system/uapp/disk-pave/pave-lib.cpp

// Copyright 2018 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include <dirent.h>
#include <fcntl.h>
#include <stdbool.h>
#include <stddef.h>
#include <string.h>

#include <block-client/cpp/client.h>
#include <crypto/bytes.h>
#include <fbl/algorithm.h>
#include <fbl/array.h>
#include <fbl/auto_call.h>
#include <fbl/unique_fd.h>
#include <fbl/vector.h>
#include <fs-management/fvm.h>
#include <fs-management/mount.h>
#include <fs-management/ramdisk.h>
#include <fvm/fvm-lz4.h>
#include <fvm/fvm-sparse.h>
#include <lib/cksum.h>
#include <lib/fzl/mapped-vmo.h>
#include <lib/zx/fifo.h>
#include <lib/zx/vmo.h>
#include <zircon/boot/image.h>
#include <zircon/device/block.h>
#include <zircon/device/device.h>
#include <zircon/status.h>
#include <zircon/syscalls.h>
#include <zxcrypt/volume.h>

#include "fvm/fvm-sparse.h"
#include "fvm/fvm.h"
#include "pave-lib.h"
#include "pave-logging.h"

namespace paver {
namespace {

// Confirm that the file descriptor to the underlying partition exists within an
// FVM, not, for example, a GPT or MBR.
//
// |out| is true if |fd| is a VPartition, else false.
zx_status_t FvmIsVirtualPartition(const fbl::unique_fd& fd, bool* out) {
    char path[PATH_MAX];
    const ssize_t r = ioctl_device_get_topo_path(fd.get(), path, sizeof(path));
    if (r < 0) {
        return ZX_ERR_IO;
    }

    *out = strstr(path, "fvm") != nullptr;
    return ZX_OK;
}

// Describes the state of a partition actively being written
// out to disk.
struct PartitionInfo {
    fvm::partition_descriptor_t* pd;
    fbl::unique_fd new_part;
    fbl::unique_fd old_part; // Or '-1' if this is a new partition
};

inline fvm::extent_descriptor_t* GetExtent(fvm::partition_descriptor_t* pd, size_t extent) {
    return reinterpret_cast<fvm::extent_descriptor_t*>(
        reinterpret_cast<uintptr_t>(pd) + sizeof(fvm::partition_descriptor_t) +
        extent * sizeof(fvm::extent_descriptor_t));
}

// Registers a FIFO
zx_status_t RegisterFastBlockIo(const fbl::unique_fd& fd, zx_handle_t vmo,
                                vmoid_t* vmoid_out, block_client::Client* client_out) {
    zx::fifo fifo;
    if (ioctl_block_get_fifos(fd.get(), fifo.reset_and_get_address()) < 0) {
        ERROR("Couldn't attach fifo to partition\n");
        return ZX_ERR_IO;
    }
    zx::vmo dup;
    if (zx_handle_duplicate(vmo, ZX_RIGHT_SAME_RIGHTS,
                            dup.reset_and_get_address()) != ZX_OK) {
        ERROR("Couldn't duplicate buffer vmo\n");
        return ZX_ERR_IO;
    }
    zx_handle_t h = dup.release();
    if (ioctl_block_attach_vmo(fd.get(), &h, vmoid_out) < 0) {
        ERROR("Couldn't attach VMO\n");
        return ZX_ERR_IO;
    }
    return block_client::Client::Create(fbl::move(fifo), client_out);
}

// Stream an FVM partition to disk.
zx_status_t StreamFvmPartition(fvm::SparseReader* reader, PartitionInfo* part,
                               fzl::MappedVmo* mvmo, const block_client::Client& client, size_t block_size,
                               block_fifo_request_t* request) {
    size_t slice_size = reader->Image()->slice_size;
    const size_t vmo_cap = mvmo->GetSize();
    for (size_t e = 0; e < part->pd->extent_count; e++) {
        LOG("Writing extent %zu... \n", e);
        fvm::extent_descriptor_t* ext = GetExtent(part->pd, e);
        size_t offset = ext->slice_start * slice_size;
        size_t bytes_left = ext->extent_length;

