blob: 42208511b5c054a20f513c350807e54d67abb724 [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>
37#include <zlib.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100038#include <assert.h>
39#include <sched.h>
Rusty Russella586d4f2008-02-04 23:49:56 -050040#include <limits.h>
41#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050042#include <signal.h>
Rusty Russellb45d8cb2007-10-22 10:56:24 +100043#include "linux/lguest_launcher.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100044#include "linux/virtio_config.h"
45#include "linux/virtio_net.h"
46#include "linux/virtio_blk.h"
47#include "linux/virtio_console.h"
Rusty Russell28fd6d72008-07-29 09:58:33 -050048#include "linux/virtio_rng.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100049#include "linux/virtio_ring.h"
Rusty Russelld5d02d62008-10-31 11:24:25 -050050#include "asm/bootparam.h"
Rusty Russell2e04ef72009-07-30 16:03:45 -060051/*L:110
Rusty Russella91d74a2009-07-30 16:03:45 -060052 * We can ignore the 42 include files we need for this program, but I do want
Rusty Russell2e04ef72009-07-30 16:03:45 -060053 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100054 *
55 * As Linus said, "C is a Spartan language, and so should your naming be." I
56 * like these abbreviations, so we define them here. Note that u64 is always
57 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060058 * use %llu in printf for any u64.
59 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100060typedef unsigned long long u64;
61typedef uint32_t u32;
62typedef uint16_t u16;
63typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070064/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070065
66#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
Rusty Russell8ca47e02007-07-19 01:49:29 -070067#define BRIDGE_PFX "bridge:"
68#ifndef SIOCBRADDIF
69#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
70#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100071/* We can have up to 256 pages for devices. */
72#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050073/* This will occupy 3 pages: it must be a power of 2. */
74#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070075
Rusty Russell2e04ef72009-07-30 16:03:45 -060076/*L:120
77 * verbose is both a global flag and a macro. The C preprocessor allows
78 * this, and although I wouldn't recommend it, it works quite nicely here.
79 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070080static bool verbose;
81#define verbose(args...) \
82 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070083/*:*/
84
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100085/* The pointer to the start of guest memory. */
86static void *guest_base;
87/* The maximum guest physical address allowed, and maximum possible. */
88static unsigned long guest_limit, guest_max;
Rusty Russell56739c82009-06-12 22:26:59 -060089/* The /dev/lguest file descriptor. */
90static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070091
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020092/* a per-cpu variable indicating whose vcpu is currently running */
93static unsigned int __thread cpu_id;
94
Rusty Russelldde79782007-07-26 10:41:03 -070095/* This is our list of devices. */
Rusty Russell1842f232009-07-30 16:03:46 -060096struct device_list {
Rusty Russell17cbca22007-10-22 11:24:22 +100097 /* Counter to assign interrupt numbers. */
98 unsigned int next_irq;
99
100 /* Counter to print out convenient device numbers. */
101 unsigned int device_num;
102
Rusty Russelldde79782007-07-26 10:41:03 -0700103 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000104 u8 *descpage;
105
Rusty Russelldde79782007-07-26 10:41:03 -0700106 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700107 struct device *dev;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600108 /* And a pointer to the last device for easy append. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500109 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700110};
111
Rusty Russell17cbca22007-10-22 11:24:22 +1000112/* The list of Guest devices, based on command line arguments. */
113static struct device_list devices;
114
Rusty Russelldde79782007-07-26 10:41:03 -0700115/* The device structure describes a single device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600116struct device {
Rusty Russelldde79782007-07-26 10:41:03 -0700117 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700118 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000119
Rusty Russell713b15b2009-06-12 22:26:58 -0600120 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700121 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000122
Rusty Russell713b15b2009-06-12 22:26:58 -0600123 /* We can't trust desc values once Guest has booted: we use these. */
124 unsigned int feature_len;
125 unsigned int num_vq;
126
Rusty Russell17cbca22007-10-22 11:24:22 +1000127 /* The name of this device, for --verbose. */
128 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700129
Rusty Russell17cbca22007-10-22 11:24:22 +1000130 /* Any queues attached to this device */
131 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700132
