+// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
* Copyright (C) 2014 Datera Inc.
#include "error.h"
#include "extents.h"
#include "journal.h"
-#include "journal_io.h"
#include "keylist.h"
#include "move.h"
+#include "recovery.h"
#include "replicas.h"
#include "super-io.h"
__gc_pos_set(c, new_pos);
}
-/* range_checks - for validating min/max pos of each btree node: */
-
-struct range_checks {
- struct range_level {
- struct bpos min;
- struct bpos max;
- } l[BTREE_MAX_DEPTH];
- unsigned depth;
-};
-
-static void btree_node_range_checks_init(struct range_checks *r, unsigned depth)
-{
- unsigned i;
-
- for (i = 0; i < BTREE_MAX_DEPTH; i++)
- r->l[i].min = r->l[i].max = POS_MIN;
- r->depth = depth;
-}
-
-static void btree_node_range_checks(struct bch_fs *c, struct btree *b,
- struct range_checks *r)
+static int bch2_gc_check_topology(struct bch_fs *c,
+ struct bkey_s_c k,
+ struct bpos *expected_start,
+ struct bpos expected_end,
+ bool is_last)
{
- struct range_level *l = &r->l[b->level];
-
- struct bpos expected_min = bkey_cmp(l->min, l->max)
- ? btree_type_successor(b->btree_id, l->max)
- : l->max;
-
- bch2_fs_inconsistent_on(bkey_cmp(b->data->min_key, expected_min), c,
- "btree node has incorrect min key: %llu:%llu != %llu:%llu",
- b->data->min_key.inode,
- b->data->min_key.offset,
- expected_min.inode,
- expected_min.offset);
-
- l->max = b->data->max_key;
+ int ret = 0;
- if (b->level > r->depth) {
- l = &r->l[b->level - 1];
+ if (k.k->type == KEY_TYPE_btree_ptr_v2) {
+ struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- bch2_fs_inconsistent_on(bkey_cmp(b->data->min_key, l->min), c,
- "btree node min doesn't match min of child nodes: %llu:%llu != %llu:%llu",
- b->data->min_key.inode,
- b->data->min_key.offset,
- l->min.inode,
- l->min.offset);
+ if (fsck_err_on(bkey_cmp(*expected_start, bp.v->min_key), c,
+ "btree node with incorrect min_key: got %llu:%llu, should be %llu:%llu",
+ bp.v->min_key.inode,
+ bp.v->min_key.offset,
+ expected_start->inode,
+ expected_start->offset)) {
+ BUG();
+ }
+ }
- bch2_fs_inconsistent_on(bkey_cmp(b->data->max_key, l->max), c,
- "btree node max doesn't match max of child nodes: %llu:%llu != %llu:%llu",
- b->data->max_key.inode,
- b->data->max_key.offset,
- l->max.inode,
- l->max.offset);
-
- if (bkey_cmp(b->data->max_key, POS_MAX))
- l->min = l->max =
- btree_type_successor(b->btree_id,
- b->data->max_key);
+ *expected_start = bkey_cmp(k.k->p, POS_MAX)
+ ? bkey_successor(k.k->p)
+ : k.k->p;
+
+ if (fsck_err_on(is_last &&
+ bkey_cmp(k.k->p, expected_end), c,
+ "btree node with incorrect max_key: got %llu:%llu, should be %llu:%llu",
+ k.k->p.inode,
+ k.k->p.offset,
+ expected_end.inode,
+ expected_end.offset)) {
+ BUG();
}
+fsck_err:
+ return ret;
}
/* marking of btree keys/nodes: */
{
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
const struct bch_extent_ptr *ptr;
- struct gc_pos pos = { 0 };
unsigned flags =
- BCH_BUCKET_MARK_GC|
- (initial ? BCH_BUCKET_MARK_NOATOMIC : 0);
+ BTREE_TRIGGER_GC|
+ (initial ? BTREE_TRIGGER_NOATOMIC : 0);
int ret = 0;
if (initial) {
BUG_ON(journal_seq_verify(c) &&
k.k->version.lo > journal_cur_seq(&c->journal));
- if (k.k->version.lo > atomic64_read(&c->key_version))
+ /* XXX change to fsck check */
+ if (fsck_err_on(k.k->version.lo > atomic64_read(&c->key_version), c,
+ "key version number higher than recorded: %llu > %llu",
+ k.k->version.lo,
+ atomic64_read(&c->key_version)))
atomic64_set(&c->key_version, k.k->version.lo);
