+// SPDX-License-Identifier: GPL-2.0
/*
* bcachefs setup/teardown code, and some metadata io - read a superblock and
* figure out what to do with it.
*/
#include "bcachefs.h"
-#include "alloc.h"
+#include "alloc_background.h"
+#include "alloc_foreground.h"
+#include "bkey_sort.h"
#include "btree_cache.h"
#include "btree_gc.h"
-#include "btree_update.h"
+#include "btree_key_cache.h"
#include "btree_update_interior.h"
#include "btree_io.h"
#include "chardev.h"
#include "clock.h"
#include "compress.h"
#include "debug.h"
+#include "disk_groups.h"
+#include "ec.h"
#include "error.h"
#include "fs.h"
+#include "fs-io.h"
#include "fsck.h"
#include "inode.h"
#include "io.h"
#include "journal.h"
-#include "keylist.h"
+#include "journal_reclaim.h"
+#include "journal_seq_blacklist.h"
#include "move.h"
#include "migrate.h"
#include "movinggc.h"
+#include "quota.h"
+#include "rebalance.h"
+#include "recovery.h"
+#include "replicas.h"
#include "super.h"
#include "super-io.h"
#include "sysfs.h"
-#include "tier.h"
#include <linux/backing-dev.h>
#include <linux/blkdev.h>
static void bch2_dev_free(struct bch_dev *);
static int bch2_dev_alloc(struct bch_fs *, unsigned);
-static int bch2_dev_sysfs_online(struct bch_dev *);
+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)
lockdep_assert_held(&bch_fs_list_lock);
list_for_each_entry(c, &bch_fs_list, list)
- if (!memcmp(&c->disk_sb->uuid, &uuid, sizeof(uuid_le)))
+ if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
return c;
return NULL;
return c;
}
-int bch2_congested(struct bch_fs *c, int bdi_bits)
+int bch2_congested(void *data, int bdi_bits)
{
+ struct bch_fs *c = data;
struct backing_dev_info *bdi;
struct bch_dev *ca;
unsigned i;
int ret = 0;
+ rcu_read_lock();
if (bdi_bits & (1 << WB_sync_congested)) {
/* Reads - check all devices: */
for_each_readable_member(ca, c, i) {
}
}
} else {
- /* Writes prefer fastest tier: */
- struct bch_tier *tier = READ_ONCE(c->fastest_tier);
- struct bch_devs_mask *devs =
- tier ? &tier->devs : &c->rw_devs[BCH_DATA_USER];
+ 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];
- rcu_read_lock();
for_each_member_device_rcu(ca, c, i, devs) {
bdi = ca->disk_sb.bdev->bd_bdi;
break;
}
}
- rcu_read_unlock();
}
+ rcu_read_unlock();
return ret;
}
-static int bch2_congested_fn(void *data, int bdi_bits)
-{
- struct bch_fs *c = data;
-
- return bch2_congested(c, bdi_bits);
-}
-
/* Filesystem RO/RW: */
/*
* For startup/shutdown of RW stuff, the dependencies are:
*
- * - foreground writes depend on copygc and tiering (to free up space)
+ * - foreground writes depend on copygc and rebalance (to free up space)
*
- * - copygc and tiering depend on mark and sweep gc (they actually probably
+ * - copygc and rebalance depend on mark and sweep gc (they actually probably
* don't because they either reserve ahead of time or don't block if
* allocations fail, but allocations can require mark and sweep gc to run
* because of generation number wraparound)
static void __bch2_fs_read_only(struct bch_fs *c)
{
struct bch_dev *ca;
- unsigned i;
+ bool wrote = false;
+ unsigned i, clean_passes = 0;
+ int ret;
- bch2_tiering_stop(c);
+ bch2_rebalance_stop(c);
for_each_member_device(ca, c, i)
- bch2_moving_gc_stop(ca);
+ bch2_copygc_stop(ca);
bch2_gc_thread_stop(c);
* Flush journal before stopping allocators, because flushing journal
* blacklist entries involves allocating new btree nodes:
*/
- bch2_journal_flush_pins(&c->journal, U64_MAX);
+ bch2_journal_flush_all_pins(&c->journal);
- if (!bch2_journal_error(&c->journal))
- bch2_btree_verify_flushed(c);
+ /*
+ * If the allocator threads didn't all start up, the btree updates to
+ * write out alloc info aren't going to work:
+ */
+ 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);
+ set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
+
+ do {
+ clean_passes++;
+
+ if (bch2_journal_flush_all_pins(&c->journal))
+ clean_passes = 0;
+
+ /*
+ * In flight interior btree updates will generate more journal
+ * updates and btree updates (alloc btree):
+ */
+ if (bch2_btree_interior_updates_nr_pending(c)) {
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_nr_pending(c));
+ clean_passes = 0;
+ }
+ flush_work(&c->btree_interior_update_work);
+
+ if (bch2_journal_flush_all_pins(&c->journal))
+ clean_passes = 0;
+ } while (clean_passes < 2);
+ bch_verbose(c, "flushing journal and stopping allocators complete");
+
+ set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
+nowrote_alloc:
+ closure_wait_event(&c->btree_interior_update_wait,
+ !bch2_btree_interior_updates_nr_pending(c));
+ flush_work(&c->btree_interior_update_work);
for_each_member_device(ca, c, i)
bch2_dev_allocator_stop(ca);
+ clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
+ clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
+
bch2_fs_journal_stop(&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);
+
+ /*
+ * After stopping journal:
+ */
for_each_member_device(ca, c, i)
bch2_dev_allocator_remove(c, ca);
}
void bch2_fs_read_only(struct bch_fs *c)
{
- if (c->state != BCH_FS_STARTING &&
- c->state != BCH_FS_RW)
+ if (!test_bit(BCH_FS_RW, &c->flags)) {
+ cancel_delayed_work_sync(&c->journal.reclaim_work);
return;
+ }
- if (test_bit(BCH_FS_ERROR, &c->flags))
- return;
+ BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
/*
* Block new foreground-end write operations from starting - any new
*/
percpu_ref_kill(&c->writes);
- del_timer(&c->foreground_write_wakeup);
+ cancel_work_sync(&c->ec_stripe_delete_work);
cancel_delayed_work(&c->pd_controllers_update);
- c->foreground_write_pd.rate.rate = UINT_MAX;
- bch2_wake_delayed_writes((unsigned long) c);
-
/*
* If we're not doing an emergency shutdown, we want to wait on
* outstanding writes to complete so they don't see spurious errors due
clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
if (!bch2_journal_error(&c->journal) &&
- !test_bit(BCH_FS_ERROR, &c->flags)) {
- mutex_lock(&c->sb_lock);
- SET_BCH_SB_CLEAN(c->disk_sb, true);
- bch2_write_super(c);
- mutex_unlock(&c->sb_lock);
+ !test_bit(BCH_FS_ERROR, &c->flags) &&
+ !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
+ test_bit(BCH_FS_STARTED, &c->flags) &&
+ test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
+ !c->opts.norecovery) {
+ bch_verbose(c, "marking filesystem clean");
+ bch2_fs_mark_clean(c);
}
- c->state = BCH_FS_RO;
+ clear_bit(BCH_FS_RW, &c->flags);
}
static void bch2_fs_read_only_work(struct work_struct *work)
struct bch_fs *c =
container_of(work, struct bch_fs, read_only_work);
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
bch2_fs_read_only(c);
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
}
static void bch2_fs_read_only_async(struct bch_fs *c)
return ret;
}
