+// 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_journal_iter.h"
+#include "btree_key_cache.h"
+#include "btree_update_interior.h"
#include "btree_io.h"
+#include "btree_write_buffer.h"
+#include "buckets_waiting_for_journal.h"
#include "chardev.h"
#include "checksum.h"
#include "clock.h"
#include "compress.h"
+#include "counters.h"
#include "debug.h"
+#include "disk_groups.h"
+#include "ec.h"
+#include "errcode.h"
#include "error.h"
#include "fs.h"
-#include "fs-gc.h"
+#include "fs-io.h"
+#include "fs-io-buffered.h"
+#include "fs-io-direct.h"
+#include "fsck.h"
#include "inode.h"
-#include "io.h"
+#include "io_read.h"
+#include "io_write.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 "nocow_locking.h"
+#include "quota.h"
+#include "rebalance.h"
+#include "recovery.h"
+#include "replicas.h"
+#include "sb-clean.h"
+#include "sb-errors.h"
+#include "sb-members.h"
+#include "snapshot.h"
+#include "subvolume.h"
#include "super.h"
#include "super-io.h"
#include "sysfs.h"
-#include "tier.h"
+#include "trace.h"
#include <linux/backing-dev.h>
#include <linux/blkdev.h>
#include <linux/debugfs.h>
#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>
#include <linux/sysfs.h>
#include <crypto/hash.h>
-#include <trace/events/bcachefs.h>
-
MODULE_LICENSE("GPL");
MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
+MODULE_DESCRIPTION("bcachefs filesystem");
+MODULE_SOFTDEP("pre: crc32c");
+MODULE_SOFTDEP("pre: crc64");
+MODULE_SOFTDEP("pre: sha256");
+MODULE_SOFTDEP("pre: chacha20");
+MODULE_SOFTDEP("pre: poly1305");
+MODULE_SOFTDEP("pre: xxhash");
+
+const char * const bch2_fs_flag_strs[] = {
+#define x(n) #n,
+ BCH_FS_FLAGS()
+#undef x
+ NULL
+};
+
+void __bch2_print(struct bch_fs *c, const char *fmt, ...)
+{
+ struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
+
+ va_list args;
+ va_start(args, fmt);
+ if (likely(!stdio)) {
+ vprintk(fmt, args);
+ } else {
+ unsigned long flags;
+
+ if (fmt[0] == KERN_SOH[0])
+ fmt += 2;
+
+ spin_lock_irqsave(&stdio->output_lock, flags);
+ prt_vprintf(&stdio->output_buf, fmt, args);
+ spin_unlock_irqrestore(&stdio->output_lock, flags);
+
+ wake_up(&stdio->output_wait);
+ }
+ va_end(args);
+}
#define KTYPE(type) \
-struct kobj_type type ## _ktype = { \
+static const struct attribute_group type ## _group = { \
+ .attrs = type ## _files \
+}; \
+ \
+static const struct attribute_group *type ## _groups[] = { \
+ &type ## _group, \
+ NULL \
+}; \
+ \
+static const struct kobj_type type ## _ktype = { \
.release = type ## _release, \
.sysfs_ops = &type ## _sysfs_ops, \
- .default_attrs = type ## _files \
+ .default_groups = type ## _groups \
}
static void bch2_fs_release(struct kobject *);
static void bch2_dev_release(struct kobject *);
+static void bch2_fs_counters_release(struct kobject *k)
+{
+}
static void bch2_fs_internal_release(struct kobject *k)
{
{
}
-static KTYPE(bch2_fs);
-static KTYPE(bch2_fs_internal);
-static KTYPE(bch2_fs_opts_dir);
-static KTYPE(bch2_fs_time_stats);
-static KTYPE(bch2_dev);
+KTYPE(bch2_fs);
+KTYPE(bch2_fs_counters);
+KTYPE(bch2_fs_internal);
+KTYPE(bch2_fs_opts_dir);
+KTYPE(bch2_fs_time_stats);
+KTYPE(bch2_dev);
static struct kset *bcachefs_kset;
static LIST_HEAD(bch_fs_list);
static DEFINE_MUTEX(bch_fs_list_lock);
-static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
+DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
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)
+struct bch_fs *bch2_dev_to_fs(dev_t dev)
{
struct bch_fs *c;
- struct bch_dev *ca;
- unsigned i;
mutex_lock(&bch_fs_list_lock);
rcu_read_lock();
list_for_each_entry(c, &bch_fs_list, list)
- for_each_member_device_rcu(ca, c, i)
- if (ca->disk_sb.bdev == bdev) {
+ for_each_member_device_rcu(c, ca, NULL)
+ if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
closure_get(&c->cl);
goto found;
}
return c;
}
-static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
+static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
{
struct bch_fs *c;
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)))
return c;
return NULL;
}
-struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
+struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
{
struct bch_fs *c;
return c;
}
-int bch2_congested(struct bch_fs *c, int bdi_bits)
+static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
{
- struct backing_dev_info *bdi;
- struct bch_dev *ca;
- unsigned i;
- int ret = 0;
+ unsigned nr = 0, u64s =
+ ((sizeof(struct jset_entry_dev_usage) +
+ sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
+ sizeof(u64);
- if (bdi_bits & (1 << WB_sync_congested)) {
- /* Reads - check all devices: */
- for_each_readable_member(ca, c, i) {
- bdi = blk_get_backing_dev_info(ca->disk_sb.bdev);
-
- if (bdi_congested(bdi, bdi_bits)) {
- ret = 1;
- break;
- }
- }
- } else {
- /* Writes prefer fastest tier: */
- struct bch_tier *tier = READ_ONCE(c->fastest_tier);
- struct dev_group *grp = tier ? &tier->devs : &c->all_devs;
-
- rcu_read_lock();
- group_for_each_dev(ca, grp, i) {
- bdi = blk_get_backing_dev_info(ca->disk_sb.bdev);
-
- if (bdi_congested(bdi, bdi_bits)) {
- ret = 1;
- break;
- }
- }
- rcu_read_unlock();
- }
-
- return ret;
-}
-
-static int bch2_congested_fn(void *data, int bdi_bits)
-{
- struct bch_fs *c = data;
+ rcu_read_lock();
+ for_each_member_device_rcu(c, ca, NULL)
+ nr++;
+ rcu_read_unlock();
- return bch2_congested(c, bdi_bits);
+ bch2_journal_entry_res_resize(&c->journal,
+ &c->dev_usage_journal_res, u64s * nr);
}
/* 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;
+ unsigned clean_passes = 0;
+ u64 seq = 0;
- bch2_tiering_stop(c);
+ bch2_fs_ec_stop(c);
+ bch2_open_buckets_stop(c, NULL, true);
+ bch2_rebalance_stop(c);
+ bch2_copygc_stop(c);
+ bch2_gc_thread_stop(c);
+ bch2_fs_ec_flush(c);
- for_each_member_device(ca, c, i)
- bch2_moving_gc_stop(ca);
+ bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
+ journal_cur_seq(&c->journal));
- bch2_gc_thread_stop(c);
+ do {
+ clean_passes++;
- bch2_btree_flush(c);
+ if (bch2_btree_interior_updates_flush(c) ||
+ bch2_journal_flush_all_pins(&c->journal) ||
+ bch2_btree_flush_all_writes(c) ||
+ seq != atomic64_read(&c->journal.seq)) {
+ seq = atomic64_read(&c->journal.seq);
+ clean_passes = 0;
+ }
+ } while (clean_passes < 2);
- for_each_member_device(ca, c, i)
- bch2_dev_allocator_stop(ca);
+ bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
+ journal_cur_seq(&c->journal));
+ if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
+ !test_bit(BCH_FS_emergency_ro, &c->flags))
+ set_bit(BCH_FS_clean_shutdown, &c->flags);
bch2_fs_journal_stop(&c->journal);
+
+ /*
+ * After stopping journal:
+ */
+ for_each_member_device(c, ca)
+ bch2_dev_allocator_remove(c, ca);
}
+#ifndef BCH_WRITE_REF_DEBUG
static void bch2_writes_disabled(struct percpu_ref *writes)
{
struct bch_fs *c = container_of(writes, struct bch_fs, writes);
- set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
- wake_up(&bch_read_only_wait);
+ set_bit(BCH_FS_write_disable_complete, &c->flags);
+ wake_up(&bch2_read_only_wait);
}
+#endif
void bch2_fs_read_only(struct bch_fs *c)
{
- mutex_lock(&c->state_lock);
- if (c->state != BCH_FS_STARTING &&
- c->state != BCH_FS_RW)
- goto out;
+ if (!test_bit(BCH_FS_rw, &c->flags)) {
+ bch2_journal_reclaim_stop(&c->journal);
+ return;
+ }
- if (test_bit(BCH_FS_ERROR, &c->flags))
- goto out;
+ BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
+
+ bch_verbose(c, "going read-only");
/*
* Block new foreground-end write operations from starting - any new
* writes will return -EROFS:
- *
- * (This is really blocking new _allocations_, writes to previously
- * allocated space can still happen until stopping the allocator in
- * bch2_dev_allocator_stop()).
