xref: /openbmc/phosphor-logging/extensions/openpower-pels/manager.cpp (revision 085efbb190a87f160901e334dc73be68039ff4fa)
1 /**
2  * Copyright © 2019 IBM Corporation
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *     http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 #include "manager.hpp"
17 
18 #include "additional_data.hpp"
19 #include "elog_serialize.hpp"
20 #include "json_utils.hpp"
21 #include "pel.hpp"
22 #include "pel_entry.hpp"
23 #include "pel_values.hpp"
24 #include "service_indicators.hpp"
25 #include "severity.hpp"
26 
27 #include <sys/inotify.h>
28 #include <unistd.h>
29 
30 #include <phosphor-logging/lg2.hpp>
31 #include <xyz/openbmc_project/Common/error.hpp>
32 #include <xyz/openbmc_project/Logging/Create/server.hpp>
33 
34 #include <filesystem>
35 #include <format>
36 #include <fstream>
37 
38 namespace openpower
39 {
40 namespace pels
41 {
42 
43 using namespace phosphor::logging;
44 namespace fs = std::filesystem;
45 namespace rg = openpower::pels::message;
46 
47 namespace common_error = sdbusplus::xyz::openbmc_project::Common::Error;
48 
49 using Create = sdbusplus::server::xyz::openbmc_project::logging::Create;
50 
51 namespace additional_data
52 {
53 constexpr auto rawPEL = "RAWPEL";
54 constexpr auto esel = "ESEL";
55 constexpr auto error = "ERROR_NAME";
56 } // namespace additional_data
57 
58 constexpr auto defaultLogMessage = "xyz.openbmc_project.Logging.Error.Default";
59 constexpr uint32_t bmcThermalCompID = 0x2700;
60 constexpr uint32_t bmcFansCompID = 0x2800;
61 
~Manager()62 Manager::~Manager()
63 {
64     if (_pelFileDeleteFD != -1)
65     {
66         if (_pelFileDeleteWatchFD != -1)
67         {
68             inotify_rm_watch(_pelFileDeleteFD, _pelFileDeleteWatchFD);
69         }
70         close(_pelFileDeleteFD);
71     }
72 }
73 
create(const std::string & message,uint32_t obmcLogID,uint64_t timestamp,Entry::Level severity,const std::map<std::string,std::string> & additionalData,const std::vector<std::string> & associations,const FFDCEntries & ffdc)74 void Manager::create(const std::string& message, uint32_t obmcLogID,
75                      uint64_t timestamp, Entry::Level severity,
76                      const std::map<std::string, std::string>& additionalData,
77                      const std::vector<std::string>& associations,
78                      const FFDCEntries& ffdc)
79 {
80     AdditionalData ad{additionalData};
81 
82     // If a PEL was passed in via a filename or in an ESEL,
83     // use that.  Otherwise, create one.
84     auto rawPelPath = ad.getValue(additional_data::rawPEL);
85     if (rawPelPath)
86     {
87         addRawPEL(*rawPelPath, obmcLogID);
88     }
89     else
90     {
91         auto esel = ad.getValue(additional_data::esel);
92         if (esel)
93         {
94             addESELPEL(*esel, obmcLogID);
95         }
96         else
97         {
98             createPEL(message, obmcLogID, timestamp, severity, additionalData,
99                       associations, ffdc);
100         }
101     }
102 
103     setEntryPath(obmcLogID);
104     setServiceProviderNotifyFlag(obmcLogID);
105 }
106 
addRawPEL(const std::string & rawPelPath,uint32_t obmcLogID)107 void Manager::addRawPEL(const std::string& rawPelPath, uint32_t obmcLogID)
108 {
109     if (fs::exists(rawPelPath))
110     {
111         std::ifstream file(rawPelPath, std::ios::in | std::ios::binary);
112 
113         auto data = std::vector<uint8_t>(std::istreambuf_iterator<char>(file),
114                                          std::istreambuf_iterator<char>());
115         if (file.fail())
116         {
117             lg2::error(
118                 "Filesystem error reading a raw PEL. File = {FILE}, obmcLogID = {LOGID}",
119                 "FILE", rawPelPath, "LOGID", obmcLogID);
120             // TODO, Decide what to do here. Maybe nothing.
121             return;
122         }
123 
124         file.close();
125 
126         addPEL(data, obmcLogID);
127 
128         std::error_code ec;
129         fs::remove(rawPelPath, ec);
130     }
131     else
132     {
133         lg2::error(
134             "Raw PEL file from BMC event log does not exit. File = {FILE}, obmcLogID = {LOGID}",
135             "FILE", rawPelPath, "LOGID", obmcLogID);
136     }
137 }
138 
addPEL(std::vector<uint8_t> & pelData,uint32_t obmcLogID)139 void Manager::addPEL(std::vector<uint8_t>& pelData, uint32_t obmcLogID)
140 {
141     auto pel = std::make_unique<openpower::pels::PEL>(pelData, obmcLogID);
142     if (pel->valid())
143     {
144         // PELs created by others still need this field set by us.
