1 /*-
2 * Copyright (c) 2011, 2012, 2013, 2014, 2015, 2016 Spectra Logic Corporation
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions, and the following disclaimer,
10 * without modification.
11 * 2. Redistributions in binary form must reproduce at minimum a disclaimer
12 * substantially similar to the "NO WARRANTY" disclaimer below
13 * ("Disclaimer") and any redistribution must be conditioned upon
14 * including a substantially similar Disclaimer requirement for further
15 * binary redistribution.
16 *
17 * NO WARRANTY
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
21 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22 * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGES.
29 *
30 * Authors: Justin T. Gibbs (Spectra Logic Corporation)
31 */
32
33 /**
34 * \file zfsd.cc
35 *
36 * The ZFS daemon consumes kernel devdctl(4) event data via devd(8)'s
37 * unix domain socket in order to react to system changes that impact
38 * the function of ZFS storage pools. The goal of this daemon is to
39 * provide similar functionality to the Solaris ZFS Diagnostic Engine
40 * (zfs-diagnosis), the Solaris ZFS fault handler (zfs-retire), and
41 * the Solaris ZFS vdev insertion agent (zfs-mod sysevent handler).
42 */
43
44 #include <sys/cdefs.h>
45 #include <sys/param.h>
46 #include <sys/fs/zfs.h>
47
48 #include <err.h>
49 #include <libgeom.h>
50 #include <libutil.h>
51 #include <poll.h>
52 #include <syslog.h>
53
54 #include <libzfs.h>
55
56 #include <list>
57 #include <map>
58 #include <string>
59
60 #include <devdctl/guid.h>
61 #include <devdctl/event.h>
62 #include <devdctl/event_factory.h>
63 #include <devdctl/exception.h>
64 #include <devdctl/consumer.h>
65
66 #include "callout.h"
67 #include "vdev_iterator.h"
68 #include "zfsd_event.h"
69 #include "case_file.h"
70 #include "vdev.h"
71 #include "vdev_iterator.h"
72 #include "zfsd.h"
73 #include "zfsd_exception.h"
74 #include "zpool_list.h"
75
76 __FBSDID("$FreeBSD$");
77
78 /*================================== Macros ==================================*/
79 #define NUM_ELEMENTS(x) (sizeof(x) / sizeof(*x))
80
81 /*============================ Namespace Control =============================*/
82 using DevdCtl::Event;
83 using DevdCtl::EventFactory;
84 using DevdCtl::EventList;
85
86 /*================================ Global Data ===============================*/
87 int g_debug = 0;
88 libzfs_handle_t *g_zfsHandle;
89
90 /*--------------------------------- ZfsDaemon --------------------------------*/
91 //- ZfsDaemon Static Private Data ----------------------------------------------
92 ZfsDaemon *ZfsDaemon::s_theZfsDaemon;
93 bool ZfsDaemon::s_logCaseFiles;
94 bool ZfsDaemon::s_terminateEventLoop;
95 char ZfsDaemon::s_pidFilePath[] = "/var/run/zfsd.pid";
96 pidfh *ZfsDaemon::s_pidFH;
97 int ZfsDaemon::s_signalPipeFD[2];
98 bool ZfsDaemon::s_systemRescanRequested(false);
99 EventFactory::Record ZfsDaemon::s_registryEntries[] =
100 {
101 { Event::NOTIFY, "GEOM", &GeomEvent::Builder },
102 { Event::NOTIFY, "ZFS", &ZfsEvent::Builder }
103 };
104
105 //- ZfsDaemon Static Public Methods --------------------------------------------
106 ZfsDaemon &
Get()107 ZfsDaemon::Get()
108 {
109 return (*s_theZfsDaemon);
110 }
111
112 void
WakeEventLoop()113 ZfsDaemon::WakeEventLoop()
114 {
115 write(s_signalPipeFD[1], "+", 1);
116 }
117
118 void
RequestSystemRescan()119 ZfsDaemon::RequestSystemRescan()
120 {
121 s_systemRescanRequested = true;
122 ZfsDaemon::WakeEventLoop();
123 }
124
125 void
Run()126 ZfsDaemon::Run()
127 {
128 ZfsDaemon daemon;
129
130 while (s_terminateEventLoop == false) {
131
132 try {
133 daemon.DisconnectFromDevd();
134
135 if (daemon.ConnectToDevd() == false) {
