xref: /freebsd/sbin/hastd/control.c (revision 4c8945a06b01a5c8122cdeb402af36bb46a06acc)
1 /*-
2  * Copyright (c) 2009-2010 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Pawel Jakub Dawidek under sponsorship from
6  * the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 
36 #include <assert.h>
37 #include <errno.h>
38 #include <pthread.h>
39 #include <signal.h>
40 #include <stdio.h>
41 #include <string.h>
42 #include <unistd.h>
43 
44 #include "hast.h"
45 #include "hastd.h"
46 #include "hast_proto.h"
47 #include "hooks.h"
48 #include "nv.h"
49 #include "pjdlog.h"
50 #include "proto.h"
51 #include "subr.h"
52 
53 #include "control.h"
54 
55 void
56 child_cleanup(struct hast_resource *res)
57 {
58 
59 	proto_close(res->hr_ctrl);
60 	res->hr_ctrl = NULL;
61 	if (res->hr_event != NULL) {
62 		proto_close(res->hr_event);
63 		res->hr_event = NULL;
64 	}
65 	res->hr_workerpid = 0;
66 }
67 
68 static void
69 control_set_role_common(struct hastd_config *cfg, struct nv *nvout,
70     uint8_t role, struct hast_resource *res, const char *name, unsigned int no)
71 {
72 	int oldrole;
73 
74 	/* Name is always needed. */
75 	if (name != NULL)
76 		nv_add_string(nvout, name, "resource%u", no);
77 
78 	if (res == NULL) {
79 		assert(cfg != NULL);
80 		assert(name != NULL);
81 
82 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
83 			if (strcmp(res->hr_name, name) == 0)
84 				break;
85 		}
86 		if (res == NULL) {
87 			nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
88 			return;
89 		}
90 	}
91 	assert(res != NULL);
92 
93 	/* Send previous role back. */
94 	nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
95 
96 	/* Nothing changed, return here. */
97 	if (role == res->hr_role)
98 		return;
99 
100 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
101 	pjdlog_info("Role changed to %s.", role2str(role));
102 
103 	/* Change role to the new one. */
104 	oldrole = res->hr_role;
105 	res->hr_role = role;
106 	pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
107 
108 	/*
109 	 * If previous role was primary or secondary we have to kill process
110 	 * doing that work.
111 	 */
112 	if (res->hr_workerpid != 0) {
113 		if (kill(res->hr_workerpid, SIGTERM) < 0) {
114 			pjdlog_errno(LOG_WARNING,
115 			    "Unable to kill worker process %u",
116 			    (unsigned int)res->hr_workerpid);
117 		} else if (waitpid(res->hr_workerpid, NULL, 0) !=
118 		    res->hr_workerpid) {
119 			pjdlog_errno(LOG_WARNING,
120 			    "Error while waiting for worker process %u",
121 			    (unsigned int)res->hr_workerpid);
122 		} else {
123 			pjdlog_debug(1, "Worker process %u stopped.",
124 			    (unsigned int)res->hr_workerpid);
125 		}
126 		child_cleanup(res);
127 	}
128 
129 	/* Start worker process if we are changing to primary. */
130 	if (role == HAST_ROLE_PRIMARY)
131 		hastd_primary(res);
132 	pjdlog_prefix_set("%s", "");
133 	hook_exec(res->hr_exec, "role", res->hr_name, role2str(oldrole),
134 	    role2str(res->hr_role), NULL);
135 }
136 
137 void
138 control_set_role(struct hast_resource *res, uint8_t role)
139 {
140 
141 	control_set_role_common(NULL, NULL, role, res, NULL, 0);
142 }
143 
144 static void
145 control_status_worker(struct hast_resource *res, struct nv *nvout,
146     unsigned int no)
147 {
148 	struct nv *cnvin, *cnvout;
149 	const char *str;
150 	int error;
151 
152 	cnvin = cnvout = NULL;
153 	error = 0;
154 
155 	/*
156 	 * Prepare and send command to worker process.
157 	 */
158 	cnvout = nv_alloc();
159 	nv_add_uint8(cnvout, HASTCTL_STATUS, "cmd");
160 	error = nv_error(cnvout);
161 	if (error != 0) {
162 		/* LOG */
163 		goto end;
164 	}
165 	if (hast_proto_send(res, res->hr_ctrl, cnvout, NULL, 0) < 0) {
166 		error = errno;
167 		/* LOG */
168 		goto end;
169 	}
170 
171 	/*
172 	 * Receive response.
