#include "rebalance.h"
#include "recovery.h"
#include "replicas.h"
+#include "subvolume.h"
#include "super.h"
#include "super-io.h"
#include "sysfs.h"
static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
-struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
+struct bch_fs *bch2_dev_to_fs(dev_t dev)
{
struct bch_fs *c;
struct bch_dev *ca;
list_for_each_entry(c, &bch_fs_list, list)
for_each_member_device_rcu(ca, c, i, NULL)
- if (ca->disk_sb.bdev == bdev) {
+ if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
closure_get(&c->cl);
goto found;
}
return c;
}
+static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
+{
+ struct bch_dev *ca;
+ 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();
+ for_each_member_device_rcu(ca, c, i, NULL)
+ nr++;
+ rcu_read_unlock();
+
+ bch2_journal_entry_res_resize(&c->journal,
+ &c->dev_usage_journal_res, u64s * nr);
+}
+
/* Filesystem RO/RW: */
/*
bch2_copygc_stop(c);
bch2_gc_thread_stop(c);
- bch2_io_timer_del(&c->io_clock[READ], &c->bucket_clock[READ].rescale);
- bch2_io_timer_del(&c->io_clock[WRITE], &c->bucket_clock[WRITE].rescale);
-
/*
* Flush journal before stopping allocators, because flushing journal
* blacklist entries involves allocating new btree nodes:
void bch2_fs_read_only(struct bch_fs *c)
{
if (!test_bit(BCH_FS_RW, &c->flags)) {
- BUG_ON(c->journal.reclaim_thread);
+ bch2_journal_reclaim_stop(&c->journal);
return;
}
percpu_ref_kill(&c->writes);
cancel_work_sync(&c->ec_stripe_delete_work);
- cancel_delayed_work(&c->pd_controllers_update);
/*
* If we're not doing an emergency shutdown, we want to wait on
return ret;
}
- schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
-
schedule_work(&c->ec_stripe_delete_work);
return 0;
unsigned i;
int ret;
+ if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
+ bch_err(c, "cannot go rw, unfixed btree errors");
+ return -EROFS;
+ }
+
if (test_bit(BCH_FS_RW, &c->flags))
return 0;
(!early || c->opts.read_only)))
return -EROFS;
+ bch_info(c, "going read-write");
+
ret = bch2_fs_mark_dirty(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)
bch2_dev_allocator_add(c, ca);
bch2_recalc_capacity(c);
- bch2_io_timer_add(&c->io_clock[READ], &c->bucket_clock[READ].rescale);
- bch2_io_timer_add(&c->io_clock[WRITE], &c->bucket_clock[WRITE].rescale);
-
for_each_rw_member(ca, c, i) {
ret = bch2_dev_allocator_start(ca);
if (ret) {
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;
- }
+ for_each_rw_member(ca, c, i)
+ bch2_wake_allocator(ca);
if (!early) {
ret = bch2_fs_read_write_late(c);
percpu_ref_reinit(&c->writes);
set_bit(BCH_FS_RW, &c->flags);
+ set_bit(BCH_FS_WAS_RW, &c->flags);
return 0;
err:
__bch2_fs_read_only(c);
for (i = 0; i < BCH_TIME_STAT_NR; i++)
bch2_time_stats_exit(&c->times[i]);
+ bch2_fs_snapshots_exit(c);
bch2_fs_quota_exit(c);
bch2_fs_fsio_exit(c);
bch2_fs_ec_exit(c);
bch2_fs_btree_iter_exit(c);
bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
bch2_fs_btree_cache_exit(c);
+ bch2_fs_replicas_exit(c);
bch2_fs_journal_exit(&c->journal);
bch2_io_clock_exit(&c->io_clock[WRITE]);
bch2_io_clock_exit(&c->io_clock[READ]);
bch2_journal_keys_free(&c->journal_keys);
bch2_journal_entries_free(&c->journal_entries);
percpu_free_rwsem(&c->mark_lock);
- kfree(c->usage_scratch);
- for (i = 0; i < ARRAY_SIZE(c->usage); i++)
- free_percpu(c->usage[i]);
- kfree(c->usage_base);
- if (c->btree_iters_bufs)
+ if (c->btree_paths_bufs)
for_each_possible_cpu(cpu)
- kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
+ kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
- free_percpu(c->btree_iters_bufs);
+ free_percpu(c->online_reserved);
+ free_percpu(c->btree_paths_bufs);
free_percpu(c->pcpu);
mempool_exit(&c->large_bkey_pool);
mempool_exit(&c->btree_bounce_pool);