*/
+ set_bit(BCH_FS_going_ro, &c->flags);
+#ifndef BCH_WRITE_REF_DEBUG
percpu_ref_kill(&c->writes);
-
- del_timer(&c->foreground_write_wakeup);
- cancel_delayed_work(&c->pd_controllers_update);
-
- c->foreground_write_pd.rate.rate = UINT_MAX;
- bch2_wake_delayed_writes((unsigned long) c);
+#else
+ for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
+ bch2_write_ref_put(c, i);
+#endif
/*
* If we're not doing an emergency shutdown, we want to wait on
* we do need to wait on them before returning and signalling
* that going RO is complete:
*/
- wait_event(bch_read_only_wait,
- test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
- test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
+ wait_event(bch2_read_only_wait,
+ test_bit(BCH_FS_write_disable_complete, &c->flags) ||
+ test_bit(BCH_FS_emergency_ro, &c->flags));
+
+ bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
+ if (writes_disabled)
+ bch_verbose(c, "finished waiting for writes to stop");
__bch2_fs_read_only(c);
- wait_event(bch_read_only_wait,
- test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
+ wait_event(bch2_read_only_wait,
+ test_bit(BCH_FS_write_disable_complete, &c->flags));
- clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
+ if (!writes_disabled)
+ bch_verbose(c, "finished waiting for writes to stop");
+
+ clear_bit(BCH_FS_write_disable_complete, &c->flags);
+ clear_bit(BCH_FS_going_ro, &c->flags);
+ clear_bit(BCH_FS_rw, &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_clean_shutdown, &c->flags) &&
+ !c->opts.norecovery) {
+ BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
+ BUG_ON(atomic_read(&c->btree_cache.dirty));
+ BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
+ BUG_ON(c->btree_write_buffer.inc.keys.nr);
+ BUG_ON(c->btree_write_buffer.flushing.keys.nr);
+
+ bch_verbose(c, "marking filesystem clean");
+ bch2_fs_mark_clean(c);
+ } else {
+ bch_verbose(c, "done going read-only, filesystem not clean");
}
-
- c->state = BCH_FS_RO;
-out:
- mutex_unlock(&c->state_lock);
}
static void bch2_fs_read_only_work(struct work_struct *work)
struct bch_fs *c =
container_of(work, struct bch_fs, read_only_work);
+ down_write(&c->state_lock);
bch2_fs_read_only(c);
+ up_write(&c->state_lock);
}
static void bch2_fs_read_only_async(struct bch_fs *c)
bool bch2_fs_emergency_read_only(struct bch_fs *c)
{
- bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
+ 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);
+ wake_up(&bch2_read_only_wait);
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;
- mutex_lock(&c->state_lock);
- if (c->state != BCH_FS_STARTING &&
- c->state != BCH_FS_RO)
- goto out;
+ /*
+ * Data move operations can't run until after check_snapshots has
+ * completed, and bch2_snapshot_is_ancestor() is available.
+ *
+ * Ideally we'd start copygc/rebalance earlier instead of waiting for
+ * all of recovery/fsck to complete:
+ */
+ ret = bch2_copygc_start(c);
+ if (ret) {
+ bch_err(c, "error starting copygc thread");
+ return ret;
+ }
- 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;
- }
+ ret = bch2_rebalance_start(c);
+ if (ret) {
+ bch_err(c, "error starting rebalance thread");
+ return ret;
+ }
- err = "error starting btree GC thread";
- if (bch2_gc_thread_start(c))
- goto err;
+ return 0;
+}
- err = "error starting moving GC thread";
- for_each_rw_member(ca, c, i)
- if (bch2_moving_gc_start(ca)) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
+static int __bch2_fs_read_write(struct bch_fs *c, bool early)
+{
+ int ret;
+
+ if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
+ bch_err(c, "cannot go rw, unfixed btree errors");
+ return -BCH_ERR_erofs_unfixed_errors;
+ }
- err = "error starting tiering thread";
- if (bch2_tiering_start(c))
+ if (test_bit(BCH_FS_rw, &c->flags))
+ return 0;
+
+ bch_info(c, "going read-write");
+
+ ret = bch2_sb_members_v2_init(c);
+ if (ret)
goto err;
- schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
+ ret = bch2_fs_mark_dirty(c);
+ if (ret)
+ goto err;
- if (c->state != BCH_FS_STARTING)
- percpu_ref_reinit(&c->writes);
+ clear_bit(BCH_FS_clean_shutdown, &c->flags);
- c->state = BCH_FS_RW;
- err = NULL;
-out:
- mutex_unlock(&c->state_lock);
- return err;
+ /*
+ * First journal write must be a flush write: after a clean shutdown we
+ * don't read the journal, so the first journal write may end up
+ * overwriting whatever was there previously, and there must always be
+ * at least one non-flush write in the journal or recovery will fail:
+ */
+ set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
+
+ for_each_rw_member(c, ca)
+ bch2_dev_allocator_add(c, ca);
+ bch2_recalc_capacity(c);
+
+ set_bit(BCH_FS_rw, &c->flags);
+ set_bit(BCH_FS_was_rw, &c->flags);
+
+#ifndef BCH_WRITE_REF_DEBUG
+ percpu_ref_reinit(&c->writes);
+#else
+ for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
+ BUG_ON(atomic_long_read(&c->writes[i]));
+ atomic_long_inc(&c->writes[i]);
+ }
+#endif
+
+ ret = bch2_gc_thread_start(c);
+ if (ret) {
+ bch_err(c, "error starting gc thread");
+ return ret;
+ }
+
+ ret = bch2_journal_reclaim_start(&c->journal);
+ if (ret)
+ goto err;
+
+ if (!early) {
+ ret = bch2_fs_read_write_late(c);
+ if (ret)
+ goto err;
+ }
+
+ bch2_do_discards(c);
+ bch2_do_invalidates(c);
+ bch2_do_stripe_deletes(c);
+ bch2_do_pending_node_rewrites(c);
+ return 0;
err:
- __bch2_fs_read_only(c);
- goto out;
+ if (test_bit(BCH_FS_rw, &c->flags))
+ bch2_fs_read_only(c);
+ else
+ __bch2_fs_read_only(c);
+ return ret;
+}
+
+int bch2_fs_read_write(struct bch_fs *c)
+{
+ if (c->opts.norecovery)
+ return -BCH_ERR_erofs_norecovery;
+
+ if (c->opts.nochanges)
+ return -BCH_ERR_erofs_nochanges;
+
+ 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)
+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_free_pending_node_rewrites(c);
+ bch2_fs_sb_errors_exit(c);
+ bch2_fs_counters_exit(c);
+ bch2_fs_snapshots_exit(c);
+ bch2_fs_quota_exit(c);
+ bch2_fs_fs_io_direct_exit(c);
+ bch2_fs_fs_io_buffered_exit(c);
+ bch2_fs_fsio_exit(c);
+ bch2_fs_ec_exit(c);
bch2_fs_encryption_exit(c);
- bch2_fs_btree_exit(c);
+ bch2_fs_nocow_locking_exit(c);
+ bch2_fs_io_write_exit(c);
+ bch2_fs_io_read_exit(c);
+ bch2_fs_buckets_waiting_for_journal_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_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_fs_compress_exit(c);
- bdi_destroy(&c->bdi);
- lg_lock_free(&c->usage_lock);
- free_percpu(c->usage_percpu);
+ bch2_journal_keys_put_initial(c);
+ BUG_ON(atomic_read(&c->journal_keys.ref));
+ bch2_fs_btree_write_buffer_exit(c);
+ percpu_free_rwsem(&c->mark_lock);
+ free_percpu(c->online_reserved);
+
+ darray_exit(&c->btree_roots_extra);
+ 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);
+#ifndef BCH_WRITE_REF_DEBUG
percpu_ref_exit(&c->writes);
-
+#endif
+ kfree(rcu_dereference_protected(c->disk_groups, 1));
+ kfree(c->journal_seq_blacklist_table);
+ kfree(c->unused_inode_hints);
+
+ if (c->write_ref_wq)
+ destroy_workqueue(c->write_ref_wq);
+ 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, c->disk_sb_order);
- kfree(c);
+ bch2_free_super(&c->disk_sb);
+ 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);
- cancel_work_sync(&c->read_retry_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;
+ bch_verbose(c, "shutting down");
- mutex_lock(&bch_fs_list_lock);
- list_del(&c->list);
- mutex_unlock(&bch_fs_list_lock);
+ 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)
+ for_each_member_device(c, ca)
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);
bch2_fs_debug_exit(c);
bch2_fs_chardev_exit(c);
+ bch2_ro_ref_put(c);
+ wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
+
+ kobject_put(&c->counters_kobj);
kobject_put(&c->time_stats);
kobject_put(&c->opts_dir);
kobject_put(&c->internal);
- __bch2_fs_read_only(c);
+ /* btree prefetch might have kicked off reads in the background: */
+ bch2_btree_flush_all_reads(c);
+
+ for_each_member_device(c, ca)
+ cancel_work_sync(&ca->io_error_work);
+
+ cancel_work_sync(&c->read_only_work);
}
