Commit | Line | Data |
---|---|---|
173fca4f | 1 | /* |
ab5be9fa MJ |
2 | * Copyright (C) 2012 David Goulet <dgoulet@efficios.com> |
3 | * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com> | |
173fca4f | 4 | * |
ab5be9fa | 5 | * SPDX-License-Identifier: GPL-2.0-only |
173fca4f | 6 | * |
173fca4f JG |
7 | */ |
8 | ||
c9e313bc SM |
9 | #include "lttng-sessiond.hpp" |
10 | #include "health-sessiond.hpp" | |
11 | #include <common/macros.hpp> | |
12 | #include <common/error.hpp> | |
13 | #include <common/utils.hpp> | |
14 | #include <common/pipe.hpp> | |
173fca4f JG |
15 | #include <inttypes.h> |
16 | #include <sys/stat.h> | |
c9e313bc SM |
17 | #include "utils.hpp" |
18 | #include "thread.hpp" | |
173fca4f | 19 | |
f1494934 | 20 | namespace { |
c78d8e86 JG |
21 | struct thread_notifiers { |
22 | struct lttng_pipe *quit_pipe; | |
23 | sem_t ready; | |
24 | }; | |
f1494934 | 25 | } /* namespace */ |
c78d8e86 JG |
26 | |
27 | static | |
28 | void mark_thread_as_ready(struct thread_notifiers *notifiers) | |
173fca4f | 29 | { |
c78d8e86 JG |
30 | DBG("Marking health management thread as ready"); |
31 | sem_post(¬ifiers->ready); | |
32 | } | |
173fca4f | 33 | |
c78d8e86 JG |
34 | static |
35 | void wait_until_thread_is_ready(struct thread_notifiers *notifiers) | |
36 | { | |
37 | DBG("Waiting for health management thread to be ready"); | |
38 | sem_wait(¬ifiers->ready); | |
39 | DBG("Health management thread is ready"); | |
40 | } | |
41 | ||
42 | static void cleanup_health_management_thread(void *data) | |
43 | { | |
7966af57 | 44 | struct thread_notifiers *notifiers = (thread_notifiers *) data; |
c78d8e86 JG |
45 | |
46 | lttng_pipe_destroy(notifiers->quit_pipe); | |
47 | sem_destroy(¬ifiers->ready); | |
48 | free(notifiers); | |
173fca4f JG |
49 | } |
50 | ||
51 | /* | |
52 | * Thread managing health check socket. | |
53 | */ | |
c78d8e86 | 54 | static void *thread_manage_health(void *data) |
173fca4f JG |
55 | { |
56 | const bool is_root = (getuid() == 0); | |
57 | int sock = -1, new_sock = -1, ret, i, pollfd, err = -1; | |
58 | uint32_t revents, nb_fd; | |
59 | struct lttng_poll_event events; | |
60 | struct health_comm_msg msg; | |
61 | struct health_comm_reply reply; | |
62 | /* Thread-specific quit pipe. */ | |
7966af57 | 63 | struct thread_notifiers *notifiers = (thread_notifiers *) data; |
c78d8e86 JG |
64 | const int quit_pipe_read_fd = lttng_pipe_get_readfd( |
65 | notifiers->quit_pipe); | |
173fca4f JG |
66 | |
67 | DBG("[thread] Manage health check started"); | |
68 | ||
69 | rcu_register_thread(); | |
70 | ||
71 | /* | |
72 | * Created with a size of two for: | |
73 | * - client socket | |
74 | * - thread quit pipe | |
75 | */ | |
76 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
77 | if (ret < 0) { | |
78 | goto error; | |
79 | } | |
80 | ||
81 | /* Create unix socket */ | |
412d7227 | 82 | sock = lttcomm_create_unix_sock(the_config.health_unix_sock_path.value); |
173fca4f JG |
83 | if (sock < 0) { |
84 | ERR("Unable to create health check Unix socket"); | |
85 | goto error; | |
86 | } | |
87 | ||
88 | if (is_root) { | |
89 | /* lttng health client socket path permissions */ | |
