]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/super.c
Update bcachefs sources to bdf6d7c135 fixup! bcachefs: Kill journal buf bloom filter
[bcachefs-tools-debian] / libbcachefs / super.c
index 505e559b48a682c6177bb6c52f82e9877fde7305..bdac2b7272334085204bf7fef5fb9e385abed9b1 100644 (file)
@@ -16,6 +16,7 @@
 #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"
@@ -43,6 +44,7 @@
 #include "super.h"
 #include "super-io.h"
 #include "sysfs.h"
+#include "counters.h"
 
 #include <linux/backing-dev.h>
 #include <linux/blkdev.h>
@@ -70,6 +72,9 @@ struct kobj_type type ## _ktype = {                                   \
 
 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)
 {
@@ -84,6 +89,7 @@ static void bch2_fs_time_stats_release(struct kobject *k)
 }
 
 static KTYPE(bch2_fs);
+static KTYPE(bch2_fs_counters);
 static KTYPE(bch2_fs_internal);
 static KTYPE(bch2_fs_opts_dir);
 static KTYPE(bch2_fs_time_stats);
@@ -187,71 +193,33 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 {
        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:
         */
@@ -279,10 +247,6 @@ void bch2_fs_read_only(struct bch_fs *c)
        /*
         * 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);
 
@@ -314,7 +278,7 @@ void bch2_fs_read_only(struct bch_fs *c)
            !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);
@@ -405,25 +369,13 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
        if (ret)
                goto err;
 
-       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);
+       bch2_do_discards(c);
 
        if (!early) {
                ret = bch2_fs_read_write_late(c);
@@ -468,6 +420,7 @@ static void __bch2_fs_free(struct bch_fs *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);
@@ -478,7 +431,7 @@ static void __bch2_fs_free(struct bch_fs *c)
        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_paths_bufs)
@@ -545,6 +498,7 @@ void __bch2_fs_stop(struct bch_fs *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);
@@ -613,6 +567,7 @@ static int bch2_fs_online(struct bch_fs *c)
            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");
@@ -661,6 +616,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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;
@@ -672,6 +628,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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]);
@@ -682,6 +639,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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);
 
@@ -695,7 +653,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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);
@@ -742,7 +699,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        if (ret)
                goto err;
 
-       scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
+       uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 
        /* Compat: */
        if (sb->version <= bcachefs_metadata_version_inode_v2 &&
@@ -754,10 +711,20 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
                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")) {
@@ -807,18 +774,17 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
            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_fsio_init(c) ?:
+           bch2_fs_counters_init(c);
        if (ret)
                goto err;
 
-       if (c->opts.nochanges)
-               set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
-
        mi = bch2_sb_get_members(c->disk_sb.sb);
        for (i = 0; i < c->sb.nr_devices; i++)
                if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
@@ -854,12 +820,9 @@ noinline_for_stack
 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");
                first = false;
@@ -869,7 +832,7 @@ static void print_mount_opts(struct bch_fs *c)
                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))
@@ -878,10 +841,14 @@ static void print_mount_opts(struct bch_fs *c)
                if (!first)
                        pr_buf(&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)
+               pr_buf(&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)
@@ -929,20 +896,6 @@ int bch2_fs_start(struct bch_fs *c)
 
        set_bit(BCH_FS_STARTED, &c->flags);
 
-       /*
-        * Allocator threads don't start filling copygc reserve until after we
-        * set BCH_FS_STARTED - wake them now:
-        *
-        * 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 {
@@ -995,7 +948,7 @@ static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
        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) <
@@ -1034,8 +987,6 @@ static void bch2_dev_release(struct kobject *kobj)
 
 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 &&
@@ -1150,8 +1101,8 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
        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) ||
@@ -1202,12 +1153,6 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 
        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);
@@ -1253,6 +1198,8 @@ static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
                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;
@@ -1391,14 +1338,13 @@ static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
        /*
         * 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 int __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);
 
@@ -1406,8 +1352,6 @@ static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
 
        bch2_dev_allocator_add(c, ca);
        bch2_recalc_capacity(c);
-
-       return bch2_dev_allocator_start(ca);
 }
 
 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
@@ -1434,7 +1378,7 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
        mutex_unlock(&c->sb_lock);
 
        if (new_state == BCH_MEMBER_STATE_rw)
-               ret = __bch2_dev_read_write(c, ca);
+               __bch2_dev_read_write(c, ca);
 
        rebalance_wakeup(c);
 
@@ -1457,30 +1401,26 @@ int bch2_dev_set_state(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 = lockrestart_do(&trans,
-                       bch2_btree_key_cache_flush(&trans,
-                               BTREE_ID_alloc, POS(ca->dev_idx, i)));
-               if (ret)
-                       break;
-       }
-       bch2_trans_exit(&trans);
-
-       if (ret) {
+       /*
+        * 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_alloc, start, end,
+                                       BTREE_TRIGGER_NORUN, NULL);
+       if (ret)
                bch_err(c, "error %i removing dev alloc info", ret);
-               return 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)
@@ -1510,28 +1450,19 @@ 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;
@@ -1545,11 +1476,11 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
        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);
+               bch2_flags_to_text(&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;
        }
@@ -1597,69 +1528,61 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
        struct bch_dev *ca = NULL;
        struct bch_sb_field_members *mi;
        struct bch_member dev_mi;
-       struct bucket_array *buckets;
-       struct bucket *g;
        unsigned dev_idx, nr_devices, u64s;
+       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;
        }
@@ -1672,7 +1595,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
                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;
 
@@ -1681,12 +1604,12 @@ have_slot:
        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: */
 
@@ -1702,27 +1625,23 @@ have_slot:
 
        bch2_dev_usage_journal_reserve(c);
 
-       /*
-        * Clear marks before marking transactionally in the btree, so that
-        * per-device accounting gets done correctly:
-        */
-       down_read(&ca->bucket_lock);
-       buckets = bucket_array(ca);
-       for_each_bucket(g, buckets)
-               atomic64_set(&g->_mark.v, 0);
-       up_read(&ca->bucket_lock);
-
-       err = "error marking superblock";
        ret = bch2_trans_mark_dev_sb(c, 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) {
-               ret = __bch2_dev_read_write(c, ca);
-               if (ret)
-                       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;
 
@@ -1733,12 +1652,12 @@ err:
        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: */
@@ -1781,11 +1700,8 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
                goto err;
        }
 
-       if (ca->mi.state == BCH_MEMBER_STATE_rw) {
-               ret = __bch2_dev_read_write(c, ca);
-               if (ret)
-                       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);
@@ -1872,20 +1788,14 @@ err:
 }
 
 /* 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:
@@ -1904,18 +1814,17 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
        struct bch_sb_field_members *mi;
        unsigned i, best_sb = 0;
        const char *err;
+       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);
@@ -1929,9 +1838,6 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
                if (ret)
                        goto err;
 
-               err = bch2_sb_validate(&sb[i]);
-               if (err)
-                       goto err_print;
        }
 
        for (i = 1; i < nr_devices; i++)
@@ -1986,8 +1892,8 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_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:
@@ -2004,81 +1910,6 @@ err:
        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 {
-               allocated_fs = true;
-               c = bch2_fs_alloc(sb->sb, opts);
-
-               err = "bch2_fs_alloc() error";
-               if (IS_ERR(c))
-                       goto err;
-       }
-
-       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 && !IS_ERR(c))
-               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)