#include "bcachefs.h"
#include "alloc_foreground.h"
+#include "backpointers.h"
#include "bkey_buf.h"
#include "bset.h"
#include "btree_gc.h"
#include "btree_update.h"
+#include "btree_write_buffer.h"
#include "buckets.h"
#include "disk_groups.h"
#include "ec.h"
/* Stripes btree keys: */
-const char *bch2_stripe_invalid(const struct bch_fs *c, struct bkey_s_c k)
+int bch2_stripe_invalid(const struct bch_fs *c, struct bkey_s_c k,
+ unsigned flags, struct printbuf *err)
{
const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
- if (!bkey_cmp(k.k->p, POS_MIN))
- return "stripe at pos 0";
+ if (bkey_eq(k.k->p, POS_MIN)) {
+ prt_printf(err, "stripe at POS_MIN");
+ return -BCH_ERR_invalid_bkey;
+ }
- if (k.k->p.inode)
- return "invalid stripe key";
+ if (k.k->p.inode) {
+ prt_printf(err, "nonzero inode field");
+ return -BCH_ERR_invalid_bkey;
+ }
- if (bkey_val_bytes(k.k) < sizeof(*s))
- return "incorrect value size";
+ if (bkey_val_bytes(k.k) < sizeof(*s)) {
+ prt_printf(err, "incorrect value size (%zu < %zu)",
+ bkey_val_bytes(k.k), sizeof(*s));
+ return -BCH_ERR_invalid_bkey;
+ }
- if (bkey_val_bytes(k.k) < sizeof(*s) ||
- bkey_val_u64s(k.k) < stripe_val_u64s(s))
- return "incorrect value size";
+ if (bkey_val_u64s(k.k) < stripe_val_u64s(s)) {
+ prt_printf(err, "incorrect value size (%zu < %u)",
+ bkey_val_u64s(k.k), stripe_val_u64s(s));
+ return -BCH_ERR_invalid_bkey;
+ }
- return bch2_bkey_ptrs_invalid(c, k);
+ return bch2_bkey_ptrs_invalid(c, k, flags, err);
}
void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
unsigned i;
- pr_buf(out, "algo %u sectors %u blocks %u:%u csum %u gran %u",
+ prt_printf(out, "algo %u sectors %u blocks %u:%u csum %u gran %u",
s->algorithm,
le16_to_cpu(s->sectors),
s->nr_blocks - s->nr_redundant,
1U << s->csum_granularity_bits);
for (i = 0; i < s->nr_blocks; i++)
- pr_buf(out, " %u:%llu:%u", s->ptrs[i].dev,
+ prt_printf(out, " %u:%llu:%u", s->ptrs[i].dev,
(u64) s->ptrs[i].offset,
stripe_blockcount_get(s, i));
}
struct bch_csum got = ec_block_checksum(buf, i, offset);
if (bch2_crc_cmp(want, got)) {
- char buf2[200];
+ struct printbuf buf2 = PRINTBUF;
- bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(&buf->key.k_i));
+ bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(&buf->key.k_i));
bch_err_ratelimited(c,
"stripe checksum error for %ps at %u:%u: csum type %u, expected %llx got %llx\n%s",
(void *) _RET_IP_, i, j, v->csum_type,
- want.lo, got.lo, buf2);
+ want.lo, got.lo, buf2.buf);
+ printbuf_exit(&buf2);
clear_bit(i, buf->valid);
break;
}
nr_iovecs << PAGE_SHIFT);
struct ec_bio *ec_bio;
- ec_bio = container_of(bio_alloc_bioset(GFP_KERNEL, nr_iovecs,
+ ec_bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev,
+ nr_iovecs,
+ rw,
+ GFP_KERNEL,
&c->ec_bioset),
struct ec_bio, bio);
ec_bio->buf = buf;
ec_bio->idx = idx;
- bio_set_dev(&ec_bio->bio, ca->disk_sb.bdev);
- bio_set_op_attrs(&ec_bio->bio, rw, 0);
-
ec_bio->bio.bi_iter.bi_sector = ptr->offset + buf->offset + (offset >> 9);
ec_bio->bio.bi_end_io = ec_block_endio;
ec_bio->bio.bi_private = cl;
if (!init_heap(&n, max(1024UL, roundup_pow_of_two(idx + 1)), gfp))
return -ENOMEM;
- spin_lock(&c->ec_stripes_heap_lock);
+ mutex_lock(&c->ec_stripes_heap_lock);
if (n.size > h->size) {
memcpy(n.data, h->data, h->used * sizeof(h->data[0]));
n.used = h->used;
swap(*h, n);
}
- spin_unlock(&c->ec_stripes_heap_lock);
+ mutex_unlock(&c->ec_stripes_heap_lock);
free_heap(&n);
}
struct btree_iter *iter)
{
size_t idx = iter->pos.offset;
- int ret = 0;
if (!__ec_stripe_mem_alloc(trans->c, idx, GFP_NOWAIT|__GFP_NOWARN))
- return ret;
+ return 0;
bch2_trans_unlock(trans);
- ret = -EINTR;
- if (!__ec_stripe_mem_alloc(trans->c, idx, GFP_KERNEL))
- return ret;
+ return __ec_stripe_mem_alloc(trans->c, idx, GFP_KERNEL) ?:
+ bch2_trans_relock(trans);
+}
- return -ENOMEM;
+/*
+ * Hash table of open stripes:
+ * Stripes that are being created or modified are kept in a hash table, so that
+ * stripe deletion can skip them.
