1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 245cabbc6bSPrashanth Sreenivasa * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 255f368aefSYuri Pankov * Copyright 2017 Nexenta Systems, Inc. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27c8811bd3SToomas Soome * Copyright 2016 Toomas Soome <tsoome@me.com> 28ce1577b0SDave Eddy * Copyright 2017 Joyent, Inc. 29fa9e4066Sahrens */ 30fa9e4066Sahrens 31fa9e4066Sahrens #include <sys/zfs_context.h> 32ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 33fa9e4066Sahrens #include <sys/spa.h> 34fa9e4066Sahrens #include <sys/spa_impl.h> 355cabbc6bSPrashanth Sreenivasa #include <sys/bpobj.h> 36fa9e4066Sahrens #include <sys/dmu.h> 37fa9e4066Sahrens #include <sys/dmu_tx.h> 385cabbc6bSPrashanth Sreenivasa #include <sys/dsl_dir.h> 39fa9e4066Sahrens #include <sys/vdev_impl.h> 40fa9e4066Sahrens #include <sys/uberblock_impl.h> 41fa9e4066Sahrens #include <sys/metaslab.h> 42fa9e4066Sahrens #include <sys/metaslab_impl.h> 43fa9e4066Sahrens #include <sys/space_map.h> 440713e232SGeorge Wilson #include <sys/space_reftree.h> 45fa9e4066Sahrens #include <sys/zio.h> 46fa9e4066Sahrens #include <sys/zap.h> 47fa9e4066Sahrens #include <sys/fs/zfs.h> 48c5904d13Seschrock #include <sys/arc.h> 49e6ca193dSGeorge Wilson #include <sys/zil.h> 503f9d6ad7SLin Ling #include <sys/dsl_scan.h> 51770499e1SDan Kimmel #include <sys/abd.h> 52fa9e4066Sahrens 53fa9e4066Sahrens /* 54fa9e4066Sahrens * Virtual device management. 55fa9e4066Sahrens */ 56fa9e4066Sahrens 57fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 58fa9e4066Sahrens &vdev_root_ops, 59fa9e4066Sahrens &vdev_raidz_ops, 60fa9e4066Sahrens &vdev_mirror_ops, 61fa9e4066Sahrens &vdev_replacing_ops, 6299653d4eSeschrock &vdev_spare_ops, 63fa9e4066Sahrens &vdev_disk_ops, 64fa9e4066Sahrens &vdev_file_ops, 65fa9e4066Sahrens &vdev_missing_ops, 6688ecc943SGeorge Wilson &vdev_hole_ops, 675cabbc6bSPrashanth Sreenivasa &vdev_indirect_ops, 68fa9e4066Sahrens NULL 69fa9e4066Sahrens }; 70fa9e4066Sahrens 71088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 72088f3894Sahrens int zfs_scrub_limit = 10; 7305b2b3b8Smishra 74bf3e216cSMatthew Ahrens /* 75bf3e216cSMatthew Ahrens * When a vdev is added, it will be divided into approximately (but no 76bf3e216cSMatthew Ahrens * more than) this number of metaslabs. 77bf3e216cSMatthew Ahrens */ 78bf3e216cSMatthew Ahrens int metaslabs_per_vdev = 200; 79bf3e216cSMatthew Ahrens 80*6f793812SPavel Zakharov boolean_t vdev_validate_skip = B_FALSE; 81*6f793812SPavel Zakharov 823ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 833ee8c80cSPavel Zakharov void 843ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 853ee8c80cSPavel Zakharov { 863ee8c80cSPavel Zakharov va_list adx; 873ee8c80cSPavel Zakharov char buf[256]; 883ee8c80cSPavel Zakharov 893ee8c80cSPavel Zakharov va_start(adx, fmt); 903ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 913ee8c80cSPavel Zakharov va_end(adx); 923ee8c80cSPavel Zakharov 933ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 943ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 953ee8c80cSPavel Zakharov vd->vdev_path, buf); 963ee8c80cSPavel Zakharov } else { 973ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 983ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 993ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1003ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1013ee8c80cSPavel Zakharov } 1023ee8c80cSPavel Zakharov } 1033ee8c80cSPavel Zakharov 104*6f793812SPavel Zakharov void 105*6f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 106*6f793812SPavel Zakharov { 107*6f793812SPavel Zakharov char state[20]; 108*6f793812SPavel Zakharov 109*6f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 110*6f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 111*6f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 112*6f793812SPavel Zakharov return; 113*6f793812SPavel Zakharov } 114*6f793812SPavel Zakharov 115*6f793812SPavel Zakharov switch (vd->vdev_state) { 116*6f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 117*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 118*6f793812SPavel Zakharov break; 119*6f793812SPavel Zakharov case VDEV_STATE_CLOSED: 120*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 121*6f793812SPavel Zakharov break; 122*6f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 123*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 124*6f793812SPavel Zakharov break; 125*6f793812SPavel Zakharov case VDEV_STATE_REMOVED: 126*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 127*6f793812SPavel Zakharov break; 128*6f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 129*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 130*6f793812SPavel Zakharov break; 131*6f793812SPavel Zakharov case VDEV_STATE_FAULTED: 132*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 133*6f793812SPavel Zakharov break; 134*6f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 135*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 136*6f793812SPavel Zakharov break; 137*6f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 138*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 139*6f793812SPavel Zakharov break; 140*6f793812SPavel Zakharov default: 141*6f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 142*6f793812SPavel Zakharov (uint_t)vd->vdev_state); 143*6f793812SPavel Zakharov } 144*6f793812SPavel Zakharov 145*6f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 146*6f793812SPavel Zakharov "", vd->vdev_id, vd->vdev_ops->vdev_op_type, 147*6f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 148*6f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 149*6f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 150*6f793812SPavel Zakharov 151*6f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 152*6f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 153*6f793812SPavel Zakharov } 154*6f793812SPavel Zakharov 155fa9e4066Sahrens /* 156fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 157fa9e4066Sahrens */ 158fa9e4066Sahrens static vdev_ops_t * 159fa9e4066Sahrens vdev_getops(const char *type) 160fa9e4066Sahrens { 161fa9e4066Sahrens vdev_ops_t *ops, **opspp; 162fa9e4066Sahrens 163fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 164fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 165fa9e4066Sahrens break; 166fa9e4066Sahrens 167fa9e4066Sahrens return (ops); 168fa9e4066Sahrens } 169fa9e4066Sahrens 170fa9e4066Sahrens /* 171fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 172fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 173fa9e4066Sahrens */ 174fa9e4066Sahrens uint64_t 175fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 176fa9e4066Sahrens { 177ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 178fa9e4066Sahrens uint64_t csize; 179fa9e4066Sahrens 180573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 181fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 182fa9e4066Sahrens asize = MAX(asize, csize); 183fa9e4066Sahrens } 184fa9e4066Sahrens 185fa9e4066Sahrens return (asize); 186fa9e4066Sahrens } 187fa9e4066Sahrens 1882a79c5feSlling /* 189573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 190573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 191573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 192573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 1932a79c5feSlling */ 1942a79c5feSlling uint64_t 195573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 1962a79c5feSlling { 197573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 1982a79c5feSlling 199573ca77eSGeorge Wilson /* 2004263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 201573ca77eSGeorge Wilson * just return our own asize. 202573ca77eSGeorge Wilson */ 203573ca77eSGeorge Wilson if (pvd == NULL) 204573ca77eSGeorge Wilson return (vd->vdev_asize); 2052a79c5feSlling 2062a79c5feSlling /* 207573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 208573ca77eSGeorge Wilson * to the nearest metaslab. 2092a79c5feSlling */ 210573ca77eSGeorge Wilson if (vd == vd->vdev_top) 211573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2122a79c5feSlling 213573ca77eSGeorge Wilson /* 214573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 215573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 216573ca77eSGeorge Wilson */ 217573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 218c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 219c040c10cSSteven Hartland pvd->vdev_children); 2202a79c5feSlling 221573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 222573ca77eSGeorge Wilson } 2232a79c5feSlling 224573ca77eSGeorge Wilson void 225573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 226573ca77eSGeorge Wilson { 227573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 228573ca77eSGeorge Wilson 229573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 230573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2312a79c5feSlling } 2322a79c5feSlling 233fa9e4066Sahrens vdev_t * 234fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 235fa9e4066Sahrens { 236fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 237fa9e4066Sahrens 238e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 239e05725b1Sbonwick 240088f3894Sahrens if (vdev < rvd->vdev_children) { 241088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 242fa9e4066Sahrens return (rvd->vdev_child[vdev]); 243088f3894Sahrens } 244fa9e4066Sahrens 245fa9e4066Sahrens return (NULL); 246fa9e4066Sahrens } 247fa9e4066Sahrens 248fa9e4066Sahrens vdev_t * 249fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 250fa9e4066Sahrens { 251fa9e4066Sahrens vdev_t *mvd; 252fa9e4066Sahrens 2530e34b6a7Sbonwick if (vd->vdev_guid == guid) 254fa9e4066Sahrens return (vd); 255fa9e4066Sahrens 256573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 257fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 258fa9e4066Sahrens NULL) 259fa9e4066Sahrens return (mvd); 260fa9e4066Sahrens 261fa9e4066Sahrens return (NULL); 262fa9e4066Sahrens } 263fa9e4066Sahrens 26412380e1eSArne Jansen static int 26512380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 26612380e1eSArne Jansen { 26712380e1eSArne Jansen int n = 0; 26812380e1eSArne Jansen 26912380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 27012380e1eSArne Jansen return (1); 27112380e1eSArne Jansen 27212380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 27312380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 27412380e1eSArne Jansen 27512380e1eSArne Jansen return (n); 27612380e1eSArne Jansen } 27712380e1eSArne Jansen 27812380e1eSArne Jansen int 27912380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 28012380e1eSArne Jansen { 28112380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 28212380e1eSArne Jansen } 28312380e1eSArne Jansen 284fa9e4066Sahrens void 285fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 286fa9e4066Sahrens { 287fa9e4066Sahrens size_t oldsize, newsize; 288fa9e4066Sahrens uint64_t id = cvd->vdev_id; 289fa9e4066Sahrens vdev_t **newchild; 29081cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 291fa9e4066Sahrens 29281cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 293fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 294fa9e4066Sahrens 295fa9e4066Sahrens cvd->vdev_parent = pvd; 296fa9e4066Sahrens 297fa9e4066Sahrens if (pvd == NULL) 298fa9e4066Sahrens return; 299fa9e4066Sahrens 300fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 301fa9e4066Sahrens 302fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 303fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 304fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 305fa9e4066Sahrens 306fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 307fa9e4066Sahrens if (pvd->vdev_child != NULL) { 308fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 309fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 310fa9e4066Sahrens } 311fa9e4066Sahrens 312fa9e4066Sahrens pvd->vdev_child = newchild; 313fa9e4066Sahrens pvd->vdev_child[id] = cvd; 314fa9e4066Sahrens 315fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 316fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 317fa9e4066Sahrens 318fa9e4066Sahrens /* 319fa9e4066Sahrens * Walk up all ancestors to update guid sum. 320fa9e4066Sahrens */ 321fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 322fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 323fa9e4066Sahrens } 324fa9e4066Sahrens 325fa9e4066Sahrens void 326fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 327fa9e4066Sahrens { 328fa9e4066Sahrens int c; 329fa9e4066Sahrens uint_t id = cvd->vdev_id; 330fa9e4066Sahrens 331fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 332fa9e4066Sahrens 333fa9e4066Sahrens if (pvd == NULL) 334fa9e4066Sahrens return; 335fa9e4066Sahrens 336fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 337fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 338fa9e4066Sahrens 339fa9e4066Sahrens pvd->vdev_child[id] = NULL; 340fa9e4066Sahrens cvd->vdev_parent = NULL; 341fa9e4066Sahrens 342fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 343fa9e4066Sahrens if (pvd->vdev_child[c]) 344fa9e4066Sahrens break; 345fa9e4066Sahrens 346fa9e4066Sahrens if (c == pvd->vdev_children) { 347fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 348fa9e4066Sahrens pvd->vdev_child = NULL; 349fa9e4066Sahrens pvd->vdev_children = 0; 350fa9e4066Sahrens } 351fa9e4066Sahrens 352fa9e4066Sahrens /* 353fa9e4066Sahrens * Walk up all ancestors to update guid sum. 354fa9e4066Sahrens */ 355fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 356fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 357fa9e4066Sahrens } 358fa9e4066Sahrens 359fa9e4066Sahrens /* 360fa9e4066Sahrens * Remove any holes in the child array. 361fa9e4066Sahrens */ 362fa9e4066Sahrens void 363fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 364fa9e4066Sahrens { 365fa9e4066Sahrens vdev_t **newchild, *cvd; 366fa9e4066Sahrens int oldc = pvd->vdev_children; 367573ca77eSGeorge Wilson int newc; 368fa9e4066Sahrens 369e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 370fa9e4066Sahrens 371573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 372fa9e4066Sahrens if (pvd->vdev_child[c]) 373fa9e4066Sahrens newc++; 374fa9e4066Sahrens 375fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 376fa9e4066Sahrens 377573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 378fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 379fa9e4066Sahrens newchild[newc] = cvd; 380fa9e4066Sahrens cvd->vdev_id = newc++; 381fa9e4066Sahrens } 382fa9e4066Sahrens } 383fa9e4066Sahrens 384fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 385fa9e4066Sahrens pvd->vdev_child = newchild; 386fa9e4066Sahrens pvd->vdev_children = newc; 387fa9e4066Sahrens } 388fa9e4066Sahrens 389fa9e4066Sahrens /* 390fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 391fa9e4066Sahrens */ 39288ecc943SGeorge Wilson vdev_t * 393fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 394fa9e4066Sahrens { 395fa9e4066Sahrens vdev_t *vd; 3965cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 397fa9e4066Sahrens 398fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 3995cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 400fa9e4066Sahrens 4010e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4020e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4030e34b6a7Sbonwick spa->spa_root_vdev = vd; 404e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4050e34b6a7Sbonwick } 4060e34b6a7Sbonwick 40788ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4080e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4090e34b6a7Sbonwick /* 4100e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4110e34b6a7Sbonwick * which must be unique among all pools. 4120e34b6a7Sbonwick */ 4131195e687SMark J Musante guid = spa_generate_guid(NULL); 4140e34b6a7Sbonwick } else { 4150e34b6a7Sbonwick /* 4160e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4170e34b6a7Sbonwick */ 4181195e687SMark J Musante guid = spa_generate_guid(spa); 4190e34b6a7Sbonwick } 4200e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4210e34b6a7Sbonwick } 4220e34b6a7Sbonwick 423fa9e4066Sahrens vd->vdev_spa = spa; 424fa9e4066Sahrens vd->vdev_id = id; 425fa9e4066Sahrens vd->vdev_guid = guid; 426fa9e4066Sahrens vd->vdev_guid_sum = guid; 427fa9e4066Sahrens vd->vdev_ops = ops; 428fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 42988ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 4305cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 4315cabbc6bSPrashanth Sreenivasa 4325cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 4335cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 4345cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); 435fa9e4066Sahrens 436fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 4375ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 438e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 4390f7643c7SGeorge Wilson mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); 4408ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 4415cabbc6bSPrashanth Sreenivasa vd->vdev_dtl[t] = range_tree_create(NULL, NULL); 4428ad4d6ddSJeff Bonwick } 443b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 444fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 445b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 446fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 447fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 4483d7072f8Seschrock vdev_queue_init(vd); 4493d7072f8Seschrock vdev_cache_init(vd); 450fa9e4066Sahrens 451fa9e4066Sahrens return (vd); 452fa9e4066Sahrens } 453fa9e4066Sahrens 454fa9e4066Sahrens /* 455fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 456fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 457fa9e4066Sahrens * different for each case. 458fa9e4066Sahrens */ 45999653d4eSeschrock int 46099653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 46199653d4eSeschrock int alloctype) 462fa9e4066Sahrens { 463fa9e4066Sahrens vdev_ops_t *ops; 464fa9e4066Sahrens char *type; 4658654d025Sperrin uint64_t guid = 0, islog, nparity; 466fa9e4066Sahrens vdev_t *vd; 4675cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 468fa9e4066Sahrens 469e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 470fa9e4066Sahrens 471fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 472be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 473fa9e4066Sahrens 474fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 475be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 476fa9e4066Sahrens 477fa9e4066Sahrens /* 478fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 479fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 480fa9e4066Sahrens */ 481fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 482fa9e4066Sahrens uint64_t label_id; 483fa9e4066Sahrens 484fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 485fa9e4066Sahrens label_id != id) 486be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 487fa9e4066Sahrens 488fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 489be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 49099653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 49199653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 492be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 493fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 494fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 495be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 49621ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 49721ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 498be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 499fa9e4066Sahrens } 500fa9e4066Sahrens 50199653d4eSeschrock /* 50299653d4eSeschrock * The first allocated vdev must be of type 'root'. 50399653d4eSeschrock */ 50499653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 505be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 50699653d4eSeschrock 5078654d025Sperrin /* 5088654d025Sperrin * Determine whether we're a log vdev. 5098654d025Sperrin */ 5108654d025Sperrin islog = 0; 5118654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 512990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 513be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 514fa9e4066Sahrens 51588ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 516be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 51788ecc943SGeorge Wilson 51899653d4eSeschrock /* 5198654d025Sperrin * Set the nparity property for RAID-Z vdevs. 52099653d4eSeschrock */ 5218654d025Sperrin nparity = -1ULL; 52299653d4eSeschrock if (ops == &vdev_raidz_ops) { 52399653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 5248654d025Sperrin &nparity) == 0) { 525b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 526be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 52799653d4eSeschrock /* 528f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 529f94275ceSAdam Leventhal * device. 53099653d4eSeschrock */ 531f94275ceSAdam Leventhal if (nparity > 1 && 532f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 533be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 534f94275ceSAdam Leventhal if (nparity > 2 && 535f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 536be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 53799653d4eSeschrock } else { 53899653d4eSeschrock /* 53999653d4eSeschrock * We require the parity to be specified for SPAs that 54099653d4eSeschrock * support multiple parity levels. 