145         pel->setCommitTime();
146 
147         // Assign Id other than to Hostbot PEL
148         if ((pel->privateHeader()).creatorID() !=
149             static_cast<uint8_t>(CreatorID::hostboot))
150         {
151             pel->assignID();
152         }
153         else
154         {
155             const Repository::LogID id{Repository::LogID::Pel(pel->id())};
156             auto result = _repo.hasPEL(id);
157             if (result)
158             {
159                 lg2::warning(
160                     "Duplicate HostBoot PEL ID {ID} found, moving it to archive folder",
161                     "ID", lg2::hex, pel->id());
162 
163                 _repo.archivePEL(*pel);
164 
165                 // No need to keep around the openBMC event log entry
166                 scheduleObmcLogDelete(obmcLogID);
167                 return;
168             }
169         }
170 
171         // Update System Info to Extended User Data
172         pel->updateSysInfoInExtendedUserDataSection(*_dataIface);
173 
174         // Check for severity 0x51 and update boot progress SRC
175         updateProgressSRC(pel);
176 
177         try
178         {
179             lg2::debug("Adding external PEL {ID} (BMC ID {BMCID}) to repo",
180                        "ID", lg2::hex, pel->id(), "BMCID", obmcLogID);
181             _repo.add(pel);
182 
183             if (_repo.sizeWarning())
184             {
185                 scheduleRepoPrune();
186             }
187 
188             // Activate any resulting service indicators if necessary
189             auto policy = service_indicators::getPolicy(*_dataIface);
190             policy->activate(*pel);
191         }
192         catch (const std::exception& e)
193         {
194             // Probably a full or r/o filesystem, not much we can do.
195             lg2::error("Unable to add PEL {ID} to Repository", "ID", lg2::hex,
196                        pel->id());
197         }
198 
199         updateEventId(pel);
200         updateResolution(*pel);
201         serializeLogEntry(obmcLogID);
202         createPELEntry(obmcLogID);
203 
204         // Check if firmware should quiesce system due to error
205         checkPelAndQuiesce(pel);
206     }
207     else
208     {
209         lg2::error("Invalid PEL received from the host. BMC ID = {ID}", "ID",
210                    obmcLogID);
211 
212         AdditionalData ad;
213         ad.add("PLID", getNumberString("0x%08X", pel->plid()));
214         ad.add("OBMC_LOG_ID", std::to_string(obmcLogID));
215         ad.add("PEL_SIZE", std::to_string(pelData.size()));
216 
217         std::string asciiString;
218         auto src = pel->primarySRC();
219         if (src)
220         {
221             asciiString = (*src)->asciiString();
222         }
223 
224         ad.add("SRC", asciiString);
225 
226         _eventLogger.log("org.open_power.Logging.Error.BadHostPEL",
227                          Entry::Level::Error, ad);
228 
229         // Save it to a file for debug in the lab.  Just keep the latest.
230         // Not adding it to the PEL because it could already be max size
231         // and don't want to truncate an already invalid PEL.
232         std::ofstream pelFile{getPELRepoPath() / "badPEL"};
233         pelFile.write(reinterpret_cast<const char*>(pelData.data()),
234                       pelData.size());
235 
236         // No need to keep around the openBMC event log entry
237         scheduleObmcLogDelete(obmcLogID);
238     }
239 }
240 
addESELPEL(const std::string & esel,uint32_t obmcLogID)241 void Manager::addESELPEL(const std::string& esel, uint32_t obmcLogID)
242 {
243     std::vector<uint8_t> data;
244 
245     lg2::debug("Adding PEL from ESEL. BMC ID = {ID}", "ID", obmcLogID);
246 
247     try
248     {
249         data = eselToRawData(esel);
250     }
251     catch (const std::exception& e)
252     {
253         // Try to add it below anyway, so it follows the usual bad data path.
254         lg2::error("Problems converting ESEL string to a byte vector");
255     }
256 
257     addPEL(data, obmcLogID);
258 }
259 
eselToRawData(const std::string & esel)260 std::vector<uint8_t> Manager::eselToRawData(const std::string& esel)
261 {
262     std::vector<uint8_t> data;
263     std::string byteString;
264 
265     // As the eSEL string looks like: "50 48 00 ab ..." there are 3
266     // characters per raw byte, and since the actual PEL data starts
267     // at the 16th byte, the code will grab the PEL data starting at
268     // offset 48 in the string.
