Merge: Fix outstanding device-mapper bugs from upstream 6.1 and 6.2 an dm-cache cleanup issue

MR: https://gitlab.com/redhat/centos-stream/src/kernel/centos-stream-9/-/merge_requests/1923

Bugzilla: https://bugzilla.redhat.com/2162536
Bugzilla: https://bugzilla.redhat.com/2163655
Tested: lvm-testsuite
Upstream Status: kernel/git/torvalds/linux.git

Pull in dm bugfixes from upstream and fix an existing dm-cache bug found while testing the changes.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>

Approved-by: Mike Snitzer <snitzer@redhat.com>
Approved-by: Mikuláš Patočka <mpatocka@redhat.com>

Signed-off-by: Herton R. Krzesinski <herton@redhat.com>

Author: Herton R. Krzesinski

drivers/md/dm-bufio.c

@@ -1857,6 +1857,8 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign
 	dm_io_client_destroy(c->dm_io);
 bad_dm_io:
 	mutex_destroy(&c->lock);
+	if (c->no_sleep)
+		static_branch_dec(&no_sleep_enabled);
 	kfree(c);
 bad_client:
 	return ERR_PTR(r);

drivers/md/dm-cache-background-tracker.c

@@ -60,6 +60,14 @@ EXPORT_SYMBOL_GPL(btracker_create);
 
 void btracker_destroy(struct background_tracker *b)
 {
+	struct bt_work *w, *tmp;
+
+	BUG_ON(!list_empty(&b->issued));
+	list_for_each_entry_safe (w, tmp, &b->queued, list) {
+		list_del(&w->list);
+		kmem_cache_free(b->work_cache, w);
+	}
+
 	kmem_cache_destroy(b->work_cache);
 	kfree(b);
 }

drivers/md/dm-cache-metadata.c

@@ -551,11 +551,13 @@ static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
 	return r;
 }
 
-static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
+static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd,
+					      bool destroy_bm)
 {
 	dm_sm_destroy(cmd->metadata_sm);
 	dm_tm_destroy(cmd->tm);
-	dm_block_manager_destroy(cmd->bm);
+	if (destroy_bm)
+		dm_block_manager_destroy(cmd->bm);
 }
 
 typedef unsigned long (*flags_mutator)(unsigned long);
@@ -826,7 +828,7 @@ static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
 		cmd2 = lookup(bdev);
 		if (cmd2) {
 			mutex_unlock(&table_lock);
-			__destroy_persistent_data_objects(cmd);
+			__destroy_persistent_data_objects(cmd, true);
 			kfree(cmd);
 			return cmd2;
 		}
@@ -874,7 +876,7 @@ void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
 		mutex_unlock(&table_lock);
 
 		if (!cmd->fail_io)
-			__destroy_persistent_data_objects(cmd);
+			__destroy_persistent_data_objects(cmd, true);
 		kfree(cmd);
 	}
 }
@@ -1807,14 +1809,52 @@ int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
 
 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
 {
-	int r;
+	int r = -EINVAL;
+	struct dm_block_manager *old_bm = NULL, *new_bm = NULL;
+
+	/* fail_io is double-checked with cmd->root_lock held below */
+	if (unlikely(cmd->fail_io))
+		return r;
+
+	/*
+	 * Replacement block manager (new_bm) is created and old_bm destroyed outside of
+	 * cmd root_lock to avoid ABBA deadlock that would result (due to life-cycle of
+	 * shrinker associated with the block manager's bufio client vs cmd root_lock).
+	 * - must take shrinker_rwsem without holding cmd->root_lock
+	 */
+	new_bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
+					 CACHE_MAX_CONCURRENT_LOCKS);
 
 	WRITE_LOCK(cmd);
-	__destroy_persistent_data_objects(cmd);
-	r = __create_persistent_data_objects(cmd, false);
+	if (cmd->fail_io) {
+		WRITE_UNLOCK(cmd);
+		goto out;
+	}
+
+	__destroy_persistent_data_objects(cmd, false);
+	old_bm = cmd->bm;
+	if (IS_ERR(new_bm)) {
+		DMERR("could not create block manager during abort");
+		cmd->bm = NULL;
+		r = PTR_ERR(new_bm);
+		goto out_unlock;
+	}
+
+	cmd->bm = new_bm;
+	r = __open_or_format_metadata(cmd, false);
+	if (r) {
+		cmd->bm = NULL;
+		goto out_unlock;
+	}
+	new_bm = NULL;
+out_unlock:
 	if (r)
 		cmd->fail_io = true;
 	WRITE_UNLOCK(cmd);
+	dm_block_manager_destroy(old_bm);
+out:
+	if (new_bm && !IS_ERR(new_bm))
+		dm_block_manager_destroy(new_bm);
 
 	return r;
 }

drivers/md/dm-cache-target.c

@@ -907,16 +907,16 @@ static void abort_transaction(struct cache *cache)
 	if (get_cache_mode(cache) >= CM_READ_ONLY)
 		return;
 
