Lines Matching +full:additional +full:- +full:devs
1 // SPDX-License-Identifier: GPL-2.0
25 return seq > j->seq_ondisk; in journal_seq_unwritten()
35 return atomic64_read(&j->seq) - j->seq_ondisk; in nr_unwritten_journal_entries()
40 return __journal_entry_is_open(j->reservations); in journal_entry_is_open()
45 union journal_res_state s = READ_ONCE(j->reservations); in bch2_journal_buf_to_text()
47 struct journal_buf *buf = j->buf + i; in bch2_journal_buf_to_text()
52 if (!buf->write_started) in bch2_journal_buf_to_text()
55 struct closure *cl = &buf->io; in bch2_journal_buf_to_text()
56 int r = atomic_read(&cl->remaining); in bch2_journal_buf_to_text()
57 prt_printf(out, "io:\t%pS r %i\n", cl->fn, r & CLOSURE_REMAINING_MASK); in bch2_journal_buf_to_text()
59 if (buf->data) { in bch2_journal_buf_to_text()
61 prt_human_readable_u64(out, vstruct_bytes(buf->data)); in bch2_journal_buf_to_text()
65 prt_printf(out, "expires:\t%li jiffies\n", buf->expires - jiffies); in bch2_journal_buf_to_text()
68 if (buf->noflush) in bch2_journal_buf_to_text()
70 if (buf->must_flush) in bch2_journal_buf_to_text()
72 if (buf->separate_flush) in bch2_journal_buf_to_text()
74 if (buf->need_flush_to_write_buffer) in bch2_journal_buf_to_text()
76 if (buf->write_started) in bch2_journal_buf_to_text()
78 if (buf->write_allocated) in bch2_journal_buf_to_text()
80 if (buf->write_done) in bch2_journal_buf_to_text()
89 lockdep_assert_held(&j->lock); in bch2_journal_bufs_to_text()
90 out->atomic++; in bch2_journal_bufs_to_text()
92 if (!out->nr_tabstops) in bch2_journal_bufs_to_text()
101 --out->atomic; in bch2_journal_bufs_to_text()
112 buf = j->buf + (seq & JOURNAL_BUF_MASK); in journal_seq_to_buf()
118 for (unsigned i = 0; i < ARRAY_SIZE(p->unflushed); i++) in journal_pin_list_init()
119 INIT_LIST_HEAD(&p->unflushed[i]); in journal_pin_list_init()
120 for (unsigned i = 0; i < ARRAY_SIZE(p->flushed); i++) in journal_pin_list_init()
121 INIT_LIST_HEAD(&p->flushed[i]); in journal_pin_list_init()
122 atomic_set(&p->count, count); in journal_pin_list_init()
123 p->devs.nr = 0; in journal_pin_list_init()
146 if (!(error == -BCH_ERR_journal_full || in journal_error_check_stuck()
147 error == -BCH_ERR_journal_pin_full) || in journal_error_check_stuck()
152 spin_lock(&j->lock); in journal_error_check_stuck()
154 if (j->can_discard) { in journal_error_check_stuck()
155 spin_unlock(&j->lock); in journal_error_check_stuck()
162 * The journal shutdown path will set ->err_seq, but do it here first to in journal_error_check_stuck()
166 if (j->err_seq) { in journal_error_check_stuck()
167 spin_unlock(&j->lock); in journal_error_check_stuck()
170 j->err_seq = journal_cur_seq(j); in journal_error_check_stuck()
173 spin_unlock(&j->lock); in journal_error_check_stuck()
174 prt_printf(&buf, bch2_fmt(c, "Journal stuck! Hava a pre-reservation but journal full (error %s)"), in journal_error_check_stuck()
195 struct journal_buf *w = j->buf + idx; in bch2_journal_do_writes()
197 if (w->write_started && !w->write_allocated) in bch2_journal_do_writes()
199 if (w->write_started) in bch2_journal_do_writes()
202 if (!journal_state_seq_count(j, j->reservations, seq)) { in bch2_journal_do_writes()
203 j->seq_write_started = seq; in bch2_journal_do_writes()
204 w->write_started = true; in bch2_journal_do_writes()
205 closure_call(&w->io, bch2_journal_write, j->wq, NULL); in bch2_journal_do_writes()
219 lockdep_assert_held(&j->lock); in bch2_journal_buf_put_final()
229 wake_up(&j->wait); in bch2_journal_buf_put_final()
236 * and can be opened again - this also initializes the next journal_buf:
248 lockdep_assert_held(&j->lock); in __journal_entry_close()
250 old.v = atomic64_read(&j->reservations.counter); in __journal_entry_close()
258 } while (!atomic64_try_cmpxchg(&j->reservations.counter, in __journal_entry_close()
265 old.cur_entry_offset = j->cur_entry_offset_if_blocked; in __journal_entry_close()
268 buf->data->u64s = cpu_to_le32(old.cur_entry_offset); in __journal_entry_close()
275 prt_human_readable_u64(&pbuf, vstruct_bytes(buf->data)); in __journal_entry_close()
282 sectors = vstruct_blocks_plus(buf->data, c->block_bits, in __journal_entry_close()
283 buf->u64s_reserved) << c->block_bits; in __journal_entry_close()
284 if (unlikely(sectors > buf->sectors)) { in __journal_entry_close()
289 sectors, buf->sectors); in __journal_entry_close()
291 le32_to_cpu(buf->data->u64s), buf->u64s_reserved, in __journal_entry_close()
292 j->cur_entry_u64s, in __journal_entry_close()
