Re: [PATCH v2 2/2] virtio-blk: add zoned storage emulation for zoned devices

[Stefan Hajnoczi]

On Thu, Sep 29, 2022 at 05:48:21PM +0800, Sam Li wrote:
> This patch extends virtio-blk emulation to handle zoned device commands
> by calling the new block layer APIs to perform zoned device I/O on
> behalf of the guest. It supports Report Zone, four zone oparations (open,
> close, finish, reset), and Append Zone.
> 
> The VIRTIO_BLK_F_ZONED feature bit will only be set if the host does
> support zoned block devices. Regular block devices(conventional zones)
> will not be set.
> 
> The guest os having zoned device support can use blkzone(8) to test those
> commands. Furthermore, using zonefs to test zone append write is also
> supported.
> 
> Signed-off-by: Sam Li <faithilikerun@gmail.com>
> ---
>  hw/block/virtio-blk.c | 393 ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 393 insertions(+)
> 
> diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
> index e9ba752f6b..1c2535bfeb 100644
> --- a/hw/block/virtio-blk.c
> +++ b/hw/block/virtio-blk.c
> @@ -26,6 +26,9 @@
>  #include "hw/virtio/virtio-blk.h"
>  #include "dataplane/virtio-blk.h"
>  #include "scsi/constants.h"
> +#if defined(CONFIG_BLKZONED)
> +#include <linux/blkzoned.h>
> +#endif

Why is this Linux-specific header file included? The virtio-blk
emulation code should only use QEMU block layer APIs, not Linux APIs.

>  #ifdef __linux__
>  # include <scsi/sg.h>
>  #endif
> @@ -46,6 +49,8 @@ static const VirtIOFeature feature_sizes[] = {
>       .end = endof(struct virtio_blk_config, discard_sector_alignment)},
>      {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES,
>       .end = endof(struct virtio_blk_config, write_zeroes_may_unmap)},
> +    {.flags = 1ULL << VIRTIO_BLK_F_ZONED,
> +     .end = endof(struct virtio_blk_config, zoned)},
>      {}
>  };
>  
> @@ -614,6 +619,340 @@ err:
>      return err_status;
>  }
>  
> +typedef struct ZoneCmdData {
> +    VirtIOBlockReq *req;
> +    union {
> +        struct {
> +            unsigned int nr_zones;
> +            BlockZoneDescriptor *zones;
> +        } zone_report_data;
> +        struct {
> +            int64_t append_sector;
> +        } zone_append_data;
> +    };
> +} ZoneCmdData;
> +
> +/*
> + * check zoned_request: error checking before issuing requests. If all checks
> + * passed, return true.
> + * append: true if only zone append requests issued.
> + */
> +static bool check_zoned_request(VirtIOBlock *s, int64_t offset, int64_t len,
> +                             bool append, uint8_t *status) {
> +    BlockDriverState *bs = blk_bs(s->blk);
> +    int index = offset / bs->bl.zone_size;

This function doesn't check that offset+len is in the same zone as
offset. Maybe that's correct because some request types allow [offset,
offset+len) to cross zones?

> +
> +    if (offset < 0 || offset + len > bs->bl.capacity) {

Other cases that are not checked:
1. len < 0
2. offset >= bs->bl.capacity
3. len > bs->bl.capacity - offset (catches integer overflow)

It may be possible to combine these cases, but be careful about integer
overflow.

> +        *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        return false;
> +    }
> +
> +    if (!virtio_has_feature(s->host_features, VIRTIO_BLK_F_ZONED)) {
> +        *status = VIRTIO_BLK_S_UNSUPP;
> +        return false;
> +    }
> +
> +    if (append) {
> +        if ((offset % bs->bl.write_granularity) != 0) {
> +            *status = VIRTIO_BLK_S_ZONE_UNALIGNED_WP;
> +            return false;
> +        }
> +
> +        if (!BDRV_ZT_IS_SWR(bs->bl.wps->wp[index])) {
> +            *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +            return false;
> +        }

Where does the virtio-blk zone spec say that only SWR zones allow zone
append commands? Should it work for SWP zones too?

