diff --git a/usr.sbin/bhyve/bhyve_config.5 b/usr.sbin/bhyve/bhyve_config.5 index 1a77b1bbacb4..a9cea3849c01 100644 --- a/usr.sbin/bhyve/bhyve_config.5 +++ b/usr.sbin/bhyve/bhyve_config.5 @@ -1,562 +1,572 @@ .\" SPDX-License-Identifier: BSD-2-Clause .\" .\" Copyright (c) 2021 John H. Baldwin .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. 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. .\" .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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. .\" -.Dd April 20, 2021 +.Dd September 17, 2021 .Dt BHYVE_CONFIG 5 .Os .Sh NAME .Nm bhyve_config .Nd "bhyve configuration variables" .Sh DESCRIPTION .Xr bhyve 8 uses a hierarchical tree of configuration variables to describe global and per-device settings. Internal nodes in this tree do not have a value, only leaf nodes have values. This manual describes the configuration variables understood by .Xr bhyve 8 . If additional variables are defined, .Xr bhyve 8 will ignore them and will not emit errors for unknown variables. However, these additional variables can be referenced by other variables as described below. .Sh VARIABLE VALUES Configuration variable values are stored as strings. A configuration variable value may refer to one or more other configuration values by name. Instances of the pattern .Sq % Ns Pq Ar var are replaced by the value of the configuration variable .Va var . To avoid unwanted expansion, .Sq % characters can be escaped by a leading .Sq % . For example, if a configuration variable .Va disk uses the value .Pa /dev/zvol/bhyve/%(name) , then the final value of the .Va disk variable will be set to the path of a ZFS volume whose name matches the name of the virtual machine on the pool .Pa bhyve . .Pp Some configuration variables may be interpreted as a boolean value. For those variables the following case-insensitive values may be used to indicate true: .Pp .Bl -bullet -offset indent -compact .It true .It on .It yes .It 1 .El .Pp The following values may be used to indicate false: .Pp .Bl -bullet -offset indent -compact .It false .It off .It no .It 0 .El .Pp Some configuration variables may be interperted as an integer. For those variables, any syntax supported by .Xr strtol 3 may be used. .Sh GLOBAL SETTINGS .Ss Architecture Neutral Settings .Bl -column "memory.guest_in_core" "integer" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va name Ta string Ta Ta The name of the VM. .It Va cpus Ta integer Ta 1 Ta The total number of virtual CPUs. .It Va cores Ta integer Ta 1 Ta The number of virtual cores in each virtual socket. .It Va threads Ta integer Ta 1 Ta The number of virtual CPUs in each virtual core. .It Va sockets Ta integer Ta 1 Ta The number of virtual sockets. .It Va memory.guest_in_core Ta bool Ta false Ta Include guest memory in core file. .It Va memory.size Ta string Ta 256M Ta Guest physical memory size in bytes. The value must be formatted as described in .Xr expand_number 3 . .It Va memory.wired Ta bool Ta false Ta Wire guest memory. .It Va acpi_tables Ta bool Ta false Ta Generate ACPI tables. .It Va destroy_on_poweroff Ta bool Ta false Ta Destroy the VM on guest-initiated power-off. .It Va gdb.port Ta integer Ta 0 Ta TCP port number for the debug server. If this is set to a non-zero value, a debug server will listen for connections on this port. .It Va gdb.wait Ta bool Ta false Ta If the debug server is enabled, wait for a debugger to connect before starting the guest. .It Va rtc.use_localtime Ta bool Ta true Ta The real time clock uses the local time of the host. If this is set to false, the real time clock uses UTC. .It Va uuid Ta string Ta Ta The universally unique identifier (UUID) to use in the guest's System Management BIOS System Information structure. If an explicit value is not set, a valid UUID is generated from the host's hostname and the VM name. .It Va virtio_msix Ta bool Ta true Ta Use MSI-X interrupts for PCI VirtIO devices. If set to false, MSI interrupts are used instead. .It Va config.dump Ta bool Ta false Ta If this value is set to true after .Xr bhyve 8 has finished parsing command line options, then .Xr bhyve 8 will