From owner-svn-src-vendor@freebsd.org Tue Jul 31 00:13:07 2018 Return-Path: Delivered-To: svn-src-vendor@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 7D9471065B79; Tue, 31 Jul 2018 00:13:07 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from mxrelay.nyi.freebsd.org (mxrelay.nyi.freebsd.org [IPv6:2610:1c1:1:606c::19:3]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client CN "mxrelay.nyi.freebsd.org", Issuer "Let's Encrypt Authority X3" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 201E88E4AA; Tue, 31 Jul 2018 00:13:07 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mxrelay.nyi.freebsd.org (Postfix) with ESMTPS id F0B0A4058; Tue, 31 Jul 2018 00:13:06 +0000 (UTC) (envelope-from mav@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id w6V0D6fB088739; Tue, 31 Jul 2018 00:13:06 GMT (envelope-from mav@FreeBSD.org) Received: (from mav@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id w6V0D48W088729; Tue, 31 Jul 2018 00:13:04 GMT (envelope-from mav@FreeBSD.org) Message-Id: <201807310013.w6V0D48W088729@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: mav set sender to mav@FreeBSD.org using -f From: Alexander Motin Date: Tue, 31 Jul 2018 00:13:04 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-vendor@freebsd.org Subject: svn commit: r336950 - vendor-sys/illumos/dist/uts/common/fs/zfs vendor-sys/illumos/dist/uts/common/fs/zfs/sys vendor/illumos/dist/cmd/zdb vendor/illumos/dist/cmd/ztest vendor/illumos/dist/lib/libzf... X-SVN-Group: vendor-sys X-SVN-Commit-Author: mav X-SVN-Commit-Paths: vendor-sys/illumos/dist/uts/common/fs/zfs vendor-sys/illumos/dist/uts/common/fs/zfs/sys vendor/illumos/dist/cmd/zdb vendor/illumos/dist/cmd/ztest vendor/illumos/dist/lib/libzfs/common X-SVN-Commit-Revision: 336950 X-SVN-Commit-Repository: base MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-vendor@freebsd.org X-Mailman-Version: 2.1.27 Precedence: list List-Id: SVN commit messages for the vendor work area tree List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 31 Jul 2018 00:13:07 -0000 Author: mav Date: Tue Jul 31 00:13:04 2018 New Revision: 336950 URL: https://svnweb.freebsd.org/changeset/base/336950 Log: 9290 device removal reduces redundancy of mirrors Mirrors are supposed to provide redundancy in the face of whole-disk failure and silent damage (e.g. some data on disk is not right, but ZFS hasn't detected the whole device as being broken). However, the current device removal implementation bypasses some of the mirror's redundancy. illumos/illumos-gate@3a4b1be953ee5601bab748afa07c26ed4996cde6 Reviewed by: George Wilson Reviewed by: Prashanth Sreenivasa Reviewed by: Sara Hartse Reviewed by: Serapheim Dimitropoulos Reviewed by: Brian Behlendorf Reviewed by: Tim Chase Approved by: Richard Lowe Author: Matthew Ahrens Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_scan.c vendor-sys/illumos/dist/uts/common/fs/zfs/metaslab.c vendor-sys/illumos/dist/uts/common/fs/zfs/spa.c vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c vendor-sys/illumos/dist/uts/common/fs/zfs/sys/vdev_removal.h vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h vendor-sys/illumos/dist/uts/common/fs/zfs/vdev.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_mirror.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_removal.c vendor-sys/illumos/dist/uts/common/fs/zfs/zio.c Changes in other areas also in this revision: Modified: vendor/illumos/dist/cmd/zdb/zdb.c vendor/illumos/dist/cmd/ztest/ztest.c vendor/illumos/dist/lib/libzfs/common/libzfs_pool.c Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_scan.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_scan.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/dsl_scan.c Tue Jul 31 00:13:04 2018 (r336950) @@ -1988,7 +1988,16 @@ dsl_scan_scrub_cb(dsl_pool_t *dp, /* if it's a resilver, this may not be in the target range */ if (!needs_io) { - if (DVA_GET_GANG(&bp->blk_dva[d])) { + if (vd->vdev_ops == &vdev_indirect_ops) { + /* + * The indirect vdev can point to multiple + * vdevs. For simplicity, always create + * the resilver zio_t. zio_vdev_io_start() + * will bypass the child resilver i/o's if + * they are on vdevs that don't have DTL's. + */ + needs_io = B_TRUE; + } else if (DVA_GET_GANG(&bp->blk_dva[d])) { /* * Gang members may be spread across multiple * vdevs, so the best estimate we have is the Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/metaslab.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/metaslab.