diff options
Diffstat (limited to 'Documentation/virtual')
| -rw-r--r-- | Documentation/virtual/00-INDEX | 10 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/api.txt | 1451 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/cpuid.txt | 45 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/locking.txt | 25 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/mmu.txt | 348 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/msr.txt | 187 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/ppc-pv.txt | 196 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/review-checklist.txt | 38 | ||||
| -rw-r--r-- | Documentation/virtual/kvm/timekeeping.txt | 612 | ||||
| -rw-r--r-- | Documentation/virtual/lguest/.gitignore | 1 | ||||
| -rw-r--r-- | Documentation/virtual/lguest/Makefile | 8 | ||||
| -rw-r--r-- | Documentation/virtual/lguest/extract | 58 | ||||
| -rw-r--r-- | Documentation/virtual/lguest/lguest.c | 2095 | ||||
| -rw-r--r-- | Documentation/virtual/lguest/lguest.txt | 129 | ||||
| -rw-r--r-- | Documentation/virtual/uml/UserModeLinux-HOWTO.txt | 4579 |
15 files changed, 9782 insertions, 0 deletions
diff --git a/Documentation/virtual/00-INDEX b/Documentation/virtual/00-INDEX new file mode 100644 index 00000000000..fe0251c4cfb --- /dev/null +++ b/Documentation/virtual/00-INDEX @@ -0,0 +1,10 @@ +Virtualization support in the Linux kernel. + +00-INDEX + - this file. +kvm/ + - Kernel Virtual Machine. See also http://linux-kvm.org +lguest/ + - Extremely simple hypervisor for experimental/educational use. +uml/ + - User Mode Linux, builds/runs Linux kernel as a userspace program. diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt new file mode 100644 index 00000000000..9bef4e4cec5 --- /dev/null +++ b/Documentation/virtual/kvm/api.txt @@ -0,0 +1,1451 @@ +The Definitive KVM (Kernel-based Virtual Machine) API Documentation +=================================================================== + +1. General description + +The kvm API is a set of ioctls that are issued to control various aspects +of a virtual machine. The ioctls belong to three classes + + - System ioctls: These query and set global attributes which affect the + whole kvm subsystem. In addition a system ioctl is used to create + virtual machines + + - VM ioctls: These query and set attributes that affect an entire virtual + machine, for example memory layout. In addition a VM ioctl is used to + create virtual cpus (vcpus). + + Only run VM ioctls from the same process (address space) that was used + to create the VM. + + - vcpu ioctls: These query and set attributes that control the operation + of a single virtual cpu. + + Only run vcpu ioctls from the same thread that was used to create the + vcpu. + +2. File descriptors + +The kvm API is centered around file descriptors. An initial +open("/dev/kvm") obtains a handle to the kvm subsystem; this handle +can be used to issue system ioctls. A KVM_CREATE_VM ioctl on this +handle will create a VM file descriptor which can be used to issue VM +ioctls. A KVM_CREATE_VCPU ioctl on a VM fd will create a virtual cpu +and return a file descriptor pointing to it. Finally, ioctls on a vcpu +fd can be used to control the vcpu, including the important task of +actually running guest code. + +In general file descriptors can be migrated among processes by means +of fork() and the SCM_RIGHTS facility of unix domain socket. These +kinds of tricks are explicitly not supported by kvm. While they will +not cause harm to the host, their actual behavior is not guaranteed by +the API. The only supported use is one virtual machine per process, +and one vcpu per thread. + +3. Extensions + +As of Linux 2.6.22, the KVM ABI has been stabilized: no backward +incompatible change are allowed. However, there is an extension +facility that allows backward-compatible extensions to the API to be +queried and used. + +The extension mechanism is not based on on the Linux version number. +Instead, kvm defines extension identifiers and a facility to query +whether a particular extension identifier is available. If it is, a +set of ioctls is available for application use. + +4. API description + +This section describes ioctls that can be used to control kvm guests. +For each ioctl, the following information is provided along with a +description: + + Capability: which KVM extension provides this ioctl. Can be 'basic', + which means that is will be provided by any kernel that supports + API version 12 (see section 4.1), or a KVM_CAP_xyz constant, which + means availability needs to be checked with KVM_CHECK_EXTENSION + (see section 4.4). + + Architectures: which instruction set architectures provide this ioctl. + x86 includes both i386 and x86_64. + + Type: system, vm, or vcpu. + + Parameters: what parameters are accepted by the ioctl. + + Returns: the return value. General error numbers (EBADF, ENOMEM, EINVAL) + are not detailed, but errors with specific meanings are. + +4.1 KVM_GET_API_VERSION + +Capability: basic +Architectures: all +Type: system ioctl +Parameters: none +Returns: the constant KVM_API_VERSION (=12) + +This identifies the API version as the stable kvm API. It is not +expected that this number will change. However, Linux 2.6.20 and +2.6.21 report earlier versions; these are not documented and not +supported. Applications should refuse to run if KVM_GET_API_VERSION +returns a value other than 12. If this check passes, all ioctls +described as 'basic' will be available. + +4.2 KVM_CREATE_VM + +Capability: basic +Architectures: all +Type: system ioctl +Parameters: none +Returns: a VM fd that can be used to control the new virtual machine. + +The new VM has no virtual cpus and no memory. An mmap() of a VM fd +will access the virtual machine's physical address space; offset zero +corresponds to guest physical address zero. Use of mmap() on a VM fd +is discouraged if userspace memory allocation (KVM_CAP_USER_MEMORY) is +available. + +4.3 KVM_GET_MSR_INDEX_LIST + +Capability: basic +Architectures: x86 +Type: system +Parameters: struct kvm_msr_list (in/out) +Returns: 0 on success; -1 on error +Errors: + E2BIG: the msr index list is to be to fit in the array specified by + the user. + +struct kvm_msr_list { + __u32 nmsrs; /* number of msrs in entries */ + __u32 indices[0]; +}; + +This ioctl returns the guest msrs that are supported. The list varies +by kvm version and host processor, but does not change otherwise. The +user fills in the size of the indices array in nmsrs, and in return +kvm adjusts nmsrs to reflect the actual number of msrs and fills in +the indices array with their numbers. + +Note: if kvm indicates supports MCE (KVM_CAP_MCE), then the MCE bank MSRs are +not returned in the MSR list, as different vcpus can have a different number +of banks, as set via the KVM_X86_SETUP_MCE ioctl. + +4.4 KVM_CHECK_EXTENSION + +Capability: basic +Architectures: all +Type: system ioctl +Parameters: extension identifier (KVM_CAP_*) +Returns: 0 if unsupported; 1 (or some other positive integer) if supported + +The API allows the application to query about extensions to the core +kvm API. Userspace passes an extension identifier (an integer) and +receives an integer that describes the extension availability. +Generally 0 means no and 1 means yes, but some extensions may report +additional information in the integer return value. + +4.5 KVM_GET_VCPU_MMAP_SIZE + +Capability: basic +Architectures: all +Type: system ioctl +Parameters: none +Returns: size of vcpu mmap area, in bytes + +The KVM_RUN ioctl (cf.) communicates with userspace via a shared +memory region. This ioctl returns the size of that region. See the +KVM_RUN documentation for details. + +4.6 KVM_SET_MEMORY_REGION + +Capability: basic +Architectures: all +Type: vm ioctl +Parameters: struct kvm_memory_region (in) +Returns: 0 on success, -1 on error + +This ioctl is obsolete and has been removed. + +4.7 KVM_CREATE_VCPU + +Capability: basic +Architectures: all +Type: vm ioctl +Parameters: vcpu id (apic id on x86) +Returns: vcpu fd on success, -1 on error + +This API adds a vcpu to a virtual machine. The vcpu id is a small integer +in the range [0, max_vcpus). + +4.8 KVM_GET_DIRTY_LOG (vm ioctl) + +Capability: basic +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_dirty_log (in/out) +Returns: 0 on success, -1 on error + +/* for KVM_GET_DIRTY_LOG */ +struct kvm_dirty_log { + __u32 slot; + __u32 padding; + union { + void __user *dirty_bitmap; /* one bit per page */ + __u64 padding; + }; +}; + +Given a memory slot, return a bitmap containing any pages dirtied +since the last call to this ioctl. Bit 0 is the first page in the +memory slot. Ensure the entire structure is cleared to avoid padding +issues. + +4.9 KVM_SET_MEMORY_ALIAS + +Capability: basic +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_memory_alias (in) +Returns: 0 (success), -1 (error) + +This ioctl is obsolete and has been removed. + +4.10 KVM_RUN + +Capability: basic +Architectures: all +Type: vcpu ioctl +Parameters: none +Returns: 0 on success, -1 on error +Errors: + EINTR: an unmasked signal is pending + +This ioctl is used to run a guest virtual cpu. While there are no +explicit parameters, there is an implicit parameter block that can be +obtained by mmap()ing the vcpu fd at offset 0, with the size given by +KVM_GET_VCPU_MMAP_SIZE. The parameter block is formatted as a 'struct +kvm_run' (see below). + +4.11 KVM_GET_REGS + +Capability: basic +Architectures: all +Type: vcpu ioctl +Parameters: struct kvm_regs (out) +Returns: 0 on success, -1 on error + +Reads the general purpose registers from the vcpu. + +/* x86 */ +struct kvm_regs { + /* out (KVM_GET_REGS) / in (KVM_SET_REGS) */ + __u64 rax, rbx, rcx, rdx; + __u64 rsi, rdi, rsp, rbp; + __u64 r8, r9, r10, r11; + __u64 r12, r13, r14, r15; + __u64 rip, rflags; +}; + +4.12 KVM_SET_REGS + +Capability: basic +Architectures: all +Type: vcpu ioctl +Parameters: struct kvm_regs (in) +Returns: 0 on success, -1 on error + +Writes the general purpose registers into the vcpu. + +See KVM_GET_REGS for the data structure. + +4.13 KVM_GET_SREGS + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_sregs (out) +Returns: 0 on success, -1 on error + +Reads special registers from the vcpu. + +/* x86 */ +struct kvm_sregs { + struct kvm_segment cs, ds, es, fs, gs, ss; + struct kvm_segment tr, ldt; + struct kvm_dtable gdt, idt; + __u64 cr0, cr2, cr3, cr4, cr8; + __u64 efer; + __u64 apic_base; + __u64 interrupt_bitmap[(KVM_NR_INTERRUPTS + 63) / 64]; +}; + +interrupt_bitmap is a bitmap of pending external interrupts. At most +one bit may be set. This interrupt has been acknowledged by the APIC +but not yet injected into the cpu core. + +4.14 KVM_SET_SREGS + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_sregs (in) +Returns: 0 on success, -1 on error + +Writes special registers into the vcpu. See KVM_GET_SREGS for the +data structures. + +4.15 KVM_TRANSLATE + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_translation (in/out) +Returns: 0 on success, -1 on error + +Translates a virtual address according to the vcpu's current address +translation mode. + +struct kvm_translation { + /* in */ + __u64 linear_address; + + /* out */ + __u64 physical_address; + __u8 valid; + __u8 writeable; + __u8 usermode; + __u8 pad[5]; +}; + +4.16 KVM_INTERRUPT + +Capability: basic +Architectures: x86, ppc +Type: vcpu ioctl +Parameters: struct kvm_interrupt (in) +Returns: 0 on success, -1 on error + +Queues a hardware interrupt vector to be injected. This is only +useful if in-kernel local APIC or equivalent is not used. + +/* for KVM_INTERRUPT */ +struct kvm_interrupt { + /* in */ + __u32 irq; +}; + +X86: + +Note 'irq' is an interrupt vector, not an interrupt pin or line. + +PPC: + +Queues an external interrupt to be injected. This ioctl is overleaded +with 3 different irq values: + +a) KVM_INTERRUPT_SET + + This injects an edge type external interrupt into the guest once it's ready + to receive interrupts. When injected, the interrupt is done. + +b) KVM_INTERRUPT_UNSET + + This unsets any pending interrupt. + + Only available with KVM_CAP_PPC_UNSET_IRQ. + +c) KVM_INTERRUPT_SET_LEVEL + + This injects a level type external interrupt into the guest context. The + interrupt stays pending until a specific ioctl with KVM_INTERRUPT_UNSET + is triggered. + + Only available with KVM_CAP_PPC_IRQ_LEVEL. + +Note that any value for 'irq' other than the ones stated above is invalid +and incurs unexpected behavior. + +4.17 KVM_DEBUG_GUEST + +Capability: basic +Architectures: none +Type: vcpu ioctl +Parameters: none) +Returns: -1 on error + +Support for this has been removed. Use KVM_SET_GUEST_DEBUG instead. + +4.18 KVM_GET_MSRS + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_msrs (in/out) +Returns: 0 on success, -1 on error + +Reads model-specific registers from the vcpu. Supported msr indices can +be obtained using KVM_GET_MSR_INDEX_LIST. + +struct kvm_msrs { + __u32 nmsrs; /* number of msrs in entries */ + __u32 pad; + + struct kvm_msr_entry entries[0]; +}; + +struct kvm_msr_entry { + __u32 index; + __u32 reserved; + __u64 data; +}; + +Application code should set the 'nmsrs' member (which indicates the +size of the entries array) and the 'index' member of each array entry. +kvm will fill in the 'data' member. + +4.19 KVM_SET_MSRS + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_msrs (in) +Returns: 0 on success, -1 on error + +Writes model-specific registers to the vcpu. See KVM_GET_MSRS for the +data structures. + +Application code should set the 'nmsrs' member (which indicates the +size of the entries array), and the 'index' and 'data' members of each +array entry. + +4.20 KVM_SET_CPUID + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_cpuid (in) +Returns: 0 on success, -1 on error + +Defines the vcpu responses to the cpuid instruction. Applications +should use the KVM_SET_CPUID2 ioctl if available. + + +struct kvm_cpuid_entry { + __u32 function; + __u32 eax; + __u32 ebx; + __u32 ecx; + __u32 edx; + __u32 padding; +}; + +/* for KVM_SET_CPUID */ +struct kvm_cpuid { + __u32 nent; + __u32 padding; + struct kvm_cpuid_entry entries[0]; +}; + +4.21 KVM_SET_SIGNAL_MASK + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_signal_mask (in) +Returns: 0 on success, -1 on error + +Defines which signals are blocked during execution of KVM_RUN. This +signal mask temporarily overrides the threads signal mask. Any +unblocked signal received (except SIGKILL and SIGSTOP, which retain +their traditional behaviour) will cause KVM_RUN to return with -EINTR. + +Note the signal will only be delivered if not blocked by the original +signal mask. + +/* for KVM_SET_SIGNAL_MASK */ +struct kvm_signal_mask { + __u32 len; + __u8 sigset[0]; +}; + +4.22 KVM_GET_FPU + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_fpu (out) +Returns: 0 on success, -1 on error + +Reads the floating point state from the vcpu. + +/* for KVM_GET_FPU and KVM_SET_FPU */ +struct kvm_fpu { + __u8 fpr[8][16]; + __u16 fcw; + __u16 fsw; + __u8 ftwx; /* in fxsave format */ + __u8 pad1; + __u16 last_opcode; + __u64 last_ip; + __u64 last_dp; + __u8 xmm[16][16]; + __u32 mxcsr; + __u32 pad2; +}; + +4.23 KVM_SET_FPU + +Capability: basic +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_fpu (in) +Returns: 0 on success, -1 on error + +Writes the floating point state to the vcpu. + +/* for KVM_GET_FPU and KVM_SET_FPU */ +struct kvm_fpu { + __u8 fpr[8][16]; + __u16 fcw; + __u16 fsw; + __u8 ftwx; /* in fxsave format */ + __u8 pad1; + __u16 last_opcode; + __u64 last_ip; + __u64 last_dp; + __u8 xmm[16][16]; + __u32 mxcsr; + __u32 pad2; +}; + +4.24 KVM_CREATE_IRQCHIP + +Capability: KVM_CAP_IRQCHIP +Architectures: x86, ia64 +Type: vm ioctl +Parameters: none +Returns: 0 on success, -1 on error + +Creates an interrupt controller model in the kernel. On x86, creates a virtual +ioapic, a virtual PIC (two PICs, nested), and sets up future vcpus to have a +local APIC. IRQ routing for GSIs 0-15 is set to both PIC and IOAPIC; GSI 16-23 +only go to the IOAPIC. On ia64, a IOSAPIC is created. + +4.25 KVM_IRQ_LINE + +Capability: KVM_CAP_IRQCHIP +Architectures: x86, ia64 +Type: vm ioctl +Parameters: struct kvm_irq_level +Returns: 0 on success, -1 on error + +Sets the level of a GSI input to the interrupt controller model in the kernel. +Requires that an interrupt controller model has been previously created with +KVM_CREATE_IRQCHIP. Note that edge-triggered interrupts require the level +to be set to 1 and then back to 0. + +struct kvm_irq_level { + union { + __u32 irq; /* GSI */ + __s32 status; /* not used for KVM_IRQ_LEVEL */ + }; + __u32 level; /* 0 or 1 */ +}; + +4.26 KVM_GET_IRQCHIP + +Capability: KVM_CAP_IRQCHIP +Architectures: x86, ia64 +Type: vm ioctl +Parameters: struct kvm_irqchip (in/out) +Returns: 0 on success, -1 on error + +Reads the state of a kernel interrupt controller created with +KVM_CREATE_IRQCHIP into a buffer provided by the caller. + +struct kvm_irqchip { + __u32 chip_id; /* 0 = PIC1, 1 = PIC2, 2 = IOAPIC */ + __u32 pad; + union { + char dummy[512]; /* reserving space */ + struct kvm_pic_state pic; + struct kvm_ioapic_state ioapic; + } chip; +}; + +4.27 KVM_SET_IRQCHIP + +Capability: KVM_CAP_IRQCHIP +Architectures: x86, ia64 +Type: vm ioctl +Parameters: struct kvm_irqchip (in) +Returns: 0 on success, -1 on error + +Sets the state of a kernel interrupt controller created with +KVM_CREATE_IRQCHIP from a buffer provided by the caller. + +struct kvm_irqchip { + __u32 chip_id; /* 0 = PIC1, 1 = PIC2, 2 = IOAPIC */ + __u32 pad; + union { + char dummy[512]; /* reserving space */ + struct kvm_pic_state pic; + struct kvm_ioapic_state ioapic; + } chip; +}; + +4.28 KVM_XEN_HVM_CONFIG + +Capability: KVM_CAP_XEN_HVM +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_xen_hvm_config (in) +Returns: 0 on success, -1 on error + +Sets the MSR that the Xen HVM guest uses to initialize its hypercall +page, and provides the starting address and size of the hypercall +blobs in userspace. When the guest writes the MSR, kvm copies one +page of a blob (32- or 64-bit, depending on the vcpu mode) to guest +memory. + +struct kvm_xen_hvm_config { + __u32 flags; + __u32 msr; + __u64 blob_addr_32; + __u64 blob_addr_64; + __u8 blob_size_32; + __u8 blob_size_64; + __u8 pad2[30]; +}; + +4.29 KVM_GET_CLOCK + +Capability: KVM_CAP_ADJUST_CLOCK +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_clock_data (out) +Returns: 0 on success, -1 on error + +Gets the current timestamp of kvmclock as seen by the current guest. In +conjunction with KVM_SET_CLOCK, it is used to ensure monotonicity on scenarios +such as migration. + +struct kvm_clock_data { + __u64 clock; /* kvmclock current value */ + __u32 flags; + __u32 pad[9]; +}; + +4.30 KVM_SET_CLOCK + +Capability: KVM_CAP_ADJUST_CLOCK +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_clock_data (in) +Returns: 0 on success, -1 on error + +Sets the current timestamp of kvmclock to the value specified in its parameter. +In conjunction with KVM_GET_CLOCK, it is used to ensure monotonicity on scenarios +such as migration. + +struct kvm_clock_data { + __u64 clock; /* kvmclock current value */ + __u32 flags; + __u32 pad[9]; +}; + +4.31 KVM_GET_VCPU_EVENTS + +Capability: KVM_CAP_VCPU_EVENTS +Extended by: KVM_CAP_INTR_SHADOW +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_vcpu_event (out) +Returns: 0 on success, -1 on error + +Gets currently pending exceptions, interrupts, and NMIs as well as related +states of the vcpu. + +struct kvm_vcpu_events { + struct { + __u8 injected; + __u8 nr; + __u8 has_error_code; + __u8 pad; + __u32 error_code; + } exception; + struct { + __u8 injected; + __u8 nr; + __u8 soft; + __u8 shadow; + } interrupt; + struct { + __u8 injected; + __u8 pending; + __u8 masked; + __u8 pad; + } nmi; + __u32 sipi_vector; + __u32 flags; +}; + +KVM_VCPUEVENT_VALID_SHADOW may be set in the flags field to signal that +interrupt.shadow contains a valid state. Otherwise, this field is undefined. + +4.32 KVM_SET_VCPU_EVENTS + +Capability: KVM_CAP_VCPU_EVENTS +Extended by: KVM_CAP_INTR_SHADOW +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_vcpu_event (in) +Returns: 0 on success, -1 on error + +Set pending exceptions, interrupts, and NMIs as well as related states of the +vcpu. + +See KVM_GET_VCPU_EVENTS for the data structure. + +Fields that may be modified asynchronously by running VCPUs can be excluded +from the update. These fields are nmi.pending and sipi_vector. Keep the +corresponding bits in the flags field cleared to suppress overwriting the +current in-kernel state. The bits are: + +KVM_VCPUEVENT_VALID_NMI_PENDING - transfer nmi.pending to the kernel +KVM_VCPUEVENT_VALID_SIPI_VECTOR - transfer sipi_vector + +If KVM_CAP_INTR_SHADOW is available, KVM_VCPUEVENT_VALID_SHADOW can be set in +the flags field to signal that interrupt.shadow contains a valid state and +shall be written into the VCPU. + +4.33 KVM_GET_DEBUGREGS + +Capability: KVM_CAP_DEBUGREGS +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_debugregs (out) +Returns: 0 on success, -1 on error + +Reads debug registers from the vcpu. + +struct kvm_debugregs { + __u64 db[4]; + __u64 dr6; + __u64 dr7; + __u64 flags; + __u64 reserved[9]; +}; + +4.34 KVM_SET_DEBUGREGS + +Capability: KVM_CAP_DEBUGREGS +Architectures: x86 +Type: vm ioctl +Parameters: struct kvm_debugregs (in) +Returns: 0 on success, -1 on error + +Writes debug registers into the vcpu. + +See KVM_GET_DEBUGREGS for the data structure. The flags field is unused +yet and must be cleared on entry. + +4.35 KVM_SET_USER_MEMORY_REGION + +Capability: KVM_CAP_USER_MEM +Architectures: all +Type: vm ioctl +Parameters: struct kvm_userspace_memory_region (in) +Returns: 0 on success, -1 on error + +struct kvm_userspace_memory_region { + __u32 slot; + __u32 flags; + __u64 guest_phys_addr; + __u64 memory_size; /* bytes */ + __u64 userspace_addr; /* start of the userspace allocated memory */ +}; + +/* for kvm_memory_region::flags */ +#define KVM_MEM_LOG_DIRTY_PAGES 1UL + +This ioctl allows the user to create or modify a guest physical memory +slot. When changing an existing slot, it may be moved in the guest +physical memory space, or its flags may be modified. It may not be +resized. Slots may not overlap in guest physical address space. + +Memory for the region is taken starting at the address denoted by the +field userspace_addr, which must point at user addressable memory for +the entire memory slot size. Any object may back this memory, including +anonymous memory, ordinary files, and hugetlbfs. + +It is recommended that the lower 21 bits of guest_phys_addr and userspace_addr +be identical. This allows large pages in the guest to be backed by large +pages in the host. + +The flags field supports just one flag, KVM_MEM_LOG_DIRTY_PAGES, which +instructs kvm to keep track of writes to memory within the slot. See +the KVM_GET_DIRTY_LOG ioctl. + +When the KVM_CAP_SYNC_MMU capability, changes in the backing of the memory +region are automatically reflected into the guest. For example, an mmap() +that affects the region will be made visible immediately. Another example +is madvise(MADV_DROP). + +It is recommended to use this API instead of the KVM_SET_MEMORY_REGION ioctl. +The KVM_SET_MEMORY_REGION does not allow fine grained control over memory +allocation and is deprecated. + +4.36 KVM_SET_TSS_ADDR + +Capability: KVM_CAP_SET_TSS_ADDR +Architectures: x86 +Type: vm ioctl +Parameters: unsigned long tss_address (in) +Returns: 0 on success, -1 on error + +This ioctl defines the physical address of a three-page region in the guest +physical address space. The region must be within the first 4GB of the +guest physical address space and must not conflict with any memory slot +or any mmio address. The guest may malfunction if it accesses this memory +region. + +This ioctl is required on Intel-based hosts. This is needed on Intel hardware +because of a quirk in the virtualization implementation (see the internals +documentation when it pops into existence). + +4.37 KVM_ENABLE_CAP + +Capability: KVM_CAP_ENABLE_CAP +Architectures: ppc +Type: vcpu ioctl +Parameters: struct kvm_enable_cap (in) +Returns: 0 on success; -1 on error + ++Not all extensions are enabled by default. Using this ioctl the application +can enable an extension, making it available to the guest. + +On systems that do not support this ioctl, it always fails. On systems that +do support it, it only works for extensions that are supported for enablement. + +To check if a capability can be enabled, the KVM_CHECK_EXTENSION ioctl should +be used. + +struct kvm_enable_cap { + /* in */ + __u32 cap; + +The capability that is supposed to get enabled. + + __u32 flags; + +A bitfield indicating future enhancements. Has to be 0 for now. + + __u64 args[4]; + +Arguments for enabling a feature. If a feature needs initial values to +function properly, this is the place to put them. + + __u8 pad[64]; +}; + +4.38 KVM_GET_MP_STATE + +Capability: KVM_CAP_MP_STATE +Architectures: x86, ia64 +Type: vcpu ioctl +Parameters: struct kvm_mp_state (out) +Returns: 0 on success; -1 on error + +struct kvm_mp_state { + __u32 mp_state; +}; + +Returns the vcpu's current "multiprocessing state" (though also valid on +uniprocessor guests). + +Possible values are: + + - KVM_MP_STATE_RUNNABLE: the vcpu is currently running + - KVM_MP_STATE_UNINITIALIZED: the vcpu is an application processor (AP) + which has not yet received an INIT signal + - KVM_MP_STATE_INIT_RECEIVED: the vcpu has received an INIT signal, and is + now ready for a SIPI + - KVM_MP_STATE_HALTED: the vcpu has executed a HLT instruction and + is waiting for an interrupt + - KVM_MP_STATE_SIPI_RECEIVED: the vcpu has just received a SIPI (vector + accessible via KVM_GET_VCPU_EVENTS) + +This ioctl is only useful after KVM_CREATE_IRQCHIP. Without an in-kernel +irqchip, the multiprocessing state must be maintained by userspace. + +4.39 KVM_SET_MP_STATE + +Capability: KVM_CAP_MP_STATE +Architectures: x86, ia64 +Type: vcpu ioctl +Parameters: struct kvm_mp_state (in) +Returns: 0 on success; -1 on error + +Sets the vcpu's current "multiprocessing state"; see KVM_GET_MP_STATE for +arguments. + +This ioctl is only useful after KVM_CREATE_IRQCHIP. Without an in-kernel +irqchip, the multiprocessing state must be maintained by userspace. + +4.40 KVM_SET_IDENTITY_MAP_ADDR + +Capability: KVM_CAP_SET_IDENTITY_MAP_ADDR +Architectures: x86 +Type: vm ioctl +Parameters: unsigned long identity (in) +Returns: 0 on success, -1 on error + +This ioctl defines the physical address of a one-page region in the guest +physical address space. The region must be within the first 4GB of the +guest physical address space and must not conflict with any memory slot +or any mmio address. The guest may malfunction if it accesses this memory +region. + +This ioctl is required on Intel-based hosts. This is needed on Intel hardware +because of a quirk in the virtualization implementation (see the internals +documentation when it pops into existence). + +4.41 KVM_SET_BOOT_CPU_ID + +Capability: KVM_CAP_SET_BOOT_CPU_ID +Architectures: x86, ia64 +Type: vm ioctl +Parameters: unsigned long vcpu_id +Returns: 0 on success, -1 on error + +Define which vcpu is the Bootstrap Processor (BSP). Values are the same +as the vcpu id in KVM_CREATE_VCPU. If this ioctl is not called, the default +is vcpu 0. + +4.42 KVM_GET_XSAVE + +Capability: KVM_CAP_XSAVE +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_xsave (out) +Returns: 0 on success, -1 on error + +struct kvm_xsave { + __u32 region[1024]; +}; + +This ioctl would copy current vcpu's xsave struct to the userspace. + +4.43 KVM_SET_XSAVE + +Capability: KVM_CAP_XSAVE +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_xsave (in) +Returns: 0 on success, -1 on error + +struct kvm_xsave { + __u32 region[1024]; +}; + +This ioctl would copy userspace's xsave struct to the kernel. + +4.44 KVM_GET_XCRS + +Capability: KVM_CAP_XCRS +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_xcrs (out) +Returns: 0 on success, -1 on error + +struct kvm_xcr { + __u32 xcr; + __u32 reserved; + __u64 value; +}; + +struct kvm_xcrs { + __u32 nr_xcrs; + __u32 flags; + struct kvm_xcr xcrs[KVM_MAX_XCRS]; + __u64 padding[16]; +}; + +This ioctl would copy current vcpu's xcrs to the userspace. + +4.45 KVM_SET_XCRS + +Capability: KVM_CAP_XCRS +Architectures: x86 +Type: vcpu ioctl +Parameters: struct kvm_xcrs (in) +Returns: 0 on success, -1 on error + +struct kvm_xcr { + __u32 xcr; + __u32 reserved; + __u64 value; +}; + +struct kvm_xcrs { + __u32 nr_xcrs; + __u32 flags; + struct kvm_xcr xcrs[KVM_MAX_XCRS]; + __u64 padding[16]; +}; + +This ioctl would set vcpu's xcr to the value userspace specified. + +4.46 KVM_GET_SUPPORTED_CPUID + +Capability: KVM_CAP_EXT_CPUID +Architectures: x86 +Type: system ioctl +Parameters: struct kvm_cpuid2 (in/out) +Returns: 0 on success, -1 on error + +struct kvm_cpuid2 { + __u32 nent; + __u32 padding; + struct kvm_cpuid_entry2 entries[0]; +}; + +#define KVM_CPUID_FLAG_SIGNIFCANT_INDEX 1 +#define KVM_CPUID_FLAG_STATEFUL_FUNC 2 +#define KVM_CPUID_FLAG_STATE_READ_NEXT 4 + +struct kvm_cpuid_entry2 { + __u32 function; + __u32 index; + __u32 flags; + __u32 eax; + __u32 ebx; + __u32 ecx; + __u32 edx; + __u32 padding[3]; +}; + +This ioctl returns x86 cpuid features which are supported by both the hardware +and kvm. Userspace can use the information returned by this ioctl to +construct cpuid information (for KVM_SET_CPUID2) that is consistent with +hardware, kernel, and userspace capabilities, and with user requirements (for +example, the user may wish to constrain cpuid to emulate older hardware, +or for feature consistency across a cluster). + +Userspace invokes KVM_GET_SUPPORTED_CPUID by passing a kvm_cpuid2 structure +with the 'nent' field indicating the number of entries in the variable-size +array 'entries'. If the number of entries is too low to describe the cpu +capabilities, an error (E2BIG) is returned. If the number is too high, +the 'nent' field is adjusted and an error (ENOMEM) is returned. If the +number is just right, the 'nent' field is adjusted to the number of valid +entries in the 'entries' array, which is then filled. + +The entries returned are the host cpuid as returned by the cpuid instruction, +with unknown or unsupported features masked out. Some features (for example, +x2apic), may not be present in the host cpu, but are exposed by kvm if it can +emulate them efficiently. The fields in each entry are defined as follows: + + function: the eax value used to obtain the entry + index: the ecx value used to obtain the entry (for entries that are + affected by ecx) + flags: an OR of zero or more of the following: + KVM_CPUID_FLAG_SIGNIFCANT_INDEX: + if the index field is valid + KVM_CPUID_FLAG_STATEFUL_FUNC: + if cpuid for this function returns different values for successive + invocations; there will be several entries with the same function, + all with this flag set + KVM_CPUID_FLAG_STATE_READ_NEXT: + for KVM_CPUID_FLAG_STATEFUL_FUNC entries, set if this entry is + the first entry to be read by a cpu + eax, ebx, ecx, edx: the values returned by the cpuid instruction for + this function/index combination + +4.47 KVM_PPC_GET_PVINFO + +Capability: KVM_CAP_PPC_GET_PVINFO +Architectures: ppc +Type: vm ioctl +Parameters: struct kvm_ppc_pvinfo (out) +Returns: 0 on success, !0 on error + +struct kvm_ppc_pvinfo { + __u32 flags; + __u32 hcall[4]; + __u8 pad[108]; +}; + +This ioctl fetches PV specific information that need to be passed to the guest +using the device tree or other means from vm context. + +For now the only implemented piece of information distributed here is an array +of 4 instructions that make up a hypercall. + +If any additional field gets added to this structure later on, a bit for that +additional piece of information will be set in the flags bitmap. + +4.48 KVM_ASSIGN_PCI_DEVICE + +Capability: KVM_CAP_DEVICE_ASSIGNMENT +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_pci_dev (in) +Returns: 0 on success, -1 on error + +Assigns a host PCI device to the VM. + +struct kvm_assigned_pci_dev { + __u32 assigned_dev_id; + __u32 busnr; + __u32 devfn; + __u32 flags; + __u32 segnr; + union { + __u32 reserved[11]; + }; +}; + +The PCI device is specified by the triple segnr, busnr, and devfn. +Identification in succeeding service requests is done via assigned_dev_id. The +following flags are specified: + +/* Depends on KVM_CAP_IOMMU */ +#define KVM_DEV_ASSIGN_ENABLE_IOMMU (1 << 0) + +4.49 KVM_DEASSIGN_PCI_DEVICE + +Capability: KVM_CAP_DEVICE_DEASSIGNMENT +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_pci_dev (in) +Returns: 0 on success, -1 on error + +Ends PCI device assignment, releasing all associated resources. + +See KVM_CAP_DEVICE_ASSIGNMENT for the data structure. Only assigned_dev_id is +used in kvm_assigned_pci_dev to identify the device. + +4.50 KVM_ASSIGN_DEV_IRQ + +Capability: KVM_CAP_ASSIGN_DEV_IRQ +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_irq (in) +Returns: 0 on success, -1 on error + +Assigns an IRQ to a passed-through device. + +struct kvm_assigned_irq { + __u32 assigned_dev_id; + __u32 host_irq; + __u32 guest_irq; + __u32 flags; + union { + struct { + __u32 addr_lo; + __u32 addr_hi; + __u32 data; + } guest_msi; + __u32 reserved[12]; + }; +}; + +The following flags are defined: + +#define KVM_DEV_IRQ_HOST_INTX (1 << 0) +#define KVM_DEV_IRQ_HOST_MSI (1 << 1) +#define KVM_DEV_IRQ_HOST_MSIX (1 << 2) + +#define KVM_DEV_IRQ_GUEST_INTX (1 << 8) +#define KVM_DEV_IRQ_GUEST_MSI (1 << 9) +#define KVM_DEV_IRQ_GUEST_MSIX (1 << 10) + +It is not valid to specify multiple types per host or guest IRQ. However, the +IRQ type of host and guest can differ or can even be null. + +4.51 KVM_DEASSIGN_DEV_IRQ + +Capability: KVM_CAP_ASSIGN_DEV_IRQ +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_irq (in) +Returns: 0 on success, -1 on error + +Ends an IRQ assignment to a passed-through device. + +See KVM_ASSIGN_DEV_IRQ for the data structure. The target device is specified +by assigned_dev_id, flags must correspond to the IRQ type specified on +KVM_ASSIGN_DEV_IRQ. Partial deassignment of host or guest IRQ is allowed. + +4.52 KVM_SET_GSI_ROUTING + +Capability: KVM_CAP_IRQ_ROUTING +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_irq_routing (in) +Returns: 0 on success, -1 on error + +Sets the GSI routing table entries, overwriting any previously set entries. + +struct kvm_irq_routing { + __u32 nr; + __u32 flags; + struct kvm_irq_routing_entry entries[0]; +}; + +No flags are specified so far, the corresponding field must be set to zero. + +struct kvm_irq_routing_entry { + __u32 gsi; + __u32 type; + __u32 flags; + __u32 pad; + union { + struct kvm_irq_routing_irqchip irqchip; + struct kvm_irq_routing_msi msi; + __u32 pad[8]; + } u; +}; + +/* gsi routing entry types */ +#define KVM_IRQ_ROUTING_IRQCHIP 1 +#define KVM_IRQ_ROUTING_MSI 2 + +No flags are specified so far, the corresponding field must be set to zero. + +struct kvm_irq_routing_irqchip { + __u32 irqchip; + __u32 pin; +}; + +struct kvm_irq_routing_msi { + __u32 address_lo; + __u32 address_hi; + __u32 data; + __u32 pad; +}; + +4.53 KVM_ASSIGN_SET_MSIX_NR + +Capability: KVM_CAP_DEVICE_MSIX +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_msix_nr (in) +Returns: 0 on success, -1 on error + +Set the number of MSI-X interrupts for an assigned device. This service can +only be called once in the lifetime of an assigned device. + +struct kvm_assigned_msix_nr { + __u32 assigned_dev_id; + __u16 entry_nr; + __u16 padding; +}; + +#define KVM_MAX_MSIX_PER_DEV 256 + +4.54 KVM_ASSIGN_SET_MSIX_ENTRY + +Capability: KVM_CAP_DEVICE_MSIX +Architectures: x86 ia64 +Type: vm ioctl +Parameters: struct kvm_assigned_msix_entry (in) +Returns: 0 on success, -1 on error + +Specifies the routing of an MSI-X assigned device interrupt to a GSI. Setting +the GSI vector to zero means disabling the interrupt. + +struct kvm_assigned_msix_entry { + __u32 assigned_dev_id; + __u32 gsi; + __u16 entry; /* The index of entry in the MSI-X table */ + __u16 padding[3]; +}; + +5. The kvm_run structure + +Application code obtains a pointer to the kvm_run structure by +mmap()ing a vcpu fd. From that point, application code can control +execution by changing fields in kvm_run prior to calling the KVM_RUN +ioctl, and obtain information about the reason KVM_RUN returned by +looking up structure members. + +struct