write all of its configuration variables to stdout and exit. No VM will be started. .El .Ss x86-Specific Settings .Bl -column "x86.vmexit_on_pause" "integer" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va x86.mptable Ta bool Ta true Ta Generate an MPTable. .It Va x86.x2apic Ta bool Ta false Ta Configure guest's local APICs in x2APIC mode. .It Va x86.strictio Ta bool Ta false Ta Exit if a guest accesses an I/O port that is not emulated. By default, writes are ignored and reads return all bits set. .It Va x86.strictmsr Ta bool Ta true Ta Inject a general protection fault if a guest accesses a Model Specific Register (MSR) that is not emulated. If this is false, writes are ignored and reads return zero. .It Va x86.vmexit_on_hlt Ta bool Ta false Ta Force a VM exit when a guest CPU executes the .Dv HLT instruction. This allows idle guest CPUs to yield the host CPU. .It Va x86.vmexit_on_pause Ta bool Ta false Ta Force a VM exit when a guest CPU executes the .Dv PAUSE instruction. .El .Sh DEVICE SETTINGS Device settings are stored under a device node. The device node's name is set by the parent bus of the device. .Ss PCI Device Settings PCI devices are described by a device node named .Dq pci Ns Ar bus . Ns Ar slot . Ns Ar function where each of .Ar bus , .Ar slot , and .Ar function are formatted as decimal values with no padding. All PCI device nodes must contain a configuration variable named .Dq device which specifies the device model to use. The following PCI device models are supported: .Bl -tag -indent .It Li hostbridge Provide a simple PCI-Host bridge device. This is usually configured at pci0:0:0 and is required by most guest operating systems. .It Li ahci AHCI storage controller. .It Li e1000 Intel e82545 network interface. .It Li fbuf VGA framebuffer device attached to VNC server. .It Li lpc LPC PCI-ISA bridge with COM1-COM4 16550 serial ports, a boot ROM, and an optional debug/test device. This device must be configured on bus 0. .It Li hda High Definition audio controller. .It Li nvme NVM Express (NVMe) controller. .It Li passthru PCI pass-through device. .It Li uart PCI 16550 serial device. .It Li virtio-9p VirtIO 9p (VirtFS) interface. .It Li virtio-blk VirtIO block storage interface. .It Li virtio-console VirtIO console interface. .It Li virtio-net VirtIO network interface. .It Li virtio-rnd VirtIO RNG interface. .It Li virtio-scsi VirtIO SCSI interface. .It Li xhci Extensible Host Controller Interface (XHCI) USB controller. .El .Ss USB Device Settings USB controller devices contain zero or more child USB devices attached to slots. Each USB device stores its settings in a node named .Dq slot. Ns Va N under the controller's device node. .Va N is the number of the slot to which the USB device is attached. Note that USB slot numbers begin at 1. All USB device nodes must contain a configuration variable named .Dq device which specifies the device model to use. The following USB device models are supported: .Bl -tag -indent .It Li tablet A USB tablet device which provides precise cursor synchronization when using VNC. .El .Ss Block Device Settings Block devices use the following settings to configure their backing store. These settings are stored in the configuration node of the respective device. .Bl -column "sectorsize" "logical[/physical]" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It path Ta string Ta Ta The path of the file or disk device to use as the backing store. .It nocache Ta bool Ta false Ta Disable caching on the backing file by opening the backing file with .Dv O_DIRECT . .It nodelete Ta bool Ta false Ta Disable emulation of guest trim requests via .Dv DIOCGDELETE requests. .It sync Ta bool Ta false Ta Write changes to the backing file with synchronous writes. .It direct Ta bool Ta false Ta An alias for .Va sync . .It ro Ta bool Ta false Ta Disable writes to the backing file. .It sectorsize Ta Va logical Ns Op / Ns Va physical Ta Ta Specify the logical and physical sector size of the emulated disk. If the physical size is not specified, it is equal to the logical size. .El .Ss Network Backend Settings Network devices use the following settings