#include "btree_update.h"
#include "btree_write_buffer.h"
#include "buckets.h"
+#include "checksum.h"
#include "disk_groups.h"
#include "ec.h"
#include "error.h"
-#include "io.h"
+#include "io_read.h"
#include "keylist.h"
#include "recovery.h"
#include "replicas.h"
/* Stripes btree keys: */
-int bch2_stripe_invalid(const struct bch_fs *c, struct bkey_s_c k,
+int bch2_stripe_invalid(struct bch_fs *c, struct bkey_s_c k,
enum bkey_invalid_flags flags,
struct printbuf *err)
{
const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
+ int ret = 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) {
- prt_printf(err, "nonzero inode field");
- return -BCH_ERR_invalid_bkey;
- }
+ bkey_fsck_err_on(bkey_eq(k.k->p, POS_MIN) ||
+ bpos_gt(k.k->p, POS(0, U32_MAX)), c, err,
+ stripe_pos_bad,
+ "stripe at bad pos");
- 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;
- }
+ bkey_fsck_err_on(bkey_val_u64s(k.k) < stripe_val_u64s(s), c, err,
+ stripe_val_size_bad,
+ "incorrect value size (%zu < %u)",
+ bkey_val_u64s(k.k), stripe_val_u64s(s));
- return bch2_bkey_ptrs_invalid(c, k, flags, err);
+ ret = bch2_bkey_ptrs_invalid(c, k, flags, err);
+fsck_err:
+ return ret;
}
void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
prt_printf(out, " %u:%llu:%u", ptr->dev, b, offset);
if (i < nr_data)
prt_printf(out, "#%u", stripe_blockcount_get(s, i));
+ prt_printf(out, " gen %u", ptr->gen);
if (ptr_stale(ca, ptr))
prt_printf(out, " stale");
}
static void ec_stripe_buf_exit(struct ec_stripe_buf *buf)
{
- unsigned i;
+ if (buf->key.k.type == KEY_TYPE_stripe) {
+ struct bkey_i_stripe *s = bkey_i_to_stripe(&buf->key);
+ unsigned i;
- for (i = 0; i < buf->key.v.nr_blocks; i++) {
- kvpfree(buf->data[i], buf->size << 9);
- buf->data[i] = NULL;
+ for (i = 0; i < s->v.nr_blocks; i++) {
+ kvpfree(buf->data[i], buf->size << 9);
+ buf->data[i] = NULL;
+ }
}
}
static int ec_stripe_buf_init(struct ec_stripe_buf *buf,
unsigned offset, unsigned size)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned csum_granularity = 1U << v->csum_granularity_bits;
unsigned end = offset + size;
unsigned i;
memset(buf->valid, 0xFF, sizeof(buf->valid));
- for (i = 0; i < buf->key.v.nr_blocks; i++) {
+ for (i = 0; i < v->nr_blocks; i++) {
buf->data[i] = kvpmalloc(buf->size << 9, GFP_KERNEL);
if (!buf->data[i])
goto err;
static struct bch_csum ec_block_checksum(struct ec_stripe_buf *buf,
unsigned block, unsigned offset)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned csum_granularity = 1 << v->csum_granularity_bits;
unsigned end = buf->offset + buf->size;
unsigned len = min(csum_granularity, end - offset);
static void ec_generate_checksums(struct ec_stripe_buf *buf)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned i, j, csums_per_device = stripe_csums_per_device(v);
if (!v->csum_type)
static void ec_validate_checksums(struct bch_fs *c, struct ec_stripe_buf *buf)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned csum_granularity = 1 << v->csum_granularity_bits;
unsigned i;
struct bch_csum got = ec_block_checksum(buf, i, offset);
if (bch2_crc_cmp(want, got)) {
- struct printbuf buf2 = PRINTBUF;
+ struct printbuf err = PRINTBUF;
+ struct bch_dev *ca = bch_dev_bkey_exists(c, v->ptrs[i].dev);
+
