diff --git a/module/zfs/dsl_scan.c b/module/zfs/dsl_scan.c index 07e3dec1ecec..453431def881 100644 --- a/module/zfs/dsl_scan.c +++ b/module/zfs/dsl_scan.c @@ -1,1819 +1,1819 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #endif typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *); static scan_cb_t dsl_scan_scrub_cb; static dsl_syncfunc_t dsl_scan_cancel_sync; static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *tx); int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */ int zfs_resilver_delay = 2; /* number of ticks to delay resilver */ int zfs_scrub_delay = 4; /* number of ticks to delay scrub */ int zfs_scan_idle = 50; /* idle window in clock ticks */ int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */ int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */ int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */ int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */ int zfs_no_scrub_prefetch = B_FALSE; /* set to disable srub prefetching */ enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE; int dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */ #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \ ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \ (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) extern int zfs_txg_timeout; /* the order has to match pool_scan_type */ static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = { NULL, dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */ dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */ }; int dsl_scan_init(dsl_pool_t *dp, uint64_t txg) { int err; dsl_scan_t *scn; spa_t *spa = dp->dp_spa; uint64_t f; scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP); scn->scn_dp = dp; err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_func", sizeof (uint64_t), 1, &f); if (err == 0) { /* * There was an old-style scrub in progress. Restart a * new-style scrub from the beginning. */ scn->scn_restart_txg = txg; zfs_dbgmsg("old-style scrub was in progress; " "restarting new-style scrub in txg %llu", scn->scn_restart_txg); /* * Load the queue obj from the old location so that it * can be freed by dsl_scan_done(). */ (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, "scrub_queue", sizeof (uint64_t), 1, &scn->scn_phys.scn_queue_obj); } else { err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys); if (err == ENOENT) return (0); else if (err) return (err); if (scn->scn_phys.scn_state == DSS_SCANNING && spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) { /* * A new-type scrub was in progress on an old * pool, and the pool was accessed by old * software. Restart from the beginning, since * the old software may have changed the pool in * the meantime. */ scn->scn_restart_txg = txg; zfs_dbgmsg("new-style scrub was modified " "by old software; restarting in txg %llu", scn->scn_restart_txg); } } spa_scan_stat_init(spa); return (0); } void dsl_scan_fini(dsl_pool_t *dp) { if (dp->dp_scan) { kmem_free(dp->dp_scan, sizeof (dsl_scan_t)); dp->dp_scan = NULL; } } /* ARGSUSED */ static int dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx) { dsl_scan_t *scn = arg1; if (scn->scn_phys.scn_state == DSS_SCANNING) return (EBUSY); return (0); } /* ARGSUSED */ static void dsl_scan_setup_sync(void *arg1, void *arg2, dmu_tx_t *tx) { dsl_scan_t *scn = arg1; pool_scan_func_t *funcp = arg2; dmu_object_type_t ot = 0; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; ASSERT(scn->scn_phys.scn_state != DSS_SCANNING); ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS); bzero(&scn->scn_phys, sizeof (scn->scn_phys)); scn->scn_phys.scn_func = *funcp; scn->scn_phys.scn_state = DSS_SCANNING; scn->scn_phys.scn_min_txg = 0; scn->scn_phys.scn_max_txg = tx->tx_txg; scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */ scn->scn_phys.scn_start_time = gethrestime_sec(); scn->scn_phys.scn_errors = 0; scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc; scn->scn_restart_txg = 0; spa_scan_stat_init(spa); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max; /* rewrite all disk labels */ vdev_config_dirty(spa->spa_root_vdev); if (vdev_resilver_needed(spa->spa_root_vdev, &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) { spa_event_notify(spa, NULL, FM_EREPORT_ZFS_RESILVER_START); } else { spa_event_notify(spa, NULL, FM_EREPORT_ZFS_SCRUB_START); } spa->spa_scrub_started = B_TRUE; /* * If this is an incremental scrub, limit the DDT scrub phase * to just the auto-ditto class (for correctness); the rest * of the scrub should go faster using top-down pruning. */ if (scn->scn_phys.scn_min_txg > TXG_INITIAL) scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO; } /* back to the generic stuff */ if (dp->dp_blkstats == NULL) { dp->dp_blkstats = kmem_alloc(sizeof (zfs_all_blkstats_t), - KM_SLEEP | KM_NODEBUG); + KM_PUSHPAGE | KM_NODEBUG); } bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) ot = DMU_OT_ZAP_OTHER; scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx); dsl_scan_sync_state(scn, tx); spa_history_log_internal(LOG_POOL_SCAN, spa, tx, "func=%u mintxg=%llu maxtxg=%llu", *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg); } /* ARGSUSED */ static void dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) { static const char *old_names[] = { "scrub_bookmark", "scrub_ddt_bookmark", "scrub_ddt_class_max", "scrub_queue", "scrub_min_txg", "scrub_max_txg", "scrub_func", "scrub_errors", NULL }; dsl_pool_t *dp = scn->scn_dp; spa_t *spa = dp->dp_spa; int i; /* Remove any remnants of an old-style scrub. */ for (i = 0; old_names[i]; i++) { (void) zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx); } if (scn->scn_phys.scn_queue_obj != 0) { VERIFY(0 == dmu_object_free(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, tx)); scn->scn_phys.scn_queue_obj = 0; } /* * If we were "restarted" from a stopped state, don't bother * with anything else. */ if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (complete) scn->scn_phys.scn_state = DSS_FINISHED; else scn->scn_phys.scn_state = DSS_CANCELED; spa_history_log_internal(LOG_POOL_SCAN_DONE, spa, tx, "complete=%u", complete); if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight > 0) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } mutex_exit(&spa->spa_scrub_lock); spa->spa_scrub_started = B_FALSE; spa->spa_scrub_active = B_FALSE; /* * If the scrub/resilver completed, update all DTLs to * reflect this. Whether it succeeded or not, vacate * all temporary scrub DTLs. */ vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE); if (complete) { spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ? FM_EREPORT_ZFS_RESILVER_FINISH : FM_EREPORT_ZFS_SCRUB_FINISH); } spa_errlog_rotate(spa); /* * We may have finished replacing a device. * Let the async thread assess this and handle the detach. */ spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); } scn->scn_phys.scn_end_time = gethrestime_sec(); } /* ARGSUSED */ static int dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx) { dsl_scan_t *scn = arg1; if (scn->scn_phys.scn_state != DSS_SCANNING) return (ENOENT); return (0); } /* ARGSUSED */ static void dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx) { dsl_scan_t *scn = arg1; dsl_scan_done(scn, B_FALSE, tx); dsl_scan_sync_state(scn, tx); } int dsl_scan_cancel(dsl_pool_t *dp) { boolean_t complete = B_FALSE; int err; err = dsl_sync_task_do(dp, dsl_scan_cancel_check, dsl_scan_cancel_sync, dp->dp_scan, &complete, 3); return (err); } static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_t *zb, dnode_phys_t *dnp, arc_buf_t *pbuf, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx); inline __attribute__((always_inline)) static void dsl_scan_visitdnode( dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, arc_buf_t *buf, uint64_t object, dmu_tx_t *tx); void dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp) { zio_free(dp->dp_spa, txg, bp); } void dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp) { ASSERT(dsl_pool_sync_context(dp)); zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags)); } int dsl_read(zio_t *pio, spa_t *spa, const blkptr_t *bpp, arc_buf_t *pbuf, arc_done_func_t *done, void *private, int priority, int zio_flags, uint32_t *arc_flags, const zbookmark_t *zb) { return (arc_read(pio, spa, bpp, pbuf, done, private, priority, zio_flags, arc_flags, zb)); } int dsl_read_nolock(zio_t *pio, spa_t *spa, const blkptr_t *bpp, arc_done_func_t *done, void *private, int priority, int zio_flags, uint32_t *arc_flags, const zbookmark_t *zb) { return (arc_read_nolock(pio, spa, bpp, done, private, priority, zio_flags, arc_flags, zb)); } static boolean_t bookmark_is_zero(const zbookmark_t *zb) { return (zb->zb_objset == 0 && zb->zb_object == 0 && zb->zb_level == 0 && zb->zb_blkid == 0); } /* dnp is the dnode for zb1->zb_object */ static boolean_t bookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, const zbookmark_t *zb2) { uint64_t zb1nextL0, zb2thisobj; ASSERT(zb1->zb_objset == zb2->zb_objset); ASSERT(zb2->zb_level == 0); /* * A bookmark in the deadlist is considered to be after * everything else. */ if (zb2->zb_object == DMU_DEADLIST_OBJECT) return (B_TRUE); /* The objset_phys_t isn't before anything. */ if (dnp == NULL) return (B_FALSE); zb1nextL0 = (zb1->zb_blkid + 1) << ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); zb2thisobj = zb2->zb_object ? zb2->zb_object : zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); if (zb1->zb_object == DMU_META_DNODE_OBJECT) { uint64_t nextobj = zb1nextL0 * (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; return (nextobj <= zb2thisobj); } if (zb1->zb_object < zb2thisobj) return (B_TRUE); if (zb1->zb_object > zb2thisobj) return (B_FALSE); if (zb2->zb_object == DMU_META_DNODE_OBJECT) return (B_FALSE); return (zb1nextL0 <= zb2->zb_blkid); } static uint64_t dsl_scan_ds_maxtxg(dsl_dataset_t *ds) { uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg; if (dsl_dataset_is_snapshot(ds)) return (MIN(smt, ds->ds_phys->ds_creation_txg)); return (smt); } static void dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx) { VERIFY(0 == zap_update(scn->scn_dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, &scn->scn_phys, tx)); } static boolean_t dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_t *zb) { uint64_t elapsed_nanosecs; int mintime; /* we never skip user/group accounting objects */ if (zb && (int64_t)zb->zb_object < 0) return (B_FALSE); if (scn->scn_pausing) return (B_TRUE); /* we're already pausing */ if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark)) return (B_FALSE); /* we're resuming */ /* We only know how to resume from level-0 blocks. */ if (zb && zb->zb_level != 0) return (B_FALSE); mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms : zfs_scan_min_time_ms; elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || (elapsed_nanosecs / MICROSEC > mintime && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)) { if (zb) { dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); scn->scn_phys.scn_bookmark = *zb; } dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n", (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); scn->scn_pausing = B_TRUE; return (B_TRUE); } return (B_FALSE); } typedef struct zil_scan_arg { dsl_pool_t *zsa_dp; zil_header_t *zsa_zh; } zil_scan_arg_t; /* ARGSUSED */ static int dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; zbookmark_t zb; if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * One block ("stubby") can be allocated a long time ago; we * want to visit that one because it has been allocated * (on-disk) even if it hasn't been claimed (even though for * scrub there's nothing to do to it). */ if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa)) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); return (0); } /* ARGSUSED */ static int dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) { if (lrc->lrc_txtype == TX_WRITE) { zil_scan_arg_t *zsa = arg; dsl_pool_t *dp = zsa->zsa_dp; dsl_scan_t *scn = dp->dp_scan; zil_header_t *zh = zsa->zsa_zh; lr_write_t *lr = (lr_write_t *)lrc; blkptr_t *bp = &lr->lr_blkptr; zbookmark_t zb; if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) return (0); /* * birth can be < claim_txg if this record's txg is * already txg sync'ed (but this log block contains * other records that are not synced) */ if (claim_txg == 0 || bp->blk_birth < claim_txg) return (0); SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], lr->lr_foid, ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); } return (0); } static void dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh) { uint64_t claim_txg = zh->zh_claim_txg; zil_scan_arg_t zsa = { dp, zh }; zilog_t *zilog; /* * We only want to visit blocks that have been claimed but not yet * replayed (or, in read-only mode, blocks that *would* be claimed). */ if (claim_txg == 0 && spa_writeable(dp->dp_spa)) return; zilog = zil_alloc(dp->dp_meta_objset, zh); (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa, claim_txg); zil_free(zilog); } /* ARGSUSED */ static void dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp, uint64_t objset, uint64_t object, uint64_t blkid) { zbookmark_t czb; uint32_t flags = ARC_NOWAIT | ARC_PREFETCH; if (zfs_no_scrub_prefetch) return; if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg || (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)) return; SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid); /* * XXX need to make sure all of these arc_read() prefetches are * done before setting xlateall (similar to dsl_read()) */ (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp, buf, NULL, NULL, ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb); } static boolean_t dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, const zbookmark_t *zb) { /* * We never skip over user/group accounting objects (obj<0) */ if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark) && (int64_t)zb->zb_object >= 0) { /* * If we already visited this bp & everything below (in * a prior txg sync), don't bother doing it again. */ if (bookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark)) return (B_TRUE); /* * If we found the block we're trying to resume from, or * we went past it to a different object, zero it out to * indicate that it's OK to start checking for pausing * again. */ if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 || zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) { dprintf("resuming at %llx/%llx/%llx/%llx\n", (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object, (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid); bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb)); } } return (B_FALSE); } /* * Return nonzero on i/o error. * Return new buf to write out in *bufp. */ inline __attribute__((always_inline)) static int dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, const blkptr_t *bp, const zbookmark_t *zb, dmu_tx_t *tx, arc_buf_t **bufp) { dsl_pool_t *dp = scn->scn_dp; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; int err; if (BP_GET_LEVEL(bp) > 0) { uint32_t flags = ARC_WAIT; int i; blkptr_t *cbp; int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; err = arc_read_nolock(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) { dsl_scan_prefetch(scn, *bufp, cbp, zb->zb_objset, zb->zb_object, zb->zb_blkid * epb + i); } for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) { zbookmark_t czb; SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, zb->zb_level - 1, zb->zb_blkid * epb + i); dsl_scan_visitbp(cbp, &czb, dnp, *bufp, ds, scn, ostype, tx); } } else if (BP_GET_TYPE(bp) == DMU_OT_USERGROUP_USED) { uint32_t flags = ARC_WAIT; err = arc_read_nolock(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { uint32_t flags = ARC_WAIT; dnode_phys_t *cdnp; int i, j; int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; err = arc_read_nolock(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } for (i = 0, cdnp = (*bufp)->b_data; i < epb; i++, cdnp++) { for (j = 0; j < cdnp->dn_nblkptr; j++) { blkptr_t *cbp = &cdnp->dn_blkptr[j]; dsl_scan_prefetch(scn, *bufp, cbp, zb->zb_objset, zb->zb_blkid * epb + i, j); } } for (i = 0, cdnp = (*bufp)->b_data; i < epb; i++, cdnp++) { dsl_scan_visitdnode(scn, ds, ostype, cdnp, *bufp, zb->zb_blkid * epb + i, tx); } } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { uint32_t flags = ARC_WAIT; objset_phys_t *osp; err = arc_read_nolock(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); if (err) { scn->scn_phys.scn_errors++; return (err); } osp = (*bufp)->b_data; dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_meta_dnode, *bufp, DMU_META_DNODE_OBJECT, tx); if (OBJSET_BUF_HAS_USERUSED(*bufp)) { /* * We also always visit user/group accounting * objects, and never skip them, even if we are * pausing. This is necessary so that the space * deltas from this txg get integrated. */ dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_groupused_dnode, *bufp, DMU_GROUPUSED_OBJECT, tx); dsl_scan_visitdnode(scn, ds, osp->os_type, &osp->os_userused_dnode, *bufp, DMU_USERUSED_OBJECT, tx); } } return (0); } inline __attribute__((always_inline)) static void dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, dnode_phys_t *dnp, arc_buf_t *buf, uint64_t object, dmu_tx_t *tx) { int j; for (j = 0; j < dnp->dn_nblkptr; j++) { zbookmark_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, dnp->dn_nlevels - 1, j); dsl_scan_visitbp(&dnp->dn_blkptr[j], &czb, dnp, buf, ds, scn, ostype, tx); } if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { zbookmark_t czb; SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, 0, DMU_SPILL_BLKID); dsl_scan_visitbp(&dnp->dn_spill, &czb, dnp, buf, ds, scn, ostype, tx); } } /* * The arguments are in this order because mdb can only print the * first 5; we want them to be useful. */ static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_t *zb, dnode_phys_t *dnp, arc_buf_t *pbuf, dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; arc_buf_t *buf = NULL; blkptr_t *bp_toread; - bp_toread = kmem_alloc(sizeof (blkptr_t), KM_SLEEP); + bp_toread = kmem_alloc(sizeof (blkptr_t), KM_PUSHPAGE); *bp_toread = *bp; /* ASSERT(pbuf == NULL || arc_released(pbuf)); */ if (dsl_scan_check_pause(scn, zb)) goto out; if (dsl_scan_check_resume(scn, dnp, zb)) goto out; if (bp->blk_birth == 0) goto out; scn->scn_visited_this_txg++; /* * This debugging is commented out to conserve stack space. This * function is called recursively and the debugging addes several * bytes to the stack for each call. It can be commented back in * if required to debug an issue in dsl_scan_visitbp(). * * dprintf_bp(bp, * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx buf=%p