T3Q was born on a highway. That's where most accidents happen.

drivers/block/amiflop.c

@@ -61,6 +61,7 @@
 #include <linux/hdreg.h>
 #include <linux/delay.h>
 #include <linux/init.h>
+#include <linux/major.h>
 #include <linux/mutex.h>
 #include <linux/fs.h>
 #include <linux/blk-mq.h>

drivers/block/ataflop.c

@@ -68,6 +68,7 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/blk-mq.h>
+#include <linux/major.h>
 #include <linux/mutex.h>
 #include <linux/completion.h>
 #include <linux/wait.h>

drivers/block/floppy.c

@@ -184,6 +184,7 @@ static int print_unex = 1;
 #include <linux/ioport.h>
 #include <linux/interrupt.h>
 #include <linux/init.h>
+#include <linux/major.h>
 #include <linux/platform_device.h>
 #include <linux/mod_devicetable.h>
 #include <linux/mutex.h>

drivers/block/swim.c

@@ -16,6 +16,7 @@
 #include <linux/fd.h>
 #include <linux/slab.h>
 #include <linux/blk-mq.h>
+#include <linux/major.h>
 #include <linux/mutex.h>
 #include <linux/hdreg.h>
 #include <linux/kernel.h>

drivers/block/virtio_blk.c

@@ -859,9 +859,15 @@ static int virtblk_probe(struct virtio_device *vdev)
 
 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 			     &v);
+
+		/*
+		 * max_discard_seg == 0 is out of spec but we always
+		 * handled it.
+		 */
+		if (!v)
+			v = sg_elems - 2;
 		blk_queue_max_discard_segments(q,
-					       min_not_zero(v,
+					       min(v, MAX_DISCARD_SEGMENTS));
-							    MAX_DISCARD_SEGMENTS));
 
 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 	}

drivers/block/xen-blkfront.c

@@ -42,6 +42,7 @@
 #include <linux/cdrom.h>
 #include <linux/module.h>
 #include <linux/slab.h>
+#include <linux/major.h>
 #include <linux/mutex.h>
 #include <linux/scatterlist.h>
 #include <linux/bitmap.h>

drivers/pinctrl/intel/pinctrl-tigerlake.c

@@ -749,7 +749,6 @@ static const struct acpi_device_id tgl_pinctrl_acpi_match[] = {
 	{ "INT34C5", (kernel_ulong_t)&tgllp_soc_data },
 	{ "INT34C6", (kernel_ulong_t)&tglh_soc_data },
 	{ "INTC1055", (kernel_ulong_t)&tgllp_soc_data },
-	{ "INTC1057", (kernel_ulong_t)&tgllp_soc_data },
 	{ }
 };
 MODULE_DEVICE_TABLE(acpi, tgl_pinctrl_acpi_match);

drivers/s390/block/dasd_genhd.c

@@ -14,6 +14,7 @@
 #define KMSG_COMPONENT "dasd"
 
 #include <linux/interrupt.h>
+#include <linux/major.h>
 #include <linux/fs.h>
 #include <linux/blkpg.h>
 

drivers/spi/spi-rockchip.c

@@ -585,6 +585,12 @@ static int rockchip_spi_slave_abort(struct spi_controller *ctlr)
 {
 	struct rockchip_spi *rs = spi_controller_get_devdata(ctlr);
 
+	if (atomic_read(&rs->state) & RXDMA)
+		dmaengine_terminate_sync(ctlr->dma_rx);
+	if (atomic_read(&rs->state) & TXDMA)
+		dmaengine_terminate_sync(ctlr->dma_tx);
+	atomic_set(&rs->state, 0);
+	spi_enable_chip(rs, false);
 	rs->slave_abort = true;
 	spi_finalize_current_transfer(ctlr);
 
@@ -654,7 +660,7 @@ static int rockchip_spi_probe(struct platform_device *pdev)
 	struct spi_controller *ctlr;
 	struct resource *mem;
 	struct device_node *np = pdev->dev.of_node;
-	u32 rsd_nsecs;
+	u32 rsd_nsecs, num_cs;
 	bool slave_mode;
 
 	slave_mode = of_property_read_bool(np, "spi-slave");
@@ -764,8 +770,9 @@ static int rockchip_spi_probe(struct platform_device *pdev)
 		 * rk spi0 has two native cs, spi1..5 one cs only
 		 * if num-cs is missing in the dts, default to 1
 		 */
-		if (of_property_read_u16(np, "num-cs", &ctlr->num_chipselect))
+		if (of_property_read_u32(np, "num-cs", &num_cs))
-			ctlr->num_chipselect = 1;
+			num_cs = 1;
+		ctlr->num_chipselect = num_cs;
 		ctlr->use_gpio_descriptors = true;
 	}
 	ctlr->dev.of_node = pdev->dev.of_node;

drivers/virtio/virtio.c

@@ -166,14 +166,13 @@ void virtio_add_status(struct virtio_device *dev, unsigned int status)
 }
 EXPORT_SYMBOL_GPL(virtio_add_status);
 
-int virtio_finalize_features(struct virtio_device *dev)
+/* Do some validation, then set FEATURES_OK */
+static int virtio_features_ok(struct virtio_device *dev)
 {
-	int ret = dev->config->finalize_features(dev);
 	unsigned status;
+	int ret;
 
 	might_sleep();
-	if (ret)
-		return ret;
 
 	ret = arch_has_restricted_virtio_memory_access();
 	if (ret) {
@@ -202,7 +201,6 @@ int virtio_finalize_features(struct virtio_device *dev)
 	}
 	return 0;
 }
-EXPORT_SYMBOL_GPL(virtio_finalize_features);
 
 static int virtio_dev_probe(struct device *_d)
 {
@@ -239,17 +237,6 @@ static int virtio_dev_probe(struct device *_d)
 		driver_features_legacy = driver_features;
 	}
 
