#include "keylist.h"
#include "move.h"
#include "rebalance.h"
+#include "subvolume.h"
#include "super.h"
#include "super-io.h"
* the time:
*/
if (abs((int) (old - io_latency)) < (old >> 1) &&
- now & ~(~0 << 5))
+ now & ~(~0U << 5))
break;
new = ewma_add(old, io_latency, 5);
int bch2_sum_sector_overwrites(struct btree_trans *trans,
struct btree_iter *extent_iter,
struct bkey_i *new,
- bool *maybe_extending,
- bool *should_check_enospc,
+ bool *usage_increasing,
s64 *i_sectors_delta,
s64 *disk_sectors_delta)
{
struct bch_fs *c = trans->c;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_s_c old;
unsigned new_replicas = bch2_bkey_replicas(c, bkey_i_to_s_c(new));
bool new_compressed = bch2_bkey_sectors_compressed(bkey_i_to_s_c(new));
int ret = 0;
- *maybe_extending = true;
- *should_check_enospc = false;
+ *usage_increasing = false;
*i_sectors_delta = 0;
*disk_sectors_delta = 0;
- iter = bch2_trans_copy_iter(trans, extent_iter);
+ bch2_trans_copy_iter(&iter, extent_iter);
- for_each_btree_key_continue(iter, BTREE_ITER_SLOTS, old, ret) {
+ for_each_btree_key_continue_norestart(iter, 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));
(bkey_extent_is_allocation(&new->k) -
bkey_extent_is_allocation(old.k));
- *disk_sectors_delta += sectors *
- (int) (bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(new)) -
- bch2_bkey_nr_ptrs_fully_allocated(old));
+ *disk_sectors_delta += sectors * bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(new));
+ *disk_sectors_delta -= new->k.p.snapshot == old.k->p.snapshot
+ ? sectors * bch2_bkey_nr_ptrs_fully_allocated(old)
+ : 0;
- if (!*should_check_enospc &&
- (new_replicas > bch2_bkey_replicas(c, old) ||
+ if (!*usage_increasing &&
+ (new->k.p.snapshot != old.k->p.snapshot ||
+ new_replicas > bch2_bkey_replicas(c, old) ||
(!new_compressed && bch2_bkey_sectors_compressed(old))))
- *should_check_enospc = true;
-
- if (bkey_cmp(old.k->p, new->k.p) >= 0) {
- /*
- * Check if there's already data above where we're
- * going to be writing to - this means we're definitely
- * not extending the file:
- *
- * Note that it's not sufficient to check if there's
- * data up to the sector offset we're going to be
- * writing to, because i_size could be up to one block
- * less:
- */
- if (!bkey_cmp(old.k->p, new->k.p))
- old = bch2_btree_iter_next(iter);
-
- if (old.k && !bkey_err(old) &&
- old.k->p.inode == extent_iter->pos.inode &&
- bkey_extent_is_data(old.k))
- *maybe_extending = false;
+ *usage_increasing = true;
+ if (bkey_cmp(old.k->p, new->k.p) >= 0)
break;
- }
}
- bch2_trans_iter_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
int bch2_extent_update(struct btree_trans *trans,
+ subvol_inum inum,
struct btree_iter *iter,
struct bkey_i *k,
struct disk_reservation *disk_res,
u64 *journal_seq,
u64 new_i_size,
- s64 *i_sectors_delta_total)
+ s64 *i_sectors_delta_total,
+ bool check_enospc)
{
- /* this must live until after bch2_trans_commit(): */
- struct bkey_inode_buf inode_p;
- bool extending = false, should_check_enospc;
+ struct btree_iter inode_iter;
+ struct bch_inode_unpacked inode_u;
+ struct bpos next_pos;
+ bool usage_increasing;
