+// SPDX-License-Identifier: GPL-2.0
#include "bcachefs.h"
+#include "alloc_foreground.h"
+#include "backpointers.h"
+#include "bkey_buf.h"
#include "btree_gc.h"
#include "btree_update.h"
-#include "buckets.h"
+#include "btree_update_interior.h"
+#include "disk_groups.h"
+#include "ec.h"
+#include "error.h"
+#include "inode.h"
#include "io.h"
+#include "journal_reclaim.h"
#include "move.h"
+#include "replicas.h"
#include "super-io.h"
#include "keylist.h"
#include <linux/ioprio.h>
+#include <linux/kthread.h>
#include <trace/events/bcachefs.h>
-static struct bch_extent_ptr *bkey_find_ptr(struct bch_fs *c,
- struct bkey_s_extent e,
- struct bch_extent_ptr ptr)
+static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
{
- struct bch_extent_ptr *ptr2;
- unsigned bucket_bits = c->devs[ptr.dev]->bucket_bits;
+ mutex_lock(&c->data_progress_lock);
+ list_add(&stats->list, &c->data_progress_list);
+ mutex_unlock(&c->data_progress_lock);
+}
+
+static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
+{
+ mutex_lock(&c->data_progress_lock);
+ list_del(&stats->list);
+ mutex_unlock(&c->data_progress_lock);
+}
+
+struct moving_io {
+ struct list_head list;
+ struct closure cl;
+ bool read_completed;
+
+ unsigned read_sectors;
+ unsigned write_sectors;
+
+ struct bch_read_bio rbio;
+
+ struct data_update write;
+ /* Must be last since it is variable size */
+ struct bio_vec bi_inline_vecs[0];
+};
- extent_for_each_ptr(e, ptr2)
- if (ptr2->dev == ptr.dev &&
- ptr2->gen == ptr.gen &&
- (ptr2->offset >> bucket_bits) ==
- (ptr.offset >> bucket_bits))
- return ptr2;
+static void move_free(struct closure *cl)
+{
+ struct moving_io *io = container_of(cl, struct moving_io, cl);
+ struct moving_context *ctxt = io->write.ctxt;
+ struct bch_fs *c = ctxt->c;
- return NULL;
+ bch2_data_update_exit(&io->write);
+ wake_up(&ctxt->wait);
+ percpu_ref_put(&c->writes);
+ kfree(io);
}
-static struct bch_extent_ptr *bch2_migrate_matching_ptr(struct migrate_write *m,
- struct bkey_s_extent e)
+static void move_write_done(struct closure *cl)
{
- const struct bch_extent_ptr *ptr;
- struct bch_extent_ptr *ret;
+ struct moving_io *io = container_of(cl, struct moving_io, cl);
+ struct moving_context *ctxt = io->write.ctxt;
- if (m->move)
- ret = bkey_find_ptr(m->op.c, e, m->move_ptr);
- else
- extent_for_each_ptr(bkey_i_to_s_c_extent(&m->key), ptr)
- if ((ret = bkey_find_ptr(m->op.c, e, *ptr)))
- break;
+ if (io->write.op.error)
+ ctxt->write_error = true;
- return ret;
+ atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
+ closure_return_with_destructor(cl, move_free);
}
-static int bch2_migrate_index_update(struct bch_write_op *op)
+static void move_write(struct closure *cl)
{
- struct bch_fs *c = op->c;
- struct migrate_write *m =
- container_of(op, struct migrate_write, op);
- struct keylist *keys = &op->insert_keys;
- struct btree_iter iter;
- int ret = 0;
+ struct moving_io *io = container_of(cl, struct moving_io, cl);
- bch2_btree_iter_init(&iter, c, BTREE_ID_EXTENTS,
- bkey_start_pos(&bch2_keylist_front(keys)->k),
- BTREE_ITER_INTENT);
+ if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
+ closure_return_with_destructor(cl, move_free);
+ return;
+ }
- while (1) {
- struct bkey_s_extent insert =
- bkey_i_to_s_extent(bch2_keylist_front(keys));
- struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
- struct bch_extent_ptr *ptr;
- struct bkey_s_extent e;
- BKEY_PADDED(k) new;
+ atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
- if (!k.k) {
- ret = bch2_btree_iter_unlock(&iter);
- break;
- }
+ bch2_data_update_read_done(&io->write, io->rbio.pick.crc, cl);
+ continue_at(cl, move_write_done, NULL);
+}
- if (!bkey_extent_is_data(k.k))
- goto nomatch;
-
- bkey_reassemble(&new.k, k);
- bch2_cut_front(iter.pos, &new.k);
- bch2_cut_back(insert.k->p, &new.k.k);
- e = bkey_i_to_s_extent(&new.k);
-
- /* hack - promotes can race: */
