]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - cddl/usr.sbin/zfsd/case_file.cc
zfsd(8), the ZFS fault management daemon
[FreeBSD/FreeBSD.git] / cddl / usr.sbin / zfsd / case_file.cc
1 /*-
2  * Copyright (c) 2011, 2012, 2013, 2014, 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 case_file.cc
35  *
36  * We keep case files for any leaf vdev that is not in the optimal state.
37  * However, we only serialize to disk those events that need to be preserved
38  * across reboots.  For now, this is just a log of soft errors which we
39  * accumulate in order to mark a device as degraded.
40  */
41 #include <sys/cdefs.h>
42 #include <sys/time.h>
43
44 #include <sys/fs/zfs.h>
45
46 #include <dirent.h>
47 #include <iomanip>
48 #include <fstream>
49 #include <functional>
50 #include <sstream>
51 #include <syslog.h>
52 #include <unistd.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 "zfsd.h"
72 #include "zfsd_exception.h"
73 #include "zpool_list.h"
74
75 __FBSDID("$FreeBSD$");
76
77 /*============================ Namespace Control =============================*/
78 using std::auto_ptr;
79 using std::hex;
80 using std::ifstream;
81 using std::stringstream;
82 using std::setfill;
83 using std::setw;
84
85 using DevdCtl::Event;
86 using DevdCtl::EventBuffer;
87 using DevdCtl::EventFactory;
88 using DevdCtl::EventList;
89 using DevdCtl::Guid;
90 using DevdCtl::ParseException;
91
92 /*--------------------------------- CaseFile ---------------------------------*/
93 //- CaseFile Static Data -------------------------------------------------------
94
95 CaseFileList  CaseFile::s_activeCases;
96 const string  CaseFile::s_caseFilePath = "/var/db/zfsd/cases";
97 const timeval CaseFile::s_removeGracePeriod = { 60 /*sec*/, 0 /*usec*/};
98
99 //- CaseFile Static Public Methods ---------------------------------------------
100 CaseFile *
101 CaseFile::Find(Guid poolGUID, Guid vdevGUID)
102 {
103         for (CaseFileList::iterator curCase = s_activeCases.begin();
104              curCase != s_activeCases.end(); curCase++) {
105
106                 if ((*curCase)->PoolGUID() != poolGUID
107                  || (*curCase)->VdevGUID() != vdevGUID)
108                         continue;
109
110                 /*
111                  * We only carry one active case per-vdev.
112                  */
113                 return (*curCase);
114         }
115         return (NULL);
116 }
117
118 CaseFile *
119 CaseFile::Find(const string &physPath)
120 {
121         CaseFile *result = NULL;
122
123         for (CaseFileList::iterator curCase = s_activeCases.begin();
124              curCase != s_activeCases.end(); curCase++) {
125
126                 if ((*curCase)->PhysicalPath() != physPath)
127                         continue;
128
129                 if (result != NULL) {
130                         syslog(LOG_WARNING, "Multiple casefiles found for "
131                             "physical path %s.  "
132                             "This is most likely a bug in zfsd",
133                             physPath.c_str());
134                 }
135                 result = *curCase;
136         }
137         return (result);
138 }
139
140
141 void
142 CaseFile::ReEvaluateByGuid(Guid poolGUID, const ZfsEvent &event)
143 {
144         CaseFileList::iterator casefile;
145         for (casefile = s_activeCases.begin(); casefile != s_activeCases.end();){
146                 CaseFileList::iterator next = casefile;
147                 next++;
148                 if (poolGUID == (*casefile)->PoolGUID())
149                         (*casefile)->ReEvaluate(event);
150                 casefile = next;
151         }
152 }
153
154 CaseFile &
155 CaseFile::Create(Vdev &vdev)
156 {
157         CaseFile *activeCase;
158
159         activeCase = Find(vdev.PoolGUID(), vdev.GUID());
160         if (activeCase == NULL)
161                 activeCase = new CaseFile(vdev);
162
163         return (*activeCase);
164 }
165
166 void
167 CaseFile::DeSerialize()
168 {
169         struct dirent **caseFiles;
170
171         int numCaseFiles(scandir(s_caseFilePath.c_str(), &caseFiles,
172                          DeSerializeSelector, /*compar*/NULL));
173
174         if (numCaseFiles == -1)
175                 return;
176         if (numCaseFiles == 0) {
177                 free(caseFiles);
178                 return;
179         }
180
181         for (int i = 0; i < numCaseFiles; i++) {
182
183                 DeSerializeFile(caseFiles[i]->d_name);
184                 free(caseFiles[i]);
185         }
186         free(caseFiles);
187 }
188
189 void
190 CaseFile::LogAll()
191 {
192         for (CaseFileList::iterator curCase = s_activeCases.begin();
193              curCase != s_activeCases.end(); curCase++)
194                 (*curCase)->Log();
195 }
196
197 void
198 CaseFile::PurgeAll()
199 {
200         /*
201          * Serialize casefiles before deleting them so that they can be reread
202          * and revalidated during BuildCaseFiles.
