blob: d9da7e148538286d78d6f11236f3a4db89218f45 [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:100
2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5 * control it.
6:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07007#define _LARGEFILE64_SOURCE
8#define _GNU_SOURCE
9#include <stdio.h>
10#include <string.h>
11#include <unistd.h>
12#include <err.h>
13#include <stdint.h>
14#include <stdlib.h>
15#include <elf.h>
16#include <sys/mman.h>
Ronald G. Minnich6649bb72007-08-28 14:35:59 -070017#include <sys/param.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070018#include <sys/types.h>
19#include <sys/stat.h>
20#include <sys/wait.h>
Rusty Russell659a0e62009-06-12 22:27:10 -060021#include <sys/eventfd.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070022#include <fcntl.h>
23#include <stdbool.h>
24#include <errno.h>
25#include <ctype.h>
26#include <sys/socket.h>
27#include <sys/ioctl.h>
28#include <sys/time.h>
29#include <time.h>
30#include <netinet/in.h>
31#include <net/if.h>
32#include <linux/sockios.h>
33#include <linux/if_tun.h>
34#include <sys/uio.h>
35#include <termios.h>
36#include <getopt.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100037#include <assert.h>
38#include <sched.h>
Rusty Russella586d4f2008-02-04 23:49:56 -050039#include <limits.h>
40#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050041#include <signal.h>
Philip Sanderson8aeb36e2011-01-20 21:37:28 -060042#include <pwd.h>
43#include <grp.h>
44
Rusty Russellf8466192010-08-27 08:39:48 -060045#include <linux/virtio_config.h>
46#include <linux/virtio_net.h>
47#include <linux/virtio_blk.h>
48#include <linux/virtio_console.h>
49#include <linux/virtio_rng.h>
50#include <linux/virtio_ring.h>
51#include <asm/bootparam.h>
52#include "../../include/linux/lguest_launcher.h"
Rusty Russell2e04ef72009-07-30 16:03:45 -060053/*L:110
Rusty Russella91d74a2009-07-30 16:03:45 -060054 * We can ignore the 42 include files we need for this program, but I do want
Rusty Russell2e04ef72009-07-30 16:03:45 -060055 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100056 *
57 * As Linus said, "C is a Spartan language, and so should your naming be." I
58 * like these abbreviations, so we define them here. Note that u64 is always
59 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060060 * use %llu in printf for any u64.
61 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100062typedef unsigned long long u64;
63typedef uint32_t u32;
64typedef uint16_t u16;
65typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070066/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070067
68#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
Rusty Russell8ca47e02007-07-19 01:49:29 -070069#define BRIDGE_PFX "bridge:"
70#ifndef SIOCBRADDIF
71#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
72#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100073/* We can have up to 256 pages for devices. */
74#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050075/* This will occupy 3 pages: it must be a power of 2. */
76#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070077
Rusty Russell2e04ef72009-07-30 16:03:45 -060078/*L:120
79 * verbose is both a global flag and a macro. The C preprocessor allows
80 * this, and although I wouldn't recommend it, it works quite nicely here.
81 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070082static bool verbose;
83#define verbose(args...) \
84 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070085/*:*/
86
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100087/* The pointer to the start of guest memory. */
88static void *guest_base;
89/* The maximum guest physical address allowed, and maximum possible. */
90static unsigned long guest_limit, guest_max;
Rusty Russell56739c802009-06-12 22:26:59 -060091/* The /dev/lguest file descriptor. */
92static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070093
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020094/* a per-cpu variable indicating whose vcpu is currently running */
95static unsigned int __thread cpu_id;
96
Rusty Russelldde79782007-07-26 10:41:03 -070097/* This is our list of devices. */
Rusty Russell1842f232009-07-30 16:03:46 -060098struct device_list {
Rusty Russell17cbca22007-10-22 11:24:22 +100099 /* Counter to assign interrupt numbers. */
100 unsigned int next_irq;
101
102 /* Counter to print out convenient device numbers. */
103 unsigned int device_num;
104
Rusty Russelldde79782007-07-26 10:41:03 -0700105 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000106 u8 *descpage;
107
Rusty Russelldde79782007-07-26 10:41:03 -0700108 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700109 struct device *dev;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600110 /* And a pointer to the last device for easy append. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500111 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700112};
113
Rusty Russell17cbca22007-10-22 11:24:22 +1000114/* The list of Guest devices, based on command line arguments. */
115static struct device_list devices;
116
Rusty Russelldde79782007-07-26 10:41:03 -0700117/* The device structure describes a single device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600118struct device {
Rusty Russelldde79782007-07-26 10:41:03 -0700119 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700120 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000121
Rusty Russell713b15b2009-06-12 22:26:58 -0600122 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700123 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000124
Rusty Russell713b15b2009-06-12 22:26:58 -0600125 /* We can't trust desc values once Guest has booted: we use these. */
126 unsigned int feature_len;
127 unsigned int num_vq;
128
Rusty Russell17cbca22007-10-22 11:24:22 +1000129 /* The name of this device, for --verbose. */
130 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700131
Rusty Russell17cbca22007-10-22 11:24:22 +1000132 /* Any queues attached to this device */
133 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700134
Rusty Russell659a0e62009-06-12 22:27:10 -0600135 /* Is it operational */
136 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500137
Rusty Russellca60a422009-09-23 22:26:47 -0600138 /* Does Guest want an intrrupt on empty? */
139 bool irq_on_empty;
140
Rusty Russell8ca47e02007-07-19 01:49:29 -0700141 /* Device-specific data. */
142 void *priv;
143};
144
Rusty Russell17cbca22007-10-22 11:24:22 +1000145/* The virtqueue structure describes a queue attached to a device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600146struct virtqueue {
Rusty Russell17cbca22007-10-22 11:24:22 +1000147 struct virtqueue *next;
148
149 /* Which device owns me. */
150 struct device *dev;
151
152 /* The configuration for this queue. */
153 struct lguest_vqconfig config;
154
155 /* The actual ring of buffers. */
156 struct vring vring;
157
158 /* Last available index we saw. */
159 u16 last_avail_idx;
160
Rusty Russell95c517c2009-06-12 22:27:11 -0600161 /* How many are used since we sent last irq? */
162 unsigned int pending_used;
163
Rusty Russell659a0e62009-06-12 22:27:10 -0600164 /* Eventfd where Guest notifications arrive. */
165 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500166
Rusty Russell659a0e62009-06-12 22:27:10 -0600167 /* Function for the thread which is servicing this virtqueue. */
168 void (*service)(struct virtqueue *vq);
169 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000170};
171
Balaji Raoec04b132007-12-28 14:26:24 +0530172/* Remember the arguments to the program so we can "reboot" */
173static char **main_args;
174
Rusty Russell659a0e62009-06-12 22:27:10 -0600175/* The original tty settings to restore on exit. */
176static struct termios orig_term;
177
Rusty Russell2e04ef72009-07-30 16:03:45 -0600178/*
179 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600180 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600181 * in precise order.
182 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600183#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russellb60da132009-06-12 22:27:12 -0600184#define mb() __asm__ __volatile__("" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000185
Rusty Russell2e04ef72009-07-30 16:03:45 -0600186/*
187 * Convert an iovec element to the given type.
Rusty Russell17cbca22007-10-22 11:24:22 +1000188 *
189 * This is a fairly ugly trick: we need to know the size of the type and
190 * alignment requirement to check the pointer is kosher. It's also nice to
191 * have the name of the type in case we report failure.
192 *
193 * Typing those three things all the time is cumbersome and error prone, so we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600194 * have a macro which sets them all up and passes to the real function.
195 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000196#define convert(iov, type) \
197 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
198
199static void *_convert(struct iovec *iov, size_t size, size_t align,
200 const char *name)
201{
202 if (iov->iov_len != size)
203 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
204 if ((unsigned long)iov->iov_base % align != 0)
205 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
206 return iov->iov_base;
207}
208
Rusty Russellb5111792008-07-29 09:58:34 -0500209/* Wrapper for the last available index. Makes it easier to change. */
210#define lg_last_avail(vq) ((vq)->last_avail_idx)
211
Rusty Russell2e04ef72009-07-30 16:03:45 -0600212/*
213 * The virtio configuration space is defined to be little-endian. x86 is
214 * little-endian too, but it's nice to be explicit so we have these helpers.
