1 files changed, 99 insertions, 92 deletions

diff --git a/fs/btrfs/dev-replace.c b/fs/btrfs/dev-replace.c
index 2aa48aecc52b..8750c835f535 100644
--- a/fs/btrfs/dev-replace.c
+++ b/fs/btrfs/dev-replace.c
@@ -59,7 +59,6 @@ no_valid_dev_replace_entry_found:
 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
 		dev_replace->cont_reading_from_srcdev_mode =
 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
-		dev_replace->replace_state = 0;
 		dev_replace->time_started = 0;
 		dev_replace->time_stopped = 0;
 		atomic64_set(&dev_replace->num_write_errors, 0);
@@ -285,13 +284,13 @@ int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 	struct btrfs_dev_replace_item *ptr;
 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 
-	btrfs_dev_replace_read_lock(dev_replace);
+	down_read(&dev_replace->rwsem);
 	if (!dev_replace->is_valid ||
 	    !dev_replace->item_needs_writeback) {
-		btrfs_dev_replace_read_unlock(dev_replace);
+		up_read(&dev_replace->rwsem);
 		return 0;
 	}
-	btrfs_dev_replace_read_unlock(dev_replace);
+	up_read(&dev_replace->rwsem);
 
 	key.objectid = 0;
 	key.type = BTRFS_DEV_REPLACE_KEY;
@@ -349,7 +348,7 @@ int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 	ptr = btrfs_item_ptr(eb, path->slots[0],
 			     struct btrfs_dev_replace_item);
 
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
 	if (dev_replace->srcdev)
 		btrfs_set_dev_replace_src_devid(eb, ptr,
 			dev_replace->srcdev->devid);
@@ -372,7 +371,7 @@ int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 	btrfs_set_dev_replace_cursor_right(eb, ptr,
 		dev_replace->cursor_right);
 	dev_replace->item_needs_writeback = 0;
-	btrfs_dev_replace_write_unlock(dev_replace);
+	up_write(&dev_replace->rwsem);
 
 	btrfs_mark_buffer_dirty(eb);
 
@@ -390,7 +389,7 @@ static char* btrfs_dev_name(struct btrfs_device *device)
 		return rcu_str_deref(device->name);
 }
 
-int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
+static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 		int read_src)
 {
@@ -407,6 +406,13 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 	if (IS_ERR(src_device))
 		return PTR_ERR(src_device);
 
+	if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
+		btrfs_warn_in_rcu(fs_info,
+	  "cannot replace device %s (devid %llu) due to active swapfile",
+			btrfs_dev_name(src_device), src_device->devid);
+		return -ETXTBSY;
+	}
+
 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 					    src_device, &tgt_device);
 	if (ret)
@@ -426,7 +432,7 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 	}
 
 	need_unlock = true;
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
 	switch (dev_replace->replace_state) {
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
@@ -464,7 +470,7 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 	dev_replace->item_needs_writeback = 1;
 	atomic64_set(&dev_replace->num_write_errors, 0);
 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
-	btrfs_dev_replace_write_unlock(dev_replace);
+	up_write(&dev_replace->rwsem);
 	need_unlock = false;
 
 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
@@ -478,7 +484,7 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 	if (IS_ERR(trans)) {
 		ret = PTR_ERR(trans);
 		need_unlock = true;
-		btrfs_dev_replace_write_lock(dev_replace);
+		down_write(&dev_replace->rwsem);
 		dev_replace->replace_state =
 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 		dev_replace->srcdev = NULL;
@@ -497,7 +503,7 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 	ret = btrfs_dev_replace_finishing(fs_info, ret);
 	if (ret == -EINPROGRESS) {
 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
-	} else {
+	} else if (ret != -ECANCELED) {
 		WARN_ON(ret);
 	}
 
@@ -505,7 +511,7 @@ int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 
 leave:
 	if (need_unlock)
-		btrfs_dev_replace_write_unlock(dev_replace);
+		up_write(&dev_replace->rwsem);
 	btrfs_destroy_dev_replace_tgtdev(tgt_device);
 	return ret;
 }
@@ -533,8 +539,9 @@ int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 					args->start.cont_reading_from_srcdev_mode);
 	args->result = ret;
 	/* don't warn if EINPROGRESS, someone else might be running scrub */
-	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
-		ret = 0;
+	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
+	    ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
+		return 0;
 
 	return ret;
 }
@@ -572,18 +579,18 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	/* don't allow cancel or unmount to disturb the finishing procedure */
 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 
-	btrfs_dev_replace_read_lock(dev_replace);
+	down_read(&dev_replace->rwsem);
 	/* was the operation canceled, or is it finished? */
 	if (dev_replace->replace_state !=
 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
-		btrfs_dev_replace_read_unlock(dev_replace);
+		up_read(&dev_replace->rwsem);
 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 		return 0;
 	}
 
 	tgt_device = dev_replace->tgtdev;
 	src_device = dev_replace->srcdev;
-	btrfs_dev_replace_read_unlock(dev_replace);
+	up_read(&dev_replace->rwsem);
 
