diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-27 18:24:20 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-27 18:24:20 +0000 |
commit | 483eb2f56657e8e7f419ab1a4fab8dce9ade8609 (patch) | |
tree | e5d88d25d870d5dedacb6bbdbe2a966086a0a5cf /src/spdk/lib/bdev/virtio | |
parent | Initial commit. (diff) | |
download | ceph-upstream.tar.xz ceph-upstream.zip |
Adding upstream version 14.2.21.upstream/14.2.21upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/spdk/lib/bdev/virtio')
-rw-r--r-- | src/spdk/lib/bdev/virtio/Makefile | 40 | ||||
-rw-r--r-- | src/spdk/lib/bdev/virtio/bdev_virtio.h | 164 | ||||
-rw-r--r-- | src/spdk/lib/bdev/virtio/bdev_virtio_blk.c | 707 | ||||
-rw-r--r-- | src/spdk/lib/bdev/virtio/bdev_virtio_rpc.c | 613 | ||||
-rw-r--r-- | src/spdk/lib/bdev/virtio/bdev_virtio_scsi.c | 2017 |
5 files changed, 3541 insertions, 0 deletions
diff --git a/src/spdk/lib/bdev/virtio/Makefile b/src/spdk/lib/bdev/virtio/Makefile new file mode 100644 index 00000000..fabe2b9f --- /dev/null +++ b/src/spdk/lib/bdev/virtio/Makefile @@ -0,0 +1,40 @@ +# +# BSD LICENSE +# +# Copyright (c) Intel Corporation. +# All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# * Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# * Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in +# the documentation and/or other materials provided with the +# distribution. +# * Neither the name of Intel Corporation nor the names of its +# contributors may be used to endorse or promote products derived +# from this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# + +SPDK_ROOT_DIR := $(abspath $(CURDIR)/../../..) +include $(SPDK_ROOT_DIR)/mk/spdk.common.mk + +C_SRCS = bdev_virtio_scsi.c bdev_virtio_blk.c bdev_virtio_rpc.c +LIBNAME = bdev_virtio + +include $(SPDK_ROOT_DIR)/mk/spdk.lib.mk diff --git a/src/spdk/lib/bdev/virtio/bdev_virtio.h b/src/spdk/lib/bdev/virtio/bdev_virtio.h new file mode 100644 index 00000000..538fab8f --- /dev/null +++ b/src/spdk/lib/bdev/virtio/bdev_virtio.h @@ -0,0 +1,164 @@ +/*- + * BSD LICENSE + * + * Copyright (c) Intel Corporation. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef SPDK_BDEV_VIRTIO_H +#define SPDK_BDEV_VIRTIO_H + +#include "spdk/bdev.h" +#include "spdk/env.h" + +/** + * Callback for creating virtio bdevs. + * + * \param ctx opaque context set by the user + * \param errnum error code. 0 on success, negative errno on error. + * \param bdevs contiguous array of created bdevs + * \param bdev_cnt number of bdevs in the `bdevs` array + */ +typedef void (*bdev_virtio_create_cb)(void *ctx, int errnum, + struct spdk_bdev **bdevs, size_t bdev_cnt); + +/** + * Callback for removing virtio devices. + * + * \param ctx opaque context set by the user + * \param errnum error code. 0 on success, negative errno on error. + */ +typedef void (*bdev_virtio_remove_cb)(void *ctx, int errnum); + +/** + * Connect to a vhost-user Unix domain socket and create a Virtio SCSI device. + * If the connection is successful, the device will be automatically scanned. + * The scan consists of probing the targets on the device and will result in + * creating possibly multiple Virtio SCSI bdevs - one for each target. Currently + * only one LUN per target is detected - LUN0. Note that the bdev creation is + * run asynchronously in the background. After it's finished, the `cb_fn` + * callback is called. + * + * \param name name for the virtio device. It will be inherited by all created + * bdevs, which are named in the following format: <name>t<target_id> + * \param path path to the socket + * \param num_queues max number of request virtqueues to use. `vdev` will be + * started successfully even if the host device supports less queues than requested. + * \param queue_size depth of each queue + * \param cb_fn function to be called after scanning all targets on the virtio + * device. It's optional, can be NULL. See \c bdev_virtio_create_cb. + * \param cb_arg argument for the `cb_fn` + * \return zero on success (device scan is started) or negative error code. + * In case of error the \c cb_fn is not called. + */ +int bdev_virtio_user_scsi_dev_create(const char *name, const char *path, + unsigned num_queues, unsigned queue_size, + bdev_virtio_create_cb cb_fn, void *cb_arg); + +/** + * Attach virtio-pci device. This creates a Virtio SCSI device with the same + * capabilities as the vhost-user equivalent. The device will be automatically + * scanned for exposed SCSI targets. This will result in creating possibly multiple + * Virtio SCSI bdevs - one for each target. Currently only one LUN per target is + * detected - LUN0. Note that the bdev creation is run asynchronously in the + * background. After it's finished, the `cb_fn` callback is called. + * + * \param name name for the virtio device. It will be inherited by all created + * bdevs, which are named in the following format: <name>t<target_id> + * \param pci_addr PCI address of the device to attach + * \param cb_fn function to be called after scanning all targets on the virtio + * device. It's optional, can be NULL. See \c bdev_virtio_create_cb. + * \param cb_arg argument for the `cb_fn` + * \return zero on success (device scan is started) or negative error code. + * In case of error the \c cb_fn is not called. + */ +int bdev_virtio_pci_scsi_dev_create(const char *name, struct spdk_pci_addr *pci_addr, + bdev_virtio_create_cb cb_fn, void *cb_arg); + +/** + * Remove a Virtio device with given name. This will destroy all bdevs exposed + * by this device. + * + * \param name virtio device name + * \param cb_fn function to be called after scanning all targets on the virtio + * device. It's optional, can be NULL. See \c bdev_virtio_create_cb. Possible + * error codes are: + * * ENODEV - couldn't find device with given name + * * EBUSY - device is already being removed + * \param cb_arg argument for the `cb_fn` + * \return zero on success or -ENODEV if scsi dev does not exist + */ +int bdev_virtio_scsi_dev_remove(const char *name, + bdev_virtio_remove_cb cb_fn, void *cb_arg); + +/** + * Remove a Virtio device with given name. + * + * \param bdev virtio blk device bdev + * \param cb_fn function to be called after removing bdev + * \param cb_arg argument for the `cb_fn` + * \return zero on success, -ENODEV if bdev with 'name' does not exist or + * -EINVAL if bdev with 'name' is not a virtio blk device. + */ +int bdev_virtio_blk_dev_remove(const char *name, + bdev_virtio_remove_cb cb_fn, void *cb_arg); + +/** + * List all created Virtio-SCSI devices. + * + * \param write_ctx JSON context to write into + */ +void bdev_virtio_scsi_dev_list(struct spdk_json_write_ctx *write_ctx); + +/** + * Connect to a vhost-user Unix domain socket and create a Virtio BLK bdev. + * + * \param name name for the virtio bdev + * \param path path to the socket + * \param num_queues max number of request virtqueues to use. `vdev` will be + * started successfully even if the host device supports less queues than requested. + * \param queue_size depth of each queue + * \return virtio-blk bdev or NULL + */ +struct spdk_bdev *bdev_virtio_user_blk_dev_create(const char *name, const char *path, + unsigned num_queues, unsigned queue_size); + +/** + * Attach virtio-pci device. This creates a Virtio BLK device with the same + * capabilities as the vhost-user equivalent. + * + * \param name name for the virtio device. It will be inherited by all created + * bdevs, which are named in the following format: <name>t<target_id> + * \param pci_addr PCI address of the device to attach + * \return virtio-blk bdev or NULL + */ +struct spdk_bdev *bdev_virtio_pci_blk_dev_create(const char *name, + struct spdk_pci_addr *pci_addr); + +#endif /* SPDK_BDEV_VIRTIO_H */ diff --git a/src/spdk/lib/bdev/virtio/bdev_virtio_blk.c b/src/spdk/lib/bdev/virtio/bdev_virtio_blk.c new file mode 100644 index 00000000..598f7f15 --- /dev/null +++ b/src/spdk/lib/bdev/virtio/bdev_virtio_blk.c @@ -0,0 +1,707 @@ +/*- + * BSD LICENSE + * + * Copyright (c) Intel Corporation. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "spdk/stdinc.h" + +#include "spdk/bdev.h" +#include "spdk/conf.h" +#include "spdk/endian.h" +#include "spdk/env.h" +#include "spdk/thread.h" +#include "spdk/string.h" +#include "spdk/util.h" +#include "spdk/json.h" + +#include "spdk_internal/assert.h" +#include "spdk/bdev_module.h" +#include "spdk_internal/log.h" +#include "spdk_internal/virtio.h" + +#include <linux/virtio_blk.h> + +#include "bdev_virtio.h" + +struct virtio_blk_dev { + struct virtio_dev vdev; + struct spdk_bdev bdev; + bool readonly; +}; + +struct virtio_blk_io_ctx { + struct iovec iov_req; + struct iovec iov_resp; + struct virtio_blk_outhdr req; + uint8_t resp; +}; + +struct bdev_virtio_blk_io_channel { + struct virtio_dev *vdev; + + /** Virtqueue exclusively assigned to this channel. */ + struct virtqueue *vq; + + /** Virtio response poller. */ + struct spdk_poller *poller; +}; + +/* Features desired/implemented by this driver. */ +#define VIRTIO_BLK_DEV_SUPPORTED_FEATURES \ + (1ULL << VIRTIO_BLK_F_BLK_SIZE | \ + 1ULL << VIRTIO_BLK_F_TOPOLOGY | \ + 1ULL << VIRTIO_BLK_F_MQ | \ + 1ULL << VIRTIO_BLK_F_RO | \ + 1ULL << VIRTIO_RING_F_EVENT_IDX | \ + 1ULL << VHOST_USER_F_PROTOCOL_FEATURES) + +static int bdev_virtio_initialize(void); +static int bdev_virtio_blk_get_ctx_size(void); + +static struct spdk_bdev_module virtio_blk_if = { + .name = "virtio_blk", + .module_init = bdev_virtio_initialize, + .get_ctx_size = bdev_virtio_blk_get_ctx_size, +}; + +SPDK_BDEV_MODULE_REGISTER(&virtio_blk_if) + +static int bdev_virtio_blk_ch_create_cb(void *io_device, void *ctx_buf); +static void bdev_virtio_blk_ch_destroy_cb(void *io_device, void *ctx_buf); + +static struct virtio_blk_io_ctx * +bdev_virtio_blk_init_io_vreq(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_blk_outhdr *req; + uint8_t *resp; + struct virtio_blk_io_ctx *io_ctx = (struct virtio_blk_io_ctx *)bdev_io->driver_ctx; + + req = &io_ctx->req; + resp = &io_ctx->resp; + + io_ctx->iov_req.iov_base = req; + io_ctx->iov_req.iov_len = sizeof(*req); + + io_ctx->iov_resp.iov_base = resp; + io_ctx->iov_resp.iov_len = sizeof(*resp); + + memset(req, 0, sizeof(*req)); + return io_ctx; +} + +static void +bdev_virtio_blk_send_io(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct bdev_virtio_blk_io_channel *virtio_channel = spdk_io_channel_get_ctx(ch); + struct virtqueue *vq = virtio_channel->vq; + struct virtio_blk_io_ctx *io_ctx = (struct virtio_blk_io_ctx *)bdev_io->driver_ctx; + int rc; + + rc = virtqueue_req_start(vq, bdev_io, bdev_io->u.bdev.iovcnt + 2); + if (rc == -ENOMEM) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); + return; + } else if (rc != 0) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + return; + } + + virtqueue_req_add_iovs(vq, &io_ctx->iov_req, 1, SPDK_VIRTIO_DESC_RO); + virtqueue_req_add_iovs(vq, bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, + bdev_io->type == SPDK_BDEV_IO_TYPE_READ ? + SPDK_VIRTIO_DESC_WR : SPDK_VIRTIO_DESC_RO); + virtqueue_req_add_iovs(vq, &io_ctx->iov_resp, 1, SPDK_VIRTIO_DESC_WR); + + virtqueue_req_flush(vq); +} + +static void +bdev_virtio_rw(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_blk_io_ctx *io_ctx = bdev_virtio_blk_init_io_vreq(ch, bdev_io); + struct virtio_blk_outhdr *req = &io_ctx->req; + + req->type = bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ? + VIRTIO_BLK_T_OUT : VIRTIO_BLK_T_IN; + + req->sector = bdev_io->u.bdev.offset_blocks * + spdk_bdev_get_block_size(bdev_io->bdev) / 512; + + bdev_virtio_blk_send_io(ch, bdev_io); +} + +static int +_bdev_virtio_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_blk_dev *bvdev = bdev_io->bdev->ctxt; + + switch (bdev_io->type) { + case SPDK_BDEV_IO_TYPE_READ: + 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: + if (bvdev->readonly) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + } else { + bdev_virtio_rw(ch, bdev_io); + } + return 0; + case SPDK_BDEV_IO_TYPE_RESET: + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS); + return 0; + case SPDK_BDEV_IO_TYPE_UNMAP: + case SPDK_BDEV_IO_TYPE_FLUSH: + default: + return -1; + } + + SPDK_UNREACHABLE(); +} + +static void +bdev_virtio_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + if (_bdev_virtio_submit_request(ch, bdev_io) < 0) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + } +} + +static bool +bdev_virtio_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type) +{ + struct virtio_blk_dev *bvdev = ctx; + + switch (io_type) { + case SPDK_BDEV_IO_TYPE_READ: + case SPDK_BDEV_IO_TYPE_RESET: + return true; + case SPDK_BDEV_IO_TYPE_WRITE: + return !bvdev->readonly; + case SPDK_BDEV_IO_TYPE_FLUSH: + case SPDK_BDEV_IO_TYPE_UNMAP: + default: + return false; + } +} + +static struct spdk_io_channel * +bdev_virtio_get_io_channel(void *ctx) +{ + struct virtio_blk_dev *bvdev = ctx; + + return spdk_get_io_channel(bvdev); +} + +static void +virtio_blk_dev_unregister_cb(void *io_device) +{ + struct virtio_blk_dev *bvdev = io_device; + struct virtio_dev *vdev = &bvdev->vdev; + + virtio_dev_stop(vdev); + virtio_dev_destruct(vdev); + spdk_bdev_destruct_done(&bvdev->bdev, 0); + free(bvdev); +} + +static int +bdev_virtio_disk_destruct(void *ctx) +{ + struct virtio_blk_dev *bvdev = ctx; + + spdk_io_device_unregister(bvdev, virtio_blk_dev_unregister_cb); + return 1; +} + +int +bdev_virtio_blk_dev_remove(const char *name, bdev_virtio_remove_cb cb_fn, void *cb_arg) +{ + struct spdk_bdev *bdev; + + bdev = spdk_bdev_get_by_name(name); + if (bdev == NULL) { + return -ENODEV; + } + + if (bdev->module != &virtio_blk_if) { + return -ENODEV; + } + + spdk_bdev_unregister(bdev, cb_fn, cb_arg); + + return 0; +} + +static int +bdev_virtio_dump_json_config(void *ctx, struct spdk_json_write_ctx *w) +{ + struct virtio_blk_dev *bvdev = ctx; + + virtio_dev_dump_json_info(&bvdev->vdev, w); + return 0; +} + +static void +bdev_virtio_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w) +{ + struct virtio_blk_dev *bvdev = bdev->ctxt; + + spdk_json_write_object_begin(w); + + spdk_json_write_named_string(w, "method", "construct_virtio_dev"); + + spdk_json_write_named_object_begin(w, "params"); + spdk_json_write_named_string(w, "name", bvdev->vdev.name); + spdk_json_write_named_string(w, "dev_type", "blk"); + + /* Write transport specific parameters. */ + bvdev->vdev.backend_ops->write_json_config(&bvdev->vdev, w); + + spdk_json_write_object_end(w); + + spdk_json_write_object_end(w); +} + +static const struct spdk_bdev_fn_table virtio_fn_table = { + .destruct = bdev_virtio_disk_destruct, + .submit_request = bdev_virtio_submit_request, + .io_type_supported = bdev_virtio_io_type_supported, + .get_io_channel = bdev_virtio_get_io_channel, + .dump_info_json = bdev_virtio_dump_json_config, + .write_config_json = bdev_virtio_write_config_json, +}; + +static void +bdev_virtio_io_cpl(struct spdk_bdev_io *bdev_io) +{ + struct virtio_blk_io_ctx *io_ctx = (struct virtio_blk_io_ctx *)bdev_io->driver_ctx; + + spdk_bdev_io_complete(bdev_io, io_ctx->resp == VIRTIO_BLK_S_OK ? + SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED); +} + +static int +bdev_virtio_poll(void *arg) +{ + struct bdev_virtio_blk_io_channel *ch = arg; + void *io[32]; + uint32_t io_len[32]; + uint16_t i, cnt; + + cnt = virtio_recv_pkts(ch->vq, io, io_len, SPDK_COUNTOF(io)); + for (i = 0; i < cnt; ++i) { + bdev_virtio_io_cpl(io[i]); + } + + return cnt; +} + +static int +bdev_virtio_blk_ch_create_cb(void *io_device, void *ctx_buf) +{ + struct virtio_blk_dev *bvdev = io_device; + struct virtio_dev *vdev = &bvdev->vdev; + struct bdev_virtio_blk_io_channel *ch = ctx_buf; + struct virtqueue *vq; + int32_t queue_idx; + + queue_idx = virtio_dev_find_and_acquire_queue(vdev, 0); + if (queue_idx < 0) { + SPDK_ERRLOG("Couldn't get an unused queue for the io_channel.\n"); + return -1; + } + + vq = vdev->vqs[queue_idx]; + + ch->vdev = vdev; + ch->vq = vq; + + ch->poller = spdk_poller_register(bdev_virtio_poll, ch, 0); + return 0; +} + +static void +bdev_virtio_blk_ch_destroy_cb(void *io_device, void *ctx_buf) +{ + struct virtio_blk_dev *bvdev = io_device; + struct virtio_dev *vdev = &bvdev->vdev; + struct bdev_virtio_blk_io_channel *ch = ctx_buf; + struct virtqueue *vq = ch->vq; + + spdk_poller_unregister(&ch->poller); + virtio_dev_release_queue(vdev, vq->vq_queue_index); +} + +static int +virtio_blk_dev_init(struct virtio_blk_dev *bvdev, uint16_t max_queues) +{ + struct virtio_dev *vdev = &bvdev->vdev; + struct spdk_bdev *bdev = &bvdev->bdev; + uint64_t capacity, num_blocks; + uint32_t block_size; + uint16_t host_max_queues; + int rc; + + if (virtio_dev_has_feature(vdev, VIRTIO_BLK_F_BLK_SIZE)) { + rc = virtio_dev_read_dev_config(vdev, offsetof(struct virtio_blk_config, blk_size), + &block_size, sizeof(block_size)); + if (rc) { + SPDK_ERRLOG("%s: config read failed: %s\n", vdev->name, spdk_strerror(-rc)); + return rc; + } + + if (block_size == 0 || block_size % 512 != 0) { + SPDK_ERRLOG("%s: invalid block size (%"PRIu32"). Must be " + "a multiple of 512.\n", vdev->name, block_size); + return -EIO; + } + } else { + block_size = 512; + } + + rc = virtio_dev_read_dev_config(vdev, offsetof(struct virtio_blk_config, capacity), + &capacity, sizeof(capacity)); + if (rc) { + SPDK_ERRLOG("%s: config read failed: %s\n", vdev->name, spdk_strerror(-rc)); + return rc; + } + + /* `capacity` is a number of 512-byte sectors. */ + num_blocks = capacity * 512 / block_size; + if (num_blocks == 0) { + SPDK_ERRLOG("%s: size too small (size: %"PRIu64", blocksize: %"PRIu32").\n", + vdev->name, capacity * 512, block_size); + return -EIO; + } + + if ((capacity * 512) % block_size != 0) { + SPDK_WARNLOG("%s: size has been rounded down to the nearest block size boundary. " + "(block size: %"PRIu32", previous size: %"PRIu64", new size: %"PRIu64")\n", + vdev->name, block_size, capacity * 512, num_blocks * block_size); + } + + if (virtio_dev_has_feature(vdev, VIRTIO_BLK_F_MQ)) { + rc = virtio_dev_read_dev_config(vdev, offsetof(struct virtio_blk_config, num_queues), + &host_max_queues, sizeof(host_max_queues)); + if (rc) { + SPDK_ERRLOG("%s: config read failed: %s\n", vdev->name, spdk_strerror(-rc)); + return rc; + } + } else { + host_max_queues = 1; + } + + if (virtio_dev_has_feature(vdev, VIRTIO_BLK_F_RO)) { + bvdev->readonly = true; + } + + if (max_queues == 0) { + SPDK_ERRLOG("%s: requested 0 request queues (%"PRIu16" available).\n", + vdev->name, host_max_queues); + return -EINVAL; + } + + if (max_queues > host_max_queues) { + SPDK_WARNLOG("%s: requested %"PRIu16" request queues " + "but only %"PRIu16" available.\n", + vdev->name, max_queues, host_max_queues); + max_queues = host_max_queues; + } + + /* bdev is tied with the virtio device; we can reuse the name */ + bdev->name = vdev->name; + rc = virtio_dev_start(vdev, max_queues, 0); + if (rc != 0) { + return rc; + } + + bdev->product_name = "VirtioBlk Disk"; + bdev->write_cache = 0; + bdev->blocklen = block_size; + bdev->blockcnt = num_blocks; + + bdev->ctxt = bvdev; + bdev->fn_table = &virtio_fn_table; + bdev->module = &virtio_blk_if; + + spdk_io_device_register(bvdev, bdev_virtio_blk_ch_create_cb, + bdev_virtio_blk_ch_destroy_cb, + sizeof(struct bdev_virtio_blk_io_channel), + vdev->name); + + rc = spdk_bdev_register(bdev); + if (rc) { + SPDK_ERRLOG("Failed to register bdev name=%s\n", bdev->name); + spdk_io_device_unregister(bvdev, NULL); + virtio_dev_stop(vdev); + return rc; + } + + return 0; +} + +static struct virtio_blk_dev * +virtio_pci_blk_dev_create(const char *name, struct virtio_pci_ctx *pci_ctx) +{ + static int pci_dev_counter = 0; + struct virtio_blk_dev *bvdev; + struct virtio_dev *vdev; + char *default_name = NULL; + uint16_t num_queues; + int rc; + + bvdev = calloc(1, sizeof(*bvdev)); + if (bvdev == NULL) { + SPDK_ERRLOG("virtio device calloc failed\n"); + return NULL; + } + vdev = &bvdev->vdev; + + if (name == NULL) { + default_name = spdk_sprintf_alloc("VirtioBlk%"PRIu32, pci_dev_counter++); + if (default_name == NULL) { + free(vdev); + return NULL; + } + name = default_name; + } + + rc = virtio_pci_dev_init(vdev, name, pci_ctx); + free(default_name); + + if (rc != 0) { + free(bvdev); + return NULL; + } + + rc = virtio_dev_reset(vdev, VIRTIO_BLK_DEV_SUPPORTED_FEATURES); + if (rc != 0) { + virtio_dev_destruct(vdev); + free(bvdev); + return NULL; + } + + /* TODO: add a way to limit usable virtqueues */ + if (virtio_dev_has_feature(vdev, VIRTIO_BLK_F_MQ)) { + rc = virtio_dev_read_dev_config(vdev, offsetof(struct virtio_blk_config, num_queues), + &num_queues, sizeof(num_queues)); + if (rc) { + SPDK_ERRLOG("%s: config read failed: %s\n", vdev->name, spdk_strerror(-rc)); + virtio_dev_destruct(vdev); + free(bvdev); + return NULL; + } + } else { + num_queues = 1; + } + + rc = virtio_blk_dev_init(bvdev, num_queues); + if (rc != 0) { + virtio_dev_destruct(vdev); + free(bvdev); + return NULL; + } + + return bvdev; +} + +static struct virtio_blk_dev * +virtio_user_blk_dev_create(const char *name, const char *path, + uint16_t num_queues, uint32_t queue_size) +{ + struct virtio_blk_dev *bvdev; + int rc; + + bvdev = calloc(1, sizeof(*bvdev)); + if (bvdev == NULL) { + SPDK_ERRLOG("calloc failed for virtio device %s: %s\n", name, path); + return NULL; + } + + rc = virtio_user_dev_init(&bvdev->vdev, name, path, queue_size); + if (rc != 0) { + SPDK_ERRLOG("Failed to create virito device %s: %s\n", name, path); + free(bvdev); + return NULL; + } + + rc = virtio_dev_reset(&bvdev->vdev, VIRTIO_BLK_DEV_SUPPORTED_FEATURES); + if (rc != 0) { + virtio_dev_destruct(&bvdev->vdev); + free(bvdev); + return NULL; + } + + rc = virtio_blk_dev_init(bvdev, num_queues); + if (rc != 0) { + virtio_dev_destruct(&bvdev->vdev); + free(bvdev); + return NULL; + } + + return bvdev; +} + +struct bdev_virtio_pci_dev_create_ctx { + const char *name; + struct virtio_blk_dev *ret; +}; + +static int +bdev_virtio_pci_blk_dev_create_cb(struct virtio_pci_ctx *pci_ctx, void *ctx) +{ + struct bdev_virtio_pci_dev_create_ctx *create_ctx = ctx; + + create_ctx->ret = virtio_pci_blk_dev_create(create_ctx->name, pci_ctx); + if (create_ctx->ret == NULL) { + return -1; + } + + return 0; +} + +struct spdk_bdev * +bdev_virtio_pci_blk_dev_create(const char *name, struct spdk_pci_addr *pci_addr) +{ + struct bdev_virtio_pci_dev_create_ctx create_ctx; + + create_ctx.name = name; + create_ctx.ret = NULL; + + virtio_pci_dev_attach(bdev_virtio_pci_blk_dev_create_cb, &create_ctx, + PCI_DEVICE_ID_VIRTIO_BLK_MODERN, pci_addr); + + if (create_ctx.ret == NULL) { + return NULL; + } + + return &create_ctx.ret->bdev; +} + +static int +virtio_pci_blk_dev_enumerate_cb(struct virtio_pci_ctx *pci_ctx, void *ctx) +{ + struct virtio_blk_dev *bvdev; + + bvdev = virtio_pci_blk_dev_create(NULL, pci_ctx); + return bvdev == NULL ? -1 : 0; +} + +static int +bdev_virtio_initialize(void) +{ + struct spdk_conf_section *sp; + struct virtio_blk_dev *bvdev; + char *default_name = NULL; + char *path, *type, *name; + unsigned vdev_num; + int num_queues; + bool enable_pci; + int rc = 0; + + for (sp = spdk_conf_first_section(NULL); sp != NULL; sp = spdk_conf_next_section(sp)) { + if (!spdk_conf_section_match_prefix(sp, "VirtioUser")) { + continue; + } + + if (sscanf(spdk_conf_section_get_name(sp), "VirtioUser%u", &vdev_num) != 1) { + SPDK_ERRLOG("Section '%s' has non-numeric suffix.