+ */
+
+static bool __bch2_stripe_is_open(struct bch_fs *c, u64 idx)
+{
+ unsigned hash = hash_64(idx, ilog2(ARRAY_SIZE(c->ec_stripes_new)));
+ struct ec_stripe_new *s;
+
+ hlist_for_each_entry(s, &c->ec_stripes_new[hash], hash)
+ if (s->idx == idx)
+ return true;
+ return false;
+}
+
+static bool bch2_stripe_is_open(struct bch_fs *c, u64 idx)
+{
+ bool ret = false;
+
+ spin_lock(&c->ec_stripes_new_lock);
+ ret = __bch2_stripe_is_open(c, idx);
+ spin_unlock(&c->ec_stripes_new_lock);
+
+ return ret;
+}
+
+static bool bch2_try_open_stripe(struct bch_fs *c,
+ struct ec_stripe_new *s,
+ u64 idx)
+{
+ bool ret;
+
+ spin_lock(&c->ec_stripes_new_lock);
+ ret = !__bch2_stripe_is_open(c, idx);
+ if (ret) {
+ unsigned hash = hash_64(idx, ilog2(ARRAY_SIZE(c->ec_stripes_new)));
+
+ s->idx = idx;
+ hlist_add_head(&s->hash, &c->ec_stripes_new[hash]);
+ }
+ spin_unlock(&c->ec_stripes_new_lock);
+
+ return ret;
+}
+
+static void bch2_stripe_close(struct bch_fs *c, struct ec_stripe_new *s)
+{
+ BUG_ON(!s->idx);
+
+ spin_lock(&c->ec_stripes_new_lock);
+ hlist_del(&s->hash);
+ spin_unlock(&c->ec_stripes_new_lock);
+
+ s->idx = 0;
}
-static ssize_t stripe_idx_to_delete(struct bch_fs *c)
+/* Heap of all existing stripes, ordered by blocks_nonempty */
+
+static u64 stripe_idx_to_delete(struct bch_fs *c)
{
ec_stripes_heap *h = &c->ec_stripes_heap;
+ size_t heap_idx;
- return h->used && h->data[0].blocks_nonempty == 0
- ? h->data[0].idx : -1;
+ lockdep_assert_held(&c->ec_stripes_heap_lock);
+
+ for (heap_idx = 0; heap_idx < h->used; heap_idx++)
+ if (h->data[heap_idx].blocks_nonempty == 0 &&
+ !bch2_stripe_is_open(c, h->data[heap_idx].idx))
+ return h->data[heap_idx].idx;
+
+ return 0;
}
static inline int ec_stripes_heap_cmp(ec_stripes_heap *h,
ec_stripes_heap *h = &c->ec_stripes_heap;
struct stripe *m = genradix_ptr(&c->stripes, idx);
- BUG_ON(!m->alive);
BUG_ON(m->heap_idx >= h->used);
BUG_ON(h->data[m->heap_idx].idx != idx);
}
void bch2_stripes_heap_del(struct bch_fs *c,
struct stripe *m, size_t idx)
{
- if (!m->on_heap)
- return;
-
- m->on_heap = false;
-
+ mutex_lock(&c->ec_stripes_heap_lock);
heap_verify_backpointer(c, idx);
heap_del(&c->ec_stripes_heap, m->heap_idx,
ec_stripes_heap_cmp,
ec_stripes_heap_set_backpointer);
+ mutex_unlock(&c->ec_stripes_heap_lock);
}
void bch2_stripes_heap_insert(struct bch_fs *c,
struct stripe *m, size_t idx)
{
- if (m->on_heap)
- return;
-
+ mutex_lock(&c->ec_stripes_heap_lock);
BUG_ON(heap_full(&c->ec_stripes_heap));
- m->on_heap = true;
-
heap_add(&c->ec_stripes_heap, ((struct ec_stripe_heap_entry) {
.idx = idx,
.blocks_nonempty = m->blocks_nonempty,
ec_stripes_heap_set_backpointer);
heap_verify_backpointer(c, idx);
+ mutex_unlock(&c->ec_stripes_heap_lock);
}
void bch2_stripes_heap_update(struct bch_fs *c,
struct stripe *m, size_t idx)
{
ec_stripes_heap *h = &c->ec_stripes_heap;
+ bool do_deletes;
size_t i;
- if (!m->on_heap)
- return;
-
+ mutex_lock(&c->ec_stripes_heap_lock);
heap_verify_backpointer(c, idx);
h->data[m->heap_idx].blocks_nonempty = m->blocks_nonempty;
heap_verify_backpointer(c, idx);
- if (stripe_idx_to_delete(c) >= 0 &&
- !percpu_ref_is_dying(&c->writes))
- schedule_work(&c->ec_stripe_delete_work);
+ do_deletes = stripe_idx_to_delete(c) != 0;
+ mutex_unlock(&c->ec_stripes_heap_lock);
+
+ if (do_deletes)
+ bch2_do_stripe_deletes(c);
}
/* stripe deletion */
-static int ec_stripe_delete(struct bch_fs *c, size_t idx)
+static int ec_stripe_delete(struct btree_trans *trans, u64 idx)
{
- return bch2_btree_delete_range(c, BTREE_ID_stripes,
- POS(0, idx),
- POS(0, idx + 1),
- 0, NULL);
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bkey_s_c_stripe s;
+ int ret;
+
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes, POS(0, idx),
+ BTREE_ITER_INTENT);
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto err;
+
+ if (k.k->type != KEY_TYPE_stripe) {