bioset_exit(&c->btree_bio);
mempool_exit(&c->fill_iter);
percpu_ref_exit(&c->writes);
- kfree(c->replicas.entries);
- 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->io_complete_wq )
+ destroy_workqueue(c->io_complete_wq );
if (c->copygc_wq)
destroy_workqueue(c->copygc_wq);
- if (c->wq)
- destroy_workqueue(c->wq);
+ if (c->btree_io_complete_wq)
+ destroy_workqueue(c->btree_io_complete_wq);
+ if (c->btree_update_wq)
+ destroy_workqueue(c->btree_update_wq);
- free_pages((unsigned long) c->disk_sb.sb,
- c->disk_sb.page_order);
+ bch2_free_super(&c->disk_sb);
kvpfree(c, sizeof(*c));
module_put(THIS_MODULE);
}
for_each_member_device(ca, c, i)
if (ca->kobj.state_in_sysfs &&
ca->disk_sb.bdev)
- sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
- "bcachefs");
+ sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
if (c->kobj.state_in_sysfs)
kobject_del(&c->kobj);
for_each_member_device(ca, c, i)
cancel_work_sync(&ca->io_error_work);
- cancel_work_sync(&c->btree_write_error_work);
- cancel_delayed_work_sync(&c->pd_controllers_update);
cancel_work_sync(&c->read_only_work);
for (i = 0; i < c->sb.nr_devices; i++)
bch2_fs_free(c);
}
-static const char *bch2_fs_online(struct bch_fs *c)
+static int bch2_fs_online(struct bch_fs *c)
{
struct bch_dev *ca;
- const char *err = NULL;
unsigned i;
- int ret;
+ int ret = 0;
lockdep_assert_held(&bch_fs_list_lock);
- if (!list_empty(&c->list))
- return NULL;
-
- if (__bch2_uuid_to_fs(c->sb.uuid))
- return "filesystem UUID already open";
+ if (__bch2_uuid_to_fs(c->sb.uuid)) {
+ bch_err(c, "filesystem UUID already open");
+ return -EINVAL;
+ }
ret = bch2_fs_chardev_init(c);
- if (ret)
- return "error creating character device";
+ if (ret) {
+ bch_err(c, "error creating character device");
+ return ret;
+ }
bch2_fs_debug_init(c);
- if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
- kobject_add(&c->internal, &c->kobj, "internal") ||
- kobject_add(&c->opts_dir, &c->kobj, "options") ||
- kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
- bch2_opts_create_sysfs_files(&c->opts_dir))
- return "error creating sysfs objects";
+ ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
+ kobject_add(&c->internal, &c->kobj, "internal") ?:
+ kobject_add(&c->opts_dir, &c->kobj, "options") ?:
+ kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
+ bch2_opts_create_sysfs_files(&c->opts_dir);
+ if (ret) {
+ bch_err(c, "error creating sysfs objects");
+ return ret;
+ }
down_write(&c->state_lock);
- err = "error creating sysfs objects";
- __for_each_member_device(ca, c, i, NULL)
- if (bch2_dev_sysfs_online(c, ca))
+ for_each_member_device(ca, c, i) {
+ ret = bch2_dev_sysfs_online(c, ca);
+ if (ret) {
+ bch_err(c, "error creating sysfs objects");
+ percpu_ref_put(&ca->ref);
goto err;
+ }
+ }
+ BUG_ON(!list_empty(&c->list));
list_add(&c->list, &bch_fs_list);
- err = NULL;
err:
up_write(&c->state_lock);
- return err;
+ return ret;
}
static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
struct bch_sb_field_members *mi;
struct bch_fs *c;
unsigned i, iter_size;
- const char *err;
+ int ret = 0;
pr_verbose_init(opts, "");
c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
- if (!c)
+ if (!c) {
+ c = ERR_PTR(-ENOMEM);
goto out;
+ }
__module_get(THIS_MODULE);
mutex_init(&c->usage_scratch_lock);
mutex_init(&c->bio_bounce_pages_lock);
+ mutex_init(&c->snapshot_table_lock);
- bio_list_init(&c->btree_write_error_list);
spin_lock_init(&c->btree_write_error_lock);
- INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
INIT_WORK(&c->journal_seq_blacklist_gc_work,
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_stripe_new_list);
mutex_init(&c->ec_stripe_new_lock);
+ INIT_LIST_HEAD(&c->data_progress_list);
+ mutex_init(&c->data_progress_lock);
+
