]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/journal.c
Update bcachefs sources to 8dbfede1d9 fixup! bcachefs: More info on check_bucket_ref...
[bcachefs-tools-debian] / libbcachefs / journal.c
index 3f1cf1ac921a064741beecce7e9daa6136fb7dec..957eeece4d98f56c827f6063d074998dd5401e4c 100644 (file)
@@ -199,12 +199,6 @@ static bool journal_entry_close(struct journal *j)
 /*
  * should _only_ called from journal_res_get() - when we actually want a
  * journal reservation - journal entry is open means journal is dirty:
- *
- * returns:
- * 0:          success
- * -ENOSPC:    journal currently full, must invoke reclaim
- * -EAGAIN:    journal blocked, must wait
- * -EROFS:     insufficient rw devices or journal error
  */
 static int journal_entry_open(struct journal *j)
 {
@@ -231,7 +225,7 @@ static int journal_entry_open(struct journal *j)
        if (!fifo_free(&j->pin))
                return JOURNAL_ERR_journal_pin_full;
 
-       if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) - 1)
+       if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
                return JOURNAL_ERR_max_in_flight;
 
        BUG_ON(!j->cur_entry_sectors);
@@ -250,7 +244,7 @@ static int journal_entry_open(struct journal *j)
                journal_entry_overhead(j);
        u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 
-       if (u64s <= 0)
+       if (u64s <= (ssize_t) j->early_journal_entries.nr)
                return JOURNAL_ERR_journal_full;
 
        if (fifo_empty(&j->pin) && j->reclaim_thread)
@@ -275,6 +269,12 @@ static int journal_entry_open(struct journal *j)
        buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
        buf->data->u64s = 0;
 
+       if (j->early_journal_entries.nr) {
+               memcpy(buf->data->_data, j->early_journal_entries.data,
+                      j->early_journal_entries.nr * sizeof(u64));
+               le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
+       }
+
        /*
         * Must be set before marking the journal entry as open:
         */
@@ -291,7 +291,9 @@ static int journal_entry_open(struct journal *j)
                BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 
                journal_state_inc(&new);
-               new.cur_entry_offset = 0;
+
+               /* Handle any already added entries */
+               new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
        } while ((v = atomic64_cmpxchg(&j->reservations.counter,
                                       old.v, new.v)) != old.v);
 
@@ -304,6 +306,9 @@ static int journal_entry_open(struct journal *j)
                         &j->write_work,
                         msecs_to_jiffies(c->opts.journal_flush_delay));
        journal_wake(j);
+
+       if (j->early_journal_entries.nr)
+               darray_exit(&j->early_journal_entries);
        return 0;
 }
 
@@ -353,7 +358,7 @@ retry:
                return 0;
 
        if (bch2_journal_error(j))
-               return -EROFS;
+               return -BCH_ERR_erofs_journal_err;
 
        spin_lock(&j->lock);
 
@@ -391,12 +396,12 @@ retry:
        ret = journal_entry_open(j);
 
        if (ret == JOURNAL_ERR_max_in_flight)
-               trace_journal_entry_full(c);
+               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_journal_full(c);
+               trace_and_count(c, journal_full, c);
        }
 
        can_discard = j->can_discard;
@@ -445,7 +450,9 @@ unlock:
                }
        }
 
-       return ret == JOURNAL_ERR_insufficient_devices ? -EROFS : -EAGAIN;
+       return ret == JOURNAL_ERR_insufficient_devices
+               ? -EROFS
+               : -BCH_ERR_journal_res_get_blocked;
 }
 
 /*
@@ -464,7 +471,8 @@ 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)) != -EAGAIN ||
+                  (ret = __journal_res_get(j, res, flags)) !=
+                  -BCH_ERR_journal_res_get_blocked||
                   (flags & JOURNAL_RES_GET_NONBLOCK));
        return ret;
 }
@@ -720,39 +728,6 @@ int bch2_journal_meta(struct journal *j)
        return bch2_journal_flush_seq(j, res.seq);
 }
 
-int bch2_journal_log_msg(struct journal *j, const char *fmt, ...)
-{
-       struct jset_entry_log *entry;
-       struct journal_res res = { 0 };
-       unsigned msglen, u64s;
-       va_list args;
-       int ret;
-
-       va_start(args, fmt);
-       msglen = vsnprintf(NULL, 0, fmt, args) + 1;
-       va_end(args);
-
-       u64s = jset_u64s(DIV_ROUND_UP(msglen, sizeof(u64)));
-
-       ret = bch2_journal_res_get(j, &res, u64s, 0);
-       if (ret)
-               return ret;
-
-       entry = container_of(journal_res_entry(j, &res),
-                            struct jset_entry_log, entry);;
-       memset(entry, 0, u64s * sizeof(u64));
-       entry->entry.type = BCH_JSET_ENTRY_log;
-       entry->entry.u64s = u64s - 1;
-
-       va_start(args, fmt);
-       vsnprintf(entry->d, INT_MAX, fmt, args);
-       va_end(args);
-
-       bch2_journal_res_put(j, &res);
-
-       return bch2_journal_flush_seq(j, res.seq);
-}
-
 /* block/unlock the journal: */
 
 void bch2_journal_unblock(struct journal *j)
@@ -794,12 +769,13 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
        if (c) {
                bch2_journal_flush_all_pins(&c->journal);
                bch2_journal_block(&c->journal);
+               mutex_lock(&c->sb_lock);
        }
 