-	/*
-	 * Some devices detect legacy solely via F_VERSION_1. Write
-	 * F_VERSION_1 to force LE config space accesses before FEATURES_OK for
-	 * these when needed.
-	 */
-	if (drv->validate && !virtio_legacy_is_little_endian()
-			  && device_features & BIT_ULL(VIRTIO_F_VERSION_1)) {
-		dev->features = BIT_ULL(VIRTIO_F_VERSION_1);
-		dev->config->finalize_features(dev);
-	}
-
 	if (device_features & (1ULL << VIRTIO_F_VERSION_1))
 		dev->features = driver_features & device_features;
 	else
@@ -260,13 +247,26 @@ static int virtio_dev_probe(struct device *_d)
 		if (device_features & (1ULL << i))
 			__virtio_set_bit(dev, i);
 
+	err = dev->config->finalize_features(dev);
+	if (err)
+		goto err;
+
 	if (drv->validate) {
+		u64 features = dev->features;
+
 		err = drv->validate(dev);
 		if (err)
 			goto err;
+
+		/* Did validation change any features? Then write them again. */
+		if (features != dev->features) {
+			err = dev->config->finalize_features(dev);
+			if (err)
+				goto err;
+		}
 	}
 
-	err = virtio_finalize_features(dev);
+	err = virtio_features_ok(dev);
 	if (err)
 		goto err;
 
@@ -490,7 +490,11 @@ int virtio_device_restore(struct virtio_device *dev)
 	/* We have a driver! */
 	virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER);
 
-	ret = virtio_finalize_features(dev);
+	ret = dev->config->finalize_features(dev);
+	if (ret)
+		goto err;
+
+	ret = virtio_features_ok(dev);
 	if (ret)
 		goto err;
 

fs/btrfs/block-group.c

@@ -1491,7 +1491,6 @@ void btrfs_reclaim_bgs_work(struct work_struct *work)
 		container_of(work, struct btrfs_fs_info, reclaim_bgs_work);
 	struct btrfs_block_group *bg;
 	struct btrfs_space_info *space_info;
-	LIST_HEAD(again_list);
 
 	if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
 		return;
@@ -1562,18 +1561,14 @@ void btrfs_reclaim_bgs_work(struct work_struct *work)
 				div64_u64(zone_unusable * 100, bg->length));
 		trace_btrfs_reclaim_block_group(bg);
 		ret = btrfs_relocate_chunk(fs_info, bg->start);
-		if (ret && ret != -EAGAIN)
+		if (ret)
 			btrfs_err(fs_info, "error relocating chunk %llu",
 				  bg->start);
 
 next:
+		btrfs_put_block_group(bg);
 		spin_lock(&fs_info->unused_bgs_lock);
-		if (ret == -EAGAIN && list_empty(&bg->bg_list))
-			list_add_tail(&bg->bg_list, &again_list);
-		else
-			btrfs_put_block_group(bg);
 	}
-	list_splice_tail(&again_list, &fs_info->reclaim_bgs);
 	spin_unlock(&fs_info->unused_bgs_lock);
 	mutex_unlock(&fs_info->reclaim_bgs_lock);
 	btrfs_exclop_finish(fs_info);

fs/btrfs/ctree.c

@@ -1566,32 +1566,13 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
 							struct btrfs_path *p,
 							int write_lock_level)
 {
-	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct extent_buffer *b;
 	int root_lock = 0;
 	int level = 0;
 
 	if (p->search_commit_root) {
-		/*
+		b = root->commit_root;
-		 * The commit roots are read only so we always do read locks,
+		atomic_inc(&b->refs);
-		 * and we always must hold the commit_root_sem when doing
-		 * searches on them, the only exception is send where we don't
-		 * want to block transaction commits for a long time, so
-		 * we need to clone the commit root in order to avoid races
-		 * with transaction commits that create a snapshot of one of
-		 * the roots used by a send operation.
-		 */
-		if (p->need_commit_sem) {
-			down_read(&fs_info->commit_root_sem);
-			b = btrfs_clone_extent_buffer(root->commit_root);
-			up_read(&fs_info->commit_root_sem);
-			if (!b)
-				return ERR_PTR(-ENOMEM);
-
-		} else {
-			b = root->commit_root;
-			atomic_inc(&b->refs);
-		}
 		level = btrfs_header_level(b);
 		/*
 		 * Ensure that all callers have set skip_locking when
@@ -1657,6 +1638,42 @@ static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
 	return b;
 }
 
+/*
+ * Replace the extent buffer at the lowest level of the path with a cloned
+ * version. The purpose is to be able to use it safely, after releasing the
+ * commit root semaphore, even if relocation is happening in parallel, the
+ * transaction used for relocation is committed and the extent buffer is
+ * reallocated in the next transaction.
+ *
+ * This is used in a context where the caller does not prevent transaction
+ * commits from happening, either by holding a transaction handle or holding
+ * some lock, while it's doing searches through a commit root.
+ * At the moment it's only used for send operations.
+ */
+static int finish_need_commit_sem_search(struct btrfs_path *path)
+{
+	const int i = path->lowest_level;
+	const int slot = path->slots[i];
+	struct extent_buffer *lowest = path->nodes[i];
+	struct extent_buffer *clone;
+
+	ASSERT(path->need_commit_sem);
+
+	if (!lowest)
+		return 0;
+
+	lockdep_assert_held_read(&lowest->fs_info->commit_root_sem);
+
+	clone = btrfs_clone_extent_buffer(lowest);
+	if (!clone)
+		return -ENOMEM;
+
+	btrfs_release_path(path);
+	path->nodes[i] = clone;
+	path->slots[i] = slot;
+
+	return 0;
+}
 