136 sleep(30);
137 continue;
138 }
139
140 daemon.DetectMissedEvents();
141
142 daemon.EventLoop();
143
144 } catch (const DevdCtl::Exception &exp) {
145 exp.Log();
146 }
147 }
148
149 daemon.DisconnectFromDevd();
150 }
151
152 //- ZfsDaemon Private Methods --------------------------------------------------
ZfsDaemon()153 ZfsDaemon::ZfsDaemon()
154 : Consumer(/*defBuilder*/NULL, s_registryEntries,
155 NUM_ELEMENTS(s_registryEntries))
156 {
157 if (s_theZfsDaemon != NULL)
158 errx(1, "Multiple ZfsDaemon instances created. Exiting");
159
160 s_theZfsDaemon = this;
161
162 if (pipe(s_signalPipeFD) != 0)
163 errx(1, "Unable to allocate signal pipe. Exiting");
164
165 if (fcntl(s_signalPipeFD[0], F_SETFL, O_NONBLOCK) == -1)
166 errx(1, "Unable to set pipe as non-blocking. Exiting");
167
168 if (fcntl(s_signalPipeFD[1], F_SETFL, O_NONBLOCK) == -1)
169 errx(1, "Unable to set pipe as non-blocking. Exiting");
170
171 signal(SIGHUP, ZfsDaemon::RescanSignalHandler);
172 signal(SIGINFO, ZfsDaemon::InfoSignalHandler);
173 signal(SIGINT, ZfsDaemon::QuitSignalHandler);
174 signal(SIGTERM, ZfsDaemon::QuitSignalHandler);
175 signal(SIGUSR1, ZfsDaemon::RescanSignalHandler);
176
177 g_zfsHandle = libzfs_init();
178 if (g_zfsHandle == NULL)
179 errx(1, "Unable to initialize ZFS library. Exiting");
180
181 Callout::Init();
182 InitializeSyslog();
183 OpenPIDFile();
184
185 if (g_debug == 0)
186 daemon(0, 0);
187
188 UpdatePIDFile();
189 }
190
~ZfsDaemon()191 ZfsDaemon::~ZfsDaemon()
192 {
193 PurgeCaseFiles();
194 ClosePIDFile();
195 }
196
197 void
PurgeCaseFiles()198 ZfsDaemon::PurgeCaseFiles()
199 {
200 CaseFile::PurgeAll();
201 }
202
203 bool
VdevAddCaseFile(Vdev & vdev,void * cbArg)204 ZfsDaemon::VdevAddCaseFile(Vdev &vdev, void *cbArg)
205 {
206 if (vdev.State() != VDEV_STATE_HEALTHY)
207 CaseFile::Create(vdev);
208
209 return (/*break early*/false);
210 }
211
212 void
BuildCaseFiles()213 ZfsDaemon::BuildCaseFiles()
214 {
215 ZpoolList zpl;
216 ZpoolList::iterator pool;
217
218 /* Add CaseFiles for vdevs with issues. */
219 for (pool = zpl.begin(); pool != zpl.end(); pool++)
220 VdevIterator(*pool).Each(VdevAddCaseFile, NULL);
221
222 /* De-serialize any saved cases. */
223 CaseFile::DeSerialize();
224
225 /* Simulate config_sync events to force CaseFile reevaluation */
226 for (pool = zpl.begin(); pool != zpl.end(); pool++) {
227 char evString[160];
228 Event *event;
229 nvlist_t *config;
230 uint64_t poolGUID;
231 const char *poolname;
232
233 poolname = zpool_get_name(*pool);
234 config = zpool_get_config(*pool, NULL);
235 if (config == NULL) {
236 syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
237 "find pool config for pool %s", poolname);
238 continue;
239 }
240 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
241 &poolGUID) != 0) {
242 syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
243 "find pool guid for pool %s", poolname);
244 continue;
245 }
246
247
248 snprintf(evString, 160, "!system=ZFS subsystem=ZFS "
249 "type=misc.fs.zfs.config_sync sub_type=synthesized "
250 "pool_name=%s pool_guid=%" PRIu64 "\n", poolname, poolGUID);
251 event = Event::CreateEvent(GetFactory(), string(evString));
252 if (event != NULL) {
253 event->Process();
254 delete event;
255 }
256 }
257 }
258
259 void
RescanSystem()260 ZfsDaemon::RescanSystem()
261 {
262 struct gmesh mesh;
263 struct gclass *mp;
264 struct ggeom *gp;
265 struct gprovider *pp;
266 int result;
267
268 /*
269 * The devdctl system doesn't replay events for new consumers
270 * of the interface. Emit manufactured DEVFS arrival events
271 * for any devices that already before we started or during
272 * periods where we've lost our connection to devd.
