ot_scan.h ot_socket.h ot_tcp.h ot_mutex.h ot_random.h ot_stats.h ot_vector.h ot_clean.h ot_udp.h ot_iovec.h ot_fullscrape.h ot_accesslist.h ot_http.h ot_livesync.h ot_rijndael.h
73SOURCES=opentracker.c trackerlogic.c ot_urlencoded.c ot_batch.c ot_buf.c ot_event.c ot_fdstate.c ot_fmt.c ot_hex.c ot_ip.c ot_config.c ot_scan.c ot_socket.c ot_mutex.c ot_random.c ot_stats.c ot_vector.c ot_clean.c ot_udp.c ot_iovec.c ot_fullscrape.c ot_accesslist.c ot_http.c ot_livesync.c ot_rijndael.c 73SOURCES=opentracker.c trackerlogic.c ot_urlencoded.c ot_batch.c ot_buf.c ot_event.c ot_fdstate.c ot_fmt.c ot_hex.c ot_ip.c ot_config.c ot_scan.c ot_socket.c ot_tcp.c ot_mutex.c ot_random.c ot_stats.c ot_vector.c ot_clean.c ot_udp.c ot_iovec.c ot_fullscrape.c ot_accesslist.c ot_http.c ot_livesync.c ot_rijndael.c
74SOURCES_proxy=proxy.c ot_batch.c ot_fdstate.c ot_vector.c ot_mutex.c ot_random.c ot_scan.c ot_ip.c ot_socket.c 74SOURCES_proxy=proxy.c ot_batch.c ot_fdstate.c ot_vector.c ot_mutex.c ot_random.c ot_scan.c ot_ip.c ot_socket.c
75 75
76all: $(BINARY) $(BINARY).debug 76all: $(BINARY) $(BINARY).debug
@@ -110,6 +110,12 @@ $(TESTBIN): $(TEST_SOURCES) $(TEST_UNITS) $(TEST_HEADERS)
110check: $(TESTBIN) $(BINARY) 110check: $(TESTBIN) $(BINARY)
111 ./$(TESTBIN) 111 ./$(TESTBIN)
112 112
113$(TESTBIN).poll: $(TEST_SOURCES) $(TEST_UNITS) $(TEST_HEADERS)
114 $(CC) -o $@ $(CFLAGS_test) -DOT_EVENT_FORCE_POLL $(TEST_SOURCES) $(TEST_UNITS)
115
116check-poll: $(TESTBIN).poll
117 ./$(TESTBIN).poll ot_event
118
113# The same unit tests under the sanitizers. Use clang over $(CC) because the 119# The same unit tests under the sanitizers. Use clang over $(CC) because the
114# two toolchains do not check the same things 120# two toolchains do not check the same things
115# Override CC_sanitize to use another compiler. 121# Override CC_sanitize to use another compiler.
@@ -153,13 +159,14 @@ check-tsan: opentracker.tsan
153# Everything, cheapest first, so a quick failure comes back quickly. 159# Everything, cheapest first, so a quick failure comes back quickly.
154check-all: 160check-all:
155 $(MAKE) check 161 $(MAKE) check
162 $(MAKE) check-poll
156 $(MAKE) check-sanitize 163 $(MAKE) check-sanitize
157 $(MAKE) check-integration 164 $(MAKE) check-integration
158 $(MAKE) check-tsan 165 $(MAKE) check-tsan
159 166
160.PHONY: all check check-sanitize check-integration check-tsan check-all clean install 167.PHONY: all check check-poll check-sanitize check-integration check-tsan check-all clean install
161 168
162BINARIES=opentracker opentracker.debug opentracker.integration opentracker.tsan proxy proxy.debug $(TESTBIN) $(TESTBIN).sanitize 169BINARIES=opentracker opentracker.debug opentracker.integration opentracker.tsan proxy proxy.debug $(TESTBIN) $(TESTBIN).sanitize $(TESTBIN).poll
163 170
164clean: 171clean:
165 rm -rf $(BINARIES) $(BINARIES:=.dSYM) *.o *~ 172 rm -rf $(BINARIES) $(BINARIES:=.dSYM) *.o *~
diff --git a/README b/README
index bb8a06e..c84c4e9 100644
--- a/README
+++ b/README
@@ -1,14 +1,8 @@
1This is opentracker. An open bittorrent tracker. 1This is opentracker. An open bittorrent tracker.
2 2
3You need libowfat (http://www.fefe.de/libowfat/).
4
5Steps to go: 3Steps to go:
6 4
7cvs -d :pserver:cvs@cvs.fefe.de:/cvs -z9 co libowfat 5git clone git://erdgeist.org/opentracker
8cd libowfat
9make
10cd ..
11cvs -d:pserver:anoncvs@cvs.erdgeist.org:/home/cvsroot co opentracker
12cd opentracker 6cd opentracker
13make 7make
14./opentracker 8./opentracker
@@ -24,7 +18,3 @@ sysctl kern.ipc.somaxconn=1024
24sysctl kern.ipc.nmbclusters=32768 18sysctl kern.ipc.nmbclusters=32768
25sysctl net.inet.tcp.msl=10000 19sysctl net.inet.tcp.msl=10000
26sysctl kern.maxfiles=10240 20sysctl kern.maxfiles=10240
27
28License information:
29
30Although the libowfat library is under GPL, Felix von Leitner agreed that the compiled binary may be distributed under the same beer ware license as the source code for opentracker. However, we like to hear from happy customers.
diff --git a/man4/opentracker.conf.4 b/man4/opentracker.conf.4
index b28879f..aa82e1d 100644
--- a/man4/opentracker.conf.4
+++ b/man4/opentracker.conf.4
@@ -100,6 +100,17 @@ group only where the answer is yes.
100.Pa tests/udp_scaling.py 100.Pa tests/udp_scaling.py
101in the source distribution sweeps the setting against a running tracker. 101in the source distribution sweeps the setting against a running tracker.
102Measure on the machine that will carry the traffic. 102Measure on the machine that will carry the traffic.
103.Pp
104On FreeBSD the
105.Xr accf_data 9
106filter has the kernel hold a new connection until data arrives on it, so the
107tracker is woken once with the request already there. opentracker uses it
108wherever it is available; enable it with
109.Cm options ACCEPT_FILTER_DATA
110in the kernel or
111.Va kld_list Ns = Ns Qq accf_data
112in
113.Xr rc.conf 5 .
103.Sh EXAMPLES 114.Sh EXAMPLES
104To specify the address opentracker will listen on for both TCP and UDP connections: 115To specify the address opentracker will listen on for both TCP and UDP connections:
105.Dl listen.tcp_udp 0.0.0.0:6969 116.Dl listen.tcp_udp 0.0.0.0:6969
diff --git a/opentracker.c b/opentracker.c
index 6160696..2eaf988 100644
--- a/opentracker.c
+++ b/opentracker.c
@@ -35,6 +35,7 @@
35#include "ot_fdstate.h" 35#include "ot_fdstate.h"
36#include "ot_socket.h" 36#include "ot_socket.h"
37#include "ot_stats.h" 37#include "ot_stats.h"
38#include "ot_tcp.h"
38#include "ot_udp.h" 39#include "ot_udp.h"
39#include "trackerlogic.h" 40#include "trackerlogic.h"
40 41
@@ -96,129 +97,6 @@ static void install_signal_handlers(void) {
96} 97}
97 98
98 99
99static ssize_t header_complete(char *request, ssize_t byte_count) {
100 ssize_t i = 0, state = 0;
101
102 for (i = 1; i < byte_count; i += 2)
103 if (request[i] <= 13) {
104 i--;
105 for (state = 0; i < byte_count; ++i) {
106 char c = request[i];
107 if (c == '\r' || c == '\n')
108 state = (state >> 2) | ((c << 6) & 0xc0);
109 else
110 break;
111 if (state >= 0xa0 || state == 0x99)
112 return i + 1;
113 }
114 }
115 return 0;
116}
117
118static void handle_read(const int64_t sock, struct ot_workstruct *ws) {
119 struct http_data *cookie = fdstate_cookie(sock);
120 ssize_t byte_count = read((int)sock, ws->inbuf, G_INBUF_SIZE);
121
122 if (byte_count == 0 || (byte_count == -1 && errno != EAGAIN && errno != EWOULDBLOCK && errno != EINTR)) {
123 http_close(sock);
124 return;
125 }
126
127 if (byte_count == -1)
128 return;
129
130 ws->request = NULL;
131 ws->request_size = 0;
132 ws->header_size = 0;
133 ws->keep_alive = 0;
134
135 /* If we get the whole request in one packet, handle it without copying */
136 if (!cookie->request.data) {
137 if ((ws->header_size = header_complete(ws->inbuf, byte_count))) {
138 ws->request = ws->inbuf;
139 ws->request_size = byte_count;
140 http_handle_request(sock, ws);
141 } else if (buf_cat(&cookie->request, ws->inbuf, (size_t)byte_count))
142 http_close(sock);
143 return;
144 }
145
146 if (buf_cat(&cookie->request, ws->inbuf, (size_t)byte_count) || cookie->request.size > G_INBUF_SIZE) {
147 http_issue_error(sock, ws, CODE_HTTPERROR_500);
148 return;
149 }
150
151 while ((ws->header_size = header_complete(cookie->request.data, (ssize_t)cookie->request.size))) {
152 ws->request = cookie->request.data;
153 ws->request_size = (ssize_t)cookie->request.size;
154 http_handle_request(sock, ws);
155#ifdef WANT_KEEPALIVE
156 if (!ws->keep_alive)
157#endif
158 return;
159 }
160}
161
162static void handle_write(const int64_t sock) {
163 struct http_data *cookie = fdstate_cookie(sock);
164 int chunked = 0;
165 int keep_alive = 0;
166
167 if (cookie) {
168 chunked = (cookie->flag & STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER) != 0;
169 keep_alive = (cookie->flag & STRUCT_HTTP_FLAG_KEEPALIVE) != 0;
170
171 if (cookie->batch) {
172 if (batch_send(sock, cookie->batch) == BATCH_FAILED) {
173 http_close(sock);
174 return;
175 }
176 if (batch_pending(cookie->batch))
177 return;
178 }
179 }
180
181 /* In a chunked transfer, wait for the next piece once everything queued has gone out */
182 if (chunked)
183 http_interest_drop(sock, cookie, EVENT_OUT);
184 else if (keep_alive) {
185 /* The reply is out and the client asked to keep the connection, so stop
186 waiting to write and go back to waiting for the next request */
187 http_interest_drop(sock, cookie, EVENT_OUT);
188 http_interest_add(sock, cookie, EVENT_IN);
189 fdstate_set_expiry(sock, time_future_seconds(OT_CLIENT_TIMEOUT));
190 } else
191 http_close(sock);
192}
193
194static void handle_accept(const int64_t serversocket, struct ot_workstruct *ws) {
195 struct http_data *cookie;
196 int64_t sock;
197 ot_ip6 ip;
198 uint16_t port;
199
200 while ((sock = socket_accept(serversocket, ip, &port)) != -1) {
201
202 socket_set_blocking((int)sock, 0);
203
204 if (!(cookie = (struct http_data *)calloc(1, sizeof(struct http_data)))) {
205 close((int)sock);
206 continue;
207 }
208 memcpy(cookie->ip, ip, sizeof(ot_ip6));