-const char *bch2_fs_read_write(struct bch_fs *c)
+static int bch2_fs_read_write_late(struct bch_fs *c)
{
struct bch_dev *ca;
- const char *err = NULL;
unsigned i;
+ int ret;
- if (c->state != BCH_FS_STARTING &&
- c->state != BCH_FS_RO)
- return NULL;
-
- for_each_rw_member(ca, c, i)
- bch2_dev_allocator_add(c, ca);
- bch2_recalc_capacity(c);
+ ret = bch2_gc_thread_start(c);
+ if (ret) {
+ bch_err(c, "error starting gc thread");
+ return ret;
+ }
- err = "error starting allocator thread";
- for_each_rw_member(ca, c, i)
- if (bch2_dev_allocator_start(ca)) {
+ 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);
- goto err;
+ return ret;
}
+ }
+
+ ret = bch2_rebalance_start(c);
+ if (ret) {
+ bch_err(c, "error starting rebalance thread");
+ return ret;
+ }
+
+ schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
- err = "error starting btree GC thread";
- if (bch2_gc_thread_start(c))
+ schedule_work(&c->ec_stripe_delete_work);
+
+ return 0;
+}
+
+static int __bch2_fs_read_write(struct bch_fs *c, bool early)
+{
+ struct bch_dev *ca;
+ unsigned i;
+ int ret;
+
+ if (test_bit(BCH_FS_RW, &c->flags))
+ return 0;
+
+ /*
+ * nochanges is used for fsck -n mode - we have to allow going rw
+ * during recovery for that to work:
+ */
+ if (c->opts.norecovery ||
+ (c->opts.nochanges &&
+ (!early || c->opts.read_only)))
+ return -EROFS;
+
+ ret = bch2_fs_mark_dirty(c);
+ if (ret)
goto err;
- err = "error starting moving GC thread";
+ clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
+
for_each_rw_member(ca, c, i)
- if (bch2_moving_gc_start(ca)) {
+ bch2_dev_allocator_add(c, ca);
+ bch2_recalc_capacity(c);
+
+ for_each_rw_member(ca, c, i) {
+ ret = bch2_dev_allocator_start(ca);
+ if (ret) {
+ bch_err(c, "error starting allocator threads");
percpu_ref_put(&ca->io_ref);
goto err;
}
+ }
- err = "error starting tiering thread";
- if (bch2_tiering_start(c))
- goto err;
+ set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
- schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
+ if (!early) {
+ ret = bch2_fs_read_write_late(c);
+ if (ret)
+ goto err;
+ }
- if (c->state != BCH_FS_STARTING)
- percpu_ref_reinit(&c->writes);
+ percpu_ref_reinit(&c->writes);
+ set_bit(BCH_FS_RW, &c->flags);
- c->state = BCH_FS_RW;
- return NULL;
+ queue_delayed_work(c->journal_reclaim_wq,
+ &c->journal.reclaim_work, 0);
+ return 0;
err:
__bch2_fs_read_only(c);
- return err;
+ return ret;
+}
+
+int bch2_fs_read_write(struct bch_fs *c)
+{
+ return __bch2_fs_read_write(c, false);
+}
+
+int bch2_fs_read_write_early(struct bch_fs *c)
+{
+ lockdep_assert_held(&c->state_lock);
+
+ return __bch2_fs_read_write(c, true);
}
/* Filesystem startup/shutdown: */
static void bch2_fs_free(struct bch_fs *c)
{
+ unsigned i;
+
+ for (i = 0; i < BCH_TIME_STAT_NR; i++)
+ bch2_time_stats_exit(&c->times[i]);
+
+ bch2_fs_quota_exit(c);
+ bch2_fs_fsio_exit(c);
+ bch2_fs_ec_exit(c);
bch2_fs_encryption_exit(c);
- bch2_fs_btree_exit(c);
+ bch2_fs_io_exit(c);
+ bch2_fs_btree_interior_update_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_journal_exit(&c->journal);
bch2_io_clock_exit(&c->io_clock[WRITE]);
bch2_io_clock_exit(&c->io_clock[READ]);
bch2_fs_compress_exit(c);
- if (c->bdi.bdi_list.next)
- bdi_destroy(&c->bdi);
- lg_lock_free(&c->usage_lock);
- free_percpu(c->usage_percpu);
+ bch2_journal_keys_free(&c->journal_keys);
+ 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]);
+ kfree(c->usage_base);
+ free_percpu(c->pcpu);
+ mempool_exit(&c->large_bkey_pool);
mempool_exit(&c->btree_bounce_pool);
- mempool_exit(&c->bio_bounce_pages);
- bioset_exit(&c->bio_write);
- bioset_exit(&c->bio_read_split);
- bioset_exit(&c->bio_read);
- bioset_exit(&c->btree_read_bio);
- mempool_exit(&c->btree_interior_update_pool);
- mempool_exit(&c->btree_reserve_pool);
+ bioset_exit(&c->btree_bio);
mempool_exit(&c->fill_iter);
percpu_ref_exit(&c->writes);
- kfree(c->replicas);
+ kfree(c->replicas.entries);
+ kfree(c->replicas_gc.entries);
+ kfree(rcu_dereference_protected(c->disk_groups, 1));
+ kfree(c->journal_seq_blacklist_table);
+ if (c->journal_reclaim_wq)
+ destroy_workqueue(c->journal_reclaim_wq);
if (c->copygc_wq)
destroy_workqueue(c->copygc_wq);
if (c->wq)
destroy_workqueue(c->wq);
- free_pages((unsigned long) c->disk_sb, c->disk_sb_order);
- kfree(c);
+ free_pages((unsigned long) c->disk_sb.sb,
+ c->disk_sb.page_order);
+ kvpfree(c, sizeof(*c));
module_put(THIS_MODULE);
}
-static void bch2_fs_exit(struct bch_fs *c)
+static void bch2_fs_release(struct kobject *kobj)
{
- unsigned i;
-
- del_timer_sync(&c->foreground_write_wakeup);
- 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_dev_free(c->devs[i]);
+ struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
- closure_debug_destroy(&c->cl);
- kobject_put(&c->kobj);
+ bch2_fs_free(c);
}
-static void bch2_fs_offline(struct bch_fs *c)
+void bch2_fs_stop(struct bch_fs *c)
{
struct bch_dev *ca;
unsigned i;
- mutex_lock(&bch_fs_list_lock);
- list_del(&c->list);
- mutex_unlock(&bch_fs_list_lock);
+ bch_verbose(c, "shutting down");
+
+ set_bit(BCH_FS_STOPPING, &c->flags);
+
+ cancel_work_sync(&c->journal_seq_blacklist_gc_work);
+
+ down_write(&c->state_lock);
+ bch2_fs_read_only(c);
+ up_write(&c->state_lock);
for_each_member_device(ca, c, i)
if (ca->kobj.state_in_sysfs &&
kobject_put(&c->opts_dir);
kobject_put(&c->internal);
- mutex_lock(&c->state_lock);
- __bch2_fs_read_only(c);
- mutex_unlock(&c->state_lock);
-}
+ mutex_lock(&bch_fs_list_lock);
+ list_del(&c->list);
+ mutex_unlock(&bch_fs_list_lock);
-static void bch2_fs_release(struct kobject *kobj)
-{
- struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
+ closure_sync(&c->cl);
+ closure_debug_destroy(&c->cl);
- bch2_fs_free(c);
+ /* btree prefetch might have kicked off reads in the background: */
+ bch2_btree_flush_all_reads(c);
+
+ 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++)
+ if (c->devs[i])
+ bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
+
+ bch_verbose(c, "shutdown complete");
+
+ kobject_put(&c->kobj);
}
-void bch2_fs_stop(struct bch_fs *c)
+static const char *bch2_fs_online(struct bch_fs *c)
{
- mutex_lock(&c->state_lock);
- BUG_ON(c->state == BCH_FS_STOPPING);
- c->state = BCH_FS_STOPPING;
- mutex_unlock(&c->state_lock);
+ struct bch_dev *ca;
+ const char *err = NULL;
+ unsigned i;
+ int ret;
- bch2_fs_offline(c);
+ lockdep_assert_held(&bch_fs_list_lock);
- closure_sync(&c->cl);
+ if (!list_empty(&c->list))
+ return NULL;
- bch2_fs_exit(c);
+ if (__bch2_uuid_to_fs(c->sb.uuid))
+ return "filesystem UUID already open";
+
+ ret = bch2_fs_chardev_init(c);
+ if (ret)
+ return "error creating character device";
+
+ 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";
+
+ 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))
+ goto err;
+
