#include <linux/device.h>
#include <linux/genhd.h>
#include <linux/idr.h>
-#include <linux/kthread.h>
#include <linux/module.h>
#include <linux/percpu.h>
#include <linux/random.h>
return c;
}
-int bch2_congested(void *data, int bdi_bits)
+static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
{
- struct bch_fs *c = data;
- struct backing_dev_info *bdi;
struct bch_dev *ca;
- unsigned i;
- int ret = 0;
+ unsigned i, nr = 0, u64s =
+ ((sizeof(struct jset_entry_dev_usage) +
+ sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
+ sizeof(u64);
rcu_read_lock();
- if (bdi_bits & (1 << WB_sync_congested)) {
- /* Reads - check all devices: */
- for_each_readable_member(ca, c, i) {
- bdi = ca->disk_sb.bdev->bd_bdi;
-
- if (bdi_congested(bdi, bdi_bits)) {
- ret = 1;
- break;
- }
- }
- } else {
- unsigned target = READ_ONCE(c->opts.foreground_target);
- const struct bch_devs_mask *devs = target
- ? bch2_target_to_mask(c, target)
- : &c->rw_devs[BCH_DATA_USER];
-
- for_each_member_device_rcu(ca, c, i, devs) {
- bdi = ca->disk_sb.bdev->bd_bdi;
-
- if (bdi_congested(bdi, bdi_bits)) {
- ret = 1;
- break;
- }
- }
- }
+ for_each_member_device_rcu(ca, c, i, NULL)
+ nr++;
rcu_read_unlock();
- return ret;
+ bch2_journal_entry_res_resize(&c->journal,
+ &c->dev_usage_journal_res, u64s * nr);
}
/* Filesystem RO/RW: */
static void __bch2_fs_read_only(struct bch_fs *c)
{
struct bch_dev *ca;
- bool wrote = false;
unsigned i, clean_passes = 0;
- int ret;
bch2_rebalance_stop(c);
-
- for_each_member_device(ca, c, i)
- bch2_copygc_stop(ca);
-
+ bch2_copygc_stop(c);
bch2_gc_thread_stop(c);
/*
if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
goto nowrote_alloc;
- bch_verbose(c, "writing alloc info");
- /*
- * This should normally just be writing the bucket read/write clocks:
- */
- ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?:
- bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
- bch_verbose(c, "writing alloc info complete");
-
- if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
- bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
-
- if (ret)
- goto nowrote_alloc;
-
bch_verbose(c, "flushing journal and stopping allocators");
bch2_journal_flush_all_pins(&c->journal);
* the journal kicks off btree writes via reclaim - wait for in flight
* writes after stopping journal:
*/
- if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
- bch2_btree_flush_all_writes(c);
- else
- bch2_btree_verify_flushed(c);
+ bch2_btree_flush_all_writes(c);
/*
* After stopping journal:
void bch2_fs_read_only(struct bch_fs *c)
{
if (!test_bit(BCH_FS_RW, &c->flags)) {
- cancel_delayed_work_sync(&c->journal.reclaim_work);
+ BUG_ON(c->journal.reclaim_thread);
return;
}
{
bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
- bch2_fs_read_only_async(c);
bch2_journal_halt(&c->journal);
+ bch2_fs_read_only_async(c);
wake_up(&bch_read_only_wait);
return ret;
static int bch2_fs_read_write_late(struct bch_fs *c)
{
- struct bch_dev *ca;
- unsigned i;
int ret;
ret = bch2_gc_thread_start(c);
return ret;
}
- for_each_rw_member(ca, c, i) {
- ret = bch2_copygc_start(c, ca);
- if (ret) {
- bch_err(c, "error starting copygc threads");
- percpu_ref_put(&ca->io_ref);
- return ret;
- }
+ ret = bch2_copygc_start(c);
+ if (ret) {
+ bch_err(c, "error starting copygc thread");
+ return ret;
}
ret = bch2_rebalance_start(c);
if (ret)
goto err;
+ /*
+ * We need to write out a journal entry before we start doing btree
+ * updates, to ensure that on unclean shutdown new journal blacklist
+ * entries are created:
+ */
+ bch2_journal_meta(&c->journal);
+
clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
for_each_rw_member(ca, c, i)
set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
+ ret = bch2_journal_reclaim_start(&c->journal);