kvm_run { + /* in */ + __u8 request_interrupt_window; + +Request that KVM_RUN return when it becomes possible to inject external +interrupts into the guest. Useful in conjunction with KVM_INTERRUPT. + + __u8 padding1[7]; + + /* out */ + __u32 exit_reason; + +When KVM_RUN has returned successfully (return value 0), this informs +application code why KVM_RUN has returned. Allowable values for this +field are detailed below. + + __u8 ready_for_interrupt_injection; + +If request_interrupt_window has been specified, this field indicates +an interrupt can be injected now with KVM_INTERRUPT. + + __u8 if_flag; + +The value of the current interrupt flag. Only valid if in-kernel +local APIC is not used. + + __u8 padding2[2]; + + /* in (pre_kvm_run), out (post_kvm_run) */ + __u64 cr8; + +The value of the cr8 register. Only valid if in-kernel local APIC is +not used. Both input and output. + + __u64 apic_base; + +The value of the APIC BASE msr. Only valid if in-kernel local +APIC is not used. Both input and output. + + union { + /* KVM_EXIT_UNKNOWN */ + struct { + __u64 hardware_exit_reason; + } hw; + +If exit_reason is KVM_EXIT_UNKNOWN, the vcpu has exited due to unknown +reasons. Further architecture-specific information is available in +hardware_exit_reason. + + /* KVM_EXIT_FAIL_ENTRY */ + struct { + __u64 hardware_entry_failure_reason; + } fail_entry; + +If exit_reason is KVM_EXIT_FAIL_ENTRY, the vcpu could not be run due +to unknown reasons. Further architecture-specific information is +available in hardware_entry_failure_reason. + + /* KVM_EXIT_EXCEPTION */ + struct { + __u32 exception; + __u32 error_code; + } ex; + +Unused. + + /* KVM_EXIT_IO */ + struct { +#define KVM_EXIT_IO_IN 0 +#define KVM_EXIT_IO_OUT 1 + __u8 direction; + __u8 size; /* bytes */ + __u16 port; + __u32 count; + __u64 data_offset; /* relative to kvm_run start */ + } io; + +If exit_reason is KVM_EXIT_IO, then the vcpu has +executed a port I/O instruction which could not be satisfied by kvm. +data_offset describes where the data is located (KVM_EXIT_IO_OUT) or +where kvm expects application code to place the data for the next +KVM_RUN invocation (KVM_EXIT_IO_IN). Data format is a packed array. + + struct { + struct kvm_debug_exit_arch arch; + } debug; + +Unused. + + /* KVM_EXIT_MMIO */ + struct { + __u64 phys_addr; + __u8 data[8]; + __u32 len; + __u8 is_write; + } mmio; + +If exit_reason is KVM_EXIT_MMIO, then the vcpu has +executed a memory-mapped I/O instruction which could not be satisfied +by kvm. The 'data' member contains the written data if 'is_write' is +true, and should be filled by application code otherwise. + +NOTE: For KVM_EXIT_IO, KVM_EXIT_MMIO and KVM_EXIT_OSI, the corresponding +operations are complete (and guest state is consistent) only after userspace +has re-entered the kernel with KVM_RUN. The kernel side will first finish +incomplete operations and then check for pending signals. Userspace +can re-enter the guest with an unmasked signal pending to complete +pending operations. + + /* KVM_EXIT_HYPERCALL */ + struct { + __u64 nr; + __u64 args[6]; + __u64 ret; + __u32 longmode; + __u32 pad; + } hypercall; + +Unused. This was once used for 'hypercall to userspace'. To implement +such functionality, use KVM_EXIT_IO (x86) or KVM_EXIT_MMIO (all except s390). +Note KVM_EXIT_IO is significantly faster than KVM_EXIT_MMIO. + + /* KVM_EXIT_TPR_ACCESS */ + struct { + __u64 rip; + __u32 is_write; + __u32 pad; + } tpr_access; + +To be documented (KVM_TPR_ACCESS_REPORTING). + + /* KVM_EXIT_S390_SIEIC */ + struct { + __u8 icptcode; + __u64 mask; /* psw upper half */ + __u64 addr; /* psw lower half */ + __u16 ipa; + __u32 ipb; + } s390_sieic; + +s390 specific. + + /* KVM_EXIT_S390_RESET */ +#define KVM_S390_RESET_POR 1 +#define KVM_S390_RESET_CLEAR 2 +#define KVM_S390_RESET_SUBSYSTEM 4 +#define KVM_S390_RESET_CPU_INIT 8 +#define KVM_S390_RESET_IPL 16 + __u64 s390_reset_flags; + +s390 specific. + + /* KVM_EXIT_DCR */ + struct { + __u32 dcrn; + __u32 data; + __u8 is_write; + } dcr; + +powerpc specific. + + /* KVM_EXIT_OSI */ + struct { + __u64 gprs[32]; + } osi; + +MOL uses a special hypercall interface it calls 'OSI'. To enable it, we catch +hypercalls and exit with this exit struct that contains all the guest gprs. + +If exit_reason is KVM_EXIT_OSI, then the vcpu has triggered such a hypercall. +Userspace can now handle the hypercall and when it's done modify the gprs as +necessary. Upon guest entry all guest GPRs will then be replaced by the values +in this struct. + + /* Fix the size of the union. */ + char padding[256]; + }; +}; diff --git a/Documentation/virtual/kvm/cpuid.txt b/Documentation/virtual/kvm/cpuid.txt new file mode 100644 index 00000000000..882068538c9 --- /dev/null +++ b/Documentation/virtual/kvm/cpuid.txt @@ -0,0 +1,45 @@ +KVM CPUID bits +Glauber Costa <glommer@redhat.com>, Red Hat Inc, 2010 +===================================================== + +A guest running on a kvm host, can check some of its features using +cpuid. This is not always guaranteed to work, since userspace can +mask-out some, or even all KVM-related cpuid features before launching +a guest. + +KVM cpuid functions are: + +function: KVM_CPUID_SIGNATURE (0x40000000) +returns : eax = 0, + ebx = 0x4b4d564b, + ecx = 0x564b4d56, + edx = 0x4d. +Note that this value in ebx, ecx and edx corresponds to the string "KVMKVMKVM". +This function queries the presence of KVM cpuid leafs. + + +function: define KVM_CPUID_FEATURES (0x40000001) +returns : ebx, ecx, edx = 0 + eax = and OR'ed group of (1 << flag), where each flags is: + + +flag || value || meaning +============================================================================= +KVM_FEATURE_CLOCKSOURCE || 0 || kvmclock available at msrs + || || 0x11 and 0x12. +------------------------------------------------------------------------------ +KVM_FEATURE_NOP_IO_DELAY || 1 || not necessary to perform delays + || || on PIO operations. +------------------------------------------------------------------------------ +KVM_FEATURE_MMU_OP || 2 || deprecated. +------------------------------------------------------------------------------ +KVM_FEATURE_CLOCKSOURCE2 || 3 || kvmclock available at msrs + || || 0x4b564d00 and 0x4b564d01 +------------------------------------------------------------------------------ +KVM_FEATURE_ASYNC_PF || 4 || async pf can be enabled by + || || writing to msr 0x4b564d02 +------------------------------------------------------------------------------ +KVM_FEATURE_CLOCKSOURCE_STABLE_BIT || 24 || host will warn if no guest-side + || || per-cpu warps are expected in + || || kvmclock. +------------------------------------------------------------------------------ diff --git a/Documentation/virtual/kvm/locking.txt b/Documentation/virtual/kvm/locking.txt new file mode 100644 index 00000000000..3b4cd3bf563 --- /dev/null +++ b/Documentation/virtual/kvm/locking.txt @@ -0,0 +1,25 @@ +KVM Lock Overview +================= + +1. Acquisition Orders +--------------------- + +(to be written) + +2. Reference +------------ + +Name: kvm_lock +Type: raw_spinlock +Arch: any +Protects: - vm_list + - hardware virtualization enable/disable +Comment: 'raw' because hardware enabling/disabling must be atomic /wrt + migration. + +Name: kvm_arch::tsc_write_lock +Type: raw_spinlock +Arch: x86 +Protects: - kvm_arch::{last_tsc_write,last_tsc_nsec,last_tsc_offset} + - tsc offset in vmcb +Comment: 'raw' because updating the tsc offsets must not be preempted. diff --git a/Documentation/virtual/kvm/mmu.txt b/Documentation/virtual/kvm/mmu.txt new file mode 100644 index 00000000000..f46aa58389c --- /dev/null +++ b/Documentation/virtual/kvm/mmu.txt @@ -0,0 +1,348 @@ +The x86 kvm shadow mmu +====================== + +The mmu (in arch/x86/kvm, files mmu.[ch] and paging_tmpl.h) is responsible +for presenting a standard x86 mmu to the guest, while translating guest +physical addresses to host physical addresses. + +The mmu code attempts to satisfy the following requirements: + +- correctness: the guest should not be able to determine that it is running + on an emulated mmu except for timing (we attempt to comply + with the specification, not emulate the characteristics of + a particular implementation such as tlb size) +- security: the guest must not be able to touch host memory not assigned + to it +- performance: minimize the performance penalty imposed by the mmu +- scaling: need to scale to large memory and large vcpu guests +- hardware: support the full range of x86 virtualization hardware +- integration: Linux memory management code must be in control of guest memory + so that swapping, page migration, page merging, transparent + hugepages, and similar features work without change +- dirty tracking: report writes to guest memory to enable live migration + and framebuffer-based displays +- footprint: keep the amount of pinned kernel memory low (most memory + should be shrinkable) +- reliability: avoid multipage or GFP_ATOMIC allocations + +Acronyms +======== + +pfn host page frame number +hpa host physical address +hva host virtual address +gfn guest frame number +gpa guest physical address +gva guest virtual address +ngpa nested guest physical address +ngva nested guest virtual address +pte page table entry (used also to refer generically to paging structure + entries) +gpte guest pte (referring to gfns) +spte shadow pte (referring to pfns) +tdp two dimensional paging (vendor neutral term for NPT and EPT) + +Virtual and real hardware supported +=================================== + +The mmu supports first-generation mmu hardware, which allows an atomic switch +of the current paging mode and cr3 during guest entry, as well as +two-dimensional paging (AMD's NPT and Intel's EPT). The emulated hardware +it exposes is the traditional 2/3/4 level x86 mmu, with support for global +pages, pae, pse, pse36, cr0.wp, and 1GB pages. Work is in progress to support +exposing NPT capable hardware on NPT capable hosts. + +Translation +=========== + +The primary job of the mmu is to program the processor's mmu to translate +addresses for the guest. Different translations are required at different +times: + +- when guest paging is disabled, we translate guest physical addresses to + host physical addresses (gpa->hpa) +- when guest paging is enabled, we translate guest virtual addresses, to + guest physical addresses, to host physical addresses (gva->gpa->hpa) +- when the guest launches a guest of its own, we translate nested guest + virtual addresses, to nested guest physical addresses, to guest physical + addresses, to host physical addresses (ngva->ngpa->gpa->hpa) + +The primary challenge is to encode between 1 and 3 translations into hardware +that support only 1 (traditional) and 2 (tdp) translations. When the +number of required translations matches the hardware, the mmu operates in +direct mode; otherwise it operates in shadow mode (see below). + +Memory +====== + +Guest memory (gpa) is part of the user address space of the process that is +using kvm. Userspace defines the translation between guest addresses and user +addresses (gpa->hva); note that two gpas may alias to the same hva, but not +vice versa. + +These hvas may be backed using any method available to the host: anonymous +memory, file backed memory, and device memory. Memory might be paged by the +host at any time. + +Events +====== + +The mmu is driven by events, some from the guest, some from the host. + +Guest generated events: +- writes to control registers (especially cr3) +- invlpg/invlpga instruction execution +- access to missing or protected translations + +Host generated events: +- changes in the gpa->hpa translation (either through gpa->hva changes or + through hva->hpa changes) +- memory pressure (the shrinker) + +Shadow pages +============ + +The principal data structure is the shadow page, 'struct kvm_mmu_page'. A +shadow page contains 512 sptes, which can be either leaf or nonleaf sptes. A +shadow page may contain a mix of leaf and nonleaf sptes. + +A nonleaf spte allows the hardware mmu to reach the leaf pages and +is not related to a translation directly. It points to other shadow pages. + +A leaf spte corresponds to either one or two translations encoded into +one paging structure entry. These are always the lowest level of the +translation stack, with optional higher level translations left to NPT/EPT. +Leaf ptes point at guest pages. + +The following table shows translations encoded by leaf ptes, with higher-level +translations in parentheses: + + Non-nested guests: + nonpaging: gpa->hpa + paging: gva->gpa->hpa + paging, tdp: (gva->)gpa->hpa + Nested guests: + non-tdp: ngva->gpa->hpa (*) + tdp: (ngva->)ngpa->gpa->hpa + +(*) the guest hypervisor will encode the ngva->gpa translation into its page + tables if npt is not present + +Shadow pages contain the following information: + role.level: + The level in the shadow paging hierarchy that this shadow page belongs to. + 1=4k sptes, 2=2M sptes, 3=1G sptes, etc. + role.direct: + If set, leaf sptes reachable from this page are for a linear range. + Examples include real mode translation, large guest pages backed by small + host pages, and gpa->hpa translations when NPT or EPT is active. + The linear range starts at (gfn << PAGE_SHIFT) and its size is determined + by role.level (2MB for first level, 1GB for second level, 0.5TB for third + level, 256TB for fourth level) + If clear, this page corresponds to a guest page table denoted by the gfn + field. + role.quadrant: + When role.cr4_pae=0, the guest uses 32-bit gptes while the host uses 64-bit + sptes. That means a guest page table contains more ptes than the host, + so multiple shadow pages are needed to shadow one guest page. + For first-level shadow pages, role.quadrant can be 0 or 1 and denotes the + first or second 512-gpte block in the guest page table. For second-level + page tables, each 32-bit gpte is converted to two 64-bit sptes + (since each first-level guest page is shadowed by two first-level + shadow pages) so role.quadrant takes values in the range 0..3. Each + quadrant maps 1GB virtual address space. + role.access: + Inherited guest access permissions in the form uwx. Note execute + permission is positive, not negative. + role.invalid: + The page is invalid and should not be used. It is a root page that is + currently pinned (by a cpu hardware register pointing to it); once it is + unpinned it will be destroyed. + role.cr4_pae: + Contains the value of cr4.pae for which the page is valid (e.g. whether + 32-bit or 64-bit gptes are in use). + role.nxe: + Contains the value of efer.nxe for which the page is valid. + role.cr0_wp: + Contains the value of cr0.wp for which the page is valid. + gfn: + Either the guest page table containing the translations shadowed by this + page, or the base page frame for linear translations. See role.direct. + spt: + A pageful of 64-bit sptes containing the translations for this page. + Accessed by both kvm and hardware. + The page pointed to by spt will have its page->private pointing back + at the shadow page structure. + sptes in spt point either at guest pages, or at lower-level shadow pages. + Specifically, if sp1 and sp2 are shadow pages, then sp1->spt[n] may point + at __pa(sp2->spt). sp2 will point back at sp1 through parent_pte. + The spt array forms a DAG structure with the shadow page as a node, and + guest pages as leaves. + gfns: + An array of 512 guest frame numbers, one for each present pte. Used to + perform a reverse map from a pte to a gfn. When role.direct is set, any + element of this array can be calculated from the gfn field when used, in + this case, the array of gfns is not allocated. See role.direct and gfn. + slot_bitmap: + A bitmap containing one bit per memory slot. If the page contains a pte + mapping a page from memory slot n, then bit n of slot_bitmap will be set + (if a page is aliased among several slots, then it is not guaranteed that + all slots will be marked). + Used during dirty logging to avoid scanning a shadow page if none if its + pages need tracking. + root_count: + A counter keeping track of how many hardware registers (guest cr3 or + pdptrs) are now pointing at the page. While this counter is nonzero, the + page cannot be destroyed. See role.invalid. + multimapped: + Whether there exist multiple sptes pointing at this page. + parent_pte/parent_ptes: + If multimapped is zero, parent_pte points at the single spte that points at + this page's spt. Otherwise, parent_ptes points at a data structure + with a list of parent_ptes. + unsync: + If true, then the translations in this page may not match the guest's + translation. This is equivalent to the state of the tlb when a pte is + changed but before the tlb entry is flushed. Accordingly, unsync ptes + are synchronized when the guest executes invlpg or flushes its tlb by + other means. Valid for leaf pages. + unsync_children: + How many sptes in the page point at pages that are unsync (or have + unsynchronized children). + unsync_child_bitmap: + A bitmap indicating which sptes in spt point (directly or indirectly) at + pages that may be unsynchronized. Used to quickly locate all unsychronized + pages reachable from a given page. + +Reverse map +=========== + +The mmu maintains a reverse mapping whereby all ptes mapping a page can be +reached given its gfn. This is used, for example, when swapping out a page. + +Synchronized and unsynchronized pages +===================================== + +The guest uses two events to synchronize its tlb and page tables: tlb flushes +and page invalidations (invlpg). + +A tlb flush means that we need to synchronize all sptes reachable from the +guest's cr3. This is expensive, so we keep all guest page tables write +protected, and synchronize sptes to gptes when a gpte is written. + +A special case is when a guest page table is reachable from the current +guest cr3. In this case, the guest is obliged to issue an invlpg instruction +before using the translation. We take advantage of that by removing write +protection from the guest page, and allowing the guest to modify it freely. +We synchronize modified gptes when the guest invokes invlpg. This reduces +the amount of emulation we have to do when the guest modifies multiple gptes, +or when the a guest page is no longer used as a page table and is used for +random guest data. + +As a side effect we have to resynchronize all reachable unsynchronized shadow +pages on a tlb flush. + + +Reaction to events +================== + +- guest page fault (or npt page fault, or ept violation) + +This is the most complicated event. The cause of a page fault can be: + + - a true guest fault (the guest translation won't allow the access) (*) + - access to a missing translation + - access to a protected translation + - when logging dirty pages, memory is write protected + - synchronized shadow pages are write protected (*) + - access to untranslatable memory (mmio) + + (*) not applicable in direct mode + +Handling a page fault is performed as follows: + + - if needed, walk the guest page tables to determine the guest translation + (gva->gpa or ngpa->gpa) + - if permissions are insufficient, reflect the fault back to the guest + - determine the host page + - if this is an mmio request, there is no host page; call the emulator + to emulate the instruction instead + - walk the shadow page table to find the spte for the translation, + instantiating missing intermediate page tables as necessary + - try to unsynchronize the page + - if successful, we can let the guest continue and modify the gpte + - emulate the instruction + - if failed, unshadow the page and let the guest continue + - update any translations that were modified by the instruction + +invlpg handling: + + - walk the shadow page hierarchy and drop affected translations + - try to reinstantiate the indicated translation in the hope that the + guest will use it in the near future + +Guest control register updates: + +- mov to cr3 + - look up new shadow roots + - synchronize newly reachable shadow pages + +- mov to cr0/cr4/efer + - set up mmu context for new paging mode + - look up new shadow roots + - synchronize newly reachable shadow pages + +Host translation updates: + + - mmu notifier called with updated hva + - look up affected sptes through reverse map + - drop (or update) translations + +Emulating cr0.wp +================ + +If tdp is not enabled, the host must keep cr0.wp=1 so page write protection +works for the guest kernel, not guest guest userspace. When the guest +cr0.wp=1, this does not present a problem. However when the guest cr0.wp=0, +we cannot map the permissions for gpte.u=1, gpte.w=0 to any spte (the +semantics require allowing any guest kernel access plus user read access). + +We handle this by mapping the permissions to two possible sptes, depending +on fault type: + +- kernel write fault: spte.u=0, spte.w=1 (allows full kernel access, + disallows user access) +- read fault: spte.u=1, spte.w=0 (allows full read access, disallows kernel + write access) + +(user write faults generate a #PF) + +Large pages +=========== + +The mmu supports all combinations of large and small guest and host pages. +Supported page sizes include 4k, 2M, 4M, and 1G. 4M pages are treated as +two separate 2M pages, on both guest and host, since the mmu always uses PAE +paging. + +To instantiate a large spte, four constraints must be satisfied: + +- the spte must point to a large host page +- the guest pte must be a large pte of at least equivalent size (if tdp is + enabled, there is no guest pte and this condition is satisified) +- if the spte will be writeable, the large page frame may not overlap any + write-protected pages +- the guest page must be wholly contained by a single memory slot + +To check the last two conditions, the mmu maintains a ->write_count set of +arrays for each memory slot and large page size. Every write protected page +causes its write_count to be incremented, thus preventing instantiation of +a large spte. The frames at the end of an unaligned memory slot have +artificically inflated ->write_counts so they can never be instantiated. + +Further reading +=============== + +- NPT presentation from KVM Forum 2008 + http://www.linux-kvm.org/wiki/images/c/c8/KvmForum2008%24kdf2008_21.pdf + diff --git a/Documentation/virtual/kvm/msr.txt b/Documentation/virtual/kvm/msr.txt new file mode 100644 index 00000000000..d079aed27e0 --- /dev/null +++ b/Documentation/virtual/kvm/msr.txt @@ -0,0 +1,187 @@ +KVM-specific MSRs. +Glauber Costa <glommer@redhat.com>, Red Hat Inc, 2010 +===================================================== + +KVM makes use of some custom MSRs to service some requests. + +Custom MSRs have a range reserved for them, that goes from +0x4b564d00 to 0x4b564dff. There are MSRs outside this area, +but they are deprecated and their use is discouraged. + +Custom MSR list +-------- + +The current supported Custom MSR list is: + +MSR_KVM_WALL_CLOCK_NEW: 0x4b564d00 + + data: 4-byte alignment physical address of a memory area which must be + in guest RAM. This memory is expected to hold a copy of the following + structure: + + struct pvclock_wall_clock { + u32 version; + u32 sec; + u32 nsec; + } __attribute__((__packed__)); + + whose data will be filled in by the hypervisor. The hypervisor is only + guaranteed to update this data at the moment of MSR write. + Users that want to reliably query this information more than once have + to write more than once to this MSR. Fields have the following meanings: + + version: guest has to check version before and after grabbing + time information and check that they are both equal and even. + An odd version indicates an in-progress update. + + sec: number of seconds for wallclock. + + nsec: number of nanoseconds for wallclock. + + Note that although MSRs are per-CPU entities, the effect of this + particular MSR is global. + + Availability of this MSR must be checked via bit 3 in 0x4000001 cpuid + leaf prior to usage. + +MSR_KVM_SYSTEM_TIME_NEW: 0x4b564d01 + + data: 4-byte aligned physical address of a memory area which must be in + guest RAM, plus an enable bit in bit 0. This memory is expected to hold + a copy of the following structure: + + struct pvclock_vcpu_time_info { + u32 version; + u32 pad0; + u64 tsc_timestamp; + u64 system_time; + u32 tsc_to_system_mul; + s8 tsc_shift; + u8 flags; + u8 pad[2]; + } __attribute__((__packed__)); /* 32 bytes */ + + whose data will be filled in by the hypervisor periodically. Only one + write, or registration, is needed for each VCPU. The interval between + updates of this structure is arbitrary and implementation-dependent. + The hypervisor may update this structure at any time it sees fit until + anything with bit0 == 0 is written to it. + + Fields have the following meanings: + + version: guest has to check version before and after grabbing + time information and check that they are both equal and even. + An odd version indicates an in-progress update. + + tsc_timestamp: the tsc value at the current VCPU at the time + of the update of this structure. Guests can subtract this value + from current tsc to derive a notion of elapsed time since the + structure update. + + system_time: a host notion of monotonic time, including sleep + time at the time this structure was last updated. Unit is + nanoseconds. + + tsc_to_system_mul: a function of the tsc frequency. One has + to multiply any tsc-related quantity by this value to get + a value in nanoseconds, besides dividing by 2^tsc_shift + + tsc_shift: cycle to nanosecond divider, as a power of two, to + allow for shift rights. One has to shift right any tsc-related + quantity by this value to get a value in nanoseconds, besides + multiplying by tsc_to_system_mul. + + With this information, guests can derive per-CPU time by + doing: + + time = (current_tsc - tsc_timestamp) + time = (time * tsc_to_system_mul) >> tsc_shift + time = time + system_time + + flags: bits in this field indicate extended capabilities + coordinated between the guest and the hypervisor. Availability + of specific flags has to be checked in 0x40000001 cpuid leaf. + Current flags are: + + flag bit | cpuid bit | meaning + ------------------------------------------------------------- + | | time measures taken across + 0 | 24 | multiple cpus are guaranteed to + | | be monotonic + ------------------------------------------------------------- + + Availability of this MSR must be checked via bit 3 in 0x4000001 cpuid + leaf prior to usage. + + +MSR_KVM_WALL_CLOCK: 0x11 + + data and functioning: same as MSR_KVM_WALL_CLOCK_NEW. Use that instead. + + This MSR falls outside the reserved KVM range and may be removed in the + future. Its usage is deprecated. + + Availability of this MSR must be checked via bit 0 in 0x4000001 cpuid + leaf prior to usage. + +MSR_KVM_SYSTEM_TIME: 0x12 + + data and functioning: same as MSR_KVM_SYSTEM_TIME_NEW. Use that instead. + + This MSR falls outside the reserved KVM range and may be removed in the + future. Its usage is deprecated. + + Availability of this MSR must be checked via bit 0 in 0x4000001 cpuid + leaf prior to usage. + + The suggested algorithm for detecting kvmclock presence is then: + + if (!kvm_para_available()) /* refer to cpuid.txt */ + return NON_PRESENT; + + flags = cpuid_eax(0x40000001); + if (flags & 3) { + msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW; + msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW; + return PRESENT; + } else if (flags & 0) { + msr_kvm_system_time = MSR_KVM_SYSTEM_TIME; + msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK; + return PRESENT; + } else + return NON_PRESENT; + +MSR_KVM_ASYNC_PF_EN: 0x4b564d02 + data: Bits 63-6 hold 64-byte aligned physical address of a + 64 byte memory area which must be in guest RAM and must be + zeroed. Bits 5-2 are reserved and should be zero. Bit 0 is 1 + when asynchronous page faults are enabled on the vcpu 0 when + disabled. Bit 2 is 1 if asynchronous page faults can be injected + when vcpu is in cpl == 0. + + First 4 byte of 64 byte memory location will be written to by + the hypervisor at the time of asynchronous page fault (APF) + injection to indicate type of asynchronous page fault. Value + of 1 means that the page referred to by the page fault is not + present. Value 2 means that the page is now available. Disabling + interrupt inhibits APFs. Guest must not enable interrupt + before the reason is read, or it may be overwritten by another + APF. Since APF uses the same exception vector as regular page + fault guest must reset the reason to 0 before it does + something that can generate normal page fault. If during page + fault APF reason is 0 it means that this is regular page + fault. + + During delivery of type 1 APF cr2 contains a token that will + be used to notify a guest when missing page becomes + available. When page becomes available type 2 APF is sent with + cr2 set to the token associated with the page. There is special + kind of token 0xffffffff which tells vcpu that it should wake + up all processes waiting for APFs and no individual type 2 APFs + will be sent. + + If APF is disabled while there are outstanding APFs, they will + not be delivered. + + Currently type 2 APF will be always delivered on the same vcpu as + type 1 was, but guest should not rely on that. diff --git a/Documentation/virtual/kvm/ppc-pv.txt b/Documentation/virtual/kvm/ppc-pv.txt new file mode 100644 index 00000000000..3ab969c5904 --- /dev/null +++ b/Documentation/virtual/kvm/ppc-pv.txt @@ -0,0 +1,196 @@ +The PPC KVM paravirtual interface +================================= + +The basic execution principle by which KVM on PowerPC works is to run all kernel +space code in PR=1 which is user space. This way we trap all privileged +instructions and can emulate them accordingly. + +Unfortunately that is also the downfall. There are quite some privileged +instructions that needlessly return us to the hypervisor even though they +could be handled differently. + +This is what the PPC PV interface helps with. It takes privileged instructions +and transforms them into unprivileged ones with some help from the hypervisor. +This cuts down virtualization costs by about 50% on some of my benchmarks. + +The code for that interface can be found in arch/powerpc/kernel/kvm* + +Querying for existence +====================== + +To find out if we're running on KVM or not, we leverage the device tree. When +Linux is running on KVM, a node /hypervisor exists. That node contains a +compatible property with the value "linux,kvm". + +Once you determined you're running under a PV capable KVM, you can now use +hypercalls as described below. + +KVM hypercalls +============== + +Inside the device tree's /hypervisor node there's a property called +'hypercall-instructions'. This property contains at most 4 opcodes that make +up the hypercall. To call a hypercall, just call these instructions. + +The parameters are as follows: + + Register IN OUT + + r0 - volatile + r3 1st parameter Return code + r4 2nd parameter 1st output value + r5 3rd parameter 2nd output value + r6 4th parameter 3rd output value + r7 5th parameter 4th output value + r8 6th parameter 5th output value + r9 7th parameter 6th output value + r10 8th parameter 7th output value + r11 hypercall number 8th output value + r12 - volatile + +Hypercall definitions are shared in generic code, so the same hypercall numbers +apply for x86 and powerpc alike with the exception that each KVM hypercall +also needs to be ORed with the KVM vendor code which is (42 << 16). + +Return codes can be as follows: + + Code Meaning + + 0 Success + 12 Hypercall not implemented + <0 Error + +The magic page +============== + +To enable communication between the hypervisor and guest there is a new shared +page that contains parts of supervisor visible register state. The guest can +map this shared page using the KVM hypercall KVM_HC_PPC_MAP_MAGIC_PAGE. + +With this hypercall issued the guest always gets the magic page mapped at the +desired location in effective and physical address space. For now, we always +map the page to -4096. This way we can access it using absolute load and store +functions. The following instruction reads the first field of the magic page: + + ld rX, -4096(0) + +The interface is designed to be extensible should there be need later to add +additional registers to the magic page. If you add fields to the magic page, +also define a new hypercall feature to indicate that the host can give you more +registers. Only if the host supports the additional features, make use of them. + +The magic page has the following layout as described in +arch/powerpc/include/asm/kvm_para.h: + +struct kvm_vcpu_arch_shared { + __u64 scratch1; + __u64 scratch2; + __u64 scratch3; + __u64 critical; /* Guest may not get interrupts if == r1 */ + __u64 sprg0; + __u64 sprg1; + __u64 sprg2; + __u64 sprg3; + __u64 srr0; + __u64 srr1; + __u64 dar; + __u64 msr; + __u32 dsisr; + __u32 int_pending; /* Tells the guest if we have an interrupt */ +}; + +Additions to the page must only occur at the end. Struct fields are always 32 +or 64 bit aligned, depending on them being 32 or 64 bit wide respectively. + +Magic page features +=================== + +When mapping the magic page using the KVM hypercall KVM_HC_PPC_MAP_MAGIC_PAGE, +a second return value is passed to the guest. This second return value contains +a bitmap of available features inside the magic page. + +The following enhancements to the magic page are currently available: + + KVM_MAGIC_FEAT_SR Maps SR registers r/w in the magic page + +For enhanced features in the magic page, please check for the existence of the +feature before using them! + +MSR bits +======== + +The MSR contains bits that require hypervisor intervention and bits that do +not require direct hypervisor intervention because they only get interpreted +when entering the guest or don't have any impact on the hypervisor's behavior. + +The following bits are safe to be set inside the guest: + + MSR_EE + MSR_RI + MSR_CR + MSR_ME + +If any other bit changes in the MSR, please still use mtmsr(d). + +Patched instructions +==================== + +The "ld" and "std" instructions are transormed to "lwz" and "stw" instructions +respectively on 32 bit systems with an added offset of 4 to accommodate for big +endianness. + +The following is a list of mapping the Linux kernel performs when running as +guest. Implementing any of those mappings is optional, as the instruction traps +also act on the shared page. So calling privileged instructions still works as +before. + +From To +==== == + +mfmsr rX ld rX, magic_page->msr +mfsprg rX, 0 ld rX, magic_page->sprg0 +mfsprg rX, 1 ld rX, magic_page->sprg1 +mfsprg rX, 2 ld rX, magic_page->sprg2 +mfsprg rX, 3 ld rX, magic_page->sprg3 +mfsrr0 rX ld rX, magic_page->srr0 +mfsrr1 rX ld rX, magic_page->srr1 +mfdar rX ld rX, magic_page->dar +mfdsisr rX lwz rX, magic_page->dsisr + +mtmsr rX std rX, magic_page->msr +mtsprg 0, rX std rX, magic_page->sprg0 +mtsprg 1, rX std rX, magic_page->sprg1 +mtsprg 2, rX std rX, magic_page->sprg2 +mtsprg 3, rX std rX, magic_page->sprg3 +mtsrr0 rX std rX, magic_page->srr0 +mtsrr1 rX std rX, magic_page->srr1 +mtdar rX std rX, magic_page->dar +mtdsisr rX stw rX, magic_page->dsisr + +tlbsync nop + +mtmsrd rX, 0 b <special mtmsr section> +mtmsr rX b <special mtmsr section> + +mtmsrd rX, 1 b <special mtmsrd section> + +[Book3S only] +mtsrin rX, rY b <special mtsrin section> + +[BookE only] +wrteei [0|1] b <special wrteei section> + + +Some instructions require more logic to determine what's going on than a load +or store instruction can deliver. To enable patching of those, we keep some +RAM around where we can live translate instructions to. What happens is the +following: + + 1) copy emulation code to memory + 2) patch that code to fit the emulated instruction + 3) patch that code to return to the original pc + 4 + 4) patch the original instruction to branch to the new code + +That way we can inject an arbitrary amount of code as replacement for a single +instruction. This allows us to check for pending interrupts when setting EE=1 +for example. diff --git a/Documentation/virtual/kvm/review-checklist.txt b/Documentation/virtual/kvm/review-checklist.txt new file mode 100644 index 00000000000..a850986ed68 --- /dev/null +++ b/Documentation/virtual/kvm/review-checklist.txt @@ -0,0 +1,38 @@ +Review checklist for kvm patches +================================ + +1. The patch must follow Documentation/CodingStyle and + Documentation/SubmittingPatches. + +2. Patches should be against kvm.git master branch. + +3. If the patch introduces or modifies a new userspace API: + - the API must be documented in Documentation/virtual/kvm/api.txt + - the API must be discoverable using KVM_CHECK_EXTENSION + +4. New state must include support for save/restore. + +5. New features must default to off (userspace should explicitly request them). + Performance improvements can and should default to on. + +6. New cpu features should be exposed via KVM_GET_SUPPORTED_CPUID2 + +7. Emulator changes should be accompanied by unit tests for qemu-kvm.git + kvm/test directory. + +8. Changes should be vendor neutral when possible. Changes to common code + are better than duplicating changes to vendor code. + +9. Similarly, prefer changes to arch independent code than to arch dependent + code. + +10. User/kernel interfaces and guest/host interfaces must be 64-bit clean + (all variables and sizes naturally aligned on 64-bit; use specific types + only - u64 rather than ulong). + +11. New guest visible features must either be documented in a hardware manual + or be accompanied by documentation. + +12. Features must be robust against reset and kexec - for example, shared + host/guest memory must be unshared to prevent the host from writing to + guest memory that the guest has not reserved for this purpose. diff --git a/Documentation/virtual/kvm/timekeeping.txt b/Documentation/virtual/kvm/timekeeping.txt new file mode 100644 index 00000000000..df8946377cb --- /dev/null +++ b/Documentation/virtual/kvm/timekeeping.txt @@ -0,0 +1,612 @@ + + Timekeeping Virtualization for X86-Based Architectures + + Zachary Amsden <zamsden@redhat.com> + Copyright (c) 2010, Red Hat. All rights reserved. + +1) Overview +2) Timing Devices +3) TSC Hardware +4) Virtualization Problems + +========================================================================= + +1) Overview + +One of the most complicated parts of the X86 platform, and specifically, +the virtualization of this platform is the plethora of timing devices available +and the complexity of emulating those devices. In addition, virtualization of +time introduces a new set of challenges because it introduces a multiplexed +division of time beyond the control of the guest CPU. + +First, we will describe the various timekeeping hardware available, then +present some of the problems which arise and solutions available, giving +specific recommendations for certain classes of KVM guests. + +The purpose of this document is to collect data and information relevant to +timekeeping which may be difficult to find elsewhere, specifically, +information relevant to KVM and hardware-based virtualization. + +========================================================================= + +2) Timing Devices + +First we discuss the basic hardware devices available. TSC and the related +KVM clock are special enough to warrant a full exposition and are described in +the following section. + +2.1) i8254 - PIT + +One of the first timer devices available is the programmable interrupt timer, +or PIT. The PIT has a fixed frequency 1.193182 MHz base clock and three +channels which can be programmed to deliver periodic or one-shot interrupts. +These three channels can be configured in different modes and have individual +counters. Channel 1 and 2 were not available for general use in the original +IBM PC, and historically were connected to control RAM refresh and the PC +speaker. Now the PIT is typically integrated as part of an emulated chipset +and a separate physical PIT is not used. + +The PIT uses I/O ports 0x40 - 0x43. Access to the 16-bit counters is done +using single or multiple byte access to the I/O ports. There are 6 modes +available, but not all modes are available to all timers, as only timer 2 +has a connected gate input, required for modes 1 and 5. The gate line is +controlled by port 61h, bit 0, as illustrated in the following diagram. + + -------------- ---------------- +| | | | +| 1.1932 MHz |---------->| CLOCK OUT | ---------> IRQ 0 +| Clock | | | | + -------------- | +->| GATE TIMER 0 | + | ---------------- + | + | ---------------- + | | | + |------>| CLOCK OUT | ---------> 66.3 KHZ DRAM + | | | (aka /dev/null) + | +->| GATE TIMER 1 | + | ---------------- + | + | ---------------- + | | | + |------>| CLOCK OUT | ---------> Port 61h, bit 5 + | | | +Port 61h, bit 0 ---------->| GATE TIMER 2 | \_.