\n", + spdk_conf_section_get_name(sp)); + return -1; + } + + path = spdk_conf_section_get_val(sp, "Path"); + if (path == NULL) { + SPDK_ERRLOG("VirtioUserBlk%u: missing Path\n", vdev_num); + return -1; + } + + type = spdk_conf_section_get_val(sp, "Type"); + if (type == NULL || strcmp(type, "Blk") != 0) { + continue; + } + + num_queues = spdk_conf_section_get_intval(sp, "Queues"); + if (num_queues < 1) { + num_queues = 1; + } + + name = spdk_conf_section_get_val(sp, "Name"); + if (name == NULL) { + default_name = spdk_sprintf_alloc("VirtioBlk%u", vdev_num); + name = default_name; + } + + bvdev = virtio_user_blk_dev_create(name, path, num_queues, 512); + free(default_name); + default_name = NULL; + + if (bvdev == NULL) { + return -1; + } + } + + sp = spdk_conf_find_section(NULL, "VirtioPci"); + if (sp == NULL) { + return 0; + } + + enable_pci = spdk_conf_section_get_boolval(sp, "Enable", false); + if (enable_pci) { + rc = virtio_pci_dev_enumerate(virtio_pci_blk_dev_enumerate_cb, NULL, + PCI_DEVICE_ID_VIRTIO_BLK_MODERN); + } + + return rc; +} + +struct spdk_bdev * +bdev_virtio_user_blk_dev_create(const char *name, const char *path, + unsigned num_queues, unsigned queue_size) +{ + struct virtio_blk_dev *bvdev; + + bvdev = virtio_user_blk_dev_create(name, path, num_queues, queue_size); + if (bvdev == NULL) { + return NULL; + } + + return &bvdev->bdev; +} + +static int +bdev_virtio_blk_get_ctx_size(void) +{ + return sizeof(struct virtio_blk_io_ctx); +} + +SPDK_LOG_REGISTER_COMPONENT("virtio_blk", SPDK_LOG_VIRTIO_BLK) diff --git a/src/spdk/lib/bdev/virtio/bdev_virtio_rpc.c b/src/spdk/lib/bdev/virtio/bdev_virtio_rpc.c new file mode 100644 index 00000000..e96fb42a --- /dev/null +++ b/src/spdk/lib/bdev/virtio/bdev_virtio_rpc.c @@ -0,0 +1,613 @@ +/*- + * BSD LICENSE + * + * Copyright (c) Intel Corporation. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "spdk/stdinc.h" + +#include "spdk/string.h" +#include "spdk/rpc.h" +#include "spdk/util.h" +#include "spdk_internal/log.h" + +#include "bdev_virtio.h" + +#define SPDK_VIRTIO_USER_DEFAULT_VQ_COUNT 1 +#define SPDK_VIRTIO_USER_DEFAULT_QUEUE_SIZE 512 + +struct rpc_construct_virtio_scsi_dev { + char *path; + char *pci_address; + char *name; + uint32_t vq_count; + uint32_t vq_size; + struct spdk_jsonrpc_request *request; + +}; + +static const struct spdk_json_object_decoder rpc_construct_virtio_user_scsi_dev[] = { + {"path", offsetof(struct rpc_construct_virtio_scsi_dev, path), spdk_json_decode_string }, + {"name", offsetof(struct rpc_construct_virtio_scsi_dev, name), spdk_json_decode_string }, + {"vq_count", offsetof(struct rpc_construct_virtio_scsi_dev, vq_size), spdk_json_decode_uint32, true }, + {"vq_size", offsetof(struct rpc_construct_virtio_scsi_dev, vq_size), spdk_json_decode_uint32, true }, +}; + +static void +free_rpc_construct_virtio_scsi_dev(struct rpc_construct_virtio_scsi_dev *req) +{ + if (!req) { + return; + } + + free(req->path); + free(req->pci_address); + free(req->name); + free(req); +} + +static void +rpc_construct_virtio_scsi_dev_cb(void *ctx, int result, struct spdk_bdev **bdevs, size_t cnt) +{ + struct rpc_construct_virtio_scsi_dev *req = ctx; + struct spdk_json_write_ctx *w; + size_t i; + + if (result) { + spdk_jsonrpc_send_error_response(req->request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-result)); + free_rpc_construct_virtio_scsi_dev(req); + return; + } + + w = spdk_jsonrpc_begin_result(req->request); + if (w) { + spdk_json_write_array_begin(w); + + for (i = 0; i < cnt; i++) { + spdk_json_write_string(w, spdk_bdev_get_name(bdevs[i])); + } + + spdk_json_write_array_end(w); + spdk_jsonrpc_end_result(req->request, w); + } + + free_rpc_construct_virtio_scsi_dev(ctx); +} + +static void +spdk_rpc_create_virtio_user_scsi_bdev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_construct_virtio_scsi_dev *req; + int rc; + + SPDK_WARNLOG("construct_virtio_user_scsi_bdev command has been deprecated and will be removed " + "in the subsequent release. Please use construct_virtio_dev instead.\n"); + + req = calloc(1, sizeof(*req)); + if (!req) { + rc = -ENOMEM; + goto invalid; + } + + req->pci_address = NULL; + req->vq_count = SPDK_VIRTIO_USER_DEFAULT_VQ_COUNT; + req->vq_size = SPDK_VIRTIO_USER_DEFAULT_QUEUE_SIZE; + + if (spdk_json_decode_object(params, rpc_construct_virtio_user_scsi_dev, + SPDK_COUNTOF(rpc_construct_virtio_user_scsi_dev), + req)) { + rc = -EINVAL; + goto invalid; + } + + req->request = request; + rc = bdev_virtio_user_scsi_dev_create(req->name, req->path, req->vq_count, req->vq_size, + rpc_construct_virtio_scsi_dev_cb, req); + if (rc < 0) { + goto invalid; + } + + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); + free_rpc_construct_virtio_scsi_dev(req); +} +SPDK_RPC_REGISTER("construct_virtio_user_scsi_bdev", spdk_rpc_create_virtio_user_scsi_bdev, + SPDK_RPC_RUNTIME); + +static const struct spdk_json_object_decoder rpc_construct_virtio_pci_scsi_dev[] = { + {"pci_address", offsetof(struct rpc_construct_virtio_scsi_dev, pci_address), spdk_json_decode_string }, + {"name", offsetof(struct rpc_construct_virtio_scsi_dev, name), spdk_json_decode_string }, +}; + +static void +spdk_rpc_construct_virtio_pci_scsi_dev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_construct_virtio_scsi_dev *req; + struct spdk_pci_addr pci_addr; + int rc; + + SPDK_WARNLOG("construct_virtio_pci_scsi_bdev command has been deprecated and will be removed " + "in the subsequent release. Please use construct_virtio_dev instead.\n"); + + req = calloc(1, sizeof(*req)); + if (!req) { + rc = -ENOMEM; + goto invalid; + } + + req->path = NULL; + + if (spdk_json_decode_object(params, rpc_construct_virtio_pci_scsi_dev, + SPDK_COUNTOF(rpc_construct_virtio_pci_scsi_dev), + req)) { + rc = -EINVAL; + goto invalid; + } + + if (spdk_pci_addr_parse(&pci_addr, req->pci_address) != 0) { + SPDK_ERRLOG("Invalid PCI address '%s'\n", req->pci_address); + rc = -EINVAL; + goto invalid; + } + + req->request = request; + rc = bdev_virtio_pci_scsi_dev_create(req->name, &pci_addr, + rpc_construct_virtio_scsi_dev_cb, req); + if (rc < 0) { + goto invalid; + } + + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); + free_rpc_construct_virtio_scsi_dev(req); +} +SPDK_RPC_REGISTER("construct_virtio_pci_scsi_bdev", spdk_rpc_construct_virtio_pci_scsi_dev, + SPDK_RPC_RUNTIME); + +struct rpc_remove_virtio_dev { + char *name; +}; + +static const struct spdk_json_object_decoder rpc_remove_virtio_dev[] = { + {"name", offsetof(struct rpc_remove_virtio_dev, name), spdk_json_decode_string }, +}; + +static void +spdk_rpc_remove_virtio_scsi_bdev_cb(void *ctx, int errnum) +{ + struct spdk_jsonrpc_request *request = ctx; + struct spdk_json_write_ctx *w; + + if (errnum != 0) { + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-errnum)); + return; + } + + w = spdk_jsonrpc_begin_result(request); + if (w == NULL) { + return; + } + + spdk_json_write_bool(w, true); + spdk_jsonrpc_end_result(request, w); +} + +static void +spdk_rpc_remove_virtio_scsi_bdev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_remove_virtio_dev req = {NULL}; + int rc; + + SPDK_WARNLOG("remove_virtio_scsi_bdev command has been deprecated and will be removed " + "in the subsequent release. Please use remove_virtio_bdev instead.\n"); + + if (spdk_json_decode_object(params, rpc_remove_virtio_dev, + SPDK_COUNTOF(rpc_remove_virtio_dev), + &req)) { + rc = -EINVAL; + goto invalid; + } + + rc = bdev_virtio_scsi_dev_remove(req.name, spdk_rpc_remove_virtio_scsi_bdev_cb, request); + if (rc != 0) { + goto invalid; + } + + free(req.name); + + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); + free(req.name); +} +SPDK_RPC_REGISTER("remove_virtio_scsi_bdev", spdk_rpc_remove_virtio_scsi_bdev, SPDK_RPC_RUNTIME); + +static void +spdk_rpc_remove_virtio_bdev_cb(void *ctx, int errnum) +{ + struct spdk_jsonrpc_request *request = ctx; + struct spdk_json_write_ctx *w; + + if (errnum != 0) { + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-errnum)); + return; + } + + w = spdk_jsonrpc_begin_result(request); + if (w == NULL) { + return; + } + + spdk_json_write_bool(w, true); + spdk_jsonrpc_end_result(request, w); +} + +static void +spdk_rpc_remove_virtio_bdev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_remove_virtio_dev req = {NULL}; + int rc; + + if (spdk_json_decode_object(params, rpc_remove_virtio_dev, + SPDK_COUNTOF(rpc_remove_virtio_dev), + &req)) { + rc = -EINVAL; + goto invalid; + } + + rc = bdev_virtio_blk_dev_remove(req.name, spdk_rpc_remove_virtio_bdev_cb, request); + if (rc == -ENODEV) { + rc = bdev_virtio_scsi_dev_remove(req.name, spdk_rpc_remove_virtio_bdev_cb, request); + } + + if (rc != 0) { + goto invalid; + } + + free(req.name); + + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); + free(req.name); +} +SPDK_RPC_REGISTER("remove_virtio_bdev", spdk_rpc_remove_virtio_bdev, SPDK_RPC_RUNTIME); + +static void +spdk_rpc_get_virtio_scsi_devs(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct spdk_json_write_ctx *w; + + if (params != NULL) { + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + "get_virtio_scsi_devs requires no parameters"); + return; + } + + w = spdk_jsonrpc_begin_result(request); + if (w == NULL) { + return; + } + + bdev_virtio_scsi_dev_list(w); + spdk_jsonrpc_end_result(request, w); +} +SPDK_RPC_REGISTER("get_virtio_scsi_devs", spdk_rpc_get_virtio_scsi_devs, SPDK_RPC_RUNTIME) + +struct rpc_construct_virtio_blk_dev { + char *path; + char *pci_address; + char *name; + uint32_t vq_count; + uint32_t vq_size; +}; + +static void +free_rpc_construct_virtio_blk_dev(struct rpc_construct_virtio_blk_dev *req) +{ + free(req->path); + free(req->pci_address); + free(req->name); +} + +static const struct spdk_json_object_decoder rpc_construct_virtio_user_blk_dev[] = { + {"path", offsetof(struct rpc_construct_virtio_blk_dev, path), spdk_json_decode_string }, + {"name", offsetof(struct rpc_construct_virtio_blk_dev, name), spdk_json_decode_string }, + {"vq_count", offsetof(struct rpc_construct_virtio_blk_dev, vq_count), spdk_json_decode_uint32, true }, + {"vq_size", offsetof(struct rpc_construct_virtio_blk_dev, vq_size), spdk_json_decode_uint32, true }, +}; + +static void +spdk_rpc_create_virtio_user_blk_bdev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_construct_virtio_blk_dev req = {0}; + struct spdk_json_write_ctx *w; + struct spdk_bdev *bdev; + int rc; + + req.pci_address = NULL; + req.vq_count = SPDK_VIRTIO_USER_DEFAULT_VQ_COUNT; + req.vq_size = SPDK_VIRTIO_USER_DEFAULT_QUEUE_SIZE; + + SPDK_WARNLOG("construct_virtio_user_blk_bdev command has been deprecated and will be removed " + "in the subsequent release. Please use construct_virtio_dev instead.