bp=%p", * ds, ds ? ds->ds_object : 0, * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid, * pbuf, bp); */ if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) goto out; if (BP_GET_TYPE(bp) != DMU_OT_USERGROUP_USED) { /* * For non-user-accounting blocks, we need to read the * new bp (from a deleted snapshot, found in * check_existing_xlation). If we used the old bp, * pointers inside this block from before we resumed * would be untranslated. * * For user-accounting blocks, we need to read the old * bp, because we will apply the entire space delta to * it (original untranslated -> translations from * deleted snap -> now). */ *bp_toread = *bp; } if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx, &buf) != 0) goto out; /* * If dsl_scan_ddt() has aready visited this block, it will have * already done any translations or scrubbing, so don't call the * callback again. */ if (ddt_class_contains(dp->dp_spa, scn->scn_phys.scn_ddt_class_max, bp)) { ASSERT(buf == NULL); goto out; } /* * If this block is from the future (after cur_max_txg), then we * are doing this on behalf of a deleted snapshot, and we will * revisit the future block on the next pass of this dataset. * Don't scan it now unless we need to because something * under it was modified. */ if (bp->blk_birth <= scn->scn_phys.scn_cur_max_txg) { scan_funcs[scn->scn_phys.scn_func](dp, bp, zb); } if (buf) (void) arc_buf_remove_ref(buf, &buf); out: kmem_free(bp_toread, sizeof(blkptr_t)); } static void dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) { zbookmark_t zb; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); dsl_scan_visitbp(bp, &zb, NULL, NULL, ds, scn, DMU_OST_NONE, tx); dprintf_ds(ds, "finished scan%s", ""); } void dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { if (dsl_dataset_is_snapshot(ds)) { /* Note, scn_cur_{min,max}_txg stays the same. */ scn->scn_phys.scn_bookmark.zb_objset = ds->ds_phys->ds_next_snap_obj; zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)ds->ds_phys->ds_next_snap_obj); scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN; } else { SET_BOOKMARK(&scn->scn_phys.scn_bookmark, ZB_DESTROYED_OBJSET, 0, 0, 0); zfs_dbgmsg("destroying ds %llu; currently traversing; " "reset bookmark to -1,0,0,0", (u_longlong_t)ds->ds_object); } } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); if (dsl_dataset_is_snapshot(ds)) { /* * We keep the same mintxg; it could be > * ds_creation_txg if the previous snapshot was * deleted too. */ VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_phys->ds_next_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("destroying ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)ds->ds_phys->ds_next_snap_obj); } else { zfs_dbgmsg("destroying ds %llu; in queue; removing", (u_longlong_t)ds->ds_object); } } else { zfs_dbgmsg("destroying ds %llu; ignoring", (u_longlong_t)ds->ds_object); } /* * dsl_scan_sync() should be called after this, and should sync * out our changed state, but just to be safe, do it here. */ dsl_scan_sync_state(scn, tx); } void dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) { dsl_pool_t *dp = ds->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; ASSERT(ds->ds_phys->ds_prev_snap_obj != 0); if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = ds->ds_phys->ds_prev_snap_obj; zfs_dbgmsg("snapshotting ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)ds->ds_phys->ds_prev_snap_obj); } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_phys->ds_prev_snap_obj, mintxg, tx) == 0); zfs_dbgmsg("snapshotting ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds->ds_object, (u_longlong_t)ds->ds_phys->ds_prev_snap_obj); } dsl_scan_sync_state(scn, tx); } void dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx) { dsl_pool_t *dp = ds1->ds_dir->dd_pool; dsl_scan_t *scn = dp->dp_scan; uint64_t mintxg; if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) { scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object; zfs_dbgmsg("clone_swap ds %llu; currently traversing; " "reset zb_objset to %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, &mintxg) == 0) { int err; ASSERT3U(mintxg, ==, ds1->ds_phys->ds_prev_snap_txg); ASSERT3U(mintxg, ==, ds2->ds_phys->ds_prev_snap_txg); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, tx)); err = zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx); VERIFY(err == 0 || err == EEXIST); if (err == EEXIST) { /* Both were there to begin with */ VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx)); } zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds1->ds_object, (u_longlong_t)ds2->ds_object); } else if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) { ASSERT3U(mintxg, ==, ds1->ds_phys->ds_prev_snap_txg); ASSERT3U(mintxg, ==, ds2->ds_phys->ds_prev_snap_txg); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds2->ds_object, tx)); VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx)); zfs_dbgmsg("clone_swap ds %llu; in queue; " "replacing with %llu", (u_longlong_t)ds2->ds_object, (u_longlong_t)ds1->ds_object); } dsl_scan_sync_state(scn, tx); } struct enqueue_clones_arg { dmu_tx_t *tx; uint64_t originobj; }; /* ARGSUSED */ static int enqueue_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) { struct enqueue_clones_arg *eca = arg; dsl_dataset_t *ds; int err; dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); if (err) return (err); if (ds->ds_dir->dd_phys->dd_origin_obj == eca->originobj) { while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); dsl_dataset_rele(ds, FTAG); if (err) return (err); ds = prev; } VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0); } dsl_dataset_rele(ds, FTAG); return (0); } static void dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; dsl_dataset_t *ds; objset_t *os; char *dsname; VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (dmu_objset_from_ds(ds, &os)) goto out; /* * Only the ZIL in the head (non-snapshot) is valid. Even though * snapshots can have ZIL block pointers (which may be the same * BP as in the head), they must be ignored. So we traverse the * ZIL here, rather than in scan_recurse(), because the regular * snapshot block-sharing rules don't apply to it. */ if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !dsl_dataset_is_snapshot(ds)) dsl_scan_zil(dp, &os->os_zil_header); /* * Iterate over the bps in this ds. */ dmu_buf_will_dirty(ds->ds_dbuf, tx); dsl_scan_visit_rootbp(scn, ds, &ds->ds_phys->ds_bp, tx); - dsname = kmem_alloc(ZFS_MAXNAMELEN, KM_SLEEP); + dsname = kmem_alloc(ZFS_MAXNAMELEN, KM_PUSHPAGE); dsl_dataset_name(ds, dsname); zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; " "pausing=%u", (longlong_t)dsobj, dsname, (longlong_t)scn->scn_phys.scn_cur_min_txg, (longlong_t)scn->scn_phys.scn_cur_max_txg, (int)scn->scn_pausing); kmem_free(dsname, ZFS_MAXNAMELEN); if (scn->scn_pausing) goto out; /* * We've finished this pass over this dataset. */ /* * If we did not completely visit this dataset, do another pass. */ if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) { zfs_dbgmsg("incomplete pass; visiting again"); scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN; VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, scn->scn_phys.scn_cur_max_txg, tx) == 0); goto out; } /* * Add descendent datasets to work queue. */ if (ds->ds_phys->ds_next_snap_obj != 0) { VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_phys->ds_next_snap_obj, ds->ds_phys->ds_creation_txg, tx) == 0); } if (ds->ds_phys->ds_num_children > 1) { boolean_t usenext = B_FALSE; if (ds->ds_phys->ds_next_clones_obj != 0) { uint64_t count; /* * A bug in a previous version of the code could * cause upgrade_clones_cb() to not set * ds_next_snap_obj when it should, leading to a * missing entry. Therefore we can only use the * next_clones_obj when its count is correct. */ int err = zap_count(dp->dp_meta_objset, ds->ds_phys->ds_next_clones_obj, &count); if (err == 0 && count == ds->ds_phys->ds_num_children - 1) usenext = B_TRUE; } if (usenext) { VERIFY(zap_join_key(dp->dp_meta_objset, ds->ds_phys->ds_next_clones_obj, scn->scn_phys.scn_queue_obj, ds->ds_phys->ds_creation_txg, tx) == 0); } else { struct enqueue_clones_arg eca; eca.tx = tx; eca.originobj = ds->ds_object; (void) dmu_objset_find_spa(ds->ds_dir->dd_pool->dp_spa, NULL, enqueue_clones_cb, &eca, DS_FIND_CHILDREN); } } out: dsl_dataset_rele(ds, FTAG); } /* ARGSUSED */ static int enqueue_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) { dmu_tx_t *tx = arg; dsl_dataset_t *ds; int err; dsl_pool_t *dp = spa->spa_dsl_pool; dsl_scan_t *scn = dp->dp_scan; err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); if (err) return (err); while (ds->ds_phys->ds_prev_snap_obj != 0) { dsl_dataset_t *prev; err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); if (err) { dsl_dataset_rele(ds, FTAG); return (err); } /* * If this is a clone, we don't need to worry about it for now. */ if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) { dsl_dataset_rele(ds, FTAG); dsl_dataset_rele(prev, FTAG); return (0); } dsl_dataset_rele(ds, FTAG); ds = prev; } VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, ds->ds_object, ds->ds_phys->ds_prev_snap_txg, tx) == 0); dsl_dataset_rele(ds, FTAG); return (0); } /* * Scrub/dedup interaction. * * If there are N references to a deduped block, we don't want to scrub it * N times -- ideally, we should