+// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
#include "alloc_foreground.h"
+#include "bkey_on_stack.h"
#include "btree_gc.h"
#include "btree_update.h"
+#include "btree_update_interior.h"
#include "buckets.h"
#include "disk_groups.h"
#include "inode.h"
static int bch2_migrate_index_update(struct bch_write_op *op)
{
struct bch_fs *c = op->c;
+ struct btree_trans trans;
+ struct btree_iter *iter;
struct migrate_write *m =
container_of(op, struct migrate_write, op);
struct keylist *keys = &op->insert_keys;
- struct btree_iter iter;
int ret = 0;
- bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS,
- bkey_start_pos(&bch2_keylist_front(keys)->k),
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+ bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+
+ iter = bch2_trans_get_iter(&trans, m->btree_id,
+ bkey_start_pos(&bch2_keylist_front(keys)->k),
+ BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
while (1) {
- struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
- struct bkey_i_extent *insert, *new =
+ struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
+ struct bkey_i *insert;
+ struct bkey_i_extent *new =
bkey_i_to_extent(bch2_keylist_front(keys));
BKEY_PADDED(k) _new, _insert;
const union bch_extent_entry *entry;
bool did_work = false;
int nr;
- if (btree_iter_err(k)) {
- ret = bch2_btree_iter_unlock(&iter);
+ ret = bkey_err(k);
+ if (ret)
break;
- }
if (bversion_cmp(k.k->version, new->k.version) ||
- !bkey_extent_is_data(k.k) ||
- !bch2_extent_matches_ptr(c, bkey_s_c_to_extent(k),
- m->ptr, m->offset))
+ !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
goto nomatch;
if (m->data_cmd == DATA_REWRITE &&
- !bch2_extent_has_device(bkey_s_c_to_extent(k),
- m->data_opts.rewrite_dev))
+ !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
goto nomatch;
bkey_reassemble(&_insert.k, k);
- insert = bkey_i_to_extent(&_insert.k);
+ insert = &_insert.k;
bkey_copy(&_new.k, bch2_keylist_front(keys));
new = bkey_i_to_extent(&_new.k);
+ bch2_cut_front(iter->pos, &new->k_i);
- bch2_cut_front(iter.pos, &insert->k_i);
- bch2_cut_back(new->k.p, &insert->k);
- bch2_cut_back(insert->k.p, &new->k);
+ bch2_cut_front(iter->pos, insert);
+ bch2_cut_back(new->k.p, insert);
+ bch2_cut_back(insert->k.p, &new->k_i);
if (m->data_cmd == DATA_REWRITE)
- bch2_bkey_drop_device(extent_i_to_s(insert).s,
+ bch2_bkey_drop_device(bkey_i_to_s(insert),
m->data_opts.rewrite_dev);
extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
- if (bch2_extent_has_device(extent_i_to_s_c(insert), p.ptr.dev)) {
+ if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
/*
* raced with another move op? extent already
* has a pointer to the device we just wrote
if (!did_work)
goto nomatch;
- bch2_extent_narrow_crcs(insert,
+ bch2_bkey_narrow_crcs(insert,
(struct bch_extent_crc_unpacked) { 0 });
- bch2_extent_normalize(c, extent_i_to_s(insert).s);
- bch2_extent_mark_replicas_cached(c, extent_i_to_s(insert),
- op->opts.background_target,
- op->opts.data_replicas);
+ bch2_extent_normalize(c, bkey_i_to_s(insert));
+ bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
+ op->opts.background_target,
+ op->opts.data_replicas);
/*
- * It's possible we race, and for whatever reason the extent now
- * has fewer replicas than when we last looked at it - meaning
- * we need to get a disk reservation here:
+ * If we're not fully overwriting @k, and it's compressed, we
+ * need a reservation for all the pointers in @insert
*/
- nr = bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(&insert->k_i)) -
- (bch2_bkey_nr_dirty_ptrs(k) + m->nr_ptrs_reserved);
- if (nr > 0) {
- /*
- * can't call bch2_disk_reservation_add() with btree
- * locks held, at least not without a song and dance
- */
- bch2_btree_iter_unlock(&iter);
+ nr = bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(insert)) -
+ m->nr_ptrs_reserved;
+ if (insert->k.size < k.k->size &&
+ bch2_bkey_sectors_compressed(k) &&
+ nr > 0) {
ret = bch2_disk_reservation_add(c, &op->res,
keylist_sectors(keys) * nr, 0);
if (ret)
goto next;
}
- ret = bch2_btree_insert_at(c, &op->res,
+ bch2_trans_update(&trans, iter, insert);
+
+ ret = bch2_trans_commit(&trans, &op->res,
op_journal_seq(op),
BTREE_INSERT_ATOMIC|
BTREE_INSERT_NOFAIL|
BTREE_INSERT_USE_RESERVE|
- m->data_opts.btree_insert_flags,
- BTREE_INSERT_ENTRY(&iter, &insert->k_i));
+ m->data_opts.btree_insert_flags);
if (!ret)
atomic_long_inc(&c->extent_migrate_done);
if (ret == -EINTR)
if (ret)
break;
next:
- while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
+ while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
bch2_keylist_pop_front(keys);
if (bch2_keylist_empty(keys))
goto out;
}
-
- bch2_cut_front(iter.pos, bch2_keylist_front(keys));
continue;
nomatch:
if (m->ctxt)
- atomic64_add(k.k->p.offset - iter.pos.offset,
+ atomic64_add(k.k->p.offset - iter->pos.offset,
&m->ctxt->stats->sectors_raced);
atomic_long_inc(&c->extent_migrate_raced);
trace_move_race(&new->k);
- bch2_btree_iter_next_slot(&iter);
+ bch2_btree_iter_next_slot(iter);
goto next;
}
out:
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_exit(&trans);
+ BUG_ON(ret == -EINTR);
return ret;
}
struct bch_io_opts io_opts,
enum data_cmd data_cmd,
struct data_opts data_opts,
+ enum btree_id btree_id,
struct bkey_s_c k)
{
int ret;
+ m->btree_id = btree_id;
m->data_cmd = data_cmd;
m->data_opts = data_opts;
m->nr_ptrs_reserved = 0;
switch (data_cmd) {
case DATA_ADD_REPLICAS: {
+ /*
+ * DATA_ADD_REPLICAS is used for moving data to a different
+ * device in the background, and due to compression the new copy
+ * might take up more space than the old copy:
+ */
+#if 0
int nr = (int) io_opts.data_replicas -
- bch2_bkey_nr_dirty_ptrs(k);
+ bch2_bkey_nr_ptrs_allocated(k);
+#endif
+ int nr = (int) io_opts.data_replicas;
if (nr > 0) {
m->op.nr_replicas = m->nr_ptrs_reserved = nr;
break;
}
case DATA_REWRITE: {
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
const union bch_extent_entry *entry;
struct extent_ptr_decoded p;
unsigned compressed_sectors = 0;
- extent_for_each_ptr_decode(bkey_s_c_to_extent(k), p, entry)
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
if (!p.ptr.cached &&
p.crc.compression_type != BCH_COMPRESSION_NONE &&
bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
{
struct moving_io *io = container_of(cl, struct moving_io, cl);
struct moving_context *ctxt = io->write.ctxt;
+ struct bvec_iter_all iter;
struct bio_vec *bv;
- int i;
bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
- bio_for_each_segment_all(bv, &io->write.op.wbio.bio, i)
+ bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
if (bv->bv_page)
__free_page(bv->bv_page);
struct moving_context *ctxt,
struct write_point_specifier wp,
struct bch_io_opts io_opts,