\n"); + + if (spdk_json_decode_object(params, rpc_construct_virtio_user_blk_dev, + SPDK_COUNTOF(rpc_construct_virtio_user_blk_dev), + &req)) { + free_rpc_construct_virtio_blk_dev(&req); + rc = -EINVAL; + goto invalid; + } + + bdev = bdev_virtio_user_blk_dev_create(req.name, req.path, req.vq_count, req.vq_size); + free_rpc_construct_virtio_blk_dev(&req); + if (bdev == NULL) { + rc = -EINVAL; + goto invalid; + } + + w = spdk_jsonrpc_begin_result(request); + if (w == NULL) { + return; + } + + spdk_json_write_string(w, spdk_bdev_get_name(bdev)); + spdk_jsonrpc_end_result(request, w); + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); +} +SPDK_RPC_REGISTER("construct_virtio_user_blk_bdev", spdk_rpc_create_virtio_user_blk_bdev, + SPDK_RPC_RUNTIME); + +static const struct spdk_json_object_decoder rpc_construct_virtio_pci_blk_dev[] = { + {"pci_address", offsetof(struct rpc_construct_virtio_blk_dev, pci_address), spdk_json_decode_string }, + {"name", offsetof(struct rpc_construct_virtio_blk_dev, name), spdk_json_decode_string }, +}; + +static void +spdk_rpc_create_virtio_pci_blk_bdev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_construct_virtio_blk_dev req = {0}; + struct spdk_json_write_ctx *w; + struct spdk_bdev *bdev; + struct spdk_pci_addr pci_addr; + int rc; + + req.pci_address = NULL; + + SPDK_WARNLOG("construct_virtio_pci_blk_bdev command has been deprecated and will be removed " + "in the subsequent release. Please use construct_virtio_dev instead.\n"); + + if (spdk_json_decode_object(params, rpc_construct_virtio_pci_blk_dev, + SPDK_COUNTOF(rpc_construct_virtio_pci_blk_dev), + &req)) { + free_rpc_construct_virtio_blk_dev(&req); + rc = -EINVAL; + goto invalid; + } + + if (spdk_pci_addr_parse(&pci_addr, req.pci_address) != 0) { + SPDK_ERRLOG("Invalid PCI address '%s'\n", req.pci_address); + free_rpc_construct_virtio_blk_dev(&req); + rc = -EINVAL; + goto invalid; + } + + bdev = bdev_virtio_pci_blk_dev_create(req.name, &pci_addr); + free_rpc_construct_virtio_blk_dev(&req); + if (bdev == NULL) { + rc = -EINVAL; + goto invalid; + } + + w = spdk_jsonrpc_begin_result(request); + if (w == NULL) { + return; + } + + spdk_json_write_string(w, spdk_bdev_get_name(bdev)); + spdk_jsonrpc_end_result(request, w); + return; + +invalid: + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-rc)); +} +SPDK_RPC_REGISTER("construct_virtio_pci_blk_bdev", spdk_rpc_create_virtio_pci_blk_bdev, + SPDK_RPC_RUNTIME); + +struct rpc_construct_virtio_dev { + char *name; + char *trtype; + char *traddr; + char *dev_type; + uint32_t vq_count; + uint32_t vq_size; + struct spdk_jsonrpc_request *request; +}; + +static const struct spdk_json_object_decoder rpc_construct_virtio_dev[] = { + {"name", offsetof(struct rpc_construct_virtio_dev, name), spdk_json_decode_string }, + {"trtype", offsetof(struct rpc_construct_virtio_dev, trtype), spdk_json_decode_string }, + {"traddr", offsetof(struct rpc_construct_virtio_dev, traddr), spdk_json_decode_string }, + {"dev_type", offsetof(struct rpc_construct_virtio_dev, dev_type), spdk_json_decode_string }, + {"vq_count", offsetof(struct rpc_construct_virtio_dev, vq_count), spdk_json_decode_uint32, true }, + {"vq_size", offsetof(struct rpc_construct_virtio_dev, vq_size), spdk_json_decode_uint32, true }, +}; + +static void +free_rpc_construct_virtio_dev(struct rpc_construct_virtio_dev *req) +{ + free(req->name); + free(req->trtype); + free(req->traddr); + free(req->dev_type); + free(req); +} + +static void +spdk_rpc_create_virtio_dev_cb(void *ctx, int result, struct spdk_bdev **bdevs, size_t cnt) +{ + struct rpc_construct_virtio_dev *req = ctx; + struct spdk_json_write_ctx *w; + size_t i; + + if (result) { + spdk_jsonrpc_send_error_response(req->request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + spdk_strerror(-result)); + free_rpc_construct_virtio_dev(req); + return; + } + + w = spdk_jsonrpc_begin_result(req->request); + if (w) { + spdk_json_write_array_begin(w); + + for (i = 0; i < cnt; i++) { + spdk_json_write_string(w, spdk_bdev_get_name(bdevs[i])); + } + + spdk_json_write_array_end(w); + spdk_jsonrpc_end_result(req->request, w); + } + + free_rpc_construct_virtio_dev(ctx); +} + +static void +spdk_rpc_create_virtio_dev(struct spdk_jsonrpc_request *request, + const struct spdk_json_val *params) +{ + struct rpc_construct_virtio_dev *req; + struct spdk_bdev *bdev; + struct spdk_pci_addr pci_addr; + bool pci; + int rc; + + req = calloc(1, sizeof(*req)); + if (!req) { + SPDK_ERRLOG("calloc() failed\n"); + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, spdk_strerror(ENOMEM)); + return; + } + + if (spdk_json_decode_object(params, rpc_construct_virtio_dev, + SPDK_COUNTOF(rpc_construct_virtio_dev), + req)) { + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, spdk_strerror(EINVAL)); + goto invalid; + } + + if (strcmp(req->trtype, "pci") == 0) { + if (req->vq_count != 0 || req->vq_size != 0) { + SPDK_ERRLOG("VQ count or size is not allowed for PCI transport type\n"); + spdk_jsonrpc_send_error_response(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + "vq_count or vq_size is not allowed for PCI transport type."); + goto invalid; + } + + if (spdk_pci_addr_parse(&pci_addr, req->traddr) != 0) { + SPDK_ERRLOG("Invalid PCI address '%s'\n", req->traddr); + spdk_jsonrpc_send_error_response_fmt(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + "Invalid PCI address '%s'", req->traddr); + goto invalid; + } + + pci = true; + } else if (strcmp(req->trtype, "user") == 0) { + req->vq_count = req->vq_count == 0 ? SPDK_VIRTIO_USER_DEFAULT_VQ_COUNT : req->vq_count; + req->vq_size = req->vq_size == 0 ? SPDK_VIRTIO_USER_DEFAULT_QUEUE_SIZE : req->vq_size; + pci = false; + } else { + SPDK_ERRLOG("Invalid trtype '%s'\n", req->trtype); + spdk_jsonrpc_send_error_response_fmt(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + "Invalid trtype '%s'", req->trtype); + goto invalid; + } + + req->request = request; + if (strcmp(req->dev_type, "blk") == 0) { + if (pci) { + bdev = bdev_virtio_pci_blk_dev_create(req->name, &pci_addr); + } else { + bdev = bdev_virtio_user_blk_dev_create(req->name, req->traddr, req->vq_count, req->vq_size); + } + + /* Virtio blk doesn't use callback so call it manually to send result. */ + rc = bdev ? 0 : -EINVAL; + spdk_rpc_create_virtio_dev_cb(req, rc, &bdev, bdev ? 1 : 0); + } else if (strcmp(req->dev_type, "scsi") == 0) { + if (pci) { + rc = bdev_virtio_pci_scsi_dev_create(req->name, &pci_addr, spdk_rpc_create_virtio_dev_cb, req); + } else { + rc = bdev_virtio_user_scsi_dev_create(req->name, req->traddr, req->vq_count, req->vq_size, + spdk_rpc_create_virtio_dev_cb, req); + } + + if (rc < 0) { + /* In case of error callback is not called so do it manually to send result. */ + spdk_rpc_create_virtio_dev_cb(req, rc, NULL, 0); + } + } else { + SPDK_ERRLOG("Invalid dev_type '%s'\n", req->dev_type); + spdk_jsonrpc_send_error_response_fmt(request, SPDK_JSONRPC_ERROR_INVALID_PARAMS, + "Invalid dev_type '%s'", req->dev_type); + goto invalid; + } + + return; +invalid: + free_rpc_construct_virtio_dev(req); +} +SPDK_RPC_REGISTER("construct_virtio_dev", spdk_rpc_create_virtio_dev, SPDK_RPC_RUNTIME); diff --git a/src/spdk/lib/bdev/virtio/bdev_virtio_scsi.c b/src/spdk/lib/bdev/virtio/bdev_virtio_scsi.c new file mode 100644 index 00000000..4ff3db4a --- /dev/null +++ b/src/spdk/lib/bdev/virtio/bdev_virtio_scsi.c @@ -0,0 +1,2017 @@ +/*- + * BSD LICENSE + * + * Copyright (c) Intel Corporation. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "spdk/stdinc.h" + +#include "spdk/bdev.h" +#include "spdk/conf.h" +#include "spdk/endian.h" +#include "spdk/env.h" +#include "spdk/thread.h" +#include "spdk/scsi_spec.h" +#include "spdk/string.h" +#include "spdk/util.h" +#include "spdk/json.h" + +#include "spdk/bdev_module.h" +#include "spdk_internal/log.h" +#include "spdk_internal/virtio.h" + +#include <linux/virtio_scsi.h> + +#include "bdev_virtio.h" + +#define BDEV_VIRTIO_MAX_TARGET 64 +#define BDEV_VIRTIO_SCAN_PAYLOAD_SIZE 256 +#define MGMT_POLL_PERIOD_US (1000 * 5) +#define CTRLQ_RING_SIZE 16 +#define SCAN_REQUEST_RETRIES 5 + +/* Number of non-request queues - eventq and controlq */ +#define SPDK_VIRTIO_SCSI_QUEUE_NUM_FIXED 2 + +#define VIRTIO_SCSI_EVENTQ_BUFFER_COUNT 16 + +#define VIRTIO_SCSI_CONTROLQ 0 +#define VIRTIO_SCSI_EVENTQ 1 +#define VIRTIO_SCSI_REQUESTQ 2 + +static int bdev_virtio_initialize(void); +static void bdev_virtio_finish(void); + +struct virtio_scsi_dev { + /* Generic virtio device data. */ + struct virtio_dev vdev; + + /** Detected SCSI LUNs */ + TAILQ_HEAD(, virtio_scsi_disk) luns; + + /** Context for the SCSI target scan. */ + struct virtio_scsi_scan_base *scan_ctx; + + /** Controlq poller. */ + struct spdk_poller *mgmt_poller; + + /** Controlq messages to be sent. */ + struct spdk_ring *ctrlq_ring; + + /** Buffers for the eventq. */ + struct virtio_scsi_eventq_io *eventq_ios; + + /** Device marked for removal. */ + bool removed; + + /** Callback to be called after vdev removal. */ + bdev_virtio_remove_cb remove_cb; + + /** Context for the `remove_cb`. */ + void *remove_ctx; + + TAILQ_ENTRY(virtio_scsi_dev) tailq; +}; + +struct virtio_scsi_io_ctx { + struct iovec iov_req; + struct iovec iov_resp; + union { + struct virtio_scsi_cmd_req req; + struct virtio_scsi_ctrl_tmf_req tmf_req; + }; + union { + struct virtio_scsi_cmd_resp resp; + struct virtio_scsi_ctrl_tmf_resp tmf_resp; + }; +}; + +struct virtio_scsi_eventq_io { + struct iovec iov; + struct virtio_scsi_event ev; +}; + +struct virtio_scsi_scan_info { + uint64_t num_blocks; + uint32_t block_size; + uint8_t target; + bool unmap_supported; + TAILQ_ENTRY(virtio_scsi_scan_info) tailq; +}; + +struct virtio_scsi_scan_base { + struct virtio_scsi_dev *svdev; + + /** I/O channel used for the scan I/O. */ + struct bdev_virtio_io_channel *channel; + + bdev_virtio_create_cb cb_fn; + void *cb_arg; + + /** Scan all targets on the device. */ + bool full_scan; + + /** Start a full rescan after receiving next scan I/O response. */ + bool restart; + + /** Additional targets to be (re)scanned. */ + TAILQ_HEAD(, virtio_scsi_scan_info) scan_queue; + + /** Remaining attempts for sending the current request. */ + unsigned retries; + + /** If set, the last scan I/O needs to be resent */ + bool needs_resend; + + struct virtio_scsi_io_ctx io_ctx; + struct iovec iov; + uint8_t payload[BDEV_VIRTIO_SCAN_PAYLOAD_SIZE]; + + /** Scan results for the current target. */ + struct virtio_scsi_scan_info info; +}; + +struct virtio_scsi_disk { + struct spdk_bdev bdev; + struct virtio_scsi_dev *svdev; + struct virtio_scsi_scan_info info; + + /** Descriptor opened just to be notified of external bdev hotremove. */ + struct spdk_bdev_desc *notify_desc; + + /** Disk marked for removal. */ + bool removed; + TAILQ_ENTRY(virtio_scsi_disk) link; +}; + +struct bdev_virtio_io_channel { + struct virtio_scsi_dev *svdev; + + /** Virtqueue exclusively assigned to this channel. */ + struct virtqueue *vq; + + /** Virtio response poller. */ + struct spdk_poller *poller; +}; + +static TAILQ_HEAD(, virtio_scsi_dev) g_virtio_scsi_devs = + TAILQ_HEAD_INITIALIZER(g_virtio_scsi_devs); + +static pthread_mutex_t g_virtio_scsi_mutex = PTHREAD_MUTEX_INITIALIZER; + +/** Module finish in progress */ +static bool g_bdev_virtio_finish = false; + +/* Features desired/implemented by this driver. */ +#define VIRTIO_SCSI_DEV_SUPPORTED_FEATURES \ + (1ULL << VIRTIO_SCSI_F_INOUT | \ + 1ULL << VIRTIO_SCSI_F_HOTPLUG | \ + 1ULL << VIRTIO_RING_F_EVENT_IDX | \ + 1ULL << VHOST_USER_F_PROTOCOL_FEATURES) + +static void virtio_scsi_dev_unregister_cb(void *io_device); +static void virtio_scsi_dev_remove(struct virtio_scsi_dev *svdev, + bdev_virtio_remove_cb cb_fn, void *cb_arg); +static int bdev_virtio_scsi_ch_create_cb(void *io_device, void *ctx_buf); +static void bdev_virtio_scsi_ch_destroy_cb(void *io_device, void *ctx_buf); +static void process_scan_resp(struct virtio_scsi_scan_base *base); +static int bdev_virtio_mgmt_poll(void *arg); + +static int +virtio_scsi_dev_send_eventq_io(struct virtqueue *vq, struct virtio_scsi_eventq_io *io) +{ + int rc; + + rc = virtqueue_req_start(vq, io, 1); + if (rc != 0) { + return -1; + } + + virtqueue_req_add_iovs(vq, &io->iov, 1, SPDK_VIRTIO_DESC_WR); + virtqueue_req_flush(vq); + + return 0; +} + +static int +virtio_scsi_dev_init(struct virtio_scsi_dev *svdev, uint16_t max_queues) +{ + struct virtio_dev *vdev = &svdev->vdev; + struct spdk_ring *ctrlq_ring; + struct virtio_scsi_eventq_io *eventq_io; + struct virtqueue *eventq; + uint16_t i, num_events; + int rc; + + rc = virtio_dev_reset(vdev, VIRTIO_SCSI_DEV_SUPPORTED_FEATURES); + if (rc != 0) { + return rc; + } + + rc = virtio_dev_start(vdev, max_queues, SPDK_VIRTIO_SCSI_QUEUE_NUM_FIXED); + if (rc != 0) { + return rc; + } + + ctrlq_ring = spdk_ring_create(SPDK_RING_TYPE_MP_SC, CTRLQ_RING_SIZE, + SPDK_ENV_SOCKET_ID_ANY); + if (ctrlq_ring == NULL) { + SPDK_ERRLOG("Failed to allocate send ring for the controlq.