293 c->block_bits); in __journal_entry_close()
294 prt_printf(&err, "fatal error - emergency read only"); in __journal_entry_close()
302 buf->sectors = sectors; in __journal_entry_close()
308 * journal reservation - the expectation being that the journal will in __journal_entry_close()
323 buf->last_seq = journal_last_seq(j); in __journal_entry_close()
324 buf->data->last_seq = cpu_to_le64(buf->last_seq); in __journal_entry_close()
325 BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq)); in __journal_entry_close()
327 cancel_delayed_work(&j->write_work); in __journal_entry_close()
331 __bch2_journal_buf_put(j, le64_to_cpu(buf->data->seq)); in __journal_entry_close()
336 spin_lock(&j->lock); in bch2_journal_halt()
338 if (!j->err_seq) in bch2_journal_halt()
339 j->err_seq = journal_cur_seq(j); in bch2_journal_halt()
341 spin_unlock(&j->lock); in bch2_journal_halt()
346 lockdep_assert_held(&j->lock); in bch2_journal_halt_locked()
349 if (!j->err_seq) in bch2_journal_halt_locked()
350 j->err_seq = journal_cur_seq(j); in bch2_journal_halt_locked()
365 if (!buf->flush_time) { in journal_entry_want_write()
366 buf->flush_time = local_clock() ?: 1; in journal_entry_want_write()
367 buf->expires = jiffies; in journal_entry_want_write()
378 spin_lock(&j->lock); in bch2_journal_entry_close()
380 spin_unlock(&j->lock); in bch2_journal_entry_close()
386 * should _only_ called from journal_res_get() - when we actually want a
387 * journal reservation - journal entry is open means journal is dirty:
392 struct journal_buf *buf = j->buf + in journal_entry_open()
397 lockdep_assert_held(&j->lock); in journal_entry_open()
399 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb)); in journal_entry_open()
401 if (j->blocked) in journal_entry_open()
402 return -BCH_ERR_journal_blocked; in journal_entry_open()
404 if (j->cur_entry_error) in journal_entry_open()
405 return j->cur_entry_error; in journal_entry_open()
411 if (!fifo_free(&j->pin)) in journal_entry_open()
412 return -BCH_ERR_journal_pin_full; in journal_entry_open()
414 if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf)) in journal_entry_open()
415 return -BCH_ERR_journal_max_in_flight; in journal_entry_open()
417 if (atomic64_read(&j->seq) - j->seq_write_started == JOURNAL_STATE_BUF_NR) in journal_entry_open()
418 return -BCH_ERR_journal_max_open; in journal_entry_open()
423 bch_err(c, "fatal error - emergency read only"); in journal_entry_open()
424 return -BCH_ERR_journal_shutdown; in journal_entry_open()
427 if (!j->free_buf && !buf->data) in journal_entry_open()
428 return -BCH_ERR_journal_buf_enomem; /* will retry after write completion frees up a buf */ in journal_entry_open()
430 BUG_ON(!j->cur_entry_sectors); in journal_entry_open()
432 if (!buf->data) { in journal_entry_open()
433 swap(buf->data, j->free_buf); in journal_entry_open()
434 swap(buf->buf_size, j->free_buf_size); in journal_entry_open()
437 buf->expires = in journal_entry_open()
438 (journal_cur_seq(j) == j->flushed_seq_ondisk in journal_entry_open()
440 : j->last_flush_write) + in journal_entry_open()
441 msecs_to_jiffies(c->opts.journal_flush_delay); in journal_entry_open()
443 buf->u64s_reserved = j->entry_u64s_reserved; in journal_entry_open()
444 buf->disk_sectors = j->cur_entry_sectors; in journal_entry_open()
445 buf->sectors = min(buf->disk_sectors, buf->buf_size >> 9); in journal_entry_open()
447 u64s = (int) (buf->sectors << 9) / sizeof(u64) - in journal_entry_open()
449 u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1); in journal_entry_open()
451 if (u64s <= (ssize_t) j->early_journal_entries.nr) in journal_entry_open()
452 return -BCH_ERR_journal_full; in journal_entry_open()
454 if (fifo_empty(&j->pin) && j->reclaim_thread) in journal_entry_open()
455 wake_up_process(j->reclaim_thread); in journal_entry_open()
458 * The fifo_push() needs to happen at the same time as j->seq is in journal_entry_open()
461 atomic64_inc(&j->seq); in journal_entry_open()
462 journal_pin_list_init(fifo_push_ref(&j->pin), 1); in journal_entry_open()
464 BUG_ON(j->pin.back - 1 != atomic64_read(&j->seq)); in journal_entry_open()
466 BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf); in journal_entry_open()
468 bkey_extent_init(&buf->key); in journal_entry_open()
469 buf->noflush = false; in journal_entry_open()
470 buf->must_flush = false; in journal_entry_open()
471 buf->separate_flush = false; in journal_entry_open()
472 buf->flush_time = 0; in journal_entry_open()
