Commit da105ed5 authored by Joe Thornber's avatar Joe Thornber Committed by Alasdair G Kergon
Browse files

dm thin metadata: introduce dm_pool_abort_metadata 



Introduce dm_pool_abort_metadata to abort the current metadata
transaction.  Generally this will only be called when bad things are
happening and dm-thin is trying to roll back to a good state for
read-only mode.

It's complicated by the fact that the metadata device may have failed
completely causing the abort to be unable to read the old transaction.
In this case the metadata object is placed in a 'fail' mode and
everything fails apart from destroying it.

Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
Signed-off-by: default avatarAlasdair G Kergon <agk@redhat.com>
parent 12ba58af

drivers/md/dm-thin-metadata.c

+146 −61
Original line number Diff line number Diff line 
/*

 * Copyright (C) 2011 Red Hat, Inc.

 * Copyright (C) 2011-2012 Red Hat, Inc.

 *

 * This file is released under the GPL.

 */
@@ -185,6 +185,13 @@ struct dm_pool_metadata { 
	unsigned long flags;

	sector_t data_block_size;

	bool read_only:1;



	/*

	 * Set if a transaction has to be aborted but the attempt to roll back

	 * to the previous (good) transaction failed.  The only pool metadata

	 * operation possible in this state is the closing of the device.

	 */

	bool fail_io:1;

};



struct dm_thin_device {
@@ -193,7 +200,8 @@ struct dm_thin_device { 
	dm_thin_id id;



	int open_count;

	int changed;

	bool changed:1;

	bool aborted_with_changes:1;

	uint64_t mapped_blocks;

	uint64_t transaction_id;

	uint32_t creation_time;
@@ -731,9 +739,6 @@ static int __write_changed_details(struct dm_pool_metadata *pmd) 


static int __commit_transaction(struct dm_pool_metadata *pmd)

{

	/*

	 * FIXME: Associated pool should be made read-only on failure.

	 */

	int r;

	size_t metadata_len, data_len;

	struct thin_disk_superblock *disk_super;
@@ -809,6 +814,7 @@ struct dm_pool_metadata *dm_pool_metadata_open(struct block_device *bdev, 
	pmd->time = 0;

	INIT_LIST_HEAD(&pmd->thin_devices);

	pmd->read_only = false;

	pmd->fail_io = false;

	pmd->bdev = bdev;

	pmd->data_block_size = data_block_size;
@@ -851,16 +857,17 @@ int dm_pool_metadata_close(struct dm_pool_metadata *pmd) 
		return -EBUSY;

	}



	if (!pmd->read_only) {

	if (!pmd->read_only && !pmd->fail_io) {

		r = __commit_transaction(pmd);

		if (r < 0)

			DMWARN("%s: __commit_transaction() failed, error = %d",

			       __func__, r);

	}



	if (!pmd->fail_io)

		__destroy_persistent_data_objects(pmd);

	kfree(pmd);



	kfree(pmd);

	return 0;

}
@@ -921,6 +928,7 @@ static int __open_device(struct dm_pool_metadata *pmd, 
	(*td)->id = dev;

	(*td)->open_count = 1;

	(*td)->changed = changed;

	(*td)->aborted_with_changes = false;

	(*td)->mapped_blocks = le64_to_cpu(details_le.mapped_blocks);

	(*td)->transaction_id = le64_to_cpu(details_le.transaction_id);

	(*td)->creation_time = le32_to_cpu(details_le.creation_time);
@@ -982,9 +990,10 @@ static int __create_thin(struct dm_pool_metadata *pmd, 


int dm_pool_create_thin(struct dm_pool_metadata *pmd, dm_thin_id dev)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __create_thin(pmd, dev);

	up_write(&pmd->root_lock);
@@ -1072,9 +1081,10 @@ int dm_pool_create_snap(struct dm_pool_metadata *pmd, 
				 dm_thin_id dev,

				 dm_thin_id origin)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __create_snap(pmd, dev, origin);

	up_write(&pmd->root_lock);
@@ -1114,9 +1124,10 @@ static int __delete_device(struct dm_pool_metadata *pmd, dm_thin_id dev) 
int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd,

			       dm_thin_id dev)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __delete_device(pmd, dev);

	up_write(&pmd->root_lock);
@@ -1127,27 +1138,40 @@ int dm_pool_set_metadata_transaction_id(struct dm_pool_metadata *pmd, 
					uint64_t current_id,

					uint64_t new_id)

