bcachefs: Turn c->state_lock into an rwsem

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

7 files changed, 50 insertions, 57 deletions

diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index a219969357bc..8d9cc7eb6ad7 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -551,8 +551,8 @@ struct bch_fs {
 	struct super_block	*vfs_sb;
 	char			name[40];
 
-	/* ro/rw, add/remove devices: */
-	struct mutex		state_lock;
+	/* ro/rw, add/remove/resize devices: */
+	struct rw_semaphore	state_lock;
 
 	/* Counts outstanding writes, for clean transition to read-only */
 	struct percpu_ref	writes;
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 6589fe0bad6c..22aa845ea630 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -800,6 +800,7 @@ int bch2_gc(struct bch_fs *c, struct journal_keys *journal_keys,
 	unsigned i, iter = 0;
 	int ret;
 
+	lockdep_assert_held(&c->state_lock);
 	trace_gc_start(c);
 
 	down_write(&c->gc_lock);
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 08e8b578fff5..5ee978c94568 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -1967,6 +1967,8 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	int ret = -ENOMEM;
 	unsigned i;
 
+	lockdep_assert_held(&c->state_lock);
+
 	memset(&free,		0, sizeof(free));
 	memset(&free_inc,	0, sizeof(free_inc));
 	memset(&alloc_heap,	0, sizeof(alloc_heap));
@@ -1993,7 +1995,6 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	bch2_copygc_stop(ca);
 
 	if (resize) {
-		down_write(&c->gc_lock);
 		down_write(&ca->bucket_lock);
 		percpu_down_write(&c->mark_lock);
 	}
@@ -2036,10 +2037,8 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 
 	nbuckets = ca->mi.nbuckets;
 
-	if (resize) {
+	if (resize)
 		up_write(&ca->bucket_lock);
-		up_write(&c->gc_lock);
-	}
 
 	if (start_copygc &&
 	    bch2_copygc_start(c, ca))
diff --git a/fs/bcachefs/error.c b/fs/bcachefs/error.c
index 1662a36244cd..cd46706fb6f5 100644
--- a/fs/bcachefs/error.c
+++ b/fs/bcachefs/error.c
@@ -37,7 +37,7 @@ void bch2_io_error_work(struct work_struct *work)
 	struct bch_fs *c = ca->fs;
 	bool dev;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_RO,
 				    BCH_FORCE_IF_DEGRADED);
 	if (dev
@@ -47,7 +47,7 @@ void bch2_io_error_work(struct work_struct *work)
 		bch_err(ca,
 			"too many IO errors, setting %s RO",
 			dev ? "device" : "filesystem");
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 }
 
 void bch2_io_error(struct bch_dev *ca)
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 6aff3203b4e1..4538551ccca3 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -1342,16 +1342,16 @@ static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * cons
 	if (IS_ERR(c))
 		return c;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	if (!test_bit(BCH_FS_STARTED, &c->flags)) {
-		mutex_unlock(&c->state_lock);
+		up_write(&c->state_lock);
 		closure_put(&c->cl);
 		pr_err("err mounting %s: incomplete filesystem", dev_name);
 		return ERR_PTR(-EINVAL);
 	}
 
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 
 	set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
 	return c;
@@ -1400,7 +1400,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data)
 		return ret;
 
 	if (opts.read_only != c->opts.read_only) {
-		mutex_lock(&c->state_lock);
+		down_write(&c->state_lock);
 
 		if (opts.read_only) {
 			bch2_fs_read_only(c);
@@ -1410,7 +1410,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data)
 			ret = bch2_fs_read_write(c);
 			if (ret) {
 				bch_err(c, "error going rw: %i", ret);
-				mutex_unlock(&c->state_lock);
+				up_write(&c->state_lock);
 				return -EINVAL;
 			}
 
@@ -1419,7 +1419,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data)
 
 		c->opts.read_only = opts.read_only;
 
-		mutex_unlock(&c->state_lock);
+		up_write(&c->state_lock);
 	}
 
 	if (opts.errors >= 0)
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 819d4392d529..3a7c48def9e8 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -340,9 +340,9 @@ static void bch2_fs_read_only_work(struct work_struct *work)
 	struct bch_fs *c =
 		container_of(work, struct bch_fs, read_only_work);
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	bch2_fs_read_only(c);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 }
 
 static void bch2_fs_read_only_async(struct bch_fs *c)
@@ -534,9 +534,9 @@ void bch2_fs_stop(struct bch_fs *c)
 
 	cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	bch2_fs_read_only(c);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 
 	for_each_member_device(ca, c, i)
 		if (ca->kobj.state_in_sysfs &&
@@ -607,7 +607,7 @@ static const char *bch2_fs_online(struct bch_fs *c)
 	    bch2_opts_create_sysfs_files(&c->opts_dir))
 		return "error creating sysfs objects";
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	err = "error creating sysfs objects";
 	__for_each_member_device(ca, c, i, NULL)
@@ -617,7 +617,7 @@ static const char *bch2_fs_online(struct bch_fs *c)
 	list_add(&c->list, &bch_fs_list);
 	err = NULL;
 err:
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return err;
 }
 
@@ -639,7 +639,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	c->minor		= -1;
 	c->disk_sb.fs_sb	= true;
 