---- ____ + ---------------- _| )--|LPF|---Speaker + / *---- \___/ +Port 61h, bit 1 -----------------------------------/ + +The timer modes are now described. + +Mode 0: Single Timeout. This is a one-shot software timeout that counts down + when the gate is high (always true for timers 0 and 1). When the count + reaches zero, the output goes high. + +Mode 1: Triggered One-shot. The output is initially set high. When the gate + line is set high, a countdown is initiated (which does not stop if the gate is + lowered), during which the output is set low. When the count reaches zero, + the output goes high. + +Mode 2: Rate Generator. The output is initially set high. When the countdown + reaches 1, the output goes low for one count and then returns high. The value + is reloaded and the countdown automatically resumes. If the gate line goes + low, the count is halted. If the output is low when the gate is lowered, the + output automatically goes high (this only affects timer 2). + +Mode 3: Square Wave. This generates a high / low square wave. The count + determines the length of the pulse, which alternates between high and low + when zero is reached. The count only proceeds when gate is high and is + automatically reloaded on reaching zero. The count is decremented twice at + each clock to generate a full high / low cycle at the full periodic rate. + If the count is even, the clock remains high for N/2 counts and low for N/2 + counts; if the clock is odd, the clock is high for (N+1)/2 counts and low + for (N-1)/2 counts. Only even values are latched by the counter, so odd + values are not observed when reading. This is the intended mode for timer 2, + which generates sine-like tones by low-pass filtering the square wave output. + +Mode 4: Software Strobe. After programming this mode and loading the counter, + the output remains high until the counter reaches zero. Then the output + goes low for 1 clock cycle and returns high. The counter is not reloaded. + Counting only occurs when gate is high. + +Mode 5: Hardware Strobe. After programming and loading the counter, the + output remains high. When the gate is raised, a countdown is initiated + (which does not stop if the gate is lowered). When the counter reaches zero, + the output goes low for 1 clock cycle and then returns high. The counter is + not reloaded. + +In addition to normal binary counting, the PIT supports BCD counting. The +command port, 0x43 is used to set the counter and mode for each of the three +timers. + +PIT commands, issued to port 0x43, using the following bit encoding: + +Bit 7-4: Command (See table below) +Bit 3-1: Mode (000 = Mode 0, 101 = Mode 5, 11X = undefined) +Bit 0 : Binary (0) / BCD (1) + +Command table: + +0000 - Latch Timer 0 count for port 0x40 + sample and hold the count to be read in port 0x40; + additional commands ignored until counter is read; + mode bits ignored. + +0001 - Set Timer 0 LSB mode for port 0x40 + set timer to read LSB only and force MSB to zero; + mode bits set timer mode + +0010 - Set Timer 0 MSB mode for port 0x40 + set timer to read MSB only and force LSB to zero; + mode bits set timer mode + +0011 - Set Timer 0 16-bit mode for port 0x40 + set timer to read / write LSB first, then MSB; + mode bits set timer mode + +0100 - Latch Timer 1 count for port 0x41 - as described above +0101 - Set Timer 1 LSB mode for port 0x41 - as described above +0110 - Set Timer 1 MSB mode for port 0x41 - as described above +0111 - Set Timer 1 16-bit mode for port 0x41 - as described above + +1000 - Latch Timer 2 count for port 0x42 - as described above +1001 - Set Timer 2 LSB mode for port 0x42 - as described above +1010 - Set Timer 2 MSB mode for port 0x42 - as described above +1011 - Set Timer 2 16-bit mode for port 0x42 as described above + +1101 - General counter latch + Latch combination of counters into corresponding ports + Bit 3 = Counter 2 + Bit 2 = Counter 1 + Bit 1 = Counter 0 + Bit 0 = Unused + +1110 - Latch timer status + Latch combination of counter mode into corresponding ports + Bit 3 = Counter 2 + Bit 2 = Counter 1 + Bit 1 = Counter 0 + + The output of ports 0x40-0x42 following this command will be: + + Bit 7 = Output pin + Bit 6 = Count loaded (0 if timer has expired) + Bit 5-4 = Read / Write mode + 01 = MSB only + 10 = LSB only + 11 = LSB / MSB (16-bit) + Bit 3-1 = Mode + Bit 0 = Binary (0) / BCD mode (1) + +2.2) RTC + +The second device which was available in the original PC was the MC146818 real +time clock. The original device is now obsolete, and usually emulated by the +system chipset, sometimes by an HPET and some frankenstein IRQ routing. + +The RTC is accessed through CMOS variables, which uses an index register to +control which bytes are read. Since there is only one index register, read +of the CMOS and read of the RTC require lock protection (in addition, it is +dangerous to allow userspace utilities such as hwclock to have direct RTC +access, as they could corrupt kernel reads and writes of CMOS memory). + +The RTC generates an interrupt which is usually routed to IRQ 8. The interrupt +can function as a periodic timer, an additional once a day alarm, and can issue +interrupts after an update of the CMOS registers by the MC146818 is complete. +The type of interrupt is signalled in the RTC status registers. + +The RTC will update the current time fields by battery power even while the +system is off. The current time fields should not be read while an update is +in progress, as indicated in the status register. + +The clock uses a 32.768kHz crystal, so bits 6-4 of register A should be +programmed to a 32kHz divider if the RTC is to count seconds. + +This is the RAM map originally used for the RTC/CMOS: + +Location Size Description +------------------------------------------ +00h byte Current second (BCD) +01h byte Seconds alarm (BCD) +02h byte Current minute (BCD) +03h byte Minutes alarm (BCD) +04h byte Current hour (BCD) +05h byte Hours alarm (BCD) +06h byte Current day of week (BCD) +07h byte Current day of month (BCD) +08h byte Current month (BCD) +09h byte Current year (BCD) +0Ah byte Register A + bit 7 = Update in progress + bit 6-4 = Divider for clock + 000 = 4.194 MHz + 001 = 1.049 MHz + 010 = 32 kHz + 10X = test modes + 110 = reset / disable + 111 = reset / disable + bit 3-0 = Rate selection for periodic interrupt + 000 = periodic timer disabled + 001 = 3.90625 uS + 010 = 7.8125 uS + 011 = .122070 mS + 100 = .244141 mS + ... + 1101 = 125 mS + 1110 = 250 mS + 1111 = 500 mS +0Bh byte Register B + bit 7 = Run (0) / Halt (1) + bit 6 = Periodic interrupt enable + bit 5 = Alarm interrupt enable + bit 4 = Update-ended interrupt enable + bit 3 = Square wave interrupt enable + bit 2 = BCD calendar (0) / Binary (1) + bit 1 = 12-hour mode (0) / 24-hour mode (1) + bit 0 = 0 (DST off) / 1 (DST enabled) +OCh byte Register C (read only) + bit 7 = interrupt request flag (IRQF) + bit 6 = periodic interrupt flag (PF) + bit 5 = alarm interrupt flag (AF) + bit 4 = update interrupt flag (UF) + bit 3-0 = reserved +ODh byte Register D (read only) + bit 7 = RTC has power + bit 6-0 = reserved +32h byte Current century BCD (*) + (*) location vendor specific and now determined from ACPI global tables + +2.3) APIC + +On Pentium and later processors, an on-board timer is available to each CPU +as part of the Advanced Programmable Interrupt Controller. The APIC is +accessed through memory-mapped registers and provides interrupt service to each +CPU, used for IPIs and local timer interrupts. + +Although in theory the APIC is a safe and stable source for local interrupts, +in practice, many bugs and glitches have occurred due to the special nature of +the APIC CPU-local memory-mapped hardware. Beware that CPU errata may affect +the use of the APIC and that workarounds may be required. In addition, some of +these workarounds pose unique constraints for virtualization - requiring either +extra overhead incurred from extra reads of memory-mapped I/O or additional +functionality that may be more computationally expensive to implement. + +Since the APIC is documented quite well in the Intel and AMD manuals, we will +avoid repetition of the detail here. It should be pointed out that the APIC +timer is programmed through the LVT (local vector timer) register, is capable +of one-shot or periodic operation, and is based on the bus clock divided down +by the programmable divider register. + +2.4) HPET + +HPET is quite complex, and was originally intended to replace the PIT / RTC +support of the X86 PC. It remains to be seen whether that will be the case, as +the de facto standard of PC hardware is to emulate these older devices. Some +systems designated as legacy free may support only the HPET as a hardware timer +device. + +The HPET spec is rather loose and vague, requiring at least 3 hardware timers, +but allowing implementation freedom to support many more. It also imposes no +fixed rate on the timer frequency, but does impose some extremal values on +frequency, error and slew. + +In general, the HPET is recommended as a high precision (compared to PIT /RTC) +time source which is independent of local variation (as there is only one HPET +in any given system). The HPET is also memory-mapped, and its presence is +indicated through ACPI tables by the BIOS. + +Detailed specification of the HPET is beyond the current scope of this +document, as it is also very well documented elsewhere. + +2.5) Offboard Timers + +Several cards, both proprietary (watchdog boards) and commonplace (e1000) have +timing chips built into the cards which may have registers which are accessible +to kernel or user drivers. To the author's knowledge, using these to generate +a clocksource for a Linux or other kernel has not yet been attempted and is in +general frowned upon as not playing by the agreed rules of the game. Such a +timer device would require additional support to be virtualized properly and is +not considered important at this time as no known operating system does this. + +========================================================================= + +3) TSC Hardware + +The TSC or time stamp counter is relatively simple in theory; it counts +instruction cycles issued by the processor, which can be used as a measure of +time. In practice, due to a number of problems, it is the most complicated +timekeeping device to use. + +The TSC is represented internally as a 64-bit MSR which can be read with the +RDMSR, RDTSC, or RDTSCP (when available) instructions. In the past, hardware +limitations made it possible to write the TSC, but generally on old hardware it +was only possible to write the low 32-bits of the 64-bit counter, and the upper +32-bits of the counter were cleared. Now, however, on Intel processors family +0Fh, for models 3, 4 and 6, and family 06h, models e and f, this restriction +has been lifted and all 64-bits are writable. On AMD systems, the ability to +write the TSC MSR is not an architectural guarantee. + +The TSC is accessible from CPL-0 and conditionally, for CPL > 0 software by +means of the CR4.TSD bit, which when enabled, disables CPL > 0 TSC access. + +Some vendors have implemented an additional instruction, RDTSCP, which returns +atomically not just the TSC, but an indicator which corresponds to the +processor number. This can be used to index into an array of TSC variables to +determine offset information in SMP systems where TSCs are not synchronized. +The presence of this instruction must be determined by consulting CPUID feature +bits. + +Both VMX and SVM provide extension fields in the virtualization hardware which +allows the guest visible TSC to be offset by a constant. Newer implementations +promise to allow the TSC to additionally be scaled, but this hardware is not +yet widely available. + +3.1) TSC synchronization + +The TSC is a CPU-local clock in most implementations. This means, on SMP +platforms, the TSCs of different CPUs may start at different times depending +on when the CPUs are powered on. Generally, CPUs on the same die will share +the same clock, however, this is not always the case. + +The BIOS may attempt to resynchronize the TSCs during the poweron process and +the operating system or other system software may attempt to do this as well. +Several hardware limitations make the problem worse - if it is not possible to +write the full 64-bits of the TSC, it may be impossible to match the TSC in +newly arriving CPUs to that of the rest of the system, resulting in +unsynchronized TSCs. This may be done by BIOS or system software, but in +practice, getting a perfectly synchronized TSC will not be possible unless all +values are read from the same clock, which generally only is possible on single +socket systems or those with special hardware support. + +3.2) TSC and CPU hotplug + +As touched on already, CPUs which arrive later than the boot time of the system +may not have a TSC value that is synchronized with the rest of the system. +Either system software, BIOS, or SMM code may actually try to establish the TSC +to a value matching the rest of the system, but a perfect match is usually not +a guarantee. This can have the effect of bringing a system from a state where +TSC is synchronized back to a state where TSC synchronization flaws, however +small, may be exposed to the OS and any virtualization environment. + +3.3) TSC and multi-socket / NUMA + +Multi-socket systems, especially large multi-socket systems are likely to have +individual clocksources rather than a single, universally distributed clock. +Since these clocks are driven by different crystals, they will not have +perfectly matched frequency, and temperature and electrical variations will +cause the CPU clocks, and thus the TSCs to drift over time. Depending on the +exact clock and bus design, the drift may or may not be fixed in absolute +error, and may accumulate over time. + +In addition, very large systems may deliberately slew the clocks of individual +cores. This technique, known as spread-spectrum clocking, reduces EMI at the +clock frequency and harmonics of it, which may be required to pass FCC +standards for telecommunications and computer equipment. + +It is recommended not to trust the TSCs to remain synchronized on NUMA or +multiple socket systems for these reasons. + +3.4) TSC and C-states + +C-states, or idling states of the processor, especially C1E and deeper sleep +states may be problematic for TSC as well. The TSC may stop advancing in such +a state, resulting in a TSC which is behind that of other CPUs when execution +is resumed. Such CPUs must be detected and flagged by the operating system +based on CPU and chipset identifications. + +The TSC in such a case may be corrected by catching it up to a known external +clocksource. + +3.5) TSC frequency change / P-states + +To make things slightly more interesting, some CPUs may change frequency. They +may or may not run the TSC at the same rate, and because the frequency change +may be staggered or slewed, at some points in time, the TSC rate may not be +known other than falling within a range of values. In this case, the TSC will +not be a stable time source, and must be calibrated against a known, stable, +external clock to be a usable source of time. + +Whether the TSC runs at a constant rate or scales with the P-state is model +dependent and must be determined by inspecting CPUID, chipset or vendor +specific MSR fields. + +In addition, some vendors have known bugs where the P-state is actually +compensated for properly during normal operation, but when the processor is +inactive, the P-state may be raised temporarily to service cache misses from +other processors. In such cases, the TSC on halted CPUs could advance faster +than that of non-halted processors. AMD Turion processors are known to have +this problem. + +3.6) TSC and STPCLK / T-states + +External signals given to the processor may also have the effect of stopping +the TSC. This is typically done for thermal emergency power control to prevent +an overheating condition, and typically, there is no way to detect that this +condition has happened. + +3.7) TSC virtualization - VMX + +VMX provides conditional trapping of RDTSC, RDMSR, WRMSR and RDTSCP +instructions, which is enough for full virtualization of TSC in any manner. In +addition, VMX allows passing through the host TSC plus an additional TSC_OFFSET +field specified in the VMCS. Special instructions must be used to read and +write the VMCS field. + +3.8) TSC virtualization - SVM + +SVM provides conditional trapping of RDTSC, RDMSR, WRMSR and RDTSCP +instructions, which is enough for full virtualization of TSC in any manner. In +addition, SVM allows passing through the host TSC plus an additional offset +field specified in the SVM control block. + +3.9) TSC feature bits in Linux + +In summary, there is no way to guarantee the TSC remains in perfect +synchronization unless it is explicitly guaranteed by the architecture. Even +if so, the TSCs in multi-sockets or NUMA systems may still run independently +despite being locally consistent. + +The following feature bits are used by Linux to signal various TSC attributes, +but they can only be taken to be meaningful for UP or single node systems. + +X86_FEATURE_TSC : The TSC is available in hardware +X86_FEATURE_RDTSCP : The RDTSCP instruction is available +X86_FEATURE_CONSTANT_TSC : The TSC rate is unchanged with P-states +X86_FEATURE_NONSTOP_TSC : The TSC does not stop in C-states +X86_FEATURE_TSC_RELIABLE : TSC sync checks are skipped (VMware) + +4) Virtualization Problems + +Timekeeping is especially problematic for virtualization because a number of +challenges arise. The most obvious problem is that time is now shared between +the host and, potentially, a number of virtual machines. Thus the virtual +operating system does not run with 100% usage of the CPU, despite the fact that +it may very well make that assumption. It may expect it to remain true to very +exacting bounds when interrupt sources are disabled, but in reality only its +virtual interrupt sources are disabled, and the machine may still be preempted +at any time. This causes problems as the passage of real time, the injection +of machine interrupts and the associated clock sources are no longer completely +synchronized with real time. + +This same problem can occur on native harware to a degree, as SMM mode may +steal cycles from the naturally on X86 systems when SMM mode is used by the +BIOS, but not in such an extreme fashion. However, the fact that SMM mode may +cause similar problems to virtualization makes it a good justification for +solving many of these problems on bare metal. + +4.1) Interrupt clocking + +One of the most immediate problems that occurs with legacy operating systems +is that the system timekeeping routines are often designed to keep track of +time by counting periodic interrupts. These interrupts may come from the PIT +or the RTC, but the problem is the same: the host virtualization engine may not +be able to deliver the proper number of interrupts per second, and so guest +time may fall behind. This is especially problematic if a high interrupt rate +is selected, such as 1000 HZ, which is unfortunately the default for many Linux +guests. + +There are three approaches to solving this problem; first, it may be possible +to simply ignore it. Guests which have a separate time source for tracking +'wall clock' or 'real time' may not need any adjustment of their interrupts to +maintain proper time. If this is not sufficient, it may be necessary to inject +additional interrupts into the guest in order to increase the effective +interrupt rate. This approach leads to complications in extreme conditions, +where host load or guest lag is too much to compensate for, and thus another +solution to the problem has risen: the guest may need to become aware of lost +ticks and compensate for them internally. Although promising in theory, the +implementation of this policy in Linux has been extremely error prone, and a +number of buggy variants of lost tick compensation are distributed across +commonly used Linux systems. + +Windows uses periodic RTC clocking as a means of keeping time internally, and +thus requires interrupt slewing to keep proper time. It does use a low enough +rate (ed: is it 18.2 Hz?) however that it has not yet been a problem in +practice. + +4.2) TSC sampling and serialization + +As the highest precision time source available, the cycle counter of the CPU +has aroused much interest from developers. As explained above, this timer has +many problems unique to its nature as a local, potentially unstable and +potentially unsynchronized source. One issue which is not unique to the TSC, +but is highlighted because of its very precise nature is sampling delay. By +definition, the counter, once read is already old. However, it is also +possible for the counter to be read ahead of the actual use of the result. +This is a consequence of the superscalar execution of the instruction stream, +which may execute instructions out of order. Such execution is called +non-serialized. Forcing serialized execution is necessary for precise +measurement with the TSC, and requires a serializing instruction, such as CPUID +or an MSR read. + +Since CPUID may actually be virtualized by a trap and emulate mechanism, this +serialization can pose a performance issue for hardware virtualization. An +accurate time stamp counter reading may therefore not always be available, and +it may be necessary for an implementation to guard against "backwards" reads of +the TSC as seen from other CPUs, even in an otherwise perfectly synchronized +system. + +4.3) Timespec aliasing + +Additionally, this lack of serialization from the TSC poses another challenge +when using results of the TSC when measured against another time source. As +the TSC is much higher precision, many possible values of the TSC may be read +while another clock is still expressing the same value. + +That is, you may read (T,T+10) while external clock C maintains the same value. +Due to non-serialized reads, you may actually end up with a range which +fluctuates - from (T-1.. T+10). Thus, any time calculated from a TSC, but +calibrated against an external value may have a range of valid values. +Re-calibrating this computation may actually cause time, as computed after the +calibration, to go backwards, compared with time computed before the +calibration. + +This problem is particularly pronounced with an internal time source in Linux, +the kernel time, which is expressed in the theoretically high resolution +timespec - but which advances in much larger granularity intervals, sometimes +at the rate of jiffies, and possibly in catchup modes, at a much larger step. + +This aliasing requires care in the computation and recalibration of kvmclock +and any other values derived from TSC computation (such as TSC virtualization +itself). + +4.4) Migration + +Migration of a virtual machine raises problems for timekeeping in two ways. +First, the migration itself may take time, during which interrupts cannot be +delivered, and after which, the guest time may need to be caught up. NTP may +be able to help to some degree here, as the clock correction required is +typically small enough to fall in the NTP-correctable window. + +An additional concern is that timers based off the TSC (or HPET, if the raw bus +clock is exposed) may now be running at different rates, requiring compensation +in some way in the hypervisor by virtualizing these timers. In addition, +migrating to a faster machine may preclude the use of a passthrough TSC, as a +faster clock cannot be made visible to a guest without the potential of time +advancing faster than usual. A slower clock is less of a problem, as it can +always be caught up to the original rate. KVM clock avoids these problems by +simply storing multipliers and offsets against the TSC for the guest to convert +back into nanosecond resolution values. + +4.5) Scheduling + +Since scheduling may be based on precise timing and firing of interrupts, the +scheduling algorithms of an operating system may be adversely affected by +virtualization. In theory, the effect is random and should be universally +distributed, but in contrived as well as real scenarios (guest device access, +causes of virtualization exits, possible context switch), this may not always +be the case. The effect of this has not been well studied. + +In an attempt to work around this, several implementations have provided a +paravirtualized scheduler clock, which reveals the true amount of CPU time for +which a virtual machine has been running. + +4.6) Watchdogs + +Watchdog timers, such as the lock detector in Linux may fire accidentally when +running under hardware virtualization due to timer interrupts being delayed or +misinterpretation of the passage of real time. Usually, these warnings are +spurious and can be ignored, but in some circumstances it may be necessary to +disable such detection. + +4.7) Delays and precision timing + +Precise timing and delays may not be possible in a virtualized system. This +can happen if the system is controlling physical hardware, or issues delays to +compensate for slower I/O to and from devices. The first issue is not solvable +in general for a virtualized system; hardware control software can't be +adequately virtualized without a full real-time operating system, which would +require an RT aware virtualization platform. + +The second issue may cause performance problems, but this is unlikely to be a +significant issue. In many cases these delays may be eliminated through +configuration or paravirtualization. + +4.8) Covert channels and leaks + +In addition to the above problems, time information will inevitably leak to the +guest about the host in anything but a perfect implementation of virtualized +time. This may allow the guest to infer the presence of a hypervisor (as in a +red-pill type detection), and it may allow information to leak between guests +by using CPU utilization itself as a signalling channel. Preventing such +problems would require completely isolated virtual time which may not track +real time any longer. This may be useful in certain security or QA contexts, +but in general isn't recommended for real-world deployment scenarios. diff --git a/Documentation/virtual/lguest/.gitignore b/Documentation/virtual/lguest/.gitignore new file mode 100644 index 00000000000..115587fd5f6 --- /dev/null +++ b/Documentation/virtual/lguest/.gitignore @@ -0,0 +1 @@ +lguest diff --git a/Documentation/virtual/lguest/Makefile b/Documentation/virtual/lguest/Makefile new file mode 100644 index 00000000000..bebac6b4f33 --- /dev/null +++ b/Documentation/virtual/lguest/Makefile @@ -0,0 +1,8 @@ +# This creates the demonstration utility "lguest" which runs a Linux guest. +# Missing headers? Add "-I../../include -I../../arch/x86/include" +CFLAGS:=-m32 -Wall -Wmissing-declarations -Wmissing-prototypes -O3 -U_FORTIFY_SOURCE + +all: lguest + +clean: + rm -f lguest diff --git a/Documentation/virtual/lguest/extract b/Documentation/virtual/lguest/extract new file mode 100644 index 00000000000..7730bb6e4b9 --- /dev/null +++ b/Documentation/virtual/lguest/extract @@ -0,0 +1,58 @@ +#! /bin/sh + +set -e + +PREFIX=$1 +shift + +trap 'rm -r $TMPDIR' 0 +TMPDIR=`mktemp -d` + +exec 3>/dev/null +for f; do + while IFS=" +" read -r LINE; do + case "$LINE" in + *$PREFIX:[0-9]*:\**) + NUM=`echo "$LINE" | sed "s/.*$PREFIX:\([0-9]*\).*/\1/"` + if [ -f $TMPDIR/$NUM ]; then + echo "$TMPDIR/$NUM already exits prior to $f" + exit 1 + fi + exec 3>>$TMPDIR/$NUM + echo $f | sed 's,\.\./,,g' > $TMPDIR/.$NUM + /bin/echo "$LINE" | sed -e "s/$PREFIX:[0-9]*//" -e "s/:\*/*/" >&3 + ;; + *$PREFIX:[0-9]*) + NUM=`echo "$LINE" | sed "s/.*$PREFIX:\([0-9]*\).*/\1/"` + if [ -f $TMPDIR/$NUM ]; then + echo "$TMPDIR/$NUM already exits prior to $f" + exit 1 + fi + exec 3>>$TMPDIR/$NUM + echo $f | sed 's,\.\./,,g' > $TMPDIR/.$NUM + /bin/echo "$LINE" | sed "s/$PREFIX:[0-9]*//" >&3 + ;; + *:\**) + /bin/echo "$LINE" | sed -e "s/:\*/*/" -e "s,/\*\*/,," >&3 + echo >&3 + exec 3>/dev/null + ;; + *) + /bin/echo "$LINE" >&3 + ;; + esac + done < $f + echo >&3 + exec 3>/dev/null +done + +LASTFILE="" +for f in $TMPDIR/*; do + if [ "$LASTFILE" != $(cat $TMPDIR/.$(basename $f) ) ]; then + LASTFILE=$(cat $TMPDIR/.$(basename $f) ) + echo "[ $LASTFILE ]" + fi + cat $f +done + diff --git a/Documentation/virtual/lguest/lguest.c b/Documentation/virtual/lguest/lguest.c new file mode 100644 index 00000000000..d9da7e14853 --- /dev/null +++ b/Documentation/virtual/lguest/lguest.c @@ -0,0 +1,2095 @@ +/*P:100 + * This is the Launcher code, a simple program which lays out the "physical" + * memory for the new Guest by mapping the kernel image and the virtual + * devices, then opens /dev/lguest to tell the kernel about the Guest and + * control it. +:*/ +#define _LARGEFILE64_SOURCE +#define _GNU_SOURCE +#include <stdio.h> +#include <string.h> +#include <unistd.h> +#include <err.h> +#include <stdint.h> +#include <stdlib.h> +#include <elf.h> +#include <sys/mman.h> +#include <sys/param.