]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/journal.c
New upstream release
[bcachefs-tools-debian] / libbcachefs / journal.c
index 80a612c0577f25d4228aabc3567d4e7aebe3de61..7d448136434bd8f2b63674298387e84c214753bb 100644 (file)
@@ -63,6 +63,7 @@ journal_seq_to_buf(struct journal *j, u64 seq)
 static void journal_pin_list_init(struct journal_entry_pin_list *p, int count)
 {
        unsigned i;
+
        for (i = 0; i < ARRAY_SIZE(p->list); i++)
                INIT_LIST_HEAD(&p->list[i]);
        INIT_LIST_HEAD(&p->flushed);
@@ -131,13 +132,21 @@ journal_error_check_stuck(struct journal *j, int error, unsigned flags)
        return stuck;
 }
 
-/* journal entry close/open: */
-
-void __bch2_journal_buf_put(struct journal *j)
+/*
+ * Final processing when the last reference of a journal buffer has been
+ * dropped. Drop the pin list reference acquired at journal entry open and write
+ * the buffer, if requested.
+ */
+void bch2_journal_buf_put_final(struct journal *j, u64 seq, bool write)
 {
        struct bch_fs *c = container_of(j, struct bch_fs, journal);
 
-       closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
+       lockdep_assert_held(&j->lock);
+
+       if (__bch2_journal_pin_put(j, seq))
+               bch2_journal_reclaim_fast(j);
+       if (write)
+               closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 }
 
 /*
@@ -203,13 +212,11 @@ static void __journal_entry_close(struct journal *j, unsigned closed_val)
        buf->data->last_seq     = cpu_to_le64(buf->last_seq);
        BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 
-       __bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
-
        cancel_delayed_work(&j->write_work);
 
        bch2_journal_space_available(j);
 
-       bch2_journal_buf_put(j, old.idx);
+       __bch2_journal_buf_put(j, old.idx, le64_to_cpu(buf->data->seq));
 }
 
 void bch2_journal_halt(struct journal *j)
@@ -354,11 +361,6 @@ static int journal_entry_open(struct journal *j)
        } while ((v = atomic64_cmpxchg(&j->reservations.counter,
                                       old.v, new.v)) != old.v);
 
-       if (j->res_get_blocked_start)
-               bch2_time_stats_update(j->blocked_time,
-                                      j->res_get_blocked_start);
-       j->res_get_blocked_start = 0;
-
        mod_delayed_work(c->io_complete_wq,
                         &j->write_work,
                         msecs_to_jiffies(c->opts.journal_flush_delay));
@@ -458,15 +460,12 @@ retry:
        __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
        ret = journal_entry_open(j);
 
-       if (ret == JOURNAL_ERR_max_in_flight)
+       if (ret == JOURNAL_ERR_max_in_flight) {
+               track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight],
+                                  &j->max_in_flight_start, true);
                trace_and_count(c, journal_entry_full, c);
-unlock:
-       if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
-           !j->res_get_blocked_start) {
-               j->res_get_blocked_start = local_clock() ?: 1;
-               trace_and_count(c, journal_full, c);
        }
-
+unlock:
        can_discard = j->can_discard;
        spin_unlock(&j->lock);
 
@@ -514,42 +513,11 @@ int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
        int ret;
 
        closure_wait_event(&j->async_wait,
-                  (ret = __journal_res_get(j, res, flags)) !=
-                  -BCH_ERR_journal_res_get_blocked||
+                  (ret = __journal_res_get(j, res, flags)) != -BCH_ERR_journal_res_get_blocked ||
                   (flags & JOURNAL_RES_GET_NONBLOCK));
        return ret;
 }
 
-/* journal_preres: */
-
-static bool journal_preres_available(struct journal *j,
-                                    struct journal_preres *res,
-                                    unsigned new_u64s,
-                                    unsigned flags)
-{
-       bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
-
-       if (!ret && mutex_trylock(&j->reclaim_lock)) {
-               bch2_journal_reclaim(j);
-               mutex_unlock(&j->reclaim_lock);
-       }
-
-       return ret;
-}
-
-int __bch2_journal_preres_get(struct journal *j,
-                             struct journal_preres *res,
-                             unsigned new_u64s,
-                             unsigned flags)
-{
-       int ret;
-
-       closure_wait_event(&j->preres_wait,
-                  (ret = bch2_journal_error(j)) ||
-                  journal_preres_available(j, res, new_u64s, flags));
-       return ret;
-}
-
 /* journal_entry_res: */
 
 void bch2_journal_entry_res_resize(struct journal *j,
@@ -588,8 +556,13 @@ out:
 
