#include "btree_key_cache.h"
#include "btree_update_interior.h"
#include "btree_io.h"
+#include "buckets_waiting_for_journal.h"
#include "chardev.h"
#include "checksum.h"
#include "clock.h"
#include "rebalance.h"
#include "recovery.h"
#include "replicas.h"
+#include "subvolume.h"
#include "super.h"
#include "super-io.h"
#include "sysfs.h"
+#include "counters.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/module.h>
#include <linux/percpu.h>
+#include <linux/pretty-printers.h>
#include <linux/random.h>
#include <linux/sysfs.h>
#include <crypto/hash.h>
MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
#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);
{
struct bch_dev *ca;
unsigned i, clean_passes = 0;
+ u64 seq = 0;
bch2_rebalance_stop(c);
bch2_copygc_stop(c);
bch2_gc_thread_stop(c);
- /*
- * Flush journal before stopping allocators, because flushing journal
- * blacklist entries involves allocating new btree nodes:
- */
- bch2_journal_flush_all_pins(&c->journal);
-
- /*
- * 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, "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));
+ 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;
}
- 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);
+ bch_verbose(c, "flushing journal and stopping allocators complete");
+ 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);
- /*
- * the journal kicks off btree writes via reclaim - wait for in flight
- * writes after stopping journal:
- */
- bch2_btree_flush_all_writes(c);
-
/*
* After stopping journal:
*/
void bch2_fs_read_only(struct bch_fs *c)
{
if (!test_bit(BCH_FS_RW, &c->flags)) {
- BUG_ON(c->journal.reclaim_thread);
+ bch2_journal_reclaim_stop(&c->journal);
return;
}
/*
* 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()).
*/
percpu_ref_kill(&c->writes);
!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) &&
+ test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
!c->opts.norecovery) {
bch_verbose(c, "marking filesystem clean");
bch2_fs_mark_clean(c);
if (ret)
goto err;
- /*
- * We need to write out a journal entry before we start doing btree
- * updates, to ensure that on unclean shutdown new journal blacklist
- * entries are created:
- */
- bch2_journal_meta(&c->journal);
-
- clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
+ clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
for_each_rw_member(ca, c, i)
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;
- }
- }
-
- set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
-
- for_each_rw_member(ca, c, i)
- bch2_wake_allocator(ca);
-
- ret = bch2_journal_reclaim_start(&c->journal);
- if (ret) {
- bch_err(c, "error starting journal reclaim: %i", ret);
- return ret;
- }
+ bch2_do_discards(c);
+ bch2_do_invalidates(c);
if (!early) {
ret = bch2_fs_read_write_late(c);
for (i = 0; i < BCH_TIME_STAT_NR; i++)
bch2_time_stats_exit(&c->times[i]);
+ bch2_fs_counters_exit(c);
+ bch2_fs_snapshots_exit(c);
bch2_fs_quota_exit(c);
bch2_fs_fsio_exit(c);
bch2_fs_ec_exit(c);
bch2_fs_encryption_exit(c);
bch2_fs_io_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_io_clock_exit(&c->io_clock[READ]);
bch2_fs_compress_exit(c);
bch2_journal_keys_free(&c->journal_keys);
- bch2_journal_entries_free(&c->journal_entries);
+ bch2_journal_entries_free(c);
percpu_free_rwsem(&c->mark_lock);
- if (c->btree_iters_bufs)
+ if (c->btree_paths_bufs)
for_each_possible_cpu(cpu)
- kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
+ kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
free_percpu(c->online_reserved);
- free_percpu(c->btree_iters_bufs);
+ free_percpu(c->btree_paths_bufs);
free_percpu(c->pcpu);
mempool_exit(&c->large_bkey_pool);
mempool_exit(&c->btree_bounce_pool);
kfree(c->unused_inode_hints);
free_heap(&c->copygc_heap);
+ if (c->io_complete_wq )
+ destroy_workqueue(c->io_complete_wq );
if (c->copygc_wq)
destroy_workqueue(c->copygc_wq);
- if (c->wq)
- destroy_workqueue(c->wq);