209 cookie->interest = EVENT_IN;
210
211 if (event_add((int)sock, EVENT_IN, fdstate_open(sock, cookie, time_future_seconds(OT_CLIENT_TIMEOUT))) == -1) {
212 fdstate_close(sock);
213 free(cookie);
214 close((int)sock);
215 continue;
216 }
217
218 stats_issue_event(ws, EVENT_ACCEPT, FLAG_TCP, (uintptr_t)ip);
219 }
220}
221
222/* How many ready descriptors one wait hands back at a time */ 100/* How many ready descriptors one wait hands back at a time */
223#define OT_EVENT_BATCH 512 101#define OT_EVENT_BATCH 512
224 102
@@ -264,7 +142,7 @@ static void *server_mainloop(void *args) {
264 cookie = fdstate_cookie((int)sock); 142 cookie = fdstate_cookie((int)sock);
265 143
266 if ((intptr_t)cookie == FLAG_TCP) 144 if ((intptr_t)cookie == FLAG_TCP)
267 handle_accept(sock, &ws); 145 tcp_accept(sock, &ws);
268 else if ((intptr_t)cookie == FLAG_UDP) 146 else if ((intptr_t)cookie == FLAG_UDP)
269 handle_udp6(sock, &ws); 147 handle_udp6(sock, &ws);
270 else if ((intptr_t)cookie == FLAG_SELFPIPE) 148 else if ((intptr_t)cookie == FLAG_SELFPIPE)
@@ -272,10 +150,10 @@ static void *server_mainloop(void *args) {
272 ; 150 ;
273 else { 151 else {
274 if (ready[i].what & EVENT_IN) 152 if (ready[i].what & EVENT_IN)
275 handle_read(sock, &ws); 153 tcp_read(sock, &ws);
276 /* The read may have closed it, so look again before writing */ 154 /* The read may have closed it, so look again before writing */
277 if ((ready[i].what & EVENT_OUT) && fdstate_generation((int)sock) == ready[i].generation) 155 if ((ready[i].what & EVENT_OUT) && fdstate_generation((int)sock) == ready[i].generation)
278 handle_write(sock); 156 tcp_write(sock);
279 } 157 }
280 } 158 }
281 159
@@ -342,6 +220,7 @@ static void listen_bind_one(ot_listen *entry) {
342 if (entry->proto == FLAG_TCP) { 220 if (entry->proto == FLAG_TCP) {
343 if ((entry->sock = socket_tcp_listen(entry->ip, entry->port, sharing)) == -1) 221 if ((entry->sock = socket_tcp_listen(entry->ip, entry->port, sharing)) == -1)
344 panic("socket_tcp_listen"); 222 panic("socket_tcp_listen");
223 socket_defer_accept((int)entry->sock);
345 } else { 224 } else {
346 if ((entry->sock = socket_udp_bind(entry->ip, entry->port, sharing)) == -1) 225 if ((entry->sock = socket_udp_bind(entry->ip, entry->port, sharing)) == -1)
347 panic("socket_udp_bind"); 226 panic("socket_udp_bind");
diff --git a/opentracker.xcodeproj/project.pbxproj b/opentracker.xcodeproj/project.pbxproj
deleted file mode 100644
index 3706a3f..0000000
--- a/opentracker.xcodeproj/project.pbxproj
+++ /dev/null
@@ -1,320 +0,0 @@
1// !$*UTF8*$!
2{
3 archiveVersion = 1;
4 classes = {
5 };
6 objectVersion = 44;
7 objects = {
8
9/* Begin PBXBuildFile section */
10 6520B7530D036AAF00A43B1F /* libowfat.a in Frameworks */ = {isa = PBXBuildFile; fileRef = 6520B7520D036AAF00A43B1F /* libowfat.a */; };
11 653A320C0CE7F475007F0D03 /* ot_accesslist.c in Sources */ = {isa = PBXBuildFile; fileRef = 653A320B0CE7F475007F0D03 /* ot_accesslist.c */; };
12 653A56B50CE28EC5000CF140 /* ot_iovec.c in Sources */ = {isa = PBXBuildFile; fileRef = 653A56B40CE28EC5000CF140 /* ot_iovec.c */; };
13 654A80890CD832FD009035DE /* opentracker.c in Sources */ = {isa = PBXBuildFile; fileRef = 654A80840CD832FC009035DE /* opentracker.c */; };
14 654A808A0CD832FD009035DE /* scan_urlencoded_query.c in Sources */ = {isa = PBXBuildFile; fileRef = 654A80850CD832FC009035DE /* scan_urlencoded_query.c */; };
15 654A808B0CD832FD009035DE /* trackerlogic.c in Sources */ = {isa = PBXBuildFile; fileRef = 654A80870CD832FC009035DE /* trackerlogic.c */; };
16 65542D8B0CE078E800469330 /* ot_vector.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542D8A0CE078E800469330 /* ot_vector.c */; };
17 65542D930CE07CED00469330 /* ot_mutex.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542D8F0CE07CED00469330 /* ot_mutex.c */; };
18 65542D940CE07CED00469330 /* ot_stats.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542D910CE07CED00469330 /* ot_stats.c */; };
19 65542E750CE08B9100469330 /* ot_clean.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542E740CE08B9100469330 /* ot_clean.c */; };
20 65542EE80CE0CA6B00469330 /* ot_udp.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542EE70CE0CA6B00469330 /* ot_udp.c */; };
21 65542F920CE17CA900469330 /* ot_fullscrape.c in Sources */ = {isa = PBXBuildFile; fileRef = 65542F910CE17CA900469330 /* ot_fullscrape.c */; };
22 65B8DF3C0D0310D20017149E /* ot_http.c in Sources */ = {isa = PBXBuildFile; fileRef = 65B8DF3B0D0310D20017149E /* ot_http.c */; };
23 65FA33990E7EF09200F7D5A5 /* ot_livesync.c in Sources */ = {isa = PBXBuildFile; fileRef = 65FA33980E7EF09200F7D5A5 /* ot_livesync.c */; };
24 8DD76FB00486AB0100D96B5E /* opentracker.1 in CopyFiles */ = {isa = PBXBuildFile; fileRef = C6A0FF2C0290799A04C91782 /* opentracker.1 */; };
25/* End PBXBuildFile section */
26
27/* Begin PBXCopyFilesBuildPhase section */
28 8DD76FAF0486AB0100D96B5E /* CopyFiles */ = {
29 isa = PBXCopyFilesBuildPhase;
30 buildActionMask = 8;
31 dstPath = /usr/share/man/man1/;
32 dstSubfolderSpec = 0;
33 files = (
34 8DD76FB00486AB0100D96B5E /* opentracker.1 in CopyFiles */,
35 );
36 runOnlyForDeploymentPostprocessing = 1;
37 };
38/* End PBXCopyFilesBuildPhase section */
39
40/* Begin PBXFileReference section */
41 6520B7520D036AAF00A43B1F /* libowfat.a */ = {isa = PBXFileReference; lastKnownFileType = archive.ar; name = libowfat.a; path = ../libowfat/libowfat.a; sourceTree = SOURCE_ROOT; };
42 653A320A0CE7F475007F0D03 /* ot_accesslist.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_accesslist.h; sourceTree = "<group>"; };
43 653A320B0CE7F475007F0D03 /* ot_accesslist.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_accesslist.c; sourceTree = "<group>"; };
44 653A56AC0CE201FF000CF140 /* opentracker */ = {isa = PBXFileReference; explicitFileType = "compiled.mach-o.executable"; includeInIndex = 0; path = opentracker; sourceTree = BUILT_PRODUCTS_DIR; };
45 653A56B30CE28EC5000CF140 /* ot_iovec.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_iovec.h; sourceTree = "<group>"; };
46 653A56B40CE28EC5000CF140 /* ot_iovec.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_iovec.c; sourceTree = "<group>"; };
47 654A80840CD832FC009035DE /* opentracker.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = opentracker.c; sourceTree = "<group>"; };
48 654A80850CD832FC009035DE /* scan_urlencoded_query.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = scan_urlencoded_query.c; sourceTree = "<group>"; };
49 654A80860CD832FC009035DE /* scan_urlencoded_query.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = scan_urlencoded_query.h; sourceTree = "<group>"; };
50 654A80870CD832FC009035DE /* trackerlogic.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = trackerlogic.c; sourceTree = "<group>"; };
51 654A80880CD832FC009035DE /* trackerlogic.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = trackerlogic.h; sourceTree = "<group>"; };
52 65542D890CE078E800469330 /* ot_vector.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_vector.h; sourceTree = "<group>"; };
53 65542D8A0CE078E800469330 /* ot_vector.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_vector.c; sourceTree = "<group>"; };
54 65542D8F0CE07CED00469330 /* ot_mutex.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_mutex.c; sourceTree = "<group>"; };
55 65542D900CE07CED00469330 /* ot_mutex.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_mutex.h; sourceTree = "<group>"; };
56 65542D910CE07CED00469330 /* ot_stats.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_stats.c; sourceTree = "<group>"; };
57 65542D920CE07CED00469330 /* ot_stats.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_stats.h; sourceTree = "<group>"; };
58 65542E730CE08B9100469330 /* ot_clean.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_clean.h; sourceTree = "<group>"; };
59 65542E740CE08B9100469330 /* ot_clean.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_clean.c; sourceTree = "<group>"; };
60 65542EE60CE0CA6B00469330 /* ot_udp.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_udp.h; sourceTree = "<group>"; };
61 65542EE70CE0CA6B00469330 /* ot_udp.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_udp.c; sourceTree = "<group>"; };
62 65542F900CE17CA900469330 /* ot_fullscrape.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_fullscrape.h; sourceTree = "<group>"; };
63 65542F910CE17CA900469330 /* ot_fullscrape.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_fullscrape.c; sourceTree = "<group>"; };
64 65B8DF3A0D0310D20017149E /* ot_http.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_http.h; sourceTree = "<group>"; };
65 65B8DF3B0D0310D20017149E /* ot_http.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_http.c; sourceTree = "<group>"; };
66 65FA33970E7EF09200F7D5A5 /* ot_livesync.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = ot_livesync.h; sourceTree = "<group>"; };
67 65FA33980E7EF09200F7D5A5 /* ot_livesync.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = ot_livesync.c; sourceTree = "<group>"; };
