From: Kent Overstreet Date: Fri, 28 May 2021 03:16:25 +0000 (-0400) Subject: bcachefs: Fix journal write error path X-Git-Url: http://git.maquefel.me/?a=commitdiff_plain;h=d797ca3d8ed3b493d6558426fb4b4980ff186c37;p=linux.git bcachefs: Fix journal write error path Journal write errors were racing with the submission path - potentially causing writes to other replicas to not get submitted. Signed-off-by: Kent Overstreet --- diff --git a/fs/bcachefs/journal_io.c b/fs/bcachefs/journal_io.c index 66a0e44a21a63..e42ca2a5e201c 100644 --- a/fs/bcachefs/journal_io.c +++ b/fs/bcachefs/journal_io.c @@ -1232,8 +1232,6 @@ static void journal_write_done(struct closure *cl) struct journal *j = container_of(cl, struct journal, io); struct bch_fs *c = container_of(j, struct bch_fs, journal); struct journal_buf *w = journal_last_unwritten_buf(j); - struct bch_devs_list devs = - bch2_bkey_devs(bkey_i_to_s_c(&w->key)); struct bch_replicas_padded replicas; union journal_res_state old, new; u64 v, seq; @@ -1241,11 +1239,12 @@ static void journal_write_done(struct closure *cl) bch2_time_stats_update(j->write_time, j->write_start_time); - if (!devs.nr) { + if (!w->devs_written.nr) { bch_err(c, "unable to write journal to sufficient devices"); err = -EIO; } else { - bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, devs); + bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal, + w->devs_written); if (bch2_mark_replicas(c, &replicas.e)) err = -EIO; } @@ -1257,7 +1256,7 @@ static void journal_write_done(struct closure *cl) seq = le64_to_cpu(w->data->seq); if (seq >= j->pin.front) - journal_seq_pin(j, seq)->devs = devs; + journal_seq_pin(j, seq)->devs = w->devs_written; j->seq_ondisk = seq; if (err && (!j->err_seq || seq < j->err_seq)) @@ -1307,15 +1306,15 @@ static void journal_write_endio(struct bio *bio) { struct bch_dev *ca = bio->bi_private; struct journal *j = &ca->fs->journal; + struct journal_buf *w = journal_last_unwritten_buf(j); + unsigned long flags; - if (bch2_dev_io_err_on(bio->bi_status, ca, "journal write error: %s", + if (bch2_dev_io_err_on(bio->bi_status, ca, "error writing journal entry %llu: %s", + le64_to_cpu(w->data->seq), bch2_blk_status_to_str(bio->bi_status)) || bch2_meta_write_fault("journal")) { - struct journal_buf *w = journal_last_unwritten_buf(j); - unsigned long flags; - spin_lock_irqsave(&j->err_lock, flags); - bch2_bkey_drop_device(bkey_i_to_s(&w->key), ca->dev_idx); + bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx); spin_unlock_irqrestore(&j->err_lock, flags); } @@ -1510,10 +1509,8 @@ retry_alloc: return; } - /* - * XXX: we really should just disable the entire journal in nochanges - * mode - */ + w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key)); + if (c->opts.nochanges) goto no_io; diff --git a/fs/bcachefs/journal_types.h b/fs/bcachefs/journal_types.h index b4e4d5e0bc93a..cce02bad850cc 100644 --- a/fs/bcachefs/journal_types.h +++ b/fs/bcachefs/journal_types.h @@ -21,6 +21,7 @@ struct journal_buf { struct jset *data; __BKEY_PADDED(key, BCH_REPLICAS_MAX); + struct bch_devs_list devs_written; struct closure_waitlist wait; u64 last_seq; /* copy of data->last_seq */