269     static constexpr size_t pelStart = 16 * 3;
270 
271     if (esel.size() <= pelStart)
272     {
273         lg2::error("ESEL data too short, length = {LEN}", "LEN", esel.size());
274         throw std::length_error("ESEL data too short");
275     }
276 
277     for (size_t i = pelStart; i < esel.size(); i += 3)
278     {
279         if (i + 1 < esel.size())
280         {
281             byteString = esel.substr(i, 2);
282             data.push_back(std::stoi(byteString, nullptr, 16));
283         }
284         else
285         {
286             lg2::error("ESEL data too short, length = {LEN}", "LEN",
287                        esel.size());
288             throw std::length_error("ESEL data too short");
289         }
290     }
291 
292     return data;
293 }
294 
erase(uint32_t obmcLogID)295 void Manager::erase(uint32_t obmcLogID)
296 {
297     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
298 
299     auto path = std::string(OBJ_ENTRY) + '/' + std::to_string(obmcLogID);
300     _pelEntries.erase(path);
301     _repo.remove(id);
302 }
303 
getLogIDWithHwIsolation(std::vector<uint32_t> & idsWithHwIsoEntry)304 void Manager::getLogIDWithHwIsolation(std::vector<uint32_t>& idsWithHwIsoEntry)
305 {
306     idsWithHwIsoEntry = _dataIface->getLogIDWithHwIsolation();
307 }
308 
isDeleteProhibited(uint32_t obmcLogID)309 bool Manager::isDeleteProhibited(uint32_t obmcLogID)
310 {
311     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
312     if (!_repo.hasPEL(id))
313     {
314         return false;
315     }
316     auto entryPath{std::string(OBJ_ENTRY) + '/' + std::to_string(obmcLogID)};
317     auto entry = _pelEntries.find(entryPath);
318     if (entry != _pelEntries.end())
319     {
320         if (entry->second->guard())
321         {
322             auto hwIsolationAssocPaths = _dataIface->getAssociatedPaths(
323                 entryPath += "/isolated_hw_entry", "/", 0,
324                 {"xyz.openbmc_project.HardwareIsolation.Entry"});
325             if (!hwIsolationAssocPaths.empty())
326             {
327                 return true;
328             }
329         }
330     }
331     return false;
332 }
333 
convertToPelFFDC(const FFDCEntries & ffdc)334 PelFFDC Manager::convertToPelFFDC(const FFDCEntries& ffdc)
335 {
336     PelFFDC pelFFDC;
337 
338     std::for_each(ffdc.begin(), ffdc.end(), [&pelFFDC](const auto& f) {
339         PelFFDCfile pf;
340         pf.subType = std::get<ffdcSubtypePos>(f);
341         pf.version = std::get<ffdcVersionPos>(f);
342         pf.fd = std::get<ffdcFDPos>(f);
343 
344         switch (std::get<ffdcFormatPos>(f))
345         {
346             case Create::FFDCFormat::JSON:
347                 pf.format = UserDataFormat::json;
348                 break;
349             case Create::FFDCFormat::CBOR:
350                 pf.format = UserDataFormat::cbor;
351                 break;
352             case Create::FFDCFormat::Text:
353                 pf.format = UserDataFormat::text;
354                 break;
355             case Create::FFDCFormat::Custom:
356                 pf.format = UserDataFormat::custom;
357                 break;
358         }
359 
360         pelFFDC.push_back(pf);
361     });
362 
363     return pelFFDC;
364 }
365 
createPEL(const std::string & message,uint32_t obmcLogID,uint64_t timestamp,phosphor::logging::Entry::Level severity,const std::map<std::string,std::string> & additionalData,const std::vector<std::string> &,const FFDCEntries & ffdc)366 void Manager::createPEL(
367     const std::string& message, uint32_t obmcLogID, uint64_t timestamp,
368     phosphor::logging::Entry::Level severity,
369     const std::map<std::string, std::string>& additionalData,
370     const std::vector<std::string>& /*associations*/, const FFDCEntries& ffdc)
371 {
372     auto start = std::chrono::steady_clock::now();
373     auto entry = _registry.lookup(message, rg::LookupType::name);
374     auto pelFFDC = convertToPelFFDC(ffdc);
375     AdditionalData ad{additionalData};
376     std::string msg;
377 
378     if (!entry)
379     {
380         // Instead, get the default entry that means there is no
381         // other matching entry.  This error will still use the
382         // AdditionalData values of the original error, and this
383         // code will add the error message value that wasn't found
384         // to this AD.  This way, there will at least be a PEL,
385         // possibly with callouts, to allow users to debug the
386         // issue that caused the error even without its own PEL.
387         lg2::error("Event not found in PEL message registry: {MSG}", "MSG",
388                    message);
389 
390         entry = _registry.lookup(defaultLogMessage, rg::LookupType::name);
391         if (!entry)
392         {
393             lg2::error("Default event not found in PEL message registry");
394             return;
395         }
396 
397         ad.add(additional_data::error, message);
398     }
399 
400     auto pel = std::make_unique<openpower::pels::PEL>(
401         *entry, obmcLogID, timestamp, severity, ad, pelFFDC, *_dataIface,
402         *_journal);
403 
404     _repo.add(pel);
405 
406     if (_repo.sizeWarning())
407     {
408         scheduleRepoPrune();
409     }
410 
411     // Check for severity 0x51 and update boot progress SRC
412     updateProgressSRC(pel);
413 
414     // Activate any resulting service indicators if necessary
415     auto policy = service_indicators::getPolicy(*_dataIface);
416     policy->activate(*pel);
417 
418     updateDBusSeverity(*pel);
419     updateEventId(pel);
420     updateResolution(*pel);
421     serializeLogEntry(obmcLogID);
422     createPELEntry(obmcLogID);
423 
424     auto src = pel->primarySRC();
425     if (src)
426     {
427         auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(