 /*
  * btrfs_search_slot - look for a key in a tree and perform necessary
@@ -1693,6 +1710,7 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		      const struct btrfs_key *key, struct btrfs_path *p,
 		      int ins_len, int cow)
 {
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct extent_buffer *b;
 	int slot;
 	int ret;
@@ -1734,6 +1752,11 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 
 	min_write_lock_level = write_lock_level;
 
+	if (p->need_commit_sem) {
+		ASSERT(p->search_commit_root);
+		down_read(&fs_info->commit_root_sem);
+	}
+
 again:
 	prev_cmp = -1;
 	b = btrfs_search_slot_get_root(root, p, write_lock_level);
@@ -1928,6 +1951,16 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 done:
 	if (ret < 0 && !p->skip_release_on_error)
 		btrfs_release_path(p);
+
+	if (p->need_commit_sem) {
+		int ret2;
+
+		ret2 = finish_need_commit_sem_search(p);
+		up_read(&fs_info->commit_root_sem);
+		if (ret2)
+			ret = ret2;
+	}
+
 	return ret;
 }
 ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO);
@@ -4396,7 +4429,9 @@ int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
 	int level;
 	struct extent_buffer *c;
 	struct extent_buffer *next;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_key key;
+	bool need_commit_sem = false;
 	u32 nritems;
 	int ret;
 	int i;
@@ -4413,14 +4448,20 @@ int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
 
 	path->keep_locks = 1;
 
-	if (time_seq)
+	if (time_seq) {
 		ret = btrfs_search_old_slot(root, &key, path, time_seq);
-	else
+	} else {
+		if (path->need_commit_sem) {
+			path->need_commit_sem = 0;
+			need_commit_sem = true;
+			down_read(&fs_info->commit_root_sem);
+		}
 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	}
 	path->keep_locks = 0;
 
 	if (ret < 0)
-		return ret;
+		goto done;
 
 	nritems = btrfs_header_nritems(path->nodes[0]);
 	/*
@@ -4543,6 +4584,15 @@ int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
 	ret = 0;
 done:
 	unlock_up(path, 0, 1, 0, NULL);
+	if (need_commit_sem) {
+		int ret2;
+
+		path->need_commit_sem = 1;
+		ret2 = finish_need_commit_sem_search(path);
+		up_read(&fs_info->commit_root_sem);
+		if (ret2)
+			ret = ret2;
+	}
 
 	return ret;
 }

fs/btrfs/ctree.h

@@ -568,7 +568,6 @@ enum {
 	/*
 	 * Indicate that relocation of a chunk has started, it's set per chunk
 	 * and is toggled between chunks.
-	 * Set, tested and cleared while holding fs_info::send_reloc_lock.
 	 */
 	BTRFS_FS_RELOC_RUNNING,
 
@@ -668,6 +667,12 @@ struct btrfs_fs_info {
 
 	u64 generation;
 	u64 last_trans_committed;
+	/*
+	 * Generation of the last transaction used for block group relocation
+	 * since the filesystem was last mounted (or 0 if none happened yet).
+	 * Must be written and read while holding btrfs_fs_info::commit_root_sem.
+	 */
+	u64 last_reloc_trans;
 	u64 avg_delayed_ref_runtime;
 
 	/*
@@ -997,13 +1002,6 @@ struct btrfs_fs_info {
 
 	struct crypto_shash *csum_shash;
 
-	spinlock_t send_reloc_lock;
-	/*
-	 * Number of send operations in progress.
-	 * Updated while holding fs_info::send_reloc_lock.
-	 */
-	int send_in_progress;
-
 	/* Type of exclusive operation running, protected by super_lock */
 	enum btrfs_exclusive_operation exclusive_operation;
 

fs/btrfs/disk-io.c

@@ -2859,6 +2859,7 @@ static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
 		/* All successful */
 		fs_info->generation = generation;
 		fs_info->last_trans_committed = generation;
+		fs_info->last_reloc_trans = 0;
 
 		/* Always begin writing backup roots after the one being used */
 		if (backup_index < 0) {
@@ -2992,9 +2993,6 @@ void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
 	spin_lock_init(&fs_info->swapfile_pins_lock);
 	fs_info->swapfile_pins = RB_ROOT;
 
-	spin_lock_init(&fs_info->send_reloc_lock);
-	fs_info->send_in_progress = 0;
-
 	fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
 	INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
 }

fs/btrfs/relocation.c

@@ -3854,25 +3854,14 @@ struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
  *   0             success
  *   -EINPROGRESS  operation is already in progress, that's probably a bug
  *   -ECANCELED    cancellation request was set before the operation started
- *   -EAGAIN       can not start because there are ongoing send operations
  */
 static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
 {
-	spin_lock(&fs_info->send_reloc_lock);
-	if (fs_info->send_in_progress) {
-		btrfs_warn_rl(fs_info,
-"cannot run relocation while send operations are in progress (%d in progress)",
-			      fs_info->send_in_progress);
-		spin_unlock(&fs_info->send_reloc_lock);
-		return -EAGAIN;
-	}
 	if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
 		/* This should not happen */
-		spin_unlock(&fs_info->send_reloc_lock);
 		btrfs_err(fs_info, "reloc already running, cannot start");
 		return -EINPROGRESS;
 	}
-	spin_unlock(&fs_info->send_reloc_lock);
 
 	if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
 		btrfs_info(fs_info, "chunk relocation canceled on start");
@@ -3894,9 +3883,7 @@ static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
 	/* Requested after start, clear bit first so any waiters can continue */
 	if (atomic_read(&fs_info->reloc_cancel_req) > 0)
 		btrfs_info(fs_info, "chunk relocation canceled during operation");
-	spin_lock(&fs_info->send_reloc_lock);
 	clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
-	spin_unlock(&fs_info->send_reloc_lock);
 	atomic_set(&fs_info->reloc_cancel_req, 0);
 }
 

fs/btrfs/send.c

@@ -24,6 +24,7 @@
 #include "transaction.h"
 #include "compression.h"
 #include "xattr.h"
+#include "print-tree.h"
 
