X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fio.c;h=092ece2c8bdee20594ad57fc34148acf4b7a87f7;hb=07ec713e056a73337bc0f53f5910f5179537b2c2;hp=5c9c3cf54edd1c70c80c857626428c428ffbed22;hpb=da730dc67c1bd10842c49a1534fe23e0d8fdb4be;p=bcachefs-tools-debian diff --git a/libbcachefs/io.c b/libbcachefs/io.c index 5c9c3cf..092ece2 100644 --- a/libbcachefs/io.c +++ b/libbcachefs/io.c @@ -7,8 +7,9 @@ */ #include "bcachefs.h" +#include "alloc_background.h" #include "alloc_foreground.h" -#include "bkey_on_stack.h" +#include "bkey_buf.h" #include "bset.h" #include "btree_update.h" #include "buckets.h" @@ -119,7 +120,7 @@ void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw) * the time: */ if (abs((int) (old - io_latency)) < (old >> 1) && - now & ~(~0 << 5)) + now & ~(~0U << 5)) break; new = ewma_add(old, io_latency, 5); @@ -170,7 +171,7 @@ void bch2_bio_alloc_pages_pool(struct bch_fs *c, struct bio *bio, while (size) { struct page *page = __bio_alloc_page_pool(c, &using_mempool); - unsigned len = min(PAGE_SIZE, size); + unsigned len = min_t(size_t, PAGE_SIZE, size); BUG_ON(!bio_add_page(bio, page, len, 0)); size -= len; @@ -182,39 +183,47 @@ void bch2_bio_alloc_pages_pool(struct bch_fs *c, struct bio *bio, /* Extent update path: */ -static int sum_sector_overwrites(struct btree_trans *trans, - struct btree_iter *extent_iter, - struct bkey_i *new, - bool may_allocate, - bool *maybe_extending, - s64 *delta) +int bch2_sum_sector_overwrites(struct btree_trans *trans, + struct btree_iter *extent_iter, + struct bkey_i *new, + bool *maybe_extending, + bool *usage_increasing, + s64 *i_sectors_delta, + s64 *disk_sectors_delta) { + struct bch_fs *c = trans->c; struct btree_iter *iter; struct bkey_s_c old; + unsigned new_replicas = bch2_bkey_replicas(c, bkey_i_to_s_c(new)); + bool new_compressed = bch2_bkey_sectors_compressed(bkey_i_to_s_c(new)); int ret = 0; - *maybe_extending = true; - *delta = 0; + *maybe_extending = true; + *usage_increasing = false; + *i_sectors_delta = 0; + *disk_sectors_delta = 0; iter = bch2_trans_copy_iter(trans, extent_iter); - if (IS_ERR(iter)) - return PTR_ERR(iter); for_each_btree_key_continue(iter, BTREE_ITER_SLOTS, old, ret) { - if (!may_allocate && - bch2_bkey_nr_ptrs_fully_allocated(old) < - bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(new))) { - ret = -ENOSPC; - break; - } + s64 sectors = min(new->k.p.offset, old.k->p.offset) - + max(bkey_start_offset(&new->k), + bkey_start_offset(old.k)); - *delta += (min(new->k.p.offset, - old.k->p.offset) - - max(bkey_start_offset(&new->k), - bkey_start_offset(old.k))) * + *i_sectors_delta += sectors * (bkey_extent_is_allocation(&new->k) - bkey_extent_is_allocation(old.k)); + *disk_sectors_delta += sectors * bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(new)); + *disk_sectors_delta -= new->k.p.snapshot == old.k->p.snapshot + ? sectors * bch2_bkey_nr_ptrs_fully_allocated(old) + : 0; + + if (!