428             std::chrono::steady_clock::now() - start);
429 
430         auto asciiString = (*src)->asciiString();
431         while (asciiString.back() == ' ')
432         {
433             asciiString.pop_back();
434         }
435         lg2::info("Created PEL {ID} (BMC ID {BMCID}) with SRC {SRC}", "ID",
436                   lg2::hex, pel->id(), "BMCID", pel->obmcLogID(), "SRC",
437                   asciiString, "PEL_CREATE_DURATION", duration.count());
438     }
439 
440     // Check if firmware should quiesce system due to error
441     checkPelAndQuiesce(pel);
442 }
443 
getPEL(uint32_t pelID)444 sdbusplus::message::unix_fd Manager::getPEL(uint32_t pelID)
445 {
446     Repository::LogID id{Repository::LogID::Pel(pelID)};
447     std::optional<int> fd;
448 
449     lg2::debug("getPEL {ID}", "ID", lg2::hex, pelID);
450 
451     try
452     {
453         fd = _repo.getPELFD(id);
454     }
455     catch (const std::exception& e)
456     {
457         throw common_error::InternalFailure();
458     }
459 
460     if (!fd)
461     {
462         throw common_error::InvalidArgument();
463     }
464 
465     scheduleFDClose(*fd);
466 
467     return *fd;
468 }
469 
scheduleFDClose(int fd)470 void Manager::scheduleFDClose(int fd)
471 {
472     _fdCloserEventSource = std::make_unique<sdeventplus::source::Defer>(
473         _event, std::bind(std::mem_fn(&Manager::closeFD), this, fd,
474                           std::placeholders::_1));
475 }
476 
closeFD(int fd,sdeventplus::source::EventBase &)477 void Manager::closeFD(int fd, sdeventplus::source::EventBase& /*source*/)
478 {
479     close(fd);
480     _fdCloserEventSource.reset();
481 }
482 
getPELFromOBMCID(uint32_t obmcLogID)483 std::vector<uint8_t> Manager::getPELFromOBMCID(uint32_t obmcLogID)
484 {
485     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
486     std::optional<std::vector<uint8_t>> data;
487 
488     lg2::debug("getPELFromOBMCID  {BMCID}", "BMCID", obmcLogID);
489 
490     try
491     {
492         data = _repo.getPELData(id);
493     }
494     catch (const std::exception& e)
495     {
496         throw common_error::InternalFailure();
497     }
498 
499     if (!data)
500     {
501         throw common_error::InvalidArgument();
502     }
503 
504     return *data;
505 }
506 
hostAck(uint32_t pelID)507 void Manager::hostAck(uint32_t pelID)
508 {
509     Repository::LogID id{Repository::LogID::Pel(pelID)};
510 
511     lg2::debug("HostHack {ID}", "ID", lg2::hex, pelID);
512 
513     if (!_repo.hasPEL(id))
514     {
515         throw common_error::InvalidArgument();
516     }
517 
518     if (_hostNotifier)
519     {
520         _hostNotifier->ackPEL(pelID);
521     }
522 }
523 
hostReject(uint32_t pelID,RejectionReason reason)524 void Manager::hostReject(uint32_t pelID, RejectionReason reason)
525 {
526     Repository::LogID id{Repository::LogID::Pel(pelID)};
527 
528     lg2::debug("HostReject {ID}, reason = {REASON}", "ID", lg2::hex, pelID,
529                "REASON", reason);
530 
531     if (!_repo.hasPEL(id))
532     {
533         throw common_error::InvalidArgument();
534     }
535 
536     if (reason == RejectionReason::BadPEL)
537     {
538         AdditionalData data;
539         data.add("BAD_ID", getNumberString("0x%08X", pelID));
540         _eventLogger.log("org.open_power.Logging.Error.SentBadPELToHost",
541                          Entry::Level::Informational, data);
542         if (_hostNotifier)
543         {
544             _hostNotifier->setBadPEL(pelID);
545         }
546     }
547     else if ((reason == RejectionReason::HostFull) && _hostNotifier)
548     {
549         _hostNotifier->setHostFull(pelID);
550     }
551 }
552 
scheduleRepoPrune()553 void Manager::scheduleRepoPrune()
554 {
555     _repoPrunerEventSource = std::make_unique<sdeventplus::source::Defer>(
556         _event, std::bind(std::mem_fn(&Manager::pruneRepo), this,
557                           std::placeholders::_1));
558 }
559 
pruneRepo(sdeventplus::source::EventBase &)560 void Manager::pruneRepo(sdeventplus::source::EventBase& /*source*/)
561 {
562     auto idsWithHwIsoEntry = _dataIface->getLogIDWithHwIsolation();
563 
564     auto idsToDelete = _repo.prune(idsWithHwIsoEntry);
565 
566     // Remove the OpenBMC event logs for the PELs that were just removed.
567     std::for_each(idsToDelete.begin(), idsToDelete.end(),
568                   [this](auto id) { this->_logManager.erase(id); });
569 
570     _repoPrunerEventSource.reset();
571 }
572 
setupPELDeleteWatch()573 void Manager::setupPELDeleteWatch()
574 {
575     _pelFileDeleteFD = inotify_init1(IN_NONBLOCK);
576     if (-1 == _pelFileDeleteFD)
577     {
578         auto e = errno;
579         lg2::error("inotify_init1 failed with errno {ERRNO}", "ERRNO", e);
580         abort();
581     }
582 
583     _pelFileDeleteWatchFD = inotify_add_watch(
584         _pelFileDeleteFD, _repo.repoPath().c_str(), IN_DELETE);
585     if (-1 == _pelFileDeleteWatchFD)
586     {
587         auto e = errno;
588         lg2::error("inotify_add_watch failed with errno {ERRNO}", "ERRNO", e);
589         abort();
590     }
591 
592     _pelFileDeleteEventSource = std::make_unique<sdeventplus::source::IO>(
593         _event, _pelFileDeleteFD, EPOLLIN,
594         std::bind(std::mem_fn(&Manager::pelFileDeleted), this,
595                   std::placeholders::_1, std::placeholders::_2,
596                   std::placeholders::_3));
597 }
598 
pelFileDeleted(sdeventplus::source::IO &,int,uint32_t revents)599 void Manager::pelFileDeleted(sdeventplus::source::IO& /*io*/, int /*fd*/,
600                              uint32_t revents)
601 {
602     if (!(revents & EPOLLIN))
603     {
604         return;
605     }
606 
607     // An event for 1 PEL uses 48B. When all PELs are deleted at once,
608     // as many events as there is room for can be handled in one callback.