+ list_add(&c->list, &bch_fs_list);
+ err = NULL;
+err:
+ up_write(&c->state_lock);
+ return err;
}
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;
+
+ pr_verbose_init(opts, "");
- c = kzalloc(sizeof(struct bch_fs), GFP_KERNEL);
+ c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
if (!c)
- return NULL;
+ goto out;
__module_get(THIS_MODULE);
c->minor = -1;
+ c->disk_sb.fs_sb = true;
- mutex_init(&c->state_lock);
+ init_rwsem(&c->state_lock);
mutex_init(&c->sb_lock);
mutex_init(&c->replicas_gc_lock);
- mutex_init(&c->btree_cache_lock);
- mutex_init(&c->bucket_lock);
mutex_init(&c->btree_root_lock);
INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
init_rwsem(&c->gc_lock);
-#define BCH_TIME_STAT(name, frequency_units, duration_units) \
- spin_lock_init(&c->name##_time.lock);
- BCH_TIME_STATS()
-#undef BCH_TIME_STAT
+ for (i = 0; i < BCH_TIME_STAT_NR; i++)
+ bch2_time_stats_init(&c->times[i]);
- bch2_fs_allocator_init(c);
- bch2_fs_tiering_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_fs_rebalance_init(c);
+ bch2_fs_quota_init(c);
INIT_LIST_HEAD(&c->list);
- INIT_LIST_HEAD(&c->btree_cache);
- INIT_LIST_HEAD(&c->btree_cache_freeable);
- INIT_LIST_HEAD(&c->btree_cache_freed);
- INIT_LIST_HEAD(&c->btree_interior_update_list);
- mutex_init(&c->btree_reserve_cache_lock);
- mutex_init(&c->btree_interior_update_lock);
+ mutex_init(&c->usage_scratch_lock);
mutex_init(&c->bio_bounce_pages_lock);
- mutex_init(&c->zlib_workspace_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->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);
+ spin_lock_init(&c->ec_stripes_heap_lock);
+
seqcount_init(&c->gc_pos_lock);
- c->prio_clock[READ].hand = 1;
- c->prio_clock[READ].min_prio = 0;
- c->prio_clock[WRITE].hand = 1;
- c->prio_clock[WRITE].min_prio = 0;
+ seqcount_init(&c->usage_lock);
+
+ sema_init(&c->io_in_flight, 64);
- init_waitqueue_head(&c->writeback_wait);
- c->writeback_pages_max = (256 << 10) / PAGE_SIZE;
+ c->copy_gc_enabled = 1;
+ c->rebalance.enabled = 1;
+ c->promote_whole_extents = true;
- c->copy_gc_enabled = 1;
- c->tiering_enabled = 1;
- c->tiering_percent = 10;
+ 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->foreground_target_percent = 20;
+ bch2_fs_btree_cache_init_early(&c->btree_cache);
- c->journal.write_time = &c->journal_write_time;
- c->journal.delay_time = &c->journal_delay_time;
- c->journal.blocked_time = &c->journal_blocked_time;
- c->journal.flush_seq_time = &c->journal_flush_seq_time;
+ if (percpu_init_rwsem(&c->mark_lock))
+ goto err;
mutex_lock(&c->sb_lock);
bch2_opts_apply(&c->opts, opts);
c->block_bits = ilog2(c->opts.block_size);
-
- c->opts.nochanges |= c->opts.noreplay;
- c->opts.read_only |= c->opts.nochanges;
+ c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
if (bch2_fs_init_fault("fs_alloc"))
goto err;
- iter_size = (btree_blocks(c) + 1) * 2 *
- sizeof(struct btree_node_iter_set);
+ iter_size = sizeof(struct sort_iter) +
+ (btree_blocks(c) + 1) * 2 *
+ sizeof(struct sort_iter_set);
if (!(c->wq = alloc_workqueue("bcachefs",
- WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
+ WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
!(c->copygc_wq = alloc_workqueue("bcache_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, 0, GFP_KERNEL) ||
- mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
- sizeof(struct btree_reserve)) ||
- mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
- sizeof(struct btree_update)) ||
+ percpu_ref_init(&c->writes, bch2_writes_disabled,
+ PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
- bioset_init(&c->btree_read_bio, 1,
- offsetof(struct btree_read_bio, bio)) ||
- bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio)) ||
- bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio)) ||
- bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio)) ||
- mempool_init_page_pool(&c->bio_bounce_pages,
- max_t(unsigned,
- c->opts.btree_node_size,
- c->sb.encoded_extent_max) /
- PAGE_SECTORS, 0) ||
- !(c->usage_percpu = alloc_percpu(struct bch_fs_usage)) ||
- lg_lock_init(&c->usage_lock) ||
- mempool_init_vp_pool(&c->btree_bounce_pool, 1, btree_bytes(c)) ||
- bdi_setup_and_register(&c->bdi, "bcachefs") ||
+ bioset_init(&c->btree_bio, 1,
+ max(offsetof(struct btree_read_bio, bio),
+ offsetof(struct btree_write_bio, wbio.bio)),
+ BIOSET_NEED_BVECS) ||
+ !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
+ mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
+ btree_bytes(c)) ||
+ mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
bch2_io_clock_init(&c->io_clock[READ]) ||
bch2_io_clock_init(&c->io_clock[WRITE]) ||
bch2_fs_journal_init(&c->journal) ||
- bch2_fs_btree_init(c) ||
+ 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_check_set_has_compressed_data(c, c->opts.compression))
+ bch2_fs_ec_init(c) ||
+ bch2_fs_fsio_init(c))
goto err;
- c->bdi.ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
- c->bdi.congested_fn = bch2_congested_fn;
- c->bdi.congested_data = c;
-
- mi = bch2_sb_get_members(c->disk_sb);
+ 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, mi, i) &&
+ if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
bch2_dev_alloc(c, i))
goto err;
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);
+
+ mutex_lock(&bch_fs_list_lock);
+ err = bch2_fs_online(c);
+ mutex_unlock(&bch_fs_list_lock);
+ if (err) {
+ bch_err(c, "bch2_fs_online() error: %s", err);
+ goto err;
+ }
+out:
+ pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
return c;
err:
bch2_fs_free(c);
- return NULL;
+ c = NULL;
+ goto out;
}
-static const char *__bch2_fs_online(struct bch_fs *c)
+noinline_for_stack
+static void print_mount_opts(struct bch_fs *c)
{
- struct bch_dev *ca;
- const char *err = NULL;
- unsigned i;
- int ret;
-
- lockdep_assert_held(&bch_fs_list_lock);
+ enum bch_opt_id i;
+ char buf[512];
+ struct printbuf p = PBUF(buf);
+ bool first = true;
- if (!list_empty(&c->list))
- return NULL;
+ strcpy(buf, "(null)");
- if (__bch2_uuid_to_fs(c->sb.uuid))
- return "filesystem UUID already open";
-
- ret = bch2_fs_chardev_init(c);
- if (ret)
- return "error creating character device";
-
- 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";
-
- mutex_lock(&c->state_lock);
+ if (c->opts.read_only) {
+ pr_buf(&p, "ro");
+ first = false;
+ }
- err = "error creating sysfs objects";
- __for_each_member_device(ca, c, i, NULL)
- if (bch2_dev_sysfs_online(ca))
- goto err;
+ for (i = 0; i < bch2_opts_nr; i++) {
+ const struct bch_option *opt = &bch2_opt_table[i];
+ u64 v = bch2_opt_get_by_id(&c->opts, i);
- list_add(&c->list, &bch_fs_list);
- err = NULL;
-err:
- mutex_unlock(&c->state_lock);
- return err;
-}
+ if (!(opt->mode & OPT_MOUNT))
+ continue;
-static const char *bch2_fs_online(struct bch_fs *c)