203          * CaseFiles remove themselves from this list on destruction.
204          */
205         while (s_activeCases.size() != 0) {
206                 CaseFile *casefile = s_activeCases.front();
207                 casefile->Serialize();
208                 delete casefile;
209         }
210
211 }
212
213 //- CaseFile Public Methods ----------------------------------------------------
214 bool
215 CaseFile::RefreshVdevState()
216 {
217         ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
218         zpool_handle_t *casePool(zpl.empty() ? NULL : zpl.front());
219         if (casePool == NULL)
220                 return (false);
221
222         Vdev vd(casePool, CaseVdev(casePool));
223         if (vd.DoesNotExist())
224                 return (false);
225
226         m_vdevState    = vd.State();
227         m_vdevPhysPath = vd.PhysicalPath();
228         return (true);
229 }
230
231 bool
232 CaseFile::ReEvaluate(const string &devPath, const string &physPath, Vdev *vdev)
233 {
234         ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
235         zpool_handle_t *pool(zpl.empty() ? NULL : zpl.front());
236
237         if (pool == NULL || !RefreshVdevState()) {
238                 /*
239                  * The pool or vdev for this case file is no longer
240                  * part of the configuration.  This can happen
241                  * if we process a device arrival notification
242                  * before seeing the ZFS configuration change
243                  * event.
244                  */
245                 syslog(LOG_INFO,
246                        "CaseFile::ReEvaluate(%s,%s) Pool/Vdev unconfigured.  "
247                        "Closing\n",
248                        PoolGUIDString().c_str(),
249                        VdevGUIDString().c_str());
250                 Close();
251
252                 /*
253                  * Since this event was not used to close this
254                  * case, do not report it as consumed.
255                  */
256                 return (/*consumed*/false);
257         }
258
259         if (VdevState() > VDEV_STATE_CANT_OPEN) {
260                 /*
261                  * For now, newly discovered devices only help for
262                  * devices that are missing.  In the future, we might
263                  * use a newly inserted spare to replace a degraded
264                  * or faulted device.
265                  */
266                 syslog(LOG_INFO, "CaseFile::ReEvaluate(%s,%s): Pool/Vdev ignored",
267                     PoolGUIDString().c_str(), VdevGUIDString().c_str());
268                 return (/*consumed*/false);
269         }
270
271         if (vdev != NULL
272          && vdev->PoolGUID() == m_poolGUID
273          && vdev->GUID() == m_vdevGUID) {
274
275                 zpool_vdev_online(pool, vdev->GUIDString().c_str(),
276                                   ZFS_ONLINE_CHECKREMOVE | ZFS_ONLINE_UNSPARE,
277                                   &m_vdevState);
278                 syslog(LOG_INFO, "Onlined vdev(%s/%s:%s).  State now %s.\n",
279                        zpool_get_name(pool), vdev->GUIDString().c_str(),
280                        devPath.c_str(),
281                        zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
282
283                 /*
284                  * Check the vdev state post the online action to see
285                  * if we can retire this case.
286                  */
287                 CloseIfSolved();
288
289                 return (/*consumed*/true);
290         }
291
292         /*
293          * If the auto-replace policy is enabled, and we have physical
294          * path information, try a physical path replacement.
