+// SPDX-License-Identifier: GPL-2.0
/*
* Moving/copying garbage collector
*
*/
#include "bcachefs.h"
+#include "alloc_background.h"
+#include "alloc_foreground.h"
#include "btree_iter.h"
#include "btree_update.h"
+#include "btree_write_buffer.h"
#include "buckets.h"
#include "clock.h"
-#include "disk_groups.h"
-#include "extents.h"
-#include "eytzinger.h"
-#include "io.h"
-#include "keylist.h"
+#include "errcode.h"
+#include "error.h"
+#include "lru.h"
#include "move.h"
#include "movinggc.h"
-#include "super-io.h"
+#include "trace.h"
-#include <trace/events/bcachefs.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include <linux/math64.h>
#include <linux/sched/task.h>
-#include <linux/sort.h>
#include <linux/wait.h>
-/*
- * We can't use the entire copygc reserve in one iteration of copygc: we may
- * need the buckets we're freeing up to go back into the copygc reserve to make
- * forward progress, but if the copygc reserve is full they'll be available for
- * any allocation - and it's possible that in a given iteration, we free up most
- * of the buckets we're going to free before we allocate most of the buckets
- * we're going to allocate.
- *
- * If we only use half of the reserve per iteration, then in steady state we'll
- * always have room in the reserve for the buckets we're going to need in the
- * next iteration:
- */
-#define COPYGC_BUCKETS_PER_ITER(ca) \
- ((ca)->free[RESERVE_MOVINGGC].size / 2)
+struct buckets_in_flight {
+ struct rhashtable table;
+ struct move_bucket_in_flight *first;
+ struct move_bucket_in_flight *last;
+ size_t nr;
+ size_t sectors;
+};
+
+static const struct rhashtable_params bch_move_bucket_params = {
+ .head_offset = offsetof(struct move_bucket_in_flight, hash),
+ .key_offset = offsetof(struct move_bucket_in_flight, bucket.k),
+ .key_len = sizeof(struct move_bucket_key),
+};
+
+static struct move_bucket_in_flight *
+move_bucket_in_flight_add(struct buckets_in_flight *list, struct move_bucket b)
+{
+ struct move_bucket_in_flight *new = kzalloc(sizeof(*new), GFP_KERNEL);
+ int ret;
-/*
- * Max sectors to move per iteration: Have to take into account internal
- * fragmentation from the multiple write points for each generation:
- */
-#define COPYGC_SECTORS_PER_ITER(ca) \
- ((ca)->mi.bucket_size * COPYGC_BUCKETS_PER_ITER(ca))
+ if (!new)
+ return ERR_PTR(-ENOMEM);
-static inline int sectors_used_cmp(copygc_heap *heap,
- struct copygc_heap_entry l,
- struct copygc_heap_entry r)
-{
- return (l.sectors > r.sectors) - (l.sectors < r.sectors);
+ new->bucket = b;
+
+ ret = rhashtable_lookup_insert_fast(&list->table, &new->hash,
+ bch_move_bucket_params);
+ if (ret) {
+ kfree(new);
+ return ERR_PTR(ret);
+ }
+
+ if (!list->first)
+ list->first = new;
+ else
+ list->last->next = new;
+
+ list->last = new;
+ list->nr++;
+ list->sectors += b.sectors;
+ return new;
}
-static int bucket_offset_cmp(const void *_l, const void *_r, size_t size)
+static int bch2_bucket_is_movable(struct btree_trans *trans,
+ struct move_bucket *b, u64 time)
{
- const struct copygc_heap_entry *l = _l;
- const struct copygc_heap_entry *r = _r;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct bch_alloc_v4 _a;
+ const struct bch_alloc_v4 *a;
+ int ret;
+
+ if (bch2_bucket_is_open(trans->c,
+ b->k.bucket.inode,
+ b->k.bucket.offset))
+ return 0;
+
+ k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
+ b->k.bucket, BTREE_ITER_CACHED);
+ ret = bkey_err(k);
+ if (ret)
+ return ret;
+
+ a = bch2_alloc_to_v4(k, &_a);
+ b->k.gen = a->gen;
+ b->sectors = bch2_bucket_sectors_dirty(*a);
+
+ ret = data_type_movable(a->data_type) &&