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <sys/wait.h> +#include <sys/eventfd.h> +#include <fcntl.h> +#include <stdbool.h> +#include <errno.h> +#include <ctype.h> +#include <sys/socket.h> +#include <sys/ioctl.h> +#include <sys/time.h> +#include <time.h> +#include <netinet/in.h> +#include <net/if.h> +#include <linux/sockios.h> +#include <linux/if_tun.h> +#include <sys/uio.h> +#include <termios.h> +#include <getopt.h> +#include <assert.h> +#include <sched.h> +#include <limits.h> +#include <stddef.h> +#include <signal.h> +#include <pwd.h> +#include <grp.h> + +#include <linux/virtio_config.h> +#include <linux/virtio_net.h> +#include <linux/virtio_blk.h> +#include <linux/virtio_console.h> +#include <linux/virtio_rng.h> +#include <linux/virtio_ring.h> +#include <asm/bootparam.h> +#include "../../include/linux/lguest_launcher.h" +/*L:110 + * We can ignore the 42 include files we need for this program, but I do want + * to draw attention to the use of kernel-style types. + * + * As Linus said, "C is a Spartan language, and so should your naming be." I + * like these abbreviations, so we define them here. Note that u64 is always + * unsigned long long, which works on all Linux systems: this means that we can + * use %llu in printf for any u64. + */ +typedef unsigned long long u64; +typedef uint32_t u32; +typedef uint16_t u16; +typedef uint8_t u8; +/*:*/ + +#define PAGE_PRESENT 0x7 /* Present, RW, Execute */ +#define BRIDGE_PFX "bridge:" +#ifndef SIOCBRADDIF +#define SIOCBRADDIF 0x89a2 /* add interface to bridge */ +#endif +/* We can have up to 256 pages for devices. */ +#define DEVICE_PAGES 256 +/* This will occupy 3 pages: it must be a power of 2. */ +#define VIRTQUEUE_NUM 256 + +/*L:120 + * verbose is both a global flag and a macro. The C preprocessor allows + * this, and although I wouldn't recommend it, it works quite nicely here. + */ +static bool verbose; +#define verbose(args...) \ + do { if (verbose) printf(args); } while(0) +/*:*/ + +/* The pointer to the start of guest memory. */ +static void *guest_base; +/* The maximum guest physical address allowed, and maximum possible. */ +static unsigned long guest_limit, guest_max; +/* The /dev/lguest file descriptor. */ +static int lguest_fd; + +/* a per-cpu variable indicating whose vcpu is currently running */ +static unsigned int __thread cpu_id; + +/* This is our list of devices. */ +struct device_list { + /* Counter to assign interrupt numbers. */ + unsigned int next_irq; + + /* Counter to print out convenient device numbers. */ + unsigned int device_num; + + /* The descriptor page for the devices. */ + u8 *descpage; + + /* A single linked list of devices. */ + struct device *dev; + /* And a pointer to the last device for easy append. */ + struct device *lastdev; +}; + +/* The list of Guest devices, based on command line arguments. */ +static struct device_list devices; + +/* The device structure describes a single device. */ +struct device { + /* The linked-list pointer. */ + struct device *next; + + /* The device's descriptor, as mapped into the Guest. */ + struct lguest_device_desc *desc; + + /* We can't trust desc values once Guest has booted: we use these. */ + unsigned int feature_len; + unsigned int num_vq; + + /* The name of this device, for --verbose. */ + const char *name; + + /* Any queues attached to this device */ + struct virtqueue *vq; + + /* Is it operational */ + bool running; + + /* Does Guest want an intrrupt on empty? */ + bool irq_on_empty; + + /* Device-specific data. */ + void *priv; +}; + +/* The virtqueue structure describes a queue attached to a device. */ +struct virtqueue { + struct virtqueue *next; + + /* Which device owns me. */ + struct device *dev; + + /* The configuration for this queue. */ + struct lguest_vqconfig config; + + /* The actual ring of buffers. */ + struct vring vring; + + /* Last available index we saw. */ + u16 last_avail_idx; + + /* How many are used since we sent last irq? */ + unsigned int pending_used; + + /* Eventfd where Guest notifications arrive. */ + int eventfd; + + /* Function for the thread which is servicing this virtqueue. */ + void (*service)(struct virtqueue *vq); + pid_t thread; +}; + +/* Remember the arguments to the program so we can "reboot" */ +static char **main_args; + +/* The original tty settings to restore on exit. */ +static struct termios orig_term; + +/* + * We have to be careful with barriers: our devices are all run in separate + * threads and so we need to make sure that changes visible to the Guest happen + * in precise order. + */ +#define wmb() __asm__ __volatile__("" : : : "memory") +#define mb() __asm__ __volatile__("" : : : "memory") + +/* + * Convert an iovec element to the given type. + * + * This is a fairly ugly trick: we need to know the size of the type and + * alignment requirement to check the pointer is kosher. It's also nice to + * have the name of the type in case we report failure. + * + * Typing those three things all the time is cumbersome and error prone, so we + * have a macro which sets them all up and passes to the real function. + */ +#define convert(iov, type) \ + ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) + +static void *_convert(struct iovec *iov, size_t size, size_t align, + const char *name) +{ + if (iov->iov_len != size) + errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); + if ((unsigned long)iov->iov_base % align != 0) + errx(1, "Bad alignment %p for %s", iov->iov_base, name); + return iov->iov_base; +} + +/* Wrapper for the last available index. Makes it easier to change. */ +#define lg_last_avail(vq) ((vq)->last_avail_idx) + +/* + * The virtio configuration space is defined to be little-endian. x86 is + * little-endian too, but it's nice to be explicit so we have these helpers. + */ +#define cpu_to_le16(v16) (v16) +#define cpu_to_le32(v32) (v32) +#define cpu_to_le64(v64) (v64) +#define le16_to_cpu(v16) (v16) +#define le32_to_cpu(v32) (v32) +#define le64_to_cpu(v64) (v64) + +/* Is this iovec empty? */ +static bool iov_empty(const struct iovec iov[], unsigned int num_iov) +{ + unsigned int i; + + for (i = 0; i < num_iov; i++) + if (iov[i].iov_len) + return false; + return true; +} + +/* Take len bytes from the front of this iovec. */ +static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) +{ + unsigned int i; + + for (i = 0; i < num_iov; i++) { + unsigned int used; + + used = iov[i].iov_len < len ? iov[i].iov_len : len; + iov[i].iov_base += used; + iov[i].iov_len -= used; + len -= used; + } + assert(len == 0); +} + +/* The device virtqueue descriptors are followed by feature bitmasks. */ +static u8 *get_feature_bits(struct device *dev) +{ + return (u8 *)(dev->desc + 1) + + dev->num_vq * sizeof(struct lguest_vqconfig); +} + +/*L:100 + * The Launcher code itself takes us out into userspace, that scary place where + * pointers run wild and free! Unfortunately, like most userspace programs, + * it's quite boring (which is why everyone likes to hack on the kernel!). + * Perhaps if you make up an Lguest Drinking Game at this point, it will get + * you through this section. Or, maybe not. + * + * The Launcher sets up a big chunk of memory to be the Guest's "physical" + * memory and stores it in "guest_base". In other words, Guest physical == + * Launcher virtual with an offset. + * + * This can be tough to get your head around, but usually it just means that we + * use these trivial conversion functions when the Guest gives us its + * "physical" addresses: + */ +static void *from_guest_phys(unsigned long addr) +{ + return guest_base + addr; +} + +static unsigned long to_guest_phys(const void *addr) +{ + return (addr - guest_base); +} + +/*L:130 + * Loading the Kernel. + * + * We start with couple of simple helper routines. open_or_die() avoids + * error-checking code cluttering the callers: + */ +static int open_or_die(const char *name, int flags) +{ + int fd = open(name, flags); + if (fd < 0) + err(1, "Failed to open %s", name); + return fd; +} + +/* map_zeroed_pages() takes a number of pages. */ +static void *map_zeroed_pages(unsigned int num) +{ + int fd = open_or_die("/dev/zero", O_RDONLY); + void *addr; + + /* + * We use a private mapping (ie. if we write to the page, it will be + * copied). We allocate an extra two pages PROT_NONE to act as guard + * pages against read/write attempts that exceed allocated space. + */ + addr = mmap(NULL, getpagesize() * (num+2), + PROT_NONE, MAP_PRIVATE, fd, 0); + + if (addr == MAP_FAILED) + err(1, "Mmapping %u pages of /dev/zero", num); + + if (mprotect(addr + getpagesize(), getpagesize() * num, + PROT_READ|PROT_WRITE) == -1) + err(1, "mprotect rw %u pages failed", num); + + /* + * One neat mmap feature is that you can close the fd, and it + * stays mapped. + */ + close(fd); + + /* Return address after PROT_NONE page */ + return addr + getpagesize(); +} + +/* Get some more pages for a device. */ +static void *get_pages(unsigned int num) +{ + void *addr = from_guest_phys(guest_limit); + + guest_limit += num * getpagesize(); + if (guest_limit > guest_max) + errx(1, "Not enough memory for devices"); + return addr; +} + +/* + * This routine is used to load the kernel or initrd. It tries mmap, but if + * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), + * it falls back to reading the memory in. + */ +static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) +{ + ssize_t r; + + /* + * We map writable even though for some segments are marked read-only. + * The kernel really wants to be writable: it patches its own + * instructions. + * + * MAP_PRIVATE means that the page won't be copied until a write is + * done to it. This allows us to share untouched memory between + * Guests. + */ + if (mmap(addr, len, PROT_READ|PROT_WRITE, + MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) + return; + + /* pread does a seek and a read in one shot: saves a few lines. */ + r = pread(fd, addr, len, offset); + if (r != len) + err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); +} + +/* + * This routine takes an open vmlinux image, which is in ELF, and maps it into + * the Guest memory. ELF = Embedded Linking Format, which is the format used + * by all modern binaries on Linux including the kernel. + * + * The ELF headers give *two* addresses: a physical address, and a virtual + * address. We use the physical address; the Guest will map itself to the + * virtual address. + * + * We return the starting address. + */ +static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) +{ + Elf32_Phdr phdr[ehdr->e_phnum]; + unsigned int i; + + /* + * Sanity checks on the main ELF header: an x86 executable with a + * reasonable number of correctly-sized program headers. + */ + if (ehdr->e_type != ET_EXEC + || ehdr->e_machine != EM_386 + || ehdr->e_phentsize != sizeof(Elf32_Phdr) + || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) + errx(1, "Malformed elf header"); + + /* + * An ELF executable contains an ELF header and a number of "program" + * headers which indicate which parts ("segments") of the program to + * load where. + */ + + /* We read in all the program headers at once: */ + if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) + err(1, "Seeking to program headers"); + if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) + err(1, "Reading program headers"); + + /* + * Try all the headers: there are usually only three. A read-only one, + * a read-write one, and a "note" section which we don't load. + */ + for (i = 0; i < ehdr->e_phnum; i++) { + /* If this isn't a loadable segment, we ignore it */ + if (phdr[i].p_type != PT_LOAD) + continue; + + verbose("Section %i: size %i addr %p\n", + i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); + + /* We map this section of the file at its physical address. */ + map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), + phdr[i].p_offset, phdr[i].p_filesz); + } + + /* The entry point is given in the ELF header. */ + return ehdr->e_entry; +} + +/*L:150 + * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed + * to jump into it and it will unpack itself. We used to have to perform some + * hairy magic because the unpacking code scared me. + * + * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote + * a small patch to jump over the tricky bits in the Guest, so now we just read + * the funky header so we know where in the file to load, and away we go! + */ +static unsigned long load_bzimage(int fd) +{ + struct boot_params boot; + int r; + /* Modern bzImages get loaded at 1M. */ + void *p = from_guest_phys(0x100000); + + /* + * Go back to the start of the file and read the header. It should be + * a Linux boot header (see Documentation/x86/i386/boot.txt) + */ + lseek(fd, 0, SEEK_SET); + read(fd, &boot, sizeof(boot)); + + /* Inside the setup_hdr, we expect the magic "HdrS" */ + if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) + errx(1, "This doesn't look like a bzImage to me"); + + /* Skip over the extra sectors of the header. */ + lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); + + /* Now read everything into memory. in nice big chunks. */ + while ((r = read(fd, p, 65536)) > 0) + p += r; + + /* Finally, code32_start tells us where to enter the kernel. */ + return boot.hdr.code32_start; +} + +/*L:140 + * Loading the kernel is easy when it's a "vmlinux", but most kernels + * come wrapped up in the self-decompressing "bzImage" format. With a little + * work, we can load those, too. + */ +static unsigned long load_kernel(int fd) +{ + Elf32_Ehdr hdr; + + /* Read in the first few bytes. */ + if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) + err(1, "Reading kernel"); + + /* If it's an ELF file, it starts with "\177ELF" */ + if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) + return map_elf(fd, &hdr); + + /* Otherwise we assume it's a bzImage, and try to load it. */ + return load_bzimage(fd); +} + +/* + * This is a trivial little helper to align pages. Andi Kleen hated it because + * it calls getpagesize() twice: "it's dumb code." + * + * Kernel guys get really het up about optimization, even when it's not + * necessary. I leave this code as a reaction against that. + */ +static inline unsigned long page_align(unsigned long addr) +{ + /* Add upwards and truncate downwards. */ + return ((addr + getpagesize()-1) & ~(getpagesize()-1)); +} + +/*L:180 + * An "initial ram disk" is a disk image loaded into memory along with the + * kernel which the kernel can use to boot from without needing any drivers. + * Most distributions now use this as standard: the initrd contains the code to + * load the appropriate driver modules for the current machine. + * + * Importantly, James Morris works for RedHat, and Fedora uses initrds for its + * kernels. He sent me this (and tells me when I break it). + */ +static unsigned long load_initrd(const char *name, unsigned long mem) +{ + int ifd; + struct stat st; + unsigned long len; + + ifd = open_or_die(name, O_RDONLY); + /* fstat() is needed to get the file size. */ + if (fstat(ifd, &st) < 0) + err(1, "fstat() on initrd '%s'", name); + + /* + * We map the initrd at the top of memory, but mmap wants it to be + * page-aligned, so we round the size up for that. + */ + len = page_align(st.st_size); + map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); + /* + * Once a file is mapped, you can close the file descriptor. It's a + * little odd, but quite useful. + */ + close(ifd); + verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); + + /* We return the initrd size. */ + return len; +} +/*:*/ + +/* + * Simple routine to roll all the commandline arguments together with spaces + * between them. + */ +static void concat(char *dst, char *args[]) +{ + unsigned int i, len = 0; + + for (i = 0; args[i]; i++) { + if (i) { + strcat(dst+len, " "); + len++; + } + strcpy(dst+len, args[i]); + len += strlen(args[i]); + } + /* In case it's empty. */ + dst[len] = '\0'; +} + +/*L:185 + * This is where we actually tell the kernel to initialize the Guest. We + * saw the arguments it expects when we looked at initialize() in lguest_user.c: + * the base of Guest "physical" memory, the top physical page to allow and the + * entry point for the Guest. + */ +static void tell_kernel(unsigned long start) +{ + unsigned long args[] = { LHREQ_INITIALIZE, + (unsigned long)guest_base, + guest_limit / getpagesize(), start }; + verbose("Guest: %p - %p (%#lx)\n", + guest_base, guest_base + guest_limit, guest_limit); + lguest_fd = open_or_die("/dev/lguest", O_RDWR); + if (write(lguest_fd, args, sizeof(args)) < 0) + err(1, "Writing to /dev/lguest"); +} +/*:*/ + +/*L:200 + * Device Handling. + * + * When the Guest gives us a buffer, it sends an array of addresses and sizes. + * We need to make sure it's not trying to reach into the Launcher itself, so + * we have a convenient routine which checks it and exits with an error message + * if something funny is going on: + */ +static void *_check_pointer(unsigned long addr, unsigned int size, + unsigned int line) +{ + /* + * Check if the requested address and size exceeds the allocated memory, + * or addr + size wraps around. + */ + if ((addr + size) > guest_limit || (addr + size) < addr) + errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); + /* + * We return a pointer for the caller's convenience, now we know it's + * safe to use. + */ + return from_guest_phys(addr); +} +/* A macro which transparently hands the line number to the real function. */ +#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) + +/* + * Each buffer in the virtqueues is actually a chain of descriptors. This + * function returns the next descriptor in the chain, or vq->vring.num if we're + * at the end. + */ +static unsigned next_desc(struct vring_desc *desc, + unsigned int i, unsigned int max) +{ + unsigned int next; + + /* If this descriptor says it doesn't chain, we're done. */ + if (!(desc[i].flags & VRING_DESC_F_NEXT)) + return max; + + /* Check they're not leading us off end of descriptors. */ + next = desc[i].next; + /* Make sure compiler knows to grab that: we don't want it changing! */ + wmb(); + + if (next >= max) + errx(1, "Desc next is %u", next); + + return next; +} + +/* + * This actually sends the interrupt for this virtqueue, if we've used a + * buffer. + */ +static void trigger_irq(struct virtqueue *vq) +{ + unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; + + /* Don't inform them if nothing used. */ + if (!vq->pending_used) + return; + vq->pending_used = 0; + + /* If they don't want an interrupt, don't send one... */ + if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) { + /* ... unless they've asked us to force one on empty. */ + if (!vq->dev->irq_on_empty + || lg_last_avail(vq) != vq->vring.avail->idx) + return; + } + + /* Send the Guest an interrupt tell them we used something up. */ + if (write(lguest_fd, buf, sizeof(buf)) != 0) + err(1, "Triggering irq %i", vq->config.irq); +} + +/* + * This looks in the virtqueue for the first available buffer, and converts + * it to an iovec for convenient access. Since descriptors consist of some + * number of output then some number of input descriptors, it's actually two + * iovecs, but we pack them into one and note how many of each there were. + * + * This function waits if necessary, and returns the descriptor number found. + */ +static unsigned wait_for_vq_desc(struct virtqueue *vq, + struct iovec iov[], + unsigned int *out_num, unsigned int *in_num) +{ + unsigned int i, head, max; + struct vring_desc *desc; + u16 last_avail = lg_last_avail(vq); + + /* There's nothing available? */ + while (last_avail == vq->vring.avail->idx) { + u64 event; + + /* + * Since we're about to sleep, now is a good time to tell the + * Guest about what we've used up to now. + */ + trigger_irq(vq); + + /* OK, now we need to know about added descriptors. */ + vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; + + /* + * They could have slipped one in as we were doing that: make + * sure it's written, then check again. + */ + mb(); + if (last_avail != vq->vring.avail->idx) { + vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; + break; + } + + /* Nothing new? Wait for eventfd to tell us they refilled. */ + if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) + errx(1, "Event read failed?"); + + /* We don't need to be notified again. */ + vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; + } + + /* Check it isn't doing very strange things with descriptor numbers. */ + if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) + errx(1, "Guest moved used index from %u to %u", + last_avail, vq->vring.avail->idx); + + /* + * Grab the next descriptor number they're advertising, and increment + * the index we've seen. + */ + head = vq->vring.avail->ring[last_avail % vq->vring.num]; + lg_last_avail(vq)++; + + /* If their number is silly, that's a fatal mistake. */ + if (head >= vq->vring.num) + errx(1, "Guest says index %u is available", head); + + /* When we start there are none of either input nor output. */ + *out_num = *in_num = 0; + + max = vq->vring.num; + desc = vq->vring.desc; + i = head; + + /* + * If this is an indirect entry, then this buffer contains a descriptor + * table which we handle as if it's any normal descriptor chain. + */ + if (desc[i].flags & VRING_DESC_F_INDIRECT) { + if (desc[i].len % sizeof(struct vring_desc)) + errx(1, "Invalid size for indirect buffer table"); + + max = desc[i].len / sizeof(struct vring_desc); + desc = check_pointer(desc[i].addr, desc[i].len); + i = 0; + } + + do { + /* Grab the first descriptor, and check it's OK. */ + iov[*out_num + *in_num].iov_len = desc[i].len; + iov[*out_num + *in_num].iov_base + = check_pointer(desc[i].addr, desc[i].len); + /* If this is an input descriptor, increment that count. */ + if (desc[i].flags & VRING_DESC_F_WRITE) + (*in_num)++; + else { + /* + * If it's an output descriptor, they're all supposed + * to come before any input descriptors. + */ + if (*in_num) + errx(1, "Descriptor has out after in"); + (*out_num)++; + } + + /* If we've got too many, that implies a descriptor loop. */ + if (*out_num + *in_num > max) + errx(1, "Looped descriptor"); + } while ((i = next_desc(desc, i, max)) != max); + + return head; +} + +/* + * After we've used one of their buffers, we tell the Guest about it. Sometime + * later we'll want to send them an interrupt using trigger_irq(); note that + * wait_for_vq_desc() does that for us if it has to wait. + */ +static void add_used(struct virtqueue *vq, unsigned int head, int len) +{ + struct vring_used_elem *used; + + /* + * The virtqueue contains a ring of used buffers. Get a pointer to the + * next entry in that used ring. + */ + used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; + used->id = head; + used->len = len; + /* Make sure buffer is written before we update index. */ + wmb(); + vq->vring.used->idx++; + vq->pending_used++; +} + +/* And here's the combo meal deal. Supersize me! */ +static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) +{ + add_used(vq, head, len); + trigger_irq(vq); +} + +/* + * The Console + * + * We associate some data with the console for our exit hack. + */ +struct console_abort { + /* How many times have they hit ^C? */ + int count; + /* When did they start? */ + struct timeval start; +}; + +/* This is the routine which handles console input (ie. stdin). */ +static void console_input(struct virtqueue *vq) +{ + int len; + unsigned int head, in_num, out_num; + struct console_abort *abort = vq->dev->priv; + struct iovec iov[vq->vring.num]; + + /* Make sure there's a descriptor available. */ + head = wait_for_vq_desc(vq, iov, &out_num, &in_num); + if (out_num) + errx(1, "Output buffers in console in queue?"); + + /* Read into it. This is where we usually wait. */ + len = readv(STDIN_FILENO, iov, in_num); + if (len <= 0) { + /* Ran out of input? */ + warnx("Failed to get console input, ignoring console."); + /* + * For simplicity, dying threads kill the whole Launcher. So + * just nap here. + */ + for (;;) + pause(); + } + + /* Tell the Guest we used a buffer. */ + add_used_and_trigger(vq, head, len); + + /* + * Three ^C within one second? Exit. + * + * This is such a hack, but works surprisingly well. Each ^C has to + * be in a buffer by itself, so they can't be too fast. But we check + * that we get three within about a second, so they can't be too + * slow. + */ + if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { + abort->count = 0; + return; + } + + abort->count++; + if (abort->count == 1) + gettimeofday(&abort->start, NULL); + else if (abort->count == 3) { + struct timeval now; + gettimeofday(&now, NULL); + /* Kill all Launcher processes with SIGINT, like normal ^C */ + if (now.tv_sec <= abort->start.tv_sec+1) + kill(0, SIGINT); + abort->count = 0; + } +} + +/* This is the routine which handles console output (ie. stdout). */ +static void console_output(struct virtqueue *vq) +{ + unsigned int head, out, in; + struct iovec iov[vq->vring.num]; + + /* We usually wait in here, for the Guest to give us something. */ + head = wait_for_vq_desc(vq, iov, &out, &in); + if (in) + errx(1, "Input buffers in console output queue?"); + + /* writev can return a partial write, so we loop here. */ + while (!iov_empty(iov, out)) { + int len = writev(STDOUT_FILENO, iov, out); + if (len <= 0) + err(1, "Write to stdout gave %i", len); + iov_consume(iov, out, len); + } + + /* + * We're finished with that buffer: if we're going to sleep, + * wait_for_vq_desc() will prod the Guest with an interrupt. + */ + add_used(vq, head, 0); +} + +/* + * The Network + * + * Handling output for network is also simple: we get all the output buffers + * and write them to /dev/net/tun. + */ +struct net_info { + int tunfd; +}; + +static void net_output(struct virtqueue *vq) +{ + struct net_info *net_info = vq->dev->priv; + unsigned int head, out, in; + struct iovec iov[vq->vring.num]; + + /* We usually wait in here for the Guest to give us a packet. */ + head = wait_for_vq_desc(vq, iov, &out, &in); + if (in) + errx(1, "Input buffers in net output queue?"); + /* + * Send the whole thing through to /dev/net/tun. It expects the exact + * same format: what a coincidence! + */ + if (writev(net_info->tunfd, iov, out) < 0) + errx(1, "Write to tun failed?"); + + /* + * Done with that one; wait_for_vq_desc() will send the interrupt if + * all packets are processed. + */ + add_used(vq, head, 0); +} + +/* + * Handling network input is a bit trickier, because I've tried to optimize it. + * + * First we have a helper routine which tells is if from this file descriptor + * (ie. the /dev/net/tun device) will block: + */ +static bool will_block(int fd) +{ + fd_set fdset; + struct timeval zero = { 0, 0 }; + FD_ZERO(&fdset); + FD_SET(fd, &fdset); + return select(fd+1, &fdset, NULL, NULL, &zero) != 1; +} + +/* + * This handles packets coming in from the tun device to our Guest. Like all + * service routines, it gets called again as soon as it returns, so you don't + * see a while(1) loop here. + */ +static void net_input(struct virtqueue *vq) +{ + int len; + unsigned int head, out, in; + struct iovec iov[vq->vring.num]; + struct net_info *net_info = vq->dev->priv; + + /* + * Get a descriptor to write an incoming packet into. This will also + * send an interrupt if they're out of descriptors. + */ + head = wait_for_vq_desc(vq, iov, &out, &in); + if (out) + errx(1, "Output buffers in net input queue?"); + + /* + * If it looks like we'll block reading from the tun device, send them + * an interrupt. + */ + if (vq->pending_used && will_block(net_info->tunfd)) + trigger_irq(vq); + + /* + * Read in the packet. This is where we normally wait (when there's no + * incoming network traffic). + */ + len = readv(net_info->tunfd, iov, in); + if (len <= 0) + err(1, "Failed to read from tun."); + + /* + * Mark that packet buffer as used, but don't interrupt here. We want + * to wait until we've done as much work as we can. + */ + add_used(vq, head, len); +} +/*:*/ + +/* This is the helper to create threads: run the service routine in a loop. */ +static int do_thread(void *_vq) +{ + struct virtqueue *vq = _vq; + + for (;;) + vq->service(vq); + return 0; +} + +/* + * When a child dies, we kill our entire process group with SIGTERM. This + * also has the side effect that the shell restores the console for us! + */ +static void kill_launcher(int signal) +{ + kill(0, SIGTERM); +} + +static void reset_device(struct device *dev) +{ + struct virtqueue *vq; + + verbose("Resetting device %s\n", dev->name); + + /* Clear any features they've acked. */ + memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); + + /* We're going to be explicitly killing threads, so ignore them. */ + signal(SIGCHLD, SIG_IGN); + + /* Zero out the virtqueues, get rid of their threads */ + for (vq = dev->vq; vq; vq = vq->next) { + if (vq->thread != (pid_t)-1) { + kill(vq->thread, SIGTERM); + waitpid(vq->thread, NULL, 0); + vq->thread = (pid_t)-1; + } + memset(vq->vring.desc, 0, + vring_size(vq->config.num, LGUEST_VRING_ALIGN)); + lg_last_avail(vq) = 0; + } + dev->running = false; + + /* Now we care if threads die. */ + signal(SIGCHLD, (void *)kill_launcher); +} + +/*L:216 + * This actually creates the thread which services the virtqueue for a device. + */ +static void create_thread(struct virtqueue *vq) +{ + /* + * Create stack for thread. Since the stack grows upwards, we point + * the stack pointer to the end of this region. + */ + char *stack = malloc(32768); + unsigned long args[] = { LHREQ_EVENTFD, + vq->config.pfn*getpagesize(), 0 }; + + /* Create a zero-initialized eventfd. */ + vq->eventfd = eventfd(0, 0); + if (vq->eventfd < 0) + err(1, "Creating eventfd"); + args[2] = vq->eventfd; + + /* + * Attach an eventfd to this virtqueue: it will go off when the Guest + * does an LHCALL_NOTIFY for this vq. + */ + if (write(lguest_fd, &args, sizeof(args)) != 0) + err(1, "Attaching eventfd"); + + /* + * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so + * we get a signal if it dies. + */ + vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); + if (vq->thread == (pid_t)-1) + err(1, "Creating clone"); + + /* We close our local copy now the child has it. */ + close(vq->eventfd); +} + +static bool accepted_feature(struct device *dev, unsigned int bit) +{ + const u8 *features = get_feature_bits(dev) + dev->feature_len; + + if (dev->feature_len < bit / CHAR_BIT) + return false; + return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT)); +} + +static void start_device(struct device *dev) +{ + unsigned int i; + struct virtqueue *vq; + + verbose("Device %s OK: offered", dev->name); + for (i = 0; i < dev->feature_len; i++) + verbose(" %02x", get_feature_bits(dev)[i]); + verbose(", accepted"); + for (i = 0; i < dev->feature_len; i++) + verbose(" %02x", get_feature_bits(dev) + [dev->feature_len+i]); + + dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); + + for (vq = dev->vq; vq; vq = vq->next) { + if (vq->service) + create_thread(vq); + } + dev->running = true; +} + +static void cleanup_devices(void) +{ + struct device *dev; + + for (dev = devices.dev; dev; dev = dev->next) + reset_device(dev); + + /* If we saved off the original terminal settings, restore them now. */ + if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) + tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); +} + +/* When the Guest tells us they updated the status field, we handle it. */ +static void update_device_status(struct device *dev) +{ + /* A zero status is a reset, otherwise it's a set of flags. */ + if (dev->desc->status == 0) + reset_device(dev); + else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { + warnx("Device %s configuration FAILED", dev->name); + if (dev->running) + reset_device(dev); + } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { + if (!dev->running) + start_device(dev); + } +} + +/*L:215 + * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In + * particular, it's used to notify us of device status changes during boot. + */ +static void handle_output(unsigned long addr) +{ + struct device *i; + + /* Check each device. */ + for (i = devices.dev; i; i = i->next) { + struct virtqueue *vq; + + /* + * Notifications to device descriptors mean they updated the + * device status. + */ + if (from_guest_phys(addr) == i->desc) { + update_device_status(i); + return; + } + + /* + * Devices *can* be used before status is set to DRIVER_OK. + * The original plan was that they would never do this: they + * would always finish setting up their status bits before + * actually touching the virtqueues. In practice, we allowed + * them to, and they do (eg. the disk probes for partition + * tables as part of initialization). + * + * If we see this, we start the device: once it's running, we + * expect the device to catch all the notifications. + */ + for (vq = i->vq; vq; vq = vq->next) { + if (addr != vq->config.pfn*getpagesize()) + continue; + if (i->running) + errx(1, "Notification on running %s", i->name); + /* This just calls create_thread() for each virtqueue */ + start_device(i); + return; + } + } + + /* + * Early console write is done using notify on a nul-terminated string + * in Guest memory. It's also great for hacking debugging messages + * into a Guest. + */ + if (addr >= guest_limit) + errx(1, "Bad NOTIFY %#lx", addr); + + write(STDOUT_FILENO, from_guest_phys(addr), + strnlen(from_guest_phys(addr), guest_limit - addr)); +} + +/*L:190 + * Device Setup + * + * All devices need a descriptor so the Guest knows it exists, and a "struct + * device" so the Launcher can keep track of it. We have common helper + * routines to allocate and manage them. + */ + +/* + * The layout of the device page is a "struct lguest_device_desc" followed by a + * number of virtqueue descriptors, then two sets of feature bits, then an + * array of configuration bytes. This routine returns the configuration + * pointer. + */ +static u8 *device_config(const struct device *dev) +{ + return (void *)(dev->desc + 1) + + dev->num_vq * sizeof(struct lguest_vqconfig) + + dev->feature_len * 2; +} + +/* + * This routine allocates a new "struct lguest_device_desc" from descriptor + * table page just above the Guest's normal memory. It returns a pointer to + * that descriptor. + */ +static struct lguest_device_desc *new_dev_desc(u16 type) +{ + struct lguest_device_desc d = { .type = type }; + void *p; + + /* Figure out where the next device config is, based on the last one. */ + if (devices.lastdev) + p = device_config(devices.lastdev) + + devices.lastdev->desc->config_len; + else + p = devices.descpage; + + /* We only have one page for all the descriptors. */ + if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) + errx(1, "Too many devices"); + + /* p might not be aligned, so we memcpy in. */ + return memcpy(p, &d, sizeof(d)); +} + +/* + * Each device descriptor is followed by the description of its virtqueues. We + * specify how many descriptors the virtqueue is to have. + */ +static void add_virtqueue(struct device *dev, unsigned int num_descs, + void (*service)(struct virtqueue *)) +{ + unsigned int pages; + struct virtqueue **i, *vq = malloc(sizeof(*vq)); + void *p; + + /* First we need some memory for this virtqueue. */ + pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) + / getpagesize(); + p = get_pages(pages); + + /* Initialize the virtqueue */ + vq->next = NULL; + vq->last_avail_idx = 0; + vq->dev = dev; + + /* + * This is the routine the service thread will run, and its Process ID + * once it's running. + */ + vq->service = service; + vq->thread = (pid_t)-1; + + /* Initialize the configuration. */ + vq->config.num = num_descs; + vq->config.irq = devices.next_irq++; + vq->config.pfn = to_guest_phys(p) / getpagesize(); + + /* Initialize the vring. */ + vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); + + /* + * Append virtqueue to this device's descriptor. We use + * device_config() to get the end of the device's current virtqueues; + * we check that we haven't added any config or feature information + * yet, otherwise we'd be overwriting them. + */ + assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); + memcpy(device_config(dev), &vq->config, sizeof(vq->config)); + dev->num_vq++; + dev->desc->num_vq++; + + verbose("Virtqueue page %#lx\n", to_guest_phys(p)); + + /* + * Add to tail of list, so dev->vq is first vq, dev->vq->next is + * second. + */ + for (i = &dev->vq; *i; i = &(*i)->next); + *i = vq; +} + +/* + * The first half of the feature bitmask is for us to advertise features. The + * second half is for the Guest to accept features. + */ +static void add_feature(struct device *dev, unsigned bit) +{ + u8 *features = get_feature_bits(dev); + + /* We can't extend the feature bits once we've added config bytes */ + if (dev->desc->feature_len <= bit / CHAR_BIT) { + assert(dev->desc->config_len == 0); + dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; + } + + features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); +} + +/* + * This routine sets the configuration fields for an existing device's + * descriptor. It only works for the last device, but that's OK because that's + * how we use it. + */ +static void set_config(struct device *dev, unsigned len, const void *conf) +{ + /* Check we haven't overflowed our single page. */ + if (device_config(dev) + len > devices.descpage + getpagesize()) + errx(1, "Too many devices"); + + /* Copy in the config information, and store the length. */ + memcpy(device_config(dev), conf, len); + dev->desc->config_len = len; + + /* Size must fit in config_len field (8 bits)! */ + assert(dev->desc->config_len == len); +} + +/* + * This routine does all the creation and setup of a new device, including + * calling new_dev_desc() to allocate the descriptor and device memory. We + * don't actually start the service threads until later. + * + * See what I mean about userspace being boring? + */ +static struct device *new_device(const char *name, u16 type) +{ + struct device *dev = malloc(sizeof(*dev)); + + /* Now we populate the fields one at a time. */ + dev->desc = new_dev_desc(type); + dev->name = name; + dev->vq = NULL; + dev->feature_len = 0; + dev->num_vq = 0; + dev->running = false; + + /* + * Append to device list. Prepending to a single-linked list is + * easier, but the user expects the devices to be arranged on the bus + * in command-line order. The first network device on the command line + * is eth0, the first block device /dev/vda, etc. + */ + if (devices.lastdev) + devices.lastdev->next = dev; + else + devices.dev = dev; + devices.lastdev = dev; + + return dev; +} + +/* + * Our first setup routine is the console. It's a fairly simple device, but + * UNIX tty handling makes it uglier than it could be. + */ +static void setup_console(void) +{ + struct device *dev; + + /* If we can save the initial standard input settings... */ + if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { + struct termios term = orig_term; + /* + * Then we turn off echo, line buffering and ^C etc: We want a + * raw input stream to the Guest. + */ + term.c_lflag &= ~(ISIG|ICANON|ECHO); + tcsetattr(STDIN_FILENO, TCSANOW, &term); + } + + dev = new_device("console", VIRTIO_ID_CONSOLE); + + /* We store the console state in dev->priv, and initialize it. */ + dev->priv = malloc(sizeof(struct console_abort)); + ((struct console_abort *)dev->priv)->count = 0; + + /* + * The console needs two virtqueues: the input then the output. When + * they put something the input queue, we make sure we're listening to + * stdin. When they put something in the output queue, we write it to + * stdout. + */ + add_virtqueue(dev, VIRTQUEUE_NUM, console_input); + add_virtqueue(dev, VIRTQUEUE_NUM, console_output); + + verbose("device %u: console\n", ++devices.device_num); +} +/*:*/ + +/*M:010 + * Inter-guest networking is an interesting area. Simplest is to have a + * --sharenet=<name> option which opens or creates a named pipe. This can be + * used to send packets to another guest in a 1:1 manner. + * + * More sopisticated is to use one of the tools developed for project like UML + * to do networking. + * + * Faster is to do virtio bonding in kernel. Doing this 1:1 would be + * completely generic ("here's my vring, attach to your vring") and would work + * for any traffic. Of course, namespace and permissions issues need to be + * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide + * multiple inter-guest channels behind one interface, although it would + * require some manner of hotplugging new virtio channels. + * + * Finally, we could implement a virtio network switch in the kernel. +:*/ + +static u32 str2ip(const char *ipaddr) +{ + unsigned int b[4]; + + if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) + errx(1, "Failed to parse IP address '%s'", ipaddr); + return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; +} + +static void str2mac(const char *macaddr, unsigned char mac[6]) +{ + unsigned int m[6]; + if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", + &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) + errx(1, "Failed to parse mac address '%s'", macaddr); + mac[0] = m[0]; + mac[1] = m[1]; + mac[2] = m[2]; + mac[3] = m[3]; + mac[4] = m[4]; + mac[5] = m[5]; +} + +/* + * This code is "adapted" from libbridge: it attaches the Host end of the + * network device to the bridge device specified by the command line. + * + * This is yet another James Morris contribution (I'm an IP-level guy, so I + * dislike bridging), and I just try not to break it. + */ +static void add_to_bridge(int fd, const char *if_name, const char *br_name) +{ + int ifidx; + struct ifreq ifr; + + if (!