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.
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.
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]
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 #include <sys/spa_impl.h>
29 #include <sys/vdev_impl.h>
32 #include <sys/fm/fs/zfs.h>
33 #include <sys/fm/protocol.h>
34 #include <sys/fm/util.h>
35 #include <sys/sysevent.h>
38 * This general routine is responsible for generating all the different ZFS
39 * ereports. The payload is dependent on the class, and which arguments are
40 * supplied to the function:
42 * EREPORT POOL VDEV IO
48 * If we are in a loading state, all errors are chained together by the same
49 * SPA-wide ENA (Error Numeric Association).
51 * For isolated I/O requests, we get the ENA from the zio_t. The propagation
52 * gets very complicated due to RAID-Z, gang blocks, and vdev caching. We want
53 * to chain together all ereports associated with a logical piece of data. For
54 * read I/Os, there are basically three 'types' of I/O, which form a roughly
58 * | Aggregate I/O | No associated logical data or device
62 * +---------------+ Reads associated with a piece of logical data.
63 * | Read I/O | This includes reads on behalf of RAID-Z,
64 * +---------------+ mirrors, gang blocks, retries, etc.
67 * +---------------+ Reads associated with a particular device, but
68 * | Physical I/O | no logical data. Issued as part of vdev caching
69 * +---------------+ and I/O aggregation.
71 * Note that 'physical I/O' here is not the same terminology as used in the rest
72 * of ZIO. Typically, 'physical I/O' simply means that there is no attached
73 * blockpointer. But I/O with no associated block pointer can still be related
74 * to a logical piece of data (i.e. RAID-Z requests).
76 * Purely physical I/O always have unique ENAs. They are not related to a
77 * particular piece of logical data, and therefore cannot be chained together.
78 * We still generate an ereport, but the DE doesn't correlate it with any
79 * logical piece of data. When such an I/O fails, the delegated I/O requests
80 * will issue a retry, which will trigger the 'real' ereport with the correct
83 * We keep track of the ENA for a ZIO chain through the 'io_logical' member.
84 * When a new logical I/O is issued, we set this to point to itself. Child I/Os
85 * then inherit this pointer, so that when it is first set subsequent failures
86 * will use the same ENA. For vdev cache fill and queue aggregation I/O,
87 * this pointer is set to NULL, and no ereport will be generated (since it
88 * doesn't actually correspond to any particular device or piece of data,
89 * and the caller will always retry without caching or queueing anyway).
92 zfs_ereport_post(const char *subclass, spa_t *spa, vdev_t *vd, zio_t *zio,
93 uint64_t stateoroffset, uint64_t size)
96 nvlist_t *ereport, *detector;
101 * If we are doing a spa_tryimport(), ignore errors.
103 if (spa->spa_load_state == SPA_LOAD_TRYIMPORT)
107 * If we are in the middle of opening a pool, and the previous attempt
108 * failed, don't bother logging any new ereports - we're just going to
109 * get the same diagnosis anyway.
111 if (spa->spa_load_state != SPA_LOAD_NONE &&
112 spa->spa_last_open_failed)
117 * If this is not a read or write zio, ignore the error. This
118 * can occur if the DKIOCFLUSHWRITECACHE ioctl fails.
120 if (zio->io_type != ZIO_TYPE_READ &&
121 zio->io_type != ZIO_TYPE_WRITE)
125 * Ignore any errors from speculative I/Os, as failure is an
128 if (zio->io_flags & ZIO_FLAG_SPECULATIVE)
132 * If this I/O is not a retry I/O, don't post an ereport.