+ a->fragmentation_lru &&
+ a->fragmentation_lru <= time;
- return (l->offset > r->offset) - (l->offset < r->offset);
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
}
-static bool __copygc_pred(struct bch_dev *ca,
- struct bkey_s_c_extent e)
+static void move_buckets_wait(struct moving_context *ctxt,
+ struct buckets_in_flight *list,
+ bool flush)
{
- copygc_heap *h = &ca->copygc_heap;
- const struct bch_extent_ptr *ptr =
- bch2_extent_has_device(e, ca->dev_idx);
+ struct move_bucket_in_flight *i;
+ int ret;
+
+ while ((i = list->first)) {
+ if (flush)
+ move_ctxt_wait_event(ctxt, !atomic_read(&i->count));
+
+ if (atomic_read(&i->count))
+ break;
- if (ptr) {
- struct copygc_heap_entry search = { .offset = ptr->offset };
+ list->first = i->next;
+ if (!list->first)
+ list->last = NULL;
- ssize_t i = eytzinger0_find_le(h->data, h->used,
- sizeof(h->data[0]),
- bucket_offset_cmp, &search);
+ list->nr--;
+ list->sectors -= i->bucket.sectors;
- return (i >= 0 &&
- ptr->offset < h->data[i].offset + ca->mi.bucket_size &&
- ptr->gen == h->data[i].gen);
+ ret = rhashtable_remove_fast(&list->table, &i->hash,
+ bch_move_bucket_params);
+ BUG_ON(ret);
+ kfree(i);
}
- return false;
+ bch2_trans_unlock_long(ctxt->trans);
}
-static enum data_cmd copygc_pred(struct bch_fs *c, void *arg,
- enum bkey_type type,
- struct bkey_s_c_extent e,
- struct bch_io_opts *io_opts,
- struct data_opts *data_opts)
+static bool bucket_in_flight(struct buckets_in_flight *list,
+ struct move_bucket_key k)
{
- struct bch_dev *ca = arg;
-
- if (!__copygc_pred(ca, e))
- return DATA_SKIP;
-
- data_opts->target = dev_to_target(ca->dev_idx);
- data_opts->btree_insert_flags = BTREE_INSERT_USE_RESERVE;
- data_opts->rewrite_dev = ca->dev_idx;
- return DATA_REWRITE;
+ return rhashtable_lookup_fast(&list->table, &k, bch_move_bucket_params);
}
-static bool have_copygc_reserve(struct bch_dev *ca)
+typedef DARRAY(struct move_bucket) move_buckets;
+
+static int bch2_copygc_get_buckets(struct moving_context *ctxt,
+ struct buckets_in_flight *buckets_in_flight,
+ move_buckets *buckets)
{
- bool ret;
+ struct btree_trans *trans = ctxt->trans;
+ struct bch_fs *c = trans->c;
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ size_t nr_to_get = max_t(size_t, 16U, buckets_in_flight->nr / 4);
+ size_t saw = 0, in_flight = 0, not_movable = 0, sectors = 0;
+ int ret;
- spin_lock(&ca->freelist_lock);
- ret = fifo_full(&ca->free[RESERVE_MOVINGGC]) ||
- ca->allocator_blocked;
- spin_unlock(&ca->freelist_lock);
+ move_buckets_wait(ctxt, buckets_in_flight, false);
+
+ ret = bch2_btree_write_buffer_tryflush(trans);
+ if (bch2_err_matches(ret, EROFS))
+ return ret;
+
+ if (bch2_fs_fatal_err_on(ret, c, "%s: error %s from bch2_btree_write_buffer_tryflush()",
+ __func__, bch2_err_str(ret)))
+ return ret;
+
+ ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
+ lru_pos(BCH_LRU_FRAGMENTATION_START, 0, 0),
+ lru_pos(BCH_LRU_FRAGMENTATION_START, U64_MAX, LRU_TIME_MAX),
+ 0, k, ({
+ struct move_bucket b = { .k.bucket = u64_to_bucket(k.k->p.offset) };
+ int ret2 = 0;
+
+ saw++;
+
+ ret2 = bch2_bucket_is_movable(trans, &b, lru_pos_time(k.k->p));
+ if (ret2 < 0)
+ goto err;
+
+ if (!ret2)
+ not_movable++;
+ else if (bucket_in_flight(buckets_in_flight, b.k))
+ in_flight++;
+ else {
+ ret2 = darray_push(buckets, b);
+ if (ret2)
+ goto err;
+ sectors += b.sectors;
+ }
- return ret;
+ ret2 = buckets->nr >= nr_to_get;
+err:
+ ret2;
+ }));
+
+ pr_debug("have: %zu (%zu) saw %zu in flight %zu not movable %zu got %zu (%zu)/%zu buckets ret %i",