Rusty Russell659a0e62009-06-12 22:27:10 -0600133 /* Is it operational */
134 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500135
Rusty Russellca60a422009-09-23 22:26:47 -0600136 /* Does Guest want an intrrupt on empty? */
137 bool irq_on_empty;
138
Rusty Russell8ca47e02007-07-19 01:49:29 -0700139 /* Device-specific data. */
140 void *priv;
141};
142
Rusty Russell17cbca22007-10-22 11:24:22 +1000143/* The virtqueue structure describes a queue attached to a device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600144struct virtqueue {
Rusty Russell17cbca22007-10-22 11:24:22 +1000145 struct virtqueue *next;
146
147 /* Which device owns me. */
148 struct device *dev;
149
150 /* The configuration for this queue. */
151 struct lguest_vqconfig config;
152
153 /* The actual ring of buffers. */
154 struct vring vring;
155
156 /* Last available index we saw. */
157 u16 last_avail_idx;
158
Rusty Russell95c517c2009-06-12 22:27:11 -0600159 /* How many are used since we sent last irq? */
160 unsigned int pending_used;
161
Rusty Russell659a0e62009-06-12 22:27:10 -0600162 /* Eventfd where Guest notifications arrive. */
163 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500164
Rusty Russell659a0e62009-06-12 22:27:10 -0600165 /* Function for the thread which is servicing this virtqueue. */
166 void (*service)(struct virtqueue *vq);
167 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000168};
169
Balaji Raoec04b132007-12-28 14:26:24 +0530170/* Remember the arguments to the program so we can "reboot" */
171static char **main_args;
172
Rusty Russell659a0e62009-06-12 22:27:10 -0600173/* The original tty settings to restore on exit. */
174static struct termios orig_term;
175
Rusty Russell2e04ef72009-07-30 16:03:45 -0600176/*
177 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600178 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600179 * in precise order.
180 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600181#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russellb60da132009-06-12 22:27:12 -0600182#define mb() __asm__ __volatile__("" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000183
Rusty Russell2e04ef72009-07-30 16:03:45 -0600184/*
185 * Convert an iovec element to the given type.
Rusty Russell17cbca22007-10-22 11:24:22 +1000186 *
187 * This is a fairly ugly trick: we need to know the size of the type and
188 * alignment requirement to check the pointer is kosher. It's also nice to
189 * have the name of the type in case we report failure.
190 *
191 * Typing those three things all the time is cumbersome and error prone, so we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600192 * have a macro which sets them all up and passes to the real function.
193 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000194#define convert(iov, type) \
195 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
196
197static void *_convert(struct iovec *iov, size_t size, size_t align,
198 const char *name)
199{
200 if (iov->iov_len != size)
201 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
202 if ((unsigned long)iov->iov_base % align != 0)
203 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
204 return iov->iov_base;
205}
206
Rusty Russellb5111792008-07-29 09:58:34 -0500207/* Wrapper for the last available index. Makes it easier to change. */
208#define lg_last_avail(vq) ((vq)->last_avail_idx)
209
Rusty Russell2e04ef72009-07-30 16:03:45 -0600210/*
211 * The virtio configuration space is defined to be little-endian. x86 is
212 * little-endian too, but it's nice to be explicit so we have these helpers.
213 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000214#define cpu_to_le16(v16) (v16)
215#define cpu_to_le32(v32) (v32)
216#define cpu_to_le64(v64) (v64)
217#define le16_to_cpu(v16) (v16)
218#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500219#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000220
Rusty Russell28fd6d72008-07-29 09:58:33 -0500221/* Is this iovec empty? */
222static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
223{
224 unsigned int i;
225
226 for (i = 0; i < num_iov; i++)
227 if (iov[i].iov_len)
228 return false;
229 return true;
230}
231
232/* Take len bytes from the front of this iovec. */
233static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
234{
235 unsigned int i;
236
237 for (i = 0; i < num_iov; i++) {
238 unsigned int used;
239
240 used = iov[i].iov_len < len ? iov[i].iov_len : len;
241 iov[i].iov_base += used;
242 iov[i].iov_len -= used;
243 len -= used;
244 }
245 assert(len == 0);
246}
247
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500248/* The device virtqueue descriptors are followed by feature bitmasks. */
249static u8 *get_feature_bits(struct device *dev)
250{
251 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600252 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500253}
254
Rusty Russell2e04ef72009-07-30 16:03:45 -0600255/*L:100
256 * The Launcher code itself takes us out into userspace, that scary place where
257 * pointers run wild and free! Unfortunately, like most userspace programs,
258 * it's quite boring (which is why everyone likes to hack on the kernel!).