173 	 */
174 	if (hast_proto_recv_hdr(res->hr_ctrl, &cnvin) < 0) {
175 		error = errno;
176 		/* LOG */
177 		goto end;
178 	}
179 
180 	error = nv_get_int64(cnvin, "error");
181 	if (error != 0)
182 		goto end;
183 
184 	if ((str = nv_get_string(cnvin, "status")) == NULL) {
185 		error = ENOENT;
186 		/* LOG */
187 		goto end;
188 	}
189 	nv_add_string(nvout, str, "status%u", no);
190 	nv_add_uint64(nvout, nv_get_uint64(cnvin, "dirty"), "dirty%u", no);
191 	nv_add_uint32(nvout, nv_get_uint32(cnvin, "extentsize"),
192 	    "extentsize%u", no);
193 	nv_add_uint32(nvout, nv_get_uint32(cnvin, "keepdirty"),
194 	    "keepdirty%u", no);
195 end:
196 	if (cnvin != NULL)
197 		nv_free(cnvin);
198 	if (cnvout != NULL)
199 		nv_free(cnvout);
200 	if (error != 0)
201 		nv_add_int16(nvout, error, "error");
202 }
203 
204 static void
205 control_status(struct hastd_config *cfg, struct nv *nvout,
206     struct hast_resource *res, const char *name, unsigned int no)
207 {
208 
209 	assert(cfg != NULL);
210 	assert(nvout != NULL);
211 	assert(name != NULL);
212 
213 	/* Name is always needed. */
214 	nv_add_string(nvout, name, "resource%u", no);
215 
216 	if (res == NULL) {
217 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
218 			if (strcmp(res->hr_name, name) == 0)
219 				break;
220 		}
221 		if (res == NULL) {
222 			nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
223 			return;
224 		}
225 	}
226 	assert(res != NULL);
227 	nv_add_string(nvout, res->hr_provname, "provname%u", no);
228 	nv_add_string(nvout, res->hr_localpath, "localpath%u", no);
229 	nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr%u", no);
230 	switch (res->hr_replication) {
231 	case HAST_REPLICATION_FULLSYNC:
232 		nv_add_string(nvout, "fullsync", "replication%u", no);
233 		break;
234 	case HAST_REPLICATION_MEMSYNC:
235 		nv_add_string(nvout, "memsync", "replication%u", no);
236 		break;
237 	case HAST_REPLICATION_ASYNC:
238 		nv_add_string(nvout, "async", "replication%u", no);
239 		break;
240 	default:
241 		nv_add_string(nvout, "unknown", "replication%u", no);
242 		break;
243 	}
244 	nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
245 
246 	switch (res->hr_role) {
247 	case HAST_ROLE_PRIMARY:
248 		assert(res->hr_workerpid != 0);
249 		/* FALLTHROUGH */
250 	case HAST_ROLE_SECONDARY:
251 		if (res->hr_workerpid != 0)
252 			break;
253 		/* FALLTHROUGH */
254 	default:
255 		return;
256 	}
257 
258 	/*
259 	 * If we are here, it means that we have a worker process, which we
260 	 * want to ask some questions.
261 	 */
262 	control_status_worker(res, nvout, no);
263 }
264 
265 void
266 control_handle(struct hastd_config *cfg)
267 {
268 	struct proto_conn *conn;
269 	struct nv *nvin, *nvout;
270 	unsigned int ii;
271 	const char *str;
272 	uint8_t cmd, role;
273 	int error;
274 
275 	if (proto_accept(cfg->hc_controlconn, &conn) < 0) {
276 		pjdlog_errno(LOG_ERR, "Unable to accept control connection");
277 		return;
278 	}
279 
280 	nvin = nvout = NULL;
281 	role = HAST_ROLE_UNDEF;
282 
283 	if (hast_proto_recv_hdr(conn, &nvin) < 0) {
284 		pjdlog_errno(LOG_ERR, "Unable to receive control header");
285 		nvin = NULL;
286 		goto close;
287 	}
288 
289 	/* Obtain command code. 0 means that nv_get_uint8() failed. */
290 	cmd = nv_get_uint8(nvin, "cmd");
291 	if (cmd == 0) {
292 		pjdlog_error("Control header is missing 'cmd' field.");
293 		error = EHAST_INVALID;
294 		goto close;
295 	}
296 
297 	/* Allocate outgoing nv structure. */
298 	nvout = nv_alloc();
299 	if (nvout == NULL) {
300 		pjdlog_error("Unable to allocate header for control response.");
301 		error = EHAST_NOMEMORY;
302 		goto close;
303 	}
304 
305 	error = 0;
306 
307 	str = nv_get_string(nvin, "resource0");
308 	if (str == NULL) {
309 		pjdlog_error("Control header is missing 'resource0' field.");
310 		error = EHAST_INVALID;
311 		goto fail;
312 	}