295          */
296         if (zpool_get_prop_int(pool, ZPOOL_PROP_AUTOREPLACE, NULL) == 0) {
297                 syslog(LOG_INFO,
298                        "CaseFile(%s:%s:%s): AutoReplace not set.  "
299                        "Ignoring device insertion.\n",
300                        PoolGUIDString().c_str(),
301                        VdevGUIDString().c_str(),
302                        zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
303                 return (/*consumed*/false);
304         }
305
306         if (PhysicalPath().empty()) {
307                 syslog(LOG_INFO,
308                        "CaseFile(%s:%s:%s): No physical path information.  "
309                        "Ignoring device insertion.\n",
310                        PoolGUIDString().c_str(),
311                        VdevGUIDString().c_str(),
312                        zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
313                 return (/*consumed*/false);
314         }
315
316         if (physPath != PhysicalPath()) {
317                 syslog(LOG_INFO,
318                        "CaseFile(%s:%s:%s): Physical path mismatch.  "
319                        "Ignoring device insertion.\n",
320                        PoolGUIDString().c_str(),
321                        VdevGUIDString().c_str(),
322                        zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
323                 return (/*consumed*/false);
324         }
325
326         /* Write a label on the newly inserted disk. */
327         if (zpool_label_disk(g_zfsHandle, pool, devPath.c_str()) != 0) {
328                 syslog(LOG_ERR,
329                        "Replace vdev(%s/%s) by physical path (label): %s: %s\n",
330                        zpool_get_name(pool), VdevGUIDString().c_str(),
331                        libzfs_error_action(g_zfsHandle),
332                        libzfs_error_description(g_zfsHandle));
333                 return (/*consumed*/false);
334         }
335
336         syslog(LOG_INFO, "CaseFile::ReEvaluate(%s/%s): Replacing with %s",
337             PoolGUIDString().c_str(), VdevGUIDString().c_str(),
338             devPath.c_str());
339         return (Replace(VDEV_TYPE_DISK, devPath.c_str(), /*isspare*/false));
340 }
341
342 bool
343 CaseFile::ReEvaluate(const ZfsEvent &event)
344 {
345         bool consumed(false);
346
347         if (event.Value("type") == "misc.fs.zfs.vdev_remove") {
348                 /*
349                  * The Vdev we represent has been removed from the
350                  * configuration.  This case is no longer of value.
351                  */
352                 Close();
353
354                 return (/*consumed*/true);
355         } else if (event.Value("type") == "misc.fs.zfs.pool_destroy") {
356                 /* This Pool has been destroyed.  Discard the case */
357                 Close();
358
359                 return (/*consumed*/true);
360         } else if (event.Value("type") == "misc.fs.zfs.config_sync") {
361                 RefreshVdevState();
362                 if (VdevState() < VDEV_STATE_HEALTHY)
363                         consumed = ActivateSpare();
364         }
365
366
367         if (event.Value("class") == "resource.fs.zfs.removed") {
368                 bool spare_activated;
369
370                 if (!RefreshVdevState()) {
371                         /*
372                          * The pool or vdev for this case file is no longer
373                          * part of the configuration.  This can happen
374                          * if we process a device arrival notification
375                          * before seeing the ZFS configuration change
376                          * event.
377                          */
378                         syslog(LOG_INFO,
379                                "CaseFile::ReEvaluate(%s,%s) Pool/Vdev "
380                                "unconfigured.  Closing\n",
381                                PoolGUIDString().c_str(),
382                                VdevGUIDString().c_str());
383                         /*
384                          * Close the case now so we won't waste cycles in the
385                          * system rescan
386                          */
387                         Close();
388
389                         /*
390                          * Since this event was not used to close this
391                          * case, do not report it as consumed.
392                          */
393                         return (/*consumed*/false);
394                 }
395
396                 /*
397                  * Discard any tentative I/O error events for
398                  * this case.  They were most likely caused by the
399                  * hot-unplug of this device.
400                  */
401                 PurgeTentativeEvents();
402
403                 /* Try to activate spares if they are available */
404                 spare_activated = ActivateSpare();
405
406                 /*
407                  * Rescan the drives in the system to see if a recent
408                  * drive arrival can be used to solve this case.
409                  */
410                 ZfsDaemon::RequestSystemRescan();
411
412                 /*
413                  * Consume the event if we successfully activated a spare.
414                  * Otherwise, leave it in the unconsumed events list so that the
415                  * future addition of a spare to this pool might be able to
416                  * close the case
417                  */
418                 consumed = spare_activated;
419         } else if (event.Value("class") == "resource.fs.zfs.statechange") {
420                 RefreshVdevState();
421                 /*
422                  * If this vdev is DEGRADED, FAULTED, or UNAVAIL, try to
423                  * activate a hotspare.  Otherwise, ignore the event
424                  */
425                 if (VdevState() == VDEV_STATE_FAULTED ||
426                     VdevState() == VDEV_STATE_DEGRADED ||