*usage_increasing && + (new_replicas > bch2_bkey_replicas(c, old) || + (!new_compressed && bch2_bkey_sectors_compressed(old)))) + *usage_increasing = true; + if (bkey_cmp(old.k->p, new->k.p) >= 0) { /* * Check if there's already data above where we're @@ -248,36 +257,53 @@ int bch2_extent_update(struct btree_trans *trans, struct disk_reservation *disk_res, u64 *journal_seq, u64 new_i_size, - s64 *i_sectors_delta) + s64 *i_sectors_delta_total, + bool check_enospc) { /* this must live until after bch2_trans_commit(): */ struct bkey_inode_buf inode_p; - bool extending = false; - s64 delta = 0; + bool extending = false, usage_increasing; + s64 i_sectors_delta = 0, disk_sectors_delta = 0; int ret; ret = bch2_extent_trim_atomic(k, iter); if (ret) return ret; - ret = sum_sector_overwrites(trans, iter, k, - disk_res && disk_res->sectors != 0, - &extending, &delta); + ret = bch2_sum_sector_overwrites(trans, iter, k, + &extending, + &usage_increasing, + &i_sectors_delta, + &disk_sectors_delta); if (ret) return ret; + if (!usage_increasing) + check_enospc = false; + + if (disk_res && + disk_sectors_delta > (s64) disk_res->sectors) { + ret = bch2_disk_reservation_add(trans->c, disk_res, + disk_sectors_delta - disk_res->sectors, + !check_enospc + ? BCH_DISK_RESERVATION_NOFAIL : 0); + if (ret) + return ret; + } + new_i_size = extending ? min(k->k.p.offset << 9, new_i_size) : 0; - if (delta || new_i_size) { + if (i_sectors_delta || new_i_size) { struct btree_iter *inode_iter; struct bch_inode_unpacked inode_u; inode_iter = bch2_inode_peek(trans, &inode_u, k->k.p.inode, BTREE_ITER_INTENT); - if (IS_ERR(inode_iter)) - return PTR_ERR(inode_iter); + ret = PTR_ERR_OR_ZERO(inode_iter); + if (ret) + return ret; /* * XXX: @@ -297,27 +323,34 @@ int bch2_extent_update(struct btree_trans *trans, else new_i_size = 0; - inode_u.bi_sectors += delta; + inode_u.bi_sectors += i_sectors_delta; - if (delta || new_i_size) { - bch2_inode_pack(&inode_p, &inode_u); - bch2_trans_update(trans, inode_iter, + if (i_sectors_delta || new_i_size) { + bch2_inode_pack(trans->c, &inode_p, &inode_u); + + inode_p.inode.k.p.snapshot = iter->snapshot; + + ret = bch2_trans_update(trans, inode_iter, &inode_p.inode.k_i, 0); } bch2_trans_iter_put(trans, inode_iter); - } - bch2_trans_update(trans, iter, k, 0); + if (ret) + return ret; + } - ret = bch2_trans_commit(trans, disk_res, journal_seq, + ret = bch2_trans_update(trans, iter, k, 0) ?: + bch2_trans_commit(trans, disk_res, journal_seq, BTREE_INSERT_NOCHECK_RW| - BTREE_INSERT_NOFAIL| - BTREE_INSERT_USE_RESERVE); - if (!ret && i_sectors_delta) - *i_sectors_delta += delta; + BTREE_INSERT_NOFAIL); + BUG_ON(ret == -ENOSPC); + if (ret) + return ret; - return ret; + if (i_sectors_delta_total) + *i_sectors_delta_total += i_sectors_delta; + return 0; } int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter, @@ -350,7 +383,7 @@ int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter, ret = bch2_extent_update(trans, iter, &delete, &disk_res, journal_seq, - 0, i_sectors_delta); + 0, i_sectors_delta, false); bch2_disk_reservation_put(c, &disk_res); btree_err: if (ret == -EINTR) { @@ -377,12 +410,14 @@ int bch2_fpunch(struct bch_fs *c, u64 inum, u64 start, u64 end, int ret = 0; bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); - iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, + iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, POS(inum, start), BTREE_ITER_INTENT); ret = bch2_fpunch_at(&trans, iter, POS(inum, end), journal_seq, i_sectors_delta); + + bch2_trans_iter_put(&trans, iter); bch2_trans_exit(&trans); if (ret == -EINTR) @@ -394,17 +429,17 @@ int bch2_fpunch(struct bch_fs *c, u64 inum, u64 start, u64 end, int bch2_write_index_default(struct bch_write_op *op) { struct bch_fs *c = op->c; - struct bkey_on_stack sk; + struct bkey_buf sk; struct keylist *keys = &op->insert_keys; struct bkey_i *k = bch2_keylist_front(keys); struct btree_trans trans; struct btree_iter *iter; int ret; - bkey_on_stack_init(&sk); + bch2_bkey_buf_init(&sk); bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024); - iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, + iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, bkey_start_pos(&k->k), BTREE_ITER_SLOTS|BTREE_ITER_INTENT); @@ -413,13 +448,16 @@ int bch2_write_index_default(struct bch_write_op *op) k = bch2_keylist_front(keys); - bkey_on_stack_realloc(&sk, c, k->k.u64s); + k->k.p.snapshot = iter->snapshot; + + bch2_bkey_buf_realloc(&sk, c, k->k.u64s); bkey_copy(sk.k, k); bch2_cut_front(iter->pos, sk.k); ret = bch2_extent_update(&trans, iter, sk.k, &op->res, op_journal_seq(op), - op->new_i_size, &op->i_sectors_delta); + op->new_i_size, &op->i_sectors_delta, + op->flags & BCH_WRITE_CHECK_ENOSPC); if (ret == -EINTR) continue; if (ret) @@ -429,8 +467,9 @@ int bch2_write_index_default(struct bch_write_op *op) bch2_keylist_pop_front(keys); } while (!bch2_keylist_empty(keys)); + bch2_trans_iter_put(&trans, iter); bch2_trans_exit(&trans); - bkey_on_stack_exit(&sk, c); + bch2_bkey_buf_exit(&sk, c); return ret; } @@ -478,9 +517,6 @@ void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c, n->submit_time = local_clock(); n->bio.bi_iter.bi_sector = ptr->offset; - if (!journal_flushes_device(ca)) - n->bio.bi_opf |= REQ_FUA; - if (likely(n->have_ioref)) { this_cpu_add(ca->io_done->sectors[WRITE][type], bio_sectors(&n->bio)); @@ -510,9 +546,6 @@ static void bch2_write_done(struct closure *cl) bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time); - if (!(op->flags & BCH_WRITE_FROM_INTERNAL)) - up(&c->io_in_flight); - if (op->end_io) { EBUG_ON(cl->parent); closure_debug_destroy(cl); @@ -577,7 +610,8 @@ static void __bch2_write_index(struct bch_write_op *op) op->written += sectors_start - keylist_sectors(keys); if (ret) { - __bcache_io_error(c, "btree IO error %i", ret); + bch_err_inum_ratelimited(c, op->pos.inode, + "write error %i from btree update", ret); op->error = ret; } } @@ -622,7 +656,10 @@ static void bch2_write_endio(struct bio *bio) struct bch_fs *c = wbio->c; struct bch_dev *ca = bch_dev_bkey_exists(c, wbio->dev); - if (bch2_dev_io_err_on(bio->bi_status, ca, "data write: %s", + if (bch2_dev_inum_io_err_on(bio->bi_status, ca, + op->pos.inode, + op->pos.offset - bio_sectors(bio), /* XXX definitely wrong */ + "data write error: %s", bch2_blk_status_to_str(bio->bi_status))) set_bit(wbio->dev, op->failed.d); @@ -1280,25 +1317,18 @@ void bch2_write(struct closure *cl) wbio_init(bio)->put_bio = false; if (bio_sectors(bio) & (c->opts.block_size - 1)) { - __bcache_io_error(c, "misaligned write"); + bch_err_inum_ratelimited(c, op->pos.inode, + "misaligned write"); op->error = -EIO; goto err; } if (c->opts.nochanges || !percpu_ref_tryget(&c->writes)) { - if (!