473 buf->need_flush_to_write_buffer = true; in journal_entry_open()
474 buf->write_started = false; in journal_entry_open()
475 buf->write_allocated = false; in journal_entry_open()
476 buf->write_done = false; in journal_entry_open()
478 memset(buf->data, 0, sizeof(*buf->data)); in journal_entry_open()
479 buf->data->seq = cpu_to_le64(journal_cur_seq(j)); in journal_entry_open()
480 buf->data->u64s = 0; in journal_entry_open()
482 if (j->early_journal_entries.nr) { in journal_entry_open()
483 memcpy(buf->data->_data, j->early_journal_entries.data, in journal_entry_open()
484 j->early_journal_entries.nr * sizeof(u64)); in journal_entry_open()
485 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr); in journal_entry_open()
491 j->cur_entry_u64s = u64s; in journal_entry_open()
493 old.v = atomic64_read(&j->reservations.counter); in journal_entry_open()
506 new.cur_entry_offset = le32_to_cpu(buf->data->u64s); in journal_entry_open()
507 } while (!atomic64_try_cmpxchg(&j->reservations.counter, in journal_entry_open()
511 mod_delayed_work(j->wq, in journal_entry_open()
512 &j->write_work, in journal_entry_open()
513 msecs_to_jiffies(c->opts.journal_flush_delay)); in journal_entry_open()
516 if (j->early_journal_entries.nr) in journal_entry_open()
517 darray_exit(&j->early_journal_entries); in journal_entry_open()
523 bool ret = atomic64_read(&j->seq) == j->seq_ondisk; in journal_quiesced()
532 wait_event(j->wait, journal_quiesced(j)); in journal_quiesce()
539 spin_lock(&j->lock); in journal_write_work()
540 if (__journal_entry_is_open(j->reservations)) { in journal_write_work()
541 long delta = journal_cur_buf(j)->expires - jiffies; in journal_write_work()
544 mod_delayed_work(j->wq, &j->write_work, delta); in journal_write_work()
548 spin_unlock(&j->lock); in journal_write_work()
553 if (j->free_buf && in journal_buf_prealloc()
554 j->free_buf_size >= j->buf_size_want) in journal_buf_prealloc()
557 unsigned buf_size = j->buf_size_want; in journal_buf_prealloc()
559 spin_unlock(&j->lock); in journal_buf_prealloc()
561 spin_lock(&j->lock); in journal_buf_prealloc()
564 (!j->free_buf || in journal_buf_prealloc()
565 buf_size > j->free_buf_size)) { in journal_buf_prealloc()
566 swap(buf, j->free_buf); in journal_buf_prealloc()
567 swap(buf_size, j->free_buf_size); in journal_buf_prealloc()
571 spin_unlock(&j->lock); in journal_buf_prealloc()
574 spin_lock(&j->lock); in journal_buf_prealloc()
593 if (j->blocked) in __journal_res_get()
594 return -BCH_ERR_journal_blocked; in __journal_res_get()
596 if ((flags & BCH_WATERMARK_MASK) < j->watermark) { in __journal_res_get()
597 ret = -BCH_ERR_journal_full; in __journal_res_get()
598 can_discard = j->can_discard; in __journal_res_get()
602 if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) && !journal_entry_is_open(j)) { in __journal_res_get()
603 ret = -BCH_ERR_journal_max_in_flight; in __journal_res_get()
607 spin_lock(&j->lock); in __journal_res_get()
628 buf->buf_size >> 9 < buf->disk_sectors && in __journal_res_get()
629 buf->buf_size < JOURNAL_ENTRY_SIZE_MAX) in __journal_res_get()
630 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1); in __journal_res_get()
633 ret = journal_entry_open(j) ?: -BCH_ERR_journal_retry_open; in __journal_res_get()
635 can_discard = j->can_discard; in __journal_res_get()
636 spin_unlock(&j->lock); in __journal_res_get()
640 if (ret == -BCH_ERR_journal_retry_open) in __journal_res_get()
644 ret = -BCH_ERR_journal_stuck; in __journal_res_get()
646 if (ret == -BCH_ERR_journal_max_in_flight && in __journal_res_get()
647 track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], true) && in __journal_res_get()
653 spin_lock(&j->lock); in __journal_res_get()
656 spin_unlock(&j->lock); in __journal_res_get()
663 if (ret == -BCH_ERR_journal_max_open && in __journal_res_get()
664 track_event_change(&c->times[BCH_TIME_blocked_journal_max_open], true) && in __journal_res_get()
670 spin_lock(&j->lock); in __journal_res_get()
673 spin_unlock(&j->lock); in __journal_res_get()
681 * Journal is full - can't rely on reclaim from work item due to in __journal_res_get()
684 if ((ret == -BCH_ERR_journal_full || in __journal_res_get()
685 ret == -BCH_ERR_journal_pin_full) && in __journal_res_get()
692 if (mutex_trylock(&j->reclaim_lock)) { in __journal_res_get()
694 mutex_unlock(&j->reclaim_lock); in __journal_res_get()
706 nsecs = max(nsecs, ca->io_latency[WRITE].stats.max_duration); in max_dev_latency()