427                     VdevState() == VDEV_STATE_CANT_OPEN)
428                         (void) ActivateSpare();
429                 consumed = true;
430         }
431         else if (event.Value("class") == "ereport.fs.zfs.io" ||
432                  event.Value("class") == "ereport.fs.zfs.checksum") {
433
434                 m_tentativeEvents.push_front(event.DeepCopy());
435                 RegisterCallout(event);
436                 consumed = true;
437         }
438
439         bool closed(CloseIfSolved());
440
441         return (consumed || closed);
442 }
443
444
445 bool
446 CaseFile::ActivateSpare() {
447         nvlist_t        *config, *nvroot;
448         nvlist_t       **spares;
449         char            *devPath, *vdev_type;
450         const char      *poolname;
451         u_int            nspares, i;
452         int              error;
453
454         ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
455         zpool_handle_t  *zhp(zpl.empty() ? NULL : zpl.front());
456         if (zhp == NULL) {
457                 syslog(LOG_ERR, "CaseFile::ActivateSpare: Could not find pool "
458                        "for pool_guid %"PRIu64".", (uint64_t)m_poolGUID);
459                 return (false);
460         }
461         poolname = zpool_get_name(zhp);
462         config = zpool_get_config(zhp, NULL);
463         if (config == NULL) {
464                 syslog(LOG_ERR, "CaseFile::ActivateSpare: Could not find pool "
465                        "config for pool %s", poolname);
466                 return (false);
467         }
468         error = nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot);
469         if (error != 0){
470                 syslog(LOG_ERR, "CaseFile::ActivateSpare: Could not find vdev "
471                        "tree for pool %s", poolname);
472                 return (false);
473         }
474         nspares = 0;
475         nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
476                                    &nspares);
477         if (nspares == 0) {
478                 /* The pool has no spares configured */
479                 syslog(LOG_INFO, "CaseFile::ActivateSpare: "
480                        "No spares available for pool %s", poolname);
481                 return (false);
482         }
483         for (i = 0; i < nspares; i++) {
484                 uint64_t    *nvlist_array;
485                 vdev_stat_t *vs;
486                 uint_t       nstats;
487
488                 if (nvlist_lookup_uint64_array(spares[i],
489                     ZPOOL_CONFIG_VDEV_STATS, &nvlist_array, &nstats) != 0) {
490                         syslog(LOG_ERR, "CaseFile::ActivateSpare: Could not "
491                                "find vdev stats for pool %s, spare %d",
492                                poolname, i);
493                         return (false);
494                 }
495                 vs = reinterpret_cast<vdev_stat_t *>(nvlist_array);
496
497                 if ((vs->vs_aux != VDEV_AUX_SPARED)
498                  && (vs->vs_state == VDEV_STATE_HEALTHY)) {
499                         /* We found a usable spare */
500                         break;
501                 }
502         }
503
504         if (i == nspares) {
505                 /* No available spares were found */
506                 return (false);
507         }
508
509         error = nvlist_lookup_string(spares[i], ZPOOL_CONFIG_PATH, &devPath);
510         if (error != 0) {
511                 syslog(LOG_ERR, "CaseFile::ActivateSpare: Cannot determine "
512                        "the path of pool %s, spare %d. Error %d",
513                        poolname, i, error);
514                 return (false);
515         }
516
517         error = nvlist_lookup_string(spares[i], ZPOOL_CONFIG_TYPE, &vdev_type);
518         if (error != 0) {
519                 syslog(LOG_ERR, "CaseFile::ActivateSpare: Cannot determine "
520                        "the vdev type of pool %s, spare %d. Error %d",
521                        poolname, i, error);
522                 return (false);
523         }
524
525         return (Replace(vdev_type, devPath, /*isspare*/true));
526 }
527
528 void
529 CaseFile::RegisterCallout(const Event &event)
530 {
531         timeval now, countdown, elapsed, timestamp, zero, remaining;
532
533         gettimeofday(&now, 0);
534         timestamp = event.GetTimestamp();
535         timersub(&now, &timestamp, &elapsed);
536         timersub(&s_removeGracePeriod, &elapsed, &countdown);
537         /*
538          * If countdown is <= zero, Reset the timer to the
539          * smallest positive time value instead
540          */
541         timerclear(&zero);
542         if (timercmp(&countdown, &zero, <=)) {
543                 timerclear(&countdown);
544                 countdown.tv_usec = 1;
545         }
546
547         remaining = m_tentativeTimer.TimeRemaining();
548
549         if (!m_tentativeTimer.IsPending()
550          || timercmp(&countdown, &remaining, <))
551                 m_tentativeTimer.Reset(countdown, OnGracePeriodEnded, this);
552 }
553
554
555 bool
556 CaseFile::CloseIfSolved()
557 {
558         if (m_events.empty()
559          && m_tentativeEvents.empty()) {
560
561                 /*
562                  * We currently do not track or take actions on
563                  * devices in the degraded or faulted state.
564                  * Once we have support for spare pools, we'll
565                  * retain these cases so that any spares added in
566                  * the future can be applied to them.
567                  */
568                 switch (VdevState()) {
569                 case VDEV_STATE_HEALTHY:
570                         /* No need to keep cases for healthy vdevs */
571                         Close();
572                         return (true);
573                 case VDEV_STATE_REMOVED:
574                 case VDEV_STATE_CANT_OPEN:
575                         /*
576                          * Keep open.  We may solve it with a newly inserted
577                          * device.