spin_lock_init(&c->ec_stripes_heap_lock);
seqcount_init(&c->gc_pos_lock);
c->rebalance.enabled = 1;
c->promote_whole_extents = true;
- c->journal.write_time = &c->times[BCH_TIME_journal_write];
- c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
- c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
- c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
+ c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
+ c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
+ c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
+ c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
bch2_fs_btree_cache_init_early(&c->btree_cache);
mutex_init(&c->sectors_available_lock);
- if (percpu_init_rwsem(&c->mark_lock))
+ ret = percpu_init_rwsem(&c->mark_lock);
+ if (ret)
goto err;
mutex_lock(&c->sb_lock);
+ ret = bch2_sb_to_fs(c, sb);
+ mutex_unlock(&c->sb_lock);
- if (bch2_sb_to_fs(c, sb)) {
- mutex_unlock(&c->sb_lock);
+ if (ret)
goto err;
- }
-
- mutex_unlock(&c->sb_lock);
scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
+ /* Compat: */
+ if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+ SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
+
+ if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
+ SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
+
c->opts = bch2_opts_default;
bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
bch2_opts_apply(&c->opts, opts);
c->block_bits = ilog2(c->opts.block_size);
c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
- if (bch2_fs_init_fault("fs_alloc"))
+ if (bch2_fs_init_fault("fs_alloc")) {
+ bch_err(c, "fs_alloc fault injected");
+ ret = -EFAULT;
goto err;
+ }
iter_size = sizeof(struct sort_iter) +
(btree_blocks(c) + 1) * 2 *
c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
- if (!(c->wq = alloc_workqueue("bcachefs",
+ if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
+ WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+ !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
!(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+ !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
+ WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 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)) ||
+ !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
+ !(c->online_reserved = alloc_percpu(u64)) ||
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_fs_replicas_init(c) ||
- bch2_fs_btree_cache_init(c) ||
- bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
- bch2_fs_btree_iter_init(c) ||
- bch2_fs_btree_interior_update_init(c) ||
- bch2_fs_io_init(c) ||
- bch2_fs_encryption_init(c) ||
- bch2_fs_compress_init(c) ||
- bch2_fs_ec_init(c) ||
- bch2_fs_fsio_init(c))
+ sizeof(u64), GFP_KERNEL))) {
+ ret = -ENOMEM;
goto err;
+ }
+
+ ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
+ bch2_io_clock_init(&c->io_clock[WRITE]) ?:
+ bch2_fs_journal_init(&c->journal) ?:
+ bch2_fs_replicas_init(c) ?:
+ bch2_fs_btree_cache_init(c) ?:
+ bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
+ bch2_fs_btree_iter_init(c) ?:
+ bch2_fs_btree_interior_update_init(c) ?:
+ bch2_fs_subvolumes_init(c) ?:
+ bch2_fs_io_init(c) ?:
+ bch2_fs_encryption_init(c) ?:
+ bch2_fs_compress_init(c) ?:
+ bch2_fs_ec_init(c) ?:
+ bch2_fs_fsio_init(c);
+ if (ret)
+ goto err;
+
+ if (c->opts.nochanges)
+ set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
mi = bch2_sb_get_members(c->disk_sb.sb);
for (i = 0; i < c->sb.nr_devices; i++)
if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
- bch2_dev_alloc(c, i))
+ bch2_dev_alloc(c, i)) {
+ ret = -EEXIST;
goto err;
+ }
+
+ 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);
+ ret = bch2_fs_online(c);
mutex_unlock(&bch_fs_list_lock);
- if (err) {
- bch_err(c, "bch2_fs_online() error: %s", err);
