bdev: allow allocating custom-length buffers for bdev_io

This is required for upcoming UNMAP
implementation in bdev_virtio.

While here, also added documentation for
spdk_bdev_io_get_buf().

Change-Id: Ia769ee9b8b132f31208ae66598b29a1c9ed37312
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/379721
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>

include/spdk_internal/bdev.h

@@ -251,6 +251,9 @@ struct spdk_bdev_io {
 	/** bdev allocated memory associated with this request */
 	void *buf;
 
+	/** requested size of the buffer associated with this I/O */
+	uint64_t buf_len;
+
 	/** Callback for when buf is allocated */
 	spdk_bdev_io_get_buf_cb get_buf_cb;
 
@@ -367,7 +370,20 @@ void spdk_bdev_poller_start(struct spdk_bdev_poller **ppoller,
 
 void spdk_bdev_poller_stop(struct spdk_bdev_poller **ppoller);
 
-void spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb);
+/**
+ * Allocate a buffer for given bdev_io.  Allocation will happen
+ * only if the bdev_io has no assigned SGL yet. The buffer will be
+ * freed automatically on \c spdk_bdev_free_io() call. This call
+ * will never fail - in case of lack of memory given callback \c cb
+ * will be deferred until enough memory is freed.
+ *
+ * \param bdev_io I/O to allocate buffer for.
+ * \param cb callback to be called when the buffer is allocated
+ * or the bdev_io has an SGL assigned already.
+ * \param len size of the buffer to allocate.
+ */
+void spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len);
+
 struct spdk_bdev_io *spdk_bdev_get_io(void);
 void spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io,
 			   enum spdk_bdev_io_status status);

lib/bdev/aio/bdev_aio.c

@@ -244,7 +244,8 @@ static int _bdev_aio_submit_request(struct spdk_io_channel *ch, struct spdk_bdev
 {
 	switch (bdev_io->type) {
 	case SPDK_BDEV_IO_TYPE_READ:
-		spdk_bdev_io_get_buf(bdev_io, bdev_aio_get_buf_cb);
+		spdk_bdev_io_get_buf(bdev_io, bdev_aio_get_buf_cb,
+				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
 		return 0;
 
 	case SPDK_BDEV_IO_TYPE_WRITE:

lib/bdev/bdev.c

@@ -230,7 +230,7 @@ spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf)
 
 	bdev_io->buf = buf;
 	bdev_io->u.bdev.iovs[0].iov_base = (void *)((unsigned long)((char *)buf + 512) & ~511UL);
-	bdev_io->u.bdev.iovs[0].iov_len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
+	bdev_io->u.bdev.iovs[0].iov_len = bdev_io->buf_len;
 	bdev_io->get_buf_cb(bdev_io->ch->channel, bdev_io);
 }
 
@@ -241,17 +241,14 @@ spdk_bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
 	struct spdk_bdev_io *tmp;
 	void *buf;
 	bdev_io_tailq_t *tailq;
-	uint64_t length;
 	struct spdk_bdev_mgmt_channel *ch;
 
 	assert(bdev_io->u.bdev.iovcnt == 1);
 
-	length = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
 	buf = bdev_io->buf;
-
 	ch = spdk_io_channel_get_ctx(bdev_io->ch->mgmt_channel);
 
-	if (length <= SPDK_BDEV_SMALL_BUF_MAX_SIZE) {
+	if (bdev_io->buf_len <= SPDK_BDEV_SMALL_BUF_MAX_SIZE) {
 		pool = g_bdev_mgr.buf_small_pool;
 		tailq = &ch->need_buf_small;
 	} else {
@@ -269,9 +266,8 @@ spdk_bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
 }
 
 void
-spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb)
+spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
 {
-	uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
 	struct spdk_mempool *pool;
 	bdev_io_tailq_t *tailq;
 	void *buf = NULL;
@@ -288,6 +284,7 @@ spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb)
 
 	ch = spdk_io_channel_get_ctx(bdev_io->ch->mgmt_channel);
 
+	bdev_io->buf_len = len;
 	bdev_io->get_buf_cb = cb;
 	if (len <= SPDK_BDEV_SMALL_BUF_MAX_SIZE) {
 		pool = g_bdev_mgr.buf_small_pool;

lib/bdev/lvol/vbdev_lvol.c

@@ -386,7 +386,8 @@ vbdev_lvol_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_
 
 	switch (bdev_io->type) {
 	case SPDK_BDEV_IO_TYPE_READ:
-		spdk_bdev_io_get_buf(bdev_io, lvol_read);
+		spdk_bdev_io_get_buf(bdev_io, lvol_read,
+				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
 		break;
 	case SPDK_BDEV_IO_TYPE_WRITE:
 		lvol_write(lvol, ch, bdev_io);

lib/bdev/nvme/bdev_nvme.c

@@ -363,7 +363,8 @@ _bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_
 
 	switch (bdev_io->type) {
 	case SPDK_BDEV_IO_TYPE_READ:
-		spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb);
+		spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
+				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
 		return 0;
 
 	case SPDK_BDEV_IO_TYPE_WRITE:

lib/bdev/rbd/bdev_rbd.c

@@ -272,7 +272,8 @@ static int _bdev_rbd_submit_request(struct spdk_io_channel *ch, struct spdk_bdev
 {
 	switch (bdev_io->type) {
 	case SPDK_BDEV_IO_TYPE_READ:
-		spdk_bdev_io_get_buf(bdev_io, bdev_rbd_get_buf_cb);
+		spdk_bdev_io_get_buf(bdev_io, bdev_rbd_get_buf_cb,
+				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
 		return 0;
 
 	case SPDK_BDEV_IO_TYPE_WRITE:

lib/bdev/virtio/bdev_virtio.c

@@ -155,7 +155,8 @@ static int _bdev_virtio_submit_request(struct spdk_io_channel *ch, struct spdk_b
 {
 	switch (bdev_io->type) {
 	case SPDK_BDEV_IO_TYPE_READ:
-		spdk_bdev_io_get_buf(bdev_io, bdev_virtio_rw);
+		spdk_bdev_io_get_buf(bdev_io, bdev_virtio_rw,
+				     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
 		return 0;
 	case SPDK_BDEV_IO_TYPE_WRITE:
 		bdev_virtio_rw(ch, bdev_io);

test/unit/lib/bdev/vbdev_lvol.c/vbdev_lvol_ut.c

@@ -198,7 +198,7 @@ struct spdk_io_channel *spdk_lvol_get_io_channel(struct spdk_lvol *lvol)
 }
 
 void
-spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb)
+spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
 {
 }