 /*
  * Maximum number of references an extent can have in order for us to attempt to
@@ -95,6 +96,15 @@ struct send_ctx {
 	struct btrfs_path *right_path;
 	struct btrfs_key *cmp_key;
 
+	/*
+	 * Keep track of the generation of the last transaction that was used
+	 * for relocating a block group. This is periodically checked in order
+	 * to detect if a relocation happened since the last check, so that we
+	 * don't operate on stale extent buffers for nodes (level >= 1) or on
+	 * stale disk_bytenr values of file extent items.
+	 */
+	u64 last_reloc_trans;
+
 	/*
 	 * infos of the currently processed inode. In case of deleted inodes,
 	 * these are the values from the deleted inode.
@@ -1415,6 +1425,26 @@ static int find_extent_clone(struct send_ctx *sctx,
 	if (ret < 0)
 		goto out;
 
+	down_read(&fs_info->commit_root_sem);
+	if (fs_info->last_reloc_trans > sctx->last_reloc_trans) {
+		/*
+		 * A transaction commit for a transaction in which block group
+		 * relocation was done just happened.
+		 * The disk_bytenr of the file extent item we processed is
+		 * possibly stale, referring to the extent's location before
+		 * relocation. So act as if we haven't found any clone sources
+		 * and fallback to write commands, which will read the correct
+		 * data from the new extent location. Otherwise we will fail
+		 * below because we haven't found our own back reference or we
+		 * could be getting incorrect sources in case the old extent
+		 * was already reallocated after the relocation.
+		 */
+		up_read(&fs_info->commit_root_sem);
+		ret = -ENOENT;
+		goto out;
+	}
+	up_read(&fs_info->commit_root_sem);
+
 	if (!backref_ctx.found_itself) {
 		/* found a bug in backref code? */
 		ret = -EIO;
@@ -6596,6 +6626,50 @@ static int changed_cb(struct btrfs_path *left_path,
 {
 	int ret = 0;
 
+	/*
+	 * We can not hold the commit root semaphore here. This is because in
+	 * the case of sending and receiving to the same filesystem, using a
+	 * pipe, could result in a deadlock:
+	 *
+	 * 1) The task running send blocks on the pipe because it's full;
+	 *
+	 * 2) The task running receive, which is the only consumer of the pipe,
+	 *    is waiting for a transaction commit (for example due to a space
+	 *    reservation when doing a write or triggering a transaction commit
+	 *    when creating a subvolume);
+	 *
+	 * 3) The transaction is waiting to write lock the commit root semaphore,
+	 *    but can not acquire it since it's being held at 1).
+	 *
+	 * Down this call chain we write to the pipe through kernel_write().
+	 * The same type of problem can also happen when sending to a file that
+	 * is stored in the same filesystem - when reserving space for a write
+	 * into the file, we can trigger a transaction commit.
+	 *
+	 * Our caller has supplied us with clones of leaves from the send and
+	 * parent roots, so we're safe here from a concurrent relocation and
+	 * further reallocation of metadata extents while we are here. Below we
+	 * also assert that the leaves are clones.
+	 */
+	lockdep_assert_not_held(&sctx->send_root->fs_info->commit_root_sem);
+
+	/*
+	 * We always have a send root, so left_path is never NULL. We will not
+	 * have a leaf when we have reached the end of the send root but have
+	 * not yet reached the end of the parent root.
+	 */
+	if (left_path->nodes[0])
+		ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED,
+				&left_path->nodes[0]->bflags));
+	/*
+	 * When doing a full send we don't have a parent root, so right_path is
+	 * NULL. When doing an incremental send, we may have reached the end of
+	 * the parent root already, so we don't have a leaf at right_path.
+	 */
+	if (right_path && right_path->nodes[0])
+		ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED,
+				&right_path->nodes[0]->bflags));
+
 	if (result == BTRFS_COMPARE_TREE_SAME) {
 		if (key->type == BTRFS_INODE_REF_KEY ||
 		    key->type == BTRFS_INODE_EXTREF_KEY) {
@@ -6642,14 +6716,46 @@ static int changed_cb(struct btrfs_path *left_path,
 	return ret;
 }
 