*br_name) + errx(1, "must specify bridge name"); + + ifidx = if_nametoindex(if_name); + if (!ifidx) + errx(1, "interface %s does not exist!", if_name); + + strncpy(ifr.ifr_name, br_name, IFNAMSIZ); + ifr.ifr_name[IFNAMSIZ-1] = '\0'; + ifr.ifr_ifindex = ifidx; + if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) + err(1, "can't add %s to bridge %s", if_name, br_name); +} + +/* + * This sets up the Host end of the network device with an IP address, brings + * it up so packets will flow, the copies the MAC address into the hwaddr + * pointer. + */ +static void configure_device(int fd, const char *tapif, u32 ipaddr) +{ + struct ifreq ifr; + struct sockaddr_in sin; + + memset(&ifr, 0, sizeof(ifr)); + strcpy(ifr.ifr_name, tapif); + + /* Don't read these incantations. Just cut & paste them like I did! */ + sin.sin_family = AF_INET; + sin.sin_addr.s_addr = htonl(ipaddr); + memcpy(&ifr.ifr_addr, &sin, sizeof(sin)); + if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) + err(1, "Setting %s interface address", tapif); + ifr.ifr_flags = IFF_UP; + if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) + err(1, "Bringing interface %s up", tapif); +} + +static int get_tun_device(char tapif[IFNAMSIZ]) +{ + struct ifreq ifr; + int netfd; + + /* Start with this zeroed. Messy but sure. */ + memset(&ifr, 0, sizeof(ifr)); + + /* + * We open the /dev/net/tun device and tell it we want a tap device. A + * tap device is like a tun device, only somehow different. To tell + * the truth, I completely blundered my way through this code, but it + * works now! + */ + netfd = open_or_die("/dev/net/tun", O_RDWR); + ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; + strcpy(ifr.ifr_name, "tap%d"); + if (ioctl(netfd, TUNSETIFF, &ifr) != 0) + err(1, "configuring /dev/net/tun"); + + if (ioctl(netfd, TUNSETOFFLOAD, + TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) + err(1, "Could not set features for tun device"); + + /* + * We don't need checksums calculated for packets coming in this + * device: trust us! + */ + ioctl(netfd, TUNSETNOCSUM, 1); + + memcpy(tapif, ifr.ifr_name, IFNAMSIZ); + return netfd; +} + +/*L:195 + * Our network is a Host<->Guest network. This can either use bridging or + * routing, but the principle is the same: it uses the "tun" device to inject + * packets into the Host as if they came in from a normal network card. We + * just shunt packets between the Guest and the tun device. + */ +static void setup_tun_net(char *arg) +{ + struct device *dev; + struct net_info *net_info = malloc(sizeof(*net_info)); + int ipfd; + u32 ip = INADDR_ANY; + bool bridging = false; + char tapif[IFNAMSIZ], *p; + struct virtio_net_config conf; + + net_info->tunfd = get_tun_device(tapif); + + /* First we create a new network device. */ + dev = new_device("net", VIRTIO_ID_NET); + dev->priv = net_info; + + /* Network devices need a recv and a send queue, just like console. */ + add_virtqueue(dev, VIRTQUEUE_NUM, net_input); + add_virtqueue(dev, VIRTQUEUE_NUM, net_output); + + /* + * We need a socket to perform the magic network ioctls to bring up the + * tap interface, connect to the bridge etc. Any socket will do! + */ + ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); + if (ipfd < 0) + err(1, "opening IP socket"); + + /* If the command line was --tunnet=bridge:<name> do bridging. */ + if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { + arg += strlen(BRIDGE_PFX); + bridging = true; + } + + /* A mac address may follow the bridge name or IP address */ + p = strchr(arg, ':'); + if (p) { + str2mac(p+1, conf.mac); + add_feature(dev, VIRTIO_NET_F_MAC); + *p = '\0'; + } + + /* arg is now either an IP address or a bridge name */ + if (bridging) + add_to_bridge(ipfd, tapif, arg); + else + ip = str2ip(arg); + + /* Set up the tun device. */ + configure_device(ipfd, tapif, ip); + + add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); + /* Expect Guest to handle everything except UFO */ + add_feature(dev, VIRTIO_NET_F_CSUM); + add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); + add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); + add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); + add_feature(dev, VIRTIO_NET_F_GUEST_ECN); + add_feature(dev, VIRTIO_NET_F_HOST_TSO4); + add_feature(dev, VIRTIO_NET_F_HOST_TSO6); + add_feature(dev, VIRTIO_NET_F_HOST_ECN); + /* We handle indirect ring entries */ + add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); + set_config(dev, sizeof(conf), &conf); + + /* We don't need the socket any more; setup is done. */ + close(ipfd); + + devices.device_num++; + + if (bridging) + verbose("device %u: tun %s attached to bridge: %s\n", + devices.device_num, tapif, arg); + else + verbose("device %u: tun %s: %s\n", + devices.device_num, tapif, arg); +} +/*:*/ + +/* This hangs off device->priv. */ +struct vblk_info { + /* The size of the file. */ + off64_t len; + + /* The file descriptor for the file. */ + int fd; + +}; + +/*L:210 + * The Disk + * + * The disk only has one virtqueue, so it only has one thread. It is really + * simple: the Guest asks for a block number and we read or write that position + * in the file. + * + * Before we serviced each virtqueue in a separate thread, that was unacceptably + * slow: the Guest waits until the read is finished before running anything + * else, even if it could have been doing useful work. + * + * We could have used async I/O, except it's reputed to suck so hard that + * characters actually go missing from your code when you try to use it. + */ +static void blk_request(struct virtqueue *vq) +{ + struct vblk_info *vblk = vq->dev->priv; + unsigned int head, out_num, in_num, wlen; + int ret; + u8 *in; + struct virtio_blk_outhdr *out; + struct iovec iov[vq->vring.num]; + off64_t off; + + /* + * Get the next request, where we normally wait. It triggers the + * interrupt to acknowledge previously serviced requests (if any). + */ + head = wait_for_vq_desc(vq, iov, &out_num, &in_num); + + /* + * Every block request should contain at least one output buffer + * (detailing the location on disk and the type of request) and one + * input buffer (to hold the result). + */ + if (out_num == 0 || in_num == 0) + errx(1, "Bad virtblk cmd %u out=%u in=%u", + head, out_num, in_num); + + out = convert(&iov[0], struct virtio_blk_outhdr); + in = convert(&iov[out_num+in_num-1], u8); + /* + * For historical reasons, block operations are expressed in 512 byte + * "sectors". + */ + off = out->sector * 512; + + /* + * In general the virtio block driver is allowed to try SCSI commands. + * It'd be nice if we supported eject, for example, but we don't. + */ + if (out->type & VIRTIO_BLK_T_SCSI_CMD) { + fprintf(stderr, "Scsi commands unsupported\n"); + *in = VIRTIO_BLK_S_UNSUPP; + wlen = sizeof(*in); + } else if (out->type & VIRTIO_BLK_T_OUT) { + /* + * Write + * + * Move to the right location in the block file. This can fail + * if they try to write past end. + */ + if (lseek64(vblk->fd, off, SEEK_SET) != off) + err(1, "Bad seek to sector %llu", out->sector); + + ret = writev(vblk->fd, iov+1, out_num-1); + verbose("WRITE to sector %llu: %i\n", out->sector, ret); + + /* + * Grr... Now we know how long the descriptor they sent was, we + * make sure they didn't try to write over the end of the block + * file (possibly extending it). + */ + if (ret > 0 && off + ret > vblk->len) { + /* Trim it back to the correct length */ + ftruncate64(vblk->fd, vblk->len); + /* Die, bad Guest, die. */ + errx(1, "Write past end %llu+%u", off, ret); + } + + wlen = sizeof(*in); + *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); + } else if (out->type & VIRTIO_BLK_T_FLUSH) { + /* Flush */ + ret = fdatasync(vblk->fd); + verbose("FLUSH fdatasync: %i\n", ret); + wlen = sizeof(*in); + *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); + } else { + /* + * Read + * + * Move to the right location in the block file. This can fail + * if they try to read past end. + */ + if (lseek64(vblk->fd, off, SEEK_SET) != off) + err(1, "Bad seek to sector %llu", out->sector); + + ret = readv(vblk->fd, iov+1, in_num-1); + verbose("READ from sector %llu: %i\n", out->sector, ret); + if (ret >= 0) { + wlen = sizeof(*in) + ret; + *in = VIRTIO_BLK_S_OK; + } else { + wlen = sizeof(*in); + *in = VIRTIO_BLK_S_IOERR; + } + } + + /* Finished that request. */ + add_used(vq, head, wlen); +} + +/*L:198 This actually sets up a virtual block device. */ +static void setup_block_file(const char *filename) +{ + struct device *dev; + struct vblk_info *vblk; + struct virtio_blk_config conf; + + /* Creat the device. */ + dev = new_device("block", VIRTIO_ID_BLOCK); + + /* The device has one virtqueue, where the Guest places requests. */ + add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); + + /* Allocate the room for our own bookkeeping */ + vblk = dev->priv = malloc(sizeof(*vblk)); + + /* First we open the file and store the length. */ + vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); + vblk->len = lseek64(vblk->fd, 0, SEEK_END); + + /* We support FLUSH. */ + add_feature(dev, VIRTIO_BLK_F_FLUSH); + + /* Tell Guest how many sectors this device has. */ + conf.capacity = cpu_to_le64(vblk->len / 512); + + /* + * Tell Guest not to put in too many descriptors at once: two are used + * for the in and out elements. + */ + add_feature(dev, VIRTIO_BLK_F_SEG_MAX); + conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); + + /* Don't try to put whole struct: we have 8 bit limit. */ + set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf); + + verbose("device %u: virtblock %llu sectors\n", + ++devices.device_num, le64_to_cpu(conf.capacity)); +} + +/*L:211 + * Our random number generator device reads from /dev/random into the Guest's + * input buffers. The usual case is that the Guest doesn't want random numbers + * and so has no buffers although /dev/random is still readable, whereas + * console is the reverse. + * + * The same logic applies, however. + */ +struct rng_info { + int rfd; +}; + +static void rng_input(struct virtqueue *vq) +{ + int len; + unsigned int head, in_num, out_num, totlen = 0; + struct rng_info *rng_info = vq->dev->priv; + struct iovec iov[vq->vring.num]; + + /* First we need a buffer from the Guests's virtqueue. */ + head = wait_for_vq_desc(vq, iov, &out_num, &in_num); + if (out_num) + errx(1, "Output buffers in rng?"); + + /* + * Just like the console write, we loop to cover the whole iovec. + * In this case, short reads actually happen quite a bit. + */ + while (!iov_empty(iov, in_num)) { + len = readv(rng_info->rfd, iov, in_num); + if (len <= 0) + err(1, "Read from /dev/random gave %i", len); + iov_consume(iov, in_num, len); + totlen += len; + } + + /* Tell the Guest about the new input. */ + add_used(vq, head, totlen); +} + +/*L:199 + * This creates a "hardware" random number device for the Guest. + */ +static void setup_rng(void) +{ + struct device *dev; + struct rng_info *rng_info = malloc(sizeof(*rng_info)); + + /* Our device's privat info simply contains the /dev/random fd. */ + rng_info->rfd = open_or_die("/dev/random", O_RDONLY); + + /* Create the new device. */ + dev = new_device("rng", VIRTIO_ID_RNG); + dev->priv = rng_info; + + /* The device has one virtqueue, where the Guest places inbufs. */ + add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); + + verbose("device %u: rng\n", devices.device_num++); +} +/* That's the end of device setup. */ + +/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ +static void __attribute__((noreturn)) restart_guest(void) +{ + unsigned int i; + + /* + * Since we don't track all open fds, we simply close everything beyond + * stderr. + */ + for (i = 3; i < FD_SETSIZE; i++) + close(i); + + /* Reset all the devices (kills all threads). */ + cleanup_devices(); + + execv(main_args[0], main_args); + err(1, "Could not exec %s", main_args[0]); +} + +/*L:220 + * Finally we reach the core of the Launcher which runs the Guest, serves + * its input and output, and finally, lays it to rest. + */ +static void __attribute__((noreturn)) run_guest(void) +{ + for (;;) { + unsigned long notify_addr; + int readval; + + /* We read from the /dev/lguest device to run the Guest. */ + readval = pread(lguest_fd, ¬ify_addr, + sizeof(notify_addr), cpu_id); + + /* One unsigned long means the Guest did HCALL_NOTIFY */ + if (readval == sizeof(notify_addr)) { + verbose("Notify on address %#lx\n", notify_addr); + handle_output(notify_addr); + /* ENOENT means the Guest died. Reading tells us why. */ + } else if (errno == ENOENT) { + char reason[1024] = { 0 }; + pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); + errx(1, "%s", reason); + /* ERESTART means that we need to reboot the guest */ + } else if (errno == ERESTART) { + restart_guest(); + /* Anything else means a bug or incompatible change. */ + } else + err(1, "Running guest failed"); + } +} +/*L:240 + * This is the end of the Launcher. The good news: we are over halfway + * through! The bad news: the most fiendish part of the code still lies ahead + * of us. + * + * Are you ready? Take a deep breath and join me in the core of the Host, in + * "make Host". +:*/ + +static struct option opts[] = { + { "verbose", 0, NULL, 'v' }, + { "tunnet", 1, NULL, 't' }, + { "block", 1, NULL, 'b' }, + { "rng", 0, NULL, 'r' }, + { "initrd", 1, NULL, 'i' }, + { "username", 1, NULL, 'u' }, + { "chroot", 1, NULL, 'c' }, + { NULL }, +}; +static void usage(void) +{ + errx(1, "Usage: lguest [--verbose] " + "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" + "|--block=<filename>|--initrd=<filename>]...\n" + "<mem-in-mb> vmlinux [args...]"); +} + +/*L:105 The main routine is where the real work begins: */ +int main(int argc, char *argv[]) +{ + /* Memory, code startpoint and size of the (optional) initrd. */ + unsigned long mem = 0, start, initrd_size = 0; + /* Two temporaries. */ + int i, c; + /* The boot information for the Guest. */ + struct boot_params *boot; + /* If they specify an initrd file to load. */ + const char *initrd_name = NULL; + + /* Password structure for initgroups/setres[gu]id */ + struct passwd *user_details = NULL; + + /* Directory to chroot to */ + char *chroot_path = NULL; + + /* Save the args: we "reboot" by execing ourselves again. */ + main_args = argv; + + /* + * First we initialize the device list. We keep a pointer to the last + * device, and the next interrupt number to use for devices (1: + * remember that 0 is used by the timer). + */ + devices.lastdev = NULL; + devices.next_irq = 1; + + /* We're CPU 0. In fact, that's the only CPU possible right now. */ + cpu_id = 0; + + /* + * We need to know how much memory so we can set up the device + * descriptor and memory pages for the devices as we parse the command + * line. So we quickly look through the arguments to find the amount + * of memory now. + */ + for (i = 1; i < argc; i++) { + if (argv[i][0] != '-') { + mem = atoi(argv[i]) * 1024 * 1024; + /* + * We start by mapping anonymous pages over all of + * guest-physical memory range. This fills it with 0, + * and ensures that the Guest won't be killed when it + * tries to access it. + */ + guest_base = map_zeroed_pages(mem / getpagesize() + + DEVICE_PAGES); + guest_limit = mem; + guest_max = mem + DEVICE_PAGES*getpagesize(); + devices.descpage = get_pages(1); + break; + } + } + + /* The options are fairly straight-forward */ + while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { + switch (c) { + case 'v': + verbose = true; + break; + case 't': + setup_tun_net(optarg); + break; + case 'b': + setup_block_file(optarg); + break; + case 'r': + setup_rng(); + break; + case 'i': + initrd_name = optarg; + break; + case 'u': + user_details = getpwnam(optarg); + if (!user_details) + err(1, "getpwnam failed, incorrect username?"); + break; + case 'c': + chroot_path = optarg; + break; + default: + warnx("Unknown argument %s", argv[optind]); + usage(); + } + } + /* + * After the other arguments we expect memory and kernel image name, + * followed by command line arguments for the kernel. + */ + if (optind + 2 > argc) + usage(); + + verbose("Guest base is at %p\n", guest_base); + + /* We always have a console device */ + setup_console(); + + /* Now we load the kernel */ + start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); + + /* Boot information is stashed at physical address 0 */ + boot = from_guest_phys(0); + + /* Map the initrd image if requested (at top of physical memory) */ + if (initrd_name) { + initrd_size = load_initrd(initrd_name, mem); + /* + * These are the location in the Linux boot header where the + * start and size of the initrd are expected to be found. + */ + boot->hdr.ramdisk_image = mem - initrd_size; + boot->hdr.ramdisk_size = initrd_size; + /* The bootloader type 0xFF means "unknown"; that's OK. */ + boot->hdr.type_of_loader = 0xFF; + } + + /* + * The Linux boot header contains an "E820" memory map: ours is a + * simple, single region. + */ + boot->e820_entries = 1; + boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); + /* + * The boot header contains a command line pointer: we put the command + * line after the boot header. + */ + boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); + /* We use a simple helper to copy the arguments separated by spaces. */ + concat((char *)(boot + 1), argv+optind+2); + + /* Boot protocol version: 2.07 supports the fields for lguest. */ + boot->hdr.version = 0x207; + + /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ + boot->hdr.hardware_subarch = 1; + + /* Tell the entry path not to try to reload segment registers. */ + boot->hdr.loadflags |= KEEP_SEGMENTS; + + /* + * We tell the kernel to initialize the Guest: this returns the open + * /dev/lguest file descriptor. + */ + tell_kernel(start); + + /* Ensure that we terminate if a device-servicing child dies. */ + signal(SIGCHLD, kill_launcher); + + /* If we exit via err(), this kills all the threads, restores tty. */ + atexit(cleanup_devices); + + /* If requested, chroot to a directory */ + if (chroot_path) { + if (chroot(chroot_path) != 0) + err(1, "chroot(\"%s\") failed", chroot_path); + + if (chdir("/") != 0) + err(1, "chdir(\"/\") failed"); + + verbose("chroot done\n"); + } + + /* If requested, drop privileges */ + if (user_details) { + uid_t u; + gid_t g; + + u = user_details->pw_uid; + g = user_details->pw_gid; + + if (initgroups(user_details->pw_name, g) != 0) + err(1, "initgroups failed"); + + if (setresgid(g, g, g) != 0) + err(1, "setresgid failed"); + + if (setresuid(u, u, u) != 0) + err(1, "setresuid failed"); + + verbose("Dropping privileges completed\n"); + } + + /* Finally, run the Guest. This doesn't return. */ + run_guest(); +} +/*:*/ + +/*M:999 + * Mastery is done: you now know everything I do. + * + * But surely you have seen code, features and bugs in your wanderings which + * you now yearn to attack? That is the real game, and I look forward to you + * patching and forking lguest into the Your-Name-Here-visor. + * + * Farewell, and good coding! + * Rusty Russell. + */ diff --git a/Documentation/virtual/lguest/lguest.txt b/Documentation/virtual/lguest/lguest.txt new file mode 100644 index 00000000000..bff0c554485 --- /dev/null +++ b/Documentation/virtual/lguest/lguest.txt @@ -0,0 +1,129 @@ + __ + (___()'`; Rusty's Remarkably Unreliable Guide to Lguest + /, /` - or, A Young Coder's Illustrated Hypervisor + \\"--\\ http://lguest.ozlabs.org + +Lguest is designed to be a minimal 32-bit x86 hypervisor for the Linux kernel, +for Linux developers and users to experiment with virtualization with the +minimum of complexity. Nonetheless, it should have sufficient features to +make it useful for specific tasks, and, of course, you are encouraged to fork +and enhance it (see drivers/lguest/README). + +Features: + +- Kernel module which runs in a normal kernel. +- Simple I/O model for communication. +- Simple program to create new guests. +- Logo contains cute puppies: http://lguest.ozlabs.org + +Developer features: + +- Fun to hack on. +- No ABI: being tied to a specific kernel anyway, you can change anything. +- Many opportunities for improvement or feature implementation. + +Running Lguest: + +- The easiest way to run lguest is to use same kernel as guest and host. + You can configure them differently, but usually it's easiest not to. + + You will need to configure your kernel with the following options: + + "General setup": + "Prompt for development and/or incomplete code/drivers" = Y + (CONFIG_EXPERIMENTAL=y) + + "Processor type and features": + "Paravirtualized guest support" = Y + "Lguest guest support" = Y + "High Memory Support" = off/4GB + "Alignment value to which kernel should be aligned" = 0x100000 + (CONFIG_PARAVIRT=y, CONFIG_LGUEST_GUEST=y, CONFIG_HIGHMEM64G=n and + CONFIG_PHYSICAL_ALIGN=0x100000) + + "Device Drivers": + "Block devices" + "Virtio block driver (EXPERIMENTAL)" = M/Y + "Network device support" + "Universal TUN/TAP device driver support" = M/Y + "Virtio network driver (EXPERIMENTAL)" = M/Y + (CONFIG_VIRTIO_BLK=m, CONFIG_VIRTIO_NET=m and CONFIG_TUN=m) + + "Virtualization" + "Linux hypervisor example code" = M/Y + (CONFIG_LGUEST=m) + +- A tool called "lguest" is available in this directory: type "make" + to build it. If you didn't build your kernel in-tree, use "make + O=<builddir>". + +- Create or find a root disk image. There are several useful ones + around, such as the xm-test tiny root image at + http://xm-test.xensource.com/ramdisks/initrd-1.1-i386.img + + For more serious work, I usually use a distribution ISO image and + install it under qemu, then make multiple copies: + + dd if=/dev/zero of=rootfile bs=1M count=2048 + qemu -cdrom image.iso -hda rootfile -net user -net nic -boot d + + Make sure that you install a getty on /dev/hvc0 if you want to log in on the + console! + +- "modprobe lg" if you built it as a module. + +- Run an lguest as root: + + Documentation/virtual/lguest/lguest 64 vmlinux --tunnet=192.168.19.1 \ + --block=rootfile root=/dev/vda + + Explanation: + 64: the amount of memory to use, in MB. + + vmlinux: the kernel image found in the top of your build directory. You + can also use a standard bzImage. + + --tunnet=192.168.19.1: configures a "tap" device for networking with this + IP address. + + --block=rootfile: a file or block device which becomes /dev/vda + inside the guest. + + root=/dev/vda: this (and anything else on the command line) are + kernel boot parameters. + +- Configuring networking. I usually have the host masquerade, using + "iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE" and "echo 1 > + /proc/sys/net/ipv4/ip_forward". In this example, I would configure + eth0 inside the guest at 192.168.19.2. + + Another method is to bridge the tap device to an external interface + using --tunnet=bridge:<bridgename>, and perhaps run dhcp on the guest + to obtain an IP address. The bridge needs to be configured first: + this option simply adds the tap interface to it. + + A simple example on my system: + + ifconfig eth0 0.0.0.0 + brctl addbr lg0 + ifconfig lg0 up + brctl addif lg0 eth0 + dhclient lg0 + + Then use --tunnet=bridge:lg0 when launching the guest. + + See: + + http://www.linuxfoundation.org/collaborate/workgroups/networking/bridge + + for general information on how to get bridging to work. + +- Random number generation. Using the --rng option will provide a + /dev/hwrng in the guest that will read from the host's /dev/random. + Use this option in conjunction with rng-tools (see ../hw_random.txt) + to provide entropy to the guest kernel's /dev/random. + +There is a helpful mailing list at http://ozlabs.org/mailman/listinfo/lguest + +Good luck! +Rusty Russell rusty@rustcorp.com.au. diff --git a/Documentation/virtual/uml/UserModeLinux-HOWTO.txt b/Documentation/virtual/uml/UserModeLinux-HOWTO.txt new file mode 100644 index 00000000000..9b7e1904db1 --- /dev/null +++ b/Documentation/virtual/uml/UserModeLinux-HOWTO.txt @@ -0,0 +1,4579 @@ + User Mode Linux HOWTO + User Mode Linux Core Team + Mon Nov 18 14:16:16 EST 2002 + + This document describes the use and abuse of Jeff Dike's User Mode + Linux: a port of the Linux kernel as a normal Intel Linux process. + ______________________________________________________________________ + + Table of Contents + + 1. Introduction + + 1.1 How is User Mode Linux Different? + 1.2 Why Would I Want User Mode Linux? + + 2. Compiling the kernel and modules + + 2.1 Compiling the kernel + 2.2 Compiling and installing kernel modules + 2.3 Compiling and installing uml_utilities + + 3. Running UML and logging in + + 3.1 Running UML + 3.2 Logging in + 3.3 Examples + + 4. UML on 2G/2G hosts + + 4.1 Introduction + 4.2 The problem + 4.3 The solution + + 5. Setting up serial lines and consoles + + 5.1 Specifying the device + 5.2 Specifying the channel + 5.3 Examples + + 6. Setting up the network + + 6.1 General setup + 6.2 Userspace daemons + 6.3 Specifying ethernet addresses + 6.4 UML interface setup + 6.5 Multicast + 6.6 TUN/TAP with the uml_net helper + 6.7 TUN/TAP with a preconfigured tap device + 6.8 Ethertap + 6.9 The switch daemon + 6.10 Slip + 6.11 Slirp + 6.12 pcap + 6.13 Setting up the host yourself + + 7. Sharing Filesystems between Virtual Machines + + 7.1 A warning + 7.2 Using layered block devices + 7.3 Note! + 7.4 Another warning + 7.5 uml_moo : Merging a COW file with its backing file + + 8. Creating filesystems + + 8.1 Create the filesystem file + 8.2 Assign the file to a UML device + 8.3 Creating and mounting the filesystem + + 9. Host file access + + 9.1 Using hostfs + 9.2 hostfs as the root filesystem + 9.3 Building hostfs + + 10. The Management Console + 10.1 version + 10.2 halt and reboot + 10.3 config + 10.4 remove + 10.5 sysrq + 10.6 help + 10.7 cad + 10.8 stop + 10.9 go + + 11. Kernel debugging + + 11.1 Starting the kernel under gdb + 11.2 Examining sleeping processes + 11.3 Running ddd on UML + 11.4 Debugging modules + 11.5 Attaching gdb to the kernel + 11.6 Using alternate debuggers + + 12. Kernel debugging examples + + 12.1 The case of the hung fsck + 12.2 Episode 2: The case of the hung fsck + + 13. What to do when UML doesn't work + + 13.1 Strange compilation errors when you build from source + 13.2 (obsolete) + 13.3 A variety of panics and hangs with /tmp on a reiserfs filesystem + 13.4 The compile fails with errors about conflicting types for 'open', 'dup', and 'waitpid' + 13.5 UML doesn't work when /tmp is an NFS filesystem + 13.6 UML hangs on boot when compiled with gprof support + 13.7 syslogd dies with a SIGTERM on startup + 13.8 TUN/TAP networking doesn't work on a 2.4 host + 13.9 You can network to the host but not to other machines on the net + 13.10 I have no root and I want to scream + 13.11 UML build conflict between ptrace.h and ucontext.h + 13.12 The UML BogoMips is exactly half the host's BogoMips + 13.13 When you run UML, it immediately segfaults + 13.14 xterms appear, then immediately disappear + 13.15 Any other panic, hang, or strange behavior + + 14. Diagnosing Problems + + 14.1 Case 1 : Normal kernel panics + 14.2 Case 2 : Tracing thread panics + 14.3 Case 3 : Tracing thread panics caused by other threads + 14.4 Case 4 : Hangs + + 15. Thanks + + 15.1 Code and Documentation + 15.2 Flushing out bugs + 15.3 Buglets and clean-ups + 15.4 Case Studies + 15.5 Other contributions + + + ______________________________________________________________________ + + 11.. IInnttrroodduuccttiioonn + + Welcome to User Mode Linux. It's going to be fun. + + + + 11..11.. HHooww iiss UUsseerr MMooddee LLiinnuuxx DDiiffffeerreenntt?? + + Normally, the Linux Kernel talks straight to your hardware (video + card, keyboard, hard drives, etc), and any programs which run ask the + kernel to operate the hardware, like so: + + + + +-----------+-----------+----+ + | Process 1 | Process 2 | ...| + +-----------+-----------+----+ + | Linux Kernel | + +----------------------------+ + | Hardware | + +----------------------------+ + + + + + The User Mode Linux Kernel is different; instead of talking to the + hardware, it talks to a `real' Linux kernel (called the `host kernel' + from now on), like any other program. Programs can then run inside + User-Mode Linux as if they were running under a normal kernel, like + so: + + + + +----------------+ + | Process 2 | ...| + +-----------+----------------+ + | Process 1 | User-Mode Linux| + +----------------------------+ + | Linux Kernel | + +----------------------------+ + | Hardware | + +----------------------------+ + + + + + + 11..22.. WWhhyy WWoouulldd II WWaanntt UUsseerr MMooddee LLiinnuuxx?? + + + 1. If User Mode Linux crashes, your host kernel is still fine. + + 2. You can run a usermode kernel as a non-root user. + + 3. You can debug the User Mode Linux like any normal process. + + 4. You can run gprof (profiling) and gcov (coverage testing). + + 5. You can play with your kernel without breaking things. + + 6. You can use it as a sandbox for testing new apps. + + 7. You can try new development kernels safely. + + 8. You can run different distributions simultaneously. + + 9. It's extremely fun. + + + + + + 22.. CCoommppiilliinngg tthhee kkeerrnneell aanndd mmoodduulleess + + + + + 22..11.. CCoommppiilliinngg tthhee kkeerrnneell + + + Compiling the user mode kernel is just like compiling any other + kernel. Let's go through the steps, using 2.4.0-prerelease (current + as of this writing) as an example: + + + 1. Download the latest UML patch from + + the download page <http://user-mode-linux.sourceforge.net/ + + In this example, the file is uml-patch-2.4.0-prerelease.bz2. + + + 2. Download the matching kernel from your favourite kernel mirror, + such as: + + ftp://ftp.ca.kernel.org/pub/kernel/v2.4/linux-2.4.0-prerelease.tar.bz2 + <ftp://ftp.ca.kernel.org/pub/kernel/v2.4/linux-2.4.0-prerelease.tar.bz2> + . + + + 3. Make a directory and unpack the kernel into it. + + + + host% + mkdir ~/uml + + + + + + + host% + cd ~/uml + + + + + + + host% + tar -xzvf linux-2.4.0-prerelease.tar.bz2 + + + + + + + 4. Apply the patch using + + + + host% + cd ~/uml/linux + + + + host% + bzcat uml-patch-2.4.0-prerelease.bz2 | patch -p1 + + + + + + + 5. Run your favorite config; `make xconfig ARCH=um' is the most + convenient. `make config ARCH=um' and 'make menuconfig ARCH=um' + will work as well. The defaults will give you a useful kernel. If + you want to change something, go ahead, it probably won't hurt + anything. + + + Note: If the host is configured with a 2G/2G address space split + rather than the usual 3G/1G split, then the packaged UML binaries + will not run. They will immediately segfault. See ``UML on 2G/2G + hosts'' for the scoop on running UML on your system. + + + + 6. Finish with `make linux ARCH=um': the result is a file called + `linux' in the top directory of your source tree. + + Make sure that you don't build this kernel in /usr/src/linux. On some + distributions, /usr/include/asm is a link into this pool. The user- + mode build changes the other end of that link, and things that include + <asm/anything.h> stop compiling. + + The sources are also available from cvs at the project's cvs page, + which has directions on getting the sources. You can also browse the + CVS pool from there. + + If you get the CVS sources, you will have to check them out into an + empty directory. You will then have to copy each file into the + corresponding directory in the appropriate kernel pool. + + If you don't have the latest kernel pool, you can get the + corresponding user-mode sources with + + + host% cvs co -r v_2_3_x linux + + + + + where 'x' is the version in your pool. Note that you will not get the + bug fixes and enhancements that have gone into subsequent releases. + + + 22..22.. CCoommppiilliinngg aanndd iinnssttaalllliinngg kkeerrnneell mmoodduulleess + + UML modules are built in the same way as the native kernel (with the + exception of the 'ARCH=um' that you always need for UML): + + + host% make modules ARCH=um + + + + + Any modules that you want to load into this kernel need to be built in + the user-mode pool. Modules from the native kernel won't work. + + You can install them by using ftp or something to copy them into the + virtual machine and dropping them into /lib/modules/`uname -r`. + + You can also get the kernel build process to install them as follows: + + 1. with the kernel not booted, mount the root filesystem in the top + level of the kernel pool: + + + host% mount root_fs mnt -o loop + + + + + + + 2. run + + + host% + make modules_install INSTALL_MOD_PATH=`pwd`/mnt ARCH=um + + + + + + + 3. unmount the filesystem + + + host% umount mnt + + + + + + + 4. boot the kernel on it + + + When the system is booted, you can use insmod as usual to get the + modules into the kernel. A number of things have been loaded into UML + as modules, especially filesystems and network protocols and filters, + so most symbols which need to be exported probably already are. + However, if you do find symbols that need exporting, let us + <http://user-mode-linux.sourceforge.net/> know, and + they'll be "taken care of". + + + + 22..33.. CCoommppiilliinngg aanndd iinnssttaalllliinngg uummll__uuttiilliittiieess + + Many features of the UML kernel require a user-space helper program, + so a uml_utilities package is distributed separately from the kernel + patch which provides these helpers. Included within this is: + + +o port-helper - Used by consoles which connect to xterms or ports + + +o tunctl - Configuration tool to create and delete tap devices + + +o uml_net - Setuid binary for automatic tap device configuration + + +o uml_switch - User-space virtual switch required for daemon + transport + + The uml_utilities tree is compiled with: + + + host# + make && make install + + + + + Note that UML kernel patches may require a specific version of the + uml_utilities distribution. If you don't keep up with the mailing + lists, ensure that you have the latest release of uml_utilities if you + are experiencing