cddl/usr.sbin/zfsd/zfsd_event.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 zfsd_event.cc
  35  */
  36 #include <sys/cdefs.h>
  37 #include <sys/time.h>
  38 #include <sys/fs/zfs.h>
  39 #include <sys/vdev_impl.h>
  40
  41 #include <syslog.h>
  42
  43 #include <libzfs.h>
  44 /*
  45  * Undefine flush, defined by cpufunc.h on sparc64, because it conflicts with
  46  * C++ flush methods
  47  */
  48 #undef   flush
  49
  50 #include <list>
  51 #include <map>
  52 #include <sstream>
  53 #include <string>
  54
  55 #include <devdctl/guid.h>
  56 #include <devdctl/event.h>
  57 #include <devdctl/event_factory.h>
  58 #include <devdctl/exception.h>
  59 #include <devdctl/consumer.h>
  60
  61 #include "callout.h"
  62 #include "vdev_iterator.h"
  63 #include "zfsd_event.h"
  64 #include "case_file.h"
  65 #include "vdev.h"
  66 #include "zfsd.h"
  67 #include "zfsd_exception.h"
  68 #include "zpool_list.h"
  69
  70 __FBSDID("$FreeBSD$");
  71 /*============================ Namespace Control =============================*/
  72 using DevdCtl::Event;
  73 using DevdCtl::Guid;
  74 using DevdCtl::NVPairMap;
  75 using std::stringstream;
  76
  77 /*=========================== Class Implementations ==========================*/
  78
  79 /*-------------------------------- DevfsEvent --------------------------------*/
  80
  81 //- DevfsEvent Static Public Methods -------------------------------------------
  82 Event *
  83 DevfsEvent::Builder(Event::Type type,
  84                     NVPairMap &nvPairs,
  85                     const string &eventString)
  86 {
  87         return (new DevfsEvent(type, nvPairs, eventString));
  88 }
  89
  90 //- DevfsEvent Static Protected Methods ----------------------------------------
  91 nvlist_t *
  92 DevfsEvent::ReadLabel(int devFd, bool &inUse, bool &degraded)
  93 {
  94         pool_state_t poolState;
  95         char        *poolName;
  96         boolean_t    b_inuse;
  97         int          nlabels;
  98
  99         inUse    = false;
 100         degraded = false;
 101         poolName = NULL;
 102         if (zpool_in_use(g_zfsHandle, devFd, &poolState,
 103                          &poolName, &b_inuse) == 0) {
 104                 nvlist_t *devLabel;
 105
 106                 inUse = b_inuse == B_TRUE;
 107                 if (poolName != NULL)
 108                         free(poolName);
 109
 110                 nlabels = zpool_read_all_labels(devFd, &devLabel);
 111                 /*
 112                  * If we find a disk with fewer than the maximum number of
 113                  * labels, it might be the whole disk of a partitioned disk
 114                  * where ZFS resides on a partition.  In that case, we should do
 115                  * nothing and wait for the partition to appear.  Or, the disk
 116                  * might be damaged.  In that case, zfsd should do nothing and
 117                  * wait for the sysadmin to decide.
 118                  */
 119                 if (nlabels != VDEV_LABELS || devLabel == NULL)
 120                         return (NULL);
 121
 122                 try {
 123                         Vdev vdev(devLabel);
 124                         degraded = vdev.State() != VDEV_STATE_HEALTHY;
 125                         return (devLabel);
 126                 } catch (ZfsdException &exp) {
 127                         string devName = fdevname(devFd);
 128                         string devPath = _PATH_DEV + devName;
 129                         string context("DevfsEvent::ReadLabel: "
 130                                      + devPath + ": ");
 131
 132                         exp.GetString().insert(0, context);
 133                         exp.Log();
 134                 }
 135         }
 136         return (NULL);
 137 }
 138
 139 bool
 140 DevfsEvent::OnlineByLabel(const string &devPath, const string& physPath,
 141                               nvlist_t *devConfig)
 142 {
 143         try {
 144                 /*
 145                  * A device with ZFS label information has been
 146                  * inserted.  If it matches a device for which we
 147                  * have a case, see if we can solve that case.
 148                  */
 149                 syslog(LOG_INFO, "Interrogating VDEV label for %s\n",
 150                        devPath.c_str());
 151                 Vdev vdev(devConfig);
 152                 CaseFile *caseFile(CaseFile::Find(vdev.PoolGUID(),
 153                                                   vdev.GUID()));
 154                 if (caseFile != NULL)
 155                         return (caseFile->ReEvaluate(devPath, physPath, &vdev));
 156
 157         } catch (ZfsdException &exp) {
 158                 string context("DevfsEvent::OnlineByLabel: " + devPath + ": ");
 159
 160                 exp.GetString().insert(0, context);
 161                 exp.Log();
 162         }
 163         return (false);
 164 }
 165
 166 //- DevfsEvent Virtual Public Methods ------------------------------------------
 167 Event *
 168 DevfsEvent::DeepCopy() const
 169 {
 170         return (new DevfsEvent(*this));
 171 }
 172
 173 bool
 174 DevfsEvent::Process() const
 175 {
 176         /*
 177          * We are only concerned with newly discovered
 178          * devices that can be ZFS vdevs.
