]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/super.c
New upstream snapshot
[bcachefs-tools-debian] / libbcachefs / super.c
index ce8e5d4843d0e8f4bd9a596d0421b9a04860bd4b..b36e6216a8a10a2ec12b0de39ba954424d81d45d 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"
@@ -39,6 +40,7 @@
 #include "rebalance.h"
 #include "recovery.h"
 #include "replicas.h"
+#include "subvolume.h"
 #include "super.h"
 #include "super-io.h"
 #include "sysfs.h"
@@ -404,13 +406,6 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
        if (ret)
                goto err;
 
-       /*
-        * We need to write out a journal entry before we start doing btree
-        * updates, to ensure that on unclean shutdown new journal blacklist
-        * entries are created:
-        */
-       bch2_journal_meta(&c->journal);
-
        clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 
        for_each_rw_member(ca, c, i)
@@ -468,11 +463,13 @@ static void __bch2_fs_free(struct bch_fs *c)
        for (i = 0; i < BCH_TIME_STAT_NR; i++)
                bch2_time_stats_exit(&c->times[i]);
 
+       bch2_fs_snapshots_exit(c);
        bch2_fs_quota_exit(c);
        bch2_fs_fsio_exit(c);
        bch2_fs_ec_exit(c);
        bch2_fs_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);
@@ -486,12 +483,12 @@ static void __bch2_fs_free(struct bch_fs *c)
        bch2_journal_entries_free(&c->journal_entries);
        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);
@@ -593,48 +590,53 @@ void bch2_fs_stop(struct bch_fs *c)
        bch2_fs_free(c);
 }
 
-static const char *bch2_fs_online(struct bch_fs *c)
+static int bch2_fs_online(struct bch_fs *c)
 {
        struct bch_dev *ca;
-       const char *err = NULL;
        unsigned i;
-       int ret;
+       int ret = 0;
 
        lockdep_assert_held(&bch_fs_list_lock);
 
-       if (!list_empty(&c->list))
-               return NULL;
-
-       if (__bch2_uuid_to_fs(c->sb.uuid))
-               return "filesystem UUID already open";
+       if (__bch2_uuid_to_fs(c->sb.uuid)) {
+               bch_err(c, "filesystem UUID already open");
+               return -EINVAL;
+       }
 
        ret = bch2_fs_chardev_init(c);
-       if (ret)
-               return "error creating character device";
+       if (ret) {
+               bch_err(c, "error creating character device");
+               return ret;
+       }
 
        bch2_fs_debug_init(c);
 
-       if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
-           kobject_add(&c->internal, &c->kobj, "internal") ||
-           kobject_add(&c->opts_dir, &c->kobj, "options") ||
-           kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
-           bch2_opts_create_sysfs_files(&c->opts_dir))
-               return "error creating sysfs objects";
+       ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
+           kobject_add(&c->internal, &c->kobj, "internal") ?:
+           kobject_add(&c->opts_dir, &c->kobj, "options") ?:
+           kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
+           bch2_opts_create_sysfs_files(&c->opts_dir);
+       if (ret) {
+               bch_err(c, "error creating sysfs objects");
+               return ret;
+       }
 
        down_write(&c->state_lock);
 
-       err = "error creating sysfs objects";
-       for_each_member_device(ca, c, i)
-               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)
@@ -642,13 +644,15 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        struct bch_sb_field_members *mi;
        struct bch_fs *c;
        unsigned i, iter_size;
-       const char *err;
+       int ret = 0;
 
        pr_verbose_init(opts, "");
 
        c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
-       if (!c)
+       if (!c) {
+               c = ERR_PTR(-ENOMEM);
                goto out;
+       }
 
        __module_get(THIS_MODULE);
 
@@ -670,6 +674,7 @@ 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]);
@@ -686,6 +691,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        mutex_init(&c->usage_scratch_lock);
 
        mutex_init(&c->bio_bounce_pages_lock);
+       mutex_init(&c->snapshot_table_lock);
 
        spin_lock_init(&c->btree_write_error_lock);
 
@@ -704,6 +710,9 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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);
@@ -716,38 +725,59 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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);
+       uuid_unparse_lower(c->sb.user_uuid.b, c->name);
+
+       /* Compat: */
+       if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+           !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
+               SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 
-       scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
+       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 *
@@ -771,33 +801,45 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
                            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);
+       if (ret)
+               goto err;
+
+       if (c->opts.nochanges)
+               set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
 
        mi = bch2_sb_get_members(c->disk_sb.sb);
        for (i = 0; i < c->sb.nr_devices; i++)
                if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
-                   bch2_dev_alloc(c, i))
+                   bch2_dev_alloc(c, i)) {
+                       ret = -EEXIST;
                        goto err;
+               }
 
        bch2_journal_entry_res_resize(&c->journal,
                        &c->btree_root_journal_res,
@@ -808,18 +850,17 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
                        (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;
 }
 
@@ -842,7 +883,7 @@ 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))
@@ -859,7 +900,6 @@ static void print_mount_opts(struct bch_fs *c)
 
 int bch2_fs_start(struct bch_fs *c)
 {
-       const char *err = "cannot allocate memory";
        struct bch_sb_field_members *mi;
        struct bch_dev *ca;
        time64_t now = ktime_get_real_seconds();
@@ -895,10 +935,11 @@ int bch2_fs_start(struct bch_fs *c)
        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);
 