+ if (c->btree_io_complete_wq)
+ destroy_workqueue(c->btree_io_complete_wq);
+ if (c->btree_update_wq)
+ destroy_workqueue(c->btree_update_wq);
bch2_free_super(&c->disk_sb);
kvpfree(c, sizeof(*c));
bch2_fs_debug_exit(c);
bch2_fs_chardev_exit(c);
+ kobject_put(&c->counters_kobj);
kobject_put(&c->time_stats);
kobject_put(&c->opts_dir);
kobject_put(&c->internal);
for_each_member_device(ca, c, i)
cancel_work_sync(&ca->io_error_work);
- cancel_work_sync(&c->btree_write_error_work);
cancel_work_sync(&c->read_only_work);
for (i = 0; i < c->sb.nr_devices; i++)
bch2_fs_free(c);
}
-static const char *bch2_fs_online(struct bch_fs *c)
+static int bch2_fs_online(struct bch_fs *c)
{
struct bch_dev *ca;
- const char *err = NULL;
unsigned i;
- int ret;
+ int ret = 0;
lockdep_assert_held(&bch_fs_list_lock);
- if (!list_empty(&c->list))
- return NULL;
-
- if (__bch2_uuid_to_fs(c->sb.uuid))
- return "filesystem UUID already open";
+ if (__bch2_uuid_to_fs(c->sb.uuid)) {
+ bch_err(c, "filesystem UUID already open");
+ return -EINVAL;
+ }
ret = bch2_fs_chardev_init(c);
- if (ret)
- return "error creating character device";
+ if (ret) {
+ bch_err(c, "error creating character device");
+ return ret;
+ }
bch2_fs_debug_init(c);
- if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
- kobject_add(&c->internal, &c->kobj, "internal") ||
- kobject_add(&c->opts_dir, &c->kobj, "options") ||
- kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
- bch2_opts_create_sysfs_files(&c->opts_dir))
- return "error creating sysfs objects";
+ ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
+ kobject_add(&c->internal, &c->kobj, "internal") ?:
+ kobject_add(&c->opts_dir, &c->kobj, "options") ?:
+ kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
+ 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);
- err = "error creating sysfs objects";
- for_each_member_device(ca, c, i)
- if (bch2_dev_sysfs_online(c, ca)) {
+ for_each_member_device(ca, c, i) {
+ ret = bch2_dev_sysfs_online(c, ca);
+ if (ret) {
+ bch_err(c, "error creating sysfs objects");
percpu_ref_put(&ca->ref);
goto err;
}
+ }
+ BUG_ON(!list_empty(&c->list));
list_add(&c->list, &bch_fs_list);
- err = NULL;
err:
up_write(&c->state_lock);
- return err;
+ return ret;
}
static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
{
struct bch_sb_field_members *mi;
struct bch_fs *c;
+ struct printbuf name = PRINTBUF;
unsigned i, iter_size;
- const char *err;
+ int ret = 0;
pr_verbose_init(opts, "");
c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
- if (!c)
+ if (!c) {
+ c = ERR_PTR(-ENOMEM);
goto out;
+ }
__module_get(THIS_MODULE);
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;
INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
init_rwsem(&c->gc_lock);
+ mutex_init(&c->gc_gens_lock);
for (i = 0; i < BCH_TIME_STAT_NR; i++)
bch2_time_stats_init(&c->times[i]);
bch2_fs_allocator_foreground_init(c);
bch2_fs_rebalance_init(c);
bch2_fs_quota_init(c);
+ bch2_fs_ec_init_early(c);
INIT_LIST_HEAD(&c->list);
mutex_init(&c->usage_scratch_lock);
mutex_init(&c->bio_bounce_pages_lock);
+ mutex_init(&c->snapshot_table_lock);
- bio_list_init(&c->btree_write_error_list);
spin_lock_init(&c->btree_write_error_lock);
- INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
INIT_WORK(&c->journal_seq_blacklist_gc_work,
bch2_blacklist_entries_gc);
- INIT_LIST_HEAD(&c->journal_entries);
INIT_LIST_HEAD(&c->journal_iters);
INIT_LIST_HEAD(&c->fsck_errors);
INIT_LIST_HEAD(&c->ec_stripe_new_list);
mutex_init(&c->ec_stripe_new_lock);
+ INIT_LIST_HEAD(&c->data_progress_list);
+ mutex_init(&c->data_progress_lock);
+
spin_lock_init(&c->ec_stripes_heap_lock);
seqcount_init(&c->gc_pos_lock);
c->rebalance.enabled = 1;
c->promote_whole_extents = true;
- c->journal.write_time = &c->times[BCH_TIME_journal_write];
- c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
- c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
- c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
