From e9b47a6d699b97b768a7bd4ea3e5d5a9a7753537 Mon Sep 17 00:00:00 2001 From: jhb Date: Tue, 1 May 2018 15:17:46 +0000 Subject: [PATCH] Initial debug server for bhyve. This commit adds a new debug server to bhyve. Unlike the existing -g option which provides an efficient connection to a debug server running in the guest OS, this debug server permits inspection and control of the guest from within the hypervisor itself without requiring any cooperation from the guest. It is similar to the debug server provided by qemu. To avoid conflicting with the existing -g option, a new -G option has been added that accepts a TCP port. An IPv4 socket is bound to this port and listens for connections from debuggers. In addition, if the port begins with the character 'w', the hypervisor will pause the guest at the first instruction until a debugger attaches and explicitly continues the guest. Note that only a single debugger can attach to a guest at a time. Virtual CPUs are exposed to the remote debugger as threads. General purpose register values can be read for each virtual CPU. Other registers cannot currently be read, and no register values can be changed by the debugger. The remote debugger can read guest memory but not write to guest memory. To facilitate source-level debugging of the guest, memory addresses from the debugger are treated as virtual addresses (rather than physical addresses) and are resolved to a physical address using the active virtual address translation of the current virtual CPU. Memory reads should honor memory mapped I/O regions, though the debug server does not attempt to honor any alignment or size constraints when accessing MMIO. The debug server provides limited support for controlling the guest. The guest is suspended when a debugger is attached and resumes when a debugger detaches. A debugger can suspend a guest by sending a Ctrl-C request (e.g. via Ctrl-C in GDB). A debugger can also continue a suspended guest while remaining attached. Breakpoints are not yet supported. Single stepping is supported on Intel CPUs that support MTRAP VM exits, but is not available on other systems. While the current debug server has limited functionality, it should at least be usable for basic debugging now. It is also a useful checkpoint to serve as a base for adding additional features. Reviewed by: grehan Differential Revision: https://reviews.freebsd.org/D15022 --- usr.sbin/bhyve/Makefile | 5 + usr.sbin/bhyve/bhyve.8 | 34 +- usr.sbin/bhyve/bhyverun.c | 33 +- usr.sbin/bhyve/gdb.c | 1313 +++++++++++++++++++++++++++++++++++++ usr.sbin/bhyve/gdb.h | 39 ++ usr.sbin/bhyve/mem.c | 67 +- usr.sbin/bhyve/mem.h | 2 + 7 files changed, 1483 insertions(+), 10 deletions(-) create mode 100644 usr.sbin/bhyve/gdb.c create mode 100644 usr.sbin/bhyve/gdb.h diff --git a/usr.sbin/bhyve/Makefile b/usr.sbin/bhyve/Makefile index 8159d7ddbd7a..bf4450f12eaf 100644 --- a/usr.sbin/bhyve/Makefile +++ b/usr.sbin/bhyve/Makefile @@ -24,6 +24,7 @@ SRCS= \ consport.c \ dbgport.c \ fwctl.c \ + gdb.c \ inout.c \ ioapic.c \ mem.c \ @@ -75,6 +76,10 @@ CFLAGS+= -I${BHYVE_SYSDIR}/sys/dev/e1000 CFLAGS+= -I${BHYVE_SYSDIR}/sys/dev/mii CFLAGS+= -I${BHYVE_SYSDIR}/sys/dev/usb/controller +.ifdef GDB_LOG +CFLAGS+=-DGDB_LOG +.endif + WARNS?= 2 .include diff --git a/usr.sbin/bhyve/bhyve.8 b/usr.sbin/bhyve/bhyve.8 index 8c259b000051..9b36576114f7 100644 --- a/usr.sbin/bhyve/bhyve.8 +++ b/usr.sbin/bhyve/bhyve.8 @@ -24,7 +24,7 @@ .\" .\" $FreeBSD$ .\" -.Dd April 6, 2018 +.Dd May 1, 2018 .Dt BHYVE 8 .Os .Sh NAME @@ -48,6 +48,7 @@ .Op Fl m Ar memsize Ns Op Ar K|k|M|m|G|g|T|t .Op Fl p Ar vcpu:hostcpu .Op Fl s Ar slot,emulation Ns Op , Ns Ar conf +.Op Fl G Ar port .Op Fl U Ar uuid .Ar vmname .Sh DESCRIPTION @@ -125,6 +126,19 @@ kernels compiled with allow a remote kernel kgdb to be relayed to the guest kernel gdb stub via a local IPv4 address and this port. This option will be deprecated in a future version. +.It Fl G Ar port +Start a debug server that uses the GDB protocol to export guest state to a +debugger. +An IPv4 TCP socket will be bound to the supplied +.Ar port +to listen for debugger connections. +Only a single debugger may be attached to the debug server at a time. +If +.Ar port +begins with +.Sq w , +.Nm +will pause execution at the first instruction waiting for a debugger to attach. .It Fl h Print help message and exit. .It Fl H @@ -438,6 +452,24 @@ Alphanumeric name of the guest. This should be the same as that created by .Xr bhyveload 8 . .El +.Sh DEBUG SERVER +The current debug server provides limited support for