 	/*
 	 * flush all outstanding I/O and inode extent mappings before the
@@ -607,7 +614,7 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	/* keep away write_all_supers() during the finishing procedure */
 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
 	mutex_lock(&fs_info->chunk_mutex);
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
 	dev_replace->replace_state =
 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
@@ -622,12 +629,13 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 								src_device,
 								tgt_device);
 	} else {
-		btrfs_err_in_rcu(fs_info,
+		if (scrub_ret != -ECANCELED)
+			btrfs_err_in_rcu(fs_info,
 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 				 btrfs_dev_name(src_device),
 				 src_device->devid,
 				 rcu_str_deref(tgt_device->name), scrub_ret);
-		btrfs_dev_replace_write_unlock(dev_replace);
+		up_write(&dev_replace->rwsem);
 		mutex_unlock(&fs_info->chunk_mutex);
 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 		btrfs_rm_dev_replace_blocked(fs_info);
@@ -663,8 +671,7 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 	fs_info->fs_devices->rw_devices++;
 
-	btrfs_dev_replace_write_unlock(dev_replace);
-
+	up_write(&dev_replace->rwsem);
 	btrfs_rm_dev_replace_blocked(fs_info);
 
 	btrfs_rm_dev_replace_remove_srcdev(src_device);
@@ -761,7 +768,7 @@ void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 {
 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 
-	btrfs_dev_replace_read_lock(dev_replace);
+	down_read(&dev_replace->rwsem);
 	/* even if !dev_replace_is_valid, the values are good enough for
 	 * the replace_status ioctl */
 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
@@ -773,7 +780,7 @@ void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 	args->status.num_uncorrectable_read_errors =
 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
-	btrfs_dev_replace_read_unlock(dev_replace);
+	up_read(&dev_replace->rwsem);
 }
 
 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
@@ -790,46 +797,74 @@ int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 		return -EROFS;
 
 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
 	switch (dev_replace->replace_state) {
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
-		btrfs_dev_replace_write_unlock(dev_replace);
-		goto leave;
+		up_write(&dev_replace->rwsem);
+		break;
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
+		tgt_device = dev_replace->tgtdev;
+		src_device = dev_replace->srcdev;
+		up_write(&dev_replace->rwsem);
+		ret = btrfs_scrub_cancel(fs_info);
+		if (ret < 0) {
+			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
+		} else {
+			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
+			/*
+			 * btrfs_dev_replace_finishing() will handle the
+			 * cleanup part
+			 */
+			btrfs_info_in_rcu(fs_info,
+				"dev_replace from %s (devid %llu) to %s canceled",
+				btrfs_dev_name(src_device), src_device->devid,
+				btrfs_dev_name(tgt_device));
+		}
+		break;
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
+		/*
+		 * Scrub doing the replace isn't running so we need to do the
+		 * cleanup step of btrfs_dev_replace_finishing() here
+		 */
 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 		tgt_device = dev_replace->tgtdev;
 		src_device = dev_replace->srcdev;
 		dev_replace->tgtdev = NULL;
 		dev_replace->srcdev = NULL;
-		break;
-	}
-	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
-	dev_replace->time_stopped = ktime_get_real_seconds();
-	dev_replace->item_needs_writeback = 1;
-	btrfs_dev_replace_write_unlock(dev_replace);
-	btrfs_scrub_cancel(fs_info);
+		dev_replace->replace_state =
+				BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
+		dev_replace->time_stopped = ktime_get_real_seconds();
+		dev_replace->item_needs_writeback = 1;
 
-	trans = btrfs_start_transaction(root, 0);
-	if (IS_ERR(trans)) {
-		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
-		return PTR_ERR(trans);
-	}
-	ret = btrfs_commit_transaction(trans);
-	WARN_ON(ret);
+		up_write(&dev_replace->rwsem);
 