+ bch2_fs_inconsistent(c, "attempting to delete nonexistent stripe %llu", idx);
+ ret = -EINVAL;
+ goto err;
+ }
+
+ s = bkey_s_c_to_stripe(k);
+ for (unsigned i = 0; i < s.v->nr_blocks; i++)
+ if (stripe_blockcount_get(s.v, i)) {
+ struct printbuf buf = PRINTBUF;
+
+ bch2_bkey_val_to_text(&buf, c, k);
+ bch2_fs_inconsistent(c, "attempting to delete nonempty stripe %s", buf.buf);
+ printbuf_exit(&buf);
+ ret = -EINVAL;
+ goto err;
+ }
+
+ ret = bch2_btree_delete_at(trans, &iter, 0);
+err:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
}
static void ec_stripe_delete_work(struct work_struct *work)
{
struct bch_fs *c =
container_of(work, struct bch_fs, ec_stripe_delete_work);
- ssize_t idx;
+ struct btree_trans trans;
+ int ret;
+ u64 idx;
+
+ bch2_trans_init(&trans, c, 0, 0);
while (1) {
- spin_lock(&c->ec_stripes_heap_lock);
+ mutex_lock(&c->ec_stripes_heap_lock);
idx = stripe_idx_to_delete(c);
- if (idx < 0) {
- spin_unlock(&c->ec_stripes_heap_lock);
- break;
- }
-
- bch2_stripes_heap_del(c, genradix_ptr(&c->stripes, idx), idx);
- spin_unlock(&c->ec_stripes_heap_lock);
+ mutex_unlock(&c->ec_stripes_heap_lock);
- if (ec_stripe_delete(c, idx))
+ if (!idx)
break;
- }
-}
-
-/* stripe creation: */
-
-static int ec_stripe_bkey_insert(struct btree_trans *trans,
- struct bkey_i_stripe *stripe,
- struct disk_reservation *res)
-{
- struct bch_fs *c = trans->c;
- struct btree_iter iter;
- struct bkey_s_c k;
- struct bpos min_pos = POS(0, 1);
- struct bpos start_pos = bpos_max(min_pos, POS(0, c->ec_stripe_hint));
- int ret;
- for_each_btree_key(trans, iter, BTREE_ID_stripes, start_pos,
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
- if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0) {
- if (start_pos.offset) {
- start_pos = min_pos;
- bch2_btree_iter_set_pos(&iter, start_pos);
- continue;
- }
-
- ret = -ENOSPC;
+ ret = commit_do(&trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+ ec_stripe_delete(&trans, idx));
+ if (ret) {
+ bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
break;
}
-
- if (bkey_deleted(k.k))
- goto found_slot;
}
- goto err;
-found_slot:
- start_pos = iter.pos;
-
- ret = ec_stripe_mem_alloc(trans, &iter);
- if (ret)
- goto err;
-
- stripe->k.p = iter.pos;
-
- ret = bch2_trans_update(trans, &iter, &stripe->k_i, 0);
+ bch2_trans_exit(&trans);
- c->ec_stripe_hint = start_pos.offset;
-err:
- bch2_trans_iter_exit(trans, &iter);
+ bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
+}
- return ret;
+void bch2_do_stripe_deletes(struct bch_fs *c)
+{
+ if (bch2_write_ref_tryget(c, BCH_WRITE_REF_stripe_delete) &&
+ !schedule_work(&c->ec_stripe_delete_work))
+ bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
}
-static int ec_stripe_bkey_update(struct btree_trans *trans,
- struct bkey_i_stripe *new,
- struct disk_reservation *res)
+/* stripe creation: */
+
+static int ec_stripe_key_update(struct btree_trans *trans,
+ struct bkey_i_stripe *new,
+ bool create)
{
+ struct bch_fs *c = trans->c;
struct btree_iter iter;
struct bkey_s_c k;
- const struct bch_stripe *existing;
- unsigned i;
int ret;
bch2_trans_iter_init(trans, &iter, BTREE_ID_stripes,
if (ret)
goto err;
- if (!k.k || k.k->type != KEY_TYPE_stripe) {
- bch_err(trans->c, "error updating stripe: not found");
- ret = -ENOENT;
+ if (k.k->type != (create ? KEY_TYPE_deleted : KEY_TYPE_stripe)) {
+ bch2_fs_inconsistent(c, "error %s stripe: got existing key type %s",
+ create ? "creating" : "updating",
+ bch2_bkey_types[k.k->type]);
+ ret = -EINVAL;
goto err;
}
- existing = bkey_s_c_to_stripe(k).v;
+ if (k.k->type == KEY_TYPE_stripe) {
+ const struct bch_stripe *old = bkey_s_c_to_stripe(k).v;
+ unsigned i;
- if (existing->nr_blocks != new->v.nr_blocks) {
- bch_err(trans->c, "error updating stripe: nr_blocks does not match");
- ret = -EINVAL;
- goto err;
- }