debuggers. +.Ss Registers +Each virtual CPU is exposed to the debugger as a thread. +.Pp +General purpose registers can be queried for each virtual CPU, but other +registers such as floating-point and system registers cannot be queried. +.Ss Memory +Memory (including memory mapped I/O regions) can be read by the debugger, +but not written. Memory operations use virtual addresses that are resolved +to physical addresses via the current virtual CPU's active address translation. +.Ss Control +The running guest can be interrupted by the debugger at any time +.Pq for example, by pressing Ctrl-C in the debugger . +.Pp +Single stepping is only supported on Intel CPUs supporting the MTRAP VM exit. +.Pp +Breakpoints are not supported. .Sh SIGNAL HANDLING .Nm deals with the following signals: diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c index 8cad2299291b..90ab1a66536c 100644 --- a/usr.sbin/bhyve/bhyverun.c +++ b/usr.sbin/bhyve/bhyverun.c @@ -71,6 +71,7 @@ __FBSDID("$FreeBSD$"); #include "inout.h" #include "dbgport.h" #include "fwctl.h" +#include "gdb.h" #include "ioapic.h" #include "mem.h" #include "mevent.h" @@ -338,6 +339,8 @@ fbsdrun_start_thread(void *param) snprintf(tname, sizeof(tname), "vcpu %d", vcpu); pthread_set_name_np(mtp->mt_thr, tname); + gdb_cpu_add(vcpu); + vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip); /* not reached */ @@ -601,6 +604,8 @@ vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) stats.vmexit_mtrap++; + gdb_cpu_mtrap(*pvcpu); + return (VMEXIT_CONTINUE); } @@ -675,6 +680,14 @@ vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) return (0); /* NOTREACHED */ } +static int +vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) +{ + + gdb_cpu_suspend(*pvcpu); + return (VMEXIT_CONTINUE); +} + static vmexit_handler_t handler[VM_EXITCODE_MAX] = { [VM_EXITCODE_INOUT] = vmexit_inout, [VM_EXITCODE_INOUT_STR] = vmexit_inout, @@ -689,6 +702,7 @@ static vmexit_handler_t handler[VM_EXITCODE_MAX] = { [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, [VM_EXITCODE_SUSPENDED] = vmexit_suspend, [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, + [VM_EXITCODE_DEBUG] = vmexit_debug, }; static void @@ -877,9 +891,10 @@ do_open(const char *vmname) int main(int argc, char *argv[]) { - int c, error, gdb_port, err, bvmcons; + int c, error, dbg_port, gdb_port, err, bvmcons; int max_vcpus, mptgen, memflags; int rtc_localtime; + bool gdb_stop; struct vmctx *ctx; uint64_t rip; size_t memsize; @@ -887,7 +902,9 @@ main(int argc, char *argv[]) bvmcons = 0; progname = basename(argv[0]); + dbg_port = 0; gdb_port = 0; + gdb_stop = false; guest_ncpus = 1; sockets = cores = threads = 1; maxcpus = 0; @@ -896,7 +913,7 @@ main(int argc, char *argv[]) rtc_localtime = 1; memflags = 0; - optstr = "abehuwxACHIPSWYp:g:c:s:m:l:U:"; + optstr = "abehuwxACHIPSWYp:g:G:c:s:m:l:U:"; while ((c = getopt(argc, argv, optstr)) != -1) { switch (c) { case 'a': @@ -924,6 +941,13 @@ main(int argc, char *argv[]) memflags |= VM_MEM_F_INCORE; break; case 'g': + dbg_port = atoi(optarg); + break; + case 'G': + if (optarg[0] == 'w') { + gdb_stop = true; + optarg++; + } gdb_port = atoi(optarg); break; case 'l': @@ -1033,8 +1057,11 @@ main(int argc, char *argv[]) if (init_pci(ctx) != 0) exit(1); + if (dbg_port != 0) + init_dbgport(dbg_port); + if (gdb_port != 0) - init_dbgport(gdb_port); + init_gdb(ctx, gdb_port, gdb_stop); if (bvmcons) init_bvmcons(); diff --git a/usr.sbin/bhyve/gdb.c b/usr.sbin/bhyve/gdb.c new file mode 100644 index 000000000000..684eb2bbec03 --- /dev/null +++ b/usr.sbin/bhyve/gdb.c @@ -0,0 +1,1313 @@ +/*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * + * Copyright (c) 2017-2018 John H. Baldwin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#ifndef WITHOUT_CAPSICUM +#include +#endif +#include +#include +#include +#include +#include +#include +#include +#include +#ifndef WITHOUT_CAPSICUM +#include +#endif +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "bhyverun.h" +#include "mem.h" +#include "mevent.h" + +/* + * GDB_SIGNAL_* numbers are part of the GDB remote protocol. Most stops + * use SIGTRAP. + */ +#define GDB_SIGNAL_TRAP 5 + +static void gdb_resume_vcpus(void); +static void check_command(int fd); + +static struct mevent *read_event, *write_event; + +static cpuset_t vcpus_active, vcpus_suspended, vcpus_waiting; +static pthread_mutex_t gdb_lock; +static pthread_cond_t idle_vcpus; +static bool stop_pending, first_stop; +static int stepping_vcpu, stopped_vcpu; + +/* + * An I/O buffer contains 'capacity' bytes of room at 'data'. For a + * read