578                          */
579                 case VDEV_STATE_FAULTED:
580                 case VDEV_STATE_DEGRADED:
581                         /*
582                          * Keep open.  We may solve it with the future
583                          * addition of a spare to the pool
584                          */
585                 case VDEV_STATE_UNKNOWN:
586                 case VDEV_STATE_CLOSED:
587                 case VDEV_STATE_OFFLINE:
588                         /*
589                          * Keep open?  This may not be the correct behavior,
590                          * but it's what we've always done
591                          */
592                         ;
593                 }
594
595                 /*
596                  * Re-serialize the case in order to remove any
597                  * previous event data.
598                  */
599                 Serialize();
600         }
601
602         return (false);
603 }
604
605 void
606 CaseFile::Log()
607 {
608         syslog(LOG_INFO, "CaseFile(%s,%s,%s)\n", PoolGUIDString().c_str(),
609                VdevGUIDString().c_str(), PhysicalPath().c_str());
610         syslog(LOG_INFO, "\tVdev State = %s\n",
611                zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
612         if (m_tentativeEvents.size() != 0) {
613                 syslog(LOG_INFO, "\t=== Tentative Events ===\n");
614                 for (EventList::iterator event(m_tentativeEvents.begin());
615                      event != m_tentativeEvents.end(); event++)
616                         (*event)->Log(LOG_INFO);
617         }
618         if (m_events.size() != 0) {
619                 syslog(LOG_INFO, "\t=== Events ===\n");
620                 for (EventList::iterator event(m_events.begin());
621                      event != m_events.end(); event++)
622                         (*event)->Log(LOG_INFO);
623         }
624 }
625
626 //- CaseFile Static Protected Methods ------------------------------------------
627 void
628 CaseFile::OnGracePeriodEnded(void *arg)
629 {
630         CaseFile &casefile(*static_cast<CaseFile *>(arg));
631
632         casefile.OnGracePeriodEnded();
633 }
634
635 int
636 CaseFile::DeSerializeSelector(const struct dirent *dirEntry)
637 {
638         uint64_t poolGUID;
639         uint64_t vdevGUID;
640
641         if (dirEntry->d_type == DT_REG
642          && sscanf(dirEntry->d_name, "pool_%"PRIu64"_vdev_%"PRIu64".case",
643                    &poolGUID, &vdevGUID) == 2)
644                 return (1);
645         return (0);
646 }
647
648 void
649 CaseFile::DeSerializeFile(const char *fileName)
650 {
651         string    fullName(s_caseFilePath + '/' + fileName);
652         CaseFile *existingCaseFile(NULL);
653         CaseFile *caseFile(NULL);
654
655         try {
656                 uint64_t poolGUID;
657                 uint64_t vdevGUID;
658                 nvlist_t *vdevConf;
659
660                 sscanf(fileName, "pool_%"PRIu64"_vdev_%"PRIu64".case",
661                        &poolGUID, &vdevGUID);
662                 existingCaseFile = Find(Guid(poolGUID), Guid(vdevGUID));
663                 if (existingCaseFile != NULL) {
664                         /*
665                          * If the vdev is already degraded or faulted,
666                          * there's no point in keeping the state around
667                          * that we use to put a drive into the degraded
668                          * state.  However, if the vdev is simply missing,
669                          * preserve the case data in the hopes that it will
670                          * return.
671                          */
672                         caseFile = existingCaseFile;
673                         vdev_state curState(caseFile->VdevState());
674                         if (curState > VDEV_STATE_CANT_OPEN
675                          && curState < VDEV_STATE_HEALTHY) {
676                                 unlink(fileName);
677                                 return;
678                         }
679                 } else {
680                         ZpoolList zpl(ZpoolList::ZpoolByGUID, &poolGUID);
681                         if (zpl.empty()
682                          || (vdevConf = VdevIterator(zpl.front())
683                                                     .Find(vdevGUID)) == NULL) {
684                                 /*
685                                  * Either the pool no longer exists
686                                  * or this vdev is no longer a member of
687                                  * the pool.
688                                  */
689                                 unlink(fullName.c_str());
690                                 return;
691                         }
692
693                         /*
694                          * Any vdev we find that does not have a case file
695                          * must be in the healthy state and thus worthy of
696                          * continued SERD data tracking.
697                          */
698                         caseFile = new CaseFile(Vdev(zpl.front(), vdevConf));
699                 }
700
701                 ifstream caseStream(fullName.c_str());
702                 if (!caseStream)
703                         throw ZfsdException("CaseFile::DeSerialize: Unable to "
704                                             "read %s.\n", fileName);
705
706                 caseFile->DeSerialize(caseStream);
707         } catch (const ParseException &exp) {
708
709                 exp.Log();
710                 if (caseFile != existingCaseFile)
711                         delete caseFile;
712
713                 /*
714                  * Since we can't parse the file, unlink it so we don't
715                  * trip over it again.
