From 38f8daa2b17627488c51f2e0c150213bd1636e00 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Tue, 28 Dec 2021 14:30:25 -0500 Subject: [PATCH] Update bcachefs sources to 078a1a596a bcachefs: Optimize bucket reuse --- .bcachefs_revision | 2 +- include/trace/events/bcachefs.h | 28 ++++ libbcachefs.c | 6 - libbcachefs/alloc_background.c | 202 +++++------------------ libbcachefs/alloc_background.h | 2 - libbcachefs/alloc_foreground.c | 158 ++++++++++++------ libbcachefs/alloc_foreground.h | 31 +++- libbcachefs/alloc_types.h | 13 +- libbcachefs/bcachefs.h | 5 +- libbcachefs/bcachefs_format.h | 15 +- libbcachefs/btree_cache.c | 21 ++- libbcachefs/btree_gc.c | 62 +++---- libbcachefs/btree_gc.h | 1 - libbcachefs/btree_iter.c | 1 + libbcachefs/btree_update_interior.c | 2 +- libbcachefs/buckets.c | 243 ++++++++++++++-------------- libbcachefs/buckets.h | 54 +++++-- libbcachefs/buckets_types.h | 7 + libbcachefs/ec.c | 48 +----- libbcachefs/ec.h | 2 - libbcachefs/inode.c | 53 +++--- libbcachefs/io.c | 22 +-- libbcachefs/journal.c | 57 +++++-- libbcachefs/journal.h | 1 + libbcachefs/journal_io.c | 7 +- libbcachefs/journal_seq_blacklist.c | 78 --------- libbcachefs/journal_seq_blacklist.h | 2 - libbcachefs/opts.c | 5 - libbcachefs/opts.h | 1 - libbcachefs/recovery.c | 106 ++++++------ libbcachefs/super-io.h | 1 - libbcachefs/super.c | 33 +--- libbcachefs/super.h | 21 +++ libbcachefs/super_types.h | 1 - libbcachefs/sysfs.c | 119 +++++++------- libbcachefs/tests.c | 8 +- 36 files changed, 633 insertions(+), 785 deletions(-) diff --git a/.bcachefs_revision b/.bcachefs_revision index 6b23844..e201254 100644 --- a/.bcachefs_revision +++ b/.bcachefs_revision @@ -1 +1 @@ -f9d2e809a8c3b3a3c6bb0f8fe8e646425f7fce8d +078a1a596a74ade60db6eee0f0be927defb7abed diff --git a/include/trace/events/bcachefs.h b/include/trace/events/bcachefs.h index fce3146..5a409ee 100644 --- a/include/trace/events/bcachefs.h +++ b/include/trace/events/bcachefs.h @@ -318,6 +318,34 @@ DEFINE_EVENT(btree_node, btree_set_root, TP_ARGS(c, b) ); +TRACE_EVENT(btree_cache_scan, + TP_PROTO(unsigned long nr_to_scan_pages, + unsigned long nr_to_scan_nodes, + unsigned long can_free_nodes, + long ret), + TP_ARGS(nr_to_scan_pages, nr_to_scan_nodes, can_free_nodes, ret), + + TP_STRUCT__entry( + __field(unsigned long, nr_to_scan_pages ) + __field(unsigned long, nr_to_scan_nodes ) + __field(unsigned long, can_free_nodes ) + __field(long, ret ) + ), + + TP_fast_assign( + __entry->nr_to_scan_pages = nr_to_scan_pages; + __entry->nr_to_scan_nodes = nr_to_scan_nodes; + __entry->can_free_nodes = can_free_nodes; + __entry->ret = ret; + ), + + TP_printk("scanned for %lu pages, %lu nodes, can free %lu nodes, ret %li", + __entry->nr_to_scan_pages, + __entry->nr_to_scan_nodes, + __entry->can_free_nodes, + __entry->ret) +); + /* Garbage collection */ DEFINE_EVENT(btree_node, btree_gc_rewrite_node, diff --git a/libbcachefs.c b/libbcachefs.c index 16f15a8..076fe66 100644 --- a/libbcachefs.c +++ b/libbcachefs.c @@ -231,7 +231,6 @@ struct bch_sb *bch2_format(struct bch_opt_strs fs_opt_strs, m->first_bucket = 0; m->bucket_size = cpu_to_le16(i->bucket_size); - SET_BCH_MEMBER_REPLACEMENT(m, BCH_CACHE_REPLACEMENT_lru); SET_BCH_MEMBER_DISCARD(m, i->discard); SET_BCH_MEMBER_DATA_ALLOWED(m, i->data_allowed); SET_BCH_MEMBER_DURABILITY(m, i->durability + 1); @@ -521,7 +520,6 @@ static void bch2_sb_print_members(struct bch_sb *sb, struct bch_sb_field *f, " Has data: %s\n" - " Replacement policy: %s\n" " Discard: %llu\n", i, member_uuid_str, pr_units(le16_to_cpu(m->bucket_size) * @@ -539,10 +537,6 @@ static void bch2_sb_print_members(struct bch_sb *sb, struct bch_sb_field *f, data_allowed_str, data_has_str, - BCH_MEMBER_REPLACEMENT(m) < BCH_CACHE_REPLACEMENT_NR - ? bch2_cache_replacement_policies[BCH_MEMBER_REPLACEMENT(m)] - : "unknown", - BCH_MEMBER_DISCARD(m)); } } diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c index ed919b4..2a36af5 100644 --- a/libbcachefs/alloc_background.c +++ b/libbcachefs/alloc_background.c @@ -354,6 +354,7 @@ static int bch2_alloc_read_fn(struct btree_trans *trans, struct bkey_s_c k) g = bucket(ca, k.k->p.offset); u = bch2_alloc_unpack(k); + *bucket_gen(ca, k.k->p.offset) = u.gen; g->_mark.gen = u.gen; g->_mark.data_type = u.data_type; g->_mark.dirty_sectors = u.dirty_sectors; @@ -513,6 +514,18 @@ static bool bch2_can_invalidate_bucket(struct bch_dev *ca, size_t b, test_bit(b, ca->buckets_nouse)) return false; + if (ca->new_fs_bucket_idx) { + /* + * Device or filesystem is still being initialized, and we + * haven't fully marked superblocks & journal: + */ + if (is_superblock_bucket(ca, b)) + return false; + + if (b < ca->new_fs_bucket_idx) + return false; + } + gc_gen = bucket_gc_gen(bucket(ca, b)); ca->inc_gen_needs_gc += gc_gen >= BUCKET_GC_GEN_MAX / 2; @@ -581,7 +594,7 @@ static void find_reclaimable_buckets_lru(struct bch_fs *c, struct bch_dev *ca) buckets = bucket_array(ca); ca->alloc_heap.used = 0; now = atomic64_read(&c->io_clock[READ].now); - last_seq_ondisk = c->journal.last_seq_ondisk; + last_seq_ondisk = c->journal.flushed_seq_ondisk; /* * Find buckets with lowest read priority, by building a maxheap sorted @@ -628,76 +641,6 @@ static void find_reclaimable_buckets_lru(struct bch_fs *c, struct bch_dev *ca) up_read(&ca->bucket_lock); } -static void find_reclaimable_buckets_fifo(struct bch_fs *c, struct bch_dev *ca) -{ - struct bucket_array *buckets = bucket_array(ca); - struct bucket_mark m; - size_t b, start; - - if (ca->fifo_last_bucket < ca->mi.first_bucket || - ca->fifo_last_bucket >= ca->mi.nbuckets) - ca->fifo_last_bucket = ca->mi.first_bucket; - - start = ca->fifo_last_bucket; - - do { - ca->fifo_last_bucket++; - if (ca->fifo_last_bucket == ca->mi.nbuckets) - ca->fifo_last_bucket = ca->mi.first_bucket; - - b = ca->fifo_last_bucket; - m = READ_ONCE(buckets->b[b].mark); - - if (bch2_can_invalidate_bucket(ca, b, m)) { - struct alloc_heap_entry e = { .bucket = b, .nr = 1, }; - - heap_add(&ca->alloc_heap, e, bucket_alloc_cmp, NULL); - if (heap_full(&ca->alloc_heap)) - break; - } - - cond_resched(); - } while (ca->fifo_last_bucket != start); -} - -static void find_reclaimable_buckets_random(struct bch_fs *c, struct bch_dev *ca) -{ - struct bucket_array *buckets = bucket_array(ca); - struct bucket_mark m; - size_t checked, i; - - for (checked = 0; - checked < ca->mi.nbuckets / 2; - checked++) { - size_t b = bch2_rand_range(ca->mi.nbuckets - - ca->mi.first_bucket) + - ca->mi.first_bucket; - - m = READ_ONCE(buckets->b[b].mark); - - if (bch2_can_invalidate_bucket(ca, b, m)) { - struct alloc_heap_entry e = { .bucket = b, .nr = 1, }; - - heap_add(&ca->alloc_heap, e, bucket_alloc_cmp, NULL); - if (heap_full(&ca->alloc_heap)) - break; - } - - cond_resched(); - } - - sort(ca->alloc_heap.data, - ca->alloc_heap.used, - sizeof(ca->alloc_heap.data[0]), - bucket_idx_cmp, NULL); - - /* remove duplicates: */ - for (i = 0; i + 1 < ca->alloc_heap.used; i++) - if (ca->alloc_heap.data[i].bucket == - ca->alloc_heap.data[i + 1].bucket) - ca->alloc_heap.data[i].nr = 0; -} - static size_t find_reclaimable_buckets(struct bch_fs *c, struct bch_dev *ca) { size_t i, nr = 0; @@ -705,17 +648,7 @@ static size_t find_reclaimable_buckets(struct bch_fs *c, struct bch_dev *ca) ca->inc_gen_needs_gc = 0; ca->inc_gen_really_needs_gc = 0; - switch (ca->mi.replacement) { - case BCH_CACHE_REPLACEMENT_lru: - find_reclaimable_buckets_lru(c, ca); - break; - case BCH_CACHE_REPLACEMENT_fifo: - find_reclaimable_buckets_fifo(c, ca); - break; - case BCH_CACHE_REPLACEMENT_random: - find_reclaimable_buckets_random(c, ca); - break; - } + find_reclaimable_buckets_lru(c, ca); heap_resort(&ca->alloc_heap, bucket_alloc_cmp, NULL); @@ -725,33 +658,11 @@ static size_t find_reclaimable_buckets(struct bch_fs *c, struct bch_dev *ca) return nr; } -/* - * returns sequence number of most recent journal entry that updated this - * bucket: - */ -static u64 bucket_journal_seq(struct bch_fs *c, struct bucket_mark m) -{ - if (m.journal_seq_valid) { - u64 journal_seq = atomic64_read(&c->journal.seq); - u64 bucket_seq = journal_seq; - - bucket_seq &= ~((u64) U16_MAX); - bucket_seq |= m.journal_seq; - - if (bucket_seq > journal_seq) - bucket_seq -= 1 << 16; - - return bucket_seq; - } else { - return 0; - } -} - static int bucket_invalidate_btree(struct btree_trans *trans, - struct bch_dev *ca, u64 b) + struct bch_dev *ca, u64 b, + struct bkey_alloc_unpacked *u) { struct bch_fs *c = trans->c; - struct bkey_alloc_unpacked u; struct btree_iter iter; int ret; @@ -765,16 +676,16 @@ static int bucket_invalidate_btree(struct btree_trans *trans, if (ret) goto err; - u = alloc_mem_to_key(c, &iter); + *u = alloc_mem_to_key(c, &iter); - u.gen++; - u.data_type = 0; - u.dirty_sectors = 0; - u.cached_sectors = 0; - u.read_time = atomic64_read(&c->io_clock[READ].now); - u.write_time = atomic64_read(&c->io_clock[WRITE].now); + u->gen++; + u->data_type = 0; + u->dirty_sectors = 0; + u->cached_sectors = 0; + u->read_time = atomic64_read(&c->io_clock[READ].now); + u->write_time = atomic64_read(&c->io_clock[WRITE].now); - ret = bch2_alloc_write(trans, &iter, &u, + ret = bch2_alloc_write(trans, &iter, u, BTREE_TRIGGER_BUCKET_INVALIDATE); err: bch2_trans_iter_exit(trans, &iter); @@ -784,21 +695,23 @@ err: static int bch2_invalidate_one_bucket(struct bch_fs *c, struct bch_dev *ca, u64 *journal_seq, unsigned flags) { - struct bucket *g; - struct bucket_mark m; + struct bkey_alloc_unpacked u; size_t b; int ret = 0; + /* + * If the read-only path is trying to shut down, we can't be generating + * new btree updates: + */ + if (test_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags)) + return 1; + BUG_ON(!ca->alloc_heap.used || !ca->alloc_heap.data[0].nr); b = ca->alloc_heap.data[0].bucket; /* first, put on free_inc and mark as owned by allocator: */ percpu_down_read(&c->mark_lock); - g = bucket(ca, b); - m = READ_ONCE(g->mark); - - BUG_ON(m.dirty_sectors); bch2_mark_alloc_bucket(c, ca, b, true); @@ -807,37 +720,15 @@ static int bch2_invalidate_one_bucket(struct bch_fs *c, struct bch_dev *ca, BUG_ON(!fifo_push(&ca->free_inc, b)); spin_unlock(&c->freelist_lock); - /* - * If