727 if (closure_wait_event_timeout(&j->async_wait, in bch2_journal_res_get_slowpath()
739 remaining_wait = max(0, remaining_wait - HZ); in bch2_journal_res_get_slowpath()
741 if (closure_wait_event_timeout(&j->async_wait, in bch2_journal_res_get_slowpath()
753 closure_wait_event(&j->async_wait, in bch2_journal_res_get_slowpath()
766 int d = new_u64s - res->u64s; in bch2_journal_entry_res_resize()
768 spin_lock(&j->lock); in bch2_journal_entry_res_resize()
770 j->entry_u64s_reserved += d; in bch2_journal_entry_res_resize()
774 j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d); in bch2_journal_entry_res_resize()
775 state = READ_ONCE(j->reservations); in bch2_journal_entry_res_resize()
778 state.cur_entry_offset > j->cur_entry_u64s) { in bch2_journal_entry_res_resize()
779 j->cur_entry_u64s += d; in bch2_journal_entry_res_resize()
785 journal_cur_buf(j)->u64s_reserved += d; in bch2_journal_entry_res_resize()
788 spin_unlock(&j->lock); in bch2_journal_entry_res_resize()
789 res->u64s += d; in bch2_journal_entry_res_resize()
795 * bch2_journal_flush_seq_async - wait for a journal entry to be written
800 * -BCH_ERR_journal_flush_err if @seq will never be flushed
811 if (seq <= j->flushed_seq_ondisk) in bch2_journal_flush_seq_async()
814 spin_lock(&j->lock); in bch2_journal_flush_seq_async()
822 if (j->err_seq && seq >= j->err_seq) { in bch2_journal_flush_seq_async()
823 ret = -BCH_ERR_journal_flush_err; in bch2_journal_flush_seq_async()
827 if (seq <= j->flushed_seq_ondisk) { in bch2_journal_flush_seq_async()
832 /* if seq was written, but not flushed - flush a newer one instead */ in bch2_journal_flush_seq_async()
842 spin_unlock(&j->lock); in bch2_journal_flush_seq_async()
845 * We're called from bch2_journal_flush_seq() -> wait_event(); in bch2_journal_flush_seq_async()
856 buf->must_flush = true; in bch2_journal_flush_seq_async()
858 if (!buf->flush_time) { in bch2_journal_flush_seq_async()
859 buf->flush_time = local_clock() ?: 1; in bch2_journal_flush_seq_async()
860 buf->expires = jiffies; in bch2_journal_flush_seq_async()
863 if (parent && !closure_wait(&buf->wait, parent)) in bch2_journal_flush_seq_async()
868 spin_lock(&j->lock); in bch2_journal_flush_seq_async()
877 if (buf->noflush) { in bch2_journal_flush_seq_async()
882 buf->must_flush = true; in bch2_journal_flush_seq_async()
883 j->flushing_seq = max(j->flushing_seq, seq); in bch2_journal_flush_seq_async()
885 if (parent && !closure_wait(&buf->wait, parent)) in bch2_journal_flush_seq_async()
891 spin_unlock(&j->lock); in bch2_journal_flush_seq_async()
903 if (seq <= j->flushed_seq_ondisk) in bch2_journal_flush_seq()
906 ret = wait_event_state(j->wait, in bch2_journal_flush_seq()
911 bch2_time_stats_update(j->flush_seq_time, start_time); in bch2_journal_flush_seq()
917 * bch2_journal_flush_async - if there is an open journal entry, or a journal
922 bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent); in bch2_journal_flush_async()
927 return bch2_journal_flush_seq(j, atomic64_read(&j->seq), TASK_UNINTERRUPTIBLE); in bch2_journal_flush()
931 * bch2_journal_noflush_seq - ask the journal not to issue any flushes in the
941 if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush))) in bch2_journal_noflush_seq()
944 if (c->journal.flushed_seq_ondisk >= start) in bch2_journal_noflush_seq()
947 spin_lock(&j->lock); in bch2_journal_noflush_seq()
948 if (c->journal.flushed_seq_ondisk >= start) in bch2_journal_noflush_seq()
957 if (buf->must_flush) in bch2_journal_noflush_seq()
960 buf->noflush = true; in bch2_journal_noflush_seq()
965 spin_unlock(&j->lock); in bch2_journal_noflush_seq()
976 struct journal_buf *buf = j->buf + (res.seq & JOURNAL_BUF_MASK); in __bch2_journal_meta()
977 buf->must_flush = true; in __bch2_journal_meta()
979 if (!buf->flush_time) { in __bch2_journal_meta()
980 buf->flush_time = local_clock() ?: 1; in __bch2_journal_meta()
981 buf->expires = jiffies; in __bch2_journal_meta()
994 return -BCH_ERR_erofs_no_writes; in bch2_journal_meta()
1005 spin_lock(&j->lock); in bch2_journal_unblock()
1006 if (!--j->blocked && in bch2_journal_unblock()
1007 j->cur_entry_offset_if_blocked < JOURNAL_ENTRY_CLOSED_VAL && in bch2_journal_unblock()
1008 j->reservations.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL) { in bch2_journal_unblock()
1011 old.v = atomic64_read(&j->reservations.counter); in bch2_journal_unblock()
1014 new.cur_entry_offset = j->cur_entry_offset_if_blocked; in bch2_journal_unblock()