716                  */
717                 unlink(fileName);
718         } catch (const ZfsdException &zfsException) {
719
720                 zfsException.Log();
721                 if (caseFile != existingCaseFile)
722                         delete caseFile;
723         }
724 }
725
726 //- CaseFile Protected Methods -------------------------------------------------
727 CaseFile::CaseFile(const Vdev &vdev)
728  : m_poolGUID(vdev.PoolGUID()),
729    m_vdevGUID(vdev.GUID()),
730    m_vdevState(vdev.State()),
731    m_vdevPhysPath(vdev.PhysicalPath())
732 {
733         stringstream guidString;
734
735         guidString << m_vdevGUID;
736         m_vdevGUIDString = guidString.str();
737         guidString.str("");
738         guidString << m_poolGUID;
739         m_poolGUIDString = guidString.str();
740
741         s_activeCases.push_back(this);
742
743         syslog(LOG_INFO, "Creating new CaseFile:\n");
744         Log();
745 }
746
747 CaseFile::~CaseFile()
748 {
749         PurgeEvents();
750         PurgeTentativeEvents();
751         m_tentativeTimer.Stop();
752         s_activeCases.remove(this);
753 }
754
755 void
756 CaseFile::PurgeEvents()
757 {
758         for (EventList::iterator event(m_events.begin());
759              event != m_events.end(); event++)
760                 delete *event;
761
762         m_events.clear();
763 }
764
765 void
766 CaseFile::PurgeTentativeEvents()
767 {
768         for (EventList::iterator event(m_tentativeEvents.begin());
769              event != m_tentativeEvents.end(); event++)
770                 delete *event;
771
772         m_tentativeEvents.clear();
773 }
774
775 void
776 CaseFile::SerializeEvList(const EventList events, int fd,
777                 const char* prefix) const
778 {
779         if (events.empty())
780                 return;
781         for (EventList::const_iterator curEvent = events.begin();
782              curEvent != events.end(); curEvent++) {
783                 const string &eventString((*curEvent)->GetEventString());
784
785                 // TODO: replace many write(2) calls with a single writev(2)
786                 if (prefix)
787                         write(fd, prefix, strlen(prefix));
788                 write(fd, eventString.c_str(), eventString.length());
789         }
790 }
791
792 void
793 CaseFile::Serialize()
794 {
795         stringstream saveFile;
796
797         saveFile << setfill('0')
798                  << s_caseFilePath << "/"
799                  << "pool_" << PoolGUIDString()
800                  << "_vdev_" << VdevGUIDString()
801                  << ".case";
802
803         if (m_events.empty() && m_tentativeEvents.empty()) {
804                 unlink(saveFile.str().c_str());
805                 return;
806         }
807
808         int fd(open(saveFile.str().c_str(), O_CREAT|O_TRUNC|O_WRONLY, 0644));
809         if (fd == -1) {
810                 syslog(LOG_ERR, "CaseFile::Serialize: Unable to open %s.\n",
811                        saveFile.str().c_str());
812                 return;
813         }
814         SerializeEvList(m_events, fd);
815         SerializeEvList(m_tentativeEvents, fd, "tentative ");
816         close(fd);
817 }
818
819 /*
820  * XXX: This method assumes that events may not contain embedded newlines.  If
821  * ever events can contain embedded newlines, then CaseFile must switch
822  * serialization formats
823  */
824 void
825 CaseFile::DeSerialize(ifstream &caseStream)
826 {
827         string        evString;
828         const EventFactory &factory(ZfsDaemon::Get().GetFactory());
829
830         caseStream >> std::noskipws >> std::ws;
831         while (caseStream.good()) {
832                 /*
833                  * Outline:
834                  * read the beginning of a line and check it for
835                  * "tentative".  If found, discard "tentative".
836                  * Create a new event
837                  * continue
838                  */
839                 EventList* destEvents;
840                 const string tentFlag("tentative ");
841                 string line;
842                 std::stringbuf lineBuf;
843
844                 caseStream.get(lineBuf);
845                 caseStream.ignore();  /*discard the newline character*/
846                 line = lineBuf.str();
847                 if (line.compare(0, tentFlag.size(), tentFlag) == 0) {
848                         /* Discard "tentative" */
849                         line.erase(0, tentFlag.size());
850                         destEvents = &m_tentativeEvents;
851                 } else {
852                         destEvents = &m_events;
853                 }
854                 Event *event(Event::CreateEvent(factory, line));
855                 if (event != NULL) {
856                         destEvents->push_back(event);
857                         RegisterCallout(*event);
858                 }
859         }
860 }
861
862 void
863 CaseFile::Close()
864 {
865         /*
866          * This case is no longer relevant.  Clean up our
867          * serialization file, and delete the case.