+static int search_key_again(const struct send_ctx *sctx,
+			    struct btrfs_root *root,
+			    struct btrfs_path *path,
+			    const struct btrfs_key *key)
+{
+	int ret;
+
+	if (!path->need_commit_sem)
+		lockdep_assert_held_read(&root->fs_info->commit_root_sem);
+
+	/*
+	 * Roots used for send operations are readonly and no one can add,
+	 * update or remove keys from them, so we should be able to find our
+	 * key again. The only exception is deduplication, which can operate on
+	 * readonly roots and add, update or remove keys to/from them - but at
+	 * the moment we don't allow it to run in parallel with send.
+	 */
+	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
+	ASSERT(ret <= 0);
+	if (ret > 0) {
+		btrfs_print_tree(path->nodes[path->lowest_level], false);
+		btrfs_err(root->fs_info,
+"send: key (%llu %u %llu) not found in %s root %llu, lowest_level %d, slot %d",
+			  key->objectid, key->type, key->offset,
+			  (root == sctx->parent_root ? "parent" : "send"),
+			  root->root_key.objectid, path->lowest_level,
+			  path->slots[path->lowest_level]);
+		return -EUCLEAN;
+	}
+
+	return ret;
+}
+
 static int full_send_tree(struct send_ctx *sctx)
 {
 	int ret;
 	struct btrfs_root *send_root = sctx->send_root;
 	struct btrfs_key key;
+	struct btrfs_fs_info *fs_info = send_root->fs_info;
 	struct btrfs_path *path;
-	struct extent_buffer *eb;
-	int slot;
 
 	path = alloc_path_for_send();
 	if (!path)
@@ -6660,6 +6766,10 @@ static int full_send_tree(struct send_ctx *sctx)
 	key.type = BTRFS_INODE_ITEM_KEY;
 	key.offset = 0;
 
+	down_read(&fs_info->commit_root_sem);
+	sctx->last_reloc_trans = fs_info->last_reloc_trans;
+	up_read(&fs_info->commit_root_sem);
+
 	ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
 	if (ret < 0)
 		goto out;
@@ -6667,15 +6777,35 @@ static int full_send_tree(struct send_ctx *sctx)
 		goto out_finish;
 
 	while (1) {
-		eb = path->nodes[0];
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
-		slot = path->slots[0];
-		btrfs_item_key_to_cpu(eb, &key, slot);
 
 		ret = changed_cb(path, NULL, &key,
 				 BTRFS_COMPARE_TREE_NEW, sctx);
 		if (ret < 0)
 			goto out;
 
+		down_read(&fs_info->commit_root_sem);
+		if (fs_info->last_reloc_trans > sctx->last_reloc_trans) {
+			sctx->last_reloc_trans = fs_info->last_reloc_trans;
+			up_read(&fs_info->commit_root_sem);
+			/*
+			 * A transaction used for relocating a block group was
+			 * committed or is about to finish its commit. Release
+			 * our path (leaf) and restart the search, so that we
+			 * avoid operating on any file extent items that are
+			 * stale, with a disk_bytenr that reflects a pre
+			 * relocation value. This way we avoid as much as
+			 * possible to fallback to regular writes when checking
+			 * if we can clone file ranges.
+			 */
+			btrfs_release_path(path);
+			ret = search_key_again(sctx, send_root, path, &key);
+			if (ret < 0)
+				goto out;
+		} else {
+			up_read(&fs_info->commit_root_sem);
+		}
+
 		ret = btrfs_next_item(send_root, path);
 		if (ret < 0)
 			goto out;
@@ -6693,6 +6823,20 @@ static int full_send_tree(struct send_ctx *sctx)
 	return ret;
 }
 
+static int replace_node_with_clone(struct btrfs_path *path, int level)
+{
+	struct extent_buffer *clone;
+
+	clone = btrfs_clone_extent_buffer(path->nodes[level]);
+	if (!clone)
+		return -ENOMEM;
+
+	free_extent_buffer(path->nodes[level]);
+	path->nodes[level] = clone;
+
+	return 0;
+}
+
 static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen)
 {
 	struct extent_buffer *eb;
@@ -6702,6 +6846,8 @@ static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen
 	u64 reada_max;
 	u64 reada_done = 0;
 
+	lockdep_assert_held_read(&parent->fs_info->commit_root_sem);
+
 	BUG_ON(*level == 0);
 	eb = btrfs_read_node_slot(parent, slot);
 	if (IS_ERR(eb))
@@ -6725,6 +6871,10 @@ static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen
 	path->nodes[*level - 1] = eb;
 	path->slots[*level - 1] = 0;
 	(*level)--;
+
+	if (*level == 0)
+		return replace_node_with_clone(path, 0);
+
 	return 0;
 }
 
@@ -6738,8 +6888,10 @@ static int tree_move_next_or_upnext(struct btrfs_path *path,
 	path->slots[*level]++;
 
 	while (path->slots[*level] >= nritems) {
-		if (*level == root_level)
+		if (*level == root_level) {
+			path->slots[*level] = nritems - 1;
 			return -1;
+		}
 
 		/* move upnext */
 		path->slots[*level] = 0;
@@ -6771,14 +6923,20 @@ static int tree_advance(struct btrfs_path *path,
 	} else {
 		ret = tree_move_down(path, level, reada_min_gen);
 	}
-	if (ret >= 0) {
+
-		if (*level == 0)
+	/*
-			btrfs_item_key_to_cpu(path->nodes[*level], key,
+	 * Even if we have reached the end of a tree, ret is -1, update the key
-					path->slots[*level]);
+	 * anyway, so that in case we need to restart due to a block group
-		else
+	 * relocation, we can assert that the last key of the root node still
-			btrfs_node_key_to_cpu(path->nodes[*level], key,
+	 * exists in the tree.
-					path->slots[*level]);
+	 */
-	}
+	if (*level == 0)
+		btrfs_item_key_to_cpu(path->nodes[*level], key,
+				      path->slots[*level]);
+	else
+		btrfs_node_key_to_cpu(path->nodes[*level], key,
+				      path->slots[*level]);
+
 	return ret;
 }
 
@@ -6807,6 +6965,97 @@ static int tree_compare_item(struct btrfs_path *left_path,
 	return 0;
 }
 
