X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fbuckets.c;h=9f2ecff55c50c70220eb1adb20bbb4aab505e792;hb=ec28e6bfa269daf953af95610c0d1af3e7df1c0a;hp=f0c3bbc76894999179574657a5671758eb679ceb;hpb=b422ff58ba8eedcfef3b67b66468660f07b0cfc1;p=bcachefs-tools-debian diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c index f0c3bbc..9f2ecff 100644 --- a/libbcachefs/buckets.c +++ b/libbcachefs/buckets.c @@ -3,76 +3,24 @@ * Code for manipulating bucket marks for garbage collection. * * Copyright 2014 Datera, Inc. - * - * Bucket states: - * - free bucket: mark == 0 - * The bucket contains no data and will not be read - * - * - allocator bucket: owned_by_allocator == 1 - * The bucket is on a free list, or it is an open bucket - * - * - cached bucket: owned_by_allocator == 0 && - * dirty_sectors == 0 && - * cached_sectors > 0 - * The bucket contains data but may be safely discarded as there are - * enough replicas of the data on other cache devices, or it has been - * written back to the backing device - * - * - dirty bucket: owned_by_allocator == 0 && - * dirty_sectors > 0 - * The bucket contains data that we must not discard (either only copy, - * or one of the 'main copies' for data requiring multiple replicas) - * - * - metadata bucket: owned_by_allocator == 0 && is_metadata == 1 - * This is a btree node, journal or gen/prio bucket - * - * Lifecycle: - * - * bucket invalidated => bucket on freelist => open bucket => - * [dirty bucket =>] cached bucket => bucket invalidated => ... - * - * Note that cache promotion can skip the dirty bucket step, as data - * is copied from a deeper tier to a shallower tier, onto a cached - * bucket. - * Note also that a cached bucket can spontaneously become dirty -- - * see below. - * - * Only a traversal of the key space can determine whether a bucket is - * truly dirty or cached. - * - * Transitions: - * - * - free => allocator: bucket was invalidated - * - cached => allocator: bucket was invalidated - * - * - allocator => dirty: open bucket was filled up - * - allocator => cached: open bucket was filled up - * - allocator => metadata: metadata was allocated - * - * - dirty => cached: dirty sectors were copied to a deeper tier - * - dirty => free: dirty sectors were overwritten or moved (copy gc) - * - cached => free: cached sectors were overwritten - * - * - metadata => free: metadata was freed - * - * Oddities: - * - cached => dirty: a device was removed so formerly replicated data - * is no longer sufficiently replicated - * - free => cached: cannot happen - * - free => dirty: cannot happen - * - free => metadata: cannot happen */ #include "bcachefs.h" #include "alloc_background.h" +#include "backpointers.h" #include "bset.h" #include "btree_gc.h" #include "btree_update.h" #include "buckets.h" +#include "buckets_waiting_for_journal.h" #include "ec.h" #include "error.h" +#include "inode.h" #include "movinggc.h" +#include "recovery.h" +#include "reflink.h" #include "replicas.h" +#include "subvolume.h" #include #include @@ -97,43 +45,6 @@ static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage, } } -/* - * Clear journal_seq_valid for buckets for which it's not needed, to prevent - * wraparound: - */ -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; - struct bch_dev *ca; - struct bucket_array *buckets; - struct bucket *g; - struct bucket_mark m; - unsigned i; - - if (journal_seq - c->last_bucket_seq_cleanup < - (1U << (BUCKET_JOURNAL_SEQ_BITS - 2))) - return; - - c->last_bucket_seq_cleanup = journal_seq; - - for_each_member_device(ca, c, i) { - down_read(&ca->bucket_lock); - buckets = bucket_array(ca); - - for_each_bucket(g, buckets) { - bucket_cmpxchg(g, m, ({ - if (!m.journal_seq_valid || - bucket_needs_journal_commit(m, last_seq_ondisk)) - break; - - m.journal_seq_valid = 0; - })); - } - up_read(&ca->bucket_lock); - } -} - void bch2_fs_usage_initialize(struct bch_fs *c) { struct bch_fs_usage *usage; @@ -171,31 +82,33 @@ static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca, unsigned journal_seq, bool gc) { + BUG_ON(!gc && !journal_seq); + return this_cpu_ptr(gc ? ca->usage_gc : ca->usage[journal_seq & JOURNAL_BUF_MASK]); } -struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca) +void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage) { struct bch_fs *c = ca->fs; - struct bch_dev_usage ret; unsigned seq, i, u64s = dev_usage_u64s(); do { seq = read_seqcount_begin(&c->usage_lock); - memcpy(&ret, ca->usage_base, u64s * sizeof(u64)); + memcpy(usage, ca->usage_base, u64s * sizeof(u64)); for (i = 0; i < ARRAY_SIZE(ca->usage); i++) - acc_u64s_percpu((u64 *) &ret, (u64 __percpu *) ca->usage[i], u64s); + acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s); } while (read_seqcount_retry(&c->usage_lock, seq)); - - return ret; } static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c, unsigned journal_seq, bool gc) { + percpu_rwsem_assert_held(&c->mark_lock); + BUG_ON(!gc && !journal_seq); + return this_cpu_ptr(gc ? c->usage_gc : c->usage[journal_seq & JOURNAL_BUF_MASK]); @@ -224,23 +137,28 @@ u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v) struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c) { struct bch_fs_usage_online *ret; - unsigned seq, i, u64s; + unsigned seq, i, v, u64s = fs_usage_u64s(c) + 1; +retry: + ret = kmalloc(u64s * sizeof(u64), GFP_NOFS); + if (unlikely(!ret)) + return NULL; percpu_down_read(&c->mark_lock); - ret = kmalloc(sizeof(struct bch_fs_usage_online) + - sizeof(u64) + c->replicas.nr, GFP_NOFS); - if (unlikely(!ret)) { + v = fs_usage_u64s(c) + 1; + if (unlikely(u64s != v)) { + u64s = v; percpu_up_read(&c->mark_lock); - return NULL; + kfree(ret); + goto retry; } ret->online_reserved = percpu_u64_get(c->online_reserved); - u64s = fs_usage_u64s(c); do { seq = read_seqcount_begin(&c->usage_lock); - memcpy(&ret->u, c->usage_base, u64s * sizeof(u64)); + unsafe_memcpy(&ret->u, c->usage_base, u64s * sizeof(u64), + "embedded variable length struct"); for (i = 0; i < ARRAY_SIZE(c->usage); i++) acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i], u64s); } while (read_seqcount_retry(&c->usage_lock, seq)); @@ -282,26 +200,26 @@ void bch2_fs_usage_to_text(struct printbuf *out, { unsigned i; - pr_buf(out, "capacity:\t\t\t%llu\n", c->capacity); + prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity); - pr_buf(out, "hidden:\t\t\t\t%llu\n", + prt_printf(out, "hidden:\t\t\t\t%llu\n", fs_usage->u.hidden); - pr_buf(out, "data:\t\t\t\t%llu\n", + prt_printf(out, "data:\t\t\t\t%llu\n", fs_usage->u.data); - pr_buf(out, "cached:\t\t\t\t%llu\n", + prt_printf(out, "cached:\t\t\t\t%llu\n", fs_usage->u.cached); - pr_buf(out, "reserved:\t\t\t%llu\n", + prt_printf(out, "reserved:\t\t\t%llu\n", fs_usage->u.reserved); - pr_buf(out, "nr_inodes:\t\t\t%llu\n", + prt_printf(out, "nr_inodes:\t\t\t%llu\n", fs_usage->u.nr_inodes); - pr_buf(out, "online reserved:\t\t%llu\n", + prt_printf(out, "online reserved:\t\t%llu\n", fs_usage->online_reserved); for (i = 0; i < ARRAY_SIZE(fs_usage->u.persistent_reserved); i++) { - pr_buf(out, "%u replicas:\n", i + 1); - pr_buf(out, "\treserved:\t\t%llu\n", + prt_printf(out, "%u replicas:\n", i + 1); + prt_printf(out, "\treserved:\t\t%llu\n", fs_usage->u.persistent_reserved[i]); } @@ -309,24 +227,17 @@ void bch2_fs_usage_to_text(struct printbuf *out, struct bch_replicas_entry *e = cpu_replicas_entry(&c->replicas, i); - pr_buf(out, "\t"); + prt_printf(out, "\t"); bch2_replicas_entry_to_text(out, e); - pr_buf(out, ":\t%llu\n", fs_usage->u.replicas[i]); + prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]); } } -#define RESERVE_FACTOR 6 - static u64 reserve_factor(u64 r) { return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR); } -static u64 avail_factor(u64 r) -{ - return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1); -} - u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage) { return min(fs_usage->u.hidden + @@ -371,111 +282,143 @@ bch2_fs_usage_read_short(struct bch_fs *c) return ret; } -static inline int is_unavailable_bucket(struct bucket_mark m) +void bch2_dev_usage_init(struct bch_dev *ca) { - return !is_available_bucket(m); + ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket; } static inline int bucket_sectors_fragmented(struct bch_dev *ca, - struct bucket_mark m) + struct bch_alloc_v4 a) { - return bucket_sectors_used(m) - ? max(0, (int) ca->mi.bucket_size - (int) bucket_sectors_used(m)) + return a.dirty_sectors + ? max(0, (int) ca->mi.bucket_size - (int) a.dirty_sectors) : 0; } -static inline int is_stripe_data_bucket(struct bucket_mark m) -{ - return m.stripe && m.data_type != BCH_DATA_parity; -} - -static inline enum bch_data_type bucket_type(struct bucket_mark m) -{ - return m.cached_sectors && !m.dirty_sectors - ? BCH_DATA_cached - : 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, - int nr, s64 size) -{ - if (type == BCH_DATA_sb || type == BCH_DATA_journal) - fs_usage->hidden += size; - - dev_usage->d[type].buckets += nr; -} - static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca, - struct bch_fs_usage *fs_usage, - struct bucket_mark old, struct bucket_mark new, + struct bch_alloc_v4 old, + struct bch_alloc_v4 new, u64 journal_seq, bool gc) { + struct bch_fs_usage *fs_usage; struct bch_dev_usage *u; - percpu_rwsem_assert_held(&c->mark_lock); - preempt_disable(); - if (!fs_usage) - fs_usage = fs_usage_ptr(c, journal_seq, gc); - u = dev_usage_ptr(ca, journal_seq, gc); + fs_usage = fs_usage_ptr(c, journal_seq, gc); - if (bucket_type(old)) - account_bucket(fs_usage, u, bucket_type(old), - -1, -ca->mi.bucket_size); + if (data_type_is_hidden(old.data_type)) + fs_usage->hidden -= ca->mi.bucket_size; + if (data_type_is_hidden(new.data_type)) + fs_usage->hidden += ca->mi.bucket_size; + + u = dev_usage_ptr(ca, journal_seq, gc); - if (bucket_type(new)) - account_bucket(fs_usage, u, bucket_type(new), - 1, ca->mi.bucket_size); + u->d[old.data_type].buckets--; + u->d[new.data_type].buckets++; - u->buckets_ec += (int) new.stripe - (int) old.stripe; - u->buckets_unavailable += - is_unavailable_bucket(new) - is_unavailable_bucket(old); + u->buckets_ec -= (int) !!old.stripe; + u->buckets_ec += (int) !!new.stripe; u->d[old.data_type].sectors -= old.dirty_sectors; u->d[new.data_type].sectors += new.dirty_sectors; - u->d[BCH_DATA_cached].sectors += - (int) new.cached_sectors - (int) old.cached_sectors; + + u->d[BCH_DATA_cached].sectors += new.cached_sectors; + u->d[BCH_DATA_cached].sectors -= old.cached_sectors; u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old); u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new); preempt_enable(); +} + +static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca, + struct bucket old, struct bucket new, + u64 journal_seq, bool gc) +{ + struct bch_alloc_v4 old_a = { + .gen = old.gen, + .data_type = old.data_type, + .dirty_sectors = old.dirty_sectors, + .cached_sectors = old.cached_sectors, + .stripe = old.stripe, + }; + struct bch_alloc_v4 new_a = { + .gen = new.gen, + .data_type = new.data_type, + .dirty_sectors = new.dirty_sectors, + .cached_sectors = new.cached_sectors, + .stripe = new.stripe, + }; - if (!is_available_bucket(old) && is_available_bucket(new)) - bch2_wake_allocator(ca); + bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc); } -static inline void update_replicas(struct bch_fs *c, - struct bch_fs_usage *fs_usage, - struct bch_replicas_entry *r, - s64 sectors) +static inline int __update_replicas(struct bch_fs *c, + struct bch_fs_usage *fs_usage, + struct bch_replicas_entry *r, + s64 sectors) { int idx = bch2_replicas_entry_idx(c, r); - BUG_ON(idx < 0); + if (idx < 0) + return -1; + + fs_usage_data_type_to_base(fs_usage, r->data_type, sectors); + fs_usage->replicas[idx] += sectors; + return 0; +} + +static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k, + struct bch_replicas_entry *r, s64 sectors, + unsigned journal_seq, bool gc) +{ + struct bch_fs_usage __percpu *fs_usage; + int idx, ret = 0; + struct printbuf buf = PRINTBUF; + + percpu_down_read(&c->mark_lock); + buf.atomic++; + + idx = bch2_replicas_entry_idx(c, r); + if (idx < 0 && + fsck_err(c, "no replicas entry\n" + " while marking %s", + (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { + percpu_up_read(&c->mark_lock); + ret = bch2_mark_replicas(c, r); + percpu_down_read(&c->mark_lock); + + if (ret) + goto err; + idx = bch2_replicas_entry_idx(c, r); + } + if (idx < 0) { + ret = -1; + goto err; + } + preempt_disable(); + fs_usage = fs_usage_ptr(c, journal_seq, gc); fs_usage_data_type_to_base(fs_usage, r->data_type, sectors); fs_usage->replicas[idx] += sectors; + preempt_enable(); +err: +fsck_err: + percpu_up_read(&c->mark_lock); + printbuf_exit(&buf); + return ret; } -static inline void update_cached_sectors(struct bch_fs *c, - struct bch_fs_usage *fs_usage, - unsigned dev, s64 sectors) +static inline int update_cached_sectors(struct bch_fs *c, + struct bkey_s_c k, + unsigned dev, s64 sectors, + unsigned journal_seq, bool gc) { struct bch_replicas_padded r; bch2_replicas_entry_cached(&r.e, dev); - update_replicas(c, fs_usage, &r.e, sectors); + return update_replicas(c, k, &r.e, sectors, journal_seq, gc); } static struct replicas_delta_list * @@ -483,10 +426,26 @@ replicas_deltas_realloc(struct btree_trans *trans, unsigned more) { struct replicas_delta_list *d = trans->fs_usage_deltas; unsigned new_size = d ? (d->size + more) * 2 : 128; + unsigned alloc_size = sizeof(*d) + new_size; + + WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX); if (!d || d->used + more > d->size) { - d = krealloc(d, sizeof(*d) + new_size, GFP_NOIO|__GFP_ZERO); - BUG_ON(!d); + d = krealloc(d, alloc_size, GFP_NOIO|__GFP_ZERO); + + BUG_ON(!d && alloc_size > REPLICAS_DELTA_LIST_MAX); + + if (!d) { + d = mempool_alloc(&trans->c->replicas_delta_pool, GFP_NOIO); + memset(d, 0, REPLICAS_DELTA_LIST_MAX); + + if (trans->fs_usage_deltas) + memcpy(d, trans->fs_usage_deltas, + trans->fs_usage_deltas->size + sizeof(*d)); + + new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d); + kfree(trans->fs_usage_deltas); + } d->size = new_size; trans->fs_usage_deltas = d; @@ -510,7 +469,8 @@ static inline void update_replicas_list(struct btree_trans *trans, n = (void *) d->d + d->used; n->delta = sectors; - memcpy(&n->r, r, replicas_entry_bytes(r)); + memcpy((void *) n + offsetof(struct replicas_delta, r), + r, replicas_entry_bytes(r)); bch2_replicas_entry_sort(&n->r); d->used += b; } @@ -525,63 +485,19 @@ 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; \ -}) - -static int __bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t b, bool owned_by_allocator, - bool gc) -{ - struct bucket *g = __bucket(ca, b, gc); - struct bucket_mark old, new; - - old = bucket_cmpxchg(g, new, ({ - new.owned_by_allocator = owned_by_allocator; - })); - - BUG_ON(!gc && - !owned_by_allocator && !old.owned_by_allocator); - - return 0; -} - -void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t b, bool owned_by_allocator, - struct gc_pos pos, unsigned flags) -{ - preempt_disable(); - - do_mark_fn(__bch2_mark_alloc_bucket, c, pos, flags, - ca, b, owned_by_allocator); - - preempt_enable(); -} - -static int bch2_mark_alloc(struct bch_fs *c, - struct bkey_s_c old, struct bkey_s_c new, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +int bch2_mark_alloc(struct btree_trans *trans, + enum btree_id btree, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) { bool gc = flags & BTREE_TRIGGER_GC; - struct bkey_alloc_unpacked u; + u64 journal_seq = trans->journal_res.seq; + u64 bucket_journal_seq; + struct bch_fs *c = trans->c; + struct bch_alloc_v4 old_a_convert, new_a_convert; + const struct bch_alloc_v4 *old_a, *new_a; struct bch_dev *ca; - struct bucket *g; - struct bucket_mark old_m, m; - - /* We don't do anything for deletions - do we?: */ - if (new.k->type != KEY_TYPE_alloc && - new.k->type != KEY_TYPE_alloc_v2) - return 0; + int ret = 0; /* * alloc btree is read in by bch2_alloc_read, not gc: @@ -590,35 +506,73 @@ static int bch2_mark_alloc(struct bch_fs *c, !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE)) return 0; + if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans, + "alloc key for invalid device or bucket")) + return -EIO; + ca = bch_dev_bkey_exists(c, new.k->p.inode); - if (new.k->p.offset >= ca->mi.nbuckets) - return 0; + old_a = bch2_alloc_to_v4(old, &old_a_convert); + new_a = bch2_alloc_to_v4(new, &new_a_convert); + + bucket_journal_seq = new_a->journal_seq; - g = __bucket(ca, new.k->p.offset, gc); - u = bch2_alloc_unpack(new); + if ((flags & BTREE_TRIGGER_INSERT) && + data_type_is_empty(old_a->data_type) != + data_type_is_empty(new_a->data_type) && + new.k->type == KEY_TYPE_alloc_v4) { + struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v; - 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; + EBUG_ON(!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 = bucket_journal_seq = + data_type_is_empty(new_a->data_type) && + (journal_seq == v->journal_seq || + bch2_journal_noflush_seq(&c->journal, v->journal_seq)) + ? 0 : journal_seq; + } - if (journal_seq) { - m.journal_seq_valid = 1; - m.journal_seq = journal_seq; + if (!data_type_is_empty(old_a->data_type) && + data_type_is_empty(new_a->data_type) && + bucket_journal_seq) { + ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, + c->journal.flushed_seq_ondisk, + new.k->p.inode, new.k->p.offset, + bucket_journal_seq); + if (ret) { + bch2_fs_fatal_error(c, + "error setting bucket_needs_journal_commit: %i", ret); + return ret; } - })); + } + + percpu_down_read(&c->mark_lock); + if (!gc && new_a->gen != old_a->gen) + *bucket_gen(ca, new.k->p.offset) = new_a->gen; + + bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc); - bch2_dev_usage_update(c, ca, fs_usage, old_m, m, journal_seq, gc); + if (gc) { + struct bucket *g = gc_bucket(ca, new.k->p.offset); + + bucket_lock(g); - g->io_time[READ] = u.read_time; - g->io_time[WRITE] = u.write_time; - g->oldest_gen = u.oldest_gen; - g->gen_valid = 1; - g->stripe = u.stripe; - g->stripe_redundancy = u.stripe_redundancy; + g->gen_valid = 1; + g->gen = new_a->gen; + g->data_type = new_a->data_type; + g->stripe = new_a->stripe; + g->stripe_redundancy = new_a->stripe_redundancy; + g->dirty_sectors = new_a->dirty_sectors; + g->cached_sectors = new_a->cached_sectors; + + bucket_unlock(g); + } + percpu_up_read(&c->mark_lock); /* * need to know if we're getting called from the invalidate path or @@ -626,251 +580,263 @@ static int bch2_mark_alloc(struct bch_fs *c, */ if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) && - old_m.cached_sectors) { - update_cached_sectors(c, fs_usage, ca->dev_idx, - -old_m.cached_sectors); - trace_invalidate(ca, bucket_to_sector(ca, new.k->p.offset), - old_m.cached_sectors); + old_a->cached_sectors) { + ret = update_cached_sectors(c, new, ca->dev_idx, + -((s64) old_a->cached_sectors), + journal_seq, gc); + if (ret) { + bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors", + __func__); + return ret; + } } - return 0; -} - -#define checked_add(a, b) \ -({ \ - unsigned _res = (unsigned) (a) + (b); \ - bool overflow = _res > U16_MAX; \ - if (overflow) \ - _res = U16_MAX; \ - (a) = _res; \ - 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) -{ - struct bucket *g = __bucket(ca, b, gc); - struct bucket_mark old, new; - bool overflow; - - BUG_ON(data_type != BCH_DATA_sb && - data_type != BCH_DATA_journal); - - old = bucket_cmpxchg(g, new, ({ - new.data_type = data_type; - overflow = checked_add(new.dirty_sectors, sectors); - })); + if (new_a->data_type == BCH_DATA_free && + (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk)) + closure_wake_up(&c->freelist_wait); - bch2_fs_inconsistent_on(old.data_type && - old.data_type != data_type, c, - "different types of data in same bucket: %s, %s", - bch2_data_types[old.data_type], - bch2_data_types[data_type]); + if (new_a->data_type == BCH_DATA_need_discard && + (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk)) + bch2_do_discards(c); - bch2_fs_inconsistent_on(overflow, c, - "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > U16_MAX", - ca->dev_idx, b, new.gen, - bch2_data_types[old.data_type ?: data_type], - old.dirty_sectors, sectors); + if (old_a->data_type != BCH_DATA_cached && + new_a->data_type == BCH_DATA_cached && + should_invalidate_buckets(ca, bch2_dev_usage_read(ca))) + bch2_do_invalidates(c); - if (c) - bch2_dev_usage_update(c, ca, fs_usage_ptr(c, 0, gc), - old, new, 0, gc); + if (new_a->data_type == BCH_DATA_need_gc_gens) + bch2_do_gc_gens(c); 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) +int 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) { - BUG_ON(type != BCH_DATA_sb && - type != BCH_DATA_journal); - - preempt_disable(); - - 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); - } + struct