From 70feb569f2ce915068ac3d2050b843322cb5218c Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Wed, 4 Dec 2024 18:14:14 -0500 Subject: [PATCH 169/233] bcachefs: Journal write path refactoring, debug improvements Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit Signed-off-by: Kent Overstreet Signed-off-by: Alexander Miroshnichenko --- fs/bcachefs/journal.c | 6 ++++ fs/bcachefs/journal_io.c | 70 ++++++++++++++++++++++------------------ 2 files changed, 45 insertions(+), 31 deletions(-) diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c index dc66521964b7..04a9ccf76d75 100644 --- a/fs/bcachefs/journal.c +++ b/fs/bcachefs/journal.c @@ -1564,6 +1564,9 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j) printbuf_indent_sub(out, 2); for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) { + if (!ca->mi.durability) + continue; + struct journal_device *ja = &ca->journal; if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d)) @@ -1573,6 +1576,7 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j) continue; prt_printf(out, "dev %u:\n", ca->dev_idx); + prt_printf(out, "durability %u:\n", ca->mi.durability); printbuf_indent_add(out, 2); prt_printf(out, "nr\t%u\n", ja->nr); prt_printf(out, "bucket size\t%u\n", ca->mi.bucket_size); @@ -1584,6 +1588,8 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j) printbuf_indent_sub(out, 2); } + prt_printf(out, "replicas want %u need %u\n", c->opts.metadata_replicas, c->opts.metadata_replicas_required); + rcu_read_unlock(); --out->atomic; diff --git a/fs/bcachefs/journal_io.c b/fs/bcachefs/journal_io.c index 9a1647297d11..2f4daa8bd498 100644 --- a/fs/bcachefs/journal_io.c +++ b/fs/bcachefs/journal_io.c @@ -1420,6 +1420,35 @@ int bch2_journal_read(struct bch_fs *c, /* journal write: */ +static void journal_advance_devs_to_next_bucket(struct journal *j, + struct dev_alloc_list *devs, + unsigned sectors, u64 seq) +{ + struct bch_fs *c = container_of(j, struct bch_fs, journal); + + darray_for_each(*devs, i) { + struct bch_dev *ca = rcu_dereference(c->devs[*i]); + if (!ca) + continue; + + struct journal_device *ja = &ca->journal; + + if (sectors > ja->sectors_free && + sectors <= ca->mi.bucket_size && + bch2_journal_dev_buckets_available(j, ja, + journal_space_discarded)) { + ja->cur_idx = (ja->cur_idx + 1) % ja->nr; + ja->sectors_free = ca->mi.bucket_size; + + /* + * ja->bucket_seq[ja->cur_idx] must always have + * something sensible: + */ + ja->bucket_seq[ja->cur_idx] = le64_to_cpu(seq); + } + } +} + static void __journal_write_alloc(struct journal *j, struct journal_buf *w, struct dev_alloc_list *devs, @@ -1429,9 +1458,6 @@ static void __journal_write_alloc(struct journal *j, { struct bch_fs *c = container_of(j, struct bch_fs, journal); - if (*replicas >= replicas_want) - return; - darray_for_each(*devs, i) { struct bch_dev *ca = rcu_dereference(c->devs[*i]); if (!ca) @@ -1491,6 +1517,7 @@ static int journal_write_alloc(struct journal *j, struct journal_buf *w) READ_ONCE(c->opts.metadata_replicas); unsigned replicas_need = min_t(unsigned, replicas_want, READ_ONCE(c->opts.metadata_replicas_required)); + bool advance_done = false; rcu_read_lock(); @@ -1502,45 +1529,26 @@ static int journal_write_alloc(struct journal *j, struct journal_buf *w) replicas += ca->mi.durability; } -retry: +retry_target: devs = target_rw_devs(c, BCH_DATA_journal, target); - devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs); - +retry_alloc: __journal_write_alloc(j, w, &devs_sorted, sectors, &replicas, replicas_want); - if (replicas >= replicas_want) + if (likely(replicas >= replicas_want)) goto done; - darray_for_each(devs_sorted, i) { - struct bch_dev *ca = rcu_dereference(c->devs[*i]); - if (!ca) - continue; - - struct journal_device *ja = &ca->journal; - - if (sectors > ja->sectors_free && - sectors <= ca->mi.bucket_size && - bch2_journal_dev_buckets_available(j, ja, - journal_space_discarded)) { - ja->cur_idx = (ja->cur_idx + 1) % ja->nr; - ja->sectors_free = ca->mi.bucket_size; - - /* - * ja->bucket_seq[ja->cur_idx] must always have - * something sensible: - */ - ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq); - } + if (!advance_done) { + journal_advance_devs_to_next_bucket(j, &devs_sorted, sectors, w->data->seq); + advance_done = true; + goto retry_alloc; } - __journal_write_alloc(j, w, &devs_sorted, - sectors, &replicas, replicas_want); - if (replicas < replicas_want && target) { /* Retry from all devices: */ target = 0; - goto retry; + advance_done = false; + goto retry_target; } done: rcu_read_unlock(); -- 2.45.2