+ buckets_in_flight->nr, buckets_in_flight->sectors,
+ saw, in_flight, not_movable, buckets->nr, sectors, nr_to_get, ret);
+
+ return ret < 0 ? ret : 0;
}
-static void bch2_copygc(struct bch_fs *c, struct bch_dev *ca)
+noinline
+static int bch2_copygc(struct moving_context *ctxt,
+ struct buckets_in_flight *buckets_in_flight,
+ bool *did_work)
{
- copygc_heap *h = &ca->copygc_heap;
- struct copygc_heap_entry e, *i;
- struct bucket_array *buckets;
- struct bch_move_stats move_stats;
- u64 sectors_to_move = 0, sectors_not_moved = 0;
- u64 buckets_to_move, buckets_not_moved = 0;
- size_t b;
- int ret;
+ struct btree_trans *trans = ctxt->trans;
+ struct bch_fs *c = trans->c;
+ struct data_update_opts data_opts = {
+ .btree_insert_flags = BCH_WATERMARK_copygc,
+ };
+ move_buckets buckets = { 0 };
+ struct move_bucket_in_flight *f;
+ struct move_bucket *i;
+ u64 moved = atomic64_read(&ctxt->stats->sectors_moved);
+ int ret = 0;
+
+ ret = bch2_copygc_get_buckets(ctxt, buckets_in_flight, &buckets);
+ if (ret)
+ goto err;
+
+ darray_for_each(buckets, i) {
+ if (kthread_should_stop() || freezing(current))
+ break;
- memset(&move_stats, 0, sizeof(move_stats));
- closure_wait_event(&c->freelist_wait, have_copygc_reserve(ca));
-
- /*
- * Find buckets with lowest sector counts, skipping completely
- * empty buckets, by building a maxheap sorted by sector count,
- * and repeatedly replacing the maximum element until all
- * buckets have been visited.
- */
- h->used = 0;
-
- /*
- * We need bucket marks to be up to date - gc can't be recalculating
- * them:
- */
- down_read(&c->gc_lock);
- down_read(&ca->bucket_lock);
- buckets = bucket_array(ca);
-
- for (b = buckets->first_bucket; b < buckets->nbuckets; b++) {
- struct bucket_mark m = READ_ONCE(buckets->b[b].mark);
- struct copygc_heap_entry e;
-
- if (m.owned_by_allocator ||
- m.data_type != BCH_DATA_USER ||
- !bucket_sectors_used(m) ||
- bucket_sectors_used(m) >= ca->mi.bucket_size)
+ f = move_bucket_in_flight_add(buckets_in_flight, *i);
+ ret = PTR_ERR_OR_ZERO(f);
+ if (ret == -EEXIST) { /* rare race: copygc_get_buckets returned same bucket more than once */
+ ret = 0;
continue;
+ }
+ if (ret == -ENOMEM) { /* flush IO, continue later */
+ ret = 0;
+ break;
+ }
+
+ ret = bch2_evacuate_bucket(ctxt, f, f->bucket.k.bucket,
+ f->bucket.k.gen, data_opts);
+ if (ret)
+ goto err;
- e = (struct copygc_heap_entry) {
- .gen = m.gen,
- .sectors = bucket_sectors_used(m),
- .offset = bucket_to_sector(ca, b),
- };
- heap_add_or_replace(h, e, -sectors_used_cmp);
+ *did_work = true;
}
- up_read(&ca->bucket_lock);
- up_read(&c->gc_lock);
+err:
+ darray_exit(&buckets);
- for (i = h->data; i < h->data + h->used; i++)
- sectors_to_move += i->sectors;
+ /* no entries in LRU btree found, or got to end: */
+ if (bch2_err_matches(ret, ENOENT))
+ ret = 0;
- while (sectors_to_move > COPYGC_SECTORS_PER_ITER(ca)) {
- BUG_ON(!heap_pop(h, e, -sectors_used_cmp));
- sectors_to_move -= e.sectors;
- }
+ if (ret < 0 && !bch2_err_matches(ret, EROFS))
+ bch_err_msg(c, ret, "from bch2_move_data()");
- buckets_to_move = h->used;
+ moved = atomic64_read(&ctxt->stats->sectors_moved) - moved;
+ trace_and_count(c, copygc, c, moved, 0, 0, 0);
+ return ret;
+}
- if (!buckets_to_move)
- return;
+/*
+ * Copygc runs when the amount of fragmented data is above some arbitrary
+ * threshold:
+ *
+ * The threshold at the limit - when the device is full - is the amount of space
+ * we reserved in bch2_recalc_capacity; we can't have more than that amount of
+ * disk space stranded due to fragmentation and store everything we have
+ * promised to store.