scrub it exactly once. * * We leverage the fact that the dde's replication class (enum ddt_class) * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order. * * To prevent excess scrubbing, the scrub begins by walking the DDT * to find all blocks with refcnt > 1, and scrubs each of these once. * Since there are two replication classes which contain blocks with * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first. * Finally the top-down scrub begins, only visiting blocks with refcnt == 1. * * There would be nothing more to say if a block's refcnt couldn't change * during a scrub, but of course it can so we must account for changes * in a block's replication class. * * Here's an example of what can occur: * * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1 * when visited during the top-down scrub phase, it will be scrubbed twice. * This negates our scrub optimization, but is otherwise harmless. * * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1 * on each visit during the top-down scrub phase, it will never be scrubbed. * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1 * while a scrub is in progress, it scrubs the block right then. */ static void dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx) { ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark; ddt_entry_t dde; int error; uint64_t n = 0; bzero(&dde, sizeof (ddt_entry_t)); while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) { ddt_t *ddt; if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max) break; dprintf("visiting ddb=%llu/%llu/%llu/%llx\n", (longlong_t)ddb->ddb_class, (longlong_t)ddb->ddb_type, (longlong_t)ddb->ddb_checksum, (longlong_t)ddb->ddb_cursor); /* There should be no pending changes to the dedup table */ ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum]; ASSERT(avl_first(&ddt->ddt_tree) == NULL); dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx); n++; if (dsl_scan_check_pause(scn, NULL)) break; } zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u", (longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_pausing); ASSERT(error == 0 || error == ENOENT); ASSERT(error != ENOENT || ddb->ddb_class > scn->scn_phys.scn_ddt_class_max); } /* ARGSUSED */ void dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, ddt_entry_t *dde, dmu_tx_t *tx) { const ddt_key_t *ddk = &dde->dde_key; ddt_phys_t *ddp = dde->dde_phys; blkptr_t bp; zbookmark_t zb = { 0 }; int p; if (scn->scn_phys.scn_state != DSS_SCANNING) return; for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp->ddp_phys_birth > scn->scn_phys.scn_cur_max_txg) continue; ddt_bp_create(checksum, ddk, ddp, &bp); scn->scn_visited_this_txg++; scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb); } } static void dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) { dsl_pool_t *dp = scn->scn_dp; zap_cursor_t *zc; zap_attribute_t *za; if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_ddt(scn, tx); if (scn->scn_pausing) return; } if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) { /* First do the MOS & ORIGIN */ scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; dsl_scan_visit_rootbp(scn, NULL, &dp->dp_meta_rootbp, tx); spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); if (scn->scn_pausing) return; if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { VERIFY(0 == dmu_objset_find_spa(dp->dp_spa, NULL, enqueue_cb, tx, DS_FIND_CHILDREN)); } else { dsl_scan_visitds(scn, dp->dp_origin_snap->ds_object, tx); } ASSERT(!scn->scn_pausing); } else if (scn->scn_phys.scn_bookmark.zb_objset != ZB_DESTROYED_OBJSET) { /* * If we were paused, continue from here. Note if the * ds we were paused on was deleted, the zb_objset may * be -1, so we will skip this and find a new objset * below. */ dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx); if (scn->scn_pausing) return; } /* * In case we were paused right at the end of the ds, zero the * bookmark so we don't think that we're still trying to resume. */ bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_t)); - zc = kmem_alloc(sizeof(zap_cursor_t), KM_SLEEP); - za = kmem_alloc(sizeof(zap_attribute_t), KM_SLEEP); + zc = kmem_alloc(sizeof(zap_cursor_t), KM_PUSHPAGE); + za = kmem_alloc(sizeof(zap_attribute_t), KM_PUSHPAGE); /* keep pulling things out of the zap-object-as-queue */ while (zap_cursor_init(zc, dp->dp_meta_objset, scn->scn_phys.scn_queue_obj), zap_cursor_retrieve(zc, za) == 0) { dsl_dataset_t *ds; uint64_t dsobj; dsobj = strtonum(za->za_name, NULL); VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, dsobj, tx)); /* Set up min/max txg */ VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); if (za->za_first_integer != 0) { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, za->za_first_integer); } else { scn->scn_phys.scn_cur_min_txg = MAX(scn->scn_phys.scn_min_txg, ds->ds_phys->ds_prev_snap_txg); } scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds); dsl_dataset_rele(ds, FTAG); dsl_scan_visitds(scn, dsobj, tx); zap_cursor_fini(zc); if (scn->scn_pausing) goto out; } zap_cursor_fini(zc); out: kmem_free(za, sizeof(zap_attribute_t)); kmem_free(zc, sizeof(zap_cursor_t)); } static int dsl_scan_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { dsl_scan_t *scn = arg; uint64_t elapsed_nanosecs; elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || (elapsed_nanosecs / MICROSEC > zfs_free_min_time_ms && txg_sync_waiting(scn->scn_dp)) || spa_shutting_down(scn->scn_dp->dp_spa)) return (ERESTART); zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, dmu_tx_get_txg(tx), bp, 0)); dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp), -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); scn->scn_visited_this_txg++; return (0); } boolean_t dsl_scan_active(dsl_scan_t *scn) { spa_t *spa = scn->scn_dp->dp_spa; uint64_t used = 0, comp, uncomp; if (spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); if (spa_shutting_down(spa)) return (B_FALSE); if (scn->scn_phys.scn_state == DSS_SCANNING) return (B_TRUE); if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) { (void) bpobj_space(&scn->scn_dp->dp_free_bpobj, &used, &comp, &uncomp); } return (used != 0); } void dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) { dsl_scan_t *scn = dp->dp_scan; spa_t *spa = dp->dp_spa; int err; /* * Check for scn_restart_txg before checking spa_load_state, so * that we can restart an old-style scan while the pool is being * imported (see dsl_scan_init). */ if (scn->scn_restart_txg != 0 && scn->scn_restart_txg <= tx->tx_txg) { pool_scan_func_t func = POOL_SCAN_SCRUB; dsl_scan_done(scn, B_FALSE, tx); if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) func = POOL_SCAN_RESILVER; zfs_dbgmsg("restarting scan func=%u txg=%llu", func, tx->tx_txg); dsl_scan_setup_sync(scn, &func, tx); } if (!dsl_scan_active(scn) || spa_sync_pass(dp->dp_spa) > 1) return; scn->scn_visited_this_txg = 0; scn->scn_pausing = B_FALSE; scn->scn_sync_start_time = gethrtime(); spa->spa_scrub_active = B_TRUE; /* * First process the free list. If we pause the free, don't do * any scanning. This ensures that there is no free list when * we are scanning, so the scan code doesn't have to worry about * traversing it. */ if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); err = bpobj_iterate(&dp->dp_free_bpobj, dsl_scan_free_cb, scn, tx); VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); if (scn->scn_visited_this_txg) { zfs_dbgmsg("freed %llu blocks in %llums from " "free_bpobj txg %llu", (longlong_t)scn->scn_visited_this_txg, (longlong_t) (gethrtime() - scn->scn_sync_start_time) / MICROSEC, (longlong_t)tx->tx_txg); scn->scn_visited_this_txg = 0; /* * Re-sync the ddt so that we can further modify * it when doing bprewrite. */ ddt_sync(spa, tx->tx_txg); } if (err == ERESTART) return; } if (scn->scn_phys.scn_state != DSS_SCANNING) return; if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= scn->scn_phys.scn_ddt_class_max) { zfs_dbgmsg("doing scan sync txg %llu; " "ddt bm=%llu/%llu/%llu/%llx", (longlong_t)tx->tx_txg, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0); ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0); } else { zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu", (longlong_t)tx->tx_txg, (longlong_t)scn->scn_phys.scn_bookmark.zb_objset, (longlong_t)scn->scn_phys.scn_bookmark.zb_object, (longlong_t)scn->scn_phys.scn_bookmark.zb_level, (longlong_t)scn->scn_phys.scn_bookmark.zb_blkid); } scn->scn_zio_root = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_CANFAIL); dsl_scan_visit(scn, tx); (void) zio_wait(scn->scn_zio_root); scn->scn_zio_root = NULL; zfs_dbgmsg("visited %llu blocks in %llums", (longlong_t)scn->scn_visited_this_txg, (longlong_t)(gethrtime() - scn->scn_sync_start_time) / MICROSEC); if (!scn->scn_pausing) { /* finished with scan. */ zfs_dbgmsg("finished scan txg %llu", (longlong_t)tx->tx_txg); dsl_scan_done(scn, B_TRUE, tx); } if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight > 0) { cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); } mutex_exit(&spa->spa_scrub_lock); } dsl_scan_sync_state(scn, tx); } /* * This will start a new scan, or restart an existing one. */ void dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg) { if (txg == 0) { dmu_tx_t *tx; tx = dmu_tx_create_dd(dp->dp_mos_dir); VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); txg = dmu_tx_get_txg(tx); dp->dp_scan->scn_restart_txg = txg; dmu_tx_commit(tx); } else { dp->dp_scan->scn_restart_txg = txg; } zfs_dbgmsg("restarting resilver txg=%llu", txg); } boolean_t dsl_scan_resilvering(dsl_pool_t *dp) { return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING && dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER); } /* * scrub consumers */ static void count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) { int i; /* * If we resume after a reboot, zab will be NULL; don't record * incomplete stats in that case. */ if (zab == NULL) return; for (i = 0; i < 4; i++) { int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS; int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL; zfs_blkstat_t *zb = &zab->zab_type[l][t]; int equal; zb->zb_count++; zb->zb_asize += BP_GET_ASIZE(bp); zb->zb_lsize += BP_GET_LSIZE(bp); zb->zb_psize += BP_GET_PSIZE(bp); zb->zb_gangs += BP_COUNT_GANG(bp); switch (BP_GET_NDVAS(bp)) { case 2: if (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) zb->zb_ditto_2_of_2_samevdev++; break; case 3: equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[1])) + (DVA_GET_VDEV(&bp->blk_dva[0]) == DVA_GET_VDEV(&bp->blk_dva[2])) + (DVA_GET_VDEV(&bp->blk_dva[1]) == DVA_GET_VDEV(&bp->blk_dva[2])); if (equal == 1) zb->zb_ditto_2_of_3_samevdev++; else if (equal == 3) zb->zb_ditto_3_of_3_samevdev++; break; } } } static void dsl_scan_scrub_done(zio_t *zio) { spa_t *spa = zio->io_spa; zio_data_buf_free(zio->io_data, zio->io_size); mutex_enter(&spa->spa_scrub_lock); spa->spa_scrub_inflight--; cv_broadcast(&spa->spa_scrub_io_cv); if (zio->io_error && (zio->io_error != ECKSUM || !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) { spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++; } mutex_exit(&spa->spa_scrub_lock); } static int dsl_scan_scrub_cb(dsl_pool_t *dp, const blkptr_t *bp, const zbookmark_t *zb) { dsl_scan_t *scn = dp->dp_scan; size_t size = BP_GET_PSIZE(bp); spa_t *spa = dp->dp_spa; uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp); boolean_t needs_io = B_FALSE; int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL; int zio_priority = 0; int scan_delay = 0; int d; if (phys_birth <= scn->scn_phys.scn_min_txg || phys_birth >= scn->scn_phys.scn_max_txg) return (0); count_block(dp->dp_blkstats, bp); ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn)); if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) { zio_flags |= ZIO_FLAG_SCRUB; zio_priority = ZIO_PRIORITY_SCRUB; needs_io = B_TRUE; scan_delay = zfs_scrub_delay; } else if (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) { zio_flags |= ZIO_FLAG_RESILVER; zio_priority = ZIO_PRIORITY_RESILVER; needs_io = B_FALSE; scan_delay = zfs_resilver_delay; } /* If it's an intent log block, failure is expected. */ if (zb->zb_level == ZB_ZIL_LEVEL) zio_flags |= ZIO_FLAG_SPECULATIVE; for (d = 0; d < BP_GET_NDVAS(bp); d++) { vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[d])); /* * Keep track of how much data we've examined so that * zpool(1M) status can make useful progress reports. */ scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]); spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]); /* if it's a resilver, this may not be in the target range */ if (!needs_io) { if (DVA_GET_GANG(&bp->blk_dva[d])) { /* * Gang members may be spread across multiple * vdevs, so the best estimate we have is the * scrub range, which has already been checked. * XXX -- it would be better to change our * allocation policy to ensure that all * gang members reside on the same vdev. */ needs_io = B_TRUE; } else { needs_io = vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1); } } } if (needs_io && !zfs_no_scrub_io) { vdev_t *rvd = spa->spa_root_vdev; uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight; void *data = zio_data_buf_alloc(size); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight >= maxinflight) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_scrub_inflight++; mutex_exit(&spa->spa_scrub_lock); /* * If we're seeing recent (zfs_scan_idle) "important" I/Os * then throttle our workload to limit the impact of a scan. */ if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle) delay(scan_delay); zio_nowait(zio_read(NULL, spa, bp, data, size, dsl_scan_scrub_done, NULL, zio_priority, zio_flags, zb)); } /* do not relocate this block */ return (0); } int dsl_scan(dsl_pool_t *dp, pool_scan_func_t func) { spa_t *spa = dp->dp_spa; /* * Purge all vdev caches and probe all devices. We do this here * rather than in sync context because this requires a writer lock * on the spa_config lock, which we can't do from sync context. The * spa_scrub_reopen flag indicates that vdev_open() should not * attempt to start another scrub. */ spa_vdev_state_enter(spa, SCL_NONE); spa->spa_scrub_reopen = B_TRUE; vdev_reopen(spa->spa_root_vdev); spa->spa_scrub_reopen = B_FALSE; (void) spa_vdev_state_exit(spa, NULL, 0); return (dsl_sync_task_do(dp, dsl_scan_setup_check, dsl_scan_setup_sync, dp->dp_scan, &func, 0)); } #if defined(_KERNEL) && defined(HAVE_SPL) module_param(zfs_top_maxinflight, int, 0644); MODULE_PARM_DESC(zfs_top_maxinflight, "Max I/Os per top-level"); module_param(zfs_resilver_delay, int, 0644); MODULE_PARM_DESC(zfs_resilver_delay, "Number of ticks to delay resilver"); module_param(zfs_scrub_delay, int, 0644); MODULE_PARM_DESC(zfs_scrub_delay, "Number of ticks to delay scrub"); module_param(zfs_scan_idle, int, 0644); MODULE_PARM_DESC(zfs_scan_idle, "Idle window in clock ticks"); module_param(zfs_scan_min_time_ms, int, 0644); MODULE_PARM_DESC(zfs_scan_min_time_ms, "Min millisecs to scrub per txg"); module_param(zfs_free_min_time_ms, int, 0644); MODULE_PARM_DESC(zfs_free_min_time_ms, "Min millisecs to free per txg"); module_param(zfs_resilver_min_time_ms, int, 0644); MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg"); module_param(zfs_no_scrub_io, int, 0644); MODULE_PARM_DESC(zfs_no_scrub_io, "Set to disable scrub I/O"); module_param(zfs_no_scrub_prefetch, int, 0644); MODULE_PARM_DESC(zfs_no_scrub_prefetch, "Set to disable scrub prefetching"); module_param(zfs_txg_timeout, int, 0644); MODULE_PARM_DESC(zfs_txg_timeout, "Max seconds worth of delta per txg"); #endif diff --git a/module/zfs/fm.c b/module/zfs/fm.c index b91516e7aa9f..ce0ebe0c1a30 100644 --- a/module/zfs/fm.c +++ b/module/zfs/fm.c @@ -1,1553 +1,1553 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. */ /* * Fault Management Architecture (FMA) Resource and Protocol Support * * The routines contained herein provide services to support kernel subsystems * in publishing fault management telemetry (see PSARC 2002/412 and 2003/089). * * Name-Value Pair Lists * * The embodiment of an FMA protocol element (event, fmri or authority) is a * name-value pair list (nvlist_t). FMA-specific nvlist construtor and * destructor functions, fm_nvlist_create() and fm_nvlist_destroy(), are used * to create an nvpair list using custom allocators. Callers may choose to * allocate either from the kernel memory allocator, or from a preallocated * buffer, useful in constrained contexts like high-level interrupt routines. * * Protocol Event and FMRI Construction * * Convenience routines are provided to construct nvlist events according to * the FMA Event Protocol and Naming Schema specification for ereports and * FMRIs for the dev, cpu, hc, mem, legacy hc and de schemes. * * ENA Manipulation * * Routines to generate ENA formats 0, 1 and 2 are available as well as * routines to increment formats 1 and 2. Individual fields within the * ENA are extractable via fm_ena_time_get(), fm_ena_id_get(), * fm_ena_format_get() and fm_ena_gen_get(). */ #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #include #include #include #include #include #include #include int zfs_zevent_len_max = 0; int zfs_zevent_cols = 80; int zfs_zevent_console = 0; static int zevent_len_cur = 0; static int zevent_waiters = 0; static int zevent_flags = 0; static kmutex_t zevent_lock; static list_t zevent_list; static kcondvar_t zevent_cv; #endif /* _KERNEL */ extern void fastreboot_disable_highpil(void); /* * Common fault management kstats to record event generation failures */ struct erpt_kstat { kstat_named_t erpt_dropped; /* num erpts dropped on post */ kstat_named_t erpt_set_failed; /* num erpt set failures */ kstat_named_t fmri_set_failed; /* num fmri set failures */ kstat_named_t payload_set_failed; /* num payload set failures */ }; static struct erpt_kstat erpt_kstat_data = { { "erpt-dropped", KSTAT_DATA_UINT64 }, { "erpt-set-failed", KSTAT_DATA_UINT64 }, { "fmri-set-failed", KSTAT_DATA_UINT64 }, { "payload-set-failed", KSTAT_DATA_UINT64 } }; kstat_t *fm_ksp; #ifdef _KERNEL /* * Formatting utility function for fm_nvprintr. We attempt to wrap chunks of * output so they aren't split across console lines, and return the end column. */ /*PRINTFLIKE4*/ static int fm_printf(int depth, int c, int cols, const char *format, ...) { va_list ap; int width; char c1; va_start(ap, format); width = vsnprintf(&c1, sizeof (c1), format, ap); va_end(ap); if (c + width >= cols) { console_printf("\n"); c = 0; if (format[0] != ' ' && depth > 0) { console_printf(" "); c++; } } va_start(ap, format); console_vprintf(format, ap); va_end(ap); return ((c + width) % cols); } /* * Recursively print a nvlist in the specified column width and return the * column we end up