Merge tag 'fuse-update-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse

Pull fuse updates from Miklos Szeredi:
 "Scalability and performance improvements, as well as minor bug fixes
  and cleanups"

* tag 'fuse-update-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse: (25 commits)
  fuse: cache readdir calls if filesystem opts out of opendir
  fuse: support clients that don't implement 'opendir'
  fuse: lift bad inode checks into callers
  fuse: multiplex cached/direct_io file operations
  fuse add copy_file_range to direct io fops
  fuse: use iov_iter based generic splice helpers
  fuse: Switch to using async direct IO for FOPEN_DIRECT_IO
  fuse: use atomic64_t for khctr
  fuse: clean up aborted
  fuse: Protect ff->reserved_req via corresponding fi->lock
  fuse: Protect fi->nlookup with fi->lock
  fuse: Introduce fi->lock to protect write related fields
  fuse: Convert fc->attr_version into atomic64_t
  fuse: Add fuse_inode argument to fuse_prepare_release()
  fuse: Verify userspace asks to requeue interrupt that we really sent
  fuse: Do some refactoring in fuse_dev_do_write()
  fuse: Wake up req->waitq of only if not background
  fuse: Optimize request_end() by not taking fiq->waitq.lock
  fuse: Kill fasync only if interrupt is queued in queue_interrupt()
  fuse: Remove stale comment in end_requests()
  ...

8 files changed, 321 insertions, 259 deletions

diff --git a/fs/fuse/control.c b/fs/fuse/control.c
index 989df5accaee..fe80bea4ad89 100644
--- a/fs/fuse/control.c
+++ b/fs/fuse/control.c
@@ -35,7 +35,9 @@ static ssize_t fuse_conn_abort_write(struct file *file, const char __user *buf,
 {
 	struct fuse_conn *fc = fuse_ctl_file_conn_get(file);
 	if (fc) {
-		fuse_abort_conn(fc, true);
+		if (fc->abort_err)
+			fc->aborted = true;
+		fuse_abort_conn(fc);
 		fuse_conn_put(fc);
 	}
 	return count;
diff --git a/fs/fuse/cuse.c b/fs/fuse/cuse.c
index 8f68181256c0..55a26f351467 100644
--- a/fs/fuse/cuse.c
+++ b/fs/fuse/cuse.c
@@ -141,10 +141,11 @@ static int cuse_open(struct inode *inode, struct file *file)
 
 static int cuse_release(struct inode *inode, struct file *file)
 {
+	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_file *ff = file->private_data;
 	struct fuse_conn *fc = ff->fc;
 
-	fuse_sync_release(ff, file->f_flags);
+	fuse_sync_release(fi, ff, file->f_flags);
 	fuse_conn_put(fc);
 
 	return 0;
@@ -407,7 +408,7 @@ err_unlock:
 err_region:
 	unregister_chrdev_region(devt, 1);
 err:
-	fuse_abort_conn(fc, false);
+	fuse_abort_conn(fc);
 	goto out;
 }
 
@@ -586,7 +587,7 @@ static ssize_t cuse_class_abort_store(struct device *dev,
 {
 	struct cuse_conn *cc = dev_get_drvdata(dev);
 
-	fuse_abort_conn(&cc->fc, false);
+	fuse_abort_conn(&cc->fc);
 	return count;
 }
 static DEVICE_ATTR(abort, 0200, NULL, cuse_class_abort_store);
diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index 809c0f2f9942..8a63e52785e9 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -251,17 +251,18 @@ static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
 					 struct file *file)
 {
 	struct fuse_req *req = NULL;
+	struct fuse_inode *fi = get_fuse_inode(file_inode(file));
 	struct fuse_file *ff = file->private_data;
 
 	do {
 		wait_event(fc->reserved_req_waitq, ff->reserved_req);
-		spin_lock(&fc->lock);
+		spin_lock(&fi->lock);
 		if (ff->reserved_req) {
 			req = ff->reserved_req;
 			ff->reserved_req = NULL;
 			req->stolen_file = get_file(file);
 		}
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 	} while (!req);
 
 	return req;
@@ -273,16 +274,17 @@ static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
 static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
 {
 	struct file *file = req->stolen_file;
+	struct fuse_inode *fi = get_fuse_inode(file_inode(file));
 	struct fuse_file *ff = file->private_data;
 
