fs,block: yield devices early

Currently a device is only really released once the umount returns to
userspace due to how file closing works. That ultimately could cause
an old umount assumption to be violated that concurrent umount and mount
don't fail. So an exclusively held device with a temporary holder should
be yielded before the filesystem is gone. Add a helper that allows
callers to do that. This also allows us to remove the two holder ops
that Linus wasn't excited about.

Link: https://lore.kernel.org/r/20240326-vfs-bdev-end_holder-v1-1-20af85202918@kernel.org
Fixes: f3a608827d ("bdev: open block device as files") # mainline only
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>

block/bdev.c

@@ -583,9 +583,6 @@ static void bd_finish_claiming(struct block_device *bdev, void *holder,
 	mutex_unlock(&bdev->bd_holder_lock);
 	bd_clear_claiming(whole, holder);
 	mutex_unlock(&bdev_lock);
-
-	if (hops && hops->get_holder)
-		hops->get_holder(holder);
 }
 
 /**
@@ -608,7 +605,6 @@ EXPORT_SYMBOL(bd_abort_claiming);
 static void bd_end_claim(struct block_device *bdev, void *holder)
 {
 	struct block_device *whole = bdev_whole(bdev);
-	const struct blk_holder_ops *hops = bdev->bd_holder_ops;
 	bool unblock = false;
 
 	/*
@@ -631,9 +627,6 @@ static void bd_end_claim(struct block_device *bdev, void *holder)
 		whole->bd_holder = NULL;
 	mutex_unlock(&bdev_lock);
 
-	if (hops && hops->put_holder)
-		hops->put_holder(holder);
-
 	/*
 	 * If this was the last claim, remove holder link and unblock evpoll if
 	 * it was a write holder.
@@ -813,6 +806,11 @@ static void bdev_claim_write_access(struct block_device *bdev, blk_mode_t mode)
 		bdev->bd_writers++;
 }
 
+static inline bool bdev_unclaimed(const struct file *bdev_file)
+{
+	return bdev_file->private_data == BDEV_I(bdev_file->f_mapping->host);
+}
+
 static void bdev_yield_write_access(struct file *bdev_file)
 {
 	struct block_device *bdev;
@@ -820,6 +818,9 @@ static void bdev_yield_write_access(struct file *bdev_file)
 	if (bdev_allow_write_mounted)
 		return;
 
+	if (bdev_unclaimed(bdev_file))
+		return;
+
 	bdev = file_bdev(bdev_file);
 
 	if (bdev_file->f_mode & FMODE_WRITE_RESTRICTED)
@@ -1012,6 +1013,20 @@ struct file *bdev_file_open_by_path(const char *path, blk_mode_t mode,
 }
 EXPORT_SYMBOL(bdev_file_open_by_path);
 
+static inline void bd_yield_claim(struct file *bdev_file)
+{
+	struct block_device *bdev = file_bdev(bdev_file);
+	void *holder = bdev_file->private_data;
+
+	lockdep_assert_held(&bdev->bd_disk->open_mutex);
+
+	if (WARN_ON_ONCE(IS_ERR_OR_NULL(holder)))
+		return;
+
+	if (!bdev_unclaimed(bdev_file))
+		bd_end_claim(bdev, holder);
+}
+
 void bdev_release(struct file *bdev_file)
 {
 	struct block_device *bdev = file_bdev(bdev_file);
@@ -1036,7 +1051,7 @@ void bdev_release(struct file *bdev_file)
 	bdev_yield_write_access(bdev_file);
 
 	if (holder)
-		bd_end_claim(bdev, holder);
+		bd_yield_claim(bdev_file);
 
 	/*
 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
@@ -1056,6 +1071,39 @@ put_no_open:
 	blkdev_put_no_open(bdev);
 }
 