if (test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) ||
bkey_for_each_ptr(ptrs, ptr) {
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);
+ struct bucket *g = PTR_BUCKET(ca, ptr, true);
+ struct bucket *g2 = PTR_BUCKET(ca, ptr, false);
if (mustfix_fsck_err_on(!g->gen_valid, c,
- "found ptr with missing gen in alloc btree,\n"
- "type %u gen %u",
- k.k->type, ptr->gen)) {
- g->_mark.gen = ptr->gen;
- g->gen_valid = 1;
- bucket_set_dirty(ca, b);
+ "bucket %u:%zu data type %s ptr gen %u missing in alloc btree",
+ ptr->dev, PTR_BUCKET_NR(ca, ptr),
+ bch2_data_types[ptr_data_type(k.k, ptr)],
+ ptr->gen)) {
+ g2->_mark.gen = g->_mark.gen = ptr->gen;
+ g2->gen_valid = g->gen_valid = true;
}
if (mustfix_fsck_err_on(gen_cmp(ptr->gen, g->mark.gen) > 0, c,
- "%u ptr gen in the future: %u > %u",
- k.k->type, ptr->gen, g->mark.gen)) {
- g->_mark.gen = ptr->gen;
- g->gen_valid = 1;
- bucket_set_dirty(ca, b);
+ "bucket %u:%zu data type %s ptr gen in the future: %u > %u",
+ ptr->dev, PTR_BUCKET_NR(ca, ptr),
+ bch2_data_types[ptr_data_type(k.k, ptr)],
+ ptr->gen, g->mark.gen)) {
+ g2->_mark.gen = g->_mark.gen = ptr->gen;
+ g2->gen_valid = g->gen_valid = true;
+ g2->_mark.data_type = 0;
+ g2->_mark.dirty_sectors = 0;
+ g2->_mark.cached_sectors = 0;
set_bit(BCH_FS_FIXED_GENS, &c->flags);
}
}
bkey_for_each_ptr(ptrs, ptr) {
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, true);
- if (gen_after(ca->oldest_gens[b], ptr->gen))
- ca->oldest_gens[b] = ptr->gen;
+ if (gen_after(g->oldest_gen, ptr->gen))
+ g->oldest_gen = ptr->gen;
*max_stale = max(*max_stale, ptr_stale(ca, ptr));
}
- bch2_mark_key(c, k, true, k.k->size, pos, NULL, 0, flags);
+ bch2_mark_key(c, k, 0, k.k->size, NULL, 0, flags);
fsck_err:
return ret;
}
-static int btree_gc_mark_node(struct bch_fs *c, struct btree *b,
- u8 *max_stale, bool initial)
+static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale,
+ bool initial)
{
+ struct bpos next_node_start = b->data->min_key;
struct btree_node_iter iter;
struct bkey unpacked;
struct bkey_s_c k;
if (!btree_node_type_needs_gc(btree_node_type(b)))
return 0;
- for_each_btree_node_key_unpack(b, k, &iter,
- &unpacked) {
+ bch2_btree_node_iter_init_from_start(&iter, b);
+
+ while ((k = bch2_btree_node_iter_peek_unpack(&iter, b, &unpacked)).k) {
bch2_bkey_debugcheck(c, b, k);
ret = bch2_gc_mark_key(c, k, max_stale, initial);
if (ret)
break;
+
+ bch2_btree_node_iter_advance(&iter, b);
+
+ if (b->level) {
+ ret = bch2_gc_check_topology(c, k,
+ &next_node_start,
+ b->data->max_key,
+ bch2_btree_node_iter_end(&iter));
+ if (ret)
+ break;
+ }
}
return ret;
}
static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id,
- bool initial)
+ bool initial, bool metadata_only)
{
- struct btree_iter iter;
+ struct btree_trans trans;
+ struct btree_iter *iter;
struct btree *b;
- struct range_checks r;
- unsigned depth = btree_node_type_needs_gc(btree_id) ? 0 : 1;
- u8 max_stale;
+ unsigned depth = metadata_only ? 1
+ : expensive_debug_checks(c) ? 0
+ : !btree_node_type_needs_gc(btree_id) ? 1
+ : 0;
+ u8 max_stale = 0;
int ret = 0;
- gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
+ bch2_trans_init(&trans, c, 0, 0);
- /*
- * if expensive_debug_checks is on, run range_checks on all leaf nodes:
- *
- * and on startup, we have to read every btree node (XXX: only if it was
- * an unclean shutdown)
- */
- if (initial || expensive_debug_checks(c))
- depth = 0;
-
- btree_node_range_checks_init(&r, depth);
+ gc_pos_set(c, gc_pos_btree(btree_id, POS_MIN, 0));