+ c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
+ c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
+ c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
+ c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
bch2_fs_btree_cache_init_early(&c->btree_cache);
mutex_init(&c->sectors_available_lock);
- if (percpu_init_rwsem(&c->mark_lock))
+ ret = percpu_init_rwsem(&c->mark_lock);
+ if (ret)
goto err;
mutex_lock(&c->sb_lock);
+ ret = bch2_sb_to_fs(c, sb);
+ mutex_unlock(&c->sb_lock);
- if (bch2_sb_to_fs(c, sb)) {
- mutex_unlock(&c->sb_lock);
+ if (ret)
goto err;
- }
- mutex_unlock(&c->sb_lock);
+ pr_uuid(&name, c->sb.user_uuid.b);
+ strlcpy(c->name, name.buf, sizeof(c->name));
+ printbuf_exit(&name);
- scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
+ ret = name.allocation_failure ? -ENOMEM : 0;
+ if (ret)
+ goto err;
+
+ /* Compat: */
+ if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+ SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
+
+ if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+ !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
+ SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
c->opts = bch2_opts_default;
- bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
+ ret = bch2_opts_from_sb(&c->opts, sb);
+ if (ret)
+ goto err;
+
bch2_opts_apply(&c->opts, opts);
- c->block_bits = ilog2(c->opts.block_size);
+ /* key cache currently disabled for inodes, because of snapshots: */
+ c->opts.inodes_use_key_cache = 0;
+
+ 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->block_bits = ilog2(block_sectors(c));
c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
- if (bch2_fs_init_fault("fs_alloc"))
+ if (bch2_fs_init_fault("fs_alloc")) {
+ bch_err(c, "fs_alloc fault injected");
+ ret = -EFAULT;
goto err;
+ }
iter_size = sizeof(struct sort_iter) +
(btree_blocks(c) + 1) * 2 *
c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
- if (!(c->wq = alloc_workqueue("bcachefs",
+ if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
+ WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+ !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
!(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+ !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
+ WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
percpu_ref_init(&c->writes, bch2_writes_disabled,
PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
offsetof(struct btree_write_bio, wbio.bio)),
BIOSET_NEED_BVECS) ||
!(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
- !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
+ !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
!(c->online_reserved = alloc_percpu(u64)) ||
mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
btree_bytes(c)) ||
mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
!(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
- sizeof(u64), GFP_KERNEL)) ||
- bch2_io_clock_init(&c->io_clock[READ]) ||
- bch2_io_clock_init(&c->io_clock[WRITE]) ||
- bch2_fs_journal_init(&c->journal) ||
- bch2_fs_replicas_init(c) ||
- bch2_fs_btree_cache_init(c) ||
- bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
- bch2_fs_btree_iter_init(c) ||
- bch2_fs_btree_interior_update_init(c) ||
- bch2_fs_io_init(c) ||
- bch2_fs_encryption_init(c) ||
- bch2_fs_compress_init(c) ||
- bch2_fs_ec_init(c) ||
- bch2_fs_fsio_init(c))
+ sizeof(u64), GFP_KERNEL))) {
+ ret = -ENOMEM;
+ goto err;
+ }
+
+ ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
+ bch2_io_clock_init(&c->io_clock[WRITE]) ?:
+ bch2_fs_journal_init(&c->journal) ?:
+ bch2_fs_replicas_init(c) ?:
+ bch2_fs_btree_cache_init(c) ?:
+ bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
+ bch2_fs_btree_iter_init(c) ?:
+ bch2_fs_btree_interior_update_init(c) ?:
+ bch2_fs_buckets_waiting_for_journal_init(c) ?:
+ bch2_fs_subvolumes_init(c) ?:
+ bch2_fs_io_init(c) ?:
+ bch2_fs_encryption_init(c) ?:
+ bch2_fs_compress_init(c) ?:
+ bch2_fs_ec_init(c) ?:
+ bch2_fs_fsio_init(c) ?:
+ bch2_fs_counters_init(c);
+ if (ret)
goto err;
mi = bch2_sb_get_members(c->disk_sb.sb);
for (i = 0; i < c->sb.nr_devices; i++)
if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
- bch2_dev_alloc(c, i))
+ bch2_dev_alloc(c, i)) {
+ ret = -EEXIST;
goto err;
+ }
bch2_journal_entry_res_resize(&c->journal,
&c->btree_root_journal_res,
(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
mutex_lock(&bch_fs_list_lock);
- err = bch2_fs_online(c);
+ ret = bch2_fs_online(c);
mutex_unlock(&bch_fs_list_lock);
- if (err) {
- bch_err(c, "bch2_fs_online() error: %s", err);
+
+ if (ret)
goto err;
- }
out:
- pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
+ pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
return c;
err:
bch2_fs_free(c);
- c = NULL;
+ c = ERR_PTR(ret);
goto out;
}
static void print_mount_opts(struct bch_fs *c)
{
enum bch_opt_id i;
- char buf[512];
- struct printbuf p = PBUF(buf);
+ struct printbuf p = PRINTBUF;
bool first = true;
- strcpy(buf, "(null)");
-
if (c->opts.read_only) {
- pr_buf(&p, "ro");
+ prt_printf(&p, "ro");
first = false;
}
const struct bch_option *opt = &bch2_opt_table[i];
u64 v = bch2_opt_get_by_id(&c->opts, i);
- if (!(opt->mode & OPT_MOUNT))
+ if (!(opt->flags & OPT_MOUNT))
continue;
if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
continue;
if (!first)
- pr_buf(&p, ",");
+ prt_printf(&p, ",");
first = false;
- bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
+ bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
}
- bch_info(c, "mounted with opts: %s", buf);
+ if (!p.pos)
+ prt_printf(&p, "(null)");
+
+ bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
+ printbuf_exit(&p);
}
int bch2_fs_start(struct bch_fs *c)
{
- const char *err = "cannot allocate memory";
struct bch_sb_field_members *mi;
struct bch_dev *ca;
time64_t now = ktime_get_real_seconds();
if (ret)
goto err;
- err = "dynamic fault";
ret = -EINVAL;
- if (bch2_fs_init_fault("fs_start"))
+ if (bch2_fs_init_fault("fs_start")) {
+ bch_err(c, "fs_start fault injected");
goto err;
+ }
set_bit(BCH_FS_STARTED, &c->flags);
- /*
- * Allocator threads don't start filling copygc reserve until after we
- * set BCH_FS_STARTED - wake them now:
- *
- * XXX ugly hack:
- * Need to set ca->allocator_state here instead of relying on the
- * allocator threads to do it to avoid racing with the copygc threads
- * checking it and thinking they have no alloc reserve:
- */
- for_each_online_member(ca, c, i) {
- ca->allocator_state = ALLOCATOR_running;
- bch2_wake_allocator(ca);
- }
-
if (c->opts.read_only || c->opts.nochanges) {
bch2_fs_read_only(c);
} else {
- err = "error going read write";
ret = !test_bit(BCH_FS_RW, &c->flags)
? bch2_fs_read_write(c)
: bch2_fs_read_write_late(c);
case BCH_FSCK_ERRORS_NOT_FIXED:
bch_err(c, "filesystem contains errors: please report this to the developers");
pr_cont("mount with -o fix_errors to repair\n");
- err = "fsck error";
break;
case BCH_FSCK_REPAIR_UNIMPLEMENTED:
bch_err(c, "filesystem contains errors: please report this to the developers");
pr_cont("repair unimplemented: inform the developers so that it can be added\n");
- err = "fsck error";
break;
case BCH_FSCK_REPAIR_IMPOSSIBLE:
bch_err(c, "filesystem contains errors, but repair impossible");
- err = "fsck error";
break;
case BCH_FSCK_UNKNOWN_VERSION:
- err = "unknown metadata version";;
+ bch_err(c, "unknown metadata version");
break;
case -ENOMEM:
- err = "cannot allocate memory";
+ bch_err(c, "cannot allocate memory");
break;
case -EIO:
- err = "IO error";
+ bch_err(c, "IO error");
break;
}
if (!sb_mi)
return "Invalid superblock: member info area missing";
- if (le16_to_cpu(sb->block_size) != c->opts.block_size)
+ if (le16_to_cpu(sb->block_size) != block_sectors(c))
return "mismatched block size";
if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
static void bch2_dev_free(struct bch_dev *ca)
{
- bch2_dev_allocator_stop(ca);