+ if (ret) {
+ bch_err(c, "error starting journal reclaim: %i", ret);
+ return ret;
+ }
+
if (!early) {
ret = bch2_fs_read_write_late(c);
if (ret)
percpu_ref_reinit(&c->writes);
set_bit(BCH_FS_RW, &c->flags);
-
- queue_delayed_work(c->journal_reclaim_wq,
- &c->journal.reclaim_work, 0);
return 0;
err:
__bch2_fs_read_only(c);
/* Filesystem startup/shutdown: */
-static void bch2_fs_free(struct bch_fs *c)
+static void __bch2_fs_free(struct bch_fs *c)
{
unsigned i;
+ int cpu;
for (i = 0; i < BCH_TIME_STAT_NR; i++)
bch2_time_stats_exit(&c->times[i]);
bch2_journal_entries_free(&c->journal_entries);
percpu_free_rwsem(&c->mark_lock);
kfree(c->usage_scratch);
- free_percpu(c->usage[1]);
- free_percpu(c->usage[0]);
+ for (i = 0; i < ARRAY_SIZE(c->usage); i++)
+ free_percpu(c->usage[i]);
kfree(c->usage_base);
+
+ if (c->btree_iters_bufs)
+ for_each_possible_cpu(cpu)
+ kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
+
+ free_percpu(c->btree_iters_bufs);
free_percpu(c->pcpu);
mempool_exit(&c->large_bkey_pool);
mempool_exit(&c->btree_bounce_pool);
kfree(c->replicas_gc.entries);
kfree(rcu_dereference_protected(c->disk_groups, 1));
kfree(c->journal_seq_blacklist_table);
+ kfree(c->unused_inode_hints);
+ free_heap(&c->copygc_heap);
- if (c->journal_reclaim_wq)
- destroy_workqueue(c->journal_reclaim_wq);
if (c->copygc_wq)
destroy_workqueue(c->copygc_wq);
if (c->wq)
{
struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
- bch2_fs_free(c);
+ __bch2_fs_free(c);
}
-void bch2_fs_stop(struct bch_fs *c)
+void __bch2_fs_stop(struct bch_fs *c)
{
struct bch_dev *ca;
unsigned i;
kobject_put(&c->opts_dir);
kobject_put(&c->internal);
- mutex_lock(&bch_fs_list_lock);
- list_del(&c->list);
- mutex_unlock(&bch_fs_list_lock);
-
- closure_sync(&c->cl);
- closure_debug_destroy(&c->cl);
-
/* btree prefetch might have kicked off reads in the background: */
bch2_btree_flush_all_reads(c);
cancel_delayed_work_sync(&c->pd_controllers_update);
cancel_work_sync(&c->read_only_work);
+ for (i = 0; i < c->sb.nr_devices; i++)
+ if (c->devs[i])
+ bch2_free_super(&c->devs[i]->disk_sb);
+}
+
+void bch2_fs_free(struct bch_fs *c)
+{
+ unsigned i;
+
+ mutex_lock(&bch_fs_list_lock);
+ list_del(&c->list);
+ mutex_unlock(&bch_fs_list_lock);
+
+ closure_sync(&c->cl);
+ closure_debug_destroy(&c->cl);
+
for (i = 0; i < c->sb.nr_devices; i++)
if (c->devs[i])
bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
kobject_put(&c->kobj);
}
+void bch2_fs_stop(struct bch_fs *c)
+{
+ __bch2_fs_stop(c);
+ bch2_fs_free(c);
+}
+
static const char *bch2_fs_online(struct bch_fs *c)
{
struct bch_dev *ca;
__module_get(THIS_MODULE);
+ closure_init(&c->cl, NULL);
+
+ c->kobj.kset = bcachefs_kset;
+ kobject_init(&c->kobj, &bch2_fs_ktype);
+ kobject_init(&c->internal, &bch2_fs_internal_ktype);
+ kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
+ kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
+
c->minor = -1;
c->disk_sb.fs_sb = true;
for (i = 0; i < BCH_TIME_STAT_NR; i++)
bch2_time_stats_init(&c->times[i]);
+ bch2_fs_copygc_init(c);
bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
bch2_fs_allocator_background_init(c);
bch2_fs_allocator_foreground_init(c);
bch2_blacklist_entries_gc);
INIT_LIST_HEAD(&c->journal_entries);
+ INIT_LIST_HEAD(&c->journal_iters);
INIT_LIST_HEAD(&c->fsck_errors);
mutex_init(&c->fsck_error_lock);
- INIT_LIST_HEAD(&c->ec_new_stripe_list);
- mutex_init(&c->ec_new_stripe_lock);
- mutex_init(&c->ec_stripe_create_lock);
+ INIT_LIST_HEAD(&c->ec_stripe_head_list);
+ mutex_init(&c->ec_stripe_head_lock);
+
+ INIT_LIST_HEAD(&c->ec_stripe_new_list);
+ mutex_init(&c->ec_stripe_new_lock);
+
spin_lock_init(&c->ec_stripes_heap_lock);
seqcount_init(&c->gc_pos_lock);
bch2_fs_btree_cache_init_early(&c->btree_cache);