54199653d4eSeschrock */ 542f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 543be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 54499653d4eSeschrock /* 54599653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 54699653d4eSeschrock */ 5478654d025Sperrin nparity = 1; 54899653d4eSeschrock } 54999653d4eSeschrock } else { 5508654d025Sperrin nparity = 0; 55199653d4eSeschrock } 5528654d025Sperrin ASSERT(nparity != -1ULL); 5538654d025Sperrin 5548654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 5555cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 5568654d025Sperrin 5578654d025Sperrin vd->vdev_islog = islog; 5588654d025Sperrin vd->vdev_nparity = nparity; 5598654d025Sperrin 5608654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 5618654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 5628654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 5638654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 5648654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 5658654d025Sperrin &vd->vdev_physpath) == 0) 5668654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 5676809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 5686809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 56999653d4eSeschrock 570afefbcddSeschrock /* 571afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 572afefbcddSeschrock * as -1. 573afefbcddSeschrock */ 574afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 575afefbcddSeschrock &vd->vdev_wholedisk) != 0) 576afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 577afefbcddSeschrock 5785cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 5795cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 5805cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 5815cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 5825cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 5835cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 5845cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 5855cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 5865cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 5875cabbc6bSPrashanth Sreenivasa 588ea8dc4b6Seschrock /* 589ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 590ea8dc4b6Seschrock * was not present at the time of import. 591ea8dc4b6Seschrock */ 5926809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 5936809eb4eSEric Schrock &vd->vdev_not_present); 594ea8dc4b6Seschrock 595ecc2d604Sbonwick /* 596ecc2d604Sbonwick * Get the alignment requirement. 597ecc2d604Sbonwick */ 598ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 599ecc2d604Sbonwick 60088ecc943SGeorge Wilson /* 60188ecc943SGeorge Wilson * Retrieve the vdev creation time. 60288ecc943SGeorge Wilson */ 60388ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 60488ecc943SGeorge Wilson &vd->vdev_crtxg); 60588ecc943SGeorge Wilson 606fa9e4066Sahrens /* 607fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 608fa9e4066Sahrens */ 6091195e687SMark J Musante if (parent && !parent->vdev_parent && 6101195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 611fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 612fa9e4066Sahrens &vd->vdev_ms_array); 613fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 614fa9e4066Sahrens &vd->vdev_ms_shift); 615fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 616fa9e4066Sahrens &vd->vdev_asize); 6173f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 6183f9d6ad7SLin Ling &vd->vdev_removing); 619215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 620215198a6SJoe Stein &vd->vdev_top_zap); 621215198a6SJoe Stein } else { 622215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 623fa9e4066Sahrens } 624fa9e4066Sahrens 625cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 626a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 6279f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 6281195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 6299f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 630a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 631a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 632a1521560SJeff Bonwick } 633a1521560SJeff Bonwick 634215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 635215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 636215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 637215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 638215198a6SJoe Stein } else { 639215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 640215198a6SJoe Stein } 641215198a6SJoe Stein 642fa9e4066Sahrens /* 6433d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 644fa9e4066Sahrens */ 645215198a6SJoe Stein 646c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 64721ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 64821ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 649c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 650c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 6510713e232SGeorge Wilson &vd->vdev_dtl_object); 652c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 653c5904d13Seschrock &vd->vdev_unspare); 654c5904d13Seschrock } 65521ecdf64SLin Ling 65621ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 65721ecdf64SLin Ling uint64_t spare = 0; 65821ecdf64SLin Ling 65921ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 66021ecdf64SLin Ling &spare) == 0 && spare) 66121ecdf64SLin Ling spa_spare_add(vd); 66221ecdf64SLin Ling } 66321ecdf64SLin Ling 664ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 665ecc2d604Sbonwick &vd->vdev_offline); 666c5904d13Seschrock 667b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 668b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 669cb04b873SMark J Musante 6703d7072f8Seschrock /* 6713d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 6723d7072f8Seschrock * state, as the diagnosis made on another system may not be 673069f55e2SEric Schrock * valid in the current context. Local vdevs will 674069f55e2SEric Schrock * remain in the faulted state. 6753d7072f8Seschrock */ 676b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 6773d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 6783d7072f8Seschrock &vd->vdev_faulted); 6793d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 6803d7072f8Seschrock &vd->vdev_degraded); 6813d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 6823d7072f8Seschrock &vd->vdev_removed); 683069f55e2SEric Schrock 684069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 685069f55e2SEric Schrock char *aux; 686069f55e2SEric Schrock 687069f55e2SEric Schrock vd->vdev_label_aux = 688069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 689069f55e2SEric Schrock if (nvlist_lookup_string(nv, 690069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 691069f55e2SEric Schrock strcmp(aux, "external") == 0) 692069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 693069f55e2SEric Schrock } 6943d7072f8Seschrock } 695fa9e4066Sahrens } 696fa9e4066Sahrens 697fa9e4066Sahrens /* 698fa9e4066Sahrens * Add ourselves to the parent's list of children. 699fa9e4066Sahrens */ 700fa9e4066Sahrens vdev_add_child(parent, vd); 701fa9e4066Sahrens 70299653d4eSeschrock *vdp = vd; 70399653d4eSeschrock 70499653d4eSeschrock return (0); 705fa9e4066Sahrens } 706fa9e4066Sahrens 707fa9e4066Sahrens void 708fa9e4066Sahrens vdev_free(vdev_t *vd) 709fa9e4066Sahrens { 7103d7072f8Seschrock spa_t *spa = vd->vdev_spa; 711fa9e4066Sahrens 712fa9e4066Sahrens /* 713fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 714fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 715fa9e4066Sahrens */ 716fa9e4066Sahrens vdev_close(vd); 717fa9e4066Sahrens 718e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 719b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 720fa9e4066Sahrens 721fa9e4066Sahrens /* 722fa9e4066Sahrens * Free all children. 723fa9e4066Sahrens */ 724573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 725fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 726fa9e4066Sahrens 727fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 728fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 729fa9e4066Sahrens 730fa9e4066Sahrens /* 731fa9e4066Sahrens * Discard allocation state. 732fa9e4066Sahrens */ 733a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 734fa9e4066Sahrens vdev_metaslab_fini(vd); 735a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 736a1521560SJeff Bonwick } 737fa9e4066Sahrens 738fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 739fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 740fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 741fa9e4066Sahrens 742fa9e4066Sahrens /* 743fa9e4066Sahrens * Remove this vdev from its parent's child list. 744fa9e4066Sahrens */ 745fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 746fa9e4066Sahrens 747fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 748fa9e4066Sahrens 7493d7072f8Seschrock /* 7503d7072f8Seschrock * Clean up vdev structure. 7513d7072f8Seschrock */ 7523d7072f8Seschrock vdev_queue_fini(vd); 7533d7072f8Seschrock vdev_cache_fini(vd); 7543d7072f8Seschrock 7553d7072f8Seschrock if (vd->vdev_path) 7563d7072f8Seschrock spa_strfree(vd->vdev_path); 7573d7072f8Seschrock if (vd->vdev_devid) 7583d7072f8Seschrock spa_strfree(vd->vdev_devid); 7593d7072f8Seschrock if (vd->vdev_physpath) 7603d7072f8Seschrock spa_strfree(vd->vdev_physpath); 7616809eb4eSEric Schrock if (vd->vdev_fru) 7626809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 7633d7072f8Seschrock 7643d7072f8Seschrock if (vd->vdev_isspare) 7653d7072f8Seschrock spa_spare_remove(vd); 766fa94a07fSbrendan if (vd->vdev_isl2cache) 767fa94a07fSbrendan spa_l2cache_remove(vd); 7683d7072f8Seschrock 7693d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 7703d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 7718ad4d6ddSJeff Bonwick 7723d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 7730713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 7748ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 7750713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 7760713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 7778ad4d6ddSJeff Bonwick } 7783d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 7798ad4d6ddSJeff Bonwick 7805cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 7815cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 7825cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 7835cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 7845cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 7855cabbc6bSPrashanth Sreenivasa } 7865cabbc6bSPrashanth Sreenivasa 7875cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 7885cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 7895cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 7905cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 7915cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 7925cabbc6bSPrashanth Sreenivasa } 7935cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 7945cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 7955cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 7965cabbc6bSPrashanth Sreenivasa 7970f7643c7SGeorge Wilson mutex_destroy(&vd->vdev_queue_lock); 7983d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 7993d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 800e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 8013d7072f8Seschrock 8023d7072f8Seschrock if (vd == spa->spa_root_vdev) 8033d7072f8Seschrock spa->spa_root_vdev = NULL; 8043d7072f8Seschrock 8053d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 806fa9e4066Sahrens } 807fa9e4066Sahrens 808fa9e4066Sahrens /* 809fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 810fa9e4066Sahrens */ 811fa9e4066Sahrens static void 812fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 813fa9e4066Sahrens { 814fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 815fa9e4066Sahrens metaslab_t *msp; 816fa9e4066Sahrens vdev_t *vd; 817fa9e4066Sahrens int t; 818fa9e4066Sahrens 819fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 820fa9e4066Sahrens 821fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 822fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 823fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 824215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 825fa9e4066Sahrens 826fa9e4066Sahrens svd->vdev_ms_array = 0; 827fa9e4066Sahrens svd->vdev_ms_shift = 0; 828fa9e4066Sahrens svd->vdev_ms_count = 0; 829215198a6SJoe Stein svd->vdev_top_zap = 0; 830fa9e4066Sahrens 831cd0837ccSGeorge Wilson if (tvd->vdev_mg) 832cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 833fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 834fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 835fa9e4066Sahrens 836fa9e4066Sahrens svd->vdev_mg = NULL; 837fa9e4066Sahrens svd->vdev_ms = NULL; 838ecc2d604Sbonwick 839ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 840ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 841fa9e4066Sahrens 842fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 843fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 84499653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 845fa9e4066Sahrens 846fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 847fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 84899653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 849fa9e4066Sahrens 850fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 851fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 852fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 853fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 854fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 855fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 856fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 857fa9e4066Sahrens } 858fa9e4066Sahrens 859e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 860fa9e4066Sahrens vdev_config_clean(svd); 861fa9e4066Sahrens vdev_config_dirty(tvd); 862fa9e4066Sahrens } 863fa9e4066Sahrens 864e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 865e14bb325SJeff Bonwick vdev_state_clean(svd); 866e14bb325SJeff Bonwick vdev_state_dirty(tvd); 867e14bb325SJeff Bonwick } 868e14bb325SJeff Bonwick 86999653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 87099653d4eSeschrock svd->vdev_deflate_ratio = 0; 8718654d025Sperrin 8728654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 8738654d025Sperrin svd->vdev_islog = 0; 874fa9e4066Sahrens } 875fa9e4066Sahrens 876fa9e4066Sahrens static void 877fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 878fa9e4066Sahrens { 879fa9e4066Sahrens if (vd == NULL) 880fa9e4066Sahrens return; 881fa9e4066Sahrens 882fa9e4066Sahrens vd->vdev_top = tvd; 883fa9e4066Sahrens 884573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 885fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 886fa9e4066Sahrens } 887fa9e4066Sahrens 888fa9e4066Sahrens /* 889fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 890fa9e4066Sahrens */ 891fa9e4066Sahrens vdev_t * 892fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 893fa9e4066Sahrens { 894fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 895fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 896fa9e4066Sahrens vdev_t *mvd; 897fa9e4066Sahrens 898e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 899fa9e4066Sahrens 900fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 901ecc2d604Sbonwick 902ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 903573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 9044263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 9055cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 906ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 907ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 90888ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 909ecc2d604Sbonwick 910fa9e4066Sahrens vdev_remove_child(pvd, cvd); 911fa9e4066Sahrens vdev_add_child(pvd, mvd); 912fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 913fa9e4066Sahrens vdev_add_child(mvd, cvd); 914fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 915fa9e4066Sahrens 916fa9e4066Sahrens if (mvd == mvd->vdev_top) 917fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 918fa9e4066Sahrens 919fa9e4066Sahrens return (mvd); 920fa9e4066Sahrens } 921fa9e4066Sahrens 922fa9e4066Sahrens /* 923fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 924fa9e4066Sahrens */ 925fa9e4066Sahrens void 926fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 927fa9e4066Sahrens { 928fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 929fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 930fa9e4066Sahrens 931e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 932fa9e4066Sahrens 933fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 934fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 93599653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 93699653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 937ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 938fa9e4066Sahrens 939fa9e4066Sahrens vdev_remove_child(mvd, cvd); 940fa9e4066Sahrens vdev_remove_child(pvd, mvd); 9418ad4d6ddSJeff Bonwick 94299653d4eSeschrock /* 943e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 944e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 945e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 946e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 94799653d4eSeschrock */ 9488ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 9498ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 9501195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 9518ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 9528ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 9538ad4d6ddSJeff Bonwick } 954e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 955e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 956fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 957fa9e4066Sahrens 958fa9e4066Sahrens if (cvd == cvd->vdev_top) 959fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 960fa9e4066Sahrens 961fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 962fa9e4066Sahrens vdev_free(mvd); 963fa9e4066Sahrens } 964fa9e4066Sahrens 965ea8dc4b6Seschrock int 966fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 967fa9e4066Sahrens { 968fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 969ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 970ecc2d604Sbonwick uint64_t m; 971fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 972fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 973ecc2d604Sbonwick metaslab_t **mspp; 974ecc2d604Sbonwick int error; 975fa9e4066Sahrens 976a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 977a1521560SJeff Bonwick 97888ecc943SGeorge Wilson /* 97988ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 98088ecc943SGeorge Wilson */ 98188ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 9820e34b6a7Sbonwick return (0); 9830e34b6a7Sbonwick 98488ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 98588ecc943SGeorge Wilson 986fa9e4066Sahrens ASSERT(oldc <= newc); 987fa9e4066Sahrens 988ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 989fa9e4066Sahrens 990ecc2d604Sbonwick if (oldc != 0) { 991ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 992ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 993ecc2d604Sbonwick } 994fa9e4066Sahrens 995ecc2d604Sbonwick vd->vdev_ms = mspp; 996ecc2d604Sbonwick vd->vdev_ms_count = newc; 997fa9e4066Sahrens 998ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 9990713e232SGeorge Wilson uint64_t object = 0; 10000713e232SGeorge Wilson 10015cabbc6bSPrashanth Sreenivasa /* 10025cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 10035cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 10045cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 10055cabbc6bSPrashanth Sreenivasa */ 10065cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1007ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 10087bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 10097bfdf011SNeil Perrin DMU_READ_PREFETCH); 10103ee8c80cSPavel Zakharov if (error != 0) { 10113ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 10123ee8c80cSPavel Zakharov "array [error=%d]", error); 1013ecc2d604Sbonwick return (error); 10143ee8c80cSPavel Zakharov } 1015fa9e4066Sahrens } 10161e9bd7ecSPrakash Surya 10171e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 10181e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 10193ee8c80cSPavel Zakharov if (error != 0) { 10203ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 10213ee8c80cSPavel Zakharov error); 10221e9bd7ecSPrakash Surya return (error); 10233ee8c80cSPavel Zakharov } 1024fa9e4066Sahrens } 1025fa9e4066Sahrens 1026a1521560SJeff Bonwick if (txg == 0) 1027a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1028a1521560SJeff Bonwick 10293f9d6ad7SLin Ling /* 10303f9d6ad7SLin Ling * If the vdev is being removed we don't activate 10313f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 10323f9d6ad7SLin Ling * allocations are performed on this device. 