-
cancel_work_sync(&ca->io_error_work);
if (ca->kobj.state_in_sysfs &&
ca->mi = bch2_mi_to_cpu(member);
ca->uuid = member->uuid;
- if (opt_defined(c->opts, discard))
- ca->mi.discard = opt_get(c->opts, discard);
+ 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_complete,
0, GFP_KERNEL) ||
ca->fs = c;
- if (ca->mi.state == BCH_MEMBER_STATE_rw &&
- bch2_dev_allocator_start(ca)) {
- bch2_dev_free(ca);
- goto err;
- }
-
bch2_dev_attach(c, ca, dev_idx);
out:
pr_verbose_init(c->opts, "ret %i", ret);
ca->disk_sb.bdev->bd_holder = ca;
memset(sb, 0, sizeof(*sb));
+ ca->dev = ca->disk_sb.bdev->bd_dev;
+
percpu_ref_reinit(&ca->io_ref);
return 0;
/*
* The allocator thread itself allocates btree nodes, so stop it first:
*/
- bch2_dev_allocator_stop(ca);
bch2_dev_allocator_remove(c, ca);
bch2_dev_journal_stop(&c->journal, ca);
bch2_copygc_start(c);
}
-static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
+static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
{
lockdep_assert_held(&c->state_lock);
bch2_dev_allocator_add(c, ca);
bch2_recalc_capacity(c);
-
- if (bch2_dev_allocator_start(ca))
- return "error starting allocator thread";
-
- return NULL;
}
int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- if (new_state == BCH_MEMBER_STATE_rw &&
- __bch2_dev_read_write(c, ca))
- ret = -ENOMEM;
+ if (new_state == BCH_MEMBER_STATE_rw)
+ __bch2_dev_read_write(c, ca);
rebalance_wakeup(c);
/* Device add/removal: */
-int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
+static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
{
- struct btree_trans trans;
- size_t i;
+ struct bpos start = POS(ca->dev_idx, 0);
+ struct bpos end = POS(ca->dev_idx, U64_MAX);
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);
-
+ /*
+ * 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);
if (ret)
- return ret;
+ bch_err(c, "error %i removing dev alloc info", ret);
- return bch2_btree_delete_range(c, BTREE_ID_alloc,
- POS(ca->dev_idx, 0),
- POS(ca->dev_idx + 1, 0),
- NULL);
+ return ret;
}
int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
goto err;
}
- ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
+ ret = bch2_dev_remove_alloc(c, ca);
if (ret) {
- bch_err(ca, "Remove failed: error %i flushing journal", ret);
+ bch_err(ca, "Remove failed, error deleting alloc info");
goto err;
}
- ret = bch2_dev_remove_alloc(c, ca);
+ ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
if (ret) {
- bch_err(ca, "Remove failed, error deleting alloc info");
+ bch_err(ca, "Remove failed: error %i flushing journal", ret);
goto err;
}
- /*
- * 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
- */
- bch2_journal_meta(&c->journal);
- ret = bch2_journal_error(&c->journal);
+ ret = bch2_journal_flush(&c->journal);
if (ret) {
bch_err(ca, "Remove failed, journal error");
goto err;
data = bch2_dev_has_data(c, ca);
if (data) {
- char data_has_str[100];
+ struct printbuf data_has = PRINTBUF;
- bch2_flags_to_text(&PBUF(data_has_str),
- bch2_data_types, data);
- bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
+ 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;
}
struct bch_sb_field_members *mi;
struct bch_member dev_mi;
unsigned dev_idx, nr_devices, u64s;
+ struct printbuf errbuf = PRINTBUF;
int ret;
ret = bch2_read_super(path, &opts, &sb);
- if (ret)
- return ret;
-
- err = bch2_sb_validate(&sb);
- if (err)
- return -EINVAL;
+ if (ret) {
+ bch_err(c, "device add error: error reading super: %i", ret);
+ goto err;
+ }
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;
+ if (err) {
+ bch_err(c, "device add error: %s", err);
+ ret = -EINVAL;
+ goto err;
+ }
ca = __bch2_dev_alloc(c, &dev_mi);
if (!ca) {
bch2_free_super(&sb);
- return -ENOMEM;
+ ret = -ENOMEM;
+ goto err;
}
+ bch2_dev_usage_init(ca);
+
ret = __bch2_dev_attach_bdev(ca, &sb);
if (ret) {
bch2_dev_free(ca);
- return ret;
+ goto err;
}
- /*
- * 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...