-       bu              = kzalloc(nr_want * sizeof(*bu), GFP_KERNEL);
-       ob              = kzalloc(nr_want * sizeof(*ob), GFP_KERNEL);
-       new_buckets     = kzalloc(nr * sizeof(u64), GFP_KERNEL);
-       new_bucket_seq  = kzalloc(nr * sizeof(u64), GFP_KERNEL);
+       bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
+       ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
+       new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
+       new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
        if (!bu || !ob || !new_buckets || !new_bucket_seq) {
                ret = -ENOMEM;
                goto err_unblock;
@@ -809,16 +785,15 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
                if (new_fs) {
                        bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
                        if (bu[nr_got] < 0) {
-                               ret = -ENOSPC;
+                               ret = -BCH_ERR_ENOSPC_bucket_alloc;
                                break;
                        }
                } else {
                        ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none,
                                               false, cl);
-                       if (IS_ERR(ob[nr_got])) {
-                               ret = cl ? -EAGAIN : -ENOSPC;
+                       ret = PTR_ERR_OR_ZERO(ob[nr_got]);
+                       if (ret)
                                break;
-                       }
 
                        bu[nr_got] = ob[nr_got]->bucket;
                }
@@ -875,6 +850,9 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
        if (!new_fs)
                spin_unlock(&c->journal.lock);
 
+       if (ja->nr != old_nr && !new_fs)
+               bch2_write_super(c);
+
        if (c)
                bch2_journal_unblock(&c->journal);
 
@@ -894,6 +872,9 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
                }
        }
 err:
+       if (c)
+               mutex_unlock(&c->sb_lock);
+
        if (ob && !new_fs)
                for (i = 0; i < nr_got; i++)
                        bch2_open_bucket_put(c, ob[i]);
@@ -919,7 +900,6 @@ int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 {
        struct journal_device *ja = &ca->journal;
        struct closure cl;
-       unsigned current_nr;
        int ret = 0;
 
        /* don't handle reducing nr of buckets yet: */
@@ -928,43 +908,44 @@ int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 
        closure_init_stack(&cl);
 
-       while (ja->nr != nr && (ret == 0 || ret == -EAGAIN)) {
+       while (ja->nr != nr) {
                struct disk_reservation disk_res = { 0, 0 };
 
-               closure_sync(&cl);
-
-               mutex_lock(&c->sb_lock);
-               current_nr = ja->nr;
-
                /*
                 * note: journal buckets aren't really counted as _sectors_ used yet, so
                 * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
                 * when space used goes up without a reservation - but we do need the
                 * reservation to ensure we'll actually be able to allocate:
+                *
+                * XXX: that's not right, disk reservations only ensure a
+                * filesystem-wide allocation will succeed, this is a device
+                * specific allocation - we can hang here:
                 */
 
-               if (bch2_disk_reservation_get(c, &disk_res,
-                                             bucket_to_sector(ca, nr - ja->nr), 1, 0)) {
-                       mutex_unlock(&c->sb_lock);
-                       return -ENOSPC;
-               }
+               ret = bch2_disk_reservation_get(c, &disk_res,
+                                               bucket_to_sector(ca, nr - ja->nr), 1, 0);
+               if (ret)
+                       break;
 
                ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 
                bch2_disk_reservation_put(c, &disk_res);
 
-               if (ja->nr != current_nr)
-                       bch2_write_super(c);
-               mutex_unlock(&c->sb_lock);
+               closure_sync(&cl);
+
+               if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
+                       break;
        }
 
+       if (ret)
+               bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
+
        return ret;
 }
 
 int bch2_dev_journal_alloc(struct bch_dev *ca)
 {
        unsigned nr;
-       int ret;
 
        if (dynamic_fault("bcachefs:add:journal_alloc"))
                return -ENOMEM;
@@ -981,15 +962,7 @@ int bch2_dev_journal_alloc(struct bch_dev *ca)
                     min(1 << 13,
                         (1 << 24) / ca->mi.bucket_size));
 
-       if (ca->fs)
-               mutex_lock(&ca->fs->sb_lock);
-
-       ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
-
-       if (ca->fs)
-               mutex_unlock(&ca->fs->sb_lock);
-
-       return ret;
+       return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 }
 
 /* startup/shutdown: */
@@ -1195,6 +1168,8 @@ void bch2_fs_journal_exit(struct journal *j)
 {
        unsigned i;
 
+       darray_exit(&j->early_journal_entries);
+
        for (i = 0; i < ARRAY_SIZE(j->buf); i++)
                kvpfree(j->buf[i].data, j->buf[i].buf_size);
        free_fifo(&j->pin);
@@ -1262,7 +1237,7 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
        rcu_read_lock();
        s = READ_ONCE(j->reservations);
 
-       prt_printf(out, "dirty journal entries:\t%llu/%llu\n",fifo_used(&j->pin), j->pin.size);
+       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));