> +
> +        if (len / 512 > bs->bl.max_append_sectors) {
> +            if (bs->bl.max_append_sectors == 0) {
> +                *status = VIRTIO_BLK_S_UNSUPP;
> +            } else {
> +                *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +            }
> +            return false;
> +        }
> +    }
> +    return true;
> +}
> +
> +static void virtio_blk_zone_report_complete(void *opaque, int ret)
> +{
> +    ZoneCmdData *data = opaque;
> +    VirtIOBlockReq *req = data->req;
> +    VirtIOBlock *s = req->dev;
> +    VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
> +    struct iovec *in_iov = req->elem.in_sg;
> +    unsigned in_num = req->elem.in_num;
> +    int64_t zrp_size, nz, n, j = 0;
> +    int8_t err_status = VIRTIO_BLK_S_OK;
> +
> +    if (ret) {
> +        err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        goto out;
> +    }
> +
> +    nz = data->zone_report_data.nr_zones;
> +    struct virtio_blk_zone_report zrp_hdr = (struct virtio_blk_zone_report) {
> +            .nr_zones = cpu_to_le64(nz),
> +    };
> +
> +    zrp_size = sizeof(struct virtio_blk_zone_report)
> +               + sizeof(struct virtio_blk_zone_descriptor) * nz;
> +    n = iov_from_buf(in_iov, in_num, 0, &zrp_hdr, sizeof(zrp_hdr));
> +    if (n != sizeof(zrp_hdr)) {
> +        virtio_error(vdev, "Driver provided intput buffer that is too 
> small!");
> +        err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        goto out;
> +    }
> +
> +    for (size_t i = sizeof(zrp_hdr); i < zrp_size; i += sizeof(struct 
> virtio_blk_zone_descriptor), ++j) {
> +        struct virtio_blk_zone_descriptor desc =
> +                (struct virtio_blk_zone_descriptor) {
> +                        .z_start = 
> cpu_to_le64(data->zone_report_data.zones[j].start) >> BDRV_SECTOR_BITS,
> +                        .z_cap = 
> cpu_to_le64(data->zone_report_data.zones[j].cap) >> BDRV_SECTOR_BITS,
> +                        .z_wp = 
> cpu_to_le64(data->zone_report_data.zones[j].wp) >> BDRV_SECTOR_BITS,
> +                };
> +
> +        switch (data->zone_report_data.zones[j].type) {
> +        case BLK_ZT_CONV:
> +            desc.z_type = BLK_ZONE_TYPE_CONVENTIONAL;
> +            break;
> +        case BLK_ZT_SWR:
> +            desc.z_type = BLK_ZONE_TYPE_SEQWRITE_REQ;
> +            break;
> +        case BLK_ZT_SWP:
> +            desc.z_type = BLK_ZONE_TYPE_SEQWRITE_PREF;
> +            break;
> +        default:
> +            g_assert_not_reached();
> +        }
> +
> +        switch (data->zone_report_data.zones[j].cond) {
> +        case BLK_ZS_RDONLY:
> +            desc.z_state = BLK_ZONE_COND_READONLY;
> +            break;
> +        case BLK_ZS_OFFLINE:
> +            desc.z_state = BLK_ZONE_COND_OFFLINE;
> +            break;
> +        case BLK_ZS_EMPTY:
> +            desc.z_state = BLK_ZONE_COND_EMPTY;
> +            break;
> +        case BLK_ZS_CLOSED:
> +            desc.z_state = BLK_ZONE_COND_CLOSED;
> +            break;
> +        case BLK_ZS_FULL:
> +            desc.z_state = BLK_ZONE_COND_FULL;
> +            break;
> +        case BLK_ZS_EOPEN:
> +            desc.z_state = BLK_ZONE_COND_EXP_OPEN;
> +            break;
> +        case BLK_ZS_IOPEN:
> +            desc.z_state = BLK_ZONE_COND_IMP_OPEN;
> +            break;
> +        case BLK_ZS_NOT_WP:
> +            desc.z_state = BLK_ZONE_COND_NOT_WP;
> +            break;
> +        default:
> +            g_assert_not_reached();
> +            break;
> +        }
> +
> +        /* TODO: it takes O(n^2) time complexity. Optimizations required 
> here. */