we're not invalidating cached data, we only increment the bucket - * gen in memory here, the incremented gen will be updated in the btree - * by bch2_trans_mark_pointer(): - */ - if (!m.cached_sectors && - !bucket_needs_journal_commit(m, c->journal.last_seq_ondisk)) { - BUG_ON(m.data_type); - bucket_cmpxchg(g, m, m.gen++); - percpu_up_read(&c->mark_lock); - goto out; - } - percpu_up_read(&c->mark_lock); - /* - * If the read-only path is trying to shut down, we can't be generating - * new btree updates: - */ - if (test_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags)) { - ret = 1; - goto out; - } - ret = bch2_trans_do(c, NULL, journal_seq, BTREE_INSERT_NOCHECK_RW| BTREE_INSERT_NOFAIL| BTREE_INSERT_JOURNAL_RESERVED| flags, - bucket_invalidate_btree(&trans, ca, b)); -out: + bucket_invalidate_btree(&trans, ca, b, &u)); + if (!ret) { /* remove from alloc_heap: */ struct alloc_heap_entry e, *top = ca->alloc_heap.data; @@ -853,7 +744,7 @@ out: * bucket (i.e. deleting the last reference) before writing to * this bucket again: */ - *journal_seq = max(*journal_seq, bucket_journal_seq(c, m)); + *journal_seq = max(*journal_seq, u.journal_seq); } else { size_t b2; @@ -1133,7 +1024,7 @@ static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca) ob++) { spin_lock(&ob->lock); if (ob->valid && !ob->on_partial_list && - ob->ptr.dev == ca->dev_idx) + ob->dev == ca->dev_idx) ret = true; spin_unlock(&ob->lock); } @@ -1280,22 +1171,3 @@ void bch2_fs_allocator_background_init(struct bch_fs *c) { spin_lock_init(&c->freelist_lock); } - -void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c) -{ - struct open_bucket *ob; - - for (ob = c->open_buckets; - ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); - ob++) { - spin_lock(&ob->lock); - if (ob->valid && !ob->on_partial_list) { - pr_buf(out, "%zu ref %u type %s\n", - ob - c->open_buckets, - atomic_read(&ob->pin), - bch2_data_types[ob->type]); - } - spin_unlock(&ob->lock); - } - -} diff --git a/libbcachefs/alloc_background.h b/libbcachefs/alloc_background.h index e3cdb8b..86b6417 100644 --- a/libbcachefs/alloc_background.h +++ b/libbcachefs/alloc_background.h @@ -142,6 +142,4 @@ int bch2_dev_allocator_start(struct bch_dev *); int bch2_alloc_write_all(struct bch_fs *, unsigned); void bch2_fs_allocator_background_init(struct bch_fs *); -void bch2_open_buckets_to_text(struct printbuf *, struct bch_fs *); - #endif /* _BCACHEFS_ALLOC_BACKGROUND_H */ diff --git a/libbcachefs/alloc_foreground.c b/libbcachefs/alloc_foreground.c index dce77cc..0a63412 100644 --- a/libbcachefs/alloc_foreground.c +++ b/libbcachefs/alloc_foreground.c @@ -43,9 +43,32 @@ * reference _after_ doing the index update that makes its allocation reachable. */ +static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob) +{ + open_bucket_idx_t idx = ob - c->open_buckets; + open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket); + + ob->hash = *slot; + *slot = idx; +} + +static void bch2_open_bucket_hash_remove(struct bch_fs *c, struct open_bucket *ob) +{ + open_bucket_idx_t idx = ob - c->open_buckets; + open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket); + + while (*slot != idx) { + BUG_ON(!*slot); + slot = &c->open_buckets[*slot].hash; + } + + *slot = ob->hash; + ob->hash = 0; +} + void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); if (ob->ec) { bch2_ec_bucket_written(c, ob); @@ -55,14 +78,16 @@ void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob) percpu_down_read(&c->mark_lock); spin_lock(&ob->lock); - bch2_mark_alloc_bucket(c, ca, PTR_BUCKET_NR(ca, &ob->ptr), false); + bch2_mark_alloc_bucket(c, ca, ob->bucket, false); ob->valid = false; - ob->type = 0; + ob->data_type = 0; spin_unlock(&ob->lock); percpu_up_read(&c->mark_lock); spin_lock(&c->freelist_lock); + bch2_open_bucket_hash_remove(c, ob); + ob->freelist = c->open_buckets_freelist; c->open_buckets_freelist = ob - c->open_buckets; @@ -81,8 +106,7 @@ void bch2_open_bucket_write_error(struct bch_fs *c, unsigned i; open_bucket_for_each(c, obs, ob, i) - if (ob->ptr.dev == dev && - ob->ec) + if (ob->dev == dev && ob->ec) bch2_ec_bucket_cancel(c, ob); } @@ -95,7 +119,7 @@ static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c) ob = c->open_buckets + c->open_buckets_freelist; c->open_buckets_freelist = ob->freelist; atomic_set(&ob->pin, 1); - ob->type = 0; + ob->data_type = 0; c->open_buckets_nr_free--; return ob; @@ -105,8 +129,8 @@ static void open_bucket_free_unused(struct bch_fs *c, struct write_point *wp, struct open_bucket *ob) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); - bool may_realloc = wp->type == BCH_DATA_user; + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); + bool may_realloc = wp->data_type == BCH_DATA_user; BUG_ON(ca->open_buckets_partial_nr > ARRAY_SIZE(ca->open_buckets_partial)); @@ -133,32 +157,28 @@ static void verify_not_stale(struct bch_fs *c, const struct open_buckets *obs) struct open_bucket *ob; unsigned i; + rcu_read_lock(); open_bucket_for_each(c, obs, ob, i) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); - BUG_ON(ptr_stale(ca, &ob->ptr)); + BUG_ON(*bucket_gen(ca, ob->bucket) != ob->gen); } + rcu_read_unlock(); #endif } /* _only_ for allocating the journal on a new device: */ long bch2_bucket_alloc_new_fs(struct bch_dev *ca) { - struct bucket_array *buckets; - ssize_t b; + while (ca->new_fs_bucket_idx < ca->mi.nbuckets) { + u64 b = ca->new_fs_bucket_idx++; - rcu_read_lock(); - buckets = bucket_array(ca); - - for (b = buckets->first_bucket; b < buckets->nbuckets; b++) - if (is_available_bucket(buckets->b[b].mark) && - (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)) && - !buckets->b[b].mark.owned_by_allocator) - goto success; - b = -1; -success: - rcu_read_unlock(); - return b; + if (!is_superblock_bucket(ca, b) && + (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse))) + return b; + } + + return -1; } static inline unsigned open_buckets_reserved(enum alloc_reserve reserve) @@ -252,15 +272,14 @@ out: ob->valid = true; ob->sectors_free = ca->mi.bucket_size; ob->alloc_reserve = reserve; - ob->ptr = (struct bch_extent_ptr) { - .type = 1 << BCH_EXTENT_ENTRY_ptr, - .gen = bucket(ca, b)->mark.gen, - .offset = bucket_to_sector(ca, b), - .dev = ca->dev_idx, - }; - + ob->dev = ca->dev_idx; + ob->gen = *bucket_gen(ca, b); + ob->bucket = b; spin_unlock(&ob->lock); + ca->nr_open_buckets++; + bch2_open_bucket_hash_add(c, ob); + if (c->blocked_allocate_open_bucket) { bch2_time_stats_update( &c->times[BCH_TIME_blocked_allocate_open_bucket], @@ -275,7 +294,6 @@ out: c->blocked_allocate = 0; } - ca->nr_open_buckets++; spin_unlock(&c->freelist_lock); bch2_wake_allocator(ca); @@ -339,9 +357,9 @@ static void add_new_bucket(struct bch_fs *c, struct open_bucket *ob) { unsigned durability = - bch_dev_bkey_exists(c, ob->ptr.dev)->mi.durability; + bch_dev_bkey_exists(c, ob->dev)->mi.durability; - __clear_bit(ob->ptr.dev, devs_may_alloc->d); + __clear_bit(ob->dev, devs_may_alloc->d); *nr_effective += (flags & BUCKET_ALLOC_USE_DURABILITY) ? durability : 1; *have_cache |= !durability; @@ -451,13 +469,13 @@ static int bucket_alloc_from_stripe(struct bch_fs *c, continue; ob = c->open_buckets + h->s->blocks[ec_idx]; - if (ob->ptr.dev == devs_sorted.devs[i] && + if (ob->dev == devs_sorted.devs[i] && !test_and_set_bit(ec_idx, h->s->blocks_allocated)) goto got_bucket; } goto out_put_head; got_bucket: - ca = bch_dev_bkey_exists(c, ob->ptr.dev); + ca = bch_dev_bkey_exists(c, ob->dev); ob->ec_idx = ec_idx; ob->ec = h->s; @@ -487,12 +505,12 @@ static void get_buckets_from_writepoint(struct bch_fs *c, unsigned i; open_bucket_for_each(c, &wp->ptrs, ob, i) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); if (*nr_effective < nr_replicas && - test_bit(ob->ptr.dev, devs_may_alloc->d) && + test_bit(ob->dev, devs_may_alloc->d) && (ca->mi.durability || - (wp->type == BCH_DATA_user && !*have_cache)) && + (wp->data_type == BCH_DATA_user && !*have_cache)) && (ob->ec || !need_ec)) { add_new_bucket(c, ptrs, devs_may_alloc, nr_effective, have_cache, @@ -524,7 +542,7 @@ static int open_bucket_add_buckets(struct bch_fs *c, unsigned i; rcu_read_lock(); - devs = target_rw_devs(c, wp->type, target); + devs = target_rw_devs(c, wp->data_type, target); rcu_read_unlock(); /* Don't allocate from devices we already have pointers to: */ @@ -532,7 +550,7 @@ static int open_bucket_add_buckets(struct bch_fs *c, __clear_bit(devs_have->devs[i], devs.d); open_bucket_for_each(c, ptrs, ob, i) - __clear_bit(ob->ptr.dev, devs.d); + __clear_bit(ob->dev, devs.d); if (erasure_code) { if (!ec_open_bucket(c, ptrs)) { @@ -592,7 +610,7 @@ void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca, unsigned i, j; open_bucket_for_each(c, obs, ob, i) { - bool drop = !ca || ob->ptr.dev == ca->dev_idx; + bool drop = !ca || ob->dev == ca->dev_idx; if (!drop && ob->ec) { mutex_lock(&ob->ec->lock); @@ -601,7 +619,7 @@ void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca, continue; ob2 = c->open_buckets + ob->ec->blocks[j]; - drop |= ob2->ptr.dev == ca->dev_idx; + drop |= ob2->dev == ca->dev_idx; } mutex_unlock(&ob->ec->lock); } @@ -785,11 +803,11 @@ retry: wp = writepoint_find(c, write_point.v); - if (wp->type == BCH_DATA_user) + if (wp->data_type == BCH_DATA_user) ob_flags |= BUCKET_MAY_ALLOC_PARTIAL; /* metadata may not allocate on cache devices: */ - if (wp->type != BCH_DATA_user) + if (wp->data_type != BCH_DATA_user) have_cache = true; if (!target || (flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) { @@ -867,12 +885,27 @@ err: } } +struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob) +{ + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); + + return (struct bch_extent_ptr) { + .type = 1 << BCH_EXTENT_ENTRY_ptr, + .gen = ob->gen, + .dev = ob->dev, + .offset = bucket_to_sector(ca, ob->bucket) + + ca->mi.bucket_size - + ob->sectors_free, + }; +} + /* * Append pointers to the space we just allocated to @k, and mark @sectors space * as allocated out of @ob */ void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp, - struct bkey_i *k, unsigned sectors) + struct bkey_i *k, unsigned sectors, + bool