+ prt_printf(&err, "stripe checksum error: expected %0llx:%0llx got %0llx:%0llx (type %s)\n",
+ want.hi, want.lo,
+ got.hi, got.lo,
+ bch2_csum_types[v->csum_type]);
+ prt_printf(&err, " for %ps at %u of\n ", (void *) _RET_IP_, i);
+ bch2_bkey_val_to_text(&err, c, bkey_i_to_s_c(&buf->key));
+ bch_err_ratelimited(ca, "%s", err.buf);
+ printbuf_exit(&err);
- 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.buf);
- printbuf_exit(&buf2);
clear_bit(i, buf->valid);
+
+ bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
break;
}
static void ec_generate_ec(struct ec_stripe_buf *buf)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned nr_data = v->nr_blocks - v->nr_redundant;
unsigned bytes = le16_to_cpu(v->sectors) << 9;
static unsigned ec_nr_failed(struct ec_stripe_buf *buf)
{
- return buf->key.v.nr_blocks -
- bitmap_weight(buf->valid, buf->key.v.nr_blocks);
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
+
+ return v->nr_blocks - bitmap_weight(buf->valid, v->nr_blocks);
}
static int ec_do_recov(struct bch_fs *c, struct ec_stripe_buf *buf)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned i, failed[BCH_BKEY_PTRS_MAX], nr_failed = 0;
unsigned nr_data = v->nr_blocks - v->nr_redundant;
unsigned bytes = buf->size << 9;
static void ec_block_endio(struct bio *bio)
{
struct ec_bio *ec_bio = container_of(bio, struct ec_bio, bio);
- struct bch_stripe *v = &ec_bio->buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&ec_bio->buf->key)->v;
struct bch_extent_ptr *ptr = &v->ptrs[ec_bio->idx];
struct bch_dev *ca = ec_bio->ca;
struct closure *cl = bio->bi_private;
- if (bch2_dev_io_err_on(bio->bi_status, ca, "erasure coding %s error: %s",
+ if (bch2_dev_io_err_on(bio->bi_status, ca,
+ bio_data_dir(bio)
+ ? BCH_MEMBER_ERROR_write
+ : BCH_MEMBER_ERROR_read,
+ "erasure coding %s error: %s",
bio_data_dir(bio) ? "write" : "read",
bch2_blk_status_to_str(bio->bi_status)))
clear_bit(ec_bio->idx, ec_bio->buf->valid);
static void ec_block_io(struct bch_fs *c, struct ec_stripe_buf *buf,
blk_opf_t opf, unsigned idx, struct closure *cl)
{
- struct bch_stripe *v = &buf->key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&buf->key)->v;
unsigned offset = 0, bytes = buf->size << 9;
struct bch_extent_ptr *ptr = &v->ptrs[idx];
struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
- enum bch_data_type data_type = idx < buf->key.v.nr_blocks - buf->key.v.nr_redundant
+ enum bch_data_type data_type = idx < v->nr_blocks - v->nr_redundant
? BCH_DATA_user
: BCH_DATA_parity;
int rw = op_is_write(opf);
ret = -ENOENT;
goto err;
}
- bkey_reassemble(&stripe->key.k_i, k);
+ bkey_reassemble(&stripe->key, k);
err:
bch2_trans_iter_exit(trans, &iter);
return ret;
}
-static int get_stripe_key(struct bch_fs *c, u64 idx, struct ec_stripe_buf *stripe)
-{
- return bch2_trans_run(c, get_stripe_key_trans(&trans, idx, stripe));
-}
-
/* recovery read path: */
-int bch2_ec_read_extent(struct bch_fs *c, struct bch_read_bio *rbio)
+int bch2_ec_read_extent(struct btree_trans *trans, struct bch_read_bio *rbio)
{
+ struct bch_fs *c = trans->c;
struct ec_stripe_buf *buf;
struct closure cl;
struct bch_stripe *v;
if (!buf)
return -BCH_ERR_ENOMEM_ec_read_extent;
- ret = get_stripe_key(c, rbio->pick.ec.idx, buf);