+ if (old->nr_blocks != new->v.nr_blocks) {
+ bch_err(c, "error updating stripe: nr_blocks does not match");
+ ret = -EINVAL;
+ goto err;
+ }
- for (i = 0; i < new->v.nr_blocks; i++)
- stripe_blockcount_set(&new->v, i,
- stripe_blockcount_get(existing, i));
+ for (i = 0; i < new->v.nr_blocks; i++)
+ stripe_blockcount_set(&new->v, i, stripe_blockcount_get(old, i));
+ }
ret = bch2_trans_update(trans, &iter, &new->k_i, 0);
err:
};
}
-static int ec_stripe_update_ptrs(struct bch_fs *c,
- struct ec_stripe_buf *s,
- struct bkey *pos)
+static int ec_stripe_update_extent(struct btree_trans *trans,
+ struct bpos bucket, u8 gen,
+ struct ec_stripe_buf *s,
+ u64 *bp_offset)
{
- struct btree_trans trans;
+ struct bch_fs *c = trans->c;
+ struct bch_backpointer bp;
struct btree_iter iter;
struct bkey_s_c k;
- struct bkey_s_extent e;
- struct bkey_buf sk;
- struct bpos next_pos;
- int ret = 0, dev, block;
-
- bch2_bkey_buf_init(&sk);
- bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+ const struct bch_extent_ptr *ptr_c;
+ struct bch_extent_ptr *ptr, *ec_ptr = NULL;
+ struct bkey_i *n;
+ int ret, dev, block;
- /* XXX this doesn't support the reflink btree */
+ ret = bch2_get_next_backpointer(trans, bucket, gen,
+ bp_offset, &bp, BTREE_ITER_CACHED);
+ if (ret)
+ return ret;
+ if (*bp_offset == U64_MAX)
+ return 0;
- bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
- bkey_start_pos(pos),
- BTREE_ITER_INTENT);
-retry:
- while (bch2_trans_begin(&trans),
- (k = bch2_btree_iter_peek(&iter)).k &&
- !(ret = bkey_err(k)) &&
- bkey_cmp(bkey_start_pos(k.k), pos->p) < 0) {
- const struct bch_extent_ptr *ptr_c;
- struct bch_extent_ptr *ptr, *ec_ptr = NULL;
-
- if (extent_has_stripe_ptr(k, s->key.k.p.offset)) {
- bch2_btree_iter_advance(&iter);
- continue;
- }
+ if (bch2_fs_inconsistent_on(bp.level, c, "found btree node in erasure coded bucket!?"))
+ return -EIO;
- ptr_c = bkey_matches_stripe(&s->key.v, k, &block);
+ k = bch2_backpointer_get_key(trans, &iter, bucket, *bp_offset, bp);
+ ret = bkey_err(k);
+ if (ret)
+ return ret;
+ if (!k.k) {
/*
- * It doesn't generally make sense to erasure code cached ptrs:
- * XXX: should we be incrementing a counter?
+ * extent no longer exists - we could flush the btree
+ * write buffer and retry to verify, but no need:
*/
- if (!ptr_c || ptr_c->cached) {
- bch2_btree_iter_advance(&iter);
- continue;
- }
+ return 0;
+ }
- dev = s->key.v.ptrs[block].dev;
+ if (extent_has_stripe_ptr(k, s->key.k.p.offset))
+ goto out;
- bch2_bkey_buf_reassemble(&sk, c, k);
- e = bkey_i_to_s_extent(sk.k);
+ ptr_c = bkey_matches_stripe(&s->key.v, k, &block);
+ /*
+ * It doesn't generally make sense to erasure code cached ptrs:
+ * XXX: should we be incrementing a counter?
+ */
+ if (!ptr_c || ptr_c->cached)
+ goto out;
- bch2_bkey_drop_ptrs(e.s, ptr, ptr->dev != dev);
- ec_ptr = (void *) bch2_bkey_has_device(e.s_c, dev);
- BUG_ON(!ec_ptr);
+ dev = s->key.v.ptrs[block].dev;
+
+ n = bch2_bkey_make_mut(trans, k);
+ ret = PTR_ERR_OR_ZERO(n);
+ if (ret)
+ goto out;
- extent_stripe_ptr_add(e, s, ec_ptr, block);
+ bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, ptr->dev != dev);
+ ec_ptr = (void *) bch2_bkey_has_device(bkey_i_to_s_c(n), dev);
+ BUG_ON(!ec_ptr);
- bch2_btree_iter_set_pos(&iter, bkey_start_pos(&sk.k->k));
- next_pos = sk.k->k.p;
+ extent_stripe_ptr_add(bkey_i_to_s_extent(n), s, ec_ptr, block);
- ret = bch2_btree_iter_traverse(&iter) ?:
- bch2_trans_update(&trans, &iter, sk.k, 0) ?:
- bch2_trans_commit(&trans, NULL, NULL,
- BTREE_INSERT_NOFAIL);
- if (!ret)
- bch2_btree_iter_set_pos(&iter, next_pos);
+ ret = bch2_trans_update(trans, &iter, n, 0);
+out:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
+
+static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_buf *s,
+ unsigned block)
+{
+ struct bch_fs *c = trans->c;
+ struct bch_extent_ptr bucket = s->key.v.ptrs[block];