 	WARN_ON(req->max_pages);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	memset(req, 0, sizeof(*req));
 	fuse_request_init(req, NULL, NULL, 0);
 	BUG_ON(ff->reserved_req);
 	ff->reserved_req = req;
 	wake_up_all(&fc->reserved_req_waitq);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 	fput(file);
 }
 
@@ -431,10 +433,16 @@ static void request_end(struct fuse_conn *fc, struct fuse_req *req)
 
 	if (test_and_set_bit(FR_FINISHED, &req->flags))
 		goto put_request;
-
-	spin_lock(&fiq->waitq.lock);
-	list_del_init(&req->intr_entry);
-	spin_unlock(&fiq->waitq.lock);
+	/*
+	 * test_and_set_bit() implies smp_mb() between bit
+	 * changing and below intr_entry check. Pairs with
+	 * smp_mb() from queue_interrupt().
+	 */
+	if (!list_empty(&req->intr_entry)) {
+		spin_lock(&fiq->waitq.lock);
+		list_del_init(&req->intr_entry);
+		spin_unlock(&fiq->waitq.lock);
+	}
 	WARN_ON(test_bit(FR_PENDING, &req->flags));
 	WARN_ON(test_bit(FR_SENT, &req->flags));
 	if (test_bit(FR_BACKGROUND, &req->flags)) {
@@ -462,27 +470,43 @@ static void request_end(struct fuse_conn *fc, struct fuse_req *req)
 		fc->active_background--;
 		flush_bg_queue(fc);
 		spin_unlock(&fc->bg_lock);
+	} else {
+		/* Wake up waiter sleeping in request_wait_answer() */
+		wake_up(&req->waitq);
 	}
-	wake_up(&req->waitq);
+
 	if (req->end)
 		req->end(fc, req);
 put_request:
 	fuse_put_request(fc, req);
 }
 
-static void queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
+static int queue_interrupt(struct fuse_iqueue *fiq, struct fuse_req *req)
 {
 	spin_lock(&fiq->waitq.lock);
-	if (test_bit(FR_FINISHED, &req->flags)) {
+	/* Check for we've sent request to interrupt this req */
+	if (unlikely(!test_bit(FR_INTERRUPTED, &req->flags))) {
 		spin_unlock(&fiq->waitq.lock);
-		return;
+		return -EINVAL;
 	}
+
 	if (list_empty(&req->intr_entry)) {
 		list_add_tail(&req->intr_entry, &fiq->interrupts);
+		/*
+		 * Pairs with smp_mb() implied by test_and_set_bit()
+		 * from request_end().
+		 */
+		smp_mb();
+		if (test_bit(FR_FINISHED, &req->flags)) {
+			list_del_init(&req->intr_entry);
+			spin_unlock(&fiq->waitq.lock);
+			return 0;
+		}
 		wake_up_locked(&fiq->waitq);
+		kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
 	}
 	spin_unlock(&fiq->waitq.lock);
-	kill_fasync(&fiq->fasync, SIGIO, POLL_IN);
+	return 0;
 }
 
 static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
@@ -1306,7 +1330,7 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
 		goto err_unlock;
 
 	if (!fiq->connected) {
-		err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
+		err = fc->aborted ? -ECONNABORTED : -ENODEV;
 		goto err_unlock;
 	}
 
@@ -1353,7 +1377,7 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
 	spin_lock(&fpq->lock);
 	clear_bit(FR_LOCKED, &req->flags);
 	if (!fpq->connected) {
-		err = (fc->aborted && fc->abort_err) ? -ECONNABORTED : -ENODEV;
+		err = fc->aborted ? -ECONNABORTED : -ENODEV;
 		goto out_end;
 	}
 	if (err) {
@@ -1900,16 +1924,17 @@ static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
 	struct fuse_req *req;
 	struct fuse_out_header oh;
 
+	err = -EINVAL;
 	if (nbytes < sizeof(struct fuse_out_header))
-		return -EINVAL;
+		goto out;
 
 	err = fuse_copy_one(cs, &oh, sizeof(oh));
 	if (err)
-		goto err_finish;
+		goto copy_finish;
 
 	err = -EINVAL;
 	if (oh.len != nbytes)
-		goto err_finish;
+		goto copy_finish;
 
