zfs_fm.c revision 212425
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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#include <sys/spa.h> 27#include <sys/spa_impl.h> 28#include <sys/vdev.h> 29#include <sys/vdev_impl.h> 30#include <sys/zio.h> 31 32#include <sys/fm/fs/zfs.h> 33#include <sys/fm/protocol.h> 34#include <sys/fm/util.h> 35 36#ifdef _KERNEL 37/* Including sys/bus.h is just too hard, so I declare what I need here. */ 38extern void devctl_notify(const char *__system, const char *__subsystem, 39 const char *__type, const char *__data); 40#endif 41 42/* 43 * This general routine is responsible for generating all the different ZFS 44 * ereports. The payload is dependent on the class, and which arguments are 45 * supplied to the function: 46 * 47 * EREPORT POOL VDEV IO 48 * block X X X 49 * data X X 50 * device X X 51 * pool X 52 * 53 * If we are in a loading state, all errors are chained together by the same 54 * SPA-wide ENA (Error Numeric Association). 55 * 56 * For isolated I/O requests, we get the ENA from the zio_t. The propagation 57 * gets very complicated due to RAID-Z, gang blocks, and vdev caching. We want 58 * to chain together all ereports associated with a logical piece of data. For 59 * read I/Os, there are basically three 'types' of I/O, which form a roughly 60 * layered diagram: 61 * 62 * +---------------+ 63 * | Aggregate I/O | No associated logical data or device 64 * +---------------+ 65 * | 66 * V 67 * +---------------+ Reads associated with a piece of logical data. 68 * | Read I/O | This includes reads on behalf of RAID-Z, 69 * +---------------+ mirrors, gang blocks, retries, etc. 70 * | 71 * V 72 * +---------------+ Reads associated with a particular device, but 73 * | Physical I/O | no logical data. Issued as part of vdev caching 74 * +---------------+ and I/O aggregation. 75 * 76 * Note that 'physical I/O' here is not the same terminology as used in the rest 77 * of ZIO. Typically, 'physical I/O' simply means that there is no attached 78 * blockpointer. But I/O with no associated block pointer can still be related 79 * to a logical piece of data (i.e. RAID-Z requests). 80 * 81 * Purely physical I/O always have unique ENAs. They are not related to a 82 * particular piece of logical data, and therefore cannot be chained together. 83 * We still generate an ereport, but the DE doesn't correlate it with any 84 * logical piece of data. When such an I/O fails, the delegated I/O requests 85 * will issue a retry, which will trigger the 'real' ereport with the correct 86 * ENA. 87 * 88 * We keep track of the ENA for a ZIO chain through the 'io_logical' member. 89 * When a new logical I/O is issued, we set this to point to itself. Child I/Os 90 * then inherit this pointer, so that when it is first set subsequent failures 91 * will use the same ENA. For vdev cache fill and queue aggregation I/O, 92 * this pointer is set to NULL, and no ereport will be generated (since it 93 * doesn't actually correspond to any particular device or piece of data, 94 * and the caller will always retry without caching or queueing anyway). 95 */ 96void 97zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio, 98 uint64_t stateoroffset, uint64_t size) 99{ 100#ifdef _KERNEL 101 char buf[1024]; 102 struct sbuf sb; 103 struct timespec ts; 104 int error; 105 106 /* 107 * If we are doing a spa_tryimport(), ignore errors. 108 */ 109 if (spa->spa_load_state == SPA_LOAD_TRYIMPORT) 110 return; 111 112 /* 113 * If we are in the middle of opening a pool, and the previous attempt 114 * failed, don't bother logging any new ereports - we're just going to 115 * get the same diagnosis anyway. 116 */ 117 if (spa->spa_load_state != SPA_LOAD_NONE && 118 spa->spa_last_open_failed) 119 return; 120 121 if (zio != NULL) { 122 /* 123 * If this is not a read or write zio, ignore the error. This 124 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails. 125 */ 126 if (zio->io_type != ZIO_TYPE_READ && 127 zio->io_type != ZIO_TYPE_WRITE) 128 return; 129 130 /* 131 * Ignore any errors from speculative I/Os, as failure is an 132 * expected result. 