- */
- bch2_mark_dev_superblock(NULL, ca, 0);
-
- err = "journal alloc failed";
ret = bch2_dev_journal_alloc(ca);
- if (ret)
+ if (ret) {
+ bch_err(c, "device add error: journal alloc failed");
goto err;
+ }
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)
+ if (ret) {
+ bch_err(c, "device add error: new device superblock too small");
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))) {
+ bch_err(c, "device add error: new device superblock too small");
ret = -ENOSPC;
goto err_unlock;
}
if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
goto have_slot;
no_slot:
- err = "no slots available in superblock";
+ bch_err(c, "device add error: already have maximum number of devices");
ret = -ENOSPC;
goto err_unlock;
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_sb_resize_members(&c->disk_sb, u64s);
- if (!mi)
+ if (!mi) {
+ bch_err(c, "device add error: no room in superblock for member info");
+ ret = -ENOSPC;
goto err_unlock;
+ }
/* success: */
bch2_dev_usage_journal_reserve(c);
- err = "error marking superblock";
ret = bch2_trans_mark_dev_sb(c, ca);
- if (ret)
+ if (ret) {
+ bch_err(c, "device add error: error marking new superblock: %i", ret);
goto err_late;
+ }
- if (ca->mi.state == BCH_MEMBER_STATE_rw) {
- err = __bch2_dev_read_write(c, ca);
- if (err)
- goto err_late;
+ ret = bch2_fs_freespace_init(c);
+ if (ret) {
+ bch_err(c, "device add error: error initializing free space: %i", ret);
+ goto err_late;
}
+ 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;
if (ca)
bch2_dev_free(ca);
bch2_free_super(&sb);
- bch_err(c, "Unable to add device: %s", err);
+ printbuf_exit(&errbuf);
return ret;
err_late:
up_write(&c->state_lock);
- bch_err(c, "Error going rw after adding device: %s", err);
- return -EINVAL;
+ ca = NULL;
+ goto err;
}
/* Hot add existing device to running filesystem: */
dev_idx = sb.sb->dev_idx;
err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
- if (err)
+ if (err) {
+ bch_err(c, "error bringing %s online: %s", path, err);
goto err;
+ }
- if (bch2_dev_attach_bdev(c, &sb)) {
- err = "bch2_dev_attach_bdev() error";
+ ret = bch2_dev_attach_bdev(c, &sb);
+ if (ret)
goto err;
- }
ca = bch_dev_locked(c, dev_idx);
- if (bch2_trans_mark_dev_sb(c, ca)) {
- err = "bch2_trans_mark_dev_sb() error";
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret) {
+ bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
+ path, ret);
goto err;
}
- if (ca->mi.state == BCH_MEMBER_STATE_rw) {
- err = __bch2_dev_read_write(c, ca);
- if (err)
- goto err;
- }
+ if (ca->mi.state == BCH_MEMBER_STATE_rw)
+ __bch2_dev_read_write(c, ca);
mutex_lock(&c->sb_lock);
mi = bch2_sb_get_members(c->disk_sb.sb);
err:
up_write(&c->state_lock);
bch2_free_super(&sb);
- bch_err(c, "error bringing %s online: %s", path, err);
return -EINVAL;
}
goto err;
}
+ ret = bch2_trans_mark_dev_sb(c, ca);
+ if (ret) {
+ goto err;
+ }
+
mutex_lock(&c->sb_lock);
mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
mi->nbuckets = cpu_to_le64(nbuckets);
}
/* return with ref on ca->ref: */
-struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
+struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
{
struct bch_dev *ca;