 	/*
 	 * Zero oh.unique indicates unsolicited notification message
@@ -1917,41 +1942,40 @@ static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
 	 */
 	if (!oh.unique) {
 		err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
-		return err ? err : nbytes;
+		goto out;
 	}
 
 	err = -EINVAL;
 	if (oh.error <= -1000 || oh.error > 0)
-		goto err_finish;
+		goto copy_finish;
 
 	spin_lock(&fpq->lock);
-	err = -ENOENT;
-	if (!fpq->connected)
-		goto err_unlock_pq;
+	req = NULL;
+	if (fpq->connected)
+		req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
 
-	req = request_find(fpq, oh.unique & ~FUSE_INT_REQ_BIT);
-	if (!req)
-		goto err_unlock_pq;
+	err = -ENOENT;
+	if (!req) {
+		spin_unlock(&fpq->lock);
+		goto copy_finish;
+	}
 
 	/* Is it an interrupt reply ID? */
 	if (oh.unique & FUSE_INT_REQ_BIT) {
 		__fuse_get_request(req);
 		spin_unlock(&fpq->lock);
 
-		err = -EINVAL;
-		if (nbytes != sizeof(struct fuse_out_header)) {
-			fuse_put_request(fc, req);
-			goto err_finish;
-		}
-
-		if (oh.error == -ENOSYS)
+		err = 0;
+		if (nbytes != sizeof(struct fuse_out_header))
+			err = -EINVAL;
+		else if (oh.error == -ENOSYS)
 			fc->no_interrupt = 1;
 		else if (oh.error == -EAGAIN)
-			queue_interrupt(&fc->iq, req);
+			err = queue_interrupt(&fc->iq, req);
+
 		fuse_put_request(fc, req);
 
-		fuse_copy_finish(cs);
-		return nbytes;
+		goto copy_finish;
 	}
 
 	clear_bit(FR_SENT, &req->flags);
@@ -1977,14 +2001,12 @@ static ssize_t fuse_dev_do_write(struct fuse_dev *fud,
 	spin_unlock(&fpq->lock);
 
 	request_end(fc, req);
-
+out:
 	return err ? err : nbytes;
 
- err_unlock_pq:
-	spin_unlock(&fpq->lock);
- err_finish:
+copy_finish:
 	fuse_copy_finish(cs);
-	return err;
+	goto out;
 }
 
 static ssize_t fuse_dev_write(struct kiocb *iocb, struct iov_iter *from)
@@ -2109,11 +2131,7 @@ static __poll_t fuse_dev_poll(struct file *file, poll_table *wait)
 	return mask;
 }
 
-/*
- * Abort all requests on the given list (pending or processing)
- *
- * This function releases and reacquires fc->lock
- */
+/* Abort all requests on the given list (pending or processing) */
 static void end_requests(struct fuse_conn *fc, struct list_head *head)
 {
 	while (!list_empty(head)) {
@@ -2159,7 +2177,7 @@ static void end_polls(struct fuse_conn *fc)
  * is OK, the request will in that case be removed from the list before we touch
  * it.
  */
-void fuse_abort_conn(struct fuse_conn *fc, bool is_abort)
+void fuse_abort_conn(struct fuse_conn *fc)
 {
 	struct fuse_iqueue *fiq = &fc->iq;
 
@@ -2175,7 +2193,6 @@ void fuse_abort_conn(struct fuse_conn *fc, bool is_abort)
 		fc->connected = 0;
 		spin_unlock(&fc->bg_lock);
 
-		fc->aborted = is_abort;
 		fuse_set_initialized(fc);
 		list_for_each_entry(fud, &fc->devices, entry) {
 			struct fuse_pqueue *fpq = &fud->pq;
@@ -2253,7 +2270,7 @@ int fuse_dev_release(struct inode *inode, struct file *file)
 		/* Are we the last open device? */
 		if (atomic_dec_and_test(&fc->dev_count)) {
 			WARN_ON(fc->iq.fasync != NULL);
-			fuse_abort_conn(fc, false);
+			fuse_abort_conn(fc);
 		}
 		fuse_dev_free(fud);
 	}
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index e909678afa2d..dd0f64f7bc06 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -149,21 +149,6 @@ static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
 	args->out.args[0].value = outarg;
 }
 
