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 2006 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. 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]; 102 103 /* 104 * If we are doing a spa_tryimport(), ignore errors. 105 */ 106 if (spa->spa_load_state == SPA_LOAD_TRYIMPORT) 107 return; 108 109 /* 110 * If we are in the middle of opening a pool, and the previous attempt 111 * failed, don't bother logging any new ereports - we're just going to 112 * get the same diagnosis anyway. 113 */ 114 if (spa->spa_load_state != SPA_LOAD_NONE && 115 spa->spa_last_open_failed) 116 return; 117 118 /* 119 * Ignore any errors from I/Os that we are going to retry anyway - we 120 * only generate errors from the final failure. 121 */ 122 if (zio && zio_should_retry(zio)) 123 return; 124 125 /* 126 * If this is not a read or write zio, ignore the error. This can occur 127 * if the DKIOCFLUSHWRITECACHE ioctl fails. 128 */ 129 if (zio && zio->io_type != ZIO_TYPE_READ && 130 zio->io_type != ZIO_TYPE_WRITE) 131 return; 132 133 if ((ereport = fm_nvlist_create(NULL)) == NULL) 134 return; 135 136 if ((detector = fm_nvlist_create(NULL)) == NULL) { 137 fm_nvlist_destroy(ereport, FM_NVA_FREE); 138 return; 139 } 140 141 /* 142 * Serialize ereport generation 143 */ 144 mutex_enter(&spa->spa_errlist_lock); 145 146 /* 147 * Determine the ENA to use for this event. If we are in a loading 148 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 149 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 150 */ 151 if (spa->spa_load_state != SPA_LOAD_NONE) { 152 if (spa->spa_ena == 0) 153 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 154 ena = spa->spa_ena; 155 } else if (zio != NULL && zio->io_logical != NULL) { 156 if (zio->io_logical->io_ena == 0) 157 zio->io_logical->io_ena = 158 fm_ena_generate(0, FM_ENA_FMT1); 159 ena = zio->io_logical->io_ena; 160 } else { 161 ena = fm_ena_generate(0, FM_ENA_FMT1); 162 } 163 164 /* 165 * Construct the full class, detector, and other standard FMA fields. 166 */ 167 (void) snprintf(class, sizeof (class), "%s.%s", 168 ZFS_ERROR_CLASS, subclass); 169 170 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa), 171 vd != NULL ? vd->vdev_guid : 0); 172 173 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL); 174 175 /* 176 * Construct the per-ereport payload, depending on which parameters are 177 * passed in. 178 */ 179 180 /* 181 * Generic payload members common to all ereports. 182 * 183 * The direct reference to spa_name is used rather than spa_name() 184 * because of the asynchronous nature of the zio pipeline. spa_name() 185 * asserts that the config lock is held in some form. This is always 186 * the case in I/O context, but because the check for RW_WRITER compares 187 * against 'curthread', we may be in an asynchronous context and blow 188 * this assert. Rather than loosen this assert, we acknowledge that all 189 * contexts in which this function is called (pool open, I/O) are safe, 190 * and dereference the name directly. 191 */ 192 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL, 193 DATA_TYPE_STRING, spa->spa_name, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 194 DATA_TYPE_UINT64, spa_guid(spa), 195 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32, 196 spa->spa_load_state, NULL); 197 198 if (vd != NULL) { 199 vdev_t *pvd = vd->vdev_parent; 200 201 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 202 DATA_TYPE_UINT64, vd->vdev_guid, 203 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 204 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL); 205 if (vd->vdev_path) 206 fm_payload_set(ereport, 207 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, 208 DATA_TYPE_STRING, vd->vdev_path, NULL); 209 if (vd->vdev_devid) 210 fm_payload_set(ereport, 211 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, 212 DATA_TYPE_STRING, vd->vdev_devid, NULL); 213 214 if (pvd != NULL) { 215 fm_payload_set(ereport, 216 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, 217 DATA_TYPE_UINT64, pvd->vdev_guid, 218 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 219 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type, 220 NULL); 221 if (pvd->vdev_path) 222 fm_payload_set(ereport, 223 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 224 DATA_TYPE_STRING, vd->vdev_path, NULL); 225 if (pvd->vdev_devid) 226 fm_payload_set(ereport, 227 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 228 DATA_TYPE_STRING, pvd->vdev_devid, NULL); 229 } 230 } 231 232 if (zio != NULL) { 233 /* 234 * Payload common to all I/Os. 235 */ 236 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 237 DATA_TYPE_INT32, zio->io_error, NULL); 238 239 /* 240 * If the 'size' parameter is non-zero, it indicates this is a 241 * RAID-Z or other I/O where the physical offset and length are 242 * provided for us, instead of within the zio_t. 243 */ 244 if (vd != NULL) { 245 if (size) 246 fm_payload_set(ereport, 247 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 248 DATA_TYPE_UINT64, stateoroffset, 249 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 250 DATA_TYPE_UINT64, size); 251 else 252 fm_payload_set(ereport, 253 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 254 DATA_TYPE_UINT64, zio->io_offset, 255 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 256 DATA_TYPE_UINT64, zio->io_size); 257 } 258 259 /* 260 * Payload for I/Os with corresponding logical information. 261 */ 262 if (zio->io_logical != NULL) 263 fm_payload_set(ereport, 264 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET, 265 DATA_TYPE_UINT64, 266 zio->io_logical->io_bookmark.zb_objset, 267 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 268 DATA_TYPE_UINT64, 269 zio->io_logical->io_bookmark.zb_object, 270 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 271 DATA_TYPE_INT32, 272 zio->io_logical->io_bookmark.zb_level, 273 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 274 DATA_TYPE_UINT64, 275 zio->io_logical->io_bookmark.zb_blkid); 276 } else if (vd != NULL) { 277 /* 278 * If we have a vdev but no zio, this is a device fault, and the 279 * 'stateoroffset' parameter indicates the previous state of the 280 * vdev. 281 */ 282 fm_payload_set(ereport, 283 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 284 DATA_TYPE_UINT64, stateoroffset, NULL); 285 } 286 mutex_exit(&spa->spa_errlist_lock); 287 288 fm_ereport_post(ereport, EVCH_SLEEP); 289 290 fm_nvlist_destroy(ereport, FM_NVA_FREE); 291 fm_nvlist_destroy(detector, FM_NVA_FREE); 292 #endif 293 } 294 295 /* 296 * The 'resource.fs.zfs.ok' event is an internal signal that the associated 297 * resource (pool or disk) has been identified by ZFS as healthy. This will 298 * then trigger the DE to close the associated case, if any. 299 */ 300 void 301 zfs_post_ok(spa_t *spa, vdev_t *vd) 302 { 303 #ifdef _KERNEL 304 nvlist_t *resource; 305 char class[64]; 306 307 if ((resource = fm_nvlist_create(NULL)) == NULL) 308 return; 309 310 (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE, 311 ZFS_ERROR_CLASS, FM_RESOURCE_OK); 312 VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0); 313 VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0); 314 VERIFY(nvlist_add_uint64(resource, 315 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0); 316 if (vd) 317 VERIFY(nvlist_add_uint64(resource, 318 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0); 319 320 fm_ereport_post(resource, EVCH_SLEEP); 321 322 fm_nvlist_destroy(resource, FM_NVA_FREE); 323 #endif 324 } 325