diff options
Diffstat (limited to 'fs/bcachefs/io_read.c')
| -rw-r--r-- | fs/bcachefs/io_read.c | 1296 |
1 files changed, 807 insertions, 489 deletions
diff --git a/fs/bcachefs/io_read.c b/fs/bcachefs/io_read.c index 8a556e6d1ab6..cd184b219a65 100644 --- a/fs/bcachefs/io_read.c +++ b/fs/bcachefs/io_read.c @@ -9,6 +9,7 @@ #include "bcachefs.h" #include "alloc_background.h" #include "alloc_foreground.h" +#include "async_objs.h" #include "btree_update.h" #include "buckets.h" #include "checksum.h" @@ -17,15 +18,31 @@ #include "data_update.h" #include "disk_groups.h" #include "ec.h" +#include "enumerated_ref.h" #include "error.h" #include "io_read.h" #include "io_misc.h" #include "io_write.h" +#include "reflink.h" #include "subvolume.h" #include "trace.h" +#include <linux/moduleparam.h> +#include <linux/random.h> #include <linux/sched/mm.h> +#ifdef CONFIG_BCACHEFS_DEBUG +static unsigned bch2_read_corrupt_ratio; +module_param_named(read_corrupt_ratio, bch2_read_corrupt_ratio, uint, 0644); +MODULE_PARM_DESC(read_corrupt_ratio, ""); +#endif + +static bool bch2_poison_extents_on_checksum_error; +module_param_named(poison_extents_on_checksum_error, + bch2_poison_extents_on_checksum_error, bool, 0644); +MODULE_PARM_DESC(poison_extents_on_checksum_error, + "Extents with checksum errors are marked as poisoned - unsafe without read fua support"); + #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT static bool bch2_target_congested(struct bch_fs *c, u16 target) @@ -39,7 +56,7 @@ static bool bch2_target_congested(struct bch_fs *c, u16 target) if (!target) return false; - rcu_read_lock(); + guard(rcu)(); devs = bch2_target_to_mask(c, target) ?: &c->rw_devs[BCH_DATA_user]; @@ -56,9 +73,8 @@ static bool bch2_target_congested(struct bch_fs *c, u16 target) total += max(congested, 0LL); nr++; } - rcu_read_unlock(); - return bch2_rand_range(nr * CONGESTED_MAX) < total; + return get_random_u32_below(nr * CONGESTED_MAX) < total; } #else @@ -72,195 +88,233 @@ static bool bch2_target_congested(struct bch_fs *c, u16 target) /* Cache promotion on read */ -struct promote_op { - struct rcu_head rcu; - u64 start_time; +static const struct rhashtable_params bch_promote_params = { + .head_offset = offsetof(struct promote_op, hash), + .key_offset = offsetof(struct promote_op, pos), + .key_len = sizeof(struct bpos), + .automatic_shrinking = true, +}; - struct rhash_head hash; - struct bpos pos; +static inline bool have_io_error(struct bch_io_failures *failed) +{ + return failed && failed->nr; +} - struct data_update write; - struct bio_vec bi_inline_vecs[]; /* must be last */ -}; +static inline struct data_update *rbio_data_update(struct bch_read_bio *rbio) +{ + EBUG_ON(rbio->split); -static const struct rhashtable_params bch_promote_params = { - .head_offset = offsetof(struct promote_op, hash), - .key_offset = offsetof(struct promote_op, pos), - .key_len = sizeof(struct bpos), -}; + return rbio->data_update + ? container_of(rbio, struct data_update, rbio) + : NULL; +} + +static bool ptr_being_rewritten(struct bch_read_bio *orig, unsigned dev) +{ + struct data_update *u = rbio_data_update(orig); + if (!u) + return false; + + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(u->k.k)); + unsigned i = 0; + bkey_for_each_ptr(ptrs, ptr) { + if (ptr->dev == dev && + u->data_opts.rewrite_ptrs & BIT(i)) + return true; + i++; + } + + return false; +} static inline int should_promote(struct bch_fs *c, struct bkey_s_c k, struct bpos pos, struct bch_io_opts opts, - unsigned flags) + unsigned flags, + struct bch_io_failures *failed) { - BUG_ON(!opts.promote_target); + if (!have_io_error(failed)) { + BUG_ON(!opts.promote_target); - if (!(flags & BCH_READ_MAY_PROMOTE)) - return -BCH_ERR_nopromote_may_not; + if (!(flags & BCH_READ_may_promote)) + return bch_err_throw(c, nopromote_may_not); - if (bch2_bkey_has_target(c, k, opts.promote_target)) - return -BCH_ERR_nopromote_already_promoted; + if (bch2_bkey_has_target(c, k, opts.promote_target)) + return bch_err_throw(c, nopromote_already_promoted); - if (bkey_extent_is_unwritten(k)) - return -BCH_ERR_nopromote_unwritten; + if (bkey_extent_is_unwritten(k)) + return bch_err_throw(c, nopromote_unwritten); - if (bch2_target_congested(c, opts.promote_target)) - return -BCH_ERR_nopromote_congested; + if (bch2_target_congested(c, opts.promote_target)) + return bch_err_throw(c, nopromote_congested); + } if (rhashtable_lookup_fast(&c->promote_table, &pos, bch_promote_params)) - return -BCH_ERR_nopromote_in_flight; + return bch_err_throw(c, nopromote_in_flight); return 0; } -static void promote_free(struct bch_fs *c, struct promote_op *op) +static noinline void promote_free(struct bch_read_bio *rbio) { - int ret; + struct promote_op *op = container_of(rbio, struct promote_op, write.rbio); + struct bch_fs *c = rbio->c; + + int ret = rhashtable_remove_fast(&c->promote_table, &op->hash, + bch_promote_params); + BUG_ON(ret); + + async_object_list_del(c, promote, op->list_idx); bch2_data_update_exit(&op->write); - ret = rhashtable_remove_fast(&c->promote_table, &op->hash, - bch_promote_params); - BUG_ON(ret); - bch2_write_ref_put(c, BCH_WRITE_REF_promote); + enumerated_ref_put(&c->writes, BCH_WRITE_REF_promote); kfree_rcu(op, rcu); } static void promote_done(struct bch_write_op *wop) { - struct promote_op *op = - container_of(wop, struct promote_op, write.op); - struct bch_fs *c = op->write.op.c; + struct promote_op *op = container_of(wop, struct promote_op, write.op); + struct bch_fs *c = op->write.rbio.c; - bch2_time_stats_update(&c->times[BCH_TIME_data_promote], - op->start_time); - promote_free(c, op); + bch2_time_stats_update(&c->times[BCH_TIME_data_promote], op->start_time); + promote_free(&op->write.rbio); } -static void promote_start(struct promote_op *op, struct bch_read_bio *rbio) +static void promote_start_work(struct work_struct *work) { - struct bio *bio = &op->write.op.wbio.bio; + struct promote_op *op = container_of(work, struct promote_op, work); - trace_and_count(op->write.op.c, read_promote, &rbio->bio); + bch2_data_update_read_done(&op->write); +} - /* we now own pages: */ - BUG_ON(!rbio->bounce); - BUG_ON(rbio->bio.bi_vcnt > bio->bi_max_vecs); +static noinline void promote_start(struct bch_read_bio *rbio) +{ + struct promote_op *op = container_of(rbio, struct promote_op, write.rbio); - memcpy(bio->bi_io_vec, rbio->bio.bi_io_vec, - sizeof(struct bio_vec) * rbio->bio.bi_vcnt); - swap(bio->bi_vcnt, rbio->bio.bi_vcnt); + trace_and_count(op->write.op.c, io_read_promote, &rbio->bio); - bch2_data_update_read_done(&op->write, rbio->pick.crc); + INIT_WORK(&op->work, promote_start_work); + queue_work(rbio->c->write_ref_wq, &op->work); } -static