- if (m->promote)
- extent_for_each_ptr(insert, ptr)
- if (bch2_extent_has_device(e.c, ptr->dev))
- goto nomatch;
-
- ptr = bch2_migrate_matching_ptr(m, e);
- if (ptr) {
- int nr_new_dirty = bch2_extent_nr_dirty_ptrs(insert.s_c);
- unsigned insert_flags =
- BTREE_INSERT_ATOMIC|
- BTREE_INSERT_NOFAIL;
-
- /* copygc uses btree node reserve: */
- if (m->move)
- insert_flags |= BTREE_INSERT_USE_RESERVE;
-
- if (m->move) {
- nr_new_dirty -= !ptr->cached;
- __bch2_extent_drop_ptr(e, ptr);
- }
+static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
+{
+ struct moving_io *io =
+ list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
- BUG_ON(nr_new_dirty < 0);
+ return io && io->read_completed ? io : NULL;
+}
- memcpy_u64s(extent_entry_last(e),
- insert.v,
- bkey_val_u64s(insert.k));
- e.k->u64s += bkey_val_u64s(insert.k);
+static void move_read_endio(struct bio *bio)
+{
+ struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
+ struct moving_context *ctxt = io->write.ctxt;
- bch2_extent_narrow_crcs(e);
- bch2_extent_drop_redundant_crcs(e);
- bch2_extent_normalize(c, e.s);
- bch2_extent_mark_replicas_cached(c, e, nr_new_dirty);
+ atomic_sub(io->read_sectors, &ctxt->read_sectors);
+ io->read_completed = true;
- ret = bch2_btree_insert_at(c, &op->res,
- NULL, op_journal_seq(op),
- insert_flags,
- BTREE_INSERT_ENTRY(&iter, &new.k));
- if (ret && ret != -EINTR)
- break;
- } else {
-nomatch:
- bch2_btree_iter_advance_pos(&iter);
- }
+ wake_up(&ctxt->wait);
+ closure_put(&ctxt->cl);
+}
- while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
- bch2_keylist_pop_front(keys);
- if (bch2_keylist_empty(keys))
- goto out;
- }
+static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
+{
+ struct moving_io *io;
+
+ if (trans)
+ bch2_trans_unlock(trans);
- bch2_cut_front(iter.pos, bch2_keylist_front(keys));
+ while ((io = next_pending_write(ctxt))) {
+ list_del(&io->list);
+ closure_call(&io->cl, move_write, NULL, &ctxt->cl);
}
-out:
- bch2_btree_iter_unlock(&iter);
- return ret;
}
-void bch2_migrate_write_init(struct bch_fs *c,
- struct migrate_write *m,
- struct write_point *wp,
- struct bkey_s_c k,
- const struct bch_extent_ptr *move_ptr,
- unsigned flags)
-{
- bkey_reassemble(&m->key, k);
+#define move_ctxt_wait_event(_ctxt, _trans, _cond) \
+do { \
+ do_pending_writes(_ctxt, _trans); \
+ \
+ if (_cond) \
+ break; \
+ __wait_event((_ctxt)->wait, \
+ next_pending_write(_ctxt) || (_cond)); \
+} while (1)
- m->promote = false;
- m->move = move_ptr != NULL;
- if (move_ptr)
- m->move_ptr = *move_ptr;
+static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
+ struct btree_trans *trans)
+{
+ unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
- if (bkey_extent_is_cached(k.k) ||
- (move_ptr && move_ptr->cached))
- flags |= BCH_WRITE_CACHED;
+ move_ctxt_wait_event(ctxt, trans,
+ !atomic_read(&ctxt->write_sectors) ||
+ atomic_read(&ctxt->write_sectors) != sectors_pending);
+}
- bch2_write_op_init(&m->op, c, &m->wbio,
- (struct disk_reservation) { 0 },
- wp,
- bkey_start_pos(k.k),
- NULL, flags);
+void bch2_moving_ctxt_exit(struct moving_context *ctxt)
+{
+ move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
+ closure_sync(&ctxt->cl);
+ EBUG_ON(atomic_read(&ctxt->write_sectors));
- if (m->move)
- m->op.alloc_reserve = RESERVE_MOVINGGC;
+ if (ctxt->stats) {
+ progress_list_del(ctxt->c, ctxt->stats);
- m->op.nonce = extent_current_nonce(bkey_s_c_to_extent(k));
- m->op.nr_replicas = 1;
- m->op.index_update_fn = bch2_migrate_index_update;
+ trace_move_data(ctxt->c,
+ atomic64_read(&ctxt->stats->sectors_moved),
+ atomic64_read(&ctxt->stats->keys_moved));
+ }
}
-static void migrate_bio_init(struct moving_io *io, struct bio *bio,
- unsigned sectors)
+void bch2_moving_ctxt_init(struct moving_context *ctxt,
+ struct bch_fs *c,
+ struct bch_ratelimit *rate,
+ struct bch_move_stats *stats,