259 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
260 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000261 *
262 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
263 * memory and stores it in "guest_base". In other words, Guest physical ==
264 * Launcher virtual with an offset.
265 *
266 * This can be tough to get your head around, but usually it just means that we
267 * use these trivial conversion functions when the Guest gives us it's
Rusty Russell2e04ef72009-07-30 16:03:45 -0600268 * "physical" addresses:
269 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000270static void *from_guest_phys(unsigned long addr)
271{
272 return guest_base + addr;
273}
274
275static unsigned long to_guest_phys(const void *addr)
276{
277 return (addr - guest_base);
278}
279
Rusty Russelldde79782007-07-26 10:41:03 -0700280/*L:130
281 * Loading the Kernel.
282 *
283 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600284 * error-checking code cluttering the callers:
285 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700286static int open_or_die(const char *name, int flags)
287{
288 int fd = open(name, flags);
289 if (fd < 0)
290 err(1, "Failed to open %s", name);
291 return fd;
292}
293
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000294/* map_zeroed_pages() takes a number of pages. */
295static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700296{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000297 int fd = open_or_die("/dev/zero", O_RDONLY);
298 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700299
Rusty Russell2e04ef72009-07-30 16:03:45 -0600300 /*
301 * We use a private mapping (ie. if we write to the page, it will be
302 * copied).
303 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000304 addr = mmap(NULL, getpagesize() * num,
305 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
306 if (addr == MAP_FAILED)
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200307 err(1, "Mmapping %u pages of /dev/zero", num);
Rusty Russella91d74a2009-07-30 16:03:45 -0600308
309 /*
310 * One neat mmap feature is that you can close the fd, and it
311 * stays mapped.
312 */
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100313 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700314
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000315 return addr;
316}
317
318/* Get some more pages for a device. */
319static void *get_pages(unsigned int num)
320{
321 void *addr = from_guest_phys(guest_limit);
322
323 guest_limit += num * getpagesize();
324 if (guest_limit > guest_max)
325 errx(1, "Not enough memory for devices");
326 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700327}
328
Rusty Russell2e04ef72009-07-30 16:03:45 -0600329/*
330 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700331 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600332 * it falls back to reading the memory in.
333 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700334static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
335{
336 ssize_t r;
337
Rusty Russell2e04ef72009-07-30 16:03:45 -0600338 /*
339 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700340 * The kernel really wants to be writable: it patches its own
341 * instructions.
342 *
343 * MAP_PRIVATE means that the page won't be copied until a write is
344 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600345 * Guests.
346 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700347 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
348 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
349 return;
350
351 /* pread does a seek and a read in one shot: saves a few lines. */
352 r = pread(fd, addr, len, offset);
353 if (r != len)
354 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
355}
356
Rusty Russell2e04ef72009-07-30 16:03:45 -0600357/*
358 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700359 * the Guest memory. ELF = Embedded Linking Format, which is the format used
360 * by all modern binaries on Linux including the kernel.
361 *
362 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000363 * address. We use the physical address; the Guest will map itself to the
364 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700365 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600366 * We return the starting address.
367 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000368static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700369{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700370 Elf32_Phdr phdr[ehdr->e_phnum];
371 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700372
Rusty Russell2e04ef72009-07-30 16:03:45 -0600373 /*
374 * Sanity checks on the main ELF header: an x86 executable with a
375 * reasonable number of correctly-sized program headers.
376 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700377 if (ehdr->e_type != ET_EXEC
378 || ehdr->e_machine != EM_386
379 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
380 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
381 errx(1, "Malformed elf header");
382
Rusty Russell2e04ef72009-07-30 16:03:45 -0600383 /*
384 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700385 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600386 * load where.
387 */
Rusty Russelldde79782007-07-26 10:41:03 -0700388
389 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700390 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
391 err(1, "Seeking to program headers");
392 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
393 err(1, "Reading program headers");
394
Rusty Russell2e04ef72009-07-30 16:03:45 -0600395 /*
396 * Try all the headers: there are usually only three. A read-only one,
397 * a read-write one, and a "note" section which we don't load.