133 */ 134 if (zio->io_flags & ZIO_FLAG_SPECULATIVE) 135 return; 136 137 if (vd != NULL) { 138 /* 139 * If the vdev has already been marked as failing due 140 * to a failed probe, then ignore any subsequent I/O 141 * errors, as the DE will automatically fault the vdev 142 * on the first such failure. This also catches cases 143 * where vdev_remove_wanted is set and the device has 144 * not yet been asynchronously placed into the REMOVED 145 * state. 146 */ 147 if (zio->io_vd == vd && 148 !vdev_accessible(vd, zio) && 149 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0) 150 return; 151 152 /* 153 * Ignore checksum errors for reads from DTL regions of 154 * leaf vdevs. 155 */ 156 if (zio->io_type == ZIO_TYPE_READ && 157 zio->io_error == ECKSUM && 158 vd->vdev_ops->vdev_op_leaf && 159 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1)) 160 return; 161 } 162 } 163 nanotime(&ts); 164 165 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 166 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 167 168 /* 169 * Serialize ereport generation 170 */ 171 mutex_enter(&spa->spa_errlist_lock); 172 173#if 0 174 /* 175 * Determine the ENA to use for this event. If we are in a loading 176 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 177 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 178 */ 179 if (spa->spa_load_state != SPA_LOAD_NONE) { 180#if 0 181 if (spa->spa_ena == 0) 182 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 183#endif 184 ena = spa->spa_ena; 185 } else if (zio != NULL && zio->io_logical != NULL) { 186#if 0 187 if (zio->io_logical->io_ena == 0) 188 zio->io_logical->io_ena = 189 fm_ena_generate(0, FM_ENA_FMT1); 190#endif 191 ena = zio->io_logical->io_ena; 192 } else { 193#if 0 194 ena = fm_ena_generate(0, FM_ENA_FMT1); 195#else 196 ena = 0; 197#endif 198 } 199#endif 200 201 /* 202 * Construct the full class, detector, and other standard FMA fields. 203 */ 204 sbuf_printf(&sb, " ereport_version=%u", FM_EREPORT_VERSION); 205 sbuf_printf(&sb, " class=%s.%s", ZFS_ERROR_CLASS, subclass); 206 207 sbuf_printf(&sb, " zfs_scheme_version=%u", FM_ZFS_SCHEME_VERSION); 208 209 /* 210 * Construct the per-ereport payload, depending on which parameters are 211 * passed in. 212 */ 213 214 /* 215 * Generic payload members common to all ereports. 216 */ 217 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL, spa_name(spa)); 218 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 219 spa_guid(spa)); 220 sbuf_printf(&sb, " %s=%d", FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, 221 spa->spa_load_state); 222 223 if (spa != NULL) { 224 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 225 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 226 FM_EREPORT_FAILMODE_WAIT : 227 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 228 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC); 229 } 230 231 if (vd != NULL) { 232 vdev_t *pvd = vd->vdev_parent; 233 234 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 235 vd->vdev_guid); 236 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 237 vd->vdev_ops->vdev_op_type); 238 if (vd->vdev_path != NULL) 239 sbuf_printf(&sb, " %s=%s", 240 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, vd->vdev_path); 241 if (vd->vdev_devid != NULL) 242 sbuf_printf(&sb, " %s=%s", 243 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, vd->vdev_devid); 244 if (vd->vdev_fru != NULL) 245 sbuf_printf(&sb, " %s=%s", 246 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, vd->vdev_fru); 247 248 if (pvd != NULL) { 249 sbuf_printf(&sb, " %s=%ju", 250 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, pvd->vdev_guid); 251 sbuf_printf(&sb, " %s=%s", 252 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 253 