+ struct write_point_specifier wp,
+ bool wait_on_copygc)
{
- bio_init(bio, io->bi_inline_vecs,
- DIV_ROUND_UP(sectors, PAGE_SECTORS));
- bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
+ memset(ctxt, 0, sizeof(*ctxt));
- bio->bi_iter.bi_size = sectors << 9;
- bio->bi_private = &io->cl;
- bch2_bio_map(bio, NULL);
+ ctxt->c = c;
+ ctxt->rate = rate;
+ ctxt->stats = stats;
+ ctxt->wp = wp;
+ ctxt->wait_on_copygc = wait_on_copygc;
+
+ closure_init_stack(&ctxt->cl);
+ INIT_LIST_HEAD(&ctxt->reads);
+ init_waitqueue_head(&ctxt->wait);
+
+ if (stats) {
+ progress_list_add(c, stats);
+ stats->data_type = BCH_DATA_user;
+ }
}
-static void moving_io_destructor(struct closure *cl)
+void bch_move_stats_init(struct bch_move_stats *stats, char *name)
{
- struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct moving_context *ctxt = io->ctxt;
- struct bio_vec *bv;
- int i;
+ memset(stats, 0, sizeof(*stats));
+ scnprintf(stats->name, sizeof(stats->name), "%s", name);
+}
- //if (io->replace.failures)
- // trace_copy_collision(q, &io->key.k);
+static int bch2_move_extent(struct btree_trans *trans,
+ struct moving_context *ctxt,
+ struct bch_io_opts io_opts,
+ enum btree_id btree_id,
+ struct bkey_s_c k,
+ struct data_update_opts data_opts)
+{
+ struct bch_fs *c = trans->c;
+ 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 = k.k->size, pages;
+ int ret = -ENOMEM;
- atomic_sub(io->write.key.k.size, &ctxt->sectors_in_flight);
- wake_up(&ctxt->wait);
+ if (!percpu_ref_tryget_live(&c->writes))
+ return -EROFS;
+
+ /* write path might have to decompress data: */
+ 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);
+ io = kzalloc(sizeof(struct moving_io) +
+ sizeof(struct bio_vec) * pages, GFP_KERNEL);
+ if (!io)
+ goto err;
- bio_for_each_segment_all(bv, &io->write.wbio.bio, i)
- if (bv->bv_page)
- __free_page(bv->bv_page);
+ io->write.ctxt = ctxt;
+ io->read_sectors = k.k->size;
+ io->write_sectors = k.k->size;
+ bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
+ bio_set_prio(&io->write.op.wbio.bio,
+ IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
+
+ if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
+ GFP_KERNEL))
+ goto err_free;
+
+ io->rbio.c = c;
+ io->rbio.opts = io_opts;
+ bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
+ 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(k.k);
+ io->rbio.bio.bi_end_io = move_read_endio;
+
+ ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
+ data_opts, btree_id, k);
+ if (ret)
+ goto err_free_pages;
+
+ io->write.ctxt = ctxt;
+
+ atomic64_inc(&ctxt->stats->keys_moved);
+ atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
+ this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
+
+ trace_move_extent(k.k);
+
+ atomic_add(io->read_sectors, &ctxt->read_sectors);
+ list_add_tail(&io->list, &ctxt->reads);
+
+ /*
+ * dropped by move_read_endio() - guards against use after free of
+ * ctxt when doing wakeup
+ */
+ closure_get(&ctxt->cl);
+ bch2_read_extent(trans, &io->rbio,
+ bkey_start_pos(k.k),
+ btree_id, k, 0,
+ BCH_READ_NODECODE|
+ BCH_READ_LAST_FRAGMENT);
+ return 0;
+err_free_pages:
+ bio_free_pages(&io->write.op.wbio.bio);
+err_free:
kfree(io);
+err:
+ percpu_ref_put(&c->writes);
+ trace_move_alloc_mem_fail(k.k);
+ return ret;
}
-static void moving_error(struct moving_context *ctxt, unsigned flag)
+static int lookup_inode(struct btree_trans *trans, struct bpos pos,
+ struct bch_inode_unpacked *inode)
{
- atomic_inc(&ctxt->error_count);
- //atomic_or(flag, &ctxt->error_flags);
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ int ret;
+
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
+ BTREE_ITER_ALL_SNAPSHOTS);
+ k = bch2_btree_iter_peek(&iter);
+ ret = bkey_err(k);
+ if (ret)
+ goto err;
+
+ if (!k.k || bkey_cmp(k.k->p, pos)) {
+ ret = -ENOENT;
+ goto err;
+ }
+
+ ret = bkey_is_inode(k.k) ? 0 : -EIO;
+ if (ret)
+ goto err;