struct promote_op *__promote_alloc(struct btree_trans *trans, - enum btree_id btree_id, - struct bkey_s_c k, - struct bpos pos, - struct extent_ptr_decoded *pick, - struct bch_io_opts opts, - unsigned sectors, - struct bch_read_bio **rbio) +static struct bch_read_bio *__promote_alloc(struct btree_trans *trans, + enum btree_id btree_id, + struct bkey_s_c k, + struct bpos pos, + struct extent_ptr_decoded *pick, + unsigned sectors, + struct bch_read_bio *orig, + struct bch_io_failures *failed) { struct bch_fs *c = trans->c; - struct promote_op *op = NULL; - struct bio *bio; - unsigned pages = DIV_ROUND_UP(sectors, PAGE_SECTORS); int ret; - if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_promote)) - return ERR_PTR(-BCH_ERR_nopromote_no_writes); - - op = kzalloc(struct_size(op, bi_inline_vecs, pages), GFP_KERNEL); - if (!op) { - ret = -BCH_ERR_nopromote_enomem; - goto err; - } + struct data_update_opts update_opts = { .write_flags = BCH_WRITE_alloc_nowait }; - op->start_time = local_clock(); - op->pos = pos; + if (!have_io_error(failed)) { + update_opts.target = orig->opts.promote_target; + update_opts.extra_replicas = 1; + update_opts.write_flags |= BCH_WRITE_cached; + update_opts.write_flags |= BCH_WRITE_only_specified_devs; + } else { + update_opts.target = orig->opts.foreground_target; + + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + unsigned ptr_bit = 1; + bkey_for_each_ptr(ptrs, ptr) { + if (bch2_dev_io_failures(failed, ptr->dev) && + !ptr_being_rewritten(orig, ptr->dev)) + update_opts.rewrite_ptrs |= ptr_bit; + ptr_bit <<= 1; + } - /* - * We don't use the mempool here because extents that aren't - * checksummed or compressed can be too big for the mempool: - */ - *rbio = kzalloc(sizeof(struct bch_read_bio) + - sizeof(struct bio_vec) * pages, - GFP_KERNEL); - if (!*rbio) { - ret = -BCH_ERR_nopromote_enomem; - goto err; + if (!update_opts.rewrite_ptrs) + return NULL; } - rbio_init(&(*rbio)->bio, opts); - bio_init(&(*rbio)->bio, NULL, (*rbio)->bio.bi_inline_vecs, pages, 0); + if (!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_promote)) + return ERR_PTR(-BCH_ERR_nopromote_no_writes); - if (bch2_bio_alloc_pages(&(*rbio)->bio, sectors << 9, GFP_KERNEL)) { - ret = -BCH_ERR_nopromote_enomem; - goto err; + struct promote_op *op = kzalloc(sizeof(*op), GFP_KERNEL); + if (!op) { + ret = bch_err_throw(c, nopromote_enomem); + goto err_put; } - (*rbio)->bounce = true; - (*rbio)->split = true; - (*rbio)->kmalloc = true; + op->start_time = local_clock(); + op->pos = pos; if (rhashtable_lookup_insert_fast(&c->promote_table, &op->hash, bch_promote_params)) { - ret = -BCH_ERR_nopromote_in_flight; + ret = bch_err_throw(c, nopromote_in_flight); goto err; } - bio = &op->write.op.wbio.bio; - bio_init(bio, NULL, bio->bi_inline_vecs, pages, 0); + ret = async_object_list_add(c, promote, op, &op->list_idx); + if (ret < 0) + goto err_remove_hash; ret = bch2_data_update_init(trans, NULL, NULL, &op->write, writepoint_hashed((unsigned long) current), - opts, - (struct data_update_opts) { - .target = opts.promote_target, - .extra_replicas = 1, - .write_flags = BCH_WRITE_ALLOC_NOWAIT|BCH_WRITE_CACHED, - }, + &orig->opts, + update_opts, btree_id, k); + op->write.type = BCH_DATA_UPDATE_promote; /* * possible errors: -BCH_ERR_nocow_lock_blocked, * -BCH_ERR_ENOSPC_disk_reservation: */ - if (ret) { - BUG_ON(rhashtable_remove_fast(&c->promote_table, &op->hash, - bch_promote_params)); - goto err; - } + if (ret) + goto err_remove_list; + rbio_init_fragment(&op->write.rbio.bio, orig); + op->write.rbio.bounce = true; + op->write.rbio.promote = true; op->write.op.end_io = promote_done; - return op; + return &op->write.rbio; +err_remove_list: + async_object_list_del(c, promote, op->list_idx); +err_remove_hash: + BUG_ON(rhashtable_remove_fast(&c->promote_table, &op->hash, + bch_promote_params)); err: - if (*rbio) - bio_free_pages(&(*rbio)->bio); - kfree(*rbio); - *rbio = NULL; - kfree(op); - bch2_write_ref_put(c, BCH_WRITE_REF_promote); + bio_free_pages(&op->write.op.wbio.bio); + /* We may have added to the rhashtable and thus need rcu freeing: */ + kfree_rcu(op, rcu); +err_put: + enumerated_ref_put(&c->writes, BCH_WRITE_REF_promote); return ERR_PTR(ret); } noinline -static struct promote_op *promote_alloc(struct btree_trans *trans, +static struct bch_read_bio *promote_alloc(struct btree_trans *trans, struct bvec_iter iter, struct bkey_s_c k, struct extent_ptr_decoded *pick, - struct bch_io_opts opts, unsigned flags, - struct bch_read_bio **rbio, + struct bch_read_bio *orig, bool *bounce, - bool *read_full) + bool *read_full, + struct bch_io_failures *failed) { + /* + * We're in the retry path, but we don't know what to repair yet, and we + * don't want to do a promote here: + */ + if (failed && !failed->nr) + return NULL; + struct bch_fs *c = trans->c; - bool promote_full = *read_full || READ_ONCE(c->promote_whole_extents); + /* + * if failed != NULL we're not actually doing a promote, we're + * recovering from an io/checksum error + */ + bool promote_full = (have_io_error(failed) || + *read_full || + READ_ONCE(c->opts.promote_whole_extents)); /* data might have to be decompressed in the write path: */ unsigned sectors = promote_full ? max(pick->crc.compressed_size, pick->crc.live_size) @@ -268,35 +322,72 @@ static struct promote_op *promote_alloc(struct btree_trans *trans, struct bpos pos = promote_full ? bkey_start_pos(k.k) : POS(k.k->p.inode, iter.bi_sector); - struct promote_op *promote; int ret; - ret = should_promote(c, k, pos, opts, flags); + ret = should_promote(c, k, pos, orig->opts, flags, failed); if (ret) goto nopromote; - promote = __promote_alloc(trans, - k.k->type == KEY_TYPE_reflink_v - ? BTREE_ID_reflink - : BTREE_ID_extents, - k, pos, pick, opts, sectors, rbio); + struct bch_read_bio *promote = + __promote_alloc(trans, + k.k->type == KEY_TYPE_reflink_v + ? BTREE_ID_reflink + : BTREE_ID_extents, + k, pos, pick, sectors, orig, failed); + if (!promote) + return NULL; + ret = PTR_ERR_OR_ZERO(promote); if (ret) goto nopromote; *bounce = true; *read_full = promote_full; + + if (have_io_error(failed)) + orig->self_healing = true; + return promote; nopromote: - trace_read_nopromote(c, ret); + trace_io_read_nopromote(c, ret); return NULL; } +void bch2_promote_op_to_text(struct printbuf *out, struct promote_op *op) +{ + if (!op->write.read_done) { + prt_printf(out, "parent read: %px\n", op->write.rbio.parent); + printbuf_indent_add(out, 2); + bch2_read_bio_to_text(out, op->write.rbio.parent); + printbuf_indent_sub(out, 2); + } + + bch2_data_update_to_text(out, &op->write); +} + /* Read */ -#define READ_RETRY_AVOID 1 -#define READ_RETRY 2 -#define READ_ERR 3 +static int bch2_read_err_msg_trans(struct btree_trans *trans, struct printbuf *out, + struct bch_read_bio *rbio, struct bpos read_pos) +{ + int ret = lockrestart_do(trans, + bch2_inum_offset_err_msg_trans(trans, out, + (subvol_inum) { rbio->subvol, read_pos.inode }, + read_pos.offset << 9)); + if (ret) + return ret; + + if (rbio->data_update) + prt_str(out, "(internal move) "); + + return 0; +} + +static void bch2_read_err_msg(struct bch_fs *c, struct printbuf *out, + struct bch_read_bio *rbio, struct bpos read_pos) +{ + bch2_trans_run(c, bch2_read_err_msg_trans(trans, out, rbio, read_pos)); +} enum rbio_context { RBIO_CONTEXT_NULL, @@ -328,20 +419,27 @@ static inline struct bch_read_bio *bch2_rbio_free(struct bch_read_bio *rbio) { BUG_ON(rbio->bounce && !rbio->split); - if (rbio->promote) - promote_free(rbio->c, rbio->promote); - rbio->promote = NULL; - - if (rbio->bounce) - bch2_bio_free_pages_pool(rbio->c, &rbio->bio); + if (rbio->have_ioref) { + struct bch_dev *ca = bch2_dev_have_ref(rbio->c, rbio->pick.ptr.dev); + enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_io_read); + } if (rbio->split) { struct bch_read_bio *parent = rbio->parent; - if (rbio->kmalloc) - kfree(rbio); - else + if (unlikely(rbio->promote)) { + if (!rbio->bio.bi_status) + promote_start(rbio); + else + promote_free(rbio); + } else { + async_object_list_del(rbio->c, rbio, rbio->list_idx); + + if (rbio->bounce) + bch2_bio_free_pages_pool(rbio->c, &rbio->bio); + bio_put(&rbio->bio); + } rbio = parent; } @@ -361,61 +459,118 @@ static void bch2_rbio_done(struct bch_read_bio *rbio) bio_endio(&rbio->bio); } -static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio, - struct bvec_iter bvec_iter, - struct bch_io_failures *failed, - unsigned flags) +static void get_rbio_extent(struct btree_trans *trans, + struct bch_read_bio *rbio, + struct bkey_buf *sk) { - struct btree_trans *trans = bch2_trans_get(c); struct btree_iter iter; - struct bkey_buf sk; struct bkey_s_c k; - int ret; + int ret = lockrestart_do(trans, + bkey_err(k = bch2_bkey_get_iter(trans, &iter, + rbio->data_btree, rbio->data_pos, 0))); + if (ret) + return; + + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + bkey_for_each_ptr(ptrs, ptr) + if (bch2_extent_ptr_eq(*ptr, rbio->pick.ptr)) { + bch2_bkey_buf_reassemble(sk, trans->c, k); + break; + } - flags &= ~BCH_READ_LAST_FRAGMENT; - flags |= BCH_READ_MUST_CLONE; + bch2_trans_iter_exit(trans, &iter); +} - bch2_bkey_buf_init(&sk); +static noinline int maybe_poison_extent(struct btree_trans *trans, struct bch_read_bio *rbio, + enum btree_id btree, struct bkey_s_c read_k) +{ + if (!bch2_poison_extents_on_checksum_error) + return 0; + + struct bch_fs *c = trans->c; + + struct data_update *u = rbio_data_update(rbio); + if (u) + read_k = bkey_i_to_s_c(u->k.k); + + u64 flags = bch2_bkey_extent_flags(read_k); + if (flags & BIT_ULL(BCH_EXTENT_FLAG_poisoned)) + return 0; - bch2_trans_iter_init(trans, &iter, rbio->data_btree, - rbio->read_pos, BTREE_ITER_SLOTS); + struct btree_iter iter; + struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, btree, bkey_start_pos(read_k.k), + BTREE_ITER_intent); + int ret = bkey_err(k); + if (ret) + return ret; + + if (!bkey_and_val_eq(k, read_k)) + goto out; + + struct bkey_i *new = bch2_trans_kmalloc(trans, + bkey_bytes(k.k) + sizeof(struct bch_extent_flags)); + ret = PTR_ERR_OR_ZERO(new) ?: + (bkey_reassemble(new, k), 0) ?: + bch2_bkey_extent_flags_set(c, new, flags|BIT_ULL(BCH_EXTENT_FLAG_poisoned)) ?: + bch2_trans_update(trans, &iter, new, BTREE_UPDATE_internal_snapshot_node) ?: + bch2_trans_commit(trans, NULL, NULL, 0); + + /* + * Propagate key change back to data update path, in particular so it + * knows the extent has been poisoned and it's safe to change the + * checksum + */ + if (u && !ret) + bch2_bkey_buf_copy(&u->k, c, new); +out: + bch2_trans_iter_exit(trans, &iter); + return ret; +} + +static noinline int bch2_read_retry_nodecode(struct btree_trans *trans, + struct bch_read_bio *rbio, + struct bvec_iter bvec_iter, + struct bch_io_failures *failed, + unsigned flags) +{ + struct data_update *u = container_of(rbio, struct data_update, rbio); retry: - rbio->bio.bi_status = 0; + bch2_trans_begin(trans); - k = bch2_btree_iter_peek_slot(&iter); - if (bkey_err(k)) + struct btree_iter iter; + struct bkey_s_c k; + int ret = lockrestart_do(trans, + bkey_err(k = bch2_bkey_get_iter(trans, &iter, + u->btree_id, bkey_start_pos(&u->k.k->k), + 0))); + if (ret) goto err; - bch2_bkey_buf_reassemble(&sk, c, k); - k = bkey_i_to_s_c(sk.k); - bch2_trans_unlock(trans); - - if (!bch2_bkey_matches_ptr(c, k, - rbio->pick.ptr, - rbio->data_pos.offset - - rbio->pick.crc.offset)) { + if (!bkey_and_val_eq(k, bkey_i_to_s_c(u->k.k))) { /* extent we wanted to read no longer exists: */ - rbio->hole = true; - goto out; + rbio->ret = bch_err_throw(trans->c, data_read_key_overwritten); + goto err; } ret = __bch2_read_extent(trans, rbio, bvec_iter, - rbio->read_pos, - rbio->data_btree, - k, 0, failed, flags); - if (ret == READ_RETRY) - goto retry; - if (ret) - goto err; -out: - bch2_rbio_done(rbio); - bch2_trans_iter_exit(trans, &iter); - bch2_trans_put(trans); - bch2_bkey_buf_exit(&sk, c); - return; + bkey_start_pos(&u->k.k->k), + u->btree_id, + bkey_i_to_s_c(u->k.k), + 0, failed, flags, -1); err: - rbio->bio.bi_status = BLK_STS_IOERR; - goto out; + bch2_trans_iter_exit(trans, &iter); + + if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || + bch2_err_matches(ret, BCH_ERR_data_read_retry)) + goto retry; + + if (ret) { + rbio->bio.bi_status = BLK_STS_IOERR; + rbio->ret = ret; + } + + BUG_ON(atomic_read(&rbio->bio.