bucket old, new, *g; + int ret = 0; - preempt_enable(); -} + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + BUG_ON(data_type != BCH_DATA_sb && + data_type != BCH_DATA_journal); -static s64 disk_sectors_scaled(unsigned n, unsigned d, unsigned sectors) -{ - return DIV_ROUND_UP(sectors * n, d); -} + /* + * Backup superblock might be past the end of our normal usable space: + */ + if (b >= ca->mi.nbuckets) + return 0; -static s64 __ptr_disk_sectors_delta(unsigned old_size, - unsigned offset, s64 delta, - unsigned flags, - unsigned n, unsigned d) -{ - BUG_ON(!n || !d); + percpu_down_read(&c->mark_lock); + g = gc_bucket(ca, b); - if (flags & BTREE_TRIGGER_OVERWRITE_SPLIT) { - BUG_ON(offset + -delta > old_size); + bucket_lock(g); + old = *g; - return -disk_sectors_scaled(n, d, old_size) + - disk_sectors_scaled(n, d, offset) + - disk_sectors_scaled(n, d, old_size - offset + delta); - } else if (flags & BTREE_TRIGGER_OVERWRITE) { - BUG_ON(offset + -delta > old_size); + if (bch2_fs_inconsistent_on(g->data_type && + g->data_type != data_type, c, + "different types of data in same bucket: %s, %s", + bch2_data_types[g->data_type], + bch2_data_types[data_type])) { + ret = -EIO; + goto err; + } - return -disk_sectors_scaled(n, d, old_size) + - disk_sectors_scaled(n, d, old_size + delta); - } else { - return disk_sectors_scaled(n, d, delta); + if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c, + "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size", + ca->dev_idx, b, g->gen, + bch2_data_types[g->data_type ?: data_type], + g->dirty_sectors, sectors)) { + ret = -EIO; + goto err; } -} -static s64 ptr_disk_sectors_delta(struct extent_ptr_decoded p, - unsigned offset, s64 delta, - unsigned flags) -{ - return __ptr_disk_sectors_delta(p.crc.live_size, - offset, delta, flags, - p.crc.compressed_size, - p.crc.uncompressed_size); + + g->data_type = data_type; + g->dirty_sectors += sectors; + new = *g; +err: + bucket_unlock(g); + if (!ret) + bch2_dev_usage_update_m(c, ca, old, new, 0, true); + percpu_up_read(&c->mark_lock); + return ret; } -static int check_bucket_ref(struct bch_fs *c, struct bkey_s_c k, +static int check_bucket_ref(struct btree_trans *trans, + struct bkey_s_c k, const struct bch_extent_ptr *ptr, s64 sectors, enum bch_data_type ptr_data_type, - u8 bucket_gen, u8 bucket_data_type, - u16 dirty_sectors, u16 cached_sectors) + u8 b_gen, u8 bucket_data_type, + u32 dirty_sectors, u32 cached_sectors) { - size_t bucket_nr = PTR_BUCKET_NR(bch_dev_bkey_exists(c, ptr->dev), ptr); + struct bch_fs *c = trans->c; + struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); + size_t bucket_nr = PTR_BUCKET_NR(ca, ptr); u16 bucket_sectors = !ptr->cached ? dirty_sectors : cached_sectors; - char buf[200]; + struct printbuf buf = PRINTBUF; + int ret = 0; - if (gen_after(ptr->gen, bucket_gen)) { + if (bucket_data_type == BCH_DATA_cached) + bucket_data_type = BCH_DATA_user; + + if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) || + (bucket_data_type == BCH_DATA_user && ptr_data_type == BCH_DATA_stripe)) + bucket_data_type = ptr_data_type = BCH_DATA_stripe; + + if (gen_after(ptr->gen, b_gen)) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n" "while marking %s", - ptr->dev, bucket_nr, bucket_gen, + ptr->dev, bucket_nr, b_gen, bch2_data_types[bucket_data_type ?: ptr_data_type], ptr->gen, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EIO; + (bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + goto err; } - if (gen_cmp(bucket_gen, ptr->gen) > BUCKET_GC_GEN_MAX) { + if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n" "while marking %s", - ptr->dev, bucket_nr, bucket_gen, + ptr->dev, bucket_nr, b_gen, bch2_data_types[bucket_data_type ?: ptr_data_type], ptr->gen, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EIO; + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + goto err; } - if (bucket_gen != ptr->gen && !ptr->cached) { + if (b_gen != ptr->gen && !ptr->cached) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "bucket %u:%zu gen %u data type %s: stale dirty ptr (gen %u)\n" + "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n" "while marking %s", - ptr->dev, bucket_nr, bucket_gen, + ptr->dev, bucket_nr, b_gen, + *bucket_gen(ca, bucket_nr), bch2_data_types[bucket_data_type ?: ptr_data_type], ptr->gen, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EIO; + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + goto err; } - if (bucket_gen != ptr->gen) - return 1; + if (b_gen != ptr->gen) { + ret = 1; + goto out; + } - if (bucket_data_type && ptr_data_type && + if (!data_type_is_empty(bucket_data_type) && + ptr_data_type && bucket_data_type != ptr_data_type) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n" "while marking %s", - ptr->dev, bucket_nr, bucket_gen, + ptr->dev, bucket_nr, b_gen, bch2_data_types[bucket_data_type], bch2_data_types[ptr_data_type], - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EIO; + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + goto err; } - if ((unsigned) (bucket_sectors + sectors) > U16_MAX) { + if ((unsigned) (bucket_sectors + sectors) > U32_MAX) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U16_MAX\n" "while marking %s", - ptr->dev, bucket_nr, bucket_gen, + ptr->dev, bucket_nr, b_gen, bch2_data_types[bucket_data_type ?: ptr_data_type], bucket_sectors, sectors, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EIO; + (printbuf_reset(&buf), + bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EIO; + goto err; } - - return 0; +out: + printbuf_exit(&buf); + return ret; +err: + bch2_dump_trans_updates(trans); + goto out; } -static int mark_stripe_bucket(struct bch_fs *c, struct bkey_s_c k, - unsigned ptr_idx, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +static int mark_stripe_bucket(struct btree_trans *trans, + struct bkey_s_c k, + unsigned ptr_idx, + unsigned flags) { + struct bch_fs *c = trans->c; + u64 journal_seq = trans->journal_res.seq; const struct bch_stripe *s = bkey_s_c_to_stripe(k).v; unsigned nr_data = s->nr_blocks - s->nr_redundant; bool parity = ptr_idx >= nr_data; + enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe; + 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 = PTR_BUCKET(ca, ptr, gc); - struct bucket_mark new, old; - char buf[200]; - int ret; + struct bucket old, new, *g; + struct printbuf buf = PRINTBUF; + int ret = 0; + + BUG_ON(!(flags & BTREE_TRIGGER_GC)); + + /* * XXX doesn't handle deletion */ + + percpu_down_read(&c->mark_lock); + buf.atomic++; + g = PTR_GC_BUCKET(ca, ptr); - if (g->stripe && g->stripe != k.k->p.offset) { + if (g->dirty_sectors || + (g->stripe && g->stripe != k.k->p.offset)) { bch2_fs_inconsistent(c, "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s", - ptr->dev, PTR_BUCKET_NR(ca, ptr), g->mark.gen, - (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); - return -EINVAL; + ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen, + (bch2_bkey_val_to_text(&buf, c, k), buf.buf)); + ret = -EINVAL; + goto err; } - old = bucket_cmpxchg(g, new, ({ - ret = check_bucket_ref(c, k, ptr, 0, 0, new.gen, new.data_type, - new.dirty_sectors, new.cached_sectors); - if (ret) - return ret; + bucket_lock(g); + old = *g; - if (parity) { - new.data_type = BCH_DATA_parity; - new.dirty_sectors = le16_to_cpu(s->sectors); - } + ret = check_bucket_ref(trans, k, ptr, sectors, data_type, + g->gen, g->data_type, + g->dirty_sectors, g->cached_sectors); + if (ret) + goto err; - if (journal_seq) { - new.journal_seq_valid = 1; - new.journal_seq = journal_seq; - } - })); + g->data_type = data_type; + g->dirty_sectors += sectors; g->stripe = k.k->p.offset; g->stripe_redundancy = s->nr_redundant; - - bch2_dev_usage_update(c, ca, fs_usage, old, new, journal_seq, gc); - return 0; + new = *g; +err: + bucket_unlock(g); + if (!ret) + bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true); + percpu_up_read(&c->mark_lock); + printbuf_exit(&buf); + return ret; } -static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k, +static int __mark_pointer(struct btree_trans *trans, + struct bkey_s_c k, const struct bch_extent_ptr *ptr, s64 sectors, enum bch_data_type ptr_data_type, u8 bucket_gen, u8 *bucket_data_type, - u16 *dirty_sectors, u16 *cached_sectors) + u32 *dirty_sectors, u32 *cached_sectors) { - u16 *dst_sectors = !ptr->cached + u32 *dst_sectors = !ptr->cached ? dirty_sectors : cached_sectors; - int ret = check_bucket_ref(c, k, ptr, sectors, ptr_data_type, + int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type, bucket_gen, *bucket_data_type, *dirty_sectors, *cached_sectors); @@ -883,141 +849,143 @@ static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k, return 0; } -static int bch2_mark_pointer(struct bch_fs *c, struct bkey_s_c k, +static int bch2_mark_pointer(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c k, struct extent_ptr_decoded p, - s64 sectors, enum bch_data_type data_type, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) + s64 sectors, + unsigned flags) { - bool gc = flags & BTREE_TRIGGER_GC; - struct bucket_mark old, new; + u64 journal_seq = trans->journal_res.seq; + struct bch_fs *c = trans->c; struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); - struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc); + struct bucket old, new, *g; + enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p); u8 bucket_data_type; - u64 v; - int ret; - - v = atomic64_read(&g->_mark.v); - do { - new.v.counter = old.v.counter = v; - bucket_data_type = new.data_type; - - ret = __mark_pointer(c, k, &p.ptr, sectors, data_type, new.gen, - &bucket_data_type, - &new.dirty_sectors, - &new.cached_sectors); - if (ret) - return ret; - - new.data_type = bucket_data_type; - - if (journal_seq) { - new.journal_seq_valid = 1; - new.journal_seq = journal_seq; - } - - if (flags & BTREE_TRIGGER_NOATOMIC) { - g->_mark = new; - break; - } - } while ((v = atomic64_cmpxchg(&g->_mark.v, - old.v.counter, - new.v.counter)) != old.v.counter); + int ret = 0; - bch2_dev_usage_update(c, ca, fs_usage, old, new, journal_seq, gc); + BUG_ON(!