\n"); + return -1; + } + + rc = virtio_dev_acquire_queue(vdev, VIRTIO_SCSI_CONTROLQ); + if (rc != 0) { + SPDK_ERRLOG("Failed to acquire the controlq.\n"); + spdk_ring_free(ctrlq_ring); + return -1; + } + + rc = virtio_dev_acquire_queue(vdev, VIRTIO_SCSI_EVENTQ); + if (rc != 0) { + SPDK_ERRLOG("Failed to acquire the eventq.\n"); + virtio_dev_release_queue(vdev, VIRTIO_SCSI_CONTROLQ); + spdk_ring_free(ctrlq_ring); + return -1; + } + + eventq = vdev->vqs[VIRTIO_SCSI_EVENTQ]; + num_events = spdk_min(eventq->vq_nentries, VIRTIO_SCSI_EVENTQ_BUFFER_COUNT); + svdev->eventq_ios = spdk_dma_zmalloc(sizeof(*svdev->eventq_ios) * num_events, + 0, NULL); + if (svdev->eventq_ios == NULL) { + SPDK_ERRLOG("cannot allocate memory for %"PRIu16" eventq buffers\n", + num_events); + virtio_dev_release_queue(vdev, VIRTIO_SCSI_EVENTQ); + virtio_dev_release_queue(vdev, VIRTIO_SCSI_CONTROLQ); + spdk_ring_free(ctrlq_ring); + return -1; + } + + for (i = 0; i < num_events; i++) { + eventq_io = &svdev->eventq_ios[i]; + eventq_io->iov.iov_base = &eventq_io->ev; + eventq_io->iov.iov_len = sizeof(eventq_io->ev); + virtio_scsi_dev_send_eventq_io(eventq, eventq_io); + } + + svdev->ctrlq_ring = ctrlq_ring; + + svdev->mgmt_poller = spdk_poller_register(bdev_virtio_mgmt_poll, svdev, + MGMT_POLL_PERIOD_US); + + TAILQ_INIT(&svdev->luns); + svdev->scan_ctx = NULL; + svdev->removed = false; + svdev->remove_cb = NULL; + svdev->remove_ctx = NULL; + + spdk_io_device_register(svdev, bdev_virtio_scsi_ch_create_cb, + bdev_virtio_scsi_ch_destroy_cb, + sizeof(struct bdev_virtio_io_channel), + svdev->vdev.name); + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_INSERT_TAIL(&g_virtio_scsi_devs, svdev, tailq); + pthread_mutex_unlock(&g_virtio_scsi_mutex); + return 0; +} + +static struct virtio_scsi_dev * +virtio_pci_scsi_dev_create(const char *name, struct virtio_pci_ctx *pci_ctx) +{ + static int pci_dev_counter = 0; + struct virtio_scsi_dev *svdev; + struct virtio_dev *vdev; + char *default_name = NULL; + uint32_t num_queues; + int rc; + + svdev = calloc(1, sizeof(*svdev)); + if (svdev == NULL) { + SPDK_ERRLOG("virtio device calloc failed\n"); + return NULL; + } + + vdev = &svdev->vdev; + if (name == NULL) { + default_name = spdk_sprintf_alloc("VirtioScsi%"PRIu32, pci_dev_counter++); + if (default_name == NULL) { + free(vdev); + return NULL; + } + name = default_name; + } + + rc = virtio_pci_dev_init(vdev, name, pci_ctx); + free(default_name); + + if (rc != 0) { + free(svdev); + return NULL; + } + + rc = virtio_dev_read_dev_config(vdev, offsetof(struct virtio_scsi_config, num_queues), + &num_queues, sizeof(num_queues)); + if (rc) { + SPDK_ERRLOG("%s: config read failed: %s\n", vdev->name, spdk_strerror(-rc)); + virtio_dev_destruct(vdev); + free(svdev); + return NULL; + } + + rc = virtio_scsi_dev_init(svdev, num_queues); + if (rc != 0) { + virtio_dev_destruct(vdev); + free(svdev); + return NULL; + } + + return svdev; +} + +static struct virtio_scsi_dev * +virtio_user_scsi_dev_create(const char *name, const char *path, + uint16_t num_queues, uint32_t queue_size) +{ + struct virtio_scsi_dev *svdev; + struct virtio_dev *vdev; + int rc; + + svdev = calloc(1, sizeof(*svdev)); + if (svdev == NULL) { + SPDK_ERRLOG("calloc failed for virtio device %s: %s\n", name, path); + return NULL; + } + + vdev = &svdev->vdev; + rc = virtio_user_dev_init(vdev, name, path, queue_size); + if (rc != 0) { + SPDK_ERRLOG("Failed to create virito device %s: %s\n", name, path); + free(svdev); + return NULL; + } + + rc = virtio_scsi_dev_init(svdev, num_queues); + if (rc != 0) { + virtio_dev_destruct(vdev); + free(svdev); + return NULL; + } + + return svdev; +} + +static struct virtio_scsi_disk * +virtio_scsi_dev_get_disk_by_id(struct virtio_scsi_dev *svdev, uint8_t target_id) +{ + struct virtio_scsi_disk *disk; + + TAILQ_FOREACH(disk, &svdev->luns, link) { + if (disk->info.target == target_id) { + return disk; + } + } + + return NULL; +} + +static int virtio_scsi_dev_scan(struct virtio_scsi_dev *svdev, + bdev_virtio_create_cb cb_fn, void *cb_arg); +static int send_scan_io(struct virtio_scsi_scan_base *base); +static void _virtio_scsi_dev_scan_tgt(struct virtio_scsi_scan_base *base, uint8_t target); +static int _virtio_scsi_dev_scan_next(struct virtio_scsi_scan_base *base, int rc); +static void _virtio_scsi_dev_scan_finish(struct virtio_scsi_scan_base *base, int errnum); +static int virtio_scsi_dev_scan_tgt(struct virtio_scsi_dev *svdev, uint8_t target); + +static int +bdev_virtio_get_ctx_size(void) +{ + return sizeof(struct virtio_scsi_io_ctx); +} + +static int +bdev_virtio_scsi_config_json(struct spdk_json_write_ctx *w) +{ + struct virtio_scsi_dev *svdev; + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_FOREACH(svdev, &g_virtio_scsi_devs, tailq) { + spdk_json_write_object_begin(w); + + spdk_json_write_named_string(w, "method", "construct_virtio_dev"); + + spdk_json_write_named_object_begin(w, "params"); + spdk_json_write_named_string(w, "name", svdev->vdev.name); + spdk_json_write_named_string(w, "dev_type", "scsi"); + + /* Write transport specific parameters. */ + svdev->vdev.backend_ops->write_json_config(&svdev->vdev, w); + + spdk_json_write_object_end(w); + + spdk_json_write_object_end(w); + + } + pthread_mutex_unlock(&g_virtio_scsi_mutex); + + return 0; +} + + +static struct spdk_bdev_module virtio_scsi_if = { + .name = "virtio_scsi", + .module_init = bdev_virtio_initialize, + .module_fini = bdev_virtio_finish, + .get_ctx_size = bdev_virtio_get_ctx_size, + .config_json = bdev_virtio_scsi_config_json, + .async_init = true, + .async_fini = true, +}; + +SPDK_BDEV_MODULE_REGISTER(&virtio_scsi_if) + +static struct virtio_scsi_io_ctx * +bdev_virtio_init_io_vreq(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_cmd_req *req; + struct virtio_scsi_cmd_resp *resp; + struct virtio_scsi_disk *disk = (struct virtio_scsi_disk *)bdev_io->bdev; + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + + req = &io_ctx->req; + resp = &io_ctx->resp; + + io_ctx->iov_req.iov_base = req; + io_ctx->iov_req.iov_len = sizeof(*req); + + io_ctx->iov_resp.iov_base = resp; + io_ctx->iov_resp.iov_len = sizeof(*resp); + + memset(req, 0, sizeof(*req)); + req->lun[0] = 1; + req->lun[1] = disk->info.target; + + return io_ctx; +} + +static struct virtio_scsi_io_ctx * +bdev_virtio_init_tmf_vreq(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_ctrl_tmf_req *tmf_req; + struct virtio_scsi_ctrl_tmf_resp *tmf_resp; + struct virtio_scsi_disk *disk = SPDK_CONTAINEROF(bdev_io->bdev, struct virtio_scsi_disk, bdev); + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + + tmf_req = &io_ctx->tmf_req; + tmf_resp = &io_ctx->tmf_resp; + + io_ctx->iov_req.iov_base = tmf_req; + io_ctx->iov_req.iov_len = sizeof(*tmf_req); + io_ctx->iov_resp.iov_base = tmf_resp; + io_ctx->iov_resp.iov_len = sizeof(*tmf_resp); + + memset(tmf_req, 0, sizeof(*tmf_req)); + tmf_req->lun[0] = 1; + tmf_req->lun[1] = disk->info.target; + + return io_ctx; +} + +static void +bdev_virtio_send_io(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct bdev_virtio_io_channel *virtio_channel = spdk_io_channel_get_ctx(ch); + struct virtqueue *vq = virtio_channel->vq; + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + int rc; + + rc = virtqueue_req_start(vq, bdev_io, bdev_io->u.bdev.iovcnt + 2); + if (rc == -ENOMEM) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); + return; + } else if (rc != 0) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + return; + } + + virtqueue_req_add_iovs(vq, &io_ctx->iov_req, 1, SPDK_VIRTIO_DESC_RO); + if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) { + virtqueue_req_add_iovs(vq, &io_ctx->iov_resp, 1, SPDK_VIRTIO_DESC_WR); + virtqueue_req_add_iovs(vq, bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, + SPDK_VIRTIO_DESC_WR); + } else { + virtqueue_req_add_iovs(vq, bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, + SPDK_VIRTIO_DESC_RO); + virtqueue_req_add_iovs(vq, &io_ctx->iov_resp, 1, SPDK_VIRTIO_DESC_WR); + } + + virtqueue_req_flush(vq); +} + +static void +bdev_virtio_rw(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_disk *disk = SPDK_CONTAINEROF(bdev_io->bdev, struct virtio_scsi_disk, bdev); + struct virtio_scsi_io_ctx *io_ctx = bdev_virtio_init_io_vreq(ch, bdev_io); + struct virtio_scsi_cmd_req *req = &io_ctx->req; + bool is_write = bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE; + + if (disk->info.num_blocks > (1ULL << 32)) { + req->cdb[0] = is_write ? SPDK_SBC_WRITE_16 : SPDK_SBC_READ_16; + to_be64(&req->cdb[2], bdev_io->u.bdev.offset_blocks); + to_be32(&req->cdb[10], bdev_io->u.bdev.num_blocks); + } else { + req->cdb[0] = is_write ? SPDK_SBC_WRITE_10 : SPDK_SBC_READ_10; + to_be32(&req->cdb[2], bdev_io->u.bdev.offset_blocks); + to_be16(&req->cdb[7], bdev_io->u.bdev.num_blocks); + } + + bdev_virtio_send_io(ch, bdev_io); +} + +static void +bdev_virtio_reset(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct bdev_virtio_io_channel *virtio_ch = spdk_io_channel_get_ctx(ch); + struct virtio_scsi_io_ctx *io_ctx = bdev_virtio_init_tmf_vreq(ch, bdev_io); + struct virtio_scsi_ctrl_tmf_req *tmf_req = &io_ctx->tmf_req; + struct virtio_scsi_dev *svdev = virtio_ch->svdev; + size_t enqueued_count; + + tmf_req->type = VIRTIO_SCSI_T_TMF; + tmf_req->subtype = VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET; + + enqueued_count = spdk_ring_enqueue(svdev->ctrlq_ring, (void **)&bdev_io, 1); + if (spdk_likely(enqueued_count == 1)) { + return; + } else { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); + } +} + +static void +bdev_virtio_unmap(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_io_ctx *io_ctx = bdev_virtio_init_io_vreq(ch, bdev_io); + struct virtio_scsi_cmd_req *req = &io_ctx->req; + struct spdk_scsi_unmap_bdesc *desc, *first_desc; + uint8_t *buf; + uint64_t offset_blocks, num_blocks; + uint16_t cmd_len; + + buf = bdev_io->u.bdev.iovs[0].iov_base; + + offset_blocks = bdev_io->u.bdev.offset_blocks; + num_blocks = bdev_io->u.bdev.num_blocks; + + /* (n-1) * 16-byte descriptors */ + first_desc = desc = (struct spdk_scsi_unmap_bdesc *)&buf[8]; + while (num_blocks > UINT32_MAX) { + to_be64(&desc->lba, offset_blocks); + to_be32(&desc->block_count, UINT32_MAX); + memset(&desc->reserved, 