cached) { struct open_bucket *ob; @@ -882,14 +915,14 @@ void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp, wp->sectors_free -= sectors; open_bucket_for_each(c, &wp->ptrs, ob, i) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); - struct bch_extent_ptr tmp = ob->ptr; + struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev); + struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob); - tmp.cached = !ca->mi.durability && - wp->type == BCH_DATA_user; + ptr.cached = cached || + (!ca->mi.durability && + wp->data_type == BCH_DATA_user); - tmp.offset += ca->mi.bucket_size - ob->sectors_free; - bch2_bkey_append_ptr(k, tmp); + bch2_bkey_append_ptr(k, ptr); BUG_ON(sectors > ob->sectors_free); ob->sectors_free -= sectors; @@ -919,7 +952,7 @@ static inline void writepoint_init(struct write_point *wp, enum bch_data_type type) { mutex_init(&wp->lock); - wp->type = type; + wp->data_type = type; } void bch2_fs_allocator_foreground_init(struct bch_fs *c) @@ -956,3 +989,22 @@ void bch2_fs_allocator_foreground_init(struct bch_fs *c) writepoint_hash(c, wp->write_point)); } } + +void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c) +{ + struct open_bucket *ob; + + for (ob = c->open_buckets; + ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); + ob++) { + spin_lock(&ob->lock); + if (ob->valid && !ob->on_partial_list) { + pr_buf(out, "%zu ref %u type %s\n", + ob - c->open_buckets, + atomic_read(&ob->pin), + bch2_data_types[ob->data_type]); + } + spin_unlock(&ob->lock); + } + +} diff --git a/libbcachefs/alloc_foreground.h b/libbcachefs/alloc_foreground.h index 2e81712..d466bda 100644 --- a/libbcachefs/alloc_foreground.h +++ b/libbcachefs/alloc_foreground.h @@ -85,12 +85,36 @@ static inline void bch2_open_bucket_get(struct bch_fs *c, unsigned i; open_bucket_for_each(c, &wp->ptrs, ob, i) { - ob->type = wp->type; + ob->data_type = wp->data_type; atomic_inc(&ob->pin); ob_push(c, ptrs, ob); } } +static inline open_bucket_idx_t *open_bucket_hashslot(struct bch_fs *c, + unsigned dev, u64 bucket) +{ + return c->open_buckets_hash + + (jhash_3words(dev, bucket, bucket >> 32, 0) & + (OPEN_BUCKETS_COUNT - 1)); +} + +static inline bool bch2_bucket_is_open(struct bch_fs *c, unsigned dev, u64 bucket) +{ + open_bucket_idx_t slot = *open_bucket_hashslot(c, dev, bucket); + + while (slot) { + struct open_bucket *ob = &c->open_buckets[slot]; + + if (ob->dev == dev && ob->bucket == bucket) + return true; + + slot = ob->hash; + } + + return false; +} + int bch2_bucket_alloc_set(struct bch_fs *, struct open_buckets *, struct dev_stripe_state *, struct bch_devs_mask *, unsigned, unsigned *, bool *, enum alloc_reserve, @@ -105,8 +129,9 @@ struct write_point *bch2_alloc_sectors_start(struct bch_fs *, unsigned, struct closure *); +struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *, struct open_bucket *); void bch2_alloc_sectors_append_ptrs(struct bch_fs *, struct write_point *, - struct bkey_i *, unsigned); + struct bkey_i *, unsigned, bool); void bch2_alloc_sectors_done(struct bch_fs *, struct write_point *); void bch2_open_buckets_stop_dev(struct bch_fs *, struct bch_dev *, @@ -127,4 +152,6 @@ static inline struct write_point_specifier writepoint_ptr(struct write_point *wp void bch2_fs_allocator_foreground_init(struct bch_fs *); +void bch2_open_buckets_to_text(struct printbuf *, struct bch_fs *); + #endif /* _BCACHEFS_ALLOC_FOREGROUND_H */ diff --git a/libbcachefs/alloc_types.h b/libbcachefs/alloc_types.h index 4a1cd8b..409232e 100644 --- a/libbcachefs/alloc_types.h +++ b/libbcachefs/alloc_types.h @@ -37,24 +37,31 @@ typedef FIFO(long) alloc_fifo; #define WRITE_POINT_HASH_NR 32 #define WRITE_POINT_MAX 32 +/* + * 0 is never a valid open_bucket_idx_t: + */ typedef u16 open_bucket_idx_t; struct open_bucket { spinlock_t lock; atomic_t pin; open_bucket_idx_t freelist; + open_bucket_idx_t hash; /* * When an open bucket has an ec_stripe attached, this is the index of * the block in the stripe this open_bucket corresponds to: */ u8 ec_idx; - u8 type; + enum bch_data_type data_type:3; unsigned valid:1; unsigned on_partial_list:1; int alloc_reserve:3; + unsigned sectors_free; - struct bch_extent_ptr ptr; + u8 dev; + u8 gen; + u64 bucket; struct ec_stripe_new *ec; }; @@ -74,7 +81,7 @@ struct write_point { struct mutex lock; u64 last_used; unsigned long write_point; - enum bch_data_type type; + enum bch_data_type data_type; /* calculated based on how many pointers we're actually going to use: */ unsigned sectors_free; diff --git a/libbcachefs/bcachefs.h b/libbcachefs/bcachefs.h index 943487f..3ada85a 100644 --- a/libbcachefs/bcachefs.h +++ b/libbcachefs/bcachefs.h @@ -445,6 +445,7 @@ struct bch_dev { * Or rcu_read_lock(), but only for ptr_stale(): */ struct bucket_array __rcu *buckets[2]; + struct bucket_gens *bucket_gens; unsigned long *buckets_nouse; struct rw_semaphore bucket_lock; @@ -453,6 +454,7 @@ struct bch_dev { struct bch_dev_usage __percpu *usage_gc; /* Allocator: */ + u64 new_fs_bucket_idx; struct task_struct __rcu *alloc_thread; /* @@ -748,17 +750,18 @@ struct bch_fs { /* JOURNAL SEQ BLACKLIST */ struct journal_seq_blacklist_table * journal_seq_blacklist_table; - struct work_struct journal_seq_blacklist_gc_work; /* ALLOCATOR */ spinlock_t freelist_lock; struct closure_waitlist freelist_wait; u64 blocked_allocate; u64 blocked_allocate_open_bucket; + open_bucket_idx_t open_buckets_freelist; open_bucket_idx_t open_buckets_nr_free; struct closure_waitlist open_buckets_wait; struct open_bucket open_buckets[OPEN_BUCKETS_COUNT]; + open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT]; struct write_point btree_write_point; struct write_point rebalance_write_point; diff --git a/libbcachefs/bcachefs_format.h b/libbcachefs/bcachefs_format.h index 495f4d1..a053fca 100644 --- a/libbcachefs/bcachefs_format.h +++ b/libbcachefs/bcachefs_format.h @@ -1063,8 +1063,7 @@ struct bch_member { }; LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags[0], 0, 4) -/* 4-10 unused, was TIER, HAS_(META)DATA */ -LE64_BITMASK(BCH_MEMBER_REPLACEMENT, struct bch_member, flags[0], 10, 14) +/* 4-14 unused, was TIER, HAS_(META)DATA, REPLACEMENT */ LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags[0], 14, 15) LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags[0], 15, 20) LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags[0], 20, 28) @@ -1088,18 +1087,6 @@ enum bch_member_state { BCH_MEMBER_STATE_NR }; -#define BCH_CACHE_REPLACEMENT_POLICIES() \ - x(lru, 0) \ - x(fifo, 1) \ - x(random, 2) - -enum bch_cache_replacement_policies { -#define x(t, n) BCH_CACHE_REPLACEMENT_##t = n, - BCH_CACHE_REPLACEMENT_POLICIES() -#undef x - BCH_CACHE_REPLACEMENT_NR -}; - struct bch_sb_field_members { struct bch_sb_field field; struct bch_member members[0]; diff --git a/libbcachefs/btree_cache.c b/libbcachefs/btree_cache.c index 3411d5a..b13563d 100644 --- a/libbcachefs/btree_cache.c +++ b/libbcachefs/btree_cache.c @@ -274,6 +274,7 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink, unsigned long touched = 0; unsigned long freed = 0; unsigned i, flags; + unsigned long ret = SHRINK_STOP; if (bch2_btree_shrinker_disabled) return SHRINK_STOP; @@ -282,7 +283,7 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink, if (sc->gfp_mask & __GFP_FS) mutex_lock(&bc->lock); else if (!mutex_trylock(&bc->lock)) - return -1; + goto out_norestore; flags = memalloc_nofs_save(); @@ -299,13 +300,19 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink, i = 0; list_for_each_entry_safe(b, t, &bc->freeable, list) { + /* + * Leave a few nodes on the freeable list, so that a btree split + * won't have to hit the system allocator: + */ + if (++i <= 3) + continue; + touched++; if (touched >= nr) break; - if (++i > 3 && - !btree_node_reclaim(c, b)) { + if (!btree_node_reclaim(c, b)) { btree_node_data_free(c, b); six_unlock_write(&b->c.lock); six_unlock_intent(&b->c.lock); @@ -351,8 +358,14 @@ restart: mutex_unlock(&bc->lock); out: + ret = (unsigned long) freed * btree_pages(c); memalloc_nofs_restore(flags); - return (unsigned long) freed * btree_pages(c); +out_norestore: + trace_btree_cache_scan(sc->nr_to_scan, + sc->nr_to_scan / btree_pages(c), + btree_cache_can_free(bc), + ret); + return ret; } static unsigned long bch2_btree_cache_count(struct shrinker *shrink, diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c index 91c69a9..d188370 100644 --- a/libbcachefs/btree_gc.c +++ b/libbcachefs/btree_gc.c @@ -170,10 +170,10 @@ static int set_node_min(struct bch_fs *c, struct btree *b, struct bpos new_min) SET_BTREE_PTR_RANGE_UPDATED(&new->v, true); ret = bch2_journal_key_insert(c, b->c.btree_id, b->c.level + 1, &new->k_i); - if (ret) { - kfree(new); + kfree(new); + + if (ret) return ret; - } bch2_btree_node_drop_keys_outside_node(b); @@ -199,10 +199,10 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max) SET_BTREE_PTR_RANGE_UPDATED(&new->v, true); ret = bch2_journal_key_insert(c, b->c.btree_id, b->c.level + 1, &new->k_i); - if (ret) { - kfree(new); + kfree(new); + + if (ret) return ret; - } bch2_btree_node_drop_keys_outside_node(b); @@ -504,8 +504,8 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, */ bkey_for_each_ptr_decode(k->k, ptrs, p, entry) { struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); - struct bucket *g = PTR_BUCKET(ca, &p.ptr, true); - struct bucket *g2 = PTR_BUCKET(ca, &p.ptr, false); + struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr); + struct bucket *g2 = PTR_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, @@ -643,14 +643,14 @@ static int bch2_check_fix_ptrs(struct bch_fs *c, enum btree_id btree_id, ptrs = bch2_bkey_ptrs(bkey_i_to_s(new)); bkey_for_each_ptr(ptrs, ptr) { struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bucket *g = PTR_BUCKET(ca, ptr, true); + struct bucket *g = PTR_GC_BUCKET(ca, ptr); ptr->gen = g->mark.gen; } } else { bch2_bkey_drop_ptrs(bkey_i_to_s(new), ptr, ({ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bucket *g = PTR_BUCKET(ca, ptr, true); + struct bucket *g = PTR_GC_BUCKET(ca, ptr); enum bch_data_type data_type = bch2_bkey_ptr_data_type(*k, ptr); (ptr->cached && @@ -691,9 +691,9 @@ found: } ret = bch2_journal_key_insert(c, btree_id, level, new); - if (ret) - kfree(new); - else + kfree(new); + + if (!ret) *k = bkey_i_to_s_c(new); } fsck_err: @@ -737,7 +737,7 @@ static int bch2_gc_mark_key(struct btree_trans *trans, enum btree_id btree_id, ptrs = bch2_bkey_ptrs_c(*k); bkey_for_each_ptr(ptrs, ptr) { struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bucket *g = PTR_BUCKET(ca, ptr, true); + struct bucket *g = PTR_GC_BUCKET(ca, ptr); if (gen_after(g->oldest_gen, ptr->gen)) g->oldest_gen = ptr->gen; @@ -1056,23 +1056,13 @@ static void mark_metadata_sectors(struct bch_fs *c, struct bch_dev *ca, } while (start < end); } -void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca, - unsigned flags) +static void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca, + unsigned flags) { struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; unsigned i; u64 b; - /* - * This conditional is kind of gross, but we may be called from the - * device add path, before the new device has actually been added to the - * running filesystem: - */ - if (c) { - lockdep_assert_held(&c->sb_lock); - percpu_down_read(&c->mark_lock); - } - for (i = 0; i < layout->nr_superblocks; i++) { u64 offset = le64_to_cpu(layout->sb_offset[i]); @@ -1091,9 +1081,6 @@ void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca, ca->mi.bucket_size, gc_phase(GC_PHASE_SB), flags); } - - if (c) - percpu_up_read(&c->mark_lock); } static void bch2_mark_superblocks(struct bch_fs *c) @@ -1283,7 +1270,6 @@ static int bch2_gc_start(struct bch_fs *c, { struct bch_dev *ca = NULL; unsigned i; - int ret; BUG_ON(c->usage_gc); @@ -1315,12 +1301,6 @@ static int bch2_gc_start(struct bch_fs *c, } } - ret = bch2_ec_mem_alloc(c, true); - if (ret) { - bch_err(c, "error allocating ec gc mem"); - return ret; - } - percpu_down_write(&c->mark_lock); /* @@ -1403,8 +1383,7 @@ static int bch2_gc_reflink_done_initial_fn(struct btree_trans *trans, } ret = bch2_journal_key_insert(c, BTREE_ID_reflink, 0, new); - if (ret) - kfree(new); + kfree(new); } fsck_err: return ret; @@ -1529,8 +1508,7 @@ inconsistent: stripe_blockcount_set(&new->v, i, m ? m->block_sectors[i] : 0); ret = bch2_journal_key_insert(c, BTREE_ID_stripes, 0, &new->k_i); - if (ret) - kfree(new); + kfree(new); } fsck_err: return ret; @@ -1768,7 +1746,7 @@ static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k) percpu_down_read(&c->mark_lock); bkey_for_each_ptr(ptrs, ptr) { struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bucket *g = PTR_BUCKET(ca, ptr, false); + struct bucket *g = PTR_BUCKET(ca, ptr); if (gen_after(g->mark.gen, ptr->gen) > 16) { percpu_up_read(&c->mark_lock); @@ -1778,7 +1756,7 @@ static bool gc_btree_gens_key(struct bch_fs *c, struct bkey_s_c k) bkey_for_each_ptr(ptrs, ptr) { struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bucket *g = PTR_BUCKET(ca, ptr, false); + struct bucket *g = PTR_BUCKET(ca, ptr); if (gen_after(g->gc_gen, ptr->gen)) g->gc_gen = ptr->gen; diff --git a/libbcachefs/btree_gc.h b/libbcachefs/btree_gc.h index 59dfb06..0665f59 100644 --- a/libbcachefs/btree_gc.h +++ b/libbcachefs/btree_gc.h @@ -8,7 +8,6 @@ int bch2_gc(struct bch_fs *, bool, bool); int bch2_gc_gens(struct bch_fs *); void bch2_gc_thread_stop(struct bch_fs *); int bch2_gc_thread_start(struct bch_fs *); -void bch2_mark_dev_superblock(struct bch_fs *, struct bch_dev *, unsigned); /* * For concurrent mark and sweep (with other index updates), we define a total diff --git a/libbcachefs/btree_iter.c b/libbcachefs/btree_iter.c index e9091f8..65ab2cd 100644 --- a/libbcachefs/btree_iter.c +++ b/libbcachefs/btree_iter.c @@ -746,6 +746,7 @@ static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k k.k->p.snapshot)); bch2_trans_iter_init(trans, ©, iter->btree_id, iter->pos, + BTREE_ITER_NOPRESERVE| BTREE_ITER_ALL_SNAPSHOTS); prev = bch2_btree_iter_prev(©); if (!prev.k) diff --git a/libbcachefs/btree_update_interior.c b/libbcachefs/btree_update_interior.c index 9dca694..6872e56 100644 --- a/libbcachefs/btree_update_interior.c +++ b/libbcachefs/btree_update_interior.c @@ -236,7 +236,7 @@ retry: } bkey_btree_ptr_v2_init(&tmp.k); - bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, btree_sectors(c)); + bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, btree_sectors(c), false); bch2_open_bucket_get(c, wp, &ob); bch2_alloc_sectors_done(c, wp); diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c index c73abe6..738ce67 100644 --- a/libbcachefs/buckets.c +++ b/libbcachefs/buckets.c @@ -50,7 +50,7 @@ static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage, void bch2_bucket_seq_cleanup(struct bch_fs *c) { u64 journal_seq = atomic64_read(&c->journal.seq); - u16 last_seq_ondisk = c->journal.last_seq_ondisk; + u16 last_seq_ondisk = c->journal.flushed_seq_ondisk; struct bch_dev *ca; struct bucket_array *buckets; struct bucket *g; @@ -340,13 +340,6 @@ static inline enum bch_data_type bucket_type(struct bucket_mark m) : m.data_type; } -static bool bucket_became_unavailable(struct bucket_mark old, - struct bucket_mark new) -{ - return is_available_bucket(old) && - !is_available_bucket(new); -} - static inline void account_bucket(struct bch_fs_usage *fs_usage, struct bch_dev_usage *dev_usage, enum bch_data_type type, @@ -532,19 +525,6 @@ static inline void update_cached_sectors_list(struct btree_trans *trans, update_replicas_list(trans, &r.e, sectors); } -#define do_mark_fn(fn, c, pos, flags, ...) \ -({ \ - int gc, ret = 0; \ - \ - percpu_rwsem_assert_held(&c->mark_lock); \ - \ - for (gc = 0; gc < 2 && !ret; gc++) \ - if (!gc == !(flags & BTREE_TRIGGER_GC) || \ - (gc && gc_visited(c, pos))) \ - ret = fn(c, __VA_ARGS__, gc); \ - ret; \ -}) - void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca, size_t b, bool owned_by_allocator) { @@ -558,6 +538,20 @@ void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca, BUG_ON(owned_by_allocator == old.owned_by_allocator); } +static inline u8 bkey_alloc_gen(struct bkey_s_c k) +{ + switch (k.k->type) { + case KEY_TYPE_alloc: + return bkey_s_c_to_alloc(k).v->gen; + case KEY_TYPE_alloc_v2: + return bkey_s_c_to_alloc_v2(k).v->gen; + case KEY_TYPE_alloc_v3: + return bkey_s_c_to_alloc_v3(k).v->gen; + default: + return 0; + } +} + static int bch2_mark_alloc(struct btree_trans *trans, struct bkey_s_c old, struct bkey_s_c new, unsigned flags) @@ -565,16 +559,13 @@ static int bch2_mark_alloc(struct btree_trans *trans, bool gc = flags & BTREE_TRIGGER_GC; u64 journal_seq = trans->journal_res.seq; struct bch_fs *c = trans->c; - struct bkey_alloc_unpacked u; + struct bkey_alloc_unpacked old_u = bch2_alloc_unpack(old); + struct bkey_alloc_unpacked new_u = bch2_alloc_unpack(new); struct bch_dev *ca; struct bucket *g; struct bucket_mark old_m, m; int ret = 0; - /* We don't do anything for deletions - do we?: */ - if (!bkey_is_alloc(new.k)) - return 0; - /* * alloc btree is read in by bch2_alloc_read, not gc: */ @@ -582,13 +573,21 @@ static int bch2_mark_alloc(struct btree_trans *trans, !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE)) return 0; - if (flags & BTREE_TRIGGER_INSERT) { + if ((flags & BTREE_TRIGGER_INSERT) && + !old_u.data_type != !new_u.data_type && + new.k->type == KEY_TYPE_alloc_v3) { struct bch_alloc_v3 *v = (struct bch_alloc_v3 *) new.v; BUG_ON(!journal_seq); - BUG_ON(new.k->type != KEY_TYPE_alloc_v3); - v->journal_seq = cpu_to_le64(journal_seq); + /* + * If the btree updates referring to a bucket weren't flushed + * before the bucket became empty again, then the we don't have + * to wait on a journal flush before we can reuse the bucket: + */ + v->journal_seq = !new_u.data_type && + bch2_journal_noflush_seq(&c->journal, journal_seq) + ? 0 : cpu_to_le64(journal_seq); } ca = bch_dev_bkey_exists(c, new.k->p.inode); @@ -597,15 +596,17 @@ static int bch2_mark_alloc(struct btree_trans *trans, return 0; percpu_down_read(&c->mark_lock); + if (!gc && new_u.gen != bkey_alloc_gen(old)) + *bucket_gen(ca, new.k->p.offset) = new_u.gen; + g = __bucket(ca, new.k->p.offset, gc); - u = bch2_alloc_unpack(new); old_m = bucket_cmpxchg(g, m, ({ - m.gen = u.gen; - m.data_type = u.data_type; - m.dirty_sectors = u.dirty_sectors; - m.cached_sectors = u.cached_sectors; - m.stripe = u.stripe != 0; + m.gen = new_u.gen; + m.data_type = new_u.data_type; + m.dirty_sectors = new_u.dirty_sectors; + m.cached_sectors = new_u.cached_sectors; + m.stripe = new_u.stripe != 0; if (journal_seq) { m.journal_seq_valid = 1; @@ -615,12 +616,12 @@ static int bch2_mark_alloc(struct btree_trans *trans, bch2_dev_usage_update(c, ca, old_m, m, journal_seq, gc); - g->io_time[READ] = u.read_time; - g->io_time[WRITE] = u.write_time; - g->oldest_gen = u.oldest_gen; + g->io_time[READ] = new_u.read_time; + g->io_time[WRITE] = new_u.write_time; + g->oldest_gen = new_u.oldest_gen; g->gen_valid = 1; - g->stripe = u.stripe; - g->stripe_redundancy = u.stripe_redundancy; + g->stripe = new_u.stripe; + g->stripe_redundancy = new_u.stripe_redundancy; percpu_up_read(&c->mark_lock); /* @@ -655,17 +656,27 @@ static int bch2_mark_alloc(struct btree_trans *trans, overflow; \ }) -static int __bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t b, enum bch_data_type data_type, - unsigned sectors, bool gc) +void bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca, + size_t b, enum bch_data_type data_type, + unsigned sectors, struct gc_pos pos, + unsigned flags) { - struct bucket *g = __bucket(ca, b, gc); + struct bucket *g; struct bucket_mark old, new; bool overflow; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); BUG_ON(data_type != BCH_DATA_sb && data_type != BCH_DATA_journal); + /* + * Backup superblock might be past the end of our normal usable space: + */ + if (b >= ca->mi.nbuckets) + return; + + percpu_down_read(&c->mark_lock); + g = gc_bucket(ca, b); old = bucket_cmpxchg(g, new, ({ new.data_type = data_type; overflow = checked_add(new.dirty_sectors, sectors); @@ -683,32 +694,8 @@ static int __bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca, bch2_data_types[old.data_type ?: data_type], old.dirty_sectors, sectors); - if (c) - bch2_dev_usage_update(c, ca, old, new, 