28ab59d0 JR |
90 | gid_t gid; |
91 | ||
412d7227 | 92 | ret = utils_get_group_id(the_config.tracing_group_name.value, true, &gid); |
28ab59d0 JR |
93 | if (ret) { |
94 | /* Default to root group. */ | |
95 | gid = 0; | |
96 | } | |
97 | ||
412d7227 | 98 | ret = chown(the_config.health_unix_sock_path.value, 0, gid); |
173fca4f | 99 | if (ret < 0) { |
412d7227 | 100 | ERR("Unable to set group on %s", the_config.health_unix_sock_path.value); |
173fca4f JG |
101 | PERROR("chown"); |
102 | goto error; | |
103 | } | |
104 | ||
412d7227 | 105 | ret = chmod(the_config.health_unix_sock_path.value, |
173fca4f JG |
106 | S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); |
107 | if (ret < 0) { | |
412d7227 | 108 | ERR("Unable to set permissions on %s", the_config.health_unix_sock_path.value); |
173fca4f JG |
109 | PERROR("chmod"); |
110 | goto error; | |
111 | } | |
112 | } | |
113 | ||
114 | /* | |
115 | * Set the CLOEXEC flag. Return code is useless because either way, the | |
116 | * show must go on. | |
117 | */ | |
118 | (void) utils_set_fd_cloexec(sock); | |
119 | ||
120 | ret = lttcomm_listen_unix_sock(sock); | |
121 | if (ret < 0) { | |
122 | goto error; | |
123 | } | |
124 | ||
125 | ret = lttng_poll_add(&events, quit_pipe_read_fd, LPOLLIN | LPOLLERR); | |
126 | if (ret < 0) { | |
127 | goto error; | |
128 | } | |
129 | ||
130 | /* Add the application registration socket */ | |
131 | ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI); | |
132 | if (ret < 0) { | |
133 | goto error; | |
134 | } | |
135 | ||
c78d8e86 | 136 | mark_thread_as_ready(notifiers); |
173fca4f JG |
137 | while (1) { |
138 | DBG("Health check ready"); | |
139 | ||
140 | /* Infinite blocking call, waiting for transmission */ | |
141 | restart: | |
142 | ret = lttng_poll_wait(&events, -1); | |
143 | if (ret < 0) { | |
144 | /* | |
145 | * Restart interrupted system call. | |
146 | */ | |
147 | if (errno == EINTR) { | |
148 | goto restart; | |
149 | } | |
150 | goto error; | |
151 | } | |
152 | ||
153 | nb_fd = ret; | |
154 | ||
155 | for (i = 0; i < nb_fd; i++) { | |
156 | /* Fetch once the poll data */ | |
157 | revents = LTTNG_POLL_GETEV(&events, i); | |
158 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
159 | ||
173fca4f JG |
160 | /* Event on the registration socket */ |
161 | if (pollfd == sock) { | |
162 | if (revents & LPOLLIN) { | |
163 | continue; | |
164 | } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
165 | ERR("Health socket poll error"); | |
166 | goto error; | |
167 | } else { | |
168 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
169 | goto error; | |
170 | } | |
171 | } else { | |
172 | /* Event on the thread's quit pipe. */ | |
173 | err = 0; | |
174 | goto exit; | |
175 | } | |
176 | } | |
177 | ||
178 | new_sock = lttcomm_accept_unix_sock(sock); | |
179 | if (new_sock < 0) { | |
180 | goto error; | |
181 | } | |
182 | ||
183 | /* | |
184 | * Set the CLOEXEC flag. Return code is useless because either way, the | |
185 | * show must go on. | |
186 | */ | |
187 | (void) utils_set_fd_cloexec(new_sock); | |
188 | ||
189 | DBG("Receiving data from client for health..."); | |