(flags & BTREE_TRIGGER_GC)); - BUG_ON(!gc && bucket_became_unavailable(old, new)); + percpu_down_read(&c->mark_lock); + g = PTR_GC_BUCKET(ca, &p.ptr); + bucket_lock(g); + old = *g; + + bucket_data_type = g->data_type; + ret = __mark_pointer(trans, k, &p.ptr, sectors, + data_type, g->gen, + &bucket_data_type, + &g->dirty_sectors, + &g->cached_sectors); + if (!ret) + g->data_type = bucket_data_type; + + new = *g; + bucket_unlock(g); + if (!ret) + bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true); + percpu_up_read(&c->mark_lock); - return 0; + return ret; } -static int bch2_mark_stripe_ptr(struct bch_fs *c, +static int bch2_mark_stripe_ptr(struct btree_trans *trans, + struct bkey_s_c k, struct bch_extent_stripe_ptr p, enum bch_data_type data_type, - struct bch_fs_usage *fs_usage, - s64 sectors, unsigned flags) + s64 sectors, + unsigned flags) { - bool gc = flags & BTREE_TRIGGER_GC; + struct bch_fs *c = trans->c; struct bch_replicas_padded r; - struct stripe *m; - unsigned i, blocks_nonempty = 0; + struct gc_stripe *m; - m = genradix_ptr(&c->stripes[gc], p.idx); + BUG_ON(!(flags & BTREE_TRIGGER_GC)); - spin_lock(&c->ec_stripes_heap_lock); + m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL); + if (!m) { + bch_err(c, "error allocating memory for gc_stripes, idx %llu", + (u64) p.idx); + return -BCH_ERR_ENOMEM_mark_stripe_ptr; + } + + mutex_lock(&c->ec_stripes_heap_lock); if (!m || !m->alive) { - spin_unlock(&c->ec_stripes_heap_lock); + mutex_unlock(&c->ec_stripes_heap_lock); 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; - - for (i = 0; i < m->nr_blocks; i++) - blocks_nonempty += m->block_sectors[i] != 0; - - if (m->blocks_nonempty != blocks_nonempty) { - m->blocks_nonempty = blocks_nonempty; - if (!gc) - bch2_stripes_heap_update(c, m, p.idx); - } - - spin_unlock(&c->ec_stripes_heap_lock); + mutex_unlock(&c->ec_stripes_heap_lock); r.e.data_type = data_type; - update_replicas(c, fs_usage, &r.e, sectors); + update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true); return 0; } -static int bch2_mark_extent(struct bch_fs *c, - struct bkey_s_c old, struct bkey_s_c new, - unsigned offset, s64 sectors, - enum bch_data_type data_type, - struct bch_fs_usage *fs_usage, - unsigned journal_seq, unsigned flags) +int bch2_mark_extent(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) { - struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old; + u64 journal_seq = trans->journal_res.seq; + struct bch_fs *c = trans->c; + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new; struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); const union bch_extent_entry *entry; struct extent_ptr_decoded p; struct bch_replicas_padded r; + enum bch_data_type data_type = bkey_is_btree_ptr(k.k) + ? BCH_DATA_btree + : BCH_DATA_user; + s64 sectors = bkey_is_btree_ptr(k.k) + ? btree_sectors(c) + : k.k->size; s64 dirty_sectors = 0; 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; - BUG_ON(!sectors); - bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { - s64 disk_sectors = data_type == BCH_DATA_btree - ? sectors - : ptr_disk_sectors_delta(p, offset, sectors, flags); + s64 disk_sectors = ptr_disk_sectors(sectors, p); - ret = bch2_mark_pointer(c, k, p, disk_sectors, data_type, - fs_usage, journal_seq, flags); + if (flags & BTREE_TRIGGER_OVERWRITE) + disk_sectors = -disk_sectors; + + ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags); if (ret < 0) return ret; stale = ret > 0; if (p.ptr.cached) { - if (!stale) - update_cached_sectors(c, fs_usage, p.ptr.dev, - disk_sectors); + if (!stale) { + ret = update_cached_sectors(c, k, p.ptr.dev, + disk_sectors, journal_seq, true); + if (ret) { + bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors", + __func__); + return ret; + } + } } else if (!p.has_ec) { dirty_sectors += disk_sectors; r.e.devs[r.e.nr_devs++] = p.ptr.dev; } else { - ret = bch2_mark_stripe_ptr(c, p.ec, data_type, - fs_usage, disk_sectors, flags); + ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type, + disk_sectors, flags); if (ret) return ret; @@ -1030,331 +998,341 @@ static int bch2_mark_extent(struct bch_fs *c, } } - if (r.e.nr_devs) - update_replicas(c, fs_usage, &r.e, dirty_sectors); + if (r.e.nr_devs) { + ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true); + if (ret) { + struct printbuf buf = PRINTBUF; + + bch2_bkey_val_to_text(&buf, c, k); + bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf); + printbuf_exit(&buf); + return ret; + } + } return 0; } -static int bch2_mark_stripe(struct bch_fs *c, - struct bkey_s_c old, struct bkey_s_c new, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +int bch2_mark_stripe(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) { bool gc = flags & BTREE_TRIGGER_GC; - size_t idx = new.k->p.offset; + u64 journal_seq = trans->journal_res.seq; + struct bch_fs *c = trans->c; + u64 idx = new.k->p.offset; const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe ? bkey_s_c_to_stripe(old).v : NULL; const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe ? bkey_s_c_to_stripe(new).v : NULL; - struct stripe *m = genradix_ptr(&c->stripes[gc], idx); unsigned i; int ret; BUG_ON(gc && old_s); - if (!m || (old_s && !m->alive)) { - bch_err_ratelimited(c, "error marking nonexistent stripe %zu", - idx); - return -1; - } + if (!gc) { + struct stripe *m = genradix_ptr(&c->stripes, idx); + + if (!m) { + struct printbuf buf1 = PRINTBUF; + struct printbuf buf2 = PRINTBUF; + + bch2_bkey_val_to_text(&buf1, c, old); + bch2_bkey_val_to_text(&buf2, c, new); + bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n" + "old %s\n" + "new %s", idx, buf1.buf, buf2.buf); + printbuf_exit(&buf2); + printbuf_exit(&buf1); + bch2_inconsistent_error(c); + return -1; + } - if (!new_s) { - spin_lock(&c->ec_stripes_heap_lock); - bch2_stripes_heap_del(c, m, idx); - spin_unlock(&c->ec_stripes_heap_lock); + if (!new_s) { + bch2_stripes_heap_del(c, m, idx); - memset(m, 0, sizeof(*m)); + memset(m, 0, sizeof(*m)); + } else { + m->sectors = le16_to_cpu(new_s->sectors); + m->algorithm = new_s->algorithm; + m->nr_blocks = new_s->nr_blocks; + m->nr_redundant = new_s->nr_redundant; + m->blocks_nonempty = 0; + + for (i = 0; i < new_s->nr_blocks; i++) + m->blocks_nonempty += !!stripe_blockcount_get(new_s, i); + + if (!old_s) + bch2_stripes_heap_insert(c, m, idx); + else + bch2_stripes_heap_update(c, m, idx); + } } else { + struct gc_stripe *m = + genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL); + + if (!m) { + bch_err(c, "error allocating memory for gc_stripes, idx %llu", + idx); + return -BCH_ERR_ENOMEM_mark_stripe; + } + /* + * This will be wrong when we bring back runtime gc: we should + * be unmarking the old key and then marking the new key + */ m->alive = true; m->sectors = le16_to_cpu(new_s->sectors); - m->algorithm = new_s->algorithm; m->nr_blocks = new_s->nr_blocks; m->nr_redundant = new_s->nr_redundant; - m->blocks_nonempty = 0; - - for (i = 0; i < new_s->nr_blocks; i++) { - m->block_sectors[i] = - stripe_blockcount_get(new_s, i); - m->blocks_nonempty += !!m->block_sectors[i]; + for (i = 0; i < new_s->nr_blocks; i++) m->ptrs[i] = new_s->ptrs[i]; - } bch2_bkey_to_replicas(&m->r.e, new); - if (!gc) { - spin_lock(&c->ec_stripes_heap_lock); - bch2_stripes_heap_update(c, m, idx); - spin_unlock(&c->ec_stripes_heap_lock); - } - } - - if (gc) { /* * gc recalculates this field from stripe ptr * references: */ memset(m->block_sectors, 0, sizeof(m->block_sectors)); - m->blocks_nonempty = 0; for (i = 0; i < new_s->nr_blocks; i++) { - ret = mark_stripe_bucket(c, new, i, fs_usage, - journal_seq, flags); + ret = mark_stripe_bucket(trans, new, i, flags); if (ret) return ret; } - update_replicas(c, fs_usage, &m->r.e, - ((s64) m->sectors * m->nr_redundant)); + ret = update_replicas(c, new, &m->r.e, + ((s64) m->sectors * m->nr_redundant), + journal_seq, gc); + if (ret) { + struct printbuf buf = PRINTBUF; + + bch2_bkey_val_to_text(&buf, c, new); + bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf); + printbuf_exit(&buf); + return ret; + } } return 0; } -static int bch2_mark_key_locked(struct bch_fs *c, - struct bkey_s_c old, - struct bkey_s_c new, - unsigned offset, s64 sectors, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +int bch2_mark_inode(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) { - struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old; - int ret = 0; + struct bch_fs *c = trans->c; + struct bch_fs_usage __percpu *fs_usage; + u64 journal_seq = trans->journal_res.seq; - BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE))); + if (flags & BTREE_TRIGGER_INSERT) { + struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v; - preempt_disable(); + BUG_ON(!journal_seq); + BUG_ON(new.k->type != KEY_TYPE_inode_v3); - if (!fs_usage || (flags & BTREE_TRIGGER_GC)) - fs_usage = fs_usage_ptr(c, journal_seq, - flags & BTREE_TRIGGER_GC); + v->bi_journal_seq = cpu_to_le64(journal_seq); + } - switch (k.k->type) { - case KEY_TYPE_alloc: - case KEY_TYPE_alloc_v2: - ret = bch2_mark_alloc(c, old, new, fs_usage, journal_seq, flags); - break; - case KEY_TYPE_btree_ptr: - case KEY_TYPE_btree_ptr_v2: - sectors = !(flags & BTREE_TRIGGER_OVERWRITE) - ? c->opts.btree_node_size - : -c->opts.btree_node_size; - - ret = bch2_mark_extent(c, old, new, offset, sectors, - BCH_DATA_btree, fs_usage, journal_seq, flags); - break; - case KEY_TYPE_extent: - case KEY_TYPE_reflink_v: - ret = bch2_mark_extent(c, old, new, offset, sectors, - BCH_DATA_user, fs_usage, journal_seq, flags); - break; - case KEY_TYPE_stripe: - ret = bch2_mark_stripe(c, old, new, fs_usage, journal_seq, flags); - break; - case KEY_TYPE_inode: - fs_usage->nr_inodes += new.k->type == KEY_TYPE_inode; - fs_usage->nr_inodes -= old.k->type == KEY_TYPE_inode; - break; - case KEY_TYPE_reservation: { - unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas; + if (flags & BTREE_TRIGGER_GC) { + percpu_down_read(&c->mark_lock); + preempt_disable(); - sectors *= replicas; - replicas = clamp_t(unsigned, replicas, 1, - ARRAY_SIZE(fs_usage->persistent_reserved)); + fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC); + fs_usage->nr_inodes += bkey_is_inode(new.k); + fs_usage->nr_inodes -= bkey_is_inode(old.k); - fs_usage->reserved += sectors; - fs_usage->persistent_reserved[replicas - 1] += sectors; - break; - } + preempt_enable(); + percpu_up_read(&c->mark_lock); } - - preempt_enable(); - - return ret; + return 0; } -int bch2_mark_key(struct bch_fs *c, struct bkey_s_c new, - unsigned offset, s64 sectors, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +int bch2_mark_reservation(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) { - struct bkey deleted; - struct bkey_s_c old = (struct bkey_s_c) { &deleted, NULL }; - int ret; + struct bch_fs *c = trans->c; + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new; + struct bch_fs_usage __percpu *fs_usage; + unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas; + s64 sectors = (s64) k.k->size; + + BUG_ON(!