to configure their backend. The backend is responsible for passing packets between the device model and a desired destination. Configuring a backend requires setting the .Va backend variable to one of the following values: .Bl -tag .It tap Ns Va N Use the named .Xr tap 4 interface as the backend. .It vmnet Ns Va N Use the named .Xr vmnet 4 interface as the backend. .It netgraph Use a .Xr netgraph 4 socket hook as the backend. This backend uses the following additional variables: .Bl -column "peerhook" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va path Ta string Ta Ta The name of the .Xr netgraph 4 destination node. .It Va peerhook Ta string Ta Ta The name of the destination hook. .It Va socket Ta string Ta Ta The name of the created .Xr ng_socket 4 node. .It Va hook Ta string Ta vmlink Ta The name of the source hook on the created .Xr ng_socket 4 node. .El .It netmap: Ns Va interface Use .Xr netmap 4 on a network interface as the backend. .It vale Ns Va bridge : Ns Va port Use a port on a .Xr vale 4 bridge as the backend. .El .Ss UART Device Settings .Bl -column "Name" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va path Ta path Ta Ta Backend device for the serial port. Either the pathname of a character device or .Dq stdio to use standard input and output of the .Xr bhyve 8 process. .El .Ss Host Bridge Settings .Bl -column "vendor" "integer" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va vendor Ta integer Ta 0x1275 Ta PCI vendor ID. .It Va devid Ta integer Ta 0x1275 Ta PCI device ID. .El .Ss AHCI Controller Settings AHCI controller devices contain zero or more ports each of which provides a storage device. Each port stores its settings in a node named .Dq port. Ns Va N under the controller's device node. The .Va N values are formatted as successive decimal values starting with 0. In addition to the block device settings described above, each port supports the following settings: .Bl -column "model" "integer" "generated" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va type Ta string Ta Ta The type of storage device to emulate. Must be set to either .Dq cd or .Dq hd . .It Va nmrr Ta integer Ta 0 Ta Nominal Media Rotation Rate, also known as RPM. A value 1 of indicates a device with no rate such as a Solid State Disk. .It Va ser Ta string Ta generated Ta Serial number of up to twenty characters. A default serial number is generated using a hash of the backing store's pathname. .It Va rev Ta string Ta 001 Ta Revision number of up to eight characters. .It Va model Ta string Ta Ta Model number of up to forty characters. Separate default model strings are used for .Dq cd and .Dq hd device types. .El .Ss e1000 Settings In addition to the network backend settings, Intel e82545 network interfaces support the following variables: .Bl -column "Name" "MAC address" "generated" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va mac Ta MAC address Ta generated Ta MAC address. If an explicit address is not provided, a MAC address is generated from a hash of the device's PCI address. .El .Ss Frame Buffer Settings .Bl -column "password" "[IP:]port" "127.0.0.1:5900" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va wait Ta bool Ta false Ta Wait for a remote connection before starting the VM. .It Va rfb Ta Oo Ar IP Ns : Oc Ns Ar port Ta 127.0.0.1:5900 Ta TCP address to listen on for remote connections. The IP address must be given as a numeric address. IPv6 addresses must be enclosed in square brackets and support scoped identifiers as described in .Xr getaddrinfo 3 . A bare port number may be given in which case the IPv4 localhost address is used. .It Va vga Ta string Ta io Ta VGA configuration. More details are provided in .Xr bhyve 8 . .It Va w Ta integer Ta 1024 Ta Frame buffer width in pixels. .It Va h Ta integer Ta 768 Ta Frame buffer height in pixels. .It Va password Ta string Ta Ta Password to use for VNC authentication. This type of authentication is known to be cryptographically weak and is not intended for use on untrusted networks. .El .Ss High Definition Audio Settings .Bl -column "Name" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va