-{
- const char *err;
+ if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
+ continue;
- mutex_lock(&bch_fs_list_lock);
- err = __bch2_fs_online(c);
- mutex_unlock(&bch_fs_list_lock);
+ if (!first)
+ pr_buf(&p, ",");
+ first = false;
+ bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
+ }
- return err;
+ bch_info(c, "mounted with opts: %s", buf);
}
-static const char *__bch2_fs_start(struct bch_fs *c)
+int bch2_fs_start(struct bch_fs *c)
{
const char *err = "cannot allocate memory";
struct bch_sb_field_members *mi;
struct bch_dev *ca;
- LIST_HEAD(journal);
- struct jset *j;
- struct closure cl;
- time64_t now;
+ time64_t now = ktime_get_real_seconds();
unsigned i;
int ret = -EINVAL;
- closure_init_stack(&cl);
+ down_write(&c->state_lock);
- mutex_lock(&c->state_lock);
-
- BUG_ON(c->state != BCH_FS_STARTING);
+ BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
mutex_lock(&c->sb_lock);
+
for_each_online_member(ca, c, i)
bch2_sb_from_fs(c, ca);
+
+ mi = bch2_sb_get_members(c->disk_sb.sb);
+ for_each_online_member(ca, c, i)
+ mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
+
mutex_unlock(&c->sb_lock);
for_each_rw_member(ca, c, i)
bch2_dev_allocator_add(c, ca);
bch2_recalc_capacity(c);
- if (BCH_SB_INITIALIZED(c->disk_sb)) {
- ret = bch2_journal_read(c, &journal);
- if (ret)
- goto err;
-
- j = &list_entry(journal.prev, struct journal_replay, list)->j;
-
- c->prio_clock[READ].hand = le16_to_cpu(j->read_clock);
- c->prio_clock[WRITE].hand = le16_to_cpu(j->write_clock);
-
- for (i = 0; i < BTREE_ID_NR; i++) {
- unsigned level;
- struct bkey_i *k;
-
- err = "missing btree root";
- k = bch2_journal_find_btree_root(c, j, i, &level);
- if (!k && i < BTREE_ID_ALLOC)
- goto err;
-
- if (!k)
- continue;
-
- err = "error reading btree root";
- if (bch2_btree_root_read(c, i, k, level))
- goto err;
- }
-
- err = "error reading allocation information";
- ret = bch2_alloc_read(c, &journal);
- if (ret)
- goto err;
-
- set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
-
- bch_verbose(c, "starting mark and sweep:");
- err = "error in recovery";
- ret = bch2_initial_gc(c, &journal);
- if (ret)
- goto err;
- bch_verbose(c, "mark and sweep done");
-
- if (c->opts.noreplay)
- goto recovery_done;
-
- err = "cannot allocate new btree root";
- for (i = 0; i < BTREE_ID_NR; i++)
- if (!c->btree_roots[i].b &&
- bch2_btree_root_alloc(c, i, &cl))
- goto err;
-
- closure_sync(&cl);
-
- /*
- * bch2_journal_start() can't happen sooner, or btree_gc_finish()
- * will give spurious errors about oldest_gen > bucket_gen -
- * this is a hack but oh well.
- */
- bch2_journal_start(c);
-
- err = "error starting allocator thread";
- for_each_rw_member(ca, c, i)
- if (bch2_dev_allocator_start(ca)) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
-
- bch_verbose(c, "starting journal replay:");
- err = "journal replay failed";
- ret = bch2_journal_replay(c, &journal);
- if (ret)
- goto err;
- bch_verbose(c, "journal replay done");
-
- if (c->opts.norecovery)
- goto recovery_done;
-
- bch_verbose(c, "starting fsck:");
- err = "error in fsck";
- ret = bch2_fsck(c, !c->opts.nofsck);
- if (ret)
- goto err;
- bch_verbose(c, "fsck done");
- } else {
- struct bch_inode_unpacked inode;
- struct bkey_inode_buf packed_inode;
-
- bch_notice(c, "initializing new filesystem");
-
- set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
-
- ret = bch2_initial_gc(c, &journal);
- if (ret)
- goto err;
-
- err = "unable to allocate journal buckets";
- for_each_rw_member(ca, c, i)
- if (bch2_dev_journal_alloc(ca)) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
-
- err = "cannot allocate new btree root";
- for (i = 0; i < BTREE_ID_NR; i++)
- if (bch2_btree_root_alloc(c, i, &cl))
- goto err;
-
- /*
- * journal_res_get() will crash if called before this has
- * set up the journal.pin FIFO and journal.cur pointer:
- */
- bch2_journal_start(c);
- bch2_journal_set_replay_done(&c->journal);
-
- err = "error starting allocator thread";
- for_each_rw_member(ca, c, i)
- if (bch2_dev_allocator_start(ca)) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
-
- /* Wait for new btree roots to be written: */
- closure_sync(&cl);
-
- bch2_inode_init(c, &inode, 0, 0,
- S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0);
- inode.bi_inum = BCACHEFS_ROOT_INO;
-
- bch2_inode_pack(&packed_inode, &inode);
+ ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
+ ? bch2_fs_recovery(c)
+ : bch2_fs_initialize(c);
+ if (ret)
+ goto err;
- err = "error creating root directory";
- if (bch2_btree_insert(c, BTREE_ID_INODES,
- &packed_inode.inode.k_i,
- NULL, NULL, NULL, 0))
- goto err;
+ ret = bch2_opts_check_may_set(c);
+ if (ret)
+ goto err;
- err = "error writing first journal entry";
- if (bch2_journal_meta(&c->journal))
- goto err;
- }
-recovery_done:
err = "dynamic fault";
+ ret = -EINVAL;
if (bch2_fs_init_fault("fs_start"))
goto err;
- if (c->opts.read_only) {
+ set_bit(BCH_FS_STARTED, &c->flags);
+
+ if (c->opts.read_only || c->opts.nochanges) {
bch2_fs_read_only(c);
} else {
- err = bch2_fs_read_write(c);
- if (err)
+ err = "error going read write";
+ ret = !test_bit(BCH_FS_RW, &c->flags)
+ ? bch2_fs_read_write(c)
+ : bch2_fs_read_write_late(c);
+ if (ret)
goto err;
}
- mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
- now = ktime_get_seconds();
-
- for_each_member_device(ca, c, i)
- mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
-
- SET_BCH_SB_INITIALIZED(c->disk_sb, true);
- SET_BCH_SB_CLEAN(c->disk_sb, false);
-
- bch2_write_super(c);
- mutex_unlock(&c->sb_lock);
-
- err = NULL;
+ print_mount_opts(c);
+ ret = 0;
out:
- mutex_unlock(&c->state_lock);
- bch2_journal_entries_free(&journal);
- return err;
+ up_write(&c->state_lock);
+ return ret;
err:
- closure_sync(&cl);
-
switch (ret) {
case BCH_FSCK_ERRORS_NOT_FIXED:
bch_err(c, "filesystem contains errors: please report this to the developers");
break;
}
- BUG_ON(!err);
- set_bit(BCH_FS_ERROR, &c->flags);
+ if (ret >= 0)
+ ret = -EIO;
goto out;
}
-const char *bch2_fs_start(struct bch_fs *c)
-{
- return __bch2_fs_start(c) ?: bch2_fs_online(c);
-}
-
static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
{
struct bch_sb_field_members *sb_mi;
return "mismatched block size";
if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
- BCH_SB_BTREE_NODE_SIZE(c->disk_sb))
+ BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
return "new cache bucket size is too small";
return NULL;
static void bch2_dev_free(struct bch_dev *ca)
{
- unsigned i;
-
cancel_work_sync(&ca->io_error_work);
if (ca->kobj.state_in_sysfs &&
free_percpu(ca->io_done);
bioset_exit(&ca->replica_set);
- free_percpu(ca->usage_percpu);
- kvpfree(ca->bucket_dirty, BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