__bi_remaining) != 1); + return ret; } static void bch2_rbio_retry(struct work_struct *work) @@ -431,44 +586,107 @@ static void bch2_rbio_retry(struct work_struct *work) }; struct bch_io_failures failed = { .nr = 0 }; - trace_and_count(c, read_retry, &rbio->bio); + struct btree_trans *trans = bch2_trans_get(c); + + struct bkey_buf sk; + bch2_bkey_buf_init(&sk); + bkey_init(&sk.k->k); - if (rbio->retry == READ_RETRY_AVOID) - bch2_mark_io_failure(&failed, &rbio->pick); + trace_io_read_retry(&rbio->bio); + this_cpu_add(c->counters[BCH_COUNTER_io_read_retry], + bvec_iter_sectors(rbio->bvec_iter)); - rbio->bio.bi_status = 0; + get_rbio_extent(trans, rbio, &sk); + + if (!bkey_deleted(&sk.k->k) && + bch2_err_matches(rbio->ret, BCH_ERR_data_read_retry_avoid)) + bch2_mark_io_failure(&failed, &rbio->pick, + rbio->ret == -BCH_ERR_data_read_retry_csum_err); + + if (!rbio->split) { + rbio->bio.bi_status = 0; + rbio->ret = 0; + } + + unsigned subvol = rbio->subvol; + struct bpos read_pos = rbio->read_pos; rbio = bch2_rbio_free(rbio); - flags |= BCH_READ_IN_RETRY; - flags &= ~BCH_READ_MAY_PROMOTE; + flags |= BCH_READ_in_retry; + flags &= ~BCH_READ_may_promote; + flags &= ~BCH_READ_last_fragment; + flags |= BCH_READ_must_clone; - if (flags & BCH_READ_NODECODE) { - bch2_read_retry_nodecode(c, rbio, iter, &failed, flags); - } else { - flags &= ~BCH_READ_LAST_FRAGMENT; - flags |= BCH_READ_MUST_CLONE; + int ret = rbio->data_update + ? bch2_read_retry_nodecode(trans, rbio, iter, &failed, flags) + : __bch2_read(trans, rbio, iter, inum, &failed, &sk, flags); - __bch2_read(c, rbio, iter, inum, &failed, flags); + if (ret) { + rbio->ret = ret; + rbio->bio.bi_status = BLK_STS_IOERR; } + + if (failed.nr || ret) { + struct printbuf buf = PRINTBUF; + bch2_log_msg_start(c, &buf); + + lockrestart_do(trans, + bch2_inum_offset_err_msg_trans(trans, &buf, + (subvol_inum) { subvol, read_pos.inode }, + read_pos.offset << 9)); + if (rbio->data_update) + prt_str(&buf, "(internal move) "); + + prt_str(&buf, "data read error, "); + if (!ret) { + prt_str(&buf, "successful retry"); + if (rbio->self_healing) + prt_str(&buf, ", self healing"); + } else + prt_str(&buf, bch2_err_str(ret)); + prt_newline(&buf); + + + if (!bkey_deleted(&sk.k->k)) { + bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(sk.k)); + prt_newline(&buf); + } + + bch2_io_failures_to_text(&buf, c, &failed); + + bch2_print_str_ratelimited(c, KERN_ERR, buf.buf); + printbuf_exit(&buf); + } + + bch2_rbio_done(rbio); + bch2_bkey_buf_exit(&sk, c); + bch2_trans_put(trans); } -static void bch2_rbio_error(struct bch_read_bio *rbio, int retry, - blk_status_t error) +static void bch2_rbio_error(struct bch_read_bio *rbio, + int ret, blk_status_t blk_error) { - rbio->retry = retry; + BUG_ON(ret >= 0); + + rbio->ret = ret; + rbio->bio.bi_status = blk_error; - if (rbio->flags & BCH_READ_IN_RETRY) + bch2_rbio_parent(rbio)->saw_error = true; + + if (rbio->flags & BCH_READ_in_retry) return; - if (retry == READ_ERR) { + if (bch2_err_matches(ret, BCH_ERR_data_read_retry)) { + bch2_rbio_punt(rbio, bch2_rbio_retry, + RBIO_CONTEXT_UNBOUND, system_unbound_wq); + } else { rbio = bch2_rbio_free(rbio); - rbio->bio.bi_status = error; + rbio->ret = ret; + rbio->bio.bi_status = blk_error; + bch2_rbio_done(rbio); - } else { - bch2_rbio_punt(rbio, bch2_rbio_retry, - RBIO_CONTEXT_UNBOUND, system_unbound_wq); } } @@ -487,11 +705,11 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans, return 0; k = bch2_bkey_get_iter(trans, &iter, rbio->data_btree, rbio->data_pos, - BTREE_ITER_SLOTS|BTREE_ITER_INTENT); + BTREE_ITER_slots|BTREE_ITER_intent); if ((ret = bkey_err(k))) goto out; - if (bversion_cmp(k.k->version, rbio->version) || + if (bversion_cmp(k.k->bversion, rbio->version) || !bch2_bkey_matches_ptr(c, k, rbio->pick.ptr, data_offset)) goto out; @@ -523,7 +741,7 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans, goto out; ret = bch2_trans_update(trans, &iter, new, - BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE); + BTREE_UPDATE_internal_snapshot_node); out: bch2_trans_iter_exit(trans, &iter); return ret; @@ -531,8 +749,48 @@ out: static noinline void bch2_rbio_narrow_crcs(struct bch_read_bio *rbio) { - bch2_trans_do(rbio->c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, - __bch2_rbio_narrow_crcs(trans, rbio)); + bch2_trans_commit_do(rbio->c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc, + __bch2_rbio_narrow_crcs(trans, rbio)); +} + +static void bch2_read_decompress_err(struct work_struct *work) +{ + struct bch_read_bio *rbio = + container_of(work, struct bch_read_bio, work); + struct bch_fs *c = rbio->c; + struct printbuf buf = PRINTBUF; + + bch2_read_err_msg(c, &buf, rbio, rbio->read_pos); + prt_str(&buf, "decompression error"); + + struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL; + if (ca) + bch_err_ratelimited(ca, "%s", buf.buf); + else + bch_err_ratelimited(c, "%s", buf.buf); + + bch2_rbio_error(rbio, -BCH_ERR_data_read_decompress_err, BLK_STS_IOERR); + printbuf_exit(&buf); +} + +static void bch2_read_decrypt_err(struct work_struct *work) +{ + struct bch_read_bio *rbio = + container_of(work, struct bch_read_bio, work); + struct bch_fs *c = rbio->c; + struct printbuf buf = PRINTBUF; + + bch2_read_err_msg(c, &buf, rbio, rbio->read_pos); + prt_str(&buf, "decrypt error"); + + struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL; + if (ca) + bch_err_ratelimited(ca, "%s", buf.buf); + else + bch_err_ratelimited(c, "%s", buf.buf); + + bch2_rbio_error(rbio, -BCH_ERR_data_read_decrypt_err, BLK_STS_IOERR); + printbuf_exit(&buf); } /* Inner part that may run in process context */ @@ -541,10 +799,11 @@ static void __bch2_read_endio(struct work_struct *work) struct bch_read_bio *rbio = container_of(work, struct bch_read_bio, work); struct bch_fs *c = rbio->c; - struct bch_dev *ca = bch_dev_bkey_exists(c, rbio->pick.ptr.dev); - struct bio *src = &rbio->bio; - struct bio *dst = &bch2_rbio_parent(rbio)->bio; - struct bvec_iter dst_iter = rbio->bvec_iter; + struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL; + struct bch_read_bio *parent = bch2_rbio_parent(rbio); + struct bio *src = &rbio->bio; + struct bio *dst = &parent->bio; + struct bvec_iter dst_iter = rbio->bvec_iter; struct bch_extent_crc_unpacked crc = rbio->pick.crc; struct nonce nonce = extent_nonce(rbio->version, crc); unsigned nofs_flags; @@ -562,8 +821,26 @@ static void __bch2_read_endio(struct work_struct *work) src->bi_iter = rbio->bvec_iter; } + bch2_maybe_corrupt_bio(src, bch2_read_corrupt_ratio); + csum = bch2_checksum_bio(c, crc.csum_type, nonce, src); - if (bch2_crc_cmp(csum, rbio->pick.crc.csum) && !c->opts.no_data_io) + bool csum_good = !bch2_crc_cmp(csum, rbio->pick.crc.csum) || c->opts.no_data_io; + + /* + * Checksum error: if the bio wasn't bounced, we may have been + * reading into buffers owned by userspace (that userspace can + * scribble over) - retry the read, bouncing it this time: + */ + if (!csum_good && !rbio->bounce && (rbio->flags & BCH_READ_user_mapped)) { + rbio->flags |= BCH_READ_must_bounce; + bch2_rbio_error(rbio, -BCH_ERR_data_read_retry_csum_err_maybe_userspace, + BLK_STS_IOERR); + goto out; + } + + bch2_account_io_completion(ca, BCH_MEMBER_ERROR_checksum, 0, csum_good); + + if (!csum_good) goto csum_err; /* @@ -576,32 +853,40 @@ static void __bch2_read_endio(struct work_struct *work) if (unlikely(rbio->narrow_crcs)) bch2_rbio_narrow_crcs(rbio); - if (rbio->flags & BCH_READ_NODECODE) - goto nodecode; + if (likely(!parent->data_update)) { + /* Adjust crc