215 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000216#define cpu_to_le16(v16) (v16)
217#define cpu_to_le32(v32) (v32)
218#define cpu_to_le64(v64) (v64)
219#define le16_to_cpu(v16) (v16)
220#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500221#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000222
Rusty Russell28fd6d72008-07-29 09:58:33 -0500223/* Is this iovec empty? */
224static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
225{
226 unsigned int i;
227
228 for (i = 0; i < num_iov; i++)
229 if (iov[i].iov_len)
230 return false;
231 return true;
232}
233
234/* Take len bytes from the front of this iovec. */
235static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
236{
237 unsigned int i;
238
239 for (i = 0; i < num_iov; i++) {
240 unsigned int used;
241
242 used = iov[i].iov_len < len ? iov[i].iov_len : len;
243 iov[i].iov_base += used;
244 iov[i].iov_len -= used;
245 len -= used;
246 }
247 assert(len == 0);
248}
249
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500250/* The device virtqueue descriptors are followed by feature bitmasks. */
251static u8 *get_feature_bits(struct device *dev)
252{
253 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600254 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500255}
256
Rusty Russell2e04ef72009-07-30 16:03:45 -0600257/*L:100
258 * The Launcher code itself takes us out into userspace, that scary place where
259 * pointers run wild and free! Unfortunately, like most userspace programs,
260 * it's quite boring (which is why everyone likes to hack on the kernel!).
261 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
262 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000263 *
264 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
265 * memory and stores it in "guest_base". In other words, Guest physical ==
266 * Launcher virtual with an offset.
267 *
268 * This can be tough to get your head around, but usually it just means that we
Francis Galieguea33f3222010-04-23 00:08:02 +0200269 * use these trivial conversion functions when the Guest gives us its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600270 * "physical" addresses:
271 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000272static void *from_guest_phys(unsigned long addr)
273{
274 return guest_base + addr;
275}
276
277static unsigned long to_guest_phys(const void *addr)
278{
279 return (addr - guest_base);
280}
281
Rusty Russelldde79782007-07-26 10:41:03 -0700282/*L:130
283 * Loading the Kernel.
284 *
285 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600286 * error-checking code cluttering the callers:
287 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700288static int open_or_die(const char *name, int flags)
289{
290 int fd = open(name, flags);
291 if (fd < 0)
292 err(1, "Failed to open %s", name);
293 return fd;
294}
295
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000296/* map_zeroed_pages() takes a number of pages. */
297static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700298{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000299 int fd = open_or_die("/dev/zero", O_RDONLY);
300 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700301
Rusty Russell2e04ef72009-07-30 16:03:45 -0600302 /*
303 * We use a private mapping (ie. if we write to the page, it will be
Philip Sanderson5230ff02011-01-20 21:37:28 -0600304 * copied). We allocate an extra two pages PROT_NONE to act as guard
305 * pages against read/write attempts that exceed allocated space.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600306 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600307 addr = mmap(NULL, getpagesize() * (num+2),
308 PROT_NONE, MAP_PRIVATE, fd, 0);
309
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000310 if (addr == MAP_FAILED)
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200311 err(1, "Mmapping %u pages of /dev/zero", num);
Rusty Russella91d74a2009-07-30 16:03:45 -0600312
Philip Sanderson5230ff02011-01-20 21:37:28 -0600313 if (mprotect(addr + getpagesize(), getpagesize() * num,
314 PROT_READ|PROT_WRITE) == -1)
315 err(1, "mprotect rw %u pages failed", num);
316
Rusty Russella91d74a2009-07-30 16:03:45 -0600317 /*
318 * One neat mmap feature is that you can close the fd, and it
319 * stays mapped.
320 */
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100321 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700322
Philip Sanderson5230ff02011-01-20 21:37:28 -0600323 /* Return address after PROT_NONE page */
324 return addr + getpagesize();
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000325}
326
327/* Get some more pages for a device. */
328static void *get_pages(unsigned int num)
329{
330 void *addr = from_guest_phys(guest_limit);
331
332 guest_limit += num * getpagesize();
333 if (guest_limit > guest_max)
334 errx(1, "Not enough memory for devices");
335 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700336}
337
Rusty Russell2e04ef72009-07-30 16:03:45 -0600338/*
339 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700340 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600341 * it falls back to reading the memory in.
342 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700343static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
344{
345 ssize_t r;
346
Rusty Russell2e04ef72009-07-30 16:03:45 -0600347 /*
348 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700349 * The kernel really wants to be writable: it patches its own
350 * instructions.
351 *
352 * MAP_PRIVATE means that the page won't be copied until a write is
353 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600354 * Guests.
355 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600356 if (mmap(addr, len, PROT_READ|PROT_WRITE,
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700357 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
358 return;
359
360 /* pread does a seek and a read in one shot: saves a few lines. */
361 r = pread(fd, addr, len, offset);
362 if (r != len)
363 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
364}
365
Rusty Russell2e04ef72009-07-30 16:03:45 -0600366/*
367 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700368 * the Guest memory. ELF = Embedded Linking Format, which is the format used
369 * by all modern binaries on Linux including the kernel.
370 *
371 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000372 * address. We use the physical address; the Guest will map itself to the
373 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700374 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600375 * We return the starting address.
376 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000377static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700378{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700379 Elf32_Phdr phdr[ehdr->e_phnum];
380 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700381
Rusty Russell2e04ef72009-07-30 16:03:45 -0600382 /*
383 * Sanity checks on the main ELF header: an x86 executable with a
384 * reasonable number of correctly-sized program headers.
385 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700386 if (ehdr->e_type != ET_EXEC
387 || ehdr->e_machine != EM_386
388 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
389 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
390 errx(1, "Malformed elf header");
391
Rusty Russell2e04ef72009-07-30 16:03:45 -0600392 /*
393 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700394 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600395 * load where.
396 */
Rusty Russelldde79782007-07-26 10:41:03 -0700397
398 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700399 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
400 err(1, "Seeking to program headers");
401 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
402 err(1, "Reading program headers");
403
Rusty Russell2e04ef72009-07-30 16:03:45 -0600404 /*
405 * Try all the headers: there are usually only three. A read-only one,
406 * a read-write one, and a "note" section which we don't load.
407 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700408 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700409 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700410 if (phdr[i].p_type != PT_LOAD)
411 continue;
412
413 verbose("Section %i: size %i addr %p\n",
414 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
415
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700416 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000417 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700418 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700419 }
420
Rusty Russell814a0e52007-10-22 11:29:44 +1000421 /* The entry point is given in the ELF header. */
422 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700423}
424
Rusty Russell2e04ef72009-07-30 16:03:45 -0600425/*L:150
426 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
427 * to jump into it and it will unpack itself. We used to have to perform some
428 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700429 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000430 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
431 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600432 * the funky header so we know where in the file to load, and away we go!
433 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000434static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700435{
Rusty Russell43d33b22007-10-22 11:29:57 +1000436 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000437 int r;
438 /* Modern bzImages get loaded at 1M. */
439 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700440
Rusty Russell2e04ef72009-07-30 16:03:45 -0600441 /*
442 * Go back to the start of the file and read the header. It should be
443 * a Linux boot header (see Documentation/x86/i386/boot.txt)
444 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000445 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000446 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000447
Rusty Russell43d33b22007-10-22 11:29:57 +1000448 /* Inside the setup_hdr, we expect the magic "HdrS" */
449 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000450 errx(1, "This doesn't look like a bzImage to me");
451
Rusty Russell43d33b22007-10-22 11:29:57 +1000452 /* Skip over the extra sectors of the header. */
453 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000454
455 /* Now read everything into memory. in nice big chunks. */
456 while ((r = read(fd, p, 65536)) > 0)
457 p += r;
458
Rusty Russell43d33b22007-10-22 11:29:57 +1000459 /* Finally, code32_start tells us where to enter the kernel. */
460 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700461}
462
Rusty Russell2e04ef72009-07-30 16:03:45 -0600463/*L:140
464 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000465 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600466 * work, we can load those, too.
467 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000468static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700469{
470 Elf32_Ehdr hdr;
471
Rusty Russelldde79782007-07-26 10:41:03 -0700472 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700473 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
474 err(1, "Reading kernel");
475
Rusty Russelldde79782007-07-26 10:41:03 -0700476 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700477 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000478 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700479
Rusty Russella6bd8e12008-03-28 11:05:53 -0500480 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000481 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700482}
483
Rusty Russell2e04ef72009-07-30 16:03:45 -0600484/*
485 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700486 * it calls getpagesize() twice: "it's dumb code."
487 *
488 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600489 * necessary. I leave this code as a reaction against that.
490 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700491static inline unsigned long page_align(unsigned long addr)
492{
Rusty Russelldde79782007-07-26 10:41:03 -0700493 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700494 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
495}
496
Rusty Russell2e04ef72009-07-30 16:03:45 -0600497/*L:180
498 * An "initial ram disk" is a disk image loaded into memory along with the
499 * kernel which the kernel can use to boot from without needing any drivers.
