X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=linux%2Fblkdev.c;h=b7f58737298fa7280f1c949312486c363f85ad14;hb=HEAD;hp=6c940f79fcaf8bce58f959ad4e80c8eac9adfa9a;hpb=ba1dbafc1fce6d37c3703e9788bbf5a72f8e55a3;p=bcachefs-tools-debian

diff --git a/linux/blkdev.c b/linux/blkdev.c
index 6c940f7..b7f5873 100644
--- a/linux/blkdev.c
+++ b/linux/blkdev.c
@@ -113,11 +113,19 @@ int submit_bio_wait(struct bio *bio)
 
 int blkdev_issue_discard(struct block_device *bdev,
 			 sector_t sector, sector_t nr_sects,
-			 gfp_t gfp_mask, unsigned long flags)
+			 gfp_t gfp_mask)
 {
 	return 0;
 }
 
+int blkdev_issue_zeroout(struct block_device *bdev,
+			 sector_t sector, sector_t nr_sects,
+			 gfp_t gfp_mask, unsigned flags)
+{
+	/* Not yet implemented: */
+	BUG();
+}
+
 unsigned bdev_logical_block_size(struct block_device *bdev)
 {
 	struct stat statbuf;
@@ -128,12 +136,10 @@ unsigned bdev_logical_block_size(struct block_device *bdev)
 	BUG_ON(ret);
 
 	if (!S_ISBLK(statbuf.st_mode))
-		return statbuf.st_blksize >> 9;
+		return statbuf.st_blksize;
 
-	ret = ioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
-	BUG_ON(ret);
-
-	return blksize >> 9;
+	xioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
+	return blksize;
 }
 
 sector_t get_capacity(struct gendisk *disk)
@@ -156,45 +162,37 @@ sector_t get_capacity(struct gendisk *disk)
 	return bytes >> 9;
 }
 
-void blkdev_put(struct block_device *bdev, fmode_t mode)
+void bdev_release(struct bdev_handle *handle)
 {
-	fdatasync(bdev->bd_fd);
-	close(bdev->bd_sync_fd);
-	close(bdev->bd_fd);
-	free(bdev);
+	fdatasync(handle->bdev->bd_fd);
+	close(handle->bdev->bd_fd);
+	free(handle->bdev);
+	free(handle);
 }
 
-struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
-					void *holder)
+struct bdev_handle *bdev_open_by_path(const char *path, blk_mode_t mode,
+				      void *holder, const struct blk_holder_ops *hop)
 {
-	struct block_device *bdev;
-	int fd, sync_fd, flags = O_DIRECT;
+	int fd, flags = 0;
 
-	if ((mode & (FMODE_READ|FMODE_WRITE)) == (FMODE_READ|FMODE_WRITE))
+	if ((mode & (BLK_OPEN_READ|BLK_OPEN_WRITE)) == (BLK_OPEN_READ|BLK_OPEN_WRITE))
 		flags = O_RDWR;
-	else if (mode & FMODE_READ)
+	else if (mode & BLK_OPEN_READ)
 		flags = O_RDONLY;
-	else if (mode & FMODE_WRITE)
+	else if (mode & BLK_OPEN_WRITE)
 		flags = O_WRONLY;
 
-#if 0
-	/* using O_EXCL doesn't work with opening twice for an O_SYNC fd: */
-	if (mode & FMODE_EXCL)
+	if (!(mode & BLK_OPEN_BUFFERED))
+		flags |= O_DIRECT;
+
+	if (mode & BLK_OPEN_EXCL)
 		flags |= O_EXCL;
-#endif
 
 	fd = open(path, flags);
 	if (fd < 0)
 		return ERR_PTR(-errno);
 