+
+ ret = bch2_inode_unpack(k, inode);
+ if (ret)
+ goto err;
+err:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
}
-static void moving_io_after_write(struct closure *cl)
+static int move_ratelimit(struct btree_trans *trans,
+ struct moving_context *ctxt)
{
- struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct moving_context *ctxt = io->ctxt;
+ struct bch_fs *c = trans->c;
+ u64 delay;
+
+ if (ctxt->wait_on_copygc) {
+ bch2_trans_unlock(trans);
+ wait_event_killable(c->copygc_running_wq,
+ !c->copygc_running ||
+ kthread_should_stop());
+ }
- if (io->write.op.error)
- moving_error(ctxt, MOVING_FLAG_WRITE);
+ do {
+ delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
+
+ if (delay) {
+ bch2_trans_unlock(trans);
+ set_current_state(TASK_INTERRUPTIBLE);
+ }
+
+ if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
+ __set_current_state(TASK_RUNNING);
+ return 1;
+ }
+
+ if (delay)
+ schedule_timeout(delay);
+
+ if (unlikely(freezing(current))) {
+ move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
+ try_to_freeze();
+ }
+ } while (delay);
+
+ move_ctxt_wait_event(ctxt, trans,
+ atomic_read(&ctxt->write_sectors) <
+ c->opts.move_bytes_in_flight >> 9);
- moving_io_destructor(cl);
+ move_ctxt_wait_event(ctxt, trans,
+ atomic_read(&ctxt->read_sectors) <
+ c->opts.move_bytes_in_flight >> 9);
+
+ return 0;
}
-static void write_moving(struct moving_io *io)
+static int move_get_io_opts(struct btree_trans *trans,
+ struct bch_io_opts *io_opts,
+ struct bkey_s_c k, u64 *cur_inum)
{
- struct bch_write_op *op = &io->write.op;
+ struct bch_inode_unpacked inode;
+ int ret;
+
+ if (*cur_inum == k.k->p.inode)
+ return 0;
+
+ *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
+
+ ret = lookup_inode(trans,
+ SPOS(0, k.k->p.inode, k.k->p.snapshot),
+ &inode);
+ if (ret == -EINTR)
+ return ret;
+
+ if (!ret)
+ bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
+
+ *cur_inum = k.k->p.inode;
+ return 0;
+}
+
+static int __bch2_move_data(struct moving_context *ctxt,
+ struct bpos start,
+ struct bpos end,
+ move_pred_fn pred, void *arg,
+ enum btree_id btree_id)
+{
+ struct bch_fs *c = ctxt->c;
+ struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+ struct bkey_buf sk;
+ struct btree_trans trans;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct data_update_opts data_opts;
+ u64 cur_inum = U64_MAX;
+ int ret = 0, ret2;
+
+ bch2_bkey_buf_init(&sk);
+ bch2_trans_init(&trans, c, 0, 0);
- if (op->error) {
- closure_return_with_destructor(&io->cl, moving_io_destructor);
- } else {
- closure_call(&op->cl, bch2_write, NULL, &io->cl);
- closure_return_with_destructor(&io->cl, moving_io_after_write);
+ ctxt->stats->data_type = BCH_DATA_user;
+ ctxt->stats->btree_id = btree_id;
+ ctxt->stats->pos = start;
+
+ bch2_trans_iter_init(&trans, &iter, btree_id, start,
+ BTREE_ITER_PREFETCH|
+ BTREE_ITER_ALL_SNAPSHOTS);
+
+ if (ctxt->rate)
+ bch2_ratelimit_reset(ctxt->rate);
+
+ while (!move_ratelimit(&trans, ctxt)) {
+ bch2_trans_begin(&trans);
+
+ k = bch2_btree_iter_peek(&iter);
+ if (!k.k)
+ break;
+
+ ret = bkey_err(k);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ break;
+
+ if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
+ break;
+
+ ctxt->stats->pos = iter.pos;
+
+ if (!bkey_extent_is_direct_data(k.k))
+ goto next_nondata;
+
+ ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
+ if (ret)
+ continue;
+
+ memset(&data_opts, 0, sizeof(data_opts));
+ if (!pred(c, arg, k, &io_opts, &data_opts))
+ goto next;
+
+ /*
+ * The iterator gets unlocked by __bch2_read_extent - need to
+ * save a copy of @k elsewhere:
+ */
+ bch2_bkey_buf_reassemble(&sk, c, k);
+ k = bkey_i_to_s_c(sk.k);
+
+ ret2 = bch2_move_extent(&trans, ctxt, io_opts,
+ btree_id, k, data_opts);
+ if (ret2) {
+ if (ret2 == -EINTR)
+ continue;
+
+ if (ret2 == -ENOMEM) {
+ /* memory allocation failure, wait for some IO to finish */
