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