1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/spa.h> 29 #include <sys/spa_impl.h> 30 #include <sys/vdev.h> 31 #include <sys/vdev_impl.h> 32 #include <sys/zio.h> 33 34 #include <sys/fm/fs/zfs.h> 35 #include <sys/fm/protocol.h> 36 #include <sys/fm/util.h> 37 #include <sys/sysevent.h> 38 39 /* 40 * This general routine is responsible for generating all the different ZFS 41 * ereports. The payload is dependent on the class, and which arguments are 42 * supplied to the function: 43 * 44 * EREPORT POOL VDEV IO 45 * block X X X 46 * data X X 47 * device X X 48 * pool X 49 * 50 * If we are in a loading state, all errors are chained together by the same 51 * SPA-wide ENA (Error Numeric Association). 52 * 53 * For isolated I/O requests, we get the ENA from the zio_t. The propagation 54 * gets very complicated due to RAID-Z, gang blocks, and vdev caching. We want 55 * to chain together all ereports associated with a logical piece of data. For 56 * read I/Os, there are basically three 'types' of I/O, which form a roughly 57 * layered diagram: 58 * 59 * +---------------+ 60 * | Aggregate I/O | No associated logical data or device 61 * +---------------+ 62 * | 63 * V 64 * +---------------+ Reads associated with a piece of logical data. 65 * | Read I/O | This includes reads on behalf of RAID-Z, 66 * +---------------+ mirrors, gang blocks, retries, etc. 67 * | 68 * V 69 * +---------------+ Reads associated with a particular device, but 70 * | Physical I/O | no logical data. Issued as part of vdev caching 71 * +---------------+ and I/O aggregation. 72 * 73 * Note that 'physical I/O' here is not the same terminology as used in the rest 74 * of ZIO. Typically, 'physical I/O' simply means that there is no attached 75 * blockpointer. But I/O with no associated block pointer can still be related 76 * to a logical piece of data (i.e. RAID-Z requests). 77 * 78 * Purely physical I/O always have unique ENAs. They are not related to a 79 * particular piece of logical data, and therefore cannot be chained together. 80 * We still generate an ereport, but the DE doesn't correlate it with any 81 * logical piece of data. When such an I/O fails, the delegated I/O requests 82 * will issue a retry, which will trigger the 'real' ereport with the correct 83 * ENA. 84 * 85 * We keep track of the ENA for a ZIO chain through the 'io_logical' member. 86 * When a new logical I/O is issued, we set this to point to itself. Child I/Os 87 * then inherit this pointer, so that when it is first set subsequent failures 88 * will use the same ENA. If a physical I/O is issued (by passing the 89 * ZIO_FLAG_NOBOOKMARK flag), then this pointer is reset, guaranteeing that a 90 * unique ENA will be generated. For an aggregate I/O, this pointer is set to 91 * NULL, and no ereport will be generated (since it doesn't actually correspond 92 * to any particular device or piece of data). 93 */ 94 void 95 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio, 96 uint64_t stateoroffset, uint64_t size) 97 { 98 #ifdef _KERNEL 99 nvlist_t *ereport, *detector; 100 uint64_t ena; 101 char class[64]; 102 int state; 103 104 /* 105 * If we are doing a spa_tryimport(), ignore errors. 106 */ 107 if (spa->spa_load_state == SPA_LOAD_TRYIMPORT) 108 return; 109 110 /* 111 * If we are in the middle of opening a pool, and the previous attempt 112 * failed, don't bother logging any new ereports - we're just going to 113 * get the same diagnosis anyway. 114 */ 115 if (spa->spa_load_state != SPA_LOAD_NONE && 116 spa->spa_last_open_failed) 117 return; 118 119 if (zio != NULL) { 120 /* 121 * Ignore any errors from I/Os that we are going to retry 122 * anyway - we only generate errors from the final failure. 123 * Checksum errors are generated after the pipeline stage 124 * responsible for retrying the I/O (VDEV_IO_ASSESS), so this 125 * only applies to standard I/O errors. 