0, gc); - - return 0; -} - -void bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t b, enum bch_data_type type, - unsigned sectors, struct gc_pos pos, - unsigned flags) -{ - BUG_ON(type != BCH_DATA_sb && - type != BCH_DATA_journal); - - /* - * Backup superblock might be past the end of our normal usable space: - */ - if (b >= ca->mi.nbuckets) - return; - - if (likely(c)) { - do_mark_fn(__bch2_mark_metadata_bucket, c, pos, flags, - ca, b, type, sectors); - } else { - __bch2_mark_metadata_bucket(c, ca, b, type, sectors, 0); - } + bch2_dev_usage_update(c, ca, old, new, 0, true); + percpu_up_read(&c->mark_lock); } static s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p) @@ -809,17 +796,18 @@ static int mark_stripe_bucket(struct btree_trans *trans, enum bch_data_type data_type = parity ? BCH_DATA_parity : 0; s64 sectors = parity ? le16_to_cpu(s->sectors) : 0; const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx; - bool gc = flags & BTREE_TRIGGER_GC; struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); struct bucket *g; struct bucket_mark new, old; char buf[200]; int ret = 0; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + /* * XXX doesn't handle deletion */ percpu_down_read(&c->mark_lock); - g = PTR_BUCKET(ca, ptr, gc); + g = PTR_GC_BUCKET(ca, ptr); if (g->mark.dirty_sectors || (g->stripe && g->stripe != k.k->p.offset)) { @@ -853,7 +841,7 @@ static int mark_stripe_bucket(struct btree_trans *trans, g->stripe = k.k->p.offset; g->stripe_redundancy = s->nr_redundant; - bch2_dev_usage_update(c, ca, old, new, journal_seq, gc); + bch2_dev_usage_update(c, ca, old, new, journal_seq, true); err: percpu_up_read(&c->mark_lock); @@ -889,18 +877,19 @@ static int bch2_mark_pointer(struct btree_trans *trans, s64 sectors, enum bch_data_type data_type, unsigned flags) { - bool gc = flags & BTREE_TRIGGER_GC; u64 journal_seq = trans->journal_res.seq; struct bch_fs *c = trans->c; struct bucket_mark old, new; struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); - struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc); + struct bucket *g; u8 bucket_data_type; u64 v; int ret = 0; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + percpu_down_read(&c->mark_lock); - g = PTR_BUCKET(ca, &p.ptr, gc); + g = PTR_GC_BUCKET(ca, &p.ptr); v = atomic64_read(&g->_mark.v); do { @@ -930,9 +919,7 @@ static int bch2_mark_pointer(struct btree_trans *trans, old.v.counter, new.v.counter)) != old.v.counter); - bch2_dev_usage_update(c, ca, old, new, journal_seq, gc); - - BUG_ON(!gc && bucket_became_unavailable(old, new)); + bch2_dev_usage_update(c, ca, old, new, journal_seq, true); err: percpu_up_read(&c->mark_lock); @@ -946,37 +933,35 @@ static int bch2_mark_stripe_ptr(struct btree_trans *trans, s64 sectors, unsigned flags) { - bool gc = flags & BTREE_TRIGGER_GC; struct bch_fs *c = trans->c; struct bch_replicas_padded r; + struct gc_stripe *m; - if (!gc) { - BUG(); - } else { - struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL); + BUG_ON(!(flags & BTREE_TRIGGER_GC)); - if (!m) - return -ENOMEM; + m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL); - spin_lock(&c->ec_stripes_heap_lock); - - if (!m || !m->alive) { - spin_unlock(&c->ec_stripes_heap_lock); - bch_err_ratelimited(c, "pointer to nonexistent stripe %llu", - (u64) p.idx); - bch2_inconsistent_error(c); - return -EIO; - } + if (!m) + return -ENOMEM; - m->block_sectors[p.block] += sectors; + spin_lock(&c->ec_stripes_heap_lock); - r = m->r; + if (!m || !m->alive) { spin_unlock(&c->ec_stripes_heap_lock); - - r.e.data_type = data_type; - update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, gc); + bch_err_ratelimited(c, "pointer to nonexistent stripe %llu", + (u64) p.idx); + bch2_inconsistent_error(c); + return -EIO; } + m->block_sectors[p.block] += sectors; + + r = m->r; + spin_unlock(&c->ec_stripes_heap_lock); + + r.e.data_type = data_type; + update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true); + return 0; } @@ -984,7 +969,6 @@ static int bch2_mark_extent(struct btree_trans *trans, struct bkey_s_c old, struct bkey_s_c new, unsigned flags) { - bool gc = flags & BTREE_TRIGGER_GC; u64 journal_seq = trans->journal_res.seq; struct bch_fs *c = trans->c; struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old: new; @@ -1002,6 +986,8 @@ static int bch2_mark_extent(struct btree_trans *trans, bool stale; int ret; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + r.e.data_type = data_type; r.e.nr_devs = 0; r.e.nr_required = 1; @@ -1022,7 +1008,7 @@ static int bch2_mark_extent(struct btree_trans *trans, if (p.ptr.cached) { if (!stale) { ret = update_cached_sectors(c, k, p.ptr.dev, - disk_sectors, journal_seq, gc); + disk_sectors, journal_seq, true); if (ret) { bch2_fs_fatal_error(c, "bch2_mark_extent(): no replicas entry while updating cached sectors"); return ret; @@ -1047,7 +1033,7 @@ static int bch2_mark_extent(struct btree_trans *trans, } if (r.e.nr_devs) { - ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, gc); + ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true); if (ret) { char buf[200]; @@ -1114,7 +1100,11 @@ static int bch2_mark_stripe(struct btree_trans *trans, spin_unlock(&c->ec_stripes_heap_lock); } } else { - struct gc_stripe *m = genradix_ptr(&c->gc_stripes, idx); + struct gc_stripe *m = + genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL); + + if (!m) + return -ENOMEM; /* * This will be wrong when we bring back runtime gc: we should @@ -1198,6 +1188,8 @@ static int bch2_mark_reservation(struct btree_trans *trans, unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas; s64 sectors = (s64) k.k->size; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + if (flags & BTREE_TRIGGER_OVERWRITE) sectors = -sectors; sectors *= replicas; @@ -1247,19 +1239,13 @@ not_found: */ if (fsck_err(c, "%llu:%llu len %u points to nonexistent indirect extent %llu", p.k->p.inode, p.k->p.offset, p.k->size, *idx)) { - struct bkey_i_error *new; - - new = kmalloc(sizeof(*new), GFP_KERNEL); - if (!new) { - bch_err(c, "%s: error allocating new key", __func__); - return -ENOMEM; - } + struct bkey_i_error new; - bkey_init(&new->k); - new->k.type = KEY_TYPE_error; - new->k.p = p.k->p; - new->k.size = p.k->size; - ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new->k_i); + bkey_init(&new.k); + new.k.type = KEY_TYPE_error; + new.k.p = p.k->p; + new.k.size = p.k->size; + ret = bch2_journal_key_insert(c, BTREE_ID_extents, 0, &new.k_i); } fsck_err: return ret; @@ -1278,6 +1264,8 @@ static int bch2_mark_reflink_p(struct btree_trans *trans, u64 end = le64_to_cpu(p.v->idx) + p.k->size; int ret = 0; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + if (c->sb.version >= bcachefs_metadata_version_reflink_p_fix) { idx -= le32_to_cpu(p.v->front_pad); end += le32_to_cpu(p.v->back_pad); @@ -2170,9 +2158,18 @@ static void buckets_free_rcu(struct rcu_head *rcu) buckets->nbuckets * sizeof(struct bucket)); } +static void bucket_gens_free_rcu(struct rcu_head *rcu) +{ + struct bucket_gens *buckets = + container_of(rcu, struct bucket_gens, rcu); + + kvpfree(buckets, sizeof(struct bucket_array) + buckets->nbuckets); +} + int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) { struct bucket_array *buckets = NULL, *old_buckets = NULL; + struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL; unsigned long *buckets_nouse = NULL; alloc_fifo free[RESERVE_NR]; alloc_fifo free_inc; @@ -2196,6 +2193,8 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) if (!(buckets = kvpmalloc(sizeof(struct bucket_array) + nbuckets * sizeof(struct bucket), GFP_KERNEL|__GFP_ZERO)) || + !(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets, + GFP_KERNEL|__GFP_ZERO)) || !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) * sizeof(unsigned long), GFP_KERNEL|__GFP_ZERO)) || @@ -2208,6 +2207,8 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) buckets->first_bucket = ca->mi.first_bucket; buckets->nbuckets = nbuckets; + bucket_gens->first_bucket = ca->mi.first_bucket; + bucket_gens->nbuckets = nbuckets; bch2_copygc_stop(c); @@ -2218,6 +2219,7 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) } old_buckets = bucket_array(ca); + old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1); if (resize) { size_t n = min(buckets->nbuckets, old_buckets->nbuckets); @@ -2225,13 +2227,18 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) memcpy(buckets->b, old_buckets->b, n * sizeof(struct bucket)); + memcpy(bucket_gens->b, + old_bucket_gens->b, + n); memcpy(buckets_nouse, ca->buckets_nouse, BITS_TO_LONGS(n) * sizeof(unsigned long)); } rcu_assign_pointer(ca->buckets[0], buckets); - buckets = old_buckets; + rcu_assign_pointer(ca->bucket_gens, bucket_gens); + buckets = old_buckets; + bucket_gens = old_bucket_gens; swap(ca->buckets_nouse, buckets_nouse); @@ -2265,6 +2272,8 @@ err: free_fifo(&free[i]); kvpfree(buckets_nouse, BITS_TO_LONGS(nbuckets) * sizeof(unsigned long)); + if (bucket_gens) + call_rcu(&old_buckets->rcu, bucket_gens_free_rcu); if (buckets) call_rcu(&old_buckets->rcu, buckets_free_rcu); diff --git a/libbcachefs/buckets.h b/libbcachefs/buckets.h index ac9b554..45c6d23 100644 --- a/libbcachefs/buckets.h +++ b/libbcachefs/buckets.h @@ -53,11 +53,34 @@ static inline struct bucket *__bucket(struct bch_dev *ca, size_t b, bool gc) return buckets->b + b; } +static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b) +{ + return __bucket(ca, b, true); +} + static inline struct bucket *bucket(struct bch_dev *ca, size_t b) { return __bucket(ca, b, false); } +static inline struct bucket_gens *bucket_gens(struct bch_dev *ca) +{ + return rcu_dereference_check(ca->bucket_gens, + !ca->fs || + percpu_rwsem_is_held(&ca->fs->mark_lock) || + lockdep_is_held(&ca->fs->gc_lock) || + lockdep_is_held(&ca->bucket_lock)); + +} + +static inline u8 *bucket_gen(struct bch_dev *ca, size_t b) +{ + struct bucket_gens *gens = bucket_gens(ca); + + BUG_ON(b < gens->first_bucket || b >= gens->nbuckets); + return gens->b + b; +} + /* * bucket_gc_gen() returns the difference between the bucket's current gen and * the oldest gen of any pointer into that bucket in the btree. @@ -75,10 +98,15 @@ static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca, } static inline struct bucket *PTR_BUCKET(struct bch_dev *ca, - const struct bch_extent_ptr *ptr, - bool gc) + const struct bch_extent_ptr *ptr) { - return __bucket(ca, PTR_BUCKET_NR(ca, ptr), gc); + return bucket(ca, PTR_BUCKET_NR(ca, ptr)); +} + +static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca, + const struct bch_extent_ptr *ptr) +{ + return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr)); } static inline enum bch_data_type ptr_data_type(const struct bkey *k, @@ -91,18 +119,6 @@ static inline enum bch_data_type ptr_data_type(const struct bkey *k, return ptr->cached ? BCH_DATA_cached : BCH_DATA_user; } -static inline struct bucket_mark ptr_bucket_mark(struct bch_dev *ca, - const struct bch_extent_ptr *ptr) -{ - struct bucket_mark m; - - rcu_read_lock(); - m = READ_ONCE(PTR_BUCKET(ca, ptr, 0)->mark); - rcu_read_unlock(); - - return m; -} - static inline int gen_cmp(u8 a, u8 b) { return (s8) (a - b); @@ -122,7 +138,13 @@ static inline int gen_after(u8 a, u8 b) static inline u8 ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr) { - return gen_after(ptr_bucket_mark(ca, ptr).gen, ptr->gen); + u8 ret; + + rcu_read_lock(); + ret = gen_after(*bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)), ptr->gen); + rcu_read_unlock(); + + return ret; } /* bucket gc marks */ diff --git a/libbcachefs/buckets_types.h b/libbcachefs/buckets_types.h index b2de299..18bca26 100644 --- a/libbcachefs/buckets_types.h +++ b/libbcachefs/buckets_types.h @@ -52,6 +52,13 @@ struct bucket_array { struct bucket b[]; }; +struct bucket_gens { + struct rcu_head rcu; + u16 first_bucket; + size_t nbuckets; + u8 b[]; +}; + struct bch_dev_usage { u64 buckets_ec; u64 buckets_unavailable; diff --git a/libbcachefs/ec.c b/libbcachefs/ec.c index 7d78672..3cccd1f 100644 --- a/libbcachefs/ec.c +++ b/libbcachefs/ec.c @@ -1063,7 +1063,7 @@ void *bch2_writepoint_ec_buf(struct bch_fs *c, struct write_point *wp) if (!ob) return NULL; - ca = bch_dev_bkey_exists(c, ob->ptr.dev); + ca = bch_dev_bkey_exists(c, ob->dev); offset = ca->mi.bucket_size - ob->sectors_free; return ob->ec->new_stripe.data[ob->ec_idx] + (offset << 9); @@ -1318,7 +1318,7 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h, BUG_ON(j >= h->s->nr_data + h->s->nr_parity); h->s->blocks[j] = buckets.v[i]; - h->s->new_stripe.key.v.ptrs[j] = ob->ptr; + h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob); __set_bit(j, h->s->blocks_gotten); } @@ -1346,7 +1346,7 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h, BUG_ON(j >= h->s->nr_data); h->s->blocks[j] = buckets.v[i]; - h->s->new_stripe.key.v.ptrs[j] = ob->ptr; + h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob); __set_bit(j, h->s->blocks_gotten); } @@ -1535,7 +1535,7 @@ void bch2_ec_stop_dev(struct bch_fs *c, struct bch_dev *ca) continue; ob = c->open_buckets + h->s->blocks[i]; - if (ob->ptr.dev == ca->dev_idx) + if (ob->dev == ca->dev_idx) goto found; } goto unlock; @@ -1608,46 +1608,6 @@ int bch2_stripes_read(struct bch_fs *c) return ret; } -int bch2_ec_mem_alloc(struct bch_fs *c, bool gc) -{ - struct btree_trans trans; - struct btree_iter iter; - struct bkey_s_c k; - size_t i, idx = 0; - int ret = 0; - - bch2_trans_init(&trans, c, 0, 0); - bch2_trans_iter_init(&trans, &iter, BTREE_ID_stripes, POS(0, U64_MAX), 0); - - k = bch2_btree_iter_prev(&iter); - ret = bkey_err(k); - if (!ret && k.k) - idx = k.k->p.offset + 1; - - bch2_trans_iter_exit(&trans, &iter); - bch2_trans_exit(&trans); - if (ret) - return ret; - - if (!idx) - return 0; - - if (!gc && - !init_heap(&c->ec_stripes_heap, roundup_pow_of_two(idx), - GFP_KERNEL)) - return -ENOMEM; -#if 0 - ret = genradix_prealloc(&c->stripes[gc], idx, GFP_KERNEL); -#else - for (i = 0; i < idx; i++) - if (!gc - ? !genradix_ptr_alloc(&c->stripes, i, GFP_KERNEL) - : !genradix_ptr_alloc(&c->gc_stripes, i, GFP_KERNEL)) - return -ENOMEM; -#endif - return 0; -} - void bch2_stripes_heap_to_text(struct printbuf *out, struct bch_fs *c) { ec_stripes_heap *h = &c->ec_stripes_heap; diff --git a/libbcachefs/ec.h b/libbcachefs/ec.h index 4681410..78d468c 100644 --- a/libbcachefs/ec.h +++ b/libbcachefs/ec.h @@ -217,8 +217,6 @@ void bch2_stripes_heap_start(struct bch_fs *); int bch2_stripes_read(struct bch_fs *); -int bch2_ec_mem_alloc(struct bch_fs *, bool); - void bch2_stripes_heap_to_text(struct printbuf *, struct bch_fs *); void bch2_new_stripes_to_text(struct printbuf *, struct bch_fs *); diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c index 99b2a77..3a7c146 100644 --- a/libbcachefs/inode.c +++ b/libbcachefs/inode.c @@ -4,6 +4,7 @@ #include "btree_key_cache.h" #include "bkey_methods.h" #include "btree_update.h" +#include "buckets.h" #include "error.h" #include "extents.h" #include "extent_update.h" @@ -584,59 +585,49 @@ found_slot: static int bch2_inode_delete_keys(struct btree_trans *trans, subvol_inum inum, enum btree_id id) { - u64 offset = 0; + struct btree_iter iter; + struct bkey_s_c k; + struct bkey_i delete; + u32 snapshot; int ret = 0; - while (!ret || ret == -EINTR) { - struct btree_iter iter; - struct bkey_s_c k; - struct bkey_i delete; - u32 snapshot; + /* + * We're never going to be deleting extents, no need to use an extent + * iterator: + */ + bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0), + BTREE_ITER_NOT_EXTENTS| + BTREE_ITER_INTENT); + while (1) { bch2_trans_begin(trans); ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); if (ret) - continue; + goto err; - bch2_trans_iter_init(trans, &iter, id, - SPOS(inum.inum, offset, snapshot), - BTREE_ITER_INTENT); - k = bch2_btree_iter_peek(&iter); - - if (!k.k || iter.pos.inode != inum.inum) { - bch2_trans_iter_exit(trans, &iter); - break; - } + bch2_btree_iter_set_snapshot(&iter, snapshot); + k = bch2_btree_iter_peek(&iter); ret = bkey_err(k); if (ret) goto err; + if (!k.k || iter.pos.inode != inum.inum) + break; + bkey_init(&delete.k); delete.k.p = iter.pos; - if (btree_node_type_is_extents(iter.btree_id)) { - unsigned max_sectors = - min_t(u64, U64_MAX - iter.pos.offset, - KEY_SIZE_MAX & (~0 << trans->c->block_bits)); - - /* create the biggest key we can */ - bch2_key_resize(&delete.k, max_sectors); - - ret = bch2_extent_trim_atomic(trans, &iter, &delete); - if (ret) - goto err; - } - ret = bch2_trans_update(trans, &iter, &delete, 0) ?: bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL); err: - offset = iter.pos.offset; - bch2_trans_iter_exit(trans, &iter); + if (ret && ret != -EINTR) + break; } + bch2_trans_iter_exit(trans, &iter); return ret; } diff --git a/libbcachefs/io.c b/libbcachefs/io.c index 1b954df..50b90b7 100644 --- a/libbcachefs/io.c +++ b/libbcachefs/io.c @@ -665,11 +665,7 @@ static void init_append_extent(struct bch_write_op *op, { struct bch_fs *c = op->c; struct bkey_i_extent *e; - struct open_bucket *ob; - unsigned i; - BUG_ON(crc.compressed_size > wp->sectors_free); - wp->sectors_free -= crc.compressed_size; op->pos.offset += crc.uncompressed_size; e = bkey_extent_init(op->insert_keys.top); @@ -682,22 +678,8 @@ static void init_append_extent(struct bch_write_op *op, crc.nonce) bch2_extent_crc_append(&e->k_i, crc); - open_bucket_for_each(c, &wp->ptrs, ob, i) { - struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev); - union bch_extent_entry *end = - bkey_val_end(bkey_i_to_s(&e->k_i)); - - end->ptr = ob->ptr; - end->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr; - end->ptr.cached = !ca->mi.durability || - (op->flags & BCH_WRITE_CACHED) != 0; - end->ptr.offset += ca->mi.bucket_size - ob->sectors_free; - - e->k.u64s++; - - BUG_ON(crc.compressed_size > ob->sectors_free); - ob->sectors_free -= crc.compressed_size; - } + bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, crc.compressed_size, + op->flags & BCH_WRITE_CACHED); bch2_keylist_push(&op->insert_keys); } diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c index 40e7cb6..158df42 100644 --- a/libbcachefs/journal.c +++ b/libbcachefs/journal.c @@ -705,6 +705,44 @@ int bch2_journal_flush(struct journal *j) return bch2_journal_flush_seq(j, seq); } +/* + * bch2_journal_noflush_seq - tell the journal not to issue any flushes before + * @seq + */ +bool bch2_journal_noflush_seq(struct journal *j, u64 seq) +{ + struct bch_fs *c = container_of(j, struct bch_fs, journal); + u64 unwritten_seq; + bool ret = false; + + if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush))) + return false; + + if (seq <= c->journal.flushed_seq_ondisk) + return false; + + spin_lock(&j->lock); + if (seq <= c->journal.flushed_seq_ondisk) + goto out; + + for (unwritten_seq = last_unwritten_seq(j); + unwritten_seq < seq; + unwritten_seq++) { + struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq); + + /* journal write is already in flight, and was a flush write: */ + if (unwritten_seq == last_unwritten_seq(j) && !buf->noflush) + goto out; + + buf->noflush = true; + } + + ret = true; +out: + spin_unlock(&j->lock); + return ret; +} + /* block/unlock the journal: */ void bch2_journal_unblock(struct journal *j) @@ -775,11 +813,8 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr, long b; if (new_fs) { - if (c) - percpu_down_read(&c->mark_lock); b = bch2_bucket_alloc_new_fs(ca); if (b < 0) { - percpu_up_read(&c->mark_lock); ret = -ENOSPC; goto err; } @@ -793,7 +828,7 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr, goto err; } - b = sector_to_bucket(ca, ob->ptr.offset); + b = ob->bucket; } if (c) @@ -827,14 +862,7 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr, if (c) spin_unlock(&c->journal.lock); - if (new_fs) { - bch2_mark_metadata_bucket(c, ca, b, BCH_DATA_journal, - ca->mi.bucket_size, - gc_phase(GC_PHASE_SB), - 0); - if (c) - percpu_up_read(&c->mark_lock); - } else { + if (!new_fs) { ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL, bch2_trans_mark_metadata_bucket(&trans, ca, b, BCH_DATA_journal, @@ -1000,11 +1028,14 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq, j->replay_journal_seq = last_seq; j->replay_journal_seq_end = cur_seq; j->last_seq_ondisk = last_seq; - j->flushed_seq_ondisk = last_seq; + j->flushed_seq_ondisk = cur_seq - 1; j->pin.front = last_seq; j->pin.back = cur_seq; atomic64_set(&j->seq, cur_seq - 1); + if (list_empty(journal_entries)) + j->last_empty_seq = cur_seq - 1; + fifo_for_each_entry_ptr(p, &j->pin, seq) journal_pin_list_init(p, 1); diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h index c39cbbf..b298873 100644 --- a/libbcachefs/journal.h +++ b/libbcachefs/journal.h @@ -477,6 +477,7 @@ void bch2_journal_flush_async(struct journal *, struct closure *); int bch2_journal_flush_seq(struct journal *, u64); int bch2_journal_flush(struct journal *); +bool bch2_journal_noflush_seq(struct journal *, u64); int bch2_journal_meta(struct journal *); void bch2_journal_halt(struct journal *); diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c index e161e86..77201a0 100644 --- a/libbcachefs/journal_io.c +++ b/libbcachefs/journal_io.c @@ -1399,9 +1399,10 @@ void bch2_journal_write(struct closure *cl) spin_lock(&j->lock); if (c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush) && - !w->must_flush && - (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && - test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)) { + (w->noflush || + (!w->must_flush && + (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) && + test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags)))) { w->noflush = true; SET_JSET_NO_FLUSH(jset, true); jset->last_seq = 0; diff --git a/libbcachefs/journal_seq_blacklist.c b/libbcachefs/journal_seq_blacklist.c index 79bc0e4..10bd23e 100644 --- a/libbcachefs/journal_seq_blacklist.c +++ b/libbcachefs/journal_seq_blacklist.c @@ -235,81 +235,3 @@ const struct bch_sb_field_ops bch_sb_field_ops_journal_seq_blacklist = { .validate = bch2_sb_journal_seq_blacklist_validate, .to_text = bch2_sb_journal_seq_blacklist_to_text }; - -void bch2_blacklist_entries_gc(struct work_struct *work) -{ - struct bch_fs *c = container_of(work, struct bch_fs, - journal_seq_blacklist_gc_work); - struct journal_seq_blacklist_table *t; - struct bch_sb_field_journal_seq_blacklist *bl; - struct journal_seq_blacklist_entry *src, *dst; - struct btree_trans trans; - unsigned i, nr, new_nr; - int ret; - - bch2_trans_init(&trans, c, 0, 0); - - for (i = 0; i < BTREE_ID_NR; i++) { - struct btree_iter iter; - struct btree *b; - - bch2_trans_node_iter_init(&trans, &iter, i, POS_MIN, - 0, 0, BTREE_ITER_PREFETCH); -retry: - bch2_trans_begin(&trans); - - b = bch2_btree_iter_peek_node(&iter); - - while (!(ret = PTR_ERR_OR_ZERO(b)) && - b && - !test_bit(BCH_FS_STOPPING, &c->flags)) - b = bch2_btree_iter_next_node(&iter); - - if (ret == -EINTR) - goto retry; - - bch2_trans_iter_exit(&trans, &iter); - } - - bch2_trans_exit(&trans); - if (ret) - return; - - mutex_lock(&c->sb_lock); - bl = bch2_sb_get_journal_seq_blacklist(c->disk_sb.sb); - if (!bl) - goto out; - - nr = blacklist_nr_entries(bl); - dst = bl->start; - - t = c->journal_seq_blacklist_table; - BUG_ON(nr != t->nr); - - for (src = bl->start, i = eytzinger0_first(t->nr); - src < bl->start + nr; - src++, i = eytzinger0_next(i, nr)) { - BUG_ON(t->entries[i].start != le64_to_cpu(src->start)); - BUG_ON(t->entries[i].end != le64_to_cpu(src->end)); - - if (t->entries[i].dirty) - *dst++ = *src; - } - - new_nr = dst - bl->start; - - bch_info(c, "nr blacklist entries was %u, now %u", nr, new_nr); - - if (new_nr != nr) { - bl = bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, - new_nr ? sb_blacklist_u64s(new_nr) : 0); - BUG_ON(new_nr && !bl); - - if (!new_nr) - c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_journal_seq_blacklist_v3)); - - bch2_write_super(c); - } -out: - mutex_unlock(&c->sb_lock); -} diff --git a/libbcachefs/journal_seq_blacklist.h b/libbcachefs/journal_seq_blacklist.h index afb886e..b4f876a 100644 --- a/libbcachefs/journal_seq_blacklist.h +++ b/libbcachefs/journal_seq_blacklist.h @@ -17,6 +17,4 @@ int bch2_blacklist_table_initialize(struct bch_fs *); extern const struct bch_sb_field_ops bch_sb_field_ops_journal_seq_blacklist; -void bch2_blacklist_entries_gc(struct work_struct *); - #endif /* _BCACHEFS_JOURNAL_SEQ_BLACKLIST_H */ diff --git a/libbcachefs/opts.c b/libbcachefs/opts.c index 9b75c85..d9ca69f 100644 --- a/libbcachefs/opts.c +++ b/libbcachefs/opts.c @@ -66,11 +66,6 @@ const char * const bch2_data_types[] = { NULL }; -const char * const bch2_cache_replacement_policies[] = { - BCH_CACHE_REPLACEMENT_POLICIES() - NULL -}; - const char * const bch2_member_states[] = { BCH_MEMBER_STATES() NULL diff --git a/libbcachefs/opts.h b/libbcachefs/opts.h index aadd395..661eb57 100644 --- a/libbcachefs/opts.h +++ b/libbcachefs/opts.h @@ -19,7 +19,6 @@ extern const char * const bch2_compression_opts[]; extern const char * const bch2_str_hash_types[]; extern const char * const bch2_str_hash_opts[]; extern const char * const bch2_data_types[]; -extern const char * const bch2_cache_replacement_policies[]; extern const char * const bch2_member_states[]; extern const char * const bch2_d_types[]; diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c index 29fe626..8b0e468 100644 --- a/libbcachefs/recovery.c +++ b/libbcachefs/recovery.c @@ -115,21 +115,12 @@ int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id, struct journal_key n = { .btree_id = id, .level = level, - .k = k, .allocated = true }; struct journal_keys *keys = &c->journal_keys; struct journal_iter *iter; unsigned idx = journal_key_search(keys, id, level, k->k.p); - if (idx < keys->nr && - journal_key_cmp(&n, &keys->d[idx]) == 0) { - if (keys->d[idx].allocated) - kfree(keys->d[idx].k); - keys->d[idx] = n; - return 0; - } - if (keys->nr == keys->size) { struct journal_keys new_keys = { .nr = keys->nr, @@ -149,10 +140,23 @@ int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id, *keys = new_keys; } - array_insert_item(keys->d, keys->nr, idx, n); + n.k = kmalloc(bkey_bytes(&k->k), GFP_KERNEL); + if (!n.k) + return -ENOMEM; + + bkey_copy(n.k, k); + + if (idx < keys->nr && + journal_key_cmp(&n, &keys->d[idx]) == 0) { + if (keys->d[idx].allocated) + kfree(keys->d[idx].k); + keys->d[idx] = n; + } else { + array_insert_item(keys->d, keys->nr, idx, n); - list_for_each_entry(iter, &c->journal_iters, list) - journal_iter_fix(c, iter, idx); + list_for_each_entry(iter, &c->journal_iters, list) + journal_iter_fix(c, iter, idx); + } return 0; } @@ -160,22 +164,12 @@ int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id, int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id, unsigned level, struct bpos pos) { - struct bkey_i *whiteout = - kmalloc(sizeof(struct bkey), GFP_KERNEL); - int ret; - - if (!whiteout) { - bch_err(c, "%s: error allocating new key", __func__); - return -ENOMEM; - } + struct bkey_i whiteout; - bkey_init(&whiteout->k); - whiteout->k.p = pos; + bkey_init(&whiteout.k); + whiteout.k.p = pos; - ret = bch2_journal_key_insert(c, id, level, whiteout); - if (ret) - kfree(whiteout); - return ret; + return bch2_journal_key_insert(c, id, level, &whiteout); } static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter) @@ -1149,16 +1143,6 @@ use_clean: if (ret) goto err; - /* - * After an unclean shutdown, skip then next few journal sequence - * numbers as they may have been referenced by btree writes that - * happened before their corresponding journal writes - those btree - * writes need to be ignored, by skipping and blacklisting the next few - * journal sequence numbers: - */ - if (!c->sb.clean) - journal_seq += 8; - if (blacklist_seq != journal_seq) { ret = bch2_journal_seq_blacklist_add(c, blacklist_seq, journal_seq); @@ -1295,24 +1279,15 @@ use_clean: bch_verbose(c, "quotas done"); } - if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) || - !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done))) { - struct bch_move_stats stats; - - bch_move_stats_init(&stats, "recovery"); - - bch_info(c, "scanning for old btree nodes"); - ret = bch2_fs_read_write(c); - if (ret) - goto err; - - ret = bch2_scan_old_btree_nodes(c, &stats); - if (ret) - goto err; - bch_info(c, "scanning for old btree nodes done"); - } - mutex_lock(&c->sb_lock); + /* + * With journal replay done, we can clear the journal seq blacklist + * table: + */ + BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)); + if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written) + bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0); + if (c->opts.version_upgrade) { c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current); c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL); @@ -1336,9 +1311,23 @@ use_clean: bch2_write_super(c); mutex_unlock(&c->sb_lock); - if (c->journal_seq_blacklist_table && - c->journal_seq_blacklist_table->nr > 128) - queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work); + if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) || + !