10333f9d6ad7SLin Ling */ 10343f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 1035a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1036a1521560SJeff Bonwick 1037a1521560SJeff Bonwick if (txg == 0) 1038a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1039a1521560SJeff Bonwick 1040ea8dc4b6Seschrock return (0); 1041fa9e4066Sahrens } 1042fa9e4066Sahrens 1043fa9e4066Sahrens void 1044fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1045fa9e4066Sahrens { 1046fa9e4066Sahrens if (vd->vdev_ms != NULL) { 10475cabbc6bSPrashanth Sreenivasa uint64_t count = vd->vdev_ms_count; 10485cabbc6bSPrashanth Sreenivasa 1049a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 10505cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 10510713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 10520713e232SGeorge Wilson 10530713e232SGeorge Wilson if (msp != NULL) 10540713e232SGeorge Wilson metaslab_fini(msp); 10550713e232SGeorge Wilson } 1056fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1057fa9e4066Sahrens vd->vdev_ms = NULL; 10585cabbc6bSPrashanth Sreenivasa 10595cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1060fa9e4066Sahrens } 10615cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1062fa9e4066Sahrens } 1063fa9e4066Sahrens 1064e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1065e14bb325SJeff Bonwick boolean_t vps_readable; 1066e14bb325SJeff Bonwick boolean_t vps_writeable; 1067e14bb325SJeff Bonwick int vps_flags; 1068e14bb325SJeff Bonwick } vdev_probe_stats_t; 1069e14bb325SJeff Bonwick 1070e14bb325SJeff Bonwick static void 1071e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 10720a4e9518Sgw { 10738ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1074a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1075e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1076a3f829aeSBill Moore 1077a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1078e14bb325SJeff Bonwick 1079e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1080e14bb325SJeff Bonwick if (zio->io_error == 0) 1081e14bb325SJeff Bonwick vps->vps_readable = 1; 10828ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1083a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1084770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1085e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1086e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1087e14bb325SJeff Bonwick } else { 1088770499e1SDan Kimmel abd_free(zio->io_abd); 1089e14bb325SJeff Bonwick } 1090e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1091e14bb325SJeff Bonwick if (zio->io_error == 0) 1092e14bb325SJeff Bonwick vps->vps_writeable = 1; 1093770499e1SDan Kimmel abd_free(zio->io_abd); 1094e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1095a3f829aeSBill Moore zio_t *pio; 1096e14bb325SJeff Bonwick 1097e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1098e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1099e14bb325SJeff Bonwick 1100e14bb325SJeff Bonwick if (vdev_readable(vd) && 11018ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1102e14bb325SJeff Bonwick zio->io_error = 0; 1103e14bb325SJeff Bonwick } else { 1104e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 11053ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1106e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 11078ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 1108be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1109e14bb325SJeff Bonwick } 1110a3f829aeSBill Moore 1111a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1112a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1113a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1114a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1115a3f829aeSBill Moore 11160f7643c7SGeorge Wilson zio_link_t *zl = NULL; 11170f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1118a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1119be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1120a3f829aeSBill Moore 1121e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1122e14bb325SJeff Bonwick } 1123e14bb325SJeff Bonwick } 11240a4e9518Sgw 1125e14bb325SJeff Bonwick /* 1126f7170741SWill Andrews * Determine whether this device is accessible. 1127f7170741SWill Andrews * 1128f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1129f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1130f7170741SWill Andrews * a VTOC. 1131e14bb325SJeff Bonwick */ 1132e14bb325SJeff Bonwick zio_t * 1133a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1134e14bb325SJeff Bonwick { 1135e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1136a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1137a3f829aeSBill Moore zio_t *pio; 1138a3f829aeSBill Moore 1139a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 11400a4e9518Sgw 1141a3f829aeSBill Moore /* 1142a3f829aeSBill Moore * Don't probe the probe. 1143a3f829aeSBill Moore */ 1144a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1145a3f829aeSBill Moore return (NULL); 1146e14bb325SJeff Bonwick 1147a3f829aeSBill Moore /* 1148a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1149a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1150a3f829aeSBill Moore * this vdev will become parents of the probe io. 1151a3f829aeSBill Moore */ 1152a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1153e14bb325SJeff Bonwick 1154a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1155a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1156a3f829aeSBill Moore 1157a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1158a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 11598956713aSEric Schrock ZIO_FLAG_TRYHARD; 1160a3f829aeSBill Moore 1161a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1162a3f829aeSBill Moore /* 1163a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1164a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1165a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1166a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1167a3f829aeSBill Moore * any zio looking at these values can assume that 1168a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1169a3f829aeSBill Moore * important because when a device has intermittent 1170a3f829aeSBill Moore * connectivity problems, we want to ensure that 1171a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1172a3f829aeSBill Moore * the zio (EIO). 1173a3f829aeSBill Moore * 1174a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1175a3f829aeSBill Moore * values so the probe can reevaluate from first 1176a3f829aeSBill Moore * principles. 1177a3f829aeSBill Moore */ 1178a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1179a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1180a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1181a3f829aeSBill Moore } 1182a3f829aeSBill Moore 1183a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1184a3f829aeSBill Moore vdev_probe_done, vps, 1185a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1186a3f829aeSBill Moore 118798d1cbfeSGeorge Wilson /* 118898d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 118998d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 119098d1cbfeSGeorge Wilson */ 1191a3f829aeSBill Moore if (zio != NULL) { 1192a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1193a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1194a3f829aeSBill Moore } 1195e14bb325SJeff Bonwick } 1196e14bb325SJeff Bonwick 1197a3f829aeSBill Moore if (zio != NULL) 1198a3f829aeSBill Moore zio_add_child(zio, pio); 1199e14bb325SJeff Bonwick 1200a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1201e14bb325SJeff Bonwick 1202a3f829aeSBill Moore if (vps == NULL) { 1203a3f829aeSBill Moore ASSERT(zio != NULL); 1204a3f829aeSBill Moore return (NULL); 1205a3f829aeSBill Moore } 1206e14bb325SJeff Bonwick 1207e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1208a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1209e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1210770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1211770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1212e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1213e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1214e14bb325SJeff Bonwick } 1215e14bb325SJeff Bonwick 1216a3f829aeSBill Moore if (zio == NULL) 1217a3f829aeSBill Moore return (pio); 1218a3f829aeSBill Moore 1219a3f829aeSBill Moore zio_nowait(pio); 1220a3f829aeSBill Moore return (NULL); 12210a4e9518Sgw } 12220a4e9518Sgw 1223f64c0e34SEric Taylor static void 1224f64c0e34SEric Taylor vdev_open_child(void *arg) 1225f64c0e34SEric Taylor { 1226f64c0e34SEric Taylor vdev_t *vd = arg; 1227f64c0e34SEric Taylor 1228f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1229f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1230f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1231f64c0e34SEric Taylor } 1232f64c0e34SEric Taylor 1233681d9761SEric Taylor boolean_t 1234681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1235681d9761SEric Taylor { 1236681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1237681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1238681d9761SEric Taylor return (B_TRUE); 1239681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1240681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1241681d9761SEric Taylor return (B_TRUE); 1242681d9761SEric Taylor return (B_FALSE); 1243681d9761SEric Taylor } 1244681d9761SEric Taylor 1245f64c0e34SEric Taylor void 1246f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1247f64c0e34SEric Taylor { 1248f64c0e34SEric Taylor taskq_t *tq; 1249f64c0e34SEric Taylor int children = vd->vdev_children; 1250f64c0e34SEric Taylor 1251681d9761SEric Taylor /* 1252681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1253681d9761SEric Taylor * in a single thread so that the same thread holds the 1254681d9761SEric Taylor * spa_namespace_lock 1255681d9761SEric Taylor */ 1256681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1257681d9761SEric Taylor for (int c = 0; c < children; c++) 1258681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1259681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1260681d9761SEric Taylor return; 1261681d9761SEric Taylor } 1262f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1263f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1264f64c0e34SEric Taylor 1265f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1266f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1267f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1268f64c0e34SEric Taylor 1269f64c0e34SEric Taylor taskq_destroy(tq); 1270f64c0e34SEric Taylor } 1271f64c0e34SEric Taylor 12725cabbc6bSPrashanth Sreenivasa /* 12735cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 12745cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 12755cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 12765cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 12775cabbc6bSPrashanth Sreenivasa */ 12785cabbc6bSPrashanth Sreenivasa static void 12795cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 12805cabbc6bSPrashanth Sreenivasa { 12815cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 12825cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 12835cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 12845cabbc6bSPrashanth Sreenivasa } 12855cabbc6bSPrashanth Sreenivasa } 12865cabbc6bSPrashanth Sreenivasa 1287fa9e4066Sahrens /* 1288fa9e4066Sahrens * Prepare a virtual device for access. 1289fa9e4066Sahrens */ 1290fa9e4066Sahrens int 1291fa9e4066Sahrens vdev_open(vdev_t *vd) 1292fa9e4066Sahrens { 12938ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1294fa9e4066Sahrens int error; 1295fa9e4066Sahrens uint64_t osize = 0; 12964263d13fSGeorge Wilson uint64_t max_osize = 0; 12974263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1298ecc2d604Sbonwick uint64_t ashift = 0; 1299fa9e4066Sahrens 1300f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1301f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1302fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1303fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1304fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1305fa9e4066Sahrens 1306fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1307e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1308e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1309573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1310fa9e4066Sahrens 1311069f55e2SEric Schrock /* 1312069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1313069f55e2SEric Schrock * faulted, bail out of the open. 1314069f55e2SEric Schrock */ 13153d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 13163d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1317069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1318069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 13193d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1320069f55e2SEric Schrock vd->vdev_label_aux); 1321be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 13223d7072f8Seschrock } else if (vd->vdev_offline) { 1323fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1324ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1325be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1326fa9e4066Sahrens } 1327fa9e4066Sahrens 13284263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1329fa9e4066Sahrens 1330095bcd66SGeorge Wilson /* 1331095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1332095bcd66SGeorge Wilson * the vdev on error. 1333095bcd66SGeorge Wilson */ 1334095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1335ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 13368956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1337ea8dc4b6Seschrock 1338fa9e4066Sahrens if (error) { 13393d7072f8Seschrock if (vd->vdev_removed && 13403d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 13413d7072f8Seschrock vd->vdev_removed = B_FALSE; 13423d7072f8Seschrock 1343*6f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 1344*6f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 1345*6f793812SPavel Zakharov vd->vdev_stat.vs_aux); 1346*6f793812SPavel Zakharov } else { 1347*6f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1348*6f793812SPavel Zakharov vd->vdev_stat.vs_aux); 1349*6f793812SPavel Zakharov } 1350fa9e4066Sahrens return (error); 1351fa9e4066Sahrens } 1352fa9e4066Sahrens 13533d7072f8Seschrock vd->vdev_removed = B_FALSE; 13543d7072f8Seschrock 1355096d22d4SEric Schrock /* 1356096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1357096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1358096d22d4SEric Schrock */ 1359096d22d4SEric Schrock if (vd->vdev_faulted) { 1360096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1361096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1362096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1363096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1364096d22d4SEric Schrock vd->vdev_label_aux); 1365be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1366096d22d4SEric Schrock } 1367096d22d4SEric Schrock 13683d7072f8Seschrock if (vd->vdev_degraded) { 13693d7072f8Seschrock ASSERT(vd->vdev_children == 0); 13703d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 13713d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 13723d7072f8Seschrock } else { 1373069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 13743d7072f8Seschrock } 1375fa9e4066Sahrens 137688ecc943SGeorge Wilson /* 137788ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 137888ecc943SGeorge Wilson */ 137988ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 138088ecc943SGeorge Wilson return (0); 138188ecc943SGeorge Wilson 1382573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1383ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1384ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1385ea8dc4b6Seschrock VDEV_AUX_NONE); 1386ea8dc4b6Seschrock break; 1387ea8dc4b6Seschrock } 1388573ca77eSGeorge Wilson } 1389fa9e4066Sahrens 1390fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 13914263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1392fa9e4066Sahrens 1393fa9e4066Sahrens if (vd->vdev_children == 0) { 1394fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1395ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1396ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1397be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1398fa9e4066Sahrens } 1399fa9e4066Sahrens psize = osize; 1400fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 14014263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 14024263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1403fa9e4066Sahrens } else { 1404ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1405fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1406ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1407ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1408be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1409fa9e4066Sahrens } 1410fa9e4066Sahrens psize = 0; 1411fa9e4066Sahrens asize = osize; 14124263d13fSGeorge Wilson max_asize = max_osize; 1413fa9e4066Sahrens } 1414fa9e4066Sahrens 1415fa9e4066Sahrens vd->vdev_psize = psize; 1416fa9e4066Sahrens 1417573ca77eSGeorge Wilson /* 1418c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1419573ca77eSGeorge Wilson */ 1420573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1421573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1422573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1423be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1424573ca77eSGeorge Wilson } 1425573ca77eSGeorge Wilson 1426fa9e4066Sahrens if (vd->vdev_asize == 0) { 1427fa9e4066Sahrens /* 1428fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1429ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1430fa9e4066Sahrens */ 1431fa9e4066Sahrens vd->vdev_asize = asize; 14324263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1433ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1434fa9e4066Sahrens } else { 1435fa9e4066Sahrens /* 14362384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 14372384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 14382384d9f8SGeorge Wilson * issue a warning instead. 1439fa9e4066Sahrens */ 14402384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 14412384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 14422384d9f8SGeorge Wilson cmn_err(CE_WARN, 14432384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 14442384d9f8SGeorge Wilson "larger than the pool's alignment\n", 14452384d9f8SGeorge Wilson vd->vdev_path); 1446fa9e4066Sahrens } 14474263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1448573ca77eSGeorge Wilson } 1449fa9e4066Sahrens 1450573ca77eSGeorge Wilson /* 1451c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1452c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1453c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1454c040c10cSSteven Hartland * making the additional space available. 1455c040c10cSSteven Hartland * 1456c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1457c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1458c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1459c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1460c040c10cSSteven Hartland * vdev_min_asize. 1461573ca77eSGeorge Wilson */ 1462c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1463c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1464c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1465c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1466573ca77eSGeorge Wilson vd->vdev_asize = asize; 1467fa9e4066Sahrens 1468573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1469fa9e4066Sahrens 14700a4e9518Sgw /* 14710a4e9518Sgw * Ensure we can issue some IO before declaring the 14720a4e9518Sgw * vdev open for business. 14730a4e9518Sgw */ 1474e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1475e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 147698d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 147798d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 14780a4e9518Sgw return (error); 14790a4e9518Sgw } 14800a4e9518Sgw 14815cabbc6bSPrashanth Sreenivasa if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 14825cabbc6bSPrashanth Sreenivasa !vd->vdev_isl2cache && !vd->vdev_islog) { 14835cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift > spa->spa_max_ashift) 14845cabbc6bSPrashanth Sreenivasa spa->spa_max_ashift = vd->vdev_ashift; 14855cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift < spa->spa_min_ashift) 14865cabbc6bSPrashanth Sreenivasa spa->spa_min_ashift = vd->vdev_ashift; 14875cabbc6bSPrashanth Sreenivasa } 14885cabbc6bSPrashanth Sreenivasa 148981cd5c55SMatthew Ahrens /* 149081cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 149181cd5c55SMatthew Ahrens */ 149281cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 149381cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 149481cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 149581cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 149681cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 149781cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 149881cd5c55SMatthew Ahrens } 149981cd5c55SMatthew Ahrens 1500088f3894Sahrens /* 1501088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 15028ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 15038ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1504088f3894Sahrens */ 15058ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 15068ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 15078ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1508088f3894Sahrens 1509fa9e4066Sahrens return (0); 1510fa9e4066Sahrens } 1511fa9e4066Sahrens 1512560e6e96Seschrock /* 1513560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 1514*6f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 15153d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1516560e6e96Seschrock * 1517560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1518560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1519560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1520560e6e96Seschrock * will be updated but the function will return 0. 