-	if (dm_cache_metadata_set_needs_check(cache->cmd)) {
-		DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
-		set_cache_mode(cache, CM_FAIL);
-	}
-
 	DMERR_LIMIT("%s: aborting current metadata transaction", dev_name);
 	if (dm_cache_metadata_abort(cache->cmd)) {
 		DMERR("%s: failed to abort metadata transaction", dev_name);
 		set_cache_mode(cache, CM_FAIL);
 	}
+
+	if (dm_cache_metadata_set_needs_check(cache->cmd)) {
+		DMERR("%s: failed to set 'needs_check' flag in metadata", dev_name);
+		set_cache_mode(cache, CM_FAIL);
+	}
 }
 
 static void metadata_operation_failed(struct cache *cache, const char *op, int r)
@@ -1887,6 +1887,7 @@ static void destroy(struct cache *cache)
 	if (cache->prison)
 		dm_bio_prison_destroy_v2(cache->prison);
 
+	cancel_delayed_work_sync(&cache->waker);
 	if (cache->wq)
 		destroy_workqueue(cache->wq);
 

drivers/md/dm-clone-target.c

@@ -1958,6 +1958,7 @@ static void clone_dtr(struct dm_target *ti)
 
 	mempool_exit(&clone->hydration_pool);
 	dm_kcopyd_client_destroy(clone->kcopyd_client);
+	cancel_delayed_work_sync(&clone->waker);
 	destroy_workqueue(clone->wq);
 	hash_table_exit(clone);
 	dm_clone_metadata_close(clone->cmd);

drivers/md/dm-integrity.c

@@ -261,6 +261,7 @@ struct dm_integrity_c {
 
 	struct completion crypto_backoff;
 
+	bool wrote_to_journal;
 	bool journal_uptodate;
 	bool just_formatted;
 	bool recalculate_flag;
@@ -2373,6 +2374,8 @@ static void integrity_commit(struct work_struct *w)
 	if (!commit_sections)
 		goto release_flush_bios;
 
+	ic->wrote_to_journal = true;
+
 	i = commit_start;
 	for (n = 0; n < commit_sections; n++) {
 		for (j = 0; j < ic->journal_section_entries; j++) {
@@ -2589,10 +2592,6 @@ static void integrity_writer(struct work_struct *w)
 
 	unsigned prev_free_sectors;
 
-	/* the following test is not needed, but it tests the replay code */
-	if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
-		return;
-
 	spin_lock_irq(&ic->endio_wait.lock);
 	write_start = ic->committed_section;
 	write_sections = ic->n_committed_sections;
@@ -3098,10 +3097,17 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 	drain_workqueue(ic->commit_wq);
 
 	if (ic->mode == 'J') {
-		if (ic->meta_dev)
-			queue_work(ic->writer_wq, &ic->writer_work);
+		queue_work(ic->writer_wq, &ic->writer_work);
 		drain_workqueue(ic->writer_wq);
 		dm_integrity_flush_buffers(ic, true);
+		if (ic->wrote_to_journal) {
+			init_journal(ic, ic->free_section,
+				     ic->journal_sections - ic->free_section, ic->commit_seq);
+			if (ic->free_section) {
+				init_journal(ic, 0, ic->free_section,
+					     next_commit_seq(ic->commit_seq));
+			}
+		}
 	}
 
 	if (ic->mode == 'B') {
@@ -3129,6 +3135,8 @@ static void dm_integrity_resume(struct dm_target *ti)
 
 	DEBUG_print("resume\n");
 
+	ic->wrote_to_journal = false;
+
 	if (ic->provided_data_sectors != old_provided_data_sectors) {
 		if (ic->provided_data_sectors > old_provided_data_sectors &&
 		    ic->mode == 'B' &&
@@ -4547,6 +4555,8 @@ static void dm_integrity_dtr(struct dm_target *ti)
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 	BUG_ON(!list_empty(&ic->wait_list));
 
+	if (ic->mode == 'B')
+		cancel_delayed_work_sync(&ic->bitmap_flush_work);
 	if (ic->metadata_wq)
 		destroy_workqueue(ic->metadata_wq);
 	if (ic->wait_wq)

drivers/md/dm-ioctl.c

@@ -655,7 +655,7 @@ static void list_version_get_needed(struct target_type *tt, void *needed_param)
     size_t *needed = needed_param;
 
     *needed += sizeof(struct dm_target_versions);
-    *needed += strlen(tt->name);
+    *needed += strlen(tt->name) + 1;
     *needed += ALIGN_MASK;
 }
 
@@ -720,7 +720,7 @@ static int __list_versions(struct dm_ioctl *param, size_t param_size, const char
 	iter_info.old_vers = NULL;
 	iter_info.vers = vers;
 	iter_info.flags = 0;
-	iter_info.end = (char *)vers+len;
+	iter_info.end = (char *)vers + needed;
 
 	/*
 	 * Now loop through filling out the names & versions.

drivers/md/dm-thin-metadata.c

@@ -724,6 +724,15 @@ static int __open_metadata(struct dm_pool_metadata *pmd)
 		goto bad_cleanup_data_sm;
 	}
 