play Ta path Ta Ta Host playback device, typically .Pa /dev/dsp0 . .It Va rec Ta path Ta Ta Host recording device, typically .Pa /dev/dsp0 . .El .Ss LPC Device Settings The LPC bridge stores its configuration under a top-level .Va lpc node rather than under the PCI LPC device's node. The following nodes are available under .Va lpc : .Bl -column "pc-testdev" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va bootrom Ta path Ta Ta Path to a boot ROM. The contents of this file are copied into the guest's memory ending just before the 4GB physical address. If a boot ROM is present, a firmware interface device is also enabled for use by the boot ROM. .It Va com1 Ta node Ta Ta Settings for the COM1 serial port device. .It Va com2 Ta node Ta Ta Settings for the COM2 serial port device. .It Va com3 Ta node Ta Ta Settings for the COM3 serial port device. .It Va com4 Ta node Ta Ta Settings for the COM4 serial port device. .It Va pc-testdev Ta bool Ta false Ta Enable the PC debug/test device. .El .Ss NVMe Controller Settings Each NVMe controller supports a single storage device. The device can be backed either by a memory disk described by the .Va ram variable, or a block device using the the block device settings described above. In addition, each controller supports the following settings: .Bl -column "ioslots" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va maxq Ta integer Ta 16 Ta Maximum number of I/O submission and completion queue pairs. .It Va qsz Ta integer Ta 2058 Ta Number of elements in each I/O queue. .It Va ioslots Ta integer Ta 8 Ta Maximum number of concurrent I/O requests. .It Va sectsz Ta integer Ta Ta Sector size. Can be one of 512, 4096, or 8192. Devices backed by a memory disk use 4096 as the default. Devices backed by a block device use the block device's sector size as the default. .It Va ser Ta string Ta Ta Serial number of up to twenty characters. A default serial number is generated using a hash of the device's PCI address. .It Va eui64 Ta integer Ta Ta IEEE Extended Unique Identifier. If an EUI is not provided, a default is generated using a checksum of the device's PCI address. .It Va dsm Ta string Ta auto Ta Whether or not to advertise DataSet Management support. One of .Dq auto , .Dq enable , or .Dq disable . The .Dq auto setting only advertises support if the backing store supports resource freeing, for example via TRIM. .It Va ram Ta integer Ta Ta If set, allocate a memory disk as the backing store. The value of this variable is the size of the memory disk in megabytes. .El .Ss PCI Passthrough Settings .Bl -column "Name" "integer" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va bus Ta integer Ta Ta Host PCI bus address of device to pass through. .It Va slot Ta integer Ta Ta Host PCI slot address of device to pass through. .It Va func Ta integer Ta Ta Host PCI function address of device to pass through. .El .Ss VirtIO 9p Settings Each VirtIO 9p device exposes a single filesystem from a host path. .Bl -column "sharename" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va sharename Ta string Ta Ta The share name exposed to the guest. .It Va path Ta path Ta Ta The path of a directory on the host to export to the guest. .It Va ro Ta bool Ta false Ta If true, the guest filesystem is read-only. .El +.Ss VirtIO Block Device Settings +In addition to the block device settings described above, each +VirtIO block device supports the following settings: +.Bl -column "model" "integer" "generated" +.It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description +.It Va ser Ta string Ta generated Ta +Serial number of up to twenty characters. +A default serial number is generated using a hash of the backing +store's pathname. +.El .Ss VirtIO Console Device Settings Each VirtIO Console device contains one or more console ports. Each port stores its settings in a node named .Dq port. Ns Va N under the controller's device node. The .Va N values are formatted as successive decimal values starting with 0. Each port supports the following settings: .Bl -column "Name" "Format" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va name Ta string Ta Ta