500 * Most distributions now use this as standard: the initrd contains the code to
501 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700502 *
503 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600504 * kernels. He sent me this (and tells me when I break it).
505 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700506static unsigned long load_initrd(const char *name, unsigned long mem)
507{
508 int ifd;
509 struct stat st;
510 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700511
512 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700513 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700514 if (fstat(ifd, &st) < 0)
515 err(1, "fstat() on initrd '%s'", name);
516
Rusty Russell2e04ef72009-07-30 16:03:45 -0600517 /*
518 * We map the initrd at the top of memory, but mmap wants it to be
519 * page-aligned, so we round the size up for that.
520 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700521 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000522 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600523 /*
524 * Once a file is mapped, you can close the file descriptor. It's a
525 * little odd, but quite useful.
526 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700527 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700528 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700529
530 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700531 return len;
532}
Rusty Russelle1e72962007-10-25 15:02:50 +1000533/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700534
Rusty Russell2e04ef72009-07-30 16:03:45 -0600535/*
536 * Simple routine to roll all the commandline arguments together with spaces
537 * between them.
538 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700539static void concat(char *dst, char *args[])
540{
541 unsigned int i, len = 0;
542
543 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100544 if (i) {
545 strcat(dst+len, " ");
546 len++;
547 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700548 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100549 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700550 }
551 /* In case it's empty. */
552 dst[len] = '\0';
553}
554
Rusty Russell2e04ef72009-07-30 16:03:45 -0600555/*L:185
556 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000557 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300558 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600559 * entry point for the Guest.
560 */
Rusty Russell56739c802009-06-12 22:26:59 -0600561static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700562{
Jes Sorensen511801d2007-10-22 11:03:31 +1000563 unsigned long args[] = { LHREQ_INITIALIZE,
564 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300565 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000566 verbose("Guest: %p - %p (%#lx)\n",
567 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c802009-06-12 22:26:59 -0600568 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
569 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700570 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700571}
Rusty Russelldde79782007-07-26 10:41:03 -0700572/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700573
Rusty Russella91d74a2009-07-30 16:03:45 -0600574/*L:200
Rusty Russelldde79782007-07-26 10:41:03 -0700575 * Device Handling.
576 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000577 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700578 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000579 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700580 * if something funny is going on:
581 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700582static void *_check_pointer(unsigned long addr, unsigned int size,
583 unsigned int line)
584{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600585 /*
Philip Sanderson5230ff02011-01-20 21:37:28 -0600586 * Check if the requested address and size exceeds the allocated memory,
587 * or addr + size wraps around.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600588 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600589 if ((addr + size) > guest_limit || (addr + size) < addr)
Rusty Russell17cbca22007-10-22 11:24:22 +1000590 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600591 /*
592 * We return a pointer for the caller's convenience, now we know it's
593 * safe to use.
594 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000595 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700596}
Rusty Russelldde79782007-07-26 10:41:03 -0700597/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700598#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
599
Rusty Russell2e04ef72009-07-30 16:03:45 -0600600/*
601 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000602 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600603 * at the end.
604 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100605static unsigned next_desc(struct vring_desc *desc,
606 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000607{
608 unsigned int next;
609
610 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100611 if (!(desc[i].flags & VRING_DESC_F_NEXT))
612 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000613
614 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100615 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000616 /* Make sure compiler knows to grab that: we don't want it changing! */
617 wmb();
618
Mark McLoughlind1f01322009-05-11 18:11:46 +0100619 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000620 errx(1, "Desc next is %u", next);
621
622 return next;
623}
624
Rusty Russella91d74a2009-07-30 16:03:45 -0600625/*
626 * This actually sends the interrupt for this virtqueue, if we've used a
627 * buffer.
628 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600629static void trigger_irq(struct virtqueue *vq)
630{
631 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
632
Rusty Russell95c517c2009-06-12 22:27:11 -0600633 /* Don't inform them if nothing used. */
634 if (!vq->pending_used)
635 return;
636 vq->pending_used = 0;
637
Rusty Russellca60a422009-09-23 22:26:47 -0600638 /* If they don't want an interrupt, don't send one... */
639 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
640 /* ... unless they've asked us to force one on empty. */
641 if (!vq->dev->irq_on_empty
642 || lg_last_avail(vq) != vq->vring.avail->idx)
643 return;
644 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600645
646 /* Send the Guest an interrupt tell them we used something up. */
647 if (write(lguest_fd, buf, sizeof(buf)) != 0)
648 err(1, "Triggering irq %i", vq->config.irq);
649}
650
Rusty Russell2e04ef72009-07-30 16:03:45 -0600651/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600652 * This looks in the virtqueue for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000653 * it to an iovec for convenient access. Since descriptors consist of some
654 * number of output then some number of input descriptors, it's actually two
655 * iovecs, but we pack them into one and note how many of each there were.
656 *
Rusty Russella91d74a2009-07-30 16:03:45 -0600657 * This function waits if necessary, and returns the descriptor number found.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600658 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600659static unsigned wait_for_vq_desc(struct virtqueue *vq,
660 struct iovec iov[],
661 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000662{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100663 unsigned int i, head, max;
664 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600665 u16 last_avail = lg_last_avail(vq);
666
Rusty Russella91d74a2009-07-30 16:03:45 -0600667 /* There's nothing available? */
Rusty Russell659a0e62009-06-12 22:27:10 -0600668 while (last_avail == vq->vring.avail->idx) {
669 u64 event;
670
Rusty Russella91d74a2009-07-30 16:03:45 -0600671 /*
672 * Since we're about to sleep, now is a good time to tell the
673 * Guest about what we've used up to now.
674 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600675 trigger_irq(vq);
676
Rusty Russellb60da132009-06-12 22:27:12 -0600677 /* OK, now we need to know about added descriptors. */
678 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
679
Rusty Russell2e04ef72009-07-30 16:03:45 -0600680 /*
681 * They could have slipped one in as we were doing that: make
682 * sure it's written, then check again.
683 */
Rusty Russellb60da132009-06-12 22:27:12 -0600684 mb();
685 if (last_avail != vq->vring.avail->idx) {
686 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
687 break;
688 }
689
Rusty Russell659a0e62009-06-12 22:27:10 -0600690 /* Nothing new? Wait for eventfd to tell us they refilled. */
691 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
692 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600693
694 /* We don't need to be notified again. */
695 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600696 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000697
698 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500699 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000700 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500701 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000702
Rusty Russell2e04ef72009-07-30 16:03:45 -0600703 /*
704 * Grab the next descriptor number they're advertising, and increment
705 * the index we've seen.
706 */
Rusty Russellb5111792008-07-29 09:58:34 -0500707 head = vq->vring.avail->ring[last_avail % vq->vring.num];
708 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000709
710 /* If their number is silly, that's a fatal mistake. */
711 if (head >= vq->vring.num)
712 errx(1, "Guest says index %u is available", head);
713
714 /* When we start there are none of either input nor output. */
715 *out_num = *in_num = 0;
716
Mark McLoughlind1f01322009-05-11 18:11:46 +0100717 max = vq->vring.num;
718 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000719 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100720
Rusty Russell2e04ef72009-07-30 16:03:45 -0600721 /*
722 * If this is an indirect entry, then this buffer contains a descriptor
723 * table which we handle as if it's any normal descriptor chain.
724 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100725 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
726 if (desc[i].len % sizeof(struct vring_desc))
727 errx(1, "Invalid size for indirect buffer table");
728
729 max = desc[i].len / sizeof(struct vring_desc);
730 desc = check_pointer(desc[i].addr, desc[i].len);
731 i = 0;
732 }
733
Rusty Russell17cbca22007-10-22 11:24:22 +1000734 do {
735 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100736 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000737 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100738 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000739 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100740 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000741 (*in_num)++;
742 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600743 /*
744 * If it's an output descriptor, they're all supposed
745 * to come before any input descriptors.
746 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000747 if (*in_num)
748 errx(1, "Descriptor has out after in");
749 (*out_num)++;
750 }
751
752 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100753 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000754 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100755 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000756
757 return head;
758}
759
Rusty Russell2e04ef72009-07-30 16:03:45 -0600760/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600761 * After we've used one of their buffers, we tell the Guest about it. Sometime
762 * later we'll want to send them an interrupt using trigger_irq(); note that
763 * wait_for_vq_desc() does that for us if it has to wait.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600764 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000765static void add_used(struct virtqueue *vq, unsigned int head, int len)
766{
767 struct vring_used_elem *used;
768
Rusty Russell2e04ef72009-07-30 16:03:45 -0600769 /*
770 * The virtqueue contains a ring of used buffers. Get a pointer to the
771 * next entry in that used ring.