133 * Otherwise, we risk making bad diagnoses based on B_FAILFAST
136 if (zio->io_error == EIO &&
137 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
142 * If the vdev has already been marked as failing due
143 * to a failed probe, then ignore any subsequent I/O
144 * errors, as the DE will automatically fault the vdev
145 * on the first such failure. This also catches cases
146 * where vdev_remove_wanted is set and the device has
147 * not yet been asynchronously placed into the REMOVED
150 if (zio->io_vd == vd &&
151 !vdev_accessible(vd, zio) &&
152 strcmp(subclass, FM_EREPORT_ZFS_PROBE_FAILURE) != 0)
156 * Ignore checksum errors for reads from DTL regions of
159 if (zio->io_type == ZIO_TYPE_READ &&
160 zio->io_error == ECKSUM &&
161 vd->vdev_ops->vdev_op_leaf &&
162 vdev_dtl_contains(vd, DTL_MISSING, zio->io_txg, 1))
167 if ((ereport = fm_nvlist_create(NULL)) == NULL)
170 if ((detector = fm_nvlist_create(NULL)) == NULL) {
171 fm_nvlist_destroy(ereport, FM_NVA_FREE);
176 * Serialize ereport generation
178 mutex_enter(&spa->spa_errlist_lock);
181 * Determine the ENA to use for this event. If we are in a loading
182 * state, use a SPA-wide ENA. Otherwise, if we are in an I/O state, use
183 * a root zio-wide ENA. Otherwise, simply use a unique ENA.
185 if (spa->spa_load_state != SPA_LOAD_NONE) {
186 if (spa->spa_ena == 0)
187 spa->spa_ena = fm_ena_generate(0, FM_ENA_FMT1);
189 } else if (zio != NULL && zio->io_logical != NULL) {
190 if (zio->io_logical->io_ena == 0)
191 zio->io_logical->io_ena =
192 fm_ena_generate(0, FM_ENA_FMT1);
193 ena = zio->io_logical->io_ena;
195 ena = fm_ena_generate(0, FM_ENA_FMT1);
199 * Construct the full class, detector, and other standard FMA fields.
201 (void) snprintf(class, sizeof (class), "%s.%s",
202 ZFS_ERROR_CLASS, subclass);
204 fm_fmri_zfs_set(detector, FM_ZFS_SCHEME_VERSION, spa_guid(spa),
205 vd != NULL ? vd->vdev_guid : 0);
207 fm_ereport_set(ereport, FM_EREPORT_VERSION, class, ena, detector, NULL);
210 * Construct the per-ereport payload, depending on which parameters are
215 * Generic payload members common to all ereports.
217 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL,
218 DATA_TYPE_STRING, spa_name(spa), FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
219 DATA_TYPE_UINT64, spa_guid(spa),
220 FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, DATA_TYPE_INT32,
221 spa->spa_load_state, NULL);
224 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE,
226 spa_get_failmode(spa) == ZIO_FAILURE_MODE_WAIT ?
227 FM_EREPORT_FAILMODE_WAIT :
228 spa_get_failmode(spa) == ZIO_FAILURE_MODE_CONTINUE ?
229 FM_EREPORT_FAILMODE_CONTINUE : FM_EREPORT_FAILMODE_PANIC,
234 vdev_t *pvd = vd->vdev_parent;
236 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
237 DATA_TYPE_UINT64, vd->vdev_guid,
238 FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
239 DATA_TYPE_STRING, vd->vdev_ops->vdev_op_type, NULL);
240 if (vd->vdev_path != NULL)
241 fm_payload_set(ereport,
242 FM_EREPORT_PAYLOAD_ZFS_VDEV_PATH,
243 DATA_TYPE_STRING, vd->vdev_path, NULL);
244 if (vd->vdev_devid != NULL)
245 fm_payload_set(ereport,
246 FM_EREPORT_PAYLOAD_ZFS_VDEV_DEVID,
247 DATA_TYPE_STRING, vd->vdev_devid, NULL);
248 if (vd->vdev_fru != NULL)
249 fm_payload_set(ereport,
250 FM_EREPORT_PAYLOAD_ZFS_VDEV_FRU,
251 DATA_TYPE_STRING, vd->vdev_fru, NULL);
254 fm_payload_set(ereport,
255 FM_EREPORT_PAYLOAD_ZFS_PARENT_GUID,
256 DATA_TYPE_UINT64, pvd->vdev_guid,
257 FM_EREPORT_PAYLOAD_ZFS_PARENT_TYPE,
258 DATA_TYPE_STRING, pvd->vdev_ops->vdev_op_type,
261 fm_payload_set(ereport,
262 FM_EREPORT_PAYLOAD_ZFS_PARENT_PATH,
263 DATA_TYPE_STRING, pvd->vdev_path, NULL);
265 fm_payload_set(ereport,
266 FM_EREPORT_PAYLOAD_ZFS_PARENT_DEVID,
267 DATA_TYPE_STRING, pvd->vdev_devid, NULL);
273 * Payload common to all I/Os.