- __for_each_btree_node(&iter, c, btree_id, POS_MIN,
+ __for_each_btree_node(&trans, iter, btree_id, POS_MIN,
0, depth, BTREE_ITER_PREFETCH, b) {
- btree_node_range_checks(c, b, &r);
-
bch2_verify_btree_nr_keys(b);
+ gc_pos_set(c, gc_pos_btree_node(b));
+
ret = btree_gc_mark_node(c, b, &max_stale, initial);
if (ret)
break;
- gc_pos_set(c, gc_pos_btree_node(b));
-
if (!initial) {
if (max_stale > 64)
- bch2_btree_node_rewrite(c, &iter,
+ bch2_btree_node_rewrite(c, iter,
b->data->keys.seq,
BTREE_INSERT_USE_RESERVE|
BTREE_INSERT_NOWAIT|
BTREE_INSERT_GC_LOCK_HELD);
else if (!btree_gc_rewrite_disabled(c) &&
(btree_gc_always_rewrite(c) || max_stale > 16))
- bch2_btree_node_rewrite(c, &iter,
+ bch2_btree_node_rewrite(c, iter,
b->data->keys.seq,
BTREE_INSERT_NOWAIT|
BTREE_INSERT_GC_LOCK_HELD);
}
- bch2_btree_iter_cond_resched(&iter);
+ bch2_trans_cond_resched(&trans);
}
- ret = bch2_btree_iter_unlock(&iter) ?: ret;
+ ret = bch2_trans_exit(&trans) ?: ret;
if (ret)
return ret;
mutex_lock(&c->btree_root_lock);
-
b = c->btree_roots[btree_id].b;
if (!btree_node_fake(b))
- bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
- &max_stale, initial);
+ ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
+ &max_stale, initial);
gc_pos_set(c, gc_pos_btree_root(b->btree_id));
-
mutex_unlock(&c->btree_root_lock);
- return 0;
+
+ return ret;
+}
+
+static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b,
+ struct journal_keys *journal_keys,
+ unsigned target_depth)
+{
+ struct btree_and_journal_iter iter;
+ struct bkey_s_c k;
+ struct bpos next_node_start = b->data->min_key;
+ u8 max_stale = 0;
+ int ret = 0;
+
+ bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
+
+ while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
+ bch2_bkey_debugcheck(c, b, k);
+
+ BUG_ON(bkey_cmp(k.k->p, b->data->min_key) < 0);
+ BUG_ON(bkey_cmp(k.k->p, b->data->max_key) > 0);
+
+ ret = bch2_gc_mark_key(c, k, &max_stale, true);
+ if (ret)
+ break;
+
+ if (b->level) {
+ struct btree *child;
+ BKEY_PADDED(k) tmp;
+
+ bkey_reassemble(&tmp.k, k);
+ k = bkey_i_to_s_c(&tmp.k);
+
+ bch2_btree_and_journal_iter_advance(&iter);
+
+ ret = bch2_gc_check_topology(c, k,
+ &next_node_start,
+ b->data->max_key,
+ !bch2_btree_and_journal_iter_peek(&iter).k);
+ if (ret)
+ break;
+
+ if (b->level > target_depth) {
+ child = bch2_btree_node_get_noiter(c, &tmp.k,
+ b->btree_id, b->level - 1);
+ ret = PTR_ERR_OR_ZERO(child);
+ if (ret)
+ break;
+
+ ret = bch2_gc_btree_init_recurse(c, child,
+ journal_keys, target_depth);
+ six_unlock_read(&child->lock);
+
+ if (ret)
+ break;
+ }
+ } else {
+ bch2_btree_and_journal_iter_advance(&iter);
+ }
+ }
+
+ return ret;
+}
+
+static int bch2_gc_btree_init(struct bch_fs *c,
+ struct journal_keys *journal_keys,
+ enum btree_id btree_id,
+ bool metadata_only)
+{
+ struct btree *b;
+ unsigned target_depth = metadata_only ? 1
+ : expensive_debug_checks(c) ? 0
+ : !btree_node_type_needs_gc(btree_id) ? 1
+ : 0;
+ u8 max_stale = 0;
+ int ret = 0;
+
+ b = c->btree_roots[btree_id].b;
+
+ if (btree_node_fake(b))
+ return 0;
+
+ six_lock_read(&b->lock);
+ if (fsck_err_on(bkey_cmp(b->data->min_key, POS_MIN), c,
+ "btree root with incorrect min_key: %llu:%llu",
+ b->data->min_key.inode,
+ b->data->min_key.offset)) {
+ BUG();
+ }
+
+ if (fsck_err_on(bkey_cmp(b->data->max_key, POS_MAX), c,
+ "btree root with incorrect min_key: %llu:%llu",
+ b->data->max_key.inode,
+ b->data->max_key.offset)) {
+ BUG();
+ }
+
