void bch2_bio_free_pages_pool(struct bch_fs *c, struct bio *bio)
{
struct bvec_iter_all iter;
- struct bio_vec *bv;
+ struct bio_vec bv;
bio_for_each_segment_all(bv, bio, iter)
- if (bv->bv_page != ZERO_PAGE(0))
- mempool_free(bv->bv_page, &c->bio_bounce_pages);
+ if (bv.bv_page != ZERO_PAGE(0))
+ mempool_free(bv.bv_page, &c->bio_bounce_pages);
bio->bi_vcnt = 0;
}
bch2_trans_copy_iter(&iter, extent_iter);
- for_each_btree_key_continue_norestart(iter, BTREE_ITER_SLOTS, old, ret) {
+ for_each_btree_key_upto_continue_norestart(iter,
+ new->k.p, BTREE_ITER_SLOTS, old, ret) {
s64 sectors = min(new->k.p.offset, old.k->p.offset) -
max(bkey_start_offset(&new->k),
bkey_start_offset(old.k));
struct open_buckets open_buckets;
struct bkey_s_c k;
struct bkey_buf old, new;
+ unsigned sectors_allocated;
bool have_reservation = false;
bool unwritten = opts.nocow &&
c->sb.version >= bcachefs_metadata_version_unwritten_extents;
closure_init_stack(&cl);
open_buckets.nr = 0;
retry:
+ sectors_allocated = 0;
+
k = bch2_btree_iter_peek_slot(iter);
ret = bkey_err(k);
if (ret)
opts.data_replicas,
opts.data_replicas,
RESERVE_none, 0, &cl, &wp);
- if (bch2_err_matches(ret, BCH_ERR_operation_blocked)) {
+ if (ret) {
bch2_trans_unlock(trans);
closure_sync(&cl);
- goto retry;
- }
- if (ret)
+ if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
+ goto retry;
return ret;
+ }
sectors = min(sectors, wp->sectors_free);
+ sectors_allocated = sectors;
bch2_key_resize(&e->k, sectors);
goto retry;
}
+ if (!ret && sectors_allocated)
+ bch2_increment_clock(c, sectors_allocated, WRITE);
+
bch2_open_buckets_put(c, &open_buckets);
bch2_disk_reservation_put(c, &disk_res);
bch2_bkey_buf_exit(&new, c);
struct bch_fs *c = op->c;
bch2_disk_reservation_put(c, &op->res);
- bch2_write_ref_put(c, BCH_WRITE_REF_write);
+ if (!(op->flags & BCH_WRITE_MOVE))
+ bch2_write_ref_put(c, BCH_WRITE_REF_write);
bch2_keylist_free(&op->insert_keys, op->inline_keys);
bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time);
}
if (dst != src)
- memmove_u64s_down(dst, src, src->u64s);
+ memmove_u64s_down(dst, src, src->k.u64s);
dst = bkey_next(dst);
}
struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
struct write_point *wp = op->wp;
struct workqueue_struct *wq = index_update_wq(op);
+ unsigned long flags;
- barrier();
+ if ((op->flags & BCH_WRITE_DONE) &&
+ (op->flags & BCH_WRITE_MOVE))
+ bch2_bio_free_pages_pool(op->c, &op->wbio.bio);
- /*
- * We're not using wp->writes_lock here, so this is racey: that's ok,
- * because this is just for diagnostic purposes, and we're running out
- * of interrupt context here so if we were to take the log we'd have to
- * switch to spin_lock_irq()/irqsave(), which is not free:
- */
+ spin_lock_irqsave(&wp->writes_lock, flags);
if (wp->state == WRITE_POINT_waiting_io)
__wp_update_state(wp, WRITE_POINT_waiting_work);
+ list_add_tail(&op->wp_list, &wp->writes);
+ spin_unlock_irqrestore (&wp->writes_lock, flags);
- op->btree_update_ready = true;
queue_work(wq, &wp->index_update_work);
}
+static inline void bch2_write_queue(struct bch_write_op *op, struct write_point *wp)
+{
+ op->wp = wp;
+
+ if (wp->state == WRITE_POINT_stopped) {
+ spin_lock_irq(&wp->writes_lock);
+ __wp_update_state(wp, WRITE_POINT_waiting_io);
+ spin_unlock_irq(&wp->writes_lock);
+ }
+}
+
void bch2_write_point_do_index_updates(struct work_struct *work)
{
struct write_point *wp =
struct bch_write_op *op;
while (1) {
- spin_lock(&wp->writes_lock);
- list_for_each_entry(op, &wp->writes, wp_list)
- if (op->btree_update_ready) {
- list_del(&op->wp_list);
- goto unlock;
- }
- op = NULL;
-unlock:
+ spin_lock_irq(&wp->writes_lock);
+ op = list_first_entry_or_null(&wp->writes, struct bch_write_op, wp_list);
+ if (op)
+ list_del(&op->wp_list);
wp_update_state(wp, op != NULL);
- spin_unlock(&wp->writes_lock);
+ spin_unlock_irq(&wp->writes_lock);
if (!op)
break;
struct btree_iter iter;
struct bkey_s_c k;
struct bkey_ptrs_c ptrs;