868          */
869         syslog(LOG_INFO, "CaseFile(%s,%s) closed - State %s\n",
870                PoolGUIDString().c_str(), VdevGUIDString().c_str(),
871                zpool_state_to_name(VdevState(), VDEV_AUX_NONE));
872
873         /*
874          * Serialization of a Case with no event data, clears the
875          * Serialization data for that event.
876          */
877         PurgeEvents();
878         Serialize();
879
880         delete this;
881 }
882
883 void
884 CaseFile::OnGracePeriodEnded()
885 {
886         bool should_fault, should_degrade;
887         ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
888         zpool_handle_t *zhp(zpl.empty() ? NULL : zpl.front());
889
890         m_events.splice(m_events.begin(), m_tentativeEvents);
891         should_fault = ShouldFault();
892         should_degrade = ShouldDegrade();
893
894         if (should_fault || should_degrade) {
895                 if (zhp == NULL
896                  || (VdevIterator(zhp).Find(m_vdevGUID)) == NULL) {
897                         /*
898                          * Either the pool no longer exists
899                          * or this vdev is no longer a member of
900                          * the pool.
901                          */
902                         Close();
903                         return;
904                 }
905
906         }
907
908         /* A fault condition has priority over a degrade condition */
909         if (ShouldFault()) {
910                 /* Fault the vdev and close the case. */
911                 if (zpool_vdev_fault(zhp, (uint64_t)m_vdevGUID,
912                                        VDEV_AUX_ERR_EXCEEDED) == 0) {
913                         syslog(LOG_INFO, "Faulting vdev(%s/%s)",
914                                PoolGUIDString().c_str(),
915                                VdevGUIDString().c_str());
916                         Close();
917                         return;
918                 }
919                 else {
920                         syslog(LOG_ERR, "Fault vdev(%s/%s): %s: %s\n",
921                                PoolGUIDString().c_str(),
922                                VdevGUIDString().c_str(),
923                                libzfs_error_action(g_zfsHandle),
924                                libzfs_error_description(g_zfsHandle));
925                 }
926         }
927         else if (ShouldDegrade()) {
928                 /* Degrade the vdev and close the case. */
929                 if (zpool_vdev_degrade(zhp, (uint64_t)m_vdevGUID,
930                                        VDEV_AUX_ERR_EXCEEDED) == 0) {
931                         syslog(LOG_INFO, "Degrading vdev(%s/%s)",
932                                PoolGUIDString().c_str(),
933                                VdevGUIDString().c_str());
934                         Close();
935                         return;
936                 }
937                 else {
938                         syslog(LOG_ERR, "Degrade vdev(%s/%s): %s: %s\n",
939                                PoolGUIDString().c_str(),
940                                VdevGUIDString().c_str(),
941                                libzfs_error_action(g_zfsHandle),
942                                libzfs_error_description(g_zfsHandle));
943                 }
944         }
945         Serialize();
946 }
947
948 Vdev
949 CaseFile::BeingReplacedBy(zpool_handle_t *zhp) {
950         Vdev vd(zhp, CaseVdev(zhp));
951         std::list<Vdev> children;
952         std::list<Vdev>::iterator children_it;
953
954         Vdev parent(vd.Parent());
955         Vdev replacing(NonexistentVdev);
956
957         /*
958          * To determine whether we are being replaced by another spare that
959          * is still working, then make sure that it is currently spared and
960          * that the spare is either resilvering or healthy.  If any of these
961          * conditions fail, then we are not being replaced by a spare.
962          *
963          * If the spare is healthy, then the case file should be closed very
964          * soon after this check.
965          */
966         if (parent.DoesNotExist()
967          || parent.Name(zhp, /*verbose*/false) != "spare")
968                 return (NonexistentVdev);
969
970         children = parent.Children();
971         children_it = children.begin();
972         for (;children_it != children.end(); children_it++) {
973                 Vdev child = *children_it;
974
975                 /* Skip our vdev. */
976                 if (child.GUID() == VdevGUID())
977                         continue;
978                 /*
979                  * Accept the first child that doesn't match our GUID, or
980                  * any resilvering/healthy device if one exists.
