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.
|
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 state;
|
104 105 /* 106 * If we are doing a spa_tryimport(), ignore errors. 107 */ 108 if (spa->spa_load_state == SPA_LOAD_TRYIMPORT) 109 return; 110 111 /* 112 * If we are in the middle of opening a pool, and the previous attempt 113 * failed, don't bother logging any new ereports - we're just going to 114 * get the same diagnosis anyway. 115 */ 116 if (spa->spa_load_state != SPA_LOAD_NONE && 117 spa->spa_last_open_failed) 118 return; 119 120 if (zio != NULL) { 121 /* 122 * If this is not a read or write zio, ignore the error. This 123 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails. 124 */ 125 if (zio->io_type != ZIO_TYPE_READ && 126 zio->io_type != ZIO_TYPE_WRITE) 127 return; 128 129 /* 130 * Ignore any errors from speculative I/Os, as failure is an 131 * expected result. 132 */ 133 if (zio->io_flags & ZIO_FLAG_SPECULATIVE) 134 return; 135
|
137 /*
138 * If the vdev has already been marked as failing due to a
139 * failed probe, then ignore any subsequent I/O errors, as the
140 * DE will automatically fault the vdev on the first such
141 * failure.
142 */
143 if (vd != NULL &&
144 (!vdev_readable(vd) || !vdev_writeable(vd)) &&
145 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0)
146 return;
|
136 if (vd != NULL) { 137 /* 138 * If the vdev has already been marked as failing due 139 * to a failed probe, then ignore any subsequent I/O 140 * errors, as the DE will automatically fault the vdev 141 * on the first such failure. This also catches cases 142 * where vdev_remove_wanted is set and the device has 143 * not yet been asynchronously placed into the REMOVED 144 * state. 145 */ 146 if (zio->io_vd == vd && 147 !vdev_accessible(vd, zio) && 148 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0) 149 return; 150 151 /* 152 * Ignore checksum errors for reads from DTL regions of 153 * leaf vdevs. 154 */ 155 if (zio->io_type == ZIO_TYPE_READ && 156 zio->io_error == ECKSUM && 157 vd->vdev_ops->vdev_op_leaf && 158 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1)) 159 return; 160 } |
161 } 162 nanotime(&ts); 163 164 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 165 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 166 167 /* 168 * Serialize ereport generation 169 */ 170 mutex_enter(&spa->spa_errlist_lock); 171 172#if 0 173 /* 174 * Determine the ENA to use for this event. If we are in a loading 175 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use 176 * a root zio-wide ENA. Otherwise, simply use a unique ENA. 177 */ 178 if (spa->spa_load_state != SPA_LOAD_NONE) { 179#if 0 180 if (spa->spa_ena == 0) 181 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1); 182#endif 183 ena = spa->spa_ena; 184 } else if (zio != NULL && zio->io_logical != NULL) { 185#if 0 186 if (zio->io_logical->io_ena == 0) 187 zio->io_logical->io_ena = 188 fm_ena_generate(0, FM_ENA_FMT1); 189#endif 190 ena = zio->io_logical->io_ena; 191 } else { 192#if 0 193 ena = fm_ena_generate(0, FM_ENA_FMT1); 194#else 195 ena = 0; 196#endif 197 } 198#endif 199 200 /* 201 * Construct the full class, detector, and other standard FMA fields. 202 */ 203 sbuf_printf(&sb, " ereport_version=%u", FM_EREPORT_VERSION); 204 sbuf_printf(&sb, " class=%s.%s", ZFS_ERROR_CLASS, subclass); 205 206 sbuf_printf(&sb, " zfs_scheme_version=%u", FM_ZFS_SCHEME_VERSION); 207 208 /* 209 * Construct the per-ereport payload, depending on which parameters are 210 * passed in. 211 */ 212 213 /*
|
200 * If we are importing a faulted pool, then we treat it like an open,
201 * not an import. Otherwise, the DE will ignore all faults during
202 * import, since the default behavior is to mark the devices as
203 * persistently unavailable, not leave them in the faulted state.