126 */ 127 if (zio_should_retry(zio) && zio->io_error != ECKSUM) 128 return; 129 130 /* 131 * If this is not a read or write zio, ignore the error. This 132 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails. 133 */ 134 if (zio->io_type != ZIO_TYPE_READ && 135 zio->io_type != ZIO_TYPE_WRITE) 136 return; 137 138 /* 139 * Ignore any errors from speculative I/Os, as failure is an 140 * expected result. 141 */ 142 if (zio->io_flags & ZIO_FLAG_SPECULATIVE) 143 return; 144 } 145 146 if ((ereport = fm_nvlist_create(NULL)) == NULL) 147 return; 148 149 if ((detector = fm_nvlist_create(NULL)) == NULL) { 150 fm_nvlist_destroy(ereport, FM_NVA_FREE); 151 return; 152 } 153 154 /* 155 * Serialize ereport generation 156 */ 157 mutex_enter(&spa->spa_errlist_lock); 158 159 /* 160 * Determine the ENA to use for this event. If we are in a loading 161 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 162 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 163 */ 164 if (spa->spa_load_state != SPA_LOAD_NONE) { 165 if (spa->spa_ena == 0) 166 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 167 ena = spa->spa_ena; 168 } else if (zio != NULL && zio->io_logical != NULL) { 169 if (zio->io_logical->io_ena == 0) 170 zio->io_logical->io_ena = 171 fm_ena_generate(0, FM_ENA_FMT1); 172 ena = zio->io_logical->io_ena; 173 } else { 174 ena = fm_ena_generate(0, FM_ENA_FMT1); 175 } 176 177 /* 178 * Construct the full class, detector, and other standard FMA fields. 179 */ 180 (void) snprintf(class, sizeof (class), "%s.%s", 181 ZFS_ERROR_CLASS, subclass); 182 183 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa), 184 vd != NULL ? vd->vdev_guid : 0); 185 186 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL); 187 188 /* 189 * Construct the per-ereport payload, depending on which parameters are 190 * passed in. 191 */ 192 193 /* 194 * If we are importing a faulted pool, then we treat it like an open, 195 * not an import. Otherwise, the DE will ignore all faults during 196 * import, since the default behavior is to mark the devices as 197 * persistently unavailable, not leave them in the faulted state. 198 */ 199 state = spa->spa_import_faulted ? SPA_LOAD_OPEN : spa->spa_load_state; 200 201 /* 202 * Generic payload members common to all ereports. 203 * 204 * The direct reference to spa_name is used rather than spa_name() 205 * because of the asynchronous nature of the zio pipeline. spa_name() 206 * asserts that the config lock is held in some form. This is always 207 * the case in I/O context, but because the check for RW_WRITER compares 208 * against 'curthread', we may be in an asynchronous context and blow 209 * this assert. Rather than loosen this assert, we acknowledge that all 210 * contexts in which this function is called (pool open, I/O) are safe, 211 * and dereference the name directly. 212 */ 213 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL, 214 DATA_TYPE_STRING, spa->spa_name, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 215 DATA_TYPE_UINT64, spa_guid(spa), 216 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32, 217 state, NULL); 218 219 if (spa != NULL) { 220 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 221 DATA_TYPE_STRING, 222 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 223 FM_EREPORT_FAILMODE_WAIT : 224 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 225 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC, 226 NULL); 227 } 228 229 if (vd != NULL) { 230 vdev_t *pvd = vd->vdev_parent; 231 232 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 233 DATA_TYPE_UINT64, vd->vdev_guid, 234 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 235 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL); 236 if (vd->vdev_path) 237 fm_payload_set(ereport, 238 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, 239 DATA_TYPE_STRING, vd->vdev_path, NULL); 240 if (vd->vdev_devid) 241 fm_payload_set(ereport, 242 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, 243 DATA_TYPE_STRING, vd->vdev_devid, NULL); 244 245 if (pvd != NULL) { 246 fm_payload_set(ereport, 247 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, 248 DATA_TYPE_UINT64, pvd->vdev_guid, 