+/*
+ * A transaction used for relocating a block group was committed or is about to
+ * finish its commit. Release our paths and restart the search, so that we are
+ * not using stale extent buffers:
+ *
+ * 1) For levels > 0, we are only holding references of extent buffers, without
+ *    any locks on them, which does not prevent them from having been relocated
+ *    and reallocated after the last time we released the commit root semaphore.
+ *    The exception are the root nodes, for which we always have a clone, see
+ *    the comment at btrfs_compare_trees();
+ *
+ * 2) For leaves, level 0, we are holding copies (clones) of extent buffers, so
+ *    we are safe from the concurrent relocation and reallocation. However they
+ *    can have file extent items with a pre relocation disk_bytenr value, so we
+ *    restart the start from the current commit roots and clone the new leaves so
+ *    that we get the post relocation disk_bytenr values. Not doing so, could
+ *    make us clone the wrong data in case there are new extents using the old
+ *    disk_bytenr that happen to be shared.
+ */
+static int restart_after_relocation(struct btrfs_path *left_path,
+				    struct btrfs_path *right_path,
+				    const struct btrfs_key *left_key,
+				    const struct btrfs_key *right_key,
+				    int left_level,
+				    int right_level,
+				    const struct send_ctx *sctx)
+{
+	int root_level;
+	int ret;
+
+	lockdep_assert_held_read(&sctx->send_root->fs_info->commit_root_sem);
+
+	btrfs_release_path(left_path);
+	btrfs_release_path(right_path);
+
+	/*
+	 * Since keys can not be added or removed to/from our roots because they
+	 * are readonly and we do not allow deduplication to run in parallel
+	 * (which can add, remove or change keys), the layout of the trees should
+	 * not change.
+	 */
+	left_path->lowest_level = left_level;
+	ret = search_key_again(sctx, sctx->send_root, left_path, left_key);
+	if (ret < 0)
+		return ret;
+
+	right_path->lowest_level = right_level;
+	ret = search_key_again(sctx, sctx->parent_root, right_path, right_key);
+	if (ret < 0)
+		return ret;
+
+	/*
+	 * If the lowest level nodes are leaves, clone them so that they can be
+	 * safely used by changed_cb() while not under the protection of the
+	 * commit root semaphore, even if relocation and reallocation happens in
+	 * parallel.
+	 */
+	if (left_level == 0) {
+		ret = replace_node_with_clone(left_path, 0);
+		if (ret < 0)
+			return ret;
+	}
+
+	if (right_level == 0) {
+		ret = replace_node_with_clone(right_path, 0);
+		if (ret < 0)
+			return ret;
+	}
+
+	/*
+	 * Now clone the root nodes (unless they happen to be the leaves we have
+	 * already cloned). This is to protect against concurrent snapshotting of
+	 * the send and parent roots (see the comment at btrfs_compare_trees()).
+	 */
+	root_level = btrfs_header_level(sctx->send_root->commit_root);
+	if (root_level > 0) {
+		ret = replace_node_with_clone(left_path, root_level);
+		if (ret < 0)
+			return ret;
+	}
+
+	root_level = btrfs_header_level(sctx->parent_root->commit_root);
+	if (root_level > 0) {
+		ret = replace_node_with_clone(right_path, root_level);
+		if (ret < 0)
+			return ret;
+	}
+
+	return 0;
+}
+
 /*
  * This function compares two trees and calls the provided callback for
  * every changed/new/deleted item it finds.
@@ -6835,10 +7084,10 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 	int right_root_level;
 	int left_level;
 	int right_level;
-	int left_end_reached;
+	int left_end_reached = 0;
-	int right_end_reached;
+	int right_end_reached = 0;
-	int advance_left;
+	int advance_left = 0;
-	int advance_right;
+	int advance_right = 0;
 	u64 left_blockptr;
 	u64 right_blockptr;
 	u64 left_gen;
@@ -6906,12 +7155,18 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 	down_read(&fs_info->commit_root_sem);
 	left_level = btrfs_header_level(left_root->commit_root);
 	left_root_level = left_level;
+	/*
+	 * We clone the root node of the send and parent roots to prevent races
+	 * with snapshot creation of these roots. Snapshot creation COWs the
+	 * root node of a tree, so after the transaction is committed the old
+	 * extent can be reallocated while this send operation is still ongoing.
+	 * So we clone them, under the commit root semaphore, to be race free.
+	 */
 	left_path->nodes[left_level] =
 			btrfs_clone_extent_buffer(left_root->commit_root);
 	if (!left_path->nodes[left_level]) {
-		up_read(&fs_info->commit_root_sem);
 		ret = -ENOMEM;
-		goto out;
+		goto out_unlock;
 	}
 
 	right_level = btrfs_header_level(right_root->commit_root);
@@ -6919,9 +7174,8 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 	right_path->nodes[right_level] =
 			btrfs_clone_extent_buffer(right_root->commit_root);
 	if (!right_path->nodes[right_level]) {
-		up_read(&fs_info->commit_root_sem);
 		ret = -ENOMEM;
-		goto out;
+		goto out_unlock;
 	}
 	/*
 	 * Our right root is the parent root, while the left root is the "send"
@@ -6931,7 +7185,6 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 	 * will need to read them at some point.
 	 */
 	reada_min_gen = btrfs_header_generation(right_root->commit_root);
-	up_read(&fs_info->commit_root_sem);
 
 	if (left_level == 0)
 		btrfs_item_key_to_cpu(left_path->nodes[left_level],
@@ -6946,11 +7199,26 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 		btrfs_node_key_to_cpu(right_path->nodes[right_level],
 				&right_key, right_path->slots[right_level]);
 
