X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fbtree_gc.c;h=d1615607480d33d49ad74e46e126f74fe203119f;hb=89b361f24a433a4a55e0032eca4e43045ea3f0d0;hp=ba81043fff51ca3013b068211efc85f27c150306;hpb=74148a8ee52526c44752f1773365963f18734ac9;p=bcachefs-tools-debian diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c index ba81043..d161560 100644 --- a/libbcachefs/btree_gc.c +++ b/libbcachefs/btree_gc.c @@ -9,6 +9,7 @@ #include "alloc_foreground.h" #include "bkey_methods.h" #include "bkey_buf.h" +#include "btree_journal_iter.h" #include "btree_key_cache.h" #include "btree_locking.h" #include "btree_update_interior.h" @@ -27,6 +28,7 @@ #include "reflink.h" #include "replicas.h" #include "super-io.h" +#include "trace.h" #include #include @@ -35,11 +37,16 @@ #include #include #include -#include #define DROP_THIS_NODE 10 #define DROP_PREV_NODE 11 +static bool should_restart_for_topology_repair(struct bch_fs *c) +{ + return c->opts.fix_errors != FSCK_FIX_no && + !(c->recovery_passes_complete & BIT_ULL(BCH_RECOVERY_PASS_check_topology)); +} + static inline void __gc_pos_set(struct bch_fs *c, struct gc_pos new_pos) { preempt_disable(); @@ -76,11 +83,11 @@ static int bch2_gc_check_topology(struct bch_fs *c, if (cur.k->k.type == KEY_TYPE_btree_ptr_v2) { struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(cur.k); - if (bpos_cmp(expected_start, bp->v.min_key)) { + if (!bpos_eq(expected_start, bp->v.min_key)) { bch2_topology_error(c); if (bkey_deleted(&prev->k->k)) { - pr_buf(&buf1, "start of node: "); + prt_printf(&buf1, "start of node: "); bch2_bpos_to_text(&buf1, node_start); } else { bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(prev->k)); @@ -88,25 +95,25 @@ static int bch2_gc_check_topology(struct bch_fs *c, bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(cur.k)); if (__fsck_err(c, - FSCK_CAN_FIX| - FSCK_CAN_IGNORE| - FSCK_NO_RATELIMIT, - "btree node with incorrect min_key at btree %s level %u:\n" - " prev %s\n" - " cur %s", - bch2_btree_ids[b->c.btree_id], b->c.level, - buf1.buf, buf2.buf) && - !test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags)) { + FSCK_CAN_FIX| + FSCK_CAN_IGNORE| + FSCK_NO_RATELIMIT, + btree_node_topology_bad_min_key, + "btree node with incorrect min_key at btree %s level %u:\n" + " prev %s\n" + " cur %s", + bch2_btree_id_str(b->c.btree_id), b->c.level, + buf1.buf, buf2.buf) && should_restart_for_topology_repair(c)) { bch_info(c, "Halting mark and sweep to start topology repair pass"); - ret = FSCK_ERR_START_TOPOLOGY_REPAIR; + ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology); goto err; } else { - set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags); + set_bit(BCH_FS_initial_gc_unfixed, &c->flags); } } } - if (is_last && bpos_cmp(cur.k->k.p, node_end)) { + if (is_last && !bpos_eq(cur.k->k.p, node_end)) { bch2_topology_error(c); printbuf_reset(&buf1); @@ -115,21 +122,19 @@ static int bch2_gc_check_topology(struct bch_fs *c, bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(cur.k)); bch2_bpos_to_text(&buf2, node_end); - if (__fsck_err(c, - FSCK_CAN_FIX| - FSCK_CAN_IGNORE| - FSCK_NO_RATELIMIT, + if (__fsck_err(c, FSCK_CAN_FIX|FSCK_CAN_IGNORE|FSCK_NO_RATELIMIT, + btree_node_topology_bad_max_key, "btree node with incorrect max_key at btree %s level %u:\n" " %s\n" " expected %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf) && - !test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags)) { + should_restart_for_topology_repair(c)) { bch_info(c, "Halting mark and sweep to start topology repair pass"); - ret = FSCK_ERR_START_TOPOLOGY_REPAIR; + ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology); goto err; } else { - set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags); + set_bit(BCH_FS_initial_gc_unfixed, &c->flags); } } @@ -165,10 +170,11 @@ static void btree_ptr_to_v2(struct btree *b, struct bkey_i_btree_ptr_v2 *dst) } } -static void bch2_btree_node_update_key_early(struct bch_fs *c, +static void bch2_btree_node_update_key_early(struct btree_trans *trans, enum btree_id btree, unsigned level, struct bkey_s_c old, struct bkey_i *new) { + struct bch_fs *c = trans->c; struct btree *b; struct bkey_buf tmp; int ret; @@ -176,7 +182,7 @@ static void bch2_btree_node_update_key_early(struct bch_fs *c, bch2_bkey_buf_init(&tmp); bch2_bkey_buf_reassemble(&tmp, c, old); - b = bch2_btree_node_get_noiter(c, tmp.k, btree, level, true); + b = bch2_btree_node_get_noiter(trans, tmp.k, btree, level, true); if (!IS_ERR_OR_NULL(b)) { mutex_lock(&c->btree_cache.lock); @@ -198,9 +204,9 @@ static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min) struct bkey_i_btree_ptr_v2 *new; int ret; - new = kmalloc(BKEY_BTREE_PTR_U64s_MAX * sizeof(u64), GFP_KERNEL); + new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL); if (!new) - return -ENOMEM; + return -BCH_ERR_ENOMEM_gc_repair_key; btree_ptr_to_v2(b, new); b->data->min_key = new_min; @@ -214,7 +220,7 @@ static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min) } bch2_btree_node_drop_keys_outside_node(b); - + bkey_copy(&b->key, &new->k_i); return 0; } @@ -227,9 +233,9 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max) if (ret) return ret; - new = kmalloc(BKEY_BTREE_PTR_U64s_MAX * sizeof(u64), GFP_KERNEL); + new = kmalloc_array(BKEY_BTREE_PTR_U64s_MAX, sizeof(u64), GFP_KERNEL); if (!new) - return -ENOMEM; + return -BCH_ERR_ENOMEM_gc_repair_key; btree_ptr_to_v2(b, new); b->data->max_key = new_max; @@ -264,7 +270,7 @@ static int btree_repair_node_boundaries(struct bch_fs *c, struct btree *b, int ret = 0; if (!prev) { - pr_buf(&buf1, "start of node: "); + prt_printf(&buf1, "start of node: "); bch2_bpos_to_text(&buf1, b->data->min_key); } else { bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(&prev->key)); @@ -273,51 +279,55 @@ static int btree_repair_node_boundaries(struct bch_fs *c, struct btree *b, bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(&cur->key)); if (prev && - bpos_cmp(expected_start, cur->data->min_key) > 0 && + bpos_gt(expected_start, cur->data->min_key) && BTREE_NODE_SEQ(cur->data) > BTREE_NODE_SEQ(prev->data)) { /* cur overwrites prev: */ - if (mustfix_fsck_err_on(bpos_cmp(prev->data->min_key, - cur->data->min_key) >= 0, c, + if (mustfix_fsck_err_on(bpos_ge(prev->data->min_key, + cur->data->min_key), c, + btree_node_topology_overwritten_by_next_node, "btree node overwritten by next node at btree %s level %u:\n" " node %s\n" " next %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf)) { ret = DROP_PREV_NODE; goto out; } - if (mustfix_fsck_err_on(bpos_cmp(prev->key.k.p, + if (mustfix_fsck_err_on(!bpos_eq(prev->key.k.p, bpos_predecessor(cur->data->min_key)), c, + btree_node_topology_bad_max_key, "btree node with incorrect max_key at btree %s level %u:\n" " node %s\n" " next %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf)) ret = set_node_max(c, prev, bpos_predecessor(cur->data->min_key)); } else { /* prev overwrites cur: */ - if (mustfix_fsck_err_on(bpos_cmp(expected_start, - cur->data->max_key) >= 0, c, + if (mustfix_fsck_err_on(bpos_ge(expected_start, + cur->data->max_key), c, + btree_node_topology_overwritten_by_prev_node, "btree node overwritten by prev node at btree %s level %u:\n" " prev %s\n" " node %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf)) { ret = DROP_THIS_NODE; goto out; } - if (mustfix_fsck_err_on(bpos_cmp(expected_start, cur->data->min_key), c, + if (mustfix_fsck_err_on(!bpos_eq(expected_start, cur->data->min_key), c, + btree_node_topology_bad_min_key, "btree node with incorrect min_key at btree %s level %u:\n" " prev %s\n" " node %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf)) - ret = set_node_min(c, cur, expected_start); + ret = set_node_min(c, cur, expected_start); } out: fsck_err: @@ -335,11 +345,12 @@ static int btree_repair_node_end(struct bch_fs *c, struct btree *b, bch2_bkey_val_to_text(&buf1, c, bkey_i_to_s_c(&child->key)); bch2_bpos_to_text(&buf2, b->key.k.p); - if (mustfix_fsck_err_on(bpos_cmp(child->key.k.p, b->key.k.p), c, + if (mustfix_fsck_err_on(!bpos_eq(child->key.k.p, b->key.k.p), c, + btree_node_topology_bad_max_key, "btree node with incorrect max_key at btree %s level %u:\n" " %s\n" " expected %s", - bch2_btree_ids[b->c.btree_id], b->c.level, + bch2_btree_id_str(b->c.btree_id), b->c.level, buf1.buf, buf2.buf)) { ret = set_node_max(c, child, b->key.k.p); if (ret) @@ -352,14 +363,15 @@ fsck_err: return ret; } -static int bch2_btree_repair_topology_recurse(struct bch_fs *c, struct btree *b) +static int bch2_btree_repair_topology_recurse(struct btree_trans *trans, struct btree *b) { + struct bch_fs *c = trans->c; struct btree_and_journal_iter iter; struct bkey_s_c k; struct bkey_buf prev_k, cur_k; struct btree *prev = NULL, *cur = NULL; bool have_child, dropped_children = false; - struct printbuf buf; + struct printbuf buf = PRINTBUF; int ret = 0; if (!b->c.level) @@ -372,13 +384,13 @@ again: bch2_btree_and_journal_iter_init_node_iter(&iter, c, b); while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { - BUG_ON(bpos_cmp(k.k->p, b->data->min_key) < 0); - BUG_ON(bpos_cmp(k.k->p, b->data->max_key) > 0); + BUG_ON(bpos_lt(k.k->p, b->data->min_key)); + BUG_ON(bpos_gt(k.k->p, b->data->max_key)); bch2_btree_and_journal_iter_advance(&iter); bch2_bkey_buf_reassemble(&cur_k, c, k); - cur = bch2_btree_node_get_noiter(c, cur_k.k, + cur = bch2_btree_node_get_noiter(trans, cur_k.k, b->c.btree_id, b->c.level - 1, false); ret = PTR_ERR_OR_ZERO(cur); @@ -387,22 +399,23 @@ again: bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(cur_k.k)); if (mustfix_fsck_err_on(ret == -EIO, c, - "Unreadable btree node at btree %s level %u:\n" + btree_node_unreadable, + "Topology repair: unreadable btree node at btree %s level %u:\n" " %s", - bch2_btree_ids[b->c.btree_id], + bch2_btree_id_str(b->c.btree_id), b->c.level - 1, buf.buf)) { - bch2_btree_node_evict(c, cur_k.k); + bch2_btree_node_evict(trans, cur_k.k); ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level, cur_k.k->k.p); + cur = NULL; if (ret) break; continue; } if (ret) { - bch_err(c, "%s: error %i getting btree node", - __func__, ret); + bch_err_msg(c, ret, "getting btree node"); break; } @@ -410,9 +423,10 @@ again: if (ret == DROP_THIS_NODE) { six_unlock_read(&cur->c.lock); - bch2_btree_node_evict(c, cur_k.k); + bch2_btree_node_evict(trans, cur_k.k); ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level, cur_k.k->k.p); + cur = NULL; if (ret) break; continue; @@ -423,7 +437,7 @@ again: prev = NULL; if (ret == DROP_PREV_NODE) { - bch2_btree_node_evict(c, prev_k.k); + bch2_btree_node_evict(trans, prev_k.k); ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level, prev_k.k->k.p); if (ret) @@ -463,23 +477,22 @@ again: bch2_bkey_buf_reassemble(&cur_k, c, k); bch2_btree_and_journal_iter_advance(&iter); - cur = bch2_btree_node_get_noiter(c, cur_k.k, + cur = bch2_btree_node_get_noiter(trans, cur_k.k, b->c.btree_id, b->c.level - 1, false); ret = PTR_ERR_OR_ZERO(cur); if (ret) { - bch_err(c, "%s: error %i getting btree node", - __func__, ret); + bch_err_msg(c, ret, "getting btree node"); goto err; } - ret = bch2_btree_repair_topology_recurse(c, cur); + ret = bch2_btree_repair_topology_recurse(trans, cur); six_unlock_read(&cur->c.lock); cur = NULL; if (ret == DROP_THIS_NODE) { - bch2_btree_node_evict(c, cur_k.k); + bch2_btree_node_evict(trans, cur_k.k); ret = bch2_journal_key_delete(c, b->c.btree_id, b->c.level, cur_k.k->k.p); dropped_children = true; @@ -495,9 +508,10 @@ again: bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key)); if (mustfix_fsck_err_on(!have_child, c, + btree_node_topology_interior_node_empty, "empty interior btree node at btree %s level %u\n" " %s", - bch2_btree_ids[b->c.btree_id], + bch2_btree_id_str(b->c.btree_id), b->c.level, buf.buf)) ret = DROP_THIS_NODE; err: @@ -518,36 +532,45 @@ fsck_err: return ret; } -static int bch2_repair_topology(struct bch_fs *c) +int bch2_check_topology(struct bch_fs *c) { + struct btree_trans *trans = bch2_trans_get(c); struct btree *b; unsigned i; int ret = 0; - for (i = 0; i < BTREE_ID_NR && !ret; i++) { - b = c->btree_roots[i].b; + for (i = 0; i < btree_id_nr_alive(c) && !ret; i++) { + struct btree_root *r = bch2_btree_id_root(c, i); + + if (!r->alive) + continue; + + b = r->b; if (btree_node_fake(b)) continue; - six_lock_read(&b->c.lock, NULL, NULL); - ret = bch2_btree_repair_topology_recurse(c, b); + btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read); + ret = bch2_btree_repair_topology_recurse(trans, b); six_unlock_read(&b->c.lock); if (ret == DROP_THIS_NODE) { bch_err(c, "empty btree root - repair unimplemented"); - ret = FSCK_ERR_EXIT; + ret = -BCH_ERR_fsck_repair_unimplemented; } } + bch2_trans_put(trans); + return ret; } -static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, +static int bch2_check_fix_ptrs(struct btree_trans *trans, enum btree_id btree_id, unsigned level, bool is_root, struct bkey_s_c *k) { - struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(*k); - const union bch_extent_entry *entry; + struct bch_fs *c = trans->c; + struct bkey_ptrs_c ptrs_c = bch2_bkey_ptrs_c(*k); + const union bch_extent_entry *entry_c; struct extent_ptr_decoded p = { 0 }; bool do_update = false; struct printbuf buf = PRINTBUF; @@ -557,19 +580,21 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, * XXX * use check_bucket_ref here */ - bkey_for_each_ptr_decode(k->k, ptrs, p, entry) { + bkey_for_each_ptr_decode(k->k, ptrs_c, p, entry_c) { struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr); - enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, &entry->ptr); - - if (fsck_err_on(!g->gen_valid, c, - "bucket %u:%zu data type %s ptr gen %u missing in alloc