+// 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"
#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)
-{
- struct bch_extent_ptr *ptr2;
- struct bch_dev *ca = c->devs[ptr.dev];
+#define SECTORS_IN_FLIGHT_PER_DEVICE 2048
- extent_for_each_ptr(e, ptr2)
- if (ptr2->dev == ptr.dev &&
- ptr2->gen == ptr.gen &&
- PTR_BUCKET_NR(ca, ptr2) ==
- PTR_BUCKET_NR(ca, &ptr))
- return ptr2;
+struct moving_io {
+ struct list_head list;
+ struct closure cl;
+ bool read_completed;
- return NULL;
-}
+ unsigned read_sectors;
+ unsigned write_sectors;
-static struct bch_extent_ptr *bch2_migrate_matching_ptr(struct migrate_write *m,
- struct bkey_s_extent e)
-{
- const struct bch_extent_ptr *ptr;
- struct bch_extent_ptr *ret;
-
- 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;
+ struct bch_read_bio rbio;
- return ret;
-}
+ struct migrate_write write;
+ /* Must be last since it is variable size */
+ struct bio_vec bi_inline_vecs[0];
+};
+
+struct moving_context {
+ /* Closure for waiting on all reads and writes to complete */
+ struct closure cl;
+
+ struct bch_move_stats *stats;
+
+ struct list_head reads;
+
+ /* in flight sectors: */
+ atomic_t read_sectors;
+ atomic_t write_sectors;
+
+ wait_queue_head_t wait;
+};
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_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_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;
-
- if (!k.k) {
- ret = bch2_btree_iter_unlock(&iter);
+ 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;
+ struct extent_ptr_decoded p;
+ bool did_work = false;
+ int nr;
+
+ ret = bkey_err(k);
+ if (ret)
break;
- }
- if (!bkey_extent_is_data(k.k))
+ if (bversion_cmp(k.k->version, new->k.version) ||
+ !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
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->data_cmd == DATA_REWRITE &&
+ !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
+ goto nomatch;
- if (m->move) {
- nr_new_dirty -= !ptr->cached;
- __bch2_extent_drop_ptr(e, ptr);
+ bkey_reassemble(&_insert.k, 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);
+ 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(bkey_i_to_s(insert),
+ m->data_opts.rewrite_dev);
+
+ extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
+ 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
+ * data to
+ */
+ continue;
}
- BUG_ON(nr_new_dirty < 0);
+ bch2_extent_ptr_decoded_append(insert, &p);
+ did_work = true;
+ }
- memcpy_u64s(extent_entry_last(e),
- insert.v,
- bkey_val_u64s(insert.k));
- e.k->u64s += bkey_val_u64s(insert.k);
+ if (!did_work)
+ goto nomatch;
- 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);
+ bch2_bkey_narrow_crcs(insert,
+ (struct bch_extent_crc_unpacked) { 0 });
+ 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);
+
+ /*
+ * 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_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 out;
- 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);
+ m->nr_ptrs_reserved += nr;
+ goto next;
}
- while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
+ 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);
+ if (!ret)
+ atomic_long_inc(&c->extent_migrate_done);
+ if (ret == -EINTR)
+ ret = 0;
+ if (ret)
+ break;
+next:
+ 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,
+ &m->ctxt->stats->sectors_raced);
+ atomic_long_inc(&c->extent_migrate_raced);
+ trace_move_race(&new->k);
+ bch2_btree_iter_next_slot(iter);
+ goto next;
}
out:
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_exit(&trans);
+ BUG_ON(ret == -EINTR);
return ret;
}
-void bch2_migrate_write_init(struct bch_fs *c,
- struct migrate_write *m,
- struct bch_devs_mask *devs,
- struct bkey_s_c k,