(op->flags & BCH_WRITE_FROM_INTERNAL)) - __bcache_io_error(c, "read only"); op->error = -EROFS; goto err; } - /* - * Can't ratelimit copygc - we'd deadlock: - */ - if (!(op->flags & BCH_WRITE_FROM_INTERNAL)) - down(&c->io_in_flight); - bch2_increment_clock(c, bio_sectors(bio), WRITE); data_len = min_t(u64, bio->bi_iter.bi_size, @@ -1413,7 +1443,7 @@ static void promote_start(struct promote_op *op, struct bch_read_bio *rbio) bch2_migrate_read_done(&op->write, rbio); closure_init(cl, NULL); - closure_call(&op->write.op.cl, bch2_write, c->wq, cl); + closure_call(&op->write.op.cl, bch2_write, c->btree_update_wq, cl); closure_return_with_destructor(cl, promote_done); } @@ -1474,7 +1504,8 @@ static struct promote_op *__promote_alloc(struct bch_fs *c, opts, DATA_PROMOTE, (struct data_opts) { - .target = opts.promote_target + .target = opts.promote_target, + .nr_replicas = 1, }, btree_id, k); BUG_ON(ret); @@ -1516,8 +1547,8 @@ static struct promote_op *promote_alloc(struct bch_fs *c, promote = __promote_alloc(c, k.k->type == KEY_TYPE_reflink_v - ? BTREE_ID_REFLINK - : BTREE_ID_EXTENTS, + ? BTREE_ID_reflink + : BTREE_ID_extents, k, pos, pick, opts, sectors, rbio); if (!promote) return NULL; @@ -1603,18 +1634,18 @@ static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio { struct btree_trans trans; struct btree_iter *iter; - struct bkey_on_stack sk; + struct bkey_buf sk; struct bkey_s_c k; int ret; flags &= ~BCH_READ_LAST_FRAGMENT; flags |= BCH_READ_MUST_CLONE; - bkey_on_stack_init(&sk); + bch2_bkey_buf_init(&sk); bch2_trans_init(&trans, c, 0, 0); - iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, - rbio->pos, BTREE_ITER_SLOTS); + iter = bch2_trans_get_iter(&trans, rbio->data_btree, + rbio->read_pos, BTREE_ITER_SLOTS); retry: rbio->bio.bi_status = 0; @@ -1622,108 +1653,38 @@ retry: if (bkey_err(k)) goto err; - bkey_on_stack_reassemble(&sk, c, k); + 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->pos.offset - + rbio->data_pos.offset - rbio->pick.crc.offset)) { /* extent we wanted to read no longer exists: */ rbio->hole = true; goto out; } - ret = __bch2_read_extent(c, rbio, bvec_iter, k, 0, failed, flags); + 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_put(&trans, iter); bch2_trans_exit(&trans); - bkey_on_stack_exit(&sk, c); + bch2_bkey_buf_exit(&sk, c); return; err: rbio->bio.bi_status = BLK_STS_IOERR; goto out; } -static void bch2_read_retry(struct bch_fs *c, struct bch_read_bio *rbio, - struct bvec_iter bvec_iter, u64 inode, - struct bch_io_failures *failed, unsigned flags) -{ - struct btree_trans trans; - struct btree_iter *iter; - struct bkey_on_stack sk; - struct bkey_s_c k; - int ret; - - flags &= ~BCH_READ_LAST_FRAGMENT; - flags |= BCH_READ_MUST_CLONE; - - bkey_on_stack_init(&sk); - bch2_trans_init(&trans, c, 0, 0); -retry: - bch2_trans_begin(&trans); - - for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, - POS(inode, bvec_iter.bi_sector), - BTREE_ITER_SLOTS, k, ret) { - unsigned bytes, sectors, offset_into_extent; - - bkey_on_stack_reassemble(&sk, c, k); - k = bkey_i_to_s_c(sk.k); - - offset_into_extent = iter->pos.offset - - bkey_start_offset(k.k); - sectors = k.k->size - offset_into_extent; - - ret = bch2_read_indirect_extent(&trans, - &offset_into_extent, &sk); - if (ret) - break; - - sectors = min(sectors, k.k->size - offset_into_extent); - - bch2_trans_unlock(&trans); - - bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9; - swap(bvec_iter.bi_size, bytes); - - ret = __bch2_read_extent(c, rbio, bvec_iter, k, - offset_into_extent, failed, flags); - switch (ret) { - case READ_RETRY: - goto retry; - case READ_ERR: - goto err; - }; - - if (bytes == bvec_iter.bi_size) - goto out; - - swap(bvec_iter.bi_size, bytes); - bio_advance_iter(&rbio->bio, &bvec_iter, bytes); - } - - if (ret == -EINTR) - goto retry; - /* - * If we get here, it better have been because there was an error - * reading a btree node - */ - BUG_ON(!ret); - __bcache_io_error(c, "btree IO error: %i", ret); -err: - rbio->bio.bi_status = BLK_STS_IOERR; -out: - bch2_trans_exit(&trans); - bkey_on_stack_exit(&sk, c); - bch2_rbio_done(rbio); -} - static void bch2_rbio_retry(struct work_struct *work) { struct bch_read_bio *rbio = @@ -1731,7 +1692,7 @@ static void bch2_rbio_retry(struct work_struct *work) struct bch_fs *c = rbio->c; struct bvec_iter iter = rbio->bvec_iter; unsigned flags = rbio->flags; - u64 inode = rbio->pos.inode; + u64 inode = rbio->read_pos.inode; struct bch_io_failures failed = { .nr = 0 }; trace_read_retry(&rbio->bio); @@ -1746,10 +1707,14 @@ static void bch2_rbio_retry(struct work_struct *work) flags |= BCH_READ_IN_RETRY; flags &= ~BCH_READ_MAY_PROMOTE; - if (flags & BCH_READ_NODECODE) + if (flags & BCH_READ_NODECODE) { bch2_read_retry_nodecode(c, rbio, iter, inode, &failed, flags); - else - bch2_read_retry(c, rbio, iter, inode, &failed, flags); + } else { + flags &= ~BCH_READ_LAST_FRAGMENT; + flags |= BCH_READ_MUST_CLONE; + + __bch2_read(c, rbio, iter, inode, &failed, flags); + } } static void bch2_rbio_error(struct bch_read_bio *rbio, int retry, @@ -1775,7 +1740,7 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans, struct bch_read_bio *rbio) { struct bch_fs *c = rbio->c; - u64 data_offset = rbio->pos.offset - rbio->pick.crc.offset; + u64 data_offset = rbio->data_pos.offset - rbio->pick.crc.offset; struct bch_extent_crc_unpacked new_crc; struct btree_iter *iter = NULL; struct bkey_i *new; @@ -1785,26 +1750,12 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans, if (crc_is_compressed(rbio->pick.crc)) return 0; - iter = bch2_trans_get_iter(trans, BTREE_ID_EXTENTS, rbio->pos, + iter = bch2_trans_get_iter(trans, rbio->data_btree, rbio->data_pos, BTREE_ITER_SLOTS|BTREE_ITER_INTENT); - if ((ret = PTR_ERR_OR_ZERO(iter))) - goto out; - k = bch2_btree_iter_peek_slot(iter); if ((ret = bkey_err(k))) goto out; - /* - * going to be temporarily appending another checksum entry: - */ - new = bch2_trans_kmalloc(trans, bkey_bytes(k.k) + - BKEY_EXTENT_U64s_MAX * 8); - if ((ret = PTR_ERR_OR_ZERO(new))) - goto out; - - bkey_reassemble(new, k); - k = bkey_i_to_s_c(new); - if (bversion_cmp(k.k->version, rbio->version) || !bch2_bkey_matches_ptr(c, k, rbio->pick.ptr, data_offset)) goto out; @@ -1823,10 +1774,20 @@ static