68 C6A0FF2C0290799A04C91782 /* opentracker.1 */ = {isa = PBXFileReference; lastKnownFileType = text.man; path = opentracker.1; sourceTree = "<group>"; };
69/* End PBXFileReference section */
70
71/* Begin PBXFrameworksBuildPhase section */
72 8DD76FAD0486AB0100D96B5E /* Frameworks */ = {
73 isa = PBXFrameworksBuildPhase;
74 buildActionMask = 2147483647;
75 files = (
76 6520B7530D036AAF00A43B1F /* libowfat.a in Frameworks */,
77 );
78 runOnlyForDeploymentPostprocessing = 0;
79 };
80/* End PBXFrameworksBuildPhase section */
81
82/* Begin PBXGroup section */
83 08FB7794FE84155DC02AAC07 /* opentracker */ = {
84 isa = PBXGroup;
85 children = (
86 65542D810CE0786F00469330 /* Headers */,
87 08FB7795FE84155DC02AAC07 /* Source */,
88 92762AC9104EDED700FDCB60 /* Libraries */,
89 C6A0FF2B0290797F04C91782 /* Documentation */,
90 653A56AD0CE201FF000CF140 /* Products */,
91 );
92 name = opentracker;
93 sourceTree = "<group>";
94 };
95 08FB7795FE84155DC02AAC07 /* Source */ = {
96 isa = PBXGroup;
97 children = (
98 654A80840CD832FC009035DE /* opentracker.c */,
99 653A320B0CE7F475007F0D03 /* ot_accesslist.c */,
100 65542E740CE08B9100469330 /* ot_clean.c */,
101 65542F910CE17CA900469330 /* ot_fullscrape.c */,
102 65B8DF3B0D0310D20017149E /* ot_http.c */,
103 65FA33980E7EF09200F7D5A5 /* ot_livesync.c */,
104 653A56B40CE28EC5000CF140 /* ot_iovec.c */,
105 65542D8F0CE07CED00469330 /* ot_mutex.c */,
106 65542D910CE07CED00469330 /* ot_stats.c */,
107 65542EE70CE0CA6B00469330 /* ot_udp.c */,
108 65542D8A0CE078E800469330 /* ot_vector.c */,
109 654A80850CD832FC009035DE /* scan_urlencoded_query.c */,
110 654A80870CD832FC009035DE /* trackerlogic.c */,
111 );
112 name = Source;
113 sourceTree = "<group>";
114 };
115 653A56AD0CE201FF000CF140 /* Products */ = {
116 isa = PBXGroup;
117 children = (
118 653A56AC0CE201FF000CF140 /* opentracker */,
119 );
120 name = Products;
121 sourceTree = "<group>";
122 };
123 65542D810CE0786F00469330 /* Headers */ = {
124 isa = PBXGroup;
125 children = (
126 653A320A0CE7F475007F0D03 /* ot_accesslist.h */,
127 65542E730CE08B9100469330 /* ot_clean.h */,
128 65542F900CE17CA900469330 /* ot_fullscrape.h */,
129 65B8DF3A0D0310D20017149E /* ot_http.h */,
130 653A56B30CE28EC5000CF140 /* ot_iovec.h */,
131 65FA33970E7EF09200F7D5A5 /* ot_livesync.h */,
132 65542D900CE07CED00469330 /* ot_mutex.h */,
133 65542D920CE07CED00469330 /* ot_stats.h */,
134 65542EE60CE0CA6B00469330 /* ot_udp.h */,
135 65542D890CE078E800469330 /* ot_vector.h */,
136 654A80860CD832FC009035DE /* scan_urlencoded_query.h */,
137 654A80880CD832FC009035DE /* trackerlogic.h */,
138 );
139 name = Headers;
140 sourceTree = "<group>";
141 };
142 92762AC9104EDED700FDCB60 /* Libraries */ = {
143 isa = PBXGroup;
144 children = (
145 6520B7520D036AAF00A43B1F /* libowfat.a */,
146 );
147 name = Libraries;
148 sourceTree = "<group>";
149 };
150 C6A0FF2B0290797F04C91782 /* Documentation */ = {
151 isa = PBXGroup;
152 children = (
153 C6A0FF2C0290799A04C91782 /* opentracker.1 */,
154 );
155 name = Documentation;
156 sourceTree = "<group>";
157 };
158/* End PBXGroup section */
159
160/* Begin PBXNativeTarget section */
161 8DD76FA90486AB0100D96B5E /* opentracker */ = {
162 isa = PBXNativeTarget;
163 buildConfigurationList = 1DEB928508733DD80010E9CD /* Build configuration list for PBXNativeTarget "opentracker" */;
164 buildPhases = (
165 8DD76FAB0486AB0100D96B5E /* Sources */,
166 8DD76FAD0486AB0100D96B5E /* Frameworks */,
167 8DD76FAF0486AB0100D96B5E /* CopyFiles */,
168 );
169 buildRules = (
170 );
171 dependencies = (
172 );
173 name = opentracker;
174 productInstallPath = "$(HOME)/bin";
175 productName = opentracker;
176 productReference = 653A56AC0CE201FF000CF140 /* opentracker */;
177 productType = "com.apple.product-type.tool";
178 };
179/* End PBXNativeTarget section */
180
181/* Begin PBXProject section */
182 08FB7793FE84155DC02AAC07 /* Project object */ = {
183 isa = PBXProject;
184 buildConfigurationList = 1DEB928908733DD80010E9CD /* Build configuration list for PBXProject "opentracker" */;
185 compatibilityVersion = "Xcode 3.0";
186 hasScannedForEncodings = 1;
187 mainGroup = 08FB7794FE84155DC02AAC07 /* opentracker */;
188 productRefGroup = 653A56AD0CE201FF000CF140 /* Products */;
189 projectDirPath = "";
190 projectRoot = "";
191 targets = (
192 8DD76FA90486AB0100D96B5E /* opentracker */,
193 );
194 };
195/* End PBXProject section */
196
197/* Begin PBXSourcesBuildPhase section */
198 8DD76FAB0486AB0100D96B5E /* Sources */ = {
199 isa = PBXSourcesBuildPhase;
200 buildActionMask = 2147483647;
201 files = (
202 654A80890CD832FD009035DE /* opentracker.c in Sources */,
203 654A808A0CD832FD009035DE /* scan_urlencoded_query.c in Sources */,
204 654A808B0CD832FD009035DE /* trackerlogic.c in Sources */,
205 65542D8B0CE078E800469330 /* ot_vector.c in Sources */,
206 65542D930CE07CED00469330 /* ot_mutex.c in Sources */,
207 65542D940CE07CED00469330 /* ot_stats.c in Sources */,
208 65542E750CE08B9100469330 /* ot_clean.c in Sources */,
209 65542EE80CE0CA6B00469330 /* ot_udp.c in Sources */,
210 65542F920CE17CA900469330 /* ot_fullscrape.c in Sources */,
211 653A56B50CE28EC5000CF140 /* ot_iovec.c in Sources */,
212 653A320C0CE7F475007F0D03 /* ot_accesslist.c in Sources */,
213 65B8DF3C0D0310D20017149E /* ot_http.c in Sources */,
214 65FA33990E7EF09200F7D5A5 /* ot_livesync.c in Sources */,
215 );
216 runOnlyForDeploymentPostprocessing = 0;
217 };
218/* End PBXSourcesBuildPhase section */
219
220/* Begin XCBuildConfiguration section */
221 1DEB928608733DD80010E9CD /* Debug */ = {
222 isa = XCBuildConfiguration;
223 buildSettings = {
224 COPY_PHASE_STRIP = NO;
225 GCC_DYNAMIC_NO_PIC = NO;
226 GCC_ENABLE_FIX_AND_CONTINUE = YES;
227 GCC_MODEL_TUNING = G5;
228 GCC_OPTIMIZATION_LEVEL = 0;
229 INSTALL_PATH = /usr/local/bin;
230 LIBRARY_SEARCH_PATHS = (
231 "$(inherited)",
232 "\"$(SRCROOT)/../libowfat\"",
233 );
234 PRODUCT_NAME = opentracker;
235 ZERO_LINK = YES;
236 };
237 name = Debug;
238 };
239 1DEB928708733DD80010E9CD /* Release */ = {
240 isa = XCBuildConfiguration;
241 buildSettings = {
242 DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
243 GCC_ENABLE_FIX_AND_CONTINUE = YES;
244 GCC_MODEL_TUNING = G5;
245 INSTALL_PATH = /usr/local/bin;
246 LIBRARY_SEARCH_PATHS = (
247 "$(inherited)",
248 "\"$(SRCROOT)/../libowfat\"",
249 );
250 PRODUCT_NAME = opentracker;
251 };
252 name = Release;
253 };
254 1DEB928A08733DD80010E9CD /* Debug */ = {
255 isa = XCBuildConfiguration;
256 buildSettings = {
257 ARCHS = "$(ONLY_ACTIVE_ARCH_PRE_XCODE_3_1)";
258 GCC_PREPROCESSOR_DEFINITIONS = (
259 WANT_ACCESSLIST_WHITE,
260 WANT_IP_FROM_QUERY_STRING,
261 WANT_FULLSCRAPE,
262 );
263 GCC_VERSION = com.apple.compilers.llvm.clang.1_0;
264 GCC_WARN_ABOUT_RETURN_TYPE = YES;
265 GCC_WARN_UNUSED_VARIABLE = YES;
266 HEADER_SEARCH_PATHS = ../libowfat/;
267 LIBRARY_SEARCH_PATHS = ../libowfat/;
268 ONLY_ACTIVE_ARCH_PRE_XCODE_3_1 = "$(NATIVE_ARCH_ACTUAL)";
269 PREBINDING = NO;
270 SDKROOT = "$(DEVELOPER_SDK_DIR)/MacOSX10.6.sdk";
271 };
272 name = Debug;
273 };
274 1DEB928B08733DD80010E9CD /* Release */ = {
275 isa = XCBuildConfiguration;
276 buildSettings = {
277 ARCHS = "$(ONLY_ACTIVE_ARCH_PRE_XCODE_3_1)";
278 DEAD_CODE_STRIPPING = NO;
279 GCC_PREPROCESSOR_DEFINITIONS = (
280 WANT_ACCESSLIST_WHITE,
281 WANT_IP_FROM_QUERY_STRING,
282 WANT_FULLSCRAPE,
283 );
284 GCC_VERSION = com.apple.compilers.llvm.clang.1_0;
285 GCC_WARN_ABOUT_RETURN_TYPE = YES;
286 GCC_WARN_UNUSED_VARIABLE = YES;
287 HEADER_SEARCH_PATHS = ../libowfat/;
288 LIBRARY_SEARCH_PATHS = ../libowfat/;
289 ONLY_ACTIVE_ARCH_PRE_XCODE_3_1 = "$(NATIVE_ARCH_ACTUAL)";
290 PREBINDING = NO;
291 SDKROOT = "$(DEVELOPER_SDK_DIR)/MacOSX10.6.sdk";
292 ZERO_LINK = NO;
293 };
294 name = Release;
295 };
296/* End XCBuildConfiguration section */
297
298/* Begin XCConfigurationList section */
299 1DEB928508733DD80010E9CD /* Build configuration list for PBXNativeTarget "opentracker" */ = {
300 isa = XCConfigurationList;
301 buildConfigurations = (
302 1DEB928608733DD80010E9CD /* Debug */,
303 1DEB928708733DD80010E9CD /* Release */,
304 );
305 defaultConfigurationIsVisible = 0;
306 defaultConfigurationName = Release;
307 };
308 1DEB928908733DD80010E9CD /* Build configuration list for PBXProject "opentracker" */ = {
309 isa = XCConfigurationList;
310 buildConfigurations = (
311 1DEB928A08733DD80010E9CD /* Debug */,
312 1DEB928B08733DD80010E9CD /* Release */,
313 );
314 defaultConfigurationIsVisible = 0;
315 defaultConfigurationName = Release;
316 };
317/* End XCConfigurationList section */
318 };
319 rootObject = 08FB7793FE84155DC02AAC07 /* Project object */;
320}
diff --git a/ot_config.c b/ot_config.c
index a0dbef1..272acf1 100644
--- a/ot_config.c
+++ b/ot_config.c
@@ -14,6 +14,7 @@
14 14
15/* Opentracker */ 15/* Opentracker */
16#include "ot_accesslist.h" 16#include "ot_accesslist.h"
17#include "ot_event.h"
17#include "ot_config.h" 18#include "ot_config.h"