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/metaslab.c Tue Jul 31 00:13:04 2018 (r336950) @@ -3582,7 +3582,7 @@ metaslab_free_impl(vdev_t *vd, uint64_t offset, uint64 return; if (spa->spa_vdev_removal != NULL && - spa->spa_vdev_removal->svr_vdev == vd && + spa->spa_vdev_removal->svr_vdev_id == vd->vdev_id && vdev_is_concrete(vd)) { /* * Note: we check if the vdev is concrete because when Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/spa.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/spa.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/spa.c Tue Jul 31 00:13:04 2018 (r336950) @@ -5509,8 +5509,7 @@ spa_vdev_add(spa_t *spa, nvlist_t *nvroot) for (int c = 0; c < vd->vdev_children; c++) { tvd = vd->vdev_child[c]; if (spa->spa_vdev_removal != NULL && - tvd->vdev_ashift != - spa->spa_vdev_removal->svr_vdev->vdev_ashift) { + tvd->vdev_ashift != spa->spa_max_ashift) { return (spa_vdev_exit(spa, vd, txg, EINVAL)); } /* Fail if top level vdev is raidz */ @@ -5626,10 +5625,8 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *n return (spa_vdev_exit(spa, NULL, txg, error)); } - if (spa->spa_vdev_removal != NULL || - spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { + if (spa->spa_vdev_removal != NULL) return (spa_vdev_exit(spa, NULL, txg, EBUSY)); - } if (oldvd == NULL) return (spa_vdev_exit(spa, NULL, txg, ENODEV)); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/spa_misc.c Tue Jul 31 00:13:04 2018 (r336950) @@ -1761,9 +1761,12 @@ spa_update_dspace(spa_t *spa) * allocated twice (on the old device and the new * device). */ - vdev_t *vd = spa->spa_vdev_removal->svr_vdev; + spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); + vdev_t *vd = + vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id); spa->spa_dspace -= spa_deflate(spa) ? vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space; + spa_config_exit(spa, SCL_VDEV, FTAG); } } Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/vdev_removal.h ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/vdev_removal.h Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/vdev_removal.h Tue Jul 31 00:13:04 2018 (r336950) @@ -30,7 +30,7 @@ extern "C" { #endif typedef struct spa_vdev_removal { - vdev_t *svr_vdev; + uint64_t svr_vdev_id; uint64_t svr_max_offset_to_sync[TXG_SIZE]; /* Thread performing a vdev removal. */ kthread_t *svr_thread; Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/sys/zio.h Tue Jul 31 00:13:04 2018 (r336950) @@ -550,7 +550,7 @@ extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t * zio_done_func_t *done, void *private); extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, - struct abd *data, uint64_t size, int type, zio_priority_t priority, + struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private); extern void zio_vdev_io_bypass(zio_t *zio); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/vdev.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/vdev.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/vdev.c Tue Jul 31 00:13:04 2018 (r336950) @@ -867,6 +867,32 @@ vdev_top_transfer(vdev_t *svd, vdev_t *tvd) svd->vdev_stat.vs_space = 0; svd->vdev_stat.vs_dspace = 0; + /* + * State which may be set on a top-level vdev that's in the + * process of being removed. + */ + ASSERT0(tvd->vdev_indirect_config.vic_births_object); + ASSERT0(tvd->vdev_indirect_config.vic_mapping_object); + ASSERT3U(tvd->vdev_indirect_config.vic_prev_indirect_vdev, ==, -1ULL); + ASSERT3P(tvd->vdev_indirect_mapping, ==, NULL); + ASSERT3P(tvd->vdev_indirect_births, ==, NULL); + ASSERT3P(tvd->vdev_obsolete_sm, ==, NULL); + ASSERT0(tvd->vdev_removing); + tvd->vdev_removing = svd->vdev_removing; + tvd->vdev_indirect_config = svd->vdev_indirect_config; + tvd->vdev_indirect_mapping = svd->vdev_indirect_mapping; + tvd->vdev_indirect_births = svd->vdev_indirect_births; + range_tree_swap(&svd->vdev_obsolete_segments, + &tvd->vdev_obsolete_segments); + tvd->vdev_obsolete_sm = svd->vdev_obsolete_sm; + svd->vdev_indirect_config.vic_mapping_object = 0; + svd->vdev_indirect_config.vic_births_object = 0; + svd->vdev_indirect_config.vic_prev_indirect_vdev = -1ULL; + svd->vdev_indirect_mapping = NULL; + svd->vdev_indirect_births = NULL; + svd->vdev_obsolete_sm = NULL; + svd->vdev_removing = 0; + for (t = 0; t < TXG_SIZE; t++) { while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) (void) txg_list_add(&tvd->vdev_ms_list, msp, t); Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect.c Tue Jul 31 00:13:04 2018 (r336950) @@ -23,6 +23,7 @@ #include #include #include +#include #include #include #include @@ -46,10 +47,11 @@ * "vdev_remap" operation that executes a callback on each contiguous * segment of the new location. This function is used in multiple ways: * - * - reads and repair writes to this device use the callback to create - * a child io for each mapped segment. + * - i/os to this vdev use the callback to determine where the + * data is now located, and issue child i/os for each segment's new + * location. * - * - frees and claims to this device use the callback to free or claim + * - frees and claims to this vdev use the callback to free or claim * each mapped segment. (Note that we don't actually need to claim * log blocks on indirect vdevs, because we don't allocate to * removing vdevs. However, zdb uses zio_claim() for its leak @@ -204,6 +206,94 @@ uint64_t zfs_condense_min_mapping_bytes = 128 * 1024; int zfs_condense_indirect_commit_entry_delay_ticks = 0; /* + * If a split block contains more than this many segments, consider it too + * computationally expensive to check all (2^num_segments) possible + * combinations. Instead, try at most 2^_segments_max randomly-selected + * combinations. + * + * This is reasonable if only a few segment copies are damaged and the + * majority of segment copies are good. This allows all the segment copies to + * participate fairly in the reconstruction and prevents the repeated use of + * one bad copy. + */ +int zfs_reconstruct_indirect_segments_max = 10; + +/* + * The indirect_child_t represents the vdev that we will read from, when we + * need to read all copies of the data (e.g. for scrub or reconstruction). + * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror), + * ic_vdev is the same as is_vdev. However, for mirror top-level vdevs, + * ic_vdev is a child of the mirror. + */ +typedef struct indirect_child { + abd_t *ic_data; + vdev_t *ic_vdev; +} indirect_child_t; + +/* + * The indirect_split_t represents one mapped segment of an i/o to the + * indirect vdev. For non-split (contiguously-mapped) blocks, there will be + * only one indirect_split_t, with is_split_offset==0 and is_size==io_size. + * For split blocks, there will be several of these. + */ +typedef struct indirect_split { + list_node_t is_node; /* link on iv_splits */ + + /* + * is_split_offset is the offset into the i/o. + * This is the sum of the previous splits' is_size's. + */ + uint64_t is_split_offset; + + vdev_t *is_vdev; /* top-level vdev */ + uint64_t is_target_offset; /* offset on is_vdev */ + uint64_t is_size; + int is_children; /* number of entries in is_child[] */ + + /* + * is_good_child is the child that we are currently using to + * attempt reconstruction. + */ + int is_good_child; + + indirect_child_t is_child[1]; /* variable-length */ +} indirect_split_t; + +/* + * The indirect_vsd_t is associated with each i/o to the indirect vdev. + * It is the "Vdev-Specific Data" in the zio_t's io_vsd. + */ +typedef struct indirect_vsd { + boolean_t iv_split_block; + boolean_t iv_reconstruct; + + list_t iv_splits; /* list of indirect_split_t's */ +} indirect_vsd_t; + +static void +vdev_indirect_map_free(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + + indirect_split_t *is; + while ((is = list_head(&iv->iv_splits)) != NULL) { + for (int c = 0; c < is->is_children; c++) { + indirect_child_t *ic = &is->is_child[c]; + if (ic->ic_data != NULL) + abd_free(ic->ic_data); + } + list_remove(&iv->iv_splits, is); + kmem_free(is, + offsetof(indirect_split_t, is_child[is->is_children])); + } + kmem_free(iv, sizeof (*iv)); +} + +static const zio_vsd_ops_t vdev_indirect_vsd_ops = { + vdev_indirect_map_free, + zio_vsd_default_cksum_report +}; +/* * Mark the given offset and size as being obsolete. */ void @@ -817,12 +907,6 @@ vdev_indirect_close(vdev_t *vd) } /* ARGSUSED */ -static void -vdev_indirect_io_done(zio_t *zio) -{ -} - -/* ARGSUSED */ static int vdev_indirect_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, uint64_t *ashift) @@ -1065,39 +1149,473 @@ vdev_indirect_child_io_done(zio_t *zio) abd_put(zio->io_abd); } +/* + * This is a callback for vdev_indirect_remap() which allocates an + * indirect_split_t for each split segment and adds it to iv_splits. + */ static void -vdev_indirect_io_start_cb(uint64_t split_offset, vdev_t *vd, uint64_t offset, +vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { zio_t *zio = arg; + indirect_vsd_t *iv = zio->io_vsd; ASSERT3P(vd, !