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