+ mutex_init(&c->sectors_available_lock);
+
if (percpu_init_rwsem(&c->mark_lock))
goto err;
(btree_blocks(c) + 1) * 2 *
sizeof(struct sort_iter_set);
+ c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
+
if (!(c->wq = alloc_workqueue("bcachefs",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
- !(c->copygc_wq = alloc_workqueue("bcache_copygc",
+ !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
- !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
- WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
percpu_ref_init(&c->writes, bch2_writes_disabled,
PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
offsetof(struct btree_write_bio, wbio.bio)),
BIOSET_NEED_BVECS) ||
!(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
+ !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
btree_bytes(c)) ||
mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
+ !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
+ sizeof(u64), GFP_KERNEL)) ||
bch2_io_clock_init(&c->io_clock[READ]) ||
bch2_io_clock_init(&c->io_clock[WRITE]) ||
bch2_fs_journal_init(&c->journal) ||
bch2_dev_alloc(c, i))
goto err;
- /*
- * Now that all allocations have succeeded, init various refcounty
- * things that let us shutdown:
- */
- closure_init(&c->cl, NULL);
-
- c->kobj.kset = bcachefs_kset;
- kobject_init(&c->kobj, &bch2_fs_ktype);
- kobject_init(&c->internal, &bch2_fs_internal_ktype);
- kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
- kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
+ bch2_journal_entry_res_resize(&c->journal,
+ &c->btree_root_journal_res,
+ BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
+ bch2_dev_usage_journal_reserve(c);
+ bch2_journal_entry_res_resize(&c->journal,
+ &c->clock_journal_res,
+ (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
mutex_lock(&bch_fs_list_lock);
err = bch2_fs_online(c);
set_bit(BCH_FS_STARTED, &c->flags);
+ /*
+ * Allocator threads don't start filling copygc reserve until after we
+ * set BCH_FS_STARTED - wake them now:
+ */
+ for_each_online_member(ca, c, i)
+ bch2_wake_allocator(ca);
+
if (c->opts.read_only || c->opts.nochanges) {
bch2_fs_read_only(c);
} else {
init_rwsem(&ca->bucket_lock);
- writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
-
- bch2_dev_copygc_init(ca);
-
INIT_WORK(&ca->io_error_work, bch2_io_error_work);
bch2_time_stats_init(&ca->io_latency[READ]);
if (ret)
return ret;
- if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
- !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
- mutex_lock(&c->sb_lock);
- bch2_mark_dev_superblock(ca->fs, ca, 0);
- mutex_unlock(&c->sb_lock);
- }
-
bch2_dev_sysfs_online(c, ca);
if (c->sb.nr_devices == 1)
enum bch_member_state new_state, int flags)
{
struct bch_devs_mask new_online_devs;
- struct replicas_status s;
struct bch_dev *ca2;
int i, nr_rw = 0, required;
new_online_devs = bch2_online_devs(c);
__clear_bit(ca->dev_idx, new_online_devs.d);
- s = __bch2_replicas_status(c, new_online_devs);
-
- return bch2_have_enough_devs(s, flags);
+ return bch2_have_enough_devs(c, new_online_devs, flags, false);
default:
BUG();
}
static bool bch2_fs_may_start(struct bch_fs *c)
{
- struct replicas_status s;
struct bch_sb_field_members *mi;
struct bch_dev *ca;
- unsigned i, flags = c->opts.degraded
- ? BCH_FORCE_IF_DEGRADED
- : 0;
+ unsigned i, flags = 0;
+
+ if (c->opts.very_degraded)
+ flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
+
+ if (c->opts.degraded)
+ flags |= BCH_FORCE_IF_DEGRADED;
- if (!c->opts.degraded) {
+ if (!c->opts.degraded &&
+ !c->opts.very_degraded) {
mutex_lock(&c->sb_lock);
mi = bch2_sb_get_members(c->disk_sb.sb);
mutex_unlock(&c->sb_lock);
}
- s = bch2_replicas_status(c);
-
- return bch2_have_enough_devs(s, flags);