- const struct bch_extent_ptr *move_ptr,
- unsigned flags)
+void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
{
- bkey_reassemble(&m->key, k);
+ /* write bio must own pages: */
+ BUG_ON(!m->op.wbio.bio.bi_vcnt);
+
+ m->ptr = rbio->pick.ptr;
+ m->offset = rbio->pos.offset - rbio->pick.crc.offset;
+ m->op.devs_have = rbio->devs_have;
+ m->op.pos = rbio->pos;
+ m->op.version = rbio->version;
+ m->op.crc = rbio->pick.crc;
+ m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
+
+ if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
+ m->op.nonce = m->op.crc.nonce + m->op.crc.offset;
+ m->op.csum_type = m->op.crc.csum_type;
+ }
- m->promote = false;
- m->move = move_ptr != NULL;
- if (move_ptr)
- m->move_ptr = *move_ptr;
+ if (m->data_cmd == DATA_REWRITE)
+ bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
+}
- if (bkey_extent_is_cached(k.k) ||
- (move_ptr && move_ptr->cached))
- flags |= BCH_WRITE_CACHED;
+int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
+ struct write_point_specifier wp,
+ 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;
- bch2_write_op_init(&m->op, c, (struct disk_reservation) { 0 },
- devs, (unsigned long) current,
- bkey_start_pos(k.k), NULL,
- flags|BCH_WRITE_ONLY_SPECIFIED_DEVS);
+ bch2_write_op_init(&m->op, c, io_opts);
+ m->op.compression_type =
+ bch2_compression_opt_to_type[io_opts.background_compression ?:
+ io_opts.compression];
+ m->op.target = data_opts.target,
+ m->op.write_point = wp;
- if (m->move)
+ if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
m->op.alloc_reserve = RESERVE_MOVINGGC;
- m->op.nonce = extent_current_nonce(bkey_s_c_to_extent(k));
+ m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
+ BCH_WRITE_PAGES_STABLE|
+ BCH_WRITE_PAGES_OWNED|
+ BCH_WRITE_DATA_ENCODED;
+
m->op.nr_replicas = 1;
+ m->op.nr_replicas_required = 1;
m->op.index_update_fn = bch2_migrate_index_update;
-}
-static void migrate_bio_init(struct moving_io *io, struct bio *bio,
- unsigned sectors)
-{
- bio_init(bio, io->bi_inline_vecs,
- DIV_ROUND_UP(sectors, PAGE_SECTORS));
- bio_set_prio(bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 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_ptrs_allocated(k);
+#endif
+ int nr = (int) io_opts.data_replicas;
+
+ if (nr > 0) {
+ m->op.nr_replicas = m->nr_ptrs_reserved = nr;
+
+ ret = bch2_disk_reservation_get(c, &m->op.res,
+ k.k->size, m->op.nr_replicas, 0);
+ if (ret)
+ return ret;
+ }
+ 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;
+
+ 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))
+ compressed_sectors += p.crc.compressed_size;
+
+ if (compressed_sectors) {
+ ret = bch2_disk_reservation_add(c, &m->op.res,
+ compressed_sectors,
+ BCH_DISK_RESERVATION_NOFAIL);
+ if (ret)
+ return ret;
+ }
+ break;
+ }
+ case DATA_PROMOTE:
+ m->op.flags |= BCH_WRITE_ALLOC_NOWAIT;
+ m->op.flags |= BCH_WRITE_CACHED;
+ break;
+ default:
+ BUG();
+ }
- bio->bi_iter.bi_size = sectors << 9;
- bio->bi_private = &io->cl;
- bch2_bio_map(bio, NULL);
+ return 0;
}
-static void moving_io_free(struct moving_io *io)
+static void move_free(struct closure *cl)
{
- struct moving_context *ctxt = io->ctxt;
+ 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;
- atomic_sub(io->write.key.k.size, &ctxt->sectors_in_flight);
- wake_up(&ctxt->wait);
+ 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);
- kfree(io);
-}
-static void moving_error(struct moving_context *ctxt, unsigned flag)
-{
- atomic_inc(&ctxt->error_count);
- //atomic_or(flag, &ctxt->error_flags);
+ wake_up(&ctxt->wait);
+
+ kfree(io);
}
-static void moving_write_done(struct closure *cl)
+static void move_write_done(struct closure *cl)
{