+ if (b->level >= target_depth)
+ ret = bch2_gc_btree_init_recurse(c, b,
+ journal_keys, target_depth);
+
+ if (!ret)
+ ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key),
+ &max_stale, true);
+fsck_err:
+ six_unlock_read(&b->lock);
+
+ return ret;
}
static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
(int) btree_id_to_gc_phase(r);
}
-static int bch2_gc_btrees(struct bch_fs *c, struct list_head *journal,
- bool initial)
+static int bch2_gc_btrees(struct bch_fs *c, struct journal_keys *journal_keys,
+ bool initial, bool metadata_only)
{
enum btree_id ids[BTREE_ID_NR];
- u8 max_stale;
unsigned i;
for (i = 0; i < BTREE_ID_NR; i++)
for (i = 0; i < BTREE_ID_NR; i++) {
enum btree_id id = ids[i];
- enum btree_node_type type = __btree_node_type(0, id);
-
- int ret = bch2_gc_btree(c, id, initial);
+ int ret = initial
+ ? bch2_gc_btree_init(c, journal_keys,
+ id, metadata_only)
+ : bch2_gc_btree(c, id, initial, metadata_only);
if (ret)
return ret;
-
- if (journal && btree_node_type_needs_gc(type)) {
- struct bkey_i *k, *n;
- struct jset_entry *j;
- struct journal_replay *r;
- int ret;
-
- list_for_each_entry(r, journal, list)
- for_each_jset_key(k, n, j, &r->j) {
- if (type == __btree_node_type(j->level, j->btree_id)) {
- ret = bch2_gc_mark_key(c,
- bkey_i_to_s_c(k),
- &max_stale, initial);
- if (ret)
- return ret;
- }
- }
- }
}
return 0;
*/
if (c) {
lockdep_assert_held(&c->sb_lock);
- percpu_down_read_preempt_disable(&c->mark_lock);
- } else {
- preempt_disable();
+ percpu_down_read(&c->mark_lock);
}
for (i = 0; i < layout->nr_superblocks; i++) {
gc_phase(GC_PHASE_SB), flags);
}
- if (c) {
- percpu_up_read_preempt_enable(&c->mark_lock);
- } else {
- preempt_enable();
- }
+ if (c)
+ percpu_up_read(&c->mark_lock);
}
static void bch2_mark_superblocks(struct bch_fs *c)
gc_pos_set(c, gc_phase(GC_PHASE_SB));
for_each_online_member(ca, c, i)
- bch2_mark_dev_superblock(c, ca, BCH_BUCKET_MARK_GC);
+ bch2_mark_dev_superblock(c, ca, BTREE_TRIGGER_GC);
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 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),
- true, 0,
- pos, NULL, 0,
- BCH_BUCKET_MARK_GC);
+ 0, 0, NULL, 0,
+ BTREE_TRIGGER_GC);
mutex_unlock(&c->btree_interior_update_lock);
}
size_t i, j, iter;
unsigned ci;
- percpu_down_read_preempt_disable(&c->mark_lock);
+ percpu_down_read(&c->mark_lock);
spin_lock(&c->freelist_lock);
gc_pos_set(c, gc_pos_alloc(c, NULL));
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_GC);
+ BTREE_TRIGGER_GC);
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_GC);
+ BTREE_TRIGGER_GC);
}
spin_unlock(&c->freelist_lock);
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_GC);
+ BTREE_TRIGGER_GC);
}
spin_unlock(&ob->lock);
}
- percpu_up_read_preempt_enable(&c->mark_lock);
+ percpu_up_read(&c->mark_lock);
}
static void bch2_gc_free(struct bch_fs *c)
ca->usage[1] = NULL;
}
- free_percpu(c->usage[1]);
- c->usage[1] = NULL;
-}
-
-static void fs_usage_reset(struct bch_fs_usage *fs_usage)
-{
- memset(&fs_usage->s.gc_start[0], 0,
- sizeof(*fs_usage) - offsetof(typeof(*fs_usage), s.gc_start));
-}
-
-static void fs_usage_cpy(struct bch_fs_usage *dst,
- struct bch_fs_usage *src)
-{
- memcpy(&dst->s.gc_start[0],
- &src->s.gc_start[0],
- sizeof(*dst) - offsetof(typeof(*dst), s.gc_start));
-}
-
-static void bch2_gc_done_nocheck(struct bch_fs *c)
-{
- struct bch_dev *ca;
- unsigned i;
- int cpu;
-
- {
- struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
- struct genradix_iter src_iter = genradix_iter_init(&c->stripes[1], 0);
