zfs_fm.c revision 213198
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 /* 138 * If this I/O is not a retry I/O, don't post an ereport. 139 * Otherwise, we risk making bad diagnoses based on B_FAILFAST 140 * I/Os. 141 */ 142 if (zio->io_error == EIO && 143 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 144 return; 145 146 if (vd != NULL) { 147 /* 148 * If the vdev has already been marked as failing due 149 * to a failed probe, then ignore any subsequent I/O 150 * errors, as the DE will automatically fault the vdev 151 * on the first such failure. This also catches cases 152 * where vdev_remove_wanted is set and the device has 153 * not yet been asynchronously placed into the REMOVED 154 * state. 155 */ 156 if (zio->io_vd == vd && 157 !vdev_accessible(vd, zio) && 158 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0) 159 return; 160 161 /* 162 * Ignore checksum errors for reads from DTL regions of 163 * leaf vdevs. 164 */ 165 if (zio->io_type == ZIO_TYPE_READ && 166 zio->io_error == ECKSUM && 167 vd->vdev_ops->vdev_op_leaf && 168 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1)) 169 return; 170 } 171 } 172 nanotime(&ts); 173 174 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 175 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 176 177 /* 178 * Serialize ereport generation 179 */ 180 mutex_enter(&spa->spa_errlist_lock); 181 182#if 0 183 /* 184 * Determine the ENA to use for this event. If we are in a loading 185 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 186 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 187 */ 188 if (spa->spa_load_state != SPA_LOAD_NONE) { 189#if 0 190 if (spa->spa_ena == 0) 191 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 192#endif 193 ena = spa->spa_ena; 194 } else if (zio != NULL && zio->io_logical != NULL) { 195#if 0 196 if (zio->io_logical->io_ena == 0) 197 zio->io_logical->io_ena = 198 fm_ena_generate(0, FM_ENA_FMT1); 199#endif 200 ena = zio->io_logical->io_ena; 201 } else { 202#if 0 203 ena = fm_ena_generate(0, FM_ENA_FMT1); 204#else 205 ena = 0; 206#endif 207 } 208#endif 209 210 /* 211 * Construct the full class, detector, and other standard FMA fields. 212 */ 213 sbuf_printf(&sb, " ereport_version=%u", FM_EREPORT_VERSION); 214 sbuf_printf(&sb, " class=%s.%s", ZFS_ERROR_CLASS, subclass); 215 216 sbuf_printf(&sb, " zfs_scheme_version=%u", FM_ZFS_SCHEME_VERSION); 217 218 /* 219 * Construct the per-ereport payload, depending on which parameters are 220 * passed in. 221 */ 222 223 /* 224 * Generic payload members common to all ereports. 225 */ 226 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL, spa_name(spa)); 227 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 228 spa_guid(spa)); 229 sbuf_printf(&sb, " %s=%d", FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, 230 spa->spa_load_state); 231 232 if (spa != NULL) { 233 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 234 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 235 FM_EREPORT_FAILMODE_WAIT : 236 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 237 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC); 238 } 239 240 if (vd != NULL) { 241 vdev_t *pvd = vd->vdev_parent; 242 243 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 244 vd->vdev_guid); 245 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 246 vd->vdev_ops->vdev_op_type); 247 if (vd->vdev_path != NULL) 248 sbuf_printf(&sb, " %s=%s", 249 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, vd->vdev_path); 250 if (vd->vdev_devid != NULL) 251 sbuf_printf(&sb, " %s=%s", 252 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, vd->vdev_devid); 253 if (vd->vdev_fru != NULL) 254 sbuf_printf(&sb, " %s=%s", 255 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, vd->vdev_fru); 256 257 if (pvd != NULL) { 258 sbuf_printf(&sb, " %s=%ju", 259 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, pvd->vdev_guid); 260 sbuf_printf(&sb, " %s=%s", 261 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 262 pvd->vdev_ops->vdev_op_type); 263 if (pvd->vdev_path) 264 sbuf_printf(&sb, " %s=%s", 265 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 266 pvd->vdev_path); 267 if (pvd->vdev_devid) 268 sbuf_printf(&sb, " %s=%s", 269 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 270 pvd->vdev_devid); 271 } 272 } 273 274 if (zio != NULL) { 275 /* 276 * Payload common to all I/Os. 277 */ 278 sbuf_printf(&sb, " %s=%u", FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 279 zio->io_error); 280 281 /* 282 * If the 'size' parameter is non-zero, it indicates this is a 283 * RAID-Z or other I/O where the physical offset and length are 284 * provided for us, instead of within the zio_t. 285 */ 286 if (vd != NULL) { 287 if (size) { 288 sbuf_printf(&sb, " %s=%ju", 289 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 290 stateoroffset); 291 sbuf_printf(&sb, " %s=%ju", 292 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, size); 293 } else { 294 sbuf_printf(&sb, " %s=%ju", 295 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 296 zio->io_offset); 297 sbuf_printf(&sb, " %s=%ju", 298 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 299 zio->io_size); 300 } 301 } 302 303 /* 304 * Payload for I/Os with corresponding logical information. 305 */ 306 if (zio->io_logical != NULL) { 307 sbuf_printf(&sb, " %s=%ju", 308 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 309 zio->io_logical->io_bookmark.zb_object); 310 sbuf_printf(&sb, " %s=%ju", 311 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 312 zio->io_logical->io_bookmark.zb_level); 313 sbuf_printf(&sb, " %s=%ju", 314 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 315 zio->io_logical->io_bookmark.zb_blkid); 316 } 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 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 324 stateoroffset); 325 } 326 mutex_exit(&spa->spa_errlist_lock); 327 328 error = sbuf_finish(&sb); 329 devctl_notify("ZFS", spa->spa_name, subclass, sbuf_data(&sb)); 330 if (error != 0) 331 printf("ZFS WARNING: sbuf overflowed\n"); 332 sbuf_delete(&sb); 333#endif 334} 335 336static void 337zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) 338{ 339#ifdef _KERNEL 340 char buf[1024]; 341 char class[64]; 342 struct sbuf sb; 343 struct timespec ts; 344 int error; 345 346 nanotime(&ts); 347 348 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 349 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 350 351 snprintf(class, sizeof(class), "%s.%s.%s", FM_RSRC_RESOURCE, 352 ZFS_ERROR_CLASS, name); 353 sbuf_printf(&sb, " %s=%hhu", FM_VERSION, FM_RSRC_VERSION); 354 sbuf_printf(&sb, " %s=%s", FM_CLASS, class); 355 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 356 spa_guid(spa)); 357 if (vd) 358 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 359 vd->vdev_guid); 360 error = sbuf_finish(&sb); 361 ZFS_LOG(1, "%s", sbuf_data(&sb)); 362 devctl_notify("ZFS", spa->spa_name, class, sbuf_data(&sb)); 363 if (error != 0) 364 printf("ZFS WARNING: sbuf overflowed\n"); 365 sbuf_delete(&sb); 366#endif 367} 368 369/* 370 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 371 * has been removed from the system. This will cause the DE to ignore any 372 * recent I/O errors, inferring that they are due to the asynchronous device 373 * removal. 374 */ 375void 376zfs_post_remove(spa_t *spa, vdev_t *vd) 377{ 378 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED); 379} 380 381/* 382 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 383 * has the 'autoreplace' property set, and therefore any broken vdevs will be 384 * handled by higher level logic, and no vdev fault should be generated. 385 */ 386void 387zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 388{ 389 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE); 390} 391