- struct bkey_s_c_extent e,
+ enum btree_id btree_id,
+ struct bkey_s_c k,
enum data_cmd data_cmd,
struct data_opts data_opts)
{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
struct moving_io *io;
const union bch_extent_entry *entry;
struct extent_ptr_decoded p;
- unsigned sectors = e.k->size, pages;
+ unsigned sectors = k.k->size, pages;
int ret = -ENOMEM;
move_ctxt_wait_event(ctxt,
SECTORS_IN_FLIGHT_PER_DEVICE);
/* write path might have to decompress data: */
- extent_for_each_ptr_decode(e, p, entry)
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
goto err;
io->write.ctxt = ctxt;
- io->read_sectors = e.k->size;
- io->write_sectors = e.k->size;
+ io->read_sectors = k.k->size;
+ io->write_sectors = k.k->size;
bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
bio_set_prio(&io->write.op.wbio.bio,
IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
- io->write.op.wbio.bio.bi_iter.bi_size = sectors << 9;
- bch2_bio_map(&io->write.op.wbio.bio, NULL);
- if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, GFP_KERNEL))
+ if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
+ GFP_KERNEL))
goto err_free;
- io->rbio.opts = io_opts;
+ io->rbio.c = c;
+ io->rbio.opts = io_opts;
bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
io->rbio.bio.bi_vcnt = pages;
bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
io->rbio.bio.bi_iter.bi_size = sectors << 9;
bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
- io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(e.k);
+ io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
io->rbio.bio.bi_end_io = move_read_endio;
ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
- data_cmd, data_opts, e.s_c);
+ data_cmd, data_opts, btree_id, k);
if (ret)
goto err_free_pages;
atomic64_inc(&ctxt->stats->keys_moved);
- atomic64_add(e.k->size, &ctxt->stats->sectors_moved);
+ atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
- trace_move_extent(e.k);
+ trace_move_extent(k.k);
atomic_add(io->read_sectors, &ctxt->read_sectors);
list_add_tail(&io->list, &ctxt->reads);
* ctxt when doing wakeup
*/
closure_get(&ctxt->cl);
- bch2_read_extent(c, &io->rbio, e.s_c,
+ bch2_read_extent(c, &io->rbio, k, 0,
BCH_READ_NODECODE|
BCH_READ_LAST_FRAGMENT);
return 0;
err_free:
kfree(io);
err:
- trace_move_alloc_fail(e.k);
+ trace_move_alloc_fail(k.k);
return ret;
}
-int bch2_move_data(struct bch_fs *c,
- struct bch_ratelimit *rate,
- struct write_point_specifier wp,
- struct bpos start,
- struct bpos end,
- move_pred_fn pred, void *arg,
- struct bch_move_stats *stats)
+static int __bch2_move_data(struct bch_fs *c,
+ struct moving_context *ctxt,
+ struct bch_ratelimit *rate,
+ struct write_point_specifier wp,
+ struct bpos start,
+ struct bpos end,
+ move_pred_fn pred, void *arg,
+ struct bch_move_stats *stats,
+ enum btree_id btree_id)
{
bool kthread = (current->flags & PF_KTHREAD) != 0;
- struct moving_context ctxt = { .stats = stats };
struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
- BKEY_PADDED(k) tmp;
+ struct bkey_on_stack sk;
+ struct btree_trans trans;
+ struct btree_iter *iter;
struct bkey_s_c k;
struct data_opts data_opts;
enum data_cmd data_cmd;
u64 delay, cur_inum = U64_MAX;
int ret = 0, ret2;
- closure_init_stack(&ctxt.cl);
- INIT_LIST_HEAD(&ctxt.reads);
- init_waitqueue_head(&ctxt.wait);
+ bkey_on_stack_init(&sk);
+ bch2_trans_init(&trans, c, 0, 0);