- struct stripe *dst, *src;
-
- c->ec_stripes_heap.used = 0;
-
- while ((dst = genradix_iter_peek(&dst_iter, &c->stripes[0])) &&
- (src = genradix_iter_peek(&src_iter, &c->stripes[1]))) {
- *dst = *src;
-
- if (dst->alive)
- bch2_stripes_heap_insert(c, dst, dst_iter.pos);
-
- genradix_iter_advance(&dst_iter, &c->stripes[0]);
- genradix_iter_advance(&src_iter, &c->stripes[1]);
- }
- }
-
- for_each_member_device(ca, c, i) {
- struct bucket_array *src = __bucket_array(ca, 1);
-
- memcpy(__bucket_array(ca, 0), src,
- sizeof(struct bucket_array) +
- sizeof(struct bucket) * src->nbuckets);
- };
-
- for_each_member_device(ca, c, i) {
- struct bch_dev_usage *p;
-
- for_each_possible_cpu(cpu) {
- p = per_cpu_ptr(ca->usage[0], cpu);
- memset(p, 0, sizeof(*p));
- }
-
- preempt_disable();
- *this_cpu_ptr(ca->usage[0]) = __bch2_dev_usage_read(ca, 1);
- preempt_enable();
- }
-
- {
- struct bch_fs_usage src = __bch2_fs_usage_read(c, 1);
-
- for_each_possible_cpu(cpu)
- fs_usage_reset(per_cpu_ptr(c->usage[0], cpu));
-
- preempt_disable();
- fs_usage_cpy(this_cpu_ptr(c->usage[0]), &src);
- preempt_enable();
- }
-
+ free_percpu(c->usage_gc);
+ c->usage_gc = NULL;
}
-static void bch2_gc_done(struct bch_fs *c, bool initial)
+static int bch2_gc_done(struct bch_fs *c,
+ bool initial, bool metadata_only)
{
struct bch_dev *ca;
+ bool verify = !metadata_only &&
+ (!initial ||
+ (c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)));
unsigned i;
- int cpu;
+ int ret = 0;
#define copy_field(_f, _msg, ...) \
- if (dst._f != src._f) { \
- bch_err(c, _msg ": got %llu, should be %llu, fixing"\
- , ##__VA_ARGS__, dst._f, src._f); \
- dst._f = src._f; \
+ if (dst->_f != src->_f) { \
+ if (verify) \
+ fsck_err(c, _msg ": got %llu, should be %llu" \
+ , ##__VA_ARGS__, dst->_f, src->_f); \
+ dst->_f = src->_f; \
}
#define copy_stripe_field(_f, _msg, ...) \
if (dst->_f != src->_f) { \
- bch_err_ratelimited(c, "stripe %zu has wrong "_msg \
- ": got %u, should be %u, fixing", \
- dst_iter.pos, ##__VA_ARGS__, \
- dst->_f, src->_f); \
+ if (verify) \
+ fsck_err(c, "stripe %zu has wrong "_msg \
+ ": got %u, should be %u", \
+ dst_iter.pos, ##__VA_ARGS__, \
+ dst->_f, src->_f); \
dst->_f = src->_f; \
+ dst->dirty = true; \
}
#define copy_bucket_field(_f) \
if (dst->b[b].mark._f != src->b[b].mark._f) { \
- bch_err_ratelimited(c, "dev %u bucket %zu has wrong " #_f\
- ": got %u, should be %u, fixing", \
- i, b, dst->b[b].mark._f, src->b[b].mark._f); \
+ if (verify) \
+ fsck_err(c, "dev %u bucket %zu has wrong " #_f \
+ ": got %u, should be %u", i, b, \
+ dst->b[b].mark._f, src->b[b].mark._f); \
dst->b[b]._mark._f = src->b[b].mark._f; \
}
#define copy_dev_field(_f, _msg, ...) \
#define copy_fs_field(_f, _msg, ...) \
copy_field(_f, "fs has wrong " _msg, ##__VA_ARGS__)
- percpu_down_write(&c->mark_lock);
-
- if (initial) {
- bch2_gc_done_nocheck(c);
- goto out;
- }
-
- {
+ if (!metadata_only) {
struct genradix_iter dst_iter = genradix_iter_init(&c->stripes[0], 0);
struct genradix_iter src_iter = genradix_iter_init(&c->stripes[1], 0);
struct stripe *dst, *src;
while ((dst = genradix_iter_peek(&dst_iter, &c->stripes[0])) &&
(src = genradix_iter_peek(&src_iter, &c->stripes[1]))) {
+ BUG_ON(src_iter.pos != dst_iter.pos);
+
copy_stripe_field(alive, "alive");
copy_stripe_field(sectors, "sectors");
copy_stripe_field(algorithm, "algorithm");
copy_stripe_field(nr_blocks, "nr_blocks");
copy_stripe_field(nr_redundant, "nr_redundant");