        // Write real data
        while (bytes_left > 0) {
            size_t vmo_sz = 0;
            size_t actual;
            zx_status_t status = reader->ReadData(
                &reinterpret_cast<uint8_t*>(mvmo->GetData())[vmo_sz],
                fbl::min(bytes_left, vmo_cap - vmo_sz), &actual);
            vmo_sz += actual;
            bytes_left -= actual;

            if (vmo_sz == 0) {
                ERROR("Read nothing from src_fd; %zu bytes left\n", bytes_left);
                return ZX_ERR_IO;
            } else if (vmo_sz % block_size != 0) {
                ERROR("Cannot write non-block size multiple: %zu\n", vmo_sz);
                return ZX_ERR_IO;
            } else if (status != ZX_OK) {
                ERROR("Error reading partition data\n");
                return status;
            }

            uint64_t length = vmo_sz / block_size;
            if (length > UINT32_MAX) {
                ERROR("Error writing partition: Too large\n");
                return ZX_ERR_OUT_OF_RANGE;
            }
            request->length = static_cast<uint32_t>(length);
            request->vmo_offset = 0;
            request->dev_offset = offset / block_size;

            ssize_t r;
            if ((r = client.Transaction(request, 1)) != ZX_OK) {
                ERROR("Error writing partition data\n");
                return static_cast<zx_status_t>(r);
            }

            offset += vmo_sz;
        }

        // Write trailing zeroes (which are implied, but were omitted from
        // transfer).
        bytes_left = (ext->slice_count * slice_size) - ext->extent_length;
        if (bytes_left > 0) {
            LOG("%zu bytes written, %zu zeroes left\n", ext->extent_length, bytes_left);
            memset(mvmo->GetData(), 0, vmo_cap);
        }
        while (bytes_left > 0) {
            uint64_t length = fbl::min(bytes_left, vmo_cap) / block_size;
            if (length > UINT32_MAX) {
                ERROR("Error writing trailing zeroes: Too large\n");
                return ZX_ERR_OUT_OF_RANGE;
            }
            request->length = static_cast<uint32_t>(length);
            request->vmo_offset = 0;
            request->dev_offset = offset / block_size;

            zx_status_t status;
            if ((status = client.Transaction(request, 1)) != ZX_OK) {
                ERROR("Error writing trailing zeroes\n");
                return status;
            }

            offset += request->length * block_size;
            bytes_left -= request->length * block_size;
        }
    }
    return ZX_OK;
}

// Stream a raw (non-FVM) partition to a vmo.
zx_status_t StreamPayloadToVmo(fzl::MappedVmo* mvmo, const fbl::unique_fd& src_fd,
                               const block_info_t& info, size_t* payload_size) {
    zx_status_t status;
    ssize_t r;
    size_t vmo_offset = 0;

    while ((r = read(src_fd.get(), &reinterpret_cast<uint8_t*>(mvmo->GetData())[vmo_offset],
                     mvmo->GetSize() - vmo_offset)) > 0) {
        vmo_offset += r;
        if (mvmo->GetSize() - vmo_offset == 0) {
            // The buffer is full, let's grow the VMO.
            if ((status = mvmo->Grow(mvmo->GetSize() << 1)) != ZX_OK) {
                ERROR("Failed to grow VMO\n");
                return status;
            }
        }
    }

    if (r < 0) {
        ERROR("Error reading partition data\n");
        return static_cast<zx_status_t>(r);
    }

    if (vmo_offset % info.block_size) {
        // We have a partial block to write.
        const size_t rounded_length = fbl::round_up(vmo_offset, info.block_size);
        memset(&reinterpret_cast<uint8_t*>(mvmo->GetData())[vmo_offset], 0,
               rounded_length - vmo_offset);
        vmo_offset = rounded_length;
    }
    *payload_size = vmo_offset;
    return ZX_OK;
}