buffer, 'start' is unused and 'len' contains the number of + * valid bytes in the buffer. For a write buffer, 'start' is set to + * the index of the next byte in 'data' to send, and 'len' contains + * the remaining number of valid bytes to send. + */ +struct io_buffer { + uint8_t *data; + size_t capacity; + size_t start; + size_t len; +}; + +static struct io_buffer cur_comm, cur_resp; +static uint8_t cur_csum; +static int cur_vcpu; +static struct vmctx *ctx; +static int cur_fd = -1; + +const int gdb_regset[] = { + VM_REG_GUEST_RAX, + VM_REG_GUEST_RBX, + VM_REG_GUEST_RCX, + VM_REG_GUEST_RDX, + VM_REG_GUEST_RSI, + VM_REG_GUEST_RDI, + VM_REG_GUEST_RBP, + VM_REG_GUEST_RSP, + VM_REG_GUEST_R8, + VM_REG_GUEST_R9, + VM_REG_GUEST_R10, + VM_REG_GUEST_R11, + VM_REG_GUEST_R12, + VM_REG_GUEST_R13, + VM_REG_GUEST_R14, + VM_REG_GUEST_R15, + VM_REG_GUEST_RIP, + VM_REG_GUEST_RFLAGS, + VM_REG_GUEST_CS, + VM_REG_GUEST_SS, + VM_REG_GUEST_DS, + VM_REG_GUEST_ES, + VM_REG_GUEST_FS, + VM_REG_GUEST_GS +}; + +const int gdb_regsize[] = { + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 8, + 4, + 4, + 4, + 4, + 4, + 4, + 4 +}; + +#ifdef GDB_LOG +#include +#include + +static void __printflike(1, 2) +debug(const char *fmt, ...) +{ + static FILE *logfile; + va_list ap; + + if (logfile == NULL) { + logfile = fopen("/tmp/bhyve_gdb.log", "w"); + if (logfile == NULL) + return; +#ifndef WITHOUT_CAPSICUM + if (caph_limit_stream(fileno(logfile), CAPH_WRITE) == -1) { + fclose(logfile); + logfile = NULL; + return; + } +#endif + setlinebuf(logfile); + } + va_start(ap, fmt); + vfprintf(logfile, fmt, ap); + va_end(ap); +} +#else +#define debug(...) +#endif + +static int +guest_paging_info(int vcpu, struct vm_guest_paging *paging) +{ + uint64_t regs[4]; + const int regset[4] = { + VM_REG_GUEST_CR0, + VM_REG_GUEST_CR3, + VM_REG_GUEST_CR4, + VM_REG_GUEST_EFER + }; + + if (vm_get_register_set(ctx, vcpu, nitems(regset), regset, regs) == -1) + return (-1); + + /* + * For the debugger, always pretend to be the kernel (CPL 0), + * and if long-mode is enabled, always parse addresses as if + * in 64-bit mode. + */ + paging->cr3 = regs[1]; + paging->cpl = 0; + if (regs[3] & EFER_LMA) + paging->cpu_mode = CPU_MODE_64BIT; + else if (regs[0] & CR0_PE) + paging->cpu_mode = CPU_MODE_PROTECTED; + else + paging->cpu_mode = CPU_MODE_REAL; + if (!(regs[0] & CR0_PG)) + paging->paging_mode = PAGING_MODE_FLAT; + else if (!(regs[2] & CR4_PAE)) + paging->paging_mode = PAGING_MODE_32; + else if (regs[3] & EFER_LME) + paging->paging_mode = PAGING_MODE_64; + else + paging->paging_mode = PAGING_MODE_PAE; + return (0); +} + +/* + * Map a guest virtual address to a physical address (for a given vcpu). + * If a guest virtual address is valid, return 1. If the address is + * not valid, return 0. If an error occurs obtaining the mapping, + * return -1. + */ +static int +guest_vaddr2paddr(int vcpu, uint64_t vaddr, uint64_t *paddr) +{ + struct vm_guest_paging paging; + int fault; + + if (guest_paging_info(vcpu, &paging) == -1) + return (-1); + + /* + * Always use PROT_READ. We really care if the VA is + * accessible, not if the current vCPU can write. + */ + if (vm_gla2gpa_nofault(ctx, vcpu, &paging, vaddr, PROT_READ, paddr, + &fault) == -1) + return (-1); + if (fault) + return (0); + return (1); +} + +static void +io_buffer_reset(struct io_buffer *io) +{ + + io->start = 0; + io->len = 0; +} + +/* Available room for adding data. */ +static size_t +io_buffer_avail(struct io_buffer *io) +{ + + return (io->capacity - (io->start + io->len)); +} + +static uint8_t * +io_buffer_head(struct io_buffer *io) +{ + + return (io->data + io->start); +} + +static uint8_t * +io_buffer_tail(struct io_buffer *io) +{ + + return (io->data + io->start + io->len); +} + +static void +io_buffer_advance(struct io_buffer *io, size_t amount) +{ + + assert(amount <= io->len); + io->start += amount; + io->len -= amount; +} + +static void +io_buffer_consume(struct io_buffer *io, size_t amount) +{ + + io_buffer_advance(io, amount); + if (io->len == 0) { + io->start = 0; + return; + } + + /* + * XXX: Consider making this move optional and compacting on a + * future read() before realloc(). + */ + memmove(io->data, io_buffer_head(io), io->len); + io->start = 0; +} + +static void +io_buffer_grow(struct io_buffer *io, size_t newsize) +{ + uint8_t *new_data; + size_t avail, new_cap; + + avail = io_buffer_avail(io); + if (newsize <= avail) + return; + + new_cap = io->capacity + (newsize - avail); + new_data = realloc(io->data, new_cap); + if (new_data == NULL) + err(1, "Failed to grow GDB