398 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700399 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700400 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700401 if (phdr[i].p_type != PT_LOAD)
402 continue;
403
404 verbose("Section %i: size %i addr %p\n",
405 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
406
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700407 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000408 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700409 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700410 }
411
Rusty Russell814a0e52007-10-22 11:29:44 +1000412 /* The entry point is given in the ELF header. */
413 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700414}
415
Rusty Russell2e04ef72009-07-30 16:03:45 -0600416/*L:150
417 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
418 * to jump into it and it will unpack itself. We used to have to perform some
419 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700420 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000421 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
422 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600423 * the funky header so we know where in the file to load, and away we go!
424 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000425static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700426{
Rusty Russell43d33b22007-10-22 11:29:57 +1000427 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000428 int r;
429 /* Modern bzImages get loaded at 1M. */
430 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700431
Rusty Russell2e04ef72009-07-30 16:03:45 -0600432 /*
433 * Go back to the start of the file and read the header. It should be
434 * a Linux boot header (see Documentation/x86/i386/boot.txt)
435 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000436 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000437 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000438
Rusty Russell43d33b22007-10-22 11:29:57 +1000439 /* Inside the setup_hdr, we expect the magic "HdrS" */
440 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000441 errx(1, "This doesn't look like a bzImage to me");
442
Rusty Russell43d33b22007-10-22 11:29:57 +1000443 /* Skip over the extra sectors of the header. */
444 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000445
446 /* Now read everything into memory. in nice big chunks. */
447 while ((r = read(fd, p, 65536)) > 0)
448 p += r;
449
Rusty Russell43d33b22007-10-22 11:29:57 +1000450 /* Finally, code32_start tells us where to enter the kernel. */
451 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700452}
453
Rusty Russell2e04ef72009-07-30 16:03:45 -0600454/*L:140
455 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000456 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600457 * work, we can load those, too.
458 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000459static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700460{
461 Elf32_Ehdr hdr;
462
Rusty Russelldde79782007-07-26 10:41:03 -0700463 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700464 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
465 err(1, "Reading kernel");
466
Rusty Russelldde79782007-07-26 10:41:03 -0700467 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700468 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000469 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700470
Rusty Russella6bd8e12008-03-28 11:05:53 -0500471 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000472 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700473}
474
Rusty Russell2e04ef72009-07-30 16:03:45 -0600475/*
476 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700477 * it calls getpagesize() twice: "it's dumb code."
478 *
479 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600480 * necessary. I leave this code as a reaction against that.
481 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700482static inline unsigned long page_align(unsigned long addr)
483{
Rusty Russelldde79782007-07-26 10:41:03 -0700484 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700485 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
486}
487
Rusty Russell2e04ef72009-07-30 16:03:45 -0600488/*L:180
489 * An "initial ram disk" is a disk image loaded into memory along with the
490 * kernel which the kernel can use to boot from without needing any drivers.
491 * Most distributions now use this as standard: the initrd contains the code to
492 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700493 *
494 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600495 * kernels. He sent me this (and tells me when I break it).
496 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700497static unsigned long load_initrd(const char *name, unsigned long mem)
498{
499 int ifd;
500 struct stat st;
501 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700502
503 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700504 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700505 if (fstat(ifd, &st) < 0)
506 err(1, "fstat() on initrd '%s'", name);
507
Rusty Russell2e04ef72009-07-30 16:03:45 -0600508 /*
509 * We map the initrd at the top of memory, but mmap wants it to be
510 * page-aligned, so we round the size up for that.
511 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700512 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000513 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600514 /*
515 * Once a file is mapped, you can close the file descriptor. It's a
516 * little odd, but quite useful.
517 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700518 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700519 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700520
521 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700522 return len;
523}
Rusty Russelle1e72962007-10-25 15:02:50 +1000524/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700525
Rusty Russell2e04ef72009-07-30 16:03:45 -0600526/*
527 * Simple routine to roll all the commandline arguments together with spaces
528 * between them.
529 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700530static void concat(char *dst, char *args[])
531{
532 unsigned int i, len = 0;
533
534 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100535 if (i) {
536 strcat(dst+len, " ");
537 len++;
538 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700539 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100540 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700541 }
542 /* In case it's empty. */
543 dst[len] = '\0';
544}
545
Rusty Russell2e04ef72009-07-30 16:03:45 -0600546/*L:185
547 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000548 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300549 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600550 * entry point for the Guest.
