#include "keylist.h"
#include "move.h"
#include "rebalance.h"
+#include "subvolume.h"
#include "super.h"
#include "super-io.h"
int bch2_sum_sector_overwrites(struct btree_trans *trans,
struct btree_iter *extent_iter,
struct bkey_i *new,
- bool *maybe_extending,
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;
*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));
: 0;
if (!*usage_increasing &&
- (new_replicas > bch2_bkey_replicas(c, old) ||
+ (new->k.p.snapshot != old.k->p.snapshot ||
+ new_replicas > bch2_bkey_replicas(c, old) ||
(!new_compressed && bch2_bkey_sectors_compressed(old))))
*usage_increasing = 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;
-
+ 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,
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, usage_increasing;
+ 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,
&usage_increasing,
&i_sectors_delta,
&disk_sectors_delta);
if (ret)
return ret;
- if (!usage_increasing)
- check_enospc = false;
-
if (disk_res &&
disk_sectors_delta > (s64) disk_res->sectors) {
ret = bch2_disk_reservation_add(trans->c, disk_res,
disk_sectors_delta - disk_res->sectors,
- !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);
- ret = PTR_ERR_OR_ZERO(inode_iter);
- if (ret)
- return ret;
-
- /*
- * 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);
-
- inode_p.inode.k.p.snapshot = iter->snapshot;
-
- ret = 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;
- if (ret)
- return ret;
- }
+ inode_u.bi_sectors += i_sectors_delta;
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);
- BUG_ON(ret == -ENOSPC);
+ 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,
+ 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);
+ ret = bch2_fpunch_at(&trans, &iter, inum, end, i_sectors_delta);
- bch2_trans_iter_put(&trans, iter);
+ bch2_trans_iter_exit(&trans, &iter);
bch2_trans_exit(&trans);
- if (ret == -EINTR)
- ret = 0;
-
- 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);
- k->k.p.snapshot = iter->snapshot;
+ ret = bch2_subvolume_get_snapshot(&trans, inum.subvol,
+ &sk.k->k.p.snapshot);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ break;
- bch2_bkey_buf_realloc(&sk, c, k->k.u64s);
- bkey_copy(sk.k, k);
- bch2_cut_front(iter->pos, sk.k);
+ 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->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_iter_put(&trans, iter);
bch2_trans_exit(&trans);
bch2_bkey_buf_exit(&sk, c);
ca = bch_dev_bkey_exists(c, ptr->dev);
if (to_entry(ptr + 1) < ptrs.end) {
- n = to_wbio(bio_clone_fast(&wbio->bio, GFP_NOIO,
- &ca->replica_set));
+ n = to_wbio(bio_alloc_clone(NULL, &wbio->bio,
+ GFP_NOIO, &ca->replica_set));
n->bio.bi_end_io = wbio->bio.bi_end_io;
n->bio.bi_private = wbio->bio.bi_private;
{
struct bch_fs *c = op->c;
struct bkey_i_extent *e;
- struct open_bucket *ob;
- unsigned i;
- BUG_ON(crc.compressed_size > wp->sectors_free);
- wp->sectors_free -= crc.compressed_size;
op->pos.offset += crc.uncompressed_size;
e = bkey_extent_init(op->insert_keys.top);
crc.nonce)
bch2_extent_crc_append(&e->k_i, crc);
- open_bucket_for_each(c, &wp->ptrs, ob, i) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, ob->ptr.dev);
- union bch_extent_entry *end =
- bkey_val_end(bkey_i_to_s(&e->k_i));
-
- end->ptr = ob->ptr;
- end->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
- end->ptr.cached = !ca->mi.durability ||
- (op->flags & BCH_WRITE_CACHED) != 0;
- end->ptr.offset += ca->mi.bucket_size - ob->sectors_free;
-
- e->k.u64s++;
-