I/O buffer"); + io->data = new_data; + io->capacity = new_cap; +} + +static bool +response_pending(void) +{ + + if (cur_resp.start == 0 && cur_resp.len == 0) + return (false); + if (cur_resp.start + cur_resp.len == 1 && cur_resp.data[0] == '+') + return (false); + return (true); +} + +static void +close_connection(void) +{ + + /* + * XXX: This triggers a warning because mevent does the close + * before the EV_DELETE. + */ + pthread_mutex_lock(&gdb_lock); + mevent_delete(write_event); + mevent_delete_close(read_event); + write_event = NULL; + read_event = NULL; + io_buffer_reset(&cur_comm); + io_buffer_reset(&cur_resp); + cur_fd = -1; + + /* Resume any stopped vCPUs. */ + gdb_resume_vcpus(); + pthread_mutex_unlock(&gdb_lock); +} + +static uint8_t +hex_digit(uint8_t nibble) +{ + + if (nibble <= 9) + return (nibble + '0'); + else + return (nibble + 'a' - 10); +} + +static uint8_t +parse_digit(uint8_t v) +{ + + if (v >= '0' && v <= '9') + return (v - '0'); + if (v >= 'a' && v <= 'f') + return (v - 'a' + 10); + if (v >= 'A' && v <= 'F') + return (v - 'A' + 10); + return (0xF); +} + +/* Parses big-endian hexadecimal. */ +static uintmax_t +parse_integer(const uint8_t *p, size_t len) +{ + uintmax_t v; + + v = 0; + while (len > 0) { + v <<= 4; + v |= parse_digit(*p); + p++; + len--; + } + return (v); +} + +static uint8_t +parse_byte(const uint8_t *p) +{ + + return (parse_digit(p[0]) << 4 | parse_digit(p[1])); +} + +static void +send_pending_data(int fd) +{ + ssize_t nwritten; + + if (cur_resp.len == 0) { + mevent_disable(write_event); + return; + } + nwritten = write(fd, io_buffer_head(&cur_resp), cur_resp.len); + if (nwritten == -1) { + warn("Write to GDB socket failed"); + close_connection(); + } else { + io_buffer_advance(&cur_resp, nwritten); + if (cur_resp.len == 0) + mevent_disable(write_event); + else + mevent_enable(write_event); + } +} + +/* Append a single character to the output buffer. */ +static void +send_char(uint8_t data) +{ + io_buffer_grow(&cur_resp, 1); + *io_buffer_tail(&cur_resp) = data; + cur_resp.len++; +} + +/* Append an array of bytes to the output buffer. */ +static void +send_data(const uint8_t *data, size_t len) +{ + + io_buffer_grow(&cur_resp, len); + memcpy(io_buffer_tail(&cur_resp), data, len); + cur_resp.len += len; +} + +static void +format_byte(uint8_t v, uint8_t *buf) +{ + + buf[0] = hex_digit(v >> 4); + buf[1] = hex_digit(v & 0xf); +} + +/* + * Append a single byte (formatted as two hex characters) to the + * output buffer. + */ +static void +send_byte(uint8_t v) +{ + uint8_t buf[2]; + + format_byte(v, buf); + send_data(buf, sizeof(buf)); +} + +static void +start_packet(void) +{ + + send_char('$'); + cur_csum = 0; +} + +static void +finish_packet(void) +{ + + send_char('#'); + send_byte(cur_csum); + debug("-> %.*s\n", (int)cur_resp.len, io_buffer_head(&cur_resp)); +} + +/* + * Append a single character (for the packet payload) and update the + * checksum. + */ +static void +append_char(uint8_t v) +{ + + send_char(v); + cur_csum += v; +} + +/* + * Append an array of bytes (for the packet payload) and update the + * checksum. + */ +static void +append_packet_data(const uint8_t *data, size_t len) +{ + + send_data(data, len); + while (len > 0) { + cur_csum += *data; + data++; + len--; + } +} + +static void +append_string(const char *str) +{ + + append_packet_data(str, strlen(str)); +} + +static void +append_byte(uint8_t v) +{ + uint8_t buf[2]; + + format_byte(v, buf); + append_packet_data(buf, sizeof(buf)); +} + +static void +append_unsigned_native(uintmax_t value, size_t len) +{ + size_t i; + + for (i = 0; i < len; i++) { + append_byte(value); + value >>= 8; + } +} + +static void +append_unsigned_be(uintmax_t value, size_t len) +{ + char buf[len * 2]; + size_t i; + + for (i = 0; i < len; i++) { + format_byte(value, buf + (len - i - 1) * 2); + value >>= 8; + } + append_packet_data(buf, sizeof(buf)); +} + +static void +append_integer(unsigned int value) +{ + + if (value == 0) + append_char('0'); + else + append_unsigned_be(value, fls(value) + 7 / 8); +} + +static void +append_asciihex(const char *str) +{ + + while (*str != '\0') { + append_byte(*str); + str++; + } +} + +static void +send_empty_response(void) +{ + + start_packet(); + finish_packet(); +} + +static void +send_error(int error) +{ + + start_packet(); + append_char('E'); + append_byte(error); + finish_packet(); +} + +static void +send_ok(void) +{ + + start_packet(); + append_string("OK"); + finish_packet(); +} + +static int +parse_threadid(const uint8_t *data, size_t len) +{ + + if (len == 1 && *data == '0') + return (0); + if (len == 2 && memcmp(data, "-1", 2) == 0) + return (-1); + if (len == 0) + return (-2); + return (parse_integer(data, len)); +} + +static void +report_stop(void) +{ + + start_packet(); + if (stopped_vcpu == -1) + append_char('S'); + else + append_char('T'); + append_byte(GDB_SIGNAL_TRAP); + if (stopped_vcpu != -1) { + append_string("thread:"); + append_integer(stopped_vcpu + 1); + append_char(';'); + } + stopped_vcpu = -1; + finish_packet(); +} + +static void +gdb_finish_suspend_vcpus(void) +{ + + if (first_stop) { + first_stop = false; + stopped_vcpu = -1; + } else if (response_pending()) + stop_pending = true; + else { + report_stop(); + send_pending_data(cur_fd); + } +} + +static void +_gdb_cpu_suspend(int vcpu, bool report_stop) +{ + + debug("$vCPU %d suspending\n", vcpu); + CPU_SET(vcpu, &vcpus_waiting); + if (report_stop && CPU_CMP(&vcpus_waiting, &vcpus_suspended) == 0) + gdb_finish_suspend_vcpus(); + while (CPU_ISSET(vcpu, &vcpus_suspended) && vcpu != stepping_vcpu) + pthread_cond_wait(&idle_vcpus, &gdb_lock); + CPU_CLR(vcpu, &vcpus_waiting); + debug("$vCPU %d resuming\n", vcpu); +} + +void +gdb_cpu_add(int vcpu) +{ + + debug("$vCPU %d starting\n", vcpu); + pthread_mutex_lock(&gdb_lock); + CPU_SET(vcpu, &vcpus_active); + + /* + * If a vcpu is added while vcpus are stopped, suspend the new + * vcpu so that it will pop back out with a debug exit before + * executing the first instruction. + */ + if (!CPU_EMPTY(&vcpus_suspended)) { + CPU_SET(vcpu, &vcpus_suspended); + _gdb_cpu_suspend(vcpu, false); + } + pthread_mutex_unlock(&gdb_lock); +} + +void +gdb_cpu_suspend(int vcpu) +{ + + pthread_mutex_lock(&gdb_lock); + _gdb_cpu_suspend(vcpu, true); + pthread_mutex_unlock(&gdb_lock); +} + +void +gdb_cpu_mtrap(int vcpu) +{ + + debug("$vCPU %d MTRAP\n", vcpu); + pthread_mutex_lock(&gdb_lock); + if (vcpu == stepping_vcpu) { + stepping_vcpu = -1; + vm_set_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, 0); + vm_suspend_cpu(ctx, vcpu); + assert(stopped_vcpu == -1); + stopped_vcpu = vcpu; + _gdb_cpu_suspend(vcpu, true); + } + pthread_mutex_unlock(&gdb_lock); +} + +static void +gdb_suspend_vcpus(void) +{ + + assert(pthread_mutex_isowned_np(&gdb_lock)); + debug("suspending all CPUs\n"); + vcpus_suspended = vcpus_active; + vm_suspend_cpu(ctx, -1); + if (CPU_CMP(&vcpus_waiting, &vcpus_suspended) == 0) + gdb_finish_suspend_vcpus(); +} + +static bool +gdb_step_vcpu(int vcpu) +{ + int error, val; + + debug("$vCPU %d step\n", vcpu); + error = vm_get_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, &val); + if (error < 0) + return (false); + error = vm_set_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, 1); + vm_resume_cpu(ctx, vcpu); + stepping_vcpu = vcpu; + pthread_cond_broadcast(&idle_vcpus); + return (true); +} + +static void +gdb_resume_vcpus(void) +{ + + assert(pthread_mutex_isowned_np(&gdb_lock)); + vm_resume_cpu(ctx, -1); + debug("resuming all CPUs\n"); + CPU_ZERO(&vcpus_suspended); + pthread_cond_broadcast(&idle_vcpus); +} + +static void +gdb_read_regs(void) +{ + uint64_t regvals[nitems(gdb_regset)]; + int i; + + if (vm_get_register_set(ctx, cur_vcpu, nitems(gdb_regset), + gdb_regset, regvals) == -1) { + send_error(errno); + return; + } + start_packet(); + for (i = 0; i < nitems(regvals); i++) + append_unsigned_native(regvals[i], gdb_regsize[i]); + finish_packet(); +} + +static void +gdb_read_mem(const uint8_t *data, size_t len) +{ + uint64_t gpa, gva, val; + uint8_t *cp; + size_t resid, todo, bytes; + bool started; + int error; + + cp = memchr(data, ',', len); + if (cp == NULL) { + send_error(EINVAL); + return; + } + gva = parse_integer(data + 1, cp - (data + 1)); + resid = parse_integer(cp + 1, len - (cp + 1 - data)); + started = false; + + while (resid > 0) { + error = guest_vaddr2paddr(cur_vcpu, gva, &gpa); + if (error == -1) { + if (started) + finish_packet(); + else + send_error(errno); + return; + } + if (error == 0) { + if (started) + finish_packet(); + else + send_error(EFAULT); + return; + } + + /* Read bytes from current page. */ + todo = getpagesize() - gpa % getpagesize(); + if (todo > resid) + todo = resid; + + cp = paddr_guest2host(ctx, gpa, todo); + if (cp != NULL) { + /* + * If this page is guest RAM, read it a byte + * at a time. + */ + if (!started) { + start_packet(); + started = true; + } + while (todo > 0) { + append_byte(*cp); + cp++; + gpa++; + gva++; + resid--; + todo--; + } + } else { + /* + * If this page isn't guest RAM, try to handle + * it via MMIO. For MMIO requests, use + * aligned reads of words