-	mutex_init(&c->state_lock);
+	init_rwsem(&c->state_lock);
 	mutex_init(&c->sb_lock);
 	mutex_init(&c->replicas_gc_lock);
 	mutex_init(&c->btree_root_lock);
@@ -832,7 +832,7 @@ int bch2_fs_start(struct bch_fs *c)
 	unsigned i;
 	int ret = -EINVAL;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 
@@ -882,7 +882,7 @@ int bch2_fs_start(struct bch_fs *c)
 	print_mount_opts(c);
 	ret = 0;
 out:
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return ret;
 err:
 	switch (ret) {
@@ -1376,9 +1376,9 @@ int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
 {
 	int ret;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 
 	return ret;
 }
@@ -1391,7 +1391,7 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	unsigned dev_idx = ca->dev_idx, data;
 	int ret = -EINVAL;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	/*
 	 * We consume a reference to ca->ref, regardless of whether we succeed
@@ -1481,13 +1481,13 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	bch2_write_super(c);
 
 	mutex_unlock(&c->sb_lock);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return 0;
 err:
 	if (ca->mi.state == BCH_MEMBER_STATE_RW &&
 	    !percpu_ref_is_zero(&ca->io_ref))
 		__bch2_dev_read_write(c, ca);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return ret;
 }
 
@@ -1563,7 +1563,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 
 	dev_usage_clear(ca);
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	mutex_lock(&c->sb_lock);
 
 	err = "insufficient space in new superblock";
@@ -1624,12 +1624,12 @@ have_slot:
 			goto err_late;
 	}
 
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return 0;
 
 err_unlock:
 	mutex_unlock(&c->sb_lock);
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 err:
 	if (ca)
 		bch2_dev_free(ca);
@@ -1652,11 +1652,11 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 	const char *err;
 	int ret;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	ret = bch2_read_super(path, &opts, &sb);
 	if (ret) {
-		mutex_unlock(&c->state_lock);
+		up_write(&c->state_lock);
 		return ret;
 	}
 
@@ -1687,10 +1687,10 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return 0;
 err:
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	bch2_free_super(&sb);
 	bch_err(c, "error bringing %s online: %s", path, err);
 	return -EINVAL;
@@ -1698,23 +1698,23 @@ err:
 
 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
 {
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	if (!bch2_dev_is_online(ca)) {
 		bch_err(ca, "Already offline");
-		mutex_unlock(&c->state_lock);
+		up_write(&c->state_lock);
 		return 0;
 	}
 
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
 		bch_err(ca, "Cannot offline required disk");
-		mutex_unlock(&c->state_lock);
+		up_write(&c->state_lock);
 		return -EINVAL;
 	}
 
 	__bch2_dev_offline(c, ca);
 
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return 0;
 }
 
@@ -1723,7 +1723,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	struct bch_member *mi;
 	int ret = 0;
 
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 
 	if (nbuckets < ca->mi.nbuckets) {
 		bch_err(ca, "Cannot shrink yet");
@@ -1754,7 +1754,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 
 	bch2_recalc_capacity(c);
 err:
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 	return ret;
 }
 
@@ -1834,13 +1834,13 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 		goto err;
 
 	err = "bch2_dev_online() error";
-	mutex_lock(&c->state_lock);
+	down_write(&c->state_lock);
 	for (i = 0; i < nr_devices; i++)
 		if (bch2_dev_attach_bdev(c, &sb[i])) {
-			mutex_unlock(&c->state_lock);
+			up_write(&c->state_lock);
 			goto err_print;
 		}
-	mutex_unlock(&c->state_lock);
+	up_write(&c->state_lock);
 
 	err = "insufficient devices";
 	if (!bch2_fs_may_start(c))
diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c
index 06b59e991312..663b59e78824 100644
--- a/fs/bcachefs/sysfs.c
+++ b/fs/bcachefs/sysfs.c
@@ -427,7 +427,7 @@ SHOW(bch2_fs)
 	return 0;
 }
 
-STORE(__bch2_fs)
+STORE(bch2_fs)
 {
 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 
@@ -485,8 +485,11 @@ STORE(__bch2_fs)
 	if (attr == &sysfs_trigger_btree_coalesce)
 		bch2_coalesce(c);
 
-	if (attr == &sysfs_trigger_gc)
+	if (attr == &sysfs_trigger_gc) {
+		down_read(&c->state_lock);
 		bch2_gc(c, NULL, false, false);
+		up_read(&c->state_lock);
+	}
 
 	if (attr == &sysfs_trigger_alloc_write) {
 		bool wrote;
@@ -501,6 +504,7 @@ STORE(__bch2_fs)
 		sc.nr_to_scan = strtoul_or_return(buf);
 		c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc);
 	}
+
 #ifdef CONFIG_BCACHEFS_TESTS
 	if (attr == &sysfs_perf_test) {
 		char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp;
@@ -522,17 +526,6 @@ STORE(__bch2_fs)
 #endif
 	return size;
 }
-
-STORE(bch2_fs)
-{
-	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
-
-	mutex_lock(&c->state_lock);
-	size = __bch2_fs_store(kobj, attr, buf, size);
-	mutex_unlock(&c->state_lock);
-
-	return size;
-}
 SYSFS_OPS(bch2_fs);
 
 struct attribute *bch2_fs_files[] = {