+
+ if (ret)
goto err;
- }
out:
- pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
+ pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
return c;
err:
bch2_fs_free(c);
- c = NULL;
+ c = ERR_PTR(ret);
goto out;
}
int bch2_fs_start(struct bch_fs *c)
{
- const char *err = "cannot allocate memory";
struct bch_sb_field_members *mi;
struct bch_dev *ca;
time64_t now = ktime_get_real_seconds();
if (ret)
goto err;
- err = "dynamic fault";
ret = -EINVAL;
- if (bch2_fs_init_fault("fs_start"))
+ if (bch2_fs_init_fault("fs_start")) {
+ bch_err(c, "fs_start fault injected");
goto err;
+ }
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:
+ *
+ * XXX ugly hack:
+ * Need to set ca->allocator_state here instead of relying on the
+ * allocator threads to do it to avoid racing with the copygc threads
+ * checking it and thinking they have no alloc reserve:
*/
- for_each_online_member(ca, c, i)
+ for_each_online_member(ca, c, i) {
+ ca->allocator_state = ALLOCATOR_running;
bch2_wake_allocator(ca);
+ }
if (c->opts.read_only || c->opts.nochanges) {
bch2_fs_read_only(c);
} else {
- err = "error going read write";
ret = !test_bit(BCH_FS_RW, &c->flags)
? bch2_fs_read_write(c)
: bch2_fs_read_write_late(c);
case BCH_FSCK_ERRORS_NOT_FIXED:
bch_err(c, "filesystem contains errors: please report this to the developers");
pr_cont("mount with -o fix_errors to repair\n");
- err = "fsck error";
break;
case BCH_FSCK_REPAIR_UNIMPLEMENTED:
bch_err(c, "filesystem contains errors: please report this to the developers");
pr_cont("repair unimplemented: inform the developers so that it can be added\n");
- err = "fsck error";
break;
case BCH_FSCK_REPAIR_IMPOSSIBLE:
bch_err(c, "filesystem contains errors, but repair impossible");
- err = "fsck error";
break;
case BCH_FSCK_UNKNOWN_VERSION:
- err = "unknown metadata version";;
+ bch_err(c, "unknown metadata version");
break;
case -ENOMEM:
- err = "cannot allocate memory";
+ bch_err(c, "cannot allocate memory");
break;
case -EIO:
- err = "IO error";
+ bch_err(c, "IO error");
break;
}
static void bch2_dev_free(struct bch_dev *ca)
{
+ bch2_dev_allocator_stop(ca);
+
cancel_work_sync(&ca->io_error_work);
if (ca->kobj.state_in_sysfs &&
ca->disk_sb.bdev)
- sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
- "bcachefs");
+ sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
if (ca->kobj.state_in_sysfs)
kobject_del(&ca->kobj);
wait_for_completion(&ca->io_ref_completion);
if (ca->kobj.state_in_sysfs) {
- struct kobject *block =
- &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
-
- sysfs_remove_link(block, "bcachefs");
+ sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
sysfs_remove_link(&ca->kobj, "block");
}
}
if (ca->disk_sb.bdev) {
- struct kobject *block =
- &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
+ struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
if (ret)
return ret;
+
ret = sysfs_create_link(&ca->kobj, block, "block");
if (ret)
return ret;
if (!ca)
goto err;
+ ca->fs = c;
+
+ if (ca->mi.state == BCH_MEMBER_STATE_rw &&
+ bch2_dev_allocator_start(ca)) {
+ bch2_dev_free(ca);
+ goto err;
+ }
+
bch2_dev_attach(c, ca, dev_idx);
out:
pr_verbose_init(c->opts, "ret %i", ret);
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;
lockdep_assert_held(&c->state_lock);
switch (new_state) {
- case BCH_MEMBER_STATE_RW:
+ case BCH_MEMBER_STATE_rw:
return true;
- case BCH_MEMBER_STATE_RO:
- if (ca->mi.state != BCH_MEMBER_STATE_RW)
+ case BCH_MEMBER_STATE_ro:
+ if (ca->mi.state != BCH_MEMBER_STATE_rw)
return true;
/* do we have enough devices to write to? */
for_each_member_device(ca2, c, i)
if (ca2 != ca)
- nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
+ nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
? c->opts.metadata_replicas
