Merge tag 'for-5.11-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs updates from David Sterba:
 "We have a mix of all kinds of changes, feature updates, core stuff,
  performance improvements and lots of cleanups and preparatory changes.

  User visible:

   - export filesystem generation in sysfs

   - new features for mount option 'rescue':
       - what's currently supported is exported in sysfs
       - 'ignorebadroots'/'ibadroots' - continue even if some essential
         tree roots are not usable (extent, uuid, data reloc, device,
         csum, free space)
       - 'ignoredatacsums'/'idatacsums' - skip checksum verification on
         data
       - 'all' - now enables 'ignorebadroots' + 'ignoredatacsums' +
         'nologreplay'

   - export read mirror policy settings to sysfs, new policies will be
     added in the future

   - remove inode number cache feature (mount -o inode_cache), obsoleted
     in 5.9

  User visible fixes:

   - async discard scheduling fixes on high loads

   - update inode byte counter atomically so stat() does not report
     wrong value in some cases

   - free space tree fixes:
       - correctly report status of v2 after remount
       - clear v1 cache inodes when v2 is newly enabled after remount

  Core:

   - switch own tree lock implementation to standard rw semaphore:
       - one-level lock nesting is not required anymore, the last use of
         this was in free space that's now loaded asynchronously
       - own implementation of adaptive spinning before taking mutex has
         been part of rwsem
       - performance seems to be better in general, much better (+tens
         of percents) for some workloads
       - lockdep does not complain

   - finish direct IO conversion to iomap infrastructure, remove
     temporary workaround for DSYNC after iomap API updates

   - preparatory work to support data and metadata blocks smaller than
     page:
       - generalize code that assumes sectorsize == PAGE_SIZE, lots of
         refactoring
       - planned namely for 64K pages (eg. arm64, ppc64)
       - scrub read-only support

   - preparatory work for zoned allocation mode (SMR/ZBC/ZNS friendly):
       - disable incompatible features
       - round-robin superblock write

   - free space cache (v1) is loaded asynchronously, remove tree path
     recursion

   - slightly improved time tacking for transaction kthread wake ups

  Performance improvements (note that the numbers depend on load type or
  other features and weren't run on the same machine):

   - skip unnecessary work:
       - do not start readahead for csum tree when scrubbing non-data
         block groups
       - do not start and wait for delalloc on snapshot roots on
         transaction commit
       - fix race when defragmenting leads to unnecessary IO

   - dbench speedups (+throughput%/-max latency%):
       - skip unnecessary searches for xattrs when logging an inode
         (+10.8/-8.2)
       - stop incrementing log batch when joining log transaction (1-2)
       - unlock path before checking if extent is shared during nocow
         writeback (+5.0/-20.5), on fio load +9.7% throughput/-9.8%
         runtime
       - several tree log improvements, eg. removing unnecessary
         operations, fixing races that lead to additional work
         (+12.7/-8.2)

   - tree-checker error branches annotated with unlikely() (+3%
     throughput)

  Other:

   - cleanups

   - lockdep fixes

   - more btrfs_inode conversions

   - error variable cleanups"

* tag 'for-5.11-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (198 commits)
  btrfs: scrub: allow scrub to work with subpage sectorsize
  btrfs: scrub: support subpage data scrub
  btrfs: scrub: support subpage tree block scrub
  btrfs: scrub: always allocate one full page for one sector for RAID56
  btrfs: scrub: reduce width of extent_len/stripe_len from 64 to 32 bits
  btrfs: refactor btrfs_lookup_bio_sums to handle out-of-order bvecs
  btrfs: remove btrfs_find_ordered_sum call from btrfs_lookup_bio_sums
  btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors
  btrfs: update num_extent_pages to support subpage sized extent buffer
  btrfs: don't allow tree block to cross page boundary for subpage support
  btrfs: calculate inline extent buffer page size based on page size
  btrfs: factor out btree page submission code to a helper
  btrfs: make btrfs_verify_data_csum follow sector size
  btrfs: pass bio_offset to check_data_csum() directly
  btrfs: rename bio_offset of extent_submit_bio_start_t to dio_file_offset
  btrfs: fix lockdep warning when creating free space tree
  btrfs: skip space_cache v1 setup when not using it
  btrfs: remove free space items when disabling space cache v1
  btrfs: warn when remount will not change the free space tree
  btrfs: use superblock state to print space_cache mount option
  ...