609     // A size of 2000 will allow 41 to be processed, with additional
610     // callbacks being needed to process the remaining ones.
611     std::array<uint8_t, 2000> data{};
612     auto bytesRead = read(_pelFileDeleteFD, data.data(), data.size());
613     if (bytesRead < 0)
614     {
615         auto e = errno;
616         lg2::error("Failed reading data from inotify event, errno = {ERRNO}",
617                    "ERRNO", e);
618         abort();
619     }
620 
621     auto offset = 0;
622     while (offset < bytesRead)
623     {
624         auto event = reinterpret_cast<inotify_event*>(&data[offset]);
625         if (event->mask & IN_DELETE)
626         {
627             std::string filename{event->name};
628 
629             // Get the PEL ID from the filename and tell the
630             // repo it's been removed, and then delete the BMC
631             // event log if it's there.
632             auto pos = filename.find_first_of('_');
633             if (pos != std::string::npos)
634             {
635                 try
636                 {
637                     auto idString = filename.substr(pos + 1);
638                     auto pelID = std::stoul(idString, nullptr, 16);
639 
640                     Repository::LogID id{Repository::LogID::Pel(pelID)};
641                     auto removedLogID = _repo.remove(id);
642                     if (removedLogID)
643                     {
644                         _logManager.erase(removedLogID->obmcID.id);
645                     }
646                 }
647                 catch (const std::exception& e)
648                 {
649                     lg2::info("Could not find PEL ID from its filename {NAME}",
650                               "NAME", filename);
651                 }
652             }
653         }
654 
655         offset += offsetof(inotify_event, name) + event->len;
656     }
657 }
658 
createPELWithFFDCFiles(std::string message,Entry::Level severity,std::map<std::string,std::string> additionalData,std::vector<std::tuple<sdbusplus::server::xyz::openbmc_project::logging::Create::FFDCFormat,uint8_t,uint8_t,sdbusplus::message::unix_fd>> fFDC)659 std::tuple<uint32_t, uint32_t> Manager::createPELWithFFDCFiles(
660     std::string message, Entry::Level severity,
661     std::map<std::string, std::string> additionalData,
662     std::vector<std::tuple<
663         sdbusplus::server::xyz::openbmc_project::logging::Create::FFDCFormat,
664         uint8_t, uint8_t, sdbusplus::message::unix_fd>>
665         fFDC)
666 {
667     _logManager.create(message, severity, additionalData, fFDC);
668 
669     return {_logManager.lastEntryID(), _repo.lastPelID()};
670 }
671 
getPELJSON(uint32_t obmcLogID)672 std::string Manager::getPELJSON(uint32_t obmcLogID)
673 {
674     // Throws InvalidArgument if not found
675     auto pelID = getPELIdFromBMCLogId(obmcLogID);
676 
677     auto cmd = std::format("/usr/bin/peltool -i {:#x}", pelID);
678 
679     FILE* pipe = popen(cmd.c_str(), "r");
680     if (!pipe)
681     {
682         lg2::error("Error running cmd: {CMD}", "CMD", cmd);
683         throw common_error::InternalFailure();
684     }
685 
686     std::string output;
687     std::array<char, 1024> buffer;
688     while (fgets(buffer.data(), buffer.size(), pipe) != nullptr)
689     {
690         output.append(buffer.data());
691     }
692 
693     int rc = pclose(pipe);
694     if (WEXITSTATUS(rc) != 0)
695     {
696         lg2::error("Error running cmd: {CMD}, rc = {RC}", "CMD", cmd, "RC", rc);
697         throw common_error::InternalFailure();
698     }
699 
700     return output;
701 }
702 
checkPelAndQuiesce(std::unique_ptr<openpower::pels::PEL> & pel)703 void Manager::checkPelAndQuiesce(std::unique_ptr<openpower::pels::PEL>& pel)
704 {
705     if ((pel->userHeader().severity() ==
706          static_cast<uint8_t>(SeverityType::nonError)) ||
707         (pel->userHeader().severity() ==
708          static_cast<uint8_t>(SeverityType::recovered)))
709     {
710         lg2::debug(
711             "PEL severity informational or recovered. no quiesce needed");
712         return;
713     }
714     if (!_logManager.isQuiesceOnErrorEnabled())
715     {
716         lg2::debug("QuiesceOnHwError not enabled, no quiesce needed");
717         return;
718     }
719 
720     CreatorID creatorID{pel->privateHeader().creatorID()};
721 
722     if ((creatorID != CreatorID::openBMC) &&
723         (creatorID != CreatorID::hostboot) &&
724         (creatorID != CreatorID::ioDrawer) && (creatorID != CreatorID::occ) &&
725         (creatorID != CreatorID::phyp))
726     {
727         return;
728     }
729 
730     // Now check if it has any type of callout
731     if (pel->isHwCalloutPresent())
732     {
733         lg2::info(
734             "QuiesceOnHwError enabled, PEL severity not nonError or recovered, "
735             "and callout is present");
736 
737         _logManager.quiesceOnError(pel->obmcLogID());
738     }
739 }
740 
getEventId(const openpower::pels::PEL & pel) const741 std::string Manager::getEventId(const openpower::pels::PEL& pel) const
742 {
743     std::string str;
744     auto src = pel.primarySRC();
745     if (src)
746     {
747         const auto& hexwords = (*src)->hexwordData();
748 
749         std::string refcode = (*src)->asciiString();
750         size_t pos = refcode.find_last_not_of(0x20);
751         if (pos != std::string::npos)