+ ret = lockrestart_do(trans, get_stripe_key_trans(trans, rbio->pick.ec.idx, buf));
if (ret) {
bch_err_ratelimited(c,
"error doing reconstruct read: error %i looking up stripe", ret);
return -EIO;
}
- v = &buf->key.v;
+ v = &bkey_i_to_stripe(&buf->key)->v;
if (!bch2_ptr_matches_stripe(v, rbio->pick)) {
bch_err_ratelimited(c,
{
struct bch_fs *c =
container_of(work, struct bch_fs, ec_stripe_delete_work);
- struct btree_trans trans;
+ struct btree_trans *trans = bch2_trans_get(c);
int ret;
u64 idx;
- bch2_trans_init(&trans, c, 0, 0);
-
while (1) {
mutex_lock(&c->ec_stripes_heap_lock);
idx = stripe_idx_to_delete(c);
if (!idx)
break;
- ret = commit_do(&trans, NULL, NULL, BTREE_INSERT_NOFAIL,
- ec_stripe_delete(&trans, idx));
+ ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+ ec_stripe_delete(trans, idx));
if (ret) {
bch_err_fn(c, ret);
break;
}
}
- bch2_trans_exit(&trans);
+ bch2_trans_put(trans);
bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
}
struct ec_stripe_buf *s,
struct bpos *bp_pos)
{
+ struct bch_stripe *v = &bkey_i_to_stripe(&s->key)->v;
struct bch_fs *c = trans->c;
struct bch_backpointer bp;
struct btree_iter iter;
if (extent_has_stripe_ptr(k, s->key.k.p.offset))
goto out;
- ptr_c = bkey_matches_stripe(&s->key.v, k, &block);
+ ptr_c = bkey_matches_stripe(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;
- dev = s->key.v.ptrs[block].dev;
+ dev = v->ptrs[block].dev;
n = bch2_trans_kmalloc(trans, bkey_bytes(k.k) + sizeof(stripe_ptr));
ret = PTR_ERR_OR_ZERO(n);
stripe_ptr = (struct bch_extent_stripe_ptr) {
.type = 1 << BCH_EXTENT_ENTRY_stripe_ptr,
.block = block,
- .redundancy = s->key.v.nr_redundant,
+ .redundancy = v->nr_redundant,
.idx = s->key.k.p.offset,
};
unsigned block)
{
struct bch_fs *c = trans->c;
- struct bch_extent_ptr bucket = s->key.v.ptrs[block];
+ struct bch_stripe *v = &bkey_i_to_stripe(&s->key)->v;
+ struct bch_extent_ptr bucket = v->ptrs[block];
struct bpos bucket_pos = PTR_BUCKET_POS(c, &bucket);
struct bpos bp_pos = POS_MIN;
int ret = 0;
while (1) {
ret = commit_do(trans, NULL, NULL,
- BTREE_INSERT_NOCHECK_RW|
- BTREE_INSERT_NOFAIL,
+ BCH_TRANS_COMMIT_no_check_rw|
+ BCH_TRANS_COMMIT_no_enospc,
ec_stripe_update_extent(trans, bucket_pos, bucket.gen,
s, &bp_pos));
if (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;
+ struct btree_trans *trans = bch2_trans_get(c);
+ struct bch_stripe *v = &bkey_i_to_stripe(&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);
+ ret = bch2_btree_write_buffer_flush_nocheck_rw(trans);
if (ret)
goto err;
for (i = 0; i < nr_data; i++) {
- ret = ec_stripe_update_bucket(&trans, s, i);
+ ret = ec_stripe_update_bucket(trans, s, i);
if (ret)
break;
}
err:
- bch2_trans_exit(&trans);
+ bch2_trans_put(trans);
return ret;
}
int ret;
if (!bch2_dev_get_ioref(ca, WRITE)) {
- s->err = -EROFS;
+ s->err = -BCH_ERR_erofs_no_writes;
return;
}
{
struct bch_fs *c = s->c;
struct open_bucket *ob;
- struct bch_stripe *v = &s->new_stripe.key.v;
+ struct bch_stripe *v = &bkey_i_to_stripe(&s->new_stripe.key)->v;
unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
int ret;
}
for (i = 0; i < nr_data; i++)
- if (stripe_blockcount_get(&s->existing_stripe.key.v, i))
+ if (stripe_blockcount_get(&bkey_i_to_stripe(&s->existing_stripe.key)->v, i))
swap(s->new_stripe.data[i],