 179          */
 180         if (Value("type") != "CREATE" || !IsDiskDev())
 181                 return (false);
 182
 183         /* Log the event since it is of interest. */
 184         Log(LOG_INFO);
 185
 186         string devPath;
 187         if (!DevPath(devPath))
 188                 return (false);
 189
 190         int devFd(open(devPath.c_str(), O_RDONLY));
 191         if (devFd == -1)
 192                 return (false);
 193
 194         bool inUse;
 195         bool degraded;
 196         nvlist_t *devLabel(ReadLabel(devFd, inUse, degraded));
 197
 198         string physPath;
 199         bool havePhysPath(PhysicalPath(physPath));
 200
 201         string devName;
 202         DevName(devName);
 203         close(devFd);
 204
 205         if (inUse && devLabel != NULL) {
 206                 OnlineByLabel(devPath, physPath, devLabel);
 207         } else if (degraded) {
 208                 syslog(LOG_INFO, "%s is marked degraded.  Ignoring "
 209                        "as a replace by physical path candidate.\n",
 210                        devName.c_str());
 211         } else if (havePhysPath && IsWholeDev()) {
 212                 /*
 213                  * TODO: attempt to resolve events using every casefile
 214                  * that matches this physpath
 215                  */
 216                 CaseFile *caseFile(CaseFile::Find(physPath));
 217                 if (caseFile != NULL) {
 218                         syslog(LOG_INFO,
 219                                "Found CaseFile(%s:%s:%s) - ReEvaluating\n",
 220                                caseFile->PoolGUIDString().c_str(),
 221                                caseFile->VdevGUIDString().c_str(),
 222                                zpool_state_to_name(caseFile->VdevState(),
 223                                                    VDEV_AUX_NONE));
 224                         caseFile->ReEvaluate(devPath, physPath, /*vdev*/NULL);
 225                 }
 226         }
 227         if (devLabel != NULL)
 228                 nvlist_free(devLabel);
 229         return (false);
 230 }
 231
 232 //- DevfsEvent Protected Methods -----------------------------------------------
 233 DevfsEvent::DevfsEvent(Event::Type type, NVPairMap &nvpairs,
 234                                const string &eventString)
 235  : DevdCtl::DevfsEvent(type, nvpairs, eventString)
 236 {
 237 }
 238
 239 DevfsEvent::DevfsEvent(const DevfsEvent &src)
 240  : DevdCtl::DevfsEvent::DevfsEvent(src)
 241 {
 242 }
 243
 244 /*-------------------------------- GeomEvent --------------------------------*/
 245
 246 //- GeomEvent Static Public Methods -------------------------------------------
 247 Event *
 248 GeomEvent::Builder(Event::Type type,
 249                    NVPairMap &nvPairs,
 250                    const string &eventString)
 251 {
 252         return (new GeomEvent(type, nvPairs, eventString));
 253 }
 254
 255 //- GeomEvent Virtual Public Methods ------------------------------------------
 256 Event *
 257 GeomEvent::DeepCopy() const
 258 {
 259         return (new GeomEvent(*this));
 260 }
 261
 262 bool
 263 GeomEvent::Process() const
 264 {
 265         /*
 266          * We are only concerned with physical path changes, because those can
 267          * be used to satisfy autoreplace operations
 268          */
 269         if (Value("type") != "GEOM::physpath" || !IsDiskDev())
 270                 return (false);
 271
 272         /* Log the event since it is of interest. */
 273         Log(LOG_INFO);
 274
 275         string devPath;
 276         if (!DevPath(devPath))
 277                 return (false);
 278
 279         string physPath;
 280         bool havePhysPath(PhysicalPath(physPath));
 281
 282         string devName;
 283         DevName(devName);
 284
 285         if (havePhysPath) {
 286                 /*
 287                  * TODO: attempt to resolve events using every casefile
 288                  * that matches this physpath
 289                  */
 290                 CaseFile *caseFile(CaseFile::Find(physPath));
 291                 if (caseFile != NULL) {
 292                         syslog(LOG_INFO,
 293                                "Found CaseFile(%s:%s:%s) - ReEvaluating\n",
 294                                caseFile->PoolGUIDString().c_str(),
 295                                caseFile->VdevGUIDString().c_str(),
 296                                zpool_state_to_name(caseFile->VdevState(),
 297                                                    VDEV_AUX_NONE));
 298                         caseFile->ReEvaluate(devPath, physPath, /*vdev*/NULL);
 299                 }
 300         }
 301         return (false);
 302 }
 303
 304 //- GeomEvent Protected Methods -----------------------------------------------
 305 GeomEvent::GeomEvent(Event::Type type, NVPairMap &nvpairs,
 306                                const string &eventString)
 307  : DevdCtl::GeomEvent(type, nvpairs, eventString)
 308 {
 309 }
 310
 311 GeomEvent::GeomEvent(const GeomEvent &src)
 312  : DevdCtl::GeomEvent::GeomEvent(src)
 313 {
 314 }
 315
 316
 317 /*--------------------------------- ZfsEvent ---------------------------------*/
 318 //- ZfsEvent Static Public Methods ---------------------------------------------