772 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000773 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
774 used->id = head;
775 used->len = len;
776 /* Make sure buffer is written before we update index. */
777 wmb();
778 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600779 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000780}
781
Rusty Russell17cbca22007-10-22 11:24:22 +1000782/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c802009-06-12 22:26:59 -0600783static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000784{
785 add_used(vq, head, len);
Rusty Russell56739c802009-06-12 22:26:59 -0600786 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000787}
788
Rusty Russelle1e72962007-10-25 15:02:50 +1000789/*
790 * The Console
791 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600792 * We associate some data with the console for our exit hack.
793 */
Rusty Russell1842f232009-07-30 16:03:46 -0600794struct console_abort {
Rusty Russelldde79782007-07-26 10:41:03 -0700795 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700796 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700797 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700798 struct timeval start;
799};
800
Rusty Russelldde79782007-07-26 10:41:03 -0700801/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600802static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700803{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700804 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000805 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600806 struct console_abort *abort = vq->dev->priv;
807 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700808
Rusty Russella91d74a2009-07-30 16:03:45 -0600809 /* Make sure there's a descriptor available. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600810 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000811 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000812 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700813
Rusty Russella91d74a2009-07-30 16:03:45 -0600814 /* Read into it. This is where we usually wait. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600815 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700816 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600817 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700818 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600819 /*
820 * For simplicity, dying threads kill the whole Launcher. So
821 * just nap here.
822 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600823 for (;;)
824 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700825 }
826
Rusty Russella91d74a2009-07-30 16:03:45 -0600827 /* Tell the Guest we used a buffer. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600828 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700829
Rusty Russell2e04ef72009-07-30 16:03:45 -0600830 /*
831 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700832 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600833 * This is such a hack, but works surprisingly well. Each ^C has to
834 * be in a buffer by itself, so they can't be too fast. But we check
835 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600836 * slow.
837 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600838 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700839 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600840 return;
841 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700842
Rusty Russell659a0e62009-06-12 22:27:10 -0600843 abort->count++;
844 if (abort->count == 1)
845 gettimeofday(&abort->start, NULL);
846 else if (abort->count == 3) {
847 struct timeval now;
848 gettimeofday(&now, NULL);
849 /* Kill all Launcher processes with SIGINT, like normal ^C */
850 if (now.tv_sec <= abort->start.tv_sec+1)
851 kill(0, SIGINT);
852 abort->count = 0;
853 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700854}
855
Rusty Russell659a0e62009-06-12 22:27:10 -0600856/* This is the routine which handles console output (ie. stdout). */
857static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700858{
Rusty Russell17cbca22007-10-22 11:24:22 +1000859 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000860 struct iovec iov[vq->vring.num];
861
Rusty Russella91d74a2009-07-30 16:03:45 -0600862 /* We usually wait in here, for the Guest to give us something. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600863 head = wait_for_vq_desc(vq, iov, &out, &in);
864 if (in)
865 errx(1, "Input buffers in console output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600866
867 /* writev can return a partial write, so we loop here. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600868 while (!iov_empty(iov, out)) {
869 int len = writev(STDOUT_FILENO, iov, out);
870 if (len <= 0)
871 err(1, "Write to stdout gave %i", len);
872 iov_consume(iov, out, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000873 }
Rusty Russella91d74a2009-07-30 16:03:45 -0600874
875 /*
876 * We're finished with that buffer: if we're going to sleep,
877 * wait_for_vq_desc() will prod the Guest with an interrupt.
878 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600879 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500880}
881
Rusty Russelle1e72962007-10-25 15:02:50 +1000882/*
883 * The Network
884 *
885 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600886 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500887 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600888struct net_info {
889 int tunfd;
890};
891
892static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700893{
Rusty Russell659a0e62009-06-12 22:27:10 -0600894 struct net_info *net_info = vq->dev->priv;
895 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000896 struct iovec iov[vq->vring.num];
897
Rusty Russella91d74a2009-07-30 16:03:45 -0600898 /* We usually wait in here for the Guest to give us a packet. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600899 head = wait_for_vq_desc(vq, iov, &out, &in);
900 if (in)
901 errx(1, "Input buffers in net output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600902 /*
903 * Send the whole thing through to /dev/net/tun. It expects the exact
904 * same format: what a coincidence!
905 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600906 if (writev(net_info->tunfd, iov, out) < 0)
907 errx(1, "Write to tun failed?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600908
909 /*
910 * Done with that one; wait_for_vq_desc() will send the interrupt if
911 * all packets are processed.
912 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600913 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700914}
915
Rusty Russella91d74a2009-07-30 16:03:45 -0600916/*
917 * Handling network input is a bit trickier, because I've tried to optimize it.
918 *
919 * First we have a helper routine which tells is if from this file descriptor
920 * (ie. the /dev/net/tun device) will block:
921 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600922static bool will_block(int fd)
923{
924 fd_set fdset;
925 struct timeval zero = { 0, 0 };
926 FD_ZERO(&fdset);
927 FD_SET(fd, &fdset);
928 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
929}
930
Rusty Russella91d74a2009-07-30 16:03:45 -0600931/*
932 * This handles packets coming in from the tun device to our Guest. Like all
933 * service routines, it gets called again as soon as it returns, so you don't
934 * see a while(1) loop here.
935 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600936static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700937{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700938 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600939 unsigned int head, out, in;
940 struct iovec iov[vq->vring.num];
941 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700942
Rusty Russella91d74a2009-07-30 16:03:45 -0600943 /*
944 * Get a descriptor to write an incoming packet into. This will also
945 * send an interrupt if they're out of descriptors.
946 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600947 head = wait_for_vq_desc(vq, iov, &out, &in);
948 if (out)
949 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600950
Rusty Russella91d74a2009-07-30 16:03:45 -0600951 /*
952 * If it looks like we'll block reading from the tun device, send them
953 * an interrupt.
954 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600955 if (vq->pending_used && will_block(net_info->tunfd))
956 trigger_irq(vq);
957
Rusty Russella91d74a2009-07-30 16:03:45 -0600958 /*
959 * Read in the packet. This is where we normally wait (when there's no
960 * incoming network traffic).
961 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600962 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700963 if (len <= 0)
Rusty Russell659a0e62009-06-12 22:27:10 -0600964 err(1, "Failed to read from tun.");
Rusty Russella91d74a2009-07-30 16:03:45 -0600965
966 /*
967 * Mark that packet buffer as used, but don't interrupt here. We want
968 * to wait until we've done as much work as we can.
969 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600970 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700971}
Rusty Russella91d74a2009-07-30 16:03:45 -0600972/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700973
Rusty Russella91d74a2009-07-30 16:03:45 -0600974/* This is the helper to create threads: run the service routine in a loop. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600975static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000976{
Rusty Russell659a0e62009-06-12 22:27:10 -0600977 struct virtqueue *vq = _vq;
978
979 for (;;)
980 vq->service(vq);
981 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000982}
983
Rusty Russell2e04ef72009-07-30 16:03:45 -0600984/*
985 * When a child dies, we kill our entire process group with SIGTERM. This
986 * also has the side effect that the shell restores the console for us!
987 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600988static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500989{
Rusty Russell659a0e62009-06-12 22:27:10 -0600990 kill(0, SIGTERM);
991}
992
993static void reset_device(struct device *dev)
994{
995 struct virtqueue *vq;
996
997 verbose("Resetting device %s\n", dev->name);
998
999 /* Clear any features they've acked. */
1000 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
1001
1002 /* We're going to be explicitly killing threads, so ignore them. */
1003 signal(SIGCHLD, SIG_IGN);
1004
1005 /* Zero out the virtqueues, get rid of their threads */
1006 for (vq = dev->vq; vq; vq = vq->next) {
1007 if (vq->thread != (pid_t)-1) {
1008 kill(vq->thread, SIGTERM);
1009 waitpid(vq->thread, NULL, 0);
1010 vq->thread = (pid_t)-1;
1011 }
1012 memset(vq->vring.desc, 0,
1013 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1014 lg_last_avail(vq) = 0;
1015 }
1016 dev->running = false;
1017
1018 /* Now we care if threads die. */
1019 signal(SIGCHLD, (void *)kill_launcher);
1020}
1021
Rusty Russella91d74a2009-07-30 16:03:45 -06001022/*L:216
1023 * This actually creates the thread which services the virtqueue for a device.