{

	int r = -EINVAL;



	down_write(&pmd->root_lock);



	if (pmd->fail_io)

		goto out;



	if (pmd->trans_id != current_id) {

		up_write(&pmd->root_lock);

		DMERR("mismatched transaction id");

		return -EINVAL;

		goto out;

	}



	pmd->trans_id = new_id;

	r = 0;



out:

	up_write(&pmd->root_lock);



	return 0;

	return r;

}



int dm_pool_get_metadata_transaction_id(struct dm_pool_metadata *pmd,

					uint64_t *result)

{

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io) {

		*result = pmd->trans_id;

		r = 0;

	}

	up_read(&pmd->root_lock);



	return 0;

	return r;

}



static int __reserve_metadata_snap(struct dm_pool_metadata *pmd)
@@ -1211,9 +1235,10 @@ static int __reserve_metadata_snap(struct dm_pool_metadata *pmd) 


int dm_pool_reserve_metadata_snap(struct dm_pool_metadata *pmd)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __reserve_metadata_snap(pmd);

	up_write(&pmd->root_lock);
@@ -1256,9 +1281,10 @@ static int __release_metadata_snap(struct dm_pool_metadata *pmd) 


int dm_pool_release_metadata_snap(struct dm_pool_metadata *pmd)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __release_metadata_snap(pmd);

	up_write(&pmd->root_lock);
@@ -1286,9 +1312,10 @@ static int __get_metadata_snap(struct dm_pool_metadata *pmd, 
int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd,

			      dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __get_metadata_snap(pmd, result);

	up_read(&pmd->root_lock);
@@ -1298,9 +1325,10 @@ int dm_pool_get_metadata_snap(struct dm_pool_metadata *pmd, 
int dm_pool_open_thin_device(struct dm_pool_metadata *pmd, dm_thin_id dev,

			     struct dm_thin_device **td)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __open_device(pmd, dev, 0, td);

	up_write(&pmd->root_lock);
@@ -1329,27 +1357,30 @@ static bool __snapshotted_since(struct dm_thin_device *td, uint32_t time) 
int dm_thin_find_block(struct dm_thin_device *td, dm_block_t block,

		       int can_block, struct dm_thin_lookup_result *result)

{

	int r;

	int r = -EINVAL;

	uint64_t block_time = 0;

	__le64 value;

	struct dm_pool_metadata *pmd = td->pmd;

	dm_block_t keys[2] = { td->id, block };

	struct dm_btree_info *info;



	if (can_block) {

		down_read(&pmd->root_lock);

		r = dm_btree_lookup(&pmd->info, pmd->root, keys, &value);

		if (!r)

			block_time = le64_to_cpu(value);

		up_read(&pmd->root_lock);

		info = &pmd->info;

	} else if (down_read_trylock(&pmd->root_lock))

		info = &pmd->nb_info;

	else

		return -EWOULDBLOCK;



	if (pmd->fail_io)

		goto out;



	} else if (down_read_trylock(&pmd->root_lock)) {

		r = dm_btree_lookup(&pmd->nb_info, pmd->root, keys, &value);

	r = dm_btree_lookup(info, pmd->root, keys, &value);

	if (!r)

		block_time = le64_to_cpu(value);

		up_read(&pmd->root_lock);



	} else

		return -EWOULDBLOCK;

out:

	up_read(&pmd->root_lock);



	if (!r) {

		dm_block_t exception_block;
@@ -1389,9 +1420,10 @@ static int __insert(struct dm_thin_device *td, dm_block_t block, 
int dm_thin_insert_block(struct dm_thin_device *td, dm_block_t block,

			 dm_block_t data_block)

{

	int r;

	int r = -EINVAL;



	down_write(&td->pmd->root_lock);

	if (!td->pmd->fail_io)

		r = __insert(td, block, data_block);

	up_write(&td->pmd->root_lock);
@@ -1416,9 +1448,10 @@ static int __remove(struct dm_thin_device *td, dm_block_t block) 


int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block)

{

	int r;

	int r = -EINVAL;



	down_write(&td->pmd->root_lock);

	if (!td->pmd->fail_io)

		r = __remove(td, block);

	up_write(&td->pmd->root_lock);
@@ -1436,11 +1469,23 @@ bool dm_thin_changed_this_transaction(struct dm_thin_device *td) 
	return r;

}



bool dm_thin_aborted_changes(struct dm_thin_device *td)

{

	bool r;



	down_read(&td->pmd->root_lock);

	r = td->aborted_with_changes;

	up_read(&td->pmd->root_lock);



	return r;