// Writes a raw (non-FVM) partition to a block device from a VMO.
zx_status_t WriteVmoToBlock(fzl::MappedVmo* mvmo, size_t vmo_size,
                            const fbl::unique_fd& partition_fd, const block_info_t& info) {
    ZX_ASSERT(vmo_size % info.block_size == 0);

    vmoid_t vmoid;
    block_client::Client client;
    zx_status_t status = RegisterFastBlockIo(partition_fd, mvmo->GetVmo(), &vmoid, &client);
    if (status != ZX_OK) {
        ERROR("Cannot register fast block I/O\n");
        return status;
    }

    block_fifo_request_t request;
    request.group = 0;
    request.vmoid = vmoid;
    request.opcode = BLOCKIO_WRITE;

    uint64_t length = vmo_size / info.block_size;
    if (length > UINT32_MAX) {
        ERROR("Error writing partition data: Too large\n");
        return ZX_ERR_OUT_OF_RANGE;
    }
    request.length = static_cast<uint32_t>(length);
    request.vmo_offset = 0;
    request.dev_offset = 0;

    if ((status = client.Transaction(&request, 1)) != ZX_OK) {
        ERROR("Error writing partition data: %d\n", status);
        return status;
    }
    return ZX_OK;
}
// Checks first few bytes of buffer to ensure it is a ZBI.
// Also validates architecture in kernel header matches the target.
bool ValidateKernelZbi(const uint8_t* buffer, size_t size, Arch arch) {
    const auto payload = reinterpret_cast<const zircon_kernel_t*>(buffer);
    const uint32_t expected_kernel = (arch == Arch::X64) ? ZBI_TYPE_KERNEL_X64
                                                         : ZBI_TYPE_KERNEL_ARM64;

    const auto crc_valid = [](const zbi_header_t* hdr) {
        const uint32_t crc = crc32(0, reinterpret_cast<const uint8_t*>(hdr + 1),
                                   hdr->length);
        return hdr->crc32 == crc;
    };

    return size >= sizeof(zircon_kernel_t) &&
           // Container header
           payload->hdr_file.type == ZBI_TYPE_CONTAINER &&
           payload->hdr_file.extra == ZBI_CONTAINER_MAGIC &&
           (payload->hdr_file.length - offsetof(zircon_kernel_t, hdr_kernel)) <= size &&
           payload->hdr_file.magic == ZBI_ITEM_MAGIC &&
           payload->hdr_file.flags == ZBI_FLAG_VERSION &&
           payload->hdr_file.crc32 == ZBI_ITEM_NO_CRC32 &&
           // Kernel header
           payload->hdr_kernel.type == expected_kernel &&
           (payload->hdr_kernel.length - offsetof(zircon_kernel_t, data_kernel)) <= size &&
           payload->hdr_kernel.magic == ZBI_ITEM_MAGIC &&
           (payload->hdr_kernel.flags & ZBI_FLAG_VERSION) == ZBI_FLAG_VERSION &&
           ((payload->hdr_kernel.flags & ZBI_FLAG_CRC32)
                ? crc_valid(&payload->hdr_kernel)
                : payload->hdr_kernel.crc32 == ZBI_ITEM_NO_CRC32);
}

// Parses a partition and validates that it matches the expected format.
zx_status_t ValidateKernelPayload(fzl::MappedVmo* mvmo, size_t vmo_size, Partition partition_type,
                                  Arch arch) {
    const auto* buffer = reinterpret_cast<uint8_t*>(mvmo->GetData());
    switch (partition_type) {
    case Partition::kZirconA:
    case Partition::kZirconB:
    case Partition::kZirconR:
        if (!ValidateKernelZbi(buffer, vmo_size, arch)) {
            ERROR("Invalid ZBI payload!");
            return ZX_ERR_BAD_STATE;
        }
        break;

    default:
        // TODO(surajmalhotra): Validate non-zbi payloads as well.
        LOG("Skipping validation as payload is not a ZBI\n");
        break;
    }

    return ZX_OK;
}

// Attempt to bind an FVM driver to a partition fd.
fbl::unique_fd TryBindToFvmDriver(const fbl::unique_fd& partition_fd,
                                  zx::duration timeout) {
    char path[PATH_MAX];
    ssize_t r = ioctl_device_get_topo_path(partition_fd.get(), path, sizeof(path));
    if (r < 0) {
        ERROR("Failed to get topological path\n");
        return fbl::unique_fd();
    }

    constexpr char kFvmDriverLib[] = "/boot/driver/fvm.so";
    r = ioctl_device_bind(partition_fd.get(), kFvmDriverLib, sizeof(kFvmDriverLib));
    if (r < 0) {
        ERROR("Could not bind fvm driver\n");
        return fbl::unique_fd();
    }

    char fvm_path[PATH_MAX];
    snprintf(fvm_path, sizeof(fvm_path), "%s/fvm", path);
    if (wait_for_device(fvm_path, timeout.get()) != ZX_OK) {
        ERROR("Error waiting for fvm driver to bind\n");
        return fbl::unique_fd();
    }
    return fbl::unique_fd(open(fvm_path, O_RDWR));
}