struct moving_io *io = container_of(cl, struct moving_io, cl);
- if (io->write.op.error)
- moving_error(io->ctxt, MOVING_FLAG_WRITE);
-
- //if (io->replace.failures)
- // trace_copy_collision(q, &io->key.k);
-
- moving_io_free(io);
+ atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
+ closure_return_with_destructor(cl, move_free);
}
-static void write_moving(struct closure *cl)
+static void move_write(struct closure *cl)
{
struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct bch_write_op *op = &io->write.op;
- closure_call(&op->cl, bch2_write, NULL, &io->cl);
- closure_return_with_destructor(&io->cl, moving_write_done);
+ if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
+ closure_return_with_destructor(cl, move_free);
+ return;
+ }
+
+ bch2_migrate_read_done(&io->write, &io->rbio);
+
+ atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
+ closure_call(&io->write.op.cl, bch2_write, NULL, cl);
+ continue_at(cl, move_write_done, NULL);
}
static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
return io && io->read_completed ? io : NULL;
}
-static void read_moving_endio(struct bio *bio)
+static void move_read_endio(struct bio *bio)
{
- struct closure *cl = bio->bi_private;
- struct moving_io *io = container_of(cl, struct moving_io, cl);
- struct moving_context *ctxt = io->ctxt;
-
- trace_move_read_done(&io->write.key.k);
-
- if (bio->bi_error)
- moving_error(io->ctxt, MOVING_FLAG_READ);
+ struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
+ struct moving_context *ctxt = io->write.ctxt;
+ atomic_sub(io->read_sectors, &ctxt->read_sectors);
io->read_completed = true;
+
if (next_pending_write(ctxt))
wake_up(&ctxt->wait);
closure_put(&ctxt->cl);
}
-int bch2_data_move(struct bch_fs *c,
- struct moving_context *ctxt,
- struct bch_devs_mask *devs,
- struct bkey_s_c k,
- const struct bch_extent_ptr *move_ptr)
+static void do_pending_writes(struct moving_context *ctxt)
{
- struct extent_pick_ptr pick;
struct moving_io *io;
- bch2_extent_pick_ptr(c, k, &ctxt->avoid, &pick);
- if (IS_ERR_OR_NULL(pick.ca))
- return pick.ca ? PTR_ERR(pick.ca) : 0;
+ while ((io = next_pending_write(ctxt))) {
+ list_del(&io->list);
+ closure_call(&io->cl, move_write, NULL, &ctxt->cl);
+ }
+}
- io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec) *
- DIV_ROUND_UP(k.k->size, PAGE_SECTORS), GFP_KERNEL);
+#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 void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
+{
+ unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
+
+ move_ctxt_wait_event(ctxt,
+ !atomic_read(&ctxt->write_sectors) ||
+ atomic_read(&ctxt->write_sectors) != sectors_pending);
+}
+
+static int bch2_move_extent(struct bch_fs *c,
+ struct moving_context *ctxt,
+ struct write_point_specifier wp,
+ struct bch_io_opts io_opts,
+ 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 = k.k->size, pages;
+ int ret = -ENOMEM;
+
+ move_ctxt_wait_event(ctxt,
+ atomic_read(&ctxt->write_sectors) <
+ SECTORS_IN_FLIGHT_PER_DEVICE);
+
+ move_ctxt_wait_event(ctxt,
+ atomic_read(&ctxt->read_sectors) <
+ SECTORS_IN_FLIGHT_PER_DEVICE);
+
+ /* 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)
- return -ENOMEM;
+ goto err;
- io->ctxt = ctxt;
+ io->write.ctxt = ctxt;
+ io->read_sectors = k.k->size;
+ io->write_sectors = k.k->size;
- migrate_bio_init(io, &io->rbio.bio, 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));
- 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 = read_moving_endio;
+ if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
+ GFP_KERNEL))
+ goto err_free;
- if (bio_alloc_pages(&io->rbio.bio, GFP_KERNEL)) {
- kfree(io);
- return -ENOMEM;
- }
+ 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;