when possible. + */ + while (todo > 0) { + if (gpa & 1 || todo == 1) + bytes = 1; + else if (gpa & 2 || todo == 2) + bytes = 2; + else + bytes = 4; + error = read_mem(ctx, cur_vcpu, gpa, &val, + bytes); + if (error == 0) { + if (!started) { + start_packet(); + started = true; + } + gpa += bytes; + gva += bytes; + resid -= bytes; + todo -= bytes; + while (bytes > 0) { + append_byte(val); + val >>= 8; + bytes--; + } + } else { + if (started) + finish_packet(); + else + send_error(EFAULT); + return; + } + } + } + assert(resid == 0 || gpa % getpagesize() == 0); + } + if (!started) + start_packet(); + finish_packet(); +} + +static bool +command_equals(const uint8_t *data, size_t len, const char *cmd) +{ + + if (strlen(cmd) > len) + return (false); + return (memcmp(data, cmd, strlen(cmd)) == 0); +} + +static void +gdb_query(const uint8_t *data, size_t len) +{ + + /* + * TODO: + * - qSearch + * - qSupported + */ + if (command_equals(data, len, "qAttached")) { + start_packet(); + append_char('1'); + finish_packet(); + } else if (command_equals(data, len, "qC")) { + start_packet(); + append_string("QC"); + append_integer(cur_vcpu + 1); + finish_packet(); + } else if (command_equals(data, len, "qfThreadInfo")) { + cpuset_t mask; + bool first; + int vcpu; + + if (CPU_EMPTY(&vcpus_active)) { + send_error(EINVAL); + return; + } + mask = vcpus_active; + start_packet(); + append_char('m'); + first = true; + while (!CPU_EMPTY(&mask)) { + vcpu = CPU_FFS(&mask) - 1; + CPU_CLR(vcpu, &mask); + if (first) + first = false; + else + append_char(','); + append_integer(vcpu + 1); + } + finish_packet(); + } else if (command_equals(data, len, "qsThreadInfo")) { + start_packet(); + append_char('l'); + finish_packet(); + } else if (command_equals(data, len, "qThreadExtraInfo")) { + char buf[16]; + int tid; + + data += strlen("qThreadExtraInfo"); + len -= strlen("qThreadExtraInfo"); + if (*data != ',') { + send_error(EINVAL); + return; + } + tid = parse_threadid(data + 1, len - 1); + if (tid <= 0 || !CPU_ISSET(tid - 1, &vcpus_active)) { + send_error(EINVAL); + return; + } + + snprintf(buf, sizeof(buf), "vCPU %d", tid - 1); + start_packet(); + append_asciihex(buf); + finish_packet(); + } else + send_empty_response(); +} + +static void +handle_command(const uint8_t *data, size_t len) +{ + + /* Reject packets with a sequence-id. */ + if (len >= 3 && data[0] >= '0' && data[0] <= '9' && + data[0] >= '0' && data[0] <= '9' && data[2] == ':') { + send_empty_response(); + return; + } + + switch (*data) { + case 'c': + if (len != 1) { + send_error(EINVAL); + break; + } + + /* Don't send a reply until a stop occurs. */ + gdb_resume_vcpus(); + break; + case 'D': + send_ok(); + + /* TODO: Resume any stopped CPUs. */ + break; + case 'g': { + gdb_read_regs(); + break; + } + case 'H': { + int tid; + + if (data[1] != 'g' && data[1] != 'c') { + send_error(EINVAL); + break; + } + tid = parse_threadid(data + 2, len - 2); + if (tid == -2) { + send_error(EINVAL); + break; + } + + if (CPU_EMPTY(&vcpus_active)) { + send_error(EINVAL); + break; + } + if (tid == -1 || tid == 0) + cur_vcpu = CPU_FFS(&vcpus_active) - 1; + else if (CPU_ISSET(tid - 1, &vcpus_active)) + cur_vcpu = tid - 1; + else { + send_error(EINVAL); + break; + } + send_ok(); + break; + } + case 'm': + gdb_read_mem(data, len); + break; + case 'T': { + int tid; + + tid = parse_threadid(data + 1, len - 1); + if (tid <= 0 || !CPU_ISSET(tid - 1, &vcpus_active)) { + send_error(EINVAL); + return; + } + send_ok(); + break; + } + case 'q': + gdb_query(data, len); + break; + case 's': + if (len != 1) { + send_error(EINVAL); + break; + } + + /* Don't send a reply until a stop occurs. */ + if (!gdb_step_vcpu(cur_vcpu)) { + send_error(EOPNOTSUPP); + break; + } + break; + case '?': + /* XXX: Only if stopped? */ + /* For now, just report that we are always stopped. */ + start_packet(); + append_char('S'); + append_byte(GDB_SIGNAL_TRAP); + finish_packet(); + break; + case 'G': /* TODO */ + case 'M': /* TODO */ + case 'v': + /* Handle 'vCont' */ + /* 'vCtrlC' */ + case 'p': /* TODO */ + case 'P': /* TODO */ + case 'Q': /* TODO */ + case 't': /* TODO */ + case 'X': /* TODO */ + case 'z': /* TODO */ + case 'Z': /* TODO */ + default: + send_empty_response(); + } +} + +/* Check for a valid packet in the command buffer. */ +static void +check_command(int fd) +{ + uint8_t *head, *hash, *p, sum; + size_t avail, plen; + + for (;;) { + avail = cur_comm.len; + if (avail == 0) + return; + head = io_buffer_head(&cur_comm); + switch (*head) { + case 0x03: + debug("<- Ctrl-C\n"); + io_buffer_consume(&cur_comm, 1); + + gdb_suspend_vcpus(); + break; + case '+': + /* ACK of previous response. */ + debug("<- +\n"); + if (response_pending()) + io_buffer_reset(&cur_resp); + io_buffer_consume(&cur_comm, 1); + if (stop_pending) { + stop_pending = false; + report_stop(); + send_pending_data(fd); + } + break; + case '-': + /* NACK of previous response. */ + debug("<- -\n"); + if (response_pending()) { + cur_resp.len += cur_resp.start; + cur_resp.start = 0; + if (cur_resp.data[0] == '+') + io_buffer_advance(&cur_resp, 1); + debug("-> %.