752         {
753             refcode.erase(pos + 1);
754         }
755         str = refcode;
756 
757         for (auto& value : hexwords)
758         {
759             str += " ";
760             str += getNumberString("%08X", value);
761         }
762     }
763     return sanitizeFieldForDBus(str);
764 }
765 
updateEventId(std::unique_ptr<openpower::pels::PEL> & pel)766 void Manager::updateEventId(std::unique_ptr<openpower::pels::PEL>& pel)
767 {
768     std::string eventIdStr = getEventId(*pel);
769 
770     auto entryN = _logManager.entries.find(pel->obmcLogID());
771     if (entryN != _logManager.entries.end())
772     {
773         entryN->second->eventId(eventIdStr, true);
774     }
775 }
776 
sanitizeFieldForDBus(std::string field)777 std::string Manager::sanitizeFieldForDBus(std::string field)
778 {
779     std::for_each(field.begin(), field.end(), [](char& ch) {
780         if (((ch < ' ') || (ch > '~')) && (ch != '\n') && (ch != '\t'))
781         {
782             ch = ' ';
783         }
784     });
785     return field;
786 }
787 
getResolution(const openpower::pels::PEL & pel) const788 std::string Manager::getResolution(const openpower::pels::PEL& pel) const
789 {
790     std::string str;
791     std::string resolution;
792     auto src = pel.primarySRC();
793     if (src)
794     {
795         // First extract the callout pointer and then go through
796         const auto& callouts = (*src)->callouts();
797         namespace pv = openpower::pels::pel_values;
798         // All PELs dont have callout, check before parsing callout data
799         if (callouts)
800         {
801             const auto& entries = callouts->callouts();
802             // Entry starts with index 1
803             uint8_t index = 1;
804             for (auto& entry : entries)
805             {
806                 resolution += std::to_string(index) + ". ";
807                 // Adding Location code to resolution
808                 if (!entry->locationCode().empty())
809                     resolution += "Location Code: " + entry->locationCode() +
810                                   ", ";
811                 if (entry->fruIdentity())
812                 {
813                     // Get priority and set the resolution string
814                     str = pv::getValue(entry->priority(),
815                                        pel_values::calloutPriorityValues,
816                                        pel_values::registryNamePos);
817                     str[0] = toupper(str[0]);
818                     resolution += "Priority: " + str + ", ";
819                     if (entry->fruIdentity()->getPN().has_value())
820                     {
821                         resolution +=
822                             "PN: " + entry->fruIdentity()->getPN().value() +
823                             ", ";
824                     }
825                     if (entry->fruIdentity()->getSN().has_value())
826                     {
827                         resolution +=
828                             "SN: " + entry->fruIdentity()->getSN().value() +
829                             ", ";
830                     }
831                     if (entry->fruIdentity()->getCCIN().has_value())
832                     {
833                         resolution +=
834                             "CCIN: " + entry->fruIdentity()->getCCIN().value() +
835                             ", ";
836                     }
837                     // Add the maintenance procedure
838                     if (entry->fruIdentity()->getMaintProc().has_value())
839                     {
840                         resolution +=
841                             "Procedure: " +
842                             entry->fruIdentity()->getMaintProc().value() + ", ";
843                     }
844                 }
845                 resolution.resize(resolution.size() - 2);
846                 resolution += "\n";
847                 index++;
848             }
849         }
850     }
851     return sanitizeFieldForDBus(resolution);
852 }
853 
updateResolution(const openpower::pels::PEL & pel)854 bool Manager::updateResolution(const openpower::pels::PEL& pel)
855 {
856     std::string callouts = getResolution(pel);
857     auto entryN = _logManager.entries.find(pel.obmcLogID());
858     if (entryN != _logManager.entries.end())
859     {
860         entryN->second->resolution(callouts, true);
861     }
862 
863     return false;
864 }
865 
serializeLogEntry(uint32_t obmcLogID)866 void Manager::serializeLogEntry(uint32_t obmcLogID)
867 {
868     auto entryN = _logManager.entries.find(obmcLogID);
869     if (entryN != _logManager.entries.end())
870     {
871         serialize(*entryN->second);
872     }
873 }
874 
updateDBusSeverity(const openpower::pels::PEL & pel)875 void Manager::updateDBusSeverity(const openpower::pels::PEL& pel)
876 {
877     // The final severity of the PEL may not agree with the
878     // original severity of the D-Bus event log.  Update the
879     // D-Bus property to match in some cases.  This is to
880     // ensure there isn't a Critical or Warning Redfish event
881     // log for an informational or recovered PEL (or vice versa).
882     // This doesn't make an explicit call to serialize the new
883     // event log property value because updateEventId() is called
884     // right after this and will do it.