The name of the port exposed to the guest. .It Va path Ta path Ta Ta The path of a UNIX domain socket providing the host connection for the port. .El .Ss VirtIO Network Interface Settings In addition to the network backend settings, VirtIO network interfaces support the following variables: .Bl -column "Name" "MAC address" "generated" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va mac Ta MAC address Ta generated Ta MAC address. If an explicit address is not provided, a MAC address is generated from a hash of the device's PCI address. .It Va mtu Ta integer Ta 1500 Ta The largest supported MTU advertised to the guest. .El .Ss VirtIO SCSI Settings .Bl -column "Name" "integer" "Default" .It Sy Name Ta Sy Format Ta Sy Default Ta Sy Description .It Va dev Ta path Ta Ta The path of a CAM target layer (CTL) device to export: .Pa /dev/cam/ctl Ns Oo Ar pp . Ns Ar vp Oc . .It Va iid Ta integer Ta 0 Ta Initiator ID to use when sending requests to the CTL port. .El .Sh SEE ALSO .Xr expand_number 3 , .Xr getaddrinfo 3 , .Xr strtol 3 , .Xr netgraph 4 , .Xr netmap 4 , .Xr ng_socket 4 , .Xr tap 4 , .Xr vale 4 , .Xr vmnet 4 , .Xr bhyve 8 diff --git a/usr.sbin/bhyve/pci_virtio_block.c b/usr.sbin/bhyve/pci_virtio_block.c index 3718b2d7eff0..385b812be84c 100644 --- a/usr.sbin/bhyve/pci_virtio_block.c +++ b/usr.sbin/bhyve/pci_virtio_block.c @@ -1,590 +1,597 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * Copyright 2020-2021 Joyent, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. 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. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``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 NETAPP, INC 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. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" #include "config.h" #include "debug.h" #include "pci_emul.h" #include "virtio.h" #include "block_if.h" #define VTBLK_BSIZE 512 #define VTBLK_RINGSZ 128 _Static_assert(VTBLK_RINGSZ <= BLOCKIF_RING_MAX, "Each ring entry must be able to queue a request"); #define VTBLK_S_OK 0 #define VTBLK_S_IOERR 1 #define VTBLK_S_UNSUPP 2 #define VTBLK_BLK_ID_BYTES 20 + 1 /* Capability bits */ #define VTBLK_F_BARRIER (1 << 0) /* Does host support barriers? */ #define VTBLK_F_SIZE_MAX (1 << 1) /* Indicates maximum segment size */ #define VTBLK_F_SEG_MAX (1 << 2) /* Indicates maximum # of segments */ #define VTBLK_F_GEOMETRY (1 << 4) /* Legacy geometry available */ #define VTBLK_F_RO (1 << 5) /* Disk is read-only */ #define VTBLK_F_BLK_SIZE (1 << 6) /* Block size of disk is available*/ #define VTBLK_F_SCSI (1 << 7) /* Supports scsi command passthru */ #define VTBLK_F_FLUSH (1 << 9) /* Writeback mode enabled after reset */ #define VTBLK_F_WCE (1 << 9) /* Legacy alias for FLUSH */ #define VTBLK_F_TOPOLOGY (1 << 10) /* Topology information is available */ #define VTBLK_F_CONFIG_WCE (1 << 11) /* Writeback mode available in config */ #define VTBLK_F_MQ (1 << 12) /* Multi-Queue */ #define VTBLK_F_DISCARD (1 << 13) /* Trim blocks */ #define VTBLK_F_WRITE_ZEROES (1 << 14) /* Write zeros */ /* * Host capabilities */ #define VTBLK_S_HOSTCAPS \ ( VTBLK_F_SEG_MAX | \ VTBLK_F_BLK_SIZE | \ VTBLK_F_FLUSH | \ VTBLK_F_TOPOLOGY | \ VIRTIO_RING_F_INDIRECT_DESC ) /* indirect descriptors */ /* * The current blockif_delete() interface only allows a single delete * request at a time. */ #define VTBLK_MAX_DISCARD_SEG 1 /* * An arbitrary limit to prevent excessive latency due to large * delete requests. */ #define VTBLK_MAX_DISCARD_SECT ((16 << 20) / VTBLK_BSIZE) /* 16 MiB */ /* * Config space "registers" */ struct vtblk_config { uint64_t vbc_capacity; uint32_t vbc_size_max; uint32_t vbc_seg_max; struct { uint16_t cylinders; uint8_t heads; uint8_t sectors; } vbc_geometry; uint32_t vbc_blk_size; struct { uint8_t physical_block_exp; uint8_t alignment_offset; uint16_t min_io_size; uint32_t opt_io_size; } vbc_topology; uint8_t vbc_writeback; uint8_t unused0[1]; uint16_t num_queues; uint32_t max_discard_sectors; uint32_t max_discard_seg; uint32_t discard_sector_alignment; uint32_t max_write_zeroes_sectors; uint32_t max_write_zeroes_seg; uint8_t write_zeroes_may_unmap; uint8_t