+ struct bpos bucket_pos = PTR_BUCKET_POS(c, &bucket);
+ u64 bp_offset = 0;
+ int ret = 0;
+
+ while (1) {
+ ret = commit_do(trans, NULL, NULL,
+ BTREE_INSERT_NOFAIL,
+ ec_stripe_update_extent(trans, bucket_pos, bucket.gen,
+ s, &bp_offset));
if (ret)
break;
+ if (bp_offset == U64_MAX)
+ break;
+
+ bp_offset++;
}
- if (ret == -EINTR)
- goto retry;
- bch2_trans_iter_exit(&trans, &iter);
+ return ret;
+}
+
+static int ec_stripe_update_extents(struct bch_fs *c, struct ec_stripe_buf *s)
+{
+ struct btree_trans trans;
+ struct bch_stripe *v = &s->key.v;
+ unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
+ int ret = 0;
+
+ bch2_trans_init(&trans, c, 0, 0);
+
+ ret = bch2_btree_write_buffer_flush(&trans);
+ if (ret)
+ goto err;
+
+ for (i = 0; i < nr_data; i++) {
+ ret = ec_stripe_update_bucket(&trans, s, i);
+ if (ret)
+ break;
+ }
+err:
bch2_trans_exit(&trans);
- bch2_bkey_buf_exit(&sk, c);
return ret;
}
{
struct bch_fs *c = s->c;
struct open_bucket *ob;
- struct bkey_i *k;
- struct stripe *m;
struct bch_stripe *v = &s->new_stripe.key.v;
unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
int ret;
closure_sync(&s->iodone);
if (s->err) {
- if (s->err != -EROFS)
+ if (!bch2_err_matches(s->err, EROFS))
bch_err(c, "error creating stripe: error writing data buckets");
goto err;
}
BUG_ON(!s->allocated);
- if (!percpu_ref_tryget(&c->writes))
- goto err;
-
ec_generate_ec(&s->new_stripe);
ec_generate_checksums(&s->new_stripe);
if (ec_nr_failed(&s->new_stripe)) {
bch_err(c, "error creating stripe: error writing redundancy buckets");
- goto err_put_writes;
+ goto err;
}
ret = bch2_trans_do(c, &s->res, NULL, BTREE_INSERT_NOFAIL,
- s->have_existing_stripe
- ? ec_stripe_bkey_update(&trans, &s->new_stripe.key, &s->res)
- : ec_stripe_bkey_insert(&trans, &s->new_stripe.key, &s->res));
+ ec_stripe_key_update(&trans, &s->new_stripe.key,
+ !s->have_existing_stripe));
if (ret) {
bch_err(c, "error creating stripe: error creating stripe key");
- goto err_put_writes;
+ goto err;
}
- for_each_keylist_key(&s->keys, k) {
- ret = ec_stripe_update_ptrs(c, &s->new_stripe, &k->k);
- if (ret) {
- bch_err(c, "error creating stripe: error %i updating pointers", ret);
- break;
- }
+ ret = ec_stripe_update_extents(c, &s->new_stripe);
+ if (ret) {
+ bch_err(c, "error creating stripe: error updating pointers: %s",
+ bch2_err_str(ret));
+ goto err;
}
- spin_lock(&c->ec_stripes_heap_lock);
- m = genradix_ptr(&c->stripes, s->new_stripe.key.k.p.offset);
-
- BUG_ON(m->on_heap);
- bch2_stripes_heap_insert(c, m, s->new_stripe.key.k.p.offset);
- spin_unlock(&c->ec_stripes_heap_lock);
-err_put_writes:
- percpu_ref_put(&c->writes);
+ bch2_stripe_close(c, s);
err:
bch2_disk_reservation_put(c, &s->res);
}
}
- bch2_keylist_free(&s->keys, s->inline_keys);
-
ec_stripe_buf_exit(&s->existing_stripe);
ec_stripe_buf_exit(&s->new_stripe);
closure_debug_destroy(&s->iodone);
kfree(s);
}
-static void ec_stripe_create_work(struct work_struct *work)
+static struct ec_stripe_new *get_pending_stripe(struct bch_fs *c)
{
- struct bch_fs *c = container_of(work,
- struct bch_fs, ec_stripe_create_work);
- struct ec_stripe_new *s, *n;
-restart:
+ struct ec_stripe_new *s;
+
mutex_lock(&c->ec_stripe_new_lock);
- list_for_each_entry_safe(s, n, &c->ec_stripe_new_list, list)
+ list_for_each_entry(s, &c->ec_stripe_new_list, list)
if (!atomic_read(&s->pin)) {
list_del(&s->list);
- mutex_unlock(&c->ec_stripe_new_lock);
- ec_stripe_create(s);
- goto restart;
+ goto out;
}
+ s = NULL;
+out:
mutex_unlock(&c->ec_stripe_new_lock);
+
+ return s;
+}
+
+static void ec_stripe_create_work(struct work_struct *work)
+{
+ struct bch_fs *c = container_of(work,
+ struct bch_fs, ec_stripe_create_work);
+ struct ec_stripe_new *s;
+
+ while ((s = get_pending_stripe(c)))
+ ec_stripe_create(s);
+
+ bch2_write_ref_put(c, BCH_WRITE_REF_stripe_create);
+}