// Formats the FVM within the provided partition if it is not already formatted.
//
// On success, returns a file descriptor to an FVM.
// On failure, returns -1
fbl::unique_fd FvmPartitionFormat(fbl::unique_fd partition_fd, size_t slice_size) {
    // Although the format (based on the magic in the FVM superblock)
    // indicates this is (or at least was) an FVM image, it may be invalid.
    //
    // Attempt to bind the FVM driver to this partition, but fall-back to
    // reinitializing the FVM image so the rest of the paving
    // process can continue successfully.
    disk_format_t df = detect_disk_format(partition_fd.get());
    if (df == DISK_FORMAT_FVM) {
        fbl::unique_fd fvm_fd = TryBindToFvmDriver(partition_fd, zx::sec(3));
        if (fvm_fd) {
            LOG("Found already formatted FVM.\n");
            return fvm_fd;
        } else {
            ERROR("Saw DISK_FORMAT_FVM, but could not bind driver\n");
        }
    }
    ERROR("Initializing partition as FVM\n");
    zx_status_t status;
    if ((status = fvm_init(partition_fd.get(), slice_size)) != ZX_OK) {
        ERROR("Failed to initialize fvm: %d\n", status);
        return fbl::unique_fd();
    }
    return TryBindToFvmDriver(partition_fd, zx::sec(3));
}

// Formats a block device as a zxcrypt volume.
//
// On success, returns a file descriptor to an FVM.
// On failure, returns -1
zx_status_t ZxcryptCreate(PartitionInfo* part) {
    zx_status_t status;

    char path[PATH_MAX];
    ssize_t r;
    if ((r = ioctl_device_get_topo_path(part->new_part.get(), path, sizeof(path))) < 0) {
        status = static_cast<zx_status_t>(r);
        ERROR("Failed to get topological path\n");
        return status;
    }
    // TODO(security): ZX-1130. We need to bind with channel in order to pass a key here.
    // TODO(security): ZX-1864. The created volume must marked as needing key rotation.
    crypto::Secret key;
    uint8_t *tmp;
    if((status = key.Allocate(zxcrypt::kZx1130KeyLen, &tmp)) != ZX_OK) {
        return status;
    }
    memset(tmp, 0, key.len());

    fbl::unique_ptr<zxcrypt::Volume> volume;
    if ((status = zxcrypt::Volume::Create(fbl::move(part->new_part), key, &volume)) != ZX_OK ||
        (status = volume->Open(zx::sec(3), &part->new_part)) != ZX_OK) {
        ERROR("Could not create zxcrypt volume\n");
        return status;
    }

    fvm::extent_descriptor_t* ext = GetExtent(part->pd, 0);
    size_t reserved = volume->reserved_slices();

    // |Create| guarantees at least |reserved| + 1 slices are allocated.  If the first extent had a
    // single slice, we're done.
    size_t allocated = fbl::max(reserved + 1, ext->slice_count);
    size_t needed = reserved + ext->slice_count;
    if (allocated >= needed) {
        return ZX_OK;
    }

    // Otherwise, extend by the number of slices we stole for metadata
    extend_request_t req;
    req.offset = allocated - reserved;
    req.length = needed - allocated;

    if ((r = ioctl_block_fvm_extend(part->new_part.get(), &req)) < 0) {
        status = static_cast<zx_status_t>(r);
        ERROR("Failed to extend zxcrypt volume: %s\n", zx_status_get_string(status));
        return status;
    }