s->existing_stripe.data[i]);
}
ret = bch2_trans_do(c, &s->res, NULL,
- BTREE_INSERT_NOCHECK_RW|
- BTREE_INSERT_NOFAIL,
- ec_stripe_key_update(&trans, &s->new_stripe.key,
- !s->have_existing_stripe));
+ BCH_TRANS_COMMIT_no_check_rw|
+ BCH_TRANS_COMMIT_no_enospc,
+ ec_stripe_key_update(trans,
+ bkey_i_to_stripe(&s->new_stripe.key),
+ !s->have_existing_stripe));
if (ret) {
bch_err(c, "error creating stripe: error creating stripe key");
goto err;
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));
+ bch_err_msg(c, ret, "creating stripe: error updating pointers");
goto err;
}
err:
}
static void ec_stripe_key_init(struct bch_fs *c,
- struct bkey_i_stripe *s,
+ struct bkey_i *k,
unsigned nr_data,
unsigned nr_parity,
unsigned stripe_size)
{
+ struct bkey_i_stripe *s = bkey_stripe_init(k);
unsigned u64s;
- bkey_stripe_init(&s->k_i);
s->v.sectors = cpu_to_le16(stripe_size);
s->v.algorithm = 0;
s->v.nr_blocks = nr_data + nr_parity;
BCH_BKEY_PTRS_MAX) - h->redundancy;
s->nr_parity = h->redundancy;
- ec_stripe_key_init(c, &s->new_stripe.key, s->nr_data,
- s->nr_parity, h->blocksize);
+ ec_stripe_key_init(c, &s->new_stripe.key,
+ s->nr_data, s->nr_parity, h->blocksize);
h->s = s;
return 0;
h->nr_active_devs++;
rcu_read_unlock();
+
+ /*
+ * If we only have redundancy + 1 devices, we're better off with just
+ * replication:
+ */
+ if (h->nr_active_devs < h->redundancy + 2)
+ bch_err(c, "insufficient devices available to create stripe (have %u, need %u) - mismatched bucket sizes?",
+ h->nr_active_devs, h->redundancy + 2);
+
list_add(&h->list, &c->ec_stripe_head_list);
return h;
}
if (ret)
return ERR_PTR(ret);
- if (test_bit(BCH_FS_GOING_RO, &c->flags)) {
- h = ERR_PTR(-EROFS);
+ if (test_bit(BCH_FS_going_ro, &c->flags)) {
+ h = ERR_PTR(-BCH_ERR_erofs_no_writes);
goto found;
}
h = ec_new_stripe_head_alloc(c, target, algo, redundancy, watermark);
found:
+ if (!IS_ERR_OR_NULL(h) &&
+ h->nr_active_devs < h->redundancy + 2) {
+ mutex_unlock(&h->lock);
+ h = NULL;
+ }
mutex_unlock(&c->ec_stripe_head_lock);
return h;
}
struct bch_devs_mask devs = h->devs;
struct open_bucket *ob;
struct open_buckets buckets;
+ struct bch_stripe *v = &bkey_i_to_stripe(&h->s->new_stripe.key)->v;
unsigned i, j, nr_have_parity = 0, nr_have_data = 0;
bool have_cache = true;
int ret = 0;
- BUG_ON(h->s->new_stripe.key.v.nr_blocks != h->s->nr_data + h->s->nr_parity);
- BUG_ON(h->s->new_stripe.key.v.nr_redundant != h->s->nr_parity);
+ BUG_ON(v->nr_blocks != h->s->nr_data + h->s->nr_parity);
+ BUG_ON(v->nr_redundant != h->s->nr_parity);
- for_each_set_bit(i, h->s->blocks_gotten, h->s->new_stripe.key.v.nr_blocks) {
- __clear_bit(h->s->new_stripe.key.v.ptrs[i].dev, devs.d);
+ for_each_set_bit(i, h->s->blocks_gotten, v->nr_blocks) {
+ __clear_bit(v->ptrs[i].dev, devs.d);
if (i < h->s->nr_data)
nr_have_data++;
else
BUG_ON(j >= h->s->nr_data + h->s->nr_parity);
h->s->blocks[j] = buckets.v[i];
- h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob);
+ v->ptrs[j] = bch2_ob_ptr(c, ob);
__set_bit(j, h->s->blocks_gotten);
}
BUG_ON(j >= h->s->nr_data);
h->s->blocks[j] = buckets.v[i];
- h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob);
+ v->ptrs[j] = bch2_ob_ptr(c, ob);
__set_bit(j, h->s->blocks_gotten);
}