1521560e6e96Seschrock */ 1522560e6e96Seschrock int 1523*6f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1524560e6e96Seschrock { 1525560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1526560e6e96Seschrock nvlist_t *label; 1527*6f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1528560e6e96Seschrock uint64_t state; 1529*6f793812SPavel Zakharov nvlist_t *nvl; 1530*6f793812SPavel Zakharov uint64_t txg; 1531560e6e96Seschrock 1532*6f793812SPavel Zakharov if (vdev_validate_skip) 1533*6f793812SPavel Zakharov return (0); 1534*6f793812SPavel Zakharov 1535*6f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 1536*6f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1537be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1538560e6e96Seschrock 1539b5989ec7Seschrock /* 1540b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1541b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1542b5989ec7Seschrock * overwrite the previous state. 1543b5989ec7Seschrock */ 1544*6f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 1545*6f793812SPavel Zakharov return (0); 1546560e6e96Seschrock 1547*6f793812SPavel Zakharov /* 1548*6f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 1549*6f793812SPavel Zakharov * was modified at a point after the current txg. 1550*6f793812SPavel Zakharov */ 1551*6f793812SPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0) 1552*6f793812SPavel Zakharov txg = UINT64_MAX; 1553*6f793812SPavel Zakharov else 1554*6f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1555560e6e96Seschrock 1556*6f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1557*6f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1558*6f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 1559*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config"); 1560*6f793812SPavel Zakharov return (0); 1561*6f793812SPavel Zakharov } 15621195e687SMark J Musante 1563*6f793812SPavel Zakharov /* 1564*6f793812SPavel Zakharov * Determine if this vdev has been split off into another 1565*6f793812SPavel Zakharov * pool. If so, then refuse to open it. 1566*6f793812SPavel Zakharov */ 1567*6f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1568*6f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1569*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1570*6f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 1571*6f793812SPavel Zakharov nvlist_free(label); 1572*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 1573*6f793812SPavel Zakharov return (0); 1574*6f793812SPavel Zakharov } 1575560e6e96Seschrock 1576*6f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 1577*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1578*6f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1579*6f793812SPavel Zakharov nvlist_free(label); 1580*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1581*6f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 1582*6f793812SPavel Zakharov return (0); 1583*6f793812SPavel Zakharov } 15841195e687SMark J Musante 1585*6f793812SPavel Zakharov /* 1586*6f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 1587*6f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 1588*6f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 1589*6f793812SPavel Zakharov * cached config. The check will be performed again once we have the 1590*6f793812SPavel Zakharov * trusted config from the MOS. 1591*6f793812SPavel Zakharov */ 1592*6f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 1593*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1594*6f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1595*6f793812SPavel Zakharov nvlist_free(label); 1596*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 1597*6f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 1598*6f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 1599*6f793812SPavel Zakharov return (0); 1600*6f793812SPavel Zakharov } 1601*6f793812SPavel Zakharov 1602*6f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1603*6f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1604*6f793812SPavel Zakharov &aux_guid) != 0) 1605*6f793812SPavel Zakharov aux_guid = 0; 1606*6f793812SPavel Zakharov 1607*6f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 1608*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1609*6f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1610*6f793812SPavel Zakharov nvlist_free(label); 1611*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1612*6f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 1613*6f793812SPavel Zakharov return (0); 1614*6f793812SPavel Zakharov } 1615*6f793812SPavel Zakharov 1616*6f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 1617*6f793812SPavel Zakharov != 0) { 1618*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1619*6f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1620*6f793812SPavel Zakharov nvlist_free(label); 1621*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1622*6f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 1623*6f793812SPavel Zakharov return (0); 1624*6f793812SPavel Zakharov } 1625*6f793812SPavel Zakharov 1626*6f793812SPavel Zakharov /* 1627*6f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 1628*6f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 1629*6f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 1630*6f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 1631*6f793812SPavel Zakharov * vdev, we can safely compare to that instead. 1632*6f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 1633*6f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 1634*6f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 1635*6f793812SPavel Zakharov * extreme rewind. 1636*6f793812SPavel Zakharov * 1637*6f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 1638*6f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 1639*6f793812SPavel Zakharov * corrupt if it is partway through a split operation. 1640*6f793812SPavel Zakharov */ 1641*6f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 1642*6f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 1643*6f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 1644*6f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 1645*6f793812SPavel Zakharov mismatch = B_TRUE; 1646*6f793812SPavel Zakharov } else { 1647*6f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 1648*6f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 1649*6f793812SPavel Zakharov mismatch = B_TRUE; 1650560e6e96Seschrock } 1651560e6e96Seschrock 1652*6f793812SPavel Zakharov if (mismatch) { 1653560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1654560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1655560e6e96Seschrock nvlist_free(label); 1656*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 1657*6f793812SPavel Zakharov "doesn't match label guid"); 1658*6f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 1659*6f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1660*6f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 1661*6f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 1662*6f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 1663*6f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1664560e6e96Seschrock return (0); 1665560e6e96Seschrock } 1666*6f793812SPavel Zakharov } 1667560e6e96Seschrock 1668*6f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1669*6f793812SPavel Zakharov &state) != 0) { 1670*6f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1671*6f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1672560e6e96Seschrock nvlist_free(label); 1673*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 1674*6f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 1675*6f793812SPavel Zakharov return (0); 1676*6f793812SPavel Zakharov } 1677560e6e96Seschrock 1678*6f793812SPavel Zakharov nvlist_free(label); 1679*6f793812SPavel Zakharov 1680*6f793812SPavel Zakharov /* 1681*6f793812SPavel Zakharov * If this is a verbatim import, no need to check the 1682*6f793812SPavel Zakharov * state of the pool. 1683*6f793812SPavel Zakharov */ 1684*6f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1685*6f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 1686*6f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 1687*6f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 1688*6f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 1689*6f793812SPavel Zakharov return (SET_ERROR(EBADF)); 1690*6f793812SPavel Zakharov } 1691*6f793812SPavel Zakharov 1692*6f793812SPavel Zakharov /* 1693*6f793812SPavel Zakharov * If we were able to open and validate a vdev that was 1694*6f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 1695*6f793812SPavel Zakharov * now. 1696*6f793812SPavel Zakharov */ 1697*6f793812SPavel Zakharov if (vd->vdev_not_present) 1698*6f793812SPavel Zakharov vd->vdev_not_present = 0; 1699*6f793812SPavel Zakharov 1700*6f793812SPavel Zakharov return (0); 1701*6f793812SPavel Zakharov } 1702*6f793812SPavel Zakharov 1703*6f793812SPavel Zakharov static void 1704*6f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 1705*6f793812SPavel Zakharov { 1706*6f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 1707*6f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 1708*6f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 1709*6f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 1710*6f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 1711*6f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 1712*6f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 17133ee8c80cSPavel Zakharov } 1714*6f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 1715*6f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 1716*6f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 1717*6f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 1718*6f793812SPavel Zakharov } 1719*6f793812SPavel Zakharov } 1720560e6e96Seschrock 1721*6f793812SPavel Zakharov /* 1722*6f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 1723*6f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 1724*6f793812SPavel Zakharov * paths from userland config into MOS config. 1725*6f793812SPavel Zakharov */ 1726*6f793812SPavel Zakharov int 1727*6f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 1728*6f793812SPavel Zakharov { 1729*6f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 1730*6f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 1731*6f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 1732*6f793812SPavel Zakharov return (0); 1733*6f793812SPavel Zakharov 1734*6f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 1735*6f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 1736*6f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 1737*6f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 1738*6f793812SPavel Zakharov } 1739*6f793812SPavel Zakharov 1740*6f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 1741*6f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 1742*6f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 1743*6f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 1744*6f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 1745*6f793812SPavel Zakharov } 1746*6f793812SPavel Zakharov 1747*6f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 1748*6f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 1749*6f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 1750*6f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 1751*6f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 175251ece835Seschrock } 1753560e6e96Seschrock 1754*6f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 1755*6f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 1756*6f793812SPavel Zakharov dvd->vdev_child[i]); 1757*6f793812SPavel Zakharov if (error != 0) 1758*6f793812SPavel Zakharov return (error); 1759*6f793812SPavel Zakharov } 1760*6f793812SPavel Zakharov 1761*6f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 1762*6f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 1763*6f793812SPavel Zakharov 1764560e6e96Seschrock return (0); 1765560e6e96Seschrock } 1766560e6e96Seschrock 1767*6f793812SPavel Zakharov static void 1768*6f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 1769*6f793812SPavel Zakharov { 1770*6f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 1771*6f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 1772*6f793812SPavel Zakharov 1773*6f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 1774*6f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 1775*6f793812SPavel Zakharov } 1776*6f793812SPavel Zakharov 1777*6f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 1778*6f793812SPavel Zakharov return; 1779*6f793812SPavel Zakharov 1780*6f793812SPavel Zakharov /* 1781*6f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 1782*6f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 1783*6f793812SPavel Zakharov * step outside of it. 1784*6f793812SPavel Zakharov */ 1785*6f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 1786*6f793812SPavel Zakharov 1787*6f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 1788*6f793812SPavel Zakharov return; 1789*6f793812SPavel Zakharov 1790*6f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 1791*6f793812SPavel Zakharov 1792*6f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 1793*6f793812SPavel Zakharov } 1794*6f793812SPavel Zakharov 1795*6f793812SPavel Zakharov /* 1796*6f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 1797*6f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 1798*6f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 1799*6f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 1800*6f793812SPavel Zakharov */ 1801*6f793812SPavel Zakharov void 1802*6f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 1803*6f793812SPavel Zakharov { 1804*6f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 1805*6f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 1806*6f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 1807*6f793812SPavel Zakharov 1808*6f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 1809*6f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 1810*6f793812SPavel Zakharov drvd->vdev_child[i]); 1811*6f793812SPavel Zakharov } 1812*6f793812SPavel Zakharov } 1813*6f793812SPavel Zakharov 1814fa9e4066Sahrens /* 1815fa9e4066Sahrens * Close a virtual device. 1816fa9e4066Sahrens */ 1817fa9e4066Sahrens void 1818fa9e4066Sahrens vdev_close(vdev_t *vd) 1819fa9e4066Sahrens { 18208ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1821095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 18228ad4d6ddSJeff Bonwick 18238ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 18248ad4d6ddSJeff Bonwick 18251195e687SMark J Musante /* 18261195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 18271195e687SMark J Musante * going offline. 18281195e687SMark J Musante */ 1829095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 18301195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1831095bcd66SGeorge Wilson 1832fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1833fa9e4066Sahrens 18343d7072f8Seschrock vdev_cache_purge(vd); 1835fa9e4066Sahrens 1836560e6e96Seschrock /* 1837573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1838560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1839560e6e96Seschrock * it's still faulted. 1840560e6e96Seschrock */ 1841560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1842560e6e96Seschrock 1843fa9e4066Sahrens if (vd->vdev_offline) 1844fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1845fa9e4066Sahrens else 1846fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1847ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1848fa9e4066Sahrens } 1849fa9e4066Sahrens 1850dcba9f3fSGeorge Wilson void 1851dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1852dcba9f3fSGeorge Wilson { 1853dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1854dcba9f3fSGeorge Wilson 1855dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1856dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1857dcba9f3fSGeorge Wilson return; 1858dcba9f3fSGeorge Wilson 1859dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1860dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1861dcba9f3fSGeorge Wilson 1862dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1863dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1864dcba9f3fSGeorge Wilson } 1865dcba9f3fSGeorge Wilson 1866dcba9f3fSGeorge Wilson void 1867dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1868dcba9f3fSGeorge Wilson { 1869dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1870dcba9f3fSGeorge Wilson 1871dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1872dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1873dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1874dcba9f3fSGeorge Wilson 1875dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1876dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1877dcba9f3fSGeorge Wilson } 1878dcba9f3fSGeorge Wilson 1879095bcd66SGeorge Wilson /* 1880095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1881095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1882095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1883095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1884095bcd66SGeorge Wilson */ 1885fa9e4066Sahrens void 1886ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1887fa9e4066Sahrens { 1888ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1889fa9e4066Sahrens 1890e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1891ea8dc4b6Seschrock 18921195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 18931195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1894fa9e4066Sahrens vdev_close(vd); 1895fa9e4066Sahrens (void) vdev_open(vd); 1896fa9e4066Sahrens 189739c23413Seschrock /* 189839c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 189939c23413Seschrock * Otherwise, a device with an invalid label could be successfully 190039c23413Seschrock * opened in response to vdev_reopen(). 190139c23413Seschrock */ 1902c5904d13Seschrock if (vd->vdev_aux) { 1903c5904d13Seschrock (void) vdev_validate_aux(vd); 1904e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 19056809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1906573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1907573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1908c5904d13Seschrock } else { 1909*6f793812SPavel Zakharov (void) vdev_validate(vd); 1910c5904d13Seschrock } 191139c23413Seschrock 1912fa9e4066Sahrens /* 19133d7072f8Seschrock * Reassess parent vdev's health. 1914fa9e4066Sahrens */ 19153d7072f8Seschrock vdev_propagate_state(vd); 1916fa9e4066Sahrens } 1917fa9e4066Sahrens 1918fa9e4066Sahrens int 191999653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1920fa9e4066Sahrens { 1921fa9e4066Sahrens int error; 1922fa9e4066Sahrens 1923fa9e4066Sahrens /* 1924fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1925fa9e4066Sahrens * For a create, however, we want to fail the request if 1926fa9e4066Sahrens * there are any components we can't open. 1927fa9e4066Sahrens */ 1928fa9e4066Sahrens error = vdev_open(vd); 1929fa9e4066Sahrens 1930fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1931fa9e4066Sahrens vdev_close(vd); 1932fa9e4066Sahrens return (error ? error : ENXIO); 1933fa9e4066Sahrens } 1934fa9e4066Sahrens 1935fa9e4066Sahrens /* 19360713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1937fa9e4066Sahrens */ 19380713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 19390713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 194039c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1941fa9e4066Sahrens vdev_close(vd); 1942fa9e4066Sahrens return (error); 1943fa9e4066Sahrens } 1944fa9e4066Sahrens 1945fa9e4066Sahrens return (0); 1946fa9e4066Sahrens } 1947fa9e4066Sahrens 19480e34b6a7Sbonwick void 1949573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1950fa9e4066Sahrens { 1951fa9e4066Sahrens /* 1952bf3e216cSMatthew Ahrens * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev. 1953fa9e4066Sahrens */ 1954bf3e216cSMatthew Ahrens vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev); 1955fa9e4066Sahrens vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1956fa9e4066Sahrens } 1957fa9e4066Sahrens 1958fa9e4066Sahrens void 1959ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1960fa9e4066Sahrens { 1961ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 19625cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 19635cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 1964ecc2d604Sbonwick ASSERT(ISP2(flags)); 1965f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 1966fa9e4066Sahrens 1967ecc2d604Sbonwick if (flags & VDD_METASLAB) 1968ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1969ecc2d604Sbonwick 1970ecc2d604Sbonwick if (flags & VDD_DTL) 1971ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1972ecc2d604Sbonwick 1973ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1974fa9e4066Sahrens } 1975fa9e4066Sahrens 19760713e232SGeorge Wilson void 19770713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 19780713e232SGeorge Wilson { 19790713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 19800713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 19810713e232SGeorge Wilson 19820713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 19830713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 19840713e232SGeorge Wilson } 19850713e232SGeorge Wilson 19868ad4d6ddSJeff Bonwick /* 19878ad4d6ddSJeff Bonwick * DTLs. 19888ad4d6ddSJeff Bonwick * 19898ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 19909fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 19918ad4d6ddSJeff Bonwick * 19928ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 19938ad4d6ddSJeff Bonwick * 19948ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 19958ad4d6ddSJeff Bonwick * 19968ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 19978ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 19988ad4d6ddSJeff Bonwick * txgs that was scrubbed. 