problems with your UML kernel, particularly when + dealing with consoles or command-line switches to the helper programs + + + + + + + + + 33.. RRuunnnniinngg UUMMLL aanndd llooggggiinngg iinn + + + + 33..11.. RRuunnnniinngg UUMMLL + + It runs on 2.2.15 or later, and all 2.4 kernels. + + + Booting UML is straightforward. Simply run 'linux': it will try to + mount the file `root_fs' in the current directory. You do not need to + run it as root. If your root filesystem is not named `root_fs', then + you need to put a `ubd0=root_fs_whatever' switch on the linux command + line. + + + You will need a filesystem to boot UML from. There are a number + available for download from here <http://user-mode- + linux.sourceforge.net/> . There are also several tools + <http://user-mode-linux.sourceforge.net/> which can be + used to generate UML-compatible filesystem images from media. + The kernel will boot up and present you with a login prompt. + + + Note: If the host is configured with a 2G/2G address space split + rather than the usual 3G/1G split, then the packaged UML binaries will + not run. They will immediately segfault. See ``UML on 2G/2G hosts'' + for the scoop on running UML on your system. + + + + 33..22.. LLooggggiinngg iinn + + + + The prepackaged filesystems have a root account with password 'root' + and a user account with password 'user'. The login banner will + generally tell you how to log in. So, you log in and you will find + yourself inside a little virtual machine. Our filesystems have a + variety of commands and utilities installed (and it is fairly easy to + add more), so you will have a lot of tools with which to poke around + the system. + + There are a couple of other ways to log in: + + +o On a virtual console + + + + Each virtual console that is configured (i.e. the device exists in + /dev and /etc/inittab runs a getty on it) will come up in its own + xterm. If you get tired of the xterms, read ``Setting up serial + lines and consoles'' to see how to attach the consoles to + something else, like host ptys. + + + + +o Over the serial line + + + In the boot output, find a line that looks like: + + + + serial line 0 assigned pty /dev/ptyp1 + + + + + Attach your favorite terminal program to the corresponding tty. I.e. + for minicom, the command would be + + + host% minicom -o -p /dev/ttyp1 + + + + + + + +o Over the net + + + If the network is running, then you can telnet to the virtual + machine and log in to it. See ``Setting up the network'' to learn + about setting up a virtual network. + + When you're done using it, run halt, and the kernel will bring itself + down and the process will exit. + + + 33..33.. EExxaammpplleess + + Here are some examples of UML in action: + + +o A login session <http://user-mode-linux.sourceforge.net/login.html> + + +o A virtual network <http://user-mode-linux.sourceforge.net/net.html> + + + + + + + + 44.. UUMMLL oonn 22GG//22GG hhoossttss + + + + + 44..11.. IInnttrroodduuccttiioonn + + + Most Linux machines are configured so that the kernel occupies the + upper 1G (0xc0000000 - 0xffffffff) of the 4G address space and + processes use the lower 3G (0x00000000 - 0xbfffffff). However, some + machine are configured with a 2G/2G split, with the kernel occupying + the upper 2G (0x80000000 - 0xffffffff) and processes using the lower + 2G (0x00000000 - 0x7fffffff). + + + + + 44..22.. TThhee pprroobblleemm + + + The prebuilt UML binaries on this site will not run on 2G/2G hosts + because UML occupies the upper .5G of the 3G process address space + (0xa0000000 - 0xbfffffff). Obviously, on 2G/2G hosts, this is right + in the middle of the kernel address space, so UML won't even load - it + will immediately segfault. + + + + + 44..33.. TThhee ssoolluuttiioonn + + + The fix for this is to rebuild UML from source after enabling + CONFIG_HOST_2G_2G (under 'General Setup'). This will cause UML to + load itself in the top .5G of that smaller process address space, + where it will run fine. See ``Compiling the kernel and modules'' if + you need help building UML from source. + + + + + + + + + + + 55.. SSeettttiinngg uupp sseerriiaall lliinneess aanndd ccoonnssoolleess + + + It is possible to attach UML serial lines and consoles to many types + of host I/O channels by specifying them on the command line. + + + You can attach them to host ptys, ttys, file descriptors, and ports. + This allows you to do things like + + +o have a UML console appear on an unused host console, + + +o hook two virtual machines together by having one attach to a pty + and having the other attach to the corresponding tty + + +o make a virtual machine accessible from the net by attaching a + console to a port on the host. + + + The general format of the command line option is device=channel. + + + + 55..11.. SSppeecciiffyyiinngg tthhee ddeevviiccee + + Devices are specified with "con" or "ssl" (console or serial line, + respectively), optionally with a device number if you are talking + about a specific device. + + + Using just "con" or "ssl" describes all of the consoles or serial + lines. If you want to talk about console #3 or serial line #10, they + would be "con3" and "ssl10", respectively. + + + A specific device name will override a less general "con=" or "ssl=". + So, for example, you can assign a pty to each of the serial lines + except for the first two like this: + + + ssl=pty ssl0=tty:/dev/tty0 ssl1=tty:/dev/tty1 + + + + + The specificity of the device name is all that matters; order on the + command line is irrelevant. + + + + 55..22.. SSppeecciiffyyiinngg tthhee cchhaannnneell + + There are a number of different types of channels to attach a UML + device to, each with a different way of specifying exactly what to + attach to. + + +o pseudo-terminals - device=pty pts terminals - device=pts + + + This will cause UML to allocate a free host pseudo-terminal for the + device. The terminal that it got will be announced in the boot + log. You access it by attaching a terminal program to the + corresponding tty: + + +o screen /dev/pts/n + + +o screen /dev/ttyxx + + +o minicom -o -p /dev/ttyxx - minicom seems not able to handle pts + devices + + +o kermit - start it up, 'open' the device, then 'connect' + + + + + + +o terminals - device=tty:tty device file + + + This will make UML attach the device to the specified tty (i.e + + + con1=tty:/dev/tty3 + + + + + will attach UML's console 1 to the host's /dev/tty3). If the tty that + you specify is the slave end of a tty/pty pair, something else must + have already opened the corresponding pty in order for this to work. + + + + + + +o xterms - device=xterm + + + UML will run an xterm and the device will be attached to it. + + + + + + +o Port - device=port:port number + + + This will attach the UML devices to the specified host port. + Attaching console 1 to the host's port 9000 would be done like + this: + + + con1=port:9000 + + + + + Attaching all the serial lines to that port would be done similarly: + + + ssl=port:9000 + + + + + You access these devices by telnetting to that port. Each active tel- + net session gets a different device. If there are more telnets to a + port than UML devices attached to it, then the extra telnet sessions + will block until an existing telnet detaches, or until another device + becomes active (i.e. by being activated in /etc/inittab). + + This channel has the advantage that you can both attach multiple UML + devices to it and know how to access them without reading the UML boot + log. It is also unique in allowing access to a UML from remote + machines without requiring that the UML be networked. This could be + useful in allowing public access to UMLs because they would be + accessible from the net, but wouldn't need any kind of network + filtering or access control because they would have no network access. + + + If you attach the main console to a portal, then the UML boot will + appear to hang. In reality, it's waiting for a telnet to connect, at + which point the boot will proceed. + + + + + + +o already-existing file descriptors - device=file descriptor + + + If you set up a file descriptor on the UML command line, you can + attach a UML device to it. This is most commonly used to put the + main console back on stdin and stdout after assigning all the other + consoles to something else: + + + con0=fd:0,fd:1 con=pts + + + + + + + + + +o Nothing - device=null + + + This allows the device to be opened, in contrast to 'none', but + reads will block, and writes will succeed and the data will be + thrown out. + + + + + + +o None - device=none + + + This causes the device to disappear. + + + + You can also specify different input and output channels for a device + by putting a comma between them: + + + ssl3=tty:/dev/tty2,xterm + + + + + will cause serial line 3 to accept input on the host's /dev/tty3 and + display output on an xterm. That's a silly example - the most common + use of this syntax is to reattach the main console to stdin and stdout + as shown above. + + + If you decide to move the main console away from stdin/stdout, the + initial boot output will appear in the terminal that you're running + UML in. However, once the console driver has been officially + initialized, then the boot output will start appearing wherever you + specified that console 0 should be. That device will receive all + subsequent output. + + + + 55..33.. EExxaammpplleess + + There are a number of interesting things you can do with this + capability. + + + First, this is how you get rid of those bleeding console xterms by + attaching them to host ptys: + + + con=pty con0=fd:0,fd:1 + + + + + This will make a UML console take over an unused host virtual console, + so that when you switch to it, you will see the UML login prompt + rather than the host login prompt: + + + con1=tty:/dev/tty6 + + + + + You can attach two virtual machines together with what amounts to a + serial line as follows: + + Run one UML with a serial line attached to a pty - + + + ssl1=pty + + + + + Look at the boot log to see what pty it got (this example will assume + that it got /dev/ptyp1). + + Boot the other UML with a serial line attached to the corresponding + tty - + + + ssl1=tty:/dev/ttyp1 + + + + + Log in, make sure that it has no getty on that serial line, attach a + terminal program like minicom to it, and you should see the login + prompt of the other virtual machine. + + + 66.. SSeettttiinngg uupp tthhee nneettwwoorrkk + + + + This page describes how to set up the various transports and to + provide a UML instance with network access to the host, other machines + on the local net, and the rest of the net. + + + As of 2.4.5, UML networking has been completely redone to make it much + easier to set up, fix bugs, and add new features. + + + There is a new helper, uml_net, which does the host setup that + requires root privileges. + + + There are currently five transport types available for a UML virtual + machine to exchange packets with other hosts: + + +o ethertap + + +o TUN/TAP + + +o Multicast + + +o a switch daemon + + +o slip + + +o slirp + + +o pcap + + The TUN/TAP, ethertap, slip, and slirp transports allow a UML + instance to exchange packets with the host. They may be directed + to the host or the host may just act as a router to provide access + to other physical or virtual machines. + + + The pcap transport is a synthetic read-only interface, using the + libpcap binary to collect packets from interfaces on the host and + filter them. This is useful for building preconfigured traffic + monitors or sniffers. + + + The daemon and multicast transports provide a completely virtual + network to other virtual machines. This network is completely + disconnected from the physical network unless one of the virtual + machines on it is acting as a gateway. + + + With so many host transports, which one should you use? Here's when + you should use each one: + + +o ethertap - if you want access to the host networking and it is + running 2.2 + + +o TUN/TAP - if you want access to the host networking and it is + running 2.4. Also, the TUN/TAP transport is able to use a + preconfigured device, allowing it to avoid using the setuid uml_net + helper, which is a security advantage. + + +o Multicast - if you want a purely virtual network and you don't want + to set up anything but the UML + + +o a switch daemon - if you want a purely virtual network and you + don't mind running the daemon in order to get somewhat better + performance + + +o slip - there is no particular reason to run the slip backend unless + ethertap and TUN/TAP are just not available for some reason + + +o slirp - if you don't have root access on the host to setup + networking, or if you don't want to allocate an IP to your UML + + +o pcap - not much use for actual network connectivity, but great for + monitoring traffic on the host + + Ethertap is available on 2.4 and works fine. TUN/TAP is preferred + to it because it has better performance and ethertap is officially + considered obsolete in 2.4. Also, the root helper only needs to + run occasionally for TUN/TAP, rather than handling every packet, as + it does with ethertap. This is a slight security advantage since + it provides fewer opportunities for a nasty UML user to somehow + exploit the helper's root privileges. + + + 66..11.. GGeenneerraall sseettuupp + + First, you must have the virtual network enabled in your UML. If are + running a prebuilt kernel from this site, everything is already + enabled. If you build the kernel yourself, under the "Network device + support" menu, enable "Network device support", and then the three + transports. + + + The next step is to provide a network device to the virtual machine. + This is done by describing it on the kernel command line. + + The general format is + + + eth <n> = <transport> , <transport args> + + + + + For example, a virtual ethernet device may be attached to a host + ethertap device as follows: + + + eth0=ethertap,tap0,fe:fd:0:0:0:1,192.168.0.254 + + + + + This sets up eth0 inside the virtual machine to attach itself to the + host /dev/tap0, assigns it an ethernet address, and assigns the host + tap0 interface an IP address. + + + + Note that the IP address you assign to the host end of the tap device + must be different than the IP you assign to the eth device inside UML. + If you are short on IPs and don't want to consume two per UML, then + you can reuse the host's eth IP address for the host ends of the tap + devices. Internally, the UMLs must still get unique IPs for their eth + devices. You can also give the UMLs non-routable IPs (192.168.x.x or + 10.x.x.x) and have the host masquerade them. This will let outgoing + connections work, but incoming connections won't without more work, + such as port forwarding from the host. + Also note that when you configure the host side of an interface, it is + only acting as a gateway. It will respond to pings sent to it + locally, but is not useful to do that since it's a host interface. + You are not talking to the UML when you ping that interface and get a + response. + + + You can also add devices to a UML and remove them at runtime. See the + ``The Management Console'' page for details. + + + The sections below describe this in more detail. + + + Once you've decided how you're going to set up the devices, you boot + UML, log in, configure the UML side of the devices, and set up routes + to the outside world. At that point, you will be able to talk to any + other machines, physical or virtual, on the net. + + + If ifconfig inside UML fails and the network refuses to come up, run + tell you what went wrong. + + + + 66..22.. UUsseerrssppaaccee ddaaeemmoonnss + + You will likely need the setuid helper, or the switch daemon, or both. + They are both installed with the RPM and deb, so if you've installed + either, you can skip the rest of this section. + + + If not, then you need to check them out of CVS, build them, and + install them. The helper is uml_net, in CVS /tools/uml_net, and the + daemon is uml_switch, in CVS /tools/uml_router. They are both built + with a plain 'make'. Both need to be installed in a directory that's + in your path - /usr/bin is recommend. On top of that, uml_net needs + to be setuid root. + + + + 66..33.. SSppeecciiffyyiinngg eetthheerrnneett aaddddrreesssseess + + Below, you will see that the TUN/TAP, ethertap, and daemon interfaces + allow you to specify hardware addresses for the virtual ethernet + devices. This is generally not necessary. If you don't have a + specific reason to do it, you probably shouldn't. If one is not + specified on the command line, the driver will assign one based on the + device IP address. It will provide the address fe:fd:nn:nn:nn:nn + where nn.nn.nn.nn is the device IP address. This is nearly always + sufficient to guarantee a unique hardware address for the device. A + couple of exceptions are: + + +o Another set of virtual ethernet devices are on the same network and + they are assigned hardware addresses using a different scheme which + may conflict with the UML IP address-based scheme + + +o You aren't going to use the device for IP networking, so you don't + assign the device an IP address + + If you let the driver provide the hardware address, you should make + sure that the device IP address is known before the interface is + brought up. So, inside UML, this will guarantee that: + + + + UML# + ifconfig eth0 192.168.0.250 up + + + + + If you decide to assign the hardware address yourself, make sure that + the first byte of the address is even. Addresses with an odd first + byte are broadcast addresses, which you don't want assigned to a + device. + + + + 66..44.. UUMMLL iinntteerrffaaccee sseettuupp + + Once the network devices have been described on the command line, you + should boot UML and log in. + + + The first thing to do is bring the interface up: + + + UML# ifconfig ethn ip-address up + + + + + You should be able to ping the host at this point. + + + To reach the rest of the world, you should set a default route to the + host: + + + UML# route add default gw host ip + + + + + Again, with host ip of 192.168.0.4: + + + UML# route add default gw 192.168.0.4 + + + + + This page used to recommend setting a network route to your local net. + This is wrong, because it will cause UML to try to figure out hardware + addresses of the local machines by arping on the interface to the + host. Since that interface is basically a single strand of ethernet + with two nodes on it (UML and the host) and arp requests don't cross + networks, they will fail to elicit any responses. So, what you want + is for UML to just blindly throw all packets at the host and let it + figure out what to do with them, which is what leaving out the network + route and adding the default route does. + + + Note: If you can't communicate with other hosts on your physical + ethernet, it's probably because of a network route that's + automatically set up. If you run 'route -n' and see a route that + looks like this: + + + + + Destination Gateway Genmask Flags Metric Ref Use Iface + 192.168.0.0 0.0.0.0 255.255.255.0 U 0 0 0 eth0 + + + + + with a mask that's not 255.255.255.255, then replace it with a route + to your host: + + + UML# + route del -net 192.168.0.0 dev eth0 netmask 255.255.255.0 + + + + + + + UML# + route add -host 192.168.0.4 dev eth0 + + + + + This, plus the default route to the host, will allow UML to exchange + packets with any machine on your ethernet. + + + + 66..55.. MMuullttiiccaasstt + + The simplest way to set up a virtual network between multiple UMLs is + to use the mcast transport. This was written by Harald Welte and is + present in UML version 2.4.5-5um and later. Your system must have + multicast enabled in the kernel and there must be a multicast-capable + network device on the host. Normally, this is eth0, but if there is + no ethernet card on the host, then you will likely get strange error + messages when you bring the device up inside UML. + + + To use it, run two UMLs with + + + eth0=mcast + + + + + on their command lines. Log in, configure the ethernet device in each + machine with different IP addresses: + + + UML1# ifconfig eth0 192.168.0.254 + + + + + + + UML2# ifconfig eth0 192.168.0.253 + + + + + and they should be able to talk to each other. + + The full set of command line options for this transport are + + + + ethn=mcast,ethernet address,multicast + address,multicast port,ttl + + + + + Harald's original README is here <http://user-mode-linux.source- + forge.net/> and explains these in detail, as well as + some other issues. + + + + 66..66.. TTUUNN//TTAAPP wwiitthh tthhee uummll__nneett hheellppeerr + + TUN/TAP is the preferred mechanism on 2.4 to exchange packets with the + host. The TUN/TAP backend has been in UML since 2.4.9-3um. + + + The easiest way to get up and running is to let the setuid uml_net + helper do the host setup for you. This involves insmod-ing the tun.o + module if necessary, configuring the device, and setting up IP + forwarding, routing, and proxy arp. If you are new to UML networking, + do this first. If you're concerned about the security implications of + the setuid helper, use it to get up and running, then read the next + section to see how to have UML use a preconfigured tap device, which + avoids the use of uml_net. + + + If you specify an IP address for the host side of the device, the + uml_net helper will do all necessary setup on the host - the only + requirement is that TUN/TAP be available, either built in to the host + kernel or as the tun.o module. + + The format of the command line switch to attach a device to a TUN/TAP + device is + + + eth <n> =tuntap,,, <IP address> + + + + + For example, this argument will attach the UML's eth0 to the next + available tap device and assign an ethernet address to it based on its + IP address + + + eth0=tuntap,,,192.168.0.254 + + + + + + + Note that the IP address that must be used for the eth device inside + UML is fixed by the routing and proxy arp that is set up on the + TUN/TAP device on the host. You can use a different one, but it won't + work because reply packets won't reach the UML. This is a feature. + It prevents a nasty UML user from doing things like setting the UML IP + to the same as the network's nameserver or mail server. + + + There are a couple potential problems with running the TUN/TAP + transport on a 2.4 host kernel + + +o TUN/TAP seems not to work on 2.4.3 and earlier. Upgrade the host + kernel or use the ethertap transport. + + +o With an upgraded kernel, TUN/TAP may fail with + + + File descriptor in bad state + + + + + This is due to a header mismatch between the upgraded kernel and the + kernel that was originally installed on the machine. The fix is to + make sure that /usr/src/linux points to the headers for the running + kernel. + + These were pointed out by Tim Robinson <timro at trkr dot net> in + <http://www.geocrawler.com/> name="this uml- + user post"> . + + + + 66..77.. TTUUNN//TTAAPP wwiitthh aa pprreeccoonnffiigguurreedd ttaapp ddeevviiccee + + If you prefer not to have UML use uml_net (which is somewhat + insecure), with UML 2.4.17-11, you can set up a TUN/TAP device + beforehand. The setup needs to be done as root, but once that's done, + there is no need for root assistance. Setting up the device is done + as follows: + + +o Create the device with tunctl (available from the UML utilities + tarball) + + + + + host# tunctl -u uid + + + + + where uid is the user id or username that UML will be run as. This + will tell you what device was created. + + +o Configure the device IP (change IP addresses and device name to + suit) + + + + + host# ifconfig tap0 192.168.0.254 up + + + + + + +o Set up routing and arping if desired - this is my recipe, there are + other ways of doing the same thing + + + host# + bash -c 'echo 1 > /proc/sys/net/ipv4/ip_forward' + + host# + route add -host 192.168.0.253 dev tap0 + + + + + + + host# + bash -c 'echo 1 > /proc/sys/net/ipv4/conf/tap0/proxy_arp' + + + + + + + host# + arp -Ds 192.168.0.253 eth0 pub + + + + + Note that this must be done every time the host boots - this configu- + ration is not stored across host reboots. So, it's probably a good + idea to stick it in an rc file. An even better idea would be a little + utility which reads the information from a config file and sets up + devices at boot time. + + +o Rather than using up two IPs and ARPing for one of them, you can + also provide direct access to your LAN by the UML by using a + bridge. + + + host# + brctl addbr br0 + + + + + + + host# + ifconfig eth0 0.0.0.0 promisc up + + + + + + + host# + ifconfig tap0 0.0.0.0 promisc up + + + + + + + host# + ifconfig br0 192.168.0.1 netmask 255.255.255.0 up + + + + + + + + host# + brctl stp br0 off + + + + + + + host# + brctl setfd br0 1 + + + + + + + host# + brctl sethello br0 1 + + + + + + + host# + brctl addif br0 eth0 + + + + + + + host# + brctl addif br0 tap0 + + + + + Note that 'br0' should be setup using ifconfig with the existing IP + address of eth0, as eth0 no longer has its own IP. + + +o + + + Also, the /dev/net/tun device must be writable by the user running + UML in order for the UML to use the device that's been configured + for it. The simplest thing to do is + + + host# chmod 666 /dev/net/tun + + + + + Making it world-writable looks bad, but it seems not to be + exploitable as a security hole. However, it does allow anyone to cre- + ate useless tap devices (useless because they can't configure them), + which is a DOS attack. A somewhat more secure alternative would to be + to create a group containing all the users who have preconfigured tap + devices and chgrp /dev/net/tun to that group with mode 664 or 660. + + + +o Once the device is set up, run UML with 'eth0=tuntap,device name' + (i.e. 'eth0=tuntap,tap0') on the command line (or do it with the + mconsole config command). + + +o Bring the eth device up in UML and you're in business. + + If you don't want that tap device any more, you can make it non- + persistent with + + + host# tunctl -d tap device + + + + + Finally, tunctl has a -b (for brief mode) switch which causes it to + output only the name of the tap device it created. This makes it + suitable for capture by a script: + + + host# TAP=`tunctl -u 1000 -b` + + + + + + + 66..88.. EEtthheerrttaapp + + Ethertap is the general mechanism on 2.2 for userspace processes to + exchange packets with the kernel. + + + + To use this transport, you need to describe the virtual network device + on the UML command line. The general format for this is + + + eth <n> =ethertap, <device> , <ethernet address> , <tap IP address> + + + + + So, the previous example + + + eth0=ethertap,tap0,fe:fd:0:0:0:1,192.168.0.254 + + + + + attaches the UML eth0 device to the host /dev/tap0, assigns it the + ethernet address fe:fd:0:0:0:1, and assigns the IP address + 192.168.0.254 to the tap device. + + + + The tap device is mandatory, but the others are optional. If the + ethernet address is omitted, one will be assigned to it. + + + The presence of the tap IP address will cause the helper to run and do + whatever host setup is needed to allow the virtual machine to + communicate with the outside world. If you're not sure you know what + you're doing, this is the way to go. + + + If it is absent, then you must configure the tap device and whatever + arping and routing you will need on the host. However, even in this + case, the uml_net helper still needs to be in your path and it must be + setuid root if you're not running UML as root. This is because the + tap device doesn't support SIGIO, which UML needs in order to use + something as a source of input. So, the helper is used as a + convenient asynchronous IO thread. + + If you're using the uml_net helper, you can ignore the following host + setup - uml_net will do it for you. You just need to make sure you + have ethertap available, either built in to the host kernel or + available as a module. + + + If you want to set things up yourself, you need to make sure that the + appropriate /dev entry exists. If it doesn't, become root and create + it as follows: + + + mknod /dev/tap <minor> c 36 <minor> + 16 + + + + + For example, this is how to create /dev/tap0: + + + mknod /dev/tap0 c 36 0 + 16 + + + + + You also need to make sure that the host kernel has ethertap support. + If ethertap is enabled as a module, you apparently need to insmod + ethertap once for each ethertap device you want to enable. So, + + + host# + insmod ethertap + + + + + will give you the tap0 interface. To get the tap1 interface, you need + to run + + + host# + insmod ethertap unit=1 -o ethertap1 + + + + + + + + 66..99.. TThhee sswwiittcchh ddaaeemmoonn + + NNoottee: This is the daemon formerly known as uml_router, but which was + renamed so the network weenies of the world would stop growling at me. + + + The switch daemon, uml_switch, provides a mechanism for creating a + totally virtual network. By default, it provides no connection to the + host network (but see -tap, below). + + + The first thing you need to do is run the daemon. Running it with no + arguments will make it listen on a default pair of unix domain + sockets. + + + If you want it to listen on a different pair of sockets, use + + + -unix control socket data socket + + + + + + If you want it to act as a hub rather than a switch, use + + + -hub + + + + + + If you want the switch to be connected to host networking (allowing + the umls to get access to the outside world through the host), use + + + -tap tap0 + + + + + + Note that the tap device must be preconfigured (see "TUN/TAP with a + preconfigured tap device", above). If you're using a different tap + device than tap0, specify that instead of tap0. + + + uml_switch can be backgrounded as follows + + + host% + uml_switch [ options ] < /dev/null > /dev/null + + + + + The reason it doesn't background by default is that it listens to + stdin for EOF. When it sees that, it exits. + + + The general format of the kernel command line switch is + + + + ethn=daemon,ethernet address,socket + type,control socket,data socket + + + + + You can leave off everything except the 'daemon'. You only need to + specify the ethernet address if the one that will be assigned to it + isn't acceptable for some reason. The rest of the arguments describe + how to communicate with the daemon. You should only specify them if + you told the daemon to use different sockets than the default. So, if + you ran the daemon with no arguments, running the UML on the same + machine with + eth0=daemon + + + + + will cause the eth0 driver to attach itself to the daemon correctly. + + + + 66..1100.. SSlliipp + + Slip is another, less general, mechanism for a process to communicate + with the host networking. In contrast to the ethertap interface, + which exchanges ethernet frames with the host and can be used to + transport any higher-level protocol, it can only be used to transport + IP. + + + The general format of the command line switch is + + + + ethn=slip,slip IP + + + + + The slip IP argument is the IP address that will be assigned to the + host end of the slip device. If it is specified, the helper will run + and will set up the host so that the virtual machine can reach it and + the rest of the network. + + + There are some oddities with this interface that you should be aware + of. You should only specify one slip device on a given virtual + machine, and its name inside UML will be 'umn', not 'eth0' or whatever + you specified on the command line. These problems will be fixed at + some point. + + + + 66..1111.. SSlliirrpp + + slirp uses an external program, usually /usr/bin/slirp, to provide IP + only networking connectivity through the host. This is similar to IP + masquerading with a firewall, although the translation is performed in + user-space, rather than by the kernel. As slirp does not set up any + interfaces on the host, or changes routing, slirp does not require + root access or setuid binaries on the host. + + + The general format of the command line switch for slirp is: + + + + ethn=slirp,ethernet address,slirp path + + + + + The ethernet address is optional, as UML will set up the interface + with an ethernet address based upon the initial IP address of the + interface. The slirp path is generally /usr/bin/slirp, although it + will depend on distribution. + + + The slirp program can have a number of options passed to the command + line and we can't add them to the UML command line, as they will be + parsed incorrectly. Instead, a wrapper shell script can be written or + the options inserted into the /.slirprc file. More information on + all of the slirp options can be found in its man pages. + + + The eth0 interface on UML should be set up with the IP 10.2.0.15, + although you can use anything as long as it is not used by a network + you will be connecting to. The default route on UML should be set to + use + + + UML# + route add default dev eth0 + + + + + slirp provides a number of useful IP addresses which can be used by + UML, such as 10.0.2.3 which is an alias for the DNS server specified + in /etc/resolv.conf on the host or the IP given in the 'dns' option + for slirp. + + + Even with a baudrate setting higher than 115200, the slirp connection + is limited to 115200. If you need it to go faster, the slirp binary + needs to be compiled with FULL_BOLT defined in config.h. + + + + 66..1122.. ppccaapp + + The pcap transport is attached to a UML ethernet device on the command + line or with uml_mconsole with the following syntax: + + + + ethn=pcap,host interface,filter + expression,option1,option2 + + + + + The expression and options are optional. + + + The interface is whatever network device on the host you want to + sniff. The expression is a pcap filter expression, which is also what + tcpdump uses, so if you know how to specify tcpdump filters, you will + use the same expressions here. The options are up to two of + 'promisc', control whether pcap puts the host interface into + promiscuous mode. 