885     auto sevType =
886         static_cast<SeverityType>(pel.userHeader().severity() & 0xF0);
887 
888     auto entryN = _logManager.entries.find(pel.obmcLogID());
889     if (entryN != _logManager.entries.end())
890     {
891         auto newSeverity =
892             fixupLogSeverity(entryN->second->severity(), sevType);
893         if (newSeverity)
894         {
895             lg2::info("Changing event log {ID} severity from {OLD} "
896                       "to {NEW} to match PEL",
897                       "ID", lg2::hex, entryN->second->id(), "OLD",
898                       Entry::convertLevelToString(entryN->second->severity()),
899                       "NEW", Entry::convertLevelToString(*newSeverity));
900 
901             entryN->second->severity(*newSeverity, true);
902         }
903     }
904 }
905 
setEntryPath(uint32_t obmcLogID)906 void Manager::setEntryPath(uint32_t obmcLogID)
907 {
908     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
909     if (auto attributes = _repo.getPELAttributes(id); attributes)
910     {
911         auto& attr = attributes.value().get();
912         auto entry = _logManager.entries.find(obmcLogID);
913         if (entry != _logManager.entries.end())
914         {
915             entry->second->path(attr.path, true);
916         }
917     }
918 }
919 
setServiceProviderNotifyFlag(uint32_t obmcLogID)920 void Manager::setServiceProviderNotifyFlag(uint32_t obmcLogID)
921 {
922     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
923     if (auto attributes = _repo.getPELAttributes(id); attributes)
924     {
925         auto& attr = attributes.value().get();
926         auto entry = _logManager.entries.find(obmcLogID);
927         if (entry != _logManager.entries.end())
928         {
929             if (attr.actionFlags.test(callHomeFlagBit))
930             {
931                 entry->second->serviceProviderNotify(Entry::Notify::Notify,
932                                                      true);
933             }
934             else
935             {
936                 entry->second->serviceProviderNotify(Entry::Notify::Inhibit,
937                                                      true);
938             }
939         }
940     }
941 }
942 
createPELEntry(uint32_t obmcLogID,bool skipIaSignal)943 void Manager::createPELEntry(uint32_t obmcLogID, bool skipIaSignal)
944 {
945     std::map<std::string, PropertiesVariant> varData;
946     Repository::LogID id{Repository::LogID::Obmc(obmcLogID)};
947     if (auto attributes = _repo.getPELAttributes(id); attributes)
948     {
949         namespace pv = openpower::pels::pel_values;
950         auto& attr = attributes.value().get();
951 
952         // get the hidden flag values
953         auto sevType = static_cast<SeverityType>(attr.severity & 0xF0);
954         auto isHidden = true;
955         if (((sevType != SeverityType::nonError) &&
956              attr.actionFlags.test(reportFlagBit) &&
957              !attr.actionFlags.test(hiddenFlagBit)) ||
958             ((sevType == SeverityType::nonError) &&
959              attr.actionFlags.test(serviceActionFlagBit)))
960         {
961             isHidden = false;
962         }
963         varData.emplace(std::string("Hidden"), isHidden);
964         varData.emplace(
965             std::string("Subsystem"),
966             pv::getValue(attr.subsystem, pel_values::subsystemValues));
967 
968         varData.emplace(
969             std::string("ManagementSystemAck"),
970             (attr.hmcState == TransmissionState::acked ? true : false));
971 
972         varData.emplace("PlatformLogID", attr.plid);
973         varData.emplace("Deconfig", attr.deconfig);
974         varData.emplace("Guard", attr.guard);
975         varData.emplace("Timestamp", attr.creationTime);
976 
977         // Path to create PELEntry Interface is same as PEL
978         auto path = std::string(OBJ_ENTRY) + '/' + std::to_string(obmcLogID);
979         // Create Interface for PELEntry and set properties
980         auto pelEntry = std::make_unique<PELEntry>(_logManager.getBus(), path,
981                                                    varData, obmcLogID, &_repo);
982         if (!skipIaSignal)
983         {
984             pelEntry->emit_added();
985         }
986         _pelEntries.emplace(std::move(path), std::move(pelEntry));
987     }
988 }
989 
getPELIdFromBMCLogId(uint32_t bmcLogId)990 uint32_t Manager::getPELIdFromBMCLogId(uint32_t bmcLogId)
991 {
992     Repository::LogID id{Repository::LogID::Obmc(bmcLogId)};
993     if (auto logId = _repo.getLogID(id); !logId.has_value())
994     {
995         throw common_error::InvalidArgument();
996     }
997     else
998     {
999         return logId->pelID.id;
1000     }
1001 }
1002 
getBMCLogIdFromPELId(uint32_t pelId)1003 uint32_t Manager::getBMCLogIdFromPELId(uint32_t pelId)
1004 {
1005     Repository::LogID id{Repository::LogID::Pel(pelId)};
1006     if (auto logId = _repo.getLogID(id); !logId.has_value())
1007     {
1008         throw common_error::InvalidArgument();
1009     }
1010     else
1011     {
1012         return logId->obmcID.id;
1013     }
1014 }
1015 
updateProgressSRC(std::unique_ptr<openpower::pels::PEL> & pel) const1016 void Manager::updateProgressSRC(
1017     std::unique_ptr<openpower::pels::PEL>& pel) const
1018 {
1019     const size_t refcodeBegin = 40;
1020     const size_t refcodeSize = 8;
1021 
1022     // Check for pel severity of type - 0x51 = critical error, system