+ *
+ * But we don't want to be running copygc unnecessarily when the device still
+ * has plenty of free space - rather, we want copygc to smoothly run every so
+ * often and continually reduce the amount of fragmented space as the device
+ * fills up. So, we increase the threshold by half the current free space.
+ */
+unsigned long bch2_copygc_wait_amount(struct bch_fs *c)
+{
+ struct bch_dev *ca;
+ unsigned dev_idx;
+ s64 wait = S64_MAX, fragmented_allowed, fragmented;
+ unsigned i;
- eytzinger0_sort(h->data, h->used,
- sizeof(h->data[0]),
- bucket_offset_cmp, NULL);
+ for_each_rw_member(ca, c, dev_idx) {
+ struct bch_dev_usage usage = bch2_dev_usage_read(ca);
- ret = bch2_move_data(c, &ca->copygc_pd.rate,
- writepoint_ptr(&ca->copygc_write_point),
- POS_MIN, POS_MAX,
- copygc_pred, ca,
- &move_stats);
+ fragmented_allowed = ((__dev_buckets_available(ca, usage, BCH_WATERMARK_stripe) *
+ ca->mi.bucket_size) >> 1);
+ fragmented = 0;
- down_read(&ca->bucket_lock);
- buckets = bucket_array(ca);
- for (i = h->data; i < h->data + h->used; i++) {
- size_t b = sector_to_bucket(ca, i->offset);
- struct bucket_mark m = READ_ONCE(buckets->b[b].mark);
+ for (i = 0; i < BCH_DATA_NR; i++)
+ if (data_type_movable(i))
+ fragmented += usage.d[i].fragmented;
- if (i->gen == m.gen && bucket_sectors_used(m)) {
- sectors_not_moved += bucket_sectors_used(m);
- buckets_not_moved++;
- }
+ wait = min(wait, max(0LL, fragmented_allowed - fragmented));
}
- up_read(&ca->bucket_lock);
- if (sectors_not_moved && !ret)
- bch_warn(c, "copygc finished but %llu/%llu sectors, %llu/%llu buckets not moved",
- sectors_not_moved, sectors_to_move,
- buckets_not_moved, buckets_to_move);
+ return wait;
+}
- trace_copygc(ca,
- atomic64_read(&move_stats.sectors_moved), sectors_not_moved,
- buckets_to_move, buckets_not_moved);
+void bch2_copygc_wait_to_text(struct printbuf *out, struct bch_fs *c)
+{
+ prt_printf(out, "Currently waiting for: ");
+ prt_human_readable_u64(out, max(0LL, c->copygc_wait -
+ atomic64_read(&c->io_clock[WRITE].now)) << 9);
+ prt_newline(out);
+
+ prt_printf(out, "Currently waiting since: ");
+ prt_human_readable_u64(out, max(0LL,
+ atomic64_read(&c->io_clock[WRITE].now) -
+ c->copygc_wait_at) << 9);
+ prt_newline(out);
+
+ prt_printf(out, "Currently calculated wait: ");
+ prt_human_readable_u64(out, bch2_copygc_wait_amount(c));
+ prt_newline(out);
}
static int bch2_copygc_thread(void *arg)
{
- struct bch_dev *ca = arg;
- struct bch_fs *c = ca->fs;
+ struct bch_fs *c = arg;
+ struct moving_context ctxt;
+ struct bch_move_stats move_stats;
struct io_clock *clock = &c->io_clock[WRITE];
- struct bch_dev_usage usage;
- unsigned long last;
- u64 available, fragmented, reserve, next;
+ struct buckets_in_flight *buckets;
+ u64 last, wait;
+ int ret = 0;
+
+ buckets = kzalloc(sizeof(struct buckets_in_flight), GFP_KERNEL);
+ if (!buckets)
+ return -ENOMEM;
+ ret = rhashtable_init(&buckets->table, &bch_move_bucket_params);
+ if (ret) {
+ kfree(buckets);
+ bch_err_msg(c, ret, "allocating copygc buckets in flight");
+ return ret;
+ }
set_freezable();
- while (!kthread_should_stop()) {
- if (kthread_wait_freezable(c->copy_gc_enabled))
- break;
+ bch2_move_stats_init(&move_stats, "copygc");
+ bch2_moving_ctxt_init(&ctxt, c, NULL, &move_stats,
+ writepoint_ptr(&c->copygc_write_point),
+ false);
- last = atomic_long_read(&clock->now);
+ while (!ret && !kthread_should_stop()) {