1015 } while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v)); in bch2_journal_unblock()
1017 spin_unlock(&j->lock); in bch2_journal_unblock()
1024 if (!j->blocked++) { in __bch2_journal_block()
1027 old.v = atomic64_read(&j->reservations.counter); in __bch2_journal_block()
1029 j->cur_entry_offset_if_blocked = old.cur_entry_offset; in __bch2_journal_block()
1031 if (j->cur_entry_offset_if_blocked >= JOURNAL_ENTRY_CLOSED_VAL) in __bch2_journal_block()
1036 } while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v)); in __bch2_journal_block()
1039 journal_cur_buf(j)->data->u64s = cpu_to_le32(old.cur_entry_offset); in __bch2_journal_block()
1045 spin_lock(&j->lock); in bch2_journal_block()
1047 spin_unlock(&j->lock); in bch2_journal_block()
1059 mutex_lock(&j->buf_lock); in __bch2_next_write_buffer_flush_journal_buf()
1060 spin_lock(&j->lock); in __bch2_next_write_buffer_flush_journal_buf()
1067 struct journal_buf *buf = j->buf + idx; in __bch2_next_write_buffer_flush_journal_buf()
1069 if (buf->need_flush_to_write_buffer) { in __bch2_next_write_buffer_flush_journal_buf()
1071 s.v = atomic64_read_acquire(&j->reservations.counter); in __bch2_next_write_buffer_flush_journal_buf()
1081 ? ERR_PTR(-EAGAIN) in __bch2_next_write_buffer_flush_journal_buf()
1087 spin_unlock(&j->lock); in __bch2_next_write_buffer_flush_journal_buf()
1089 mutex_unlock(&j->buf_lock); in __bch2_next_write_buffer_flush_journal_buf()
1099 wait_event(j->wait, (ret = __bch2_next_write_buffer_flush_journal_buf(j, in bch2_next_write_buffer_flush_journal_buf()
1100 max_seq, blocked)) != ERR_PTR(-EAGAIN)); in bch2_next_write_buffer_flush_journal_buf()
1112 struct bch_fs *c = ca->fs; in bch2_set_nr_journal_buckets_iter()
1113 struct journal_device *ja = &ca->journal; in bch2_set_nr_journal_buckets_iter()
1117 unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr; in bch2_set_nr_journal_buckets_iter()
1120 BUG_ON(nr <= ja->nr); in bch2_set_nr_journal_buckets_iter()
1127 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets; in bch2_set_nr_journal_buckets_iter()
1145 ob[nr_got]->bucket, BCH_DATA_journal, in bch2_set_nr_journal_buckets_iter()
1146 ca->mi.bucket_size, BTREE_TRIGGER_transactional)); in bch2_set_nr_journal_buckets_iter()
1154 bu[nr_got] = ob[nr_got]->bucket; in bch2_set_nr_journal_buckets_iter()
1164 bch2_journal_flush_all_pins(&c->journal); in bch2_set_nr_journal_buckets_iter()
1165 bch2_journal_block(&c->journal); in bch2_set_nr_journal_buckets_iter()
1166 mutex_lock(&c->sb_lock); in bch2_set_nr_journal_buckets_iter()
1169 memcpy(new_buckets, ja->buckets, ja->nr * sizeof(u64)); in bch2_set_nr_journal_buckets_iter()
1170 memcpy(new_bucket_seq, ja->bucket_seq, ja->nr * sizeof(u64)); in bch2_set_nr_journal_buckets_iter()
1172 BUG_ON(ja->discard_idx > ja->nr); in bch2_set_nr_journal_buckets_iter()
1174 pos = ja->discard_idx ?: ja->nr; in bch2_set_nr_journal_buckets_iter()
1178 sizeof(new_buckets[0]) * (ja->nr - pos)); in bch2_set_nr_journal_buckets_iter()
1181 sizeof(new_bucket_seq[0]) * (ja->nr - pos)); in bch2_set_nr_journal_buckets_iter()
1188 nr = ja->nr + nr_got; in bch2_set_nr_journal_buckets_iter()
1198 spin_lock(&c->journal.lock); in bch2_set_nr_journal_buckets_iter()
1200 swap(new_buckets, ja->buckets); in bch2_set_nr_journal_buckets_iter()
1201 swap(new_bucket_seq, ja->bucket_seq); in bch2_set_nr_journal_buckets_iter()
1202 ja->nr = nr; in bch2_set_nr_journal_buckets_iter()
1204 if (pos <= ja->discard_idx) in bch2_set_nr_journal_buckets_iter()
1205 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr; in bch2_set_nr_journal_buckets_iter()
1206 if (pos <= ja->dirty_idx_ondisk) in bch2_set_nr_journal_buckets_iter()
1207 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr; in bch2_set_nr_journal_buckets_iter()
1208 if (pos <= ja->dirty_idx) in bch2_set_nr_journal_buckets_iter()
1209 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr; in bch2_set_nr_journal_buckets_iter()
1210 if (pos <= ja->cur_idx) in bch2_set_nr_journal_buckets_iter()
1211 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr; in bch2_set_nr_journal_buckets_iter()
1214 spin_unlock(&c->journal.lock); in bch2_set_nr_journal_buckets_iter()
1217 bch2_journal_unblock(&c->journal); in bch2_set_nr_journal_buckets_iter()
1218 mutex_unlock(&c->sb_lock); in bch2_set_nr_journal_buckets_iter()
1241 struct journal_device *ja = &ca->journal; in bch2_set_nr_journal_buckets_loop()
1248 if (nr < ja->nr) in bch2_set_nr_journal_buckets_loop()