-u64 fuse_get_attr_version(struct fuse_conn *fc)
-{
-	u64 curr_version;
-
-	/*
-	 * The spin lock isn't actually needed on 64bit archs, but we
-	 * don't yet care too much about such optimizations.
-	 */
-	spin_lock(&fc->lock);
-	curr_version = fc->attr_version;
-	spin_unlock(&fc->lock);
-
-	return curr_version;
-}
-
 /*
  * Check whether the dentry is still valid
  *
@@ -222,9 +207,9 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
 				fuse_queue_forget(fc, forget, outarg.nodeid, 1);
 				goto invalid;
 			}
-			spin_lock(&fc->lock);
+			spin_lock(&fi->lock);
 			fi->nlookup++;
-			spin_unlock(&fc->lock);
+			spin_unlock(&fi->lock);
 		}
 		kfree(forget);
 		if (ret == -ENOMEM)
@@ -400,6 +385,7 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry,
 	struct fuse_create_in inarg;
 	struct fuse_open_out outopen;
 	struct fuse_entry_out outentry;
+	struct fuse_inode *fi;
 	struct fuse_file *ff;
 
 	/* Userspace expects S_IFREG in create mode */
@@ -451,7 +437,7 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry,
 			  &outentry.attr, entry_attr_timeout(&outentry), 0);
 	if (!inode) {
 		flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
-		fuse_sync_release(ff, flags);
+		fuse_sync_release(NULL, ff, flags);
 		fuse_queue_forget(fc, forget, outentry.nodeid, 1);
 		err = -ENOMEM;
 		goto out_err;
@@ -462,7 +448,8 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry,
 	fuse_dir_changed(dir);
 	err = finish_open(file, entry, generic_file_open);
 	if (err) {
-		fuse_sync_release(ff, flags);
+		fi = get_fuse_inode(inode);
+		fuse_sync_release(fi, ff, flags);
 	} else {
 		file->private_data = ff;
 		fuse_finish_open(inode, file);
@@ -671,8 +658,8 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
 		struct inode *inode = d_inode(entry);
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
-		fi->attr_version = ++fc->attr_version;
+		spin_lock(&fi->lock);
+		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		/*
 		 * If i_nlink == 0 then unlink doesn't make sense, yet this can
 		 * happen if userspace filesystem is careless.  It would be
@@ -681,7 +668,7 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
 		 */
 		if (inode->i_nlink > 0)
 			drop_nlink(inode);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		fuse_dir_changed(dir);
 		fuse_invalidate_entry_cache(entry);
@@ -825,10 +812,10 @@ static int fuse_link(struct dentry *entry, struct inode *newdir,
 	if (!err) {
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
-		fi->attr_version = ++fc->attr_version;
+		spin_lock(&fi->lock);
+		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		inc_nlink(inode);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		fuse_update_ctime(inode);
 	} else if (err == -EINTR) {
@@ -1356,15 +1343,14 @@ static void iattr_to_fattr(struct fuse_conn *fc, struct iattr *iattr,
  */
 void fuse_set_nowrite(struct inode *inode)
 {
-	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	BUG_ON(!inode_is_locked(inode));
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	BUG_ON(fi->writectr < 0);
 	fi->writectr += FUSE_NOWRITE;
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 	wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
 }
 
@@ -1385,11 +1371,11 @@ static void __fuse_release_nowrite(struct inode *inode)
 
 void fuse_release_nowrite(struct inode *inode)
 {
-	struct fuse_conn *fc = get_fuse_conn(inode);
+	struct fuse_inode *fi = get_fuse_inode(inode);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	__fuse_release_nowrite(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
@@ -1524,7 +1510,7 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 		goto error;
 	}
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	/* the kernel maintains i_mtime locally */
 	if (trust_local_cmtime) {
 		if (attr->ia_valid & ATTR_MTIME)
@@ -1542,10 +1528,10 @@ int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
 		i_size_write(inode, outarg.attr.size);
 
 	if (is_truncate) {
-		/* NOTE: this may release/reacquire fc->lock */
+		/* NOTE: this may release/reacquire fi->lock */
 		__fuse_release_nowrite(inode);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	/*
 	 * Only call invalidate_inode_pages2() after removing
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index a59c16bd90ac..06096b60f1df 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -19,8 +19,6 @@
 #include <linux/falloc.h>
 #include <linux/uio.h>
 