- kvpfree(ca->buckets, ca->mi.nbuckets * sizeof(struct bucket));
- kvpfree(ca->oldest_gens, ca->mi.nbuckets * sizeof(u8));
- free_heap(&ca->copygc_heap);
- free_heap(&ca->alloc_heap);
- free_fifo(&ca->free_inc);
+ bch2_dev_buckets_free(ca);
+ free_page((unsigned long) ca->sb_read_scratch);
- for (i = 0; i < RESERVE_NR; i++)
- free_fifo(&ca->free[i]);
+ bch2_time_stats_exit(&ca->io_latency[WRITE]);
+ bch2_time_stats_exit(&ca->io_latency[READ]);
percpu_ref_exit(&ca->io_ref);
percpu_ref_exit(&ca->ref);
kobject_put(&ca->kobj);
}
-static void bch2_dev_io_ref_release(struct percpu_ref *ref)
+static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
{
- struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
-
- complete(&ca->offline_complete);
-}
-
-static void __bch2_dev_offline(struct bch_dev *ca)
-{
- struct bch_fs *c = ca->fs;
lockdep_assert_held(&c->state_lock);
+ if (percpu_ref_is_zero(&ca->io_ref))
+ return;
+
__bch2_dev_read_only(c, ca);
- reinit_completion(&ca->offline_complete);
+ reinit_completion(&ca->io_ref_completion);
percpu_ref_kill(&ca->io_ref);
- wait_for_completion(&ca->offline_complete);
+ wait_for_completion(&ca->io_ref_completion);
if (ca->kobj.state_in_sysfs) {
struct kobject *block =
bch2_dev_journal_exit(ca);
}
-static void bch2_dev_ref_release(struct percpu_ref *ref)
+static void bch2_dev_ref_complete(struct percpu_ref *ref)
{
struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
- complete(&ca->stop_complete);
+ complete(&ca->ref_completion);
}
-static void bch2_dev_stop(struct bch_dev *ca)
+static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
{
- struct bch_fs *c = ca->fs;
-
- lockdep_assert_held(&c->state_lock);
-
- BUG_ON(rcu_access_pointer(c->devs[ca->dev_idx]) != ca);
- rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
-
- synchronize_rcu();
+ struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
- reinit_completion(&ca->stop_complete);
- percpu_ref_kill(&ca->ref);
- wait_for_completion(&ca->stop_complete);
+ complete(&ca->io_ref_completion);
}
-static int bch2_dev_sysfs_online(struct bch_dev *ca)
+static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
{
- struct bch_fs *c = ca->fs;
int ret;
if (!c->kobj.state_in_sysfs)
return 0;
}
-static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
+static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
+ struct bch_member *member)
{
- struct bch_member *member;
- size_t reserve_none, movinggc_reserve, free_inc_reserve, total_reserve;
- size_t heap_size;
- unsigned i, btree_node_reserve_buckets;
struct bch_dev *ca;
- if (bch2_fs_init_fault("dev_alloc"))
- return -ENOMEM;
-
ca = kzalloc(sizeof(*ca), GFP_KERNEL);
if (!ca)
- return -ENOMEM;
+ return NULL;
kobject_init(&ca->kobj, &bch2_dev_ktype);
- init_completion(&ca->stop_complete);
- init_completion(&ca->offline_complete);
+ init_completion(&ca->ref_completion);
+ init_completion(&ca->io_ref_completion);
+
+ 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]);
+ bch2_time_stats_init(&ca->io_latency[WRITE]);
+
+ ca->mi = bch2_mi_to_cpu(member);
+ ca->uuid = member->uuid;
+
+ if (opt_defined(c->opts, discard))
+ ca->mi.discard = opt_get(c->opts, discard);
+
+ if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
+ 0, GFP_KERNEL) ||
+ percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
+ PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
+ !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
+ bch2_dev_buckets_alloc(c, ca) ||
+ bioset_init(&ca->replica_set, 4,
+ offsetof(struct bch_write_bio, bio), 0) ||
+ !(ca->io_done = alloc_percpu(*ca->io_done)))
+ goto err;
+
+ return ca;
+err:
+ bch2_dev_free(ca);
+ return NULL;
+}
+
+static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
+ unsigned dev_idx)
+{
ca->dev_idx = dev_idx;
__set_bit(ca->dev_idx, ca->self.d);
+ scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
- ca->copygc_write_point.type = BCH_DATA_USER;
+ ca->fs = c;
+ rcu_assign_pointer(c->devs[ca->dev_idx], ca);
- spin_lock_init(&ca->freelist_lock);
- bch2_dev_moving_gc_init(ca);
+ if (bch2_dev_sysfs_online(c, ca))
+ pr_warn("error creating sysfs objects");
+}
- INIT_WORK(&ca->io_error_work, bch2_io_error_work);
+static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
+{
+ struct bch_member *member =
+ bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
+ struct bch_dev *ca = NULL;
+ int ret = 0;
+
+ pr_verbose_init(c->opts, "");
if (bch2_fs_init_fault("dev_alloc"))
goto err;
- member = bch2_sb_get_members(c->disk_sb)->members + dev_idx;
+ ca = __bch2_dev_alloc(c, member);
+ if (!ca)
+ goto err;
+
+ bch2_dev_attach(c, ca, dev_idx);
+out:
+ pr_verbose_init(c->opts, "ret %i", ret);
+ return ret;
+err:
+ if (ca)
+ bch2_dev_free(ca);
+ ret = -ENOMEM;
+ goto out;
+}
- ca->mi = bch2_mi_to_cpu(member);
- ca->uuid = member->uuid;
- scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
+static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
+{
+ unsigned ret;
- /* XXX: tune these */
- movinggc_reserve = max_t(size_t, 16, ca->mi.nbuckets >> 7);
- reserve_none = max_t(size_t, 4, ca->mi.nbuckets >> 9);
- /*
- * free_inc must be smaller than the copygc reserve: if it was bigger,
- * one copygc iteration might not make enough buckets available to fill
- * up free_inc and allow the allocator to make forward progress
- */
- free_inc_reserve = movinggc_reserve / 2;
- heap_size = movinggc_reserve * 8;
+ if (bch2_dev_is_online(ca)) {
+ bch_err(ca, "already have device online in slot %u",
+ sb->sb->dev_idx);
+ return -EINVAL;
+ }
- btree_node_reserve_buckets =
- DIV_ROUND_UP(BTREE_NODE_RESERVE,
- ca->mi.bucket_size / c->opts.btree_node_size);
+ if (get_capacity(sb->bdev->bd_disk) <
+ ca->mi.bucket_size * ca->mi.nbuckets) {
+ bch_err(ca, "cannot online: device too small");
+ return -EINVAL;
+ }
- if (percpu_ref_init(&ca->ref, bch2_dev_ref_release,
- 0, GFP_KERNEL) ||
- percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_release,
- PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
- !init_fifo(&ca->free[RESERVE_BTREE], btree_node_reserve_buckets,
- GFP_KERNEL) ||
- !init_fifo(&ca->free[RESERVE_MOVINGGC],
- movinggc_reserve, GFP_KERNEL) ||
- !init_fifo(&ca->free[RESERVE_NONE], reserve_none, GFP_KERNEL) ||
- !init_fifo(&ca->free_inc, free_inc_reserve, GFP_KERNEL) ||
- !init_heap(&ca->alloc_heap, free_inc_reserve, GFP_KERNEL) ||
- !init_heap(&ca->copygc_heap,heap_size, GFP_KERNEL) ||
- !(ca->oldest_gens = kvpmalloc(ca->mi.nbuckets *
- sizeof(u8),
- GFP_KERNEL|__GFP_ZERO)) ||
- !(ca->buckets = kvpmalloc(ca->mi.nbuckets *
- sizeof(struct bucket),
- GFP_KERNEL|__GFP_ZERO)) ||
- !(ca->bucket_dirty = kvpmalloc(BITS_TO_LONGS(ca->mi.nbuckets) *
- sizeof(unsigned long),
- GFP_KERNEL|__GFP_ZERO)) ||