-static void bch2_fs_release(struct kobject *kobj)
+void bch2_fs_free(struct bch_fs *c)
{
- struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
+ unsigned i;
- bch2_fs_free(c);
+ 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++) {
+ struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
+
+ if (ca) {
+ bch2_free_super(&ca->disk_sb);
+ bch2_dev_free(ca);
+ }
+ }
+
+ bch_verbose(c, "shutdown complete");
+
+ kobject_put(&c->kobj);
}
void bch2_fs_stop(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);
+ __bch2_fs_stop(c);
+ bch2_fs_free(c);
+}
- bch2_fs_offline(c);
+static int bch2_fs_online(struct bch_fs *c)
+{
+ int ret = 0;
- closure_sync(&c->cl);
+ lockdep_assert_held(&bch_fs_list_lock);
- bch2_fs_exit(c);
-}
+ 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) {
+ bch_err(c, "error creating character device");
+ return ret;
+ }
+
+ bch2_fs_debug_init(c);
+
+ 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") ?:
+#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
+ kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
+#endif
+ kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
+ bch2_opts_create_sysfs_files(&c->opts_dir);
+ if (ret) {
+ bch_err(c, "error creating sysfs objects");
+ return ret;
+ }
+
+ down_write(&c->state_lock);
-#define alloc_bucket_pages(gfp, ca) \
- ((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(ca))))
+ for_each_member_device(c, ca) {
+ 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:
+ up_write(&c->state_lock);
+ 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, journal_entry_bytes;
+ struct printbuf name = PRINTBUF;
+ unsigned i, iter_size;
+ int ret = 0;
- c = kzalloc(sizeof(struct bch_fs), GFP_KERNEL);
- if (!c)
- return NULL;
+ c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
+ if (!c) {
+ c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
+ goto out;
+ }
+
+ c->stdio = (void *)(unsigned long) opts.stdio;
__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);
+ kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
+
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->btree_cache_lock);
- mutex_init(&c->bucket_lock);
+ mutex_init(&c->replicas_gc_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
+ refcount_set(&c->ro_ref, 1);
+ init_waitqueue_head(&c->ro_ref_wait);
+ sema_init(&c->online_fsck_mutex, 1);
- bch2_fs_allocator_init(c);
- bch2_fs_tiering_init(c);
+ init_rwsem(&c->gc_lock);
+ mutex_init(&c->gc_gens_lock);
+ atomic_set(&c->journal_keys.ref, 1);
+ c->journal_keys.initial_ref_held = 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_btree_iter_init_early(c);
+ bch2_fs_btree_interior_update_init_early(c);
+ bch2_fs_allocator_background_init(c);
+ bch2_fs_allocator_foreground_init(c);
+ bch2_fs_rebalance_init(c);
+ bch2_fs_quota_init(c);
+ bch2_fs_ec_init_early(c);
+ bch2_fs_move_init(c);
+ bch2_fs_sb_errors_init_early(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);
- bio_list_init(&c->read_retry_list);
- spin_lock_init(&c->read_retry_lock);
- INIT_WORK(&c->read_retry_work, bch2_read_retry_work);
- mutex_init(&c->zlib_workspace_lock);
+ mutex_init(&c->snapshot_table_lock);
+ init_rwsem(&c->snapshot_create_lock);
+
+ spin_lock_init(&c->btree_write_error_lock);
+
+ INIT_WORK(&c->journal_seq_blacklist_gc_work,
+ bch2_blacklist_entries_gc);
+
+ INIT_LIST_HEAD(&c->journal_iters);
+
+ INIT_LIST_HEAD(&c->fsck_error_msgs);
+ mutex_init(&c->fsck_error_msgs_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);
- init_waitqueue_head(&c->writeback_wait);
- c->writeback_pages_max = (256 << 10) / PAGE_SIZE;
+ sema_init(&c->io_in_flight, 128);
- c->copy_gc_enabled = 1;
- c->tiering_enabled = 1;
- c->tiering_percent = 10;
+ INIT_LIST_HEAD(&c->vfs_inodes_list);
+ mutex_init(&c->vfs_inodes_lock);
- c->foreground_target_percent = 20;
+ c->copy_gc_enabled = 1;
+ c->rebalance.enabled = 1;
+ c->promote_whole_extents = true;
- 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;
+ 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.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
- mutex_lock(&c->sb_lock);
+ bch2_fs_btree_cache_init_early(&c->btree_cache);
- if (bch2_sb_to_fs(c, sb)) {
- mutex_unlock(&c->sb_lock);
+ mutex_init(&c->sectors_available_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);
- scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
+ if (ret)
+ goto err;
+
+ pr_uuid(&name, c->sb.user_uuid.b);
+ strscpy(c->name, name.buf, sizeof(c->name));
+ printbuf_exit(&name);
- bch2_opts_apply(&c->opts, bch2_sb_opts(sb));
- bch2_opts_apply(&c->opts, opts);
+ ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
+ if (ret)
+ goto err;
- c->opts.nochanges |= c->opts.noreplay;
- c->opts.read_only |= c->opts.nochanges;
+ /* Compat: */
+ if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+ SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
- c->block_bits = ilog2(c->sb.block_size);
+ if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
+ SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
- if (bch2_fs_init_fault("fs_alloc"))
+ c->opts = bch2_opts_default;
+ ret = bch2_opts_from_sb(&c->opts, sb);
+ if (ret)
goto err;
- iter_size = (btree_blocks(c) + 1) * 2 *
- sizeof(struct btree_node_iter_set);
+ bch2_opts_apply(&c->opts, opts);
+
+ c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
+ if (c->opts.inodes_use_key_cache)
+ c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
+ c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
+
+ c->block_bits = ilog2(block_sectors(c));
+ c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
- journal_entry_bytes = 512U << BCH_SB_JOURNAL_ENTRY_SIZE(sb);
+ if (bch2_fs_init_fault("fs_alloc")) {
+ bch_err(c, "fs_alloc fault injected");
+ ret = -EFAULT;
+ goto err;
+ }
- if (!(c->wq = alloc_workqueue("bcachefs",
- WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
- !(c->copygc_wq = alloc_workqueue("bcache_copygc",
- 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_interior_update)) ||
+ iter_size = sizeof(struct sort_iter) +
+ (btree_blocks(c) + 1) * 2 *
+ sizeof(struct sort_iter_set);
+
+ c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
+
+ if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
+ WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
+ !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
+ WQ_FREEZABLE|WQ_MEM_RECLAIM, 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, 512)) ||
+ !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
+ WQ_FREEZABLE, 0)) ||
+#ifndef BCH_WRITE_REF_DEBUG
+ percpu_ref_init(&c->writes, bch2_writes_disabled,
+ PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
+#endif
mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
- bioset_init(&c->btree_read_bio, 1, 0) ||
- 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->sb.btree_node_size,
- BCH_ENCODED_EXTENT_MAX) /
- PAGE_SECTORS, 0) ||
- !(c->usage_percpu = alloc_percpu(struct bch_fs_usage)) ||
- lg_lock_init(&c->usage_lock) ||
- mempool_init_page_pool(&c->btree_bounce_pool, 1,
- ilog2(btree_pages(c))) ||
- bdi_setup_and_register(&c->bdi, "bcachefs") ||
- bch2_io_clock_init(&c->io_clock[READ]) ||
- bch2_io_clock_init(&c->io_clock[WRITE]) ||
- bch2_fs_journal_init(&c->journal, journal_entry_bytes) ||
- bch2_fs_btree_init(c) ||
- bch2_fs_encryption_init(c) ||
- bch2_fs_compress_init(c) ||
- bch2_check_set_has_compressed_data(c, c->opts.compression))
+ 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)) ||
+ !(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))) {
+ ret = -BCH_ERR_ENOMEM_fs_other_alloc;
goto err;
+ }
- c->bdi.ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
- c->bdi.congested_fn = bch2_congested_fn;
- c->bdi.congested_data = c;
+ ret = bch2_fs_counters_init(c) ?:
+ bch2_fs_sb_errors_init(c) ?:
+ 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_buckets_waiting_for_journal_init(c) ?:
+ bch2_fs_btree_write_buffer_init(c) ?:
+ bch2_fs_subvolumes_init(c) ?:
+ bch2_fs_io_read_init(c) ?:
+ bch2_fs_io_write_init(c) ?:
+ bch2_fs_nocow_locking_init(c) ?:
+ bch2_fs_encryption_init(c) ?:
+ bch2_fs_compress_init(c) ?:
+ bch2_fs_ec_init(c) ?:
+ bch2_fs_fsio_init(c) ?:
+ bch2_fs_fs_io_buffered_init(c) ?:
+ bch2_fs_fs_io_direct_init(c);
+ if (ret)
+ goto err;
- mi = bch2_sb_get_members(c->disk_sb);
for (i = 0; i < c->sb.nr_devices; i++)
- if (!bch2_is_zero(mi->members[i].uuid.b, sizeof(uuid_le)) &&
- bch2_dev_alloc(c, i))
+ if (bch2_dev_exists(c->disk_sb.sb, i) &&
+ bch2_dev_alloc(c, i)) {
+ ret = -EEXIST;
goto err;
+ }
- /*
- * Now that all allocations have succeeded, init various refcounty
- * things that let us shutdown:
- */
- closure_init(&c->cl, NULL);
+ 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);
- 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);
+ mutex_lock(&bch_fs_list_lock);
+ ret = bch2_fs_online(c);
+ mutex_unlock(&bch_fs_list_lock);
+
+ if (ret)
+ goto err;
+out:
return c;
err:
bch2_fs_free(c);
- return NULL;
+ c = ERR_PTR(ret);
+ 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;
+ enum bch_opt_id i;
+ struct printbuf p = PRINTBUF;
+ bool first = true;
- lockdep_assert_held(&bch_fs_list_lock);
+ prt_str(&p, "mounting version ");
+ bch2_version_to_text(&p, c->sb.version);
- if (!list_empty(&c->list))
- return 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";
+ if (c->opts.read_only) {
+ prt_str(&p, " opts=");
+ first = false;
+ prt_printf(&p, "ro");
+ }
- bch2_fs_debug_init(c);
+ 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);
- 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"))
- return "error creating sysfs objects";
+ if (!(opt->flags & OPT_MOUNT))
+ continue;
- mutex_lock(&c->state_lock);
+ if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
+ continue;
- err = "error creating sysfs objects";
- __for_each_member_device(ca, c, i)
- if (bch2_dev_sysfs_online(ca))
- goto err;
+ prt_str(&p, first ? " opts=" : ",");
+ first = false;
+ bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
+ }
- list_add(&c->list, &bch_fs_list);
- err = NULL;
-err:
- mutex_unlock(&c->state_lock);
- return err;
+ bch_info(c, "%s", p.buf);
+ printbuf_exit(&p);
}
-static const char *bch2_fs_online(struct bch_fs *c)
+int bch2_fs_start(struct bch_fs *c)
{
- const char *err;
-
- mutex_lock(&bch_fs_list_lock);
- err = __bch2_fs_online(c);
- mutex_unlock(&bch_fs_list_lock);
+ time64_t now = ktime_get_real_seconds();
+ int ret;
- return err;
-}
+ print_mount_opts(c);
-static const char *__bch2_fs_start(struct bch_fs *c)
-{
- const char *err = "cannot allocate memory";
- struct bch_sb_field_members *mi;
- struct bch_dev *ca;
- unsigned i, id;
- time64_t now;
- LIST_HEAD(journal);
- struct jset *j;
- int ret = -EINVAL;
+ down_write(&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);
- mutex_unlock(&c->sb_lock);
-
- 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);
-
- err = "error reading priorities";
- for_each_readable_member(ca, c, i) {
- ret = bch2_prio_read(ca);
- if (ret) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
- }
-
- for (id = 0; id < BTREE_ID_NR; id++) {
- unsigned level;
- struct bkey_i *k;
-
- err = "bad btree root";
- k = bch2_journal_find_btree_root(c, j, id, &level);
- if (!k && id == BTREE_ID_EXTENTS)
- goto err;
- if (!k) {
- pr_debug("missing btree root: %d", id);
- continue;
- }
-
- err = "error reading btree root";
- if (bch2_btree_root_read(c, id, k, level))
- goto err;
- }
-
- bch_verbose(c, "starting mark and sweep:");
-
- err = "error in recovery";
- ret = bch2_initial_gc(c, &journal);
- if (ret)
- goto err;
-
- if (c->opts.noreplay)
- goto recovery_done;
-
- bch_verbose(c, "mark and sweep done");
-
- /*
- * 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;
-
- for_each_rw_member(ca, c, i)
- if (ca->need_prio_write) {
- ret = bch2_prio_write(ca);
- if (ret) {
- percpu_ref_put(&ca->io_ref);
- goto err;
- }
- }
-
- bch_verbose(c, "fsck done");
- } else {
- struct bch_inode_unpacked inode;
- struct bkey_inode_buf packed_inode;
- struct closure cl;
-
- closure_init_stack(&cl);
-
- bch_notice(c, "initializing new filesystem");
-
- 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;
- }
- /*
- * 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;
- }
+ ret = bch2_sb_members_v2_init(c);
+ if (ret) {
+ mutex_unlock(&c->sb_lock);
+ goto err;
+ }
- err = "cannot allocate new btree root";
- for (id = 0; id < BTREE_ID_NR; id++)
- if (bch2_btree_root_alloc(c, id, &cl)) {
- closure_sync(&cl);
- goto err;
- }
+ for_each_online_member(c, ca)
+ bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
- /* Wait for new btree roots to be written: */
- closure_sync(&cl);
+ mutex_unlock(&c->sb_lock);
- bch2_inode_init(c, &inode, 0, 0,
- S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0);
- inode.inum = BCACHE_ROOT_INO;
+ for_each_rw_member(c, ca)
+ bch2_dev_allocator_add(c, ca);
+ bch2_recalc_capacity(c);
- 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";
- if (bch2_fs_init_fault("fs_start"))
+ if (bch2_fs_init_fault("fs_start")) {
+ bch_err(c, "fs_start fault injected");
+ ret = -EINVAL;
goto err;
+ }
+
+ set_bit(BCH_FS_started, &c->flags);
if (c->opts.read_only) {
bch2_fs_read_only(c);
} else {
- err = bch2_fs_read_write(c);
- if (err)
+ 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);
- c->disk_sb->version = BCACHE_SB_VERSION_CDEV;
-
- bch2_write_super(c);
- mutex_unlock(&c->sb_lock);
-
- err = NULL;
-out:
- bch2_journal_entries_free(&journal);
- return err;
+ ret = 0;
err:
- switch (ret) {
- 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");
- 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");
- 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";;
- break;
- case -ENOMEM:
- err = "cannot allocate memory";
- break;
- case -EIO:
- err = "IO error";
- break;
- }
-
- BUG_ON(!err);
- set_bit(BCH_FS_ERROR, &c->flags);
- goto out;
-}
-
-const char *bch2_fs_start(struct bch_fs *c)
-{
- return __bch2_fs_start(c) ?: bch2_fs_online(c);
+ if (ret)
+ bch_err_msg(c, ret, "starting filesystem");
+ else
+ bch_verbose(c, "done starting filesystem");
+ up_write(&c->state_lock);
+ return ret;
}
-static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
+static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
{
- struct bch_sb_field_members *sb_mi;
+ struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
- sb_mi = bch2_sb_get_members(sb);
- if (!sb_mi)
- return "Invalid superblock: member info area missing";
+ if (le16_to_cpu(sb->block_size) != block_sectors(c))
+ return -BCH_ERR_mismatched_block_size;
- if (le16_to_cpu(sb->block_size) != c->sb.block_size)
- return "mismatched block size";
+ if (le16_to_cpu(m.bucket_size) <
+ BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
+ return -BCH_ERR_bucket_size_too_small;
- if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
- BCH_SB_BTREE_NODE_SIZE(c->disk_sb))
- return "new cache bucket size is too small";
-
- return NULL;
+ return 0;
}
-static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
+static int bch2_dev_in_fs(struct bch_sb_handle *fs,
+ struct bch_sb_handle *sb)
{
- struct bch_sb *newest =
- le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
- struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
+ if (fs == sb)
+ return 0;
+
+ if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
+ return -BCH_ERR_device_not_a_member_of_filesystem;
- if (uuid_le_cmp(fs->uuid, sb->uuid))
- return "device not a member of filesystem";
+ if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
+ return -BCH_ERR_device_has_been_removed;
- if (sb->dev_idx >= newest->nr_devices)
- return "device has invalid dev_idx";
+ if (fs->sb->block_size != sb->sb->block_size)
+ return -BCH_ERR_mismatched_block_size;
- if (bch2_is_zero(mi->members[sb->dev_idx].uuid.b, sizeof(uuid_le)))
- return "device has been removed";
+ if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
+ le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
+ return 0;
- if (fs->block_size != sb->block_size)
- return "mismatched block size";
+ if (fs->sb->seq == sb->sb->seq &&
+ fs->sb->write_time != sb->sb->write_time) {