- copy_stripe_field(blocks_nonempty.counter,
+ copy_stripe_field(blocks_nonempty,
"blocks_nonempty");
for (i = 0; i < ARRAY_SIZE(dst->block_sectors); i++)
- copy_stripe_field(block_sectors[i].counter,
+ copy_stripe_field(block_sectors[i],
"block_sectors[%u]", i);
if (dst->alive)
struct bucket_array *src = __bucket_array(ca, 1);
size_t b;
- if (initial) {
- memcpy(dst, src,
- sizeof(struct bucket_array) +
- sizeof(struct bucket) * dst->nbuckets);
- }
-
for (b = 0; b < src->nbuckets; b++) {
copy_bucket_field(gen);
copy_bucket_field(data_type);
copy_bucket_field(stripe);
copy_bucket_field(dirty_sectors);
copy_bucket_field(cached_sectors);
+
+ dst->b[b].oldest_gen = src->b[b].oldest_gen;
}
};
- for_each_member_device(ca, c, i) {
- struct bch_dev_usage dst = __bch2_dev_usage_read(ca, 0);
- struct bch_dev_usage src = __bch2_dev_usage_read(ca, 1);
- struct bch_dev_usage *p;
- unsigned b;
-
- for (b = 0; b < BCH_DATA_NR; b++)
- copy_dev_field(buckets[b],
- "buckets[%s]", bch2_data_types[b]);
- copy_dev_field(buckets_alloc, "buckets_alloc");
- copy_dev_field(buckets_ec, "buckets_ec");
-
- for (b = 0; b < BCH_DATA_NR; b++)
- copy_dev_field(sectors[b],
- "sectors[%s]", bch2_data_types[b]);
- copy_dev_field(sectors_fragmented,
- "sectors_fragmented");
-
- for_each_possible_cpu(cpu) {
- p = per_cpu_ptr(ca->usage[0], cpu);
- memset(p, 0, sizeof(*p));
- }
+ bch2_fs_usage_acc_to_base(c, 0);
+ bch2_fs_usage_acc_to_base(c, 1);
- preempt_disable();
- p = this_cpu_ptr(ca->usage[0]);
- *p = dst;
- preempt_enable();
- }
+ bch2_dev_usage_from_buckets(c);
{
- struct bch_fs_usage dst = __bch2_fs_usage_read(c, 0);
- struct bch_fs_usage src = __bch2_fs_usage_read(c, 1);
- unsigned r, b;
-
- copy_fs_field(s.hidden, "hidden");
- copy_fs_field(s.data, "data");
- copy_fs_field(s.cached, "cached");
- copy_fs_field(s.reserved, "reserved");
- copy_fs_field(s.nr_inodes, "nr_inodes");
-
- for (r = 0; r < BCH_REPLICAS_MAX; r++) {
- for (b = 0; b < BCH_DATA_NR; b++)
- copy_fs_field(replicas[r].data[b],
- "replicas[%i].data[%s]",
- r, bch2_data_types[b]);
- copy_fs_field(replicas[r].ec_data,
- "replicas[%i].ec_data", r);
- copy_fs_field(replicas[r].persistent_reserved,
- "replicas[%i].persistent_reserved", r);
+ unsigned nr = fs_usage_u64s(c);
+ struct bch_fs_usage *dst = c->usage_base;
+ struct bch_fs_usage *src = (void *)
+ bch2_acc_percpu_u64s((void *) c->usage_gc, nr);
+
+ copy_fs_field(hidden, "hidden");
+ copy_fs_field(btree, "btree");
+
+ if (!metadata_only) {
+ copy_fs_field(data, "data");
+ copy_fs_field(cached, "cached");
+ copy_fs_field(reserved, "reserved");
+ copy_fs_field(nr_inodes,"nr_inodes");
+
+ for (i = 0; i < BCH_REPLICAS_MAX; i++)
+ copy_fs_field(persistent_reserved[i],
+ "persistent_reserved[%i]", i);
}
- for (b = 0; b < BCH_DATA_NR; b++)
- copy_fs_field(buckets[b],
- "buckets[%s]", bch2_data_types[b]);
+ for (i = 0; i < c->replicas.nr; i++) {
+ struct bch_replicas_entry *e =
+ cpu_replicas_entry(&c->replicas, i);
+ char buf[80];
+
+ if (metadata_only &&
+ (e->data_type == BCH_DATA_USER ||
+ e->data_type == BCH_DATA_CACHED))
+ continue;
- for_each_possible_cpu(cpu)
- fs_usage_reset(per_cpu_ptr(c->usage[0], cpu));
+ bch2_replicas_entry_to_text(&PBUF(buf), e);
- preempt_disable();
- fs_usage_cpy(this_cpu_ptr(c->usage[0]), &dst);
- preempt_enable();
+ copy_fs_field(replicas[i], "%s", buf);
+ }
}
-out:
- percpu_up_write(&c->mark_lock);
#undef copy_fs_field
#undef copy_dev_field
#undef copy_bucket_field
#undef copy_stripe_field
#undef copy_field