1024 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001025static void create_thread(struct virtqueue *vq)
1026{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001027 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001028 * Create stack for thread. Since the stack grows upwards, we point
1029 * the stack pointer to the end of this region.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001030 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001031 char *stack = malloc(32768);
1032 unsigned long args[] = { LHREQ_EVENTFD,
1033 vq->config.pfn*getpagesize(), 0 };
1034
1035 /* Create a zero-initialized eventfd. */
1036 vq->eventfd = eventfd(0, 0);
1037 if (vq->eventfd < 0)
1038 err(1, "Creating eventfd");
1039 args[2] = vq->eventfd;
1040
Rusty Russella91d74a2009-07-30 16:03:45 -06001041 /*
1042 * Attach an eventfd to this virtqueue: it will go off when the Guest
1043 * does an LHCALL_NOTIFY for this vq.
1044 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001045 if (write(lguest_fd, &args, sizeof(args)) != 0)
1046 err(1, "Attaching eventfd");
1047
Rusty Russella91d74a2009-07-30 16:03:45 -06001048 /*
1049 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1050 * we get a signal if it dies.
1051 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001052 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1053 if (vq->thread == (pid_t)-1)
1054 err(1, "Creating clone");
Rusty Russella91d74a2009-07-30 16:03:45 -06001055
1056 /* We close our local copy now the child has it. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001057 close(vq->eventfd);
1058}
1059
Rusty Russellca60a422009-09-23 22:26:47 -06001060static bool accepted_feature(struct device *dev, unsigned int bit)
1061{
1062 const u8 *features = get_feature_bits(dev) + dev->feature_len;
1063
1064 if (dev->feature_len < bit / CHAR_BIT)
1065 return false;
1066 return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT));
1067}
1068
Rusty Russell659a0e62009-06-12 22:27:10 -06001069static void start_device(struct device *dev)
1070{
1071 unsigned int i;
1072 struct virtqueue *vq;
1073
1074 verbose("Device %s OK: offered", dev->name);
1075 for (i = 0; i < dev->feature_len; i++)
1076 verbose(" %02x", get_feature_bits(dev)[i]);
1077 verbose(", accepted");
1078 for (i = 0; i < dev->feature_len; i++)
1079 verbose(" %02x", get_feature_bits(dev)
1080 [dev->feature_len+i]);
1081
Rusty Russellca60a422009-09-23 22:26:47 -06001082 dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1083
Rusty Russell659a0e62009-06-12 22:27:10 -06001084 for (vq = dev->vq; vq; vq = vq->next) {
1085 if (vq->service)
1086 create_thread(vq);
1087 }
1088 dev->running = true;
1089}
1090
1091static void cleanup_devices(void)
1092{
1093 struct device *dev;
1094
1095 for (dev = devices.dev; dev; dev = dev->next)
1096 reset_device(dev);
1097
1098 /* If we saved off the original terminal settings, restore them now. */
1099 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1100 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001101}
1102
Rusty Russella007a752008-05-02 21:50:53 -05001103/* When the Guest tells us they updated the status field, we handle it. */
1104static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001105{
Rusty Russell659a0e62009-06-12 22:27:10 -06001106 /* A zero status is a reset, otherwise it's a set of flags. */
1107 if (dev->desc->status == 0)
1108 reset_device(dev);
1109 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001110 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001111 if (dev->running)
1112 reset_device(dev);
Rusty Russella007a752008-05-02 21:50:53 -05001113 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001114 if (!dev->running)
1115 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001116 }
1117}
1118
Rusty Russella91d74a2009-07-30 16:03:45 -06001119/*L:215
1120 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1121 * particular, it's used to notify us of device status changes during boot.
1122 */
Rusty Russell56739c802009-06-12 22:26:59 -06001123static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001124{
1125 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001126
Rusty Russell659a0e62009-06-12 22:27:10 -06001127 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001128 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001129 struct virtqueue *vq;
1130
Rusty Russella91d74a2009-07-30 16:03:45 -06001131 /*
1132 * Notifications to device descriptors mean they updated the
1133 * device status.
1134 */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001135 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001136 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001137 return;
1138 }
1139
Rusty Russella91d74a2009-07-30 16:03:45 -06001140 /*
1141 * Devices *can* be used before status is set to DRIVER_OK.
1142 * The original plan was that they would never do this: they
1143 * would always finish setting up their status bits before
1144 * actually touching the virtqueues. In practice, we allowed
1145 * them to, and they do (eg. the disk probes for partition
1146 * tables as part of initialization).
1147 *
1148 * If we see this, we start the device: once it's running, we
1149 * expect the device to catch all the notifications.
1150 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001151 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001152 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001153 continue;
Rusty Russell659a0e62009-06-12 22:27:10 -06001154 if (i->running)
1155 errx(1, "Notification on running %s", i->name);
Rusty Russella91d74a2009-07-30 16:03:45 -06001156 /* This just calls create_thread() for each virtqueue */
Rusty Russell659a0e62009-06-12 22:27:10 -06001157 start_device(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001158 return;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001159 }
1160 }
Rusty Russelldde79782007-07-26 10:41:03 -07001161
Rusty Russell2e04ef72009-07-30 16:03:45 -06001162 /*
1163 * Early console write is done using notify on a nul-terminated string
1164 * in Guest memory. It's also great for hacking debugging messages
1165 * into a Guest.
1166 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001167 if (addr >= guest_limit)
1168 errx(1, "Bad NOTIFY %#lx", addr);
1169
1170 write(STDOUT_FILENO, from_guest_phys(addr),
1171 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001172}
1173
Rusty Russelldde79782007-07-26 10:41:03 -07001174/*L:190
1175 * Device Setup
1176 *
1177 * All devices need a descriptor so the Guest knows it exists, and a "struct
1178 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001179 * routines to allocate and manage them.
1180 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001181
Rusty Russell2e04ef72009-07-30 16:03:45 -06001182/*
1183 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001184 * number of virtqueue descriptors, then two sets of feature bits, then an
1185 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001186 * pointer.
1187 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001188static u8 *device_config(const struct device *dev)
1189{
1190 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001191 + dev->num_vq * sizeof(struct lguest_vqconfig)
1192 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001193}
1194
Rusty Russell2e04ef72009-07-30 16:03:45 -06001195/*
1196 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001197 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001198 * that descriptor.
1199 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001200static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001201{
Rusty Russella586d4f2008-02-04 23:49:56 -05001202 struct lguest_device_desc d = { .type = type };
1203 void *p;
1204
1205 /* Figure out where the next device config is, based on the last one. */
1206 if (devices.lastdev)
1207 p = device_config(devices.lastdev)
1208 + devices.lastdev->desc->config_len;
1209 else
1210 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001211
Rusty Russell17cbca22007-10-22 11:24:22 +10001212 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001213 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001214 errx(1, "Too many devices");
1215
Rusty Russella586d4f2008-02-04 23:49:56 -05001216 /* p might not be aligned, so we memcpy in. */
1217 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001218}
1219
Rusty Russell2e04ef72009-07-30 16:03:45 -06001220/*
1221 * Each device descriptor is followed by the description of its virtqueues. We
1222 * specify how many descriptors the virtqueue is to have.
1223 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001224static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001225 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001226{
1227 unsigned int pages;
1228 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1229 void *p;
1230
Rusty Russella6bd8e12008-03-28 11:05:53 -05001231 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001232 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001233 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001234 p = get_pages(pages);
1235
Rusty Russelld1c856e2007-11-19 11:20:40 -05001236 /* Initialize the virtqueue */
1237 vq->next = NULL;
1238 vq->last_avail_idx = 0;
1239 vq->dev = dev;
Rusty Russella91d74a2009-07-30 16:03:45 -06001240
1241 /*
1242 * This is the routine the service thread will run, and its Process ID
1243 * once it's running.
1244 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001245 vq->service = service;
1246 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001247
Rusty Russell17cbca22007-10-22 11:24:22 +10001248 /* Initialize the configuration. */
1249 vq->config.num = num_descs;
1250 vq->config.irq = devices.next_irq++;
1251 vq->config.pfn = to_guest_phys(p) / getpagesize();
1252
1253 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001254 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001255
Rusty Russell2e04ef72009-07-30 16:03:45 -06001256 /*
1257 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001258 * device_config() to get the end of the device's current virtqueues;
1259 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001260 * yet, otherwise we'd be overwriting them.
1261 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001262 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1263 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001264 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001265 dev->desc->num_vq++;
1266
1267 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001268
Rusty Russell2e04ef72009-07-30 16:03:45 -06001269 /*
1270 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1271 * second.
1272 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001273 for (i = &dev->vq; *i; i = &(*i)->next);
1274 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001275}
1276
Rusty Russell2e04ef72009-07-30 16:03:45 -06001277/*
1278 * The first half of the feature bitmask is for us to advertise features. The
1279 * second half is for the Guest to accept features.