- dev_t dev;
unsigned i;
- int ret;
-
- ret = lookup_bdev(path, &dev);
- if (ret)
- return ERR_PTR(ret);
rcu_read_lock();
for_each_member_device_rcu(ca, c, i, NULL)
- if (ca->disk_sb.bdev->bd_dev == dev)
+ if (!strcmp(name, ca->name))
goto found;
ca = ERR_PTR(-ENOENT);
found:
struct bch_sb_field_members *mi;
unsigned i, best_sb = 0;
const char *err;
- int ret = -ENOMEM;
+ struct printbuf errbuf = PRINTBUF;
+ int ret = 0;
+
+ if (!try_module_get(THIS_MODULE))
+ return ERR_PTR(-ENODEV);
pr_verbose_init(opts, "");
if (!nr_devices) {
- c = ERR_PTR(-EINVAL);
- goto out2;
- }
-
- if (!try_module_get(THIS_MODULE)) {
- c = ERR_PTR(-ENODEV);
- goto out2;
+ ret = -EINVAL;
+ goto err;
}
sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
- if (!sb)
+ if (!sb) {
+ ret = -ENOMEM;
goto err;
+ }
for (i = 0; i < nr_devices; i++) {
ret = bch2_read_super(devices[i], &opts, &sb[i]);
if (ret)
goto err;
- err = bch2_sb_validate(&sb[i]);
- if (err)
- goto err_print;
}
for (i = 1; i < nr_devices; i++)
i++;
}
- ret = -ENOMEM;
c = bch2_fs_alloc(sb[best_sb].sb, opts);
- if (!c)
+ if (IS_ERR(c)) {
+ ret = PTR_ERR(c);
goto err;
+ }
- err = "bch2_dev_online() error";
down_write(&c->state_lock);
- for (i = 0; i < nr_devices; i++)
- if (bch2_dev_attach_bdev(c, &sb[i])) {
+ for (i = 0; i < nr_devices; i++) {
+ ret = bch2_dev_attach_bdev(c, &sb[i]);
+ if (ret) {
up_write(&c->state_lock);
- goto err_print;
+ goto err;
}
+ }
up_write(&c->state_lock);
err = "insufficient devices";
}
out:
kfree(sb);
+ printbuf_exit(&errbuf);
module_put(THIS_MODULE);
-out2:
pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
return c;
err_print:
devices[0], err);
ret = -EINVAL;
err:
- if (c)
+ if (!IS_ERR_OR_NULL(c))
bch2_fs_stop(c);
- for (i = 0; i < nr_devices; i++)
- bch2_free_super(&sb[i]);
+ if (sb)
+ for (i = 0; i < nr_devices; i++)
+ bch2_free_super(&sb[i]);
c = ERR_PTR(ret);
goto out;
}
-static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
- struct bch_opts opts)
-{
- const char *err;
- struct bch_fs *c;
- bool allocated_fs = false;
- int ret;
-
- err = bch2_sb_validate(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);
-
- err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
- if (err)
- goto err;
- } else {
- c = bch2_fs_alloc(sb->sb, opts);
- err = "cannot allocate memory";
- if (!c)
- goto err;
-
- allocated_fs = true;
- }
-
- err = "bch2_dev_online() error";
-
- mutex_lock(&c->sb_lock);
- 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 = "error starting filesystem";
- ret = bch2_fs_start(c);
- if (ret)
- goto err;
- }
-
- closure_put(&c->cl);
- mutex_unlock(&bch_fs_list_lock);
-
- return NULL;
-err:
- mutex_unlock(&bch_fs_list_lock);
-
- if (allocated_fs)
- bch2_fs_stop(c);
- else if (c)
- closure_put(&c->cl);
-
- return err;
-}
-
-const char *bch2_fs_open_incremental(const char *path)
-{
- struct bch_sb_handle sb;
- struct bch_opts opts = bch2_opts_empty();
- const char *err;
-
- if (bch2_read_super(path, &opts, &sb))
- return "error reading superblock";
-
- err = __bch2_fs_open_incremental(&sb, opts);
- bch2_free_super(&sb);
-
- return err;
-}
-
/* Global interfaces/init */
static void bcachefs_exit(void)