+fsck_err:
+ return ret;
}
-static int bch2_gc_start(struct bch_fs *c)
+static int bch2_gc_start(struct bch_fs *c,
+ bool metadata_only)
{
struct bch_dev *ca;
unsigned i;
+ int ret;
- /*
- * indicate to stripe code that we need to allocate for the gc stripes
- * radix tree, too
- */
- gc_pos_set(c, gc_phase(GC_PHASE_START));
-
- BUG_ON(c->usage[1]);
+ BUG_ON(c->usage_gc);
- c->usage[1] = alloc_percpu(struct bch_fs_usage);
- if (!c->usage[1])
+ c->usage_gc = __alloc_percpu_gfp(fs_usage_u64s(c) * sizeof(u64),
+ sizeof(u64), GFP_KERNEL);
+ if (!c->usage_gc) {
+ bch_err(c, "error allocating c->usage_gc");
return -ENOMEM;
+ }
for_each_member_device(ca, c, i) {
BUG_ON(ca->buckets[1]);
GFP_KERNEL|__GFP_ZERO);
if (!ca->buckets[1]) {
percpu_ref_put(&ca->ref);
+ bch_err(c, "error allocating ca->buckets[gc]");
return -ENOMEM;
}
ca->usage[1] = alloc_percpu(struct bch_dev_usage);
if (!ca->usage[1]) {
+ bch_err(c, "error allocating ca->usage[gc]");
percpu_ref_put(&ca->ref);
return -ENOMEM;
}
}
+ ret = bch2_ec_mem_alloc(c, true);
+ if (ret) {
+ bch_err(c, "error allocating ec gc mem");
+ return ret;
+ }
+
percpu_down_write(&c->mark_lock);
+ /*
+ * indicate to stripe code that we need to allocate for the gc stripes
+ * radix tree, too
+ */
+ gc_pos_set(c, gc_phase(GC_PHASE_START));
+
for_each_member_device(ca, c, i) {
struct bucket_array *dst = __bucket_array(ca, 1);
struct bucket_array *src = __bucket_array(ca, 0);
dst->first_bucket = src->first_bucket;
dst->nbuckets = src->nbuckets;
- for (b = 0; b < src->nbuckets; b++)
- dst->b[b]._mark.gen = src->b[b].mark.gen;
+ for (b = 0; b < src->nbuckets; b++) {
+ struct bucket *d = &dst->b[b];
+ struct bucket *s = &src->b[b];
+
+ d->_mark.gen = dst->b[b].oldest_gen = s->mark.gen;
+ d->gen_valid = s->gen_valid;
+
+ if (metadata_only &&
+ (s->mark.data_type == BCH_DATA_USER ||
+ s->mark.data_type == BCH_DATA_CACHED)) {
+ d->_mark = s->mark;
+ d->_mark.owned_by_allocator = 0;
+ }
+ }
};
percpu_up_write(&c->mark_lock);
- return bch2_ec_mem_alloc(c, true);
+ return 0;
}
/**
* move around - if references move backwards in the ordering GC
* uses, GC could skip past them
*/
-int bch2_gc(struct bch_fs *c, struct list_head *journal, bool initial)
+int bch2_gc(struct bch_fs *c, struct journal_keys *journal_keys,
+ bool initial, bool metadata_only)
{
struct bch_dev *ca;
u64 start_time = local_clock();
down_write(&c->gc_lock);
again:
- ret = bch2_gc_start(c);
+ ret = bch2_gc_start(c, metadata_only);
if (ret)
goto out;
bch2_mark_superblocks(c);
- ret = bch2_gc_btrees(c, journal, initial);
+ ret = bch2_gc_btrees(c, journal_keys, initial, metadata_only);
if (ret)
goto out;
c->gc_count++;
out:
- if (!ret && test_bit(BCH_FS_FIXED_GENS, &c->flags)) {
+ if (!ret &&
+ (test_bit(BCH_FS_FIXED_GENS, &c->flags) ||
+ (!iter && test_restart_gc(c)))) {
/*
* XXX: make sure gens we fixed got saved
*/
if (iter++ <= 2) {
bch_info(c, "Fixed gens, restarting mark and sweep:");
clear_bit(BCH_FS_FIXED_GENS, &c->flags);
+ __gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
+
+ percpu_down_write(&c->mark_lock);
+ bch2_gc_free(c);
+ percpu_up_write(&c->mark_lock);
+ /* flush fsck errors, reset counters */
+ bch2_flush_fsck_errs(c);
+
goto again;
}
ret = -EINVAL;
}
- if (!ret)
- bch2_gc_done(c, initial);
+ if (!ret) {
+ bch2_journal_block(&c->journal);
+
+ percpu_down_write(&c->mark_lock);
+ ret = bch2_gc_done(c, initial, metadata_only);
+
+ bch2_journal_unblock(&c->journal);
+ } else {
+ percpu_down_write(&c->mark_lock);
+ }