=, NULL); if (vd->vdev_ops == &vdev_indirect_ops) return; - zio_nowait(zio_vdev_child_io(zio, NULL, vd, offset, - abd_get_offset(zio->io_abd, split_offset), - size, zio->io_type, zio->io_priority, - 0, vdev_indirect_child_io_done, zio)); + int n = 1; + if (vd->vdev_ops == &vdev_mirror_ops) + n = vd->vdev_children; + + indirect_split_t *is = + kmem_zalloc(offsetof(indirect_split_t, is_child[n]), KM_SLEEP); + + is->is_children = n; + is->is_size = size; + is->is_split_offset = split_offset; + is->is_target_offset = offset; + is->is_vdev = vd; + + /* + * Note that we only consider multiple copies of the data for + * *mirror* vdevs. We don't for "replacing" or "spare" vdevs, even + * though they use the same ops as mirror, because there's only one + * "good" copy under the replacing/spare. + */ + if (vd->vdev_ops == &vdev_mirror_ops) { + for (int i = 0; i < n; i++) { + is->is_child[i].ic_vdev = vd->vdev_child[i]; + } + } else { + is->is_child[0].ic_vdev = vd; + } + + list_insert_tail(&iv->iv_splits, is); } static void +vdev_indirect_read_split_done(zio_t *zio) +{ + indirect_child_t *ic = zio->io_private; + + if (zio->io_error != 0) { + /* + * Clear ic_data to indicate that we do not have data for this + * child. + */ + abd_free(ic->ic_data); + ic->ic_data = NULL; + } +} + +/* + * Issue reads for all copies (mirror children) of all splits. + */ +static void +vdev_indirect_read_all(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + for (int i = 0; i < is->is_children; i++) { + indirect_child_t *ic = &is->is_child[i]; + + if (!vdev_readable(ic->ic_vdev)) + continue; + + /* + * Note, we may read from a child whose DTL + * indicates that the data may not be present here. + * While this might result in a few i/os that will + * likely return incorrect data, it simplifies the + * code since we can treat scrub and resilver + * identically. (The incorrect data will be + * detected and ignored when we verify the + * checksum.) + */ + + ic->ic_data = abd_alloc_sametype(zio->io_abd, + is->is_size); + + zio_nowait(zio_vdev_child_io(zio, NULL, + ic->ic_vdev, is->is_target_offset, ic->ic_data, + is->is_size, zio->io_type, zio->io_priority, 0, + vdev_indirect_read_split_done, ic)); + } + } + iv->iv_reconstruct = B_TRUE; +} + +static void vdev_indirect_io_start(zio_t *zio) { spa_t *spa = zio->io_spa; + indirect_vsd_t *iv = kmem_zalloc(sizeof (*iv), KM_SLEEP); + list_create(&iv->iv_splits, + sizeof (indirect_split_t), offsetof(indirect_split_t, is_node)); + zio->io_vsd = iv; + zio->io_vsd_ops = &vdev_indirect_vsd_ops; + ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (zio->io_type != ZIO_TYPE_READ) { ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); - ASSERT((zio->io_flags & - (ZIO_FLAG_SELF_HEAL | ZIO_FLAG_INDUCE_DAMAGE)) != 0); + /* + * Note: this code can handle other kinds of writes, + * but we don't expect them. + */ + ASSERT((zio->io_flags & (ZIO_FLAG_SELF_HEAL | + ZIO_FLAG_RESILVER | ZIO_FLAG_INDUCE_DAMAGE)) != 0); } vdev_indirect_remap(zio->io_vd, zio->io_offset, zio->io_size, - vdev_indirect_io_start_cb, zio); + vdev_indirect_gather_splits, zio); + indirect_split_t *first = list_head(&iv->iv_splits); + if (first->is_size == zio->io_size) { + /* + * This is not a split block; we are pointing to the entire + * data, which will checksum the same as the original data. + * Pass the BP down so that the child i/o can verify the + * checksum, and try a different location if available + * (e.g. on a mirror). + * + * While this special case could be handled the same as the + * general (split block) case, doing it this way ensures + * that the vast majority of blocks on indirect vdevs + * (which are not split) are handled identically to blocks + * on non-indirect vdevs. This allows us to be less strict + * about performance in the general (but rare) case. + */ + ASSERT0(first->is_split_offset); + ASSERT3P(list_next(&iv->iv_splits, first), ==, NULL); + zio_nowait(zio_vdev_child_io(zio, zio->io_bp, + first->is_vdev, first->is_target_offset, + abd_get_offset(zio->io_abd, 0), + zio->io_size, zio->io_type, zio->io_priority, 0, + vdev_indirect_child_io_done, zio)); + } else { + iv->iv_split_block = B_TRUE; + if (zio->io_flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER)) { + /* + * Read all copies. Note that for simplicity, + * we don't bother consulting the DTL in the + * resilver case. + */ + vdev_indirect_read_all(zio); + } else { + /* + * Read one copy of each split segment, from the + * top-level vdev. Since we don't know the + * checksum