- BUG_ON(crc.compressed_size > ob->sectors_free);
- ob->sectors_free -= crc.compressed_size;
- }
+ bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, crc.compressed_size,
+ op->flags & BCH_WRITE_CACHED);
bch2_keylist_push(&op->insert_keys);
}
? ((unsigned long) buf & (PAGE_SIZE - 1))
: 0), PAGE_SIZE);
- bio = bio_alloc_bioset(GFP_NOIO, pages, &c->bio_write);
+ pages = min(pages, BIO_MAX_VECS);
+
+ bio = bio_alloc_bioset(NULL, pages, 0,
+ GFP_NOIO, &c->bio_write);
wbio = wbio_init(bio);
wbio->put_bio = true;
/* copy WRITE_SYNC flag */
*/
bch2_bio_alloc_pages_pool(c, bio,
min_t(unsigned, output_available,
- c->sb.encoded_extent_max << 9));
+ c->opts.encoded_extent_max));
if (bio->bi_iter.bi_size < output_available)
*page_alloc_failed =
struct bch_fs *c = op->c;
struct nonce nonce = extent_nonce(op->version, op->crc);
struct bch_csum csum;
+ int ret;
if (!bch2_csum_type_is_encryption(op->crc.csum_type))
return 0;
if (bch2_crc_cmp(op->crc.csum, csum))
return -EIO;
- bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
+ ret = bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
op->crc.csum_type = 0;
op->crc.csum = (struct bch_csum) { 0, 0 };
- return 0;
+ return ret;
}
static enum prep_encoded_ret {
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 */
size_t dst_len, src_len;
if (page_alloc_failed &&
- bio_sectors(dst) < wp->sectors_free &&
- bio_sectors(dst) < c->sb.encoded_extent_max)
+ dst->bi_iter.bi_size < (wp->sectors_free << 9) &&
+ dst->bi_iter.bi_size < c->opts.encoded_extent_max)
break;
BUG_ON(op->compression_type &&
if (op->csum_type)
dst_len = min_t(unsigned, dst_len,
- c->sb.encoded_extent_max << 9);
+ c->opts.encoded_extent_max);
if (bounce) {
swap(dst->bi_iter.bi_size, dst_len);
crc.live_size = src_len >> 9;
swap(dst->bi_iter.bi_size, dst_len);
- bch2_encrypt_bio(c, op->csum_type,
- extent_nonce(version, crc), dst);
+ ret = bch2_encrypt_bio(c, op->csum_type,
+ extent_nonce(version, crc), dst);
+ if (ret)
+ goto err;
+
crc.csum = bch2_checksum_bio(c, op->csum_type,
extent_nonce(version, crc), dst);
crc.csum_type = op->csum_type;
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:
- bch_err(c, "error verifying existing checksum while "
- "rewriting existing data (memory corruption?)");
+ bch_err(c, "error verifying existing checksum while rewriting existing data (memory corruption?)");
ret = -EIO;
err:
if (to_wbio(dst)->bounce)
struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
struct bch_fs *c = op->c;
struct write_point *wp;
- struct bio *bio;
+ struct bio *bio = NULL;
bool skip_put = true;
unsigned nofs_flags;
int ret;
BKEY_EXTENT_U64s_MAX))
goto flush_io;
- if ((op->flags & BCH_WRITE_FROM_INTERNAL) &&
- percpu_ref_is_dying(&c->writes)) {
- ret = -EROFS;
- goto err;
- }
-
/*
* The copygc thread is now global, which means it's no longer
* freeing up space on specific disks, which means that
*/
wp = bch2_alloc_sectors_start(c,
op->target,
- op->opts.erasure_code,
+ op->opts.erasure_code && !(op->flags & BCH_WRITE_CACHED),
op->write_point,
&op->devs_have,
op->nr_replicas,
bch2_keylist_init(&op->insert_keys, op->inline_keys);
wbio_init(bio)->put_bio = false;
- if (bio_sectors(bio) & (c->opts.block_size - 1)) {
+ if (bio->bi_iter.bi_size & (c->opts.block_size - 1)) {
bch_err_inum_ratelimited(c, op->pos.inode,
"misaligned write");
op->error = -EIO;
}
if (c->opts.nochanges ||
- !percpu_ref_tryget(&c->writes)) {
+ !percpu_ref_tryget_live(&c->writes)) {
op->error = -EROFS;
goto err;
}
+ this_cpu_add(c->counters[BCH_COUNTER_io_write], bio_sectors(bio));