'optimize' and 'nooptimize' control whether the pcap + expression optimizer is used. + + + Example: + + + + eth0=pcap,eth0,tcp + + eth1=pcap,eth0,!tcp + + + + will cause the UML eth0 to emit all tcp packets on the host eth0 and + the UML eth1 to emit all non-tcp packets on the host eth0. + + + + 66..1133.. SSeettttiinngg uupp tthhee hhoosstt yyoouurrsseellff + + If you don't specify an address for the host side of the ethertap or + slip device, UML won't do any setup on the host. So this is what is + needed to get things working (the examples use a host-side IP of + 192.168.0.251 and a UML-side IP of 192.168.0.250 - adjust to suit your + own network): + + +o The device needs to be configured with its IP address. Tap devices + are also configured with an mtu of 1484. Slip devices are + configured with a point-to-point address pointing at the UML ip + address. + + + host# ifconfig tap0 arp mtu 1484 192.168.0.251 up + + + + + + + host# + ifconfig sl0 192.168.0.251 pointopoint 192.168.0.250 up + + + + + + +o If a tap device is being set up, a route is set to the UML IP. + + + UML# route add -host 192.168.0.250 gw 192.168.0.251 + + + + + + +o To allow other hosts on your network to see the virtual machine, + proxy arp is set up for it. + + + host# arp -Ds 192.168.0.250 eth0 pub + + + + + + +o Finally, the host is set up to route packets. + + + host# echo 1 > /proc/sys/net/ipv4/ip_forward + + + + + + + + + + + 77.. SShhaarriinngg FFiilleessyysstteemmss bbeettwweeeenn VViirrttuuaall MMaacchhiinneess + + + + + 77..11.. AA wwaarrnniinngg + + Don't attempt to share filesystems simply by booting two UMLs from the + same file. That's the same thing as booting two physical machines + from a shared disk. It will result in filesystem corruption. + + + + 77..22.. UUssiinngg llaayyeerreedd bblloocckk ddeevviicceess + + The way to share a filesystem between two virtual machines is to use + the copy-on-write (COW) layering capability of the ubd block driver. + As of 2.4.6-2um, the driver supports layering a read-write private + device over a read-only shared device. A machine's writes are stored + in the private device, while reads come from either device - the + private one if the requested block is valid in it, the shared one if + not. Using this scheme, the majority of data which is unchanged is + shared between an arbitrary number of virtual machines, each of which + has a much smaller file containing the changes that it has made. With + a large number of UMLs booting from a large root filesystem, this + leads to a huge disk space saving. It will also help performance, + since the host will be able to cache the shared data using a much + smaller amount of memory, so UML disk requests will be served from the + host's memory rather than its disks. + + + + + To add a copy-on-write layer to an existing block device file, simply + add the name of the COW file to the appropriate ubd switch: + + + ubd0=root_fs_cow,root_fs_debian_22 + + + + + where 'root_fs_cow' is the private COW file and 'root_fs_debian_22' is + the existing shared filesystem. The COW file need not exist. If it + doesn't, the driver will create and initialize it. Once the COW file + has been initialized, it can be used on its own on the command line: + + + ubd0=root_fs_cow + + + + + The name of the backing file is stored in the COW file header, so it + would be redundant to continue specifying it on the command line. + + + + 77..33.. NNoottee!! + + When checking the size of the COW file in order to see the gobs of + space that you're saving, make sure you use 'ls -ls' to see the actual + disk consumption rather than the length of the file. The COW file is + sparse, so the length will be very different from the disk usage. + Here is a 'ls -l' of a COW file and backing file from one boot and + shutdown: + host% ls -l cow.debian debian2.2 + -rw-r--r-- 1 jdike jdike 492504064 Aug 6 21:16 cow.debian + -rwxrw-rw- 1 jdike jdike 537919488 Aug 6 20:42 debian2.2 + + + + + Doesn't look like much saved space, does it? Well, here's 'ls -ls': + + + host% ls -ls cow.debian debian2.2 + 880 -rw-r--r-- 1 jdike jdike 492504064 Aug 6 21:16 cow.debian + 525832 -rwxrw-rw- 1 jdike jdike 537919488 Aug 6 20:42 debian2.2 + + + + + Now, you can see that the COW file has less than a meg of disk, rather + than 492 meg. + + + + 77..44.. AAnnootthheerr wwaarrnniinngg + + Once a filesystem is being used as a readonly backing file for a COW + file, do not boot directly from it or modify it in any way. Doing so + will invalidate any COW files that are using it. The mtime and size + of the backing file are stored in the COW file header at its creation, + and they must continue to match. If they don't, the driver will + refuse to use the COW file. + + + + + If you attempt to evade this restriction by changing either the + backing file or the COW header by hand, you will get a corrupted + filesystem. + + + + + Among other things, this means that upgrading the distribution in a + backing file and expecting that all of the COW files using it will see + the upgrade will not work. + + + + + 77..55.. uummll__mmoooo :: MMeerrggiinngg aa CCOOWW ffiillee wwiitthh iittss bbaacckkiinngg ffiillee + + Depending on how you use UML and COW devices, it may be advisable to + merge the changes in the COW file into the backing file every once in + a while. + + + + + The utility that does this is uml_moo. Its usage is + + + host% uml_moo COW file new backing file + + + + + There's no need to specify the backing file since that information is + already in the COW file header. If you're paranoid, boot the new + merged file, and if you're happy with it, move it over the old backing + file. + + + + + uml_moo creates a new backing file by default as a safety measure. It + also has a destructive merge option which will merge the COW file + directly into its current backing file. This is really only usable + when the backing file only has one COW file associated with it. If + there are multiple COWs associated with a backing file, a -d merge of + one of them will invalidate all of the others. However, it is + convenient if you're short of disk space, and it should also be + noticeably faster than a non-destructive merge. + + + + + uml_moo is installed with the UML deb and RPM. If you didn't install + UML from one of those packages, you can also get it from the UML + utilities <http://user-mode-linux.sourceforge.net/ + utilities> tar file in tools/moo. + + + + + + + + + 88.. CCrreeaattiinngg ffiilleessyysstteemmss + + + You may want to create and mount new UML filesystems, either because + your root filesystem isn't large enough or because you want to use a + filesystem other than ext2. + + + This was written on the occasion of reiserfs being included in the + 2.4.1 kernel pool, and therefore the 2.4.1 UML, so the examples will + talk about reiserfs. This information is generic, and the examples + should be easy to translate to the filesystem of your choice. + + + 88..11.. CCrreeaattee tthhee ffiilleessyysstteemm ffiillee + + dd is your friend. All you need to do is tell dd to create an empty + file of the appropriate size. I usually make it sparse to save time + and to avoid allocating disk space until it's actually used. For + example, the following command will create a sparse 100 meg file full + of zeroes. + + + host% + dd if=/dev/zero of=new_filesystem seek=100 count=1 bs=1M + + + + + + + 88..22.. AAssssiiggnn tthhee ffiillee ttoo aa UUMMLL ddeevviiccee + + Add an argument like the following to the UML command line: + + ubd4=new_filesystem + + + + + making sure that you use an unassigned ubd device number. + + + + 88..33.. CCrreeaattiinngg aanndd mmoouunnttiinngg tthhee ffiilleessyysstteemm + + Make sure that the filesystem is available, either by being built into + the kernel, or available as a module, then boot up UML and log in. If + the root filesystem doesn't have the filesystem utilities (mkfs, fsck, + etc), then get them into UML by way of the net or hostfs. + + + Make the new filesystem on the device assigned to the new file: + + + host# mkreiserfs /dev/ubd/4 + + + <----------- MKREISERFSv2 -----------> + + ReiserFS version 3.6.25 + Block size 4096 bytes + Block count 25856 + Used blocks 8212 + Journal - 8192 blocks (18-8209), journal header is in block 8210 + Bitmaps: 17 + Root block 8211 + Hash function "r5" + ATTENTION: ALL DATA WILL BE LOST ON '/dev/ubd/4'! (y/n)y + journal size 8192 (from 18) + Initializing journal - 0%....20%....40%....60%....80%....100% + Syncing..done. + + + + + Now, mount it: + + + UML# + mount /dev/ubd/4 /mnt + + + + + and you're in business. + + + + + + + + + + 99.. HHoosstt ffiillee aacccceessss + + + If you want to access files on the host machine from inside UML, you + can treat it as a separate machine and either nfs mount directories + from the host or copy files into the virtual machine with scp or rcp. + However, since UML is running on the host, it can access those + files just like any other process and make them available inside the + virtual machine without needing to use the network. + + + This is now possible with the hostfs virtual filesystem. With it, you + can mount a host directory into the UML filesystem and access the + files contained in it just as you would on the host. + + + 99..11.. UUssiinngg hhoossttffss + + To begin with, make sure that hostfs is available inside the virtual + machine with + + + UML# cat /proc/filesystems + + + + . hostfs should be listed. If it's not, either rebuild the kernel + with hostfs configured into it or make sure that hostfs is built as a + module and available inside the virtual machine, and insmod it. + + + Now all you need to do is run mount: + + + UML# mount none /mnt/host -t hostfs + + + + + will mount the host's / on the virtual machine's /mnt/host. + + + If you don't want to mount the host root directory, then you can + specify a subdirectory to mount with the -o switch to mount: + + + UML# mount none /mnt/home -t hostfs -o /home + + + + + will mount the hosts's /home on the virtual machine's /mnt/home. + + + + 99..22.. hhoossttffss aass tthhee rroooott ffiilleessyysstteemm + + It's possible to boot from a directory hierarchy on the host using + hostfs rather than using the standard filesystem in a file. + + To start, you need that hierarchy. The easiest way is to loop mount + an existing root_fs file: + + + host# mount root_fs uml_root_dir -o loop + + + + + You need to change the filesystem type of / in etc/fstab to be + 'hostfs', so that line looks like this: + + /dev/ubd/0 / hostfs defaults 1 1 + + + + + Then you need to chown to yourself all the files in that directory + that are owned by root. This worked for me: + + + host# find . -uid 0 -exec chown jdike {} \; + + + + + Next, make sure that your UML kernel has hostfs compiled in, not as a + module. Then run UML with the boot device pointing at that directory: + + + ubd0=/path/to/uml/root/directory + + + + + UML should then boot as it does normally. + + + 99..33.. BBuuiillddiinngg hhoossttffss + + If you need to build hostfs because it's not in your kernel, you have + two choices: + + + + +o Compiling hostfs into the kernel: + + + Reconfigure the kernel and set the 'Host filesystem' option under + + + +o Compiling hostfs as a module: + + + Reconfigure the kernel and set the 'Host filesystem' option under + be in arch/um/fs/hostfs/hostfs.o. Install that in + /lib/modules/`uname -r`/fs in the virtual machine, boot it up, and + + + UML# insmod hostfs + + + + + + + + + + + + + 1100.. TThhee MMaannaaggeemmeenntt CCoonnssoollee + + + + The UML management console is a low-level interface to the kernel, + somewhat like the i386 SysRq interface. Since there is a full-blown + operating system under UML, there is much greater flexibility possible + than with the SysRq mechanism. + + + There are a number of things you can do with the mconsole interface: + + +o get the kernel version + + +o add and remove devices + + +o halt or reboot the machine + + +o Send SysRq commands + + +o Pause and resume the UML + + + You need the mconsole client (uml_mconsole) which is present in CVS + (/tools/mconsole) in 2.4.5-9um and later, and will be in the RPM in + 2.4.6. + + + You also need CONFIG_MCONSOLE (under 'General Setup') enabled in UML. + When you boot UML, you'll see a line like: + + + mconsole initialized on /home/jdike/.uml/umlNJ32yL/mconsole + + + + + If you specify a unique machine id one the UML command line, i.e. + + + umid=debian + + + + + you'll see this + + + mconsole initialized on /home/jdike/.uml/debian/mconsole + + + + + That file is the socket that uml_mconsole will use to communicate with + UML. Run it with either the umid or the full path as its argument: + + + host% uml_mconsole debian + + + + + or + + + host% uml_mconsole /home/jdike/.uml/debian/mconsole + + + + + You'll get a prompt, at which you can run one of these commands: + + +o version + + +o halt + + +o reboot + + +o config + + +o remove + + +o sysrq + + +o help + + +o cad + + +o stop + + +o go + + + 1100..11.. vveerrssiioonn + + This takes no arguments. It prints the UML version. + + + (mconsole) version + OK Linux usermode 2.4.5-9um #1 Wed Jun 20 22:47:08 EDT 2001 i686 + + + + + There are a couple actual uses for this. It's a simple no-op which + can be used to check that a UML is running. It's also a way of + sending an interrupt to the UML. This is sometimes useful on SMP + hosts, where there's a bug which causes signals to UML to be lost, + often causing it to appear to hang. Sending such a UML the mconsole + version command is a good way to 'wake it up' before networking has + been enabled, as it does not do anything to the function of the UML. + + + + 1100..22.. hhaalltt aanndd rreebboooott + + These take no arguments. They shut the machine down immediately, with + no syncing of disks and no clean shutdown of userspace. So, they are + pretty close to crashing the machine. + + + (mconsole) halt + OK + + + + + + + 1100..33.. ccoonnffiigg + + "config" adds a new device to the virtual machine. Currently the ubd + and network drivers support this. It takes one argument, which is the + device to add, with the same syntax as the kernel command line. + + + + + (mconsole) + config ubd3=/home/jdike/incoming/roots/root_fs_debian22 + + OK + (mconsole) config eth1=mcast + OK + + + + + + + 1100..44.. rreemmoovvee + + "remove" deletes a device from the system. Its argument is just the + name of the device to be removed. The device must be idle in whatever + sense the driver considers necessary. In the case of the ubd driver, + the removed block device must not be mounted, swapped on, or otherwise + open, and in the case of the network driver, the device must be down. + + + (mconsole) remove ubd3 + OK + (mconsole) remove eth1 + OK + + + + + + + 1100..55.. ssyyssrrqq + + This takes one argument, which is a single letter. It calls the + generic kernel's SysRq driver, which does whatever is called for by + that argument. See the SysRq documentation in Documentation/sysrq.txt + in your favorite kernel tree to see what letters are valid and what + they do. + + + + 1100..66.. hheellpp + + "help" returns a string listing the valid commands and what each one + does. + + + + 1100..77.. ccaadd + + This invokes the Ctl-Alt-Del action on init. What exactly this ends + up doing is up to /etc/inittab. Normally, it reboots the machine. + With UML, this is usually not desired, so if a halt would be better, + then find the section of inittab that looks like this + + + # What to do when CTRL-ALT-DEL is pressed. + ca:12345:ctrlaltdel:/sbin/shutdown -t1 -a -r now + + + + + and change the command to halt. + + + + 1100..88.. ssttoopp + + This puts the UML in a loop reading mconsole requests until a 'go' + mconsole command is received. This is very useful for making backups + of UML filesystems, as the UML can be stopped, then synced via 'sysrq + s', so that everything is written to the filesystem. You can then copy + the filesystem and then send the UML 'go' via mconsole. + + + Note that a UML running with more than one CPU will have problems + after you send the 'stop' command, as only one CPU will be held in a + mconsole loop and all others will continue as normal. This is a bug, + and will be fixed. + + + + 1100..99.. ggoo + + This resumes a UML after being paused by a 'stop' command. Note that + when the UML has resumed, TCP connections may have timed out and if + the UML is paused for a long period of time, crond might go a little + crazy, running all the jobs it didn't do earlier. + + + + + + + + + 1111.. KKeerrnneell ddeebbuuggggiinngg + + + NNoottee:: The interface that makes debugging, as described here, possible + is present in 2.4.0-test6 kernels and later. + + + Since the user-mode kernel runs as a normal Linux process, it is + possible to debug it with gdb almost like any other process. It is + slightly different because the kernel's threads are already being + ptraced for system call interception, so gdb can't ptrace them. + However, a mechanism has been added to work around that problem. + + + In order to debug the kernel, you need build it from source. See + ``Compiling the kernel and modules'' for information on doing that. + Make sure that you enable CONFIG_DEBUGSYM and CONFIG_PT_PROXY during + the config. These will compile the kernel with -g, and enable the + ptrace proxy so that gdb works with UML, respectively. + + + + + 1111..11.. SSttaarrttiinngg tthhee kkeerrnneell uunnddeerr ggddbb + + You can have the kernel running under the control of gdb from the + beginning by putting 'debug' on the command line. You will get an + xterm with gdb running inside it. The kernel will send some commands + to gdb which will leave it stopped at the beginning of start_kernel. + At this point, you can get things going with 'next', 'step', or + 'cont'. + + + There is a transcript of a debugging session here <debug- + session.html> , with breakpoints being set in the scheduler and in an + interrupt handler. + 1111..22.. EExxaammiinniinngg sslleeeeppiinngg pprroocceesssseess + + Not every bug is evident in the currently running process. Sometimes, + processes hang in the kernel when they shouldn't because they've + deadlocked on a semaphore or something similar. In this case, when + you ^C gdb and get a backtrace, you will see the idle thread, which + isn't very relevant. + + + What you want is the stack of whatever process is sleeping when it + shouldn't be. You need to figure out which process that is, which is + generally fairly easy. Then you need to get its host process id, + which you can do either by looking at ps on the host or at + task.thread.extern_pid in gdb. + + + Now what you do is this: + + +o detach from the current thread + + + (UML gdb) det + + + + + + +o attach to the thread you are interested in + + + (UML gdb) att <host pid> + + + + + + +o look at its stack and anything else of interest + + + (UML gdb) bt + + + + + Note that you can't do anything at this point that requires that a + process execute, e.g. calling a function + + +o when you're done looking at that process, reattach to the current + thread and continue it + + + (UML gdb) + att 1 + + + + + + + (UML gdb) + c + + + + + Here, specifying any pid which is not the process id of a UML thread + will cause gdb to reattach to the current thread. I commonly use 1, + but any other invalid pid would work. + + + + 1111..33.. RRuunnnniinngg dddddd oonn UUMMLL + + ddd works on UML, but requires a special kludge. The process goes + like this: + + +o Start ddd + + + host% ddd linux + + + + + + +o With ps, get the pid of the gdb that ddd started. You can ask the + gdb to tell you, but for some reason that confuses things and + causes a hang. + + +o run UML with 'debug=parent gdb-pid=<pid>' added to the command line + - it will just sit there after you hit return + + +o type 'att 1' to the ddd gdb and you will see something like + + + 0xa013dc51 in __kill () + + + (gdb) + + + + + + +o At this point, type 'c', UML will boot up, and you can use ddd just + as you do on any other process. + + + + 1111..44.. DDeebbuuggggiinngg mmoodduulleess + + gdb has support for debugging code which is dynamically loaded into + the process. This support is what is needed to debug kernel modules + under UML. + + + Using that support is somewhat complicated. You have to tell gdb what + object file you just loaded into UML and where in memory it is. Then, + it can read the symbol table, and figure out where all the symbols are + from the load address that you provided. It gets more interesting + when you load the module again (i.e. after an rmmod). You have to + tell gdb to forget about all its symbols, including the main UML ones + for some reason, then load then all back in again. + + + There's an easy way and a hard way to do this. The easy way is to use + the umlgdb expect script written by Chandan Kudige. It basically + automates the process for you. + + + First, you must tell it where your modules are. There is a list in + the script that looks like this: + set MODULE_PATHS { + "fat" "/usr/src/uml/linux-2.4.18/fs/fat/fat.o" + "isofs" "/usr/src/uml/linux-2.4.18/fs/isofs/isofs.o" + "minix" "/usr/src/uml/linux-2.4.18/fs/minix/minix.o" + } + + + + + You change that to list the names and paths of the modules that you + are going to debug. Then you run it from the toplevel directory of + your UML pool and it basically tells you what to do: + + + + + ******** GDB pid is 21903 ******** + Start UML as: ./linux <kernel switches> debug gdb-pid=21903 + + + + GNU gdb 5.0rh-5 Red Hat Linux 7.1 + Copyright 2001 Free Software Foundation, Inc. + GDB is free software, covered by the GNU General Public License, and you are + welcome to change it and/or distribute copies of it under certain conditions. + Type "show copying" to see the conditions. + There is absolutely no warranty for GDB. Type "show warranty" for details. + This GDB was configured as "i386-redhat-linux"... + (gdb) b sys_init_module + Breakpoint 1 at 0xa0011923: file module.c, line 349. + (gdb) att 1 + + + + + After you run UML and it sits there doing nothing, you hit return at + the 'att 1' and continue it: + + + Attaching to program: /home/jdike/linux/2.4/um/./linux, process 1 + 0xa00f4221 in __kill () + (UML gdb) c + Continuing. + + + + + At this point, you debug normally. When you insmod something, the + expect magic will kick in and you'll see something like: + + + + + + + + + + + + + + + + + + *** Module hostfs loaded *** + Breakpoint 1, sys_init_module (name_user=0x805abb0 "hostfs", + mod_user=0x8070e00) at module.c:349 + 349 char *name, *n_name, *name_tmp = NULL; + (UML gdb) finish + Run till exit from #0 sys_init_module (name_user=0x805abb0 "hostfs", + mod_user=0x8070e00) at module.c:349 + 0xa00e2e23 in execute_syscall (r=0xa8140284) at syscall_kern.c:411 + 411 else res = EXECUTE_SYSCALL(syscall, regs); + Value returned is $1 = 0 + (UML gdb) + p/x (int)module_list + module_list->size_of_struct + + $2 = 0xa9021054 + (UML gdb) symbol-file ./linux + Load new symbol table from "./linux"? (y or n) y + Reading symbols from ./linux... + done. + (UML gdb) + add-symbol-file /home/jdike/linux/2.4/um/arch/um/fs/hostfs/hostfs.o 0xa9021054 + + add symbol table from file "/home/jdike/linux/2.4/um/arch/um/fs/hostfs/hostfs.o" at + .text_addr = 0xa9021054 + (y or n) y + + Reading symbols from /home/jdike/linux/2.4/um/arch/um/fs/hostfs/hostfs.o... + done. + (UML gdb) p *module_list + $1 = {size_of_struct = 84, next = 0xa0178720, name = 0xa9022de0 "hostfs", + size = 9016, uc = {usecount = {counter = 0}, pad = 0}, flags = 1, + nsyms = 57, ndeps = 0, syms = 0xa9023170, deps = 0x0, refs = 0x0, + init = 0xa90221f0 <init_hostfs>, cleanup = 0xa902222c <exit_hostfs>, + ex_table_start = 0x0, ex_table_end = 0x0, persist_start = 0x0, + persist_end = 0x0, can_unload = 0, runsize = 0, kallsyms_start = 0x0, + kallsyms_end = 0x0, + archdata_start = 0x1b855 <Address 0x1b855 out of bounds>, + archdata_end = 0xe5890000 <Address 0xe5890000 out of bounds>, + kernel_data = 0xf689c35d <Address 0xf689c35d out of bounds>} + >> Finished loading symbols for hostfs ... + + + + + That's the easy way. It's highly recommended. The hard way is + described below in case you're interested in what's going on. + + + Boot the kernel under the debugger and load the module with insmod or + modprobe. With gdb, do: + + + (UML gdb) p module_list + + + + + This is a list of modules that have been loaded into the kernel, with + the most recently loaded module first. Normally, the module you want + is at module_list. If it's not, walk down the next links, looking at + the name fields until find the module you want to debug. Take the + address of that structure, and add module.size_of_struct (which in + 2.4.10 kernels is 96 (0x60)) to it. Gdb can make this hard addition + for you :-): + + + + (UML gdb) + printf "%#x\n", (int)module_list module_list->size_of_struct + + + + + The offset from the module start occasionally changes (before 2.4.0, + it was module.size_of_struct + 4), so it's a good idea to check the + init and cleanup addresses once in a while, as describe below. Now + do: + + + (UML gdb) + add-symbol-file /path/to/module/on/host that_address + + + + + Tell gdb you really want to do it, and you're in business. + + + If there's any doubt that you got the offset right, like breakpoints + appear not to work, or they're appearing in the wrong place, you can + check it by looking at the module structure. The init and cleanup + fields should look like: + + + init = 0x588066b0 <init_hostfs>, cleanup = 0x588066c0 <exit_hostfs> + + + + + with no offsets on the symbol names. If the names are right, but they + are offset, then the offset tells you how much you need to add to the + address you gave to add-symbol-file. + + + When you want to load in a new version of the module, you need to get + gdb to forget about the old one. The only way I've found to do that + is to tell gdb to forget about all symbols that it knows about: + + + (UML gdb) symbol-file + + + + + Then reload the symbols from the kernel binary: + + + (UML gdb) symbol-file /path/to/kernel + + + + + and repeat the process above. You'll also need to re-enable break- + points. They were disabled when you dumped all the symbols because + gdb couldn't figure out where they should go. + + + + 1111..55.. AAttttaacchhiinngg ggddbb ttoo tthhee kkeerrnneell + + If you don't have the kernel running under gdb, you can attach gdb to + it later by sending the tracing thread a SIGUSR1. The first line of + the console output identifies its pid: + tracing thread pid = 20093 + + + + + When you send it the signal: + + + host% kill -USR1 20093 + + + + + you will get an xterm with gdb running in it. + + + If you have the mconsole compiled into UML, then the mconsole client + can be used to start gdb: + + + (mconsole) (mconsole) config gdb=xterm + + + + + will fire up an xterm with gdb running in it. + + + + 1111..66.. UUssiinngg aalltteerrnnaattee ddeebbuuggggeerrss + + UML has support for attaching to an already running debugger rather + than starting gdb itself. This is present in CVS as of 17 Apr 2001. + I sent it to Alan for inclusion in the ac tree, and it will be in my + 2.4.4 release. + + + This is useful when gdb is a subprocess of some UI, such as emacs or + ddd. It can also be used to run debuggers other than gdb on UML. + Below is an example of using strace as an alternate debugger. + + + To do this, you need to get the pid of the debugger and pass it in + with the + + + If you are using gdb under some UI, then tell it to 'att 1', and + you'll find yourself attached to UML. + + + If you are using something other than gdb as your debugger, then + you'll need to get it to do the equivalent of 'att 1' if it doesn't do + it automatically. + + + An example of an alternate debugger is strace. You can strace the + actual kernel as follows: + + +o Run the following in a shell + + + host% + sh -c 'echo pid=$$; echo -n hit return; read x; exec strace -p 1 -o strace.out' + + + + +o Run UML with 'debug' and 'gdb-pid=<pid>' with the pid printed out + by the previous command + + +o Hit return in the shell, and UML will start running, and strace + output will start accumulating in the output file. + + Note that this is different from running + + + host% strace ./linux + + + + + That will strace only the main UML thread, the tracing thread, which + doesn't do any of the actual kernel work. It just oversees the vir- + tual machine. In contrast, using strace as described above will show + you the low-level activity of the virtual machine. + + + + + + 1122.. KKeerrnneell ddeebbuuggggiinngg eexxaammpplleess + + 1122..11.. TThhee ccaassee ooff tthhee hhuunngg ffsscckk + + When booting up the kernel, fsck failed, and dropped me into a shell + to fix things up. I ran fsck -y, which hung: + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + Setting hostname uml [ OK ] + Checking root filesystem + /dev/fhd0 was not cleanly unmounted, check forced. + Error reading block 86894 (Attempt to read block from filesystem resulted in short read) while reading indirect blocks of inode 19780. + + /dev/fhd0: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY. + (i.e., without -a or -p options) + [ FAILED ] + + *** An error occurred during the file system check. + *** Dropping you to a shell; the system will reboot + *** when you leave the shell. + Give root password for maintenance + (or type Control-D for normal startup): + + [root@uml /root]# fsck -y /dev/fhd0 + fsck -y /dev/fhd0 + Parallelizing fsck version 1.14 (9-Jan-1999) + e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09 + /dev/fhd0 contains a file system with errors, check forced. + Pass 1: Checking inodes, blocks, and sizes + Error reading block 86894 (Attempt to read block from filesystem resulted in short read) while reading indirect blocks of inode 19780. Ignore error? yes + + Inode 19780, i_blocks is 1548, should be 540. Fix? yes + + Pass 2: Checking directory structure + Error reading block 49405 (Attempt to read block from filesystem resulted in short read). Ignore error? yes + + Directory inode 11858, block 0, offset 0: directory corrupted + Salvage? yes + + Missing '.' in directory inode 11858. + Fix? yes + + Missing '..' in directory inode 11858. + Fix? yes + + + + + + The standard drill in this sort of situation is to fire up gdb on the + signal thread, which, in this case, was pid 1935. In another window, + I run gdb and attach pid 1935. + + + + + ~/linux/2.3.26/um 1016: gdb linux + GNU gdb 4.17.0.11 with Linux support + Copyright 1998 Free Software Foundation, Inc. + GDB is free software, covered by the GNU General Public License, and you are + welcome to change it and/or distribute copies of it under certain conditions. + Type "show copying" to see the conditions. + There is absolutely no warranty for GDB. Type "show warranty" for details. + This GDB was configured as "i386-redhat-linux"... + + (gdb) att 1935 + Attaching to program `/home/dike/linux/2.3.26/um/linux', Pid 1935 + 0x100756d9 in __wait4 () + + + + + + + Let's see what's currently running: + + + + (gdb) p current_task.pid + $1 = 0 + + + + + + It's the idle thread, which means that fsck went to sleep for some + reason and never woke up. + + + Let's guess that the last process in the process list is fsck: + + + + (gdb) p current_task.prev_task.comm + $13 = "fsck.ext2\000\000\000\000\000\000" + + + + + + It is, so let's see what it thinks it's up to: + + + + (gdb) p current_task.prev_task.thread + $14 = {extern_pid = 1980, tracing = 0, want_tracing = 0, forking = 0, + kernel_stack_page = 0, signal_stack = 1342627840, syscall = {id = 4, args = { + 3, 134973440, 1024, 0, 1024}, have_result = 0, result = 50590720}, + request = {op = 2, u = {exec = {ip = 1350467584, sp = 2952789424}, fork = { + regs = {1350467584, 2952789424, 0 <repeats 15 times>}, sigstack = 0, + pid = 0}, switch_to = 0x507e8000, thread = {proc = 0x507e8000, + arg = 0xaffffdb0, flags = 0, new_pid = 0}, input_request = { + op = 1350467584, fd = -1342177872, proc = 0, pid = 0}}}} + + + + + + The interesting things here are the fact that its .thread.syscall.id + is __NR_write (see the big switch in arch/um/kernel/syscall_kern.c or + the defines in include/asm-um/arch/unistd.h), and that it never + returned. Also, its .request.op is OP_SWITCH (see + arch/um/include/user_util.h). These mean that it went into a write, + and, for some reason, called schedule(). + + + The fact that it never returned from write means that its stack should + be fairly interesting. Its pid is 1980 (.thread.extern_pid). That + process is being ptraced by the signal thread, so it must be detached + before gdb can attach it: + + + + + + + + + + + (gdb) call detach(1980) + + Program received signal SIGSEGV, Segmentation fault. + <function called from gdb> + The program being debugged stopped while in a function called from GDB. + When the function (detach) is done executing, GDB will silently + stop (instead of continuing to evaluate the expression containing + the function call). + (gdb) call detach(1980) + $15 = 0 + + + + + + The first detach segfaults for some reason, and the second one + succeeds. + + + Now I detach from the signal thread, attach to the fsck thread, and + look at its stack: + + + (gdb) det + Detaching from program: /home/dike/linux/2.3.26/um/linux Pid 1935 + (gdb) att 1980 + Attaching to program `/home/dike/linux/2.3.26/um/linux', Pid 1980 + 0x10070451 in __kill () + (gdb) bt + #0 0x10070451 in __kill () + #1 0x10068ccd in usr1_pid (pid=1980) at process.c:30 + #2 0x1006a03f in _switch_to (prev=0x50072000, next=0x507e8000) + at process_kern.c:156 + #3 0x1006a052 in switch_to (prev=0x50072000, next=0x507e8000, last=0x50072000) + at process_kern.c:161 + #4 0x10001d12 in schedule () at sched.c:777 + #5 0x1006a744 in __down (sem=0x507d241c) at semaphore.c:71 + #6 0x1006aa10 in __down_failed () at semaphore.c:157 + #7 0x1006c5d8 in segv_handler (sc=0x5006e940) at trap_user.c:174 + #8 0x1006c5ec in kern_segv_handler (sig=11) at trap_user.c:182 + #9 <signal handler called> + #10 0x10155404 in errno () + #11 0x1006c0aa in segv (address=1342179328, is_write=2) at trap_kern.c:50 + #12 0x1006c5d8 in segv_handler (sc=0x5006eaf8) at trap_user.c:174 + #13 0x1006c5ec in kern_segv_handler (sig=11) at trap_user.c:182 + #14 <signal handler called> + #15 0xc0fd in ?? () + #16 0x10016647 in sys_write (fd=3, + buf=0x80b8800 <Address 0x80b8800 out of bounds>, count=1024) + at read_write.c:159 + #17 0x1006d5b3 in execute_syscall (syscall=4, args=0x5006ef08) + at syscall_kern.c:254 + #18 0x1006af87 in really_do_syscall (sig=12) at syscall_user.c:35 + #19 <signal handler called> + #20 0x400dc8b0 in ?? () + + + + + + The interesting things here are : + + +o There are two segfaults on this stack (frames 9 and 14) + + +o The first faulting address (frame 11) is 0x50000800 + + (gdb) p (void *)1342179328 + $16 = (void *) 0x50000800 + + + + + + The initial faulting address is interesting because it is on the idle + thread's stack. I had been seeing the idle thread segfault for no + apparent reason, and the cause looked like stack corruption. In hopes + of catching the culprit in the act, I had turned off all protections + to that stack while the idle thread wasn't running. This apparently + tripped that trap. + + + However, the more immediate problem is that second segfault and I'm + going to concentrate on that. First, I want to see where the fault + happened, so I have to go look at the sigcontent struct in frame 8: + + + + (gdb) up + #1 0x10068ccd in usr1_pid (pid=1980) at process.c:30 + 30 kill(pid, SIGUSR1); + (gdb) + #2 0x1006a03f in _switch_to (prev=0x50072000, next=0x507e8000) + at process_kern.c:156 + 156 usr1_pid(getpid()); + (gdb) + #3 0x1006a052 in switch_to (prev=0x50072000, next=0x507e8000, last=0x50072000) + at process_kern.c:161 + 161 _switch_to(prev, next); + (gdb) + #4 0x10001d12 in schedule () at sched.c:777 + 777 switch_to(prev, next, prev); + (gdb) + #5 0x1006a744 in __down (sem=0x507d241c) at semaphore.c:71 + 71 schedule(); + (gdb) + #6 0x1006aa10 in __down_failed () at semaphore.c:157 + 157 } + (gdb) + #7 0x1006c5d8 in segv_handler (sc=0x5006e940) at trap_user.c:174 + 174 segv(sc->cr2, sc->err & 2); + (gdb) + #8 0x1006c5ec in kern_segv_handler (sig=11) at trap_user.c:182 + 182 segv_handler(sc); + (gdb) p *sc + Cannot access memory at address 0x0. + + + + + That's not very useful, so I'll try a more manual method: + + + (gdb) p *((struct sigcontext *) (&sig + 1)) + $19 = {gs = 0, __gsh = 0, fs = 0, __fsh = 0, es = 43, __esh = 0, ds = 43, + __dsh = 0, edi = 1342179328, esi = 1350378548, ebp = 1342630440, + esp = 1342630420, ebx = 1348150624, edx = 1280, ecx = 0, eax = 0, + trapno = 14, err = 4, eip = 268480945, cs = 35, __csh = 0, eflags = 66118, + esp_at_signal = 1342630420, ss = 43, __ssh = 0, fpstate = 0x0, oldmask = 0, + cr2 = 1280} + + + + The ip is in handle_mm_fault: + + + (gdb) p (void *)268480945 + $20 = (void *) 0x1000b1b1 + (gdb) i sym $20 + handle_mm_fault + 57 in section .text + + + + + + Specifically, it's in pte_alloc: + + + (gdb) i line *$20 + Line 124 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b1b1 <handle_mm_fault+57> + and ends at 0x1000b1b7 <handle_mm_fault+63>. + + + + + + To find where in handle_mm_fault this is, I'll jump forward in the + code until I see an address in that procedure: + + + + (gdb) i line *0x1000b1c0 + Line 126 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b1b7 <handle_mm_fault+63> + and ends at 0x1000b1c3 <handle_mm_fault+75>. + (gdb) i line *0x1000b1d0 + Line 131 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b1d0 <handle_mm_fault+88> + and ends at 0x1000b1da <handle_mm_fault+98>. + (gdb) i line *0x1000b1e0 + Line 61 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b1da <handle_mm_fault+98> + and ends at 0x1000b1e1 <handle_mm_fault+105>. + (gdb) i line *0x1000b1f0 + Line 134 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b1f0 <handle_mm_fault+120> + and ends at 0x1000b200 <handle_mm_fault+136>. + (gdb) i line *0x1000b200 + Line 135 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b200 <handle_mm_fault+136> + and ends at 0x1000b208 <handle_mm_fault+144>. + (gdb) i line *0x1000b210 + Line 139 of "/home/dike/linux/2.3.26/um/include/asm/pgalloc.h" + starts at address 0x1000b210 <handle_mm_fault+152> + and ends at 0x1000b219 <handle_mm_fault+161>. + (gdb) i line *0x1000b220 + Line 1168 of "memory.c" starts at address 0x1000b21e <handle_mm_fault+166> + and ends at 0x1000b222 <handle_mm_fault+170>. + + + + + + Something is apparently wrong with the page tables or vma_structs, so + lets go back to frame 11 and have a look at them: + + + + #11 0x1006c0aa in