to point to subset of data we want: */ + crc.offset += rbio->offset_into_extent; + crc.live_size = bvec_iter_sectors(rbio->bvec_iter); - /* Adjust crc to point to subset of data we want: */ - crc.offset += rbio->offset_into_extent; - crc.live_size = bvec_iter_sectors(rbio->bvec_iter); + if (crc_is_compressed(crc)) { + ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src); + if (ret) + goto decrypt_err; - if (crc_is_compressed(crc)) { - ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src); - if (ret) - goto decrypt_err; + if (bch2_bio_uncompress(c, src, dst, dst_iter, crc) && + !c->opts.no_data_io) + goto decompression_err; + } else { + /* don't need to decrypt the entire bio: */ + nonce = nonce_add(nonce, crc.offset << 9); + bio_advance(src, crc.offset << 9); - if (bch2_bio_uncompress(c, src, dst, dst_iter, crc) && - !c->opts.no_data_io) - goto decompression_err; - } else { - /* don't need to decrypt the entire bio: */ - nonce = nonce_add(nonce, crc.offset << 9); - bio_advance(src, crc.offset << 9); + BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size); + src->bi_iter.bi_size = dst_iter.bi_size; - BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size); - src->bi_iter.bi_size = dst_iter.bi_size; + ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src); + if (ret) + goto decrypt_err; - ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src); - if (ret) - goto decrypt_err; + if (rbio->bounce) { + struct bvec_iter src_iter = src->bi_iter; + + bio_copy_data_iter(dst, &dst_iter, src, &src_iter); + } + } + } else { + if (rbio->split) + rbio->parent->pick = rbio->pick; if (rbio->bounce) { struct bvec_iter src_iter = src->bi_iter; @@ -618,12 +903,9 @@ static void __bch2_read_endio(struct work_struct *work) ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src); if (ret) goto decrypt_err; - - promote_start(rbio->promote, rbio); - rbio->promote = NULL; } -nodecode: - if (likely(!(rbio->flags & BCH_READ_IN_RETRY))) { + + if (likely(!(rbio->flags & BCH_READ_in_retry))) { rbio = bch2_rbio_free(rbio); bch2_rbio_done(rbio); } @@ -631,42 +913,13 @@ out: memalloc_nofs_restore(nofs_flags); return; csum_err: - /* - * Checksum error: if the bio wasn't bounced, we may have been - * reading into buffers owned by userspace (that userspace can - * scribble over) - retry the read, bouncing it this time: - */ - if (!rbio->bounce && (rbio->flags & BCH_READ_USER_MAPPED)) { - rbio->flags |= BCH_READ_MUST_BOUNCE; - bch2_rbio_error(rbio, READ_RETRY, BLK_STS_IOERR); - goto out; - } - - struct printbuf buf = PRINTBUF; - buf.atomic++; - prt_str(&buf, "data "); - bch2_csum_err_msg(&buf, crc.csum_type, rbio->pick.crc.csum, csum); - - bch_err_inum_offset_ratelimited(ca, - rbio->read_pos.inode, - rbio->read_pos.offset << 9, - "data %s", buf.buf); - printbuf_exit(&buf); - - bch2_io_error(ca, BCH_MEMBER_ERROR_checksum); - bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR); + bch2_rbio_error(rbio, -BCH_ERR_data_read_retry_csum_err, BLK_STS_IOERR); goto out; decompression_err: - bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode, - rbio->read_pos.offset << 9, - "decompression error"); - bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR); + bch2_rbio_punt(rbio, bch2_read_decompress_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq); goto out; decrypt_err: - bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode, - rbio->read_pos.offset << 9, - "decrypt error"); - bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR); + bch2_rbio_punt(rbio, bch2_read_decrypt_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq); goto out; } @@ -675,35 +928,29 @@ static void bch2_read_endio(struct bio *bio) struct bch_read_bio *rbio = container_of(bio, struct bch_read_bio, bio); struct bch_fs *c = rbio->c; - struct bch_dev *ca = bch_dev_bkey_exists(c, rbio->pick.ptr.dev); + struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL; struct workqueue_struct *wq = NULL; enum rbio_context context = RBIO_CONTEXT_NULL; - if (rbio->have_ioref) { - bch2_latency_acct(ca, rbio->submit_time, READ); - percpu_ref_put(&ca->io_ref); - } + bch2_account_io_completion(ca, BCH_MEMBER_ERROR_read, + rbio->submit_time, !bio->bi_status); if (!rbio->split) rbio->bio.bi_end_io = rbio->end_io; - if (bch2_dev_inum_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_read, - rbio->read_pos.inode, - rbio->read_pos.offset, - "data read error: %s", - bch2_blk_status_to_str(bio->bi_status))) { - bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status); + if (unlikely(bio->bi_status)) { + bch2_rbio_error(rbio, -BCH_ERR_data_read_retry_io_err, bio->bi_status); return; } - if (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) || - ptr_stale(ca, &rbio->pick.ptr)) { - trace_and_count(c, read_reuse_race, &rbio->bio); + if (((rbio->flags & BCH_READ_retry_if_stale) && race_fault()) || + (ca && dev_ptr_stale(ca, &rbio->pick.ptr))) { + trace_and_count(c, io_read_reuse_race, &rbio->bio); - if (rbio->flags & BCH_READ_RETRY_IF_STALE) - bch2_rbio_error(rbio, READ_RETRY, BLK_STS_AGAIN); + if (rbio->flags & BCH_READ_retry_if_stale) + bch2_rbio_error(rbio, -BCH_ERR_data_read_ptr_stale_retry, BLK_STS_AGAIN); else - bch2_rbio_error(rbio, READ_ERR, BLK_STS_AGAIN); + bch2_rbio_error(rbio, -BCH_ERR_data_read_ptr_stale_race, BLK_STS_AGAIN); return; } @@ -718,72 +965,45 @@ static void bch2_read_endio(struct bio *bio) bch2_rbio_punt(rbio, __bch2_read_endio, context, wq); } -int __bch2_read_indirect_extent(struct btree_trans *trans, - unsigned *offset_into_extent, - struct bkey_buf *orig_k) -{ - struct btree_iter iter; - struct bkey_s_c k; - u64 reflink_offset; - int ret; - - reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) + - *offset_into_extent; - - k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_reflink, - POS(0, reflink_offset), 0); - ret = bkey_err(k); - if (ret) - goto err; - - if (k.k->type != KEY_TYPE_reflink_v && - k.k->type != KEY_TYPE_indirect_inline_data) { - bch_err_inum_offset_ratelimited(trans->c, - orig_k->k->k.p.inode, - orig_k->k->k.p.offset << 9, - "%llu len %u points to nonexistent indirect extent %llu", - orig_k->k->k.p.offset, - orig_k->k->k.size, - reflink_offset); - bch2_inconsistent_error(trans->c); - ret = -EIO; - goto err; - } - - *offset_into_extent = iter.pos.offset - bkey_start_offset(k.k); - bch2_bkey_buf_reassemble(orig_k, trans->c, k); -err: - bch2_trans_iter_exit(trans, &iter); - return ret; -} - static noinline void read_from_stale_dirty_pointer(struct btree_trans *trans, + struct bch_dev *ca, struct bkey_s_c k, struct bch_extent_ptr ptr) { struct bch_fs *c = trans->c; - struct bch_dev *ca = bch_dev_bkey_exists(c, ptr.dev); struct btree_iter iter; struct printbuf buf = PRINTBUF; int ret; bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, - PTR_BUCKET_POS(c, &ptr), - BTREE_ITER_CACHED); + PTR_BUCKET_POS(ca, &ptr), + BTREE_ITER_cached); - prt_printf(&buf, "Attempting to read from stale dirty pointer:"); - printbuf_indent_add(&buf, 2); - prt_newline(&buf); + int gen = bucket_gen_get(ca, iter.pos.offset); + if (gen >= 0) { + prt_printf(&buf, "Attempting to read from stale dirty pointer:\n"); + printbuf_indent_add(&buf, 2); - bch2_bkey_val_to_text(&buf, c, k); - prt_newline(&buf); + bch2_bkey_val_to_text(&buf, c, k); + prt_newline(&buf); - prt_printf(&buf, "memory gen: %u", *bucket_gen(ca, iter.pos.offset)); + prt_printf(&buf, "memory gen: %u", gen); + + ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(trans, &iter))); + if (!ret) { + prt_newline(&buf); + bch2_bkey_val_to_text(&buf, c, k); + } + } else { + prt_printf(&buf, "Attempting to read from invalid bucket %llu:%llu:\n", + iter.pos.inode, iter.pos.offset); + printbuf_indent_add(&buf, 2); + + prt_printf(&buf, "first bucket %u nbuckets %llu\n", + ca->mi.first_bucket, ca->mi.nbuckets); - ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter))); - if (!ret) { - prt_newline(&buf); bch2_bkey_val_to_text(&buf, c, k); + prt_newline(&buf); } bch2_fs_inconsistent(c, "%s", buf.buf); @@ -796,16 +1016,15 @@ int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig, struct bvec_iter iter, struct bpos read_pos, enum btree_id data_btree, struct bkey_s_c k, unsigned offset_into_extent, - struct bch_io_failures *failed, unsigned flags) + struct bch_io_failures *failed, unsigned flags, int dev) { struct bch_fs *c = trans->c; struct extent_ptr_decoded pick; struct bch_read_bio *rbio = NULL; - struct bch_dev *ca = NULL; - struct promote_op *promote = NULL; bool bounce = false, read_full = false, narrow_crcs = false; struct bpos data_pos = bkey_start_pos(k.k); - int pick_ret; + struct data_update *u = rbio_data_update(orig); + int ret = 0; if (bkey_extent_is_inline_data(k.k)) { unsigned bytes = min_t(unsigned, iter.bi_size, @@ -816,23 +1035,57 @@ int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig, swap(iter.bi_size, bytes); bio_advance_iter(&orig->bio, &iter, bytes); zero_fill_bio_iter(&orig->bio, iter); + this_cpu_add(c->counters[BCH_COUNTER_io_read_inline], + bvec_iter_sectors(iter)); goto out_read_done; } + + if ((bch2_bkey_extent_flags(k) & BIT_ULL(BCH_EXTENT_FLAG_poisoned)) && + !orig->data_update) + return bch_err_throw(c, extent_poisoned); retry_pick: - pick_ret = bch2_bkey_pick_read_device(c, k, failed, &pick); + ret = bch2_bkey_pick_read_device(c, k, failed, &pick, dev); /* hole or reservation - just zero fill: */ - if (!pick_ret) + if (!ret) goto hole; - if (pick_ret < 0) { - bch_err_inum_offset_ratelimited(c, - read_pos.inode, read_pos.offset << 9, - "no device to read from"); + if (unlikely(ret < 0)) { + if (ret == -BCH_ERR_data_read_csum_err) { + int ret2 = maybe_poison_extent(trans, orig, data_btree, k); + if (ret2) { + ret = ret2; + goto err; + } + + trace_and_count(c, io_read_fail_and_poison, &orig->bio); + } + + struct printbuf buf = PRINTBUF; + bch2_read_err_msg_trans(trans, &buf, orig, read_pos); + prt_printf(&buf, "%s\n ", bch2_err_str(ret)); + bch2_bkey_val_to_text(&buf, c, k); + + bch_err_ratelimited(c, "%s", buf.buf); + printbuf_exit(&buf); goto err; } - ca = bch_dev_bkey_exists(c, pick.ptr.dev); + if (unlikely(bch2_csum_type_is_encryption(pick.crc.csum_type)) && + !c->chacha20_key_set) { + struct printbuf buf = PRINTBUF; + bch2_read_err_msg_trans(trans, &buf, orig, read_pos); + prt_printf(&buf, "attempting to read encrypted data without encryption key\n "); + bch2_bkey_val_to_text(&buf, c, k); + + bch_err_ratelimited(c, "%s", buf.buf); + printbuf_exit(&buf); + ret = bch_err_throw(c, data_read_no_encryption_key); + goto err; + } + + struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ, + BCH_DEV_READ_REF_io_read); /* * Stale dirty pointers are treated as IO errors, but @failed isn't @@ -840,57 +1093,58 @@ retry_pick: * retry path, don't check here, it'll be caught in bch2_read_endio() * and we'll end up in the retry path: */ - if ((flags & BCH_READ_IN_RETRY) && + if ((flags & BCH_READ_in_retry) && !pick.ptr.cached && - unlikely(ptr_stale(ca, &pick.ptr))) { - read_from_stale_dirty_pointer(trans, k, pick.ptr); - bch2_mark_io_failure(failed, &pick); + ca && + unlikely(dev_ptr_stale(ca, &pick.ptr))) { + read_from_stale_dirty_pointer(trans, ca, k, pick.ptr); + bch2_mark_io_failure(failed, &pick, false); + enumerated_ref_put(&ca->io_ref[READ], BCH_DEV_READ_REF_io_read); goto retry_pick; } - /* - * Unlock the iterator while the btree node's lock is still in - * cache, before doing the IO: - */ - bch2_trans_unlock(trans); + if (likely(!u)) { + if (!(flags & BCH_READ_last_fragment) || + bio_flagged(&orig->bio, BIO_CHAIN)) + flags |= BCH_READ_must_clone; + + narrow_crcs = !(flags & BCH_READ_in_retry) && + bch2_can_narrow_extent_crcs(k, pick.crc); + + if (narrow_crcs && (flags & BCH_READ_user_mapped)) + flags |= BCH_READ_must_bounce; - if (flags & BCH_READ_NODECODE) { + EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size); + + if (crc_is_compressed(pick.crc) || + (pick.crc.csum_type != BCH_CSUM_none && + (bvec_iter_sectors(iter) != pick.crc.uncompressed_size || + (bch2_csum_type_is_encryption(pick.crc.csum_type) && + (flags & BCH_READ_user_mapped)) || + (flags & BCH_READ_must_bounce)))) { + read_full = true; + bounce = true; + } + } else { /* * can happen if we retry, and the extent we were going to read * has been merged in the meantime: */ - if (pick.crc.compressed_size > orig->bio.bi_vcnt * PAGE_SECTORS) - goto hole; + if (pick.crc.compressed_size > u->op.wbio.bio.bi_iter.bi_size) { + if (ca) + enumerated_ref_put(&ca->io_ref[READ], + BCH_DEV_READ_REF_io_read); + rbio->ret = bch_err_throw(c, data_read_buffer_too_small); + goto out_read_done; + } iter.bi_size = pick.crc.compressed_size << 9; - goto get_bio; - } - - if (!(flags & BCH_READ_LAST_FRAGMENT) || - bio_flagged(&orig->bio, BIO_CHAIN)) - flags |= BCH_READ_MUST_CLONE; - - narrow_crcs = !