19998ad4d6ddSJeff Bonwick * 20008ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 20018ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 20028ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 20038ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 20048ad4d6ddSJeff Bonwick * determine whether a device can be detached. 20058ad4d6ddSJeff Bonwick * 20068ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 20078ad4d6ddSJeff Bonwick * either has the data or it doesn't. 20088ad4d6ddSJeff Bonwick * 20098ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 20108ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 20118ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 20128ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 20138ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 20148ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 20158ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 20168ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 20178ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 20188ad4d6ddSJeff Bonwick * 20198ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 20208ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 20218ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 20228ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 20238ad4d6ddSJeff Bonwick */ 2024fa9e4066Sahrens void 20258ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2026fa9e4066Sahrens { 20270713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 20288ad4d6ddSJeff Bonwick 20298ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 20308ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2031f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 20328ad4d6ddSJeff Bonwick 20335cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 20340713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 20350713e232SGeorge Wilson range_tree_add(rt, txg, size); 20365cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2037fa9e4066Sahrens } 2038fa9e4066Sahrens 20398ad4d6ddSJeff Bonwick boolean_t 20408ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2041fa9e4066Sahrens { 20420713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 20438ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2044fa9e4066Sahrens 20458ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 20468ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2047fa9e4066Sahrens 20485cabbc6bSPrashanth Sreenivasa /* 20495cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 20505cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 20515cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 20525cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 20535cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 20545cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 20555cabbc6bSPrashanth Sreenivasa */ 20565cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 20575cabbc6bSPrashanth Sreenivasa return (B_FALSE); 20585cabbc6bSPrashanth Sreenivasa 20595cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 20600713e232SGeorge Wilson if (range_tree_space(rt) != 0) 20610713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 20625cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2063fa9e4066Sahrens 2064fa9e4066Sahrens return (dirty); 2065fa9e4066Sahrens } 2066fa9e4066Sahrens 20678ad4d6ddSJeff Bonwick boolean_t 20688ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 20698ad4d6ddSJeff Bonwick { 20700713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 20718ad4d6ddSJeff Bonwick boolean_t empty; 20728ad4d6ddSJeff Bonwick 20735cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 20740713e232SGeorge Wilson empty = (range_tree_space(rt) == 0); 20755cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 20768ad4d6ddSJeff Bonwick 20778ad4d6ddSJeff Bonwick return (empty); 20788ad4d6ddSJeff Bonwick } 20798ad4d6ddSJeff Bonwick 2080b4952e17SGeorge Wilson /* 2081b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2082b4952e17SGeorge Wilson */ 2083b4952e17SGeorge Wilson static uint64_t 2084b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2085b4952e17SGeorge Wilson { 20860713e232SGeorge Wilson range_seg_t *rs; 2087b4952e17SGeorge Wilson 2088b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 20890713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2090b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2091b4952e17SGeorge Wilson 20920713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 20930713e232SGeorge Wilson return (rs->rs_start - 1); 2094b4952e17SGeorge Wilson } 2095b4952e17SGeorge Wilson 2096b4952e17SGeorge Wilson /* 2097b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2098b4952e17SGeorge Wilson */ 2099b4952e17SGeorge Wilson static uint64_t 2100b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 2101b4952e17SGeorge Wilson { 21020713e232SGeorge Wilson range_seg_t *rs; 2103b4952e17SGeorge Wilson 2104b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 21050713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2106b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2107b4952e17SGeorge Wilson 21080713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 21090713e232SGeorge Wilson return (rs->rs_end); 2110b4952e17SGeorge Wilson } 2111b4952e17SGeorge Wilson 2112b4952e17SGeorge Wilson /* 2113b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2114b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2115b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2116b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2117b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2118b4952e17SGeorge Wilson * excise the DTLs. 2119b4952e17SGeorge Wilson */ 2120b4952e17SGeorge Wilson static boolean_t 2121b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2122b4952e17SGeorge Wilson { 2123b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2124b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2125b4952e17SGeorge Wilson 2126b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 2127b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2128b4952e17SGeorge Wilson 21292d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 21302d2f193aSMatthew Ahrens return (B_FALSE); 21312d2f193aSMatthew Ahrens 2132b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 21330713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 2134b4952e17SGeorge Wilson return (B_TRUE); 2135b4952e17SGeorge Wilson 2136b4952e17SGeorge Wilson /* 2137b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2138b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2139b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2140b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2141b4952e17SGeorge Wilson * for excision. 2142b4952e17SGeorge Wilson */ 2143b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2144b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2145b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2146b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2147b4952e17SGeorge Wilson return (B_TRUE); 2148b4952e17SGeorge Wilson } 2149b4952e17SGeorge Wilson return (B_FALSE); 2150b4952e17SGeorge Wilson } 2151b4952e17SGeorge Wilson 2152fa9e4066Sahrens /* 2153fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2154fa9e4066Sahrens */ 2155fa9e4066Sahrens void 2156fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2157fa9e4066Sahrens { 2158ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 21598ad4d6ddSJeff Bonwick avl_tree_t reftree; 21608ad4d6ddSJeff Bonwick int minref; 2161fa9e4066Sahrens 21628ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2163fa9e4066Sahrens 21648ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 21658ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 21668ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 21678ad4d6ddSJeff Bonwick 21685cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 21698ad4d6ddSJeff Bonwick return; 21708ad4d6ddSJeff Bonwick 21718ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 21723f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 21733f9d6ad7SLin Ling 2174fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2175b4952e17SGeorge Wilson 2176b4952e17SGeorge Wilson /* 2177b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2178b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2179b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2180b4952e17SGeorge Wilson * the entire duration of this scan. 2181b4952e17SGeorge Wilson */ 2182088f3894Sahrens if (scrub_txg != 0 && 21833f9d6ad7SLin Ling (spa->spa_scrub_started || 2184b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2185b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2186088f3894Sahrens /* 2187088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2188088f3894Sahrens * did it without rebooting, then the scrub dtl 2189088f3894Sahrens * will be valid, so excise the old region and 2190088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2191088f3894Sahrens * dtl as-is if there was an error. 21928ad4d6ddSJeff Bonwick * 21938ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 21948ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 21958ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 21968ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 21978ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 21988ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 21998ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 22008ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 22018ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2202088f3894Sahrens */ 22030713e232SGeorge Wilson space_reftree_create(&reftree); 22040713e232SGeorge Wilson space_reftree_add_map(&reftree, 22050713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 22060713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 22070713e232SGeorge Wilson space_reftree_add_map(&reftree, 22080713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 22090713e232SGeorge Wilson space_reftree_generate_map(&reftree, 22100713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 22110713e232SGeorge Wilson space_reftree_destroy(&reftree); 2212fa9e4066Sahrens } 22130713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 22140713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 22150713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2216fa9e4066Sahrens if (scrub_done) 22170713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 22180713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 22198ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 22200713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 22218ad4d6ddSJeff Bonwick else 22220713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 22230713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2224b4952e17SGeorge Wilson 2225b4952e17SGeorge Wilson /* 2226b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2227b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2228b4952e17SGeorge Wilson */ 2229b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 22300713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 22310713e232SGeorge Wilson range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) 2232b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2233b4952e17SGeorge Wilson 2234fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2235088f3894Sahrens 2236ecc2d604Sbonwick if (txg != 0) 2237ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2238fa9e4066Sahrens return; 2239fa9e4066Sahrens } 2240fa9e4066Sahrens 2241fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 22428ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 224399bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 224499bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 22458ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 22468ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 22478ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 22488ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 22498ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 22508ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 22518ad4d6ddSJeff Bonwick else 22528ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 22530713e232SGeorge Wilson space_reftree_create(&reftree); 22548ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 22558ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 22568ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 22570713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 22588ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 22598ad4d6ddSJeff Bonwick } 22600713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 22610713e232SGeorge Wilson space_reftree_destroy(&reftree); 2262fa9e4066Sahrens } 22638ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2264fa9e4066Sahrens } 2265fa9e4066Sahrens 22660713e232SGeorge Wilson int 2267fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2268fa9e4066Sahrens { 2269fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2270ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 22710713e232SGeorge Wilson int error = 0; 2272fa9e4066Sahrens 22730713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 22745cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2275fa9e4066Sahrens 22760713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 22775cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 22780713e232SGeorge Wilson if (error) 22790713e232SGeorge Wilson return (error); 22800713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2281fa9e4066Sahrens 22820713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 228388ecc943SGeorge Wilson 22840713e232SGeorge Wilson /* 22850713e232SGeorge Wilson * Now that we've opened the space_map we need to update 22860713e232SGeorge Wilson * the in-core DTL. 22870713e232SGeorge Wilson */ 22880713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 2289ecc2d604Sbonwick 22900713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 22910713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 22920713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2293fa9e4066Sahrens 22940713e232SGeorge Wilson return (error); 22950713e232SGeorge Wilson } 22960713e232SGeorge Wilson 22970713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 22980713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 22990713e232SGeorge Wilson if (error != 0) 23000713e232SGeorge Wilson break; 23010713e232SGeorge Wilson } 2302fa9e4066Sahrens 2303fa9e4066Sahrens return (error); 2304fa9e4066Sahrens } 2305fa9e4066Sahrens 2306215198a6SJoe Stein void 2307215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2308215198a6SJoe Stein { 2309215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2310215198a6SJoe Stein 2311215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2312215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2313215198a6SJoe Stein zapobj, tx)); 2314215198a6SJoe Stein } 2315215198a6SJoe Stein 2316215198a6SJoe Stein uint64_t 2317215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2318215198a6SJoe Stein { 2319215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2320215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2321215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2322215198a6SJoe Stein 2323215198a6SJoe Stein ASSERT(zap != 0); 2324215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2325215198a6SJoe Stein zap, tx)); 2326215198a6SJoe Stein 2327215198a6SJoe Stein return (zap); 2328215198a6SJoe Stein } 2329215198a6SJoe Stein 2330215198a6SJoe Stein void 2331215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2332215198a6SJoe Stein { 2333215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2334215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2335215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2336215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2337215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2338215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2339215198a6SJoe Stein } 2340215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2341215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2342215198a6SJoe Stein } 2343215198a6SJoe Stein } 2344215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2345215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2346215198a6SJoe Stein } 2347215198a6SJoe Stein } 2348215198a6SJoe Stein 2349fa9e4066Sahrens void 2350fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2351fa9e4066Sahrens { 2352fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 23530713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2354ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 23550713e232SGeorge Wilson range_tree_t *rtsync; 2356fa9e4066Sahrens dmu_tx_t *tx; 23570713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2358fa9e4066Sahrens 23595cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 23600713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 236188ecc943SGeorge Wilson 2362fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2363fa9e4066Sahrens 23640713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 23650713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 23660713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 23670713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 23680713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 23690713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2370215198a6SJoe Stein 2371215198a6SJoe Stein /* 2372215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2373215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2374215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2375215198a6SJoe Stein */ 2376215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2377215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2378215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2379215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2380215198a6SJoe Stein } 2381215198a6SJoe Stein 2382fa9e4066Sahrens dmu_tx_commit(tx); 2383fa9e4066Sahrens return; 2384fa9e4066Sahrens } 2385fa9e4066Sahrens 23860713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 23870713e232SGeorge Wilson uint64_t new_object; 23880713e232SGeorge Wilson 23890713e232SGeorge Wilson new_object = space_map_alloc(mos, tx); 23900713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 23910713e232SGeorge Wilson 23920713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 23935cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 23940713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2395fa9e4066Sahrens } 2396fa9e4066Sahrens 23975cabbc6bSPrashanth Sreenivasa rtsync = range_tree_create(NULL, NULL); 2398fa9e4066Sahrens 2399fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 24000713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2401fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2402fa9e4066Sahrens 24030713e232SGeorge Wilson space_map_truncate(vd->vdev_dtl_sm, tx); 24040713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 24050713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2406fa9e4066Sahrens 24070713e232SGeorge Wilson range_tree_destroy(rtsync); 2408fa9e4066Sahrens 24090713e232SGeorge Wilson /* 24100713e232SGeorge Wilson * If the object for the space map has changed then dirty 24110713e232SGeorge Wilson * the top level so that we update the config. 24120713e232SGeorge Wilson */ 24130713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 24143ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 24153ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 24163ee8c80cSPavel Zakharov (u_longlong_t)object, 24173ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 24180713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 24190713e232SGeorge Wilson } 2420fa9e4066Sahrens 2421fa9e4066Sahrens dmu_tx_commit(tx); 24220713e232SGeorge Wilson 24230713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 24240713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 24250713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2426fa9e4066Sahrens } 2427fa9e4066Sahrens 24288ad4d6ddSJeff Bonwick /* 24298ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 24308ad4d6ddSJeff Bonwick * without losing data. 24318ad4d6ddSJeff Bonwick */ 24328ad4d6ddSJeff Bonwick boolean_t 24338ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 24348ad4d6ddSJeff Bonwick { 24358ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 24368ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 24378ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 24388ad4d6ddSJeff Bonwick boolean_t required; 24398ad4d6ddSJeff Bonwick 24408ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 24418ad4d6ddSJeff Bonwick 24428ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 24438ad4d6ddSJeff Bonwick return (B_TRUE); 24448ad4d6ddSJeff Bonwick 24458ad4d6ddSJeff Bonwick /* 24468ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 24478ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 24488ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 24498ad4d6ddSJeff Bonwick */ 24508ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 24518ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 24528ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 24538ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 24548ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 24558ad4d6ddSJeff Bonwick 2456cb04b873SMark J Musante if (!required && zio_injection_enabled) 2457cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2458cb04b873SMark J Musante 24598ad4d6ddSJeff Bonwick return (required); 24608ad4d6ddSJeff Bonwick } 24618ad4d6ddSJeff Bonwick 2462088f3894Sahrens /* 2463088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2464088f3894Sahrens */ 2465088f3894Sahrens boolean_t 2466088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2467088f3894Sahrens { 2468088f3894Sahrens boolean_t needed = B_FALSE; 2469088f3894Sahrens uint64_t thismin = UINT64_MAX; 2470088f3894Sahrens uint64_t thismax = 0; 2471088f3894Sahrens 2472088f3894Sahrens if (vd->vdev_children == 0) { 2473088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 24740713e232SGeorge Wilson if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 24758ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2476088f3894Sahrens 2477b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2478b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2479088f3894Sahrens needed = B_TRUE; 2480088f3894Sahrens } 2481088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2482088f3894Sahrens } else { 24838ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2484088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2485088f3894Sahrens uint64_t cmin, cmax; 2486088f3894Sahrens 2487088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2488088f3894Sahrens thismin = MIN(thismin, cmin); 2489088f3894Sahrens thismax = MAX(thismax, cmax); 2490088f3894Sahrens needed = B_TRUE; 2491088f3894Sahrens } 2492088f3894Sahrens } 2493088f3894Sahrens } 2494088f3894Sahrens 2495088f3894Sahrens if (needed && minp) { 2496088f3894Sahrens *minp = thismin; 2497088f3894Sahrens *maxp = thismax; 2498088f3894Sahrens } 2499088f3894Sahrens return (needed); 2500088f3894Sahrens } 2501088f3894Sahrens 25025cabbc6bSPrashanth Sreenivasa int 2503ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2504fa9e4066Sahrens { 25055cabbc6bSPrashanth Sreenivasa int error = 0; 2506fa9e4066Sahrens /* 2507fa9e4066Sahrens * Recursively load all children. 