18#include "ot_http.h" 19#include "ot_http.h"
19#include "ot_ip.h" 20#include "ot_ip.h"
@@ -32,7 +33,8 @@ const char *g_statefile;
32static ot_listen *g_listen; 33static ot_listen *g_listen;
33static size_t g_listen_count; 34static size_t g_listen_count;
34static int g_listen_shared; 35static int g_listen_shared;
35static int g_listen_workers = 1; 36static int g_listen_workers_udp = 1;
37static int g_listen_workers_tcp = 0;
36 38
37static void panic(const char *routine) __attribute__((noreturn)); 39static void panic(const char *routine) __attribute__((noreturn));
38static void panic(const char *routine) { 40static void panic(const char *routine) {
@@ -196,6 +198,7 @@ typedef enum {
196 CONFIG_LISTEN_TCP, 198 CONFIG_LISTEN_TCP,
197 CONFIG_LISTEN_UDP, 199 CONFIG_LISTEN_UDP,
198 CONFIG_LISTEN_UDP_WORKERS, 200 CONFIG_LISTEN_UDP_WORKERS,
201 CONFIG_LISTEN_TCP_WORKERS,
199 CONFIG_LISTEN_SHARED, 202 CONFIG_LISTEN_SHARED,
200 CONFIG_ACCESS_WHITELIST, 203 CONFIG_ACCESS_WHITELIST,
201 CONFIG_ACCESS_BLACKLIST, 204 CONFIG_ACCESS_BLACKLIST,
@@ -224,6 +227,7 @@ static const config_key config_keys[] = {
224 CONFIG_KEY("listen.tcp", CONFIG_LISTEN_TCP), 227 CONFIG_KEY("listen.tcp", CONFIG_LISTEN_TCP),
225 CONFIG_KEY("listen.udp", CONFIG_LISTEN_UDP), 228 CONFIG_KEY("listen.udp", CONFIG_LISTEN_UDP),
226 CONFIG_KEY("listen.udp.workers", CONFIG_LISTEN_UDP_WORKERS), 229 CONFIG_KEY("listen.udp.workers", CONFIG_LISTEN_UDP_WORKERS),
230 CONFIG_KEY("listen.tcp.workers", CONFIG_LISTEN_TCP_WORKERS),
227 CONFIG_KEY("listen.shared", CONFIG_LISTEN_SHARED), 231 CONFIG_KEY("listen.shared", CONFIG_LISTEN_SHARED),
228#ifdef WANT_ACCESSLIST_WHITE 232#ifdef WANT_ACCESSLIST_WHITE
229 CONFIG_KEY("access.whitelist", CONFIG_ACCESS_WHITELIST), 233 CONFIG_KEY("access.whitelist", CONFIG_ACCESS_WHITELIST),
@@ -318,7 +322,7 @@ static int parse_configfile(char *config_filename) {
318 if (!off || !listen_scan_options(p + off, &shared, &workers)) 322 if (!off || !listen_scan_options(p + off, &shared, &workers))
319 goto parse_error; 323 goto parse_error;
320 if (kind != CONFIG_LISTEN_UDP) 324 if (kind != CONFIG_LISTEN_UDP)
321 listen_append(tmpip, tmpport, FLAG_TCP, shared, OT_LISTEN_INHERIT), ++bound; 325 listen_append(tmpip, tmpport, FLAG_TCP, shared, workers), ++bound;
322 if (kind != CONFIG_LISTEN_TCP) 326 if (kind != CONFIG_LISTEN_TCP)
323 listen_append(tmpip, tmpport, FLAG_UDP, shared, workers), ++bound; 327 listen_append(tmpip, tmpport, FLAG_UDP, shared, workers), ++bound;
324 break; 328 break;
@@ -326,7 +330,12 @@ static int parse_configfile(char *config_filename) {
326 case CONFIG_LISTEN_UDP_WORKERS: 330 case CONFIG_LISTEN_UDP_WORKERS:
327 if (!scan_uint(p, &value, UINT_MAX)) 331 if (!scan_uint(p, &value, UINT_MAX))
328 goto parse_error; 332 goto parse_error;
329 g_listen_workers = (int)value; 333 g_listen_workers_udp = (int)value;
334 break;
335 case CONFIG_LISTEN_TCP_WORKERS:
336 if (!scan_uint(p, &value, UINT_MAX))
337 goto parse_error;
338 g_listen_workers_tcp = (int)value;
330 break; 339 break;
331 case CONFIG_LISTEN_SHARED: 340 case CONFIG_LISTEN_SHARED:
332 if (!scan_uint(p, &value, 1)) 341 if (!scan_uint(p, &value, 1))
@@ -446,7 +455,16 @@ static void listen_resolve_defaults(void) {
446 if (entry->shared == OT_LISTEN_INHERIT) 455 if (entry->shared == OT_LISTEN_INHERIT)
447 entry->shared = g_listen_shared; 456 entry->shared = g_listen_shared;
448 if (entry->workers == OT_LISTEN_INHERIT) 457 if (entry->workers == OT_LISTEN_INHERIT)
449 entry->workers = g_listen_workers; 458 entry->workers = entry->proto == FLAG_TCP ? g_listen_workers_tcp : g_listen_workers_udp;
459
460 /* A tcp worker hands a connection it cannot finish to the event loop,
461 which only one thread may touch where the set is ours rather than the
462 kernel's */
463 if (entry->proto == FLAG_TCP && entry->workers > 1 && !event_can_share()) {
464 fprintf(stderr, "Warning: Multiple workers impossible on this platform with event mechanism %s chosen.\n",
465 event_mechanism());
466 entry->workers = 1;
467 }
450 } 468 }
451} 469}
452 470
diff --git a/ot_event.c b/ot_event.c
index 8f55817..61cc07b 100644
--- a/ot_event.c
+++ b/ot_event.c
@@ -40,6 +40,9 @@ static struct epoll_event *g_ready;
40static int g_ready_space; 40static int g_ready_space;
41 41
42const char *event_mechanism(void) { return "epoll"; } 42const char *event_mechanism(void) { return "epoll"; }
43/* The set lives in the kernel, and epoll_ctl during another thread's
44 epoll_wait is what it is for */
45int event_can_share(void) { return 1; }
43 46
44int event_init(void) { 47int event_init(void) {
45 g_ready = NULL; 48 g_ready = NULL;
@@ -118,6 +121,8 @@ static struct kevent *g_ready;
118static int g_ready_space; 121static int g_ready_space;
119 122
120const char *event_mechanism(void) { return "kqueue"; } 123const char *event_mechanism(void) { return "kqueue"; }
124/* The set lives in the kernel, and kevent may be called from any thread */
125int event_can_share(void) { return 1; }
121 126
122int event_init(void) { 127int event_init(void) {
123 g_ready = NULL; 128 g_ready = NULL;
@@ -171,7 +176,7 @@ static int event_change(int fd, int what, uint32_t generation, int deleting) {
171 return 0; 176 return 0;
172} 177}
173 178
174int event_add(int fd, int what, uint32_t generation) { return event_change(fd, what, generation, 0); } 179int event_add(int fd, int what, uint32_t generation) { return what ? event_change(fd, what, generation, 0) : 0; }
175int event_modify(int fd, int what, uint32_t generation) { return event_change(fd, what, generation, 0); } 180int event_modify(int fd, int what, uint32_t generation) { return event_change(fd, what, generation, 0); }
176int event_del(int fd) { return event_change(fd, 0, 0, 1); } 181int event_del(int fd) { return event_change(fd, 0, 0, 1); }
177 182
@@ -229,6 +234,8 @@ static int *g_where;
229static int g_count, g_space, g_where_space; 234static int g_count, g_space, g_where_space;
230 235
231const char *event_mechanism(void) { return "poll"; } 236const char *event_mechanism(void) { return "poll"; }
237/* The arrays below are ours and nothing guards them */
238int event_can_share(void) { return 0; }
232 239
233int event_init(void) { 240int event_init(void) {
234 g_pfds = NULL, g_generation = NULL, g_where = NULL; 241 g_pfds = NULL, g_generation = NULL, g_where = NULL;
diff --git a/ot_event.h b/ot_event.h
index aa007d0..efdefff 100644
--- a/ot_event.h
+++ b/ot_event.h
@@ -38,6 +38,11 @@ int event_close(int fd);
38 out with up to max ready descriptors. Returns how many, or -1. */ 38 out with up to max ready descriptors. Returns how many, or -1. */
39int event_wait(ot_event *out, int max, int timeout_ms); 39int event_wait(ot_event *out, int max, int timeout_ms);
40 40
41/* Whether the registered set may be changed from another thread while this
42 one is waiting on it. True where the kernel keeps the set, false where
43 this module keeps it in memory of its own. */
44int event_can_share(void);
45
41/* Which mechanism was compiled in, for the version string */ 46/* Which mechanism was compiled in, for the version string */
42const char *event_mechanism(void); 47const char *event_mechanism(void);
43 48
diff --git a/ot_fdstate.c b/ot_fdstate.c
index 537bcb0..7d742da 100644
--- a/ot_fdstate.c
+++ b/ot_fdstate.c
@@ -186,6 +186,13 @@ void *fdstate_cookie(int fd) {
186 return slot ? atomic_load_explicit(&slot->cookie, memory_order_relaxed) : NULL; 186 return slot ? atomic_load_explicit(&slot->cookie, memory_order_relaxed) : NULL;
187} 187}
188 188
189void fdstate_set_cookie(int fd, void *cookie) {
190 ot_fdslot *slot = fdstate_slot(fd);
191
192 if (slot)
193 atomic_store_explicit(&slot->cookie, cookie, memory_order_relaxed);
194}
195
189void fdstate_set_expiry(int fd, ot_time expiry) { 196void fdstate_set_expiry(int fd, ot_time expiry) {
190 ot_fdslot *slot = fdstate_slot(fd); 197 ot_fdslot *slot = fdstate_slot(fd);
191 198
diff --git a/ot_fdstate.h b/ot_fdstate.h
index c09f018..93890d1 100644
--- a/ot_fdstate.h
+++ b/ot_fdstate.h
@@ -56,6 +56,9 @@ int fdstate_next_expired(ot_time now, int *cursor);
56 56
57/* Both only valid while the caller holds the descriptor. */ 57/* Both only valid while the caller holds the descriptor. */
58void *fdstate_cookie(int fd); 58void *fdstate_cookie(int fd);
59
60/* Hand the slot a different cookie, for a connection whose state has moved */
61void fdstate_set_cookie(int fd, void *cookie);
59void fdstate_set_expiry(int fd, ot_time expiry); 62void fdstate_set_expiry(int fd, ot_time expiry);
60 63
61#endif 64#endif
diff --git a/ot_http.c b/ot_http.c
index d4c7363..bb6ff24 100644
--- a/ot_http.c
+++ b/ot_http.c
@@ -6,6 +6,7 @@
6/* System */ 6/* System */
7#define _GNU_SOURCE 7#define _GNU_SOURCE
8#include <arpa/inet.h> 8#include <arpa/inet.h>
9#include <errno.h>
9#include <pthread.h> 10#include <pthread.h>
10#include <stdio.h> 11#include <stdio.h>
11#include <stdlib.h> 12#include <stdlib.h>
@@ -38,9 +39,25 @@ ssize_t g_stats_path_len;
38 39