981                  */
982                 if (replacing.DoesNotExist() || child.IsResilvering()
983                  || child.State() == VDEV_STATE_HEALTHY)
984                         replacing = child;
985         }
986
987         return (replacing);
988 }
989
990 bool
991 CaseFile::Replace(const char* vdev_type, const char* path, bool isspare) {
992         nvlist_t *nvroot, *newvd;
993         const char *poolname;
994         string oldstr(VdevGUIDString());
995         bool retval = true;
996
997         /* Figure out what pool we're working on */
998         ZpoolList zpl(ZpoolList::ZpoolByGUID, &m_poolGUID);
999         zpool_handle_t *zhp(zpl.empty() ? NULL : zpl.front());
1000         if (zhp == NULL) {
1001                 syslog(LOG_ERR, "CaseFile::Replace: could not find pool for "
1002                        "pool_guid %"PRIu64".", (uint64_t)m_poolGUID);
1003                 return (false);
1004         }
1005         poolname = zpool_get_name(zhp);
1006         Vdev vd(zhp, CaseVdev(zhp));
1007         Vdev replaced(BeingReplacedBy(zhp));
1008
1009         if (isspare && !vd.IsSpare() && !replaced.DoesNotExist()) {
1010                 /* If we are already being replaced by a working spare, pass. */
1011                 if (replaced.IsResilvering()
1012                  || replaced.State() == VDEV_STATE_HEALTHY) {
1013                         syslog(LOG_INFO, "CaseFile::Replace(%s->%s): already "
1014                             "replaced", VdevGUIDString().c_str(), path);
1015                         return (/*consumed*/false);
1016                 }
1017                 /*
1018                  * If we have already been replaced by a spare, but that spare
1019                  * is broken, we must spare the spare, not the original device.
1020                  */
1021                 oldstr = replaced.GUIDString();
1022                 syslog(LOG_INFO, "CaseFile::Replace(%s->%s): sparing "
1023                     "broken spare %s instead", VdevGUIDString().c_str(),
1024                     path, oldstr.c_str());
1025         }
1026
1027         /*
1028          * Build a root vdev/leaf vdev configuration suitable for
1029          * zpool_vdev_attach. Only enough data for the kernel to find
1030          * the device (i.e. type and disk device node path) are needed.
1031          */
1032         nvroot = NULL;
1033         newvd = NULL;
1034
1035         if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0
1036          || nvlist_alloc(&newvd, NV_UNIQUE_NAME, 0) != 0) {
1037                 syslog(LOG_ERR, "Replace vdev(%s/%s): Unable to allocate "
1038                     "configuration data.", poolname, oldstr.c_str());
1039                 if (nvroot != NULL)
1040                         nvlist_free(nvroot);
1041                 return (false);
1042         }
1043         if (nvlist_add_string(newvd, ZPOOL_CONFIG_TYPE, vdev_type) != 0
1044          || nvlist_add_string(newvd, ZPOOL_CONFIG_PATH, path) != 0
1045          || nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) != 0
1046          || nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1047                                     &newvd, 1) != 0) {
1048                 syslog(LOG_ERR, "Replace vdev(%s/%s): Unable to initialize "
1049                     "configuration data.", poolname, oldstr.c_str());
1050                 nvlist_free(newvd);
1051                 nvlist_free(nvroot);
1052                 return (true);
1053         }
1054
1055         /* Data was copied when added to the root vdev. */
1056         nvlist_free(newvd);
1057
1058         retval = (zpool_vdev_attach(zhp, oldstr.c_str(), path, nvroot,
1059             /*replace*/B_TRUE) == 0);
1060         if (retval)
1061                 syslog(LOG_INFO, "Replacing vdev(%s/%s) with %s\n",
1062                     poolname, oldstr.c_str(), path);
1063         else
1064                 syslog(LOG_ERR, "Replace vdev(%s/%s): %s: %s\n",
1065                     poolname, oldstr.c_str(), libzfs_error_action(g_zfsHandle),
1066                     libzfs_error_description(g_zfsHandle));
1067         nvlist_free(nvroot);
1068
1069         return (retval);
1070 }
1071
1072 /* Does the argument event refer to a checksum error? */
1073 static bool
1074 IsChecksumEvent(const Event* const event)
1075 {
1076         return ("ereport.fs.zfs.checksum" == event->Value("type"));
1077 }
1078
1079 /* Does the argument event refer to an IO error? */
1080 static bool
1081 IsIOEvent(const Event* const event)
1082 {
1083         return ("ereport.fs.zfs.io" == event->Value("type"));
1084 }
1085
1086 bool
1087 CaseFile::ShouldDegrade() const
1088 {
1089         return (std::count_if(m_events.begin(), m_events.end(),
1090                               IsChecksumEvent) > ZFS_DEGRADE_IO_COUNT);
1091 }
1092
1093 bool
1094 CaseFile::ShouldFault() const
1095 {
1096         return (std::count_if(m_events.begin(), m_events.end(),
1097                               IsIOEvent) > ZFS_DEGRADE_IO_COUNT);
1098 }
1099
1100 nvlist_t *
1101 CaseFile::CaseVdev(zpool_handle_t *zhp) const
1102 {
1103         return (VdevIterator(zhp).Find(VdevGUID()));
1104 }