1 files changed, 38 insertions, 421 deletions

diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index 66e02ebdd340..5fafc5e89bb7 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -17,404 +17,89 @@
  * Extent buffer locking
  * =====================
  *
- * The locks use a custom scheme that allows to do more operations than are
- * available fromt current locking primitives. The building blocks are still
- * rwlock and wait queues.
- *
- * Required semantics:
+ * We use a rw_semaphore for tree locking, and the semantics are exactly the
+ * same:
  *
  * - reader/writer exclusion
  * - writer/writer exclusion
  * - reader/reader sharing
- * - spinning lock semantics
- * - blocking lock semantics
  * - try-lock semantics for readers and writers
- * - one level nesting, allowing read lock to be taken by the same thread that
- *   already has write lock
- *
- * The extent buffer locks (also called tree locks) manage access to eb data
- * related to the storage in the b-tree (keys, items, but not the individual
- * members of eb).
- * We want concurrency of many readers and safe updates. The underlying locking
- * is done by read-write spinlock and the blocking part is implemented using
- * counters and wait queues.
- *
- * spinning semantics - the low-level rwlock is held so all other threads that
- *                      want to take it are spinning on it.
- *
- * blocking semantics - the low-level rwlock is not held but the counter
- *                      denotes how many times the blocking lock was held;
- *                      sleeping is possible
- *
- * Write lock always allows only one thread to access the data.
- *
- *
- * Debugging
- * ---------
- *
- * There are additional state counters that are asserted in various contexts,
- * removed from non-debug build to reduce extent_buffer size and for
- * performance reasons.
- *
- *
- * Lock recursion
- * --------------
- *
- * A write operation on a tree might indirectly start a look up on the same
- * tree.  This can happen when btrfs_cow_block locks the tree and needs to
- * lookup free extents.
- *
- * btrfs_cow_block
- *   ..
- *   alloc_tree_block_no_bg_flush
- *     btrfs_alloc_tree_block
- *       btrfs_reserve_extent
- *         ..
- *         load_free_space_cache
- *           ..
- *           btrfs_lookup_file_extent
- *             btrfs_search_slot
- *
- *
- * Locking pattern - spinning
- * --------------------------
- *
- * The simple locking scenario, the +--+ denotes the spinning section.
- *
- * +- btrfs_tree_lock
- * | - extent_buffer::rwlock is held
- * | - no heavy operations should happen, eg. IO, memory allocations, large
- * |   structure traversals
- * +- btrfs_tree_unock
-*
-*
- * Locking pattern - blocking
- * --------------------------
- *
- * The blocking write uses the following scheme.  The +--+ denotes the spinning
- * section.
- *
- * +- btrfs_tree_lock
- * |
- * +- btrfs_set_lock_blocking_write
- *
- *   - allowed: IO, memory allocations, etc.
- *
- * -- btrfs_tree_unlock - note, no explicit unblocking necessary
- *
- *
- * Blocking read is similar.
- *
- * +- btrfs_tree_read_lock
- * |
- * +- btrfs_set_lock_blocking_read
- *
- *  - heavy operations allowed
- *
- * +- btrfs_tree_read_unlock_blocking
- * |
- * +- btrfs_tree_read_unlock
- *
- */
-
-#ifdef CONFIG_BTRFS_DEBUG
-static inline void btrfs_assert_spinning_writers_get(struct extent_buffer *eb)
-{
-	WARN_ON(eb->spinning_writers);
-	eb->spinning_writers++;
-}
-
-static inline void btrfs_assert_spinning_writers_put(struct extent_buffer *eb)
-{
-	WARN_ON(eb->spinning_writers != 1);