s64 i_sectors_delta = 0, disk_sectors_delta = 0;
int ret;
- ret = bch2_extent_trim_atomic(k, iter);
+ /*
+ * This traverses us the iterator without changing iter->path->pos to
+ * search_key() (which is pos + 1 for extents): we want there to be a
+ * path already traversed at iter->pos because
+ * bch2_trans_extent_update() will use it to attempt extent merging
+ */
+ ret = __bch2_btree_iter_traverse(iter);
+ if (ret)
+ return ret;
+
+ ret = bch2_extent_trim_atomic(trans, iter, k);
if (ret)
return ret;
+ new_i_size = min(k->k.p.offset << 9, new_i_size);
+ next_pos = k->k.p;
+
ret = bch2_sum_sector_overwrites(trans, iter, k,
- &extending,
- &should_check_enospc,
+ &usage_increasing,
&i_sectors_delta,
&disk_sectors_delta);
if (ret)
disk_sectors_delta > (s64) disk_res->sectors) {
ret = bch2_disk_reservation_add(trans->c, disk_res,
disk_sectors_delta - disk_res->sectors,
- !should_check_enospc
+ !check_enospc || !usage_increasing
? BCH_DISK_RESERVATION_NOFAIL : 0);
if (ret)
return ret;
}
- new_i_size = extending
- ? min(k->k.p.offset << 9, new_i_size)
- : 0;
-
- if (i_sectors_delta || new_i_size) {
- struct btree_iter *inode_iter;
- struct bch_inode_unpacked inode_u;
-
- inode_iter = bch2_inode_peek(trans, &inode_u,
- k->k.p.inode, BTREE_ITER_INTENT);
- if (IS_ERR(inode_iter))
- return PTR_ERR(inode_iter);
-
- /*
- * XXX:
- * writeback can race a bit with truncate, because truncate
- * first updates the inode then truncates the pagecache. This is
- * ugly, but lets us preserve the invariant that the in memory
- * i_size is always >= the on disk i_size.
- *
- BUG_ON(new_i_size > inode_u.bi_size &&
- (inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY));
- */
- BUG_ON(new_i_size > inode_u.bi_size && !extending);
-
- if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
- new_i_size > inode_u.bi_size)
- inode_u.bi_size = new_i_size;
- else
- new_i_size = 0;
-
- inode_u.bi_sectors += i_sectors_delta;
-
- if (i_sectors_delta || new_i_size) {
- bch2_inode_pack(trans->c, &inode_p, &inode_u);
- bch2_trans_update(trans, inode_iter,
- &inode_p.inode.k_i, 0);
- }
+ ret = bch2_inode_peek(trans, &inode_iter, &inode_u, inum,
+ BTREE_ITER_INTENT);
+ if (ret)
+ return ret;
- bch2_trans_iter_put(trans, inode_iter);
- }
+ if (!(inode_u.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
+ new_i_size > inode_u.bi_size)
+ inode_u.bi_size = new_i_size;
- bch2_trans_update(trans, iter, k, 0);
+ inode_u.bi_sectors += i_sectors_delta;
- ret = bch2_trans_commit(trans, disk_res, journal_seq,
+ ret = bch2_trans_update(trans, iter, k, 0) ?:
+ bch2_inode_write(trans, &inode_iter, &inode_u) ?:
+ bch2_trans_commit(trans, disk_res, journal_seq,
BTREE_INSERT_NOCHECK_RW|
BTREE_INSERT_NOFAIL);
+ bch2_trans_iter_exit(trans, &inode_iter);
+
if (ret)
return ret;
if (i_sectors_delta_total)
*i_sectors_delta_total += i_sectors_delta;
+ bch2_btree_iter_set_pos(iter, next_pos);
+
return 0;
}
+/*
+ * Returns -EINTR if we had to drop locks:
+ */
int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
- struct bpos end, u64 *journal_seq,
+ subvol_inum inum, u64 end,
s64 *i_sectors_delta)
{
struct bch_fs *c = trans->c;
unsigned max_sectors = KEY_SIZE_MAX & (~0 << c->block_bits);