+
+void bch2_ec_do_stripe_creates(struct bch_fs *c)
+{
+ bch2_write_ref_get(c, BCH_WRITE_REF_stripe_create);
+
+ if (!queue_work(system_long_wq, &c->ec_stripe_create_work))
+ bch2_write_ref_put(c, BCH_WRITE_REF_stripe_create);
}
static void ec_stripe_new_put(struct bch_fs *c, struct ec_stripe_new *s)
{
BUG_ON(atomic_read(&s->pin) <= 0);
- if (atomic_dec_and_test(&s->pin)) {
- BUG_ON(!s->pending);
- queue_work(system_long_wq, &c->ec_stripe_create_work);
- }
+ if (atomic_dec_and_test(&s->pin))
+ bch2_ec_do_stripe_creates(c);
}
static void ec_stripe_set_pending(struct bch_fs *c, struct ec_stripe_head *h)
return ob->ec->new_stripe.data[ob->ec_idx] + (offset << 9);
}
-void bch2_ob_add_backpointer(struct bch_fs *c, struct open_bucket *ob,
- struct bkey *k)
-{
- struct ec_stripe_new *ec = ob->ec;
-
- if (!ec)
- return;
-
- mutex_lock(&ec->lock);
-
- if (bch2_keylist_realloc(&ec->keys, ec->inline_keys,
- ARRAY_SIZE(ec->inline_keys),
- BKEY_U64s)) {
- BUG();
- }
-
- bkey_init(&ec->keys.top->k);
- ec->keys.top->k.p = k->p;
- ec->keys.top->k.size = k->size;
- bch2_keylist_push(&ec->keys);
-
- mutex_unlock(&ec->lock);
-}
-
static int unsigned_cmp(const void *_l, const void *_r)
{
unsigned l = *((const unsigned *) _l);
BCH_BKEY_PTRS_MAX) - h->redundancy;
s->nr_parity = h->redundancy;
- bch2_keylist_init(&s->keys, s->inline_keys);
-
ec_stripe_key_init(c, &s->new_stripe.key, s->nr_data,
s->nr_parity, h->blocksize);
return NULL;
mutex_init(&h->lock);
- mutex_lock(&h->lock);
+ BUG_ON(!mutex_trylock(&h->lock));
h->target = target;
h->algo = algo;
mutex_unlock(&h->lock);
}
-struct ec_stripe_head *__bch2_ec_stripe_head_get(struct bch_fs *c,
+struct ec_stripe_head *__bch2_ec_stripe_head_get(struct btree_trans *trans,
unsigned target,
unsigned algo,
unsigned redundancy,
bool copygc)
{
+ struct bch_fs *c = trans->c;
struct ec_stripe_head *h;
+ int ret;
if (!redundancy)
return NULL;
- mutex_lock(&c->ec_stripe_head_lock);
+ ret = bch2_trans_mutex_lock(trans, &c->ec_stripe_head_lock);
+ if (ret)
+ return ERR_PTR(ret);
+
list_for_each_entry(h, &c->ec_stripe_head_list, list)
if (h->target == target &&
h->algo == algo &&
h->redundancy == redundancy &&
h->copygc == copygc) {
- mutex_lock(&h->lock);
+ ret = bch2_trans_mutex_lock(trans, &h->lock);
+ if (ret)
+ h = ERR_PTR(ret);
goto found;
}
return h;
}
-static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
+static int new_stripe_alloc_buckets(struct btree_trans *trans, struct ec_stripe_head *h,
struct closure *cl)
{
+ struct bch_fs *c = trans->c;
struct bch_devs_mask devs = h->devs;
struct open_bucket *ob;
struct open_buckets buckets;
BUG_ON(nr_have_data > h->s->nr_data);
BUG_ON(nr_have_parity > h->s->nr_parity);
- percpu_down_read(&c->mark_lock);
- rcu_read_lock();
-
buckets.nr = 0;
if (nr_have_parity < h->s->nr_parity) {
- ret = bch2_bucket_alloc_set(c, &buckets,
+ ret = bch2_bucket_alloc_set_trans(trans, &buckets,
&h->parity_stripe,
&devs,
h->s->nr_parity,
&nr_have_parity,
&have_cache,
h->copygc
- ? RESERVE_MOVINGGC
- : RESERVE_NONE,
+ ? RESERVE_movinggc
+ : RESERVE_none,
0,
cl);
}
if (ret)
- goto err;
+ return ret;
}
buckets.nr = 0;
if (nr_have_data < h->s->nr_data) {
- ret = bch2_bucket_alloc_set(c, &buckets,
+ ret = bch2_bucket_alloc_set_trans(trans, &buckets,
&h->block_stripe,
&devs,
h->s->nr_data,
&nr_have_data,
&have_cache,
h->copygc
- ? RESERVE_MOVINGGC
- : RESERVE_NONE,
+ ? RESERVE_movinggc
+ : RESERVE_none,
0,
cl);
}
if (ret)
- goto err;
+ return ret;
}
-err:
- rcu_read_unlock();
- percpu_up_read(&c->mark_lock);
- return ret;
+
+ return 0;
}
/* XXX: doesn't obey target: */
if (may_create_new_stripe(c))
return -1;
- spin_lock(&c->ec_stripes_heap_lock);
+ mutex_lock(&c->ec_stripes_heap_lock);
for (heap_idx = 0; heap_idx < h->used; heap_idx++) {
/* No blocks worth reusing, stripe will just be deleted: */
if (!h->data[heap_idx].blocks_nonempty)