of each split individually, the child + * zio can't ensure that we get the right data. + * E.g. if it's a mirror, it will just read from a + * random (healthy) leaf vdev. We have to verify + * the checksum in vdev_indirect_io_done(). + */ + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + zio_nowait(zio_vdev_child_io(zio, NULL, + is->is_vdev, is->is_target_offset, + abd_get_offset(zio->io_abd, + is->is_split_offset), + is->is_size, zio->io_type, + zio->io_priority, 0, + vdev_indirect_child_io_done, zio)); + } + } + } + zio_execute(zio); +} + +/* + * Report a checksum error for a child. + */ +static void +vdev_indirect_checksum_error(zio_t *zio, + indirect_split_t *is, indirect_child_t *ic) +{ + vdev_t *vd = ic->ic_vdev; + + if (zio->io_flags & ZIO_FLAG_SPECULATIVE) + return; + + mutex_enter(&vd->vdev_stat_lock); + vd->vdev_stat.vs_checksum_errors++; + mutex_exit(&vd->vdev_stat_lock); + + zio_bad_cksum_t zbc = { 0 }; + void *bad_buf = abd_borrow_buf_copy(ic->ic_data, is->is_size); + abd_t *good_abd = is->is_child[is->is_good_child].ic_data; + void *good_buf = abd_borrow_buf_copy(good_abd, is->is_size); + zfs_ereport_post_checksum(zio->io_spa, vd, zio, + is->is_target_offset, is->is_size, good_buf, bad_buf, &zbc); + abd_return_buf(ic->ic_data, bad_buf, is->is_size); + abd_return_buf(good_abd, good_buf, is->is_size); +} + +/* + * Issue repair i/os for any incorrect copies. We do this by comparing + * each split segment's correct data (is_good_child's ic_data) with each + * other copy of the data. If they differ, then we overwrite the bad data + * with the good copy. Note that we do this without regard for the DTL's, + * which simplifies this code and also issues the optimal number of writes + * (based on which copies actually read bad data, as opposed to which we + * think might be wrong). For the same reason, we always use + * ZIO_FLAG_SELF_HEAL, to bypass the DTL check in zio_vdev_io_start(). + */ +static void +vdev_indirect_repair(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + + enum zio_flag flags = ZIO_FLAG_IO_REPAIR; + + if (!(zio->io_flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER))) + flags |= ZIO_FLAG_SELF_HEAL; + + if (!spa_writeable(zio->io_spa)) + return; + + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + indirect_child_t *good_child = &is->is_child[is->is_good_child]; + + for (int c = 0; c < is->is_children; c++) { + indirect_child_t *ic = &is->is_child[c]; + if (ic == good_child) + continue; + if (ic->ic_data == NULL) + continue; + if (abd_cmp(good_child->ic_data, ic->ic_data, + is->is_size) == 0) + continue; + + zio_nowait(zio_vdev_child_io(zio, NULL, + ic->ic_vdev, is->is_target_offset, + good_child->ic_data, is->is_size, + ZIO_TYPE_WRITE, ZIO_PRIORITY_ASYNC_WRITE, + ZIO_FLAG_IO_REPAIR | ZIO_FLAG_SELF_HEAL, + NULL, NULL)); + + vdev_indirect_checksum_error(zio, is, ic); + } + } +} + +/* + * Report checksum errors on all children that we read from. + */ +static void +vdev_indirect_all_checksum_errors(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + + if (zio->io_flags & ZIO_FLAG_SPECULATIVE) + return; + + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + for (int c = 0; c < is->is_children; c++) { + indirect_child_t *ic = &is->is_child[c]; + + if (ic->ic_data == NULL) + continue; + + vdev_t *vd = ic->ic_vdev; + + mutex_enter(&vd->vdev_stat_lock); + vd->vdev_stat.vs_checksum_errors++; + mutex_exit(&vd->vdev_stat_lock); + + zfs_ereport_post_checksum(zio->io_spa, vd, zio, + is->is_target_offset, is->is_size, + NULL, NULL, NULL); + } + } +} + +/* + * This function is called when we have read all copies of the data and need + * to try to find a combination of copies that gives us the right checksum. + * + * If we pointed to any mirror vdevs, this effectively does the job of the + * mirror. The mirror vdev code can't do its own job because we don't know + * the checksum of each split segment individually. We have to try every + * combination of copies of split segments, until we find one that checksums + * correctly. (Or until we have tried all combinations, or have tried + * 2^zfs_reconstruct_indirect_segments_max combinations. In these cases we + * set io_error to ECKSUM to propagate the error up to the user.) + * + * For example, if we have 3 segments in the split, + * and each points to a 2-way mirror, we will have the following pieces of + * data: + * + * | mirror child + * split | [0] [1] + * ======|===================== + * A | data_A_0 data_A_1 + * B | data_B_0 data_B_1 + * C | data_C_0 data_C_1 + * + * We will try the following (mirror children)^(number of splits) (2^3=8) + * combinations, which is similar to bitwise-little-endian counting in + * binary. In general each "digit" corresponds to a split segment, and the + * base of each digit is is_children, which can be different for each + * digit. + * + * "low bit" "high bit" + * v v + * data_A_0 data_B_0 data_C_0 + * data_A_1 data_B_0 data_C_0 + * data_A_0 data_B_1 data_C_0 + * data_A_1 data_B_1 data_C_0 + * data_A_0 data_B_0 data_C_1 + * data_A_1 data_B_0 data_C_1 + * data_A_0 data_B_1 data_C_1 + * data_A_1 data_B_1 data_C_1 + * + * Note that the split segments may be on the same or different top-level + * vdevs. In either case, we try lots of combinations (see + * zfs_reconstruct_indirect_segments_max). This ensures that if a mirror has + * small silent errors on all of its children, we can still reconstruct the + * correct data, as long as those errors are at sufficiently-separated + * offsets (specifically, separated by the largest block size - default of + * 128KB, but up to 16MB). + */ +static void +vdev_indirect_reconstruct_io_done(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + uint64_t attempts = 0; + uint64_t attempts_max = 1ULL << zfs_reconstruct_indirect_segments_max; + int segments = 0; + + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) + segments++; + + for (;;) { + /* copy data from splits to main zio */ + int ret; + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + + /* + * If this child failed, its ic_data will be NULL. + * Skip this combination. + */ + if (is->is_child[is->is_good_child].ic_data == NULL) { + ret = EIO; + goto next; + } + + abd_copy_off(zio->io_abd, + is->is_child[is->is_good_child].ic_data, + is->is_split_offset, 0, is->is_size); + } + + /* See if this checksum matches. */ + zio_bad_cksum_t zbc; + ret = zio_checksum_error(zio, &zbc); + if (ret == 0) { + /* Found a matching checksum. Issue repair i/os. */ + vdev_indirect_repair(zio); + zio_checksum_verified(zio); + return; + } + + /* + * Checksum failed; try a different combination of split + * children. + */ + boolean_t more; +next: + more = B_FALSE; + if (segments <= zfs_reconstruct_indirect_segments_max) { + /* + * There are relatively few segments, so + * deterministically check all combinations. We do + * this by by adding one to the first split's + * good_child. If it overflows, then "carry over" to + * the next split (like counting in base is_children, + * but each digit can have a different base). + */ + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + is->is_good_child++; + if (is->is_good_child < is->is_children) { + more = B_TRUE; + break; + } + is->is_good_child = 0; + } + } else if (++attempts < attempts_max) { + /* + * There are too many combinations to try all of them + * in a reasonable amount of time, so try a fixed + * number of random combinations, after which we'll + * consider the block unrecoverable. + */ + for (indirect_split_t *is = list_head(&iv->iv_splits); + is != NULL; is = list_next(&iv->iv_splits, is)) { + is->is_good_child = + spa_get_random(is->is_children); + } + more = B_TRUE; + } + if (!more) { + /* All combinations failed. */ + zio->io_error = ret; + vdev_indirect_all_checksum_errors(zio); + zio_checksum_verified(zio); + return; + } + } +} + +static void +vdev_indirect_io_done(zio_t *zio) +{ + indirect_vsd_t *iv = zio->io_vsd; + + if (iv->iv_reconstruct) { + /* + * We have read all copies of the data (e.g. from mirrors), + * either because this was a scrub/resilver, or because the + * one-copy read didn't checksum correctly. + */ + vdev_indirect_reconstruct_io_done(zio); + return; + } + + if (!iv->iv_split_block) { + /* + * This was not a split block, so we passed the BP down, + * and the checksum was handled by the (one) child zio. + */ + return; + } + + zio_bad_cksum_t zbc; + int ret = zio_checksum_error(zio, &zbc); + if (ret == 0) { + zio_checksum_verified(zio); + return; + } + + /* + * The checksum didn't match. Read all copies of all splits, and + * then we will try to reconstruct. The next time + * vdev_indirect_io_done() is called, iv_reconstruct will be set. + */ + vdev_indirect_read_all(zio); + + zio_vdev_io_redone(zio); } vdev_ops_t vdev_indirect_ops = { Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_mirror.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_mirror.