1280 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001281static void add_feature(struct device *dev, unsigned bit)
1282{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001283 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001284
1285 /* We can't extend the feature bits once we've added config bytes */
1286 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1287 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001288 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001289 }
1290
Rusty Russella586d4f2008-02-04 23:49:56 -05001291 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1292}
1293
Rusty Russell2e04ef72009-07-30 16:03:45 -06001294/*
1295 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001296 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001297 * how we use it.
1298 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001299static void set_config(struct device *dev, unsigned len, const void *conf)
1300{
1301 /* Check we haven't overflowed our single page. */
1302 if (device_config(dev) + len > devices.descpage + getpagesize())
1303 errx(1, "Too many devices");
1304
1305 /* Copy in the config information, and store the length. */
1306 memcpy(device_config(dev), conf, len);
1307 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001308
1309 /* Size must fit in config_len field (8 bits)! */
1310 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001311}
1312
Rusty Russell2e04ef72009-07-30 16:03:45 -06001313/*
1314 * This routine does all the creation and setup of a new device, including
Rusty Russella91d74a2009-07-30 16:03:45 -06001315 * calling new_dev_desc() to allocate the descriptor and device memory. We
1316 * don't actually start the service threads until later.
Rusty Russella6bd8e12008-03-28 11:05:53 -05001317 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001318 * See what I mean about userspace being boring?
1319 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001320static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001321{
1322 struct device *dev = malloc(sizeof(*dev));
1323
Rusty Russelldde79782007-07-26 10:41:03 -07001324 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001325 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001326 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001327 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001328 dev->feature_len = 0;
1329 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001330 dev->running = false;
Rusty Russella586d4f2008-02-04 23:49:56 -05001331
Rusty Russell2e04ef72009-07-30 16:03:45 -06001332 /*
1333 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001334 * easier, but the user expects the devices to be arranged on the bus
1335 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001336 * is eth0, the first block device /dev/vda, etc.
1337 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001338 if (devices.lastdev)
1339 devices.lastdev->next = dev;
1340 else
1341 devices.dev = dev;
1342 devices.lastdev = dev;
1343
Rusty Russell8ca47e02007-07-19 01:49:29 -07001344 return dev;
1345}
1346
Rusty Russell2e04ef72009-07-30 16:03:45 -06001347/*
1348 * Our first setup routine is the console. It's a fairly simple device, but
1349 * UNIX tty handling makes it uglier than it could be.
1350 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001351static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001352{
1353 struct device *dev;
1354
Rusty Russelldde79782007-07-26 10:41:03 -07001355 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001356 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1357 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001358 /*
1359 * Then we turn off echo, line buffering and ^C etc: We want a
1360 * raw input stream to the Guest.
1361 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001362 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1363 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001364 }
1365
Rusty Russell659a0e62009-06-12 22:27:10 -06001366 dev = new_device("console", VIRTIO_ID_CONSOLE);
1367
Rusty Russelldde79782007-07-26 10:41:03 -07001368 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001369 dev->priv = malloc(sizeof(struct console_abort));
1370 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001371
Rusty Russell2e04ef72009-07-30 16:03:45 -06001372 /*
1373 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001374 * they put something the input queue, we make sure we're listening to
1375 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001376 * stdout.
1377 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001378 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1379 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001380
Rusty Russell659a0e62009-06-12 22:27:10 -06001381 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001382}
Rusty Russelldde79782007-07-26 10:41:03 -07001383/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001384
Rusty Russell2e04ef72009-07-30 16:03:45 -06001385/*M:010
1386 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001387 * --sharenet=<name> option which opens or creates a named pipe. This can be
1388 * used to send packets to another guest in a 1:1 manner.
1389 *
1390 * More sopisticated is to use one of the tools developed for project like UML
1391 * to do networking.
1392 *
1393 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1394 * completely generic ("here's my vring, attach to your vring") and would work
1395 * for any traffic. Of course, namespace and permissions issues need to be
1396 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1397 * multiple inter-guest channels behind one interface, although it would
1398 * require some manner of hotplugging new virtio channels.
1399 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001400 * Finally, we could implement a virtio network switch in the kernel.
1401:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001402
Rusty Russell8ca47e02007-07-19 01:49:29 -07001403static u32 str2ip(const char *ipaddr)
1404{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001405 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001406
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001407 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1408 errx(1, "Failed to parse IP address '%s'", ipaddr);
1409 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1410}
1411
1412static void str2mac(const char *macaddr, unsigned char mac[6])
1413{
1414 unsigned int m[6];
1415 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1416 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1417 errx(1, "Failed to parse mac address '%s'", macaddr);
1418 mac[0] = m[0];
1419 mac[1] = m[1];
1420 mac[2] = m[2];
1421 mac[3] = m[3];
1422 mac[4] = m[4];
1423 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001424}
1425
Rusty Russell2e04ef72009-07-30 16:03:45 -06001426/*
1427 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001428 * network device to the bridge device specified by the command line.
1429 *
1430 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001431 * dislike bridging), and I just try not to break it.
1432 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001433static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1434{
1435 int ifidx;
1436 struct ifreq ifr;
1437
1438 if (!*br_name)
1439 errx(1, "must specify bridge name");
1440
1441 ifidx = if_nametoindex(if_name);
1442 if (!ifidx)
1443 errx(1, "interface %s does not exist!", if_name);
1444
1445 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001446 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001447 ifr.ifr_ifindex = ifidx;
1448 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1449 err(1, "can't add %s to bridge %s", if_name, br_name);
1450}
1451
Rusty Russell2e04ef72009-07-30 16:03:45 -06001452/*
1453 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001454 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001455 * pointer.
1456 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001457static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001458{
1459 struct ifreq ifr;
Rusty Russellf8466192010-08-27 08:39:48 -06001460 struct sockaddr_in sin;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001461
1462 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001463 strcpy(ifr.ifr_name, tapif);
1464
1465 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russellf8466192010-08-27 08:39:48 -06001466 sin.sin_family = AF_INET;
1467 sin.sin_addr.s_addr = htonl(ipaddr);
1468 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001469 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001470 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001471 ifr.ifr_flags = IFF_UP;
1472 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001473 err(1, "Bringing interface %s up", tapif);
1474}
1475
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001476static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001477{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001478 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001479 int netfd;
1480
1481 /* Start with this zeroed. Messy but sure. */
1482 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001483
Rusty Russell2e04ef72009-07-30 16:03:45 -06001484 /*
1485 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001486 * tap device is like a tun device, only somehow different. To tell
1487 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001488 * works now!
1489 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001490 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001491 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001492 strcpy(ifr.ifr_name, "tap%d");
1493 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1494 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001495
Rusty Russell398f1872008-07-29 09:58:37 -05001496 if (ioctl(netfd, TUNSETOFFLOAD,
1497 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1498 err(1, "Could not set features for tun device");
1499
Rusty Russell2e04ef72009-07-30 16:03:45 -06001500 /*
1501 * We don't need checksums calculated for packets coming in this
1502 * device: trust us!
1503 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001504 ioctl(netfd, TUNSETNOCSUM, 1);
1505
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001506 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1507 return netfd;
1508}
1509
Rusty Russell2e04ef72009-07-30 16:03:45 -06001510/*L:195
1511 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001512 * routing, but the principle is the same: it uses the "tun" device to inject
1513 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001514 * just shunt packets between the Guest and the tun device.
1515 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001516static void setup_tun_net(char *arg)
1517{
1518 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001519 struct net_info *net_info = malloc(sizeof(*net_info));
1520 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001521 u32 ip = INADDR_ANY;
1522 bool bridging = false;
1523 char tapif[IFNAMSIZ], *p;
1524 struct virtio_net_config conf;
1525
Rusty Russell659a0e62009-06-12 22:27:10 -06001526 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001527
Rusty Russell17cbca22007-10-22 11:24:22 +10001528 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001529 dev = new_device("net", VIRTIO_ID_NET);
1530 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001531
Rusty Russell2e04ef72009-07-30 16:03:45 -06001532 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001533 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1534 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001535
Rusty Russell2e04ef72009-07-30 16:03:45 -06001536 /*
1537 * We need a socket to perform the magic network ioctls to bring up the
1538 * tap interface, connect to the bridge etc. Any socket will do!
