#include "journal.h"
#include "keylist.h"
#include "move.h"
+#include "replicas.h"
#include "super-io.h"
#include <linux/slab.h>
#include <linux/bitops.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
+#include <linux/preempt.h>
#include <linux/rcupdate.h>
#include <trace/events/bcachefs.h>
struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
extent_for_each_ptr(e, ptr) {
- struct bch_dev *ca = c->devs[ptr->dev];
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
size_t b = PTR_BUCKET_NR(ca, ptr);
if (gen_after(ca->oldest_gens[b], ptr->gen))
/*
* For runtime mark and sweep:
*/
-static u8 bch2_btree_mark_key(struct bch_fs *c, enum bkey_type type,
- struct bkey_s_c k, unsigned flags)
+static u8 bch2_gc_mark_key(struct bch_fs *c, enum bkey_type type,
+ struct bkey_s_c k, unsigned flags)
{
+ struct gc_pos pos = { 0 };
+ struct bch_fs_usage *stats;
+ u8 ret = 0;
+
+ preempt_disable();
+ stats = this_cpu_ptr(c->usage_percpu);
switch (type) {
case BKEY_TYPE_BTREE:
- bch2_gc_mark_key(c, k, c->opts.btree_node_size, true, flags);
- return 0;
+ bch2_mark_key(c, k, c->opts.btree_node_size, true, pos, stats,
+ 0, flags|
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
+ break;
case BKEY_TYPE_EXTENTS:
- bch2_gc_mark_key(c, k, k.k->size, false, flags);
- return bch2_btree_key_recalc_oldest_gen(c, k);
+ bch2_mark_key(c, k, k.k->size, false, pos, stats,
+ 0, flags|
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
+ ret = bch2_btree_key_recalc_oldest_gen(c, k);
+ break;
default:
BUG();
}
+ preempt_enable();
+
+ return ret;
}
int bch2_btree_mark_key_initial(struct bch_fs *c, enum bkey_type type,
? BCH_DATA_BTREE : BCH_DATA_USER;
int ret = 0;
+ if (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
+ fsck_err_on(!bch2_bkey_replicas_marked(c, data_type, k), c,
+ "superblock not marked as containing replicas (type %u)",
+ data_type)) {
+ ret = bch2_mark_bkey_replicas(c, data_type, k);
+ if (ret)
+ return ret;
+ }
+
switch (k.k->type) {
case BCH_EXTENT:
case BCH_EXTENT_CACHED: {
struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
const struct bch_extent_ptr *ptr;
- if (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
- (!c->opts.nofsck &&
- fsck_err_on(!bch2_sb_has_replicas(c, e, data_type), c,
- "superblock not marked as containing replicas"))) {
- ret = bch2_check_mark_super(c, e, data_type);
- if (ret)
- return ret;
- }
-
extent_for_each_ptr(e, ptr) {
- struct bch_dev *ca = c->devs[ptr->dev];
+ struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+ size_t b = PTR_BUCKET_NR(ca, ptr);
struct bucket *g = PTR_BUCKET(ca, ptr);
if (mustfix_fsck_err_on(!g->mark.gen_valid, c,
ptr->gen)) {
g->_mark.gen = ptr->gen;
g->_mark.gen_valid = 1;
- set_bit(g - ca->buckets, ca->bucket_dirty);
+ set_bit(b, ca->buckets_dirty);
}
if (mustfix_fsck_err_on(gen_cmp(ptr->gen, g->mark.gen) > 0, c,
ptr->gen, g->mark.gen)) {
g->_mark.gen = ptr->gen;
g->_mark.gen_valid = 1;
- set_bit(g - ca->buckets, ca->bucket_dirty);
+ set_bit(b, ca->buckets_dirty);
set_bit(BCH_FS_FIXED_GENS, &c->flags);
}
}
}
-
atomic64_set(&c->key_version,
max_t(u64, k.k->version.lo,
atomic64_read(&c->key_version)));
- bch2_btree_mark_key(c, type, k, BCH_BUCKET_MARK_NOATOMIC);
+ bch2_gc_mark_key(c, type, k, BCH_BUCKET_MARK_NOATOMIC);
fsck_err:
return ret;
}
btree_node_is_extents(b),
&unpacked) {