0, sizeof(desc->reserved)); + offset_blocks += UINT32_MAX; + num_blocks -= UINT32_MAX; + desc++; + } + + /* The last descriptor with block_count <= UINT32_MAX */ + to_be64(&desc->lba, offset_blocks); + to_be32(&desc->block_count, num_blocks); + memset(&desc->reserved, 0, sizeof(desc->reserved)); + + /* 8-byte header + n * 16-byte block descriptor */ + cmd_len = 8 + (desc - first_desc + 1) * sizeof(struct spdk_scsi_unmap_bdesc); + + req->cdb[0] = SPDK_SBC_UNMAP; + to_be16(&req->cdb[7], cmd_len); + + /* 8-byte header */ + to_be16(&buf[0], cmd_len - 2); /* total length (excluding the length field) */ + to_be16(&buf[2], cmd_len - 8); /* length of block descriptors */ + memset(&buf[4], 0, 4); /* reserved */ + + bdev_virtio_send_io(ch, bdev_io); +} + +static int _bdev_virtio_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_disk *disk = SPDK_CONTAINEROF(bdev_io->bdev, struct virtio_scsi_disk, bdev); + + switch (bdev_io->type) { + case SPDK_BDEV_IO_TYPE_READ: + 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); + return 0; + case SPDK_BDEV_IO_TYPE_RESET: + bdev_virtio_reset(ch, bdev_io); + return 0; + case SPDK_BDEV_IO_TYPE_UNMAP: { + uint64_t buf_len = 8 /* header size */ + + (bdev_io->u.bdev.num_blocks + UINT32_MAX - 1) / + UINT32_MAX * sizeof(struct spdk_scsi_unmap_bdesc); + + if (!disk->info.unmap_supported) { + return -1; + } + + if (buf_len > SPDK_BDEV_LARGE_BUF_MAX_SIZE) { + SPDK_ERRLOG("Trying to UNMAP too many blocks: %"PRIu64"\n", + bdev_io->u.bdev.num_blocks); + return -1; + } + spdk_bdev_io_get_buf(bdev_io, bdev_virtio_unmap, buf_len); + return 0; + } + case SPDK_BDEV_IO_TYPE_FLUSH: + default: + return -1; + } + return 0; +} + +static void bdev_virtio_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) +{ + if (_bdev_virtio_submit_request(ch, bdev_io) < 0) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + } +} + +static bool +bdev_virtio_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type) +{ + struct virtio_scsi_disk *disk = ctx; + + switch (io_type) { + case SPDK_BDEV_IO_TYPE_READ: + case SPDK_BDEV_IO_TYPE_WRITE: + case SPDK_BDEV_IO_TYPE_FLUSH: + case SPDK_BDEV_IO_TYPE_RESET: + return true; + + case SPDK_BDEV_IO_TYPE_UNMAP: + return disk->info.unmap_supported; + + default: + return false; + } +} + +static struct spdk_io_channel * +bdev_virtio_get_io_channel(void *ctx) +{ + struct virtio_scsi_disk *disk = ctx; + + return spdk_get_io_channel(disk->svdev); +} + +static int +bdev_virtio_disk_destruct(void *ctx) +{ + struct virtio_scsi_disk *disk = ctx; + struct virtio_scsi_dev *svdev = disk->svdev; + + TAILQ_REMOVE(&svdev->luns, disk, link); + free(disk->bdev.name); + free(disk); + + if (svdev->removed && TAILQ_EMPTY(&svdev->luns)) { + spdk_io_device_unregister(svdev, virtio_scsi_dev_unregister_cb); + } + + return 0; +} + +static int +bdev_virtio_dump_info_json(void *ctx, struct spdk_json_write_ctx *w) +{ + struct virtio_scsi_disk *disk = ctx; + + virtio_dev_dump_json_info(&disk->svdev->vdev, w); + return 0; +} + +static void +bdev_virtio_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w) +{ + /* SCSI targets and LUNS are discovered during scan process so nothing + * to save here. + */ +} + +static const struct spdk_bdev_fn_table virtio_fn_table = { + .destruct = bdev_virtio_disk_destruct, + .submit_request = bdev_virtio_submit_request, + .io_type_supported = bdev_virtio_io_type_supported, + .get_io_channel = bdev_virtio_get_io_channel, + .dump_info_json = bdev_virtio_dump_info_json, + .write_config_json = bdev_virtio_write_config_json, +}; + +static void +get_scsi_status(struct virtio_scsi_cmd_resp *resp, int *sk, int *asc, int *ascq) +{ + /* see spdk_scsi_task_build_sense_data() for sense data details */ + *sk = 0; + *asc = 0; + *ascq = 0; + + if (resp->sense_len < 3) { + return; + } + + *sk = resp->sense[2] & 0xf; + + if (resp->sense_len < 13) { + return; + } + + *asc = resp->sense[12]; + + if (resp->sense_len < 14) { + return; + } + + *ascq = resp->sense[13]; +} + +static void +bdev_virtio_io_cpl(struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + int sk, asc, ascq; + + get_scsi_status(&io_ctx->resp, &sk, &asc, &ascq); + spdk_bdev_io_complete_scsi_status(bdev_io, io_ctx->resp.status, sk, asc, ascq); +} + +static int +bdev_virtio_poll(void *arg) +{ + struct bdev_virtio_io_channel *ch = arg; + struct virtio_scsi_dev *svdev = ch->svdev; + struct virtio_scsi_scan_base *scan_ctx = svdev->scan_ctx; + void *io[32]; + uint32_t io_len[32]; + uint16_t i, cnt; + int rc; + + cnt = virtio_recv_pkts(ch->vq, (void **)io, io_len, SPDK_COUNTOF(io)); + for (i = 0; i < cnt; ++i) { + if (spdk_unlikely(scan_ctx && io[i] == &scan_ctx->io_ctx)) { + if (svdev->removed) { + _virtio_scsi_dev_scan_finish(scan_ctx, -EINTR); + return -1; + } + + if (scan_ctx->restart) { + scan_ctx->restart = false; + scan_ctx->full_scan = true; + _virtio_scsi_dev_scan_tgt(scan_ctx, 0); + continue; + } + + process_scan_resp(scan_ctx); + continue; + } + + bdev_virtio_io_cpl(io[i]); + } + + if (spdk_unlikely(scan_ctx && scan_ctx->needs_resend)) { + if (svdev->removed) { + _virtio_scsi_dev_scan_finish(scan_ctx, -EINTR); + return -1; + } else if (cnt == 0) { + return 0; + } + + rc = send_scan_io(scan_ctx); + if (rc != 0) { + assert(scan_ctx->retries > 0); + scan_ctx->retries--; + if (scan_ctx->retries == 0) { + SPDK_ERRLOG("Target scan failed unrecoverably with rc = %d.\n", rc); + _virtio_scsi_dev_scan_finish(scan_ctx, rc); + } + } + } + + return cnt; +} + +static void +bdev_virtio_tmf_cpl_cb(void *ctx) +{ + struct spdk_bdev_io *bdev_io = ctx; + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + + if (io_ctx->tmf_resp.response == VIRTIO_SCSI_S_OK) { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS); + } else { + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); + } +} + +static void +bdev_virtio_tmf_cpl(struct spdk_bdev_io *bdev_io) +{ + spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), bdev_virtio_tmf_cpl_cb, bdev_io); +} + +static void +bdev_virtio_eventq_io_cpl(struct virtio_scsi_dev *svdev, struct virtio_scsi_eventq_io *io) +{ + struct virtio_scsi_event *ev = &io->ev; + struct virtio_scsi_disk *disk; + + if (ev->lun[0] != 1) { + SPDK_WARNLOG("Received an event with invalid data layout.\n"); + goto out; + } + + if (ev->event & VIRTIO_SCSI_T_EVENTS_MISSED) { + ev->event &= ~VIRTIO_SCSI_T_EVENTS_MISSED; + virtio_scsi_dev_scan(svdev, NULL, NULL); + } + + switch (ev->event) { + case VIRTIO_SCSI_T_NO_EVENT: + break; + case VIRTIO_SCSI_T_TRANSPORT_RESET: + switch (ev->reason) { + case VIRTIO_SCSI_EVT_RESET_RESCAN: + virtio_scsi_dev_scan_tgt(svdev, ev->lun[1]); + break; + case VIRTIO_SCSI_EVT_RESET_REMOVED: + disk = virtio_scsi_dev_get_disk_by_id(svdev, ev->lun[1]); + if (disk != NULL) { + spdk_bdev_unregister(&disk->bdev, NULL, NULL); + } + break; + default: + break; + } + break; + default: + break; + } + +out: + virtio_scsi_dev_send_eventq_io(svdev->vdev.vqs[VIRTIO_SCSI_EVENTQ], io); +} + +static void +bdev_virtio_tmf_abort_nomem_cb(void *ctx) +{ + struct spdk_bdev_io *bdev_io = ctx; + + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); +} + +static void +bdev_virtio_tmf_abort_ioerr_cb(void *ctx) +{ + struct spdk_bdev_io *bdev_io = ctx; + + spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); +} + +static void +bdev_virtio_tmf_abort(struct spdk_bdev_io *bdev_io, int status) +{ + spdk_thread_fn fn; + + if (status == -ENOMEM) { + fn = bdev_virtio_tmf_abort_nomem_cb; + } else { + fn = bdev_virtio_tmf_abort_ioerr_cb; + } + + spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), fn, bdev_io); +} + +static int +bdev_virtio_send_tmf_io(struct virtqueue *ctrlq, struct spdk_bdev_io *bdev_io) +{ + struct virtio_scsi_io_ctx *io_ctx = (struct virtio_scsi_io_ctx *)bdev_io->driver_ctx; + int rc; + + rc = virtqueue_req_start(ctrlq, bdev_io, 2); + if (rc != 0) { + return rc; + } + + virtqueue_req_add_iovs(ctrlq, &io_ctx->iov_req, 1, SPDK_VIRTIO_DESC_RO); + virtqueue_req_add_iovs(ctrlq, &io_ctx->iov_resp, 1, SPDK_VIRTIO_DESC_WR); + + virtqueue_req_flush(ctrlq); + return 0; +} + +static int +bdev_virtio_mgmt_poll(void *arg) +{ + struct virtio_scsi_dev *svdev = arg; + struct virtio_dev *vdev = &svdev->vdev; + struct virtqueue *eventq = vdev->vqs[VIRTIO_SCSI_EVENTQ]; + struct virtqueue *ctrlq = vdev->vqs[VIRTIO_SCSI_CONTROLQ]; + struct spdk_ring *send_ring = svdev->ctrlq_ring; + void *io[16]; + uint32_t io_len[16]; + uint16_t i, cnt; + int rc; + int total = 0; + + cnt = spdk_ring_dequeue(send_ring, io, SPDK_COUNTOF(io)); + total += cnt; + for (i = 0; i < cnt; ++i) { + rc = bdev_virtio_send_tmf_io(ctrlq, io[i]); + if (rc != 0) { + bdev_virtio_tmf_abort(io[i], rc); + } + } + + cnt = virtio_recv_pkts(ctrlq, io, io_len, SPDK_COUNTOF(io)); + total += cnt; + for (i = 0; i < cnt; ++i) { + bdev_virtio_tmf_cpl(io[i]); + } + + cnt = virtio_recv_pkts(eventq, io, io_len, SPDK_COUNTOF(io)); + total += cnt; + for (i = 0; i < cnt; ++i) { + bdev_virtio_eventq_io_cpl(svdev, io[i]); + } + + return total; +} + +static int +bdev_virtio_scsi_ch_create_cb(void *io_device, void *ctx_buf) +{ + struct virtio_scsi_dev *svdev = io_device; + struct virtio_dev *vdev = &svdev->vdev; + struct bdev_virtio_io_channel *ch = ctx_buf; + struct virtqueue *vq; + int32_t queue_idx; + + queue_idx = virtio_dev_find_and_acquire_queue(vdev, VIRTIO_SCSI_REQUESTQ); + if (queue_idx < 0) { + SPDK_ERRLOG("Couldn't get an unused queue for the io_channel.\n"); + return -1; + } + + vq = vdev->vqs[queue_idx]; + + ch->svdev = svdev; + ch->vq = vq; + + ch->poller = spdk_poller_register(bdev_virtio_poll, ch, 0); + + return 0; +} + +static void +bdev_virtio_scsi_ch_destroy_cb(void *io_device, void *ctx_buf) +{ + struct bdev_virtio_io_channel *ch = ctx_buf; + struct virtio_scsi_dev *svdev = ch->svdev; + struct virtio_dev *vdev = &svdev->vdev; + struct virtqueue *vq = ch->vq; + + spdk_poller_unregister(&ch->poller); + virtio_dev_release_queue(vdev, vq->vq_queue_index); +} + +static void +_virtio_scsi_dev_scan_finish(struct virtio_scsi_scan_base *base, int errnum) +{ + struct virtio_scsi_dev *svdev = base->svdev; + size_t bdevs_cnt; + struct spdk_bdev *bdevs[BDEV_VIRTIO_MAX_TARGET]; + struct virtio_scsi_disk *disk; + struct virtio_scsi_scan_info *tgt, *next_tgt; + + spdk_put_io_channel(spdk_io_channel_from_ctx(base->channel)); + base->svdev->scan_ctx = NULL; + + TAILQ_FOREACH_SAFE(tgt, &base->scan_queue, tailq, next_tgt) { + TAILQ_REMOVE(&base->scan_queue, tgt, tailq); + free(tgt); + } + + if (base->cb_fn == NULL) { + spdk_dma_free(base); + return; + } + + bdevs_cnt = 0; + if (errnum == 0) { + TAILQ_FOREACH(disk, &svdev->luns, link) { + bdevs[bdevs_cnt] = &disk->bdev; + bdevs_cnt++; + } + } + + base->cb_fn(base->cb_arg, errnum, bdevs, bdevs_cnt); + spdk_dma_free(base); +} + +static int +send_scan_io(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_io_ctx *io_ctx = &base->io_ctx; + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct virtqueue *vq = base->channel->vq; + int payload_iov_cnt = base->iov.iov_len > 0 ? 