+ bch2_move_ctxt_wait_for_io(ctxt, &trans);
+ continue;
+ }
+
+ /* XXX signal failure */
+ goto next;
+ }
+
+ if (ctxt->rate)
+ bch2_ratelimit_increment(ctxt->rate, k.k->size);
+next:
+ atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
+next_nondata:
+ bch2_btree_iter_advance(&iter);
}
+
+ bch2_trans_iter_exit(&trans, &iter);
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&sk, c);
+
+ return ret;
}
-static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
+int bch2_move_data(struct bch_fs *c,
+ enum btree_id start_btree_id, struct bpos start_pos,
+ enum btree_id end_btree_id, struct bpos end_pos,
+ struct bch_ratelimit *rate,
+ struct bch_move_stats *stats,
+ struct write_point_specifier wp,
+ bool wait_on_copygc,
+ move_pred_fn pred, void *arg)
{
- struct moving_io *io =
- list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
+ struct moving_context ctxt;
+ enum btree_id id;
+ int ret;
+
+ bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
+
+ for (id = start_btree_id;
+ id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
+ id++) {
+ stats->btree_id = id;
+
+ if (id != BTREE_ID_extents &&
+ id != BTREE_ID_reflink)
+ continue;
+
+ ret = __bch2_move_data(&ctxt,
+ id == start_btree_id ? start_pos : POS_MIN,
+ id == end_btree_id ? end_pos : POS_MAX,
+ pred, arg, id);
+ if (ret)
+ break;
+ }
- return io && io->read_completed ? io : NULL;
+ bch2_moving_ctxt_exit(&ctxt);
+
+ return ret;
}
-static void read_moving_endio(struct bio *bio)
+static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
{
- struct closure *cl = bio->bi_private;
- struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct moving_context *ctxt = io->ctxt;
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ int ret;
+
+ bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
+ bucket, BTREE_ITER_CACHED);
+again:
+ k = bch2_btree_iter_peek_slot(&iter);
+ ret = bkey_err(k);
+
+ if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
+ struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
+
+ if (a.v->gen == gen &&
+ a.v->dirty_sectors) {
+ struct printbuf buf = PRINTBUF;
+
+ if (a.v->data_type == BCH_DATA_btree) {
+ bch2_trans_unlock(trans);
+ if (bch2_btree_interior_updates_flush(c))
+ goto again;
+ }
- trace_move_read_done(&io->write.key.k);
+ prt_str(&buf, "failed to evacuate bucket ");
+ bch2_bkey_val_to_text(&buf, c, k);
- if (bio->bi_error) {
- io->write.op.error = bio->bi_error;
- moving_error(io->ctxt, MOVING_FLAG_READ);
+ bch2_trans_inconsistent(trans, "%s", buf.buf);
+ printbuf_exit(&buf);
+ }
}
- io->read_completed = true;
- if (next_pending_write(ctxt))
- wake_up(&ctxt->wait);
-
- closure_put(&ctxt->cl);
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
}
-static void __bch2_data_move(struct closure *cl)
+int __bch2_evacuate_bucket(struct moving_context *ctxt,
+ struct bpos bucket, int gen,
+ struct data_update_opts _data_opts)
{
- struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct bch_fs *c = io->write.op.c;
- struct extent_pick_ptr pick;
+ struct bch_fs *c = ctxt->c;
+ struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+ struct btree_trans trans;
+ struct btree_iter iter;
+ struct bkey_buf sk;
+ struct bch_backpointer bp;
+ struct data_update_opts data_opts;
+ u64 bp_offset = 0, cur_inum = U64_MAX;
+ int ret = 0;
- bch2_extent_pick_ptr_avoiding(c, bkey_i_to_s_c(&io->write.key),
- io->ctxt->avoid, &pick);
- if (IS_ERR_OR_NULL(pick.ca))
- closure_return_with_destructor(cl, moving_io_destructor);
+ bch2_bkey_buf_init(&sk);
+ bch2_trans_init(&trans, c, 0, 0);
- bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
- io->rbio.bio.bi_iter.bi_sector = bkey_start_offset(&io->write.key.k);
- io->rbio.bio.bi_end_io = read_moving_endio;
+ while (!(ret = move_ratelimit(&trans, ctxt))) {
+ bch2_trans_begin(&trans);
- /*
- * dropped by read_moving_endio() - guards against use after free of