-static const struct file_operations fuse_direct_io_file_operations;
-
 static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 			  int opcode, struct fuse_open_out *outargp)
 {
@@ -64,9 +62,7 @@ struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
 	RB_CLEAR_NODE(&ff->polled_node);
 	init_waitqueue_head(&ff->poll_wait);
 
-	spin_lock(&fc->lock);
-	ff->kh = ++fc->khctr;
-	spin_unlock(&fc->lock);
+	ff->kh = atomic64_inc_return(&fc->khctr);
 
 	return ff;
 }
@@ -94,7 +90,7 @@ static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir)
 	if (refcount_dec_and_test(&ff->count)) {
 		struct fuse_req *req = ff->reserved_req;
 
-		if (ff->fc->no_open && !isdir) {
+		if (isdir ? ff->fc->no_opendir : ff->fc->no_open) {
 			/*
 			 * Drop the release request when client does not
 			 * implement 'open'
@@ -128,8 +124,9 @@ int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 		return -ENOMEM;
 
 	ff->fh = 0;
-	ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
-	if (!fc->no_open || isdir) {
+	/* Default for no-open */
+	ff->open_flags = FOPEN_KEEP_CACHE | (isdir ? FOPEN_CACHE_DIR : 0);
+	if (isdir ? !fc->no_opendir : !fc->no_open) {
 		struct fuse_open_out outarg;
 		int err;
 
@@ -138,11 +135,14 @@ int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 			ff->fh = outarg.fh;
 			ff->open_flags = outarg.open_flags;
 
-		} else if (err != -ENOSYS || isdir) {
+		} else if (err != -ENOSYS) {
 			fuse_file_free(ff);
 			return err;
 		} else {
-			fc->no_open = 1;
+			if (isdir)
+				fc->no_opendir = 1;
+			else
+				fc->no_open = 1;
 		}
 	}
 
@@ -159,17 +159,16 @@ EXPORT_SYMBOL_GPL(fuse_do_open);
 static void fuse_link_write_file(struct file *file)
 {
 	struct inode *inode = file_inode(file);
-	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_file *ff = file->private_data;
 	/*
 	 * file may be written through mmap, so chain it onto the
 	 * inodes's write_file list
 	 */
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if (list_empty(&ff->write_entry))
 		list_add(&ff->write_entry, &fi->write_files);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 void fuse_finish_open(struct inode *inode, struct file *file)
@@ -177,8 +176,6 @@ void fuse_finish_open(struct inode *inode, struct file *file)
 	struct fuse_file *ff = file->private_data;
 	struct fuse_conn *fc = get_fuse_conn(inode);
 
-	if (ff->open_flags & FOPEN_DIRECT_IO)
-		file->f_op = &fuse_direct_io_file_operations;
 	if (!(ff->open_flags & FOPEN_KEEP_CACHE))
 		invalidate_inode_pages2(inode->i_mapping);
 	if (ff->open_flags & FOPEN_NONSEEKABLE)
@@ -186,10 +183,10 @@ void fuse_finish_open(struct inode *inode, struct file *file)
 	if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
 		struct fuse_inode *fi = get_fuse_inode(inode);
 
-		spin_lock(&fc->lock);
-		fi->attr_version = ++fc->attr_version;
+		spin_lock(&fi->lock);
+		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		i_size_write(inode, 0);
-		spin_unlock(&fc->lock);
+		spin_unlock(&fi->lock);
 		fuse_invalidate_attr(inode);
 		if (fc->writeback_cache)
 			file_update_time(file);
@@ -224,14 +221,20 @@ int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
 	return err;
 }
 
-static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
+static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
+				 int flags, int opcode)
 {
 	struct fuse_conn *fc = ff->fc;
 	struct fuse_req *req = ff->reserved_req;
 	struct fuse_release_in *inarg = &req->misc.release.in;
 