204 */
205 state = spa->spa_import_faulted ? SPA_LOAD_OPEN : spa->spa_load_state;
206
207 /*
|
214 * Generic payload members common to all ereports. 215 */ 216 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL, spa_name(spa)); 217 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 218 spa_guid(spa));
|
213 sbuf_printf(&sb, " %s=%d", FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, state);
|
219 sbuf_printf(&sb, " %s=%d", FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, 220 spa->spa_load_state); |
221 222 if (spa != NULL) { 223 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, 224 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ? 225 FM_EREPORT_FAILMODE_WAIT : 226 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ? 227 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC); 228 } 229 230 if (vd != NULL) { 231 vdev_t *pvd = vd->vdev_parent; 232 233 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 234 vd->vdev_guid); 235 sbuf_printf(&sb, " %s=%s", FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE, 236 vd->vdev_ops->vdev_op_type);
|
230 if (vd->vdev_path)
|
237 if (vd->vdev_path != NULL) |
238 sbuf_printf(&sb, " %s=%s", 239 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH, vd->vdev_path);
|
233 if (vd->vdev_devid)
|
240 if (vd->vdev_devid != NULL) |
241 sbuf_printf(&sb, " %s=%s", 242 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID, vd->vdev_devid);
|
243 if (vd->vdev_fru != NULL) 244 sbuf_printf(&sb, " %s=%s", 245 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU, vd->vdev_fru); |
246 247 if (pvd != NULL) { 248 sbuf_printf(&sb, " %s=%ju", 249 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID, pvd->vdev_guid); 250 sbuf_printf(&sb, " %s=%s", 251 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE, 252 pvd->vdev_ops->vdev_op_type); 253 if (pvd->vdev_path) 254 sbuf_printf(&sb, " %s=%s", 255 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH, 256 pvd->vdev_path); 257 if (pvd->vdev_devid) 258 sbuf_printf(&sb, " %s=%s", 259 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID, 260 pvd->vdev_devid); 261 } 262 } 263 264 if (zio != NULL) { 265 /* 266 * Payload common to all I/Os. 267 */ 268 sbuf_printf(&sb, " %s=%u", FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR, 269 zio->io_error); 270 271 /* 272 * If the 'size' parameter is non-zero, it indicates this is a 273 * RAID-Z or other I/O where the physical offset and length are 274 * provided for us, instead of within the zio_t. 275 */ 276 if (vd != NULL) { 277 if (size) { 278 sbuf_printf(&sb, " %s=%ju", 279 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 280 stateoroffset); 281 sbuf_printf(&sb, " %s=%ju", 282 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, size); 283 } else { 284 sbuf_printf(&sb, " %s=%ju", 285 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET, 286 zio->io_offset); 287 sbuf_printf(&sb, " %s=%ju", 288 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE, 289 zio->io_size); 290 } 291 } 292 293 /* 294 * Payload for I/Os with corresponding logical information. 295 */ 296 if (zio->io_logical != NULL) { 297 sbuf_printf(&sb, " %s=%ju", 298 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT, 299 zio->io_logical->io_bookmark.zb_object); 300 sbuf_printf(&sb, " %s=%ju", 301 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL, 302 zio->io_logical->io_bookmark.zb_level); 303 sbuf_printf(&sb, " %s=%ju", 304 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID, 305 zio->io_logical->io_bookmark.zb_blkid); 306 } 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 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_PREV_STATE, 314 stateoroffset); 315 } 316 mutex_exit(&spa->spa_errlist_lock); 317 318 sbuf_finish(&sb); 319 devctl_notify("ZFS", spa->spa_name, subclass, sbuf_data(&sb)); 320 if (sbuf_overflowed(&sb)) 321 printf("ZFS WARNING: sbuf overflowed\n"); 322 sbuf_delete(&sb); 323#endif 324} 325 326static void 327zfs_post_common(spa_t *spa, vdev_t *vd, const char *name) 328{ 329#ifdef _KERNEL 330 char buf[1024]; 331 char class[64]; 332 struct sbuf sb; 333 struct timespec ts; 334 335 nanotime(&ts); 336 337 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN); 338 sbuf_printf(&sb, "time=%ju.%ld", (uintmax_t)ts.tv_sec, ts.tv_nsec); 339 340 snprintf(class, sizeof(class), "%s.%s.%s", FM_RSRC_RESOURCE, 341 ZFS_ERROR_CLASS, name); 342 sbuf_printf(&sb, " %s=%hhu", FM_VERSION, FM_RSRC_VERSION); 343 sbuf_printf(&sb, " %s=%s", FM_CLASS, class); 344 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, 345 spa_guid(spa)); 346 if (vd) 347 sbuf_printf(&sb, " %s=%ju", FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, 348 vd->vdev_guid); 349 sbuf_finish(&sb); 350 ZFS_LOG(1, "%s", sbuf_data(&sb)); 351 devctl_notify("ZFS", spa->spa_name, class, sbuf_data(&sb)); 352 if (sbuf_overflowed(&sb)) 353 printf("ZFS WARNING: sbuf overflowed\n"); 354 sbuf_delete(&sb); 355#endif 356} 357 358/* 359 * The 'resource.fs.zfs.removed' event is an internal signal that the given vdev 360 * has been removed from the system. This will cause the DE to ignore any 361 * recent I/O errors, inferring that they are due to the asynchronous device 362 * removal. 363 */ 364void 365zfs_post_remove(spa_t *spa, vdev_t *vd) 366{ 367 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED); 368} 369 370/* 371 * The 'resource.fs.zfs.autoreplace' event is an internal signal that the pool 372 * has the 'autoreplace' property set, and therefore any broken vdevs will be 373 * handled by higher level logic, and no vdev fault should be generated. 374 */ 375void 376zfs_post_autoreplace(spa_t *spa, vdev_t *vd) 377{ 378 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE); 379}
|