-	left_end_reached = right_end_reached = 0;
+	sctx->last_reloc_trans = fs_info->last_reloc_trans;
-	advance_left = advance_right = 0;
 
 	while (1) {
-		cond_resched();
+		if (need_resched() ||
+		    rwsem_is_contended(&fs_info->commit_root_sem)) {
+			up_read(&fs_info->commit_root_sem);
+			cond_resched();
+			down_read(&fs_info->commit_root_sem);
+		}
+
+		if (fs_info->last_reloc_trans > sctx->last_reloc_trans) {
+			ret = restart_after_relocation(left_path, right_path,
+						       &left_key, &right_key,
+						       left_level, right_level,
+						       sctx);
+			if (ret < 0)
+				goto out_unlock;
+			sctx->last_reloc_trans = fs_info->last_reloc_trans;
+		}
+
 		if (advance_left && !left_end_reached) {
 			ret = tree_advance(left_path, &left_level,
 					left_root_level,
@@ -6959,7 +7227,7 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 			if (ret == -1)
 				left_end_reached = ADVANCE;
 			else if (ret < 0)
-				goto out;
+				goto out_unlock;
 			advance_left = 0;
 		}
 		if (advance_right && !right_end_reached) {
@@ -6970,54 +7238,55 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 			if (ret == -1)
 				right_end_reached = ADVANCE;
 			else if (ret < 0)
-				goto out;
+				goto out_unlock;
 			advance_right = 0;
 		}
 
 		if (left_end_reached && right_end_reached) {
 			ret = 0;
-			goto out;
+			goto out_unlock;
 		} else if (left_end_reached) {
 			if (right_level == 0) {
+				up_read(&fs_info->commit_root_sem);
 				ret = changed_cb(left_path, right_path,
 						&right_key,
 						BTRFS_COMPARE_TREE_DELETED,
 						sctx);
 				if (ret < 0)
 					goto out;
+				down_read(&fs_info->commit_root_sem);
 			}
 			advance_right = ADVANCE;
 			continue;
 		} else if (right_end_reached) {
 			if (left_level == 0) {
+				up_read(&fs_info->commit_root_sem);
 				ret = changed_cb(left_path, right_path,
 						&left_key,
 						BTRFS_COMPARE_TREE_NEW,
 						sctx);
 				if (ret < 0)
 					goto out;
+				down_read(&fs_info->commit_root_sem);
 			}
 			advance_left = ADVANCE;
 			continue;
 		}
 
 		if (left_level == 0 && right_level == 0) {
+			up_read(&fs_info->commit_root_sem);
 			cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
 			if (cmp < 0) {
 				ret = changed_cb(left_path, right_path,
 						&left_key,
 						BTRFS_COMPARE_TREE_NEW,
 						sctx);
-				if (ret < 0)
-					goto out;
 				advance_left = ADVANCE;
 			} else if (cmp > 0) {
 				ret = changed_cb(left_path, right_path,
 						&right_key,
 						BTRFS_COMPARE_TREE_DELETED,
 						sctx);
-				if (ret < 0)
-					goto out;
 				advance_right = ADVANCE;
 			} else {
 				enum btrfs_compare_tree_result result;
@@ -7031,11 +7300,13 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 					result = BTRFS_COMPARE_TREE_SAME;
 				ret = changed_cb(left_path, right_path,
 						 &left_key, result, sctx);
-				if (ret < 0)
-					goto out;
 				advance_left = ADVANCE;
 				advance_right = ADVANCE;
 			}
+
+			if (ret < 0)
+				goto out;
+			down_read(&fs_info->commit_root_sem);
 		} else if (left_level == right_level) {
 			cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
 			if (cmp < 0) {
@@ -7075,6 +7346,8 @@ static int btrfs_compare_trees(struct btrfs_root *left_root,
 		}
 	}
 
+out_unlock:
+	up_read(&fs_info->commit_root_sem);
 out:
 	btrfs_free_path(left_path);
 	btrfs_free_path(right_path);
@@ -7413,21 +7686,7 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
 	if (ret)
 		goto out;
 
-	spin_lock(&fs_info->send_reloc_lock);
-	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
-		spin_unlock(&fs_info->send_reloc_lock);
-		btrfs_warn_rl(fs_info,
-		"cannot run send because a relocation operation is in progress");
-		ret = -EAGAIN;
-		goto out;
-	}
-	fs_info->send_in_progress++;
-	spin_unlock(&fs_info->send_reloc_lock);
-
 	ret = send_subvol(sctx);
-	spin_lock(&fs_info->send_reloc_lock);
-	fs_info->send_in_progress--;
-	spin_unlock(&fs_info->send_reloc_lock);
 	if (ret < 0)
 		goto out;
 

fs/btrfs/transaction.c

@@ -163,6 +163,10 @@ static noinline void switch_commit_roots(struct btrfs_trans_handle *trans)
 	struct btrfs_caching_control *caching_ctl, *next;
 
 	down_write(&fs_info->commit_root_sem);
+
+	if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
+		fs_info->last_reloc_trans = trans->transid;
+
 	list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits,
 				 dirty_list) {
 		list_del_init(&root->dirty_list);

fs/fuse/dev.c

@@ -941,7 +941,17 @@ static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
 
 	while (count) {
 		if (cs->write && cs->pipebufs && page) {
-			return fuse_ref_page(cs, page, offset, count);
+			/*
+			 * Can't control lifetime of pipe buffers, so always
+			 * copy user pages.
+			 */
+			if (cs->req->args->user_pages) {
+				err = fuse_copy_fill(cs);
+				if (err)
+					return err;
+			} else {
+				return fuse_ref_page(cs, page, offset, count);
+			}
 		} else if (!cs->len) {
 			if (cs->move_pages && page &&
 			    offset == 0 && count == PAGE_SIZE) {

fs/fuse/file.c

@@ -1417,6 +1417,7 @@ static int fuse_get_user_pages(struct fuse_args_pages *ap, struct iov_iter *ii,
 			(PAGE_SIZE - ret) & (PAGE_SIZE - 1);
 	}
 