(flags & BCH_READ_IN_RETRY) && - bch2_can_narrow_extent_crcs(k, pick.crc); - - if (narrow_crcs && (flags & BCH_READ_USER_MAPPED)) - flags |= BCH_READ_MUST_BOUNCE; - - EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size); - - if (crc_is_compressed(pick.crc) || - (pick.crc.csum_type != BCH_CSUM_none && - (bvec_iter_sectors(iter) != pick.crc.uncompressed_size || - (bch2_csum_type_is_encryption(pick.crc.csum_type) && - (flags & BCH_READ_USER_MAPPED)) || - (flags & BCH_READ_MUST_BOUNCE)))) { read_full = true; - bounce = true; } - if (orig->opts.promote_target) - promote = promote_alloc(trans, iter, k, &pick, orig->opts, flags, - &rbio, &bounce, &read_full); + if (orig->opts.promote_target || have_io_error(failed)) + rbio = promote_alloc(trans, iter, k, &pick, flags, orig, + &bounce, &read_full, failed); if (!read_full) { EBUG_ON(crc_is_compressed(pick.crc)); @@ -909,7 +1163,7 @@ retry_pick: pick.crc.offset = 0; pick.crc.live_size = bvec_iter_sectors(iter); } -get_bio: + if (rbio) { /* * promote already allocated bounce rbio: @@ -924,17 +1178,16 @@ get_bio: } else if (bounce) { unsigned sectors = pick.crc.compressed_size; - rbio = rbio_init(bio_alloc_bioset(NULL, + rbio = rbio_init_fragment(bio_alloc_bioset(NULL, DIV_ROUND_UP(sectors, PAGE_SECTORS), 0, GFP_NOFS, &c->bio_read_split), - orig->opts); + orig); bch2_bio_alloc_pages_pool(c, &rbio->bio, sectors << 9); rbio->bounce = true; - rbio->split = true; - } else if (flags & BCH_READ_MUST_CLONE) { + } else if (flags & BCH_READ_must_clone) { /* * Have to clone if there were any splits, due to error * reporting issues (if a split errored, and retrying didn't @@ -943,11 +1196,10 @@ get_bio: * from the whole bio, in which case we don't want to retry and * lose the error) */ - rbio = rbio_init(bio_alloc_clone(NULL, &orig->bio, GFP_NOFS, + rbio = rbio_init_fragment(bio_alloc_clone(NULL, &orig->bio, GFP_NOFS, &c->bio_read_split), - orig->opts); + orig); rbio->bio.bi_iter = iter; - rbio->split = true; } else { rbio = orig; rbio->bio.bi_iter = iter; @@ -956,61 +1208,66 @@ get_bio: EBUG_ON(bio_sectors(&rbio->bio) != pick.crc.compressed_size); - rbio->c = c; rbio->submit_time = local_clock(); - if (rbio->split) - rbio->parent = orig; - else + if (!rbio->split) rbio->end_io = orig->bio.bi_end_io; rbio->bvec_iter = iter; rbio->offset_into_extent= offset_into_extent; rbio->flags = flags; - rbio->have_ioref = pick_ret > 0 && bch2_dev_get_ioref(ca, READ); + rbio->have_ioref = ca != NULL; rbio->narrow_crcs = narrow_crcs; - rbio->hole = 0; - rbio->retry = 0; + rbio->ret = 0; rbio->context = 0; - /* XXX: only initialize this if needed */ - rbio->devs_have = bch2_bkey_devs(k); rbio->pick = pick; rbio->subvol = orig->subvol; rbio->read_pos = read_pos; rbio->data_btree = data_btree; rbio->data_pos = data_pos; - rbio->version = k.k->version; - rbio->promote = promote; + rbio->version = k.k->bversion; INIT_WORK(&rbio->work, NULL); rbio->bio.bi_opf = orig->bio.bi_opf; rbio->bio.bi_iter.bi_sector = pick.ptr.offset; rbio->bio.bi_end_io = bch2_read_endio; + async_object_list_add(c, rbio, rbio, &rbio->list_idx); + if (rbio->bounce) - trace_and_count(c, read_bounce, &rbio->bio); + trace_and_count(c, io_read_bounce, &rbio->bio); - this_cpu_add(c->counters[BCH_COUNTER_io_read], bio_sectors(&rbio->bio)); + if (!u) + this_cpu_add(c->counters[BCH_COUNTER_io_read], bio_sectors(&rbio->bio)); + else + this_cpu_add(c->counters[BCH_COUNTER_io_move_read], bio_sectors(&rbio->bio)); bch2_increment_clock(c, bio_sectors(&rbio->bio), READ); /* * If it's being moved internally, we don't want to flag it as a cache * hit: */ - if (pick.ptr.cached && !(flags & BCH_READ_NODECODE)) + if (ca && pick.ptr.cached && !u) bch2_bucket_io_time_reset(trans, pick.ptr.dev, PTR_BUCKET_NR(ca, &pick.ptr), READ); - if (!(flags & (BCH_READ_IN_RETRY|BCH_READ_LAST_FRAGMENT))) { + if (!(flags & (BCH_READ_in_retry|BCH_READ_last_fragment))) { bio_inc_remaining(&orig->bio); - trace_and_count(c, read_split, &orig->bio); + trace_and_count(c, io_read_split, &orig->bio); } - if (!rbio->pick.idx) { - if (!rbio->have_ioref) { - bch_err_inum_offset_ratelimited(c, - read_pos.inode, - read_pos.offset << 9, - "no device to read from"); - bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR); + /* + * Unlock the iterator while the btree node's lock is still in + * cache, before doing the IO: + */ + if (!(flags & BCH_READ_in_retry)) + bch2_trans_unlock(trans); + else + bch2_trans_unlock_long(trans); + + if (likely(!rbio->pick.do_ec_reconstruct)) { + if (unlikely(!rbio->have_ioref)) { + bch2_rbio_error(rbio, + -BCH_ERR_data_read_retry_device_offline, + BLK_STS_IOERR); goto out; } @@ -1019,10 +1276,10 @@ get_bio: bio_set_dev(&rbio->bio, ca->disk_sb.bdev); if (unlikely(c->opts.no_data_io)) { - if (likely(!(flags & BCH_READ_IN_RETRY))) + if (likely(!(flags & BCH_READ_in_retry))) bio_endio(&rbio->bio); } else { - if (likely(!(flags & BCH_READ_IN_RETRY))) + if (likely(!(flags & BCH_READ_in_retry))) submit_bio(&rbio->bio); else submit_bio_wait(&rbio->bio); @@ -1035,164 +1292,217 @@ get_bio: trans->notrace_relock_fail = true; } else { /* Attempting reconstruct read: */ - if (bch2_ec_read_extent(trans, rbio)) { - bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR); + if (bch2_ec_read_extent(trans, rbio, k)) { + bch2_rbio_error(rbio, -BCH_ERR_data_read_retry_ec_reconstruct_err, + BLK_STS_IOERR); goto out; } - if (likely(!(flags & BCH_READ_IN_RETRY))) + if (likely(!(flags & BCH_READ_in_retry))) bio_endio(&rbio->bio); } out: - if (likely(!(flags & BCH_READ_IN_RETRY))) { + if (likely(!(flags & BCH_READ_in_retry))) { return 0; } else { + bch2_trans_unlock(trans); + int ret; rbio->context = RBIO_CONTEXT_UNBOUND; bch2_read_endio(&rbio->bio); - ret = rbio->retry; + ret = rbio->ret; rbio = bch2_rbio_free(rbio); - if (ret == READ_RETRY_AVOID) { - bch2_mark_io_failure(failed, &pick); - ret = READ_RETRY; - } - - if (!ret) - goto out_read_done; + if (bch2_err_matches(ret, BCH_ERR_data_read_retry_avoid)) + bch2_mark_io_failure(failed, &pick, + ret == -BCH_ERR_data_read_retry_csum_err); return ret; } err: - if (flags & BCH_READ_IN_RETRY) - return READ_ERR; + if (flags & BCH_READ_in_retry) + return ret; - orig->bio.bi_status = BLK_STS_IOERR; + orig->bio.bi_status = BLK_STS_IOERR; + orig->ret = ret; goto out_read_done; hole: + this_cpu_add(c->counters[BCH_COUNTER_io_read_hole], + bvec_iter_sectors(iter)); /* - * won't normally happen in the BCH_READ_NODECODE - * (bch2_move_extent()) path, but if we retry and the extent we wanted - * to read no longer exists we have to signal that: + * won't normally happen in the data update (bch2_move_extent()) path, + * but if we retry and the extent we wanted to read no longer exists we + * have to signal that: */ - if (flags & BCH_READ_NODECODE) - orig->hole = true; + if (u) + orig->ret = bch_err_throw(c, data_read_key_overwritten); zero_fill_bio_iter(&orig->bio, iter); out_read_done: - if (flags & BCH_READ_LAST_FRAGMENT) + if ((flags & BCH_READ_last_fragment) && + !