segv (address=1342179328, is_write=2) at trap_kern.c:50 + 50 handle_mm_fault(current, vma, address, is_write); + (gdb) call pgd_offset_proc(vma->vm_mm, address) + $22 = (pgd_t *) 0x80a548c + + + + + + That's pretty bogus. Page tables aren't supposed to be in process + text or data areas. Let's see what's in the vma: + + + (gdb) p *vma + $23 = {vm_mm = 0x507d2434, vm_start = 0, vm_end = 134512640, + vm_next = 0x80a4f8c, vm_page_prot = {pgprot = 0}, vm_flags = 31200, + vm_avl_height = 2058, vm_avl_left = 0x80a8c94, vm_avl_right = 0x80d1000, + vm_next_share = 0xaffffdb0, vm_pprev_share = 0xaffffe63, + vm_ops = 0xaffffe7a, vm_pgoff = 2952789626, vm_file = 0xafffffec, + vm_private_data = 0x62} + (gdb) p *vma.vm_mm + $24 = {mmap = 0x507d2434, mmap_avl = 0x0, mmap_cache = 0x8048000, + pgd = 0x80a4f8c, mm_users = {counter = 0}, mm_count = {counter = 134904288}, + map_count = 134909076, mmap_sem = {count = {counter = 135073792}, + sleepers = -1342177872, wait = {lock = <optimized out or zero length>, + task_list = {next = 0xaffffe63, prev = 0xaffffe7a}, + __magic = -1342177670, __creator = -1342177300}, __magic = 98}, + page_table_lock = {}, context = 138, start_code = 0, end_code = 0, + start_data = 0, end_data = 0, start_brk = 0, brk = 0, start_stack = 0, + arg_start = 0, arg_end = 0, env_start = 0, env_end = 0, rss = 1350381536, + total_vm = 0, locked_vm = 0, def_flags = 0, cpu_vm_mask = 0, swap_cnt = 0, + swap_address = 0, segments = 0x0} + + + + + + This also pretty bogus. With all of the 0x80xxxxx and 0xaffffxxx + addresses, this is looking like a stack was plonked down on top of + these structures. Maybe it's a stack overflow from the next page: + + + + (gdb) p vma + $25 = (struct vm_area_struct *) 0x507d2434 + + + + + + That's towards the lower quarter of the page, so that would have to + have been pretty heavy stack overflow: + + + + + + + + + + + + + + + (gdb) x/100x $25 + 0x507d2434: 0x507d2434 0x00000000 0x08048000 0x080a4f8c + 0x507d2444: 0x00000000 0x080a79e0 0x080a8c94 0x080d1000 + 0x507d2454: 0xaffffdb0 0xaffffe63 0xaffffe7a 0xaffffe7a + 0x507d2464: 0xafffffec 0x00000062 0x0000008a 0x00000000 + 0x507d2474: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2484: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2494: 0x00000000 0x00000000 0x507d2fe0 0x00000000 + 0x507d24a4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d24b4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d24c4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d24d4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d24e4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d24f4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2504: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2514: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2524: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2534: 0x00000000 0x00000000 0x507d25dc 0x00000000 + 0x507d2544: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2554: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2564: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2574: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2584: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d2594: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d25a4: 0x00000000 0x00000000 0x00000000 0x00000000 + 0x507d25b4: 0x00000000 0x00000000 0x00000000 0x00000000 + + + + + + It's not stack overflow. The only "stack-like" piece of this data is + the vma_struct itself. + + + At this point, I don't see any avenues to pursue, so I just have to + admit that I have no idea what's going on. What I will do, though, is + stick a trap on the segfault handler which will stop if it sees any + writes to the idle thread's stack. That was the thing that happened + first, and it may be that if I can catch it immediately, what's going + on will be somewhat clearer. + + + 1122..22.. EEppiissooddee 22:: TThhee ccaassee ooff tthhee hhuunngg ffsscckk + + After setting a trap in the SEGV handler for accesses to the signal + thread's stack, I reran the kernel. + + + fsck hung again, this time by hitting the trap: + + + + + + + + + + + + + + + + + Setting hostname uml [ OK ] + Checking root filesystem + /dev/fhd0 contains a file system with errors, check forced. + Error reading block 86894 (Attempt to read block from filesystem resulted in short read) while reading indirect blocks of inode 19780. + + /dev/fhd0: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY. + (i.e., without -a or -p options) + [ FAILED ] + + *** An error occurred during the file system check. + *** Dropping you to a shell; the system will reboot + *** when you leave the shell. + Give root password for maintenance + (or type Control-D for normal startup): + + [root@uml /root]# fsck -y /dev/fhd0 + fsck -y /dev/fhd0 + Parallelizing fsck version 1.14 (9-Jan-1999) + e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09 + /dev/fhd0 contains a file system with errors, check forced. + Pass 1: Checking inodes, blocks, and sizes + Error reading block 86894 (Attempt to read block from filesystem resulted in short read) while reading indirect blocks of inode 19780. Ignore error? yes + + Pass 2: Checking directory structure + Error reading block 49405 (Attempt to read block from filesystem resulted in short read). Ignore error? yes + + Directory inode 11858, block 0, offset 0: directory corrupted + Salvage? yes + + Missing '.' in directory inode 11858. + Fix? yes + + Missing '..' in directory inode 11858. + Fix? yes + + Untested (4127) [100fe44c]: trap_kern.c line 31 + + + + + + I need to get the signal thread to detach from pid 4127 so that I can + attach to it with gdb. This is done by sending it a SIGUSR1, which is + caught by the signal thread, which detaches the process: + + + kill -USR1 4127 + + + + + + Now I can run gdb on it: + + + + + + + + + + + + + + ~/linux/2.3.26/um 1034: gdb linux + GNU gdb 4.17.0.11 with Linux support + Copyright 1998 Free Software Foundation, Inc. + GDB is free software, covered by the GNU General Public License, and you are + welcome to change it and/or distribute copies of it under certain conditions. + Type "show copying" to see the conditions. + There is absolutely no warranty for GDB. Type "show warranty" for details. + This GDB was configured as "i386-redhat-linux"... + (gdb) att 4127 + Attaching to program `/home/dike/linux/2.3.26/um/linux', Pid 4127 + 0x10075891 in __libc_nanosleep () + + + + + + The backtrace shows that it was in a write and that the fault address + (address in frame 3) is 0x50000800, which is right in the middle of + the signal thread's stack page: + + + (gdb) bt + #0 0x10075891 in __libc_nanosleep () + #1 0x1007584d in __sleep (seconds=1000000) + at ../sysdeps/unix/sysv/linux/sleep.c:78 + #2 0x1006ce9a in stop () at user_util.c:191 + #3 0x1006bf88 in segv (address=1342179328, is_write=2) at trap_kern.c:31 + #4 0x1006c628 in segv_handler (sc=0x5006eaf8) at trap_user.c:174 + #5 0x1006c63c in kern_segv_handler (sig=11) at trap_user.c:182 + #6 <signal handler called> + #7 0xc0fd in ?? () + #8 0x10016647 in sys_write (fd=3, buf=0x80b8800 "R.", count=1024) + at read_write.c:159 + #9 0x1006d603 in execute_syscall (syscall=4, args=0x5006ef08) + at syscall_kern.c:254 + #10 0x1006af87 in really_do_syscall (sig=12) at syscall_user.c:35 + #11 <signal handler called> + #12 0x400dc8b0 in ?? () + #13 <signal handler called> + #14 0x400dc8b0 in ?? () + #15 0x80545fd in ?? () + #16 0x804daae in ?? () + #17 0x8054334 in ?? () + #18 0x804d23e in ?? () + #19 0x8049632 in ?? () + #20 0x80491d2 in ?? () + #21 0x80596b5 in ?? () + (gdb) p (void *)1342179328 + $3 = (void *) 0x50000800 + + + + + + Going up the stack to the segv_handler frame and looking at where in + the code the access happened shows that it happened near line 110 of + block_dev.c: + + + + + + + + + + (gdb) up + #1 0x1007584d in __sleep (seconds=1000000) + at ../sysdeps/unix/sysv/linux/sleep.c:78 + ../sysdeps/unix/sysv/linux/sleep.c:78: No such file or directory. + (gdb) + #2 0x1006ce9a in stop () at user_util.c:191 + 191 while(1) sleep(1000000); + (gdb) + #3 0x1006bf88 in segv (address=1342179328, is_write=2) at trap_kern.c:31 + 31 KERN_UNTESTED(); + (gdb) + #4 0x1006c628 in segv_handler (sc=0x5006eaf8) at trap_user.c:174 + 174 segv(sc->cr2, sc->err & 2); + (gdb) p *sc + $1 = {gs = 0, __gsh = 0, fs = 0, __fsh = 0, es = 43, __esh = 0, ds = 43, + __dsh = 0, edi = 1342179328, esi = 134973440, ebp = 1342631484, + esp = 1342630864, ebx = 256, edx = 0, ecx = 256, eax = 1024, trapno = 14, + err = 6, eip = 268550834, cs = 35, __csh = 0, eflags = 66070, + esp_at_signal = 1342630864, ss = 43, __ssh = 0, fpstate = 0x0, oldmask = 0, + cr2 = 1342179328} + (gdb) p (void *)268550834 + $2 = (void *) 0x1001c2b2 + (gdb) i sym $2 + block_write + 1090 in section .text + (gdb) i line *$2 + Line 209 of "/home/dike/linux/2.3.26/um/include/asm/arch/string.h" + starts at address 0x1001c2a1 <block_write+1073> + and ends at 0x1001c2bf <block_write+1103>. + (gdb) i line *0x1001c2c0 + Line 110 of "block_dev.c" starts at address 0x1001c2bf <block_write+1103> + and ends at 0x1001c2e3 <block_write+1139>. + + + + + + Looking at the source shows that the fault happened during a call to + copy_to_user to copy the data into the kernel: + + + 107 count -= chars; + 108 copy_from_user(p,buf,chars); + 109 p += chars; + 110 buf += chars; + + + + + + p is the pointer which must contain 0x50000800, since buf contains + 0x80b8800 (frame 8 above). It is defined as: + + + p = offset + bh->b_data; + + + + + + I need to figure out what bh is, and it just so happens that bh is + passed as an argument to mark_buffer_uptodate and mark_buffer_dirty a + few lines later, so I do a little disassembly: + + + + + (gdb) disas 0x1001c2bf 0x1001c2e0 + Dump of assembler code from 0x1001c2bf to 0x1001c2d0: + 0x1001c2bf <block_write+1103>: addl %eax,0xc(%ebp) + 0x1001c2c2 <block_write+1106>: movl 0xfffffdd4(%ebp),%edx + 0x1001c2c8 <block_write+1112>: btsl $0x0,0x18(%edx) + 0x1001c2cd <block_write+1117>: btsl $0x1,0x18(%edx) + 0x1001c2d2 <block_write+1122>: sbbl %ecx,%ecx + 0x1001c2d4 <block_write+1124>: testl %ecx,%ecx + 0x1001c2d6 <block_write+1126>: jne 0x1001c2e3 <block_write+1139> + 0x1001c2d8 <block_write+1128>: pushl $0x0 + 0x1001c2da <block_write+1130>: pushl %edx + 0x1001c2db <block_write+1131>: call 0x1001819c <__mark_buffer_dirty> + End of assembler dump. + + + + + + At that point, bh is in %edx (address 0x1001c2da), which is calculated + at 0x1001c2c2 as %ebp + 0xfffffdd4, so I figure exactly what that is, + taking %ebp from the sigcontext_struct above: + + + (gdb) p (void *)1342631484 + $5 = (void *) 0x5006ee3c + (gdb) p 0x5006ee3c+0xfffffdd4 + $6 = 1342630928 + (gdb) p (void *)$6 + $7 = (void *) 0x5006ec10 + (gdb) p *((void **)$7) + $8 = (void *) 0x50100200 + + + + + + Now, I look at the structure to see what's in it, and particularly, + what its b_data field contains: + + + (gdb) p *((struct buffer_head *)0x50100200) + $13 = {b_next = 0x50289380, b_blocknr = 49405, b_size = 1024, b_list = 0, + b_dev = 15872, b_count = {counter = 1}, b_rdev = 15872, b_state = 24, + b_flushtime = 0, b_next_free = 0x501001a0, b_prev_free = 0x50100260, + b_this_page = 0x501001a0, b_reqnext = 0x0, b_pprev = 0x507fcf58, + b_data = 0x50000800 "", b_page = 0x50004000, + b_end_io = 0x10017f60 <end_buffer_io_sync>, b_dev_id = 0x0, + b_rsector = 98810, b_wait = {lock = <optimized out or zero length>, + task_list = {next = 0x50100248, prev = 0x50100248}, __magic = 1343226448, + __creator = 0}, b_kiobuf = 0x0} + + + + + + The b_data field is indeed 0x50000800, so the question becomes how + that happened. The rest of the structure looks fine, so this probably + is not a case of data corruption. It happened on purpose somehow. + + + The b_page field is a pointer to the page_struct representing the + 0x50000000 page. Looking at it shows the kernel's idea of the state + of that page: + + + + (gdb) p *$13.b_page + $17 = {list = {next = 0x50004a5c, prev = 0x100c5174}, mapping = 0x0, + index = 0, next_hash = 0x0, count = {counter = 1}, flags = 132, lru = { + next = 0x50008460, prev = 0x50019350}, wait = { + lock = <optimized out or zero length>, task_list = {next = 0x50004024, + prev = 0x50004024}, __magic = 1342193708, __creator = 0}, + pprev_hash = 0x0, buffers = 0x501002c0, virtual = 1342177280, + zone = 0x100c5160} + + + + + + Some sanity-checking: the virtual field shows the "virtual" address of + this page, which in this kernel is the same as its "physical" address, + and the page_struct itself should be mem_map[0], since it represents + the first page of memory: + + + + (gdb) p (void *)1342177280 + $18 = (void *) 0x50000000 + (gdb) p mem_map + $19 = (mem_map_t *) 0x50004000 + + + + + + These check out fine. + + + Now to check out the page_struct itself. In particular, the flags + field shows whether the page is considered free or not: + + + (gdb) p (void *)132 + $21 = (void *) 0x84 + + + + + + The "reserved" bit is the high bit, which is definitely not set, so + the kernel considers the signal stack page to be free and available to + be used. + + + At this point, I jump to conclusions and start looking at my early + boot code, because that's where that page is supposed to be reserved. + + + In my setup_arch procedure, I have the following code which looks just + fine: + + + + bootmap_size = init_bootmem(start_pfn, end_pfn - start_pfn); + free_bootmem(__pa(low_physmem) + bootmap_size, high_physmem - low_physmem); + + + + + + Two stack pages have already been allocated, and low_physmem points to + the third page, which is the beginning of free memory. + The init_bootmem call declares the entire memory to the boot memory + manager, which marks it all reserved. The free_bootmem call frees up + all of it, except for the first two pages. This looks correct to me. + + + So, I decide to see init_bootmem run and make sure that it is marking + those first two pages as reserved. I never get that far. + + + Stepping into init_bootmem, and looking at bootmem_map before looking + at what it contains shows the following: + + + + (gdb) p bootmem_map + $3 = (void *) 0x50000000 + + + + + + Aha! The light dawns. That first page is doing double duty as a + stack and as the boot memory map. The last thing that the boot memory + manager does is to free the pages used by its memory map, so this page + is getting freed even its marked as reserved. + + + The fix was to initialize the boot memory manager before allocating + those two stack pages, and then allocate them through the boot memory + manager. After doing this, and fixing a couple of subsequent buglets, + the stack corruption problem disappeared. + + + + + + 1133.. WWhhaatt ttoo ddoo wwhheenn UUMMLL ddooeessnn''tt wwoorrkk + + + + + 1133..11.. SSttrraannggee ccoommppiillaattiioonn eerrrroorrss wwhheenn yyoouu bbuuiilldd ffrroomm ssoouurrccee + + As of test11, it is necessary to have "ARCH=um" in the environment or + on the make command line for all steps in building UML, including + clean, distclean, or mrproper, config, menuconfig, or xconfig, dep, + and linux. If you forget for any of them, the i386 build seems to + contaminate the UML build. If this happens, start from scratch with + + + host% + make mrproper ARCH=um + + + + + and repeat the build process with ARCH=um on all the steps. + + + See ``Compiling the kernel and modules'' for more details. + + + Another cause of strange compilation errors is building UML in + /usr/src/linux. If you do this, the first thing you need to do is + clean up the mess you made. The /usr/src/linux/asm link will now + point to /usr/src/linux/asm-um. Make it point back to + /usr/src/linux/asm-i386. Then, move your UML pool someplace else and + build it there. Also see below, where a more specific set of symptoms + is described. + + + + 1133..33.. AA vvaarriieettyy ooff ppaanniiccss aanndd hhaannggss wwiitthh //ttmmpp oonn aa rreeiisseerrffss ffiilleessyyss-- + tteemm + + I saw this on reiserfs 3.5.21 and it seems to be fixed in 3.5.27. + Panics preceded by + + + Detaching pid nnnn + + + + are diagnostic of this problem. This is a reiserfs bug which causes a + thread to occasionally read stale data from a mmapped page shared with + another thread. The fix is to upgrade the filesystem or to have /tmp + be an ext2 filesystem. + + + + 1133..44.. TThhee ccoommppiillee ffaaiillss wwiitthh eerrrroorrss aabboouutt ccoonnfflliiccttiinngg ttyyppeess ffoorr + ''ooppeenn'',, ''dduupp'',, aanndd ''wwaaiittppiidd'' + + This happens when you build in /usr/src/linux. The UML build makes + the include/asm link point to include/asm-um. /usr/include/asm points + to /usr/src/linux/include/asm, so when that link gets moved, files + which need to include the asm-i386 versions of headers get the + incompatible asm-um versions. The fix is to move the include/asm link + back to include/asm-i386 and to do UML builds someplace else. + + + + 1133..55.. UUMMLL ddooeessnn''tt wwoorrkk wwhheenn //ttmmpp iiss aann NNFFSS ffiilleessyysstteemm + + This seems to be a similar situation with the ReiserFS problem above. + Some versions of NFS seems not to handle mmap correctly, which UML + depends on. The workaround is have /tmp be a non-NFS directory. + + + 1133..66.. UUMMLL hhaannggss oonn bboooott wwhheenn ccoommppiilleedd wwiitthh ggpprrooff ssuuppppoorrtt + + If you build UML with gprof support and, early in the boot, it does + this + + + kernel BUG at page_alloc.c:100! + + + + + you have a buggy gcc. You can work around the problem by removing + UM_FASTCALL from CFLAGS in arch/um/Makefile-i386. This will open up + another bug, but that one is fairly hard to reproduce. + + + + 1133..77.. ssyyssllooggdd ddiieess wwiitthh aa SSIIGGTTEERRMM oonn ssttaarrttuupp + + The exact boot error depends on the distribution that you're booting, + but Debian produces this: + + + /etc/rc2.d/S10sysklogd: line 49: 93 Terminated + start-stop-daemon --start --quiet --exec /sbin/syslogd -- $SYSLOGD + + + + + This is a syslogd bug. There's a race between a parent process + installing a signal handler and its child sending the signal. See + this uml-devel post <http://www.geocrawler.com/lists/3/Source- + Forge/709/0/6612801> for the details. + + + + 1133..88.. TTUUNN//TTAAPP nneettwwoorrkkiinngg ddooeessnn''tt wwoorrkk oonn aa 22..44 hhoosstt + + There are a couple of problems which were + <http://www.geocrawler.com/lists/3/SourceForge/597/0/> name="pointed + out"> by Tim Robinson <timro at trkr dot net> + + +o It doesn't work on hosts running 2.4.7 (or thereabouts) or earlier. + The fix is to upgrade to something more recent and then read the + next item. + + +o If you see + + + File descriptor in bad state + + + + when you bring up the device inside UML, you have a header mismatch + between the original kernel and the upgraded one. Make /usr/src/linux + point at the new headers. This will only be a problem if you build + uml_net yourself. + + + + 1133..99.. YYoouu ccaann nneettwwoorrkk ttoo tthhee hhoosstt bbuutt nnoott ttoo ootthheerr mmaacchhiinneess oonn tthhee + nneett + + If you can connect to the host, and the host can connect to UML, but + you cannot connect to any other machines, then you may need to enable + IP Masquerading on the host. Usually this is only experienced when + using private IP addresses (192.168.x.x or 10.x.x.x) for host/UML + networking, rather than the public address space that your host is + connected to. UML does not enable IP Masquerading, so you will need + to create a static rule to enable it: + + + host% + iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE + + + + + Replace eth0 with the interface that you use to talk to the rest of + the world. + + + Documentation on IP Masquerading, and SNAT, can be found at + www.netfilter.org <http://www.netfilter.org> . + + + If you can reach the local net, but not the outside Internet, then + that is usually a routing problem. The UML needs a default route: + + + UML# + route add default gw gateway IP + + + + + The gateway IP can be any machine on the local net that knows how to + reach the outside world. Usually, this is the host or the local net- + work's gateway. + + + Occasionally, we hear from someone who can reach some machines, but + not others on the same net, or who can reach some ports on other + machines, but not others. These are usually caused by strange + firewalling somewhere between the UML and the other box. You track + this down by running tcpdump on every interface the packets travel + over and see where they disappear. When you find a machine that takes + the packets in, but does not send them onward, that's the culprit. + + + + 1133..1100.. II hhaavvee nnoo rroooott aanndd II wwaanntt ttoo ssccrreeaamm + + Thanks to Birgit Wahlich for telling me about this strange one. It + turns out that there's a limit of six environment variables on the + kernel command line. When that limit is reached or exceeded, argument + processing stops, which means that the 'root=' argument that UML + usually adds is not seen. So, the filesystem has no idea what the + root device is, so it panics. + + + The fix is to put less stuff on the command line. Glomming all your + setup variables into one is probably the best way to go. + + + + 1133..1111.. UUMMLL bbuuiilldd ccoonnfflliicctt bbeettwweeeenn ppttrraaccee..hh aanndd uuccoonntteexxtt..hh + + On some older systems, /usr/include/asm/ptrace.h and + /usr/include/sys/ucontext.h define the same names. So, when they're + included together, the defines from one completely mess up the parsing + of the other, producing errors like: + /usr/include/sys/ucontext.h:47: parse error before + `10' + + + + + plus a pile of warnings. + + + This is a libc botch, which has since been fixed, and I don't see any + way around it besides upgrading. + + + + 1133..1122.. TThhee UUMMLL BBooggooMMiippss iiss eexxaaccttllyy hhaallff tthhee hhoosstt''ss BBooggooMMiippss + + On i386 kernels, there are two ways of running the loop that is used + to calculate the BogoMips rating, using the TSC if it's there or using + a one-instruction loop. The TSC produces twice the BogoMips as the + loop. UML uses the loop, since it has nothing resembling a TSC, and + will get almost exactly the same BogoMips as a host using the loop. + However, on a host with a TSC, its BogoMips will be double the loop + BogoMips, and therefore double the UML BogoMips. + + + + 1133..1133.. WWhheenn yyoouu rruunn UUMMLL,, iitt iimmmmeeddiiaatteellyy sseeggffaauullttss + + If the host is configured with the 2G/2G address space split, that's + why. See ``UML on 2G/2G hosts'' for the details on getting UML to + run on your host. + + + + 1133..1144.. xxtteerrmmss aappppeeaarr,, tthheenn iimmmmeeddiiaatteellyy ddiissaappppeeaarr + + If you're running an up to date kernel with an old release of + uml_utilities, the port-helper program will not work properly, so + xterms will exit straight after they appear. The solution is to + upgrade to the latest release of uml_utilities. Usually this problem + occurs when you have installed a packaged release of UML then compiled + your own development kernel without upgrading the uml_utilities from + the source distribution. + + + + 1133..1155.. AAnnyy ootthheerr ppaanniicc,, hhaanngg,, oorr ssttrraannggee bbeehhaavviioorr + + If you're seeing truly strange behavior, such as hangs or panics that + happen in random places, or you try running the debugger to see what's + happening and it acts strangely, then it could be a problem in the + host kernel. If you're not running a stock Linus or -ac kernel, then + try that. An early version of the preemption patch and a 2.4.10 SuSE + kernel have caused very strange problems in UML. + + + Otherwise, let me know about it. Send a message to one of the UML + mailing lists - either the developer list - user-mode-linux-devel at + lists dot sourceforge dot net (subscription info) or the user list - + user-mode-linux-user at lists dot sourceforge do net (subscription + info), whichever you prefer. Don't assume that everyone knows about + it and that a fix is imminent. + + + If you want to be super-helpful, read ``Diagnosing Problems'' and + follow the instructions contained therein. + 1144.. DDiiaaggnnoossiinngg PPrroobblleemmss + + + If you get UML to crash, hang, or otherwise misbehave, you should + report this on one of the project mailing lists, either the developer + list - user-mode-linux-devel at lists dot sourceforge dot net + (subscription info) or the user list - user-mode-linux-user at lists + dot sourceforge dot net (subscription info). When you do, it is + likely that I will want more information. So, it would be helpful to + read the stuff below, do whatever is applicable in your case, and + report the results to the list. + + + For any diagnosis, you're going to need to build a debugging kernel. + The binaries from this site aren't debuggable. If you haven't done + this before, read about ``Compiling the kernel and modules'' and + ``Kernel debugging'' UML first. + + + 1144..11.. CCaassee 11 :: NNoorrmmaall kkeerrnneell ppaanniiccss + + The most common case is for a normal thread to panic. To debug this, + you will need to run it under the debugger (add 'debug' to the command + line). An xterm will start up with gdb running inside it. Continue + it when it stops in start_kernel and make it crash. Now ^C gdb and + + + If the panic was a "Kernel mode fault", then there will be a segv + frame on the stack and I'm going to want some more information. The + stack might look something like this: + + + (UML gdb) backtrace + #0 0x1009bf76 in __sigprocmask (how=1, set=0x5f347940, oset=0x0) + at ../sysdeps/unix/sysv/linux/sigprocmask.c:49 + #1 0x10091411 in change_sig (signal=10, on=1) at process.c:218 + #2 0x10094785 in timer_handler (sig=26) at time_kern.c:32 + #3 0x1009bf38 in __restore () + at ../sysdeps/unix/sysv/linux/i386/sigaction.c:125 + #4 0x1009534c in segv (address=8, ip=268849158, is_write=2, is_user=0) + at trap_kern.c:66 + #5 0x10095c04 in segv_handler (sig=11) at trap_user.c:285 + #6 0x1009bf38 in __restore () + + + + + I'm going to want to see the symbol and line information for the value + of ip in the segv frame. In this case, you would do the following: + + + (UML gdb) i sym 268849158 + + + + + and + + + (UML gdb) i line *268849158 + + + + + The reason for this is the __restore frame right above the segv_han- + dler frame is hiding the frame that actually segfaulted. So, I have + to get that information from the faulting ip. + + + 1144..22.. CCaassee 22 :: TTrraacciinngg tthhrreeaadd ppaanniiccss + + The less common and more painful case is when the tracing thread + panics. In this case, the kernel debugger will be useless because it + needs a healthy tracing thread in order to work. The first thing to + do is get a backtrace from the tracing thread. This is done by + figuring out what its pid is, firing up gdb, and attaching it to that + pid. You can figure out the tracing thread pid by looking at the + first line of the console output, which will look like this: + + + tracing thread pid = 15851 + + + + + or by running ps on the host and finding the line that looks like + this: + + + jdike 15851 4.5 0.4 132568 1104 pts/0 S 21:34 0:05 ./linux [(tracing thread)] + + + + + If the panic was 'segfault in signals', then follow the instructions + above for collecting information about the location of the seg fault. + + + If the tracing thread flaked out all by itself, then send that + backtrace in and wait for our crack debugging team to fix the problem. + + + 1144..33.. CCaassee 33 :: TTrraacciinngg tthhrreeaadd ppaanniiccss ccaauusseedd bbyy ootthheerr tthhrreeaaddss + + However, there are cases where the misbehavior of another thread + caused the problem. The most common panic of this type is: + + + wait_for_stop failed to wait for <pid> to stop with <signal number> + + + + + In this case, you'll need to get a backtrace from the process men- + tioned in the panic, which is complicated by the fact that the kernel + debugger is defunct and without some fancy footwork, another gdb can't + attach to it. So, this is how the fancy footwork goes: + + In a shell: + + + host% kill -STOP pid + + + + + Run gdb on the tracing thread as described in case 2 and do: + + + (host gdb) call detach(pid) + + + If you get a segfault, do it again. It always works the second time. + + Detach from the tracing thread and attach to that other thread: + + + (host gdb) detach + + + + + + + (host gdb) attach pid + + + + + If gdb hangs when attaching to that process, go back to a shell and + do: + + + host% + kill -CONT pid + + + + + And then get the backtrace: + + + (host gdb) backtrace + + + + + + 1144..44.. CCaassee 44 :: HHaannggss + + Hangs seem to be fairly rare, but they sometimes happen. When a hang + happens, we need a backtrace from the offending process. Run the + kernel debugger as described in case 1 and get a backtrace. If the + current process is not the idle thread, then send in the backtrace. + You can tell that it's the idle thread if the stack looks like this: + + + #0 0x100b1401 in __libc_nanosleep () + #1 0x100a2885 in idle_sleep (secs=10) at time.c:122 + #2 0x100a546f in do_idle () at process_kern.c:445 + #3 0x100a5508 in cpu_idle () at process_kern.c:471 + #4 0x100ec18f in start_kernel () at init/main.c:592 + #5 0x100a3e10 in start_kernel_proc (unused=0x0) at um_arch.c:71 + #6 0x100a383f in signal_tramp (arg=0x100a3dd8) at trap_user.c:50 + + + + + If this is the case, then some other process is at fault, and went to + sleep when it shouldn't have. Run ps on the host and figure out which + process should not have gone to sleep and stayed asleep. Then attach + to it with gdb and get a backtrace as described in case 3. + + + + + + + 1155.. TThhaannkkss + + + A number of people have helped this project in various ways, and this + page gives recognition where recognition is due. + + + If you're listed here and you would prefer a real link on your name, + or no link at all, instead of the despammed email address pseudo-link, + let me know. + + + If you're not listed here and you think maybe you should be, please + let me know that as well. I try to get everyone, but sometimes my + bookkeeping lapses and I forget about contributions. + + + 1155..11.. CCooddee aanndd DDooccuummeennttaattiioonn + + Rusty Russell <rusty at linuxcare.com.au> - + + +o wrote the HOWTO <http://user-mode- + linux.sourceforge.net/UserModeLinux-HOWTO.html> + + +o prodded me into making this project official and putting it on + SourceForge + + +o came up with the way cool UML logo <http://user-mode- + linux.sourceforge.net/uml-small.png> + + +o redid the config process + + + Peter Moulder <reiter at netspace.net.au> - Fixed my config and build + processes, and added some useful code to the block driver + + + Bill Stearns <wstearns at pobox.com> - + + +o HOWTO updates + + +o lots of bug reports + + +o lots of testing + + +o dedicated a box (uml.ists.dartmouth.edu) to support UML development + + +o wrote the mkrootfs script, which allows bootable filesystems of + RPM-based distributions to be cranked out + + +o cranked out a large number of filesystems with said script + + + Jim Leu <jleu at mindspring.com> - Wrote the virtual ethernet driver + and associated usermode tools + + Lars Brinkhoff <http://lars.nocrew.org/> - Contributed the ptrace + proxy from his own project <http://a386.nocrew.org/> to allow easier + kernel debugging + + + Andrea Arcangeli <andrea at suse.de> - Redid some of the early boot + code so that it would work on machines with Large File Support + + + Chris Emerson <http://www.chiark.greenend.org.uk/~cemerson/> - Did + the first UML port to Linux/ppc + + + Harald Welte <laforge at gnumonks.org> - Wrote the multicast + transport for the network driver + + + Jorgen Cederlof - Added special file support to hostfs + + + Greg Lonnon <glonnon at ridgerun dot com> - Changed the ubd driver + to allow it to layer a COW file on a shared read-only filesystem and + wrote the iomem emulation support + + + Henrik Nordstrom <http://hem.passagen.se/hno/> - Provided a variety + of patches, fixes, and clues + + + Lennert Buytenhek - Contributed various patches, a rewrite of the + network driver, the first implementation of the mconsole driver, and + did the bulk of the work needed to get SMP working again. + + + Yon Uriarte - Fixed the TUN/TAP network backend while I slept. + + + Adam Heath - Made a bunch of nice cleanups to the initialization code, + plus various other small patches. + + + Matt Zimmerman - Matt volunteered to be the UML Debian maintainer and + is doing a real nice job of it. He also noticed and fixed a number of + actually and potentially exploitable security holes in uml_net. Plus + the occasional patch. I like patches. + + + James McMechan - James seems to have taken over maintenance of the ubd + driver and is doing a nice job of it. + + + Chandan Kudige - wrote the umlgdb script which automates the reloading + of module symbols. + + + Steve Schmidtke - wrote the UML slirp transport and hostaudio drivers, + enabling UML processes to access audio devices on the host. He also + submitted patches for the slip transport and lots of other things. + + + David Coulson <http://davidcoulson.net> - + + +o Set up the usermodelinux.org <http://usermodelinux.org> site, + which is a great way of keeping the UML user community on top of + UML goings-on. + + +o Site documentation and updates + + +o Nifty little UML management daemon UMLd + <http://uml.openconsultancy.com/umld/> + + +o Lots of testing and bug reports + + + + + 1155..22.. FFlluusshhiinngg oouutt bbuuggss + + + + +o Yuri Pudgorodsky + + +o Gerald Britton + + +o Ian Wehrman + + +o Gord Lamb + + +o Eugene Koontz + + +o John H. Hartman + + +o Anders Karlsson + + +o Daniel Phillips + + +o John Fremlin + + +o Rainer Burgstaller + + +o James Stevenson + + +o Matt Clay + + +o Cliff Jefferies + + +o Geoff Hoff + + +o Lennert Buytenhek + + +o Al Viro + + +o Frank Klingenhoefer + + +o Livio Baldini Soares + + +o Jon Burgess + + +o Petru Paler + + +o Paul + + +o Chris Reahard + + +o Sverker Nilsson + + +o Gong Su + + +o johan verrept + + +o Bjorn Eriksson + + +o Lorenzo Allegrucci + + +o Muli Ben-Yehuda + + +o David Mansfield + + +o Howard Goff + + +o Mike Anderson + + +o John Byrne + + +o Sapan J. Batia + + +o Iris Huang + + +o Jan Hudec + + +o Voluspa + + + + + 1155..33.. BBuugglleettss aanndd cclleeaann--uuppss + + + + +o Dave Zarzycki + + +o Adam Lazur + + +o Boria Feigin + + +o Brian J. Murrell + + +o JS + + +o Roman Zippel + + +o Wil Cooley + + +o Ayelet Shemesh + + +o Will Dyson + + +o Sverker Nilsson + + +o dvorak + + +o v.naga srinivas + + +o Shlomi Fish + + +o Roger Binns + + +o johan verrept + + +o MrChuoi + + +o Peter Cleve + + +o Vincent Guffens + + +o Nathan Scott + + +o Patrick Caulfield + + +o jbearce + + +o Catalin Marinas + + +o Shane Spencer + + +o Zou Min + + + +o Ryan Boder + + +o Lorenzo Colitti + + +o Gwendal Grignou + + +o Andre' Breiler + + +o Tsutomu Yasuda + + + + 1155..44.. CCaassee SSttuuddiieess + + + +o Jon Wright + + +o William McEwan + + +o Michael Richardson + + + + 1155..55.. OOtthheerr ccoonnttrriibbuuttiioonnss + + + Bill Carr <Bill.Carr at compaq.com> made the Red Hat mkrootfs script + work with RH 6.2. + + Michael Jennings <mikejen at hevanet.com> sent in some material which + is now gracing the top of the index page <http://user-mode- + linux.sourceforge.net/> of this site. + + SGI <http://www.sgi.com> (and more specifically Ralf Baechle <ralf at + uni-koblenz.de> ) gave me an account on oss.sgi.com + <http://www.oss.sgi.com> . The bandwidth there made it possible to + produce most of the filesystems available on the project download + page. + + Laurent Bonnaud <Laurent.Bonnaud at inpg.fr> took the old grotty + Debian filesystem that I've been distributing and updated it to 2.2. + It is now available by itself here. + + Rik van Riel gave me some ftp space on ftp.nl.linux.org so I can make + releases even when Sourceforge is broken. + + Rodrigo de Castro looked at my broken pte code and told me what was + wrong with it, letting me fix a long-standing (several weeks) and + serious set of bugs. + + Chris Reahard built a specialized root filesystem for running a DNS + server jailed inside UML. It's available from the download + <http://user-mode-linux.sourceforge.net/dl-sf.html> page in the Jail + Filesystems section. + + + + + + + + + + + + |