+ struct printbuf buf = PRINTBUF;
+
+ prt_str(&buf, "Split brain detected between ");
+ prt_bdevname(&buf, sb->bdev);
+ prt_str(&buf, " and ");
+ prt_bdevname(&buf, fs->bdev);
+ prt_char(&buf, ':');
+ prt_newline(&buf);
+ prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
+ prt_newline(&buf);
+
+ prt_bdevname(&buf, fs->bdev);
+ prt_char(&buf, ' ');
+ bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
+ prt_newline(&buf);
+
+ prt_bdevname(&buf, sb->bdev);
+ prt_char(&buf, ' ');
+ bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
+ prt_newline(&buf);
+
+ prt_printf(&buf, "Not using older sb");
+
+ pr_err("%s", buf.buf);
+ printbuf_exit(&buf);
+ return -BCH_ERR_device_splitbrain;
+ }
- return NULL;
+ struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
+ u64 seq_from_fs = le64_to_cpu(m.seq);
+ u64 seq_from_member = le64_to_cpu(sb->sb->seq);
+
+ if (seq_from_fs && seq_from_fs < seq_from_member) {
+ struct printbuf buf = PRINTBUF;
+
+ prt_str(&buf, "Split brain detected between ");
+ prt_bdevname(&buf, sb->bdev);
+ prt_str(&buf, " and ");
+ prt_bdevname(&buf, fs->bdev);
+ prt_char(&buf, ':');
+ prt_newline(&buf);
+
+ prt_bdevname(&buf, fs->bdev);
+ prt_str(&buf, "believes seq of ");
+ prt_bdevname(&buf, sb->bdev);
+ prt_printf(&buf, " to be %llu, but ", seq_from_fs);
+ prt_bdevname(&buf, sb->bdev);
+ prt_printf(&buf, " has %llu\n", seq_from_member);
+ prt_str(&buf, "Not using ");
+ prt_bdevname(&buf, sb->bdev);
+
+ pr_err("%s", buf.buf);
+ printbuf_exit(&buf);
+ return -BCH_ERR_device_splitbrain;
+ }
+
+ return 0;
}
/* Device startup/shutdown: */
static void bch2_dev_free(struct bch_dev *ca)
{
- unsigned i;
-
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);
bch2_free_super(&ca->disk_sb);
bch2_dev_journal_exit(ca);
- free_percpu(ca->sectors_written);
+ free_percpu(ca->io_done);
bioset_exit(&ca->replica_set);
- free_percpu(ca->usage_percpu);
- free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
- kfree(ca->prio_buckets);
- kfree(ca->bio_prio);
- vfree(ca->buckets);
- vfree(ca->oldest_gens);
- free_heap(&ca->heap);
- free_fifo(&ca->free_inc);
-
- for (i = 0; i < RESERVE_NR; i++)
- free_fifo(&ca->free[i]);
+ bch2_dev_buckets_free(ca);
+ free_page((unsigned long) ca->sb_read_scratch);
+
+ 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);
- __bch2_dev_read_only(ca->fs, ca);
+ if (percpu_ref_is_zero(&ca->io_ref))
+ return;
- reinit_completion(&ca->offline_complete);
+ __bch2_dev_read_only(c, ca);
+
+ 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 =
- &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");
}
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;
if (!ca->kobj.state_in_sysfs) {
- ret = kobject_add(&ca->kobj, &ca->fs->kobj,
+ ret = kobject_add(&ca->kobj, &c->kobj,
"dev-%u", ca->dev_idx);
if (ret)
return ret;
}
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;
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;
struct bch_dev *ca;
-
- if (bch2_fs_init_fault("dev_alloc"))
- return -ENOMEM;
+ unsigned i;
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);
- spin_lock_init(&ca->self.lock);
- ca->self.nr = 1;
- rcu_assign_pointer(ca->self.d[0].dev, ca);
- ca->dev_idx = dev_idx;
+ init_rwsem(&ca->bucket_lock);
- spin_lock_init(&ca->freelist_lock);
- spin_lock_init(&ca->prio_buckets_lock);
- mutex_init(&ca->heap_lock);
- mutex_init(&ca->prio_write_lock);
- bch2_dev_moving_gc_init(ca);
+ INIT_WORK(&ca->io_error_work, bch2_io_error_work);
- INIT_WORK(&ca->io_error_work, bch2_nonfatal_io_error_work);
+ bch2_time_stats_init(&ca->io_latency[READ]);
+ bch2_time_stats_init(&ca->io_latency[WRITE]);
- if (bch2_fs_init_fault("dev_alloc"))
- goto err;
+ ca->mi = bch2_mi_to_cpu(member);
- member = bch2_sb_get_members(c->disk_sb)->members + dev_idx;
+ for (i = 0; i < ARRAY_SIZE(member->errors); i++)
+ atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
- ca->mi = bch2_mi_to_cpu(member);
ca->uuid = member->uuid;
- ca->bucket_bits = ilog2(ca->mi.bucket_size);
- scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
- /* 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;
+ ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
+ ca->mi.bucket_size / btree_sectors(c));
- if (percpu_ref_init(&ca->ref, bch2_dev_ref_release,
+ if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
0, GFP_KERNEL) ||
- percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_release,
+ percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
- !init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
- !init_fifo(&ca->free[RESERVE_BTREE], BTREE_NODE_RESERVE, 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->heap, heap_size, GFP_KERNEL) ||
- !(ca->oldest_gens = vzalloc(sizeof(u8) *
- ca->mi.nbuckets)) ||
- !(ca->buckets = vzalloc(sizeof(struct bucket) *
- ca->mi.nbuckets)) ||
- !(ca->prio_buckets = kzalloc(sizeof(u64) * prio_buckets(ca) *
- 2, GFP_KERNEL)) ||
- !(ca->disk_buckets = alloc_bucket_pages(GFP_KERNEL, ca)) ||
- !(ca->usage_percpu = alloc_percpu(struct bch_dev_usage)) ||
- !(ca->bio_prio = bio_kmalloc(GFP_NOIO, bucket_pages(ca))) ||
+ !(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)) ||
- !(ca->sectors_written = alloc_percpu(*ca->sectors_written)))
+ offsetof(struct bch_write_bio, bio), 0) ||
+ !(ca->io_done = alloc_percpu(*ca->io_done)))
goto err;
- ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
-
- total_reserve = ca->free_inc.size;
- for (i = 0; i < RESERVE_NR; i++)
- total_reserve += ca->free[i].size;
+ return ca;
+err:
+ bch2_dev_free(ca);
+ return NULL;
+}
- ca->copygc_write_point.group = &ca->self;
- ca->tiering_write_point.group = &ca->self;
+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->fs = c;
rcu_assign_pointer(c->devs[ca->dev_idx], ca);
- if (bch2_dev_sysfs_online(ca))
+ if (bch2_dev_sysfs_online(c, ca))
pr_warn("error creating sysfs objects");
-
- return 0;
-err:
- bch2_dev_free(ca);
- return -ENOMEM;
}
-static int __bch2_dev_online(struct bch_fs *c, struct bcache_superblock *sb)
+static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
{
- struct bch_dev *ca;
- int ret;
+ struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
+ struct bch_dev *ca = NULL;
+ int ret = 0;
- lockdep_assert_held(&c->sb_lock);
+ if (bch2_fs_init_fault("dev_alloc"))
+ goto err;
- if (le64_to_cpu(sb->sb->seq) >
- le64_to_cpu(c->disk_sb->seq))
- bch2_sb_to_fs(c, sb->sb);
+ ca = __bch2_dev_alloc(c, &member);
+ if (!ca)
+ goto err;
- BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
- !c->devs[sb->sb->dev_idx]);
+ ca->fs = c;
- ca = c->devs[sb->sb->dev_idx];
- if (ca->disk_sb.bdev) {
- bch_err(c, "already have device online in slot %u",
+ bch2_dev_attach(c, ca, dev_idx);
+ return ret;
+err:
+ if (ca)
+ bch2_dev_free(ca);
+ return -BCH_ERR_ENOMEM_dev_alloc;
+}
+
+static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
+{
+ unsigned ret;
+
+ if (bch2_dev_is_online(ca)) {
+ bch_err(ca, "already have device online in slot %u",
sb->sb->dev_idx);
- return -EINVAL;
+ return -BCH_ERR_device_already_online;
}
+ if (get_capacity(sb->bdev->bd_disk) <
+ ca->mi.bucket_size * ca->mi.nbuckets) {
+ bch_err(ca, "cannot online: device too small");
+ return -BCH_ERR_device_size_too_small;
+ }
+
+ BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
+
ret = bch2_dev_journal_init(ca, sb->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);
-
/* Commit: */
ca->disk_sb = *sb;
- if (sb->mode & FMODE_EXCL)
- ca->disk_sb.bdev->bd_holder = ca;
memset(sb, 0, sizeof(*sb));
- 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(ca->fs, ca);
- lg_local_unlock(&c->usage_lock);
+ ca->dev = ca->disk_sb.bdev->bd_dev;
percpu_ref_reinit(&ca->io_ref);
+
return 0;
}
-/* Device management: */
-
-bool bch2_fs_may_start(struct bch_fs *c, int flags)
+static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
{
- struct bch_sb_field_members *mi;
- unsigned meta_missing = 0;
- unsigned data_missing = 0;
- bool degraded = false;
- unsigned i;
+ struct bch_dev *ca;
+ int ret;
- mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