-	eb->spinning_writers--;
-}
-
-static inline void btrfs_assert_no_spinning_writers(struct extent_buffer *eb)
-{
-	WARN_ON(eb->spinning_writers);
-}
-
-static inline void btrfs_assert_spinning_readers_get(struct extent_buffer *eb)
-{
-	atomic_inc(&eb->spinning_readers);
-}
-
-static inline void btrfs_assert_spinning_readers_put(struct extent_buffer *eb)
-{
-	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
-	atomic_dec(&eb->spinning_readers);
-}
-
-static inline void btrfs_assert_tree_read_locks_get(struct extent_buffer *eb)
-{
-	atomic_inc(&eb->read_locks);
-}
-
-static inline void btrfs_assert_tree_read_locks_put(struct extent_buffer *eb)
-{
-	atomic_dec(&eb->read_locks);
-}
-
-static inline void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
-{
-	BUG_ON(!atomic_read(&eb->read_locks));
-}
-
-static inline void btrfs_assert_tree_write_locks_get(struct extent_buffer *eb)
-{
-	eb->write_locks++;
-}
-
-static inline void btrfs_assert_tree_write_locks_put(struct extent_buffer *eb)
-{
-	eb->write_locks--;
-}
-
-#else
-static void btrfs_assert_spinning_writers_get(struct extent_buffer *eb) { }
-static void btrfs_assert_spinning_writers_put(struct extent_buffer *eb) { }
-static void btrfs_assert_no_spinning_writers(struct extent_buffer *eb) { }
-static void btrfs_assert_spinning_readers_put(struct extent_buffer *eb) { }
-static void btrfs_assert_spinning_readers_get(struct extent_buffer *eb) { }
-static void btrfs_assert_tree_read_locked(struct extent_buffer *eb) { }
-static void btrfs_assert_tree_read_locks_get(struct extent_buffer *eb) { }
-static void btrfs_assert_tree_read_locks_put(struct extent_buffer *eb) { }
-static void btrfs_assert_tree_write_locks_get(struct extent_buffer *eb) { }
-static void btrfs_assert_tree_write_locks_put(struct extent_buffer *eb) { }
-#endif
-
-/*
- * Mark already held read lock as blocking. Can be nested in write lock by the
- * same thread.
- *
- * Use when there are potentially long operations ahead so other thread waiting
- * on the lock will not actively spin but sleep instead.
- *
- * The rwlock is released and blocking reader counter is increased.
- */
-void btrfs_set_lock_blocking_read(struct extent_buffer *eb)
-{
-	trace_btrfs_set_lock_blocking_read(eb);
-	/*
-	 * No lock is required.  The lock owner may change if we have a read
-	 * lock, but it won't change to or away from us.  If we have the write
-	 * lock, we are the owner and it'll never change.
-	 */
-	if (eb->lock_recursed && current->pid == eb->lock_owner)
-		return;
-	btrfs_assert_tree_read_locked(eb);
-	atomic_inc(&eb->blocking_readers);
-	btrfs_assert_spinning_readers_put(eb);
-	read_unlock(&eb->lock);
-}
-
-/*
- * Mark already held write lock as blocking.
- *
- * Use when there are potentially long operations ahead so other threads
- * waiting on the lock will not actively spin but sleep instead.
  *
- * The rwlock is released and blocking writers is set.
+ * The rwsem implementation does opportunistic spinning which reduces number of
+ * times the locking task needs to sleep.
  */
-void btrfs_set_lock_blocking_write(struct extent_buffer *eb)
-{
-	trace_btrfs_set_lock_blocking_write(eb);
-	/*
-	 * No lock is required.  The lock owner may change if we have a read
-	 * lock, but it won't change to or away from us.  If we have the write
-	 * lock, we are the owner and it'll never change.
-	 */
-	if (eb->lock_recursed && current->pid == eb->lock_owner)
-		return;
-	if (eb->blocking_writers == 0) {
-		btrfs_assert_spinning_writers_put(eb);
-		btrfs_assert_tree_locked(eb);
-		WRITE_ONCE(eb->blocking_writers, 1);
-		write_unlock(&eb->lock);
-	}
-}
 