39enum { SUCCESS_HTTP_HEADER_LENGTH = 80, SUCCESS_HTTP_SIZE_OFF = 17 }; 40enum { SUCCESS_HTTP_HEADER_LENGTH = 80, SUCCESS_HTTP_SIZE_OFF = 17 };
40 41
41/* Give up a connection: the one place that knows everything hanging off it. 42int http_adopt(const int64_t sock, struct http_data *cookie) {
42 The slot is released before anything is freed, so that a late event or the 43 struct http_data *heap = malloc(sizeof(*heap));
43 expiry sweep finds a closed descriptor rather than a freed cookie. */ 44
45 if (!heap)
46 return -1;
47
48 *heap = *cookie;
49 heap->flag &= ~STRUCT_HTTP_FLAG_ONSTACK;
50 fdstate_set_cookie((int)sock, heap);
51
52 if (event_add((int)sock, heap->interest, fdstate_generation((int)sock)) == -1) {
53 fdstate_set_cookie((int)sock, cookie);
54 free(heap);
55 return -1;
56 }
57 return 0;
58}
59
60/* Give up a connection: the one place that knows everything hanging off it. */
44void http_close(const int64_t sock) { 61void http_close(const int64_t sock) {
45 struct http_data *cookie = fdstate_cookie(sock); 62 struct http_data *cookie = fdstate_cookie(sock);
46 63
@@ -53,7 +70,8 @@ void http_close(const int64_t sock) {
53 buf_reset(&cookie->request); 70 buf_reset(&cookie->request);
54 if (cookie->flag & (STRUCT_HTTP_FLAG_WAITINGFORTASK | STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER)) 71 if (cookie->flag & (STRUCT_HTTP_FLAG_WAITINGFORTASK | STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER))
55 mutex_workqueue_canceltask(sock); 72 mutex_workqueue_canceltask(sock);
56 free(cookie); 73 if (!(cookie->flag & STRUCT_HTTP_FLAG_ONSTACK))
74 free(cookie);
57 } 75 }
58 76
59 event_close((int)sock); 77 event_close((int)sock);
@@ -85,6 +103,12 @@ static void http_senddata(const int64_t sock, struct ot_workstruct *ws) {
85 buf_reset(&cookie->request); 103 buf_reset(&cookie->request);
86 104
87 written_size = write(sock, ws->reply, ws->reply_size); 105 written_size = write(sock, ws->reply, ws->reply_size);
106 /* A send buffer with no room in it is not a failure: nothing went out, so
107 the whole reply goes into the batch below and leaves when the socket
108 drains. Only a real error ends the connection here. */
109 if (written_size < 0 && (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR))
110 written_size = 0;
111
88 if ((written_size < 0) || ((written_size == ws->reply_size) && !ws->keep_alive)) { 112 if ((written_size < 0) || ((written_size == ws->reply_size) && !ws->keep_alive)) {
89 http_close(sock); 113 http_close(sock);
90 return; 114 return;
@@ -111,11 +135,9 @@ static void http_senddata(const int64_t sock, struct ot_workstruct *ws) {
111 batch_add(cookie->batch, outbuf, ws->reply_size - written_size); 135 batch_add(cookie->batch, outbuf, ws->reply_size - written_size);
112 136
113 /* writeable short data sockets just have a tcp timeout */ 137 /* writeable short data sockets just have a tcp timeout */
114 if (!ws->keep_alive) { 138 if (!ws->keep_alive)
115 fdstate_set_expiry(sock, 0); 139 fdstate_set_expiry(sock, 0);
116 http_interest_drop(sock, cookie, EVENT_IN); 140 http_interest_set(sock, cookie, EVENT_OUT);
117 }
118 http_interest_add(sock, cookie, EVENT_OUT);
119 } 141 }
120} 142}
121 143
@@ -128,21 +150,20 @@ static void http_senddata(const int64_t sock, struct ot_workstruct *ws) {
128#define HTTPERROR_403_IP return http_issue_error(sock, ws, CODE_HTTPERROR_403_IP) 150#define HTTPERROR_403_IP return http_issue_error(sock, ws, CODE_HTTPERROR_403_IP)
129#define HTTPERROR_404 return http_issue_error(sock, ws, CODE_HTTPERROR_404) 151#define HTTPERROR_404 return http_issue_error(sock, ws, CODE_HTTPERROR_404)
130#define HTTPERROR_500 return http_issue_error(sock, ws, CODE_HTTPERROR_500) 152#define HTTPERROR_500 return http_issue_error(sock, ws, CODE_HTTPERROR_500)
131static void http_interest(const int64_t sock, struct http_data *cookie, int what) { 153/* A connection is either being read or being written, never both, so what it
154 waits for is set outright rather than added to and taken away again. The
155 event layer only hears about it when the answer changes. */
156void http_interest_set(const int64_t sock, struct http_data *cookie, int what) {
132 if (cookie->interest == what) 157 if (cookie->interest == what)
133 return; 158 return;
134 cookie->interest = what; 159 cookie->interest = what;
160 /* Still on a stack, so not in the event loop yet: whoever adopts this
161 connection registers whatever it asked for last */
162 if (cookie->flag & STRUCT_HTTP_FLAG_ONSTACK)
163 return;
135 event_modify((int)sock, what, fdstate_generation((int)sock)); 164 event_modify((int)sock, what, fdstate_generation((int)sock));
136} 165}
137 166
138void http_interest_add(const int64_t sock, struct http_data *cookie, int what) {
139 http_interest(sock, cookie, cookie->interest | what);
140}
141
142void http_interest_drop(const int64_t sock, struct http_data *cookie, int what) {
143 http_interest(sock, cookie, cookie->interest & ~what);
144}
145
146ssize_t http_issue_error(const int64_t sock, struct ot_workstruct *ws, int code) { 167ssize_t http_issue_error(const int64_t sock, struct ot_workstruct *ws, int code) {
147 const char *error_code[] = {"302 Found", "400 Invalid Request", "400 Invalid Request", "400 Invalid Request", "402 Payment Required", 168 const char *error_code[] = {"302 Found", "400 Invalid Request", "400 Invalid Request", "400 Invalid Request", "402 Payment Required",
148 "403 Not Modest", "403 Access Denied", "404 Not Found", "500 Internal Server Error"}; 169 "403 Not Modest", "403 Access Denied", "404 Not Found", "500 Internal Server Error"};
@@ -187,7 +208,7 @@ ssize_t http_sendiovecdata(const int64_t sock, struct ot_workstruct *ws, size_t
187 208
188 /* If we came here, wait for the answer is over */ 209 /* If we came here, wait for the answer is over */
189 if (cookie->flag & STRUCT_HTTP_FLAG_WAITINGFORTASK) { 210 if (cookie->flag & STRUCT_HTTP_FLAG_WAITINGFORTASK) {
190 http_interest_drop(sock, cookie, EVENT_IN); 211 http_interest_set(sock, cookie, 0);
191 cookie->flag &= ~STRUCT_HTTP_FLAG_WAITINGFORTASK; 212 cookie->flag &= ~STRUCT_HTTP_FLAG_WAITINGFORTASK;
192 } 213 }
193 214
@@ -238,7 +259,7 @@ ssize_t http_sendiovecdata(const int64_t sock, struct ot_workstruct *ws, size_t
238 } 259 }
239 260
240 fdstate_set_expiry(sock, time_future_seconds(OT_CLIENT_TIMEOUT_SEND)); 261 fdstate_set_expiry(sock, time_future_seconds(OT_CLIENT_TIMEOUT_SEND));
241 http_interest_add(sock, cookie, EVENT_OUT); 262 http_interest_set(sock, cookie, EVENT_OUT);
242 return 0; 263 return 0;
243} 264}
244 265
@@ -333,7 +354,7 @@ static ssize_t http_handle_stats(const int64_t sock, struct ot_workstruct *ws, c
333 /* Clients waiting for us should not easily timeout */ 354 /* Clients waiting for us should not easily timeout */
334 fdstate_set_expiry(sock, 0); 355 fdstate_set_expiry(sock, 0);
335 fullscrape_deliver(sock, format); 356 fullscrape_deliver(sock, format);
336 http_interest_drop(sock, cookie, EVENT_IN); 357 http_interest_set(sock, cookie, 0);
337 return ws->reply_size = -2; 358 return ws->reply_size = -2;
338 } 359 }
339#endif 360#endif
@@ -428,7 +449,7 @@ static ssize_t http_handle_fullscrape(const int64_t sock, struct ot_workstruct *
428 /* Clients waiting for us should not easily timeout */ 449 /* Clients waiting for us should not easily timeout */
429 fdstate_set_expiry(sock, 0); 450 fdstate_set_expiry(sock, 0);
430 fullscrape_deliver(sock, format); 451 fullscrape_deliver(sock, format);
431 http_interest_drop(sock, cookie, EVENT_IN); 452 http_interest_set(sock, cookie, 0);
432 return ws->reply_size = -2; 453 return ws->reply_size = -2;
433} 454}
434#endif 455#endif
diff --git a/ot_http.h b/ot_http.h
index 0f351c1..82d645b 100644
--- a/ot_http.h
+++ b/ot_http.h
@@ -22,7 +22,8 @@ typedef enum {
22 STRUCT_HTTP_FLAG_ZSTD = 8, 22 STRUCT_HTTP_FLAG_ZSTD = 8,
23 STRUCT_HTTP_FLAG_CHUNKED = 16, 23 STRUCT_HTTP_FLAG_CHUNKED = 16,
24 STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER = 32, 24 STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER = 32,
25 STRUCT_HTTP_FLAG_KEEPALIVE = 64 25 STRUCT_HTTP_FLAG_KEEPALIVE = 64,
26 STRUCT_HTTP_FLAG_ONSTACK = 128
26} STRUCT_HTTP_FLAG; 27} STRUCT_HTTP_FLAG;
27 28
28struct http_data { 29struct http_data {
@@ -33,13 +34,16 @@ struct http_data {
33 STRUCT_HTTP_FLAG flag; 34 STRUCT_HTTP_FLAG flag;
34}; 35};
35 36
36/* Add to or drop from what this connection is waiting to be told about */ 37/* What this connection is waiting for: EVENT_IN, EVENT_OUT, or nothing */
37void http_interest_add(const int64_t s, struct http_data *cookie, int what); 38void http_interest_set(const int64_t s, struct http_data *cookie, int what);
38void http_interest_drop(const int64_t s, struct http_data *cookie, int what);
39 39
40/* Close a connection and release everything that belongs to it */ 40/* Close a connection and release everything that belongs to it */
41void http_close(const int64_t s); 41void http_close(const int64_t s);
42 42
43/* Promote a stack cookie to a heap one for persistence and handing it over