- * ctxt when doing wakeup
- */
- closure_get(&io->ctxt->cl);
+ ret = bch2_get_next_backpointer(&trans, bucket, gen,
+ &bp_offset, &bp);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ goto err;
+ if (bp_offset == U64_MAX)
+ break;
+
+ if (!bp.level) {
+ const struct bch_extent_ptr *ptr;
+ struct bkey_s_c k;
+ unsigned i = 0;
+
+ k = bch2_backpointer_get_key(&trans, &iter,
+ bucket, bp_offset, bp);
+ ret = bkey_err(k);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ goto err;
+ if (!k.k)
+ continue;
+
+ bch2_bkey_buf_reassemble(&sk, c, k);
+ k = bkey_i_to_s_c(sk.k);
+ bch2_trans_iter_exit(&trans, &iter);
+
+ ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
+ if (ret)
+ continue;
+
+ data_opts = _data_opts;
+ data_opts.target = io_opts.background_target;
+ data_opts.rewrite_ptrs = 0;
+
+ bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
+ if (ptr->dev == bucket.inode)
+ data_opts.rewrite_ptrs |= 1U << i;
+ i++;
+ }
+
+ ret = bch2_move_extent(&trans, ctxt, io_opts,
+ bp.btree_id, k, data_opts);
+ if (ret == -EINTR)
+ continue;
+ if (ret == -ENOMEM) {
+ /* memory allocation failure, wait for some IO to finish */
+ bch2_move_ctxt_wait_for_io(ctxt, &trans);
+ continue;
+ }
+ if (ret)
+ goto err;
- bch2_read_extent(c, &io->rbio,
- bkey_i_to_s_c(&io->write.key),
- &pick, BCH_READ_IS_LAST);
+ if (ctxt->rate)
+ bch2_ratelimit_increment(ctxt->rate, k.k->size);
+ atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
+ } else {
+ struct btree *b;
+
+ b = bch2_backpointer_get_node(&trans, &iter,
+ bucket, bp_offset, bp);
+ ret = PTR_ERR_OR_ZERO(b);
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ goto err;
+ if (!b)
+ continue;
+
+ ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
+ bch2_trans_iter_exit(&trans, &iter);
+
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ goto err;
+
+ if (ctxt->rate)
+ bch2_ratelimit_increment(ctxt->rate,
+ c->opts.btree_node_size >> 9);
+ atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
+ atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
+ }
+
+ bp_offset++;
+ }
+
+ if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
+ bch2_trans_unlock(&trans);
+ move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
+ closure_sync(&ctxt->cl);
+ if (!ctxt->write_error)
+ lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
+ }
+err:
+ bch2_trans_exit(&trans);
+ bch2_bkey_buf_exit(&sk, c);
+ return ret;
}
-int bch2_data_move(struct bch_fs *c,
- struct moving_context *ctxt,
- struct write_point *wp,
- struct bkey_s_c k,
- const struct bch_extent_ptr *move_ptr)
+int bch2_evacuate_bucket(struct bch_fs *c,
+ struct bpos bucket, int gen,
+ struct data_update_opts data_opts,
+ struct bch_ratelimit *rate,
+ struct bch_move_stats *stats,
+ struct write_point_specifier wp,
+ bool wait_on_copygc)
{
- struct moving_io *io;
+ struct moving_context ctxt;
+ int ret;
- io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec) *
- DIV_ROUND_UP(k.k->size, PAGE_SECTORS),
- GFP_KERNEL);
- if (!io)
- return -ENOMEM;
+ bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
+ ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
+ bch2_moving_ctxt_exit(&ctxt);
- io->ctxt = ctxt;
+ return ret;
+}
- migrate_bio_init(io, &io->rbio.bio, k.k->size);
+typedef bool (*move_btree_pred)(struct bch_fs *, void *,
+ struct btree *, struct bch_io_opts *,
+ struct data_update_opts *);
- if (bio_alloc_pages(&io->rbio.bio, GFP_KERNEL)) {
- kfree(io);
- return -ENOMEM;
- }
+static int bch2_move_btree(struct bch_fs *c,
+ enum btree_id start_btree_id, struct bpos start_pos,
+ enum btree_id end_btree_id, struct bpos end_pos,
+ move_btree_pred pred, void *arg,
+ struct bch_move_stats *stats)
+{
+ bool kthread = (current->flags & PF_KTHREAD) != 0;
+ struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+ struct btree_trans trans;
+ struct btree_iter iter;