btree\n" - "while marking %s", - p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), - bch2_data_types[ptr_data_type(k->k, &p.ptr)], - p.ptr.gen, - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) { + enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, &entry_c->ptr); + + if (!g->gen_valid && + (c->opts.reconstruct_alloc || + fsck_err(c, ptr_to_missing_alloc_key, + "bucket %u:%zu data type %s ptr gen %u missing in alloc btree\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), + bch2_data_types[ptr_data_type(k->k, &p.ptr)], + p.ptr.gen, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, *k), buf.buf)))) { if (!p.ptr.cached) { g->gen_valid = true; g->gen = p.ptr.gen; @@ -578,52 +603,58 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, } } - if (fsck_err_on(gen_cmp(p.ptr.gen, g->gen) > 0, c, - "bucket %u:%zu data type %s ptr gen in the future: %u > %u\n" - "while marking %s", - p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), - bch2_data_types[ptr_data_type(k->k, &p.ptr)], - p.ptr.gen, g->gen, - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) { + if (gen_cmp(p.ptr.gen, g->gen) > 0 && + (c->opts.reconstruct_alloc || + fsck_err(c, ptr_gen_newer_than_bucket_gen, + "bucket %u:%zu data type %s ptr gen in the future: %u > %u\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), + bch2_data_types[ptr_data_type(k->k, &p.ptr)], + p.ptr.gen, g->gen, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, *k), buf.buf)))) { if (!p.ptr.cached) { g->gen_valid = true; g->gen = p.ptr.gen; g->data_type = 0; g->dirty_sectors = 0; g->cached_sectors = 0; - set_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags); + set_bit(BCH_FS_need_another_gc, &c->flags); } else { do_update = true; } } - if (fsck_err_on(gen_cmp(g->gen, p.ptr.gen) > BUCKET_GC_GEN_MAX, c, - "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n" - "while marking %s", - p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), g->gen, - bch2_data_types[ptr_data_type(k->k, &p.ptr)], - p.ptr.gen, - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) + if (gen_cmp(g->gen, p.ptr.gen) > BUCKET_GC_GEN_MAX && + (c->opts.reconstruct_alloc || + fsck_err(c, ptr_gen_newer_than_bucket_gen, + "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), g->gen, + bch2_data_types[ptr_data_type(k->k, &p.ptr)], + p.ptr.gen, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, *k), buf.buf)))) do_update = true; - if (fsck_err_on(!p.ptr.cached && - gen_cmp(p.ptr.gen, g->gen) < 0, c, - "bucket %u:%zu data type %s stale dirty ptr: %u < %u\n" - "while marking %s", - p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), - bch2_data_types[ptr_data_type(k->k, &p.ptr)], - p.ptr.gen, g->gen, - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) + if (!p.ptr.cached && gen_cmp(p.ptr.gen, g->gen) < 0 && + (c->opts.reconstruct_alloc || + fsck_err(c, stale_dirty_ptr, + "bucket %u:%zu data type %s stale dirty ptr: %u < %u\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), + bch2_data_types[ptr_data_type(k->k, &p.ptr)], + p.ptr.gen, g->gen, + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, *k), buf.buf)))) do_update = true; if (data_type != BCH_DATA_btree && p.ptr.gen != g->gen) continue; - if (fsck_err_on(g->data_type && - g->data_type != data_type, c, + if (fsck_err_on(bucket_data_type(g->data_type) && + bucket_data_type(g->data_type) != data_type, c, + ptr_bucket_data_type_mismatch, "bucket %u:%zu different types of data in same bucket: %s, %s\n" "while marking %s", p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), @@ -633,7 +664,7 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) { if (data_type == BCH_DATA_btree) { g->data_type = data_type; - set_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags); + set_bit(BCH_FS_need_another_gc, &c->flags); } else { do_update = true; } @@ -643,6 +674,7 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, struct gc_stripe *m = genradix_ptr(&c->gc_stripes, p.ec.idx); if (fsck_err_on(!m || !m->alive, c, + ptr_to_missing_stripe, "pointer to nonexistent stripe %llu\n" "while marking %s", (u64) p.ec.idx, @@ -650,7 +682,8 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, bch2_bkey_val_to_text(&buf, c, *k), buf.buf))) do_update = true; - if (fsck_err_on(!bch2_ptr_matches_stripe_m(m, p), c, + if (fsck_err_on(m && m->alive && !bch2_ptr_matches_stripe_m(m, p), c, + ptr_to_incorrect_stripe, "pointer does not match stripe %llu\n" "while marking %s", (u64) p.ec.idx, @@ -674,8 +707,8 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, new = kmalloc(bkey_bytes(k->k), GFP_KERNEL); if (!new) { - bch_err(c, "%s: error allocating new key", __func__); - ret = -ENOMEM; + bch_err_msg(c, ret, "allocating new key"); + ret = -BCH_ERR_ENOMEM_gc_repair_key; goto err; } @@ -744,15 +777,17 @@ found: } if (level) - bch2_btree_node_update_key_early(c, btree_id, level - 1, *k, new); + bch2_btree_node_update_key_early(trans, btree_id, level - 1, *k, new); - printbuf_reset(&buf); - bch2_bkey_val_to_text(&buf, c, *k); - bch_info(c, "updated %s", buf.buf); + if (0) { + printbuf_reset(&buf); + bch2_bkey_val_to_text(&buf, c, *k); + bch_info(c, "updated %s", buf.buf); - printbuf_reset(&buf); - bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(new)); - bch_info(c, "new key %s", buf.buf); + printbuf_reset(&buf); + bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(new)); + bch_info(c, "new key %s", buf.buf); + } *k = bkey_i_to_s_c(new); } @@ -783,23 +818,24 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id, BUG_ON(bch2_journal_seq_verify && k->k->version.lo > atomic64_read(&c->journal.seq)); - ret = bch2_check_fix_ptrs(c, btree_id, level, is_root, k); + ret = bch2_check_fix_ptrs(trans, btree_id, level, is_root, k); if (ret) goto err; if (fsck_err_on(k->k->version.lo > atomic64_read(&c->key_version), c, + bkey_version_in_future, "key version number higher than recorded: %llu > %llu", k->k->version.lo, atomic64_read(&c->key_version))) atomic64_set(&c->key_version, k->k->version.lo); } - ret = __bch2_trans_do(trans, NULL, NULL, 0, - bch2_mark_key(trans, old, *k, flags)); + ret = commit_do(trans, NULL, NULL, 0, + bch2_mark_key(trans, btree_id, level, old, *k, flags)); fsck_err: err: if (ret) - bch_err(c, "%s: ret %i", __func__, ret); + bch_err_fn(c, ret); return ret; } @@ -849,10 +885,7 @@ static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree_id, struct bch_fs *c = trans->c; struct btree_iter iter; struct btree *b; - unsigned depth = metadata_only ? 1 - : bch2_expensive_debug_checks ? 0 - : !btree_node_type_needs_gc(btree_id) ? 1 - : 0; + unsigned depth = metadata_only ? 