+ struct bpos end_pos = POS(inum.inum, end);
struct bkey_s_c k;
int ret = 0, ret2 = 0;
+ u32 snapshot;
- while ((k = bch2_btree_iter_peek(iter)).k &&
- bkey_cmp(iter->pos, end) < 0) {
+ while (!ret || ret == -EINTR) {
struct disk_reservation disk_res =
bch2_disk_reservation_init(c, 0);
struct bkey_i delete;
+ if (ret)
+ ret2 = ret;
+
bch2_trans_begin(trans);
+ ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
+ if (ret)
+ continue;
+
+ bch2_btree_iter_set_snapshot(iter, snapshot);
+
+ k = bch2_btree_iter_peek(iter);
+ if (bkey_cmp(iter->pos, end_pos) >= 0) {
+ bch2_btree_iter_set_pos(iter, end_pos);
+ break;
+ }
+
ret = bkey_err(k);
if (ret)
- goto btree_err;
+ continue;
bkey_init(&delete.k);
delete.k.p = iter->pos;
/* create the biggest key we can */
bch2_key_resize(&delete.k, max_sectors);
- bch2_cut_back(end, &delete);
+ bch2_cut_back(end_pos, &delete);
- ret = bch2_extent_update(trans, iter, &delete,
- &disk_res, journal_seq,
- 0, i_sectors_delta);
+ ret = bch2_extent_update(trans, inum, iter, &delete,
+ &disk_res, NULL,
+ 0, i_sectors_delta, false);
bch2_disk_reservation_put(c, &disk_res);
-btree_err:
- if (ret == -EINTR) {
- ret2 = ret;
- ret = 0;
- }
- if (ret)
- break;
- }
-
- if (bkey_cmp(iter->pos, end) > 0) {
- bch2_btree_iter_set_pos(iter, end);
- ret = bch2_btree_iter_traverse(iter);
}
return ret ?: ret2;
}
-int bch2_fpunch(struct bch_fs *c, u64 inum, u64 start, u64 end,
- u64 *journal_seq, s64 *i_sectors_delta)
+int bch2_fpunch(struct bch_fs *c, subvol_inum inum, u64 start, u64 end,
+ s64 *i_sectors_delta)
{
struct btree_trans trans;
- struct btree_iter *iter;
- int ret = 0;
+ struct btree_iter iter;
+ int ret;
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
- POS(inum, start),
- BTREE_ITER_INTENT);
+ bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
+ POS(inum.inum, start),
+ BTREE_ITER_INTENT);
- ret = bch2_fpunch_at(&trans, iter, POS(inum, end),
- journal_seq, i_sectors_delta);
- bch2_trans_exit(&trans);
+ ret = bch2_fpunch_at(&trans, &iter, inum, end, i_sectors_delta);
- if (ret == -EINTR)
- ret = 0;
+ bch2_trans_iter_exit(&trans, &iter);
+ bch2_trans_exit(&trans);
- return ret;
+ return ret == -EINTR ? 0 : ret;
}
int bch2_write_index_default(struct bch_write_op *op)
{
struct bch_fs *c = op->c;
struct bkey_buf sk;
+ struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
struct keylist *keys = &op->insert_keys;
struct bkey_i *k = bch2_keylist_front(keys);
struct btree_trans trans;
- struct btree_iter *iter;
+ struct btree_iter iter;
+ subvol_inum inum = {
+ .subvol = op->subvol,
+ .inum = k->k.p.inode,
+ };
int ret;
+ BUG_ON(!inum.subvol);
+
bch2_bkey_buf_init(&sk);
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
- bkey_start_pos(&k->k),
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
-
do {
bch2_trans_begin(&trans);
k = bch2_keylist_front(keys);
+ bch2_bkey_buf_copy(&sk, c, k);
- bch2_bkey_buf_realloc(&sk, c, k->k.u64s);
- bkey_copy(sk.k, k);
- bch2_cut_front(iter->pos, sk.k);
+ ret = bch2_subvolume_get_snapshot(&trans, inum.subvol,
+ &sk.k->k.p.snapshot);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ break;
+
+ bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
+ bkey_start_pos(&sk.k->k),
+ BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
- ret = bch2_extent_update(&trans, iter, sk.k,
+ ret = bch2_extent_update(&trans, inum, &iter, sk.k,
&op->res, op_journal_seq(op),
- op->new_i_size, &op->i_sectors_delta);
+ op->new_i_size, &op->i_sectors_delta,
+ op->flags & BCH_WRITE_CHECK_ENOSPC);
+ bch2_trans_iter_exit(&trans, &iter);
+
if (ret == -EINTR)
continue;
if (ret)
break;
- if (bkey_cmp(iter->pos, k->k.p) >= 0)
- bch2_keylist_pop_front(keys);
+ if (ec_ob)
+ bch2_ob_add_backpointer(c, ec_ob, &sk.k->k);
+
+ if (bkey_cmp(iter.pos, k->k.p) >= 0)
+ bch2_keylist_pop_front(&op->insert_keys);
+ else
+ bch2_cut_front(iter.pos, k);
} while (!bch2_keylist_empty(keys));
bch2_trans_exit(&trans);
n->submit_time = local_clock();
n->bio.bi_iter.bi_sector = ptr->offset;
- if (!journal_flushes_device(ca))
- n->bio.bi_opf |= REQ_FUA;
-
if (likely(n->have_ioref)) {
this_cpu_add(ca->io_done->sectors[WRITE][type],
bio_sectors(&n->bio));
bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time);
- if (!(op->flags & BCH_WRITE_FROM_INTERNAL))
- up(&c->io_in_flight);
-
if (op->end_io) {
EBUG_ON(cl->parent);
closure_debug_destroy(cl);
? ((unsigned long) buf & (PAGE_SIZE - 1))
: 0), PAGE_SIZE);
+ pages = min(pages, BIO_MAX_VECS);
+
bio = bio_alloc_bioset(GFP_NOIO, pages, &c->bio_write);
wbio = wbio_init(bio);
wbio->put_bio = true;
struct bio *src = &op->wbio.bio, *dst = src;
struct bvec_iter saved_iter;
void *ec_buf;
- struct bpos ec_pos = op->pos;
unsigned total_output = 0, total_input = 0;
bool bounce = false;
bool page_alloc_failed = false;
ret = -EIO;
goto err;
case PREP_ENCODED_CHECKSUM_ERR:
- BUG();
goto csum_err;
case PREP_ENCODED_DO_WRITE:
/* XXX look for bug here */
dst->bi_iter.bi_size = total_output;
do_write:
- /* might have done a realloc... */
- bch2_ec_add_backpointer(c, wp, ec_pos, total_input >> 9);
-
*_dst = dst;
return more;
csum_err:
goto err;
}
- /*
- * Can't ratelimit copygc - we'd deadlock:
- */
- if (!(op->flags & BCH_WRITE_FROM_INTERNAL))
- down(&c->io_in_flight);
-
bch2_increment_clock(c, bio_sectors(bio), WRITE);
data_len = min_t(u64, bio->bi_iter.bi_size,
bch2_migrate_read_done(&op->write, rbio);
closure_init(cl, NULL);
- closure_call(&op->write.op.cl, bch2_write, c->wq, cl);
+ closure_call(&op->write.op.cl, bch2_write, c->btree_update_wq, cl);
closure_return_with_destructor(cl, promote_done);
}
promote = __promote_alloc(c,
k.k->type == KEY_TYPE_reflink_v
- ? BTREE_ID_REFLINK
- : BTREE_ID_EXTENTS,
+ ? BTREE_ID_reflink
+ : BTREE_ID_extents,
k, pos, pick, opts, sectors, rbio);
if (!promote)
return NULL;
}
static void bch2_read_retry_nodecode(struct bch_fs *c, struct bch_read_bio *rbio,
- struct bvec_iter bvec_iter, u64 inode,
+ struct bvec_iter bvec_iter,
struct bch_io_failures *failed,
unsigned flags)
{
struct btree_trans trans;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_buf sk;
struct bkey_s_c k;
int ret;
bch2_bkey_buf_init(&sk);
bch2_trans_init(&trans, c, 0, 0);
- iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
- rbio->pos, BTREE_ITER_SLOTS);
+ bch2_trans_iter_init(&trans, &iter, rbio->data_btree,