@@ -919,7 +960,6 @@ int bch2_fs_start(struct bch_fs *c)
        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);
@@ -937,25 +977,22 @@ err:
        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;
        }
 
@@ -972,7 +1009,7 @@ 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) <
@@ -1230,6 +1267,8 @@ 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;
@@ -1375,7 +1414,7 @@ static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
        bch2_copygc_start(c);
 }
 
-static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
+static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
 {
        lockdep_assert_held(&c->state_lock);
 
@@ -1384,10 +1423,7 @@ static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
        bch2_dev_allocator_add(c, ca);
        bch2_recalc_capacity(c);
 
-       if (bch2_dev_allocator_start(ca))
-               return "error starting allocator thread";
-
-       return NULL;
+       return bch2_dev_allocator_start(ca);
 }
 
 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
@@ -1413,9 +1449,8 @@ 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)
+               ret = __bch2_dev_read_write(c, ca);
 
        rebalance_wakeup(c);
 
@@ -1445,20 +1480,23 @@ static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
        bch2_trans_init(&trans, c, 0, 0);
 
        for (i = 0; i < ca->mi.nbuckets; i++) {
-               ret = bch2_btree_key_cache_flush(&trans,
-                               BTREE_ID_alloc, POS(ca->dev_idx, i));
+               ret = lockrestart_do(&trans,
+                       bch2_btree_key_cache_flush(&trans,
+                               BTREE_ID_alloc, POS(ca->dev_idx, i)));
                if (ret)
                        break;
        }
        bch2_trans_exit(&trans);
 
-       if (ret)
+       if (ret) {
+               bch_err(c, "error %i removing dev alloc info", ret);
                return ret;
+       }
 
        return bch2_btree_delete_range(c, BTREE_ID_alloc,
                                       POS(ca->dev_idx, 0),
                                       POS(ca->dev_idx + 1, 0),
-                                      NULL);
+                                      0, NULL);
 }
 
 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
@@ -1576,66 +1614,65 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
        struct bch_sb_field_members *mi;
        struct bch_member dev_mi;
        unsigned dev_idx, nr_devices, u64s;
+       char *_errbuf;
+       struct printbuf errbuf;
        int ret;
 
-       ret = bch2_read_super(path, &opts, &sb);
-       if (ret)
-               return ret;
+       _errbuf = kmalloc(4096, GFP_KERNEL);
+       if (!_errbuf)
+               return -ENOMEM;
 
-       err = bch2_sb_validate(&sb);
-       if (err)
-               return -EINVAL;
+       errbuf = _PBUF(_errbuf, 4096);
+
+       ret = bch2_read_super(path, &opts, &sb);
+       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;
        }
 
        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;
        }
@@ -1648,7 +1685,7 @@ 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;
 
@@ -1657,12 +1694,12 @@ 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: */
 
@@ -1678,15 +1715,20 @@ have_slot:
 
        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;
+       }
+
+       ca->new_fs_bucket_idx = 0;
 
        if (ca->mi.state == BCH_MEMBER_STATE_rw) {
-               err = __bch2_dev_read_write(c, ca);
-               if (err)
+               ret = __bch2_dev_read_write(c, ca);
+               if (ret) {
+                       bch_err(c, "device add error: error going RW on new device: %i", ret);
                        goto err_late;
+               }
        }
 
        up_write(&c->state_lock);
@@ -1699,12 +1741,12 @@ err:
        if (ca)
                bch2_dev_free(ca);
        bch2_free_super(&sb);
-       bch_err(c, "Unable to add device: %s", err);
+       kfree(_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: */
@@ -1729,24 +1771,27 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
        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)
+               ret = __bch2_dev_read_write(c, ca);
+               if (ret)
                        goto err;
        }
 
@@ -1764,7 +1809,6 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 err:
        up_write(&c->state_lock);
        bch2_free_super(&sb);
-       bch_err(c, "error bringing %s online: %s", path, err);
        return -EINVAL;
 }
 
@@ -1836,20 +1880,14 @@ 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:
@@ -1868,32 +1906,39 @@ 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;
-       int ret = -ENOMEM;
+       char *_errbuf = NULL;
+       struct printbuf errbuf;
+       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;
+               ret = -EINVAL;
+               goto err;
        }
 
-       if (!try_module_get(THIS_MODULE)) {
-               c = ERR_PTR(-ENODEV);
-               goto out2;
+       _errbuf = kmalloc(4096, GFP_KERNEL);
+       if (!_errbuf) {
+               ret = -ENOMEM;
+               goto err;
        }
 
+       errbuf = _PBUF(_errbuf, 4096);
+
        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++)
@@ -1921,18 +1966,20 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
                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";
@@ -1946,8 +1993,8 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
        }
 out:
        kfree(sb);
+       kfree(_errbuf);
        module_put(THIS_MODULE);
-out2:
        pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
        return c;
 err_print:
@@ -1955,89 +2002,15 @@ 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)