+ struct btree *b;
+ enum btree_id id;
+ struct data_update_opts data_opts;
+ int ret = 0;
- migrate_bio_init(io, &io->write.wbio.bio, k.k->size);
- bio_get(&io->write.wbio.bio);
- io->write.wbio.bio.bi_iter.bi_sector = bkey_start_offset(k.k);
+ bch2_trans_init(&trans, c, 0, 0);
+ progress_list_add(c, stats);
- bch2_migrate_write_init(c, &io->write, wp, k, move_ptr, 0);
+ stats->data_type = BCH_DATA_btree;
- trace_move_read(&io->write.key.k);
+ for (id = start_btree_id;
+ id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
+ id++) {
+ stats->btree_id = id;
- ctxt->keys_moved++;
- ctxt->sectors_moved += k.k->size;
- if (ctxt->rate)
- bch2_ratelimit_increment(ctxt->rate, k.k->size);
+ bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
+ BTREE_ITER_PREFETCH);
+retry:
+ ret = 0;
+ while (bch2_trans_begin(&trans),
+ (b = bch2_btree_iter_peek_node(&iter)) &&
+ !(ret = PTR_ERR_OR_ZERO(b))) {
+ if (kthread && kthread_should_stop())
+ break;
- atomic_add(k.k->size, &ctxt->sectors_in_flight);
- list_add_tail(&io->list, &ctxt->reads);
+ if ((cmp_int(id, end_btree_id) ?:
+ bpos_cmp(b->key.k.p, end_pos)) > 0)
+ break;
- closure_call(&io->cl, __bch2_data_move, NULL, &ctxt->cl);
- return 0;
+ stats->pos = iter.pos;
+
+ if (!pred(c, arg, b, &io_opts, &data_opts))
+ goto next;
+
+ ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
+ if (ret == -EINTR)
+ continue;
+ if (ret)
+ break;
+next:
+ bch2_btree_iter_next_node(&iter);
+ }
+ if (ret == -EINTR)
+ goto retry;
+
+ bch2_trans_iter_exit(&trans, &iter);
+
+ if (kthread && kthread_should_stop())
+ break;
+ }
+
+ bch2_trans_exit(&trans);
+
+ if (ret)
+ bch_err(c, "error %i in bch2_move_btree", ret);
+
+ bch2_btree_interior_updates_flush(c);
+
+ progress_list_del(c, stats);
+ return ret;
}
-static void do_pending_writes(struct moving_context *ctxt)
+static bool rereplicate_pred(struct bch_fs *c, void *arg,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts,
+ struct data_update_opts *data_opts)
{
- struct moving_io *io;
+ unsigned nr_good = bch2_bkey_durability(c, k);
+ unsigned replicas = bkey_is_btree_ptr(k.k)
+ ? c->opts.metadata_replicas
+ : io_opts->data_replicas;
+
+ if (!nr_good || nr_good >= replicas)
+ return false;
+
+ data_opts->target = 0;
+ data_opts->extra_replicas = replicas - nr_good;
+ data_opts->btree_insert_flags = 0;
+ return true;
+}
- while ((io = next_pending_write(ctxt))) {
- list_del(&io->list);
- trace_move_write(&io->write.key.k);
- write_moving(io);
+static bool migrate_pred(struct bch_fs *c, void *arg,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts,
+ struct data_update_opts *data_opts)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const struct bch_extent_ptr *ptr;
+ struct bch_ioctl_data *op = arg;
+ unsigned i = 0;
+
+ data_opts->rewrite_ptrs = 0;
+ data_opts->target = 0;
+ data_opts->extra_replicas = 0;
+ data_opts->btree_insert_flags = 0;
+
+ bkey_for_each_ptr(ptrs, ptr) {
+ if (ptr->dev == op->migrate.dev)
+ data_opts->rewrite_ptrs |= 1U << i;
+ i++;
}
+
+ return data_opts->rewrite_ptrs != 0;;
}
-#define move_ctxt_wait_event(_ctxt, _cond) \
-do { \
- do_pending_writes(_ctxt); \
- \
- if (_cond) \
- break; \
- __wait_event((_ctxt)->wait, \
- next_pending_write(_ctxt) || (_cond)); \
-} while (1)
+static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
+ struct btree *b,
+ struct bch_io_opts *io_opts,
+ struct data_update_opts *data_opts)
+{
+ return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
+}
-int bch2_move_ctxt_wait(struct moving_context *ctxt)
+static bool migrate_btree_pred(struct bch_fs *c, void *arg,
+ struct btree *b,
+ struct bch_io_opts *io_opts,
+ struct data_update_opts *data_opts)
{
- move_ctxt_wait_event(ctxt,
- atomic_read(&ctxt->sectors_in_flight) <
- ctxt->max_sectors_in_flight);
+ return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
+}
+
+static bool bformat_needs_redo(struct bkey_format *f)