+ rbio->read_pos, BTREE_ITER_SLOTS);
retry:
rbio->bio.bi_status = 0;
- k = bch2_btree_iter_peek_slot(iter);
+ k = bch2_btree_iter_peek_slot(&iter);
if (bkey_err(k))
goto err;
if (!bch2_bkey_matches_ptr(c, k,
rbio->pick.ptr,
- rbio->pos.offset -
+ rbio->data_pos.offset -
rbio->pick.crc.offset)) {
/* extent we wanted to read no longer exists: */
rbio->hole = true;
goto out;
}
- ret = __bch2_read_extent(&trans, rbio, bvec_iter, k, 0, failed, flags);
+ ret = __bch2_read_extent(&trans, rbio, bvec_iter,
+ rbio->read_pos,
+ rbio->data_btree,
+ k, 0, failed, flags);
if (ret == READ_RETRY)
goto retry;
if (ret)
goto err;
out:
bch2_rbio_done(rbio);
+ bch2_trans_iter_exit(&trans, &iter);
bch2_trans_exit(&trans);
bch2_bkey_buf_exit(&sk, c);
return;
goto out;
}
-static void bch2_read_retry(struct bch_fs *c, struct bch_read_bio *rbio,
- struct bvec_iter bvec_iter, u64 inode,
- struct bch_io_failures *failed, unsigned flags)
-{
- struct btree_trans trans;
- struct btree_iter *iter;
- struct bkey_buf sk;
- struct bkey_s_c k;
- int ret;
-
- flags &= ~BCH_READ_LAST_FRAGMENT;
- flags |= BCH_READ_MUST_CLONE;
-
- bch2_bkey_buf_init(&sk);
- bch2_trans_init(&trans, c, 0, 0);
-retry:
- bch2_trans_begin(&trans);
-
- for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS,
- POS(inode, bvec_iter.bi_sector),
- BTREE_ITER_SLOTS, k, ret) {
- unsigned bytes, sectors, offset_into_extent;
-
- bch2_bkey_buf_reassemble(&sk, c, k);
-
- offset_into_extent = iter->pos.offset -
- bkey_start_offset(k.k);
- sectors = k.k->size - offset_into_extent;
-
- ret = bch2_read_indirect_extent(&trans,
- &offset_into_extent, &sk);
- if (ret)
- break;
-
- k = bkey_i_to_s_c(sk.k);
-
- sectors = min(sectors, k.k->size - offset_into_extent);
-
- bch2_trans_unlock(&trans);
-
- bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
- swap(bvec_iter.bi_size, bytes);
-
- ret = __bch2_read_extent(&trans, rbio, bvec_iter, k,
- offset_into_extent, failed, flags);
- switch (ret) {
- case READ_RETRY:
- goto retry;
- case READ_ERR:
- goto err;
- };
-
- if (bytes == bvec_iter.bi_size)
- goto out;
-
- swap(bvec_iter.bi_size, bytes);
- bio_advance_iter(&rbio->bio, &bvec_iter, bytes);
- }
-
- if (ret == -EINTR)
- goto retry;
- /*
- * If we get here, it better have been because there was an error
- * reading a btree node
- */
- BUG_ON(!ret);
- bch_err_inum_ratelimited(c, inode,
- "read error %i from btree lookup", ret);
-err:
- rbio->bio.bi_status = BLK_STS_IOERR;
-out:
- bch2_trans_exit(&trans);
- bch2_bkey_buf_exit(&sk, c);
- bch2_rbio_done(rbio);
-}
-
static void bch2_rbio_retry(struct work_struct *work)
{
struct bch_read_bio *rbio =
struct bch_fs *c = rbio->c;
struct bvec_iter iter = rbio->bvec_iter;
unsigned flags = rbio->flags;
- u64 inode = rbio->pos.inode;
+ subvol_inum inum = {
+ .subvol = rbio->subvol,
+ .inum = rbio->read_pos.inode,
+ };
struct bch_io_failures failed = { .nr = 0 };
trace_read_retry(&rbio->bio);
flags |= BCH_READ_IN_RETRY;
flags &= ~BCH_READ_MAY_PROMOTE;
- if (flags & BCH_READ_NODECODE)
- bch2_read_retry_nodecode(c, rbio, iter, inode, &failed, flags);
- else
- bch2_read_retry(c, rbio, iter, inode, &failed, flags);
+ if (flags & BCH_READ_NODECODE) {