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_mirror.c Tue Jul 31 00:13:04 2018 (r336950) @@ -24,7 +24,7 @@ */ /* - * Copyright (c) 2012, 2015 by Delphix. All rights reserved. + * Copyright (c) 2012, 2018 by Delphix. All rights reserved. */ #include @@ -345,12 +345,15 @@ vdev_mirror_io_start(zio_t *zio) } if (zio->io_type == ZIO_TYPE_READ) { - if ((zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_resilvering) { + if (zio->io_bp != NULL && + (zio->io_flags & ZIO_FLAG_SCRUB) && !mm->mm_resilvering) { /* - * For scrubbing reads we need to allocate a read - * buffer for each child and issue reads to all - * children. If any child succeeds, it will copy its - * data into zio->io_data in vdev_mirror_scrub_done. + * For scrubbing reads (if we can verify the + * checksum here, as indicated by io_bp being + * non-NULL) we need to allocate a read buffer for + * each child and issue reads to all children. If + * any child succeeds, it will copy its data into + * zio->io_data in vdev_mirror_scrub_done. */ for (c = 0; c < mm->mm_children; c++) { mc = &mm->mm_child[c]; @@ -502,7 +505,21 @@ vdev_mirror_io_done(zio_t *zio) if (mc->mc_error == 0) { if (mc->mc_tried) continue; + /* + * We didn't try this child. We need to + * repair it if: + * 1. it's a scrub (in which case we have + * tried everything that was healthy) + * - or - + * 2. it's an indirect vdev (in which case + * it could point to any other vdev, which + * might have a bad DTL) + * - or - + * 3. the DTL indicates that this data is + * missing from this vdev + */ if (!(zio->io_flags & ZIO_FLAG_SCRUB) && + mc->mc_vd->vdev_ops != &vdev_indirect_ops && !vdev_dtl_contains(mc->mc_vd, DTL_PARTIAL, zio->io_txg, 1)) continue; Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_removal.c ============================================================================== --- vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_removal.c Tue Jul 31 00:02:42 2018 (r336949) +++ vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_removal.c Tue Jul 31 00:13:04 2018 (r336950) @@ -83,18 +83,12 @@ typedef struct vdev_copy_arg { kmutex_t vca_lock; } vdev_copy_arg_t; -typedef struct vdev_copy_seg_arg { - vdev_copy_arg_t *vcsa_copy_arg; - uint64_t vcsa_txg; - dva_t *vcsa_dest_dva; - blkptr_t *vcsa_dest_bp; -} vdev_copy_seg_arg_t; - /* - * The maximum amount of allowed data we're allowed to copy from a device - * at a time when removing it. + * The maximum amount of memory we can use for outstanding i/o while + * doing a device removal. This determines how much i/o we can have + * in flight concurrently. */ -int zfs_remove_max_copy_bytes = 8 * 1024 * 1024; +int zfs_remove_max_copy_bytes = 64 * 1024 * 1024; /* * The largest contiguous segment that we will attempt to allocate when @@ -176,7 +170,7 @@ spa_vdev_removal_create(vdev_t *vd) mutex_init(&svr->svr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&svr->svr_cv, NULL, CV_DEFAULT, NULL); svr->svr_allocd_segs = range_tree_create(NULL, NULL); - svr->svr_vdev = vd; + svr->svr_vdev_id = vd->vdev_id; for (int i = 0; i < TXG_SIZE; i++) { svr->svr_frees[i] = range_tree_create(NULL, NULL); @@ -218,9 +212,10 @@ spa_vdev_removal_destroy(spa_vdev_removal_t *svr) static void vdev_remove_initiate_sync(void *arg, dmu_tx_t *tx) { - vdev_t *vd = arg; + int vdev_id = (uintptr_t)arg; + spa_t *spa = dmu_tx_pool(tx)->dp_spa; + vdev_t *vd = vdev_lookup_top(spa, vdev_id); vdev_indirect_config_t *vic = &vd->vdev_indirect_config; - spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_dsl_pool->dp_meta_objset; spa_vdev_removal_t *svr = NULL; uint64_t txg = dmu_tx_get_txg(tx); @@ -342,7 +337,7 @@ vdev_remove_initiate_sync(void *arg, dmu_tx_t *tx) ASSERT3P(spa->spa_vdev_removal, ==, NULL); spa->spa_vdev_removal = svr; svr->svr_thread = thread_create(NULL, 0, - spa_vdev_remove_thread, vd, 0, &p0, TS_RUN, minclsyspri); + spa_vdev_remove_thread, spa, 0, &p0, TS_RUN, minclsyspri); } /* @@ -383,21 +378,24 @@ spa_remove_init(spa_t *spa) spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); vdev_t *vd = vdev_lookup_top(spa, spa->spa_removing_phys.sr_removing_vdev); - spa_config_exit(spa, SCL_STATE, FTAG); - if (vd == NULL) + if (vd == NULL) { + spa_config_exit(spa, SCL_STATE, FTAG); return (EINVAL); + } vdev_indirect_config_t *vic = &vd->vdev_indirect_config; ASSERT(vdev_is_concrete(vd)); spa_vdev_removal_t *svr = spa_vdev_removal_create(vd); - ASSERT(svr->svr_vdev->vdev_removing); + ASSERT3U(svr->svr_vdev_id, ==, vd->vdev_id); + ASSERT(vd->vdev_removing); vd->vdev_indirect_mapping = vdev_indirect_mapping_open( spa->spa_meta_objset, vic->vic_mapping_object); vd->vdev_indirect_births = vdev_indirect_births_open( spa->spa_meta_objset, vic->vic_births_object); + spa_config_exit(spa, SCL_STATE, FTAG); spa->spa_vdev_removal = svr; } @@ -450,15 +448,8 @@ spa_restart_removal(spa_t *spa) if (!spa_writeable(spa)) return; - vdev_t *vd = svr->svr_vdev; - vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; - - ASSERT3P(vd, !