- !(ca->usage_percpu = alloc_percpu(struct bch_dev_usage)) ||
- bioset_init(&ca->replica_set, 4,
- offsetof(struct bch_write_bio, bio)) ||
- !(ca->io_done = alloc_percpu(*ca->io_done)))
- goto err;
+ BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
- total_reserve = ca->free_inc.size;
- for (i = 0; i < RESERVE_NR; i++)
- total_reserve += ca->free[i].size;
+ if (get_capacity(sb->bdev->bd_disk) <
+ ca->mi.bucket_size * ca->mi.nbuckets) {
+ bch_err(ca, "device too small");
+ return -EINVAL;
+ }
- ca->copygc_write_point.group = &ca->self;
+ ret = bch2_dev_journal_init(ca, sb->sb);
+ if (ret)
+ return ret;
- ca->fs = c;
- rcu_assign_pointer(c->devs[ca->dev_idx], ca);
+ /* Commit: */
+ ca->disk_sb = *sb;
+ if (sb->mode & FMODE_EXCL)
+ ca->disk_sb.bdev->bd_holder = ca;
+ memset(sb, 0, sizeof(*sb));
- if (bch2_dev_sysfs_online(ca))
- pr_warn("error creating sysfs objects");
+ percpu_ref_reinit(&ca->io_ref);
return 0;
-err:
- bch2_dev_free(ca);
- return -ENOMEM;
}
-static int __bch2_dev_online(struct bch_fs *c, struct bch_sb_handle *sb)
+static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
{
struct bch_dev *ca;
int ret;
- lockdep_assert_held(&c->sb_lock);
+ lockdep_assert_held(&c->state_lock);
if (le64_to_cpu(sb->sb->seq) >
- le64_to_cpu(c->disk_sb->seq))
+ le64_to_cpu(c->disk_sb.sb->seq))
bch2_sb_to_fs(c, sb->sb);
BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
!c->devs[sb->sb->dev_idx]);
- ca = c->devs[sb->sb->dev_idx];
- if (ca->disk_sb.bdev) {
- bch_err(c, "already have device online in slot %u",
- sb->sb->dev_idx);
- return -EINVAL;
- }
+ ca = bch_dev_locked(c, sb->sb->dev_idx);
- ret = bch2_dev_journal_init(ca, sb->sb);
+ ret = __bch2_dev_attach_bdev(ca, sb);
if (ret)
return ret;
- /*
- * Increase journal write timeout if flushes to this device are
- * expensive:
- */
- if (!blk_queue_nonrot(bdev_get_queue(sb->bdev)) &&
- journal_flushes_device(ca))
- c->journal.write_delay_ms =
- max(c->journal.write_delay_ms, 1000U);
+ 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);
+ }
- /* Commit: */
- ca->disk_sb = *sb;
- if (sb->mode & FMODE_EXCL)
- ca->disk_sb.bdev->bd_holder = ca;
- memset(sb, 0, sizeof(*sb));
+ bch2_dev_sysfs_online(c, ca);
if (c->sb.nr_devices == 1)
bdevname(ca->disk_sb.bdev, c->name);
bdevname(ca->disk_sb.bdev, ca->name);
- if (bch2_dev_sysfs_online(ca))
- pr_warn("error creating sysfs objects");
-
- lg_local_lock(&c->usage_lock);
- if (!gc_will_visit(c, gc_phase(GC_PHASE_SB_METADATA)))
- bch2_mark_dev_metadata(c, ca);
- lg_local_unlock(&c->usage_lock);
-
- if (ca->mi.state == BCH_MEMBER_STATE_RW)
- bch2_dev_allocator_add(c, ca);
-
- percpu_ref_reinit(&ca->io_ref);
+ rebalance_wakeup(c);
return 0;
}
/* do we have enough devices to write to? */
for_each_member_device(ca2, c, i)
- nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
+ if (ca2 != ca)
+ 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
: c->opts.data_replicas_required);
- return nr_rw - 1 <= required;
+ return nr_rw >= required;
case BCH_MEMBER_STATE_FAILED:
case BCH_MEMBER_STATE_SPARE:
if (ca->mi.state != BCH_MEMBER_STATE_RW &&
s = __bch2_replicas_status(c, new_online_devs);
- return bch2_have_enough_devs(c, s, flags);
+ return bch2_have_enough_devs(s, flags);
default:
BUG();
}
{
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;
if (!c->opts.degraded) {
mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
+ mi = bch2_sb_get_members(c->disk_sb.sb);
+
+ for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
+ if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
+ continue;
- for (i = 0; i < c->disk_sb->nr_devices; i++)
- if (bch2_dev_exists(c->disk_sb, mi, i) &&
- !bch2_dev_is_online(c->devs[i]) &&
- (c->devs[i]->mi.state == BCH_MEMBER_STATE_RW ||
- c->devs[i]->mi.state == BCH_MEMBER_STATE_RO)) {
+ 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)) {
mutex_unlock(&c->sb_lock);
return false;
}
+ }
mutex_unlock(&c->sb_lock);
}
s = bch2_replicas_status(c);
- return bch2_have_enough_devs(c, s, flags);
+ return bch2_have_enough_devs(s, flags);
}
static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
{
- bch2_moving_gc_stop(ca);
+ bch2_copygc_stop(ca);
/*
- * This stops new data writes (e.g. to existing open data
- * buckets) and then waits for all existing writes to
- * complete.
+ * 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);
}
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_moving_gc_start(ca))
- return "error starting moving GC thread";
-
- if (bch2_tiering_start(c))
- return "error starting tiering thread";
+ if (bch2_copygc_start(c, ca))
+ return "error starting copygc thread";
return NULL;
}
enum bch_member_state new_state, int flags)
{
struct bch_sb_field_members *mi;
+ int ret = 0;
if (ca->mi.state == new_state)
return 0;
if (!bch2_dev_state_allowed(c, ca, new_state, flags))
return -EINVAL;
- if (new_state == BCH_MEMBER_STATE_RW) {
- if (__bch2_dev_read_write(c, ca))
- return -ENOMEM;
- } else {
+ if (new_state != BCH_MEMBER_STATE_RW)
__bch2_dev_read_only(c, ca);
- }
bch_notice(ca, "%s", bch2_dev_state[new_state]);
mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
+ mi = bch2_sb_get_members(c->disk_sb.sb);
SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- return 0;
+ if (new_state == BCH_MEMBER_STATE_RW &&
+ __bch2_dev_read_write(c, ca))
+ ret = -ENOMEM;
+
+ rebalance_wakeup(c);
+
+ return ret;
}
int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
{
int ret;
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
ret = __bch2_dev_set_state(c, ca, new_state, flags);
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
return ret;
}
/* Device add/removal: */
+int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
+{
+ struct btree_trans trans;
+ size_t i;
+ int ret;
+
+ 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));
+ if (ret)
+ break;
+ }
+ bch2_trans_exit(&trans);
+
+ if (ret)
+ return ret;
+
+ return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
+ POS(ca->dev_idx, 0),
+ POS(ca->dev_idx + 1, 0),
+ NULL);
+}
+
int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
{
struct bch_sb_field_members *mi;
unsigned dev_idx = ca->dev_idx, data;
int ret = -EINVAL;
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
- percpu_ref_put(&ca->ref); /* XXX */
+ /*
+ * We consume a reference to ca->ref, regardless of whether we succeed
+ * or fail:
+ */
+ percpu_ref_put(&ca->ref);
- if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- bch_err(ca, "Cannot remove RW device");
+ if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
+ bch_err(ca, "Cannot remove without losing data");
goto err;
}
- if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
- bch_err(ca, "Cannot remove without losing data");
+ __bch2_dev_read_only(c, ca);
+
+ ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