bch2_bkey_debugcheck(c, b, k);
- stale = max(stale, bch2_btree_mark_key(c, type, k, 0));
+ stale = max(stale, bch2_gc_mark_key(c, type, k, 0));
}
return stale;
mutex_lock(&c->btree_root_lock);
b = c->btree_roots[btree_id].b;
- bch2_btree_mark_key(c, BKEY_TYPE_BTREE, bkey_i_to_s_c(&b->key), 0);
+ if (!btree_node_fake(b))
+ bch2_gc_mark_key(c, BKEY_TYPE_BTREE, bkey_i_to_s_c(&b->key), 0);
gc_pos_set(c, gc_pos_btree_root(b->btree_id));
mutex_unlock(&c->btree_root_lock);
return 0;
}
-static void bch2_mark_allocator_buckets(struct bch_fs *c)
-{
- struct bch_dev *ca;
- struct open_bucket *ob;
- const struct open_bucket_ptr *ptr;
- size_t i, j, iter;
- unsigned ci;
-
- down_write(&c->alloc_gc_lock);
-
- for_each_member_device(ca, c, ci) {
- spin_lock(&ca->freelist_lock);
-
- fifo_for_each_entry(i, &ca->free_inc, iter)
- bch2_mark_alloc_bucket(ca, &ca->buckets[i], true);
-
- for (j = 0; j < RESERVE_NR; j++)
- fifo_for_each_entry(i, &ca->free[j], iter)
- bch2_mark_alloc_bucket(ca, &ca->buckets[i], true);
-
- for (ptr = ca->open_buckets_partial;
- ptr < ca->open_buckets_partial + ca->open_buckets_partial_nr;
- ptr++)
- bch2_mark_alloc_bucket(ca, PTR_BUCKET(ca, &ptr->ptr), true);
-
- spin_unlock(&ca->freelist_lock);
- }
-
- for (ob = c->open_buckets;
- ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
- ob++) {
- spin_lock(&ob->lock);
- open_bucket_for_each_ptr(ob, ptr) {
- ca = c->devs[ptr->ptr.dev];
- bch2_mark_alloc_bucket(ca, PTR_BUCKET(ca, &ptr->ptr), true);
- }
- spin_unlock(&ob->lock);
- }
-
- up_write(&c->alloc_gc_lock);
-}
-
-static void mark_metadata_sectors(struct bch_dev *ca, u64 start, u64 end,
- enum bucket_data_type type)
+static void mark_metadata_sectors(struct bch_fs *c, struct bch_dev *ca,
+ u64 start, u64 end,
+ enum bch_data_type type,
+ unsigned flags)
{
u64 b = sector_to_bucket(ca, start);
do {
- bch2_mark_metadata_bucket(ca, ca->buckets + b, type, true);
+ unsigned sectors =
+ min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
+
+ bch2_mark_metadata_bucket(c, ca, b, type, sectors,
+ gc_phase(GC_PHASE_SB), flags);
b++;
- } while (b < sector_to_bucket(ca, end));
+ start += sectors;
+ } while (start < end);
}
-static void bch2_dev_mark_superblocks(struct bch_dev *ca)
+void bch2_mark_dev_superblock(struct bch_fs *c, struct bch_dev *ca,
+ unsigned flags)
{
struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
unsigned i;
+ u64 b;
- for (i = 0; i < layout->nr_superblocks; i++) {
- if (layout->sb_offset[i] == BCH_SB_SECTOR)
- mark_metadata_sectors(ca, 0, BCH_SB_SECTOR,
- BUCKET_SB);
-
- mark_metadata_sectors(ca,
- layout->sb_offset[i],
- layout->sb_offset[i] +
- (1 << layout->sb_max_size_bits),
- BUCKET_SB);
- }
-}
+ if (c)
+ lockdep_assert_held(&c->sb_lock);
-/*
- * Mark non btree metadata - prios, journal
- */
-void bch2_mark_dev_metadata(struct bch_fs *c, struct bch_dev *ca)
-{
- unsigned i;
- u64 b;
+ for (i = 0; i < layout->nr_superblocks; i++) {
+ u64 offset = le64_to_cpu(layout->sb_offset[i]);
- lockdep_assert_held(&c->sb_lock);
+ if (offset == BCH_SB_SECTOR)
+ mark_metadata_sectors(c, ca, 0, BCH_SB_SECTOR,
+ BCH_DATA_SB, flags);
- bch2_dev_mark_superblocks(ca);
+ mark_metadata_sectors(c, ca, offset,
+ offset + (1 << layout->sb_max_size_bits),
+ BCH_DATA_SB, flags);
+ }
- spin_lock(&c->journal.lock);
+ if (c)
+ spin_lock(&c->journal.lock);