1251 while (!ret && ja->nr < nr) { in bch2_set_nr_journal_buckets_loop()
1257 * when space used goes up without a reservation - but we do need the in bch2_set_nr_journal_buckets_loop()
1261 * filesystem-wide allocation will succeed, this is a device in bch2_set_nr_journal_buckets_loop()
1262 * specific allocation - we can hang here: in bch2_set_nr_journal_buckets_loop()
1266 bucket_to_sector(ca, nr - ja->nr), 1, 0); in bch2_set_nr_journal_buckets_loop()
1273 if (ret == -BCH_ERR_bucket_alloc_blocked || in bch2_set_nr_journal_buckets_loop()
1274 ret == -BCH_ERR_open_buckets_empty) in bch2_set_nr_journal_buckets_loop()
1285 * Allocate more journal space at runtime - not currently making use if it, but
1291 down_write(&c->state_lock); in bch2_set_nr_journal_buckets()
1293 up_write(&c->state_lock); in bch2_set_nr_journal_buckets()
1305 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets; in bch2_dev_journal_alloc()
1310 nr = ca->mi.nbuckets >> 7; in bch2_dev_journal_alloc()
1319 (1 << 24) / ca->mi.bucket_size)); in bch2_dev_journal_alloc()
1321 ret = bch2_set_nr_journal_buckets_loop(ca->fs, ca, nr, new_fs); in bch2_dev_journal_alloc()
1330 if (ca->journal.nr) in bch2_fs_journal_alloc()
1335 percpu_ref_put(&ca->io_ref[READ]); in bch2_fs_journal_alloc()
1350 spin_lock(&j->lock); in bch2_journal_writing_to_device()
1356 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx)) in bch2_journal_writing_to_device()
1359 spin_unlock(&j->lock); in bch2_journal_writing_to_device()
1366 wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx)); in bch2_dev_journal_stop()
1371 if (!test_bit(JOURNAL_running, &j->flags)) in bch2_fs_journal_stop()
1377 wait_event(j->wait, bch2_journal_entry_close(j)); in bch2_fs_journal_stop()
1386 cancel_delayed_work_sync(&j->write_work); in bch2_fs_journal_stop()
1389 test_bit(JOURNAL_replay_done, &j->flags) && in bch2_fs_journal_stop()
1390 j->last_empty_seq != journal_cur_seq(j), in bch2_fs_journal_stop()
1392 journal_cur_seq(j), j->last_empty_seq); in bch2_fs_journal_stop()
1395 clear_bit(JOURNAL_running, &j->flags); in bch2_fs_journal_stop()
1409 return -EINVAL; in bch2_fs_journal_start()
1412 genradix_for_each_reverse(&c->journal_entries, iter, _i) { in bch2_fs_journal_start()
1418 last_seq = le64_to_cpu(i->j.last_seq); in bch2_fs_journal_start()
1422 nr = cur_seq - last_seq; in bch2_fs_journal_start()
1426 * nearly or completely full. We'll need to be able to open additional in bch2_fs_journal_start()
1432 if (nr + 1 > j->pin.size) { in bch2_fs_journal_start()
1433 free_fifo(&j->pin); in bch2_fs_journal_start()
1434 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL); in bch2_fs_journal_start()
1435 if (!j->pin.data) { in bch2_fs_journal_start()
1437 return -BCH_ERR_ENOMEM_journal_pin_fifo; in bch2_fs_journal_start()
1441 j->replay_journal_seq = last_seq; in bch2_fs_journal_start()
1442 j->replay_journal_seq_end = cur_seq; in bch2_fs_journal_start()
1443 j->last_seq_ondisk = last_seq; in bch2_fs_journal_start()
1444 j->flushed_seq_ondisk = cur_seq - 1; in bch2_fs_journal_start()
1445 j->seq_write_started = cur_seq - 1; in bch2_fs_journal_start()
1446 j->seq_ondisk = cur_seq - 1; in bch2_fs_journal_start()
1447 j->pin.front = last_seq; in bch2_fs_journal_start()
1448 j->pin.back = cur_seq; in bch2_fs_journal_start()
1449 atomic64_set(&j->seq, cur_seq - 1); in bch2_fs_journal_start()
1451 fifo_for_each_entry_ptr(p, &j->pin, seq) in bch2_fs_journal_start()
1454 genradix_for_each(&c->journal_entries, iter, _i) { in bch2_fs_journal_start()
1460 seq = le64_to_cpu(i->j.seq); in bch2_fs_journal_start()
1466 if (journal_entry_empty(&i->j)) in bch2_fs_journal_start()
1467 j->last_empty_seq = le64_to_cpu(i->j.seq); in bch2_fs_journal_start()
1471 p->devs.nr = 0; in bch2_fs_journal_start()
1472 darray_for_each(i->ptrs, ptr) in bch2_fs_journal_start()
1473 bch2_dev_list_add_dev(&p->devs, ptr->dev); in bch2_fs_journal_start()
1479 j->last_empty_seq = cur_seq - 1; /* to match j->seq */ in bch2_fs_journal_start()
1481 spin_lock(&j->lock); in bch2_fs_journal_start()
1482 j->last_flush_write = jiffies; in bch2_fs_journal_start()
1484 j->reservations.idx = journal_cur_seq(j); in bch2_fs_journal_start()
1486 c->last_bucket_seq_cleanup = journal_cur_seq(j); in bch2_fs_journal_start()
1487 spin_unlock(&j->lock); in bch2_fs_journal_start()
1498 spin_lock(&j->lock); in bch2_journal_set_replay_done()
1501 set_bit(JOURNAL_need_flush_write, &j->flags); in bch2_journal_set_replay_done()
1502 set_bit(JOURNAL_running, &j->flags); in bch2_journal_set_replay_done()