int __bch2_rbio_narrow_crcs(struct btree_trans *trans, goto out; } + /* + * going to be temporarily appending another checksum entry: + */ + new = bch2_trans_kmalloc(trans, bkey_bytes(k.k) + + sizeof(struct bch_extent_crc128)); + if ((ret = PTR_ERR_OR_ZERO(new))) + goto out; + + bkey_reassemble(new, k); + if (!bch2_bkey_narrow_crcs(new, new_crc)) goto out; - bch2_trans_update(trans, iter, new, 0); + ret = bch2_trans_update(trans, iter, new, 0); out: bch2_trans_iter_put(trans, iter); return ret; @@ -1850,8 +1811,11 @@ static void __bch2_read_endio(struct work_struct *work) 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; struct bch_csum csum; + nofs_flags = memalloc_nofs_save(); + /* Reset iterator for checksumming and copying bounced data: */ if (rbio->bounce) { src->bi_iter.bi_size = crc.compressed_size << 9; @@ -1865,6 +1829,13 @@ static void __bch2_read_endio(struct work_struct *work) if (bch2_crc_cmp(csum, rbio->pick.crc.csum)) goto csum_err; + /* + * XXX + * We need to rework the narrow_crcs path to deliver the read completion + * first, and then punt to a different workqueue, otherwise we're + * holding up reads while doing btree updates which is bad for memory + * reclaim. + */ if (unlikely(rbio->narrow_crcs)) bch2_rbio_narrow_crcs(rbio); @@ -1909,6 +1880,8 @@ nodecode: rbio = bch2_rbio_free(rbio); bch2_rbio_done(rbio); } +out: + memalloc_nofs_restore(nofs_flags); return; csum_err: /* @@ -1919,22 +1892,20 @@ csum_err: if (!rbio->bounce && (rbio->flags & BCH_READ_USER_MAPPED)) { rbio->flags |= BCH_READ_MUST_BOUNCE; bch2_rbio_error(rbio, READ_RETRY, BLK_STS_IOERR); - return; + goto out; } - bch2_dev_io_error(ca, - "data checksum error, inode %llu offset %llu: expected %0llx:%0llx got %0llx:%0llx (type %u)", - rbio->pos.inode, (u64) rbio->bvec_iter.bi_sector, + bch2_dev_inum_io_error(ca, rbio->read_pos.inode, (u64) rbio->bvec_iter.bi_sector, + "data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %u)", rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo, csum.hi, csum.lo, crc.csum_type); bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR); - return; + goto out; decompression_err: - __bcache_io_error(c, "decompression error, inode %llu offset %llu", - rbio->pos.inode, - (u64) rbio->bvec_iter.bi_sector); + bch_err_inum_ratelimited(c, rbio->read_pos.inode, + "decompression error"); bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR); - return; + goto out; } static void bch2_read_endio(struct bio *bio) @@ -1954,7 +1925,10 @@ static void bch2_read_endio(struct bio *bio) if (!rbio->split) rbio->bio.bi_end_io = rbio->end_io; - if (bch2_dev_io_err_on(bio->bi_status, ca, "data read; %s", + if (bch2_dev_inum_io_err_on(bio->bi_status, ca, + 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); return; @@ -1984,7 +1958,7 @@ static void bch2_read_endio(struct bio *bio) int __bch2_read_indirect_extent(struct btree_trans *trans, unsigned *offset_into_extent, - struct bkey_on_stack *orig_k) + struct bkey_buf *orig_k) { struct btree_iter *iter; struct bkey_s_c k; @@ -1994,52 +1968,54 @@ int __bch2_read_indirect_extent(struct btree_trans *trans, reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) + *offset_into_extent; - iter = bch2_trans_get_iter(trans, BTREE_ID_REFLINK, + iter = bch2_trans_get_iter(trans, BTREE_ID_reflink, POS(0, reflink_offset), BTREE_ITER_SLOTS); - ret = PTR_ERR_OR_ZERO(iter); - if (ret) - return ret; - k = bch2_btree_iter_peek_slot(iter); ret = bkey_err(k); if (ret) goto err; - if (k.k->type != KEY_TYPE_reflink_v) { - __bcache_io_error(trans->c, - "pointer to nonexistent indirect extent"); + if (k.k->type != KEY_TYPE_reflink_v && + k.k->type != KEY_TYPE_indirect_inline_data) { + bch_err_inum_ratelimited(trans->c, orig_k->k->k.p.inode, + "%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); - bkey_on_stack_reassemble(orig_k, trans->c, k); + bch2_bkey_buf_reassemble(orig_k, trans->c, k); err: bch2_trans_iter_put(trans, iter); return ret; } -int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig, - struct bvec_iter iter, struct bkey_s_c k, +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_fs *c = trans->c; struct extent_ptr_decoded pick; struct bch_read_bio *rbio = NULL; struct bch_dev *ca; struct promote_op *promote = NULL; bool bounce = false, read_full = false, narrow_crcs = false; - struct bpos pos = bkey_start_pos(k.k); + struct bpos data_pos = bkey_start_pos(k.k); int pick_ret; - if (k.k->type == KEY_TYPE_inline_data) { - struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); + if (bkey_extent_is_inline_data(k.k)) { unsigned bytes = min_t(unsigned, iter.bi_size, - bkey_val_bytes(d.k)); + bkey_inline_data_bytes(k.k)); swap(iter.bi_size, bytes); - memcpy_to_bio(&orig->bio, iter, d.v->data); + memcpy_to_bio(&orig->bio, iter, bkey_inline_data_p(k)); swap(iter.bi_size, bytes); bio_advance_iter(&orig->bio, &iter, bytes); zero_fill_bio_iter(&orig->bio, iter); @@ -2053,7 +2029,8 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig, goto hole; if (pick_ret < 0) { - __bcache_io_error(c, "no device to read from"); + bch_err_inum_ratelimited(c, k.k->p.inode, + "no device to read from"); goto err; } @@ -2106,7 +2083,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig, pick.crc.offset || offset_into_extent)); - pos.offset += offset_into_extent; + data_pos.offset += offset_into_extent; pick.ptr.offset += pick.crc.offset + offset_into_extent; offset_into_extent = 0; @@ -2178,7 +2155,9 @@ get_bio: /* XXX: only initialize this if needed */ rbio->devs_have = bch2_bkey_devs(k); rbio->pick = pick; - rbio->pos = pos; + rbio->read_pos = read_pos; + rbio->data_btree = data_btree; + rbio->data_pos = data_pos; rbio->version = k.k->version; rbio->promote = promote; INIT_WORK(&rbio->work, NULL); @@ -2192,9 +2171,13 @@ get_bio: bch2_increment_clock(c, bio_sectors(&rbio->bio), READ); - rcu_read_lock(); - bucket_io_clock_reset(c, ca, PTR_BUCKET_NR(ca, &pick.ptr), READ); - rcu_read_unlock(); + /* + * 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)) + 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))) { bio_inc_remaining(&orig->bio); @@ -2203,7 +2186,8 @@ get_bio: if (!rbio->pick.idx) { if (!rbio->have_ioref) { - __bcache_io_error(c, "no device to read from"); + bch_err_inum_ratelimited(c, k.k->p.inode, + "no device to read from"); bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR); goto