pvd->vdev_ops->vdev_op_type); 254 if (pvd->vdev_path) 255 sbuf_printf(&sb, " %s=%s", 256 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 257 pvd->vdev_path); 258 if (pvd->vdev_devid) 259 sbuf_printf(&sb, " %s=%s", 260 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 261 pvd->vdev_devid); 262 } 263 } 264 265 if (zio != NULL) { 266 /* 267 * Payload common to all I/Os. 268 */ 269 sbuf_printf(&sb, " %s=%u", FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 270 zio->io_error); 271 272 /* 273 * If the 'size' parameter is non-zero, it indicates this is a 274 * RAID-Z or other I/O where the physical offset and length are 275 * provided for us, instead of within the zio_t. 276 */ 277 if (vd != NULL) { 278 if (size) { 279 sbuf_printf(&sb, " %s=%ju", 280 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 281 stateoroffset); 282 sbuf_printf(&sb, " %s=%ju", 283 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, size); 284 } else { 285 sbuf_printf(&sb, " %s=%ju", 286 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 287 zio->io_offset); 288 sbuf_printf(&sb, " %s=%ju", 289 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 290 zio->io_size); 291 } 292 } 293 294 /* 295 * Payload for I/Os with corresponding logical information. 296 */ 297 if (zio->io_logical != NULL) { 298 sbuf_printf(&sb, " %s=%ju", 299 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 300 zio->io_logical->io_bookmark.zb_object); 301 sbuf_printf(&sb, " %s=%ju", 302 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 303 zio->io_logical->io_bookmark.zb_level); 304 sbuf_printf(&sb, " %s=%ju", 305 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 306 zio->io_logical->io_bookmark.zb_blkid); 307 } 308 } else if (vd != NULL) { 309 /* 310 * If we have a vdev but no zio, this is a device fault, and the 311 * 'stateoroffset' parameter indicates the previous state of the 312 * vdev. 313 */ 314 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 315 stateoroffset); 316 } 317 mutex_exit(&spa->spa_errlist_lock); 318 319 error = sbuf_finish(&sb); 320 devctl_notify("ZFS", spa->spa_name, subclass, sbuf_data(&sb)); 321 if (error != 0) 322 printf("ZFS WARNING: sbuf overflowed\n"); 323 sbuf_delete(&sb); 324#endif 325} 326 327static void 328zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) 329{ 330#ifdef _KERNEL 331 char buf[1024]; 332 char class[64]; 333 struct sbuf sb; 334 struct timespec ts; 335 int error; 336 337 nanotime(&ts); 338 339 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 340 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 341 342 snprintf(class, sizeof(class), "%s.%s.%s", FM_RSRC_RESOURCE, 343 ZFS_ERROR_CLASS, name); 344 sbuf_printf(&sb, " %s=%hhu", FM_VERSION, FM_RSRC_VERSION); 345 sbuf_printf(&sb, " %s=%s", FM_CLASS, class); 346 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 347 spa_guid(spa)); 348 if (vd) 349 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 350 vd->vdev_guid); 351 error = sbuf_finish(&sb); 352 ZFS_LOG(1, "%s", sbuf_data(&sb)); 353 devctl_notify("ZFS", spa->spa_name, class, sbuf_data(&sb)); 354 if (error != 0) 355 printf("ZFS WARNING: sbuf overflowed\n"); 356 sbuf_delete(&sb); 357#endif 358} 359 360/* 361 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 362 * has been removed from the system. This will cause the DE to ignore any 363 * recent I/O errors, inferring that they are due to the asynchronous device 364 * removal. 365 */ 366void 367zfs_post_remove(spa_t *spa, vdev_t *vd) 368{ 369 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED); 370} 371 372/* 373 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 374 * has the 'autoreplace' property set, and therefore any broken vdevs will be 375 * handled by higher level logic, and no vdev fault should be generated. 376 */ 377void 378zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 379{ 380 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE); 381} 382