44 to the event loop. */
45int http_adopt(const int64_t s, struct http_data *cookie);
46
43ssize_t http_handle_request(const int64_t s, struct ot_workstruct *ws); 47ssize_t http_handle_request(const int64_t s, struct ot_workstruct *ws);
44ssize_t http_sendiovecdata(const int64_t s, struct ot_workstruct *ws, size_t iovec_entries, struct iovec *iovector, int is_partial); 48ssize_t http_sendiovecdata(const int64_t s, struct ot_workstruct *ws, size_t iovec_entries, struct iovec *iovector, int is_partial);
45ssize_t http_issue_error(const int64_t s, struct ot_workstruct *ws, int code); 49ssize_t http_issue_error(const int64_t s, struct ot_workstruct *ws, int code);
diff --git a/ot_socket.c b/ot_socket.c
index 47ed26d..48ffc27 100644
--- a/ot_socket.c
+++ b/ot_socket.c
@@ -8,6 +8,7 @@
8#include <errno.h> 8#include <errno.h>
9#include <fcntl.h> 9#include <fcntl.h>
10#include <netinet/in.h> 10#include <netinet/in.h>
11#include <netinet/tcp.h>
11#include <string.h> 12#include <string.h>
12#include <sys/socket.h> 13#include <sys/socket.h>
13#include <unistd.h> 14#include <unistd.h>
@@ -120,6 +121,26 @@ static int socket_bind_v6(int s, const ot_ip6 ip, uint16_t port, socket_sharing
120 return bind(s, (struct sockaddr *)&sa, sizeof(sa)); 121 return bind(s, (struct sockaddr *)&sa, sizeof(sa));
121} 122}
122 123
124/* Defer the listening socket reporting it to be ready to accept until enough
125 sensible data is available for an accept+read pair. Where available. */
126int socket_defer_accept(int s) {
127#if defined(TCP_DEFER_ACCEPT)
128 int patience = OT_CLIENT_TIMEOUT;
129
130 return setsockopt(s, IPPROTO_TCP, TCP_DEFER_ACCEPT, &patience, sizeof(patience));
131#elif defined(SO_ACCEPTFILTER)
132 struct accept_filter_arg filter;
133 const char name[sizeof(filter.af_name)] = "dataready";
134
135 memset(&filter, 0, sizeof(filter));
136 memcpy(filter.af_name, name, sizeof(name));
137 return setsockopt(s, SOL_SOCKET, SO_ACCEPTFILTER, &filter, sizeof(filter));
138#else
139 (void)s;
140 return -1;
141#endif
142}
143
123int socket_tcp_listen(const ot_ip6 ip, uint16_t port, socket_sharing sharing) { 144int socket_tcp_listen(const ot_ip6 ip, uint16_t port, socket_sharing sharing) {
124 int s = socket_v6(SOCK_STREAM); 145 int s = socket_v6(SOCK_STREAM);
125 146
diff --git a/ot_socket.h b/ot_socket.h
index f73a5c9..5bbffe1 100644
--- a/ot_socket.h
+++ b/ot_socket.h
@@ -38,6 +38,7 @@ int socket_tcp_connect(const ot_ip6 srcip, uint16_t srcport, const ot_ip6 ds
38int socket_is_connected(int s); 38int socket_is_connected(int s);
39int socket_accept(int s, ot_ip6 ip, uint16_t *port); 39int socket_accept(int s, ot_ip6 ip, uint16_t *port);
40int socket_set_blocking(int s, int blocking); 40int socket_set_blocking(int s, int blocking);
41int socket_defer_accept(int s);
41 42
42ssize_t socket_recv(int s, void *buf, size_t len, ot_ip6 ip, uint16_t *port); 43ssize_t socket_recv(int s, void *buf, size_t len, ot_ip6 ip, uint16_t *port);
43ssize_t socket_send(int s, const void *buf, size_t len, const ot_ip6 ip, uint16_t port); 44ssize_t socket_send(int s, const void *buf, size_t len, const ot_ip6 ip, uint16_t port);
diff --git a/ot_tcp.c b/ot_tcp.c
new file mode 100644
index 0000000..8647a47
--- /dev/null
+++ b/ot_tcp.c
@@ -0,0 +1,157 @@
1/* This software was written by Dirk Engling <erdgeist@erdgeist.org>
2 It is considered beerware. Prost. Skol. Cheers or whatever.
3
4 $id$ */
5
6/* System */
7#include <errno.h>
8#include <string.h>
9#include <unistd.h>
10
11/* Opentracker */
12#include "ot_batch.h"
13#include "ot_buf.h"
14#include "ot_event.h"
15#include "ot_fdstate.h"
16#include "ot_http.h"
17#include "ot_mutex.h"
18#include "ot_socket.h"
19#include "ot_stats.h"
20#include "ot_tcp.h"
21#include "trackerlogic.h"
22
23static ssize_t header_complete(char *request, ssize_t byte_count) {
24 ssize_t i = 0, state = 0;
25
26 for (i = 1; i < byte_count; i += 2)
27 if (request[i] <= 13) {
28 i--;
29 for (state = 0; i < byte_count; ++i) {
30 char c = request[i];
31 if (c == '\r' || c == '\n')
32 state = (state >> 2) | ((c << 6) & 0xc0);
33 else
34 break;
35 if (state >= 0xa0 || state == 0x99)
36 return i + 1;
37 }
38 }
39 return 0;
40}
41
42void tcp_read(const int64_t sock, struct ot_workstruct *ws) {
43 struct http_data *cookie = fdstate_cookie(sock);
44 ssize_t byte_count = read((int)sock, ws->inbuf, G_INBUF_SIZE);
45
46 if (byte_count == 0 || (byte_count == -1 && errno != EAGAIN && errno != EWOULDBLOCK && errno != EINTR)) {
47 http_close(sock);
48 return;
49 }
50
51 if (byte_count == -1)
52 return;
53
54 ws->request = NULL;
55 ws->request_size = 0;
56 ws->header_size = 0;
57 ws->keep_alive = 0;
58
59 /* If we get the whole request in one packet, handle it without copying */
60 if (!cookie->request.data) {
61 if ((ws->header_size = header_complete(ws->inbuf, byte_count))) {
62 ws->request = ws->inbuf;
63 ws->request_size = byte_count;
64 http_handle_request(sock, ws);
65 } else if (buf_cat(&cookie->request, ws->inbuf, (size_t)byte_count))
66 http_close(sock);
67 return;
68 }
69
70 if (buf_cat(&cookie->request, ws->inbuf, (size_t)byte_count) || cookie->request.size > G_INBUF_SIZE) {
71 http_issue_error(sock, ws, CODE_HTTPERROR_500);
72 return;
73 }
74
75 while ((ws->header_size = header_complete(cookie->request.data, (ssize_t)cookie->request.size))) {
76 ws->request = cookie->request.data;
77 ws->request_size = (ssize_t)cookie->request.size;
78 http_handle_request(sock, ws);
79#ifdef WANT_KEEPALIVE
80 if (!ws->keep_alive)
81#endif
82 return;
83 }
84}
85
86void tcp_write(const int64_t sock) {
87 struct http_data *cookie = fdstate_cookie(sock);
88 int chunked = 0;
89 int keep_alive = 0;
90
91 if (cookie) {
92 chunked = (cookie->flag & STRUCT_HTTP_FLAG_CHUNKED_IN_TRANSFER) != 0;
93 keep_alive = (cookie->flag & STRUCT_HTTP_FLAG_KEEPALIVE) != 0;
94
95 if (cookie->batch) {
96 if (batch_send(sock, cookie->batch) == BATCH_FAILED) {
97 http_close(sock);
98 return;
99 }
100 if (batch_pending(cookie->batch))
101 return;
102 }
103 }
104
105 /* In a chunked transfer, wait for the next piece once everything queued has gone out */
106 if (chunked)
107 http_interest_set(sock, cookie, 0);
108 else if (keep_alive) {
109 /* The reply is out and the client asked to keep the connection, so go
110 back to waiting for the next request */
111 http_interest_set(sock, cookie, EVENT_IN);
112 fdstate_set_expiry(sock, time_future_seconds(OT_CLIENT_TIMEOUT));
113 } else
114 http_close(sock);
115}
116
117/* The fast path, assuming that the complete request has usually arrived by the time a
118 connection is accepted, so it is answered here outside the event loop. The more load
119 a tracker has, the more likely it is that by the time it accepts, the roundtrip to
120 the peer has completed.
121 If the request could not be completely handled, the stack cookie is promoted to heap
122 and the connection handed over to the event loop to finish */
123void tcp_accept(const int64_t serversocket, struct ot_workstruct *ws) {
124 int64_t sock;
125 ot_ip6 ip;
126 uint16_t port;
127
128 while ((sock = socket_accept(serversocket, ip, &port)) != -1) {
129 struct http_data cookie;
130 uint32_t generation;
131
132 memset(&cookie, 0, sizeof(cookie));
133 memcpy(cookie.ip, ip, sizeof(ot_ip6));
134 cookie.interest = EVENT_IN;
135 cookie.flag = STRUCT_HTTP_FLAG_ONSTACK;
136
137 generation = fdstate_open(sock, &cookie, time_future_seconds(OT_CLIENT_TIMEOUT));
138 if (!fdstate_claim((int)sock, generation)) {
139 fdstate_close(sock);
140 event_close((int)sock);
141 continue;
142 }
143
144 stats_issue_event(ws, EVENT_ACCEPT, FLAG_TCP, (uintptr_t)ip);
145
146 tcp_read(sock, ws);
147
148 /* Gone already if it was answered and closed inline */
149 if (fdstate_generation((int)sock) != generation)
150 continue;
151
152 if (http_adopt(sock, &cookie) == -1)
153 http_close(sock);
154 else
155 fdstate_release((int)sock);
156 }
157}
diff --git a/ot_tcp.h b/ot_tcp.h
new file mode 100644
index 0000000..b0a93f4
--- /dev/null
+++ b/ot_tcp.h
@@ -0,0 +1,22 @@
1/* This software was written by Dirk Engling <erdgeist@erdgeist.org>
2 It is considered beerware. Prost. Skol. Cheers or whatever.
3
4 $id$ */
5
6#ifndef OT_TCP_H__
7#define OT_TCP_H__
8
9#include <stdint.h>
10
11#include "trackerlogic.h"
12
13/* Take whatever connections are waiting on a listening socket and answer
14 them, each on this thread's stack, handing a connection that cannot be
15 finished to the event loop */
16void tcp_accept(const int64_t serversocket, struct ot_workstruct *ws);
17
18/* Handler connections in the event loop has become readable or writable */
19void tcp_read(const int64_t sock, struct ot_workstruct *ws);
20void tcp_write(const int64_t sock);
21
22#endif
diff --git a/tests/accept_filter_probe.py b/tests/accept_filter_probe.py
new file mode 100644
index 0000000..22f585a
--- /dev/null
+++ b/tests/accept_filter_probe.py
@@ -0,0 +1,110 @@
1"""What a deferred accept does to connections that never send anything.