stats->data_type = BCH_DATA_USER;
- bch2_btree_iter_init(&stats->iter, c, BTREE_ID_EXTENTS, start,
- BTREE_ITER_PREFETCH);
+ stats->btree_id = btree_id;
+ stats->pos = POS_MIN;
+
+ iter = bch2_trans_get_iter(&trans, btree_id, start,
+ BTREE_ITER_PREFETCH);
if (rate)
bch2_ratelimit_reset(rate);
delay = rate ? bch2_ratelimit_delay(rate) : 0;
if (delay) {
- bch2_btree_iter_unlock(&stats->iter);
+ bch2_trans_unlock(&trans);
set_current_state(TASK_INTERRUPTIBLE);
}
schedule_timeout(delay);
if (unlikely(freezing(current))) {
- bch2_btree_iter_unlock(&stats->iter);
- move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
+ bch2_trans_unlock(&trans);
+ move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
try_to_freeze();
}
} while (delay);
peek:
- k = bch2_btree_iter_peek(&stats->iter);
+ k = bch2_btree_iter_peek(iter);
+
+ stats->pos = iter->pos;
+
if (!k.k)
break;
- ret = btree_iter_err(k);
+ ret = bkey_err(k);
if (ret)
break;
if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
break;
- if (!bkey_extent_is_data(k.k))
+ if (!bkey_extent_is_direct_data(k.k))
goto next_nondata;
- if (cur_inum != k.k->p.inode) {
+ if (btree_id == BTREE_ID_EXTENTS &&
+ cur_inum != k.k->p.inode) {
struct bch_inode_unpacked inode;
/* don't hold btree locks while looking up inode: */
- bch2_btree_iter_unlock(&stats->iter);
+ bch2_trans_unlock(&trans);
io_opts = bch2_opts_to_inode_opts(c->opts);
if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
}
/* unlock before doing IO: */
- bkey_reassemble(&tmp.k, k);
- k = bkey_i_to_s_c(&tmp.k);
- bch2_btree_iter_unlock(&stats->iter);
+ bkey_on_stack_reassemble(&sk, c, k);
+ k = bkey_i_to_s_c(sk.k);
+ bch2_trans_unlock(&trans);
- ret2 = bch2_move_extent(c, &ctxt, wp, io_opts,
- bkey_s_c_to_extent(k),
+ ret2 = bch2_move_extent(c, ctxt, wp, io_opts, btree_id, k,
data_cmd, data_opts);
if (ret2) {
if (ret2 == -ENOMEM) {
/* memory allocation failure, wait for some IO to finish */
- bch2_move_ctxt_wait_for_io(&ctxt);
+ bch2_move_ctxt_wait_for_io(ctxt);
continue;
}
if (rate)
bch2_ratelimit_increment(rate, k.k->size);
next:
- atomic64_add(k.k->size * bch2_bkey_nr_dirty_ptrs(k),
+ atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
&stats->sectors_seen);
next_nondata:
- bch2_btree_iter_next(&stats->iter);
- bch2_btree_iter_cond_resched(&stats->iter);
+ bch2_btree_iter_next(iter);
+ bch2_trans_cond_resched(&trans);
}
out:
- bch2_btree_iter_unlock(&stats->iter);
-
- move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
- closure_sync(&ctxt.cl);
-
- EBUG_ON(atomic_read(&ctxt.write_sectors));
-
- trace_move_data(c,
- atomic64_read(&stats->sectors_moved),
- atomic64_read(&stats->keys_moved));
+ ret = bch2_trans_exit(&trans) ?: ret;
+ bkey_on_stack_exit(&sk, c);
return ret;
}
-static int bch2_gc_data_replicas(struct bch_fs *c)
+int bch2_move_data(struct bch_fs *c,
+ struct bch_ratelimit *rate,
+ struct write_point_specifier wp,
+ struct bpos start,
+ struct bpos end,
+ move_pred_fn pred, void *arg,
+ struct bch_move_stats *stats)
{
- struct btree_iter iter;
- struct bkey_s_c k;
+ struct moving_context ctxt = { .stats = stats };
int ret;
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, (1 << BCH_DATA_USER)|(1 << BCH_DATA_CACHED));
-
- for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, POS_MIN,
- BTREE_ITER_PREFETCH, k) {
- ret = bch2_mark_bkey_replicas(c, k);
- if (ret)