+	ap->args.user_pages = true;
 	if (write)
 		ap->args.in_pages = true;
 	else

fs/fuse/fuse_i.h

@@ -256,6 +256,7 @@ struct fuse_args {
 	bool nocreds:1;
 	bool in_pages:1;
 	bool out_pages:1;
+	bool user_pages:1;
 	bool out_argvar:1;
 	bool page_zeroing:1;
 	bool page_replace:1;

fs/fuse/ioctl.c

@@ -394,9 +394,12 @@ static int fuse_priv_ioctl(struct inode *inode, struct fuse_file *ff,
 	args.out_args[1].value = ptr;
 
 	err = fuse_simple_request(fm, &args);
-	if (!err && outarg.flags & FUSE_IOCTL_RETRY)
+	if (!err) {
-		err = -EIO;
+		if (outarg.result < 0)
-
+			err = outarg.result;
+		else if (outarg.flags & FUSE_IOCTL_RETRY)
+			err = -EIO;
+	}
 	return err;
 }
 

fs/pipe.c

@@ -252,7 +252,8 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)
 	 */
 	was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage);
 	for (;;) {
-		unsigned int head = pipe->head;
+		/* Read ->head with a barrier vs post_one_notification() */
+		unsigned int head = smp_load_acquire(&pipe->head);
 		unsigned int tail = pipe->tail;
 		unsigned int mask = pipe->ring_size - 1;
 
@@ -830,10 +831,8 @@ void free_pipe_info(struct pipe_inode_info *pipe)
 	int i;
 
 #ifdef CONFIG_WATCH_QUEUE
-	if (pipe->watch_queue) {
+	if (pipe->watch_queue)
 		watch_queue_clear(pipe->watch_queue);
-		put_watch_queue(pipe->watch_queue);
-	}
 #endif
 
 	(void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0);
@@ -843,6 +842,10 @@ void free_pipe_info(struct pipe_inode_info *pipe)
 		if (buf->ops)
 			pipe_buf_release(pipe, buf);
 	}
+#ifdef CONFIG_WATCH_QUEUE
+	if (pipe->watch_queue)
+		put_watch_queue(pipe->watch_queue);
+#endif
 	if (pipe->tmp_page)
 		__free_page(pipe->tmp_page);
 	kfree(pipe->bufs);

include/linux/virtio_config.h

@@ -62,8 +62,9 @@ struct virtio_shm_region {
  *	Returns the first 64 feature bits (all we currently need).
  * @finalize_features: confirm what device features we'll be using.
  *	vdev: the virtio_device
- *	This gives the final feature bits for the device: it can change
+ *	This sends the driver feature bits to the device: it can change
  *	the dev->feature bits if it wants.
+ * Note: despite the name this can be called any number of times.
  *	Returns 0 on success or error status
  * @bus_name: return the bus name associated with the device (optional)
  *	vdev: the virtio_device

include/linux/watch_queue.h

@@ -28,7 +28,8 @@ struct watch_type_filter {
 struct watch_filter {
 	union {
 		struct rcu_head	rcu;
-		unsigned long	type_filter[2];	/* Bitmask of accepted types */
+		/* Bitmask of accepted types */
+		DECLARE_BITMAP(type_filter, WATCH_TYPE__NR);
 	};
 	u32			nr_filters;	/* Number of filters */
 	struct watch_type_filter filters[];

include/net/dsa.h

@@ -1056,7 +1056,6 @@ void dsa_unregister_switch(struct dsa_switch *ds);
 int dsa_register_switch(struct dsa_switch *ds);
 void dsa_switch_shutdown(struct dsa_switch *ds);
 struct dsa_switch *dsa_switch_find(int tree_index, int sw_index);
-void dsa_flush_workqueue(void);
 #ifdef CONFIG_PM_SLEEP
 int dsa_switch_suspend(struct dsa_switch *ds);
 int dsa_switch_resume(struct dsa_switch *ds);

include/net/esp.h

@@ -4,6 +4,8 @@
 
 #include <linux/skbuff.h>
 
+#define ESP_SKB_FRAG_MAXSIZE (PAGE_SIZE << SKB_FRAG_PAGE_ORDER)
+
 struct ip_esp_hdr;
 
 static inline struct ip_esp_hdr *ip_esp_hdr(const struct sk_buff *skb)

include/net/xfrm.h

@@ -1679,14 +1679,15 @@ int km_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
 	       const struct xfrm_migrate *m, int num_bundles,
 	       const struct xfrm_kmaddress *k,
 	       const struct xfrm_encap_tmpl *encap);
-struct xfrm_state *xfrm_migrate_state_find(struct xfrm_migrate *m, struct net *net);
+struct xfrm_state *xfrm_migrate_state_find(struct xfrm_migrate *m, struct net *net,
+						u32 if_id);
 struct xfrm_state *xfrm_state_migrate(struct xfrm_state *x,
 				      struct xfrm_migrate *m,
 				      struct xfrm_encap_tmpl *encap);
 int xfrm_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
 		 struct xfrm_migrate *m, int num_bundles,
 		 struct xfrm_kmaddress *k, struct net *net,
-		 struct xfrm_encap_tmpl *encap);
+		 struct xfrm_encap_tmpl *encap, u32 if_id);
 #endif
 
 int km_new_mapping(struct xfrm_state *x, xfrm_address_t *ipaddr, __be16 sport);