(flags & BCH_READ_in_retry)) bch2_rbio_done(orig); return 0; } -void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio, - struct bvec_iter bvec_iter, subvol_inum inum, - struct bch_io_failures *failed, unsigned flags) +int __bch2_read(struct btree_trans *trans, struct bch_read_bio *rbio, + struct bvec_iter bvec_iter, subvol_inum inum, + struct bch_io_failures *failed, + struct bkey_buf *prev_read, + unsigned flags) { - struct btree_trans *trans = bch2_trans_get(c); + struct bch_fs *c = trans->c; struct btree_iter iter; struct bkey_buf sk; struct bkey_s_c k; - u32 snapshot; + enum btree_id data_btree; int ret; - BUG_ON(flags & BCH_READ_NODECODE); + EBUG_ON(rbio->data_update); bch2_bkey_buf_init(&sk); -retry: - bch2_trans_begin(trans); - iter = (struct btree_iter) { NULL }; - - ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); - if (ret) - goto err; - bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, - SPOS(inum.inum, bvec_iter.bi_sector, snapshot), - BTREE_ITER_SLOTS); + POS(inum.inum, bvec_iter.bi_sector), + BTREE_ITER_slots); + while (1) { - unsigned bytes, sectors, offset_into_extent; - enum btree_id data_btree = BTREE_ID_extents; + data_btree = BTREE_ID_extents; - /* - * read_extent -> io_time_reset may cause a transaction restart - * without returning an error, we need to check for that here: - */ - ret = bch2_trans_relock(trans); + bch2_trans_begin(trans); + + u32 snapshot; + ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); if (ret) - break; + goto err; + + bch2_btree_iter_set_snapshot(trans, &iter, snapshot); - bch2_btree_iter_set_pos(&iter, + bch2_btree_iter_set_pos(trans, &iter, POS(inum.inum, bvec_iter.bi_sector)); - k = bch2_btree_iter_peek_slot(&iter); + k = bch2_btree_iter_peek_slot(trans, &iter); ret = bkey_err(k); if (ret) - break; + goto err; - offset_into_extent = iter.pos.offset - + s64 offset_into_extent = iter.pos.offset - bkey_start_offset(k.k); - sectors = k.k->size - offset_into_extent; + unsigned sectors = k.k->size - offset_into_extent; bch2_bkey_buf_reassemble(&sk, c, k); ret = bch2_read_indirect_extent(trans, &data_btree, &offset_into_extent, &sk); if (ret) - break; + goto err; k = bkey_i_to_s_c(sk.k); + if (unlikely(flags & BCH_READ_in_retry)) { + if (!bkey_and_val_eq(k, bkey_i_to_s_c(prev_read->k))) + failed->nr = 0; + bch2_bkey_buf_copy(prev_read, c, sk.k); + } + /* * With indirect extents, the amount of data to read is the min * of the original extent and the indirect extent: */ - sectors = min(sectors, k.k->size - offset_into_extent); + sectors = min_t(unsigned, sectors, k.k->size - offset_into_extent); - bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9; + unsigned bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9; swap(bvec_iter.bi_size, bytes); if (bvec_iter.bi_size == bytes) - flags |= BCH_READ_LAST_FRAGMENT; + flags |= BCH_READ_last_fragment; ret = __bch2_read_extent(trans, rbio, bvec_iter, iter.pos, data_btree, k, - offset_into_extent, failed, flags); + offset_into_extent, failed, flags, -1); + swap(bvec_iter.bi_size, bytes); + if (ret) - break; + goto err; - if (flags & BCH_READ_LAST_FRAGMENT) + if (flags & BCH_READ_last_fragment) break; - swap(bvec_iter.bi_size, bytes); bio_advance_iter(&rbio->bio, &bvec_iter, bytes); +err: + if (ret == -BCH_ERR_data_read_retry_csum_err_maybe_userspace) + flags |= BCH_READ_must_bounce; - ret = btree_trans_too_many_iters(trans); - if (ret) + if (ret && + !bch2_err_matches(ret, BCH_ERR_transaction_restart) && + !bch2_err_matches(ret, BCH_ERR_data_read_retry)) break; } -err: - bch2_trans_iter_exit(trans, &iter); - if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || - ret == READ_RETRY || - ret == READ_RETRY_AVOID) - goto retry; + if (unlikely(ret)) { + if (ret != -BCH_ERR_extent_poisoned) { + struct printbuf buf = PRINTBUF; + lockrestart_do(trans, + bch2_inum_offset_err_msg_trans(trans, &buf, inum, + bvec_iter.bi_sector << 9)); + prt_printf(&buf, "data read error: %s", bch2_err_str(ret)); + bch_err_ratelimited(c, "%s", buf.buf); + printbuf_exit(&buf); + } - bch2_trans_put(trans); - bch2_bkey_buf_exit(&sk, c); + rbio->bio.bi_status = BLK_STS_IOERR; + rbio->ret = ret; - if (ret) { - bch_err_inum_offset_ratelimited(c, inum.inum, - bvec_iter.bi_sector << 9, - "read error %i from btree lookup", ret); - rbio->bio.bi_status = BLK_STS_IOERR; - bch2_rbio_done(rbio); + if (!(flags & BCH_READ_in_retry)) + bch2_rbio_done(rbio); } + + bch2_trans_iter_exit(trans, &iter); + bch2_bkey_buf_exit(&sk, c); + return ret; +} + +static const char * const bch2_read_bio_flags[] = { +#define x(n) #n, + BCH_READ_FLAGS() +#undef x + NULL +}; + +void bch2_read_bio_to_text(struct printbuf *out, struct bch_read_bio *rbio) +{ + u64 now = local_clock(); + prt_printf(out, "start_time:\t%llu\n", rbio->start_time ? now - rbio->start_time : 0); + prt_printf(out, "submit_time:\t%llu\n", rbio->submit_time ? now - rbio->submit_time : 0); + + if (!rbio->split) + prt_printf(out, "end_io:\t%ps\n", rbio->end_io); + else + prt_printf(out, "parent:\t%px\n", rbio->parent); + + prt_printf(out, "bi_end_io:\t%ps\n", rbio->bio.bi_end_io); + + prt_printf(out, "promote:\t%u\n", rbio->promote); + prt_printf(out, "bounce:\t%u\n", rbio->bounce); + prt_printf(out, "split:\t%u\n", rbio->split); + prt_printf(out, "have_ioref:\t%u\n", rbio->have_ioref); + prt_printf(out, "narrow_crcs:\t%u\n", rbio->narrow_crcs); + prt_printf(out, "context:\t%u\n", rbio->context); + + int ret = READ_ONCE(rbio->ret); + if (ret < 0) + prt_printf(out, "ret:\t%s\n", bch2_err_str(ret)); + else + prt_printf(out, "ret:\t%i\n", ret); + + prt_printf(out, "flags:\t"); + bch2_prt_bitflags(out, bch2_read_bio_flags, rbio->flags); + prt_newline(out); + + bch2_bio_to_text(out, &rbio->bio); } void bch2_fs_io_read_exit(struct bch_fs *c) @@ -1201,20 +1511,28 @@ void bch2_fs_io_read_exit(struct bch_fs *c) rhashtable_destroy(&c->promote_table); bioset_exit(&c->bio_read_split); bioset_exit(&c->bio_read); + mempool_exit(&c->bio_bounce_pages); } int bch2_fs_io_read_init(struct bch_fs *c) { + if (mempool_init_page_pool(&c->bio_bounce_pages, + max_t(unsigned, + c->opts.btree_node_size, + c->opts.encoded_extent_max) / + PAGE_SIZE, 0)) + return bch_err_throw(c, ENOMEM_bio_bounce_pages_init); + if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio), BIOSET_NEED_BVECS)) - return -BCH_ERR_ENOMEM_bio_read_init; + return bch_err_throw(c, ENOMEM_bio_read_init); if (bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio), BIOSET_NEED_BVECS)) - return -BCH_ERR_ENOMEM_bio_read_split_init; + return bch_err_throw(c, ENOMEM_bio_read_split_init); if (rhashtable_init(&c->promote_table, &bch_promote_params)) - return -BCH_ERR_ENOMEM_promote_table_init; + return bch_err_throw(c, ENOMEM_promote_table_init); return 0; } |