-
- for (i = 0; i < c->disk_sb->nr_devices; i++)
- if (!c->devs[i] &&
- !bch2_is_zero(mi->members[i].uuid.b, sizeof(uuid_le))) {
- degraded = true;
- if (BCH_MEMBER_HAS_METADATA(&mi->members[i]))
- meta_missing++;
- if (BCH_MEMBER_HAS_DATA(&mi->members[i]))
- data_missing++;
- }
- mutex_unlock(&c->sb_lock);
+ lockdep_assert_held(&c->state_lock);
- if (degraded &&
- !(flags & BCH_FORCE_IF_DEGRADED))
- return false;
+ if (le64_to_cpu(sb->sb->seq) >
+ le64_to_cpu(c->disk_sb.sb->seq))
+ bch2_sb_to_fs(c, sb->sb);
- if (meta_missing &&
- !(flags & BCH_FORCE_IF_METADATA_DEGRADED))
- return false;
+ BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
+ !c->devs[sb->sb->dev_idx]);
- if (meta_missing >= BCH_SB_META_REPLICAS_HAVE(c->disk_sb) &&
- !(flags & BCH_FORCE_IF_METADATA_LOST))
- return false;
+ ca = bch_dev_locked(c, sb->sb->dev_idx);
- if (data_missing && !(flags & BCH_FORCE_IF_DATA_DEGRADED))
- return false;
+ ret = __bch2_dev_attach_bdev(ca, sb);
+ if (ret)
+ return ret;
- if (data_missing >= BCH_SB_DATA_REPLICAS_HAVE(c->disk_sb) &&
- !(flags & BCH_FORCE_IF_DATA_LOST))
- return false;
+ bch2_dev_sysfs_online(c, ca);
- return true;
+ struct printbuf name = PRINTBUF;
+ prt_bdevname(&name, ca->disk_sb.bdev);
+
+ if (c->sb.nr_devices == 1)
+ strlcpy(c->name, name.buf, sizeof(c->name));
+ strlcpy(ca->name, name.buf, sizeof(ca->name));
+
+ printbuf_exit(&name);
+
+ rebalance_wakeup(c);
+ return 0;
}
+/* Device management: */
+
/*
* Note: this function is also used by the error paths - when a particular
* device sees an error, we call it to determine whether we can just set the
bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
enum bch_member_state new_state, int flags)
{
+ struct bch_devs_mask new_online_devs;
+ int nr_rw = 0, required;
+
lockdep_assert_held(&c->state_lock);
- if (new_state == BCH_MEMBER_STATE_RW)
+ switch (new_state) {
+ case BCH_MEMBER_STATE_rw:
return true;
+ 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(c, ca2)
+ 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.metadata_replicas_required,
+ !(flags & BCH_FORCE_IF_DATA_DEGRADED)
+ ? c->opts.data_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)
+ 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);
+
+ return bch2_have_enough_devs(c, new_online_devs, flags, false);
+ default:
+ BUG();
+ }
+}
- if (ca->mi.state == BCH_MEMBER_STATE_FAILED)
- return true;
+static bool bch2_fs_may_start(struct bch_fs *c)
+{
+ struct bch_dev *ca;
+ unsigned i, flags = 0;
- /*
- * If the device is already offline - whatever is going on with it can't
- * possible make the FS need to go RO:
- */
- if (!bch2_dev_is_online(ca))
- return true;
+ if (c->opts.very_degraded)
+ flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
- if (ca->mi.has_data &&
- !(flags & BCH_FORCE_IF_DATA_DEGRADED))
- return false;
+ if (c->opts.degraded)
+ flags |= BCH_FORCE_IF_DEGRADED;
- if (ca->mi.has_data &&
- c->sb.data_replicas_have <= 1 &&
- !(flags & BCH_FORCE_IF_DATA_LOST))
- return false;
+ if (!c->opts.degraded &&
+ !c->opts.very_degraded) {
+ mutex_lock(&c->sb_lock);
- if (ca->mi.has_metadata &&
- !(flags & BCH_FORCE_IF_METADATA_DEGRADED))
- return false;
+ for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
+ if (!bch2_dev_exists(c->disk_sb.sb, i))
+ continue;
- if (ca->mi.has_metadata &&
- c->sb.meta_replicas_have <= 1 &&
- !(flags & BCH_FORCE_IF_METADATA_LOST))
- return false;
+ ca = bch_dev_locked(c, i);
- return true;
+ 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);
+ }
+
+ 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_moving_gc_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_group_remove(&c->journal.devs, 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)
+static void __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);
-
- if (bch2_dev_allocator_start(ca))
- return "error starting allocator thread";
-
- if (bch2_moving_gc_start(ca))
- return "error starting moving GC thread";
+ BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
- if (bch2_tiering_start(c))
- return "error starting tiering thread";
-
- return NULL;
+ bch2_dev_allocator_add(c, ca);
+ bch2_recalc_capacity(c);
}
int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
enum bch_member_state new_state, int flags)
{
- struct bch_sb_field_members *mi;
+ struct bch_member *m;
+ int ret = 0;
if (ca->mi.state == new_state)
return 0;
if (!bch2_dev_state_allowed(c, ca, new_state, flags))
- return -EINVAL;
+ return -BCH_ERR_device_state_not_allowed;
- 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]);
+ bch_notice(ca, "%s", bch2_member_states[new_state]);
mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
- SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
+ m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
+ SET_BCH_MEMBER_STATE(m, 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);
+
+ 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: */
+static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
+{
+ struct bpos start = POS(ca->dev_idx, 0);
+ struct bpos end = POS(ca->dev_idx, U64_MAX);
+ int ret;
+
+ /*
+ * We clear the LRU and need_discard btrees first so that we don't race
+ * with bch2_do_invalidates() and bch2_do_discards()
+ */
+ ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
+ BTREE_TRIGGER_NORUN, NULL) ?:
+ bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
+ BTREE_TRIGGER_NORUN, NULL) ?:
+ bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
+ BTREE_TRIGGER_NORUN, NULL) ?:
+ bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
+ BTREE_TRIGGER_NORUN, NULL) ?:
+ bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
+ BTREE_TRIGGER_NORUN, NULL) ?:
+ bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
+ BTREE_TRIGGER_NORUN, NULL);
+ bch_err_msg(c, ret, "removing dev alloc info");
+ return ret;
+}
+
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;
- int ret = -EINVAL;
+ struct bch_member *m;
+ unsigned dev_idx = ca->dev_idx, data;
+ int ret;
- 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");
+ ret = -BCH_ERR_device_state_not_allowed;
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);
+ bch_err_msg(ca, ret, "dropping data");
+ if (ret)
goto err;
- }
- /*
- * XXX: verify that dev_idx is really not in use anymore, anywhere
- *
- * flag_data_bad() does not check btree pointers
- */
- ret = bch2_flag_data_bad(ca);
- if (ret) {
- bch_err(ca, "Remove failed");
+ ret = bch2_dev_remove_alloc(c, ca);
+ bch_err_msg(ca, ret, "deleting alloc info");
+ if (ret)
goto err;
- }
- if (ca->mi.has_data || ca->mi.has_metadata) {
- bch_err(ca, "Remove failed, still has data");
+ ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
+ bch_err_msg(ca, ret, "flushing journal");
+ if (ret)
+ goto err;
+
+ ret = bch2_journal_flush(&c->journal);
+ bch_err(ca, "journal error");
+ if (ret)
+ goto err;
+
+ ret = bch2_replicas_gc2(c);
+ bch_err_msg(ca, ret, "in replicas_gc2()");
+ if (ret)
+ goto err;
+
+ data = bch2_dev_has_data(c, ca);
+ if (data) {
+ struct printbuf data_has = PRINTBUF;
+
+ prt_bitflags(&data_has, bch2_data_types, data);
+ bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
+ printbuf_exit(&data_has);
+ ret = -EBUSY;
goto err;
}
- /*
- * Ok, really doing the remove:
- * Drop device's prio pointer before removing it from superblock:
- */
- spin_lock(&c->journal.lock);
- c->journal.prio_buckets[dev_idx] = 0;
- spin_unlock(&c->journal.lock);
+ __bch2_dev_offline(c, ca);
- bch2_journal_meta(&c->journal);
+ 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);
+ /*
+ * At this point the device object has been removed in-core, but the
+ * on-disk journal might still refer to the device index via sb device
+ * usage entries. Recovery fails if it sees usage information for an
+ * invalid device. Flush journal pins to push the back of the journal
+ * past now invalid device index references before we update the
+ * superblock, but after the device object has been removed so any
+ * further journal writes elide usage info for the device.