continue;
stripe_idx = h->data[heap_idx].idx;
+
m = genradix_ptr(&c->stripes, stripe_idx);
if (m->algorithm == head->algo &&
m->nr_redundant == head->redundancy &&
m->sectors == head->blocksize &&
- m->blocks_nonempty < m->nr_blocks - m->nr_redundant) {
- bch2_stripes_heap_del(c, m, stripe_idx);
+ m->blocks_nonempty < m->nr_blocks - m->nr_redundant &&
+ bch2_try_open_stripe(c, head->s, stripe_idx)) {
ret = stripe_idx;
break;
}
}
- spin_unlock(&c->ec_stripes_heap_lock);
+ mutex_unlock(&c->ec_stripes_heap_lock);
return ret;
}
-static int __bch2_ec_stripe_head_reuse(struct bch_fs *c,
- struct ec_stripe_head *h)
+static int __bch2_ec_stripe_head_reuse(struct bch_fs *c, struct ec_stripe_head *h)
{
unsigned i;
s64 idx;
int ret;
idx = get_existing_stripe(c, h);
- if (idx < 0) {
- bch_err(c, "failed to find an existing stripe");
- return -ENOSPC;
- }
+ if (idx < 0)
+ return -BCH_ERR_ENOSPC_stripe_reuse;
h->s->have_existing_stripe = true;
ret = get_stripe_key(c, idx, &h->s->existing_stripe);
}
bkey_copy(&h->s->new_stripe.key.k_i,
- &h->s->existing_stripe.key.k_i);
+ &h->s->existing_stripe.key.k_i);
return 0;
}
-static int __bch2_ec_stripe_head_reserve(struct bch_fs *c,
- struct ec_stripe_head *h)
+static int __bch2_ec_stripe_head_reserve(struct btree_trans *trans, struct ec_stripe_head *h)
{
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bpos min_pos = POS(0, 1);
+ struct bpos start_pos = bpos_max(min_pos, POS(0, c->ec_stripe_hint));
int ret;
+ BUG_ON(h->s->res.sectors);
+
ret = bch2_disk_reservation_get(c, &h->s->res,
- h->blocksize,
- h->s->nr_parity, 0);
+ h->blocksize,
+ h->s->nr_parity, 0);
+ if (ret)
+ return ret;
+
+ for_each_btree_key_norestart(trans, iter, BTREE_ID_stripes, start_pos,
+ BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
+ if (bkey_gt(k.k->p, POS(0, U32_MAX))) {
+ if (start_pos.offset) {
+ start_pos = min_pos;
+ bch2_btree_iter_set_pos(&iter, start_pos);
+ continue;
+ }
+
+ ret = -BCH_ERR_ENOSPC_stripe_create;
+ break;
+ }
+ if (bkey_deleted(k.k) &&
+ bch2_try_open_stripe(c, h->s, k.k->p.offset))
+ break;
+ }
+
+ c->ec_stripe_hint = iter.pos.offset;
+
+ if (ret)
+ goto err;
+
+ ret = ec_stripe_mem_alloc(trans, &iter);
if (ret) {
- /*
- * This means we need to wait for copygc to
- * empty out buckets from existing stripes:
- */
- bch_err(c, "failed to reserve stripe");
+ bch2_stripe_close(c, h->s);
+ goto err;
}
+ h->s->new_stripe.key.k.p = iter.pos;
+out:
+ bch2_trans_iter_exit(trans, &iter);
return ret;
+err:
+ bch2_disk_reservation_put(c, &h->s->res);
+ goto out;
}
-struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *c,
+struct ec_stripe_head *bch2_ec_stripe_head_get(struct btree_trans *trans,
unsigned target,
unsigned algo,
unsigned redundancy,
bool copygc,
struct closure *cl)
{
+ struct bch_fs *c = trans->c;
struct ec_stripe_head *h;
int ret;
bool needs_stripe_new;
- h = __bch2_ec_stripe_head_get(c, target, algo, redundancy, copygc);
- if (!h) {
+ h = __bch2_ec_stripe_head_get(trans, target, algo, redundancy, copygc);
+ if (!h)
bch_err(c, "no stripe head");
- return NULL;
- }
+ if (IS_ERR_OR_NULL(h))
+ return h;
needs_stripe_new = !h->s;
if (needs_stripe_new) {
*/
ret = 0;
if (!h->s->allocated && !h->s->res.sectors && !h->s->have_existing_stripe)
- ret = __bch2_ec_stripe_head_reserve(c, h);
+ ret = __bch2_ec_stripe_head_reserve(trans, h);
+ if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+ goto err;
+
if (ret && needs_stripe_new)
ret = __bch2_ec_stripe_head_reuse(c, h);
- if (ret)
+ if (ret) {
+ bch_err_ratelimited(c, "failed to get stripe: %s", bch2_err_str(ret));
goto err;
+ }
if (!h->s->allocated) {
- ret = new_stripe_alloc_buckets(c, h, cl);
+ ret = new_stripe_alloc_buckets(trans, h, cl);
if (ret)
goto err;
h->s->allocated = true;
}
+ BUG_ON(trans->restarted);
return h;
-
err:
bch2_ec_stripe_head_put(c, h);