    return ZX_OK;
}

// Returns |ZX_OK| if |partition_fd| is a child of |fvm_fd|.
zx_status_t FvmPartitionIsChild(const fbl::unique_fd& fvm_fd, const fbl::unique_fd& partition_fd) {
    char fvm_path[PATH_MAX];
    char part_path[PATH_MAX];
    ssize_t r;
    if ((r = ioctl_device_get_topo_path(fvm_fd.get(), fvm_path, sizeof(fvm_path))) < 0) {
        ERROR("Couldn't get topological path of FVM\n");
        return static_cast<zx_status_t>(r);
    } else if ((r = ioctl_device_get_topo_path(partition_fd.get(), part_path,
                                               sizeof(part_path))) < 0) {
        ERROR("Couldn't get topological path of partition\n");
        return static_cast<zx_status_t>(r);
    }
    if (strncmp(fvm_path, part_path, strlen(fvm_path))) {
        ERROR("Partition does not exist within FVM\n");
        return ZX_ERR_BAD_STATE;
    }
    return ZX_OK;
}

void RecommendWipe(const char* reason) {
    ERROR("-----------------------------------------------------\n");
    ERROR("\n");
    ERROR("%s: Please run 'install-disk-image wipe' to wipe your partitions\n", reason);
    ERROR("\n");
    ERROR("-----------------------------------------------------\n");
}

zx_status_t FvmInitSparseReader(fbl::unique_fd src_fd, fbl::unique_ptr<fvm::SparseReader>* reader) {
    zx_status_t status;
    if ((status = fvm::SparseReader::Create(fbl::move(src_fd), reader)) != ZX_OK) {
        return status;
    }

    fvm::sparse_image_t* hdr = (*reader)->Image();
    // Verify the header, then allocate and stream the remaining metadata
    if (hdr->magic != fvm::kSparseFormatMagic) {
        ERROR("Bad magic\n");
        return ZX_ERR_IO;
    } else if (hdr->version != fvm::kSparseFormatVersion) {
        ERROR("Unexpected sparse file version\n");
        return ZX_ERR_IO;
    }

    return ZX_OK;
}

// Given an fd representing a "sparse FVM format", fill the FVM with the
// provided partitions described by |src_fd|.
//
// Decides to overwrite or create new partitions based on the type
// GUID, not the instance GUID.
zx_status_t FvmStreamPartitions(fbl::unique_fd partition_fd, fbl::unique_fd src_fd) {
    fbl::unique_ptr<fvm::SparseReader> reader;
    zx_status_t status = FvmInitSparseReader(fbl::move(src_fd), &reader);
    if (status != ZX_OK) {
        return status;
    }

    LOG("Header Validated - OK\n");

    fvm::sparse_image_t* hdr = reader->Image();
    // Acquire an fd to the fvm, either by finding one that already
    // exists, or formatting a new one.
    fbl::unique_fd fvm_fd(FvmPartitionFormat(fbl::move(partition_fd), hdr->slice_size));
    if (!fvm_fd) {
        ERROR("Couldn't find FVM partition\n");
        return ZX_ERR_IO;
    }

    // TODO(smklein): In this case, we could actually unbind the FVM driver,
    // create a new FVM with the updated slice size, and rebind.

    fvm_info_t info;
    if (ioctl_block_fvm_query(fvm_fd.get(), &info) < 0) {
        ERROR("Couldn't query underlying FVM\n");
        return ZX_ERR_IO;
    } else if (info.slice_size != hdr->slice_size) {
        ERROR("Unexpected slice size (%zu vs %zu)\n", info.slice_size, hdr->slice_size);
        return ZX_ERR_IO;
    }

    fbl::Array<PartitionInfo> parts(new PartitionInfo[hdr->partition_count],
                                    hdr->partition_count);

    size_t block_size = 0;
    fvm::partition_descriptor_t* part = reader->Partitions();

    for (size_t p = 0; p < hdr->partition_count; p++) {
        parts[p].pd = part;
        parts[p].old_part.reset(open_partition(nullptr, part->type, ZX_SEC(2), nullptr));
        if (parts[p].pd->magic != fvm::kPartitionDescriptorMagic) {
            ERROR("Bad partition magic\n");
            return ZX_ERR_IO;
        }

        if (parts[p].old_part) {
            bool is_vpartition;
            if (FvmIsVirtualPartition(parts[p].old_part, &is_vpartition) != ZX_OK) {
                ERROR("Couldn't confirm old vpartition type\n");
                return ZX_ERR_IO;
            }
            if (FvmPartitionIsChild(fvm_fd, parts[p].old_part) != ZX_OK) {
                RecommendWipe("Streaming a partition type which also exists outside FVM");
                return ZX_ERR_BAD_STATE;
            }
            if (!is_vpartition) {
                RecommendWipe("Streaming a partition type which also exists in a GPT");
                return ZX_ERR_BAD_STATE;
            }
        }