551 */
Rusty Russell56739c82009-06-12 22:26:59 -0600552static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700553{
Jes Sorensen511801d2007-10-22 11:03:31 +1000554 unsigned long args[] = { LHREQ_INITIALIZE,
555 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300556 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000557 verbose("Guest: %p - %p (%#lx)\n",
558 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c82009-06-12 22:26:59 -0600559 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
560 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700561 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700562}
Rusty Russelldde79782007-07-26 10:41:03 -0700563/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700564
Rusty Russella91d74a2009-07-30 16:03:45 -0600565/*L:200
Rusty Russelldde79782007-07-26 10:41:03 -0700566 * Device Handling.
567 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000568 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700569 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000570 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700571 * if something funny is going on:
572 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700573static void *_check_pointer(unsigned long addr, unsigned int size,
574 unsigned int line)
575{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600576 /*
577 * We have to separately check addr and addr+size, because size could
578 * be huge and addr + size might wrap around.
579 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000580 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000581 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600582 /*
583 * We return a pointer for the caller's convenience, now we know it's
584 * safe to use.
585 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000586 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700587}
Rusty Russelldde79782007-07-26 10:41:03 -0700588/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700589#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
590
Rusty Russell2e04ef72009-07-30 16:03:45 -0600591/*
592 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000593 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600594 * at the end.
595 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100596static unsigned next_desc(struct vring_desc *desc,
597 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000598{
599 unsigned int next;
600
601 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100602 if (!(desc[i].flags & VRING_DESC_F_NEXT))
603 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000604
605 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100606 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000607 /* Make sure compiler knows to grab that: we don't want it changing! */
608 wmb();
609
Mark McLoughlind1f01322009-05-11 18:11:46 +0100610 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000611 errx(1, "Desc next is %u", next);
612
613 return next;
614}
615
Rusty Russella91d74a2009-07-30 16:03:45 -0600616/*
617 * This actually sends the interrupt for this virtqueue, if we've used a
618 * buffer.
619 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600620static void trigger_irq(struct virtqueue *vq)
621{
622 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
623
Rusty Russell95c517c2009-06-12 22:27:11 -0600624 /* Don't inform them if nothing used. */
625 if (!vq->pending_used)
626 return;
627 vq->pending_used = 0;
628
Rusty Russellca60a422009-09-23 22:26:47 -0600629 /* If they don't want an interrupt, don't send one... */
630 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
631 /* ... unless they've asked us to force one on empty. */
632 if (!vq->dev->irq_on_empty
633 || lg_last_avail(vq) != vq->vring.avail->idx)
634 return;
635 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600636
637 /* Send the Guest an interrupt tell them we used something up. */
638 if (write(lguest_fd, buf, sizeof(buf)) != 0)
639 err(1, "Triggering irq %i", vq->config.irq);
640}
641
Rusty Russell2e04ef72009-07-30 16:03:45 -0600642/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600643 * This looks in the virtqueue for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000644 * it to an iovec for convenient access. Since descriptors consist of some
645 * number of output then some number of input descriptors, it's actually two
646 * iovecs, but we pack them into one and note how many of each there were.
647 *
Rusty Russella91d74a2009-07-30 16:03:45 -0600648 * This function waits if necessary, and returns the descriptor number found.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600649 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600650static unsigned wait_for_vq_desc(struct virtqueue *vq,
651 struct iovec iov[],
652 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000653{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100654 unsigned int i, head, max;
655 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600656 u16 last_avail = lg_last_avail(vq);
657
Rusty Russella91d74a2009-07-30 16:03:45 -0600658 /* There's nothing available? */
Rusty Russell659a0e62009-06-12 22:27:10 -0600659 while (last_avail == vq->vring.avail->idx) {
660 u64 event;
661
Rusty Russella91d74a2009-07-30 16:03:45 -0600662 /*
663 * Since we're about to sleep, now is a good time to tell the
664 * Guest about what we've used up to now.
665 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600666 trigger_irq(vq);
667
Rusty Russellb60da132009-06-12 22:27:12 -0600668 /* OK, now we need to know about added descriptors. */
669 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
670
Rusty Russell2e04ef72009-07-30 16:03:45 -0600671 /*
672 * They could have slipped one in as we were doing that: make
673 * sure it's written, then check again.