 319 DevdCtl::Event *
 320 ZfsEvent::Builder(Event::Type type, NVPairMap &nvpairs,
 321                   const string &eventString)
 322 {
 323         return (new ZfsEvent(type, nvpairs, eventString));
 324 }
 325
 326 //- ZfsEvent Virtual Public Methods --------------------------------------------
 327 Event *
 328 ZfsEvent::DeepCopy() const
 329 {
 330         return (new ZfsEvent(*this));
 331 }
 332
 333 bool
 334 ZfsEvent::Process() const
 335 {
 336         string logstr("");
 337
 338         if (!Contains("class") && !Contains("type")) {
 339                 syslog(LOG_ERR,
 340                        "ZfsEvent::Process: Missing class or type data.");
 341                 return (false);
 342         }
 343
 344         /* On config syncs, replay any queued events first. */
 345         if (Value("type").find("misc.fs.zfs.config_sync") == 0) {
 346                 /*
 347                  * Even if saved events are unconsumed the second time
 348                  * around, drop them.  Any events that still can't be
 349                  * consumed are probably referring to vdevs or pools that
 350                  * no longer exist.
 351                  */
 352                 ZfsDaemon::Get().ReplayUnconsumedEvents(/*discard*/true);
 353                 CaseFile::ReEvaluateByGuid(PoolGUID(), *this);
 354         }
 355
 356         if (Value("type").find("misc.fs.zfs.") == 0) {
 357                 /* Configuration changes, resilver events, etc. */
 358                 ProcessPoolEvent();
 359                 return (false);
 360         }
 361
 362         if (!Contains("pool_guid") || !Contains("vdev_guid")) {
 363                 /* Only currently interested in Vdev related events. */
 364                 return (false);
 365         }
 366
 367         CaseFile *caseFile(CaseFile::Find(PoolGUID(), VdevGUID()));
 368         if (caseFile != NULL) {
 369                 Log(LOG_INFO);
 370                 syslog(LOG_INFO, "Evaluating existing case file\n");
 371                 caseFile->ReEvaluate(*this);
 372                 return (false);
 373         }
 374
 375         /* Skip events that can't be handled. */
 376         Guid poolGUID(PoolGUID());
 377         /* If there are no replicas for a pool, then it's not manageable. */
 378         if (Value("class").find("fs.zfs.vdev.no_replicas") == 0) {
 379                 stringstream msg;
 380                 msg << "No replicas available for pool "  << poolGUID;
 381                 msg << ", ignoring";
 382                 Log(LOG_INFO);
 383                 syslog(LOG_INFO, "%s", msg.str().c_str());
 384                 return (false);
 385         }
 386
 387         /*
 388          * Create a case file for this vdev, and have it
 389          * evaluate the event.
 390          */
 391         ZpoolList zpl(ZpoolList::ZpoolByGUID, &poolGUID);
 392         if (zpl.empty()) {
 393                 stringstream msg;
 394                 int priority = LOG_INFO;
 395                 msg << "ZfsEvent::Process: Event for unknown pool ";
 396                 msg << poolGUID << " ";
 397                 msg << "queued";
 398                 Log(LOG_INFO);
 399                 syslog(priority, "%s", msg.str().c_str());
 400                 return (true);
 401         }
 402
 403         nvlist_t *vdevConfig = VdevIterator(zpl.front()).Find(VdevGUID());
 404         if (vdevConfig == NULL) {
 405                 stringstream msg;
 406                 int priority = LOG_INFO;
 407                 msg << "ZfsEvent::Process: Event for unknown vdev ";
 408                 msg << VdevGUID() << " ";
 409                 msg << "queued";
 410                 Log(LOG_INFO);
 411                 syslog(priority, "%s", msg.str().c_str());
 412                 return (true);
 413         }
 414
 415         Vdev vdev(zpl.front(), vdevConfig);
 416         caseFile = &CaseFile::Create(vdev);
 417         if (caseFile->ReEvaluate(*this) == false) {
 418                 stringstream msg;
 419                 int priority = LOG_INFO;
 420                 msg << "ZfsEvent::Process: Unconsumed event for vdev(";
 421                 msg << zpool_get_name(zpl.front()) << ",";
 422                 msg << vdev.GUID() << ") ";
 423                 msg << "queued";
 424                 Log(LOG_INFO);
 425                 syslog(priority, "%s", msg.str().c_str());
 426                 return (true);
 427         }
 428         return (false);
 429 }
 430
 431 //- ZfsEvent Protected Methods -------------------------------------------------
 432 ZfsEvent::ZfsEvent(Event::Type type, NVPairMap &nvpairs,
 433                            const string &eventString)
 434  : DevdCtl::ZfsEvent(type, nvpairs, eventString)
 435 {
 436 }
 437
 438 ZfsEvent::ZfsEvent(const ZfsEvent &src)
 439  : DevdCtl::ZfsEvent(src)
 440 {
 441 }
 442
 443 /*
 444  * Sometimes the kernel won't detach a spare when it is no longer needed.  This
 445  * can happen for example if a drive is removed, then either the pool is
 446  * exported or the machine is powered off, then the drive is reinserted, then
 447  * the machine is powered on or the pool is imported.  ZFSD must detach these
 448  * spares itself.