275 fm_payload_set(ereport, FM_EREPORT_PAYLOAD_ZFS_ZIO_ERR,
276 DATA_TYPE_INT32, zio->io_error, NULL);
279 * If the 'size' parameter is non-zero, it indicates this is a
280 * RAID-Z or other I/O where the physical offset and length are
281 * provided for us, instead of within the zio_t.
285 fm_payload_set(ereport,
286 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
287 DATA_TYPE_UINT64, stateoroffset,
288 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
289 DATA_TYPE_UINT64, size, NULL);
291 fm_payload_set(ereport,
292 FM_EREPORT_PAYLOAD_ZFS_ZIO_OFFSET,
293 DATA_TYPE_UINT64, zio->io_offset,
294 FM_EREPORT_PAYLOAD_ZFS_ZIO_SIZE,
295 DATA_TYPE_UINT64, zio->io_size, NULL);
299 * Payload for I/Os with corresponding logical information.
301 if (zio->io_logical != NULL)
302 fm_payload_set(ereport,
303 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJSET,
305 zio->io_logical->io_bookmark.zb_objset,
306 FM_EREPORT_PAYLOAD_ZFS_ZIO_OBJECT,
308 zio->io_logical->io_bookmark.zb_object,
309 FM_EREPORT_PAYLOAD_ZFS_ZIO_LEVEL,
311 zio->io_logical->io_bookmark.zb_level,
312 FM_EREPORT_PAYLOAD_ZFS_ZIO_BLKID,
314 zio->io_logical->io_bookmark.zb_blkid, NULL);
315 } else if (vd != NULL) {
317 * If we have a vdev but no zio, this is a device fault, and the
318 * 'stateoroffset' parameter indicates the previous state of the
321 fm_payload_set(ereport,
322 FM_EREPORT_PAYLOAD_ZFS_PREV_STATE,
323 DATA_TYPE_UINT64, stateoroffset, NULL);
325 mutex_exit(&spa->spa_errlist_lock);
327 fm_ereport_post(ereport, EVCH_SLEEP);
329 fm_nvlist_destroy(ereport, FM_NVA_FREE);
330 fm_nvlist_destroy(detector, FM_NVA_FREE);
335 zfs_post_common(spa_t *spa, vdev_t *vd, const char *name)
341 if ((resource = fm_nvlist_create(NULL)) == NULL)
344 (void) snprintf(class, sizeof (class), "%s.%s.%s", FM_RSRC_RESOURCE,
345 ZFS_ERROR_CLASS, name);
346 VERIFY(nvlist_add_uint8(resource, FM_VERSION, FM_RSRC_VERSION) == 0);
347 VERIFY(nvlist_add_string(resource, FM_CLASS, class) == 0);
348 VERIFY(nvlist_add_uint64(resource,
349 FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, spa_guid(spa)) == 0);
351 VERIFY(nvlist_add_uint64(resource,
352 FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, vd->vdev_guid) == 0);
354 fm_ereport_post(resource, EVCH_SLEEP);
356 fm_nvlist_destroy(resource, FM_NVA_FREE);
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
367 zfs_post_remove(spa_t *spa, vdev_t *vd)
369 zfs_post_common(spa, vd, FM_RESOURCE_REMOVED);
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.
378 zfs_post_autoreplace(spa_t *spa, vdev_t *vd)
380 zfs_post_common(spa, vd, FM_RESOURCE_AUTOREPLACE);