Comment 2 for bug 1893691

This regression is specific to the "socket" chardev and was caused by this commit:

commit 271094474b65de1ad7aaf729938de3d9b9d0d36f (refs/bisect/bad)
Author: Dima Stepanov <email address hidden>
Date: Thu May 28 12:11:18 2020 +0300

    char-socket: return -1 in case of disconnect during tcp_chr_write

    During testing of the vhost-user-blk reconnect functionality the qemu
    SIGSEGV was triggered:
     start qemu as:
     x86_64-softmmu/qemu-system-x86_64 -m 1024M -M q35 \
       -object memory-backend-file,id=ram-node0,size=1024M,mem-path=/dev/shm/qemu,share=on \
       -numa node,cpus=0,memdev=ram-node0 \
       -chardev socket,id=chardev0,path=./vhost.sock,noserver,reconnect=1 \
       -device vhost-user-blk-pci,chardev=chardev0,num-queues=4 --enable-kvm
     start vhost-user-blk daemon:
     ./vhost-user-blk -s ./vhost.sock -b test-img.raw

    If vhost-user-blk will be killed during the vhost initialization
    process, for instance after getting VHOST_SET_VRING_CALL command, then
    QEMU will fail with the following backtrace:

    Thread 1 "qemu-system-x86" received signal SIGSEGV, Segmentation fault.
    0x00005555559272bb in vhost_user_read (dev=0x7fffef2d53e0, msg=0x7fffffffd5b0)
        at ./hw/virtio/vhost-user.c:260
    260 CharBackend *chr = u->user->chr;

     #0 0x00005555559272bb in vhost_user_read (dev=0x7fffef2d53e0, msg=0x7fffffffd5b0)
        at ./hw/virtio/vhost-user.c:260
     #1 0x000055555592acb8 in vhost_user_get_config (dev=0x7fffef2d53e0, config=0x7fffef2d5394 "", config_len=60)
        at ./hw/virtio/vhost-user.c:1645
     #2 0x0000555555925525 in vhost_dev_get_config (hdev=0x7fffef2d53e0, config=0x7fffef2d5394 "", config_len=60)
        at ./hw/virtio/vhost.c:1490
     #3 0x00005555558cc46b in vhost_user_blk_device_realize (dev=0x7fffef2d51a0, errp=0x7fffffffd8f0)
        at ./hw/block/vhost-user-blk.c:429
     #4 0x0000555555920090 in virtio_device_realize (dev=0x7fffef2d51a0, errp=0x7fffffffd948)
        at ./hw/virtio/virtio.c:3615
     #5 0x0000555555a9779c in device_set_realized (obj=0x7fffef2d51a0, value=true, errp=0x7fffffffdb88)
        at ./hw/core/qdev.c:891
     ...

    The problem is that vhost_user_write doesn't get an error after
    disconnect and try to call vhost_user_read(). The tcp_chr_write()
    routine should return -1 in case of disconnect. Indicate the EIO error
    if this routine is called in the disconnected state.

    Signed-off-by: Dima Stepanov <email address hidden>
    Reviewed-by: Marc-André Lureau <email address hidden>
    Message-Id: <email address hidden>
    Reviewed-by: Michael S. Tsirkin <email address hidden>
    Signed-off-by: Michael S. Tsirkin <email address hidden>