*s\n", (int)cur_resp.len, + io_buffer_head(&cur_resp)); + } + io_buffer_consume(&cur_comm, 1); + send_pending_data(fd); + break; + case '$': + /* Packet. */ + + if (response_pending()) { + warnx("New GDB command while response in " + "progress"); + io_buffer_reset(&cur_resp); + } + + /* Is packet complete? */ + hash = memchr(head, '#', avail); + if (hash == NULL) + return; + plen = (hash - head + 1) + 2; + if (avail < plen) + return; + debug("<- %.*s\n", (int)plen, head); + + /* Verify checksum. */ + for (sum = 0, p = head + 1; p < hash; p++) + sum += *p; + if (sum != parse_byte(hash + 1)) { + io_buffer_consume(&cur_comm, plen); + debug("-> -\n"); + send_char('-'); + send_pending_data(fd); + break; + } + send_char('+'); + + handle_command(head + 1, hash - (head + 1)); + io_buffer_consume(&cur_comm, plen); + if (!response_pending()) + debug("-> +\n"); + send_pending_data(fd); + break; + default: + /* XXX: Possibly drop connection instead. */ + debug("-> %02x\n", *head); + io_buffer_consume(&cur_comm, 1); + break; + } + } +} + +static void +gdb_readable(int fd, enum ev_type event, void *arg) +{ + ssize_t nread; + int pending; + + if (ioctl(fd, FIONREAD, &pending) == -1) { + warn("FIONREAD on GDB socket"); + return; + } + + /* + * 'pending' might be zero due to EOF. We need to call read + * with a non-zero length to detect EOF. + */ + if (pending == 0) + pending = 1; + + /* Ensure there is room in the command buffer. */ + io_buffer_grow(&cur_comm, pending); + assert(io_buffer_avail(&cur_comm) >= pending); + + nread = read(fd, io_buffer_tail(&cur_comm), io_buffer_avail(&cur_comm)); + if (nread == 0) { + close_connection(); + } else if (nread == -1) { + if (errno == EAGAIN) + return; + + warn("Read from GDB socket"); + close_connection(); + } else { + cur_comm.len += nread; + pthread_mutex_lock(&gdb_lock); + check_command(fd); + pthread_mutex_unlock(&gdb_lock); + } +} + +static void +gdb_writable(int fd, enum ev_type event, void *arg) +{ + + send_pending_data(fd); +} + +static void +new_connection(int fd, enum ev_type event, void *arg) +{ + int optval, s; + + s = accept4(fd, NULL, NULL, SOCK_NONBLOCK); + if (s == -1) { + if (arg != NULL) + err(1, "Failed accepting initial GDB connection"); + + /* Silently ignore errors post-startup. */ + return; + } + + optval = 1; + if (setsockopt(s, SOL_SOCKET, SO_NOSIGPIPE, &optval, sizeof(optval)) == + -1) { + warn("Failed to disable SIGPIPE for GDB connection"); + close(s); + return; + } + + pthread_mutex_lock(&gdb_lock); + if (cur_fd != -1) { + close(s); + warnx("Ignoring additional GDB connection."); + } + + read_event = mevent_add(s, EVF_READ, gdb_readable, NULL); + if (read_event == NULL) { + if (arg != NULL) + err(1, "Failed to setup initial GDB connection"); + pthread_mutex_unlock(&gdb_lock); + return; + } + write_event = mevent_add(s, EVF_WRITE, gdb_writable, NULL); + if (write_event == NULL) { + if (arg != NULL) + err(1, "Failed to setup initial GDB connection"); + mevent_delete_close(read_event); + read_event = NULL; + } + + cur_fd = s; + cur_vcpu = 0; + stepping_vcpu = -1; + stopped_vcpu = -1; + stop_pending = false; + + /* Break on attach. */ + first_stop = true; + gdb_suspend_vcpus(); + pthread_mutex_unlock(&gdb_lock); +} + +#ifndef WITHOUT_CAPSICUM +void +limit_gdb_socket(int s) +{ + cap_rights_t rights; + unsigned long ioctls[] = { FIONREAD }; + + cap_rights_init(&rights, CAP_ACCEPT, CAP_EVENT, CAP_READ, CAP_WRITE, + CAP_SETSOCKOPT, CAP_IOCTL); + if (cap_rights_limit(s, &rights) == -1 && errno != ENOSYS) + errx(EX_OSERR, "Unable to apply rights for sandbox"); + if (cap_ioctls_limit(s, ioctls, nitems(ioctls)) == -1 && errno != ENOSYS) + errx(EX_OSERR, "Unable to apply rights for sandbox"); +} +#endif + +void +init_gdb(struct vmctx *_ctx, int sport, bool wait) +{ + struct sockaddr_in sin; + int error, flags, s; + + debug("==> starting on %d, %swaiting\n", sport, wait ? "" : "not "); + + error = pthread_mutex_init(&gdb_lock, NULL); + if (error != 0) + errc(1, error, "gdb mutex init"); + error = pthread_cond_init(&idle_vcpus, NULL); + if (error != 0) + errc(1, error, "gdb cv init"); + + ctx = _ctx; + s = socket(PF_INET, SOCK_STREAM, 0); + if (s < 0) + err(1, "gdb socket create"); + + sin.sin_len = sizeof(sin); + sin.sin_family = AF_INET; + sin.sin_addr.s_addr = htonl(INADDR_ANY); + sin.sin_port = htons(sport); + + if (bind(s, (struct sockaddr *)&sin, sizeof(sin)) < 0) + err(1, "gdb socket bind"); + + if (listen(s, 1) < 0) + err(1, "gdb socket listen"); + + if (wait) { + /* + * Set vcpu 0 in vcpus_suspended. This will trigger the + * logic in gdb_cpu_add() to suspend the first vcpu before + * it starts execution. The vcpu will remain suspended + * until a debugger connects. + */ + stepping_vcpu = -1; + stopped_vcpu = -1; + CPU_SET(0, &vcpus_suspended); + } + + flags = fcntl(s, F_GETFL); + if (fcntl(s, F_SETFL, flags | O_NONBLOCK) == -1) + err(1, "Failed to mark gdb socket non-blocking"); + +#ifndef WITHOUT_CAPSICUM + limit_gdb_socket(s); +#endif + mevent_add(s, EVF_READ, new_connection, NULL); +} diff --git a/usr.sbin/bhyve/gdb.h b/usr.sbin/bhyve/gdb.h new file mode 100644 index 000000000000..fa2184df1667 --- /dev/null +++ b/usr.sbin/bhyve/gdb.h @@ -0,0 +1,39 @@ +/*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * + * Copyright (c) 2017 John H. Baldwin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef __GDB_H__ +#define __GDB_H__ + +void gdb_cpu_add(int vcpu); +void gdb_cpu_mtrap(int vcpu); +void gdb_cpu_suspend(int vcpu); +void init_gdb(struct vmctx *ctx, int sport, bool wait); + +#endif /* !__GDB_H__ */ diff --git a/usr.sbin/bhyve/mem.c b/usr.sbin/bhyve/mem.c index e01d617a895f..105d37cf56e5 100644 --- a/usr.sbin/bhyve/mem.c +++ b/usr.sbin/bhyve/mem.c @@ -136,6 +136,9 @@ mmio_rb_dump(struct mmio_rb_tree *rbt) RB_GENERATE(mmio_rb_tree, mmio_rb_range, mr_link, mmio_rb_range_compare); +typedef int (mem_cb_t)(struct vmctx *ctx, int vcpu, uint64_t gpa, + struct mem_range *mr, void *arg); + static int mem_read(void *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size, void *arg) { @@ -158,10 +161,9 @@ mem_write(void *ctx, int vcpu, uint64_t gpa, uint64_t wval, int size, void *arg) return (error); } -int -emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie, - struct vm_guest_paging *paging) - +static int +access_memory(struct vmctx *ctx, int vcpu, uint64_t paddr, mem_cb_t *cb, + void *arg) { struct mmio_rb_range *entry; int err, immutable; @@ -204,8 +206,7 @@ emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie, if (immutable) pthread_rwlock_unlock(&mmio_rwlock); - err = vmm_emulate_instruction(ctx, vcpu, paddr, vie, paging, - mem_read, mem_write, &entry->mr_param); + err = cb(ctx, vcpu, paddr, &entry->mr_param, arg); if (!immutable) pthread_rwlock_unlock(&mmio_rwlock); @@ -213,6 +214,60 @@ emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie, return (err); } +struct emulate_mem_args { + struct vie *vie; + struct vm_guest_paging *paging; +}; + +static int +emulate_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr, + void *arg) +{ + struct emulate_mem_args *ema; + + ema = arg; + return (vmm_emulate_instruction(ctx, vcpu, paddr, ema->vie, ema->paging, + mem_read, mem_write, mr)); +} + +int +emulate_mem(struct vmctx *ctx, int vcpu, uint64_t paddr, struct vie *vie, + struct vm_guest_paging *paging) + +{ + struct emulate_mem_args ema; + + ema.vie = vie; + ema.paging = paging; + return (access_memory(ctx, vcpu, paddr, emulate_mem_cb, &ema)); +} + +struct read_mem_args { + uint64_t *rval; + int size; +}; + +static int +read_mem_cb(struct vmctx *ctx, int vcpu, uint64_t paddr, struct mem_range *mr, + void *arg) +{ + struct read_mem_args *rma; + + rma = arg; + return (mr->handler(ctx, vcpu, MEM_F_READ, paddr, rma->size, + rma->rval, mr->arg1, mr->arg2)); +} + +int +read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, int size) +{ + struct read_mem_args rma; + + rma.rval = rval; + rma.size = size; + return (access_memory(ctx, vcpu, gpa, read_mem_cb, &rma)); +} + static int register_mem_int(struct mmio_rb_tree *rbt, struct mem_range *memp) { diff --git a/usr.sbin/bhyve/mem.h b/usr.sbin/bhyve/mem.h index f9f86fa4a004..f386d67749db 100644 --- a/usr.sbin/bhyve/mem.h +++ b/usr.sbin/bhyve/mem.h @@ -56,6 +56,8 @@ void init_mem(void); int emulate_mem(struct vmctx *, int vcpu, uint64_t paddr, struct vie *vie, struct vm_guest_paging *paging); +int read_mem(struct vmctx *ctx, int vcpu, uint64_t gpa, uint64_t *rval, + int size); int register_mem(struct mem_range *memp); int register_mem_fallback(struct mem_range *memp); int unregister_mem(struct mem_range *memp);