- const struct bch_extent_ptr *ptr, *ptr2;
+ const struct bch_extent_ptr *ptr;
struct {
struct bpos b;
unsigned gen;
bucket_to_u64(buckets[nr_buckets].b));
prefetch(buckets[nr_buckets].l);
- nr_buckets++;
if (unlikely(!bch2_dev_get_ioref(bch_dev_bkey_exists(c, ptr->dev), WRITE)))
goto err_get_ioref;
+ nr_buckets++;
+
if (ptr->unwritten)
op->flags |= BCH_WRITE_CONVERT_UNWRITTEN;
}
}
return;
err_get_ioref:
- bkey_for_each_ptr(ptrs, ptr2) {
- if (ptr2 == ptr)
- break;
-
- percpu_ref_put(&bch_dev_bkey_exists(c, ptr2->dev)->io_ref);
- }
+ for (i = 0; i < nr_buckets; i++)
+ percpu_ref_put(&bch_dev_bkey_exists(c, buckets[i].b.inode)->io_ref);
/* Fall back to COW path: */
goto out;
bch2_bucket_nocow_unlock(&c->nocow_locks,
buckets[i].b,
BUCKET_NOCOW_LOCK_UPDATE);
-
- bkey_for_each_ptr(ptrs, ptr2)
- percpu_ref_put(&bch_dev_bkey_exists(c, ptr2->dev)->io_ref);
+ for (i = 0; i < nr_buckets; i++)
+ percpu_ref_put(&bch_dev_bkey_exists(c, buckets[i].b.inode)->io_ref);
/* We can retry this: */
ret = BCH_ERR_transaction_restart;
nofs_flags = memalloc_nofs_save();
- if (unlikely(op->opts.nocow)) {
+ if (unlikely(op->opts.nocow && c->opts.nocow_enabled)) {
bch2_nocow_write(op);
if (op->flags & BCH_WRITE_DONE)
goto out_nofs_restore;
}
again:
memset(&op->failed, 0, sizeof(op->failed));
- op->btree_update_ready = false;
do {
struct bkey_i *key_to_write;
bch2_alloc_sectors_done_inlined(c, wp);
err:
if (ret <= 0) {
- if (!(op->flags & BCH_WRITE_SYNC)) {
- spin_lock(&wp->writes_lock);
- op->wp = wp;
- list_add_tail(&op->wp_list, &wp->writes);
- if (wp->state == WRITE_POINT_stopped)
- __wp_update_state(wp, WRITE_POINT_waiting_io);
- spin_unlock(&wp->writes_lock);
- }
-
op->flags |= BCH_WRITE_DONE;
if (ret < 0) {
goto again;
bch2_write_done(&op->cl);
} else {
+ bch2_write_queue(op, wp);
continue_at(&op->cl, bch2_write_index, NULL);
}
out_nofs_restore:
goto err;
}
- if (c->opts.nochanges ||
+ if (c->opts.nochanges) {
+ op->error = -BCH_ERR_erofs_no_writes;
+ goto err;
+ }
+
+ if (!(op->flags & BCH_WRITE_MOVE) &&
!bch2_write_ref_tryget(c, BCH_WRITE_REF_write)) {
op->error = -BCH_ERR_erofs_no_writes;
goto err;
op->end_io(op);
}
+const char * const bch2_write_flags[] = {
+#define x(f) #f,
+ BCH_WRITE_FLAGS()
+#undef x
+ NULL
+};
+
+void bch2_write_op_to_text(struct printbuf *out, struct bch_write_op *op)
+{
+ prt_str(out, "pos: ");
+ bch2_bpos_to_text(out, op->pos);
+ prt_newline(out);
+ printbuf_indent_add(out, 2);
+
+ prt_str(out, "started: ");
+ bch2_pr_time_units(out, local_clock() - op->start_time);
+ prt_newline(out);
+
+ prt_str(out, "flags: ");
+ prt_bitflags(out, bch2_write_flags, op->flags);
+ prt_newline(out);
+
+ prt_printf(out, "ref: %u", closure_nr_remaining(&op->cl));
+ prt_newline(out);
+
+ printbuf_indent_sub(out, 2);
+}
+
/* Cache promotion on read */
struct promote_op {
int bch2_fs_io_init(struct bch_fs *c)
{
if (bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio),
- BIOSET_NEED_BVECS) ||
- bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
- BIOSET_NEED_BVECS) ||
- bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio),
- BIOSET_NEED_BVECS) ||
- mempool_init_page_pool(&c->bio_bounce_pages,
+ BIOSET_NEED_BVECS))
+ return -BCH_ERR_ENOMEM_bio_read_init;
+
+ if (bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio),
+ BIOSET_NEED_BVECS))
+ return -BCH_ERR_ENOMEM_bio_read_split_init;
+
+ if (bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio),
+ BIOSET_NEED_BVECS))
+ return -BCH_ERR_ENOMEM_bio_write_init;
+
+ if (mempool_init_page_pool(&c->bio_bounce_pages,
max_t(unsigned,
c->opts.btree_node_size,
c->opts.encoded_extent_max) /
- PAGE_SIZE, 0) ||
- rhashtable_init(&c->promote_table, &bch_promote_params))
- return -ENOMEM;
+ PAGE_SIZE, 0))
+ return -BCH_ERR_ENOMEM_bio_bounce_pages_init;
+
+ if (rhashtable_init(&c->promote_table, &bch_promote_params))
+ return -BCH_ERR_ENOMEM_promote_table_init;
return 0;
}
projects / bcachefs-tools-debian / blobdiff