in. This function is called recursively by fm_nvprint(), * below. We generically format the entire nvpair using hexadecimal * integers and strings, and elide any integer arrays. Arrays are basically * used for cache dumps right now, so we suppress them so as not to overwhelm * the amount of console output we produce at panic time. This can be further * enhanced as FMA technology grows based upon the needs of consumers. All * FMA telemetry is logged using the dump device transport, so the console * output serves only as a fallback in case this procedure is unsuccessful. */ static int fm_nvprintr(nvlist_t *nvl, int d, int c, int cols) { nvpair_t *nvp; for (nvp = nvlist_next_nvpair(nvl, NULL); nvp != NULL; nvp = nvlist_next_nvpair(nvl, nvp)) { data_type_t type = nvpair_type(nvp); const char *name = nvpair_name(nvp); boolean_t b; uint8_t i8; uint16_t i16; uint32_t i32; uint64_t i64; char *str; nvlist_t *cnv; if (strcmp(name, FM_CLASS) == 0) continue; /* already printed by caller */ c = fm_printf(d, c, cols, " %s=", name); switch (type) { case DATA_TYPE_BOOLEAN: c = fm_printf(d + 1, c, cols, " 1"); break; case DATA_TYPE_BOOLEAN_VALUE: (void) nvpair_value_boolean_value(nvp, &b); c = fm_printf(d + 1, c, cols, b ? "1" : "0"); break; case DATA_TYPE_BYTE: (void) nvpair_value_byte(nvp, &i8); c = fm_printf(d + 1, c, cols, "0x%x", i8); break; case DATA_TYPE_INT8: (void) nvpair_value_int8(nvp, (void *)&i8); c = fm_printf(d + 1, c, cols, "0x%x", i8); break; case DATA_TYPE_UINT8: (void) nvpair_value_uint8(nvp, &i8); c = fm_printf(d + 1, c, cols, "0x%x", i8); break; case DATA_TYPE_INT16: (void) nvpair_value_int16(nvp, (void *)&i16); c = fm_printf(d + 1, c, cols, "0x%x", i16); break; case DATA_TYPE_UINT16: (void) nvpair_value_uint16(nvp, &i16); c = fm_printf(d + 1, c, cols, "0x%x", i16); break; case DATA_TYPE_INT32: (void) nvpair_value_int32(nvp, (void *)&i32); c = fm_printf(d + 1, c, cols, "0x%x", i32); break; case DATA_TYPE_UINT32: (void) nvpair_value_uint32(nvp, &i32); c = fm_printf(d + 1, c, cols, "0x%x", i32); break; case DATA_TYPE_INT64: (void) nvpair_value_int64(nvp, (void *)&i64); c = fm_printf(d + 1, c, cols, "0x%llx", (u_longlong_t)i64); break; case DATA_TYPE_UINT64: (void) nvpair_value_uint64(nvp, &i64); c = fm_printf(d + 1, c, cols, "0x%llx", (u_longlong_t)i64); break; case DATA_TYPE_HRTIME: (void) nvpair_value_hrtime(nvp, (void *)&i64); c = fm_printf(d + 1, c, cols, "0x%llx", (u_longlong_t)i64); break; case DATA_TYPE_STRING: (void) nvpair_value_string(nvp, &str); c = fm_printf(d + 1, c, cols, "\"%s\"", str ? str : ""); break; case DATA_TYPE_NVLIST: c = fm_printf(d + 1, c, cols, "["); (void) nvpair_value_nvlist(nvp, &cnv); c = fm_nvprintr(cnv, d + 1, c, cols); c = fm_printf(d + 1, c, cols, " ]"); break; case DATA_TYPE_NVLIST_ARRAY: { nvlist_t **val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "["); (void) nvpair_value_nvlist_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) { c = fm_nvprintr(val[i], d + 1, c, cols); } c = fm_printf(d + 1, c, cols, " ]"); } break; case DATA_TYPE_INT8_ARRAY: { int8_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_int8_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_UINT8_ARRAY: { uint8_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_uint8_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_INT16_ARRAY: { int16_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_int16_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_UINT16_ARRAY: { uint16_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_uint16_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_INT32_ARRAY: { int32_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_int32_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_UINT32_ARRAY: { uint32_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_uint32_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_INT64_ARRAY: { int64_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_int64_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_UINT64_ARRAY: { uint64_t *val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "[ "); (void) nvpair_value_uint64_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) c = fm_printf(d + 1, c, cols, "0x%llx ", (u_longlong_t)val[i]); c = fm_printf(d + 1, c, cols, "]"); break; } case DATA_TYPE_STRING_ARRAY: case DATA_TYPE_BOOLEAN_ARRAY: case DATA_TYPE_BYTE_ARRAY: c = fm_printf(d + 1, c, cols, "[...]"); break; case DATA_TYPE_UNKNOWN: c = fm_printf(d + 1, c, cols, ""); break; } } return (c); } void fm_nvprint(nvlist_t *nvl) { char *class; int c = 0; console_printf("\n"); if (nvlist_lookup_string(nvl, FM_CLASS, &class) == 0) c = fm_printf(0, c, zfs_zevent_cols, "%s", class); if (fm_nvprintr(nvl, 0, c, zfs_zevent_cols) != 0) console_printf("\n"); console_printf("\n"); } static zevent_t * zfs_zevent_alloc(void) { zevent_t *ev; - ev = kmem_zalloc(sizeof(zevent_t), KM_SLEEP); + ev = kmem_zalloc(sizeof(zevent_t), KM_PUSHPAGE); if (ev == NULL) return NULL; list_create(&ev->ev_ze_list, sizeof(zfs_zevent_t), offsetof(zfs_zevent_t, ze_node)); list_link_init(&ev->ev_node); return ev; } static void zfs_zevent_free(zevent_t *ev) { /* Run provided cleanup callback */ ev->ev_cb(ev->ev_nvl, ev->ev_detector); list_destroy(&ev->ev_ze_list); kmem_free(ev, sizeof(zevent_t)); } static void zfs_zevent_drain(zevent_t *ev) { zfs_zevent_t *ze; ASSERT(MUTEX_HELD(&zevent_lock)); list_remove(&zevent_list, ev); /* Remove references to this event in all private file data */ while ((ze = list_head(&ev->ev_ze_list)) != NULL) { list_remove(&ev->ev_ze_list, ze); ze->ze_zevent = NULL; ze->ze_dropped++; } zfs_zevent_free(ev); } void zfs_zevent_drain_all(int *count) { zevent_t *ev; mutex_enter(&zevent_lock); while ((ev = list_head(&zevent_list)) != NULL) zfs_zevent_drain(ev); *count = zevent_len_cur; zevent_len_cur = 0; mutex_exit(&zevent_lock); } /* * New zevents are inserted at the head. If the maximum queue * length is exceeded a zevent will be drained from the tail. * As part of this any user space processes which currently have * a reference to this zevent_t in their private data will have * this reference set to NULL. */ static void zfs_zevent_insert(zevent_t *ev) { mutex_enter(&zevent_lock); list_insert_head(&zevent_list, ev); if (zevent_len_cur >= zfs_zevent_len_max) zfs_zevent_drain(list_tail(&zevent_list)); else zevent_len_cur++; mutex_exit(&zevent_lock); } /* * Post a zevent */ void zfs_zevent_post(nvlist_t *nvl, nvlist_t *detector, zevent_cb_t *cb) { int64_t tv_array[2]; timestruc_t tv; size_t nvl_size = 0; zevent_t *ev; gethrestime(&tv); tv_array[0] = tv.tv_sec; tv_array[1] = tv.tv_nsec; if (nvlist_add_int64_array(nvl, FM_EREPORT_TIME, tv_array, 2)) { atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); return; } (void) nvlist_size(nvl, &nvl_size, NV_ENCODE_NATIVE); if (nvl_size > ERPT_DATA_SZ || nvl_size == 0) { atomic_add_64(&erpt_kstat_data.erpt_dropped.value.ui64, 1); return; } if (zfs_zevent_console) fm_nvprint(nvl); ev = zfs_zevent_alloc(); if (ev == NULL) { atomic_add_64(&erpt_kstat_data.erpt_dropped.value.ui64, 1); return; } ev->ev_nvl = nvl; ev->ev_detector = detector; ev->ev_cb = cb; zfs_zevent_insert(ev); cv_broadcast(&zevent_cv); } static int zfs_zevent_minor_to_state(minor_t minor, zfs_zevent_t **ze) { *ze = zfsdev_get_state(minor, ZST_ZEVENT); if (*ze == NULL) return (EBADF); return (0); } int zfs_zevent_fd_hold(int fd, minor_t *minorp, zfs_zevent_t **ze) { file_t *fp; int error; fp = getf(fd); if (fp == NULL) return (EBADF); *minorp = zfsdev_getminor(fp->f_file); error = zfs_zevent_minor_to_state(*minorp, ze); if (error) zfs_zevent_fd_rele(fd); return (error); } void zfs_zevent_fd_rele(int fd) { releasef(fd); } /* * Get the next zevent in the stream and place a copy in 'event'. This * may fail with ENOMEM if the encoded nvlist size exceeds the passed * 'event_size'. In this case the stream pointer is not advanced and * and 'event_size' is set to the minimum required buffer size. */ int zfs_zevent_next(zfs_zevent_t *ze, nvlist_t **event, uint64_t *event_size, uint64_t *dropped) { zevent_t *ev; size_t size; int error = 0; mutex_enter(&zevent_lock); if (ze->ze_zevent == NULL) { /* New stream start at the beginning/tail */ ev = list_tail(&zevent_list); if (ev == NULL) { error = ENOENT; goto out; } } else { /* Existing stream continue with the next element and remove * ourselves from the wait queue for the previous element */ ev = list_prev(&zevent_list, ze->ze_zevent); if (ev == NULL) { error = ENOENT; goto out; } } VERIFY(nvlist_size(ev->ev_nvl, &size, NV_ENCODE_NATIVE) == 0); if (size > *event_size) { *event_size = size; error = ENOMEM; goto out; } if (ze->ze_zevent) list_remove(&ze->ze_zevent->ev_ze_list, ze); ze->ze_zevent = ev; list_insert_head(&ev->ev_ze_list, ze); nvlist_dup(ev->ev_nvl, event, KM_SLEEP); *dropped = ze->ze_dropped; ze->ze_dropped = 0; out: mutex_exit(&zevent_lock); return error; } int zfs_zevent_wait(zfs_zevent_t *ze) { int error = 0; mutex_enter(&zevent_lock); if (zevent_flags & ZEVENT_SHUTDOWN) { error = ESHUTDOWN; goto out; } zevent_waiters++; cv_wait_interruptible(&zevent_cv, &zevent_lock); if (issig(JUSTLOOKING)) error = EINTR; zevent_waiters--; out: mutex_exit(&zevent_lock); return error; } void zfs_zevent_init(zfs_zevent_t **zep) { zfs_zevent_t *ze; ze = *zep = kmem_zalloc(sizeof (zfs_zevent_t), KM_SLEEP); list_link_init(&ze->ze_node); } void