1539 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001540 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1541 if (ipfd < 0)
1542 err(1, "opening IP socket");
1543
Rusty Russelldde79782007-07-26 10:41:03 -07001544 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001545 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001546 arg += strlen(BRIDGE_PFX);
1547 bridging = true;
1548 }
1549
1550 /* A mac address may follow the bridge name or IP address */
1551 p = strchr(arg, ':');
1552 if (p) {
1553 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001554 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001555 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001556 }
1557
1558 /* arg is now either an IP address or a bridge name */
1559 if (bridging)
1560 add_to_bridge(ipfd, tapif, arg);
1561 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001562 ip = str2ip(arg);
1563
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001564 /* Set up the tun device. */
1565 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001566
Rusty Russell20887612008-05-30 15:09:46 -05001567 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
Rusty Russell398f1872008-07-29 09:58:37 -05001568 /* Expect Guest to handle everything except UFO */
1569 add_feature(dev, VIRTIO_NET_F_CSUM);
1570 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001571 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1572 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1573 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1574 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1575 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1576 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001577 /* We handle indirect ring entries */
1578 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russella586d4f2008-02-04 23:49:56 -05001579 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001580
Rusty Russella586d4f2008-02-04 23:49:56 -05001581 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001582 close(ipfd);
1583
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001584 devices.device_num++;
1585
1586 if (bridging)
1587 verbose("device %u: tun %s attached to bridge: %s\n",
1588 devices.device_num, tapif, arg);
1589 else
1590 verbose("device %u: tun %s: %s\n",
1591 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001592}
Rusty Russella91d74a2009-07-30 16:03:45 -06001593/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001594
Rusty Russelle1e72962007-10-25 15:02:50 +10001595/* This hangs off device->priv. */
Rusty Russell1842f232009-07-30 16:03:46 -06001596struct vblk_info {
Rusty Russell17cbca22007-10-22 11:24:22 +10001597 /* The size of the file. */
1598 off64_t len;
1599
1600 /* The file descriptor for the file. */
1601 int fd;
1602
Rusty Russell17cbca22007-10-22 11:24:22 +10001603};
1604
Rusty Russelle1e72962007-10-25 15:02:50 +10001605/*L:210
1606 * The Disk
1607 *
Rusty Russella91d74a2009-07-30 16:03:45 -06001608 * The disk only has one virtqueue, so it only has one thread. It is really
1609 * simple: the Guest asks for a block number and we read or write that position
1610 * in the file.
1611 *
1612 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1613 * slow: the Guest waits until the read is finished before running anything
1614 * else, even if it could have been doing useful work.
1615 *
1616 * We could have used async I/O, except it's reputed to suck so hard that
1617 * characters actually go missing from your code when you try to use it.
Rusty Russelle1e72962007-10-25 15:02:50 +10001618 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001619static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001620{
Rusty Russell659a0e62009-06-12 22:27:10 -06001621 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001622 unsigned int head, out_num, in_num, wlen;
1623 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001624 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001625 struct virtio_blk_outhdr *out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001626 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001627 off64_t off;
1628
Rusty Russella91d74a2009-07-30 16:03:45 -06001629 /*
1630 * Get the next request, where we normally wait. It triggers the
1631 * interrupt to acknowledge previously serviced requests (if any).
1632 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001633 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001634
Rusty Russell2e04ef72009-07-30 16:03:45 -06001635 /*
1636 * Every block request should contain at least one output buffer
Rusty Russelle1e72962007-10-25 15:02:50 +10001637 * (detailing the location on disk and the type of request) and one
Rusty Russell2e04ef72009-07-30 16:03:45 -06001638 * input buffer (to hold the result).
1639 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001640 if (out_num == 0 || in_num == 0)
1641 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1642 head, out_num, in_num);
1643
1644 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001645 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russella91d74a2009-07-30 16:03:45 -06001646 /*
1647 * For historical reasons, block operations are expressed in 512 byte
1648 * "sectors".
1649 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001650 off = out->sector * 512;
1651
Rusty Russell2e04ef72009-07-30 16:03:45 -06001652 /*
Rusty Russell2e04ef72009-07-30 16:03:45 -06001653 * In general the virtio block driver is allowed to try SCSI commands.
1654 * It'd be nice if we supported eject, for example, but we don't.
1655 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001656 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1657 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001658 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001659 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001660 } else if (out->type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001661 /*
1662 * Write
1663 *
1664 * Move to the right location in the block file. This can fail
1665 * if they try to write past end.
1666 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001667 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1668 err(1, "Bad seek to sector %llu", out->sector);
1669
1670 ret = writev(vblk->fd, iov+1, out_num-1);
1671 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1672
Rusty Russell2e04ef72009-07-30 16:03:45 -06001673 /*
1674 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001675 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001676 * file (possibly extending it).
1677 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001678 if (ret > 0 && off + ret > vblk->len) {
1679 /* Trim it back to the correct length */
1680 ftruncate64(vblk->fd, vblk->len);
1681 /* Die, bad Guest, die. */
1682 errx(1, "Write past end %llu+%u", off, ret);
1683 }
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001684
1685 wlen = sizeof(*in);
1686 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1687 } else if (out->type & VIRTIO_BLK_T_FLUSH) {
1688 /* Flush */
1689 ret = fdatasync(vblk->fd);
1690 verbose("FLUSH fdatasync: %i\n", ret);
Anthony Liguori1200e642007-11-08 21:13:44 -06001691 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001692 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001693 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001694 /*
1695 * Read
1696 *
1697 * Move to the right location in the block file. This can fail
1698 * if they try to read past end.
1699 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001700 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1701 err(1, "Bad seek to sector %llu", out->sector);
1702
1703 ret = readv(vblk->fd, iov+1, in_num-1);
1704 verbose("READ from sector %llu: %i\n", out->sector, ret);
1705 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001706 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001707 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001708 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001709 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001710 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001711 }
1712 }
1713
Rusty Russella91d74a2009-07-30 16:03:45 -06001714 /* Finished that request. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001715 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001716}
1717
Rusty Russelle1e72962007-10-25 15:02:50 +10001718/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001719static void setup_block_file(const char *filename)
1720{
Rusty Russell17cbca22007-10-22 11:24:22 +10001721 struct device *dev;
1722 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001723 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001724
Rusty Russell2e04ef72009-07-30 16:03:45 -06001725 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001726 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001727
Rusty Russelle1e72962007-10-25 15:02:50 +10001728 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001729 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001730
1731 /* Allocate the room for our own bookkeeping */
1732 vblk = dev->priv = malloc(sizeof(*vblk));
1733
1734 /* First we open the file and store the length. */
1735 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1736 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1737
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001738 /* We support FLUSH. */
1739 add_feature(dev, VIRTIO_BLK_F_FLUSH);
Rusty Russella586d4f2008-02-04 23:49:56 -05001740
Rusty Russell17cbca22007-10-22 11:24:22 +10001741 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001742 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001743
Rusty Russell2e04ef72009-07-30 16:03:45 -06001744 /*
1745 * Tell Guest not to put in too many descriptors at once: two are used
1746 * for the in and out elements.
1747 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001748 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1749 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1750
Rusty Russell8ef562d2009-07-30 16:03:43 -06001751 /* Don't try to put whole struct: we have 8 bit limit. */
1752 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001753
Rusty Russell17cbca22007-10-22 11:24:22 +10001754 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001755 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001756}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001757
Rusty Russell2e04ef72009-07-30 16:03:45 -06001758/*L:211
1759 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001760 * input buffers. The usual case is that the Guest doesn't want random numbers
1761 * and so has no buffers although /dev/random is still readable, whereas
1762 * console is the reverse.
1763 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001764 * The same logic applies, however.
1765 */
1766struct rng_info {
1767 int rfd;
1768};
1769
Rusty Russell659a0e62009-06-12 22:27:10 -06001770static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001771{
1772 int len;
1773 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001774 struct rng_info *rng_info = vq->dev->priv;
1775 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001776
1777 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001778 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001779 if (out_num)
1780 errx(1, "Output buffers in rng?");
1781
Rusty Russell2e04ef72009-07-30 16:03:45 -06001782 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001783 * Just like the console write, we loop to cover the whole iovec.
1784 * In this case, short reads actually happen quite a bit.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001785 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001786 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001787 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001788 if (len <= 0)
1789 err(1, "Read from /dev/random gave %i", len);
1790 iov_consume(iov, in_num, len);
1791 totlen += len;
1792 }
1793
1794 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001795 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001796}
1797
Rusty Russell2e04ef72009-07-30 16:03:45 -06001798/*L:199
1799 * This creates a "hardware" random number device for the Guest.
1800 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001801static void setup_rng(void)
1802{
1803 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001804 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001805
Rusty Russell2e04ef72009-07-30 16:03:45 -06001806 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001807 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001808
Rusty Russell2e04ef72009-07-30 16:03:45 -06001809 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001810 dev = new_device("rng", VIRTIO_ID_RNG);
1811 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001812
1813 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001814 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001815
1816 verbose("device %u: rng\n", devices.device_num++);
1817}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001818/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301819
Rusty Russella6bd8e12008-03-28 11:05:53 -05001820/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301821static void __attribute__((noreturn)) restart_guest(void)
1822{
1823 unsigned int i;
1824
Rusty Russell2e04ef72009-07-30 16:03:45 -06001825 /*
1826 * Since we don't track all open fds, we simply close everything beyond
1827 * stderr.