1 : 0; int ret = 0; gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0)); @@ -873,11 +906,11 @@ static int bch2_gc_btree(struct btree_trans *trans, enum btree_id btree_id, return ret; mutex_lock(&c->btree_root_lock); - b = c->btree_roots[btree_id].b; + b = bch2_btree_id_root(c, btree_id)->b; if (!btree_node_fake(b)) { struct bkey_s_c k = bkey_i_to_s_c(&b->key); - ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level, + ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level + 1, true, &k, initial); } gc_pos_set(c, gc_pos_btree_root(b->c.btree_id)); @@ -902,15 +935,13 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b bkey_init(&prev.k->k); while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { - BUG_ON(bpos_cmp(k.k->p, b->data->min_key) < 0); - BUG_ON(bpos_cmp(k.k->p, b->data->max_key) > 0); + BUG_ON(bpos_lt(k.k->p, b->data->min_key)); + BUG_ON(bpos_gt(k.k->p, b->data->max_key)); ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level, false, &k, true); - if (ret) { - bch_err(c, "%s: error %i from bch2_gc_mark_key", __func__, ret); + if (ret) goto fsck_err; - } if (b->c.level) { bch2_bkey_buf_reassemble(&cur, c, k); @@ -938,7 +969,7 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b bch2_bkey_buf_reassemble(&cur, c, k); bch2_btree_and_journal_iter_advance(&iter); - child = bch2_btree_node_get_noiter(c, cur.k, + child = bch2_btree_node_get_noiter(trans, cur.k, b->c.btree_id, b->c.level - 1, false); ret = PTR_ERR_OR_ZERO(child); @@ -950,26 +981,26 @@ static int bch2_gc_btree_init_recurse(struct btree_trans *trans, struct btree *b FSCK_CAN_FIX| FSCK_CAN_IGNORE| FSCK_NO_RATELIMIT, + btree_node_read_error, "Unreadable btree node at btree %s level %u:\n" " %s", - bch2_btree_ids[b->c.btree_id], + bch2_btree_id_str(b->c.btree_id), b->c.level - 1, (printbuf_reset(&buf), bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(cur.k)), buf.buf)) && - !test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags)) { - ret = FSCK_ERR_START_TOPOLOGY_REPAIR; + should_restart_for_topology_repair(c)) { bch_info(c, "Halting mark and sweep to start topology repair pass"); + ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology); goto fsck_err; } else { /* Continue marking when opted to not * fix the error: */ ret = 0; - set_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags); + set_bit(BCH_FS_initial_gc_unfixed, &c->flags); continue; } } else if (ret) { - bch_err(c, "%s: error %i getting btree node", - __func__, ret); + bch_err_msg(c, ret, "getting btree node"); break; } @@ -995,14 +1026,11 @@ static int bch2_gc_btree_init(struct btree_trans *trans, { struct bch_fs *c = trans->c; struct btree *b; - unsigned target_depth = metadata_only ? 1 - : bch2_expensive_debug_checks ? 0 - : !btree_node_type_needs_gc(btree_id) ? 1 - : 0; + unsigned target_depth = metadata_only ? 1 : 0; struct printbuf buf = PRINTBUF; int ret = 0; - b = c->btree_roots[btree_id].b; + b = bch2_btree_id_root(c, btree_id)->b; if (btree_node_fake(b)) return 0; @@ -1010,19 +1038,21 @@ static int bch2_gc_btree_init(struct btree_trans *trans, six_lock_read(&b->c.lock, NULL, NULL); printbuf_reset(&buf); bch2_bpos_to_text(&buf, b->data->min_key); - if (mustfix_fsck_err_on(bpos_cmp(b->data->min_key, POS_MIN), c, + if (mustfix_fsck_err_on(!bpos_eq(b->data->min_key, POS_MIN), c, + btree_root_bad_min_key, "btree root with incorrect min_key: %s", buf.buf)) { bch_err(c, "repair unimplemented"); - ret = FSCK_ERR_EXIT; + ret = -BCH_ERR_fsck_repair_unimplemented; goto fsck_err; } printbuf_reset(&buf); bch2_bpos_to_text(&buf, b->data->max_key); - if (mustfix_fsck_err_on(bpos_cmp(b->data->max_key, SPOS_MAX), c, + if (mustfix_fsck_err_on(!bpos_eq(b->data->max_key, SPOS_MAX), c, + btree_root_bad_max_key, "btree root with incorrect max_key: %s", buf.buf)) { bch_err(c, "repair unimplemented"); - ret = FSCK_ERR_EXIT; + ret = -BCH_ERR_fsck_repair_unimplemented; goto fsck_err; } @@ -1032,14 +1062,14 @@ static int bch2_gc_btree_init(struct btree_trans *trans, if (!ret) { struct bkey_s_c k = bkey_i_to_s_c(&b->key); - ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level, true, + ret = bch2_gc_mark_key(trans, b->c.btree_id, b->c.level + 1, true, &k, true); } fsck_err: six_unlock_read(&b->c.lock); if (ret < 0) - bch_err(c, "%s: ret %i", __func__, ret); + bch_err_fn(c, ret); printbuf_exit(&buf); return ret; } @@ -1052,15 +1082,13 @@ static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r) static int bch2_gc_btrees(struct bch_fs *c, bool initial, bool metadata_only) { - struct btree_trans trans; + struct btree_trans *trans = bch2_trans_get(c); enum btree_id ids[BTREE_ID_NR]; unsigned i; int ret = 0; - bch2_trans_init(&trans, c, 0, 0); - if (initial) - trans.is_initial_gc = true; + trans->is_initial_gc = true; for (i = 0; i < BTREE_ID_NR; i++) ids[i] = i; @@ -1068,13 +1096,22 @@ static int bch2_gc_btrees(struct bch_fs *c, bool initial, bool metadata_only) for (i = 0; i < BTREE_ID_NR && !ret; i++) ret = initial - ? bch2_gc_btree_init(&trans, ids[i], metadata_only) - : bch2_gc_btree(&trans, ids[i], initial, metadata_only); + ? bch2_gc_btree_init(trans, ids[i], metadata_only) + : bch2_gc_btree(trans, ids[i], initial, metadata_only); + + for (i = BTREE_ID_NR; i < btree_id_nr_alive(c) && !ret; i++) { + if (!bch2_btree_id_root(c, i)->alive) + continue; + + ret = initial + ? bch2_gc_btree_init(trans, i, metadata_only) + : bch2_gc_btree(trans, i, initial, metadata_only); + } if (ret < 0) - bch_err(c, "%s: ret %i", __func__, ret); + bch_err_fn(c, ret); - bch2_trans_exit(&trans); + bch2_trans_put(trans); return ret; } @@ -1181,33 +1218,24 @@ static int bch2_gc_done(struct bch_fs *c, { struct bch_dev *ca = NULL; struct printbuf buf = PRINTBUF; - bool verify = !metadata_only && (!initial || - (c->sb.compat & (1ULL << BCH_COMPAT_alloc_info))); + bool verify = !metadata_only && + !c->opts.reconstruct_alloc && + (!initial || (c->sb.compat & (1ULL << BCH_COMPAT_alloc_info))); unsigned i, dev; int ret = 0; percpu_down_write(&c->mark_lock); -#define copy_field(_f, _msg, ...) \ - if (dst->_f != src->_f) { \ - if (verify) \ - fsck_err(c, _msg ": got %llu, should be %llu" \ - , ##__VA_ARGS__, dst->_f, src->_f); \ - dst->_f = src->_f; \ - } -#define copy_stripe_field(_f, _msg, ...) \ - if (dst->_f != src->_f) { \ - if (verify) \ - fsck_err(c, "stripe %zu has wrong "_msg \ - ": got %u, should be %u", \ - iter.pos, ##__VA_ARGS__, \ - dst->_f, src->_f); \ - dst->_f = src->_f; \ - } -#define copy_dev_field(_f, _msg, ...) \ - copy_field(_f, "dev %u has wrong " _msg, dev, ##__VA_ARGS__) -#define copy_fs_field(_f, _msg, ...) \ - copy_field(_f, "fs has wrong " _msg, ##__VA_ARGS__) +#define copy_field(_err, _f, _msg, ...) \ + if (dst->_f != src->_f && \ + (!verify || \ + fsck_err(c, _err, _msg ": got %llu, should be %llu" \ + , ##__VA_ARGS__, dst->_f, src->_f))) \ + dst->_f = src->_f +#define copy_dev_field(_err, _f, _msg, ...) \ + copy_field(_err, _f, "dev %u has wrong " _msg, dev, ##__VA_ARGS__) +#define copy_fs_field(_err, _f, _msg, ...) \ + copy_field(_err, _f, "fs has wrong " _msg, ##__VA_ARGS__) for (i = 0; i < ARRAY_SIZE(c->usage); i++) bch2_fs_usage_acc_to_base(c, i); @@ -1215,41 +1243,48 @@ static int bch2_gc_done(struct bch_fs *c, for_each_member_device(ca, c, dev) { struct bch_dev_usage *dst = ca->usage_base; struct bch_dev_usage *src = (void *) - bch2_acc_percpu_u64s((void *) ca->usage_gc, + bch2_acc_percpu_u64s((u64 __percpu *) ca->usage_gc, dev_usage_u64s()); - copy_dev_field(buckets_ec, "buckets_ec"); - copy_dev_field(buckets_unavailable, "buckets_unavailable"); - for (i = 0; i < BCH_DATA_NR; i++) { - copy_dev_field(d[i].buckets, "%s buckets", bch2_data_types[i]); - copy_dev_field(d[i].sectors, "%s sectors", bch2_data_types[i]); - copy_dev_field(d[i].fragmented, "%s fragmented", bch2_data_types[i]); + copy_dev_field(dev_usage_buckets_wrong, + d[i].buckets, "%s buckets", bch2_data_types[i]); + copy_dev_field(dev_usage_sectors_wrong, + d[i].sectors, "%s sectors", bch2_data_types[i]); + copy_dev_field(dev_usage_fragmented_wrong, + d[i].fragmented, "%s fragmented", bch2_data_types[i]); } - }; + } { unsigned nr = fs_usage_u64s(c); struct bch_fs_usage *dst = c->usage_base; struct bch_fs_usage *src = (void *) - bch2_acc_percpu_u64s((void *) c->usage_gc, nr); + bch2_acc_percpu_u64s((u64 __percpu *) c->usage_gc, nr); - copy_fs_field(hidden, "hidden"); - copy_fs_field(btree, "btree"); + copy_fs_field(fs_usage_hidden_wrong, + hidden, "hidden"); + copy_fs_field(fs_usage_btree_wrong, + btree, "btree"); if (!metadata_only) { - copy_fs_field(data, "data"); - copy_fs_field(cached, "cached"); - copy_fs_field(reserved, "reserved"); - copy_fs_field(nr_inodes,"nr_inodes"); + copy_fs_field(fs_usage_data_wrong, + data, "data"); + copy_fs_field(fs_usage_cached_wrong, + cached, "cached"); + copy_fs_field(fs_usage_reserved_wrong, + reserved, "reserved"); + copy_fs_field(fs_usage_nr_inodes_wrong, + nr_inodes,"nr_inodes"); for (i = 0; i < BCH_REPLICAS_MAX; i++) - copy_fs_field(persistent_reserved[i], + copy_fs_field(fs_usage_persistent_reserved_wrong, + persistent_reserved[i], "persistent_reserved[%i]", i); } for (i = 0; i < c->replicas.nr; i++) { - struct bch_replicas_entry *e = + struct bch_replicas_entry_v1 *e = cpu_replicas_entry(&c->replicas, i); if (metadata_only && @@ -1260,7 +1295,8 @@ static int bch2_gc_done(struct bch_fs *c, printbuf_reset(&buf); bch2_replicas_entry_to_text(&buf, e); - copy_fs_field(replicas[i], "%s", buf.buf); + copy_fs_field(fs_usage_replicas_wrong, + replicas[i], "%s", buf.buf); } } @@ -1272,15 +1308,14 @@ fsck_err: if (ca) percpu_ref_put(&ca->ref); if (ret) - bch_err(c, "%s: ret %i", __func__, ret); + bch_err_fn(c, ret); percpu_up_write(&c->mark_lock); printbuf_exit(&buf); return ret; } -static int bch2_gc_start(struct bch_fs *c, - bool metadata_only) +static int bch2_gc_start(struct bch_fs *c) { struct bch_dev *ca = NULL; unsigned i; @@ -1291,45 +1326,99 @@ static int bch2_gc_start(struct bch_fs *c, sizeof(u64), GFP_KERNEL); if (!c->usage_gc) { bch_err(c, "error allocating c->usage_gc"); - return -ENOMEM; + return -BCH_ERR_ENOMEM_gc_start; } for_each_member_device(ca, c, i) { - BUG_ON(ca->buckets_gc); BUG_ON(ca->usage_gc); ca->usage_gc = alloc_percpu(struct bch_dev_usage); if (!ca->usage_gc) { bch_err(c, "error allocating ca->usage_gc"); percpu_ref_put(&ca->ref); - return -ENOMEM; + return -BCH_ERR_ENOMEM_gc_start; } + + this_cpu_write(ca->usage_gc->d[BCH_DATA_free].buckets, + ca->mi.nbuckets - ca->mi.first_bucket); } return 0; } +static int bch2_gc_reset(struct bch_fs *c) +{ + struct bch_dev *ca; + unsigned i; + + for_each_member_device(ca, c, i) { + free_percpu(ca->usage_gc); + ca->usage_gc = NULL; + } + + free_percpu(c->usage_gc); + c->usage_gc = NULL; + + return bch2_gc_start(c); +} + +/* returns true if not equal */ +static inline bool bch2_alloc_v4_cmp(struct bch_alloc_v4 l, + struct bch_alloc_v4 r) +{ + return l.gen != r.gen || + l.oldest_gen != r.oldest_gen || + l.data_type != r.data_type || + l.dirty_sectors != r.dirty_sectors || + l.cached_sectors != r.cached_sectors || + l.stripe_redundancy != r.stripe_redundancy || + l.stripe != r.stripe; +} + static int bch2_alloc_write_key(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k, bool metadata_only) { struct bch_fs *c = trans->c; struct bch_dev *ca = bch_dev_bkey_exists(c, iter->pos.inode); - struct bucket gc; - struct bkey_s_c k; - struct bkey_alloc_unpacked old_u, new_u; - struct bkey_alloc_buf *a; + struct bucket gc, *b; + struct bkey_i_alloc_v4 *a; + struct bch_alloc_v4 old_convert, new; + const struct bch_alloc_v4 *old; + enum bch_data_type type; int ret; - k = bch2_btree_iter_peek_slot(iter); - ret = bkey_err(k); - if (ret) - return ret; + if (bkey_ge(iter->pos, POS(ca->dev_idx, ca->mi.nbuckets))) + return 1; - old_u = new_u = bch2_alloc_unpack(k); + old = bch2_alloc_to_v4(k, &old_convert); + new = *old; percpu_down_read(&c->mark_lock); - gc = *gc_bucket(ca, iter->pos.offset); + b = gc_bucket(ca, iter->pos.offset); + + /* + * b->data_type doesn't yet include need_discard & need_gc_gen states - + * fix that here: + */ + type = __alloc_data_type(b->dirty_sectors, + b->cached_sectors, + b->stripe, + *old, + b->data_type); + if (b->data_type != type) { + struct bch_dev_usage *u; + + preempt_disable(); + u = this_cpu_ptr(ca->usage_gc); + u->d[b->data_type].buckets--; + b->data_type = type; + u->d[b->data_type].buckets++; + preempt_enable(); + } + + gc = *b; percpu_up_read(&c->mark_lock); if (metadata_only && @@ -1338,87 +1427,101 @@ static int bch2_alloc_write_key(struct btree_trans *trans, gc.data_type != BCH_DATA_btree) return 0; - if (gen_after(old_u.gen, gc.gen)) + if (gen_after(old->gen, gc.gen)) return 0; -#define copy_bucket_field(_f) \ - if (fsck_err_on(new_u._f != gc._f, c, \ + if (c->opts.reconstruct_alloc || + fsck_err_on(new.data_type != gc.data_type, c, + alloc_key_data_type_wrong, + "bucket %llu:%llu gen %u has wrong data_type" + ": got %s, should be %s", + iter->pos.inode, iter->pos.offset, + gc.gen, + bch2_data_types[new.data_type], + bch2_data_types[gc.data_type])) + new.data_type = gc.data_type; + +#define copy_bucket_field(_errtype, _f) \ + if (c->opts.reconstruct_alloc || \ + fsck_err_on(new._f != gc._f, c, _errtype, \ "bucket %llu:%llu gen %u data type %s has wrong " #_f \ ": got %u, should be %u", \ iter->pos.inode, iter->pos.offset, \ gc.gen, \ bch2_data_types[gc.data_type], \ - new_u._f, gc._f)) \ - new_u._f = gc._f; \ - - copy_bucket_field(gen); - copy_bucket_field(data_type); - copy_bucket_field(stripe); - copy_bucket_field(dirty_sectors); - copy_bucket_field(cached_sectors); - copy_bucket_field(stripe_redundancy); - copy_bucket_field(stripe); + new._f, gc._f)) \ + new._f = gc._f; \ + + copy_bucket_field(alloc_key_gen_wrong, + gen); + copy_bucket_field(alloc_key_dirty_sectors_wrong, + dirty_sectors); + copy_bucket_field(alloc_key_cached_sectors_wrong, + cached_sectors); + copy_bucket_field(alloc_key_stripe_wrong, + stripe); + copy_bucket_field(alloc_key_stripe_redundancy_wrong, + stripe_redundancy); #undef