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 the vdev has already been marked as failing due to a 147 * failed probe, then ignore any subsequent I/O errors, as the 148 * DE will automatically fault the vdev on the first such 149 * failure. 150 */ 151 if (vd != NULL && vd->vdev_is_failing && 152 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0) 153 return; 154 } 155 156 if ((ereport = fm_nvlist_create(NULL)) == NULL) 157 return; 158 159 if ((detector = fm_nvlist_create(NULL)) == NULL) { 160 fm_nvlist_destroy(ereport, FM_NVA_FREE); 161 return; 162 } 163 164 /* 165 * Serialize ereport generation 166 */ 167 mutex_enter(&spa->spa_errlist_lock); 168 169 /* 170 * Determine the ENA to use for this event. If we are in a loading 171 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 172 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 173 */ 174 if (spa->spa_load_state != SPA_LOAD_NONE) { 175 if (spa->spa_ena == 0) 176 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 177 ena = spa->spa_ena; 178 } else if (zio != NULL && zio->io_logical != NULL) { 179 if (zio->io_logical->io_ena == 0) 180 zio->io_logical->io_ena = 181 fm_ena_generate(0, FM_ENA_FMT1); 182 ena = zio->io_logical->io_ena; 183 } else { 184 ena = fm_ena_generate(0, FM_ENA_FMT1); 185 } 186 187 /* 188 * Construct the full class, detector, and other standard FMA fields. 189 */ 190 (void) snprintf(class, sizeof (class), "%s.%s", 191 ZFS_ERROR_CLASS, subclass); 192 193 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa), 194 vd != NULL ? vd->vdev_guid : 0); 195 196 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL); 197 198 /* 199 * Construct the per-ereport payload, depending on which parameters are 200 * passed in. 201 */ 202 203 /* 204 * If we are importing a faulted pool, then we treat it like an open, 205 * not an import. Otherwise, the DE will ignore all faults during 206 * import, since the default behavior is to mark the devices as 207 * persistently unavailable, not leave them in the faulted state. 208 */ 209 state = spa->spa_import_faulted ? SPA_LOAD_OPEN : spa->spa_load_state; 210 211 /* 212 * Generic payload members common to all ereports. 213 * 214 * The direct reference to spa_name is used rather than spa_name() 215 * because of the asynchronous nature of the zio pipeline. spa_name() 216 * asserts that the config lock is held in some form. This is always 217 * the case in I/O context, but because the check for RW_WRITER compares 218 * against 'curthread', we may be in an asynchronous context and blow 219 * this assert. Rather than loosen this assert, we acknowledge that all 220 * contexts in which this function is called (pool open, I/O) are safe, 221 * and dereference the name directly. 222 */ 223 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL, 224 DATA_TYPE_STRING, spa->spa_name, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 225 DATA_TYPE_UINT64, spa_guid(spa), 226 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32, 227 state, NULL); 228 229 if (spa != NULL) { 230 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 231 DATA_TYPE_STRING, 232 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 233 FM_EREPORT_FAILMODE_WAIT : 234 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 235 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC, 236 NULL); 237 } 238 239 if (vd != NULL) { 240 vdev_t *pvd = vd->vdev_parent; 241 242 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 243 DATA_TYPE_UINT64, vd->vdev_guid, 244 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 245 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL); 246 if (vd->vdev_path) 247 fm_payload_set(ereport, 248 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, 249 DATA_TYPE_STRING, vd->vdev_path, NULL); 250 if (vd->vdev_devid) 251 fm_payload_set(ereport, 252 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, 253 DATA_TYPE_STRING, vd->vdev_devid, NULL); 254 255 if (pvd != NULL) { 256 fm_payload_set(ereport, 257 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, 258 DATA_TYPE_UINT64, pvd->vdev_guid, 259 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 260 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type, 261 NULL); 262 if (pvd->vdev_path) 263 fm_payload_set(ereport, 264 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 265 DATA_TYPE_STRING, pvd->vdev_path, NULL); 266 if (pvd->vdev_devid) 267 fm_payload_set(ereport, 268 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 269 DATA_TYPE_STRING, pvd->vdev_devid, NULL); 270 } 271 } 272 273 if (zio != NULL) { 274 /* 275 * Payload common to all I/Os. 276 */ 277 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 278 DATA_TYPE_INT32, zio->io_error, NULL); 279 280 /* 281 * If the 'size' parameter is non-zero, it indicates this is a 282 * RAID-Z or other I/O where the physical offset and length are 283 * provided for us, instead of within the zio_t. 284 */ 285 if (vd != NULL) { 286 if (size) 287 fm_payload_set(ereport, 288 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 289 DATA_TYPE_UINT64, stateoroffset, 290 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 291 DATA_TYPE_UINT64, size, NULL); 292 else 293 fm_payload_set(ereport, 294 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 295 DATA_TYPE_UINT64, zio->io_offset, 296 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 297 DATA_TYPE_UINT64, zio->io_size, NULL); 298 } 299 300 /* 301 * Payload for I/Os with corresponding logical information. 302 */ 303 if (zio->io_logical != NULL) 304 fm_payload_set(ereport, 305 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET, 306 DATA_TYPE_UINT64, 307 zio->io_logical->io_bookmark.zb_objset, 308 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 309 DATA_TYPE_UINT64, 310 zio->io_logical->io_bookmark.zb_object, 311 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 312 DATA_TYPE_INT64, 313 zio->io_logical->io_bookmark.zb_level, 314 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 315 DATA_TYPE_UINT64, 316 zio->io_logical->io_bookmark.zb_blkid, NULL); 317 } else if (vd != NULL) { 318 /* 319 * If we have a vdev but no zio, this is a device fault, and the 320 * 'stateoroffset' parameter indicates the previous state of the 321 * vdev. 322 */ 323 fm_payload_set(ereport, 324 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 325 DATA_TYPE_UINT64, stateoroffset, NULL); 326 } 327 mutex_exit(&spa->spa_errlist_lock); 328 329 fm_ereport_post(ereport, EVCH_SLEEP); 330 331 fm_nvlist_destroy(ereport, FM_NVA_FREE); 332 fm_nvlist_destroy(detector, FM_NVA_FREE); 333 #endif 334 } 335 336 static void 337 zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) 338 { 339 #ifdef _KERNEL 340 nvlist_t *resource; 341 char class[64]; 342 343 if ((resource = fm_nvlist_create(NULL)) == NULL) 344 return; 345 346 (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE, 347 ZFS_ERROR_CLASS, name); 348 VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0); 349 VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0); 350 VERIFY(nvlist_add_uint64(resource, 351 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0); 352 if (vd) 353 VERIFY(nvlist_add_uint64(resource, 354 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0); 355 356 fm_ereport_post(resource, EVCH_SLEEP); 357 358 fm_nvlist_destroy(resource, FM_NVA_FREE); 359 #endif 360 } 361 362 /* 363 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 364 * has been removed from the system. This will cause the DE to ignore any 365 * recent I/O errors, inferring that they are due to the asynchronous device 366 * removal. 367 */ 368 void 369 zfs_post_remove(spa_t *spa, vdev_t *vd) 370 { 371 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED); 372 } 373 374 /* 375 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 376 * has the 'autoreplace' property set, and therefore any broken vdevs will be 377 * handled by higher level logic, and no vdev fault should be generated. 378 */ 379 void 380 zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 381 { 382 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE); 383 } 384