+ */
+ bch2_journal_flush_all_pins(&c->journal);
+
/*
* Free this device's slot in the bch_member array - all pointers to
* this device must be gone:
*/
mutex_lock(&c->sb_lock);
- mi = bch2_sb_get_members(c->disk_sb);
- memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
+ m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
+ memset(&m->uuid, 0, sizeof(m->uuid));
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- mutex_unlock(&c->state_lock);
+ up_write(&c->state_lock);
+
+ bch2_dev_usage_journal_reserve(c);
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;
}
+/* Add new device to running filesystem: */
int bch2_dev_add(struct bch_fs *c, const char *path)
{
- struct bcache_superblock sb;
- const char *err;
+ struct bch_opts opts = bch2_opts_empty();
+ struct bch_sb_handle sb;
struct bch_dev *ca = NULL;
- struct bch_sb_field_members *mi, *dev_mi;
- struct bch_member saved_mi;
+ struct bch_sb_field_members_v2 *mi;
+ struct bch_member dev_mi;
unsigned dev_idx, nr_devices, u64s;
- int ret = -EINVAL;
+ struct printbuf errbuf = PRINTBUF;
+ struct printbuf label = PRINTBUF;
+ int ret;
- err = bch2_read_super(&sb, bch2_opts_empty(), path);
- if (err)
- return -EINVAL;
+ ret = bch2_read_super(path, &opts, &sb);
+ bch_err_msg(c, ret, "reading super");
+ if (ret)
+ goto err;
- err = bch2_validate_cache_super(&sb);
- if (err)
- return -EINVAL;
+ dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
- err = bch2_dev_may_add(sb.sb, c);
- if (err)
- return -EINVAL;
+ if (BCH_MEMBER_GROUP(&dev_mi)) {
+ bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
+ if (label.allocation_failure) {
+ ret = -ENOMEM;
+ goto err;
+ }
+ }
+
+ ret = bch2_dev_may_add(sb.sb, c);
+ if (ret)
+ goto err;
- mutex_lock(&c->state_lock);
+ ca = __bch2_dev_alloc(c, &dev_mi);
+ if (!ca) {
+ ret = -ENOMEM;
+ goto err;
+ }
+
+ bch2_dev_usage_init(ca);
+
+ ret = __bch2_dev_attach_bdev(ca, &sb);
+ if (ret)
+ goto err;
+
+ ret = bch2_dev_journal_alloc(ca);
+ bch_err_msg(c, ret, "allocating journal");
+ if (ret)
+ goto err;
+
+ down_write(&c->state_lock);
mutex_lock(&c->sb_lock);
- /*
- * Preserve the old cache member information (esp. tier)
- * before we start bashing the disk stuff.
- */
- 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());
+ ret = bch2_sb_from_fs(c, ca);
+ bch_err_msg(c, ret, "setting up new superblock");
+ if (ret)
+ goto err_unlock;
if (dynamic_fault("bcachefs:add:no_slot"))
goto no_slot;
- mi = bch2_sb_get_members(c->disk_sb);
for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
- if (dev_idx >= c->sb.nr_devices ||
- bch2_is_zero(mi->members[dev_idx].uuid.b,
- sizeof(uuid_le)))
+ if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
goto have_slot;
no_slot:
- err = "no slots available in superblock";
- ret = -ENOSPC;
+ ret = -BCH_ERR_ENOSPC_sb_members;
+ bch_err_msg(c, ret, "setting up new superblock");
goto err_unlock;
have_slot:
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";
- mi = bch2_fs_sb_resize_members(c, u64s);
- if (!mi)
- goto err_unlock;
+ mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
+ u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
+ le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
- dev_mi = bch2_sb_resize_members(&sb, u64s);
- if (!dev_mi)
+ mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
+ if (!mi) {
+ ret = -BCH_ERR_ENOSPC_sb_members;
+ bch_err_msg(c, ret, "setting up new superblock");
goto err_unlock;
+ }
+ struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
- 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;
+ *m = dev_mi;
+ m->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;
+ if (BCH_MEMBER_GROUP(&dev_mi)) {
+ ret = __bch2_dev_group_set(c, ca, label.buf);
+ bch_err_msg(c, ret, "creating new label");
+ if (ret)
+ goto err_unlock;
}
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;
+ bch2_dev_usage_journal_reserve(c);
- err = __bch2_dev_read_write(c, ca);
- if (err)
- goto err;
- }
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ bch_err_msg(ca, ret, "marking new superblock");
+ if (ret)
+ goto err_late;
+
+ ret = bch2_fs_freespace_init(c);
+ bch_err_msg(ca, ret, "initializing free space");
+ if (ret)
+ goto err_late;
- mutex_unlock(&c->state_lock);
+ ca->new_fs_bucket_idx = 0;
+
+ if (ca->mi.state == BCH_MEMBER_STATE_rw)
+ __bch2_dev_read_write(c, ca);
+
+ 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;
+ printbuf_exit(&label);
+ printbuf_exit(&errbuf);
+ bch_err_fn(c, ret);
+ return ret;
+err_late:
+ up_write(&c->state_lock);
+ ca = NULL;
+ goto err;
}
+/* Hot add existing device to running filesystem: */
int bch2_dev_online(struct bch_fs *c, const char *path)
{
- struct bcache_superblock sb = { 0 };
+ struct bch_opts opts = bch2_opts_empty();
+ struct bch_sb_handle sb = { NULL };
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(&sb, bch2_opts_empty(), path);
- 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);
- if (err)
+ ret = bch2_dev_in_fs(&c->disk_sb, &sb);
+ bch_err_msg(c, ret, "bringing %s online", path);
+ if (ret)
goto err;
- mutex_lock(&c->sb_lock);
- if (__bch2_dev_online(c, &sb)) {
- err = "__bch2_dev_online() error";
- mutex_unlock(&c->sb_lock);
+ ret = bch2_dev_attach_bdev(c, &sb);
+ if (ret)
goto err;
- }
- mutex_unlock(&c->sb_lock);
- ca = c->devs[dev_idx];
- ret = bch2_prio_read(ca);
- if (ret) {
- err = "error reading priorities";
+ ca = bch_dev_locked(c, dev_idx);
+
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
+ if (ret)
goto err;
+
+ if (ca->mi.state == BCH_MEMBER_STATE_rw)
+ __bch2_dev_read_write(c, ca);
+
+ if (!ca->mi.freespace_initialized) {
+ ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
+ bch_err_msg(ca, ret, "initializing free space");
+ if (ret)
+ goto err;
}
- if (ca->mi.state == BCH_MEMBER_STATE_RW) {
- err = __bch2_dev_read_write(c, ca);
- if (err)
+ if (!ca->journal.nr) {
+ ret = bch2_dev_journal_alloc(ca);
+ bch_err_msg(ca, ret, "allocating journal");
+ if (ret)
goto err;
}
- mutex_unlock(&c->state_lock);
+ mutex_lock(&c->sb_lock);
+ bch2_members_v2_get_mut(c->disk_sb.sb, 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;
+ return ret;
}
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_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
+ 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);
- return -EINVAL;
+ up_write(&c->state_lock);
+ return -BCH_ERR_device_state_not_allowed;
}
- __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)
{
- int ret;
+ struct bch_member *m;
+ u64 old_nbuckets;
+ int ret = 0;
- mutex_lock(&c->state_lock);
+ down_write(&c->state_lock);
+ old_nbuckets = ca->mi.nbuckets;
- 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 = -BCH_ERR_device_size_too_small;
+ goto err;
}
- ret = bch2_move_metadata_off_device(ca);
- if (ret) {