copy_bucket_field - if (!bkey_alloc_unpacked_cmp(old_u, new_u)) + if (!bch2_alloc_v4_cmp(*old, new)) return 0; - a = bch2_alloc_pack(trans, new_u); - if (IS_ERR(a)) - return PTR_ERR(a); + a = bch2_alloc_to_v4_mut(trans, k); + ret = PTR_ERR_OR_ZERO(a); + if (ret) + return ret; + + a->v = new; - ret = bch2_trans_update(trans, iter, &a->k, BTREE_TRIGGER_NORUN); + /* + * The trigger normally makes sure this is set, but we're not running + * triggers: + */ + if (a->v.data_type == BCH_DATA_cached && !a->v.io_time[READ]) + a->v.io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now)); + + ret = bch2_trans_update(trans, iter, &a->k_i, BTREE_TRIGGER_NORUN); fsck_err: return ret; } static int bch2_gc_alloc_done(struct bch_fs *c, bool metadata_only) { - struct btree_trans trans; + struct btree_trans *trans = bch2_trans_get(c); struct btree_iter iter; struct bkey_s_c k; struct bch_dev *ca; unsigned i; int ret = 0; - bch2_trans_init(&trans, c, 0, 0); - for_each_member_device(ca, c, i) { - for_each_btree_key(&trans, iter, BTREE_ID_alloc, - POS(ca->dev_idx, ca->mi.first_bucket), - BTREE_ITER_SLOTS| - BTREE_ITER_PREFETCH, k, ret) { - if (bkey_cmp(iter.pos, POS(ca->dev_idx, ca->mi.nbuckets)) >= 0) - break; - - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_LAZY_RW, - bch2_alloc_write_key(&trans, &iter, - metadata_only)); - if (ret) - break; - } - bch2_trans_iter_exit(&trans, &iter); - - if (ret) { - bch_err(c, "error writing alloc info: %i", ret); + ret = for_each_btree_key_commit(trans, iter, BTREE_ID_alloc, + POS(ca->dev_idx, ca->mi.first_bucket), + BTREE_ITER_SLOTS|BTREE_ITER_PREFETCH, k, + NULL, NULL, BCH_TRANS_COMMIT_lazy_rw, + bch2_alloc_write_key(trans, &iter, k, metadata_only)); + + if (ret < 0) { + bch_err_fn(c, ret); percpu_ref_put(&ca->ref); break; } } - bch2_trans_exit(&trans); - return ret; + bch2_trans_put(trans); + return ret < 0 ? ret : 0; } static int bch2_gc_alloc_start(struct bch_fs *c, bool metadata_only) { struct bch_dev *ca; - struct btree_trans trans; + struct btree_trans *trans = bch2_trans_get(c); struct btree_iter iter; struct bkey_s_c k; struct bucket *g; - struct bkey_alloc_unpacked u; + struct bch_alloc_v4 a_convert; + const struct bch_alloc_v4 *a; unsigned i; int ret; @@ -1429,43 +1532,42 @@ static int bch2_gc_alloc_start(struct bch_fs *c, bool metadata_only) if (!buckets) { percpu_ref_put(&ca->ref); bch_err(c, "error allocating ca->buckets[gc]"); - return -ENOMEM; + ret = -BCH_ERR_ENOMEM_gc_alloc_start; + goto err; } buckets->first_bucket = ca->mi.first_bucket; buckets->nbuckets = ca->mi.nbuckets; rcu_assign_pointer(ca->buckets_gc, buckets); - }; - - bch2_trans_init(&trans, c, 0, 0); + } - for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { + ret = for_each_btree_key2(trans, iter, BTREE_ID_alloc, POS_MIN, + BTREE_ITER_PREFETCH, k, ({ ca = bch_dev_bkey_exists(c, k.k->p.inode); g = gc_bucket(ca, k.k->p.offset); - u = bch2_alloc_unpack(k); + + a = bch2_alloc_to_v4(k, &a_convert); g->gen_valid = 1; - g->gen = u.gen; + g->gen = a->gen; if (metadata_only && - (u.data_type == BCH_DATA_user || - u.data_type == BCH_DATA_cached || - u.data_type == BCH_DATA_parity)) { - g->data_type = u.data_type; - g->dirty_sectors = u.dirty_sectors; - g->cached_sectors = u.cached_sectors; - g->stripe = u.stripe; - g->stripe_redundancy = u.stripe_redundancy; + (a->data_type == BCH_DATA_user || + a->data_type == BCH_DATA_cached || + a->data_type == BCH_DATA_parity)) { + g->data_type = a->data_type; + g->dirty_sectors = a->dirty_sectors; + g->cached_sectors = a->cached_sectors; + g->stripe = a->stripe; + g->stripe_redundancy = a->stripe_redundancy; } - } - bch2_trans_iter_exit(&trans, &iter); - - bch2_trans_exit(&trans); + 0; + })); +err: + bch2_trans_put(trans); if (ret) - bch_err(c, "error reading alloc info at gc start: %i", ret); - + bch_err_fn(c, ret); return ret; } @@ -1484,85 +1586,89 @@ static void bch2_gc_alloc_reset(struct bch_fs *c, bool metadata_only) g->data_type == BCH_DATA_cached || g->data_type == BCH_DATA_parity)) continue; + g->data_type = 0; g->dirty_sectors = 0; g->cached_sectors = 0; } - }; + } } -static int bch2_gc_reflink_done(struct bch_fs *c, bool metadata_only) +static int bch2_gc_write_reflink_key(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k, + size_t *idx) { - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - struct reflink_gc *r; - size_t idx = 0; + struct bch_fs *c = trans->c; + const __le64 *refcount = bkey_refcount_c(k); struct printbuf buf = PRINTBUF; + struct reflink_gc *r; int ret = 0; - if (metadata_only) + if (!refcount) return 0; - bch2_trans_init(&trans, c, 0, 0); + while ((r = genradix_ptr(&c->reflink_gc_table, *idx)) && + r->offset < k.k->p.offset) + ++*idx; - for_each_btree_key(&trans, iter, BTREE_ID_reflink, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { - const __le64 *refcount = bkey_refcount_c(k); + if (!r || + r->offset != k.k->p.offset || + r->size != k.k->size) { + bch_err(c, "unexpected inconsistency walking reflink table at gc finish"); + return -EINVAL; + } - if (!refcount) - continue; + if (fsck_err_on(r->refcount != le64_to_cpu(*refcount), c, + reflink_v_refcount_wrong, + "reflink key has wrong refcount:\n" + " %s\n" + " should be %u", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf), + r->refcount)) { + struct bkey_i *new = bch2_bkey_make_mut(trans, iter, &k, 0); - r = genradix_ptr(&c->reflink_gc_table, idx++); - if (!r || - r->offset != k.k->p.offset || - r->size != k.k->size) { - bch_err(c, "unexpected inconsistency walking reflink table at gc finish"); - ret = -EINVAL; - break; - } + ret = PTR_ERR_OR_ZERO(new); + if (ret) + return ret; - if (fsck_err_on(r->refcount != le64_to_cpu(*refcount), c, - "reflink key has wrong refcount:\n" - " %s\n" - " should be %u", - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, k), buf.buf), - r->refcount)) { - struct bkey_i *new; + if (!r->refcount) + new->k.type = KEY_TYPE_deleted; + else + *bkey_refcount(new) = cpu_to_le64(r->refcount); + } +fsck_err: + printbuf_exit(&buf); + return ret; +} - new = kmalloc(bkey_bytes(k.k), GFP_KERNEL); - if (!new) { - ret = -ENOMEM; - break; - } +static int bch2_gc_reflink_done(struct bch_fs *c, bool metadata_only) +{ + struct btree_trans *trans; + struct btree_iter iter; + struct bkey_s_c k; + size_t idx = 0; + int ret = 0; - bkey_reassemble(new, k); + if (metadata_only) + return 0; - if (!r->refcount) - new->k.type = KEY_TYPE_deleted; - else - *bkey_refcount(new) = cpu_to_le64(r->refcount); + trans = bch2_trans_get(c); - ret = __bch2_trans_do(&trans, NULL, NULL, 0, - __bch2_btree_insert(&trans, BTREE_ID_reflink, new)); - kfree(new); + ret = for_each_btree_key_commit(trans, iter, + BTREE_ID_reflink, POS_MIN, + BTREE_ITER_PREFETCH, k, + NULL, NULL, BCH_TRANS_COMMIT_no_enospc, + bch2_gc_write_reflink_key(trans, &iter, k, &idx)); - if (ret) - break; - } - } -fsck_err: - bch2_trans_iter_exit(&trans, &iter); c->reflink_gc_nr = 0; - bch2_trans_exit(&trans); - printbuf_exit(&buf); + bch2_trans_put(trans); return ret; } static int bch2_gc_reflink_start(struct bch_fs *c, bool metadata_only) { - struct btree_trans trans; + struct btree_trans *trans; struct btree_iter iter; struct bkey_s_c k; struct reflink_gc *r; @@ -1571,10 +1677,10 @@ static int bch2_gc_reflink_start(struct bch_fs *c, if (metadata_only) return 0; - bch2_trans_init(&trans, c, 0, 0); + trans = bch2_trans_get(c); c->reflink_gc_nr = 0; - for_each_btree_key(&trans, iter, BTREE_ID_reflink, POS_MIN, + for_each_btree_key(trans, iter, BTREE_ID_reflink, POS_MIN, BTREE_ITER_PREFETCH, k, ret) { const __le64 *refcount = bkey_refcount_c(k); @@ -1584,7 +1690,7 @@ static int bch2_gc_reflink_start(struct bch_fs *c, r = genradix_ptr_alloc(&c->reflink_gc_table, c->reflink_gc_nr++, GFP_KERNEL); if (!r) { - ret = -ENOMEM; + ret = -BCH_ERR_ENOMEM_gc_reflink_start; break; } @@ -1592,9 +1698,9 @@ static int bch2_gc_reflink_start(struct bch_fs *c, r->size = k.k->size; r->refcount = 0; } - bch2_trans_iter_exit(&trans, &iter); + bch2_trans_iter_exit(trans, &iter); - bch2_trans_exit(&trans); + bch2_trans_put(trans); return ret; } @@ -1607,66 +1713,78 @@ static void bch2_gc_reflink_reset(struct bch_fs *c, bool metadata_only) r->refcount = 0; } -static int bch2_gc_stripes_done(struct bch_fs *c, bool metadata_only) +static int bch2_gc_write_stripes_key(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k) { - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - struct gc_stripe *m; - const struct bch_stripe *s; + struct bch_fs *c = trans->c; struct printbuf buf = PRINTBUF; + const struct bch_stripe *s; + struct gc_stripe *m; + bool bad = false; unsigned i; int ret = 0; - if (metadata_only) + if (k.k->type != KEY_TYPE_stripe) return 0; - bch2_trans_init(&trans, c, 0, 0); + s = bkey_s_c_to_stripe(k).v; + m = genradix_ptr(&c->gc_stripes, k.k->p.offset); - for_each_btree_key(&trans, iter, BTREE_ID_stripes, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { - if (k.k->type != KEY_TYPE_stripe) - continue; + for (i = 0; i < s->nr_blocks; i++) { + u32 old = stripe_blockcount_get(s, i); + u32 new = (m ? m->block_sectors[i] : 0); - s = bkey_s_c_to_stripe(k).v; - m = genradix_ptr(&c->gc_stripes, k.k->p.offset); - - for (i = 0; i < s->nr_blocks; i++) - if (stripe_blockcount_get(s, i) != (m ? m->block_sectors[i] : 0)) - goto inconsistent; - continue; -inconsistent: - if (fsck_err_on(true, c, - "stripe has wrong block sector count %u:\n" - " %s\n" - " should be %u", i, - (printbuf_reset(&buf), - bch2_bkey_val_to_text(&buf, c, k), buf.buf), - m ? m->block_sectors[i] : 0)) { - struct bkey_i_stripe *new; + if (old != new) { + prt_printf(&buf, "stripe block %u has wrong sector count: got %u, should be %u\n", + i, old, new); + bad = true; + } + } - new = kmalloc(bkey_bytes(k.k), GFP_KERNEL); - if (!new) { - ret = -ENOMEM; - break; - } + if (bad) + bch2_bkey_val_to_text(&buf, c, k); - bkey_reassemble(&new->k_i, k); + if (fsck_err_on(bad, c, stripe_sector_count_wrong, + "%s", buf.buf)) { + struct bkey_i_stripe *new; - for (i = 0; i < new->v.nr_blocks; i++) - stripe_blockcount_set(&new->v, i, m ? m->block_sectors[i] : 0); + new = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); + ret = PTR_ERR_OR_ZERO(new); + if (ret) + return ret; - ret = __bch2_trans_do(&trans, NULL, NULL, 0, - __bch2_btree_insert(&trans, BTREE_ID_reflink, &new->k_i)); - kfree(new); - } + bkey_reassemble(&new->k_i, k); + + for (i = 0; i < new->v.nr_blocks; i++) + stripe_blockcount_set(&new->v, i, m ? m->block_sectors[i] : 0); + + ret = bch2_trans_update(trans, iter, &new->k_i, 0); } fsck_err: - bch2_trans_iter_exit(&trans, &iter); + printbuf_exit(&buf); + return ret; +} + +static int bch2_gc_stripes_done(struct bch_fs *c, bool metadata_only) +{ + struct btree_trans *trans; + struct btree_iter iter; + struct bkey_s_c k; + int ret = 0; - bch2_trans_exit(&trans); + if (metadata_only) + return 0; - printbuf_exit(&buf); + trans = bch2_trans_get(c); + + ret = for_each_btree_key_commit(trans, iter, + BTREE_ID_stripes, POS_MIN, + BTREE_ITER_PREFETCH, k, + NULL, NULL, BCH_TRANS_COMMIT_no_enospc, + bch2_gc_write_stripes_key(trans, &iter, k)); + + bch2_trans_put(trans); return ret; } @@ -1678,6 +1796,12 @@ static void bch2_gc_stripes_reset(struct bch_fs *c, bool metadata_only) /** * bch2_gc - walk _all_ references to buckets, and recompute them: * + * @c: filesystem object + * @initial: are we in recovery? + * @metadata_only: are we just checking metadata references, or everything? + * + * Returns: 0 on success, or standard errcode on failure + * * Order matters here: * - Concurrent GC relies on the fact that we have a total ordering for * everything that GC walks - see gc_will_visit_node(), @@ -1695,20 +1819,16 @@ static void bch2_gc_stripes_reset(struct bch_fs *c, bool metadata_only) */ int bch2_gc(struct bch_fs *c, bool initial, bool metadata_only) { - u64 start_time = local_clock(); unsigned iter = 0; int ret; lockdep_assert_held(&c->state_lock); - trace_gc_start(c); down_write(&c->gc_lock); - /* flush interior btree updates: */ - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c)); + bch2_btree_interior_updates_flush(c); - ret = bch2_gc_start(c, metadata_only) ?: + ret = bch2_gc_start(c) ?: bch2_gc_alloc_start(c, metadata_only) ?: bch2_gc_reflink_start(c, metadata_only); if (ret) @@ -1718,30 +1838,8 @@ again: bch2_mark_superblocks(c); - if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb) && - !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags) && - c->opts.fix_errors != FSCK_OPT_NO) { - bch_info(c, "starting topology repair pass"); - ret = bch2_repair_topology(c); - if (ret) - goto out; - bch_info(c, "topology repair pass done"); - - set_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags); - } - ret = bch2_gc_btrees(c, initial, metadata_only); - if (ret == FSCK_ERR_START_TOPOLOGY_REPAIR && - !test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags) && - !