(c->sb.compat & (1ULL << BCH_COMPAT_bformat_overflow_done)) || + le16_to_cpu(c->sb.version_min) < bcachefs_metadata_version_btree_ptr_sectors_written) { + struct bch_move_stats stats; + + bch_move_stats_init(&stats, "recovery"); + + bch_info(c, "scanning for old btree nodes"); + ret = bch2_fs_read_write(c); + if (ret) + goto err; + + ret = bch2_scan_old_btree_nodes(c, &stats); + if (ret) + goto err; + bch_info(c, "scanning for old btree nodes done"); + } ret = 0; out: @@ -1383,9 +1372,6 @@ int bch2_fs_initialize(struct bch_fs *c) c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL); bch2_write_super(c); } - - for_each_online_member(ca, c, i) - bch2_mark_dev_superblock(c, ca, 0); mutex_unlock(&c->sb_lock); set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags); @@ -1429,6 +1415,8 @@ int bch2_fs_initialize(struct bch_fs *c) percpu_ref_put(&ca->ref); goto err; } + + ca->new_fs_bucket_idx = 0; } err = "error creating root snapshot node"; diff --git a/libbcachefs/super-io.h b/libbcachefs/super-io.h index b64ac2f..5c26487 100644 --- a/libbcachefs/super-io.h +++ b/libbcachefs/super-io.h @@ -110,7 +110,6 @@ static inline struct bch_member_cpu bch2_mi_to_cpu(struct bch_member *mi) .bucket_size = le16_to_cpu(mi->bucket_size), .group = BCH_MEMBER_GROUP(mi), .state = BCH_MEMBER_STATE(mi), - .replacement = BCH_MEMBER_REPLACEMENT(mi), .discard = BCH_MEMBER_DISCARD(mi), .data_allowed = BCH_MEMBER_DATA_ALLOWED(mi), .durability = BCH_MEMBER_DURABILITY(mi) diff --git a/libbcachefs/super.c b/libbcachefs/super.c index 58bc290..df6bffe 100644 --- a/libbcachefs/super.c +++ b/libbcachefs/super.c @@ -528,8 +528,6 @@ void __bch2_fs_stop(struct bch_fs *c) set_bit(BCH_FS_STOPPING, &c->flags); - cancel_work_sync(&c->journal_seq_blacklist_gc_work); - down_write(&c->state_lock); bch2_fs_read_only(c); up_write(&c->state_lock); @@ -692,9 +690,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) spin_lock_init(&c->btree_write_error_lock); - INIT_WORK(&c->journal_seq_blacklist_gc_work, - bch2_blacklist_entries_gc); - INIT_LIST_HEAD(&c->journal_entries); INIT_LIST_HEAD(&c->journal_iters); @@ -1600,8 +1595,6 @@ int bch2_dev_add(struct bch_fs *c, const char *path) struct bch_dev *ca = NULL; struct bch_sb_field_members *mi; struct bch_member dev_mi; - struct bucket_array *buckets; - struct bucket *g; unsigned dev_idx, nr_devices, u64s; int ret; @@ -1631,20 +1624,6 @@ int bch2_dev_add(struct bch_fs *c, const char *path) return ret; } - /* - * We want to allocate journal on the new device before adding the new - * device to the filesystem because allocating after we attach requires - * spinning up the allocator thread, and the allocator thread requires - * doing btree writes, which if the existing devices are RO isn't going - * to work - * - * So we have to mark where the superblocks are, but marking allocated - * data normally updates the filesystem usage too, so we have to mark, - * allocate the journal, reset all the marks, then remark after we - * attach... - */ - bch2_mark_dev_superblock(NULL, ca, 0); - err = "journal alloc failed"; ret = bch2_dev_journal_alloc(ca); if (ret) @@ -1705,21 +1684,13 @@ have_slot: bch2_dev_usage_journal_reserve(c); - /* - * Clear marks before marking transactionally in the btree, so that - * per-device accounting gets done correctly: - */ - down_read(&ca->bucket_lock); - buckets = bucket_array(ca); - for_each_bucket(g, buckets) - atomic64_set(&g->_mark.v, 0); - up_read(&ca->bucket_lock); - err = "error marking superblock"; ret = bch2_trans_mark_dev_sb(c, ca); if (ret) goto err_late; + ca->new_fs_bucket_idx = 0; + if (ca->mi.state == BCH_MEMBER_STATE_rw) { ret = __bch2_dev_read_write(c, ca); if (ret) diff --git a/libbcachefs/super.h b/libbcachefs/super.h index 739e8fd..c3273e9 100644 --- a/libbcachefs/super.h +++ b/libbcachefs/super.h @@ -194,6 +194,27 @@ static inline struct bch_devs_mask bch2_online_devs(struct bch_fs *c) return devs; } +static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b) +{ + struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; + u64 b_offset = bucket_to_sector(ca, b); + u64 b_end = bucket_to_sector(ca, b + 1); + unsigned i; + + if (!b) + return true; + + for (i = 0; i < layout->nr_superblocks; i++) { + u64 offset = le64_to_cpu(layout->sb_offset[i]); + u64 end = offset + (1 << layout->sb_max_size_bits); + + if (!(offset >= b_end || end <= b_offset)) + return true; + } + + return false; +} + struct bch_fs *bch2_dev_to_fs(dev_t); struct bch_fs *bch2_uuid_to_fs(uuid_le); diff --git a/libbcachefs/super_types.h b/libbcachefs/super_types.h index 96023f3..d8b159a 100644 --- a/libbcachefs/super_types.h +++ b/libbcachefs/super_types.h @@ -29,7 +29,6 @@ struct bch_member_cpu { u16 bucket_size; /* sectors */ u16 group; u8 state; - u8 replacement; u8 discard; u8 data_allowed; u8 durability; diff --git a/libbcachefs/sysfs.c b/libbcachefs/sysfs.c index 0a0798b..6d15963 100644 --- a/libbcachefs/sysfs.c +++ b/libbcachefs/sysfs.c @@ -10,6 +10,7 @@ #include "bcachefs.h" #include "alloc_background.h" +#include "alloc_foreground.h" #include "sysfs.h" #include "btree_cache.h" #include "btree_io.h" @@ -131,7 +132,6 @@ do { \ return strtoi_h(buf, &var) ?: (ssize_t) size; \ } while (0) -write_attribute(trigger_journal_flush); write_attribute(trigger_gc); write_attribute(prune_cache); rw_attribute(btree_gc_periodic); @@ -177,7 +177,6 @@ read_attribute(extent_migrate_done); read_attribute(extent_migrate_raced); rw_attribute(discard); -rw_attribute(cache_replacement_policy); rw_attribute(label); rw_attribute(copy_gc_enabled); @@ -267,8 +266,12 @@ static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c struct btree_trans trans; struct btree_iter iter; struct bkey_s_c k; - u64 nr_uncompressed_extents = 0, uncompressed_sectors = 0, + enum btree_id id; + u64 nr_uncompressed_extents = 0, nr_compressed_extents = 0, + nr_incompressible_extents = 0, + uncompressed_sectors = 0, + incompressible_sectors = 0, compressed_sectors_compressed = 0, compressed_sectors_uncompressed = 0; int ret; @@ -278,47 +281,72 @@ static int bch2_compression_stats_to_text(struct printbuf *out, struct bch_fs *c bch2_trans_init(&trans, c, 0, 0); - for_each_btree_key(&trans, iter, BTREE_ID_extents, POS_MIN, 0, k, ret) - if (k.k->type == KEY_TYPE_extent) { - struct bkey_s_c_extent e = bkey_s_c_to_extent(k); + for (id = 0; id < BTREE_ID_NR; id++) { + if (!((1U << id) & BTREE_ID_HAS_PTRS)) + continue; + + for_each_btree_key(&trans, iter, id, POS_MIN, + BTREE_ITER_ALL_SNAPSHOTS, k, ret) { + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); const union bch_extent_entry *entry; struct extent_ptr_decoded p; - - extent_for_each_ptr_decode(e, p, entry) { - if (!crc_is_compressed(p.crc)) { - nr_uncompressed_extents++; - uncompressed_sectors += e.k->size; - } else { - nr_compressed_extents++; + bool compressed = false, uncompressed = false, incompressible = false; + + bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { + switch (p.crc.compression_type) { + case BCH_COMPRESSION_TYPE_none: + uncompressed = true; + uncompressed_sectors += k.k->size; + break; + case BCH_COMPRESSION_TYPE_incompressible: + incompressible = true; + incompressible_sectors += k.k->size; + break; + default: compressed_sectors_compressed += p.crc.compressed_size; compressed_sectors_uncompressed += p.crc.uncompressed_size; + compressed = true; + break; } - - /* only looking at the first ptr */ - break; } + + if (incompressible) + nr_incompressible_extents++; + else if (uncompressed) + nr_uncompressed_extents++; + else if (compressed) + nr_compressed_extents++; } - bch2_trans_iter_exit(&trans, &iter); + bch2_trans_iter_exit(&trans, &iter); + } bch2_trans_exit(&trans); + if (ret) return ret; - pr_buf(out, - "uncompressed data:\n" - " nr extents: %llu\n" - " size (bytes): %llu\n" - "compressed data:\n" - " nr extents: %llu\n" - " compressed size (bytes): %llu\n" - " uncompressed size (bytes): %llu\n", - nr_uncompressed_extents, - uncompressed_sectors << 9, - nr_compressed_extents, - compressed_sectors_compressed << 9, - compressed_sectors_uncompressed << 9); + pr_buf(out, "uncompressed:\n"); + pr_buf(out, " nr extents: %llu\n", nr_uncompressed_extents); + pr_buf(out, " size: "); + bch2_hprint(out, uncompressed_sectors << 9); + pr_buf(out, "\n"); + + pr_buf(out, "compressed:\n"); + pr_buf(out, " nr extents: %llu\n", nr_compressed_extents); + pr_buf(out, " compressed size: "); + bch2_hprint(out, compressed_sectors_compressed << 9); + pr_buf(out, "\n"); + pr_buf(out, " uncompressed size: "); + bch2_hprint(out, compressed_sectors_uncompressed << 9); + pr_buf(out, "\n"); + + pr_buf(out, "incompressible:\n"); + pr_buf(out, " nr extents: %llu\n", nr_incompressible_extents); + pr_buf(out, " size: "); + bch2_hprint(out, incompressible_sectors << 9); + pr_buf(out, "\n"); return 0; } @@ -483,9 +511,6 @@ STORE(bch2_fs) /* Debugging: */ - if (attr == &sysfs_trigger_journal_flush) - bch2_journal_meta(&c->journal); - if (attr == &sysfs_trigger_gc) { /* * Full gc is currently incompatible with btree key cache: @@ -575,7 +600,6 @@ struct attribute *bch2_fs_internal_files[] = { &sysfs_io_timers_read, &sysfs_io_timers_write, - &sysfs_trigger_journal_flush, &sysfs_trigger_gc, &sysfs_prune_cache, @@ -729,7 +753,7 @@ static void dev_alloc_debug_to_text(struct printbuf *out, struct bch_dev *ca) memset(nr, 0, sizeof(nr)); for (i = 0; i < ARRAY_SIZE(c->open_buckets); i++) - nr[c->open_buckets[i].type]++; + nr[c->open_buckets[i].data_type]++; pr_buf(out, "\t\t buckets\t sectors fragmented\n" @@ -826,14 +850,6 @@ SHOW(bch2_dev) return out.pos - buf; } - if (attr == &sysfs_cache_replacement_policy) { - bch2_string_opt_to_text(&out, - bch2_cache_replacement_policies, - ca->mi.replacement); - pr_buf(&out, "\n"); - return out.pos - buf; - } - if (attr == &sysfs_state_rw) { bch2_string_opt_to_text(&out, bch2_member_states, ca->mi.state); @@ -893,22 +909,6 @@ STORE(bch2_dev) mutex_unlock(&c->sb_lock); } - if (attr == &sysfs_cache_replacement_policy) { - ssize_t v = __sysfs_match_string(bch2_cache_replacement_policies, -1, buf); - - if (v < 0) - return v; - - mutex_lock(&c->sb_lock); - mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx]; - - if ((unsigned) v != BCH_MEMBER_REPLACEMENT(mi)) { - SET_BCH_MEMBER_REPLACEMENT(mi, v); - bch2_write_super(c); - } - mutex_unlock(&c->sb_lock); - } - if (attr == &sysfs_label) { char *tmp; int ret; @@ -939,7 +939,6 @@ struct attribute *bch2_dev_files[] = { /* settings: */ &sysfs_discard, - &sysfs_cache_replacement_policy, &sysfs_state_rw, &sysfs_label, diff --git a/libbcachefs/tests.c b/libbcachefs/tests.c index 478c00a..60ccb94 100644 --- a/libbcachefs/tests.c +++ b/libbcachefs/tests.c @@ -618,7 +618,6 @@ static int rand_mixed(struct bch_fs *c, u64 nr) static int __do_delete(struct btree_trans *trans, struct bpos pos) { struct btree_iter iter; - struct bkey_i delete; struct bkey_s_c k; int ret = 0; @@ -632,10 +631,7 @@ static int __do_delete(struct btree_trans *trans, struct bpos pos) if (!k.k) goto err; - bkey_init(&delete.k); - delete.k.p = k.k->p; - - ret = bch2_trans_update(trans, &iter, &delete, 0); + ret = bch2_btree_delete_at(trans, &iter, 0); err: bch2_trans_iter_exit(trans, &iter); return ret; @@ -650,7 +646,7 @@ static int rand_delete(struct bch_fs *c, u64 nr) bch2_trans_init(&trans, c, 0, 0); for (i = 0; i < nr; i++) { - struct bpos pos = POS(0, test_rand()); + struct bpos pos = SPOS(0, test_rand(), U32_MAX); ret = __bch2_trans_do(&trans, NULL, NULL, 0, __do_delete(&trans, pos)); -- 2.39.5