-	sync_fd = open(path, flags|O_SYNC);
-	if (sync_fd < 0) {
-		assert(0);
-		close(fd);
-		return ERR_PTR(-errno);
-	}
-
-	bdev = malloc(sizeof(*bdev));
+	struct block_device *bdev = malloc(sizeof(*bdev));
 	memset(bdev, 0, sizeof(*bdev));
 
 	strncpy(bdev->name, path, sizeof(bdev->name));
@@ -202,18 +200,17 @@ struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
 
 	bdev->bd_dev		= xfstat(fd).st_rdev;
 	bdev->bd_fd		= fd;
-	bdev->bd_sync_fd	= sync_fd;
 	bdev->bd_holder		= holder;
 	bdev->bd_disk		= &bdev->__bd_disk;
-	bdev->bd_bdi		= &bdev->__bd_bdi;
-	bdev->queue.backing_dev_info = bdev->bd_bdi;
+	bdev->bd_disk->bdi	= &bdev->bd_disk->__bdi;
+	bdev->queue.backing_dev_info = bdev->bd_disk->bdi;
 
-	return bdev;
-}
+	struct bdev_handle *handle = calloc(sizeof(*handle), 1);
+	handle->bdev	= bdev;
+	handle->holder	= holder;
+	handle->mode	= mode;
 
-void bdput(struct block_device *bdev)
-{
-	BUG();
+	return handle;
 }
 
 int lookup_bdev(const char *path, dev_t *dev)
@@ -254,22 +251,21 @@ static void sync_cleanup(void)
 static void sync_read(struct bio *bio, struct iovec * iov, unsigned i)
 {
 
-	int fd = bio->bi_opf & REQ_FUA
-			? bio->bi_bdev->bd_sync_fd
-			: bio->bi_bdev->bd_fd;
-	ssize_t ret = preadv(fd, iov, i, bio->bi_iter.bi_sector << 9);
+	ssize_t ret = preadv(bio->bi_bdev->bd_fd, iov, i,
+			     bio->bi_iter.bi_sector << 9);
 	sync_check(bio, ret);
 }
 
 static void sync_write(struct bio *bio, struct iovec * iov, unsigned i)
 {
-	int fd = bio->bi_opf & REQ_FUA
-			? bio->bi_bdev->bd_sync_fd
-			: bio->bi_bdev->bd_fd;
-	ssize_t ret = pwritev(fd, iov, i, bio->bi_iter.bi_sector << 9);
+	ssize_t ret = pwritev2(bio->bi_bdev->bd_fd, iov, i,
+			       bio->bi_iter.bi_sector << 9,
+			       bio->bi_opf & REQ_FUA ? RWF_SYNC : 0);
 	sync_check(bio, ret);
 }
 
+static DECLARE_WAIT_QUEUE_HEAD(aio_events_completed);
+
 static int aio_completion_thread(void *arg)
 {
 	struct io_event events[8], *ev;
@@ -284,6 +280,8 @@ static int aio_completion_thread(void *arg)
 			continue;
 		if (ret < 0)
 			die("io_getevents() error: %s", strerror(-ret));
+		if (ret)
+			wake_up(&aio_events_completed);
 
 		for (ev = events; ev < events + ret; ev++) {
 			struct bio *bio = (struct bio *) ev->data;
@@ -364,18 +362,20 @@ static void aio_op(struct bio *bio, struct iovec *iov, unsigned i, int opcode)
 	ssize_t ret;
 	struct iocb iocb = {
 		.data		= bio,
-		.aio_fildes	= bio->bi_opf & REQ_FUA
-			? bio->bi_bdev->bd_sync_fd
-			: bio->bi_bdev->bd_fd,
+		.aio_fildes	= bio->bi_bdev->bd_fd,
+		.aio_rw_flags	= bio->bi_opf & REQ_FUA ? RWF_SYNC : 0,
 		.aio_lio_opcode	= opcode,
-		.u.v.vec	= iov,
-		.u.v.nr		= i,
-		.u.v.offset	= bio->bi_iter.bi_sector << 9,
+		.u.c.buf        = iov,
+		.u.c.nbytes     = i,
+		.u.c.offset     = bio->bi_iter.bi_sector << 9,
 
 	}, *iocbp = &iocb;
 
 	atomic_inc(&running_requests);
-	ret = io_submit(aio_ctx, 1, &iocbp);
+
+	wait_event(aio_events_completed,
+		   (ret = io_submit(aio_ctx, 1, &iocbp)) != -EAGAIN);;
+
 	if (ret != 1)
 		die("io_submit err: %s", strerror(-ret));
 }