-	btrfs_info_in_rcu(fs_info,
-		"dev_replace from %s (devid %llu) to %s canceled",
-		btrfs_dev_name(src_device), src_device->devid,
-		btrfs_dev_name(tgt_device));
+		/* Scrub for replace must not be running in suspended state */
+		ret = btrfs_scrub_cancel(fs_info);
+		ASSERT(ret != -ENOTCONN);
+
+		trans = btrfs_start_transaction(root, 0);
+		if (IS_ERR(trans)) {
+			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
+			return PTR_ERR(trans);
+		}
+		ret = btrfs_commit_transaction(trans);
+		WARN_ON(ret);
 
-	if (tgt_device)
-		btrfs_destroy_dev_replace_tgtdev(tgt_device);
+		btrfs_info_in_rcu(fs_info,
+		"suspended dev_replace from %s (devid %llu) to %s canceled",
+			btrfs_dev_name(src_device), src_device->devid,
+			btrfs_dev_name(tgt_device));
+
+		if (tgt_device)
+			btrfs_destroy_dev_replace_tgtdev(tgt_device);
+		break;
+	default:
+		result = -EINVAL;
+	}
 
-leave:
 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 	return result;
 }
@@ -839,7 +874,8 @@ void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 
 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
+
 	switch (dev_replace->replace_state) {
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
@@ -855,7 +891,7 @@ void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 		break;
 	}
 
-	btrfs_dev_replace_write_unlock(dev_replace);
+	up_write(&dev_replace->rwsem);
 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 }
 
@@ -865,12 +901,13 @@ int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 	struct task_struct *task;
 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 
-	btrfs_dev_replace_write_lock(dev_replace);
+	down_write(&dev_replace->rwsem);
+
 	switch (dev_replace->replace_state) {
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
-		btrfs_dev_replace_write_unlock(dev_replace);
+		up_write(&dev_replace->rwsem);
 		return 0;
 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 		break;
@@ -884,10 +921,12 @@ int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 			   "cannot continue dev_replace, tgtdev is missing");
 		btrfs_info(fs_info,
 			   "you may cancel the operation after 'mount -o degraded'");
-		btrfs_dev_replace_write_unlock(dev_replace);
+		dev_replace->replace_state =
+					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
+		up_write(&dev_replace->rwsem);
 		return 0;
 	}
-	btrfs_dev_replace_write_unlock(dev_replace);
+	up_write(&dev_replace->rwsem);
 
 	/*
 	 * This could collide with a paused balance, but the exclusive op logic
@@ -895,6 +934,10 @@ int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 	 * dev-replace to start anyway.
 	 */
 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
+		down_write(&dev_replace->rwsem);
+		dev_replace->replace_state =
+					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
+		up_write(&dev_replace->rwsem);
 		btrfs_info(fs_info,
 		"cannot resume dev-replace, other exclusive operation running");
 		return 0;
@@ -925,7 +968,7 @@ static int btrfs_dev_replace_kthread(void *data)
 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
 			      &dev_replace->scrub_progress, 0, 1);
 	ret = btrfs_dev_replace_finishing(fs_info, ret);
-	WARN_ON(ret);
+	WARN_ON(ret && ret != -ECANCELED);
 
 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 	return 0;
@@ -948,7 +991,7 @@ int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 		 * something that can happen if the dev_replace
 		 * procedure is suspended by an umount and then
 		 * the tgtdev is missing (or "btrfs dev scan") was
-		 * not called and the the filesystem is remounted
+		 * not called and the filesystem is remounted
 		 * in degraded state. This does not stop the
 		 * dev_replace procedure. It needs to be canceled
 		 * manually if the cancellation is wanted.
@@ -958,42 +1001,6 @@ int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 	return 1;
 }
 
-void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
-{
-	read_lock(&dev_replace->lock);
-}
-
-void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
-{
-	read_unlock(&dev_replace->lock);
-}
-
-void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
-{
-again:
-	wait_event(dev_replace->read_lock_wq,
-		   atomic_read(&dev_replace->blocking_readers) == 0);
-	write_lock(&dev_replace->lock);
-	if (atomic_read(&dev_replace->blocking_readers)) {
-		write_unlock(&dev_replace->lock);
-		goto again;
-	}
-}
-
-void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
-{
-	write_unlock(&dev_replace->lock);
-}
-
-/* inc blocking cnt and release read lock */
-void btrfs_dev_replace_set_lock_blocking(
-					struct btrfs_dev_replace *dev_replace)
-{
-	/* only set blocking for read lock */
-	atomic_inc(&dev_replace->blocking_readers);
-	read_unlock(&dev_replace->lock);
-}
-
 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
 {
 	percpu_counter_inc(&fs_info->dev_replace.bio_counter);