(flags & BTREE_TRIGGER_GC)); - bkey_init(&deleted); + if (flags & BTREE_TRIGGER_OVERWRITE) + sectors = -sectors; + sectors *= replicas; percpu_down_read(&c->mark_lock); - ret = bch2_mark_key_locked(c, old, new, offset, sectors, - fs_usage, journal_seq, - BTREE_TRIGGER_INSERT|flags); - percpu_up_read(&c->mark_lock); + preempt_disable(); - return ret; -} + fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC); + replicas = clamp_t(unsigned, replicas, 1, + ARRAY_SIZE(fs_usage->persistent_reserved)); -int bch2_mark_update(struct btree_trans *trans, - struct btree_iter *iter, - struct bkey_i *new, - struct bch_fs_usage *fs_usage, - unsigned flags) -{ - struct bch_fs *c = trans->c; - struct bkey_s_c old; - struct bkey unpacked; - int ret = 0; + fs_usage->reserved += sectors; + fs_usage->persistent_reserved[replicas - 1] += sectors; - if (unlikely(flags & BTREE_TRIGGER_NORUN)) - return 0; + preempt_enable(); + percpu_up_read(&c->mark_lock); - if (!btree_node_type_needs_gc(iter->btree_id)) - return 0; + return 0; +} - bkey_init(&unpacked); - old = (struct bkey_s_c) { &unpacked, NULL }; +static s64 __bch2_mark_reflink_p(struct btree_trans *trans, + struct bkey_s_c_reflink_p p, + u64 start, u64 end, + u64 *idx, unsigned flags, size_t r_idx) +{ + struct bch_fs *c = trans->c; + struct reflink_gc *r; + int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1; + u64 next_idx = end; + s64 ret = 0; + struct printbuf buf = PRINTBUF; - if (!btree_node_type_is_extents(iter->btree_id)) { - /* iterators should be uptodate, shouldn't get errors here: */ - if (btree_iter_type(iter) != BTREE_ITER_CACHED) { - old = bch2_btree_iter_peek_slot(iter); - BUG_ON(bkey_err(old)); - } else { - struct bkey_cached *ck = (void *) iter->l[0].b; + if (r_idx >= c->reflink_gc_nr) + goto not_found; - if (ck->valid) - old = bkey_i_to_s_c(ck->k); - } + r = genradix_ptr(&c->reflink_gc_table, r_idx); + next_idx = min(next_idx, r->offset - r->size); + if (*idx < next_idx) + goto not_found; - if (old.k->type == new->k.type) { - bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0, - fs_usage, trans->journal_res.seq, - BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags); + BUG_ON((s64) r->refcount + add < 0); - } else { - bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0, - fs_usage, trans->journal_res.seq, - BTREE_TRIGGER_INSERT|flags); - bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), 0, 0, - fs_usage, trans->journal_res.seq, - BTREE_TRIGGER_OVERWRITE|flags); - } - } else { - struct btree_iter *copy; - - BUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED); - bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), - 0, new->k.size, - fs_usage, trans->journal_res.seq, - BTREE_TRIGGER_INSERT|flags); + r->refcount += add; + *idx = r->offset; + return 0; +not_found: + if (fsck_err(c, "pointer to missing indirect extent\n" + " %s\n" + " missing range %llu-%llu", + (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf), + *idx, next_idx)) { + struct bkey_i_error *new; + + new = bch2_trans_kmalloc(trans, sizeof(*new)); + ret = PTR_ERR_OR_ZERO(new); + if (ret) + goto err; - copy = bch2_trans_copy_iter(trans, iter); + bkey_init(&new->k); + new->k.type = KEY_TYPE_error; + new->k.p = bkey_start_pos(p.k); + new->k.p.offset += *idx - start; + bch2_key_resize(&new->k, next_idx - *idx); + ret = __bch2_btree_insert(trans, BTREE_ID_extents, &new->k_i, + BTREE_TRIGGER_NORUN); + } - for_each_btree_key_continue(copy, 0, old, ret) { - unsigned offset = 0; - s64 sectors = -((s64) old.k->size); + *idx = next_idx; +err: +fsck_err: + printbuf_exit(&buf); + return ret; +} - flags |= BTREE_TRIGGER_OVERWRITE; +int bch2_mark_reflink_p(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_s_c new, + unsigned flags) +{ + struct bch_fs *c = trans->c; + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new; + struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k); + struct reflink_gc *ref; + size_t l, r, m; + u64 idx = le64_to_cpu(p.v->idx), start = idx; + u64 end = le64_to_cpu(p.v->idx) + p.k->size; + int ret = 0; - if (bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0) - break; + BUG_ON(!(flags & BTREE_TRIGGER_GC)); - switch (bch2_extent_overlap(&new->k, old.k)) { - case BCH_EXTENT_OVERLAP_ALL: - offset = 0; - sectors = -((s64) old.k->size); - break; - case BCH_EXTENT_OVERLAP_BACK: - offset = bkey_start_offset(&new->k) - - bkey_start_offset(old.k); - sectors = bkey_start_offset(&new->k) - - old.k->p.offset; - break; - case BCH_EXTENT_OVERLAP_FRONT: - offset = 0; - sectors = bkey_start_offset(old.k) - - new->k.p.offset; - break; - case BCH_EXTENT_OVERLAP_MIDDLE: - offset = bkey_start_offset(&new->k) - - bkey_start_offset(old.k); - sectors = -((s64) new->k.size); - flags |= BTREE_TRIGGER_OVERWRITE_SPLIT; - break; - } + 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); + } - BUG_ON(sectors >= 0); + l = 0; + r = c->reflink_gc_nr; + while (l < r) { + m = l + (r - l) / 2; - ret = bch2_mark_key_locked(c, old, bkey_i_to_s_c(new), - offset, sectors, fs_usage, - trans->journal_res.seq, flags) ?: 1; - if (ret <= 0) - break; - } - bch2_trans_iter_put(trans, copy); + ref = genradix_ptr(&c->reflink_gc_table, m); + if (ref->offset <= idx) + l = m + 1; + else + r = m; } + while (idx < end && !ret) + ret = __bch2_mark_reflink_p(trans, p, start, end, + &idx, flags, l++); + return ret; } -static noinline __cold -void fs_usage_apply_warn(struct btree_trans *trans, - unsigned disk_res_sectors) +void bch2_trans_fs_usage_revert(struct btree_trans *trans, + struct replicas_delta_list *deltas) { struct bch_fs *c = trans->c; - struct btree_insert_entry *i; - char buf[200]; - - bch_err(c, "disk usage increased more than %u sectors reserved", - disk_res_sectors); - - trans_for_each_update(trans, i) { - pr_err("while inserting"); - bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(i->k)); - pr_err("%s", buf); - pr_err("overlapping with"); - - if (btree_iter_type(i->iter) != BTREE_ITER_CACHED) { - struct btree_iter *copy = bch2_trans_copy_iter(trans, i->iter); - struct bkey_s_c k; - int ret; - - for_each_btree_key_continue(copy, 0, k, ret) { - if (btree_node_type_is_extents(i->iter->btree_id) - ? bkey_cmp(i->k->k.p, bkey_start_pos(k.k)) <= 0 - : bkey_cmp(i->k->k.p, k.k->p)) - break; - - bch2_bkey_val_to_text(&PBUF(buf), c, k); - pr_err("%s", buf); - } - bch2_trans_iter_put(trans, copy); - } else { - struct bkey_cached *ck = (void *) i->iter->l[0].b; + struct bch_fs_usage *dst; + struct replicas_delta *d, *top = (void *) deltas->d + deltas->used; + s64 added = 0; + unsigned i; - if (ck->valid) { - bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(ck->k)); - pr_err("%s", buf); - } + percpu_down_read(&c->mark_lock); + preempt_disable(); + dst = fs_usage_ptr(c, trans->journal_res.seq, false); + + /* revert changes: */ + for (d = deltas->d; d != top; d = replicas_delta_next(d)) { + switch (d->r.data_type) { + case BCH_DATA_btree: + case BCH_DATA_user: + case BCH_DATA_parity: + added += d->delta; } + BUG_ON(__update_replicas(c, dst, &d->r, -d->delta)); + } + + dst->nr_inodes -= deltas->nr_inodes; + + for (i = 0; i < BCH_REPLICAS_MAX; i++) { + added -= deltas->persistent_reserved[i]; + dst->reserved -= deltas->persistent_reserved[i]; + dst->persistent_reserved[i] -= deltas->persistent_reserved[i]; } + + if (added > 0) { + trans->disk_res->sectors += added; + this_cpu_add(*c->online_reserved, added); + } + + preempt_enable(); + percpu_up_read(&c->mark_lock); } -void bch2_trans_fs_usage_apply(struct btree_trans *trans, - struct replicas_delta_list *deltas) +int bch2_trans_fs_usage_apply(struct btree_trans *trans, + struct replicas_delta_list *deltas) { struct bch_fs *c = trans->c; static int warned_disk_usage = 0; bool warn = false; unsigned disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0; - struct replicas_delta *d = deltas->d; + struct replicas_delta *d = deltas->d, *d2; struct replicas_delta *top = (void *) deltas->d + deltas->used; struct bch_fs_usage *dst; s64 added = 0, should_not_have_added; unsigned i; - percpu_rwsem_assert_held(&c->mark_lock); - + percpu_down_read(&c->mark_lock); preempt_disable(); dst = fs_usage_ptr(c, trans->journal_res.seq, false); @@ -1366,7 +1344,8 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans, added += d->delta; } - update_replicas(c, dst, &d->r, d->delta); + if (__update_replicas(c, dst, &d->r, d->delta)) + goto need_mark; } dst->nr_inodes += deltas->nr_inodes; @@ -1383,7 +1362,14 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans, */ should_not_have_added = added - (s64) disk_res_sectors; if (unlikely(should_not_have_added > 0)) { - atomic64_sub(should_not_have_added, &c->sectors_available); + u64 old, new, v = atomic64_read(&c->sectors_available); + + do { + old = v; + new = max_t(s64, 0, old - should_not_have_added); + } while ((v = atomic64_cmpxchg(&c->sectors_available, + old, new)) != old); + added -= should_not_have_added; warn = true; } @@ -1394,124 +1380,62 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans, } preempt_enable(); + percpu_up_read(&c->mark_lock); if (unlikely(warn) && !xchg(&warned_disk_usage, 1)) - fs_usage_apply_warn(trans, disk_res_sectors); -} - -/* trans_mark: */ - -static struct btree_iter *trans_get_update(struct btree_trans *trans, - enum btree_id btree_id, struct bpos pos, - struct bkey_s_c *k) -{ - struct btree_insert_entry *i; - - trans_for_each_update(trans, i) - if (i->iter->btree_id == btree_id && - (btree_node_type_is_extents(btree_id) - ? bkey_cmp(pos, bkey_start_pos(&i->k->k)) >= 0 && - bkey_cmp(pos, i->k->k.p) < 0 - : !bkey_cmp(pos, i->iter->pos))) { - *k = bkey_i_to_s_c(i->k); - - /* ugly hack.. */ - BUG_ON(btree_iter_live(trans, i->iter)); - trans->iters_live |= 1ULL << i->iter->idx; - return i->iter; - } + bch2_trans_inconsistent(trans, + "disk usage increased %lli more than %u sectors reserved)", + should_not_have_added, disk_res_sectors); + return 0; +need_mark: + /* revert changes: */ + for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2)) + BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta)); - return NULL; + preempt_enable(); + percpu_up_read(&c->mark_lock); + return -1; } -static int trans_get_key(struct btree_trans *trans, - enum btree_id btree_id, struct bpos pos, - struct btree_iter **iter, - struct bkey_s_c *k) -{ - unsigned flags = btree_id != BTREE_ID_alloc - ? BTREE_ITER_SLOTS - : BTREE_ITER_CACHED; - int ret; - - *iter = trans_get_update(trans, btree_id, pos, k); - if (*iter) - return 1; - - *iter = bch2_trans_get_iter(trans, btree_id, pos, - flags|BTREE_ITER_INTENT); - *k = __bch2_btree_iter_peek(*iter, flags); - ret = bkey_err(*k); - if (ret) - bch2_trans_iter_put(trans, *iter); - return ret; -} +/* trans_mark: */ -static struct bkey_alloc_buf * -bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter **_iter, - const struct bch_extent_ptr *ptr, - struct bkey_alloc_unpacked *u) -{ - struct bch_fs *c = trans->c; - struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); - struct bpos pos = POS(ptr->dev, PTR_BUCKET_NR(ca, ptr)); - struct bucket *g; - struct btree_iter *iter; - struct bkey_s_c k; - struct bkey_alloc_buf *a; +static inline int bch2_trans_mark_pointer(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c k, struct extent_ptr_decoded p, + unsigned flags) +{ + bool insert = !