1 : 0; + int rc; + + req->lun[0] = 1; + req->lun[1] = base->info.target; + + rc = virtqueue_req_start(vq, io_ctx, 2 + payload_iov_cnt); + if (rc != 0) { + base->needs_resend = true; + return -1; + } + + virtqueue_req_add_iovs(vq, &io_ctx->iov_req, 1, SPDK_VIRTIO_DESC_RO); + virtqueue_req_add_iovs(vq, &io_ctx->iov_resp, 1, SPDK_VIRTIO_DESC_WR); + virtqueue_req_add_iovs(vq, &base->iov, payload_iov_cnt, SPDK_VIRTIO_DESC_WR); + + virtqueue_req_flush(vq); + return 0; +} + +static int +send_inquiry(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct spdk_scsi_cdb_inquiry *cdb; + + memset(req, 0, sizeof(*req)); + + base->iov.iov_len = BDEV_VIRTIO_SCAN_PAYLOAD_SIZE; + cdb = (struct spdk_scsi_cdb_inquiry *)req->cdb; + cdb->opcode = SPDK_SPC_INQUIRY; + to_be16(cdb->alloc_len, BDEV_VIRTIO_SCAN_PAYLOAD_SIZE); + + return send_scan_io(base); +} + +static int +send_inquiry_vpd(struct virtio_scsi_scan_base *base, uint8_t page_code) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct spdk_scsi_cdb_inquiry *inquiry_cdb = (struct spdk_scsi_cdb_inquiry *)req->cdb; + + memset(req, 0, sizeof(*req)); + + base->iov.iov_len = BDEV_VIRTIO_SCAN_PAYLOAD_SIZE; + inquiry_cdb->opcode = SPDK_SPC_INQUIRY; + inquiry_cdb->evpd = 1; + inquiry_cdb->page_code = page_code; + to_be16(inquiry_cdb->alloc_len, base->iov.iov_len); + + return send_scan_io(base); +} + +static int +send_read_cap_10(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + + memset(req, 0, sizeof(*req)); + + base->iov.iov_len = 8; + req->cdb[0] = SPDK_SBC_READ_CAPACITY_10; + + return send_scan_io(base); +} + +static int +send_read_cap_16(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + + memset(req, 0, sizeof(*req)); + + base->iov.iov_len = 32; + req->cdb[0] = SPDK_SPC_SERVICE_ACTION_IN_16; + req->cdb[1] = SPDK_SBC_SAI_READ_CAPACITY_16; + to_be32(&req->cdb[10], base->iov.iov_len); + + return send_scan_io(base); +} + +static int +send_test_unit_ready(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + + memset(req, 0, sizeof(*req)); + req->cdb[0] = SPDK_SPC_TEST_UNIT_READY; + base->iov.iov_len = 0; + + return send_scan_io(base); +} + +static int +send_start_stop_unit(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + + memset(req, 0, sizeof(*req)); + req->cdb[0] = SPDK_SBC_START_STOP_UNIT; + req->cdb[4] = SPDK_SBC_START_STOP_UNIT_START_BIT; + base->iov.iov_len = 0; + + return send_scan_io(base); +} + +static int +process_scan_start_stop_unit(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + + if (resp->status == SPDK_SCSI_STATUS_GOOD) { + return send_inquiry_vpd(base, SPDK_SPC_VPD_SUPPORTED_VPD_PAGES); + } + + return -1; +} + +static int +process_scan_test_unit_ready(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + int sk, asc, ascq; + + get_scsi_status(resp, &sk, &asc, &ascq); + + /* check response, get VPD if spun up otherwise send SSU */ + if (resp->status == SPDK_SCSI_STATUS_GOOD) { + return send_inquiry_vpd(base, SPDK_SPC_VPD_SUPPORTED_VPD_PAGES); + } else if (resp->response == VIRTIO_SCSI_S_OK && + resp->status == SPDK_SCSI_STATUS_CHECK_CONDITION && + sk == SPDK_SCSI_SENSE_UNIT_ATTENTION && + asc == SPDK_SCSI_ASC_LOGICAL_UNIT_NOT_READY) { + return send_start_stop_unit(base); + } else { + return -1; + } +} + +static int +process_scan_inquiry_standard(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + struct spdk_scsi_cdb_inquiry_data *inquiry_data = + (struct spdk_scsi_cdb_inquiry_data *)base->payload; + + if (resp->status != SPDK_SCSI_STATUS_GOOD) { + return -1; + } + + /* check to make sure its a supported device */ + if (inquiry_data->peripheral_device_type != SPDK_SPC_PERIPHERAL_DEVICE_TYPE_DISK || + inquiry_data->peripheral_qualifier != SPDK_SPC_PERIPHERAL_QUALIFIER_CONNECTED) { + SPDK_WARNLOG("Unsupported peripheral device type 0x%02x (qualifier 0x%02x)\n", + inquiry_data->peripheral_device_type, + inquiry_data->peripheral_qualifier); + return -1; + } + + return send_test_unit_ready(base); +} + +static int +process_scan_inquiry_vpd_supported_vpd_pages(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + bool block_provisioning_page_supported = false; + + if (resp->status == SPDK_SCSI_STATUS_GOOD) { + const uint8_t *vpd_data = base->payload; + const uint8_t *supported_vpd_pages = vpd_data + 4; + uint16_t page_length; + uint16_t num_supported_pages; + uint16_t i; + + page_length = from_be16(vpd_data + 2); + num_supported_pages = spdk_min(page_length, base->iov.iov_len - 4); + + for (i = 0; i < num_supported_pages; i++) { + if (supported_vpd_pages[i] == SPDK_SPC_VPD_BLOCK_THIN_PROVISION) { + block_provisioning_page_supported = true; + break; + } + } + } + + if (block_provisioning_page_supported) { + return send_inquiry_vpd(base, SPDK_SPC_VPD_BLOCK_THIN_PROVISION); + } else { + return send_read_cap_10(base); + } +} + +static int +process_scan_inquiry_vpd_block_thin_provision(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + + base->info.unmap_supported = false; + + if (resp->status == SPDK_SCSI_STATUS_GOOD) { + uint8_t *vpd_data = base->payload; + + base->info.unmap_supported = !!(vpd_data[5] & SPDK_SCSI_UNMAP_LBPU); + } + + SPDK_INFOLOG(SPDK_LOG_VIRTIO, "Target %u: unmap supported = %d\n", + base->info.target, (int)base->info.unmap_supported); + + return send_read_cap_10(base); +} + +static int +process_scan_inquiry(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct spdk_scsi_cdb_inquiry *inquiry_cdb = (struct spdk_scsi_cdb_inquiry *)req->cdb; + + if ((inquiry_cdb->evpd & 1) == 0) { + return process_scan_inquiry_standard(base); + } + + switch (inquiry_cdb->page_code) { + case SPDK_SPC_VPD_SUPPORTED_VPD_PAGES: + return process_scan_inquiry_vpd_supported_vpd_pages(base); + case SPDK_SPC_VPD_BLOCK_THIN_PROVISION: + return process_scan_inquiry_vpd_block_thin_provision(base); + default: + SPDK_DEBUGLOG(SPDK_LOG_VIRTIO, "Unexpected VPD page 0x%02x\n", inquiry_cdb->page_code); + return -1; + } +} + +static void +bdev_virtio_disc_notify_remove(void *remove_ctx) +{ + struct virtio_scsi_disk *disk = remove_ctx; + + disk->removed = true; + spdk_bdev_close(disk->notify_desc); +} + +/* To be called only from the thread performing target scan */ +static int +virtio_scsi_dev_add_tgt(struct virtio_scsi_dev *svdev, struct virtio_scsi_scan_info *info) +{ + struct virtio_scsi_disk *disk; + struct spdk_bdev *bdev; + int rc; + + TAILQ_FOREACH(disk, &svdev->luns, link) { + if (disk->info.target == info->target) { + /* Target is already attached and param change is not supported */ + return 0; + } + } + + if (info->block_size == 0 || info->num_blocks == 0) { + SPDK_ERRLOG("%s: invalid target %u: bs=%"PRIu32" blocks=%"PRIu64"\n", + svdev->vdev.name, info->target, info->block_size, info->num_blocks); + return -EINVAL; + } + + disk = calloc(1, sizeof(*disk)); + if (disk == NULL) { + SPDK_ERRLOG("could not allocate disk\n"); + return -ENOMEM; + } + + disk->svdev = svdev; + memcpy(&disk->info, info, sizeof(*info)); + + bdev = &disk->bdev; + bdev->name = spdk_sprintf_alloc("%st%"PRIu8, svdev->vdev.name, info->target); + if (bdev->name == NULL) { + SPDK_ERRLOG("Couldn't alloc memory for the bdev name.\n"); + free(disk); + return -ENOMEM; + } + + bdev->product_name = "Virtio SCSI Disk"; + bdev->write_cache = 0; + bdev->blocklen = disk->info.block_size; + bdev->blockcnt = disk->info.num_blocks; + + bdev->ctxt = disk; + bdev->fn_table = &virtio_fn_table; + bdev->module = &virtio_scsi_if; + + rc = spdk_bdev_register(&disk->bdev); + if (rc) { + SPDK_ERRLOG("Failed to register bdev name=%s\n", disk->bdev.name); + free(bdev->name); + free(disk); + return rc; + } + + rc = spdk_bdev_open(bdev, false, bdev_virtio_disc_notify_remove, disk, &disk->notify_desc); + if (rc) { + assert(false); + } + + TAILQ_INSERT_TAIL(&svdev->luns, disk, link); + return 0; +} + +static int +process_read_cap_10(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + uint64_t max_block; + uint32_t block_size; + uint8_t target_id = req->lun[1]; + int rc; + + if (resp->response != VIRTIO_SCSI_S_OK || resp->status != SPDK_SCSI_STATUS_GOOD) { + SPDK_ERRLOG("READ CAPACITY (10) failed for target %"PRIu8".\n", target_id); + return -1; + } + + block_size = from_be32(base->payload + 4); + max_block = from_be32(base->payload); + + if (max_block == 0xffffffff) { + return send_read_cap_16(base); + } + + base->info.num_blocks = (uint64_t)max_block + 1; + base->info.block_size = block_size; + + rc = virtio_scsi_dev_add_tgt(base->svdev, &base->info); + if (rc != 0) { + return rc; + } + + return _virtio_scsi_dev_scan_next(base, 0); +} + +static int +process_read_cap_16(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + uint8_t target_id = req->lun[1]; + int rc; + + if (resp->response != VIRTIO_SCSI_S_OK || resp->status != SPDK_SCSI_STATUS_GOOD) { + SPDK_ERRLOG("READ CAPACITY (16) failed for target %"PRIu8".\n", target_id); + return -1; + } + + base->info.num_blocks = from_be64(base->payload) + 1; + base->info.block_size = from_be32(base->payload + 8); + rc = virtio_scsi_dev_add_tgt(base->svdev, &base->info); + if (rc != 0) { + return rc; + } + + return _virtio_scsi_dev_scan_next(base, 0); +} + +static void +process_scan_resp(struct virtio_scsi_scan_base *base) +{ + struct virtio_scsi_cmd_req *req = &base->io_ctx.req; + struct virtio_scsi_cmd_resp *resp = &base->io_ctx.resp; + int rc, sk, asc, ascq; + uint8_t target_id; + + if (base->io_ctx.iov_req.iov_len < sizeof(struct virtio_scsi_cmd_req) || + base->io_ctx.iov_resp.iov_len < sizeof(struct virtio_scsi_cmd_resp)) { + SPDK_ERRLOG("Received target scan message with invalid length.\n"); + _virtio_scsi_dev_scan_next(base, -EIO); + return; + } + + get_scsi_status(resp, &sk, &asc, &ascq); + target_id = req->lun[1]; + + if (resp->response == VIRTIO_SCSI_S_BAD_TARGET || + resp->response == VIRTIO_SCSI_S_INCORRECT_LUN) { + _virtio_scsi_dev_scan_next(base, -ENODEV); + return; + } + + if (resp->response != VIRTIO_SCSI_S_OK || + (resp->status == SPDK_SCSI_STATUS_CHECK_CONDITION && + sk != SPDK_SCSI_SENSE_ILLEGAL_REQUEST)) { + assert(base->retries > 0); + base->retries--; + if (base->retries == 0) { + SPDK_NOTICELOG("Target %"PRIu8" is present, but unavailable.\n", target_id); + SPDK_TRACEDUMP(SPDK_LOG_VIRTIO, "CDB", req->cdb, sizeof(req->cdb)); + SPDK_TRACEDUMP(SPDK_LOG_VIRTIO, "SENSE DATA", resp->sense, sizeof(resp->sense)); + _virtio_scsi_dev_scan_next(base, -EBUSY); + return; + } + + /* resend the same request */ + rc = send_scan_io(base); + if (rc != 0) { + /* Let response poller do the resend */ + } + return; + } + + base->retries = SCAN_REQUEST_RETRIES; + + switch (req->cdb[0]) { + case SPDK_SPC_INQUIRY: + rc = process_scan_inquiry(base); + break; + case SPDK_SPC_TEST_UNIT_READY: + rc = process_scan_test_unit_ready(base); + break; + case SPDK_SBC_START_STOP_UNIT: + rc = process_scan_start_stop_unit(base); + break; + case SPDK_SBC_READ_CAPACITY_10: + rc = process_read_cap_10(base); + break; + case SPDK_SPC_SERVICE_ACTION_IN_16: + rc = process_read_cap_16(base); + break; + default: + SPDK_ERRLOG("Received invalid target scan message: cdb[0] = %"PRIu8".