        fvm::extent_descriptor_t* ext = GetExtent(part, 0);
        if (ext->magic != fvm::kExtentDescriptorMagic) {
            ERROR("Bad extent magic\n");
            return ZX_ERR_IO;
        }
        if (ext->slice_start != 0) {
            ERROR("First slice must start at zero\n");
            return ZX_ERR_IO;
        }
        if (ext->slice_count == 0) {
            ERROR("Extents must have > 0 slices\n");
            return ZX_ERR_IO;
        }
        if (ext->extent_length > ext->slice_count * hdr->slice_size) {
            ERROR("Extent length must fit within allocated slice count\n");
            return ZX_ERR_IO;
        }

        alloc_req_t alloc;
        // Allocate this partition as inactive so it gets deleted on the next
        // reboot if this stream fails.
        alloc.flags = fvm::kVPartFlagInactive;
        alloc.slice_count = ext->slice_count;
        memcpy(&alloc.type, parts[p].pd->type, sizeof(alloc.type));
        zx_cprng_draw(alloc.guid, GPT_GUID_LEN);
        memcpy(&alloc.name, parts[p].pd->name, sizeof(alloc.name));
        LOG("Allocating partition %s consisting of %zu slices\n", alloc.name, alloc.slice_count);
        parts[p].new_part.reset(fvm_allocate_partition(fvm_fd.get(), &alloc));
        if (!parts[p].new_part) {
            ERROR("Couldn't allocate partition\n");
            return ZX_ERR_BAD_STATE;
        }

        // Add filter drivers
        if ((part->flags & fvm::kSparseFlagZxcrypt) != 0) {
            LOG("Creating zxcrypt volume\n");
            if ((status = ZxcryptCreate(&parts[p])) != ZX_OK) {
                return status;
            }
        }

        if (block_size == 0) {
            block_info_t binfo;
            if ((ioctl_block_get_info(parts[p].new_part.get(), &binfo)) < 0) {
                ERROR("Couldn't get partition block info\n");
                return ZX_ERR_IO;
            }
            block_size = binfo.block_size;
        }

        for (size_t e = 1; e < parts[p].pd->extent_count; e++) {
            ext = GetExtent(parts[p].pd, e);
            if (ext->magic != fvm::kExtentDescriptorMagic) {
                ERROR("Bad extent magic\n");
                return ZX_ERR_IO;
            } else if (ext->slice_count == 0) {
                ERROR("Extents must have > 0 slices\n");
                return ZX_ERR_IO;
            } else if (ext->extent_length > ext->slice_count * hdr->slice_size) {
                ERROR("Extent must fit within allocated slice count\n");
                return ZX_ERR_IO;
            }

            extend_request_t request;
            request.offset = ext->slice_start;
            request.length = ext->slice_count;
            LOG("Extending partition[%zu] at offset %zu by length %zu\n", p, request.offset,
                request.length);
            if (ioctl_block_fvm_extend(parts[p].new_part.get(), &request) < 0) {
                ERROR("Failed to extend partition\n");
                return ZX_ERR_BAD_STATE;
            }
        }
        part = reinterpret_cast<fvm::partition_descriptor*>(
            reinterpret_cast<uintptr_t>(ext) + sizeof(fvm::extent_descriptor_t));
    }

    LOG("Partition space pre-allocated\n");

    constexpr size_t vmo_size = 1 << 20;

    fbl::unique_ptr<fzl::MappedVmo> mvmo;
    if ((status = fzl::MappedVmo::Create(vmo_size, "fvm-stream", &mvmo)) != ZX_OK) {
        ERROR("Failed to create stream VMO\n");
        return ZX_ERR_NO_MEMORY;
    }

    // Now that all partitions are preallocated, begin streaming data to them.
    for (size_t p = 0; p < hdr->partition_count; p++) {
        vmoid_t vmoid;
        block_client::Client client;
        zx_status_t status = RegisterFastBlockIo(parts[p].new_part,
                                                 mvmo->GetVmo(), &vmoid, &client);
        if (status != ZX_OK) {
            ERROR("Failed to register fast block IO\n");
            return status;
        }

        block_fifo_request_t request;
        request.group = 0;
        request.vmoid = vmoid;
        request.opcode = BLOCKIO_WRITE;