bch2_increment_clock(c, bio_sectors(bio), WRITE);
data_len = min_t(u64, bio->bi_iter.bi_size,
struct rhash_head hash;
struct bpos pos;
- struct migrate_write write;
+ struct data_update write;
struct bio_vec bi_inline_vecs[0]; /* must be last */
};
bch2_time_stats_update(&c->times[BCH_TIME_data_promote],
op->start_time);
- bch2_bio_free_pages_pool(c, &op->write.op.wbio.bio);
+ bch2_data_update_exit(&op->write);
promote_free(c, op);
}
static void promote_start(struct promote_op *op, struct bch_read_bio *rbio)
{
- struct bch_fs *c = rbio->c;
struct closure *cl = &op->cl;
struct bio *bio = &op->write.op.wbio.bio;
sizeof(struct bio_vec) * rbio->bio.bi_vcnt);
swap(bio->bi_vcnt, rbio->bio.bi_vcnt);
- bch2_migrate_read_done(&op->write, rbio);
-
closure_init(cl, NULL);
- closure_call(&op->write.op.cl, bch2_write, c->btree_update_wq, cl);
+ bch2_data_update_read_done(&op->write, rbio->pick.crc, cl);
closure_return_with_destructor(cl, promote_done);
}
unsigned pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
int ret;
- if (!percpu_ref_tryget(&c->writes))
+ if (!percpu_ref_tryget_live(&c->writes))
return NULL;
op = kzalloc(sizeof(*op) + sizeof(struct bio_vec) * pages, GFP_NOIO);
goto err;
rbio_init(&(*rbio)->bio, opts);
- bio_init(&(*rbio)->bio, (*rbio)->bio.bi_inline_vecs, pages);
+ bio_init(&(*rbio)->bio, NULL, (*rbio)->bio.bi_inline_vecs, pages, 0);
if (bch2_bio_alloc_pages(&(*rbio)->bio, sectors << 9,
GFP_NOIO))
goto err;
bio = &op->write.op.wbio.bio;
- bio_init(bio, bio->bi_inline_vecs, pages);
+ bio_init(bio, NULL, bio->bi_inline_vecs, pages, 0);
- ret = bch2_migrate_write_init(c, &op->write,
+ ret = bch2_data_update_init(c, &op->write,
writepoint_hashed((unsigned long) current),
opts,
- DATA_PROMOTE,
- (struct data_opts) {
+ (struct data_update_opts) {
.target = opts.promote_target,
- .nr_replicas = 1,
+ .extra_replicas = 1,
+ .write_flags = BCH_WRITE_ALLOC_NOWAIT|BCH_WRITE_CACHED,
},
btree_id, k);
BUG_ON(ret);
}
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, rbio->data_btree,
- rbio->read_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;
goto err;
out:
bch2_rbio_done(rbio);
- bch2_trans_iter_put(&trans, iter);
+ bch2_trans_iter_exit(&trans, &iter);
bch2_trans_exit(&trans);
bch2_bkey_buf_exit(&sk, c);
return;
struct bch_fs *c = rbio->c;
struct bvec_iter iter = rbio->bvec_iter;
unsigned flags = rbio->flags;
- u64 inode = rbio->read_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_MAY_PROMOTE;
if (flags & BCH_READ_NODECODE) {
- bch2_read_retry_nodecode(c, rbio, iter, inode, &failed, flags);
+ 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, inode, &failed, flags);
+ __bch2_read(c, rbio, iter, inum, &failed, flags);
}
}
struct bch_fs *c = rbio->c;
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, rbio->data_btree, rbio->data_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;
- ret = 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 nonce nonce = extent_nonce(rbio->version, crc);
unsigned nofs_flags;
struct bch_csum csum;
+ int ret;
nofs_flags = memalloc_nofs_save();
crc.live_size = bvec_iter_sectors(rbio->bvec_iter);
if (crc_is_compressed(crc)) {
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
+
if (bch2_bio_uncompress(c, src, dst, dst_iter, crc))
goto decompression_err;
} else {
BUG_ON(src->bi_iter.bi_size < dst_iter.bi_size);
src->bi_iter.bi_size = dst_iter.bi_size;
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
if (rbio->bounce) {
struct bvec_iter src_iter = src->bi_iter;
* Re encrypt data we decrypted, so it's consistent with