+ bch2_read_retry_nodecode(c, rbio, iter, &failed, flags);
+ } else {
+ flags &= ~BCH_READ_LAST_FRAGMENT;
+ flags |= BCH_READ_MUST_CLONE;
+
+ __bch2_read(c, rbio, iter, inum, &failed, flags);
+ }
}
static void bch2_rbio_error(struct bch_read_bio *rbio, int retry,
struct bch_read_bio *rbio)
{
struct bch_fs *c = rbio->c;
- u64 data_offset = rbio->pos.offset - rbio->pick.crc.offset;
+ u64 data_offset = rbio->data_pos.offset - rbio->pick.crc.offset;
struct bch_extent_crc_unpacked new_crc;
- struct btree_iter *iter = NULL;
+ struct btree_iter iter;
struct bkey_i *new;
struct bkey_s_c k;
int ret = 0;
if (crc_is_compressed(rbio->pick.crc))
return 0;
- iter = bch2_trans_get_iter(trans, BTREE_ID_EXTENTS, rbio->pos,
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(iter);
+ bch2_trans_iter_init(trans, &iter, rbio->data_btree, rbio->data_pos,
+ BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+ k = bch2_btree_iter_peek_slot(&iter);
if ((ret = bkey_err(k)))
goto out;
if (!bch2_bkey_narrow_crcs(new, new_crc))
goto out;
- bch2_trans_update(trans, iter, new, 0);
+ ret = bch2_trans_update(trans, &iter, new,
+ BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
out:
- bch2_trans_iter_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
struct bvec_iter dst_iter = rbio->bvec_iter;
struct bch_extent_crc_unpacked crc = rbio->pick.crc;
struct nonce nonce = extent_nonce(rbio->version, crc);
+ unsigned nofs_flags;
struct bch_csum csum;
+ nofs_flags = memalloc_nofs_save();
+
/* Reset iterator for checksumming and copying bounced data: */
if (rbio->bounce) {
src->bi_iter.bi_size = crc.compressed_size << 9;
if (bch2_crc_cmp(csum, rbio->pick.crc.csum))
goto csum_err;
+ /*
+ * XXX
+ * We need to rework the narrow_crcs path to deliver the read completion
+ * first, and then punt to a different workqueue, otherwise we're
+ * holding up reads while doing btree updates which is bad for memory
+ * reclaim.
+ */
if (unlikely(rbio->narrow_crcs))
bch2_rbio_narrow_crcs(rbio);
rbio = bch2_rbio_free(rbio);
bch2_rbio_done(rbio);
}
+out:
+ memalloc_nofs_restore(nofs_flags);
return;
csum_err:
/*
if (!rbio->bounce && (rbio->flags & BCH_READ_USER_MAPPED)) {
rbio->flags |= BCH_READ_MUST_BOUNCE;
bch2_rbio_error(rbio, READ_RETRY, BLK_STS_IOERR);
- return;
+ goto out;
}
- bch2_dev_inum_io_error(ca, rbio->pos.inode, (u64) rbio->bvec_iter.bi_sector,
+ bch2_dev_inum_io_error(ca, rbio->read_pos.inode, (u64) rbio->bvec_iter.bi_sector,
"data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %u)",
rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo,
csum.hi, csum.lo, crc.csum_type);
bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
- return;
+ goto out;
decompression_err:
- bch_err_inum_ratelimited(c, rbio->pos.inode,
+ bch_err_inum_ratelimited(c, rbio->read_pos.inode,
"decompression error");
bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
- return;
+ goto out;
}
static void bch2_read_endio(struct bio *bio)
if (!rbio->split)
rbio->bio.bi_end_io = rbio->end_io;
- /*
- * XXX: rbio->pos is not what we want here when reading from indirect
- * extents
- */
if (bch2_dev_inum_io_err_on(bio->bi_status, ca,
- rbio->pos.inode,
- rbio->pos.offset,