        LOG("Streaming partition %zu\n", p);
        status = StreamFvmPartition(reader.get(), &parts[p], mvmo.get(), client, block_size,
                                    &request);
        LOG("Done streaming partition %zu\n", p);
        if (status != ZX_OK) {
            ERROR("Failed to stream partition\n");
            return status;
        }
    }

    for (size_t p = 0; p < hdr->partition_count; p++) {
        // Upgrade the old partition (currently active) to the new partition (currently
        // inactive), so when the new partition becomes active, the old
        // partition is destroyed.
        upgrade_req_t upgrade;
        memset(&upgrade, 0, sizeof(upgrade));
        if (parts[p].old_part) {
            if (ioctl_block_get_partition_guid(parts[p].old_part.get(), &upgrade.old_guid,
                                               GUID_LEN) < 0) {
                ERROR("Failed to get unique GUID of old partition\n");
                return ZX_ERR_BAD_STATE;
            }
        }
        if (ioctl_block_get_partition_guid(parts[p].new_part.get(), &upgrade.new_guid,
                                           GUID_LEN) < 0) {
            ERROR("Failed to get unique GUID of new partition\n");
            return ZX_ERR_BAD_STATE;
        }

        if (ioctl_block_fvm_upgrade(fvm_fd.get(), &upgrade) < 0) {
            ERROR("Failed to upgrade partition\n");
            return ZX_ERR_IO;
        }

        if (parts[p].old_part) {
            // This would fail if the old part was on GPT, not FVM. However,
            // we checked earlier and verified that parts[p].old_part, if it exists,
            // is a vpartition.
            ssize_t r = ioctl_block_fvm_destroy_partition(parts[p].old_part.get());
            if (r < 0) {
                ERROR("Couldn't destroy partition: %ld\n", r);
                return static_cast<zx_status_t>(r);
            }
        }
    }

    return ZX_OK;
}

} // namespace

zx_status_t FvmPave(fbl::unique_ptr<DevicePartitioner> device_partitioner,
                    fbl::unique_fd payload_fd) {
    LOG("Paving FVM\n");

    fbl::unique_fd partition_fd;
    zx_status_t status;
    status = device_partitioner->FindPartition(Partition::kFuchsiaVolumeManager, &partition_fd);
    if (status == ZX_OK) {
        LOG("FVM partition already exists\n");
    } else if (status != ZX_ERR_NOT_FOUND) {
        return status;
    } else {
        status = device_partitioner->AddPartition(Partition::kFuchsiaVolumeManager, &partition_fd);
        if (status != ZX_OK) {
            ERROR("Failure creating partition: %d\n", status);
            return status;
        }
    }

    LOG("Streaming partitions...\n");
    if ((status = FvmStreamPartitions(fbl::move(partition_fd), fbl::move(payload_fd))) != ZX_OK) {
        ERROR("Failed to stream partitions: %d\n", status);
        return status;
    }
    LOG("Completed successfully\n");
    return ZX_OK;
}

zx_status_t PartitionPave(fbl::unique_ptr<DevicePartitioner> partitioner,
                          fbl::unique_fd payload_fd, Partition partition_type, Arch arch) {
    LOG("Paving partition.\n");

    zx_status_t status;
    fbl::unique_fd partition_fd;
    if ((status = partitioner->FindPartition(partition_type, &partition_fd)) != ZX_OK) {
        if (status != ZX_ERR_NOT_FOUND) {
            ERROR("Failure looking for partition: %d\n", status);
            return status;
        }
        if ((status = partitioner->AddPartition(partition_type, &partition_fd)) != ZX_OK) {
            ERROR("Failure creating partition: %d\n", status);
            return status;
        }
    } else {
        LOG("Partition already exists\n");
    }

    block_info_t info;
    if ((status = partitioner->GetBlockInfo(partition_fd, &info)) != ZX_OK) {
        ERROR("Couldn't get partition block info\n");
        return status;
    }