}



int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = dm_sm_new_block(pmd->data_sm, result);

	up_write(&pmd->root_lock);
@@ -1449,9 +1494,11 @@ int dm_pool_alloc_data_block(struct dm_pool_metadata *pmd, dm_block_t *result) 


int dm_pool_commit_metadata(struct dm_pool_metadata *pmd)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (pmd->fail_io)

		goto out;



	r = __commit_transaction(pmd);

	if (r <= 0)
@@ -1466,11 +1513,40 @@ out: 
	return r;

}



static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)

{

	struct dm_thin_device *td;



	list_for_each_entry(td, &pmd->thin_devices, list)

		td->aborted_with_changes = td->changed;

}



int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)

{

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (pmd->fail_io)

		goto out;



	__set_abort_with_changes_flags(pmd);

	__destroy_persistent_data_objects(pmd);

	r = __create_persistent_data_objects(pmd, false);

	if (r)

		pmd->fail_io = true;



out:

	up_write(&pmd->root_lock);



	return r;

}



int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = dm_sm_get_nr_free(pmd->data_sm, result);

	up_read(&pmd->root_lock);
@@ -1480,9 +1556,10 @@ int dm_pool_get_free_block_count(struct dm_pool_metadata *pmd, dm_block_t *resul 
int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd,

					  dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = dm_sm_get_nr_free(pmd->metadata_sm, result);

	up_read(&pmd->root_lock);
@@ -1492,9 +1569,10 @@ int dm_pool_get_free_metadata_block_count(struct dm_pool_metadata *pmd, 
int dm_pool_get_metadata_dev_size(struct dm_pool_metadata *pmd,

				  dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = dm_sm_get_nr_blocks(pmd->metadata_sm, result);

	up_read(&pmd->root_lock);
@@ -1512,9 +1590,10 @@ int dm_pool_get_data_block_size(struct dm_pool_metadata *pmd, sector_t *result) 


int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result)

{

	int r;

	int r = -EINVAL;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = dm_sm_get_nr_blocks(pmd->data_sm, result);

	up_read(&pmd->root_lock);
@@ -1523,13 +1602,17 @@ int dm_pool_get_data_dev_size(struct dm_pool_metadata *pmd, dm_block_t *result) 


int dm_thin_get_mapped_count(struct dm_thin_device *td, dm_block_t *result)

{

	int r = -EINVAL;

	struct dm_pool_metadata *pmd = td->pmd;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io) {

		*result = td->mapped_blocks;

		r = 0;

	}

	up_read(&pmd->root_lock);



	return 0;

	return r;

}



static int __highest_block(struct dm_thin_device *td, dm_block_t *result)
@@ -1551,10 +1634,11 @@ static int __highest_block(struct dm_thin_device *td, dm_block_t *result) 
int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,

				     dm_block_t *result)

{

	int r;

	int r = -EINVAL;

	struct dm_pool_metadata *pmd = td->pmd;



	down_read(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __highest_block(td, result);

	up_read(&pmd->root_lock);
@@ -1583,9 +1667,10 @@ static int __resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count) 


int dm_pool_resize_data_dev(struct dm_pool_metadata *pmd, dm_block_t new_count)

{

	int r;

	int r = -EINVAL;



	down_write(&pmd->root_lock);

	if (!pmd->fail_io)

		r = __resize_data_dev(pmd, new_count);

	up_write(&pmd->root_lock);

drivers/md/dm-thin-metadata.h

+12 −0
Original line number Diff line number Diff line
@@ -79,6 +79,16 @@ int dm_pool_delete_thin_device(struct dm_pool_metadata *pmd, 
 */

int dm_pool_commit_metadata(struct dm_pool_metadata *pmd);



/*

 * Discards all uncommitted changes.  Rereads the superblock, rolling back

 * to the last good transaction.  Thin devices remain open.

 * dm_thin_aborted_changes() tells you if they had uncommitted changes.

 *

 * If this call fails it's only useful to call dm_pool_metadata_close().

 * All other methods will fail with -EINVAL.

 */

int dm_pool_abort_metadata(struct dm_pool_metadata *pmd);



/*

 * Set/get userspace transaction id.

 */
@@ -150,6 +160,8 @@ int dm_thin_remove_block(struct dm_thin_device *td, dm_block_t block); 
 */

bool dm_thin_changed_this_transaction(struct dm_thin_device *td);



bool dm_thin_aborted_changes(struct dm_thin_device *td);



int dm_thin_get_highest_mapped_block(struct dm_thin_device *td,

				     dm_block_t *highest_mapped);

CRA Git | Maintained and supported by SUSTech CRA and CCSE