+ rbio->read_pos.inode,
+ rbio->read_pos.offset,
"data read error: %s",
bch2_blk_status_to_str(bio->bi_status))) {
bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
unsigned *offset_into_extent,
struct bkey_buf *orig_k)
{
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_s_c k;
u64 reflink_offset;
int ret;
reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) +
*offset_into_extent;
- iter = bch2_trans_get_iter(trans, BTREE_ID_REFLINK,
- POS(0, reflink_offset),
- BTREE_ITER_SLOTS);
- k = bch2_btree_iter_peek_slot(iter);
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink,
+ POS(0, reflink_offset),
+ BTREE_ITER_SLOTS);
+ k = bch2_btree_iter_peek_slot(&iter);
ret = bkey_err(k);
if (ret)
goto err;
if (k.k->type != KEY_TYPE_reflink_v &&
k.k->type != KEY_TYPE_indirect_inline_data) {
bch_err_inum_ratelimited(trans->c, orig_k->k->k.p.inode,
- "pointer to nonexistent indirect extent");
+ "%llu len %u points to nonexistent indirect extent %llu",
+ orig_k->k->k.p.offset,
+ orig_k->k->k.size,
+ reflink_offset);
+ bch2_inconsistent_error(trans->c);
ret = -EIO;
goto err;
}
- *offset_into_extent = iter->pos.offset - bkey_start_offset(k.k);
+ *offset_into_extent = iter.pos.offset - bkey_start_offset(k.k);
bch2_bkey_buf_reassemble(orig_k, trans->c, k);
err:
- bch2_trans_iter_put(trans, iter);
+ bch2_trans_iter_exit(trans, &iter);
return ret;
}
int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
- struct bvec_iter iter, struct bkey_s_c k,
+ struct bvec_iter iter, struct bpos read_pos,
+ enum btree_id data_btree, struct bkey_s_c k,
unsigned offset_into_extent,
struct bch_io_failures *failed, unsigned flags)
{
struct bch_dev *ca;
struct promote_op *promote = NULL;
bool bounce = false, read_full = false, narrow_crcs = false;
- struct bpos pos = bkey_start_pos(k.k);
+ struct bpos data_pos = bkey_start_pos(k.k);
int pick_ret;
if (bkey_extent_is_inline_data(k.k)) {
EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size);
if (crc_is_compressed(pick.crc) ||
- (pick.crc.csum_type != BCH_CSUM_NONE &&
+ (pick.crc.csum_type != BCH_CSUM_none &&
(bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
(bch2_csum_type_is_encryption(pick.crc.csum_type) &&
(flags & BCH_READ_USER_MAPPED)) ||
pick.crc.offset ||
offset_into_extent));
- pos.offset += offset_into_extent;
+ data_pos.offset += offset_into_extent;
pick.ptr.offset += pick.crc.offset +
offset_into_extent;
offset_into_extent = 0;
/* XXX: only initialize this if needed */
rbio->devs_have = bch2_bkey_devs(k);
rbio->pick = pick;
- rbio->pos = pos;
+ rbio->subvol = orig->subvol;
+ rbio->read_pos = read_pos;
+ rbio->data_btree = data_btree;
+ rbio->data_pos = data_pos;
rbio->version = k.k->version;
rbio->promote = promote;
INIT_WORK(&rbio->work, NULL);
ret = READ_RETRY;
}
+ if (!ret)
+ goto out_read_done;
+
return ret;
}
return 0;
}
-void bch2_read(struct bch_fs *c, struct bch_read_bio *rbio, u64 inode)
+void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
+ struct bvec_iter bvec_iter, subvol_inum inum,
+ struct bch_io_failures *failed, unsigned flags)
{
struct btree_trans trans;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bkey_buf sk;
struct bkey_s_c k;