for (i = 0; i < ca->journal.nr; i++) {
b = ca->journal.buckets[i];
- bch2_mark_metadata_bucket(ca, ca->buckets + b,
- BUCKET_JOURNAL, true);
+ bch2_mark_metadata_bucket(c, ca, b, BCH_DATA_JOURNAL,
+ ca->mi.bucket_size,
+ gc_phase(GC_PHASE_SB), flags);
}
- spin_unlock(&c->journal.lock);
+ if (c)
+ spin_unlock(&c->journal.lock);
}
-static void bch2_mark_metadata(struct bch_fs *c)
+static void bch2_mark_superblocks(struct bch_fs *c)
{
struct bch_dev *ca;
unsigned i;
mutex_lock(&c->sb_lock);
- gc_pos_set(c, gc_phase(GC_PHASE_SB_METADATA));
+ gc_pos_set(c, gc_phase(GC_PHASE_SB));
for_each_online_member(ca, c, i)
- bch2_mark_dev_metadata(c, ca);
+ bch2_mark_dev_superblock(c, ca,
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
mutex_unlock(&c->sb_lock);
}
/* Also see bch2_pending_btree_node_free_insert_done() */
static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
{
+ struct gc_pos pos = { 0 };
struct bch_fs_usage stats = { 0 };
struct btree_update *as;
struct pending_btree_node_free *d;
for_each_pending_btree_node_free(c, as, d)
if (d->index_update_done)
- __bch2_mark_key(c, bkey_i_to_s_c(&d->key),
- c->opts.btree_node_size, true,
- &stats, 0,
- BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
+ bch2_mark_key(c, bkey_i_to_s_c(&d->key),
+ c->opts.btree_node_size, true, pos,
+ &stats, 0,
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
/*
* Don't apply stats - pending deletes aren't tracked in
* bch_alloc_stats:
mutex_unlock(&c->btree_interior_update_lock);
}
-void bch2_gc_start(struct bch_fs *c)
+static void bch2_mark_allocator_buckets(struct bch_fs *c)
{
struct bch_dev *ca;
- struct bucket *g;
+ struct open_bucket *ob;
+ size_t i, j, iter;
+ unsigned ci;
+
+ spin_lock(&c->freelist_lock);
+ gc_pos_set(c, gc_pos_alloc(c, NULL));
+
+ for_each_member_device(ca, c, ci) {
+ fifo_for_each_entry(i, &ca->free_inc, iter)
+ bch2_mark_alloc_bucket(c, ca, i, true,
+ gc_pos_alloc(c, NULL),
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
+
+
+
+ for (j = 0; j < RESERVE_NR; j++)
+ fifo_for_each_entry(i, &ca->free[j], iter)
+ bch2_mark_alloc_bucket(c, ca, i, true,
+ gc_pos_alloc(c, NULL),
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
+ }
+
+ spin_unlock(&c->freelist_lock);
+
+ for (ob = c->open_buckets;
+ ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
+ ob++) {
+ spin_lock(&ob->lock);
+ if (ob->valid) {
+ gc_pos_set(c, gc_pos_alloc(c, ob));
+ ca = bch_dev_bkey_exists(c, ob->ptr.dev);
+ bch2_mark_alloc_bucket(c, ca, PTR_BUCKET_NR(ca, &ob->ptr), true,
+ gc_pos_alloc(c, ob),
+ BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE|
+ BCH_BUCKET_MARK_GC_LOCK_HELD);
+ }
+ spin_unlock(&ob->lock);
+ }
+}
+
+static void bch2_gc_start(struct bch_fs *c)
+{
+ struct bch_dev *ca;
+ struct bucket_array *buckets;
struct bucket_mark new;
unsigned i;
+ size_t b;
int cpu;
lg_global_lock(&c->usage_lock);
lg_global_unlock(&c->usage_lock);
/* Clear bucket marks: */
- for_each_member_device(ca, c, i)
- for_each_bucket(g, ca) {
- bucket_cmpxchg(g, new, ({
+ for_each_member_device(ca, c, i) {
+ down_read(&ca->bucket_lock);
+ buckets = bucket_array(ca);
+
+ for (b = buckets->first_bucket; b < buckets->nbuckets; b++) {
+ bucket_cmpxchg(buckets->b + b, new, ({
new.owned_by_allocator = 0;
new.data_type = 0;
new.cached_sectors = 0;
new.dirty_sectors = 0;