273 */
274 result = geom_gettree(&mesh);
275 if (result != 0) {
276 syslog(LOG_ERR, "ZfsDaemon::RescanSystem: "
277 "geom_gettree faild with error %d\n", result);
278 return;
279 }
280
281 const string evStart("!system=DEVFS subsystem=CDEV type=CREATE "
282 "sub_type=synthesized cdev=");
283 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
284 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
285 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
286 Event *event;
287
288 string evString(evStart + pp->lg_name + "\n");
289 event = Event::CreateEvent(GetFactory(),
290 evString);
291 if (event != NULL) {
292 if (event->Process())
293 SaveEvent(*event);
294 delete event;
295 }
296 }
297 }
298 }
299 geom_deletetree(&mesh);
300 }
301
302 void
DetectMissedEvents()303 ZfsDaemon::DetectMissedEvents()
304 {
305 do {
306 PurgeCaseFiles();
307
308 /*
309 * Discard any events waiting for us. We don't know
310 * if they still apply to the current state of the
311 * system.
312 */
313 FlushEvents();
314
315 BuildCaseFiles();
316
317 /*
318 * If the system state has changed during our
319 * interrogation, start over.
320 */
321 } while (s_terminateEventLoop == false && EventsPending());
322
323 RescanSystem();
324 }
325
326 void
EventLoop()327 ZfsDaemon::EventLoop()
328 {
329 while (s_terminateEventLoop == false) {
330 struct pollfd fds[2];
331 int result;
332
333 if (s_logCaseFiles == true) {
334 EventList::iterator event(m_unconsumedEvents.begin());
335 s_logCaseFiles = false;
336 CaseFile::LogAll();
337 while (event != m_unconsumedEvents.end())
338 (*event++)->Log(LOG_INFO);
339 }
340
341 Callout::ExpireCallouts();
342
343 /* Wait for data. */
344 fds[0].fd = m_devdSockFD;
345 fds[0].events = POLLIN;
346 fds[0].revents = 0;
347 fds[1].fd = s_signalPipeFD[0];
348 fds[1].events = POLLIN;
349 fds[1].revents = 0;
350 result = poll(fds, NUM_ELEMENTS(fds), /*timeout*/INFTIM);
351 if (result == -1) {
352 if (errno == EINTR)
353 continue;
354 else
355 err(1, "Polling for devd events failed");
356 } else if (result == 0) {
357 errx(1, "Unexpected result of 0 from poll. Exiting");
358 }
359
360 if ((fds[0].revents & POLLIN) != 0)
361 ProcessEvents();
362
363 if ((fds[1].revents & POLLIN) != 0) {
364 static char discardBuf[128];
365
366 /*
367 * This pipe exists just to close the signal
368 * race. Its contents are of no interest to
369 * us, but we must ensure that future signals
370 * have space in the pipe to write.
371 */
372 while (read(s_signalPipeFD[0], discardBuf,
373 sizeof(discardBuf)) > 0)
374 ;
375 }
376
377 if (s_systemRescanRequested == true) {
378 s_systemRescanRequested = false;
379 syslog(LOG_INFO, "System Rescan request processed.");
380 RescanSystem();
381 }
382
383 if ((fds[0].revents & POLLERR) != 0) {
384 syslog(LOG_INFO, "POLLERROR detected on devd socket.");
385 break;
386 }
387
388 if ((fds[0].revents & POLLHUP) != 0) {
389 syslog(LOG_INFO, "POLLHUP detected on devd socket.");
390 break;
391 }
392 }
393 }
394 //- ZfsDaemon staic Private Methods --------------------------------------------
395 void
InfoSignalHandler(int)396 ZfsDaemon::InfoSignalHandler(int)
397 {
398 s_logCaseFiles = true;
399 ZfsDaemon::WakeEventLoop();
400 }
401
402 void
RescanSignalHandler(int)403 ZfsDaemon::RescanSignalHandler(int)
404 {
405 RequestSystemRescan();
406 }
407
408 void
QuitSignalHandler(int)409 ZfsDaemon::QuitSignalHandler(int)
410 {
411 s_terminateEventLoop = true;
412 ZfsDaemon::WakeEventLoop();
413 }
414
415 void
OpenPIDFile()416 ZfsDaemon::OpenPIDFile()
417 {
418 pid_t otherPID;
419
420 s_pidFH = pidfile_open(s_pidFilePath, 0600, &otherPID);
421 if (s_pidFH == NULL) {
422 if (errno == EEXIST)
423 errx(1, "already running as PID %d. Exiting", otherPID);
424 warn("cannot open PID file");
425 }
426 }
427
428 void
UpdatePIDFile()429 ZfsDaemon::UpdatePIDFile()
430 {
431 if (s_pidFH != NULL)
432 pidfile_write(s_pidFH);
433 }
434
435 void
ClosePIDFile()436 ZfsDaemon::ClosePIDFile()
437 {
438 if (s_pidFH != NULL)
439 pidfile_remove(s_pidFH);
440 }
441
442 void
InitializeSyslog()443 ZfsDaemon::InitializeSyslog()
444 {
445 openlog("zfsd", LOG_NDELAY, LOG_DAEMON);
446 }
447
448