- break;
- }
- ret = bch2_btree_iter_unlock(&iter) ?: ret;
-
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
-
- return ret;
-}
-
-static int bch2_gc_btree_replicas(struct bch_fs *c)
-{
- struct btree_iter iter;
- struct btree *b;
- unsigned id;
- int ret = 0;
+ closure_init_stack(&ctxt.cl);
+ INIT_LIST_HEAD(&ctxt.reads);
+ init_waitqueue_head(&ctxt.wait);
- mutex_lock(&c->replicas_gc_lock);
- bch2_replicas_gc_start(c, 1 << BCH_DATA_BTREE);
+ stats->data_type = BCH_DATA_USER;
- for (id = 0; id < BTREE_ID_NR; id++) {
- for_each_btree_node(&iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
- ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
+ ret = __bch2_move_data(c, &ctxt, rate, wp, start, end,
+ pred, arg, stats, BTREE_ID_EXTENTS) ?:
+ __bch2_move_data(c, &ctxt, rate, wp, start, end,
+ pred, arg, stats, BTREE_ID_REFLINK);
- bch2_btree_iter_cond_resched(&iter);
- }
+ move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
+ closure_sync(&ctxt.cl);
- ret = bch2_btree_iter_unlock(&iter) ?: ret;
- }
+ EBUG_ON(atomic_read(&ctxt.write_sectors));
- bch2_replicas_gc_end(c, ret);
- mutex_unlock(&c->replicas_gc_lock);
+ trace_move_data(c,
+ atomic64_read(&stats->sectors_moved),
+ atomic64_read(&stats->keys_moved));
return ret;
}
struct bch_move_stats *stats)
{
struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+ struct btree_trans trans;
+ struct btree_iter *iter;
struct btree *b;
unsigned id;
struct data_opts data_opts;
enum data_cmd cmd;
int ret = 0;
+ bch2_trans_init(&trans, c, 0, 0);
+
stats->data_type = BCH_DATA_BTREE;
for (id = 0; id < BTREE_ID_NR; id++) {
- for_each_btree_node(&stats->iter, c, id, POS_MIN, BTREE_ITER_PREFETCH, b) {
+ stats->btree_id = id;
+
+ for_each_btree_node(&trans, iter, id, POS_MIN,
+ BTREE_ITER_PREFETCH, b) {
+ stats->pos = iter->pos;
+
switch ((cmd = pred(c, arg,
bkey_i_to_s_c(&b->key),
&io_opts, &data_opts))) {
BUG();
}
- ret = bch2_btree_node_rewrite(c, &stats->iter,
+ ret = bch2_btree_node_rewrite(c, iter,
b->data->keys.seq, 0) ?: ret;
next:
- bch2_btree_iter_cond_resched(&stats->iter);
+ bch2_trans_cond_resched(&trans);
}
- ret = bch2_btree_iter_unlock(&stats->iter) ?: ret;
+ ret = bch2_trans_iter_free(&trans, iter) ?: ret;
}
+ bch2_trans_exit(&trans);
+
return ret;
}
ret = bch2_journal_flush_device_pins(&c->journal, -1);
ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
- ret = bch2_gc_btree_replicas(c) ?: ret;
+
+ while (1) {
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_nr_pending(c) ||
+ c->btree_roots_dirty);
+ if (!bch2_btree_interior_updates_nr_pending(c))
+ break;
+ bch2_journal_meta(&c->journal);
+ }
+
+ ret = bch2_replicas_gc2(c) ?: ret;
ret = bch2_move_data(c, NULL,
writepoint_hashed((unsigned long) current),
op.start,
op.end,
rereplicate_pred, c, stats) ?: ret;
- ret = bch2_gc_data_replicas(c) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
break;
case BCH_DATA_OP_MIGRATE:
if (op.migrate.dev >= c->sb.nr_devices)
ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
- ret = bch2_gc_btree_replicas(c) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
ret = bch2_move_data(c, NULL,
writepoint_hashed((unsigned long) current),
op.start,
op.end,
migrate_pred, &op, stats) ?: ret;
- ret = bch2_gc_data_replicas(c) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
break;
default:
ret = -EINVAL;
projects / bcachefs-tools-debian / blobdiff