+/**
+ * bdev_fput - yield claim to the block device and put the file
+ * @bdev_file: open block device
+ *
+ * Yield claim on the block device and put the file. Ensure that the
+ * block device can be reclaimed before the file is closed which is a
+ * deferred operation.
+ */
+void bdev_fput(struct file *bdev_file)
+{
+	if (WARN_ON_ONCE(bdev_file->f_op != &def_blk_fops))
+		return;
+
+	if (bdev_file->private_data) {
+		struct block_device *bdev = file_bdev(bdev_file);
+		struct gendisk *disk = bdev->bd_disk;
+
+		mutex_lock(&disk->open_mutex);
+		bdev_yield_write_access(bdev_file);
+		bd_yield_claim(bdev_file);
+		/*
+		 * Tell release we already gave up our hold on the
+		 * device and if write restrictions are available that
+		 * we already gave up write access to the device.
+		 */
+		bdev_file->private_data = BDEV_I(bdev_file->f_mapping->host);
+		mutex_unlock(&disk->open_mutex);
+	}
+
+	fput(bdev_file);
+}
+EXPORT_SYMBOL(bdev_fput);
+
 /**
  * lookup_bdev() - Look up a struct block_device by name.
  * @pathname: Name of the block device in the filesystem.

drivers/mtd/devices/block2mtd.c

@@ -209,7 +209,7 @@ static void block2mtd_free_device(struct block2mtd_dev *dev)
 
 	if (dev->bdev_file) {
 		invalidate_mapping_pages(dev->bdev_file->f_mapping, 0, -1);
-		fput(dev->bdev_file);
+		bdev_fput(dev->bdev_file);
 	}
 
 	kfree(dev);

fs/bcachefs/super-io.c

@@ -143,7 +143,7 @@ void bch2_free_super(struct bch_sb_handle *sb)
 {
 	kfree(sb->bio);
 	if (!IS_ERR_OR_NULL(sb->s_bdev_file))
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	kfree(sb->holder);
 	kfree(sb->sb_name);
 

fs/cramfs/inode.c

@@ -495,7 +495,7 @@ static void cramfs_kill_sb(struct super_block *sb)
 		sb->s_mtd = NULL;
 	} else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV) && sb->s_bdev) {
 		sync_blockdev(sb->s_bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 	kfree(sbi);
 }

fs/ext4/super.c

@@ -5668,7 +5668,7 @@ failed_mount:
 	brelse(sbi->s_sbh);
 	if (sbi->s_journal_bdev_file) {
 		invalidate_bdev(file_bdev(sbi->s_journal_bdev_file));
-		fput(sbi->s_journal_bdev_file);
+		bdev_fput(sbi->s_journal_bdev_file);
 	}
 out_fail:
 	invalidate_bdev(sb->s_bdev);
@@ -5913,7 +5913,7 @@ static struct file *ext4_get_journal_blkdev(struct super_block *sb,
 out_bh:
 	brelse(bh);
 out_bdev:
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 	return ERR_PTR(errno);
 }
 
@@ -5952,7 +5952,7 @@ static journal_t *ext4_open_dev_journal(struct super_block *sb,
 out_journal:
 	jbd2_journal_destroy(journal);
 out_bdev:
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 	return ERR_PTR(errno);
 }
 
@@ -7327,7 +7327,7 @@ static void ext4_kill_sb(struct super_block *sb)
 	kill_block_super(sb);
 
 	if (bdev_file)
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 }
 
 static struct file_system_type ext4_fs_type = {

fs/f2fs/super.c

@@ -1558,7 +1558,7 @@ static void destroy_device_list(struct f2fs_sb_info *sbi)
 
 	for (i = 0; i < sbi->s_ndevs; i++) {
 		if (i > 0)
-			fput(FDEV(i).bdev_file);
+			bdev_fput(FDEV(i).bdev_file);
 #ifdef CONFIG_BLK_DEV_ZONED
 		kvfree(FDEV(i).blkz_seq);
 #endif

fs/jfs/jfs_logmgr.c

@@ -1141,7 +1141,7 @@ journal_found:
 	lbmLogShutdown(log);
 
       close:		/* close external log device */
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 
       free:		/* free log descriptor */
 	mutex_unlock(&jfs_log_mutex);
@@ -1485,7 +1485,7 @@ int lmLogClose(struct super_block *sb)
 	bdev_file = log->bdev_file;
 	rc = lmLogShutdown(log);
 