1023     // termination
1024     if (pel->userHeader().severity() == 0x51)
1025     {
1026         auto src = pel->primarySRC();
1027         if (src)
1028         {
1029             std::vector<uint8_t> asciiSRC = (*src)->getSrcStruct();
1030 
1031             if (asciiSRC.size() < (refcodeBegin + refcodeSize))
1032             {
1033                 lg2::error(
1034                     "SRC struct is too short to get progress code ({SIZE})",
1035                     "SIZE", asciiSRC.size());
1036                 return;
1037             }
1038 
1039             // Pull the ASCII SRC from offset [40-47] e.g. BD8D1001
1040             std::vector<uint8_t> srcRefCode(
1041                 asciiSRC.begin() + refcodeBegin,
1042                 asciiSRC.begin() + refcodeBegin + refcodeSize);
1043 
1044             try
1045             {
1046                 _dataIface->createProgressSRC(srcRefCode, asciiSRC);
1047             }
1048             catch (const std::exception&)
1049             {
1050                 // Exception - may be no boot progress interface on dbus
1051             }
1052         }
1053     }
1054 }
1055 
scheduleObmcLogDelete(uint32_t obmcLogID)1056 void Manager::scheduleObmcLogDelete(uint32_t obmcLogID)
1057 {
1058     _obmcLogDeleteEventSource = std::make_unique<sdeventplus::source::Defer>(
1059         _event, std::bind(std::mem_fn(&Manager::deleteObmcLog), this,
1060                           std::placeholders::_1, obmcLogID));
1061 }
1062 
deleteObmcLog(sdeventplus::source::EventBase &,uint32_t obmcLogID)1063 void Manager::deleteObmcLog(sdeventplus::source::EventBase&, uint32_t obmcLogID)
1064 {
1065     lg2::info("Removing event log with no PEL: {BMCID}", "BMCID", obmcLogID);
1066     _logManager.erase(obmcLogID);
1067     _obmcLogDeleteEventSource.reset();
1068 }
1069 
clearPowerThermalDeconfigFlag(const std::string & locationCode,openpower::pels::PEL & pel)1070 bool Manager::clearPowerThermalDeconfigFlag(const std::string& locationCode,
1071                                             openpower::pels::PEL& pel)
1072 {
1073     // The requirements state that only power-thermal or
1074     // fan PELs need their deconfig flag cleared.
1075     static const std::vector<uint32_t> compIDs{bmcThermalCompID, bmcFansCompID};
1076 
1077     if (std::find(compIDs.begin(), compIDs.end(),
1078                   pel.privateHeader().header().componentID) == compIDs.end())
1079     {
1080         return false;
1081     }
1082 
1083     auto src = pel.primarySRC();
1084     const auto& callouts = (*src)->callouts();
1085     if (!callouts)
1086     {
1087         return false;
1088     }
1089 
1090     for (const auto& callout : callouts->callouts())
1091     {
1092         // Look for the passed in location code in a callout that
1093         // is either a normal HW callout or a symbolic FRU with
1094         // a trusted location code callout.
1095         if ((callout->locationCode() != locationCode) ||
1096             !callout->fruIdentity())
1097         {
1098             continue;
1099         }
1100 
1101         if ((callout->fruIdentity()->failingComponentType() !=
1102              src::FRUIdentity::hardwareFRU) &&
1103             (callout->fruIdentity()->failingComponentType() !=
1104              src::FRUIdentity::symbolicFRUTrustedLocCode))
1105         {
1106             continue;
1107         }
1108 
1109         lg2::info(
1110             "Clearing deconfig flag in PEL {ID} with SRC {SRC} because {LOC} was replaced",
1111             "ID", lg2::hex, pel.id(), "SRC", (*src)->asciiString().substr(0, 8),
1112             "LOC", locationCode);
1113         (*src)->clearErrorStatusFlag(SRC::ErrorStatusFlags::deconfigured);
1114         return true;
1115     }
1116     return false;
1117 }
1118 
hardwarePresent(const std::string & locationCode)1119 void Manager::hardwarePresent(const std::string& locationCode)
1120 {
1121     Repository::PELUpdateFunc handlePowerThermalHardwarePresent =
1122         [locationCode](openpower::pels::PEL& pel) {
1123             return Manager::clearPowerThermalDeconfigFlag(locationCode, pel);
1124         };
1125 
1126     // If the PEL was created by the BMC and has the deconfig flag set,
1127     // it's a candidate to have the deconfig flag cleared.
1128     for (const auto& [id, attributes] : _repo.getAttributesMap())
1129     {
1130         if ((attributes.creator == static_cast<uint8_t>(CreatorID::openBMC)) &&
1131             attributes.deconfig)
1132         {
1133             auto updated = _repo.updatePEL(attributes.path,
1134                                            handlePowerThermalHardwarePresent);
1135 
1136             if (updated)
1137             {
1138                 // Also update the property on D-Bus
1139                 auto objPath = std::string(OBJ_ENTRY) + '/' +
1140                                std::to_string(id.obmcID.id);
1141                 auto entryN = _pelEntries.find(objPath);
1142                 if (entryN != _pelEntries.end())
1143                 {
1144                     entryN->second->deconfig(false);
1145                 }
1146                 else
1147                 {
1148                     lg2::error(
1149                         "Could not find PEL Entry D-Bus object for {PATH}",
1150                         "PATH", objPath);
1151                 }
1152             }
1153         }
1154     }
1155 }
1156 
1157 } // namespace pels
1158 } // namespace openpower
1159