> +        n = iov_from_buf(in_iov, in_num, i, &desc, sizeof(desc));
> +        if (n != sizeof(desc)) {
> +            virtio_error(vdev, "Driver provided input buffer "
> +                               "for descriptors that is too small!");
> +            err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +            goto out;
> +        }
> +    }
> +    goto out;
> +
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    g_free(data->zone_report_data.zones);
> +    g_free(data);
> +}
> +
> +static int virtio_blk_handle_zone_report(VirtIOBlockReq *req) {
> +    VirtIOBlock *s = req->dev;
> +    VirtIODevice *vdev = VIRTIO_DEVICE(s);
> +    unsigned int nr_zones;
> +    ZoneCmdData *data;
> +    int64_t zone_size, offset;
> +    uint8_t err_status;
> +
> +    if (req->in_len < sizeof(struct virtio_blk_inhdr) +
> +            sizeof(struct virtio_blk_zone_report) +
> +            sizeof(struct virtio_blk_zone_descriptor)) {
> +        virtio_error(vdev, "in buffer too small for zone report");
> +        return -1;
> +    }
> +
> +    /* start byte offset of the zone report */
> +    offset = virtio_ldq_p(vdev, &req->out.sector) * 512;
> +    if (!check_zoned_request(s, offset, 0, false, &err_status)) {
> +        goto out;
> +    }
> +
> +    nr_zones = (req->in_len - sizeof(struct virtio_blk_inhdr) -
> +                sizeof(struct virtio_blk_zone_report)) /
> +               sizeof(struct virtio_blk_zone_descriptor);
> +
> +    zone_size = sizeof(BlockZoneDescriptor) * nr_zones;
> +    data = g_malloc(sizeof(ZoneCmdData));
> +    data->req = req;
> +    data->zone_report_data.nr_zones = nr_zones;
> +    data->zone_report_data.zones = g_malloc(zone_size),
> +
> +    blk_aio_zone_report(s->blk, offset, &data->zone_report_data.nr_zones,
> +                        data->zone_report_data.zones,
> +                        virtio_blk_zone_report_complete, data);
> +    return 0;
> +
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    return err_status;
> +}
> +
> +static void virtio_blk_zone_mgmt_complete(void *opaque, int ret) {
> +    ZoneCmdData *data = opaque;
> +    VirtIOBlockReq *req = data->req;
> +    VirtIOBlock *s = req->dev;
> +    int8_t err_status = VIRTIO_BLK_S_OK;
> +
> +    if (ret) {
> +        err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        goto out;
> +    }
> +    goto out;
> +
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    g_free(data);
> +}
> +
> +static int virtio_blk_handle_zone_mgmt(VirtIOBlockReq *req, BlockZoneOp op) {
> +    VirtIOBlock *s = req->dev;
> +    VirtIODevice *vdev = VIRTIO_DEVICE(s);
> +    BlockDriverState *bs = blk_bs(s->blk);
> +    int64_t offset = virtio_ldq_p(vdev, &req->out.sector) * 512;
> +    uint64_t len;
> +    uint32_t type;
> +    uint8_t err_status = VIRTIO_BLK_S_OK;
> +
> +    if (!check_zoned_request(s, offset, 0, false, &err_status)) {
> +        goto out;
> +    }
> +
> +    ZoneCmdData *data = g_malloc(sizeof(ZoneCmdData));
> +    data->req = req;
> +
> +    type = virtio_ldl_p(vdev, &req->out.type);
> +    if (type == VIRTIO_BLK_T_ZONE_RESET_ALL) {
> +        /* Entire drive capacity */
> +        offset = 0;
> +        len = bs->bl.capacity;
> +    } else {
> +        if (bs->bl.zone_size * bs->bl.nr_zones == bs->bl.capacity) {
> +            len = bs->bl.zone_size;
> +        } else {
> +            /* when the SWR drive has one last small zone, calculate its len 
> */
> +            len = bs->bl.capacity - bs->bl.zone_size * (bs->bl.nr_zones - 1);
> +        }
> +        if (offset + len > bs->bl.capacity) {
> +            err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +            goto out;