-	fput(bdev_file);
+	bdev_fput(bdev_file);
 
 	kfree(log);
 

fs/reiserfs/journal.c

@@ -2589,7 +2589,7 @@ static void journal_list_init(struct super_block *sb)
 static void release_journal_dev(struct reiserfs_journal *journal)
 {
 	if (journal->j_bdev_file) {
-		fput(journal->j_bdev_file);
+		bdev_fput(journal->j_bdev_file);
 		journal->j_bdev_file = NULL;
 	}
 }

fs/romfs/super.c

@@ -594,7 +594,7 @@ static void romfs_kill_sb(struct super_block *sb)
 #ifdef CONFIG_ROMFS_ON_BLOCK
 	if (sb->s_bdev) {
 		sync_blockdev(sb->s_bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 #endif
 }

fs/super.c

@@ -1515,29 +1515,11 @@ static int fs_bdev_thaw(struct block_device *bdev)
 	return error;
 }
 
-static void fs_bdev_super_get(void *data)
-{
-	struct super_block *sb = data;
-
-	spin_lock(&sb_lock);
-	sb->s_count++;
-	spin_unlock(&sb_lock);
-}
-
-static void fs_bdev_super_put(void *data)
-{
-	struct super_block *sb = data;
-
-	put_super(sb);
-}
-
 const struct blk_holder_ops fs_holder_ops = {
 	.mark_dead		= fs_bdev_mark_dead,
 	.sync			= fs_bdev_sync,
 	.freeze			= fs_bdev_freeze,
 	.thaw			= fs_bdev_thaw,
-	.get_holder		= fs_bdev_super_get,
-	.put_holder		= fs_bdev_super_put,
 };
 EXPORT_SYMBOL_GPL(fs_holder_ops);
 
@@ -1562,7 +1544,7 @@ int setup_bdev_super(struct super_block *sb, int sb_flags,
 	 * writable from userspace even for a read-only block device.
 	 */
 	if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 		return -EACCES;
 	}
 
@@ -1573,7 +1555,7 @@ int setup_bdev_super(struct super_block *sb, int sb_flags,
 	if (atomic_read(&bdev->bd_fsfreeze_count) > 0) {
 		if (fc)
 			warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
-		fput(bdev_file);
+		bdev_fput(bdev_file);
 		return -EBUSY;
 	}
 	spin_lock(&sb_lock);
@@ -1693,7 +1675,7 @@ void kill_block_super(struct super_block *sb)
 	generic_shutdown_super(sb);
 	if (bdev) {
 		sync_blockdev(bdev);
-		fput(sb->s_bdev_file);
+		bdev_fput(sb->s_bdev_file);
 	}
 }
 

fs/xfs/xfs_buf.c

@@ -2030,7 +2030,7 @@ xfs_free_buftarg(
 	fs_put_dax(btp->bt_daxdev, btp->bt_mount);
 	/* the main block device is closed by kill_block_super */
 	if (btp->bt_bdev != btp->bt_mount->m_super->s_bdev)
-		fput(btp->bt_bdev_file);
+		bdev_fput(btp->bt_bdev_file);
 	kfree(btp);
 }
 

fs/xfs/xfs_super.c

@@ -485,7 +485,7 @@ xfs_open_devices(
 		mp->m_logdev_targp = mp->m_ddev_targp;
 		/* Handle won't be used, drop it */
 		if (logdev_file)
-			fput(logdev_file);
+			bdev_fput(logdev_file);
 	}
 
 	return 0;
@@ -497,10 +497,10 @@ xfs_open_devices(
 	xfs_free_buftarg(mp->m_ddev_targp);
  out_close_rtdev:
 	 if (rtdev_file)
-		fput(rtdev_file);
+		bdev_fput(rtdev_file);
  out_close_logdev:
 	if (logdev_file)
-		fput(logdev_file);
+		bdev_fput(logdev_file);
 	return error;
 }
 

include/linux/blkdev.h

@@ -1505,16 +1505,6 @@ struct blk_holder_ops {
 	 * Thaw the file system mounted on the block device.
 	 */
 	int (*thaw)(struct block_device *bdev);
-
-	/*
-	 * If needed, get a reference to the holder.
-	 */
-	void (*get_holder)(void *holder);
-
-	/*
-	 * Release the holder.
-	 */
-	void (*put_holder)(void *holder);
 };
 
 /*
@@ -1585,6 +1575,7 @@ static inline int early_lookup_bdev(const char *pathname, dev_t *dev)
 
 int bdev_freeze(struct block_device *bdev);
 int bdev_thaw(struct block_device *bdev);
+void bdev_fput(struct file *bdev_file);
 
 struct io_comp_batch {
 	struct request *req_list;