249 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 250 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type, 251 NULL); 252 if (pvd->vdev_path) 253 fm_payload_set(ereport, 254 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 255 DATA_TYPE_STRING, pvd->vdev_path, NULL); 256 if (pvd->vdev_devid) 257 fm_payload_set(ereport, 258 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 259 DATA_TYPE_STRING, pvd->vdev_devid, NULL); 260 } 261 } 262 263 if (zio != NULL) { 264 /* 265 * Payload common to all I/Os. 266 */ 267 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 268 DATA_TYPE_INT32, zio->io_error, NULL); 269 270 /* 271 * If the 'size' parameter is non-zero, it indicates this is a 272 * RAID-Z or other I/O where the physical offset and length are 273 * provided for us, instead of within the zio_t. 274 */ 275 if (vd != NULL) { 276 if (size) 277 fm_payload_set(ereport, 278 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 279 DATA_TYPE_UINT64, stateoroffset, 280 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 281 DATA_TYPE_UINT64, size, NULL); 282 else 283 fm_payload_set(ereport, 284 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 285 DATA_TYPE_UINT64, zio->io_offset, 286 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 287 DATA_TYPE_UINT64, zio->io_size, NULL); 288 } 289 290 /* 291 * Payload for I/Os with corresponding logical information. 292 */ 293 if (zio->io_logical != NULL) 294 fm_payload_set(ereport, 295 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET, 296 DATA_TYPE_UINT64, 297 zio->io_logical->io_bookmark.zb_objset, 298 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 299 DATA_TYPE_UINT64, 300 zio->io_logical->io_bookmark.zb_object, 301 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 302 DATA_TYPE_INT64, 303 zio->io_logical->io_bookmark.zb_level, 304 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 305 DATA_TYPE_UINT64, 306 zio->io_logical->io_bookmark.zb_blkid, NULL); 307 } else if (vd != NULL) { 308 /* 309 * If we have a vdev but no zio, this is a device fault, and the 310 * 'stateoroffset' parameter indicates the previous state of the 311 * vdev. 312 */ 313 fm_payload_set(ereport, 314 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 315 DATA_TYPE_UINT64, stateoroffset, NULL); 316 } 317 mutex_exit(&spa->spa_errlist_lock); 318 319 fm_ereport_post(ereport, EVCH_SLEEP); 320 321 fm_nvlist_destroy(ereport, FM_NVA_FREE); 322 fm_nvlist_destroy(detector, FM_NVA_FREE); 323 #endif 324 } 325 326 static void 327 zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) 328 { 329 #ifdef _KERNEL 330 nvlist_t *resource; 331 char class[64]; 332 333 if ((resource = fm_nvlist_create(NULL)) == NULL) 334 return; 335 336 (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE, 337 ZFS_ERROR_CLASS, name); 338 VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0); 339 VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0); 340 VERIFY(nvlist_add_uint64(resource, 341 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0); 342 if (vd) 343 VERIFY(nvlist_add_uint64(resource, 344 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0); 345 346 fm_ereport_post(resource, EVCH_SLEEP); 347 348 fm_nvlist_destroy(resource, FM_NVA_FREE); 349 #endif 350 } 351 352 /* 353 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 354 * has been removed from the system. This will cause the DE to ignore any 355 * recent I/O errors, inferring that they are due to the asynchronous device 356 * removal. 357 */ 358 void 359 zfs_post_remove(spa_t *spa, vdev_t *vd) 360 { 361 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED); 362 } 363 364 /* 365 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 366 * has the 'autoreplace' property set, and therefore any broken vdevs will be 367 * handled by higher level logic, and no vdev fault should be generated. 368 */ 369 void 370 zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 371 { 372 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE); 373 } 374