\n", req->cdb[0]); + rc = -1; + break; + } + + if (rc != 0) { + if (base->needs_resend) { + return; /* Let response poller do the resend */ + } + + _virtio_scsi_dev_scan_next(base, rc); + } +} + +static int +_virtio_scsi_dev_scan_next(struct virtio_scsi_scan_base *base, int rc) +{ + struct virtio_scsi_scan_info *next; + struct virtio_scsi_disk *disk; + uint8_t target_id; + + if (base->full_scan) { + if (rc != 0) { + disk = virtio_scsi_dev_get_disk_by_id(base->svdev, + base->info.target); + if (disk != NULL) { + spdk_bdev_unregister(&disk->bdev, NULL, NULL); + } + } + + target_id = base->info.target + 1; + if (target_id < BDEV_VIRTIO_MAX_TARGET) { + _virtio_scsi_dev_scan_tgt(base, target_id); + return 0; + } + + base->full_scan = false; + } + + next = TAILQ_FIRST(&base->scan_queue); + if (next == NULL) { + _virtio_scsi_dev_scan_finish(base, 0); + return 0; + } + + TAILQ_REMOVE(&base->scan_queue, next, tailq); + target_id = next->target; + free(next); + + _virtio_scsi_dev_scan_tgt(base, target_id); + return 0; +} + +static int +virtio_pci_scsi_dev_enumerate_cb(struct virtio_pci_ctx *pci_ctx, void *ctx) +{ + struct virtio_scsi_dev *svdev; + + svdev = virtio_pci_scsi_dev_create(NULL, pci_ctx); + return svdev == NULL ? -1 : 0; +} + +static int +bdev_virtio_process_config(void) +{ + struct spdk_conf_section *sp; + struct virtio_scsi_dev *svdev; + char *default_name = NULL; + char *path, *type, *name; + unsigned vdev_num; + int num_queues; + bool enable_pci; + int rc = 0; + + for (sp = spdk_conf_first_section(NULL); sp != NULL; sp = spdk_conf_next_section(sp)) { + if (!spdk_conf_section_match_prefix(sp, "VirtioUser")) { + continue; + } + + if (sscanf(spdk_conf_section_get_name(sp), "VirtioUser%u", &vdev_num) != 1) { + SPDK_ERRLOG("Section '%s' has non-numeric suffix.\n", + spdk_conf_section_get_name(sp)); + rc = -1; + goto out; + } + + path = spdk_conf_section_get_val(sp, "Path"); + if (path == NULL) { + SPDK_ERRLOG("VirtioUser%u: missing Path\n", vdev_num); + rc = -1; + goto out; + } + + type = spdk_conf_section_get_val(sp, "Type"); + if (type != NULL && strcmp(type, "SCSI") != 0) { + continue; + } + + num_queues = spdk_conf_section_get_intval(sp, "Queues"); + if (num_queues < 1) { + num_queues = 1; + } else if (num_queues > SPDK_VIRTIO_MAX_VIRTQUEUES) { + num_queues = SPDK_VIRTIO_MAX_VIRTQUEUES; + } + + name = spdk_conf_section_get_val(sp, "Name"); + if (name == NULL) { + default_name = spdk_sprintf_alloc("VirtioScsi%u", vdev_num); + name = default_name; + } + + svdev = virtio_user_scsi_dev_create(name, path, num_queues, 512); + free(default_name); + default_name = NULL; + + if (svdev == NULL) { + rc = -1; + goto out; + } + } + + sp = spdk_conf_find_section(NULL, "VirtioPci"); + if (sp == NULL) { + return 0; + } + + enable_pci = spdk_conf_section_get_boolval(sp, "Enable", false); + if (enable_pci) { + rc = virtio_pci_dev_enumerate(virtio_pci_scsi_dev_enumerate_cb, NULL, + PCI_DEVICE_ID_VIRTIO_SCSI_MODERN); + } + +out: + return rc; +} + +static int +_virtio_scsi_dev_scan_init(struct virtio_scsi_dev *svdev) +{ + struct virtio_scsi_scan_base *base; + struct spdk_io_channel *io_ch; + struct virtio_scsi_io_ctx *io_ctx; + struct virtio_scsi_cmd_req *req; + struct virtio_scsi_cmd_resp *resp; + + io_ch = spdk_get_io_channel(svdev); + if (io_ch == NULL) { + return -EBUSY; + } + + base = spdk_dma_zmalloc(sizeof(*base), 64, NULL); + if (base == NULL) { + SPDK_ERRLOG("couldn't allocate memory for scsi target scan.\n"); + return -ENOMEM; + } + + base->svdev = svdev; + + base->channel = spdk_io_channel_get_ctx(io_ch); + TAILQ_INIT(&base->scan_queue); + svdev->scan_ctx = base; + + base->iov.iov_base = base->payload; + io_ctx = &base->io_ctx; + req = &io_ctx->req; + resp = &io_ctx->resp; + io_ctx->iov_req.iov_base = req; + io_ctx->iov_req.iov_len = sizeof(*req); + io_ctx->iov_resp.iov_base = resp; + io_ctx->iov_resp.iov_len = sizeof(*resp); + + base->retries = SCAN_REQUEST_RETRIES; + return 0; +} + +static void +_virtio_scsi_dev_scan_tgt(struct virtio_scsi_scan_base *base, uint8_t target) +{ + int rc; + + memset(&base->info, 0, sizeof(base->info)); + base->info.target = target; + + rc = send_inquiry(base); + if (rc) { + /* Let response poller do the resend */ + } +} + +static int +virtio_scsi_dev_scan(struct virtio_scsi_dev *svdev, bdev_virtio_create_cb cb_fn, + void *cb_arg) +{ + struct virtio_scsi_scan_base *base; + struct virtio_scsi_scan_info *tgt, *next_tgt; + int rc; + + if (svdev->scan_ctx) { + if (svdev->scan_ctx->full_scan) { + return -EEXIST; + } + + /* We're about to start a full rescan, so there's no need + * to scan particular targets afterwards. + */ + TAILQ_FOREACH_SAFE(tgt, &svdev->scan_ctx->scan_queue, tailq, next_tgt) { + TAILQ_REMOVE(&svdev->scan_ctx->scan_queue, tgt, tailq); + free(tgt); + } + + svdev->scan_ctx->cb_fn = cb_fn; + svdev->scan_ctx->cb_arg = cb_arg; + svdev->scan_ctx->restart = true; + return 0; + } + + rc = _virtio_scsi_dev_scan_init(svdev); + if (rc != 0) { + return rc; + } + + base = svdev->scan_ctx; + base->cb_fn = cb_fn; + base->cb_arg = cb_arg; + base->full_scan = true; + + _virtio_scsi_dev_scan_tgt(base, 0); + return 0; +} + +static int +virtio_scsi_dev_scan_tgt(struct virtio_scsi_dev *svdev, uint8_t target) +{ + struct virtio_scsi_scan_base *base; + struct virtio_scsi_scan_info *info; + int rc; + + base = svdev->scan_ctx; + if (base) { + info = calloc(1, sizeof(*info)); + if (info == NULL) { + SPDK_ERRLOG("calloc failed\n"); + return -ENOMEM; + } + + info->target = target; + TAILQ_INSERT_TAIL(&base->scan_queue, info, tailq); + return 0; + } + + rc = _virtio_scsi_dev_scan_init(svdev); + if (rc != 0) { + return rc; + } + + base = svdev->scan_ctx; + base->full_scan = true; + _virtio_scsi_dev_scan_tgt(base, target); + return 0; +} + +static void +bdev_virtio_initial_scan_complete(void *ctx, int result, + struct spdk_bdev **bdevs, size_t bdevs_cnt) +{ + struct virtio_scsi_dev *svdev; + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_FOREACH(svdev, &g_virtio_scsi_devs, tailq) { + if (svdev->scan_ctx) { + /* another device is still being scanned */ + pthread_mutex_unlock(&g_virtio_scsi_mutex); + return; + } + } + + pthread_mutex_unlock(&g_virtio_scsi_mutex); + spdk_bdev_module_init_done(&virtio_scsi_if); +} + +static int +bdev_virtio_initialize(void) +{ + struct virtio_scsi_dev *svdev, *next_svdev; + int rc; + + rc = bdev_virtio_process_config(); + pthread_mutex_lock(&g_virtio_scsi_mutex); + + if (rc != 0) { + goto err_unlock; + } + + if (TAILQ_EMPTY(&g_virtio_scsi_devs)) { + goto out_unlock; + } + + /* Initialize all created devices and scan available targets */ + TAILQ_FOREACH(svdev, &g_virtio_scsi_devs, tailq) { + rc = virtio_scsi_dev_scan(svdev, bdev_virtio_initial_scan_complete, NULL); + if (rc != 0) { + goto err_unlock; + } + } + + pthread_mutex_unlock(&g_virtio_scsi_mutex); + return 0; + +err_unlock: + /* Remove any created devices */ + TAILQ_FOREACH_SAFE(svdev, &g_virtio_scsi_devs, tailq, next_svdev) { + virtio_scsi_dev_remove(svdev, NULL, NULL); + } + +out_unlock: + pthread_mutex_unlock(&g_virtio_scsi_mutex); + spdk_bdev_module_init_done(&virtio_scsi_if); + return rc; +} + +static void +_virtio_scsi_dev_unregister_cb(void *io_device) +{ + struct virtio_scsi_dev *svdev = io_device; + struct virtio_dev *vdev = &svdev->vdev; + bool finish_module; + bdev_virtio_remove_cb remove_cb; + void *remove_ctx; + + assert(spdk_ring_count(svdev->ctrlq_ring) == 0); + spdk_ring_free(svdev->ctrlq_ring); + spdk_poller_unregister(&svdev->mgmt_poller); + + virtio_dev_release_queue(vdev, VIRTIO_SCSI_EVENTQ); + virtio_dev_release_queue(vdev, VIRTIO_SCSI_CONTROLQ); + + virtio_dev_stop(vdev); + virtio_dev_destruct(vdev); + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_REMOVE(&g_virtio_scsi_devs, svdev, tailq); + pthread_mutex_unlock(&g_virtio_scsi_mutex); + + remove_cb = svdev->remove_cb; + remove_ctx = svdev->remove_ctx; + spdk_dma_free(svdev->eventq_ios); + free(svdev); + + if (remove_cb) { + remove_cb(remove_ctx, 0); + } + + finish_module = TAILQ_EMPTY(&g_virtio_scsi_devs); + + if (g_bdev_virtio_finish && finish_module) { + spdk_bdev_module_finish_done(); + } +} + +static void +virtio_scsi_dev_unregister_cb(void *io_device) +{ + struct virtio_scsi_dev *svdev = io_device; + struct spdk_thread *thread; + + thread = virtio_dev_queue_get_thread(&svdev->vdev, VIRTIO_SCSI_CONTROLQ); + spdk_thread_send_msg(thread, _virtio_scsi_dev_unregister_cb, io_device); +} + +static void +virtio_scsi_dev_remove(struct virtio_scsi_dev *svdev, + bdev_virtio_remove_cb cb_fn, void *cb_arg) +{ + struct virtio_scsi_disk *disk, *disk_tmp; + bool do_remove = true; + + if (svdev->removed) { + if (cb_fn) { + cb_fn(cb_arg, -EBUSY); + } + return; + } + + svdev->remove_cb = cb_fn; + svdev->remove_ctx = cb_arg; + svdev->removed = true; + + if (svdev->scan_ctx) { + /* The removal will continue after we receive a pending scan I/O. */ + return; + } + + TAILQ_FOREACH_SAFE(disk, &svdev->luns, link, disk_tmp) { + if (!disk->removed) { + spdk_bdev_unregister(&disk->bdev, NULL, NULL); + } + do_remove = false; + } + + if (do_remove) { + spdk_io_device_unregister(svdev, virtio_scsi_dev_unregister_cb); + } +} + +static void +bdev_virtio_finish(void) +{ + struct virtio_scsi_dev *svdev, *next; + + g_bdev_virtio_finish = true; + + pthread_mutex_lock(&g_virtio_scsi_mutex); + if (TAILQ_EMPTY(&g_virtio_scsi_devs)) { + pthread_mutex_unlock(&g_virtio_scsi_mutex); + spdk_bdev_module_finish_done(); + return; + } + + /* Defer module finish until all controllers are removed. */ + TAILQ_FOREACH_SAFE(svdev, &g_virtio_scsi_devs, tailq, next) { + virtio_scsi_dev_remove(svdev, NULL, NULL); + } + pthread_mutex_unlock(&g_virtio_scsi_mutex); +} + +int +bdev_virtio_user_scsi_dev_create(const char *base_name, const char *path, + unsigned num_queues, unsigned queue_size, + bdev_virtio_create_cb cb_fn, void *cb_arg) +{ + struct virtio_scsi_dev *svdev; + int rc; + + svdev = virtio_user_scsi_dev_create(base_name, path, num_queues, queue_size); + if (svdev == NULL) { + return -1; + } + + rc = virtio_scsi_dev_scan(svdev, cb_fn, cb_arg); + if (rc) { + virtio_scsi_dev_remove(svdev, NULL, NULL); + } + + return rc; +} + +struct bdev_virtio_pci_dev_create_ctx { + const char *name; + bdev_virtio_create_cb cb_fn; + void *cb_arg; +}; + +static int +bdev_virtio_pci_scsi_dev_create_cb(struct virtio_pci_ctx *pci_ctx, void *ctx) +{ + struct virtio_scsi_dev *svdev; + struct bdev_virtio_pci_dev_create_ctx *create_ctx = ctx; + int rc; + + svdev = virtio_pci_scsi_dev_create(create_ctx->name, pci_ctx); + if (svdev == NULL) { + return -1; + } + + rc = virtio_scsi_dev_scan(svdev, create_ctx->cb_fn, create_ctx->cb_arg); + if (rc) { + virtio_scsi_dev_remove(svdev, NULL, NULL); + } + + return rc; +} + +int +bdev_virtio_pci_scsi_dev_create(const char *name, struct spdk_pci_addr *pci_addr, + bdev_virtio_create_cb cb_fn, void *cb_arg) +{ + struct bdev_virtio_pci_dev_create_ctx create_ctx; + + create_ctx.name = name; + create_ctx.cb_fn = cb_fn; + create_ctx.cb_arg = cb_arg; + + return virtio_pci_dev_attach(bdev_virtio_pci_scsi_dev_create_cb, &create_ctx, + PCI_DEVICE_ID_VIRTIO_SCSI_MODERN, pci_addr); +} + +int +bdev_virtio_scsi_dev_remove(const char *name, bdev_virtio_remove_cb cb_fn, void *cb_arg) +{ + struct virtio_scsi_dev *svdev; + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_FOREACH(svdev, &g_virtio_scsi_devs, tailq) { + if (strcmp(svdev->vdev.name, name) == 0) { + break; + } + } + + if (svdev == NULL) { + pthread_mutex_unlock(&g_virtio_scsi_mutex); + SPDK_ERRLOG("Cannot find Virtio-SCSI device named '%s'\n", name); + return -ENODEV; + } + + virtio_scsi_dev_remove(svdev, cb_fn, cb_arg); + pthread_mutex_unlock(&g_virtio_scsi_mutex); + + return 0; +} + +void +bdev_virtio_scsi_dev_list(struct spdk_json_write_ctx *w) +{ + struct virtio_scsi_dev *svdev; + + spdk_json_write_array_begin(w); + + pthread_mutex_lock(&g_virtio_scsi_mutex); + TAILQ_FOREACH(svdev, &g_virtio_scsi_devs, tailq) { + spdk_json_write_object_begin(w); + + spdk_json_write_name(w, "name"); + spdk_json_write_string(w, svdev->vdev.name); + + virtio_dev_dump_json_info(&svdev->vdev, w); + + spdk_json_write_object_end(w); + } + pthread_mutex_unlock(&g_virtio_scsi_mutex); + + spdk_json_write_array_end(w); +} + +SPDK_LOG_REGISTER_COMPONENT("virtio", SPDK_LOG_VIRTIO) |