+	/* Inode is NULL on error path of fuse_create_open() */
+	if (likely(fi)) {
+		spin_lock(&fi->lock);
+		list_del(&ff->write_entry);
+		spin_unlock(&fi->lock);
+	}
 	spin_lock(&fc->lock);
-	list_del(&ff->write_entry);
 	if (!RB_EMPTY_NODE(&ff->polled_node))
 		rb_erase(&ff->polled_node, &fc->polled_files);
 	spin_unlock(&fc->lock);
@@ -249,11 +252,12 @@ static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
 
 void fuse_release_common(struct file *file, bool isdir)
 {
+	struct fuse_inode *fi = get_fuse_inode(file_inode(file));
 	struct fuse_file *ff = file->private_data;
 	struct fuse_req *req = ff->reserved_req;
 	int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
 
-	fuse_prepare_release(ff, file->f_flags, opcode);
+	fuse_prepare_release(fi, ff, file->f_flags, opcode);
 
 	if (ff->flock) {
 		struct fuse_release_in *inarg = &req->misc.release.in;
@@ -295,10 +299,10 @@ static int fuse_release(struct inode *inode, struct file *file)
 	return 0;
 }
 
-void fuse_sync_release(struct fuse_file *ff, int flags)
+void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags)
 {
 	WARN_ON(refcount_read(&ff->count) > 1);
-	fuse_prepare_release(ff, flags, FUSE_RELEASE);
+	fuse_prepare_release(fi, ff, flags, FUSE_RELEASE);
 	/*
 	 * iput(NULL) is a no-op and since the refcount is 1 and everything's
 	 * synchronous, we are fine with not doing igrab() here"
@@ -329,33 +333,39 @@ u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
 	return (u64) v0 + ((u64) v1 << 32);
 }
 
-/*
- * Check if any page in a range is under writeback
- *
- * This is currently done by walking the list of writepage requests
- * for the inode, which can be pretty inefficient.
- */
-static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
-				   pgoff_t idx_to)
+static struct fuse_req *fuse_find_writeback(struct fuse_inode *fi,
+					    pgoff_t idx_from, pgoff_t idx_to)
 {
-	struct fuse_conn *fc = get_fuse_conn(inode);
-	struct fuse_inode *fi = get_fuse_inode(inode);
 	struct fuse_req *req;
-	bool found = false;
 
-	spin_lock(&fc->lock);
 	list_for_each_entry(req, &fi->writepages, writepages_entry) {
 		pgoff_t curr_index;
 
-		BUG_ON(req->inode != inode);
+		WARN_ON(get_fuse_inode(req->inode) != fi);
 		curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
 		if (idx_from < curr_index + req->num_pages &&
 		    curr_index <= idx_to) {
-			found = true;
-			break;
+			return req;
 		}
 	}
-	spin_unlock(&fc->lock);
+	return NULL;
+}
+
+/*
+ * Check if any page in a range is under writeback
+ *
+ * This is currently done by walking the list of writepage requests
+ * for the inode, which can be pretty inefficient.
+ */
+static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
+				   pgoff_t idx_to)
+{
+	struct fuse_inode *fi = get_fuse_inode(inode);
+	bool found;
+
+	spin_lock(&fi->lock);
+	found = fuse_find_writeback(fi, idx_from, idx_to);
+	spin_unlock(&fi->lock);
 
 	return found;
 }
@@ -598,9 +608,9 @@ static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
 			struct fuse_conn *fc = get_fuse_conn(inode);
 			struct fuse_inode *fi = get_fuse_inode(inode);
 
-			spin_lock(&fc->lock);
-			fi->attr_version = ++fc->attr_version;
-			spin_unlock(&fc->lock);
+			spin_lock(&fi->lock);
+			fi->attr_version = atomic64_inc_return(&fc->attr_version);
+			spin_unlock(&fi->lock);
 		}
 
 		io->iocb->ki_complete(io->iocb, res, 0);
@@ -675,13 +685,13 @@ static void fuse_read_update_size(struct inode *inode, loff_t size,
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if (attr_ver == fi->attr_version && size < inode->i_size &&
 	    !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
-		fi->attr_version = ++fc->attr_version;
+		fi->attr_version = atomic64_inc_return(&fc->attr_version);
 		i_size_write(inode, size);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 }
 
 static void fuse_short_read(struct fuse_req *req, struct inode *inode,
@@ -919,7 +929,7 @@ out:
 	return err;
 }
 