- migrate_bio_init(io, &io->write.op.wbio.bio, k.k->size);
+ 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;
- bch2_migrate_write_init(c, &io->write, devs, k, move_ptr, 0);
+ ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
+ data_cmd, data_opts, btree_id, k);
+ if (ret)
+ goto err_free_pages;
- trace_move_read(&io->write.key.k);
+ atomic64_inc(&ctxt->stats->keys_moved);
+ atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
- ctxt->keys_moved++;
- ctxt->sectors_moved += k.k->size;
- if (ctxt->rate)
- bch2_ratelimit_increment(ctxt->rate, k.k->size);
+ trace_move_extent(k.k);
- atomic_add(k.k->size, &ctxt->sectors_in_flight);
+ atomic_add(io->read_sectors, &ctxt->read_sectors);
list_add_tail(&io->list, &ctxt->reads);
/*
- * dropped by read_moving_endio() - guards against use after free of
+ * dropped by move_read_endio() - guards against use after free of
* ctxt when doing wakeup
*/
- closure_get(&io->ctxt->cl);
- bch2_read_extent(c, &io->rbio, k, &pick, 0);
+ closure_get(&ctxt->cl);
+ bch2_read_extent(c, &io->rbio, 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:
+ trace_move_alloc_fail(k.k);
+ return ret;
}
-static void do_pending_writes(struct moving_context *ctxt)
+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)
{
- struct moving_io *io;
+ bool kthread = (current->flags & PF_KTHREAD) != 0;
+ struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+ 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;
+
+ bkey_on_stack_init(&sk);
+ bch2_trans_init(&trans, c, 0, 0);
+
+ stats->data_type = BCH_DATA_USER;
+ 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);
- while ((io = next_pending_write(ctxt))) {
- list_del(&io->list);
+ while (1) {
+ do {
+ delay = rate ? bch2_ratelimit_delay(rate) : 0;
+
+ if (delay) {
+ bch2_trans_unlock(&trans);
+ set_current_state(TASK_INTERRUPTIBLE);
+ }
+
+ if (kthread && (ret = kthread_should_stop())) {
+ __set_current_state(TASK_RUNNING);
+ goto out;
+ }
+
+ if (delay)
+ schedule_timeout(delay);
- if (io->rbio.bio.bi_error) {
- moving_io_free(io);
- continue;
+ if (unlikely(freezing(current))) {
+ bch2_trans_unlock(&trans);
+ move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
+ try_to_freeze();
+ }
+ } while (delay);
+peek:
+ k = bch2_btree_iter_peek(iter);
+
+ stats->pos = iter->pos;
+
+ if (!k.k)
+ break;
+ ret = bkey_err(k);
+ if (ret)
+ break;
+ if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
+ break;
+
+ if (!bkey_extent_is_direct_data(k.k))
+ goto next_nondata;
+
+ 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_trans_unlock(&trans);
+
+ io_opts = bch2_opts_to_inode_opts(c->opts);
+ if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
+ bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
+ cur_inum = k.k->p.inode;
+ goto peek;
+ }
+
+ switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
+ case DATA_SKIP:
+ goto next;
+ case DATA_SCRUB:
+ BUG();
+ case DATA_ADD_REPLICAS:
+ case DATA_REWRITE:
+ case DATA_PROMOTE:
+ break;
+ default:
+ BUG();
+ }
+
+ /* unlock before doing IO: */
+ 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, 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);
+ continue;
+ }
+
+ /* XXX signal failure */
+ goto next;
}
- trace_move_write(&io->write.key.k);
- closure_call(&io->cl, write_moving, NULL, &ctxt->cl);
+ if (rate)
+ bch2_ratelimit_increment(rate, k.k->size);
+next:
+ atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
+ &stats->sectors_seen);
+next_nondata:
+ bch2_btree_iter_next(iter);
+ bch2_trans_cond_resched(&trans);
}
+out:
+ ret = bch2_trans_exit(&trans) ?: ret;
+ bkey_on_stack_exit(&sk, c);
+
+ return ret;
}
-#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)