* rbio->crc:
*/
- bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ ret = bch2_encrypt_bio(c, crc.csum_type, nonce, src);
+ if (ret)
+ goto decrypt_err;
+
promote_start(rbio->promote, rbio);
rbio->promote = NULL;
}
}
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)",
+ "data checksum error: expected %0llx:%0llx got %0llx:%0llx (type %s)",
rbio->pick.crc.csum.hi, rbio->pick.crc.csum.lo,
- csum.hi, csum.lo, crc.csum_type);
+ csum.hi, csum.lo, bch2_csum_types[crc.csum_type]);
bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
goto out;
decompression_err:
"decompression error");
bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
goto out;
+decrypt_err:
+ bch_err_inum_ratelimited(c, rbio->read_pos.inode,
+ "decrypt error");
+ bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+ goto out;
}
static void bch2_read_endio(struct bio *bio)
return;
}
- if (rbio->pick.ptr.cached &&
- (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) ||
- ptr_stale(ca, &rbio->pick.ptr))) {
+ if (((rbio->flags & BCH_READ_RETRY_IF_STALE) && race_fault()) ||
+ ptr_stale(ca, &rbio->pick.ptr)) {
atomic_long_inc(&c->read_realloc_races);
if (rbio->flags & BCH_READ_RETRY_IF_STALE)
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;
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;
}
+static noinline void read_from_stale_dirty_pointer(struct btree_trans *trans,
+ struct bkey_s_c k,
+ struct bch_extent_ptr ptr)
+{
+ struct bch_fs *c = trans->c;
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr.dev);
+ struct btree_iter iter;
+ struct printbuf buf = PRINTBUF;
+ int ret;
+
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
+ PTR_BUCKET_POS(c, &ptr),
+ BTREE_ITER_CACHED);
+
+ prt_printf(&buf, "Attempting to read from stale dirty pointer:");
+ printbuf_indent_add(&buf, 2);
+ prt_newline(&buf);
+
+ bch2_bkey_val_to_text(&buf, c, k);
+ prt_newline(&buf);
+
+ prt_printf(&buf, "memory gen: %u", *bucket_gen(ca, iter.pos.offset));
+
+ ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
+ if (!ret) {
+ prt_newline(&buf);
+ bch2_bkey_val_to_text(&buf, c, k);
+ }
+
+ bch2_fs_inconsistent(c, "%s", buf.buf);
+
+ bch2_trans_iter_exit(trans, &iter);
+ printbuf_exit(&buf);
+}
+
int __bch2_read_extent(struct btree_trans *trans, struct bch_read_bio *orig,
struct bvec_iter iter, struct bpos read_pos,
enum btree_id data_btree, struct bkey_s_c k,
struct bch_fs *c = trans->c;
struct extent_ptr_decoded pick;
struct bch_read_bio *rbio = NULL;
- struct bch_dev *ca;
+ struct bch_dev *ca = NULL;
struct promote_op *promote = NULL;
bool bounce = false, read_full = false, narrow_crcs = false;
struct bpos data_pos = bkey_start_pos(k.k);
zero_fill_bio_iter(&orig->bio, iter);
goto out_read_done;
}
-
+retry_pick:
pick_ret = bch2_bkey_pick_read_device(c, k, failed, &pick);
/* hole or reservation - just zero fill: */
goto err;
}
- if (pick_ret > 0)
- ca = bch_dev_bkey_exists(c, pick.ptr.dev);
+ ca = bch_dev_bkey_exists(c, pick.ptr.dev);
+
+ /*
+ * Stale dirty pointers are treated as IO errors, but @failed isn't
+ * allocated unless we're in the retry path - so if we're not in the
+ * retry path, don't check here, it'll be caught in bch2_read_endio()
+ * and we'll end up in the retry path:
+ */
+ if ((flags & BCH_READ_IN_RETRY) &&
+ !pick.ptr.cached &&
+ unlikely(ptr_stale(ca, &pick.ptr))) {
+ read_from_stale_dirty_pointer(trans, k, pick.ptr);
+ bch2_mark_io_failure(failed, &pick);
+ goto retry_pick;
+ }
+
+ /*
+ * Unlock the iterator while the btree node's lock is still in