=, NULL); - ASSERT(vd->vdev_removing); - - zfs_dbgmsg("restarting removal of %llu at count=%llu", - vd->vdev_id, vdev_indirect_mapping_num_entries(vim)); - svr->svr_thread = thread_create(NULL, 0, spa_vdev_remove_thread, vd, + zfs_dbgmsg("restarting removal of %llu", svr->svr_vdev_id); + svr->svr_thread = thread_create(NULL, 0, spa_vdev_remove_thread, spa, 0, &p0, TS_RUN, minclsyspri); } @@ -479,7 +470,7 @@ free_from_removing_vdev(vdev_t *vd, uint64_t offset, u ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0); ASSERT3U(vd->vdev_indirect_config.vic_mapping_object, ==, vdev_indirect_mapping_object(vim)); - ASSERT3P(vd, ==, svr->svr_vdev); + ASSERT3U(vd->vdev_id, ==, svr->svr_vdev_id); mutex_enter(&svr->svr_lock); @@ -662,7 +653,7 @@ spa_finish_removal(spa_t *spa, dsl_scan_state_t state, if (state == DSS_FINISHED) { spa_removing_phys_t *srp = &spa->spa_removing_phys; - vdev_t *vd = svr->svr_vdev; + vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id); vdev_indirect_config_t *vic = &vd->vdev_indirect_config; if (srp->sr_prev_indirect_vdev != UINT64_MAX) { @@ -705,7 +696,7 @@ vdev_mapping_sync(void *arg, dmu_tx_t *tx) { spa_vdev_removal_t *svr = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; - vdev_t *vd = svr->svr_vdev; + vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id); vdev_indirect_config_t *vic = &vd->vdev_indirect_config; uint64_t txg = dmu_tx_get_txg(tx); vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; @@ -733,64 +724,128 @@ vdev_mapping_sync(void *arg, dmu_tx_t *tx) spa_sync_removing_state(spa, tx); } +/* + * All reads and writes associated with a call to spa_vdev_copy_segment() + * are done. + */ static void +spa_vdev_copy_nullzio_done(zio_t *zio) +{ + spa_config_exit(zio->io_spa, SCL_STATE, zio->io_spa); +} + +/* + * The write of the new location is done. + */ +static void spa_vdev_copy_segment_write_done(zio_t *zio) { - vdev_copy_seg_arg_t *vcsa = zio->io_private; - vdev_copy_arg_t *vca = vcsa->vcsa_copy_arg; - spa_config_exit(zio->io_spa, SCL_STATE, FTAG); + vdev_copy_arg_t *vca = zio->io_private; + abd_free(zio->io_abd); mutex_enter(&vca->vca_lock); vca->vca_outstanding_bytes -= zio->io_size; cv_signal(&vca->vca_cv); mutex_exit(&vca->vca_lock); - - ASSERT0(zio->io_error); - kmem_free(vcsa->vcsa_dest_bp, sizeof (blkptr_t)); - kmem_free(vcsa, sizeof (vdev_copy_seg_arg_t)); } +/* + * The read of the old location is done. The parent zio is the write to + * the new location. Allow it to start. + */ static void spa_vdev_copy_segment_read_done(zio_t *zio) { - vdev_copy_seg_arg_t *vcsa = zio->io_private; - dva_t *dest_dva = vcsa->vcsa_dest_dva; - uint64_t txg = vcsa->vcsa_txg; - spa_t *spa = zio->io_spa; - vdev_t *dest_vd = vdev_lookup_top(spa, DVA_GET_VDEV(dest_dva)); - blkptr_t *bp = NULL; - dva_t *dva = NULL; - uint64_t size = zio->io_size; + zio_nowait(zio_unique_parent(zio)); +} - ASSERT3P(dest_vd, !=, NULL); - ASSERT0(zio->io_error); +/* + * If the old and new vdevs are mirrors, we will read both sides of the old + * mirror, and write each copy to the corresponding side of the new mirror. + * If the old and new vdevs have a different number of children, we will do + * this as best as possible. Since we aren't verifying checksums, this + * ensures that as long as there's a good copy of the data, we'll have a + * good copy after the removal, even if there's silent damage to one side + * of the mirror. If we're removing a mirror that has some silent damage, + * we'll have exactly the same damage in the new location (assuming that + * the new location is also a mirror). + * + * We accomplish this by creating a tree of zio_t's, with as many writes as + * there are "children" of the new vdev (a non-redundant vdev counts as one + * child, a 2-way mirror has 2 children, etc). Each write has an associated + * read from a child of the old vdev. Typically there will be the same + * number of children of the old and new vdevs. However, if there are more + * children of the new vdev, some child(ren) of the old vdev will be issued *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***