+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 moving_context ctxt = { .stats = stats };
+ int ret;
+
+ closure_init_stack(&ctxt.cl);
+ INIT_LIST_HEAD(&ctxt.reads);
+ init_waitqueue_head(&ctxt.wait);
+
+ stats->data_type = BCH_DATA_USER;
+
+ 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);
-int bch2_move_ctxt_wait(struct moving_context *ctxt)
+ 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));
+
+ return ret;
+}
+
+static int bch2_move_btree(struct bch_fs *c,
+ move_pred_fn pred,
+ void *arg,
+ struct bch_move_stats *stats)
{
- move_ctxt_wait_event(ctxt,
- atomic_read(&ctxt->sectors_in_flight) <
- ctxt->max_sectors_in_flight);
+ 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++) {
+ 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))) {
+ case DATA_SKIP:
+ goto next;
+ case DATA_SCRUB:
+ BUG();
+ case DATA_ADD_REPLICAS:
+ case DATA_REWRITE:
+ break;
+ default:
+ BUG();
+ }
- return ctxt->rate
- ? bch2_ratelimit_wait_freezable_stoppable(ctxt->rate)
- : 0;
+ ret = bch2_btree_node_rewrite(c, iter,
+ b->data->keys.seq, 0) ?: ret;
+next:
+ bch2_trans_cond_resched(&trans);
+ }
+
+ ret = bch2_trans_iter_free(&trans, iter) ?: ret;
+ }
+
+ bch2_trans_exit(&trans);
+
+ return ret;
}
-void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
+#if 0
+static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts,
+ struct data_opts *data_opts)
{
- unsigned sectors_pending = atomic_read(&ctxt->sectors_in_flight);
+ return DATA_SCRUB;
+}
+#endif
- move_ctxt_wait_event(ctxt,
- !atomic_read(&ctxt->sectors_in_flight) ||
- atomic_read(&ctxt->sectors_in_flight) != sectors_pending);
+static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts,
+ struct data_opts *data_opts)
+{
+ unsigned nr_good = bch2_bkey_durability(c, k);
+ unsigned replicas = 0;
+
+ switch (k.k->type) {
+ case KEY_TYPE_btree_ptr:
+ replicas = c->opts.metadata_replicas;
+ break;
+ case KEY_TYPE_extent:
+ replicas = io_opts->data_replicas;
+ break;
+ }
+
+ if (!nr_good || nr_good >= replicas)
+ return DATA_SKIP;
+
+ data_opts->target = 0;
+ data_opts->btree_insert_flags = 0;
+ return DATA_ADD_REPLICAS;
}
-void bch2_move_ctxt_exit(struct moving_context *ctxt)
+static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
+ struct bkey_s_c k,
+ struct bch_io_opts *io_opts,
+ struct data_opts *data_opts)
{
- move_ctxt_wait_event(ctxt, !atomic_read(&ctxt->sectors_in_flight));
- closure_sync(&ctxt->cl);
+ struct bch_ioctl_data *op = arg;
+
+ if (!bch2_bkey_has_device(k, op->migrate.dev))
+ return DATA_SKIP;
- EBUG_ON(!list_empty(&ctxt->reads));
- EBUG_ON(atomic_read(&ctxt->sectors_in_flight));
+ data_opts->target = 0;
+ data_opts->btree_insert_flags = 0;
+ data_opts->rewrite_dev = op->migrate.dev;
+ return DATA_REWRITE;
}
-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;
+
+ switch (op.op) {
+ case BCH_DATA_OP_REREPLICATE:
+ stats->data_type = BCH_DATA_JOURNAL;
+ ret = bch2_journal_flush_device_pins(&c->journal, -1);
+
+ ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
- ctxt->rate = rate;
- ctxt->max_sectors_in_flight = max_sectors_in_flight;
+ 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);
+ }
- INIT_LIST_HEAD(&ctxt->reads);
- init_waitqueue_head(&ctxt->wait);
+ 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_replicas_gc2(c) ?: ret;
+ break;
+ case BCH_DATA_OP_MIGRATE:
+ if (op.migrate.dev >= c->sb.nr_devices)
+ return -EINVAL;
+
+ stats->data_type = BCH_DATA_JOURNAL;
+ ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
+
+ ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: 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_replicas_gc2(c) ?: ret;
+ break;
+ default:
+ ret = -EINVAL;
+ }
+
+ return ret;
}