2508fa9e4066Sahrens */ 25095cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 25105cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 25115cabbc6bSPrashanth Sreenivasa if (error != 0) { 25125cabbc6bSPrashanth Sreenivasa return (error); 25135cabbc6bSPrashanth Sreenivasa } 25145cabbc6bSPrashanth Sreenivasa } 25155cabbc6bSPrashanth Sreenivasa 25165cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2517fa9e4066Sahrens 2518fa9e4066Sahrens /* 25190e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2520fa9e4066Sahrens */ 25215cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 25225cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 25235cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 25245cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 25253ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 25263ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 25273ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 25285cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 25295cabbc6bSPrashanth Sreenivasa } else if ((error = vdev_metaslab_init(vd, 0)) != 0) { 25303ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 25313ee8c80cSPavel Zakharov "[error=%d]", error); 25325cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 25335cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 25345cabbc6bSPrashanth Sreenivasa return (error); 25355cabbc6bSPrashanth Sreenivasa } 25365cabbc6bSPrashanth Sreenivasa } 2537fa9e4066Sahrens 2538fa9e4066Sahrens /* 2539fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2540fa9e4066Sahrens */ 25415cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2542560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2543560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 25443ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 25453ee8c80cSPavel Zakharov "[error=%d]", error); 25465cabbc6bSPrashanth Sreenivasa return (error); 25475cabbc6bSPrashanth Sreenivasa } 25485cabbc6bSPrashanth Sreenivasa 25495cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 25505cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 25515cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 25525cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 25535cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_obsolete_sm == NULL); 25545cabbc6bSPrashanth Sreenivasa 25555cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 25565cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 25575cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 25585cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 25593ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 25603ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 25613ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 25625cabbc6bSPrashanth Sreenivasa return (error); 25635cabbc6bSPrashanth Sreenivasa } 25645cabbc6bSPrashanth Sreenivasa space_map_update(vd->vdev_obsolete_sm); 25655cabbc6bSPrashanth Sreenivasa } 25665cabbc6bSPrashanth Sreenivasa 25675cabbc6bSPrashanth Sreenivasa return (0); 2568fa9e4066Sahrens } 2569fa9e4066Sahrens 257099653d4eSeschrock /* 2571fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2572fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2573fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2574fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2575fa94a07fSbrendan * repurposed to another pool. 257699653d4eSeschrock */ 257799653d4eSeschrock int 2578fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 257999653d4eSeschrock { 258099653d4eSeschrock nvlist_t *label; 258199653d4eSeschrock uint64_t guid, version; 258299653d4eSeschrock uint64_t state; 258399653d4eSeschrock 2584e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2585c5904d13Seschrock return (0); 2586c5904d13Seschrock 2587dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 258899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 258999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 259099653d4eSeschrock return (-1); 259199653d4eSeschrock } 259299653d4eSeschrock 259399653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2594ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 259599653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 259699653d4eSeschrock guid != vd->vdev_guid || 259799653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 259899653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 259999653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 260099653d4eSeschrock nvlist_free(label); 260199653d4eSeschrock return (-1); 260299653d4eSeschrock } 260399653d4eSeschrock 260499653d4eSeschrock /* 260599653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 260699653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 260799653d4eSeschrock */ 260899653d4eSeschrock nvlist_free(label); 260999653d4eSeschrock return (0); 261099653d4eSeschrock } 261199653d4eSeschrock 26125cabbc6bSPrashanth Sreenivasa /* 26135cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 26145cabbc6bSPrashanth Sreenivasa * that points to them. 26155cabbc6bSPrashanth Sreenivasa */ 261688ecc943SGeorge Wilson void 26175cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 26185cabbc6bSPrashanth Sreenivasa { 26195cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 26205cabbc6bSPrashanth Sreenivasa return; 26215cabbc6bSPrashanth Sreenivasa 26225cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 26235cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 26245cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 26255cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 26265cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 26275cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 26285cabbc6bSPrashanth Sreenivasa 26295cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 26305cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 26315cabbc6bSPrashanth Sreenivasa if (smobj == 0) 26325cabbc6bSPrashanth Sreenivasa continue; 26335cabbc6bSPrashanth Sreenivasa 26345cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 26355cabbc6bSPrashanth Sreenivasa } 26365cabbc6bSPrashanth Sreenivasa 26375cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 26385cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 26395cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 26405cabbc6bSPrashanth Sreenivasa } 26415cabbc6bSPrashanth Sreenivasa 26425cabbc6bSPrashanth Sreenivasa static void 26435cabbc6bSPrashanth Sreenivasa vdev_remove_empty(vdev_t *vd, uint64_t txg) 264488ecc943SGeorge Wilson { 264588ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 264688ecc943SGeorge Wilson dmu_tx_t *tx; 264788ecc943SGeorge Wilson 2648215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 2649215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 265088ecc943SGeorge Wilson 265188ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 26522e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 26532e4c9986SGeorge Wilson 26542e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 26552e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 26562e4c9986SGeorge Wilson 265788ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 265888ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 265988ecc943SGeorge Wilson 26600713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 266188ecc943SGeorge Wilson continue; 266288ecc943SGeorge Wilson 26630713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 26642e4c9986SGeorge Wilson /* 26652e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 26662e4c9986SGeorge Wilson * was removed then the histogram accounting may 26672e4c9986SGeorge Wilson * not be accurate. Update the histogram information 26682e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 26692e4c9986SGeorge Wilson * and metaslab class are up-to-date. 26702e4c9986SGeorge Wilson */ 26712e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 26722e4c9986SGeorge Wilson 26730713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 26740713e232SGeorge Wilson space_map_close(msp->ms_sm); 26750713e232SGeorge Wilson msp->ms_sm = NULL; 26760713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 267788ecc943SGeorge Wilson } 26782e4c9986SGeorge Wilson 26792e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 26802e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 26812e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 26822e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 268388ecc943SGeorge Wilson } 268488ecc943SGeorge Wilson 26855cabbc6bSPrashanth Sreenivasa tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 26865cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vd, tx); 2687215198a6SJoe Stein 2688215198a6SJoe Stein if (vd->vdev_islog && vd->vdev_top_zap != 0) { 2689215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 2690215198a6SJoe Stein vd->vdev_top_zap = 0; 2691215198a6SJoe Stein } 269288ecc943SGeorge Wilson dmu_tx_commit(tx); 269388ecc943SGeorge Wilson } 269488ecc943SGeorge Wilson 2695fa9e4066Sahrens void 2696fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2697fa9e4066Sahrens { 2698fa9e4066Sahrens metaslab_t *msp; 269980eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2700fa9e4066Sahrens 27015cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 270288ecc943SGeorge Wilson 2703fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2704fa9e4066Sahrens metaslab_sync_done(msp, txg); 270580eb36f2SGeorge Wilson 270680eb36f2SGeorge Wilson if (reassess) 270780eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2708fa9e4066Sahrens } 2709fa9e4066Sahrens 2710fa9e4066Sahrens void 2711fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2712fa9e4066Sahrens { 2713fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2714fa9e4066Sahrens vdev_t *lvd; 2715fa9e4066Sahrens metaslab_t *msp; 2716ecc2d604Sbonwick dmu_tx_t *tx; 2717fa9e4066Sahrens 27185cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 27195cabbc6bSPrashanth Sreenivasa dmu_tx_t *tx; 27205cabbc6bSPrashanth Sreenivasa 27215cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 27225cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 272388ecc943SGeorge Wilson 27245cabbc6bSPrashanth Sreenivasa tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 27255cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 27265cabbc6bSPrashanth Sreenivasa dmu_tx_commit(tx); 27275cabbc6bSPrashanth Sreenivasa 27285cabbc6bSPrashanth Sreenivasa /* 27295cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 27305cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 27315cabbc6bSPrashanth Sreenivasa */ 27325cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 27335cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 27345cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 27355cabbc6bSPrashanth Sreenivasa return; 27365cabbc6bSPrashanth Sreenivasa } 27375cabbc6bSPrashanth Sreenivasa } 27385cabbc6bSPrashanth Sreenivasa 27395cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 27405cabbc6bSPrashanth Sreenivasa 27415cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 27425cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 2743ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 27445cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 2745ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2746ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2747ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2748ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2749ecc2d604Sbonwick vdev_config_dirty(vd); 2750ecc2d604Sbonwick dmu_tx_commit(tx); 2751ecc2d604Sbonwick } 2752fa9e4066Sahrens 2753ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2754fa9e4066Sahrens metaslab_sync(msp, txg); 2755ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2756ecc2d604Sbonwick } 2757fa9e4066Sahrens 2758fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2759fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2760fa9e4066Sahrens 27615cabbc6bSPrashanth Sreenivasa /* 27625cabbc6bSPrashanth Sreenivasa * Remove the metadata associated with this vdev once it's empty. 27635cabbc6bSPrashanth Sreenivasa * Note that this is typically used for log/cache device removal; 27645cabbc6bSPrashanth Sreenivasa * we don't empty toplevel vdevs when removing them. But if 27655cabbc6bSPrashanth Sreenivasa * a toplevel happens to be emptied, this is not harmful. 27665cabbc6bSPrashanth Sreenivasa */ 27675cabbc6bSPrashanth Sreenivasa if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) { 27685cabbc6bSPrashanth Sreenivasa vdev_remove_empty(vd, txg); 27695cabbc6bSPrashanth Sreenivasa } 27705cabbc6bSPrashanth Sreenivasa 2771fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2772fa9e4066Sahrens } 2773fa9e4066Sahrens 2774fa9e4066Sahrens uint64_t 2775fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2776fa9e4066Sahrens { 2777fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2778fa9e4066Sahrens } 2779fa9e4066Sahrens 27803d7072f8Seschrock /* 27813d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 27823d7072f8Seschrock * not be opened, and no I/O is attempted. 27833d7072f8Seschrock */ 2784fa9e4066Sahrens int 2785069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2786fa9e4066Sahrens { 27874b964adaSGeorge Wilson vdev_t *vd, *tvd; 2788fa9e4066Sahrens 27898f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2790fa9e4066Sahrens 2791c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2792e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2793e14bb325SJeff Bonwick 27943d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2795e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2796fa9e4066Sahrens 27974b964adaSGeorge Wilson tvd = vd->vdev_top; 27984b964adaSGeorge Wilson 2799069f55e2SEric Schrock /* 2800069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2801069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2802069f55e2SEric Schrock * were faulted. 2803069f55e2SEric Schrock */ 2804069f55e2SEric Schrock vd->vdev_label_aux = aux; 2805069f55e2SEric Schrock 28063d7072f8Seschrock /* 28073d7072f8Seschrock * Faulted state takes precedence over degraded. 28083d7072f8Seschrock */ 280998d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 28103d7072f8Seschrock vd->vdev_faulted = 1ULL; 28113d7072f8Seschrock vd->vdev_degraded = 0ULL; 2812069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 28133d7072f8Seschrock 28143d7072f8Seschrock /* 2815c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2816c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 28173d7072f8Seschrock */ 28184b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 28193d7072f8Seschrock vd->vdev_degraded = 1ULL; 28203d7072f8Seschrock vd->vdev_faulted = 0ULL; 28213d7072f8Seschrock 28223d7072f8Seschrock /* 28233d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 28243d7072f8Seschrock * mark it degraded. 28253d7072f8Seschrock */ 28264b964adaSGeorge Wilson vdev_reopen(tvd); 28273d7072f8Seschrock 2828069f55e2SEric Schrock if (vdev_readable(vd)) 2829069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 28303d7072f8Seschrock } 28313d7072f8Seschrock 2832e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 28333d7072f8Seschrock } 28343d7072f8Seschrock 28353d7072f8Seschrock /* 28363d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 28373d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 28383d7072f8Seschrock * as I/O is concerned. 28393d7072f8Seschrock */ 28403d7072f8Seschrock int 2841069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 28423d7072f8Seschrock { 2843c5904d13Seschrock vdev_t *vd; 28440a4e9518Sgw 28458f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 28463d7072f8Seschrock 2847c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2848e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2849e14bb325SJeff Bonwick 28500e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2851e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 28520e34b6a7Sbonwick 28533d7072f8Seschrock /* 28543d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 28553d7072f8Seschrock */ 2856e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2857e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 28583d7072f8Seschrock 28593d7072f8Seschrock vd->vdev_degraded = 1ULL; 28603d7072f8Seschrock if (!vdev_is_dead(vd)) 28613d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2862069f55e2SEric Schrock aux); 28633d7072f8Seschrock 2864e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 28653d7072f8Seschrock } 28663d7072f8Seschrock 28673d7072f8Seschrock /* 2868f7170741SWill Andrews * Online the given vdev. 2869f7170741SWill Andrews * 2870f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2871f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2872f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2873f7170741SWill Andrews * events are generated if the device fails to open. 28743d7072f8Seschrock */ 28753d7072f8Seschrock int 2876e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 28773d7072f8Seschrock { 2878573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 28795f368aefSYuri Pankov boolean_t wasoffline; 28805f368aefSYuri Pankov vdev_state_t oldstate; 28813d7072f8Seschrock 28828f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 28833d7072f8Seschrock 2884c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2885e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 28863d7072f8Seschrock 28873d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2888e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2889fa9e4066Sahrens 28905f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 28915f368aefSYuri Pankov oldstate = vd->vdev_state; 289214372834SHans Rosenfeld 2893573ca77eSGeorge Wilson tvd = vd->vdev_top; 2894fa9e4066Sahrens vd->vdev_offline = B_FALSE; 2895441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 2896e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2897e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2898573ca77eSGeorge Wilson 2899573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2900573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2901573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2902573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2903573ca77eSGeorge Wilson } 2904573ca77eSGeorge Wilson 2905573ca77eSGeorge Wilson vdev_reopen(tvd); 29063d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 29073d7072f8Seschrock 2908573ca77eSGeorge Wilson if (!vd->vdev_aux) { 2909573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2910573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 2911573ca77eSGeorge Wilson } 2912573ca77eSGeorge Wilson 29133d7072f8Seschrock if (newstate) 29143d7072f8Seschrock *newstate = vd->vdev_state; 29153d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 29163d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 29173d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 29183d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 29193d7072f8Seschrock vd->vdev_unspare = B_TRUE; 2920fa9e4066Sahrens 2921573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2922573ca77eSGeorge Wilson 2923573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 2924573ca77eSGeorge Wilson if (vd->vdev_aux) 2925573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2926573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2927573ca77eSGeorge Wilson } 292814372834SHans Rosenfeld 29295f368aefSYuri Pankov if (wasoffline || 29305f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 29315f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 2932ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 293314372834SHans Rosenfeld 29348ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 2935fa9e4066Sahrens } 2936fa9e4066Sahrens 2937a1521560SJeff Bonwick static int 2938a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2939fa9e4066Sahrens { 2940e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 29418f18d1faSGeorge Wilson int error = 0; 29428f18d1faSGeorge Wilson uint64_t generation; 29438f18d1faSGeorge Wilson metaslab_group_t *mg; 29440a4e9518Sgw 29458f18d1faSGeorge Wilson top: 29468f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 2947fa9e4066Sahrens 2948c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2949e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2950fa9e4066Sahrens 29510e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2952e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 29530e34b6a7Sbonwick 2954e6ca193dSGeorge Wilson tvd = vd->vdev_top; 29558f18d1faSGeorge Wilson mg = tvd->vdev_mg; 29568f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 2957e6ca193dSGeorge Wilson 2958fa9e4066Sahrens /* 2959ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 2960fa9e4066Sahrens */ 2961ecc2d604Sbonwick if (!vd->vdev_offline) { 2962ecc2d604Sbonwick /* 29638ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 2964e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 2965e6ca193dSGeorge Wilson * expendable. 