static int __bch2_ec_stripe_head_reuse(struct btree_trans *trans, struct ec_stripe_head *h)
{
struct bch_fs *c = trans->c;
+ struct bch_stripe *new_v = &bkey_i_to_stripe(&h->s->new_stripe.key)->v;
+ struct bch_stripe *existing_v;
unsigned i;
s64 idx;
int ret;
return ret;
}
- BUG_ON(h->s->existing_stripe.key.v.nr_redundant != h->s->nr_parity);
- h->s->nr_data = h->s->existing_stripe.key.v.nr_blocks -
- h->s->existing_stripe.key.v.nr_redundant;
+ existing_v = &bkey_i_to_stripe(&h->s->existing_stripe.key)->v;
+
+ BUG_ON(existing_v->nr_redundant != h->s->nr_parity);
+ h->s->nr_data = existing_v->nr_blocks -
+ existing_v->nr_redundant;
ret = ec_stripe_buf_init(&h->s->existing_stripe, 0, h->blocksize);
if (ret) {
}
BUG_ON(h->s->existing_stripe.size != h->blocksize);
- BUG_ON(h->s->existing_stripe.size != le16_to_cpu(h->s->existing_stripe.key.v.sectors));
+ BUG_ON(h->s->existing_stripe.size != le16_to_cpu(existing_v->sectors));
/*
* Free buckets we initially allocated - they might conflict with
* blocks from the stripe we're reusing:
*/
- for_each_set_bit(i, h->s->blocks_gotten, h->s->new_stripe.key.v.nr_blocks) {
+ for_each_set_bit(i, h->s->blocks_gotten, new_v->nr_blocks) {
bch2_open_bucket_put(c, c->open_buckets + h->s->blocks[i]);
h->s->blocks[i] = 0;
}
memset(h->s->blocks_gotten, 0, sizeof(h->s->blocks_gotten));
memset(h->s->blocks_allocated, 0, sizeof(h->s->blocks_allocated));
- for (i = 0; i < h->s->existing_stripe.key.v.nr_blocks; i++) {
- if (stripe_blockcount_get(&h->s->existing_stripe.key.v, i)) {
+ for (i = 0; i < existing_v->nr_blocks; i++) {
+ if (stripe_blockcount_get(existing_v, i)) {
__set_bit(i, h->s->blocks_gotten);
__set_bit(i, h->s->blocks_allocated);
}
ec_block_io(c, &h->s->existing_stripe, READ, i, &h->s->iodone);
}
- bkey_copy(&h->s->new_stripe.key.k_i, &h->s->existing_stripe.key.k_i);
+ bkey_copy(&h->s->new_stripe.key, &h->s->existing_stripe.key);
h->s->have_existing_stripe = true;
return 0;
int ret;
h = __bch2_ec_stripe_head_get(trans, target, algo, redundancy, watermark);
- if (!h)
- bch_err(c, "no stripe head");
if (IS_ERR_OR_NULL(h))
return h;
if (!ca)
goto found;
- for (i = 0; i < h->s->new_stripe.key.v.nr_blocks; i++) {
+ for (i = 0; i < bkey_i_to_stripe(&h->s->new_stripe.key)->v.nr_blocks; i++) {
if (!h->s->blocks[i])
continue;
}
goto unlock;
found:
- h->s->err = -EROFS;
+ h->s->err = -BCH_ERR_erofs_no_writes;
ec_stripe_set_pending(c, h);
unlock:
mutex_unlock(&h->lock);
int bch2_stripes_read(struct bch_fs *c)
{
- struct btree_trans trans;
+ struct btree_trans *trans = bch2_trans_get(c);
struct btree_iter iter;
struct bkey_s_c k;
const struct bch_stripe *s;
unsigned i;
int ret;
- bch2_trans_init(&trans, c, 0, 0);
-
- for_each_btree_key(&trans, iter, BTREE_ID_stripes, POS_MIN,
+ for_each_btree_key(trans, iter, BTREE_ID_stripes, POS_MIN,
BTREE_ITER_PREFETCH, k, ret) {
if (k.k->type != KEY_TYPE_stripe)
continue;
bch2_stripes_heap_insert(c, m, k.k->p.offset);
}
- bch2_trans_iter_exit(&trans, &iter);
+ bch2_trans_iter_exit(trans, &iter);
- bch2_trans_exit(&trans);
+ bch2_trans_put(trans);
if (ret)
bch_err_fn(c, ret);
break;
if (h->s) {
- for (i = 0; i < h->s->new_stripe.key.v.nr_blocks; i++)
+ for (i = 0; i < bkey_i_to_stripe(&h->s->new_stripe.key)->v.nr_blocks; i++)
BUG_ON(h->s->blocks[i]);
kfree(h->s);