313 	if (cmd == HASTCTL_SET_ROLE) {
314 		role = nv_get_uint8(nvin, "role");
315 		switch (role) {
316 		case HAST_ROLE_INIT:	/* Is that valid to set, hmm? */
317 		case HAST_ROLE_PRIMARY:
318 		case HAST_ROLE_SECONDARY:
319 			break;
320 		default:
321 			pjdlog_error("Invalid role received (%hhu).", role);
322 			error = EHAST_INVALID;
323 			goto fail;
324 		}
325 	}
326 	if (strcmp(str, "all") == 0) {
327 		struct hast_resource *res;
328 
329 		/* All configured resources. */
330 
331 		ii = 0;
332 		TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
333 			switch (cmd) {
334 			case HASTCTL_SET_ROLE:
335 				control_set_role_common(cfg, nvout, role, res,
336 				    res->hr_name, ii++);
337 				break;
338 			case HASTCTL_STATUS:
339 				control_status(cfg, nvout, res, res->hr_name,
340 				    ii++);
341 				break;
342 			default:
343 				pjdlog_error("Invalid command received (%hhu).",
344 				    cmd);
345 				error = EHAST_UNIMPLEMENTED;
346 				goto fail;
347 			}
348 		}
349 	} else {
350 		/* Only selected resources. */
351 
352 		for (ii = 0; ; ii++) {
353 			str = nv_get_string(nvin, "resource%u", ii);
354 			if (str == NULL)
355 				break;
356 			switch (cmd) {
357 			case HASTCTL_SET_ROLE:
358 				control_set_role_common(cfg, nvout, role, NULL,
359 				    str, ii);
360 				break;
361 			case HASTCTL_STATUS:
362 				control_status(cfg, nvout, NULL, str, ii);
363 				break;
364 			default:
365 				pjdlog_error("Invalid command received (%hhu).",
366 				    cmd);
367 				error = EHAST_UNIMPLEMENTED;
368 				goto fail;
369 			}
370 		}
371 	}
372 	if (nv_error(nvout) != 0)
373 		goto close;
374 fail:
375 	if (error != 0)
376 		nv_add_int16(nvout, error, "error");
377 
378 	if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0)
379 		pjdlog_errno(LOG_ERR, "Unable to send control response");
380 close:
381 	if (nvin != NULL)
382 		nv_free(nvin);
383 	if (nvout != NULL)
384 		nv_free(nvout);
385 	proto_close(conn);
386 }
387 
388 /*
389  * Thread handles control requests from the parent.
390  */
391 void *
392 ctrl_thread(void *arg)
393 {
394 	struct hast_resource *res = arg;
395 	struct nv *nvin, *nvout;
396 	uint8_t cmd;
397 
398 	for (;;) {
399 		if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
400 			if (sigexit_received)
401 				pthread_exit(NULL);
402 			pjdlog_errno(LOG_ERR,
403 			    "Unable to receive control message");
404 			kill(getpid(), SIGTERM);
405 			pthread_exit(NULL);
406 		}
407 		cmd = nv_get_uint8(nvin, "cmd");
408 		if (cmd == 0) {
409 			pjdlog_error("Control message is missing 'cmd' field.");
410 			nv_free(nvin);
411 			continue;
412 		}
413 		nv_free(nvin);
414 		nvout = nv_alloc();
415 		switch (cmd) {
416 		case HASTCTL_STATUS:
417 			if (res->hr_remotein != NULL &&
418 			    res->hr_remoteout != NULL) {
419 				nv_add_string(nvout, "complete", "status");
420 			} else {
421 				nv_add_string(nvout, "degraded", "status");
422 			}
423 			nv_add_uint32(nvout, (uint32_t)res->hr_extentsize,
424 			    "extentsize");
425 			if (res->hr_role == HAST_ROLE_PRIMARY) {
426 				nv_add_uint32(nvout,
427 				    (uint32_t)res->hr_keepdirty, "keepdirty");
428 				nv_add_uint64(nvout,
429 				    (uint64_t)(activemap_ndirty(res->hr_amp) *
430 				    res->hr_extentsize), "dirty");
431 			} else {
432 				nv_add_uint32(nvout, (uint32_t)0, "keepdirty");
433 				nv_add_uint64(nvout, (uint64_t)0, "dirty");
434 			}
435 			break;
436 		default:
437 			nv_add_int16(nvout, EINVAL, "error");
438 			break;
439 		}
440 		if (nv_error(nvout) != 0) {
441 			pjdlog_error("Unable to create answer on control message.");
442 			nv_free(nvout);
443 			continue;
444 		}
445 		if (hast_proto_send(NULL, res->hr_ctrl, nvout, NULL, 0) < 0) {
446 			pjdlog_errno(LOG_ERR,
447 			    "Unable to send reply to control message");
448 		}
449 		nv_free(nvout);
450 	}
451 	/* NOTREACHED */
452 	return (NULL);
453 }
454