data is leaked here.

> +        }
> +    }
> +
> +    blk_aio_zone_mgmt(s->blk, op, offset, len,
> +                      virtio_blk_zone_mgmt_complete, data);
> +
> +    return 0;
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    return err_status;
> +}
> +
> +static void virtio_blk_zone_append_complete(void *opaque, int ret) {
> +    ZoneCmdData *data = opaque;
> +    VirtIOBlockReq *req = data->req;
> +    VirtIOBlock *s = req->dev;
> +    VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
> +    int64_t append_sector, n;
> +    struct iovec *out_iov = req->elem.out_sg;
> +    unsigned out_num = req->elem.out_num;
> +    uint8_t err_status = VIRTIO_BLK_S_OK;
> +
> +    if (ret) {
> +        err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        goto out;
> +    }
> +
> +    virtio_stl_p(vdev, &append_sector, data->zone_append_data.append_sector);
> +    n = iov_to_buf(out_iov, out_num, 0, &append_sector, 
> sizeof(append_sector));

out_iov contains the driver->device buffers. The device is only allowed
to read from out_iov, not write to it.

The device->driver buffers are in in_iov.

According to the spec the zone append in hdr looks like this:

  struct {
      u8 status;
      u8 reserved[7];
      le64 append_sector;
  } virtio_blk_zone_append_inhdr;

In virtio_blk_handle_request() we used iov_discard_back_undoable() to
take the last byte (the status field for non-zone append requests) from
in_iov[]. This is incorrect for zone append requests because they have
the larger struct zone_append_inhdr instead of struct
virtio_blk_inhdr.

I think it might be time to stop using req->in in virtio-blk.c and
instead use iov_from_buf() to write the status byte. For zone append
requests we also need to write reserved[] and append_sector:

  iov_discard_undo(&req->inhdr_undo);
  inhdr_len = is_zone_append ?
               sizeof(struct virtio_blk_zone_append_inhdr) :
               sizeof(struct virtio_blk_inhdr);
  iov_from_buf(req->elem.in_sg, req->elem.in_num,
               req->in_len - inhdr_len,
               &req->in, inhdr_len);

where req->in changes to:

  union {
      struct virtio_blk_inhdr inhdr;
      struct virtio_blk_zone_append_inhdr zone_append_inhdr;
  } in;

Most requests will just use in.inhdr but zone append will fill out the
full in.zone_append_inhdr struct.

> +    if (n != sizeof(append_sector)) {
> +        virtio_error(vdev, "Driver provided input buffer less than size of "
> +                     "append_sector");
> +        err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
> +        goto out;
> +    }
> +    goto out;
> +
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    g_free(data);
> +}
> +
> +static int virtio_blk_handle_zone_append(VirtIOBlockReq *req) {
> +    VirtIOBlock *s = req->dev;
> +    VirtIODevice *vdev = VIRTIO_DEVICE(s);
> +    uint64_t niov = req->elem.out_num;
> +    struct iovec *out_iov = req->elem.out_sg;
> +    uint8_t err_status = VIRTIO_BLK_S_OK;
> +
> +    int64_t offset = virtio_ldq_p(vdev, &req->out.sector) * 512;
> +    int64_t len = 0;
> +    for (int i = 1; i < niov; ++i) {
> +        len += out_iov[i].iov_len;

Please pass in out_iov and out_num instead of using req->elem.out_sg and
req->elem.out_num. virtio_blk_handle_request() modifies the iovecs
pointed to by req->elem.out_sg using iov_discard_front_undoable() and it
is not safe to access req->elem.out_sg directly.

Also, VIRTIO devices are not allowed to make assumptions about the iovec
layout. That means skipping the first iovec in the for loop violates the
spec. The driver could send struct virtio_blk_req as two or more iovecs
instead of putting it into just 1 iovec. This is why the device is not
allowed to assume out_iov[0] is struct virtio_blk_req.

The for loop can be replaced with:

  len = iov_size(out_iov, out_num);

and out_iov[1]/niov-1 can be replaced with just out_iov and out_num (if
you pass them in from virtio_blk_handle_request()).

> +    }
> +
> +    if (!check_zoned_request(s, offset, len, true, &err_status)) {
> +        goto out;
> +    }
> +
> +    ZoneCmdData *data = g_malloc(sizeof(ZoneCmdData));
> +    data->req = req;
> +    data->zone_append_data.append_sector = offset;
> +    qemu_iovec_init_external(&req->qiov, &out_iov[1], niov-1);
> +    blk_aio_zone_append(s->blk, &data->zone_append_data.append_sector, 
> &req->qiov, 0,
> +                        virtio_blk_zone_append_complete, data);
> +    return 0;
> +
> +out:
> +    aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
> +    virtio_blk_req_complete(req, err_status);
> +    virtio_blk_free_request(req);
> +    aio_context_release(blk_get_aio_context(s->conf.conf.blk));
> +    return err_status;
> +}
> +
>  static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer 
> *mrb)
>  {
>      uint32_t type;
> @@ -700,6 +1039,24 @@ static int virtio_blk_handle_request(VirtIOBlockReq 
> *req, MultiReqBuffer *mrb)
>      case VIRTIO_BLK_T_FLUSH:
>          virtio_blk_handle_flush(req, mrb);
>          break;
> +    case VIRTIO_BLK_T_ZONE_REPORT:
> +        virtio_blk_handle_zone_report(req);
> +        break;
> +    case VIRTIO_BLK_T_ZONE_OPEN:
> +        virtio_blk_handle_zone_mgmt(req, BLK_ZO_OPEN);
> +        break;
> +    case VIRTIO_BLK_T_ZONE_CLOSE:
> +        virtio_blk_handle_zone_mgmt(req, BLK_ZO_CLOSE);
> +        break;
> +    case VIRTIO_BLK_T_ZONE_FINISH:
> +        virtio_blk_handle_zone_mgmt(req, BLK_ZO_FINISH);
> +        break;
> +    case VIRTIO_BLK_T_ZONE_RESET:
> +        virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET);
> +        break;
> +    case VIRTIO_BLK_T_ZONE_RESET_ALL:
> +        virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET_ALL);
> +        break;
>      case VIRTIO_BLK_T_SCSI_CMD:
>          virtio_blk_handle_scsi(req);
>          break;
> @@ -718,6 +1075,9 @@ static int virtio_blk_handle_request(VirtIOBlockReq 
> *req, MultiReqBuffer *mrb)
>          virtio_blk_free_request(req);
>          break;
>      }
> +   case VIRTIO_BLK_T_ZONE_APPEND & ~VIRTIO_BLK_T_OUT:

Indentation is off. QEMU uses 4-space indentation.

> +       virtio_blk_handle_zone_append(req);
> +       break;
>      /*
>       * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
>       * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch 
> statement,
> @@ -917,6 +1277,7 @@ static void virtio_blk_update_config(VirtIODevice *vdev, 
> uint8_t *config)
>  {
>      VirtIOBlock *s = VIRTIO_BLK(vdev);
>      BlockConf *conf = &s->conf.conf;
> +    BlockDriverState *bs = blk_bs(s->blk);
>      struct virtio_blk_config blkcfg;
>      uint64_t capacity;
>      int64_t length;
> @@ -976,6 +1337,30 @@ static void virtio_blk_update_config(VirtIODevice 
> *vdev, uint8_t *config)
>          blkcfg.write_zeroes_may_unmap = 1;
>          virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
>      }
> +    if (bs->bl.zoned != BLK_Z_NONE) {
> +        switch (bs->bl.zoned) {
> +        case BLK_Z_HM:
> +            blkcfg.zoned.model = VIRTIO_BLK_Z_HM;
> +            break;
> +        case BLK_Z_HA:
> +            blkcfg.zoned.model = VIRTIO_BLK_Z_HA;
> +            break;
> +        default:
> +            g_assert_not_reached();
> +        }
> +
> +        virtio_stl_p(vdev, &blkcfg.zoned.zone_sectors,
> +                     bs->bl.zone_size / 512);
> +        virtio_stl_p(vdev, &blkcfg.zoned.max_active_zones,
> +                     bs->bl.max_active_zones);
> +        virtio_stl_p(vdev, &blkcfg.zoned.max_open_zones,
> +                     bs->bl.max_open_zones);
> +        virtio_stl_p(vdev, &blkcfg.zoned.write_granularity, blk_size);
> +        virtio_stl_p(vdev, &blkcfg.zoned.max_append_sectors,
> +                     bs->bl.max_append_sectors);
> +    } else {
> +        blkcfg.zoned.model = VIRTIO_BLK_Z_NONE;
> +    }
>      memcpy(config, &blkcfg, s->config_size);
>  }
>  
> @@ -1140,6 +1525,7 @@ static void virtio_blk_device_realize(DeviceState *dev, 
> Error **errp)
>      VirtIODevice *vdev = VIRTIO_DEVICE(dev);
>      VirtIOBlock *s = VIRTIO_BLK(dev);
>      VirtIOBlkConf *conf = &s->conf;
> +    BlockDriverState *bs = blk_bs(conf->conf.blk);
>      Error *err = NULL;
>      unsigned i;
>  
> @@ -1185,6 +1571,13 @@ static void virtio_blk_device_realize(DeviceState 
> *dev, Error **errp)
>          return;
>      }
>  
> +    if (bs->bl.zoned != BLK_Z_NONE) {
> +        virtio_add_feature(&s->host_features, VIRTIO_BLK_F_ZONED);
> +        if (bs->bl.zoned == BLK_Z_HM) {
> +            virtio_clear_feature(&s->host_features, VIRTIO_BLK_F_DISCARD);
> +        }
> +    }
> +
>      if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
>          (!conf->max_discard_sectors ||
>           conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
> -- 
> 2.37.3
>

signature.asc
Description: PGP signature