out; } @@ -2243,6 +2227,9 @@ out: ret = READ_RETRY; } + if (!ret) + goto out_read_done; + return ret; } @@ -2269,54 +2256,50 @@ out_read_done: return 0; } -void bch2_read(struct bch_fs *c, struct bch_read_bio *rbio, u64 inode) +void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio, + struct bvec_iter bvec_iter, u64 inode, + struct bch_io_failures *failed, unsigned flags) { struct btree_trans trans; struct btree_iter *iter; - struct bkey_on_stack sk; + struct bkey_buf sk; struct bkey_s_c k; - unsigned flags = BCH_READ_RETRY_IF_STALE| - BCH_READ_MAY_PROMOTE| - BCH_READ_USER_MAPPED; int ret; - BUG_ON(rbio->_state); BUG_ON(flags & BCH_READ_NODECODE); - BUG_ON(flags & BCH_READ_IN_RETRY); - - rbio->c = c; - rbio->start_time = local_clock(); - bkey_on_stack_init(&sk); + bch2_bkey_buf_init(&sk); bch2_trans_init(&trans, c, 0, 0); retry: bch2_trans_begin(&trans); - iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, - POS(inode, rbio->bio.bi_iter.bi_sector), + iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, + POS(inode, bvec_iter.bi_sector), BTREE_ITER_SLOTS); while (1) { unsigned bytes, sectors, offset_into_extent; + enum btree_id data_btree = BTREE_ID_extents; bch2_btree_iter_set_pos(iter, - POS(inode, rbio->bio.bi_iter.bi_sector)); + POS(inode, bvec_iter.bi_sector)); k = bch2_btree_iter_peek_slot(iter); ret = bkey_err(k); if (ret) - goto err; + break; offset_into_extent = iter->pos.offset - bkey_start_offset(k.k); sectors = k.k->size - offset_into_extent; - bkey_on_stack_reassemble(&sk, c, k); - k = bkey_i_to_s_c(sk.k); + bch2_bkey_buf_reassemble(&sk, c, k); - ret = bch2_read_indirect_extent(&trans, + ret = bch2_read_indirect_extent(&trans, &data_btree, &offset_into_extent, &sk); if (ret) - goto err; + break; + + k = bkey_i_to_s_c(sk.k); /* * With indirect extents, the amount of data to read is the min @@ -2330,31 +2313,37 @@ retry: */ bch2_trans_unlock(&trans); - bytes = min(sectors, bio_sectors(&rbio->bio)) << 9; - swap(rbio->bio.bi_iter.bi_size, bytes); + bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9; + swap(bvec_iter.bi_size, bytes); - if (rbio->bio.bi_iter.bi_size == bytes) + if (bvec_iter.bi_size == bytes) flags |= BCH_READ_LAST_FRAGMENT; - bch2_read_extent(c, rbio, k, offset_into_extent, flags); + ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter->pos, + data_btree, k, + offset_into_extent, failed, flags); + if (ret) + break; if (flags & BCH_READ_LAST_FRAGMENT) break; - swap(rbio->bio.bi_iter.bi_size, bytes); - bio_advance(&rbio->bio, bytes); + swap(bvec_iter.bi_size, bytes); + bio_advance_iter(&rbio->bio, &bvec_iter, bytes); } -out: - bch2_trans_exit(&trans); - bkey_on_stack_exit(&sk, c); - return; -err: - if (ret == -EINTR) + bch2_trans_iter_put(&trans, iter); + + if (ret == -EINTR || ret == READ_RETRY || ret == READ_RETRY_AVOID) goto retry; - bcache_io_error(c, &rbio->bio, "btree IO error: %i", ret); - bch2_rbio_done(rbio); - goto out; + if (ret) { + bch_err_inum_ratelimited(c, inode, + "read error %i from btree lookup", ret); + rbio->bio.bi_status = BLK_STS_IOERR; + bch2_rbio_done(rbio); + } + bch2_trans_exit(&trans); + bch2_bkey_buf_exit(&sk, c); } void bch2_fs_io_exit(struct bch_fs *c)