 /**
  * bch2_journal_flush_seq_async - wait for a journal entry to be written
+ * @j:         journal object
+ * @seq:       seq to flush
+ * @parent:    closure object to wait with
+ * Returns:    1 if @seq has already been flushed, 0 if @seq is being flushed,
+ *             -EIO if @seq will never be flushed
  *
- * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
+ * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if
  * necessary
  */
 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
@@ -829,12 +802,12 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
                                break;
 
                        ret = bch2_trans_run(c,
-                               bch2_trans_mark_metadata_bucket(&trans, ca,
+                               bch2_trans_mark_metadata_bucket(trans, ca,
                                                ob[nr_got]->bucket, BCH_DATA_journal,
                                                ca->mi.bucket_size));
                        if (ret) {
                                bch2_open_bucket_put(c, ob[nr_got]);
-                               bch_err(c, "error marking new journal buckets: %s", bch2_err_str(ret));
+                               bch_err_msg(c, ret, "marking new journal buckets");
                                break;
                        }
 
@@ -910,7 +883,7 @@ err_unblock:
        if (ret && !new_fs)
                for (i = 0; i < nr_got; i++)
                        bch2_trans_run(c,
-                               bch2_trans_mark_metadata_bucket(&trans, ca,
+                               bch2_trans_mark_metadata_bucket(trans, ca,
                                                bu[i], BCH_DATA_free, 0));
 err_free:
        if (!new_fs)
@@ -944,7 +917,7 @@ int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
                goto unlock;
 
        while (ja->nr < nr) {
-               struct disk_reservation disk_res = { 0, 0 };
+               struct disk_reservation disk_res = { 0, 0, 0 };
 
                /*
                 * note: journal buckets aren't really counted as _sectors_ used yet, so
@@ -1008,6 +981,25 @@ err:
        return ret;
 }
 
+int bch2_fs_journal_alloc(struct bch_fs *c)
+{
+       struct bch_dev *ca;
+       unsigned i;
+
+       for_each_online_member(ca, c, i) {
+               if (ca->journal.nr)
+                       continue;
+
+               int ret = bch2_dev_journal_alloc(ca);
+               if (ret) {
+                       percpu_ref_put(&ca->io_ref);
+                       return ret;
+               }
+       }
+
+       return 0;
+}
+
 /* startup/shutdown: */
 
 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
@@ -1159,9 +1151,9 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 {
        struct journal_device *ja = &ca->journal;
        struct bch_sb_field_journal *journal_buckets =
-               bch2_sb_get_journal(sb);
+               bch2_sb_field_get(sb, journal);
        struct bch_sb_field_journal_v2 *journal_buckets_v2 =
-               bch2_sb_get_journal_v2(sb);
+               bch2_sb_field_get(sb, journal_v2);
        unsigned i, nr_bvecs;
 
        ja->nr = 0;
@@ -1260,6 +1252,7 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
        union journal_res_state s;
        struct bch_dev *ca;
        unsigned long now = jiffies;
+       u64 nr_writes = j->nr_flush_writes + j->nr_noflush_writes;
        u64 seq;
        unsigned i;
 
@@ -1273,21 +1266,23 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
        prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
        prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
        prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
-       prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
+       prt_printf(out, "last_seq:\t\t%llu\n",                  journal_last_seq(j));
        prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
-       prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
-       prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
-       prt_printf(out, "watermark:\t\t%s\n",           bch2_watermarks[j->watermark]);
-       prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
+       prt_printf(out, "flushed_seq_ondisk:\t%llu\n",          j->flushed_seq_ondisk);
+       prt_printf(out, "watermark:\t\t%s\n",                   bch2_watermarks[j->watermark]);
+       prt_printf(out, "each entry reserved:\t%u\n",           j->entry_u64s_reserved);
        prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
-       prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
-       prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
+       prt_printf(out, "nr noflush writes:\t%llu\n",           j->nr_noflush_writes);
+       prt_printf(out, "average write size:\t");
+       prt_human_readable_u64(out, nr_writes ? div64_u64(j->entry_bytes_written, nr_writes) : 0);
+       prt_newline(out);
+       prt_printf(out, "nr direct reclaim:\t%llu\n",           j->nr_direct_reclaim);
        prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
        prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
-       prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
+       prt_printf(out, "reclaim runs in:\t%u ms\n",            time_after(j->next_reclaim, now)
               ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
-       prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
-       prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
+       prt_printf(out, "current entry sectors:\t%u\n",         j->cur_entry_sectors);
+       prt_printf(out, "current entry error:\t%s\n",           bch2_journal_errors[j->cur_entry_error]);
        prt_printf(out, "current entry:\t\t");
 
        switch (s.cur_entry_offset) {