2
3Those connections never reach opentracker, so its own client timeout cannot
4clean them up: whatever bound exists is the kernel's. This measures it, and
5then asks the question that follows from it -- whether enough of them keep
6the tracker from answering anybody else.
7
8 make opentracker
9 python3 tests/accept_filter_probe.py
10
11On FreeBSD the filter does nothing unless accf_data is loaded, so run it both
12ways to see whether the numbers differ:
13
14 kldload accf_data
15"""
16
17import os
18import socket
19import signal
20import subprocess
21import sys
22import tempfile
23import time
24import urllib.error
25import urllib.request
26
27TRACKER = os.environ.get("OPENTRACKER", "./opentracker")
28IDLE = int(os.environ.get("IDLE", "300"))
29PATIENCE = float(os.environ.get("PATIENCE", "90"))
30HASH = "".join("%%%02x" % b for b in range(1, 21))
31
32
33def start(port):
34 directory = tempfile.mkdtemp(prefix="ot-accf-")
35 with open(os.path.join(directory, "conf"), "w") as handle:
36 handle.write("listen.tcp 127.0.0.1:%d\nlisten.udp 127.0.0.1:%d\n" % (port, port))
37 log = open(os.path.join(directory, "log"), "w+b")
38 tracker = subprocess.Popen([TRACKER, "-f", os.path.join(directory, "conf")], stdout=log, stderr=log)
39 for _ in range(300):
40 try:
41 socket.create_connection(("127.0.0.1", port), timeout=0.2).close()
42 return tracker, log
43 except OSError:
44 time.sleep(0.02)
45 tracker.kill()
46 raise SystemExit("tracker never listened on %d" % port)
47
48
49def announce(port, timeout=5):
50 """One ordinary request, so we can see whether the tracker still serves."""
51 started = time.time()
52 try:
53 urllib.request.urlopen(
54 "http://127.0.0.1:%d/announce?info_hash=%s&peer_id=-OT0001-probe0000001&port=6881&left=0" % (port, HASH),
55 timeout=timeout,
56 ).read()
57 return time.time() - started, None
58 except Exception as problem: # noqa: BLE001 - reported, not swallowed
59 return time.time() - started, problem
60
61
62def main():
63 port = int(os.environ.get("PORT", "28000"))
64 tracker, log = start(port)
65 try:
66 print(" how long a connection that says nothing is held")
67 quiet = socket.create_connection(("127.0.0.1", port), timeout=5)
68 quiet.settimeout(PATIENCE)
69 started = time.time()
70 try:
71 held = "dropped after %.1f s" % (time.time() - started) if quiet.recv(1) == b"" else "sent us something"
72 except socket.timeout:
73 held = "still open after %.0f s" % PATIENCE
74 except OSError as problem:
75 held = "ended after %.1f s: %s" % (time.time() - started, problem)
76 quiet.close()
77 print(" %s" % held)
78
79 print(" and whether %d of them keep anybody else out" % IDLE)
80 before, problem = announce(port)
81 print(" a request before: %.0f ms%s" % (before * 1000, "" if problem is None else " (%s)" % problem))
82
83 idlers = []
84 for _ in range(IDLE):
85 try:
86 idlers.append(socket.create_connection(("127.0.0.1", port), timeout=2))
87 except OSError as problem:
88 print(" could only open %d of them: %s" % (len(idlers), problem))
89 break
90 print(" holding %d connections that have sent nothing" % len(idlers))
91
92 during, problem = announce(port, timeout=10)
93 print(" a request during: %.0f ms%s" % (during * 1000, "" if problem is None else " REFUSED: %s" % problem))
94
95 for idler in idlers:
96 idler.close()
97 after, problem = announce(port)
98 print(" a request after: %.0f ms%s" % (after * 1000, "" if problem is None else " (%s)" % problem))
99 finally:
100 tracker.send_signal(signal.SIGINT)
101 try:
102 tracker.wait(timeout=10)
103 except subprocess.TimeoutExpired:
104 tracker.kill()
105 tracker.wait()
106 log.close()
107
108
109if __name__ == "__main__":
110 main()
diff --git a/tests/integration/test_config.py b/tests/integration/test_config.py
new file mode 100644
index 0000000..1a84fa7
--- /dev/null
+++ b/tests/integration/test_config.py
@@ -0,0 +1,29 @@
1"""Settings the tracker has to accept, and still work with.
2
3An unrecognised key makes opentracker refuse to start, so a test that it
4comes up at all is a test that the key was understood.
5"""
6
7import unittest
8
9from harness import Tracker, TrackerTestCase, free_port
10
11
12class TcpWorkers(TrackerTestCase):
13 config = "listen.tcp.workers 2\n"
14
15 def test_the_tracker_serves_with_tcp_workers_asked_for(self):
16 self.assertEqual(self.tracker.announce(self.hash_for("tcpworkers"))[b"complete"], 1)
17
18
19class WorkersOnAListenStatement(unittest.TestCase):
20 def test_a_listen_statement_may_carry_a_worker_count(self):
21 """The option is read off the statement rather than inherited."""
22 second = free_port()
23 tracker = Tracker("listen.tcp 127.0.0.1:%d workers=2\n" % second)
24 try:
25 self.assertEqual(tracker.announce(b"listenopts\0\0\0\0\0\0\0\0\0\0")[b"complete"], 1)
26 complaints = tracker.stop()
27 self.assertEqual(complaints, [], "the tracker reported:\n" + "\n".join(complaints))
28 finally:
29 tracker._cleanup()
diff --git a/tests/integration/test_expiry.py b/tests/integration/test_expiry.py
index 5a21e57..5e6dc0a 100644
--- a/tests/integration/test_expiry.py
+++ b/tests/integration/test_expiry.py
@@ -27,6 +27,11 @@ class ExpirySweep(TrackerTestCase):
27 def test_the_sweep_takes_the_idle_and_leaves_the_answered(self): 27 def test_the_sweep_takes_the_idle_and_leaves_the_answered(self):
28 idle = socket.create_connection(("127.0.0.1", self.tracker.port), timeout=5) 28 idle = socket.create_connection(("127.0.0.1", self.tracker.port), timeout=5)
29 try: 29 try:
30 # Enough for a kernel that holds a connection back until data
31 # arrives to hand it over, and far too little to be answered, so
32 # the tracker has accepted it and is waiting for the rest
33 idle.sendall(b"GET /ann")
34
30 self.assertEqual(self.tracker.announce(self.hash_for("expiry-sweep"))[b"complete"], 1) 35 self.assertEqual(self.tracker.announce(self.hash_for("expiry-sweep"))[b"complete"], 1)
31 36
32 time.sleep(CLIENT_TIMEOUT + 3 * CHECK_INTERVAL) 37 time.sleep(CLIENT_TIMEOUT + 3 * CHECK_INTERVAL)
@@ -34,6 +39,6 @@ class ExpirySweep(TrackerTestCase):
34 self.assertEqual(self.tracker.torrent_count(), 1, "the tracker stopped serving while sweeping") 39 self.assertEqual(self.tracker.torrent_count(), 1, "the tracker stopped serving while sweeping")
35 40
36 idle.settimeout(5) 41 idle.settimeout(5)
37 self.assertEqual(idle.recv(64), b"", "a connection that never sent a request was not swept") 42 self.assertEqual(idle.recv(64), b"", "a connection that went quiet mid-request was not swept")
38 finally: 43 finally:
39 idle.close() 44 idle.close()
diff --git a/tests/syscount.c b/tests/syscount.c
new file mode 100644
index 0000000..ec45c15
--- /dev/null
+++ b/tests/syscount.c
@@ -0,0 +1,144 @@
1/* Counts the system calls a run makes, by standing in front of libc.