unused1[3]; } __packed; /* * Fixed-size block header */ struct virtio_blk_hdr { #define VBH_OP_READ 0 #define VBH_OP_WRITE 1 #define VBH_OP_SCSI_CMD 2 #define VBH_OP_SCSI_CMD_OUT 3 #define VBH_OP_FLUSH 4 #define VBH_OP_FLUSH_OUT 5 #define VBH_OP_IDENT 8 #define VBH_OP_DISCARD 11 #define VBH_OP_WRITE_ZEROES 13 #define VBH_FLAG_BARRIER 0x80000000 /* OR'ed into vbh_type */ uint32_t vbh_type; uint32_t vbh_ioprio; uint64_t vbh_sector; } __packed; /* * Debug printf */ static int pci_vtblk_debug; #define DPRINTF(params) if (pci_vtblk_debug) PRINTLN params #define WPRINTF(params) PRINTLN params struct pci_vtblk_ioreq { struct blockif_req io_req; struct pci_vtblk_softc *io_sc; uint8_t *io_status; uint16_t io_idx; }; struct virtio_blk_discard_write_zeroes { uint64_t sector; uint32_t num_sectors; struct { uint32_t unmap:1; uint32_t reserved:31; } flags; }; /* * Per-device softc */ struct pci_vtblk_softc { struct virtio_softc vbsc_vs; pthread_mutex_t vsc_mtx; struct vqueue_info vbsc_vq; struct vtblk_config vbsc_cfg; struct virtio_consts vbsc_consts; struct blockif_ctxt *bc; char vbsc_ident[VTBLK_BLK_ID_BYTES]; struct pci_vtblk_ioreq vbsc_ios[VTBLK_RINGSZ]; }; static void pci_vtblk_reset(void *); static void pci_vtblk_notify(void *, struct vqueue_info *); static int pci_vtblk_cfgread(void *, int, int, uint32_t *); static int pci_vtblk_cfgwrite(void *, int, int, uint32_t); #ifdef BHYVE_SNAPSHOT static void pci_vtblk_pause(void *); static void pci_vtblk_resume(void *); static int pci_vtblk_snapshot(void *, struct vm_snapshot_meta *); #endif static struct virtio_consts vtblk_vi_consts = { "vtblk", /* our name */ 1, /* we support 1 virtqueue */ sizeof(struct vtblk_config), /* config reg size */ pci_vtblk_reset, /* reset */ pci_vtblk_notify, /* device-wide qnotify */ pci_vtblk_cfgread, /* read PCI config */ pci_vtblk_cfgwrite, /* write PCI config */ NULL, /* apply negotiated features */ VTBLK_S_HOSTCAPS, /* our capabilities */ #ifdef BHYVE_SNAPSHOT pci_vtblk_pause, /* pause blockif threads */ pci_vtblk_resume, /* resume blockif threads */ pci_vtblk_snapshot, /* save / restore device state */ #endif }; static void pci_vtblk_reset(void *vsc) { struct pci_vtblk_softc *sc = vsc; DPRINTF(("vtblk: device reset requested !")); vi_reset_dev(&sc->vbsc_vs); } static void pci_vtblk_done_locked(struct pci_vtblk_ioreq *io, int err) { struct pci_vtblk_softc *sc = io->io_sc; /* convert errno into a virtio block error return */ if (err == EOPNOTSUPP || err == ENOSYS) *io->io_status = VTBLK_S_UNSUPP; else if (err != 0) *io->io_status = VTBLK_S_IOERR; else *io->io_status = VTBLK_S_OK; /* * Return the descriptor back to the host. * We wrote 1 byte (our status) to host. */ vq_relchain(&sc->vbsc_vq, io->io_idx, 1); vq_endchains(&sc->vbsc_vq, 0); } #ifdef BHYVE_SNAPSHOT static void pci_vtblk_pause(void *vsc) { struct pci_vtblk_softc *sc = vsc; DPRINTF(("vtblk: device pause requested !\n")); blockif_pause(sc->bc); } static void pci_vtblk_resume(void *vsc) { struct pci_vtblk_softc *sc = vsc; DPRINTF(("vtblk: device resume requested !\n")); blockif_resume(sc->bc); } static int pci_vtblk_snapshot(void *vsc, struct vm_snapshot_meta *meta) { int ret; struct pci_vtblk_softc *sc = vsc; SNAPSHOT_VAR_OR_LEAVE(sc->vbsc_cfg, meta, ret, done); SNAPSHOT_BUF_OR_LEAVE(sc->vbsc_ident, sizeof(sc->vbsc_ident), meta, ret, done); done: return (ret); } #endif static void pci_vtblk_done(struct blockif_req *br, int err) { struct pci_vtblk_ioreq *io = br->br_param; struct pci_vtblk_softc *sc = io->io_sc; pthread_mutex_lock(&sc->vsc_mtx); pci_vtblk_done_locked(io, err); pthread_mutex_unlock(&sc->vsc_mtx); } static void pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vqueue_info *vq) { struct virtio_blk_hdr *vbh; struct pci_vtblk_ioreq *io; int i, n; int err; ssize_t iolen; int writeop, type; struct vi_req req; struct iovec iov[BLOCKIF_IOV_MAX + 2]; struct virtio_blk_discard_write_zeroes *discard; n = vq_getchain(vq, iov, BLOCKIF_IOV_MAX + 2, &req); /* * The first descriptor will be the read-only fixed header, * and the last is for status (hence +2 above and below). * The remaining iov's are the actual data I/O vectors. * * XXX - note - this fails on crash dump, which does a * VIRTIO_BLK_T_FLUSH with a zero transfer length */ assert(n >= 2 && n <= BLOCKIF_IOV_MAX + 2); io = &sc->vbsc_ios[req.idx]; assert(req.readable != 0); assert(iov[0].iov_len == sizeof(struct virtio_blk_hdr)); vbh = (struct virtio_blk_hdr *)iov[0].iov_base; memcpy(&io->io_req.br_iov, &iov[1], sizeof(struct iovec) * (n - 2)); io->io_req.br_iovcnt = n - 2; io->io_req.br_offset = vbh->vbh_sector * VTBLK_BSIZE; io->io_status = (uint8_t *)iov[--n].iov_base; assert(req.writable != 0); assert(iov[n].iov_len == 1); /* * XXX * The guest should not be setting the BARRIER flag because * we don't advertise the capability. */ type = vbh->vbh_type & ~VBH_FLAG_BARRIER; writeop = (type == VBH_OP_WRITE || type == VBH_OP_DISCARD); /* * - Write op implies read-only descriptor * - Read/ident op implies write-only descriptor * * By taking away either the read-only fixed header or the write-only * status iovec, the following condition should hold true. */ assert(n == (writeop ? req.readable : req.writable)); iolen = 0; for (i = 1; i < n; i++) { iolen += iov[i].iov_len; } io->io_req.br_resid = iolen; DPRINTF(("virtio-block: %s op, %zd bytes, %d segs, offset %ld", writeop ? "write/discard" : "read/ident", iolen, i - 1, io->io_req.br_offset)); switch (type) { case VBH_OP_READ: err = blockif_read(sc->bc, &io->io_req); break; case VBH_OP_WRITE: err = blockif_write(sc->bc, &io->io_req); break; case VBH_OP_DISCARD: /* * We currently only support a single request, if the guest * has submitted a request that doesn't conform to the * requirements, we return a error. */ if (iov[1].iov_len != sizeof (*discard)) { pci_vtblk_done_locked(io, EINVAL); return; } /* The segments to discard are provided rather than data */ discard = (struct virtio_blk_discard_write_zeroes *) iov[1].iov_base; /* * virtio v1.1 5.2.6.2: * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP * for discard and write zeroes commands if any unknown flag is * set. Furthermore, the device MUST set the status byte to * VIRTIO_BLK_S_UNSUPP for discard commands if the unmap flag * is set. * * Currently there are no known flags for a DISCARD request. */ if (discard->flags.unmap != 0 || discard->flags.reserved != 0) { pci_vtblk_done_locked(io, ENOTSUP); return; } /* Make sure the request doesn't exceed our size limit */ if (discard->num_sectors > VTBLK_MAX_DISCARD_SECT) { pci_vtblk_done_locked(io, EINVAL); return; } io->io_req.br_offset = discard->sector * VTBLK_BSIZE; io->io_req.br_resid = discard->num_sectors * VTBLK_BSIZE; err = blockif_delete(sc->bc, &io->io_req); break; case VBH_OP_FLUSH: case VBH_OP_FLUSH_OUT: err = blockif_flush(sc->bc, &io->io_req); break; case VBH_OP_IDENT: /* Assume a single buffer */ /* S/n equal to buffer is not zero-terminated. */ memset(iov[1].iov_base, 0, iov[1].iov_len); strncpy(iov[1].iov_base, sc->vbsc_ident, MIN(iov[1].iov_len, sizeof(sc->vbsc_ident))); pci_vtblk_done_locked(io, 0); return; default: pci_vtblk_done_locked(io, EOPNOTSUPP); return; } assert(err == 0); } static void pci_vtblk_notify(void *vsc, struct vqueue_info *vq) { struct pci_vtblk_softc *sc = vsc; while (vq_has_descs(vq)) pci_vtblk_proc(sc, vq); } static void pci_vtblk_resized(struct blockif_ctxt *bctxt, void *arg, size_t new_size) { struct pci_vtblk_softc *sc; sc = arg; sc->vbsc_cfg.vbc_capacity = new_size / VTBLK_BSIZE; /* 512-byte units */ vi_interrupt(&sc->vbsc_vs, VIRTIO_PCI_ISR_CONFIG, sc->vbsc_vs.vs_msix_cfg_idx); } static int pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, nvlist_t *nvl) { char bident[sizeof("XX:X:X")]; struct blockif_ctxt *bctxt; - const char *path; + const char *path, *serial; MD5_CTX mdctx; u_char digest[16]; struct pci_vtblk_softc *sc; off_t size; int i, sectsz, sts, sto; /* * The supplied backing file has to exist */ snprintf(bident, sizeof(bident), "%d:%d", pi->pi_slot, pi->pi_func); bctxt = blockif_open(nvl, bident); if (bctxt == NULL) { perror("Could not open backing file"); return (1); } size = blockif_size(bctxt); sectsz = blockif_sectsz(bctxt); blockif_psectsz(bctxt, &sts, &sto); sc = calloc(1, sizeof(struct pci_vtblk_softc)); sc->bc = bctxt; for (i = 0; i < VTBLK_RINGSZ; i++) { struct pci_vtblk_ioreq *io = &sc->vbsc_ios[i]; io->io_req.br_callback = pci_vtblk_done; io->io_req.br_param = io; io->io_sc = sc; io->io_idx = i; } bcopy(&vtblk_vi_consts, &sc->vbsc_consts, sizeof (vtblk_vi_consts)); if (blockif_candelete(sc->bc)) sc->vbsc_consts.vc_hv_caps |= VTBLK_F_DISCARD; pthread_mutex_init(&sc->vsc_mtx, NULL); /* init virtio softc and virtqueues */ vi_softc_linkup(&sc->vbsc_vs, &sc->vbsc_consts, sc, pi, &sc->vbsc_vq); sc->vbsc_vs.vs_mtx = &sc->vsc_mtx; sc->vbsc_vq.vq_qsize = VTBLK_RINGSZ; /* sc->vbsc_vq.vq_notify = we have no per-queue notify */ /* - * Create an identifier for the backing file. Use parts of the - * md5 sum of the filename + * If an explicit identifier is not given, create an + * identifier using parts of the md5 sum of the filename. */ - path = get_config_value_node(nvl, "path"); - MD5Init(&mdctx); - MD5Update(&mdctx, path, strlen(path)); - MD5Final(digest, &mdctx); - snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES, - "BHYVE-%02X%02X-%02X%02X-%02X%02X", - digest[0], digest[1], digest[2], digest[3], digest[4], digest[5]); + bzero(sc->vbsc_ident, VTBLK_BLK_ID_BYTES); + if ((serial = get_config_value_node(nvl, "serial")) != NULL || + (serial = get_config_value_node(nvl, "ser")) != NULL) { + strlcpy(sc->vbsc_ident, serial, VTBLK_BLK_ID_BYTES); + } else { + path = get_config_value_node(nvl, "path"); + MD5Init(&mdctx); + MD5Update(&mdctx, path, strlen(path)); + MD5Final(digest, &mdctx); + snprintf(sc->vbsc_ident, VTBLK_BLK_ID_BYTES, + "BHYVE-%02X%02X-%02X%02X-%02X%02X", + digest[0], digest[1], digest[2], digest[3], digest[4], + digest[5]); + } /* setup virtio block config space */ sc->vbsc_cfg.vbc_capacity = size / VTBLK_BSIZE; /* 512-byte units */ sc->vbsc_cfg.vbc_size_max = 0; /* not negotiated */ /* * If Linux is presented with a seg_max greater than the virtio queue * size, it can stumble into situations where it violates its own * invariants and panics. For safety, we keep seg_max clamped, paying * heed to the two extra descriptors needed for the header and status * of a request. */ sc->vbsc_cfg.vbc_seg_max = MIN(VTBLK_RINGSZ - 2, BLOCKIF_IOV_MAX); sc->vbsc_cfg.vbc_geometry.cylinders = 0; /* no geometry */ sc->vbsc_cfg.vbc_geometry.heads = 0; sc->vbsc_cfg.vbc_geometry.sectors = 0; sc->vbsc_cfg.vbc_blk_size = sectsz; sc->vbsc_cfg.vbc_topology.physical_block_exp = (sts > sectsz) ? (ffsll(sts / sectsz) - 1) : 0; sc->vbsc_cfg.vbc_topology.alignment_offset = (sto != 0) ? ((sts - sto) / sectsz) : 0; sc->vbsc_cfg.vbc_topology.min_io_size = 0; sc->vbsc_cfg.vbc_topology.opt_io_size = 0; sc->vbsc_cfg.vbc_writeback = 0; sc->vbsc_cfg.max_discard_sectors = VTBLK_MAX_DISCARD_SECT; sc->vbsc_cfg.max_discard_seg = VTBLK_MAX_DISCARD_SEG; sc->vbsc_cfg.discard_sector_alignment = MAX(sectsz, sts) / VTBLK_BSIZE; /* * Should we move some of this into virtio.c? Could * have the device, class, and subdev_0 as fields in * the virtio constants structure. */ pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK); pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR); pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE); pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_BLOCK); pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR); if (vi_intr_init(&sc->vbsc_vs, 1, fbsdrun_virtio_msix())) { blockif_close(sc->bc); free(sc); return (1); } vi_set_io_bar(&sc->vbsc_vs, 0); blockif_register_resize_callback(sc->bc, pci_vtblk_resized, sc); return (0); } static int pci_vtblk_cfgwrite(void *vsc, int offset, int size, uint32_t value) { DPRINTF(("vtblk: write to readonly reg %d", offset)); return (1); } static int pci_vtblk_cfgread(void *vsc, int offset, int size, uint32_t *retval) { struct pci_vtblk_softc *sc = vsc; void *ptr; /* our caller has already verified offset and size */ ptr = (uint8_t *)&sc->vbsc_cfg + offset; memcpy(retval, ptr, size); return (0); } struct pci_devemu pci_de_vblk = { .pe_emu = "virtio-blk", .pe_init = pci_vtblk_init, .pe_legacy_config = blockif_legacy_config, .pe_barwrite = vi_pci_write, .pe_barread = vi_pci_read, #ifdef BHYVE_SNAPSHOT .pe_snapshot = vi_pci_snapshot, #endif }; PCI_EMUL_SET(pci_de_vblk);