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags)) { - set_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags); - ret = 0; - } - - if (ret == FSCK_ERR_START_TOPOLOGY_REPAIR) - ret = FSCK_ERR_EXIT; - if (ret) goto out; @@ -1750,7 +1848,7 @@ again: #endif c->gc_count++; - if (test_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags) || + if (test_bit(BCH_FS_need_another_gc, &c->flags) || (!iter && bch2_test_restart_gc)) { if (iter++ > 2) { bch_info(c, "Unable to fix bucket gens, looping"); @@ -1762,12 +1860,15 @@ again: * XXX: make sure gens we fixed got saved */ bch_info(c, "Second GC pass needed, restarting:"); - clear_bit(BCH_FS_NEED_ANOTHER_GC, &c->flags); + clear_bit(BCH_FS_need_another_gc, &c->flags); __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING)); bch2_gc_stripes_reset(c, metadata_only); bch2_gc_alloc_reset(c, metadata_only); bch2_gc_reflink_reset(c, metadata_only); + ret = bch2_gc_reset(c); + if (ret) + goto out; /* flush fsck errors, reset counters */ bch2_flush_fsck_errs(c); @@ -1794,21 +1895,26 @@ out: up_write(&c->gc_lock); - trace_gc_end(c); - bch2_time_stats_update(&c->times[BCH_TIME_btree_gc], start_time); - /* * At startup, allocations can happen directly instead of via the * allocator thread - issue wakeup in case they blocked on gc_lock: */ closure_wake_up(&c->freelist_wait); + + if (ret) + bch_err_fn(c, ret); return ret; } -static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k) +static int gc_btree_gens_key(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_s_c k) { + struct bch_fs *c = trans->c; struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); const struct bch_extent_ptr *ptr; + struct bkey_i *u; + int ret; percpu_down_read(&c->mark_lock); bkey_for_each_ptr(ptrs, ptr) { @@ -1816,7 +1922,7 @@ static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k) if (ptr_stale(ca, ptr) > 16) { percpu_up_read(&c->mark_lock); - return true; + goto update; } } @@ -1828,89 +1934,43 @@ static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k) *gen = ptr->gen; } percpu_up_read(&c->mark_lock); + return 0; +update: + u = bch2_bkey_make_mut(trans, iter, &k, 0); + ret = PTR_ERR_OR_ZERO(u); + if (ret) + return ret; - return false; -} - -/* - * For recalculating oldest gen, we only need to walk keys in leaf nodes; btree - * node pointers currently never have cached pointers that can become stale: - */ -static int bch2_gc_btree_gens(struct btree_trans *trans, enum btree_id btree_id) -{ - struct bch_fs *c = trans->c; - struct btree_iter iter; - struct bkey_s_c k; - struct bkey_buf sk; - int ret = 0, commit_err = 0; - - bch2_bkey_buf_init(&sk); - - bch2_trans_iter_init(trans, &iter, btree_id, POS_MIN, - BTREE_ITER_PREFETCH| - BTREE_ITER_NOT_EXTENTS| - BTREE_ITER_ALL_SNAPSHOTS); - - while ((bch2_trans_begin(trans), - k = bch2_btree_iter_peek(&iter)).k) { - ret = bkey_err(k); - - if (ret == -EINTR) - continue; - if (ret) - break; - - c->gc_gens_pos = iter.pos; - - if (gc_btree_gens_key(c, k) && !commit_err) { - bch2_bkey_buf_reassemble(&sk, c, k); - bch2_extent_normalize(c, bkey_i_to_s(sk.k)); - - commit_err = - bch2_trans_update(trans, &iter, sk.k, 0) ?: - bch2_trans_commit(trans, NULL, NULL, - BTREE_INSERT_NOWAIT| - BTREE_INSERT_NOFAIL); - if (commit_err == -EINTR) { - commit_err = 0; - continue; - } - } - - bch2_btree_iter_advance(&iter); - } - bch2_trans_iter_exit(trans, &iter); - - bch2_bkey_buf_exit(&sk, c); - - return ret; + bch2_extent_normalize(c, bkey_i_to_s(u)); + return 0; } -static int bch2_alloc_write_oldest_gen(struct btree_trans *trans, struct btree_iter *iter) +static int bch2_alloc_write_oldest_gen(struct btree_trans *trans, struct btree_iter *iter, + struct bkey_s_c k) { struct bch_dev *ca = bch_dev_bkey_exists(trans->c, iter->pos.inode); - struct bkey_s_c k; - struct bkey_alloc_unpacked u; + struct bch_alloc_v4 a_convert; + const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert); + struct bkey_i_alloc_v4 *a_mut; int ret; - k = bch2_btree_iter_peek_slot(iter); - ret = bkey_err(k); + if (a->oldest_gen == ca->oldest_gen[iter->pos.offset]) + return 0; + + a_mut = bch2_alloc_to_v4_mut(trans, k); + ret = PTR_ERR_OR_ZERO(a_mut); if (ret) return ret; - u = bch2_alloc_unpack(k); - - if (u.oldest_gen == ca->oldest_gen[iter->pos.offset]) - return 0; - - u.oldest_gen = ca->oldest_gen[iter->pos.offset]; + a_mut->v.oldest_gen = ca->oldest_gen[iter->pos.offset]; + a_mut->v.data_type = alloc_data_type(a_mut->v, a_mut->v.data_type); - return bch2_alloc_write(trans, iter, &u, 0); + return bch2_trans_update(trans, iter, &a_mut->k_i, 0); } int bch2_gc_gens(struct bch_fs *c) { - struct btree_trans trans; + struct btree_trans *trans; struct btree_iter iter; struct bkey_s_c k; struct bch_dev *ca; @@ -1926,50 +1986,56 @@ int bch2_gc_gens(struct bch_fs *c) if (!mutex_trylock(&c->gc_gens_lock)) return 0; + trace_and_count(c, gc_gens_start, c); down_read(&c->gc_lock); - bch2_trans_init(&trans, c, 0, 0); + trans = bch2_trans_get(c); for_each_member_device(ca, c, i) { - struct bucket_gens *gens; + struct bucket_gens *gens = bucket_gens(ca); BUG_ON(ca->oldest_gen); - ca->oldest_gen = kvmalloc(ca->mi.nbuckets, GFP_KERNEL); + ca->oldest_gen = kvmalloc(gens->nbuckets, GFP_KERNEL); if (!ca->oldest_gen) { percpu_ref_put(&ca->ref); - ret = -ENOMEM; + ret = -BCH_ERR_ENOMEM_gc_gens; goto err; } - gens = bucket_gens(ca); - for (b = gens->first_bucket; b < gens->nbuckets; b++) ca->oldest_gen[b] = gens->b[b]; } for (i = 0; i < BTREE_ID_NR; i++) - if ((1 << i) & BTREE_ID_HAS_PTRS) { + if (btree_type_has_ptrs(i)) { c->gc_gens_btree = i; c->gc_gens_pos = POS_MIN; - ret = bch2_gc_btree_gens(&trans, i); - if (ret) { - bch_err(c, "error recalculating oldest_gen: %i", ret); + + ret = for_each_btree_key_commit(trans, iter, i, + POS_MIN, + BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, + k, + NULL, NULL, + BCH_TRANS_COMMIT_no_enospc, + gc_btree_gens_key(trans, &iter, k)); + if (ret && !bch2_err_matches(ret, EROFS)) + bch_err_fn(c, ret); + if (ret) goto err; - } } - for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN, - BTREE_ITER_PREFETCH, k, ret) { - ret = __bch2_trans_do(&trans, NULL, NULL, - BTREE_INSERT_NOFAIL, - bch2_alloc_write_oldest_gen(&trans, &iter)); - if (ret) { - bch_err(c, "error writing oldest_gen: %i", ret); - break; - } - } - bch2_trans_iter_exit(&trans, &iter); + ret = for_each_btree_key_commit(trans, iter, BTREE_ID_alloc, + POS_MIN, + BTREE_ITER_PREFETCH, + k, + NULL, NULL, + BCH_TRANS_COMMIT_no_enospc, + bch2_alloc_write_oldest_gen(trans, &iter, k)); + if (ret && !bch2_err_matches(ret, EROFS)) + bch_err_fn(c, ret); + if (ret) + goto err; c->gc_gens_btree = 0; c->gc_gens_pos = POS_MIN; @@ -1977,13 +2043,14 @@ int bch2_gc_gens(struct bch_fs *c) c->gc_count++; bch2_time_stats_update(&c->times[BCH_TIME_btree_gc], start_time); + trace_and_count(c, gc_gens_end, c); err: for_each_member_device(ca, c, i) { kvfree(ca->oldest_gen); ca->oldest_gen = NULL; } - bch2_trans_exit(&trans); + bch2_trans_put(trans); up_read(&c->gc_lock); mutex_unlock(&c->gc_gens_lock); return ret; @@ -2038,7 +2105,7 @@ static int bch2_gc_thread(void *arg) ret = bch2_gc_gens(c); #endif if (ret < 0) - bch_err(c, "btree gc failed: %i", ret); + bch_err_fn(c, ret); debug_check_no_locks_held(); } @@ -2068,7 +2135,7 @@ int bch2_gc_thread_start(struct bch_fs *c) p = kthread_create(bch2_gc_thread, c, "bch-gc/%s", c->name); if (IS_ERR(p)) { - bch_err(c, "error creating gc thread: %li", PTR_ERR(p)); + bch_err_fn(c, PTR_ERR(p)); return PTR_ERR(p); }