 /*
- * Lock the extent buffer for read. Wait for any writers (spinning or blocking).
- * Can be nested in write lock by the same thread.
+ * __btrfs_tree_read_lock - lock extent buffer for read
+ * @eb:		the eb to be locked
+ * @nest:	the nesting level to be used for lockdep
  *
- * Use when the locked section does only lightweight actions and busy waiting
- * would be cheaper than making other threads do the wait/wake loop.
- *
- * The rwlock is held upon exit.
+ * This takes the read lock on the extent buffer, using the specified nesting
+ * level for lockdep purposes.
  */
-void __btrfs_tree_read_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest,
-			    bool recurse)
+void __btrfs_tree_read_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
 {
 	u64 start_ns = 0;
 
 	if (trace_btrfs_tree_read_lock_enabled())
 		start_ns = ktime_get_ns();
-again:
-	read_lock(&eb->lock);
-	BUG_ON(eb->blocking_writers == 0 &&
-	       current->pid == eb->lock_owner);
-	if (eb->blocking_writers) {
-		if (current->pid == eb->lock_owner) {
-			/*
-			 * This extent is already write-locked by our thread.
-			 * We allow an additional read lock to be added because
-			 * it's for the same thread. btrfs_find_all_roots()
-			 * depends on this as it may be called on a partly
-			 * (write-)locked tree.
-			 */
-			WARN_ON(!recurse);
-			BUG_ON(eb->lock_recursed);
-			eb->lock_recursed = true;
-			read_unlock(&eb->lock);
-			trace_btrfs_tree_read_lock(eb, start_ns);
-			return;
-		}
-		read_unlock(&eb->lock);
-		wait_event(eb->write_lock_wq,
-			   READ_ONCE(eb->blocking_writers) == 0);
-		goto again;
-	}
-	btrfs_assert_tree_read_locks_get(eb);
-	btrfs_assert_spinning_readers_get(eb);
+
+	down_read_nested(&eb->lock, nest);
+	eb->lock_owner = current->pid;
 	trace_btrfs_tree_read_lock(eb, start_ns);
 }
 
 void btrfs_tree_read_lock(struct extent_buffer *eb)
 {
-	__btrfs_tree_read_lock(eb, BTRFS_NESTING_NORMAL, false);
-}
-
-/*
- * Lock extent buffer for read, optimistically expecting that there are no
- * contending blocking writers. If there are, don't wait.
- *
- * Return 1 if the rwlock has been taken, 0 otherwise
- */
-int btrfs_tree_read_lock_atomic(struct extent_buffer *eb)
-{
-	if (READ_ONCE(eb->blocking_writers))
-		return 0;
-
-	read_lock(&eb->lock);
-	/* Refetch value after lock */
-	if (READ_ONCE(eb->blocking_writers)) {
-		read_unlock(&eb->lock);
-		return 0;
-	}
-	btrfs_assert_tree_read_locks_get(eb);
-	btrfs_assert_spinning_readers_get(eb);
-	trace_btrfs_tree_read_lock_atomic(eb);
-	return 1;
+	__btrfs_tree_read_lock(eb, BTRFS_NESTING_NORMAL);
 }
 
 /*
- * Try-lock for read. Don't block or wait for contending writers.
+ * Try-lock for read.
  *
  * Retrun 1 if the rwlock has been taken, 0 otherwise
  */
 int btrfs_try_tree_read_lock(struct extent_buffer *eb)
 {
-	if (READ_ONCE(eb->blocking_writers))
-		return 0;
-
-	if (!read_trylock(&eb->lock))
-		return 0;
-
-	/* Refetch value after lock */
-	if (READ_ONCE(eb->blocking_writers)) {
-		read_unlock(&eb->lock);
-		return 0;
+	if (down_read_trylock(&eb->lock)) {
+		eb->lock_owner = current->pid;
+		trace_btrfs_try_tree_read_lock(eb);
+		return 1;
 	}
-	btrfs_assert_tree_read_locks_get(eb);
-	btrfs_assert_spinning_readers_get(eb);
-	trace_btrfs_try_tree_read_lock(eb);
-	return 1;
+	return 0;
 }
 