+	/*
+	 * For pool metadata opening process, root setting is redundant
+	 * because it will be set again in __begin_transaction(). But dm
+	 * pool aborting process really needs to get last transaction's
+	 * root to avoid accessing broken btree.
+	 */
+	pmd->root = le64_to_cpu(disk_super->data_mapping_root);
+	pmd->details_root = le64_to_cpu(disk_super->device_details_root);
+
 	__setup_btree_details(pmd);
 	dm_bm_unlock(sblock);
 
@@ -776,13 +785,15 @@ static int __create_persistent_data_objects(struct dm_pool_metadata *pmd, bool f
 	return r;
 }
 
-static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd)
+static void __destroy_persistent_data_objects(struct dm_pool_metadata *pmd,
+					      bool destroy_bm)
 {
 	dm_sm_destroy(pmd->data_sm);
 	dm_sm_destroy(pmd->metadata_sm);
 	dm_tm_destroy(pmd->nb_tm);
 	dm_tm_destroy(pmd->tm);
-	dm_block_manager_destroy(pmd->bm);
+	if (destroy_bm)
+		dm_block_manager_destroy(pmd->bm);
 }
 
 static int __begin_transaction(struct dm_pool_metadata *pmd)
@@ -989,7 +1000,7 @@ int dm_pool_metadata_close(struct dm_pool_metadata *pmd)
 	}
 	pmd_write_unlock(pmd);
 	if (!pmd->fail_io)
-		__destroy_persistent_data_objects(pmd);
+		__destroy_persistent_data_objects(pmd, true);
 
 	kfree(pmd);
 	return 0;
@@ -1860,19 +1871,52 @@ static void __set_abort_with_changes_flags(struct dm_pool_metadata *pmd)
 int dm_pool_abort_metadata(struct dm_pool_metadata *pmd)
 {
 	int r = -EINVAL;
+	struct dm_block_manager *old_bm = NULL, *new_bm = NULL;
+
+	/* fail_io is double-checked with pmd->root_lock held below */
+	if (unlikely(pmd->fail_io))
+		return r;
+
+	/*
+	 * Replacement block manager (new_bm) is created and old_bm destroyed outside of
+	 * pmd root_lock to avoid ABBA deadlock that would result (due to life-cycle of
+	 * shrinker associated with the block manager's bufio client vs pmd root_lock).
+	 * - must take shrinker_rwsem without holding pmd->root_lock
+	 */
+	new_bm = dm_block_manager_create(pmd->bdev, THIN_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
+					 THIN_MAX_CONCURRENT_LOCKS);
 
 	pmd_write_lock(pmd);
-	if (pmd->fail_io)
+	if (pmd->fail_io) {
+		pmd_write_unlock(pmd);
 		goto out;
+	}
 
 	__set_abort_with_changes_flags(pmd);
-	__destroy_persistent_data_objects(pmd);
-	r = __create_persistent_data_objects(pmd, false);
+	__destroy_persistent_data_objects(pmd, false);
+	old_bm = pmd->bm;
+	if (IS_ERR(new_bm)) {
+		DMERR("could not create block manager during abort");
+		pmd->bm = NULL;
+		r = PTR_ERR(new_bm);
+		goto out_unlock;
+	}
+
+	pmd->bm = new_bm;
+	r = __open_or_format_metadata(pmd, false);
+	if (r) {
+		pmd->bm = NULL;
+		goto out_unlock;
+	}
+	new_bm = NULL;
+out_unlock:
 	if (r)
 		pmd->fail_io = true;
-
-out:
 	pmd_write_unlock(pmd);
+	dm_block_manager_destroy(old_bm);
+out:
+	if (new_bm && !IS_ERR(new_bm))
+		dm_block_manager_destroy(new_bm);
 
 	return r;
 }

drivers/md/dm-thin.c

@@ -2889,6 +2889,8 @@ static void __pool_destroy(struct pool *pool)
 	dm_bio_prison_destroy(pool->prison);
 	dm_kcopyd_client_destroy(pool->copier);
 
+	cancel_delayed_work_sync(&pool->waker);
+	cancel_delayed_work_sync(&pool->no_space_timeout);
 	if (pool->wq)
 		destroy_workqueue(pool->wq);
 
@@ -3540,20 +3542,28 @@ static int pool_preresume(struct dm_target *ti)
 	 */
 	r = bind_control_target(pool, ti);
 	if (r)
-		return r;
+		goto out;
 
 	r = maybe_resize_data_dev(ti, &need_commit1);
 	if (r)
-		return r;
+		goto out;
 
 	r = maybe_resize_metadata_dev(ti, &need_commit2);
 	if (r)
-		return r;
+		goto out;
 
 	if (need_commit1 || need_commit2)
 		(void) commit(pool);
+out:
+	/*
+	 * When a thin-pool is PM_FAIL, it cannot be rebuilt if
+	 * bio is in deferred list. Therefore need to return 0
+	 * to allow pool_resume() to flush IO.
+	 */
+	if (r && get_pool_mode(pool) == PM_FAIL)
+		r = 0;
 
-	return 0;
+	return r;
 }
 
 static void pool_suspend_active_thins(struct pool *pool)