2966ecc2d604Sbonwick */ 2967e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2968e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 2969e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2970fa9e4066Sahrens 29718f18d1faSGeorge Wilson /* 2972b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 2973b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 2974b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 2975b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 29768f18d1faSGeorge Wilson */ 29778f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 29788f18d1faSGeorge Wilson /* 29798f18d1faSGeorge Wilson * Prevent any future allocations. 29808f18d1faSGeorge Wilson */ 2981a1521560SJeff Bonwick metaslab_group_passivate(mg); 29828f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 29838f18d1faSGeorge Wilson 29845cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 29858f18d1faSGeorge Wilson 29868f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 29878f18d1faSGeorge Wilson 29888f18d1faSGeorge Wilson /* 29898f18d1faSGeorge Wilson * Check to see if the config has changed. 29908f18d1faSGeorge Wilson */ 29918f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 2992a1521560SJeff Bonwick metaslab_group_activate(mg); 29938f18d1faSGeorge Wilson if (error) 29948f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 29958f18d1faSGeorge Wilson vd, error)); 29968f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 29978f18d1faSGeorge Wilson goto top; 29988f18d1faSGeorge Wilson } 2999fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 30008f18d1faSGeorge Wilson } 30018f18d1faSGeorge Wilson 3002ecc2d604Sbonwick /* 3003ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3004e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3005e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3006e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3007ecc2d604Sbonwick */ 3008ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3009e6ca193dSGeorge Wilson vdev_reopen(tvd); 3010e6ca193dSGeorge Wilson 3011e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3012e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3013ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3014e6ca193dSGeorge Wilson vdev_reopen(tvd); 3015e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3016ecc2d604Sbonwick } 30178f18d1faSGeorge Wilson 30188f18d1faSGeorge Wilson /* 30198f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 30208f18d1faSGeorge Wilson * once we online the device it's open for business. 30218f18d1faSGeorge Wilson */ 30228f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3023a1521560SJeff Bonwick metaslab_group_activate(mg); 3024fa9e4066Sahrens } 3025fa9e4066Sahrens 3026e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3027ecc2d604Sbonwick 30288f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3029fa9e4066Sahrens } 3030fa9e4066Sahrens 3031a1521560SJeff Bonwick int 3032a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3033a1521560SJeff Bonwick { 3034a1521560SJeff Bonwick int error; 3035a1521560SJeff Bonwick 3036a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3037a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3038a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3039a1521560SJeff Bonwick 3040a1521560SJeff Bonwick return (error); 3041a1521560SJeff Bonwick } 3042a1521560SJeff Bonwick 3043ea8dc4b6Seschrock /* 3044ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3045ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3046ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3047ea8dc4b6Seschrock */ 3048ea8dc4b6Seschrock void 3049e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3050fa9e4066Sahrens { 3051e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3052e14bb325SJeff Bonwick 3053e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3054fa9e4066Sahrens 3055ea8dc4b6Seschrock if (vd == NULL) 3056e14bb325SJeff Bonwick vd = rvd; 3057fa9e4066Sahrens 3058ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3059ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3060ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3061fa9e4066Sahrens 3062e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3063e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 30643d7072f8Seschrock 30655cabbc6bSPrashanth Sreenivasa /* 30665cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 30675cabbc6bSPrashanth Sreenivasa */ 30685cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 30695cabbc6bSPrashanth Sreenivasa return; 30705cabbc6bSPrashanth Sreenivasa 30713d7072f8Seschrock /* 30728a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 30738a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 30748a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 30758a79c1b5Sek * written out to disk. 30763d7072f8Seschrock */ 3077e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3078e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 30798a79c1b5Sek 3080096d22d4SEric Schrock /* 3081096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3082096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3083096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3084096d22d4SEric Schrock */ 3085096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3086096d22d4SEric Schrock 30874b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3088e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3089e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3090e14bb325SJeff Bonwick 3091f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 30923d7072f8Seschrock 3093096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3094096d22d4SEric Schrock 3095f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3096e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3097e14bb325SJeff Bonwick 3098e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 3099bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 31003d7072f8Seschrock 3101ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 31023d7072f8Seschrock } 3103096d22d4SEric Schrock 3104096d22d4SEric Schrock /* 3105096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3106096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3107096d22d4SEric Schrock * done in response to the FMA fault. 3108096d22d4SEric Schrock */ 3109096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3110096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3111096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3112096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3113fa9e4066Sahrens } 3114fa9e4066Sahrens 3115e14bb325SJeff Bonwick boolean_t 3116e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 31170a4e9518Sgw { 311888ecc943SGeorge Wilson /* 311988ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 312088ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 312188ecc943SGeorge Wilson * these types of devices in the various code paths. 312288ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 312388ecc943SGeorge Wilson * before issuing I/O to them. 312488ecc943SGeorge Wilson */ 31255cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 31265cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 312788ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 31280a4e9518Sgw } 31290a4e9518Sgw 3130e14bb325SJeff Bonwick boolean_t 3131e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 31320a4e9518Sgw { 3133e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 31340a4e9518Sgw } 31350a4e9518Sgw 3136e14bb325SJeff Bonwick boolean_t 3137e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3138fa9e4066Sahrens { 31395cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 31405cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3141fa9e4066Sahrens } 3142fa9e4066Sahrens 3143a31e6787SGeorge Wilson boolean_t 3144a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3145a31e6787SGeorge Wilson { 31468ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 31478ad4d6ddSJeff Bonwick 3148a31e6787SGeorge Wilson /* 31498ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3150a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3151a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 31528ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 31538ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 31548ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3155a31e6787SGeorge Wilson */ 31568ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 31575cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 31580f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3159a31e6787SGeorge Wilson } 3160a31e6787SGeorge Wilson 3161e14bb325SJeff Bonwick boolean_t 3162e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3163fa9e4066Sahrens { 3164e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3165fa9e4066Sahrens 3166e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3167e14bb325SJeff Bonwick return (B_FALSE); 3168fa9e4066Sahrens 3169e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3170e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3171fa9e4066Sahrens 3172e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3173e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3174fa9e4066Sahrens 3175e14bb325SJeff Bonwick return (B_TRUE); 3176fa9e4066Sahrens } 3177fa9e4066Sahrens 3178fa9e4066Sahrens /* 3179fa9e4066Sahrens * Get statistics for the given vdev. 3180fa9e4066Sahrens */ 3181fa9e4066Sahrens void 3182fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3183fa9e4066Sahrens { 31842e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 31852e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 3186c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 31872e4c9986SGeorge Wilson 31882e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3189fa9e4066Sahrens 3190fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 3191fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3192fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3193fa9e4066Sahrens vs->vs_state = vd->vdev_state; 3194573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 3195573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 3196573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3197c39a2aaeSGeorge Wilson /* 3198c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 3199c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 3200c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 3201c39a2aaeSGeorge Wilson */ 3202c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 32037855d95bSToomas Soome vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 32047855d95bSToomas Soome spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3205c39a2aaeSGeorge Wilson } 32065cabbc6bSPrashanth Sreenivasa if (vd->vdev_aux == NULL && vd == vd->vdev_top && 32075cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 32082e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 32092986efa8SAlex Reece } 3210fa9e4066Sahrens 3211fa9e4066Sahrens /* 3212fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 3213fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 3214fa9e4066Sahrens */ 3215fa9e4066Sahrens if (vd == rvd) { 3216e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 3217fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 3218fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 3219fa9e4066Sahrens 3220e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 3221fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 3222fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 3223fa9e4066Sahrens } 32243f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 3225fa9e4066Sahrens } 3226fa9e4066Sahrens } 32272e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3228fa9e4066Sahrens } 3229fa9e4066Sahrens 3230fa94a07fSbrendan void 3231fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3232fa94a07fSbrendan { 3233fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3234fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3235fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3236fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3237fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3238fa94a07fSbrendan } 3239fa94a07fSbrendan 32403f9d6ad7SLin Ling void 32413f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 32423f9d6ad7SLin Ling { 32433f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 32443f9d6ad7SLin Ling 32453f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 32463f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 32473f9d6ad7SLin Ling 32483f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 32493f9d6ad7SLin Ling vs->vs_scan_processed = 0; 32503f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 32513f9d6ad7SLin Ling } 32523f9d6ad7SLin Ling 3253fa9e4066Sahrens void 3254e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3255fa9e4066Sahrens { 32568ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 32578ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3258e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3259fa9e4066Sahrens vdev_t *pvd; 3260fa9e4066Sahrens uint64_t txg = zio->io_txg; 3261fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3262fa9e4066Sahrens zio_type_t type = zio->io_type; 3263fa9e4066Sahrens int flags = zio->io_flags; 3264fa9e4066Sahrens 3265e14bb325SJeff Bonwick /* 3266e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3267e14bb325SJeff Bonwick */ 3268e14bb325SJeff Bonwick if (zio->io_gang_tree) 3269e14bb325SJeff Bonwick return; 3270e14bb325SJeff Bonwick 3271fa9e4066Sahrens if (zio->io_error == 0) { 3272e14bb325SJeff Bonwick /* 3273e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3274e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3275e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3276e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3277e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3278e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3279e14bb325SJeff Bonwick * remain zero, which is what we want.) 3280e14bb325SJeff Bonwick * 3281e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3282e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3283e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3284e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3285e14bb325SJeff Bonwick */ 3286e14bb325SJeff Bonwick if (vd == rvd) 3287e14bb325SJeff Bonwick return; 3288e14bb325SJeff Bonwick 3289e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 32908ad4d6ddSJeff Bonwick 32918ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 32928ad4d6ddSJeff Bonwick return; 32938ad4d6ddSJeff Bonwick 32948ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 32958ad4d6ddSJeff Bonwick 3296e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 329744ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 32983f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 32993f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 33003f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 33013f9d6ad7SLin Ling 33023f9d6ad7SLin Ling /* XXX cleanup? */ 33033f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 33043f9d6ad7SLin Ling atomic_add_64(processed, psize); 33053f9d6ad7SLin Ling vs->vs_scan_processed += psize; 33063f9d6ad7SLin Ling } 33073f9d6ad7SLin Ling 33088ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3309e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3310fa9e4066Sahrens } 33118ad4d6ddSJeff Bonwick 33128ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 33138ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 33148ad4d6ddSJeff Bonwick 33158ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3316fa9e4066Sahrens return; 3317fa9e4066Sahrens } 3318fa9e4066Sahrens 3319fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3320fa9e4066Sahrens return; 3321fa9e4066Sahrens 33228956713aSEric Schrock /* 33238956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 33248956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 33258956713aSEric Schrock * hard errors, when in reality they can happen for any number of 33268956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 33278956713aSEric Schrock */ 33288956713aSEric Schrock if (zio->io_error == EIO && 33298956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 33308956713aSEric Schrock return; 33318956713aSEric Schrock 33328f18d1faSGeorge Wilson /* 33338f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 33348f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 33358f18d1faSGeorge Wilson * errors. 33368f18d1faSGeorge Wilson */ 33378f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 33388f18d1faSGeorge Wilson return; 33398f18d1faSGeorge Wilson 3340e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3341b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3342e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3343e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3344e14bb325SJeff Bonwick else 3345e14bb325SJeff Bonwick vs->vs_read_errors++; 3346fa9e4066Sahrens } 3347b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3348e14bb325SJeff Bonwick vs->vs_write_errors++; 3349e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3350fa9e4066Sahrens 33515cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 33525cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 33538ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 335444ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3355b24ab676SJeff Bonwick spa->spa_claiming)) { 33568ad4d6ddSJeff Bonwick /* 3357b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 3358b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 3359b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 3360b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 3361b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 3362b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 3363b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 3364b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 33658ad4d6ddSJeff Bonwick * 33668ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 33678ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 33688ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 33698ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 33708ad4d6ddSJeff Bonwick */ 33718ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 33728ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 337344ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 33748ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 33758ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 33768ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3377b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 3378b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 3379b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3380b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 33818ad4d6ddSJeff Bonwick } 3382b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 33838ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3384fa9e4066Sahrens return; 33858ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 33868ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 33878ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3388fa9e4066Sahrens } 33898ad4d6ddSJeff Bonwick if (vd != rvd) 33908ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3391fa9e4066Sahrens } 3392fa9e4066Sahrens } 3393fa9e4066Sahrens 3394fa9e4066Sahrens /* 3395b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 3396b24ab676SJeff Bonwick * and the root vdev. 3397fa9e4066Sahrens */ 3398fa9e4066Sahrens void 3399b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3400b24ab676SJeff Bonwick int64_t space_delta) 3401fa9e4066Sahrens { 340299653d4eSeschrock int64_t dspace_delta = space_delta; 34038654d025Sperrin spa_t *spa = vd->vdev_spa; 34048654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 3405b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 3406b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 3407fa9e4066Sahrens 34088654d025Sperrin ASSERT(vd == vd->vdev_top); 340999653d4eSeschrock 34108654d025Sperrin /* 34118654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 34128654d025Sperrin * factor. We must calculate this here and not at the root vdev 34138654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 34148654d025Sperrin * childrens', thus not accurate enough for us. 34158654d025Sperrin */ 34168654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 3417e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 34188654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 34198654d025Sperrin vd->vdev_deflate_ratio; 34208654d025Sperrin 34218654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 34228654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 3423b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 34248654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 34258654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 34268654d025Sperrin 3427b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 3428fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 3429fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 