 /*
- * Try-lock for write. May block until the lock is uncontended, but does not
- * wait until it is free.
+ * Try-lock for write.
  *
  * Retrun 1 if the rwlock has been taken, 0 otherwise
  */
 int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 {
-	if (READ_ONCE(eb->blocking_writers) || atomic_read(&eb->blocking_readers))
-		return 0;
-
-	write_lock(&eb->lock);
-	/* Refetch value after lock */
-	if (READ_ONCE(eb->blocking_writers) || atomic_read(&eb->blocking_readers)) {
-		write_unlock(&eb->lock);
-		return 0;
+	if (down_write_trylock(&eb->lock)) {
+		eb->lock_owner = current->pid;
+		trace_btrfs_try_tree_write_lock(eb);
+		return 1;
 	}
-	btrfs_assert_tree_write_locks_get(eb);
-	btrfs_assert_spinning_writers_get(eb);
-	eb->lock_owner = current->pid;
-	trace_btrfs_try_tree_write_lock(eb);
-	return 1;
+	return 0;
 }
 
 /*
- * Release read lock. Must be used only if the lock is in spinning mode.  If
- * the read lock is nested, must pair with read lock before the write unlock.
- *
- * The rwlock is not held upon exit.
+ * Release read lock.
  */
 void btrfs_tree_read_unlock(struct extent_buffer *eb)
 {
 	trace_btrfs_tree_read_unlock(eb);
-	/*
-	 * if we're nested, we have the write lock.  No new locking
-	 * is needed as long as we are the lock owner.
-	 * The write unlock will do a barrier for us, and the lock_recursed
-	 * field only matters to the lock owner.
-	 */
-	if (eb->lock_recursed && current->pid == eb->lock_owner) {
-		eb->lock_recursed = false;
-		return;
-	}
-	btrfs_assert_tree_read_locked(eb);
-	btrfs_assert_spinning_readers_put(eb);
-	btrfs_assert_tree_read_locks_put(eb);
-	read_unlock(&eb->lock);
-}
-
-/*
- * Release read lock, previously set to blocking by a pairing call to
- * btrfs_set_lock_blocking_read(). Can be nested in write lock by the same
- * thread.
- *
- * State of rwlock is unchanged, last reader wakes waiting threads.
- */
-void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
-{
-	trace_btrfs_tree_read_unlock_blocking(eb);
-	/*
-	 * if we're nested, we have the write lock.  No new locking
-	 * is needed as long as we are the lock owner.
-	 * The write unlock will do a barrier for us, and the lock_recursed
-	 * field only matters to the lock owner.
-	 */
-	if (eb->lock_recursed && current->pid == eb->lock_owner) {
-		eb->lock_recursed = false;
-		return;
-	}
-	btrfs_assert_tree_read_locked(eb);
-	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
-	/* atomic_dec_and_test implies a barrier */
-	if (atomic_dec_and_test(&eb->blocking_readers))
-		cond_wake_up_nomb(&eb->read_lock_wq);
-	btrfs_assert_tree_read_locks_put(eb);
+	eb->lock_owner = 0;
+	up_read(&eb->lock);
 }
 
 /*
- * Lock for write. Wait for all blocking and spinning readers and writers. This
- * starts context where reader lock could be nested by the same thread.
+ * __btrfs_tree_lock - lock eb for write
+ * @eb:		the eb to lock
+ * @nest:	the nesting to use for the lock
  *
- * The rwlock is held for write upon exit.
+ * Returns with the eb->lock write locked.
  */
 void __btrfs_tree_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
 	__acquires(&eb->lock)
@@ -424,19 +109,7 @@ void __btrfs_tree_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
 	if (trace_btrfs_tree_lock_enabled())
 		start_ns = ktime_get_ns();
 