zfs_zevent_destroy(zfs_zevent_t *ze) { mutex_enter(&zevent_lock); if (ze->ze_zevent) list_remove(&ze->ze_zevent->ev_ze_list, ze); mutex_exit(&zevent_lock); kmem_free(ze, sizeof (zfs_zevent_t)); } #endif /* _KERNEL */ /* * Wrapppers for FM nvlist allocators */ /* ARGSUSED */ static void * i_fm_alloc(nv_alloc_t *nva, size_t size) { return (kmem_zalloc(size, KM_PUSHPAGE)); } /* ARGSUSED */ static void i_fm_free(nv_alloc_t *nva, void *buf, size_t size) { kmem_free(buf, size); } const nv_alloc_ops_t fm_mem_alloc_ops = { NULL, NULL, i_fm_alloc, i_fm_free, NULL }; /* * Create and initialize a new nv_alloc_t for a fixed buffer, buf. A pointer * to the newly allocated nv_alloc_t structure is returned upon success or NULL * is returned to indicate that the nv_alloc structure could not be created. */ nv_alloc_t * fm_nva_xcreate(char *buf, size_t bufsz) { nv_alloc_t *nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_SLEEP); if (bufsz == 0 || nv_alloc_init(nvhdl, nv_fixed_ops, buf, bufsz) != 0) { kmem_free(nvhdl, sizeof (nv_alloc_t)); return (NULL); } return (nvhdl); } /* * Destroy a previously allocated nv_alloc structure. The fixed buffer * associated with nva must be freed by the caller. */ void fm_nva_xdestroy(nv_alloc_t *nva) { nv_alloc_fini(nva); kmem_free(nva, sizeof (nv_alloc_t)); } /* * Create a new nv list. A pointer to a new nv list structure is returned * upon success or NULL is returned to indicate that the structure could * not be created. The newly created nv list is created and managed by the * operations installed in nva. If nva is NULL, the default FMA nva * operations are installed and used. * * When called from the kernel and nva == NULL, this function must be called * from passive kernel context with no locks held that can prevent a * sleeping memory allocation from occurring. Otherwise, this function may * be called from other kernel contexts as long a valid nva created via * fm_nva_create() is supplied. */ nvlist_t * fm_nvlist_create(nv_alloc_t *nva) { int hdl_alloced = 0; nvlist_t *nvl; nv_alloc_t *nvhdl; if (nva == NULL) { nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_PUSHPAGE); if (nv_alloc_init(nvhdl, &fm_mem_alloc_ops, NULL, 0) != 0) { kmem_free(nvhdl, sizeof (nv_alloc_t)); return (NULL); } hdl_alloced = 1; } else { nvhdl = nva; } if (nvlist_xalloc(&nvl, NV_UNIQUE_NAME, nvhdl) != 0) { if (hdl_alloced) { nv_alloc_fini(nvhdl); kmem_free(nvhdl, sizeof (nv_alloc_t)); } return (NULL); } return (nvl); } /* * Destroy a previously allocated nvlist structure. flag indicates whether * or not the associated nva structure should be freed (FM_NVA_FREE) or * retained (FM_NVA_RETAIN). Retaining the nv alloc structure allows * it to be re-used for future nvlist creation operations. */ void fm_nvlist_destroy(nvlist_t *nvl, int flag) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(nvl); nvlist_free(nvl); if (nva != NULL) { if (flag == FM_NVA_FREE) fm_nva_xdestroy(nva); } } int i_fm_payload_set(nvlist_t *payload, const char *name, va_list ap) { int nelem, ret = 0; data_type_t type; while (ret == 0 && name != NULL) { type = va_arg(ap, data_type_t); switch (type) { case DATA_TYPE_BYTE: ret = nvlist_add_byte(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_BYTE_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_byte_array(payload, name, va_arg(ap, uchar_t *), nelem); break; case DATA_TYPE_BOOLEAN_VALUE: ret = nvlist_add_boolean_value(payload, name, va_arg(ap, boolean_t)); break; case DATA_TYPE_BOOLEAN_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_boolean_array(payload, name, va_arg(ap, boolean_t *), nelem); break; case DATA_TYPE_INT8: ret = nvlist_add_int8(payload, name, va_arg(ap, int)); break; case DATA_TYPE_INT8_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int8_array(payload, name, va_arg(ap, int8_t *), nelem); break; case DATA_TYPE_UINT8: ret = nvlist_add_uint8(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_UINT8_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint8_array(payload, name, va_arg(ap, uint8_t *), nelem); break; case DATA_TYPE_INT16: ret = nvlist_add_int16(payload, name, va_arg(ap, int)); break; case DATA_TYPE_INT16_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int16_array(payload, name, va_arg(ap, int16_t *), nelem); break; case DATA_TYPE_UINT16: ret = nvlist_add_uint16(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_UINT16_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint16_array(payload, name, va_arg(ap, uint16_t *), nelem); break; case DATA_TYPE_INT32: ret = nvlist_add_int32(payload, name, va_arg(ap, int32_t)); break; case DATA_TYPE_INT32_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int32_array(payload, name, va_arg(ap, int32_t *), nelem); break; case DATA_TYPE_UINT32: ret = nvlist_add_uint32(payload, name, va_arg(ap, uint32_t)); break; case DATA_TYPE_UINT32_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint32_array(payload, name, va_arg(ap, uint32_t *), nelem); break; case DATA_TYPE_INT64: ret = nvlist_add_int64(payload, name, va_arg(ap, int64_t)); break; case DATA_TYPE_INT64_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int64_array(payload, name, va_arg(ap, int64_t *), nelem); break; case DATA_TYPE_UINT64: ret = nvlist_add_uint64(payload, name, va_arg(ap, uint64_t)); break; case DATA_TYPE_UINT64_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint64_array(payload, name, va_arg(ap, uint64_t *), nelem); break; case DATA_TYPE_STRING: ret = nvlist_add_string(payload, name, va_arg(ap, char *)); break; case DATA_TYPE_STRING_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_string_array(payload, name, va_arg(ap, char **), nelem); break; case DATA_TYPE_NVLIST: ret = nvlist_add_nvlist(payload, name, va_arg(ap, nvlist_t *)); break; case DATA_TYPE_NVLIST_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_nvlist_array(payload, name, va_arg(ap, nvlist_t **), nelem); break; default: ret = EINVAL; } name = va_arg(ap, char *); } return (ret); } void fm_payload_set(nvlist_t *payload, ...) { int ret; const char *name; va_list ap; va_start(ap, payload); name = va_arg(ap, char *); ret = i_fm_payload_set(payload, name, ap); va_end(ap); if (ret) atomic_add_64( &erpt_kstat_data.payload_set_failed.value.ui64, 1); } /* * Set-up and validate the members of an ereport event according to: * * Member name Type Value * ==================================================== * class string ereport * version uint8_t 0 * ena uint64_t * detector nvlist_t * ereport-payload nvlist_t * * We don't actually add a 'version' member to the payload. Really, * the version quoted to us by our caller is that of the category 1 * "ereport" event class (and we require FM_EREPORT_VERS0) but * the payload version of the actual leaf class event under construction * may be something else. Callers should supply a version in the varargs, * or (better) we could take two version arguments - one for the * ereport category 1 classification (expect FM_EREPORT_VERS0) and one * for the leaf class. */ void fm_ereport_set(nvlist_t *ereport, int version, const char *erpt_class, uint64_t ena, const nvlist_t *detector, ...) { char ereport_class[FM_MAX_CLASS]; const char *name; va_list ap; int ret; if (version != FM_EREPORT_VERS0) { atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); return; } (void) snprintf(ereport_class, FM_MAX_CLASS, "%s.%s", FM_EREPORT_CLASS, erpt_class); if (nvlist_add_string(ereport, FM_CLASS, ereport_class) != 0) { atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); return; } if (nvlist_add_uint64(ereport, FM_EREPORT_ENA, ena)) { atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); } if (nvlist_add_nvlist(ereport, FM_EREPORT_DETECTOR, (nvlist_t *)detector) != 0) { atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); } va_start(ap, detector); name = va_arg(ap, const char *); ret = i_fm_payload_set(ereport, name, ap); va_end(ap); if (ret) atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); } /* * Set-up and validate the members of an hc fmri according to; * * Member name Type Value * =================================================== * version uint8_t 0 * auth nvlist_t * hc-name string * hc-id string * * Note that auth and hc-id are optional members. */ #define HC_MAXPAIRS 20 #define HC_MAXNAMELEN 50 static int fm_fmri_hc_set_common(nvlist_t *fmri, int version, const nvlist_t *auth) { if (version != FM_HC_SCHEME_VERSION) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return (0); } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0 || nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_HC) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return (0); } if (auth != NULL && nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return (0); } return (1); } void fm_fmri_hc_set(nvlist_t *fmri, int version, const nvlist_t *auth, nvlist_t *snvl, int npairs, ...) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri); nvlist_t *pairs[HC_MAXPAIRS]; va_list ap; int i; if (!fm_fmri_hc_set_common(fmri, version, auth)) return; npairs = MIN(npairs, HC_MAXPAIRS); va_start(ap, npairs); for (i = 0; i < npairs; i++) { const char *name = va_arg(ap, const char *); uint32_t id = va_arg(ap, uint32_t); char idstr[11]; (void) snprintf(idstr, sizeof (idstr), "%u", id); pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); } } va_end(ap); if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST, pairs, npairs) != 0) atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); for (i = 0; i < npairs; i++) fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN); if (snvl != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); } } } void fm_fmri_hc_create(nvlist_t *fmri, int version, const nvlist_t *auth, nvlist_t *snvl, nvlist_t *bboard, int npairs, ...) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri); nvlist_t *pairs[HC_MAXPAIRS]; nvlist_t **hcl; uint_t n; int i, j; va_list ap; char *hcname, *hcid; if (!fm_fmri_hc_set_common(fmri, version, auth)) return; /* * copy the bboard nvpairs to the pairs array */ if (nvlist_lookup_nvlist_array(bboard, FM_FMRI_HC_LIST, &hcl, &n) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } for (i = 0; i < n; i++) { if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_NAME, &hcname) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_ID, &hcid) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, hcname) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, hcid) != 0) { for (j = 0; j <= i; j++) { if (pairs[j] != NULL) fm_nvlist_destroy(pairs[j], FM_NVA_RETAIN); } atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } } /* * create the pairs from passed in pairs */ npairs = MIN(npairs, HC_MAXPAIRS); va_start(ap, npairs); for (i = n; i < npairs + n; i++) { const char *name = va_arg(ap, const char *); uint32_t id = va_arg(ap, uint32_t); char idstr[11]; (void) snprintf(idstr, sizeof (idstr), "%u", id); pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) { for (j = 0; j <= i; j++) { if (pairs[j] != NULL) fm_nvlist_destroy(pairs[j], FM_NVA_RETAIN); } atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } } va_end(ap); /* * Create the fmri hc list */ if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST, pairs, npairs + n) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } for (i = 0; i < npairs + n; i++) { fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN); } if (snvl != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } } } /* * Set-up and validate the members of an dev fmri according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t * devpath string * [devid] string * [target-port-l0id] string * * Note that auth and devid are optional members. */ void fm_fmri_dev_set(nvlist_t *fmri_dev, int version, const nvlist_t *auth, const char *devpath, const char *devid, const char *tpl0) { int err = 0; if (version != DEV_SCHEME_VERSION0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } err |= nvlist_add_uint8(fmri_dev, FM_VERSION, version); err |= nvlist_add_string(fmri_dev, FM_FMRI_SCHEME, FM_FMRI_SCHEME_DEV); if (auth != NULL) { err |= nvlist_add_nvlist(fmri_dev, FM_FMRI_AUTHORITY, (nvlist_t *)auth); } err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_PATH, devpath); if (devid != NULL) err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_ID, devid); if (tpl0 != NULL) err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_TGTPTLUN0, tpl0); if (err) atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); } /* * Set-up and validate the members of an cpu fmri according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t * cpuid uint32_t * cpumask uint8_t * serial uint64_t * * Note that auth, cpumask, serial are optional members. * */ void fm_fmri_cpu_set(nvlist_t *fmri_cpu, int version, const nvlist_t *auth, uint32_t cpu_id, uint8_t *cpu_maskp, const char *serial_idp) { uint64_t *failedp = &erpt_kstat_data.fmri_set_failed.value.ui64; if (version < CPU_SCHEME_VERSION1) { atomic_add_64(failedp, 1); return; } if (nvlist_add_uint8(fmri_cpu, FM_VERSION, version) != 0) { atomic_add_64(failedp, 1); return; } if (nvlist_add_string(fmri_cpu, FM_FMRI_SCHEME, FM_FMRI_SCHEME_CPU) != 0) { atomic_add_64(failedp, 1); return; } if (auth != NULL && nvlist_add_nvlist(fmri_cpu, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) atomic_add_64(failedp, 1); if (nvlist_add_uint32(fmri_cpu, FM_FMRI_CPU_ID, cpu_id) != 0) atomic_add_64(failedp, 1); if (cpu_maskp != NULL && nvlist_add_uint8(fmri_cpu, FM_FMRI_CPU_MASK, *cpu_maskp) != 0) atomic_add_64(failedp, 1); if (serial_idp == NULL || nvlist_add_string(fmri_cpu, FM_FMRI_CPU_SERIAL_ID, (char *)serial_idp) != 0) atomic_add_64(failedp, 1); } /* * Set-up and validate the members of a mem according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t [optional] * unum string * serial string [optional*] * offset uint64_t [optional] * * * serial is required if offset is present */ void fm_fmri_mem_set(nvlist_t *fmri, int version, const nvlist_t *auth, const char *unum, const char *serial, uint64_t offset) { if (version != MEM_SCHEME_VERSION0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (!serial && (offset != (uint64_t)-1)) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_MEM) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (auth != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); } } if (nvlist_add_string(fmri, FM_FMRI_MEM_UNUM, unum) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); } if (serial != NULL) { if (nvlist_add_string_array(fmri, FM_FMRI_MEM_SERIAL_ID, (char **)&serial, 1) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); } if (offset != (uint64_t)-1) { if (nvlist_add_uint64(fmri, FM_FMRI_MEM_OFFSET, offset) != 0) { atomic_add_64(&erpt_kstat_data. fmri_set_failed.value.ui64, 1); } } } } void fm_fmri_zfs_set(nvlist_t *fmri, int version, uint64_t pool_guid, uint64_t vdev_guid) { if (version != ZFS_SCHEME_VERSION0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); return; } if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_POOL, pool_guid) != 0) { atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); } if (vdev_guid != 0) { if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_VDEV, vdev_guid) != 0) { atomic_add_64( &erpt_kstat_data.fmri_set_failed.value.ui64, 1); } } } uint64_t fm_ena_increment(uint64_t ena) { uint64_t new_ena; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: new_ena = ena + (1 << ENA_FMT1_GEN_SHFT); break; case FM_ENA_FMT2: new_ena = ena + (1 << ENA_FMT2_GEN_SHFT); break; default: new_ena = 0; } return (new_ena); } uint64_t fm_ena_generate_cpu(uint64_t timestamp, processorid_t cpuid, uchar_t format) { uint64_t ena = 0; switch (format) { case FM_ENA_FMT1: if (timestamp) { ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((cpuid << ENA_FMT1_CPUID_SHFT) & ENA_FMT1_CPUID_MASK) | ((timestamp << ENA_FMT1_TIME_SHFT) & ENA_FMT1_TIME_MASK)); } else { ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((cpuid << ENA_FMT1_CPUID_SHFT) & ENA_FMT1_CPUID_MASK) | ((gethrtime() << ENA_FMT1_TIME_SHFT) & ENA_FMT1_TIME_MASK)); } break; case FM_ENA_FMT2: ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((timestamp << ENA_FMT2_TIME_SHFT) & ENA_FMT2_TIME_MASK)); break; default: break; } return (ena); } uint64_t fm_ena_generate(uint64_t timestamp, uchar_t format) { uint64_t ena; kpreempt_disable(); ena = fm_ena_generate_cpu(timestamp, getcpuid(), format); kpreempt_enable(); return (ena); } uint64_t fm_ena_generation_get(uint64_t ena) { uint64_t gen; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: gen = (ena & ENA_FMT1_GEN_MASK) >> ENA_FMT1_GEN_SHFT; break; case FM_ENA_FMT2: gen = (ena & ENA_FMT2_GEN_MASK) >> ENA_FMT2_GEN_SHFT; break; default: gen = 0; break; } return (gen); } uchar_t fm_ena_format_get(uint64_t ena) { return (ENA_FORMAT(ena)); } uint64_t fm_ena_id_get(uint64_t ena) { uint64_t id; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: id = (ena & ENA_FMT1_ID_MASK) >> ENA_FMT1_ID_SHFT; break; case FM_ENA_FMT2: id = (ena & ENA_FMT2_ID_MASK) >> ENA_FMT2_ID_SHFT; break; default: id = 0; } return (id); } uint64_t fm_ena_time_get(uint64_t ena) { uint64_t time; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: time = (ena & ENA_FMT1_TIME_MASK) >> ENA_FMT1_TIME_SHFT; break; case FM_ENA_FMT2: time = (ena & ENA_FMT2_TIME_MASK) >> ENA_FMT2_TIME_SHFT; break; default: time = 0; } return (time); } #ifdef _KERNEL void fm_init(void) { zevent_len_cur = 0; zevent_flags = 0; if (zfs_zevent_len_max == 0) zfs_zevent_len_max = ERPT_MAX_ERRS * MAX(max_ncpus, 4); /* Initialize zevent allocation and generation kstats */ fm_ksp = kstat_create("zfs", 0, "fm", "misc", KSTAT_TYPE_NAMED, sizeof (struct erpt_kstat) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (fm_ksp != NULL) { fm_ksp->ks_data = &erpt_kstat_data; kstat_install(fm_ksp); } else { cmn_err(CE_NOTE, "failed to create fm/misc kstat\n"); } mutex_init(&zevent_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&zevent_list, sizeof(zevent_t), offsetof(zevent_t, ev_node)); cv_init(&zevent_cv, NULL, CV_DEFAULT, NULL); } void fm_fini(void) { int count; zfs_zevent_drain_all(&count); cv_broadcast(&zevent_cv); mutex_enter(&zevent_lock); zevent_flags |= ZEVENT_SHUTDOWN; while (zevent_waiters > 0) { mutex_exit(&zevent_lock); schedule(); mutex_enter(&zevent_lock); } mutex_exit(&zevent_lock); cv_destroy(&zevent_cv); list_destroy(&zevent_list); mutex_destroy(&zevent_lock); if (fm_ksp != NULL) { kstat_delete(fm_ksp); fm_ksp = NULL; } } module_param(zfs_zevent_len_max, int, 0644); MODULE_PARM_DESC(zfs_zevent_len_max, "Max event queue length"); module_param(zfs_zevent_cols, int, 0644); MODULE_PARM_DESC(zfs_zevent_cols, "Max event column width"); module_param(zfs_zevent_console, int, 0644); MODULE_PARM_DESC(zfs_zevent_console, "Log events to the console"); #endif /* _KERNEL */