Merge tag 'for-6.2/dm-changes' of...

Merge tag 'for-6.2/dm-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper updates from Mike Snitzer:

 - Fix use-after-free races due to missing resource cleanup during DM
   target destruction in DM targets: thin-pool, cache, integrity and
   clone.

 - Fix ABBA deadlocks in DM thin-pool and cache targets due to their use
   of a bufio client (that has a shrinker whose locking can cause the
   incorrect locking order).

 - Fix DM cache target to set its needs_check flag after first aborting
   the metadata (whereby using reset persistent-data objects to update
   the superblock with, otherwise the superblock update could be dropped
   due to aborting metadata). This was found with code-inspection when
   comparing with the equivalent in DM thinp code.

 - Fix DM thin-pool's presume to continue resuming the device even if
   the pool in is fail mode -- otherwise bios may never be failed up the
   IO stack (which will prevent resetting the thin-pool target via table
   reload)

 - Fix DM thin-pool's metadata to use proper btree root (from previous
   transaction) if metadata commit failed.

 - Add 'waitfor' module param to DM module (dm_mod) to allow dm-init to
   wait for the specified device before continuing with its DM target
   initialization.

* tag 'for-6.2/dm-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm thin: Use last transaction's pmd->root when commit failed
  dm init: add dm-mod.waitfor to wait for asynchronously probed block devices
  dm ioctl: fix a couple ioctl codes
  dm ioctl: a small code cleanup in list_version_get_info
  dm thin: resume even if in FAIL mode
  dm cache: set needs_check flag after aborting metadata
  dm cache: Fix ABBA deadlock between shrink_slab and dm_cache_metadata_abort
  dm thin: Fix ABBA deadlock between shrink_slab and dm_pool_abort_metadata
  dm integrity: Fix UAF in dm_integrity_dtr()
  dm cache: Fix UAF in destroy()
  dm clone: Fix UAF in clone_dtr()
  dm thin: Fix UAF in run_timer_softirq()

Documentation/admin-guide/device-mapper/dm-init.rst

@@ -123,3 +123,11 @@ Other examples (per target):
     0 1638400 verity 1 8:1 8:2 4096 4096 204800 1 sha256
     fb1a5a0f00deb908d8b53cb270858975e76cf64105d412ce764225d53b8f3cfd
     51934789604d1b92399c52e7cb149d1b3a1b74bbbcb103b2a0aaacbed5c08584
+
+For setups using device-mapper on top of asynchronously probed block
+devices (MMC, USB, ..), it may be necessary to tell dm-init to
+explicitly wait for them to become available before setting up the
+device-mapper tables. This can be done with the "dm-mod.waitfor="
+module parameter, which takes a list of devices to wait for::
+
+  dm-mod.waitfor=<device1>[,..,<deviceN>]

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-init.c

@@ -8,6 +8,7 @@
  */
 
 #include <linux/ctype.h>
+#include <linux/delay.h>
 #include <linux/device.h>
 #include <linux/device-mapper.h>
 #include <linux/init.h>
@@ -18,12 +19,17 @@
 #define DM_MAX_DEVICES 256
 #define DM_MAX_TARGETS 256
 #define DM_MAX_STR_SIZE 4096
+#define DM_MAX_WAITFOR 256
 
 static char *create;
 
+static char *waitfor[DM_MAX_WAITFOR];
+
 /*
  * Format: dm-mod.create=<name>,<uuid>,<minor>,<flags>,<table>[,<table>+][;<name>,<uuid>,<minor>,<flags>,<table>[,<table>+]+]
  * Table format: <start_sector> <num_sectors> <target_type> <target_args>
+ * Block devices to wait for to become available before setting up tables:
+ * dm-mod.waitfor=<device1>[,..,<deviceN>]
  *
  * See Documentation/admin-guide/device-mapper/dm-init.rst for dm-mod.create="..." format
  * details.
@@ -266,7 +272,7 @@ static int __init dm_init_init(void)
 	struct dm_device *dev;
 	LIST_HEAD(devices);
 	char *str;
-	int r;
+	int i, r;
 
 	if (!create)
 		return 0;
@@ -286,6 +292,17 @@ static int __init dm_init_init(void)
 	DMINFO("waiting for all devices to be available before creating mapped devices");
 	wait_for_device_probe();
 
+	for (i = 0; i < ARRAY_SIZE(waitfor); i++) {
+		if (waitfor[i]) {
+			DMINFO("waiting for device %s ...", waitfor[i]);
+			while (!dm_get_dev_t(waitfor[i]))
+				msleep(5);
+		}
+	}
+
+	if (waitfor[0])
+		DMINFO("all devices available");
+
 	list_for_each_entry(dev, &devices, list) {
 		if (dm_early_create(&dev->dmi, dev->table,
 				    dev->target_args_array))
@@ -301,3 +318,6 @@ late_initcall(dm_init_init);
 
 module_param(create, charp, 0);
 MODULE_PARM_DESC(create, "Create a mapped device in early boot");
+
+module_param_array(waitfor, charp, NULL, 0);
+MODULE_PARM_DESC(waitfor, "Devices to wait for before setting up tables");

drivers/md/dm-integrity.c

@@ -4558,6 +4558,8 @@ 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

@@ -681,7 +681,7 @@ static void list_version_get_info(struct target_type *tt, void *param)
     strcpy(info->vers->name, tt->name);
 
     info->old_vers = info->vers;
-    info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1);
+    info->vers = align_ptr((void *)(info->vers + 1) + strlen(tt->name) + 1);
 }
 
 static int __list_versions(struct dm_ioctl *param, size_t param_size, const char *name)
@@ -1788,8 +1788,8 @@ static ioctl_fn lookup_ioctl(unsigned int cmd, int *ioctl_flags)
 
 		{DM_TARGET_MSG_CMD, 0, target_message},
 		{DM_DEV_SET_GEOMETRY_CMD, 0, dev_set_geometry},
-		{DM_DEV_ARM_POLL, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
-		{DM_GET_TARGET_VERSION, 0, get_target_version},
+		{DM_DEV_ARM_POLL_CMD, IOCTL_FLAGS_NO_PARAMS, dev_arm_poll},
+		{DM_GET_TARGET_VERSION_CMD, 0, get_target_version},
 	};
 
 	if (unlikely(cmd >= ARRAY_SIZE(_ioctls)))

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)