+ return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
}
static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
{
- bch2_copygc_stop(ca);
+ /*
+ * Device going read only means the copygc reserve get smaller, so we
+ * don't want that happening while copygc is in progress:
+ */
+ bch2_copygc_stop(c);
/*
* The allocator thread itself allocates btree nodes, so stop it first:
bch2_dev_allocator_stop(ca);
bch2_dev_allocator_remove(c, ca);
bch2_dev_journal_stop(&c->journal, ca);
+
+ bch2_copygc_start(c);
}
static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
if (bch2_dev_allocator_start(ca))
return "error starting allocator thread";
- if (bch2_copygc_start(c, ca))
- return "error starting copygc thread";
-
return NULL;
}
mutex_unlock(&c->sb_lock);
up_write(&c->state_lock);
+
+ bch2_dev_usage_journal_reserve(c);
return 0;
err:
if (ca->mi.state == BCH_MEMBER_STATE_RW &&
return ret;
}
-static void dev_usage_clear(struct bch_dev *ca)
-{
- struct bucket_array *buckets;
-
- percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
-
- down_read(&ca->bucket_lock);
- buckets = bucket_array(ca);
-
- memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
- up_read(&ca->bucket_lock);
-}
-
/* Add new device to running filesystem: */
int bch2_dev_add(struct bch_fs *c, const char *path)
{
* allocate the journal, reset all the marks, then remark after we
* attach...
*/
- bch2_mark_dev_superblock(ca->fs, ca, 0);
+ bch2_mark_dev_superblock(NULL, ca, 0);
err = "journal alloc failed";
ret = bch2_dev_journal_alloc(ca);
if (ret)
goto err;
- dev_usage_clear(ca);
-
down_write(&c->state_lock);
mutex_lock(&c->sb_lock);
ca->disk_sb.sb->dev_idx = dev_idx;
bch2_dev_attach(c, ca, dev_idx);
- bch2_mark_dev_superblock(c, ca, 0);
-
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
+ bch2_dev_usage_journal_reserve(c);
+
+ err = "error marking superblock";
+ ret = bch2_trans_mark_dev_sb(c, NULL, ca);
+ if (ret)
+ goto err_late;
+
if (ca->mi.state == BCH_MEMBER_STATE_RW) {
err = __bch2_dev_read_write(c, ca);
if (err)
bch_err(c, "Unable to add device: %s", err);
return ret;
err_late:
+ up_write(&c->state_lock);
bch_err(c, "Error going rw after adding device: %s", err);
return -EINVAL;
}
}
ca = bch_dev_locked(c, dev_idx);
+
+ if (bch2_trans_mark_dev_sb(c, NULL, ca)) {
+ err = "bch2_trans_mark_dev_sb() error";
+ goto err;
+ }
+
if (ca->mi.state == BCH_MEMBER_STATE_RW) {
err = __bch2_dev_read_write(c, ca);
if (err)
/* return with ref on ca->ref: */
struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
{
-
struct block_device *bdev = lookup_bdev(path);
struct bch_dev *ca;
unsigned i;
{
struct bch_sb_handle *sb = NULL;
struct bch_fs *c = NULL;
+ struct bch_sb_field_members *mi;
unsigned i, best_sb = 0;
const char *err;
int ret = -ENOMEM;
le64_to_cpu(sb[best_sb].sb->seq))
best_sb = i;
- for (i = 0; i < nr_devices; i++) {
+ mi = bch2_sb_get_members(sb[best_sb].sb);
+
+ i = 0;
+ while (i < nr_devices) {
+ if (i != best_sb &&
+ !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
+ char buf[BDEVNAME_SIZE];
+ pr_info("%s has been removed, skipping",
+ bdevname(sb[i].bdev, buf));
+ bch2_free_super(&sb[i]);
+ array_remove_item(sb, nr_devices, i);
+ continue;
+ }
+
err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
if (err)
goto err_print;
+ i++;
}
ret = -ENOMEM;
bch2_debug_exit();
bch2_vfs_exit();
bch2_chardev_exit();
+ bch2_btree_key_cache_exit();
if (bcachefs_kset)
kset_unregister(bcachefs_kset);
}
static int __init bcachefs_init(void)
{
bch2_bkey_pack_test();
- bch2_inode_pack_test();
if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
+ bch2_btree_key_cache_init() ||
bch2_chardev_init() ||
bch2_vfs_init() ||
bch2_debug_init())
projects / bcachefs-tools-debian / blobdiff