- unsigned flags = BCH_READ_RETRY_IF_STALE|
- BCH_READ_MAY_PROMOTE|
- BCH_READ_USER_MAPPED;
+ u32 snapshot;
int ret;
- BUG_ON(rbio->_state);
BUG_ON(flags & BCH_READ_NODECODE);
- BUG_ON(flags & BCH_READ_IN_RETRY);
-
- rbio->c = c;
- rbio->start_time = local_clock();
bch2_bkey_buf_init(&sk);
bch2_trans_init(&trans, c, 0, 0);
retry:
bch2_trans_begin(&trans);
+ iter = (struct btree_iter) { NULL };
- iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
- POS(inode, rbio->bio.bi_iter.bi_sector),
- BTREE_ITER_SLOTS);
+ ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
+ if (ret)
+ goto err;
+
+ bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
+ SPOS(inum.inum, bvec_iter.bi_sector, snapshot),
+ BTREE_ITER_SLOTS|BTREE_ITER_FILTER_SNAPSHOTS);
while (1) {
unsigned bytes, sectors, offset_into_extent;
+ enum btree_id data_btree = BTREE_ID_extents;
+
+ /*
+ * read_extent -> io_time_reset may cause a transaction restart
+ * without returning an error, we need to check for that here:
+ */
+ if (!bch2_trans_relock(&trans)) {
+ ret = -EINTR;
+ break;
+ }
- bch2_btree_iter_set_pos(iter,
- POS(inode, rbio->bio.bi_iter.bi_sector));
+ bch2_btree_iter_set_pos(&iter,
+ POS(inum.inum, bvec_iter.bi_sector));
- k = bch2_btree_iter_peek_slot(iter);
+ k = bch2_btree_iter_peek_slot(&iter);
ret = bkey_err(k);
if (ret)
- goto err;
+ break;
- offset_into_extent = iter->pos.offset -
+ offset_into_extent = iter.pos.offset -
bkey_start_offset(k.k);
sectors = k.k->size - offset_into_extent;
bch2_bkey_buf_reassemble(&sk, c, k);
- ret = bch2_read_indirect_extent(&trans,
+ ret = bch2_read_indirect_extent(&trans, &data_btree,
&offset_into_extent, &sk);
if (ret)
- goto err;
+ break;
k = bkey_i_to_s_c(sk.k);
*/
bch2_trans_unlock(&trans);
- bytes = min(sectors, bio_sectors(&rbio->bio)) << 9;
- swap(rbio->bio.bi_iter.bi_size, bytes);
+ bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
+ swap(bvec_iter.bi_size, bytes);
- if (rbio->bio.bi_iter.bi_size == bytes)
+ if (bvec_iter.bi_size == bytes)
flags |= BCH_READ_LAST_FRAGMENT;
- bch2_read_extent(&trans, rbio, k, offset_into_extent, flags);
+ ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter.pos,
+ data_btree, k,
+ offset_into_extent, failed, flags);
+ if (ret)
+ break;
if (flags & BCH_READ_LAST_FRAGMENT)
break;
- swap(rbio->bio.bi_iter.bi_size, bytes);
- bio_advance(&rbio->bio, bytes);
+ swap(bvec_iter.bi_size, bytes);
+ bio_advance_iter(&rbio->bio, &bvec_iter, bytes);
+
+ ret = btree_trans_too_many_iters(&trans);
+ if (ret)
+ break;
}
-out:
- bch2_trans_exit(&trans);
- bch2_bkey_buf_exit(&sk, c);
- return;
err:
- if (ret == -EINTR)
+ bch2_trans_iter_exit(&trans, &iter);
+
+ if (ret == -EINTR || ret == READ_RETRY || ret == READ_RETRY_AVOID)
goto retry;
- bch_err_inum_ratelimited(c, inode,
- "read error %i from btree lookup", ret);
- rbio->bio.bi_status = BLK_STS_IOERR;
- bch2_rbio_done(rbio);
- goto out;
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&sk, c);
+
+ if (ret) {
+ bch_err_inum_ratelimited(c, inum.inum,
+ "read error %i from btree lookup", ret);
+ rbio->bio.bi_status = BLK_STS_IOERR;
+ bch2_rbio_done(rbio);
+ }
}
void bch2_fs_io_exit(struct bch_fs *c)