1503 set_bit(JOURNAL_replay_done, &j->flags); in bch2_journal_set_replay_done()
1504 spin_unlock(&j->lock); in bch2_journal_set_replay_done()
1511 struct journal_device *ja = &ca->journal; in bch2_dev_journal_exit()
1513 for (unsigned i = 0; i < ARRAY_SIZE(ja->bio); i++) { in bch2_dev_journal_exit()
1514 kfree(ja->bio[i]); in bch2_dev_journal_exit()
1515 ja->bio[i] = NULL; in bch2_dev_journal_exit()
1518 kfree(ja->buckets); in bch2_dev_journal_exit()
1519 kfree(ja->bucket_seq); in bch2_dev_journal_exit()
1520 ja->buckets = NULL; in bch2_dev_journal_exit()
1521 ja->bucket_seq = NULL; in bch2_dev_journal_exit()
1526 struct journal_device *ja = &ca->journal; in bch2_dev_journal_init()
1532 ja->nr = 0; in bch2_dev_journal_init()
1538 ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr); in bch2_dev_journal_init()
1540 ja->nr = bch2_nr_journal_buckets(journal_buckets); in bch2_dev_journal_init()
1543 ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL); in bch2_dev_journal_init()
1544 if (!ja->bucket_seq) in bch2_dev_journal_init()
1545 return -BCH_ERR_ENOMEM_dev_journal_init; in bch2_dev_journal_init()
1549 for (unsigned i = 0; i < ARRAY_SIZE(ja->bio); i++) { in bch2_dev_journal_init()
1550 ja->bio[i] = kzalloc(struct_size(ja->bio[i], bio.bi_inline_vecs, in bch2_dev_journal_init()
1552 if (!ja->bio[i]) in bch2_dev_journal_init()
1553 return -BCH_ERR_ENOMEM_dev_journal_init; in bch2_dev_journal_init()
1555 ja->bio[i]->ca = ca; in bch2_dev_journal_init()
1556 ja->bio[i]->buf_idx = i; in bch2_dev_journal_init()
1557 bio_init(&ja->bio[i]->bio, NULL, ja->bio[i]->bio.bi_inline_vecs, nr_bvecs, 0); in bch2_dev_journal_init()
1560 ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL); in bch2_dev_journal_init()
1561 if (!ja->buckets) in bch2_dev_journal_init()
1562 return -BCH_ERR_ENOMEM_dev_journal_init; in bch2_dev_journal_init()
1569 for (unsigned j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++) in bch2_dev_journal_init()
1570 ja->buckets[dst++] = in bch2_dev_journal_init()
1571 le64_to_cpu(journal_buckets_v2->d[i].start) + j; in bch2_dev_journal_init()
1573 for (unsigned i = 0; i < ja->nr; i++) in bch2_dev_journal_init()
1574 ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]); in bch2_dev_journal_init()
1582 if (j->wq) in bch2_fs_journal_exit()
1583 destroy_workqueue(j->wq); in bch2_fs_journal_exit()
1585 darray_exit(&j->early_journal_entries); in bch2_fs_journal_exit()
1587 for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++) in bch2_fs_journal_exit()
1588 kvfree(j->buf[i].data); in bch2_fs_journal_exit()
1589 kvfree(j->free_buf); in bch2_fs_journal_exit()
1590 free_fifo(&j->pin); in bch2_fs_journal_exit()
1597 mutex_init(&j->buf_lock); in bch2_fs_journal_init()
1598 spin_lock_init(&j->lock); in bch2_fs_journal_init()
1599 spin_lock_init(&j->err_lock); in bch2_fs_journal_init()
1600 init_waitqueue_head(&j->wait); in bch2_fs_journal_init()
1601 INIT_DELAYED_WORK(&j->write_work, journal_write_work); in bch2_fs_journal_init()
1602 init_waitqueue_head(&j->reclaim_wait); in bch2_fs_journal_init()
1603 init_waitqueue_head(&j->pin_flush_wait); in bch2_fs_journal_init()
1604 mutex_init(&j->reclaim_lock); in bch2_fs_journal_init()
1605 mutex_init(&j->discard_lock); in bch2_fs_journal_init()
1607 lockdep_init_map(&j->res_map, "journal res", &res_key, 0); in bch2_fs_journal_init()
1609 atomic64_set(&j->reservations.counter, in bch2_fs_journal_init()
1613 if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) in bch2_fs_journal_init()
1614 return -BCH_ERR_ENOMEM_journal_pin_fifo; in bch2_fs_journal_init()
1616 j->free_buf_size = j->buf_size_want = JOURNAL_ENTRY_SIZE_MIN; in bch2_fs_journal_init()
1617 j->free_buf = kvmalloc(j->free_buf_size, GFP_KERNEL); in bch2_fs_journal_init()
1618 if (!j->free_buf) in bch2_fs_journal_init()
1619 return -BCH_ERR_ENOMEM_journal_buf; in bch2_fs_journal_init()
1621 for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++) in bch2_fs_journal_init()
1622 j->buf[i].idx = i; in bch2_fs_journal_init()
1624 j->pin.front = j->pin.back = 1; in bch2_fs_journal_init()
1626 j->wq = alloc_workqueue("bcachefs_journal", in bch2_fs_journal_init()
1628 if (!j->wq) in bch2_fs_journal_init()
1629 return -BCH_ERR_ENOMEM_fs_other_alloc; in bch2_fs_journal_init()
1647 u64 nr_writes = j->nr_flush_writes + j->nr_noflush_writes; in __bch2_journal_debug_to_text()
1651 out->atomic++; in __bch2_journal_debug_to_text()