- bch_err(ca, "Error migrating metadata: %i", ret);
- return ret;
- }
+ ret = bch2_dev_buckets_resize(c, ca, nbuckets);
+ bch_err_msg(ca, ret, "resizing buckets");
+ if (ret)
+ goto err;
- if (ca->mi.has_data || ca->mi.has_metadata) {
- bch_err(ca, "Migrate error: data still present");
- return -EINVAL;
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret)
+ goto err;
+
+ mutex_lock(&c->sb_lock);
+ m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
+ m->nbuckets = cpu_to_le64(nbuckets);
+
+ bch2_write_super(c);
+ mutex_unlock(&c->sb_lock);
+
+ if (ca->mi.freespace_initialized) {
+ ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
+ if (ret)
+ goto err;
+
+ /*
+ * XXX: this is all wrong transactionally - we'll be able to do
+ * this correctly after the disk space accounting rewrite
+ */
+ ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
}
- return 0;
+ 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 *name)
+{
+ rcu_read_lock();
+ for_each_member_device_rcu(c, ca, NULL)
+ if (!strcmp(name, ca->name)) {
+ rcu_read_unlock();
+ return ca;
+ }
+ rcu_read_unlock();
+ return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
}
/* Filesystem open: */
-const char *bch2_fs_open(char * const *devices, unsigned nr_devices,
- struct bch_opts opts, struct bch_fs **ret)
+static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
{
- const char *err;
- struct bch_fs *c = NULL;
- struct bcache_superblock *sb;
- unsigned i, best_sb = 0;
+ return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
+ cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
+}
- if (!nr_devices)
- return "need at least one device";
+struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
+ struct bch_opts opts)
+{
+ DARRAY(struct bch_sb_handle) sbs = { 0 };
+ struct bch_fs *c = NULL;
+ struct bch_sb_handle *best = NULL;
+ struct printbuf errbuf = PRINTBUF;
+ int ret = 0;
if (!try_module_get(THIS_MODULE))
- return "module unloading";
+ return ERR_PTR(-ENODEV);
- err = "cannot allocate memory";
- sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
- if (!sb)
+ if (!nr_devices) {
+ ret = -EINVAL;
goto err;
+ }
- for (i = 0; i < nr_devices; i++) {
- err = bch2_read_super(&sb[i], opts, devices[i]);
- if (err)
- goto err;
+ ret = darray_make_room(&sbs, nr_devices);
+ if (ret)
+ goto err;
- err = "attempting to register backing device";
- if (__SB_IS_BDEV(le64_to_cpu(sb[i].sb->version)))
- goto err;
+ for (unsigned i = 0; i < nr_devices; i++) {
+ struct bch_sb_handle sb = { NULL };
- err = bch2_validate_cache_super(&sb[i]);
- if (err)
+ ret = bch2_read_super(devices[i], &opts, &sb);
+ if (ret)
goto err;
- }
- for (i = 1; i < nr_devices; i++)
- if (le64_to_cpu(sb[i].sb->seq) >
- le64_to_cpu(sb[best_sb].sb->seq))
- best_sb = i;
+ BUG_ON(darray_push(&sbs, sb));
+ }
- for (i = 0; i < nr_devices; i++) {
- err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
- if (err)
- goto err;
+ if (opts.nochanges && !opts.read_only) {
+ ret = -BCH_ERR_erofs_nochanges;
+ goto err_print;
}
- err = "cannot allocate memory";
- c = bch2_fs_alloc(sb[best_sb].sb, opts);
- if (!c)
- goto err;
+ darray_for_each(sbs, sb)
+ if (!best || sb_cmp(sb->sb, best->sb) > 0)
+ best = sb;
- err = "bch2_dev_online() error";
- mutex_lock(&c->sb_lock);
- for (i = 0; i < nr_devices; i++)
- if (__bch2_dev_online(c, &sb[i])) {
- mutex_unlock(&c->sb_lock);
- goto err;
- }
- mutex_unlock(&c->sb_lock);
+ darray_for_each_reverse(sbs, sb) {
+ ret = bch2_dev_in_fs(best, sb);
- err = "insufficient devices";
- if (!bch2_fs_may_start(c, 0))
- goto err;
+ if (ret == -BCH_ERR_device_has_been_removed ||
+ ret == -BCH_ERR_device_splitbrain) {
+ bch2_free_super(sb);
+ darray_remove_item(&sbs, sb);
+ best -= best > sb;
+ ret = 0;
+ continue;
+ }
- if (!c->opts.nostart) {
- err = __bch2_fs_start(c);
- if (err)
- goto err;
+ if (ret)
+ goto err_print;
}
- err = bch2_fs_online(c);
- if (err)
- goto err;
-
+ c = bch2_fs_alloc(best->sb, opts);
+ ret = PTR_ERR_OR_ZERO(c);
if (ret)
- *ret = c;
- else
- closure_put(&c->cl);
-
- err = NULL;
-out:
- kfree(sb);
- module_put(THIS_MODULE);
- if (err)
- c = NULL;
- return err;
-err:
- if (c)
- bch2_fs_stop(c);
-
- for (i = 0; i < nr_devices; i++)
- bch2_free_super(&sb[i]);
- goto out;
-}
-
-static const char *__bch2_fs_open_incremental(struct bcache_superblock *sb,
- struct bch_opts opts)
-{
- const char *err;
- struct bch_fs *c;
- bool allocated_fs = false;
-
- err = bch2_validate_cache_super(sb);
- if (err)
- return err;
-
- mutex_lock(&bch_fs_list_lock);
- c = __bch2_uuid_to_fs(sb->sb->uuid);
- if (c) {
- closure_get(&c->cl);
+ goto err;
- err = bch2_dev_in_fs(c->disk_sb, sb->sb);
- if (err)
- goto err;
- } else {
- c = bch2_fs_alloc(sb->sb, opts);
- err = "cannot allocate memory";
- if (!c)
+ down_write(&c->state_lock);
+ darray_for_each(sbs, sb) {
+ ret = bch2_dev_attach_bdev(c, sb);
+ if (ret) {
+ up_write(&c->state_lock);
goto err;
-
- allocated_fs = true;
+ }
}
+ up_write(&c->state_lock);
- err = "bch2_dev_online() error";
-
- mutex_lock(&c->sb_lock);
- if (__bch2_dev_online(c, sb)) {
- mutex_unlock(&c->sb_lock);
- goto err;
+ if (!bch2_fs_may_start(c)) {
+ ret = -BCH_ERR_insufficient_devices_to_start;
+ goto err_print;
}
- mutex_unlock(&c->sb_lock);
- if (!c->opts.nostart && bch2_fs_may_start(c, 0)) {
- err = __bch2_fs_start(c);
- if (err)
+ if (!c->opts.nostart) {
+ 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);
-
- return NULL;
+out:
+ darray_for_each(sbs, sb)
+ bch2_free_super(sb);
+ darray_exit(&sbs);
+ printbuf_exit(&errbuf);
+ module_put(THIS_MODULE);
+ return c;
+err_print:
+ pr_err("bch_fs_open err opening %s: %s",
+ devices[0], bch2_err_str(ret));
err:
- mutex_unlock(&bch_fs_list_lock);
-
- if (allocated_fs)
+ if (!IS_ERR_OR_NULL(c))
bch2_fs_stop(c);
- else if (c)
- closure_put(&c->cl);
-
- return err;
-}
-
-const char *bch2_fs_open_incremental(const char *path)
-{
- struct bcache_superblock sb;
- struct bch_opts opts = bch2_opts_empty();
- const char *err;
-
- err = bch2_read_super(&sb, opts, path);
- if (err)
- return err;
-
- if (!__SB_IS_BDEV(le64_to_cpu(sb.sb->version)))
- err = __bch2_fs_open_incremental(&sb, opts);
- else
- err = "not a bcachefs superblock";
-
- bch2_free_super(&sb);
-
- return err;
+ c = ERR_PTR(ret);
+ goto out;
}
/* Global interfaces/init */
bch2_debug_exit();
bch2_vfs_exit();
bch2_chardev_exit();
+ bch2_btree_key_cache_exit();
if (bcachefs_kset)
kset_unregister(bcachefs_kset);
}
bch2_bkey_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())
BCH_DEBUG_PARAMS()
#undef BCH_DEBUG_PARAM
+__maybe_unused
+static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
+module_param_named(version, bch2_metadata_version, uint, 0400);
+
module_exit(bcachefs_exit);
module_init(bcachefs_init);
projects / bcachefs-tools-debian / blobdiff