: c->opts.data_replicas_required);
return nr_rw >= required;
- case BCH_MEMBER_STATE_FAILED:
- case BCH_MEMBER_STATE_SPARE:
- if (ca->mi.state != BCH_MEMBER_STATE_RW &&
- ca->mi.state != BCH_MEMBER_STATE_RO)
+ case BCH_MEMBER_STATE_failed:
+ case BCH_MEMBER_STATE_spare:
+ if (ca->mi.state != BCH_MEMBER_STATE_rw &&
+ ca->mi.state != BCH_MEMBER_STATE_ro)
return true;
/* do we have enough devices to read from? */
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);
ca = bch_dev_locked(c, i);
if (!bch2_dev_is_online(ca) &&
- (ca->mi.state == BCH_MEMBER_STATE_RW ||
- ca->mi.state == BCH_MEMBER_STATE_RO)) {
+ (ca->mi.state == BCH_MEMBER_STATE_rw ||
+ ca->mi.state == BCH_MEMBER_STATE_ro)) {
mutex_unlock(&c->sb_lock);
return false;
}
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_start(c);
}
-static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
+static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
{
lockdep_assert_held(&c->state_lock);
- BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
+ BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
bch2_dev_allocator_add(c, ca);
bch2_recalc_capacity(c);
- if (bch2_dev_allocator_start(ca))
- return "error starting allocator thread";
-
- return NULL;
+ return bch2_dev_allocator_start(ca);
}
int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
if (!bch2_dev_state_allowed(c, ca, new_state, flags))
return -EINVAL;
- if (new_state != BCH_MEMBER_STATE_RW)
+ if (new_state != BCH_MEMBER_STATE_rw)
__bch2_dev_read_only(c, ca);
- bch_notice(ca, "%s", bch2_dev_state[new_state]);
+ bch_notice(ca, "%s", bch2_member_states[new_state]);
mutex_lock(&c->sb_lock);
mi = bch2_sb_get_members(c->disk_sb.sb);
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- if (new_state == BCH_MEMBER_STATE_RW &&
- __bch2_dev_read_write(c, ca))
- ret = -ENOMEM;
+ if (new_state == BCH_MEMBER_STATE_rw)
+ ret = __bch2_dev_read_write(c, ca);
rebalance_wakeup(c);
/* Device add/removal: */
-int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
+static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
{
struct btree_trans trans;
size_t i;
bch2_trans_init(&trans, c, 0, 0);
for (i = 0; i < ca->mi.nbuckets; i++) {
- ret = bch2_btree_key_cache_flush(&trans,
- BTREE_ID_ALLOC, POS(ca->dev_idx, i));
+ ret = lockrestart_do(&trans,
+ bch2_btree_key_cache_flush(&trans,
+ BTREE_ID_alloc, POS(ca->dev_idx, i)));
if (ret)
break;
}
bch2_trans_exit(&trans);
- if (ret)
+ if (ret) {
+ bch_err(c, "error %i removing dev alloc info", ret);
return ret;
+ }
- return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
+ return bch2_btree_delete_range(c, BTREE_ID_alloc,
POS(ca->dev_idx, 0),
POS(ca->dev_idx + 1, 0),
NULL);
*/
percpu_ref_put(&ca->ref);
- if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
+ if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
bch_err(ca, "Cannot remove without losing data");
goto err;
}
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 &&
+ if (ca->mi.state == BCH_MEMBER_STATE_rw &&
!percpu_ref_is_zero(&ca->io_ref))
__bch2_dev_read_write(c, ca);
up_write(&c->state_lock);
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)
{
struct bch_dev *ca = NULL;
struct bch_sb_field_members *mi;
struct bch_member dev_mi;
+ struct bucket_array *buckets;
+ struct bucket *g;
unsigned dev_idx, nr_devices, u64s;
int ret;
if (ret)
goto err;
- dev_usage_clear(ca);
-
down_write(&c->state_lock);
mutex_lock(&c->sb_lock);
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
+ bch2_dev_usage_journal_reserve(c);
+
+ /*
+ * Clear marks before marking transactionally in the btree, so that
+ * per-device accounting gets done correctly:
+ */
+ down_read(&ca->bucket_lock);
+ buckets = bucket_array(ca);
+ for_each_bucket(g, buckets)
+ atomic64_set(&g->_mark.v, 0);