-static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
+static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to)
 {
 	struct inode *inode = iocb->ki_filp->f_mapping->host;
 	struct fuse_conn *fc = get_fuse_conn(inode);
@@ -996,13 +1006,13 @@ bool fuse_write_update_size(struct inode *inode, loff_t pos)
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	bool ret = false;
 
-	spin_lock(&fc->lock);
-	fi->attr_version = ++fc->attr_version;
+	spin_lock(&fi->lock);
+	fi->attr_version = atomic64_inc_return(&fc->attr_version);
 	if (pos > inode->i_size) {
 		i_size_write(inode, pos);
 		ret = true;
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	return ret;
 }
@@ -1125,9 +1135,6 @@ static ssize_t fuse_perform_write(struct kiocb *iocb,
 	int err = 0;
 	ssize_t res = 0;
 
-	if (is_bad_inode(inode))
-		return -EIO;
-
 	if (inode->i_size < pos + iov_iter_count(ii))
 		set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
 
@@ -1173,7 +1180,7 @@ static ssize_t fuse_perform_write(struct kiocb *iocb,
 	return res > 0 ? res : err;
 }
 
-static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
+static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
 {
 	struct file *file = iocb->ki_filp;
 	struct address_space *mapping = file->f_mapping;
@@ -1416,9 +1423,6 @@ static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
 	ssize_t res;
 	struct inode *inode = file_inode(io->iocb->ki_filp);
 
-	if (is_bad_inode(inode))
-		return -EIO;
-
 	res = fuse_direct_io(io, iter, ppos, 0);
 
 	fuse_invalidate_atime(inode);
@@ -1426,10 +1430,21 @@ static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
 	return res;
 }
 
+static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter);
+
 static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
 {
-	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
-	return __fuse_direct_read(&io, to, &iocb->ki_pos);
+	ssize_t res;
+
+	if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
+		res = fuse_direct_IO(iocb, to);
+	} else {
+		struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
+
+		res = __fuse_direct_read(&io, to, &iocb->ki_pos);
+	}
+
+	return res;
 }
 
 static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
@@ -1438,14 +1453,17 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
 	ssize_t res;
 
-	if (is_bad_inode(inode))
-		return -EIO;
-
 	/* Don't allow parallel writes to the same file */
 	inode_lock(inode);
 	res = generic_write_checks(iocb, from);
-	if (res > 0)
-		res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE);
+	if (res > 0) {
+		if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
+			res = fuse_direct_IO(iocb, from);
+		} else {
+			res = fuse_direct_io(&io, from, &iocb->ki_pos,
+					     FUSE_DIO_WRITE);
+		}
+	}
 	fuse_invalidate_attr(inode);
 	if (res > 0)
 		fuse_write_update_size(inode, iocb->ki_pos);
@@ -1454,6 +1472,34 @@ static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	return res;
 }
 
+static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
+{
+	struct file *file = iocb->ki_filp;
+	struct fuse_file *ff = file->private_data;
+
+	if (is_bad_inode(file_inode(file)))
+		return -EIO;
+
+	if (!(ff->open_flags & FOPEN_DIRECT_IO))
+		return fuse_cache_read_iter(iocb, to);
+	else
+		return fuse_direct_read_iter(iocb, to);
+}
+
+static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct fuse_file *ff = file->private_data;
+
+	if (is_bad_inode(file_inode(file)))
+		return -EIO;
+
+	if (!(ff->open_flags & FOPEN_DIRECT_IO))
+		return fuse_cache_write_iter(iocb, from);
+	else
+		return fuse_direct_write_iter(iocb, from);
+}
+
 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
 {
 	int i;
@@ -1481,20 +1527,18 @@ static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
 	wake_up(&fi->page_waitq);
 }
 
-/* Called under fc->lock, may release and reacquire it */
+/* Called under fi->lock, may release and reacquire it */
 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
 				loff_t size)
-__releases(fc->lock)
-__acquires(fc->lock)
+__releases(fi->lock)
+__acquires(fi->lock)
 {
+	struct fuse_req *aux, *next;
 	struct fuse_inode *fi = get_fuse_inode(req->inode);
 	struct fuse_write_in *inarg = &req->misc.write.in;
 	__u64 data_size = req->num_pages * PAGE_SIZE;
 	bool queued;
 
