]> git.sesse.net Git - bcachefs-tools-debian/blobdiff - libbcachefs/super.c
Update bcachefs sources to 0e705f5944 fixup! bcachefs: Refactor bch2_btree_node_mem_a...
[bcachefs-tools-debian] / libbcachefs / super.c
index 586ba60d03ea7a2e0cb5b4ae9abeeb0c4bff0638..4a071711d363bcaa80cea1f982cf3ffeb287f4cb 100644 (file)
@@ -199,17 +199,9 @@ static void __bch2_fs_read_only(struct bch_fs *c)
         */
        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++;
@@ -234,17 +226,11 @@ static void __bch2_fs_read_only(struct bch_fs *c)
        bch_verbose(c, "flushing journal and stopping allocators complete");
 
        set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
-nowrote_alloc:
+
        closure_wait_event(&c->btree_interior_update_wait,
                           !bch2_btree_interior_updates_nr_pending(c));
        flush_work(&c->btree_interior_update_work);
 
-       for_each_member_device(ca, c, i)
-               bch2_dev_allocator_stop(ca);
-
-       clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
-       clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
-
        bch2_fs_journal_stop(&c->journal);
 
        /*
@@ -280,10 +266,6 @@ 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);
 
@@ -412,19 +394,7 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
                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);
@@ -674,6 +644,7 @@ 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]);
@@ -762,6 +733,13 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
        bch2_opts_apply(&c->opts, opts);
 
+       /* 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);
 
@@ -822,9 +800,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
        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) &&
@@ -860,12 +835,9 @@ 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;
@@ -884,10 +856,14 @@ 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)
@@ -935,20 +911,6 @@ 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 {
@@ -1040,8 +1002,6 @@ 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 &&
@@ -1156,8 +1116,8 @@ 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) ||
@@ -1208,12 +1168,6 @@ 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);
@@ -1399,14 +1353,13 @@ 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);
 
@@ -1414,8 +1367,6 @@ 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,
@@ -1442,7 +1393,7 @@ 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);
 
@@ -1465,30 +1416,20 @@ 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) {
+       ret =   bch2_btree_delete_range(c, BTREE_ID_alloc, 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_need_discard, 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),
-                                      0, NULL);
+       return ret;
 }
 
 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
@@ -1553,11 +1494,11 @@ 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;
        }
@@ -1606,16 +1547,9 @@ 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;
+       struct printbuf errbuf = PRINTBUF;
        int ret;
 
-       _errbuf = kmalloc(4096, GFP_KERNEL);
-       if (!_errbuf)
-               return -ENOMEM;
-
-       errbuf = _PBUF(_errbuf, 4096);
-
        ret = bch2_read_super(path, &opts, &sb);
        if (ret) {
                bch_err(c, "device add error: error reading super: %i", ret);
@@ -1713,15 +1647,16 @@ have_slot:
                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) {
-               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;
-               }
-       }
+       if (ca->mi.state == BCH_MEMBER_STATE_rw)
+               __bch2_dev_read_write(c, ca);
 
        up_write(&c->state_lock);
        return 0;
@@ -1733,7 +1668,7 @@ err:
        if (ca)
                bch2_dev_free(ca);
        bch2_free_super(&sb);
-       kfree(_errbuf);
+       printbuf_exit(&errbuf);
        return ret;
 err_late:
        up_write(&c->state_lock);
@@ -1781,11 +1716,8 @@ 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 +1804,14 @@ 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->dev == dev)
+               if (!strcmp(name, ca->name))
                        goto found;
        ca = ERR_PTR(-ENOENT);
 found:
@@ -1904,8 +1830,7 @@ 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;
-       char *_errbuf = NULL;
-       struct printbuf errbuf;
+       struct printbuf errbuf = PRINTBUF;
        int ret = 0;
 
        if (!try_module_get(THIS_MODULE))
@@ -1918,14 +1843,6 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
                goto err;
        }
 
-       _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) {
                ret = -ENOMEM;
@@ -1991,7 +1908,7 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
        }
 out:
        kfree(sb);
-       kfree(_errbuf);
+       printbuf_exit(&errbuf);
        module_put(THIS_MODULE);
        pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
        return c;