+ if (ret) {
+ bch_err(ca, "Remove failed: error %i dropping data", ret);
+ goto err;
+ }
+
+ ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
+ if (ret) {
+ bch_err(ca, "Remove failed: error %i flushing journal", ret);
+ goto err;
+ }
+
+ ret = bch2_dev_remove_alloc(c, ca);
+ if (ret) {
+ bch_err(ca, "Remove failed, error deleting alloc info");
goto err;
}
/*
- * XXX: verify that dev_idx is really not in use anymore, anywhere
- *
- * flag_data_bad() does not check btree pointers
+ * must flush all existing journal entries, they might have
+ * (overwritten) keys that point to the device we're removing:
+ */
+ bch2_journal_flush_all_pins(&c->journal);
+ /*
+ * hack to ensure bch2_replicas_gc2() clears out entries to this device
*/
- ret = bch2_flag_data_bad(ca);
+ bch2_journal_meta(&c->journal);
+ ret = bch2_journal_error(&c->journal);
+ if (ret) {
+ bch_err(ca, "Remove failed, journal error");
+ goto err;
+ }
+
+ ret = bch2_replicas_gc2(c);
if (ret) {
- bch_err(ca, "Remove failed");
+ bch_err(ca, "Remove failed: error %i from replicas gc", ret);
goto err;
}
data = bch2_dev_has_data(c, ca);
if (data) {
- bch_err(ca, "Remove failed, still has data (%x)", data);
+ char data_has_str[100];
+
+ bch2_flags_to_text(&PBUF(data_has_str),
+ bch2_data_types, data);
+ bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
+ ret = -EBUSY;
goto err;
}
- bch2_journal_meta(&c->journal);
+ __bch2_dev_offline(c, ca);
+
+ mutex_lock(&c->sb_lock);
+ rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
+ mutex_unlock(&c->sb_lock);
+
+ percpu_ref_kill(&ca->ref);
+ wait_for_completion(&ca->ref_completion);
- __bch2_dev_offline(ca);
- bch2_dev_stop(ca);
bch2_dev_free(ca);
/*
* this device must be gone:
*/
mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
+ mi = bch2_sb_get_members(c->disk_sb.sb);
memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
return 0;
err:
- mutex_unlock(&c->state_lock);
+ 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_opts opts = bch2_opts_empty();
struct bch_sb_handle sb;
const char *err;
struct bch_dev *ca = NULL;
- struct bch_sb_field_members *mi, *dev_mi;
- struct bch_member saved_mi;
+ struct bch_sb_field_members *mi;
+ struct bch_member dev_mi;
unsigned dev_idx, nr_devices, u64s;
- int ret = -EINVAL;
+ int ret;
- err = bch2_read_super(path, bch2_opts_empty(), &sb);
- if (err)
- return -EINVAL;
+ ret = bch2_read_super(path, &opts, &sb);
+ if (ret)
+ return ret;
err = bch2_sb_validate(&sb);
if (err)
return -EINVAL;
+ dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
+
err = bch2_dev_may_add(sb.sb, c);
if (err)
return -EINVAL;
- mutex_lock(&c->state_lock);
- mutex_lock(&c->sb_lock);
+ ca = __bch2_dev_alloc(c, &dev_mi);
+ if (!ca) {
+ bch2_free_super(&sb);
+ return -ENOMEM;
+ }
+
+ ret = __bch2_dev_attach_bdev(ca, &sb);
+ if (ret) {
+ bch2_dev_free(ca);
+ return ret;
+ }
/*
- * Preserve the old cache member information (esp. tier)
- * before we start bashing the disk stuff.
+ * We want to allocate journal on the new device before adding the new
+ * device to the filesystem because allocating after we attach requires
+ * spinning up the allocator thread, and the allocator thread requires
+ * doing btree writes, which if the existing devices are RO isn't going
+ * to work
+ *
+ * So we have to mark where the superblocks are, but marking allocated
+ * data normally updates the filesystem usage too, so we have to mark,
+ * allocate the journal, reset all the marks, then remark after we
+ * attach...
*/
- dev_mi = bch2_sb_get_members(sb.sb);
- saved_mi = dev_mi->members[sb.sb->dev_idx];
- saved_mi.last_mount = cpu_to_le64(ktime_get_seconds());
+ bch2_mark_dev_superblock(ca->fs, 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);
+
+ err = "insufficient space in new superblock";
+ ret = bch2_sb_from_fs(c, ca);
+ if (ret)
+ goto err_unlock;
+
+ mi = bch2_sb_get_members(ca->disk_sb.sb);
+
+ if (!bch2_sb_resize_members(&ca->disk_sb,
+ le32_to_cpu(mi->field.u64s) +
+ sizeof(dev_mi) / sizeof(u64))) {
+ ret = -ENOSPC;
+ goto err_unlock;
+ }
if (dynamic_fault("bcachefs:add:no_slot"))
goto no_slot;
- mi = bch2_sb_get_members(c->disk_sb);
+ mi = bch2_sb_get_members(c->disk_sb.sb);
for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
- if (!bch2_dev_exists(c->disk_sb, mi, dev_idx))
+ if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
goto have_slot;
no_slot:
err = "no slots available in superblock";
nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
u64s = (sizeof(struct bch_sb_field_members) +
sizeof(struct bch_member) * nr_devices) / sizeof(u64);
+
err = "no space in superblock for member info";
+ ret = -ENOSPC;
- mi = bch2_fs_sb_resize_members(c, u64s);
+ mi = bch2_sb_resize_members(&c->disk_sb, u64s);
if (!mi)
goto err_unlock;
- dev_mi = bch2_sb_resize_members(&sb, u64s);
- if (!dev_mi)
- goto err_unlock;
-
- memcpy(dev_mi, mi, u64s * sizeof(u64));
- dev_mi->members[dev_idx] = saved_mi;
-
- sb.sb->uuid = c->disk_sb->uuid;
- sb.sb->dev_idx = dev_idx;
- sb.sb->nr_devices = nr_devices;
+ /* success: */
- /* commit new member info */
- memcpy(mi, dev_mi, u64s * sizeof(u64));
- c->disk_sb->nr_devices = nr_devices;
- c->sb.nr_devices = nr_devices;
+ mi->members[dev_idx] = dev_mi;
+ mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
+ c->disk_sb.sb->nr_devices = nr_devices;
- if (bch2_dev_alloc(c, dev_idx)) {
- err = "cannot allocate memory";
- ret = -ENOMEM;
- goto err_unlock;
- }
+ ca->disk_sb.sb->dev_idx = dev_idx;
+ bch2_dev_attach(c, ca, dev_idx);
- if (__bch2_dev_online(c, &sb)) {
- err = "bch2_dev_online() error";
- ret = -ENOMEM;
- goto err_unlock;
- }
+ bch2_mark_dev_superblock(c, ca, 0);
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- ca = c->devs[dev_idx];
if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- err = "journal alloc failed";
- if (bch2_dev_journal_alloc(ca))
- goto err;
-
err = __bch2_dev_read_write(c, ca);
if (err)
- goto err;
+ goto err_late;
}
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
return 0;
+
err_unlock:
mutex_unlock(&c->sb_lock);
+ up_write(&c->state_lock);
err:
- mutex_unlock(&c->state_lock);
+ if (ca)
+ bch2_dev_free(ca);
bch2_free_super(&sb);
-
bch_err(c, "Unable to add device: %s", err);
- return ret ?: -EINVAL;
+ return ret;
+err_late:
+ bch_err(c, "Error going rw after adding device: %s", err);
+ return -EINVAL;
}
/* Hot add existing device to running filesystem: */
int bch2_dev_online(struct bch_fs *c, const char *path)
{
- struct bch_sb_handle sb = { 0 };
+ struct bch_opts opts = bch2_opts_empty();
+ struct bch_sb_handle sb = { NULL };
+ struct bch_sb_field_members *mi;
struct bch_dev *ca;
unsigned dev_idx;
const char *err;
+ int ret;
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