3430b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 3431fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 3432fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 3433fa94a07fSbrendan } 3434b24ab676SJeff Bonwick 3435b24ab676SJeff Bonwick if (mc != NULL) { 3436b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 3437b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 3438b24ab676SJeff Bonwick 3439b24ab676SJeff Bonwick metaslab_class_space_update(mc, 3440b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 3441b24ab676SJeff Bonwick } 3442fa9e4066Sahrens } 3443fa9e4066Sahrens 3444fa9e4066Sahrens /* 3445fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 3446fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 3447fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3448fa9e4066Sahrens */ 3449fa9e4066Sahrens void 3450fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3451fa9e4066Sahrens { 3452fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3453fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3454fa9e4066Sahrens int c; 3455fa9e4066Sahrens 3456f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3457f9af39baSGeorge Wilson 3458c5904d13Seschrock /* 34596809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 34606809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3461c5904d13Seschrock */ 3462c5904d13Seschrock if (vd->vdev_aux != NULL) { 3463c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3464c5904d13Seschrock nvlist_t **aux; 3465c5904d13Seschrock uint_t naux; 3466c5904d13Seschrock 3467c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3468c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3469c5904d13Seschrock break; 3470c5904d13Seschrock } 3471c5904d13Seschrock 3472e14bb325SJeff Bonwick if (c == sav->sav_count) { 3473e14bb325SJeff Bonwick /* 3474e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3475e14bb325SJeff Bonwick */ 3476e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3477e14bb325SJeff Bonwick return; 3478e14bb325SJeff Bonwick } 3479e14bb325SJeff Bonwick 3480c5904d13Seschrock sav->sav_sync = B_TRUE; 3481c5904d13Seschrock 34826809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 34836809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 34846809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 34856809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 34866809eb4eSEric Schrock } 3487c5904d13Seschrock 3488c5904d13Seschrock ASSERT(c < naux); 3489c5904d13Seschrock 3490c5904d13Seschrock /* 3491c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3492c5904d13Seschrock * sketchy, but it will work. 3493c5904d13Seschrock */ 3494c5904d13Seschrock nvlist_free(aux[c]); 34953f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3496c5904d13Seschrock 3497c5904d13Seschrock return; 3498c5904d13Seschrock } 3499c5904d13Seschrock 35005dabedeeSbonwick /* 3501e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3502e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3503e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 35045dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 35055dabedeeSbonwick */ 3506e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3507e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3508e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 35095dabedeeSbonwick 3510fa9e4066Sahrens if (vd == rvd) { 3511fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 3512fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 3513fa9e4066Sahrens } else { 3514fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 3515fa9e4066Sahrens 351688ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 35175cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3518e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 35195cabbc6bSPrashanth Sreenivasa } 3520fa9e4066Sahrens } 3521fa9e4066Sahrens } 3522fa9e4066Sahrens 3523fa9e4066Sahrens void 3524fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 3525fa9e4066Sahrens { 35265dabedeeSbonwick spa_t *spa = vd->vdev_spa; 35275dabedeeSbonwick 3528e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3529e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3530e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 35315dabedeeSbonwick 3532e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3533e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 3534e14bb325SJeff Bonwick } 3535e14bb325SJeff Bonwick 3536e14bb325SJeff Bonwick /* 3537e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 3538e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3539e14bb325SJeff Bonwick * the state changes from larger config changes because they require 3540e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3541e14bb325SJeff Bonwick */ 3542e14bb325SJeff Bonwick void 3543e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3544e14bb325SJeff Bonwick { 3545e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3546e14bb325SJeff Bonwick 3547f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3548e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3549e14bb325SJeff Bonwick 3550e14bb325SJeff Bonwick /* 3551e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3552e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3553e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3554e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3555e14bb325SJeff Bonwick */ 3556e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3557e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3558e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3559e14bb325SJeff Bonwick 35605cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 35615cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 3562e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3563e14bb325SJeff Bonwick } 3564e14bb325SJeff Bonwick 3565e14bb325SJeff Bonwick void 3566e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3567e14bb325SJeff Bonwick { 3568e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3569e14bb325SJeff Bonwick 3570e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3571e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3572e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3573e14bb325SJeff Bonwick 3574e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3575e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3576fa9e4066Sahrens } 3577fa9e4066Sahrens 357832b87932Sek /* 357932b87932Sek * Propagate vdev state up from children to parent. 358032b87932Sek */ 358144cd46caSbillm void 358244cd46caSbillm vdev_propagate_state(vdev_t *vd) 358344cd46caSbillm { 35848ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 35858ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 358644cd46caSbillm int degraded = 0, faulted = 0; 358744cd46caSbillm int corrupted = 0; 358844cd46caSbillm vdev_t *child; 358944cd46caSbillm 35903d7072f8Seschrock if (vd->vdev_children > 0) { 3591573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 35923d7072f8Seschrock child = vd->vdev_child[c]; 359351ece835Seschrock 359488ecc943SGeorge Wilson /* 35955cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 35965cabbc6bSPrashanth Sreenivasa * decision. 359788ecc943SGeorge Wilson */ 35985cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 359988ecc943SGeorge Wilson continue; 360088ecc943SGeorge Wilson 3601e14bb325SJeff Bonwick if (!vdev_readable(child) || 36028ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 360351ece835Seschrock /* 360451ece835Seschrock * Root special: if there is a top-level log 360551ece835Seschrock * device, treat the root vdev as if it were 360651ece835Seschrock * degraded. 360751ece835Seschrock */ 360851ece835Seschrock if (child->vdev_islog && vd == rvd) 360951ece835Seschrock degraded++; 361051ece835Seschrock else 361151ece835Seschrock faulted++; 361251ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 36133d7072f8Seschrock degraded++; 361451ece835Seschrock } 361544cd46caSbillm 36163d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 36173d7072f8Seschrock corrupted++; 36183d7072f8Seschrock } 361944cd46caSbillm 36203d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 36213d7072f8Seschrock 36223d7072f8Seschrock /* 3623e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 36243d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 36253d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 36263d7072f8Seschrock * replicas'. 36273d7072f8Seschrock */ 36283d7072f8Seschrock if (corrupted && vd == rvd && 36293d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 36303d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 36313d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 36323d7072f8Seschrock } 36333d7072f8Seschrock 363451ece835Seschrock if (vd->vdev_parent) 36353d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 363644cd46caSbillm } 363744cd46caSbillm 3638fa9e4066Sahrens /* 3639ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3640ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3641ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3642ea8dc4b6Seschrock * 3643ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3644ea8dc4b6Seschrock * generated. 3645fa9e4066Sahrens */ 3646fa9e4066Sahrens void 3647ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3648fa9e4066Sahrens { 3649560e6e96Seschrock uint64_t save_state; 3650c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3651ea8dc4b6Seschrock 3652ea8dc4b6Seschrock if (state == vd->vdev_state) { 3653ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3654fa9e4066Sahrens return; 3655ea8dc4b6Seschrock } 3656ea8dc4b6Seschrock 3657560e6e96Seschrock save_state = vd->vdev_state; 3658fa9e4066Sahrens 3659fa9e4066Sahrens vd->vdev_state = state; 3660fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3661fa9e4066Sahrens 36623d7072f8Seschrock /* 36633d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 366498d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 366598d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 366698d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 366798d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 366898d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 366998d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 367098d1cbfeSGeorge Wilson * children. 367198d1cbfeSGeorge Wilson */ 367298d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 367398d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 36743d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 36753d7072f8Seschrock 3676069f55e2SEric Schrock /* 3677069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3678069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3679069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3680069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3681069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3682069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3683069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3684069f55e2SEric Schrock */ 3685069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3686069f55e2SEric Schrock vd->vdev_prevstate != state) 3687069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3688069f55e2SEric Schrock 36893d7072f8Seschrock if (vd->vdev_removed && 36903d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 36913d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 36923d7072f8Seschrock /* 36933d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 36943d7072f8Seschrock * device was previously marked removed and someone attempted to 36953d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 36963d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 36973d7072f8Seschrock * it is one of our special test online cases, which is only 36983d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 36993d7072f8Seschrock * fault. 37003d7072f8Seschrock */ 37013d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 37023d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 37033d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 37043d7072f8Seschrock vd->vdev_removed = B_TRUE; 37053d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3706ea8dc4b6Seschrock /* 3707cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3708cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3709cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3710cb04b873SMark J Musante * is not considered an error. 3711ea8dc4b6Seschrock */ 3712cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3713cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3714560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3715560e6e96Seschrock vd->vdev_not_present = 1; 3716560e6e96Seschrock 3717560e6e96Seschrock /* 3718560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3719560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3720560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3721560e6e96Seschrock * want to post the ereport if the device was already in the 3722560e6e96Seschrock * CANT_OPEN state beforehand. 37233d7072f8Seschrock * 37243d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 37253d7072f8Seschrock * online the device in response to an insertion event. If we 37263d7072f8Seschrock * hit this case, then we have detected an insertion event for a 37273d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 37283d7072f8Seschrock * In this scenario, we don't post an ereport because we are 37293d7072f8Seschrock * about to replace the device, or attempt an online with 37303d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3731560e6e96Seschrock */ 37323d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 37333d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3734c5904d13Seschrock vd != spa->spa_root_vdev) { 3735ea8dc4b6Seschrock const char *class; 3736ea8dc4b6Seschrock 3737ea8dc4b6Seschrock switch (aux) { 3738ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3739ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3740ea8dc4b6Seschrock break; 3741ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3742ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3743ea8dc4b6Seschrock break; 3744ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3745ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3746ea8dc4b6Seschrock break; 3747ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3748ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3749ea8dc4b6Seschrock break; 3750ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3751ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3752ea8dc4b6Seschrock break; 3753ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3754ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3755ea8dc4b6Seschrock break; 3756ea8dc4b6Seschrock default: 3757ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3758ea8dc4b6Seschrock } 3759ea8dc4b6Seschrock 3760c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3761ea8dc4b6Seschrock } 3762ea8dc4b6Seschrock 37633d7072f8Seschrock /* Erase any notion of persistent removed state */ 37643d7072f8Seschrock vd->vdev_removed = B_FALSE; 37653d7072f8Seschrock } else { 37663d7072f8Seschrock vd->vdev_removed = B_FALSE; 37673d7072f8Seschrock } 3768ea8dc4b6Seschrock 37698b33d774STim Haley if (!isopen && vd->vdev_parent) 37708b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3771fa9e4066Sahrens } 377215e6edf1Sgw 3773*6f793812SPavel Zakharov boolean_t 3774*6f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 3775*6f793812SPavel Zakharov { 3776*6f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 3777*6f793812SPavel Zakharov 3778*6f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 3779*6f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 3780*6f793812SPavel Zakharov return (B_FALSE); 3781*6f793812SPavel Zakharov } 3782*6f793812SPavel Zakharov 3783*6f793812SPavel Zakharov return (B_TRUE); 3784*6f793812SPavel Zakharov } 3785*6f793812SPavel Zakharov 378615e6edf1Sgw /* 378715e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 3788c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 3789c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 379015e6edf1Sgw */ 379115e6edf1Sgw boolean_t 379215e6edf1Sgw vdev_is_bootable(vdev_t *vd) 379315e6edf1Sgw { 379415e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 379515e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 379615e6edf1Sgw 379715e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 379815e6edf1Sgw vd->vdev_children > 1) { 379915e6edf1Sgw return (B_FALSE); 38005cabbc6bSPrashanth Sreenivasa } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || 38015cabbc6bSPrashanth Sreenivasa strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { 380215e6edf1Sgw return (B_FALSE); 380315e6edf1Sgw } 380415e6edf1Sgw } 380515e6edf1Sgw 3806573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 380715e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 380815e6edf1Sgw return (B_FALSE); 380915e6edf1Sgw } 381015e6edf1Sgw return (B_TRUE); 381115e6edf1Sgw } 3812e6ca193dSGeorge Wilson 38135cabbc6bSPrashanth Sreenivasa boolean_t 38145cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 38155cabbc6bSPrashanth Sreenivasa { 38165cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 38175cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 38185cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 38195cabbc6bSPrashanth Sreenivasa return (B_FALSE); 38205cabbc6bSPrashanth Sreenivasa } else { 38215cabbc6bSPrashanth Sreenivasa return (B_TRUE); 38225cabbc6bSPrashanth Sreenivasa } 38235cabbc6bSPrashanth Sreenivasa } 38245cabbc6bSPrashanth Sreenivasa 38254b964adaSGeorge Wilson /* 38264b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 38274b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 38284b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 38294b964adaSGeorge Wilson */ 38304b964adaSGeorge Wilson boolean_t 38314b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 38324b964adaSGeorge Wilson { 38334b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 38344b964adaSGeorge Wilson !vd->vdev_removed) 38354b964adaSGeorge Wilson return (B_TRUE); 38364b964adaSGeorge Wilson 38374b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 38384b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 38394b964adaSGeorge Wilson return (B_TRUE); 38404b964adaSGeorge Wilson 38414b964adaSGeorge Wilson return (B_FALSE); 38424b964adaSGeorge Wilson } 38434b964adaSGeorge Wilson 3844573ca77eSGeorge Wilson /* 3845573ca77eSGeorge Wilson * Expand a vdev if possible. 3846573ca77eSGeorge Wilson */ 3847573ca77eSGeorge Wilson void 3848573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3849573ca77eSGeorge Wilson { 3850573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3851573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3852573ca77eSGeorge Wilson 38535cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 38545cabbc6bSPrashanth Sreenivasa 38555cabbc6bSPrashanth Sreenivasa if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 38565cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3857573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3858573ca77eSGeorge Wilson vdev_config_dirty(vd); 3859573ca77eSGeorge Wilson } 3860573ca77eSGeorge Wilson } 38611195e687SMark J Musante 38621195e687SMark J Musante /* 38631195e687SMark J Musante * Split a vdev. 38641195e687SMark J Musante */ 38651195e687SMark J Musante void 38661195e687SMark J Musante vdev_split(vdev_t *vd) 38671195e687SMark J Musante { 38681195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 38691195e687SMark J Musante 38701195e687SMark J Musante vdev_remove_child(pvd, vd); 38711195e687SMark J Musante vdev_compact_children(pvd); 38721195e687SMark J Musante 38731195e687SMark J Musante cvd = pvd->vdev_child[0]; 38741195e687SMark J Musante if (pvd->vdev_children == 1) { 38751195e687SMark J Musante vdev_remove_parent(cvd); 38761195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 38771195e687SMark J Musante } 38781195e687SMark J Musante vdev_propagate_state(cvd); 38791195e687SMark J Musante } 3880283b8460SGeorge.Wilson 3881283b8460SGeorge.Wilson void 3882283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 3883283b8460SGeorge.Wilson { 3884283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 3885283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 3886283b8460SGeorge.Wilson 3887283b8460SGeorge.Wilson vdev_deadman(cvd); 3888283b8460SGeorge.Wilson } 3889283b8460SGeorge.Wilson 3890283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 3891283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 3892283b8460SGeorge.Wilson 3893283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 389469962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 3895283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 3896283b8460SGeorge.Wilson zio_t *fio; 3897283b8460SGeorge.Wilson uint64_t delta; 3898283b8460SGeorge.Wilson 3899283b8460SGeorge.Wilson /* 3900283b8460SGeorge.Wilson * Look at the head of all the pending queues, 3901283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 3902283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 3903283b8460SGeorge.Wilson */ 390469962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 3905c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 3906c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 39073ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 39083ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 39093ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 3910283b8460SGeorge.Wilson vq->vq_io_complete_ts); 3911283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 3912283b8460SGeorge.Wilson "hung.", spa_name(spa)); 3913283b8460SGeorge.Wilson } 3914283b8460SGeorge.Wilson } 3915283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 3916283b8460SGeorge.Wilson } 3917283b8460SGeorge.Wilson } 3918