(flags & BTREE_TRIGGER_OVERWRITE); + struct btree_iter iter; + struct bkey_i_alloc_v4 *a; + struct bpos bucket_pos; + struct bch_backpointer bp; + s64 sectors; int ret; - a = bch2_trans_kmalloc(trans, sizeof(struct bkey_alloc_buf)); - if (IS_ERR(a)) - return a; - - iter = trans_get_update(trans, BTREE_ID_alloc, pos, &k); - if (iter) { - *u = bch2_alloc_unpack(k); - } else { - iter = bch2_trans_get_iter(trans, BTREE_ID_alloc, pos, - BTREE_ITER_CACHED| - BTREE_ITER_CACHED_NOFILL| - BTREE_ITER_INTENT); - ret = bch2_btree_iter_traverse(iter); - if (ret) { - bch2_trans_iter_put(trans, iter); - return ERR_PTR(ret); - } - - percpu_down_read(&c->mark_lock); - g = bucket(ca, pos.offset); - *u = alloc_mem_to_key(iter, g, READ_ONCE(g->mark)); - percpu_up_read(&c->mark_lock); - } - - *_iter = iter; - return a; -} + bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket_pos, &bp); + sectors = bp.bucket_len; + if (!insert) + sectors = -sectors; -static int bch2_trans_mark_pointer(struct btree_trans *trans, - struct bkey_s_c k, struct extent_ptr_decoded p, - s64 sectors, enum bch_data_type data_type) -{ - struct bch_fs *c = trans->c; - struct btree_iter *iter; - struct bkey_alloc_unpacked u; - struct bkey_alloc_buf *a; - int ret; - - a = bch2_trans_start_alloc_update(trans, &iter, &p.ptr, &u); + a = bch2_trans_start_alloc_update(trans, &iter, bucket_pos); if (IS_ERR(a)) return PTR_ERR(a); - ret = __mark_pointer(c, k, &p.ptr, sectors, data_type, u.gen, &u.data_type, - &u.dirty_sectors, &u.cached_sectors); + ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type, + a->v.gen, &a->v.data_type, + &a->v.dirty_sectors, &a->v.cached_sectors); if (ret) - goto out; + goto err; - bch2_alloc_pack(c, a, u); - bch2_trans_update(trans, iter, &a->k, 0); -out: - bch2_trans_iter_put(trans, iter); + if (!p.ptr.cached) { + ret = bch2_bucket_backpointer_mod(trans, a, bp, k, insert); + if (ret) + goto err; + } + + ret = bch2_trans_update(trans, &iter, &a->k_i, 0); +err: + bch2_trans_iter_exit(trans, &iter); return ret; } @@ -1519,61 +1443,66 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans, struct extent_ptr_decoded p, s64 sectors, enum bch_data_type data_type) { - struct bch_fs *c = trans->c; - struct btree_iter *iter; - struct bkey_s_c k; + struct btree_iter iter; struct bkey_i_stripe *s; struct bch_replicas_padded r; int ret = 0; - ret = trans_get_key(trans, BTREE_ID_stripes, POS(0, p.ec.idx), &iter, &k); - if (ret < 0) - return ret; - - if (k.k->type != KEY_TYPE_stripe) { - bch2_fs_inconsistent(c, + bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes, POS(0, p.ec.idx), + BTREE_ITER_INTENT| + BTREE_ITER_WITH_UPDATES); + s = bch2_bkey_get_mut_typed(trans, &iter, stripe); + ret = PTR_ERR_OR_ZERO(s); + if (unlikely(ret)) { + bch2_trans_inconsistent_on(ret == -ENOENT, trans, "pointer to nonexistent stripe %llu", (u64) p.ec.idx); - ret = -EIO; - goto out; + goto err; } - if (!bch2_ptr_matches_stripe(bkey_s_c_to_stripe(k).v, p)) { - bch2_fs_inconsistent(c, + if (!bch2_ptr_matches_stripe(&s->v, p)) { + bch2_trans_inconsistent(trans, "stripe pointer doesn't match stripe %llu", (u64) p.ec.idx); ret = -EIO; - goto out; + goto err; } - s = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); - ret = PTR_ERR_OR_ZERO(s); - if (ret) - goto out; - - bkey_reassemble(&s->k_i, k); stripe_blockcount_set(&s->v, p.ec.block, stripe_blockcount_get(&s->v, p.ec.block) + sectors); - bch2_trans_update(trans, iter, &s->k_i, 0); + + ret = bch2_trans_update(trans, &iter, &s->k_i, 0); + if (ret) + goto err; bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i)); r.e.data_type = data_type; update_replicas_list(trans, &r.e, sectors); -out: - bch2_trans_iter_put(trans, iter); +err: + bch2_trans_iter_exit(trans, &iter); return ret; } -static int bch2_trans_mark_extent(struct btree_trans *trans, - struct bkey_s_c k, unsigned offset, - s64 sectors, unsigned flags, - enum bch_data_type data_type) +int bch2_trans_mark_extent(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_i *new, + unsigned flags) { + struct bch_fs *c = trans->c; + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE + ? old + : bkey_i_to_s_c(new); struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); const union bch_extent_entry *entry; struct extent_ptr_decoded p; struct bch_replicas_padded r; + enum bch_data_type data_type = bkey_is_btree_ptr(k.k) + ? BCH_DATA_btree + : BCH_DATA_user; + s64 sectors = bkey_is_btree_ptr(k.k) + ? btree_sectors(c) + : k.k->size; s64 dirty_sectors = 0; bool stale; int ret; @@ -1582,15 +1511,13 @@ static int bch2_trans_mark_extent(struct btree_trans *trans, r.e.nr_devs = 0; r.e.nr_required = 1; - BUG_ON(!sectors); - bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { - s64 disk_sectors = data_type == BCH_DATA_btree - ? sectors - : ptr_disk_sectors_delta(p, offset, sectors, flags); + s64 disk_sectors = ptr_disk_sectors(sectors, p); + + if (flags & BTREE_TRIGGER_OVERWRITE) + disk_sectors = -disk_sectors; - ret = bch2_trans_mark_pointer(trans, k, p, disk_sectors, - data_type); + ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags); if (ret < 0) return ret; @@ -1619,356 +1546,291 @@ static int bch2_trans_mark_extent(struct btree_trans *trans, return 0; } -static int bch2_trans_mark_stripe_alloc_ref(struct btree_trans *trans, - struct bkey_s_c_stripe s, - unsigned idx, bool deleting) +static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans, + struct bkey_s_c_stripe s, + unsigned idx, bool deleting) { struct bch_fs *c = trans->c; const struct bch_extent_ptr *ptr = &s.v->ptrs[idx]; - struct bkey_alloc_buf *a; - struct btree_iter *iter; - struct bkey_alloc_unpacked u; - bool parity = idx >= s.v->nr_blocks - s.v->nr_redundant; + struct btree_iter iter; + struct bkey_i_alloc_v4 *a; + enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant + ? BCH_DATA_parity : 0; + s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0; int ret = 0; - a = bch2_trans_start_alloc_update(trans, &iter, ptr, &u); + if (deleting) + sectors = -sectors; + + a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr)); if (IS_ERR(a)) return PTR_ERR(a); - if (parity) { - s64 sectors = le16_to_cpu(s.v->sectors); - - if (deleting) - sectors = -sectors; - - u.dirty_sectors += sectors; - u.data_type = u.dirty_sectors - ? BCH_DATA_parity - : 0; - } + ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type, + a->v.gen, a->v.data_type, + a->v.dirty_sectors, a->v.cached_sectors); + if (ret) + goto err; if (!deleting) { - if (bch2_fs_inconsistent_on(u.stripe && u.stripe != s.k->p.offset, c, - "bucket %llu:%llu gen %u: multiple stripes using same bucket (%u, %llu)", - iter->pos.inode, iter->pos.offset, u.gen, - u.stripe, s.k->p.offset)) { + if (bch2_trans_inconsistent_on(a->v.stripe || + a->v.stripe_redundancy, trans, + "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)", + iter.pos.inode, iter.pos.offset, a->v.gen, + bch2_data_types[a->v.data_type], + a->v.dirty_sectors, + a->v.stripe, s.k->p.offset)) { ret = -EIO; goto err; } - u.stripe = s.k->p.offset; - u.stripe_redundancy = s.v->nr_redundant; + if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans, + "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu", + iter.pos.inode, iter.pos.offset, a->v.gen, + bch2_data_types[a->v.data_type], + a->v.dirty_sectors, + s.k->p.offset)) { + ret = -EIO; + goto err; + } + + a->v.stripe = s.k->p.offset; + a->v.stripe_redundancy = s.v->nr_redundant; + a->v.data_type = BCH_DATA_stripe; } else { - u.stripe = 0; - u.stripe_redundancy = 0; + if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset || + a->v.stripe_redundancy != s.v->nr_redundant, trans, + "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)", + iter.pos.inode, iter.pos.offset, a->v.gen, + s.k->p.offset, a->v.stripe)) { + ret = -EIO; + goto err; + } + + a->v.stripe = 0; + a->v.stripe_redundancy = 0; + a->v.data_type = alloc_data_type(a->v, BCH_DATA_user); } - bch2_alloc_pack(c, a, u); - bch2_trans_update(trans, iter, &a->k, 0); + a->v.dirty_sectors += sectors; + if (data_type) + a->v.data_type = !deleting ? data_type : 0; + + ret = bch2_trans_update(trans, &iter, &a->k_i, 0); + if (ret) + goto err; err: - bch2_trans_iter_put(trans, iter); + bch2_trans_iter_exit(trans, &iter); return ret; } -static int bch2_trans_mark_stripe(struct btree_trans *trans, - struct bkey_s_c old, struct bkey_s_c new, - unsigned flags) +int bch2_trans_mark_stripe(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, struct bkey_i *new, + unsigned flags) { - struct bkey_s_c_stripe old_s = { NULL }; - struct bkey_s_c_stripe new_s = { NULL }; + const struct bch_stripe *old_s = NULL; + struct bch_stripe *new_s = NULL; struct bch_replicas_padded r; - unsigned i; + unsigned i, nr_blocks; int ret = 0; if (old.k->type == KEY_TYPE_stripe) - old_s = bkey_s_c_to_stripe(old); - if (new.k->type == KEY_TYPE_stripe) - new_s = bkey_s_c_to_stripe(new); + old_s = bkey_s_c_to_stripe(old).v; + if (new->k.type == KEY_TYPE_stripe) + new_s = &bkey_i_to_stripe(new)->v; /* * If the pointers aren't changing, we don't need to do anything: */ - if (new_s.k && old_s.k && - new_s.v->nr_blocks == old_s.v->nr_blocks && - new_s.v->nr_redundant == old_s.v->nr_redundant && - !memcmp(old_s.v->ptrs, new_s.v->ptrs, - new_s.v->nr_blocks * sizeof(struct bch_extent_ptr))) + if (new_s && old_s && + new_s->nr_blocks == old_s->nr_blocks && + new_s->nr_redundant == old_s->nr_redundant && + !memcmp(old_s->ptrs, new_s->ptrs, + new_s->nr_blocks * sizeof(struct bch_extent_ptr))) return 0; - if (new_s.k) { - s64 sectors = le16_to_cpu(new_s.v->sectors); + BUG_ON(new_s && old_s && + (new_s->nr_blocks != old_s->nr_blocks || + new_s->nr_redundant != old_s->nr_redundant)); - bch2_bkey_to_replicas(&r.e, new); - update_replicas_list(trans, &r.e, sectors * new_s.v->nr_redundant); + nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks; - for (i = 0; i < new_s.v->nr_blocks; i++) { - ret = bch2_trans_mark_stripe_alloc_ref(trans, new_s, - i, false); - if (ret) - return ret; - } + if (new_s) { + s64 sectors = le16_to_cpu(new_s->sectors); + + bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new)); + update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant); } - if (old_s.k) { - s64 sectors = -((s64) le16_to_cpu(old_s.v->sectors)); + if (old_s) { + s64 sectors = -((s64) le16_to_cpu(old_s->sectors)); bch2_bkey_to_replicas(&r.e, old); - update_replicas_list(trans, &r.e, sectors * old_s.v->nr_redundant); + update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant); + } + + for (i = 0; i < nr_blocks; i++) { + if (new_s && old_s && + !memcmp(&new_s->ptrs[i], + &old_s->ptrs[i], + sizeof(new_s->ptrs[i]))) + continue; - for (i = 0; i < old_s.v->nr_blocks; i++) { - ret = bch2_trans_mark_stripe_alloc_ref(trans, old_s, - i, true); + if (new_s) { + ret = bch2_trans_mark_stripe_bucket(trans, + bkey_i_to_s_c_stripe(new), i, false); if (ret) - return ret; + break; + } + + if (old_s) { + ret = bch2_trans_mark_stripe_bucket(trans, + bkey_s_c_to_stripe(old), i, true); + if (ret) + break; } } return ret; } -static __le64 *bkey_refcount(struct bkey_i *k) +int bch2_trans_mark_inode(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, + struct bkey_i *new, + unsigned flags) { - switch (k->k.type) { - case KEY_TYPE_reflink_v: - return &bkey_i_to_reflink_v(k)->v.refcount; - case KEY_TYPE_indirect_inline_data: - return &bkey_i_to_indirect_inline_data(k)->v.refcount; - default: - return NULL; + int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k); + + if (nr) { + struct replicas_delta_list *d = + replicas_deltas_realloc(trans, 0); + d->nr_inodes += nr; } + + return 0; +} + +int bch2_trans_mark_reservation(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, + struct bkey_i *new, + unsigned flags) +{ + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE + ? old + : bkey_i_to_s_c(new); + unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas; + s64 sectors = (s64) k.k->size; + struct replicas_delta_list *d; + + if (flags & BTREE_TRIGGER_OVERWRITE) + sectors = -sectors; + sectors *= replicas; + + d = replicas_deltas_realloc(trans, 0); + + replicas = clamp_t(unsigned, replicas, 1, + ARRAY_SIZE(d->persistent_reserved)); + + d->persistent_reserved[replicas - 1] += sectors; + return 0; } static int __bch2_trans_mark_reflink_p(struct btree_trans *trans, struct bkey_s_c_reflink_p p, - u64 idx, unsigned sectors, - unsigned flags) + u64 *idx, unsigned flags) { struct bch_fs *c = trans->c; - struct btree_iter *iter; - struct bkey_s_c k; - struct bkey_i *n; + struct btree_iter iter; + struct bkey_i *k; __le64 *refcount; - s64 ret; - - ret = trans_get_key(trans, BTREE_ID_reflink, - POS(0, idx), &iter, &k); - if (ret < 0) - return ret; - - if ((flags & BTREE_TRIGGER_OVERWRITE) && - (bkey_start_offset(k.k) < idx || - k.k->p.offset > idx + sectors)) - goto out; - - sectors = k.k->p.offset - idx; + int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1; + struct printbuf buf = PRINTBUF; + int ret; - n = bch2_trans_kmalloc(trans, bkey_bytes(k.k)); - ret = PTR_ERR_OR_ZERO(n); + bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, *idx), + BTREE_ITER_INTENT| + BTREE_ITER_WITH_UPDATES); + k = bch2_bkey_get_mut(trans, &iter); + ret = PTR_ERR_OR_ZERO(k); if (ret) goto err; - bkey_reassemble(n, k); - - refcount = bkey_refcount(n); + refcount = bkey_refcount(k); if (!refcount) { - bch2_fs_inconsistent(c, - "%llu:%llu len %u points to nonexistent indirect extent %llu", - p.k->p.inode, p.k->p.offset, p.k->size, idx); + bch2_bkey_val_to_text(&buf, c, p.s_c); + bch2_trans_inconsistent(trans, + "nonexistent indirect extent at %llu while marking\n %s", + *idx, buf.buf); ret = -EIO; goto err; } - le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1); - - if (!*refcount) { - n->k.type = KEY_TYPE_deleted; - set_bkey_val_u64s(&n->k, 0); + if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) { + bch2_bkey_val_to_text(&buf, c, p.s_c); + bch2_trans_inconsistent(trans, + "indirect extent refcount underflow at %llu while marking\n %s", + *idx, buf.buf); + ret = -EIO; + goto err; } - bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k)); - bch2_trans_update(trans, iter, n, 0); -out: - ret = sectors; -err: - bch2_trans_iter_put(trans, iter); - return ret; -} + if (flags & BTREE_TRIGGER_INSERT) { + struct bch_reflink_p *v = (struct bch_reflink_p *) p.v; + u64 pad; -static int bch2_trans_mark_reflink_p(struct btree_trans *trans, - struct bkey_s_c_reflink_p p, unsigned offset, - s64 sectors, unsigned flags) -{ - u64 idx = le64_to_cpu(p.v->idx) + offset; - s64 ret = 0; + pad = max_t(s64, le32_to_cpu(v->front_pad), + le64_to_cpu(v->idx) - bkey_start_offset(&k->k)); + BUG_ON(pad > U32_MAX); + v->front_pad = cpu_to_le32(pad); - sectors = abs(sectors); - BUG_ON(offset + sectors > p.k->size); + pad = max_t(s64, le32_to_cpu(v->back_pad), + k->k.p.offset - p.k->size - le64_to_cpu(v->idx)); + BUG_ON(pad > U32_MAX); + v->back_pad = cpu_to_le32(pad); + } - while (sectors) { - ret = __bch2_trans_mark_reflink_p(trans, p, idx, sectors, flags); - if (ret < 0) - break; + le64_add_cpu(refcount, add); - idx += ret; - sectors = max_t(s64, 0LL, sectors - ret); - ret = 0; - } + bch2_btree_iter_set_pos_to_extent_start(&iter); + ret = bch2_trans_update(trans, &iter, k, 0); + if (ret) + goto err; + *idx = k->k.p.offset; +err: + bch2_trans_iter_exit(trans, &iter); + printbuf_exit(&buf); return ret; } -int bch2_trans_mark_key(struct btree_trans *trans, - struct bkey_s_c old, - struct bkey_s_c new, - unsigned offset, s64 sectors, unsigned flags) +int bch2_trans_mark_reflink_p(struct btree_trans *trans, + enum btree_id btree_id, unsigned level, + struct bkey_s_c old, + struct bkey_i *new, + unsigned flags) { - struct bch_fs *c = trans->c; - struct bkey_s_c k = flags & BTREE_TRIGGER_INSERT ? new : old; - struct replicas_delta_list *d; - - BUG_ON(!(flags & (BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE))); - - switch (k.k->type) { - case KEY_TYPE_btree_ptr: - case KEY_TYPE_btree_ptr_v2: - sectors = !(flags & BTREE_TRIGGER_OVERWRITE) - ? c->opts.btree_node_size - : -c->opts.btree_node_size; - - return bch2_trans_mark_extent(trans, k, offset, sectors, - flags, BCH_DATA_btree); - case KEY_TYPE_extent: - case KEY_TYPE_reflink_v: - return bch2_trans_mark_extent(trans, k, offset, sectors, - flags, BCH_DATA_user); - case KEY_TYPE_stripe: - return bch2_trans_mark_stripe(trans, old, new, flags); - case KEY_TYPE_inode: { - int nr = (new.k->type == KEY_TYPE_inode) - - (old.k->type == KEY_TYPE_inode); - - if (nr) { - d = replicas_deltas_realloc(trans, 0); - d->nr_inodes += nr; - } - - return 0; - } - case KEY_TYPE_reservation: { - unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas; - - d = replicas_deltas_realloc(trans, 0); + struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE + ? old + : bkey_i_to_s_c(new); + struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k); + u64 idx, end_idx; + int ret = 0; - sectors *= replicas; - replicas = clamp_t(unsigned, replicas, 1, - ARRAY_SIZE(d->persistent_reserved)); + if (flags & BTREE_TRIGGER_INSERT) { + struct bch_reflink_p *v = (struct bch_reflink_p *) p.v; - d->persistent_reserved[replicas - 1] += sectors; - return 0; - } - case KEY_TYPE_reflink_p: - return bch2_trans_mark_reflink_p(trans, - bkey_s_c_to_reflink_p(k), - offset, sectors, flags); - default: - return 0; + v->front_pad = v->back_pad = 0; } -} - -int bch2_trans_mark_update(struct btree_trans *trans, - struct btree_iter *iter, - struct bkey_i *new, - unsigned flags) -{ - struct bkey_s_c old; - int ret; - - if (unlikely(flags & BTREE_TRIGGER_NORUN)) - return 0; - - if (!btree_node_type_needs_gc(iter->btree_id)) - return 0; - - if (!btree_node_type_is_extents(iter->btree_id)) { - /* iterators should be uptodate, shouldn't get errors here: */ - if (btree_iter_type(iter) != BTREE_ITER_CACHED) { - old = bch2_btree_iter_peek_slot(iter); - BUG_ON(bkey_err(old)); - } else { - struct bkey_cached *ck = (void *) iter->l[0].b; - - BUG_ON(!ck->valid); - old = bkey_i_to_s_c(ck->k); - } - - if (old.k->type == new->k.type) { - ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0, - BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags); - } else { - ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0, - BTREE_TRIGGER_INSERT|flags) ?: - bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), 0, 0, - BTREE_TRIGGER_OVERWRITE|flags); - } - } else { - struct btree_iter *copy; - struct bkey _old; - - EBUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED); - - bkey_init(&_old); - old = (struct bkey_s_c) { &_old, NULL }; - - ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), - 0, new->k.size, - BTREE_TRIGGER_INSERT); - if (ret) - return ret; - - copy = bch2_trans_copy_iter(trans, iter); - - for_each_btree_key_continue(copy, 0, old, ret) { - unsigned offset = 0; - s64 sectors = -((s64) old.k->size); - - flags |= BTREE_TRIGGER_OVERWRITE; - - if (bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0) - break; - - switch (bch2_extent_overlap(&new->k, old.k)) { - case BCH_EXTENT_OVERLAP_ALL: - offset = 0; - sectors = -((s64) old.k->size); - break; - case BCH_EXTENT_OVERLAP_BACK: - offset = bkey_start_offset(&new->k) - - bkey_start_offset(old.k); - sectors = bkey_start_offset(&new->k) - - old.k->p.offset; - break; - case BCH_EXTENT_OVERLAP_FRONT: - offset = 0; - sectors = bkey_start_offset(old.k) - - new->k.p.offset; - break; - case BCH_EXTENT_OVERLAP_MIDDLE: - offset = bkey_start_offset(&new->k) - - bkey_start_offset(old.k); - sectors = -((s64) new->k.size); - flags |= BTREE_TRIGGER_OVERWRITE_SPLIT; - break; - } - BUG_ON(sectors >= 0); + idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad); + end_idx = le64_to_cpu(p.v->idx) + p.k->size + + le32_to_cpu(p.v->back_pad); - ret = bch2_trans_mark_key(trans, old, bkey_i_to_s_c(new), - offset, sectors, flags); - if (ret) - break; - } - bch2_trans_iter_put(trans, copy); - } + while (idx < end_idx && !ret) + ret = __bch2_trans_mark_reflink_p(trans, p, &idx, flags); return ret; } @@ -1979,105 +1841,82 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans, unsigned sectors) { struct bch_fs *c = trans->c; - struct btree_iter *iter; - struct bkey_alloc_unpacked u; - struct bkey_alloc_buf *a; - struct bch_extent_ptr ptr = { - .dev = ca->dev_idx, - .offset = bucket_to_sector(ca, b), - }; + struct btree_iter iter; + struct bkey_i_alloc_v4 *a; int ret = 0; - a = bch2_trans_start_alloc_update(trans, &iter, &ptr, &u); + /* + * Backup superblock might be past the end of our normal usable space: + */ + if (b >= ca->mi.nbuckets) + return 0; + + a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b)); if (IS_ERR(a)) return PTR_ERR(a); - if (u.data_type && u.data_type != type) { + if (a->v.data_type && type && a->v.data_type != type) { bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n" "while marking %s", - iter->pos.inode, iter->pos.offset, u.gen, - bch2_data_types[u.data_type], + iter.pos.inode, iter.pos.offset, a->v.gen, + bch2_data_types[a->v.data_type], bch2_data_types[type], bch2_data_types[type]); ret = -EIO; goto out; } - if ((unsigned) (u.dirty_sectors + sectors) > ca->mi.bucket_size) { - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "bucket %llu:%llu gen %u data type %s sector count overflow: %u + %u > %u\n" - "while marking %s", - iter->pos.inode, iter->pos.offset, u.gen, - bch2_data_types[u.data_type ?: type], - u.dirty_sectors, sectors, ca->mi.bucket_size, - bch2_data_types[type]); - ret = -EIO; - goto out; - } + a->v.data_type = type; + a->v.dirty_sectors = sectors; - if (u.data_type == type && - u.dirty_sectors == sectors) + ret = bch2_trans_update(trans, &iter, &a->k_i, 0); + if (ret) goto out; - - u.data_type = type; - u.dirty_sectors = sectors; - - bch2_alloc_pack(c, a, u); - bch2_trans_update(trans, iter, &a->k, 0); out: - bch2_trans_iter_put(trans, iter); + bch2_trans_iter_exit(trans, &iter); return ret; } int bch2_trans_mark_metadata_bucket(struct btree_trans *trans, - struct disk_reservation *res, struct bch_dev *ca, size_t b, enum bch_data_type type, unsigned sectors) { - return __bch2_trans_do(trans, res, NULL, 0, - __bch2_trans_mark_metadata_bucket(trans, ca, b, BCH_DATA_journal, - ca->mi.bucket_size)); - + return commit_do(trans, NULL, NULL, 0, + __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors)); } static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans, - struct disk_reservation *res, struct bch_dev *ca, u64 start, u64 end, enum bch_data_type type, u64 *bucket, unsigned *bucket_sectors) { - int ret; - do { u64 b = sector_to_bucket(ca, start); unsigned sectors = min_t(u64, bucket_to_sector(ca, b + 1), end) - start; - if (b != *bucket) { - if (*bucket_sectors) { - ret = bch2_trans_mark_metadata_bucket(trans, res, ca, - *bucket, type, *bucket_sectors); - if (ret) - return ret; - } + if (b != *bucket && *bucket_sectors) { + int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket, + type, *bucket_sectors); + if (ret) + return ret; - *bucket = b; - *bucket_sectors = 0; + *bucket_sectors = 0; } + *bucket = b; *bucket_sectors += sectors; start += sectors; - } while (!ret && start < end); + } while (start < end); return 0; } static int __bch2_trans_mark_dev_sb(struct btree_trans *trans, - struct disk_reservation *res, - struct bch_dev *ca) + struct bch_dev *ca) { struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; u64 bucket = 0; @@ -2088,14 +1927,14 @@ static int __bch2_trans_mark_dev_sb(struct btree_trans *trans, u64 offset = le64_to_cpu(layout->sb_offset[i]); if (offset == BCH_SB_SECTOR) { - ret = bch2_trans_mark_metadata_sectors(trans, res, ca, + ret = bch2_trans_mark_metadata_sectors(trans, ca, 0, BCH_SB_SECTOR, BCH_DATA_sb, &bucket, &bucket_sectors); if (ret) return ret; } - ret = bch2_trans_mark_metadata_sectors(trans, res, ca, offset, + ret = bch2_trans_mark_metadata_sectors(trans, ca, offset, offset + (1 << layout->sb_max_size_bits), BCH_DATA_sb, &bucket, &bucket_sectors); if (ret) @@ -2103,14 +1942,14 @@ static int __bch2_trans_mark_dev_sb(struct btree_trans *trans, } if (bucket_sectors) { - ret = bch2_trans_mark_metadata_bucket(trans, res, ca, + ret = bch2_trans_mark_metadata_bucket(trans, ca, bucket, BCH_DATA_sb, bucket_sectors); if (ret) return ret; } for (i = 0; i < ca->journal.nr; i++) { - ret = bch2_trans_mark_metadata_bucket(trans, res, ca, + ret = bch2_trans_mark_metadata_bucket(trans, ca, ca->journal.buckets[i], BCH_DATA_journal, ca->mi.bucket_size); if (ret) @@ -2120,19 +1959,16 @@ static int __bch2_trans_mark_dev_sb(struct btree_trans *trans, return 0; } -int bch2_trans_mark_dev_sb(struct bch_fs *c, - struct disk_reservation *res, - struct bch_dev *ca) +int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca) { - return bch2_trans_do(c, res, NULL, 0, - __bch2_trans_mark_dev_sb(&trans, res, ca)); + return bch2_trans_run(c, __bch2_trans_mark_dev_sb(&trans, ca)); } /* Disk reservations: */ #define SECTORS_CACHE 1024 -int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, +int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, u64 sectors, int flags) { struct bch_fs_pcpu *pcpu; @@ -2185,7 +2021,7 @@ recalculate: ret = 0; } else { atomic64_set(&c->sectors_available, sectors_available); - ret = -ENOSPC; + ret = -BCH_ERR_ENOSPC_disk_reservation; } mutex_unlock(&c->sectors_available_lock); @@ -2196,54 +2032,37 @@ recalculate: /* Startup/shutdown: */ -static void buckets_free_rcu(struct rcu_head *rcu) +static void bucket_gens_free_rcu(struct rcu_head *rcu) { - struct bucket_array *buckets = - container_of(rcu, struct bucket_array, rcu); + struct bucket_gens *buckets = + container_of(rcu, struct bucket_gens, rcu); - kvpfree(buckets, - sizeof(struct bucket_array) + - buckets->nbuckets * sizeof(struct bucket)); + kvpfree(buckets, sizeof(*buckets) + 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; - alloc_heap alloc_heap; - - size_t btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, - ca->mi.bucket_size / c->opts.btree_node_size); - /* XXX: these should be tunable */ - size_t reserve_none = max_t(size_t, 1, nbuckets >> 9); - size_t copygc_reserve = max_t(size_t, 2, nbuckets >> 6); - size_t free_inc_nr = max(max_t(size_t, 1, nbuckets >> 12), - btree_reserve * 2); - bool resize = ca->buckets[0] != NULL; - int ret = -ENOMEM; - unsigned i; + bool resize = ca->bucket_gens != NULL; + int ret; - memset(&free, 0, sizeof(free)); - memset(&free_inc, 0, sizeof(free_inc)); - memset(&alloc_heap, 0, sizeof(alloc_heap)); + if (!(bucket_gens = kvpmalloc(sizeof(struct bucket_gens) + nbuckets, + GFP_KERNEL|__GFP_ZERO))) { + ret = -BCH_ERR_ENOMEM_bucket_gens; + goto err; + } - if (!(buckets = kvpmalloc(sizeof(struct bucket_array) + - nbuckets * sizeof(struct bucket), - GFP_KERNEL|__GFP_ZERO)) || - !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) * + if ((c->opts.buckets_nouse && + !(buckets_nouse = kvpmalloc(BITS_TO_LONGS(nbuckets) * sizeof(unsigned long), - GFP_KERNEL|__GFP_ZERO)) || - !init_fifo(&free[RESERVE_MOVINGGC], - copygc_reserve, GFP_KERNEL) || - !init_fifo(&free[RESERVE_NONE], reserve_none, GFP_KERNEL) || - !init_fifo(&free_inc, free_inc_nr, GFP_KERNEL) || - !init_heap(&alloc_heap, ALLOC_SCAN_BATCH(ca) << 1, GFP_KERNEL)) + GFP_KERNEL|__GFP_ZERO)))) { + ret = -BCH_ERR_ENOMEM_buckets_nouse; goto err; + } - 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); @@ -2253,56 +2072,39 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) percpu_down_write(&c->mark_lock); } - 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); - - memcpy(buckets->b, - old_buckets->b, - n * sizeof(struct bucket)); - memcpy(buckets_nouse, - ca->buckets_nouse, - BITS_TO_LONGS(n) * sizeof(unsigned long)); + size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets); + + memcpy(bucket_gens->b, + old_bucket_gens->b, + n); + if (buckets_nouse) + 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); + bucket_gens = old_bucket_gens; swap(ca->buckets_nouse, buckets_nouse); + nbuckets = ca->mi.nbuckets; + if (resize) { percpu_up_write(&c->mark_lock); + up_write(&ca->bucket_lock); up_write(&c->gc_lock); } - spin_lock(&c->freelist_lock); - for (i = 0; i < RESERVE_NR; i++) { - fifo_move(&free[i], &ca->free[i]); - swap(ca->free[i], free[i]); - } - fifo_move(&free_inc, &ca->free_inc); - swap(ca->free_inc, free_inc); - spin_unlock(&c->freelist_lock); - - /* with gc lock held, alloc_heap can't be in use: */ - swap(ca->alloc_heap, alloc_heap); - - nbuckets = ca->mi.nbuckets; - - if (resize) - up_write(&ca->bucket_lock); - ret = 0; err: - free_heap(&alloc_heap); - free_fifo(&free_inc); - for (i = 0; i < RESERVE_NR; i++) - free_fifo(&free[i]); kvpfree(buckets_nouse, BITS_TO_LONGS(nbuckets) * sizeof(unsigned long)); - if (buckets) - call_rcu(&old_buckets->rcu, buckets_free_rcu); + if (bucket_gens) + call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu); return ret; } @@ -2311,15 +2113,10 @@ void bch2_dev_buckets_free(struct bch_dev *ca) { unsigned i; - free_heap(&ca->alloc_heap); - free_fifo(&ca->free_inc); - for (i = 0; i < RESERVE_NR; i++) - free_fifo(&ca->free[i]); kvpfree(ca->buckets_nouse, BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long)); - kvpfree(rcu_dereference_protected(ca->buckets[0], 1), - sizeof(struct bucket_array) + - ca->mi.nbuckets * sizeof(struct bucket)); + kvpfree(rcu_dereference_protected(ca->bucket_gens, 1), + sizeof(struct bucket_gens) + ca->mi.nbuckets); for (i = 0; i < ARRAY_SIZE(ca->usage); i++) free_percpu(ca->usage[i]); @@ -2332,13 +2129,13 @@ int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca) ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL); if (!ca->usage_base) - return -ENOMEM; + return -BCH_ERR_ENOMEM_usage_init; for (i = 0; i < ARRAY_SIZE(ca->usage); i++) { ca->usage[i] = alloc_percpu(struct bch_dev_usage); if (!ca->usage[i]) - return -ENOMEM; + return -BCH_ERR_ENOMEM_usage_init; } - return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);; + return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets); }