2
3 cc -O2 -shared -fPIC -o syscount.so tests/syscount.c -ldl
4 LD_PRELOAD=./syscount.so ./opentracker -f conf (DYLD_INSERT_LIBRARIES on macOS)
5
6 The tally goes to stderr when the process exits. Reading the source tells
7 you which calls a path should make; this tells you which it does make,
8 which is not always the same thing. */
9
10#define _GNU_SOURCE
11#include <dlfcn.h>
12#include <stdarg.h>
13#include <stdatomic.h>
14#include <stdio.h>
15#include <sys/socket.h>
16#include <sys/types.h>
17#include <unistd.h>
18
19#ifdef __linux__
20#include <sys/epoll.h>
21#endif
22
23#define COUNTER(name) static _Atomic long g_##name
24#define BUMP(name) atomic_fetch_add_explicit(&g_##name, 1, memory_order_relaxed)
25#define REPORT(name) fprintf(stderr, " %-14s %10ld\n", #name, atomic_load(&g_##name))
26
27COUNTER(accept);
28COUNTER(accept4);
29COUNTER(read);
30COUNTER(write);
31COUNTER(close);
32COUNTER(fcntl);
33COUNTER(recvfrom);
34COUNTER(sendto);
35COUNTER(epoll_ctl);
36COUNTER(epoll_wait);
37
38typedef int (*fn_accept)(int, struct sockaddr *, socklen_t *);
39typedef int (*fn_accept4)(int, struct sockaddr *, socklen_t *, int);
40typedef ssize_t (*fn_read)(int, void *, size_t);
41typedef ssize_t (*fn_write)(int, const void *, size_t);
42typedef int (*fn_close)(int);
43typedef int (*fn_fcntl)(int, int, ...);
44typedef ssize_t (*fn_recvfrom)(int, void *, size_t, int, struct sockaddr *, socklen_t *);
45typedef ssize_t (*fn_sendto)(int, const void *, size_t, int, const struct sockaddr *, socklen_t);
46
47static fn_accept real_accept;
48static fn_accept4 real_accept4;
49static fn_read real_read;
50static fn_write real_write;
51static fn_close real_close;
52static fn_fcntl real_fcntl;
53static fn_recvfrom real_recvfrom;
54static fn_sendto real_sendto;
55
56/* Resolving lazily keeps the library out of the way until something is
57 actually called, which matters because dlsym itself allocates */
58#define WRAP(name) \
59 if (!real_##name) \
60 real_##name = (fn_##name)dlsym(RTLD_NEXT, #name); \
61 BUMP(name)
62
63int accept(int s, struct sockaddr *addr, socklen_t *len) {
64 WRAP(accept);
65 return real_accept(s, addr, len);
66}
67
68#ifdef SOCK_NONBLOCK
69int accept4(int s, struct sockaddr *addr, socklen_t *len, int flags) {
70 WRAP(accept4);
71 return real_accept4(s, addr, len, flags);
72}
73#endif
74
75ssize_t read(int fd, void *buf, size_t count) {
76 WRAP(read);
77 return real_read(fd, buf, count);
78}
79
80ssize_t write(int fd, const void *buf, size_t count) {
81 WRAP(write);
82 return real_write(fd, buf, count);
83}
84
85int close(int fd) {
86 WRAP(close);
87 return real_close(fd);
88}
89
90int fcntl(int fd, int cmd, ...) {
91 va_list args;
92 long arg;
93
94 va_start(args, cmd);
95 arg = va_arg(args, long);
96 va_end(args);
97 WRAP(fcntl);
98 return real_fcntl(fd, cmd, arg);
99}
100
101ssize_t recvfrom(int s, void *buf, size_t len, int flags, struct sockaddr *from, socklen_t *fromlen) {
102 WRAP(recvfrom);
103 return real_recvfrom(s, buf, len, flags, from, fromlen);
104}
105
106ssize_t sendto(int s, const void *buf, size_t len, int flags, const struct sockaddr *to, socklen_t tolen) {
107 WRAP(sendto);
108 return real_sendto(s, buf, len, flags, to, tolen);
109}
110
111#ifdef __linux__
112typedef int (*fn_epoll_ctl)(int, int, int, struct epoll_event *);
113typedef int (*fn_epoll_wait)(int, struct epoll_event *, int, int);
114static fn_epoll_ctl real_epoll_ctl;
115static fn_epoll_wait real_epoll_wait;
116
117int epoll_ctl(int ep, int op, int fd, struct epoll_event *ev) {
118 WRAP(epoll_ctl);
119 return real_epoll_ctl(ep, op, fd, ev);
120}
121
122int epoll_wait(int ep, struct epoll_event *events, int max, int timeout) {
123 WRAP(epoll_wait);
124 return real_epoll_wait(ep, events, max, timeout);
125}
126#endif
127
128__attribute__((destructor)) static void report(void) {
129 fprintf(stderr, "\n system calls this run\n");
130 REPORT(accept);
131#ifdef SOCK_NONBLOCK
132 REPORT(accept4);
133#endif
134 REPORT(read);
135 REPORT(write);
136 REPORT(close);
137 REPORT(fcntl);
138 REPORT(recvfrom);
139 REPORT(sendto);
140#ifdef __linux__
141 REPORT(epoll_ctl);
142 REPORT(epoll_wait);
143#endif
144}
diff --git a/tests/t_event.c b/tests/t_event.c
index 5509f19..1908dc9 100644
--- a/tests/t_event.c
+++ b/tests/t_event.c
@@ -4,6 +4,7 @@
4 $id$ */ 4 $id$ */
5 5
6/* System */ 6/* System */
7#include <fcntl.h>
7#include <string.h> 8#include <string.h>
8#include <sys/socket.h> 9#include <sys/socket.h>
9#include <sys/time.h> 10#include <sys/time.h>
@@ -205,9 +206,75 @@ static void case_unknown_descriptor(void) {
205 CHECK_INT(event_del(4096), -1); 206 CHECK_INT(event_del(4096), -1);
206} 207}
207 208
209/* event_close exists so that the mechanisms which drop a closed descriptor
210 by themselves are not asked to do it twice. The one that keeps its own
211 list has to be told, and forgetting would leave the entry there to be
212 reported for ever, so what is checked here is the silence afterwards. */
213static void case_close_takes_it_out(void) {
214 int other[2], closed, count, i, strays = 0;
215
216 test_case("a closed descriptor is gone from the set and really closed");
217 pair_open();
218 CHECK(sv[0] != -1);
219 CHECK_INT(socketpair(AF_UNIX, SOCK_STREAM, 0, other), 0);
220 CHECK_INT(event_add(sv[0], EVENT_IN, 1), 0);
221 CHECK_INT(event_add(other[0], EVENT_IN, 2), 0);
222
223 closed = sv[0];
224 CHECK_INT(event_close(closed), 0);
225 sv[0] = -1;
226 CHECK_INT(fcntl(closed, F_GETFD), -1);
227
228 /* The survivor is made ready so the wait has something to come back with,
229 and anything left over from the closed one shows up beside it */
230 CHECK_INT((int)write(other[1], "x", 1), 1);
231 count = event_wait(ready, EVENTS, 200);
232 CHECK(count >= 1);
233 for (i = 0; i < count; ++i)
234 if (ready[i].fd != other[0])
235 ++strays;
236 CHECK_INT(strays, 0);
237
238 event_del(other[0]);
239 close(other[0]);
240 close(other[1]);
241 pair_close();
242}
243
244/* A descriptor can be taken on before anybody waits for anything on it: the
245 full scrape parks a connection while a worker thread builds the reply. The
246 mechanisms disagree about what registering nothing means, so what is
247 checked is that asking for a filter afterwards still works. */
248static void case_add_nothing_then_ask(void) {
249 test_case("a descriptor added with no interest can be given some later");
250 pair_open();
251 CHECK(sv[0] != -1);
252 CHECK_INT(event_add(sv[0], 0, 7), 0);
253 CHECK_INT((int)write(sv[1], "x", 1), 1);
254 CHECK_INT(wait_for(sv[0], EVENT_IN, 200, NULL), 0);
255
256 CHECK_INT(event_modify(sv[0], EVENT_IN, 7), 0);
257 {
258 uint32_t generation = 0;
259 CHECK_INT(wait_for(sv[0], EVENT_IN, 500, &generation), 1);
260 CHECK_INT((int)generation, 7);
261 }
262 event_del(sv[0]);
263 pair_close();
264}
265
266/* Whether a worker thread may register a descriptor while this one waits
267 depends on who keeps the set. Tying the answer to the mechanism's name
268 here means a backend added later cannot quietly forget to say. */
269static void case_sharing_matches_the_mechanism(void) {
270 test_case("only a set the kernel keeps may be changed from another thread");
271 CHECK_INT(event_can_share(), strcmp(event_mechanism(), "poll") ? 1 : 0);
272}
273
208void test_event(void) { 274void test_event(void) {
209 CHECK_INT(event_init(), 0); 275 CHECK_INT(event_init(), 0);
210 CHECK(event_mechanism() != NULL); 276 CHECK(event_mechanism() != NULL);
277 case_sharing_matches_the_mechanism();
211 278
212 case_readable(); 279 case_readable();
213 case_writable(); 280 case_writable();
@@ -217,6 +284,8 @@ void test_event(void) {
217 case_delete_from_the_middle(); 284 case_delete_from_the_middle();
218 case_timeout(); 285 case_timeout();
219 case_unknown_descriptor(); 286 case_unknown_descriptor();
287 case_close_takes_it_out();
288 case_add_nothing_then_ask();
220 289
221 event_deinit(); 290 event_deinit();
222} 291}
diff --git a/tests/t_socket.c b/tests/t_socket.c
index ab4b785..20a0c0a 100644
--- a/tests/t_socket.c
+++ b/tests/t_socket.c
@@ -78,15 +78,17 @@ static int sock_flag(int s, int level, int option) {
78} 78}
79 79
80/* Wait for one event, so a test never blocks forever on a lost packet */ 80/* Wait for one event, so a test never blocks forever on a lost packet */
81static int wait_for(int s, short events) { 81static int wait_for_ms(int s, short events, int timeout_ms) {
82 struct pollfd pfd; 82 struct pollfd pfd;
83 83
84 pfd.fd = s; 84 pfd.fd = s;
85 pfd.events = events; 85 pfd.events = events;
86 pfd.revents = 0; 86 pfd.revents = 0;
87 return poll(&pfd, 1, 2000) == 1; 87 return poll(&pfd, 1, timeout_ms) == 1;
88} 88}
89 89
90static int wait_for(int s, short events) { return wait_for_ms(s, events, 2000); }
91
90/* ------------------------------------------------------------ invariants */ 92/* ------------------------------------------------------------ invariants */
91 93
92/* Asked for the wildcard address, because a socket bound to one specific 94/* Asked for the wildcard address, because a socket bound to one specific
@@ -200,6 +202,42 @@ static void case_a_signal_ends_a_blocking_read(void) {
200 close(reader_socket); 202 close(reader_socket);
201} 203}
202 204
205/* What a deferred accept is for is a descriptor that already has the request
206 in it, so what is checked is the silence before: a client that has
207 connected but said nothing must not make the listener readable. Kernels
208 that cannot do this say so, and then there is nothing to check. */
209static void case_deferred_accept(void) {
210 struct sockaddr_in6 address;
211 char ip[16];
212 uint16_t port = 0;
213 int listener, client;
214
215 test_case("a connection that has sent nothing is held back from accept");
216 listener = socket_tcp_listen(ip_loopback, 0, SOCKET_EXCLUSIVE);
217 CHECK(listener != -1);
218 if (socket_defer_accept(listener) == -1) {
219 close(listener);
220 return;
221 }
222 bound_address(listener, ip, &port);
223
224 client = socket(AF_INET6, SOCK_STREAM, 0);
225 CHECK(client != -1);
226 memset(&address, 0, sizeof(address));
227 address.sin6_family = AF_INET6;
228 address.sin6_port = htons(port);
229 memcpy(&address.sin6_addr, ip_loopback, 16);
230 CHECK_INT(connect(client, (struct sockaddr *)&address, sizeof(address)), 0);
231
232 CHECK_INT(wait_for_ms(listener, POLLIN, 300), 0);
233
234 CHECK_INT((int)write(client, "GET / HTTP/1.0\r\n\r\n", 18), 18);
235 CHECK_INT(wait_for(listener, POLLIN), 1);
236
237 close(client);
238 close(listener);
239}
240
203/* An IPv4 peer must arrive in the v4 mapped form opentracker stores, however 241/* An IPv4 peer must arrive in the v4 mapped form opentracker stores, however
204 the kernel chose to report it. Linux fills a v4 mapped sockaddr_in6 here 242 the kernel chose to report it. Linux fills a v4 mapped sockaddr_in6 here
205 and the BSDs fill a plain sockaddr_in, so this is the case that tells the 243 and the BSDs fill a plain sockaddr_in, so this is the case that tells the
@@ -352,6 +390,7 @@ void test_socket(void) {
352 case_a_signal_ends_a_blocking_read(); 390 case_a_signal_ends_a_blocking_read();
353 case_udp_v4mapped(); 391 case_udp_v4mapped();
354 case_tcp_roundtrip(); 392 case_tcp_roundtrip();
393 case_deferred_accept();
355 case_connect_refused(); 394 case_connect_refused();
356 case_bind_failure(); 395 case_bind_failure();
357#ifdef WANT_SYNC_LIVE 396#ifdef WANT_SYNC_LIVE
generated by cgit v1.3 (git 2.53.0) at 2026-10-31 06:51:55 +0000