    const size_t vmo_sz = fbl::round_up(1LU << 20, info.block_size);
    fbl::unique_ptr<fzl::MappedVmo> mvmo;
    if ((status = fzl::MappedVmo::Create(vmo_sz, "partition-pave", &mvmo)) != ZX_OK) {
        ERROR("Failed to create stream VMO\n");
        return status;
    }
    // The streamed partition size may not line up with the mapped vmo size.
    size_t payload_size = 0;
    if ((status = StreamPayloadToVmo(mvmo.get(), payload_fd, info, &payload_size)) != ZX_OK) {
        ERROR("Failed to stream partition to VMO\n");
        return status;
    }
    if ((status = ValidateKernelPayload(mvmo.get(), payload_size, partition_type, arch)) != ZX_OK) {
        ERROR("Failed to validate partition\n");
        return status;
    }
    status = WriteVmoToBlock(mvmo.get(), payload_size, partition_fd, info);
    if (status != ZX_OK) {
        ERROR("Failed to write partition to block\n");
        return status;
    }

    if ((status = partitioner->FinalizePartition(partition_type)) != ZX_OK) {
        ERROR("Failed to finalize partition\n");
        return status;
    }

    LOG("Completed successfully\n");
    return ZX_OK;
}

zx_status_t FvmClean(fbl::unique_ptr<DevicePartitioner> partitioner) {
    const fbl::Vector<Partition> partition_list = {
        Partition::kEfi,
        Partition::kFuchsiaVolumeManager,
        Partition::kSystem,
        Partition::kBlob,
        Partition::kData,
    };
    return partitioner->WipePartitions(partition_list);
}

void Drain(fbl::unique_fd fd) {
    char buf[8192];
    while (read(fd.get(), &buf, sizeof(buf)) > 0)
        continue;
}

zx_status_t RealMain(Flags flags) {
    auto device_partitioner = DevicePartitioner::Create();
    if (!device_partitioner) {
        ERROR("Unable to initialize a partitioner.");
        return ZX_ERR_BAD_STATE;
    }
    const bool is_cros_device = device_partitioner->IsCros();

    switch (flags.cmd) {
    case Command::kInstallEfi:
        if ((is_cros_device || flags.arch == Arch::ARM64) && !flags.force) {
            LOG("SKIPPING EFI install on ARM64/CROS device, pass --force if desired.\n");
            Drain(fbl::move(flags.payload_fd));
            return ZX_OK;
        }
        return PartitionPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd),
                             Partition::kEfi, flags.arch);

    case Command::kInstallKernc:
        if (!is_cros_device && !flags.force) {
            LOG("SKIPPING KERNC install on non-CROS device, pass --force if desired.\n");
            Drain(fbl::move(flags.payload_fd));
            return ZX_OK;
        }
        return PartitionPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd),
                             Partition::kKernelC, flags.arch);

    case Command::kInstallZirconA:
        // TODO(ZX-2220): At some point x64 devices will want to pave A/B/R partitions as well.
        if (flags.arch == Arch::X64 && !flags.force) {
            LOG("SKIPPING ZIRCON-A install on x64 device, pass --force if desired.\n");
            Drain(fbl::move(flags.payload_fd));
            return ZX_OK;
        }
        return PartitionPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd),
                             Partition::kZirconA, flags.arch);

    case Command::kInstallZirconB:
        if (flags.arch == Arch::X64 && !flags.force) {
            LOG("SKIPPING ZIRCON-B install on x64 device, pass --force if desired.\n");
            Drain(fbl::move(flags.payload_fd));
            return ZX_OK;
        }
        return PartitionPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd),
                             Partition::kZirconB, flags.arch);

    case Command::kInstallZirconR:
        if (flags.arch == Arch::X64 && !flags.force) {
            LOG("SKIPPING ZIRCON-R install on x64 device, pass --force if desired.\n");
            Drain(fbl::move(flags.payload_fd));
            return ZX_OK;
        }
        return PartitionPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd),
                             Partition::kZirconR, flags.arch);
    case Command::kInstallFvm:
        return FvmPave(fbl::move(device_partitioner), fbl::move(flags.payload_fd));

    case Command::kWipe:
        return FvmClean(fbl::move(device_partitioner));

    default:
        ERROR("Unsupported command.");
        return ZX_ERR_NOT_SUPPORTED;
    }
}

} //  namespace paver