+ bool did_work = false;
- reserve = ca->copygc_threshold;
+ bch2_trans_unlock_long(ctxt.trans);
+ cond_resched();
- usage = bch2_dev_usage_read(c, ca);
+ if (!c->copy_gc_enabled) {
+ move_buckets_wait(&ctxt, buckets, true);
+ kthread_wait_freezable(c->copy_gc_enabled ||
+ kthread_should_stop());
+ }
- available = __dev_buckets_available(ca, usage) *
- ca->mi.bucket_size;
- if (available > reserve) {
- next = last + available - reserve;
- bch2_kthread_io_clock_wait(clock, next,
- MAX_SCHEDULE_TIMEOUT);
+ if (unlikely(freezing(current))) {
+ move_buckets_wait(&ctxt, buckets, true);
+ __refrigerator(false);
continue;
}
- /*
- * don't start copygc until there's more than half the copygc
- * reserve of fragmented space:
- */
- fragmented = usage.sectors_fragmented;
- if (fragmented < reserve) {
- next = last + reserve - fragmented;
- bch2_kthread_io_clock_wait(clock, next,
+ last = atomic64_read(&clock->now);
+ wait = bch2_copygc_wait_amount(c);
+
+ if (wait > clock->max_slop) {
+ c->copygc_wait_at = last;
+ c->copygc_wait = last + wait;
+ move_buckets_wait(&ctxt, buckets, true);
+ trace_and_count(c, copygc_wait, c, wait, last + wait);
+ bch2_kthread_io_clock_wait(clock, last + wait,
MAX_SCHEDULE_TIMEOUT);
continue;
}
- bch2_copygc(c, ca);
+ c->copygc_wait = 0;
+
+ c->copygc_running = true;
+ ret = bch2_copygc(&ctxt, buckets, &did_work);
+ c->copygc_running = false;
+
+ wake_up(&c->copygc_running_wq);
+
+ if (!wait && !did_work) {
+ u64 min_member_capacity = bch2_min_rw_member_capacity(c);
+
+ if (min_member_capacity == U64_MAX)
+ min_member_capacity = 128 * 2048;
+
+ bch2_trans_unlock_long(ctxt.trans);
+ bch2_kthread_io_clock_wait(clock, last + (min_member_capacity >> 6),
+ MAX_SCHEDULE_TIMEOUT);
+ }
}
+ move_buckets_wait(&ctxt, buckets, true);
+
+ rhashtable_destroy(&buckets->table);
+ kfree(buckets);
+ bch2_moving_ctxt_exit(&ctxt);
+ bch2_move_stats_exit(&move_stats, c);
+
return 0;
}
-void bch2_copygc_stop(struct bch_dev *ca)
+void bch2_copygc_stop(struct bch_fs *c)
{
- ca->copygc_pd.rate.rate = UINT_MAX;
- bch2_ratelimit_reset(&ca->copygc_pd.rate);
-
- if (ca->copygc_thread) {
- kthread_stop(ca->copygc_thread);
- put_task_struct(ca->copygc_thread);
+ if (c->copygc_thread) {
+ kthread_stop(c->copygc_thread);
+ put_task_struct(c->copygc_thread);
}
- ca->copygc_thread = NULL;
+ c->copygc_thread = NULL;
}
-int bch2_copygc_start(struct bch_fs *c, struct bch_dev *ca)
+int bch2_copygc_start(struct bch_fs *c)
{
struct task_struct *t;
+ int ret;
- BUG_ON(ca->copygc_thread);
+ if (c->copygc_thread)
+ return 0;
if (c->opts.nochanges)
return 0;
if (bch2_fs_init_fault("copygc_start"))
return -ENOMEM;
- t = kthread_create(bch2_copygc_thread, ca,
- "bch_copygc[%s]", ca->name);
- if (IS_ERR(t))
- return PTR_ERR(t);
+ t = kthread_create(bch2_copygc_thread, c, "bch-copygc/%s", c->name);
+ ret = PTR_ERR_OR_ZERO(t);
+ if (ret) {
+ bch_err_msg(c, ret, "creating copygc thread");
+ return ret;
+ }
get_task_struct(t);
- ca->copygc_thread = t;
- wake_up_process(ca->copygc_thread);
+ c->copygc_thread = t;
+ wake_up_process(c->copygc_thread);
return 0;
}
-void bch2_dev_copygc_init(struct bch_dev *ca)
+void bch2_fs_copygc_init(struct bch_fs *c)
{
- bch2_pd_controller_init(&ca->copygc_pd);
- ca->copygc_pd.d_term = 0;
+ init_waitqueue_head(&c->copygc_running_wq);
+ c->copygc_running = false;
}
projects / bcachefs-tools-debian / blobdiff