}));
- ca->oldest_gens[g - ca->buckets] = new.gen;
+ ca->oldest_gens[b] = new.gen;
}
+ up_read(&ca->bucket_lock);
+ }
}
/**
bch2_gc_start(c);
- /* Walk allocator's references: */
- bch2_mark_allocator_buckets(c);
-
/* Walk btree: */
while (c->gc_pos.phase < (int) BTREE_ID_NR) {
int ret = c->btree_roots[c->gc_pos.phase].b
gc_pos_set(c, gc_phase(c->gc_pos.phase + 1));
}
- bch2_mark_metadata(c);
+ bch2_mark_superblocks(c);
bch2_mark_pending_btree_node_frees(c);
+ bch2_mark_allocator_buckets(c);
for_each_member_device(ca, c, i)
atomic_long_set(&ca->saturated_count, 0);
static void bch2_coalesce_nodes(struct bch_fs *c, struct btree_iter *iter,
struct btree *old_nodes[GC_MERGE_NODES])
{
- struct btree *parent = iter->nodes[old_nodes[0]->level + 1];
+ struct btree *parent = btree_node_parent(iter, old_nodes[0]);
unsigned i, nr_old_nodes, nr_new_nodes, u64s = 0;
unsigned blocks = btree_blocks(c) * 2 / 3;
struct btree *new_nodes[GC_MERGE_NODES];
struct bkey_format new_format;
memset(new_nodes, 0, sizeof(new_nodes));
- bch2_keylist_init(&keylist, NULL, 0);
+ bch2_keylist_init(&keylist, NULL);
/* Count keys that are not deleted */
for (i = 0; i < GC_MERGE_NODES && old_nodes[i]; i++)
bch2_btree_build_aux_trees(n);
six_unlock_write(&n->lock);
- bch2_btree_node_write(c, n, &as->cl, SIX_LOCK_intent);
+ bch2_btree_node_write(c, n, SIX_LOCK_intent);
}
/*
BUG_ON(!bch2_keylist_empty(&keylist));
- BUG_ON(iter->nodes[old_nodes[0]->level] != old_nodes[0]);
+ BUG_ON(iter->l[old_nodes[0]->level].b != old_nodes[0]);
BUG_ON(!bch2_btree_iter_node_replace(iter, new_nodes[0]));
* and the nodes in our sliding window might not have the same
* parent anymore - blow away the sliding window:
*/
- if (iter.nodes[iter.level + 1] &&
+ if (btree_iter_node(&iter, iter.level + 1) &&
!btree_node_intent_locked(&iter, iter.level + 1))
memset(merge + 1, 0,
(GC_MERGE_NODES - 1) * sizeof(merge[0]));
if (!c->btree_roots[id].b)
return 0;
- ret = bch2_btree_mark_key_initial(c, BKEY_TYPE_BTREE,
- bkey_i_to_s_c(&c->btree_roots[id].b->key));
+ b = c->btree_roots[id].b;
+ if (!btree_node_fake(b))
+ ret = bch2_btree_mark_key_initial(c, BKEY_TYPE_BTREE,
+ bkey_i_to_s_c(&b->key));
if (ret)
return ret;
return bch2_btree_iter_unlock(&iter) ?: ret;
}
-int bch2_initial_gc(struct bch_fs *c, struct list_head *journal)
+static int __bch2_initial_gc(struct bch_fs *c, struct list_head *journal)
{
unsigned iter = 0;
enum btree_id id;
int ret;
mutex_lock(&c->sb_lock);
- if (!bch2_sb_get_replicas(c->disk_sb)) {
- if (BCH_SB_INITIALIZED(c->disk_sb))
+ if (!bch2_sb_get_replicas(c->disk_sb.sb)) {
+ if (BCH_SB_INITIALIZED(c->disk_sb.sb))
bch_info(c, "building replicas info");
set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
}
ret = bch2_journal_mark(c, journal);
if (ret)
- return ret;
-
- bch2_mark_metadata(c);
+ return ret;
if (test_bit(BCH_FS_FIXED_GENS, &c->flags)) {
if (iter++ > 2) {
if (c->sb.encryption_type)
atomic64_add(1 << 16, &c->key_version);
+ bch2_mark_superblocks(c);
+
gc_pos_set(c, gc_phase(GC_PHASE_DONE));
set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
return 0;
}
+
+int bch2_initial_gc(struct bch_fs *c, struct list_head *journal)
+{
+ int ret;
+
+ down_write(&c->gc_lock);
+ ret = __bch2_initial_gc(c, journal);
+ up_write(&c->gc_lock);
+
+ return ret;
+}
projects / bcachefs-tools-debian / blobdiff