- err = bch2_read_super(path, bch2_opts_empty(), &sb);
- if (err)
- goto err;
+ ret = bch2_read_super(path, &opts, &sb);
+ if (ret) {
+ up_write(&c->state_lock);
+ return ret;
+ }
dev_idx = sb.sb->dev_idx;
- err = bch2_dev_in_fs(c->disk_sb, sb.sb);
+ err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
if (err)
goto err;
- mutex_lock(&c->sb_lock);
- if (__bch2_dev_online(c, &sb)) {
- err = "__bch2_dev_online() error";
- mutex_unlock(&c->sb_lock);
+ if (bch2_dev_attach_bdev(c, &sb)) {
+ err = "bch2_dev_attach_bdev() error";
goto err;
}
- mutex_unlock(&c->sb_lock);
- ca = c->devs[dev_idx];
+ ca = bch_dev_locked(c, dev_idx);
if (ca->mi.state == BCH_MEMBER_STATE_RW) {
err = __bch2_dev_read_write(c, ca);
if (err)
goto err;
}
- mutex_unlock(&c->state_lock);
+ mutex_lock(&c->sb_lock);
+ mi = bch2_sb_get_members(c->disk_sb.sb);
+
+ mi->members[ca->dev_idx].last_mount =
+ cpu_to_le64(ktime_get_real_seconds());
+
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+
+ up_write(&c->state_lock);
return 0;
err:
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
bch2_free_super(&sb);
bch_err(c, "error bringing %s online: %s", path, err);
return -EINVAL;
int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
{
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
+
+ if (!bch2_dev_is_online(ca)) {
+ bch_err(ca, "Already offline");
+ up_write(&c->state_lock);
+ return 0;
+ }
if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
bch_err(ca, "Cannot offline required disk");
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
return -EINVAL;
}
- __bch2_dev_read_only(c, ca);
- __bch2_dev_offline(ca);
+ __bch2_dev_offline(c, ca);
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
return 0;
}
-int bch2_dev_evacuate(struct bch_fs *c, struct bch_dev *ca)
+int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
{
- unsigned data;
- int ret;
+ struct bch_member *mi;
+ int ret = 0;
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
- if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- bch_err(ca, "Cannot migrate data off RW device");
- mutex_unlock(&c->state_lock);
- return -EINVAL;
+ if (nbuckets < ca->mi.nbuckets) {
+ bch_err(ca, "Cannot shrink yet");
+ ret = -EINVAL;
+ goto err;
}
- mutex_unlock(&c->state_lock);
-
- ret = bch2_move_data_off_device(ca);
- if (ret) {
- bch_err(ca, "Error migrating data: %i", ret);
- return ret;
+ if (bch2_dev_is_online(ca) &&
+ get_capacity(ca->disk_sb.bdev->bd_disk) <
+ ca->mi.bucket_size * nbuckets) {
+ bch_err(ca, "New size larger than device");
+ ret = -EINVAL;
+ goto err;
}
- ret = bch2_move_metadata_off_device(ca);
+ ret = bch2_dev_buckets_resize(c, ca, nbuckets);
if (ret) {
- bch_err(ca, "Error migrating metadata: %i", ret);
- return ret;
+ bch_err(ca, "Resize error: %i", ret);
+ goto err;
}
- data = bch2_dev_has_data(c, ca);
- if (data) {
- bch_err(ca, "Migrate error: data still present (%x)", data);
- return -EINVAL;
- }
+ 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 0;
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+
+ bch2_recalc_capacity(c);
+err:
+ up_write(&c->state_lock);
+ return ret;
+}
+
+/* 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;
+
+ if (IS_ERR(bdev))
+ return ERR_CAST(bdev);
+
+ for_each_member_device(ca, c, i)
+ if (ca->disk_sb.bdev == bdev)
+ goto found;
+
+ ca = ERR_PTR(-ENOENT);
+found:
+ bdput(bdev);
+ return ca;
}
/* Filesystem open: */
-const char *bch2_fs_open(char * const *devices, unsigned nr_devices,
- struct bch_opts opts, struct bch_fs **ret)
+struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
+ struct bch_opts opts)
{
- const char *err;
+ struct bch_sb_handle *sb = NULL;
struct bch_fs *c = NULL;
- struct bch_sb_handle *sb;
unsigned i, best_sb = 0;
+ const char *err;
+ int ret = -ENOMEM;
+
+ pr_verbose_init(opts, "");
- if (!nr_devices)
- return "need at least one device";
+ if (!nr_devices) {
+ c = ERR_PTR(-EINVAL);
+ goto out2;
+ }
- if (!try_module_get(THIS_MODULE))
- return "module unloading";
+ if (!try_module_get(THIS_MODULE)) {
+ c = ERR_PTR(-ENODEV);
+ goto out2;
+ }
- err = "cannot allocate memory";
sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
if (!sb)
goto err;
for (i = 0; i < nr_devices; i++) {
- err = bch2_read_super(devices[i], opts, &sb[i]);
- if (err)
+ ret = bch2_read_super(devices[i], &opts, &sb[i]);
+ if (ret)
goto err;
err = bch2_sb_validate(&sb[i]);
if (err)
- goto err;
+ goto err_print;
}
for (i = 1; i < nr_devices; i++)
for (i = 0; i < nr_devices; i++) {
err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
if (err)
- goto err;
+ goto err_print;
}
- err = "cannot allocate memory";
+ ret = -ENOMEM;
c = bch2_fs_alloc(sb[best_sb].sb, opts);
if (!c)
goto err;
err = "bch2_dev_online() error";
- mutex_lock(&c->sb_lock);
+ down_write(&c->state_lock);
for (i = 0; i < nr_devices; i++)
- if (__bch2_dev_online(c, &sb[i])) {
- mutex_unlock(&c->sb_lock);
- goto err;
+ if (bch2_dev_attach_bdev(c, &sb[i])) {
+ up_write(&c->state_lock);
+ goto err_print;
}
- mutex_unlock(&c->sb_lock);
+ up_write(&c->state_lock);
err = "insufficient devices";
if (!bch2_fs_may_start(c))
- goto err;
+ goto err_print;
if (!c->opts.nostart) {
- err = __bch2_fs_start(c);
- if (err)
+ ret = bch2_fs_start(c);
+ if (ret)
goto err;
}
-
- err = bch2_fs_online(c);
- if (err)
- goto err;
-
- if (ret)
- *ret = c;
- else
- closure_put(&c->cl);
-
- err = NULL;
out:
kfree(sb);
module_put(THIS_MODULE);
- if (err)
- c = NULL;
- return err;
+out2:
+ pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
+ return c;
+err_print:
+ pr_err("bch_fs_open err opening %s: %s",
+ devices[0], err);
+ ret = -EINVAL;
err:
if (c)
bch2_fs_stop(c);
-
for (i = 0; i < nr_devices; i++)
bch2_free_super(&sb[i]);
+ c = ERR_PTR(ret);
goto out;
}
const char *err;
struct bch_fs *c;
bool allocated_fs = false;
+ int ret;
err = bch2_sb_validate(sb);
if (err)
if (c) {
closure_get(&c->cl);
- err = bch2_dev_in_fs(c->disk_sb, sb->sb);
+ err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
if (err)
goto err;
} else {
err = "bch2_dev_online() error";
mutex_lock(&c->sb_lock);
- if (__bch2_dev_online(c, sb)) {
+ if (bch2_dev_attach_bdev(c, sb)) {
mutex_unlock(&c->sb_lock);
goto err;
}
mutex_unlock(&c->sb_lock);
if (!c->opts.nostart && bch2_fs_may_start(c)) {
- err = __bch2_fs_start(c);
- if (err)
+ err = "error starting filesystem";
+ ret = bch2_fs_start(c);
+ if (ret)
goto err;
}
- err = __bch2_fs_online(c);
- if (err)
- goto err;
-
closure_put(&c->cl);
mutex_unlock(&bch_fs_list_lock);
struct bch_opts opts = bch2_opts_empty();
const char *err;
- err = bch2_read_super(path, opts, &sb);
- if (err)
- return err;
+ if (bch2_read_super(path, &opts, &sb))
+ return "error reading superblock";
err = __bch2_fs_open_incremental(&sb, opts);
bch2_free_super(&sb);