 449  */
 450 void
 451 ZfsEvent::CleanupSpares() const
 452 {
 453         Guid poolGUID(PoolGUID());
 454         ZpoolList zpl(ZpoolList::ZpoolByGUID, &poolGUID);
 455         if (!zpl.empty()) {
 456                 zpool_handle_t* hdl;
 457
 458                 hdl = zpl.front();
 459                 VdevIterator(hdl).Each(TryDetach, (void*)hdl);
 460         }
 461 }
 462
 463 void
 464 ZfsEvent::ProcessPoolEvent() const
 465 {
 466         bool degradedDevice(false);
 467
 468         /* The pool is destroyed.  Discard any open cases */
 469         if (Value("type") == "misc.fs.zfs.pool_destroy") {
 470                 Log(LOG_INFO);
 471                 CaseFile::ReEvaluateByGuid(PoolGUID(), *this);
 472                 return;
 473         }
 474
 475         CaseFile *caseFile(CaseFile::Find(PoolGUID(), VdevGUID()));
 476         if (caseFile != NULL) {
 477                 if (caseFile->VdevState() != VDEV_STATE_UNKNOWN
 478                  && caseFile->VdevState() < VDEV_STATE_HEALTHY)
 479                         degradedDevice = true;
 480
 481                 Log(LOG_INFO);
 482                 caseFile->ReEvaluate(*this);
 483         }
 484         else if (Value("type") == "misc.fs.zfs.resilver_finish")
 485         {
 486                 /*
 487                  * It's possible to get a resilver_finish event with no
 488                  * corresponding casefile.  For example, if a damaged pool were
 489                  * exported, repaired, then reimported.
 490                  */
 491                 Log(LOG_INFO);
 492                 CleanupSpares();
 493         }
 494
 495         if (Value("type") == "misc.fs.zfs.vdev_remove"
 496          && degradedDevice == false) {
 497
 498                 /* See if any other cases can make use of this device. */
 499                 Log(LOG_INFO);
 500                 ZfsDaemon::RequestSystemRescan();
 501         }
 502 }
 503
 504 bool
 505 ZfsEvent::TryDetach(Vdev &vdev, void *cbArg)
 506 {
 507         /*
 508          * Outline:
 509          * if this device is a spare, and its parent includes one healthy,
 510          * non-spare child, then detach this device.
 511          */
 512         zpool_handle_t *hdl(static_cast<zpool_handle_t*>(cbArg));
 513
 514         if (vdev.IsSpare()) {
 515                 std::list<Vdev> siblings;
 516                 std::list<Vdev>::iterator siblings_it;
 517                 boolean_t cleanup = B_FALSE;
 518
 519                 Vdev parent = vdev.Parent();
 520                 siblings = parent.Children();
 521
 522                 /* Determine whether the parent should be cleaned up */
 523                 for (siblings_it = siblings.begin();
 524                      siblings_it != siblings.end();
 525                      siblings_it++) {
 526                         Vdev sibling = *siblings_it;
 527
 528                         if (!sibling.IsSpare() &&
 529                              sibling.State() == VDEV_STATE_HEALTHY) {
 530                                 cleanup = B_TRUE;
 531                                 break;
 532                         }
 533                 }
 534
 535                 if (cleanup) {
 536                         syslog(LOG_INFO, "Detaching spare vdev %s from pool %s",
 537                                vdev.Path().c_str(), zpool_get_name(hdl));
 538                         zpool_vdev_detach(hdl, vdev.Path().c_str());
 539                 }
 540
 541         }
 542
 543         /* Always return false, because there may be other spares to detach */
 544         return (false);
 545 }