190 | ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg)); | |
191 | if (ret <= 0) { | |
192 | DBG("Nothing recv() from client... continuing"); | |
193 | ret = close(new_sock); | |
194 | if (ret) { | |
195 | PERROR("close"); | |
196 | } | |
197 | continue; | |
198 | } | |
199 | ||
200 | rcu_thread_online(); | |
201 | ||
202 | memset(&reply, 0, sizeof(reply)); | |
203 | for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) { | |
204 | /* | |
205 | * health_check_state returns 0 if health is | |
206 | * bad. | |
207 | */ | |
412d7227 | 208 | if (!health_check_state(the_health_sessiond, i)) { |
173fca4f JG |
209 | reply.ret_code |= 1ULL << i; |
210 | } | |
211 | } | |
212 | ||
213 | DBG2("Health check return value %" PRIx64, reply.ret_code); | |
214 | ||
215 | ret = lttcomm_send_unix_sock(new_sock, (void *) &reply, | |
216 | sizeof(reply)); | |
217 | if (ret < 0) { | |
218 | ERR("Failed to send health data back to client"); | |
219 | } | |
220 | ||
221 | /* End of transmission */ | |
222 | ret = close(new_sock); | |
223 | if (ret) { | |
224 | PERROR("close"); | |
225 | } | |
226 | } | |
227 | ||
228 | exit: | |
229 | error: | |
230 | if (err) { | |
231 | ERR("Health error occurred in %s", __func__); | |
232 | } | |
233 | DBG("Health check thread dying"); | |
412d7227 | 234 | unlink(the_config.health_unix_sock_path.value); |
173fca4f JG |
235 | if (sock >= 0) { |
236 | ret = close(sock); | |
237 | if (ret) { | |
238 | PERROR("close"); | |
239 | } | |
240 | } | |
241 | ||
242 | lttng_poll_clean(&events); | |
243 | rcu_unregister_thread(); | |
244 | return NULL; | |
245 | } | |
246 | ||
c78d8e86 | 247 | static bool shutdown_health_management_thread(void *data) |
173fca4f | 248 | { |
7966af57 | 249 | struct thread_notifiers *notifiers = ( thread_notifiers *) data; |
c78d8e86 | 250 | const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe); |
173fca4f | 251 | |
c78d8e86 | 252 | return notify_thread_pipe(write_fd) == 1; |
173fca4f JG |
253 | } |
254 | ||
255 | bool launch_health_management_thread(void) | |
256 | { | |
c78d8e86 | 257 | struct thread_notifiers *notifiers; |
173fca4f | 258 | struct lttng_thread *thread; |
173fca4f | 259 | |
64803277 | 260 | notifiers = zmalloc<thread_notifiers>(); |
c78d8e86 | 261 | if (!notifiers) { |
21fa020e | 262 | goto error_alloc; |
173fca4f JG |
263 | } |
264 | ||
c78d8e86 JG |
265 | sem_init(¬ifiers->ready, 0, 0); |
266 | notifiers->quit_pipe = lttng_pipe_open(FD_CLOEXEC); | |
267 | if (!notifiers->quit_pipe) { | |
268 | goto error; | |
269 | } | |
173fca4f JG |
270 | thread = lttng_thread_create("Health management", |
271 | thread_manage_health, | |
272 | shutdown_health_management_thread, | |
273 | cleanup_health_management_thread, | |
c78d8e86 | 274 | notifiers); |
173fca4f JG |
275 | if (!thread) { |
276 | goto error; | |
277 | } | |
c78d8e86 JG |
278 | |
279 | wait_until_thread_is_ready(notifiers); | |
173fca4f JG |
280 | lttng_thread_put(thread); |
281 | return true; | |
282 | error: | |
c78d8e86 | 283 | cleanup_health_management_thread(notifiers); |
21fa020e | 284 | error_alloc: |
173fca4f JG |
285 | return false; |
286 | } |