+ up_read(&ca->bucket_lock);
+
err = "error marking superblock";
- ret = bch2_trans_mark_dev_sb(c, NULL, ca);
+ ret = bch2_trans_mark_dev_sb(c, ca);
if (ret)
goto err_late;
- if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- err = __bch2_dev_read_write(c, ca);
- if (err)
+ if (ca->mi.state == BCH_MEMBER_STATE_rw) {
+ ret = __bch2_dev_read_write(c, ca);
+ if (ret)
goto err_late;
}
dev_idx = sb.sb->dev_idx;
err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
- if (err)
+ if (err) {
+ bch_err(c, "error bringing %s online: %s", path, err);
goto err;
+ }
- if (bch2_dev_attach_bdev(c, &sb)) {
- err = "bch2_dev_attach_bdev() error";
+ ret = bch2_dev_attach_bdev(c, &sb);
+ if (ret)
goto err;
- }
- if (bch2_trans_mark_dev_sb(c, NULL, ca)) {
- err = "bch2_trans_mark_dev_sb() error";
+ ca = bch_dev_locked(c, dev_idx);
+
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret) {
+ bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
+ path, ret);
goto err;
}
- ca = bch_dev_locked(c, dev_idx);
- if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- err = __bch2_dev_read_write(c, ca);
- if (err)
+ if (ca->mi.state == BCH_MEMBER_STATE_rw) {
+ ret = __bch2_dev_read_write(c, ca);
+ if (ret)
goto err;
}
err:
up_write(&c->state_lock);
bch2_free_super(&sb);
- bch_err(c, "error bringing %s online: %s", path, err);
return -EINVAL;
}
return 0;
}
- if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
+ if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
bch_err(ca, "Cannot offline required disk");
up_write(&c->state_lock);
return -EINVAL;
goto err;
}
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret) {
+ goto err;
+ }
+
mutex_lock(&c->sb_lock);
mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
mi->nbuckets = cpu_to_le64(nbuckets);
/* 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;
+ dev_t dev;
unsigned i;
+ int ret;
- if (IS_ERR(bdev))
- return ERR_CAST(bdev);
+ ret = lookup_bdev(path, &dev);
+ if (ret)
+ return ERR_PTR(ret);
- for_each_member_device(ca, c, i)
- if (ca->disk_sb.bdev == bdev)
+ rcu_read_lock();
+ for_each_member_device_rcu(ca, c, i, NULL)
+ if (ca->disk_sb.bdev->bd_dev == dev)
goto found;
-
ca = ERR_PTR(-ENOENT);
found:
- bdput(bdev);
+ rcu_read_unlock();
+
return ca;
}
struct bch_sb_field_members *mi;
unsigned i, best_sb = 0;
const char *err;
- int ret = -ENOMEM;
+ int ret = 0;
pr_verbose_init(opts, "");
}
sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
- if (!sb)
+ if (!sb) {
+ ret = -ENOMEM;
goto err;
+ }
for (i = 0; i < nr_devices; i++) {
ret = bch2_read_super(devices[i], &opts, &sb[i]);
i++;
}
- ret = -ENOMEM;
c = bch2_fs_alloc(sb[best_sb].sb, opts);
- if (!c)
+ if (IS_ERR(c)) {
+ ret = PTR_ERR(c);
goto err;
+ }
- err = "bch2_dev_online() error";
down_write(&c->state_lock);
- for (i = 0; i < nr_devices; i++)
- if (bch2_dev_attach_bdev(c, &sb[i])) {
+ for (i = 0; i < nr_devices; i++) {
+ ret = bch2_dev_attach_bdev(c, &sb[i]);
+ if (ret) {
up_write(&c->state_lock);
- goto err_print;
+ goto err;
}
+ }
up_write(&c->state_lock);
err = "insufficient devices";
devices[0], err);
ret = -EINVAL;
err:
- if (c)
+ if (!IS_ERR_OR_NULL(c))
bch2_fs_stop(c);
- for (i = 0; i < nr_devices; i++)
- bch2_free_super(&sb[i]);
+ if (sb)
+ for (i = 0; i < nr_devices; i++)
+ bch2_free_super(&sb[i]);
c = ERR_PTR(ret);
goto out;
}
if (err)
goto err;
} else {
+ allocated_fs = true;
c = bch2_fs_alloc(sb->sb, opts);
- err = "cannot allocate memory";
- if (!c)
- goto err;
- allocated_fs = true;
+ err = "bch2_fs_alloc() error";
+ if (IS_ERR(c))
+ goto err;
}
err = "bch2_dev_online() error";
err:
mutex_unlock(&bch_fs_list_lock);
- if (allocated_fs)
+ if (allocated_fs && !IS_ERR(c))
bch2_fs_stop(c);
else if (c)
closure_put(&c->cl);
projects / bcachefs-tools-debian / blobdiff