return ERR_PTR(ret);
mutex_unlock(&c->ec_stripe_head_lock);
}
-void bch2_stripes_heap_start(struct bch_fs *c)
-{
- struct genradix_iter iter;
- struct stripe *m;
-
- genradix_for_each(&c->stripes, iter, m)
- if (m->alive)
- bch2_stripes_heap_insert(c, m, iter.pos);
-}
-
-static int bch2_stripes_read_fn(struct btree_trans *trans, struct bkey_s_c k)
+int bch2_stripes_read(struct bch_fs *c)
{
+ struct btree_trans trans;
+ struct btree_iter iter;
+ struct bkey_s_c k;
const struct bch_stripe *s;
- struct bch_fs *c = trans->c;
struct stripe *m;
unsigned i;
- int ret = 0;
-
- if (k.k->type != KEY_TYPE_stripe)
- return 0;
+ int ret;
- ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL);
- if (ret)
- return ret;
+ bch2_trans_init(&trans, c, 0, 0);
- s = bkey_s_c_to_stripe(k).v;
+ for_each_btree_key(&trans, iter, BTREE_ID_stripes, POS_MIN,
+ BTREE_ITER_PREFETCH, k, ret) {
+ if (k.k->type != KEY_TYPE_stripe)
+ continue;
- m = genradix_ptr(&c->stripes, k.k->p.offset);
- m->alive = true;
- m->sectors = le16_to_cpu(s->sectors);
- m->algorithm = s->algorithm;
- m->nr_blocks = s->nr_blocks;
- m->nr_redundant = s->nr_redundant;
- m->blocks_nonempty = 0;
+ ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL);
+ if (ret)
+ break;
- for (i = 0; i < s->nr_blocks; i++)
- m->blocks_nonempty += !!stripe_blockcount_get(s, i);
+ s = bkey_s_c_to_stripe(k).v;
- spin_lock(&c->ec_stripes_heap_lock);
- bch2_stripes_heap_update(c, m, k.k->p.offset);
- spin_unlock(&c->ec_stripes_heap_lock);
+ m = genradix_ptr(&c->stripes, k.k->p.offset);
+ m->sectors = le16_to_cpu(s->sectors);
+ m->algorithm = s->algorithm;
+ m->nr_blocks = s->nr_blocks;
+ m->nr_redundant = s->nr_redundant;
+ m->blocks_nonempty = 0;
- return ret;
-}
+ for (i = 0; i < s->nr_blocks; i++)
+ m->blocks_nonempty += !!stripe_blockcount_get(s, i);
-int bch2_stripes_read(struct bch_fs *c)
-{
- struct btree_trans trans;
- int ret;
+ bch2_stripes_heap_insert(c, m, k.k->p.offset);
+ }
+ bch2_trans_iter_exit(&trans, &iter);
- bch2_trans_init(&trans, c, 0, 0);
- ret = bch2_btree_and_journal_walk(&trans, BTREE_ID_stripes,
- bch2_stripes_read_fn);
bch2_trans_exit(&trans);
+
if (ret)
bch_err(c, "error reading stripes: %i", ret);
struct stripe *m;
size_t i;
- spin_lock(&c->ec_stripes_heap_lock);
+ mutex_lock(&c->ec_stripes_heap_lock);
for (i = 0; i < min_t(size_t, h->used, 20); i++) {
m = genradix_ptr(&c->stripes, h->data[i].idx);
- pr_buf(out, "%zu %u/%u+%u\n", h->data[i].idx,
+ prt_printf(out, "%zu %u/%u+%u\n", h->data[i].idx,
h->data[i].blocks_nonempty,
m->nr_blocks - m->nr_redundant,
m->nr_redundant);
}
- spin_unlock(&c->ec_stripes_heap_lock);
+ mutex_unlock(&c->ec_stripes_heap_lock);
}
void bch2_new_stripes_to_text(struct printbuf *out, struct bch_fs *c)
mutex_lock(&c->ec_stripe_head_lock);
list_for_each_entry(h, &c->ec_stripe_head_list, list) {
- pr_buf(out, "target %u algo %u redundancy %u:\n",
+ prt_printf(out, "target %u algo %u redundancy %u:\n",
h->target, h->algo, h->redundancy);
if (h->s)
- pr_buf(out, "\tpending: blocks %u+%u allocated %u\n",
+ prt_printf(out, "\tpending: blocks %u+%u allocated %u\n",
h->s->nr_data, h->s->nr_parity,
bitmap_weight(h->s->blocks_allocated,
h->s->nr_data));
mutex_lock(&c->ec_stripe_new_lock);
list_for_each_entry(s, &c->ec_stripe_new_list, list) {
- pr_buf(out, "\tin flight: blocks %u+%u pin %u\n",
+ prt_printf(out, "\tin flight: blocks %u+%u pin %u\n",
s->nr_data, s->nr_parity,
atomic_read(&s->pin));
}
bioset_exit(&c->ec_bioset);
}
-int bch2_fs_ec_init(struct bch_fs *c)
+void bch2_fs_ec_init_early(struct bch_fs *c)
{
INIT_WORK(&c->ec_stripe_create_work, ec_stripe_create_work);
INIT_WORK(&c->ec_stripe_delete_work, ec_stripe_delete_work);
+}
+int bch2_fs_ec_init(struct bch_fs *c)
+{
return bioset_init(&c->ec_bioset, 1, offsetof(struct ec_bio, bio),
BIOSET_NEED_BVECS);
}
projects / bcachefs-tools-debian / blobdiff