-	WARN_ON(eb->lock_owner == current->pid);
-again:
-	wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
-	wait_event(eb->write_lock_wq, READ_ONCE(eb->blocking_writers) == 0);
-	write_lock(&eb->lock);
-	/* Refetch value after lock */
-	if (atomic_read(&eb->blocking_readers) ||
-	    READ_ONCE(eb->blocking_writers)) {
-		write_unlock(&eb->lock);
-		goto again;
-	}
-	btrfs_assert_spinning_writers_get(eb);
-	btrfs_assert_tree_write_locks_get(eb);
+	down_write_nested(&eb->lock, nest);
 	eb->lock_owner = current->pid;
 	trace_btrfs_tree_lock(eb, start_ns);
 }
@@ -447,68 +120,13 @@ void btrfs_tree_lock(struct extent_buffer *eb)
 }
 
 /*
- * Release the write lock, either blocking or spinning (ie. there's no need
- * for an explicit blocking unlock, like btrfs_tree_read_unlock_blocking).
- * This also ends the context for nesting, the read lock must have been
- * released already.
- *
- * Tasks blocked and waiting are woken, rwlock is not held upon exit.
+ * Release the write lock.
  */
 void btrfs_tree_unlock(struct extent_buffer *eb)
 {
-	/*
-	 * This is read both locked and unlocked but always by the same thread
-	 * that already owns the lock so we don't need to use READ_ONCE
-	 */
-	int blockers = eb->blocking_writers;
-
-	BUG_ON(blockers > 1);
-
-	btrfs_assert_tree_locked(eb);
 	trace_btrfs_tree_unlock(eb);
 	eb->lock_owner = 0;
-	btrfs_assert_tree_write_locks_put(eb);
-
-	if (blockers) {
-		btrfs_assert_no_spinning_writers(eb);
-		/* Unlocked write */
-		WRITE_ONCE(eb->blocking_writers, 0);
-		/*
-		 * We need to order modifying blocking_writers above with
-		 * actually waking up the sleepers to ensure they see the
-		 * updated value of blocking_writers
-		 */
-		cond_wake_up(&eb->write_lock_wq);
-	} else {
-		btrfs_assert_spinning_writers_put(eb);
-		write_unlock(&eb->lock);
-	}
-}
-
-/*
- * Set all locked nodes in the path to blocking locks.  This should be done
- * before scheduling
- */
-void btrfs_set_path_blocking(struct btrfs_path *p)
-{
-	int i;
-
-	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
-		if (!p->nodes[i] || !p->locks[i])
-			continue;
-		/*
-		 * If we currently have a spinning reader or writer lock this
-		 * will bump the count of blocking holders and drop the
-		 * spinlock.
-		 */
-		if (p->locks[i] == BTRFS_READ_LOCK) {
-			btrfs_set_lock_blocking_read(p->nodes[i]);
-			p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
-		} else if (p->locks[i] == BTRFS_WRITE_LOCK) {
-			btrfs_set_lock_blocking_write(p->nodes[i]);
-			p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
-		}
-	}
+	up_write(&eb->lock);
 }
 
 /*
@@ -564,14 +182,13 @@ struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
  *
  * Return: root extent buffer with read lock held
  */
-struct extent_buffer *__btrfs_read_lock_root_node(struct btrfs_root *root,
-						  bool recurse)
+struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
 {
 	struct extent_buffer *eb;
 
 	while (1) {
 		eb = btrfs_root_node(root);
-		__btrfs_tree_read_lock(eb, BTRFS_NESTING_NORMAL, recurse);
+		btrfs_tree_read_lock(eb);
 		if (eb == root->node)
 			break;
 		btrfs_tree_read_unlock(eb);