1654 s = READ_ONCE(j->reservations); in __bch2_journal_debug_to_text()
1657 prt_bitflags(out, bch2_journal_flags_strs, j->flags); in __bch2_journal_debug_to_text()
1659 prt_printf(out, "dirty journal entries:\t%llu/%llu\n", fifo_used(&j->pin), j->pin.size); in __bch2_journal_debug_to_text()
1661 prt_printf(out, "seq_ondisk:\t%llu\n", j->seq_ondisk); in __bch2_journal_debug_to_text()
1663 prt_printf(out, "last_seq_ondisk:\t%llu\n", j->last_seq_ondisk); in __bch2_journal_debug_to_text()
1664 prt_printf(out, "flushed_seq_ondisk:\t%llu\n", j->flushed_seq_ondisk); in __bch2_journal_debug_to_text()
1665 prt_printf(out, "watermark:\t%s\n", bch2_watermarks[j->watermark]); in __bch2_journal_debug_to_text()
1666 prt_printf(out, "each entry reserved:\t%u\n", j->entry_u64s_reserved); in __bch2_journal_debug_to_text()
1667 prt_printf(out, "nr flush writes:\t%llu\n", j->nr_flush_writes); in __bch2_journal_debug_to_text()
1668 prt_printf(out, "nr noflush writes:\t%llu\n", j->nr_noflush_writes); in __bch2_journal_debug_to_text()
1670 prt_human_readable_u64(out, nr_writes ? div64_u64(j->entry_bytes_written, nr_writes) : 0); in __bch2_journal_debug_to_text()
1672 prt_printf(out, "free buf:\t%u\n", j->free_buf ? j->free_buf_size : 0); in __bch2_journal_debug_to_text()
1673 prt_printf(out, "nr direct reclaim:\t%llu\n", j->nr_direct_reclaim); in __bch2_journal_debug_to_text()
1674 prt_printf(out, "nr background reclaim:\t%llu\n", j->nr_background_reclaim); in __bch2_journal_debug_to_text()
1675 prt_printf(out, "reclaim kicked:\t%u\n", j->reclaim_kicked); in __bch2_journal_debug_to_text()
1676 prt_printf(out, "reclaim runs in:\t%u ms\n", time_after(j->next_reclaim, now) in __bch2_journal_debug_to_text()
1677 ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0); in __bch2_journal_debug_to_text()
1678 prt_printf(out, "blocked:\t%u\n", j->blocked); in __bch2_journal_debug_to_text()
1679 prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors); in __bch2_journal_debug_to_text()
1680 prt_printf(out, "current entry error:\t%s\n", bch2_err_str(j->cur_entry_error)); in __bch2_journal_debug_to_text()
1694 prt_printf(out, "%u/%u\n", s.cur_entry_offset, j->cur_entry_u64s); in __bch2_journal_debug_to_text()
1704 j->space[journal_space_discarded].next_entry, in __bch2_journal_debug_to_text()
1705 j->space[journal_space_discarded].total); in __bch2_journal_debug_to_text()
1707 j->space[journal_space_clean_ondisk].next_entry, in __bch2_journal_debug_to_text()
1708 j->space[journal_space_clean_ondisk].total); in __bch2_journal_debug_to_text()
1710 j->space[journal_space_clean].next_entry, in __bch2_journal_debug_to_text()
1711 j->space[journal_space_clean].total); in __bch2_journal_debug_to_text()
1713 j->space[journal_space_total].next_entry, in __bch2_journal_debug_to_text()
1714 j->space[journal_space_total].total); in __bch2_journal_debug_to_text()
1717 for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) { in __bch2_journal_debug_to_text()
1718 if (!ca->mi.durability) in __bch2_journal_debug_to_text()
1721 struct journal_device *ja = &ca->journal; in __bch2_journal_debug_to_text()
1723 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d)) in __bch2_journal_debug_to_text()
1726 if (!ja->nr) in __bch2_journal_debug_to_text()
1729 prt_printf(out, "dev %u:\n", ca->dev_idx); in __bch2_journal_debug_to_text()
1730 prt_printf(out, "durability %u:\n", ca->mi.durability); in __bch2_journal_debug_to_text()
1732 prt_printf(out, "nr\t%u\n", ja->nr); in __bch2_journal_debug_to_text()
1733 prt_printf(out, "bucket size\t%u\n", ca->mi.bucket_size); in __bch2_journal_debug_to_text()
1734 …\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free); in __bch2_journal_debug_to_text()
1735 prt_printf(out, "discard_idx\t%u\n", ja->discard_idx); in __bch2_journal_debug_to_text()
1736 …prt_printf(out, "dirty_ondisk\t%u (seq %llu)\n",ja->dirty_idx_ondisk, ja->bucket_seq[ja->dirty_idx… in __bch2_journal_debug_to_text()
1737 prt_printf(out, "dirty_idx\t%u (seq %llu)\n", ja->dirty_idx, ja->bucket_seq[ja->dirty_idx]); in __bch2_journal_debug_to_text()
1738 prt_printf(out, "cur_idx\t%u (seq %llu)\n", ja->cur_idx, ja->bucket_seq[ja->cur_idx]); in __bch2_journal_debug_to_text()
1742 …prt_printf(out, "replicas want %u need %u\n", c->opts.metadata_replicas, c->opts.metadata_replicas… in __bch2_journal_debug_to_text()
1746 --out->atomic; in __bch2_journal_debug_to_text()
1751 spin_lock(&j->lock); in bch2_journal_debug_to_text()
1753 spin_unlock(&j->lock); in bch2_journal_debug_to_text()