-	if (!fc->connected)
-		goto out_free;
-
 	if (inarg->offset + data_size <= size) {
 		inarg->size = data_size;
 	} else if (inarg->offset < size) {
@@ -1505,28 +1549,40 @@ __acquires(fc->lock)
 	}
 
 	req->in.args[1].size = inarg->size;
-	fi->writectr++;
 	queued = fuse_request_queue_background(fc, req);
-	WARN_ON(!queued);
+	/* Fails on broken connection only */
+	if (unlikely(!queued))
+		goto out_free;
+
+	fi->writectr++;
 	return;
 
  out_free:
 	fuse_writepage_finish(fc, req);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
+
+	/* After fuse_writepage_finish() aux request list is private */
+	for (aux = req->misc.write.next; aux; aux = next) {
+		next = aux->misc.write.next;
+		aux->misc.write.next = NULL;
+		fuse_writepage_free(fc, aux);
+		fuse_put_request(fc, aux);
+	}
+
 	fuse_writepage_free(fc, req);
 	fuse_put_request(fc, req);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 }
 
 /*
  * If fi->writectr is positive (no truncate or fsync going on) send
  * all queued writepage requests.
  *
- * Called with fc->lock
+ * Called with fi->lock
  */
 void fuse_flush_writepages(struct inode *inode)
-__releases(fc->lock)
-__acquires(fc->lock)
+__releases(fi->lock)
+__acquires(fi->lock)
 {
 	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
@@ -1546,7 +1602,7 @@ static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
 	struct fuse_inode *fi = get_fuse_inode(inode);
 
 	mapping_set_error(inode->i_mapping, req->out.h.error);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	while (req->misc.write.next) {
 		struct fuse_conn *fc = get_fuse_conn(inode);
 		struct fuse_write_in *inarg = &req->misc.write.in;
@@ -1583,7 +1639,7 @@ static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
 	}
 	fi->writectr--;
 	fuse_writepage_finish(fc, req);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 	fuse_writepage_free(fc, req);
 }
 
@@ -1592,13 +1648,13 @@ static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
 {
 	struct fuse_file *ff = NULL;
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	if (!list_empty(&fi->write_files)) {
 		ff = list_entry(fi->write_files.next, struct fuse_file,
 				write_entry);
 		fuse_file_get(ff);
 	}
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	return ff;
 }
@@ -1669,11 +1725,11 @@ static int fuse_writepage_locked(struct page *page)
 	inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
 	inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
 
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	list_add(&req->writepages_entry, &fi->writepages);
 	list_add_tail(&req->list, &fi->queued_writes);
 	fuse_flush_writepages(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	end_page_writeback(page);
 
@@ -1722,21 +1778,27 @@ static void fuse_writepages_send(struct fuse_fill_wb_data *data)
 {
 	struct fuse_req *req = data->req;
 	struct inode *inode = data->inode;
-	struct fuse_conn *fc = get_fuse_conn(inode);
 	struct fuse_inode *fi = get_fuse_inode(inode);
 	int num_pages = req->num_pages;
 	int i;
 
 	req->ff = fuse_file_get(data->ff);
-	spin_lock(&fc->lock);
+	spin_lock(&fi->lock);
 	list_add_tail(&req->list, &fi->queued_writes);
 	fuse_flush_writepages(inode);
-	spin_unlock(&fc->lock);
+	spin_unlock(&fi->lock);
 
 	for (i = 0; i < num_pages; i++)
 		end_page_writeback(data->orig_pages[i]);
 }
 
+/*
+ * First recheck under fi->lock if the offending offset is still under
+ * writeback.  If yes, then iterate auxiliary write requests, to see if there's
+ * one already added for a page at this offset.  If there's none, then insert
+ * this new request onto the auxiliary list, otherwise reuse the existing one by
+ * copying the new page contents over to the old temporary page.
+ */
 static bool fuse_writepage_in_flight(struct fuse_req *new_req,
 				     struct page *page)
 {
@@ -1744,57 +1806,50 @@ static bool fuse_wr