/* Indicates that gc is no longer in progress: */
__gc_pos_set(c, gc_phase(GC_PHASE_NOT_RUNNING));
bch2_gc_free(c);
- up_write(&c->gc_lock);
+ percpu_up_write(&c->mark_lock);
- if (!ret && initial)
- set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
+ up_write(&c->gc_lock);
trace_gc_end(c);
bch2_time_stats_update(&c->times[BCH_TIME_btree_gc], start_time);
return;
}
- as = bch2_btree_update_start(c, iter->btree_id,
+ as = bch2_btree_update_start(iter->trans, iter->btree_id,
btree_update_reserve_required(c, parent) + nr_old_nodes,
BTREE_INSERT_NOFAIL|
BTREE_INSERT_USE_RESERVE,
k < vstruct_last(s2) &&
vstruct_blocks_plus(n1->data, c->block_bits,
u64s + k->u64s) <= blocks;
- k = bkey_next(k)) {
+ k = bkey_next_skip_noops(k, vstruct_last(s2))) {
last = k;
u64s += k->u64s;
}
n1->key.k.p = n1->data->max_key =
bkey_unpack_pos(n1, last);
- n2->data->min_key =
- btree_type_successor(iter->btree_id,
- n1->data->max_key);
+ n2->data->min_key = bkey_successor(n1->data->max_key);
memcpy_u64s(vstruct_last(s1),
s2->start, u64s);
old_nodes[i] = new_nodes[i];
} else {
old_nodes[i] = NULL;
- if (new_nodes[i])
- six_unlock_intent(&new_nodes[i]->lock);
}
}
+ for (i = 0; i < nr_new_nodes; i++)
+ six_unlock_intent(&new_nodes[i]->lock);
+
bch2_btree_update_done(as);
bch2_keylist_free(&keylist, NULL);
}
static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id)
{
- struct btree_iter iter;
+ struct btree_trans trans;
+ struct btree_iter *iter;
struct btree *b;
bool kthread = (current->flags & PF_KTHREAD) != 0;
unsigned i;
struct btree *merge[GC_MERGE_NODES];
u32 lock_seq[GC_MERGE_NODES];
+ bch2_trans_init(&trans, c, 0, 0);
+
/*
* XXX: We don't have a good way of positively matching on sibling nodes
* that have the same parent - this code works by handling the cases
*/
memset(merge, 0, sizeof(merge));
- __for_each_btree_node(&iter, c, btree_id, POS_MIN,
+ __for_each_btree_node(&trans, iter, btree_id, POS_MIN,
BTREE_MAX_DEPTH, 0,
BTREE_ITER_PREFETCH, b) {
memmove(merge + 1, merge,
}
memset(merge + i, 0, (GC_MERGE_NODES - i) * sizeof(merge[0]));
- bch2_coalesce_nodes(c, &iter, merge);
+ bch2_coalesce_nodes(c, iter, merge);
for (i = 1; i < GC_MERGE_NODES && merge[i]; i++) {
lock_seq[i] = merge[i]->lock.state.seq;
lock_seq[0] = merge[0]->lock.state.seq;
if (kthread && kthread_should_stop()) {
- bch2_btree_iter_unlock(&iter);
+ bch2_trans_exit(&trans);
return -ESHUTDOWN;
}
- bch2_btree_iter_cond_resched(&iter);
+ bch2_trans_cond_resched(&trans);
/*
* If the parent node wasn't relocked, it might have been split
* and the nodes in our sliding window might not have the same
* parent anymore - blow away the sliding window:
*/
- if (btree_iter_node(&iter, iter.level + 1) &&
- !btree_node_intent_locked(&iter, 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]));
}
- return bch2_btree_iter_unlock(&iter);
+ return bch2_trans_exit(&trans);
}
/**
last = atomic_long_read(&clock->now);
last_kick = atomic_read(&c->kick_gc);
- ret = bch2_gc(c, NULL, false);
+ ret = bch2_gc(c, NULL, false, false);
if (ret)
bch_err(c, "btree gc failed: %i", ret);
wake_up_process(p);
return 0;
}
-
-/* Initial GC computes bucket marks during startup */
-
-int bch2_initial_gc(struct bch_fs *c, struct list_head *journal)
-{
- int ret = bch2_gc(c, journal, true);
-
- /*
- * Skip past versions that might have possibly been used (as nonces),
- * but hadn't had their pointers written:
- */
- if (c->sb.encryption_type)
- atomic64_add(1 << 16, &c->key_version);
-
- return ret;
-}