+ * cache, before doing the IO:
+ */
+ bch2_trans_unlock(trans);
if (flags & BCH_READ_NODECODE) {
/*
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)) ||
} else if (bounce) {
unsigned sectors = pick.crc.compressed_size;
- rbio = rbio_init(bio_alloc_bioset(GFP_NOIO,
+ rbio = rbio_init(bio_alloc_bioset(NULL,
DIV_ROUND_UP(sectors, PAGE_SECTORS),
+ 0,
+ GFP_NOIO,
&c->bio_read_split),
orig->opts);
* from the whole bio, in which case we don't want to retry and
* lose the error)
*/
- rbio = rbio_init(bio_clone_fast(&orig->bio, GFP_NOIO,
- &c->bio_read_split),
+ rbio = rbio_init(bio_alloc_clone(NULL, &orig->bio, GFP_NOIO,
+ &c->bio_read_split),
orig->opts);
rbio->bio.bi_iter = iter;
rbio->split = true;
/* XXX: only initialize this if needed */
rbio->devs_have = bch2_bkey_devs(k);
rbio->pick = pick;
+ rbio->subvol = orig->subvol;
rbio->read_pos = read_pos;
rbio->data_btree = data_btree;
rbio->data_pos = data_pos;
if (rbio->bounce)
trace_read_bounce(&rbio->bio);
+ this_cpu_add(c->counters[BCH_COUNTER_io_read], bio_sectors(&rbio->bio));
bch2_increment_clock(c, bio_sectors(&rbio->bio), READ);
/*
}
void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
- struct bvec_iter bvec_iter, u64 inode,
+ 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;
+ u32 snapshot;
int ret;
BUG_ON(flags & BCH_READ_NODECODE);
bch2_trans_init(&trans, c, 0, 0);
retry:
bch2_trans_begin(&trans);
+ iter = (struct btree_iter) { NULL };
+
+ ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
+ if (ret)
+ goto err;
- iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
- POS(inode, bvec_iter.bi_sector),
- BTREE_ITER_SLOTS);
+ bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
+ SPOS(inum.inum, bvec_iter.bi_sector, snapshot),
+ BTREE_ITER_SLOTS);
while (1) {
unsigned bytes, sectors, offset_into_extent;
enum btree_id data_btree = BTREE_ID_extents;
- bch2_btree_iter_set_pos(iter,
- POS(inode, bvec_iter.bi_sector));
+ /*
+ * 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(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)
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;
*/
sectors = min(sectors, k.k->size - offset_into_extent);
- /*
- * Unlock the iterator while the btree node's lock is still in
- * cache, before doing the IO:
- */
- bch2_trans_unlock(&trans);
-
bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
swap(bvec_iter.bi_size, bytes);
if (bvec_iter.bi_size == bytes)
flags |= BCH_READ_LAST_FRAGMENT;
- ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter->pos,
+ ret = __bch2_read_extent(&trans, rbio, bvec_iter, iter.pos,
data_btree, k,
offset_into_extent, failed, flags);
if (ret)
swap(bvec_iter.bi_size, bytes);
bio_advance_iter(&rbio->bio, &bvec_iter, bytes);
+
+ ret = btree_trans_too_many_iters(&trans);
+ if (ret)
+ break;
}
- bch2_trans_iter_put(&trans, iter);
+err:
+ bch2_trans_iter_exit(&trans, &iter);
if (ret == -EINTR || ret == READ_RETRY || ret == READ_RETRY_AVOID)
goto retry;
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&sk, c);
+
if (ret) {
- bch_err_inum_ratelimited(c, inode,
+ 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);
}
- bch2_trans_exit(&trans);
- bch2_bkey_buf_exit(&sk, c);
}
void bch2_fs_io_exit(struct bch_fs *c)
mempool_init_page_pool(&c->bio_bounce_pages,
max_t(unsigned,
c->opts.btree_node_size,
- c->sb.encoded_extent_max) /
- PAGE_SECTORS, 0) ||
+ c->opts.encoded_extent_max) /
+ PAGE_SIZE, 0) ||
rhashtable_init(&c->promote_table, &bch_promote_params))
return -ENOMEM;
projects / bcachefs-tools-debian / blobdiff