1828 */
Balaji Raoec04b132007-12-28 14:26:24 +05301829 for (i = 3; i < FD_SETSIZE; i++)
1830 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001831
Rusty Russell659a0e62009-06-12 22:27:10 -06001832 /* Reset all the devices (kills all threads). */
1833 cleanup_devices();
1834
Balaji Raoec04b132007-12-28 14:26:24 +05301835 execv(main_args[0], main_args);
1836 err(1, "Could not exec %s", main_args[0]);
1837}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001838
Rusty Russell2e04ef72009-07-30 16:03:45 -06001839/*L:220
1840 * Finally we reach the core of the Launcher which runs the Guest, serves
1841 * its input and output, and finally, lays it to rest.
1842 */
Rusty Russell56739c802009-06-12 22:26:59 -06001843static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001844{
1845 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001846 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001847 int readval;
1848
1849 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001850 readval = pread(lguest_fd, &notify_addr,
1851 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001852
Rusty Russell17cbca22007-10-22 11:24:22 +10001853 /* One unsigned long means the Guest did HCALL_NOTIFY */
1854 if (readval == sizeof(notify_addr)) {
1855 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c802009-06-12 22:26:59 -06001856 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001857 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001858 } else if (errno == ENOENT) {
1859 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001860 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001861 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301862 /* ERESTART means that we need to reboot the guest */
1863 } else if (errno == ERESTART) {
1864 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001865 /* Anything else means a bug or incompatible change. */
1866 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001867 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001868 }
1869}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001870/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001871 * This is the end of the Launcher. The good news: we are over halfway
1872 * through! The bad news: the most fiendish part of the code still lies ahead
1873 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001874 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001875 * Are you ready? Take a deep breath and join me in the core of the Host, in
1876 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001877:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001878
1879static struct option opts[] = {
1880 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001881 { "tunnet", 1, NULL, 't' },
1882 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001883 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001884 { "initrd", 1, NULL, 'i' },
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001885 { "username", 1, NULL, 'u' },
1886 { "chroot", 1, NULL, 'c' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001887 { NULL },
1888};
1889static void usage(void)
1890{
1891 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001892 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001893 "|--block=<filename>|--initrd=<filename>]...\n"
1894 "<mem-in-mb> vmlinux [args...]");
1895}
1896
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001897/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001898int main(int argc, char *argv[])
1899{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001900 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001901 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c802009-06-12 22:26:59 -06001902 /* Two temporaries. */
1903 int i, c;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001904 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001905 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001906 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001907 const char *initrd_name = NULL;
1908
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001909 /* Password structure for initgroups/setres[gu]id */
1910 struct passwd *user_details = NULL;
1911
1912 /* Directory to chroot to */
1913 char *chroot_path = NULL;
1914
Balaji Raoec04b132007-12-28 14:26:24 +05301915 /* Save the args: we "reboot" by execing ourselves again. */
1916 main_args = argv;
Balaji Raoec04b132007-12-28 14:26:24 +05301917
Rusty Russell2e04ef72009-07-30 16:03:45 -06001918 /*
1919 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001920 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001921 * remember that 0 is used by the timer).
1922 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001923 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001924 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001925
Rusty Russella91d74a2009-07-30 16:03:45 -06001926 /* We're CPU 0. In fact, that's the only CPU possible right now. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001927 cpu_id = 0;
Rusty Russella91d74a2009-07-30 16:03:45 -06001928
Rusty Russell2e04ef72009-07-30 16:03:45 -06001929 /*
1930 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001931 * descriptor and memory pages for the devices as we parse the command
1932 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001933 * of memory now.
1934 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001935 for (i = 1; i < argc; i++) {
1936 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001937 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001938 /*
1939 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001940 * guest-physical memory range. This fills it with 0,
1941 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001942 * tries to access it.
1943 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001944 guest_base = map_zeroed_pages(mem / getpagesize()
1945 + DEVICE_PAGES);
1946 guest_limit = mem;
1947 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001948 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001949 break;
1950 }
1951 }
Rusty Russelldde79782007-07-26 10:41:03 -07001952
1953 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001954 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1955 switch (c) {
1956 case 'v':
1957 verbose = true;
1958 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001959 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001960 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001961 break;
1962 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001963 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001964 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001965 case 'r':
1966 setup_rng();
1967 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001968 case 'i':
1969 initrd_name = optarg;
1970 break;
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001971 case 'u':
1972 user_details = getpwnam(optarg);
1973 if (!user_details)
1974 err(1, "getpwnam failed, incorrect username?");
1975 break;
1976 case 'c':
1977 chroot_path = optarg;
1978 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001979 default:
1980 warnx("Unknown argument %s", argv[optind]);
1981 usage();
1982 }
1983 }
Rusty Russell2e04ef72009-07-30 16:03:45 -06001984 /*
1985 * After the other arguments we expect memory and kernel image name,
1986 * followed by command line arguments for the kernel.
1987 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001988 if (optind + 2 > argc)
1989 usage();
1990
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001991 verbose("Guest base is at %p\n", guest_base);
1992
Rusty Russelldde79782007-07-26 10:41:03 -07001993 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001994 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001995
Rusty Russell8ca47e02007-07-19 01:49:29 -07001996 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001997 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001998
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001999 /* Boot information is stashed at physical address 0 */
2000 boot = from_guest_phys(0);
2001
Rusty Russelldde79782007-07-26 10:41:03 -07002002 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07002003 if (initrd_name) {
2004 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06002005 /*
2006 * These are the location in the Linux boot header where the
2007 * start and size of the initrd are expected to be found.
2008 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002009 boot->hdr.ramdisk_image = mem - initrd_size;
2010 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07002011 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002012 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002013 }
2014
Rusty Russell2e04ef72009-07-30 16:03:45 -06002015 /*
2016 * The Linux boot header contains an "E820" memory map: ours is a
2017 * simple, single region.
2018 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002019 boot->e820_entries = 1;
2020 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06002021 /*
2022 * The boot header contains a command line pointer: we put the command
2023 * line after the boot header.
2024 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002025 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10002026 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002027 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07002028
Rusty Russell814a0e52007-10-22 11:29:44 +10002029 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002030 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002031
2032 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002033 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002034
Rusty Russell43d33b22007-10-22 11:29:57 +10002035 /* Tell the entry path not to try to reload segment registers. */
2036 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002037
Rusty Russell2e04ef72009-07-30 16:03:45 -06002038 /*
2039 * We tell the kernel to initialize the Guest: this returns the open
2040 * /dev/lguest file descriptor.
2041 */
Rusty Russell56739c802009-06-12 22:26:59 -06002042 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002043
Rusty Russella91d74a2009-07-30 16:03:45 -06002044 /* Ensure that we terminate if a device-servicing child dies. */
Rusty Russell659a0e62009-06-12 22:27:10 -06002045 signal(SIGCHLD, kill_launcher);
2046
2047 /* If we exit via err(), this kills all the threads, restores tty. */
2048 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002049
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06002050 /* If requested, chroot to a directory */
2051 if (chroot_path) {
2052 if (chroot(chroot_path) != 0)
2053 err(1, "chroot(\"%s\") failed", chroot_path);
2054
2055 if (chdir("/") != 0)
2056 err(1, "chdir(\"/\") failed");
2057
2058 verbose("chroot done\n");
2059 }
2060
2061 /* If requested, drop privileges */
2062 if (user_details) {
2063 uid_t u;
2064 gid_t g;
2065
2066 u = user_details->pw_uid;
2067 g = user_details->pw_gid;
2068
2069 if (initgroups(user_details->pw_name, g) != 0)
2070 err(1, "initgroups failed");
2071
2072 if (setresgid(g, g, g) != 0)
2073 err(1, "setresgid failed");
2074
2075 if (setresuid(u, u, u) != 0)
2076 err(1, "setresuid failed");
2077
2078 verbose("Dropping privileges completed\n");
2079 }
2080
Rusty Russelldde79782007-07-26 10:41:03 -07002081 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c802009-06-12 22:26:59 -06002082 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07002083}
Rusty Russellf56a3842007-07-26 10:41:05 -07002084/*:*/
2085
2086/*M:999
2087 * Mastery is done: you now know everything I do.
2088 *
2089 * But surely you have seen code, features and bugs in your wanderings which
2090 * you now yearn to attack? That is the real game, and I look forward to you
2091 * patching and forking lguest into the Your-Name-Here-visor.
2092 *
2093 * Farewell, and good coding!
2094 * Rusty Russell.
2095 */