+{
+ unsigned i;
+
+ for (i = 0; i < f->nr_fields; i++) {
+ unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
+ u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
+ u64 field_offset = le64_to_cpu(f->field_offset[i]);
- return ctxt->rate
- ? bch2_ratelimit_wait_freezable_stoppable(ctxt->rate)
- : 0;
+ if (f->bits_per_field[i] > unpacked_bits)
+ return true;
+
+ if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
+ return true;
+
+ if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
+ unpacked_mask) <
+ field_offset)
+ return true;
+ }
+
+ return false;
}
-void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
+static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
+ struct btree *b,
+ struct bch_io_opts *io_opts,
+ struct data_update_opts *data_opts)
{
- unsigned sectors_pending = atomic_read(&ctxt->sectors_in_flight);
+ if (b->version_ondisk != c->sb.version ||
+ btree_node_need_rewrite(b) ||
+ bformat_needs_redo(&b->format)) {
+ data_opts->target = 0;
+ data_opts->extra_replicas = 0;
+ data_opts->btree_insert_flags = 0;
+ return true;
+ }
- move_ctxt_wait_event(ctxt,
- !atomic_read(&ctxt->sectors_in_flight) ||
- atomic_read(&ctxt->sectors_in_flight) != sectors_pending);
+ return false;
}
-void bch2_move_ctxt_exit(struct moving_context *ctxt)
+int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
{
- move_ctxt_wait_event(ctxt, !atomic_read(&ctxt->sectors_in_flight));
- closure_sync(&ctxt->cl);
+ int ret;
+
+ ret = bch2_move_btree(c,
+ 0, POS_MIN,
+ BTREE_ID_NR, SPOS_MAX,
+ rewrite_old_nodes_pred, c, stats);
+ if (!ret) {
+ mutex_lock(&c->sb_lock);
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
+ c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
+ c->disk_sb.sb->version_min = c->disk_sb.sb->version;
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+ }
- EBUG_ON(!list_empty(&ctxt->reads));
- EBUG_ON(atomic_read(&ctxt->sectors_in_flight));
+ return ret;
}
-void bch2_move_ctxt_init(struct moving_context *ctxt,
- struct bch_ratelimit *rate,
- unsigned max_sectors_in_flight)
+int bch2_data_job(struct bch_fs *c,
+ struct bch_move_stats *stats,
+ struct bch_ioctl_data op)
{
- memset(ctxt, 0, sizeof(*ctxt));
- closure_init_stack(&ctxt->cl);
+ int ret = 0;
- ctxt->rate = rate;
- ctxt->max_sectors_in_flight = max_sectors_in_flight;
+ switch (op.op) {
+ case BCH_DATA_OP_REREPLICATE:
+ bch_move_stats_init(stats, "rereplicate");
+ stats->data_type = BCH_DATA_journal;
+ ret = bch2_journal_flush_device_pins(&c->journal, -1);
+
+ ret = bch2_move_btree(c,
+ op.start_btree, op.start_pos,
+ op.end_btree, op.end_pos,
+ rereplicate_btree_pred, c, stats) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
+
+ ret = bch2_move_data(c,
+ op.start_btree, op.start_pos,
+ op.end_btree, op.end_pos,
+ NULL,
+ stats,
+ writepoint_hashed((unsigned long) current),
+ true,
+ rereplicate_pred, c) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
+ break;
+ case BCH_DATA_OP_MIGRATE:
+ if (op.migrate.dev >= c->sb.nr_devices)
+ return -EINVAL;
+
+ bch_move_stats_init(stats, "migrate");
+ stats->data_type = BCH_DATA_journal;
+ ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
+
+ ret = bch2_move_btree(c,
+ op.start_btree, op.start_pos,
+ op.end_btree, op.end_pos,
+ migrate_btree_pred, &op, stats) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
+
+ ret = bch2_move_data(c,
+ op.start_btree, op.start_pos,
+ op.end_btree, op.end_pos,
+ NULL,
+ stats,
+ writepoint_hashed((unsigned long) current),
+ true,
+ migrate_pred, &op) ?: ret;
+ ret = bch2_replicas_gc2(c) ?: ret;
+ break;
+ case BCH_DATA_OP_REWRITE_OLD_NODES:
+ bch_move_stats_init(stats, "rewrite_old_nodes");
+ ret = bch2_scan_old_btree_nodes(c, stats);
+ break;
+ default:
+ ret = -EINVAL;
+ }
- INIT_LIST_HEAD(&ctxt->reads);
- init_waitqueue_head(&ctxt->wait);
+ return ret;
}
projects / bcachefs-tools-debian / blobdiff