674 */
Rusty Russellb60da132009-06-12 22:27:12 -0600675 mb();
676 if (last_avail != vq->vring.avail->idx) {
677 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
678 break;
679 }
680
Rusty Russell659a0e62009-06-12 22:27:10 -0600681 /* Nothing new? Wait for eventfd to tell us they refilled. */
682 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
683 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600684
685 /* We don't need to be notified again. */
686 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600687 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000688
689 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500690 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000691 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500692 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000693
Rusty Russell2e04ef72009-07-30 16:03:45 -0600694 /*
695 * Grab the next descriptor number they're advertising, and increment
696 * the index we've seen.
697 */
Rusty Russellb5111792008-07-29 09:58:34 -0500698 head = vq->vring.avail->ring[last_avail % vq->vring.num];
699 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000700
701 /* If their number is silly, that's a fatal mistake. */
702 if (head >= vq->vring.num)
703 errx(1, "Guest says index %u is available", head);
704
705 /* When we start there are none of either input nor output. */
706 *out_num = *in_num = 0;
707
Mark McLoughlind1f01322009-05-11 18:11:46 +0100708 max = vq->vring.num;
709 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000710 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100711
Rusty Russell2e04ef72009-07-30 16:03:45 -0600712 /*
713 * If this is an indirect entry, then this buffer contains a descriptor
714 * table which we handle as if it's any normal descriptor chain.
715 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100716 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
717 if (desc[i].len % sizeof(struct vring_desc))
718 errx(1, "Invalid size for indirect buffer table");
719
720 max = desc[i].len / sizeof(struct vring_desc);
721 desc = check_pointer(desc[i].addr, desc[i].len);
722 i = 0;
723 }
724
Rusty Russell17cbca22007-10-22 11:24:22 +1000725 do {
726 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100727 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000728 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100729 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000730 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100731 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000732 (*in_num)++;
733 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600734 /*
735 * If it's an output descriptor, they're all supposed
736 * to come before any input descriptors.
737 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000738 if (*in_num)
739 errx(1, "Descriptor has out after in");
740 (*out_num)++;
741 }
742
743 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100744 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000745 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100746 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000747
748 return head;
749}
750
Rusty Russell2e04ef72009-07-30 16:03:45 -0600751/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600752 * After we've used one of their buffers, we tell the Guest about it. Sometime
753 * later we'll want to send them an interrupt using trigger_irq(); note that
754 * wait_for_vq_desc() does that for us if it has to wait.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600755 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000756static void add_used(struct virtqueue *vq, unsigned int head, int len)
757{
758 struct vring_used_elem *used;
759
Rusty Russell2e04ef72009-07-30 16:03:45 -0600760 /*
761 * The virtqueue contains a ring of used buffers. Get a pointer to the
762 * next entry in that used ring.
763 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000764 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
765 used->id = head;
766 used->len = len;
767 /* Make sure buffer is written before we update index. */
768 wmb();
769 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600770 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000771}
772
Rusty Russell17cbca22007-10-22 11:24:22 +1000773/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c82009-06-12 22:26:59 -0600774static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000775{
776 add_used(vq, head, len);
Rusty Russell56739c82009-06-12 22:26:59 -0600777 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000778}
779
Rusty Russelle1e72962007-10-25 15:02:50 +1000780/*
781 * The Console
782 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600783 * We associate some data with the console for our exit hack.
784 */
Rusty Russell1842f232009-07-30 16:03:46 -0600785struct console_abort {
Rusty Russelldde79782007-07-26 10:41:03 -0700786 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700787 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700788 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700789 struct timeval start;
790};
791
Rusty Russelldde79782007-07-26 10:41:03 -0700792/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600793static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700794{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700795 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000796 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600797 struct console_abort *abort = vq->dev->priv;
798 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700799
Rusty Russella91d74a2009-07-30 16:03:45 -0600800 /* Make sure there's a descriptor available. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600801 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000802 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000803 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700804
Rusty Russella91d74a2009-07-30 16:03:45 -0600805 /* Read into it. This is where we usually wait. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600806 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700807 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600808 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700809 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600810 /*
811 * For simplicity, dying threads kill the whole Launcher. So
812 * just nap here.
813 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600814 for (;;)
815 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700816 }
817
Rusty Russella91d74a2009-07-30 16:03:45 -0600818 /* Tell the Guest we used a buffer. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600819 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700820
Rusty Russell2e04ef72009-07-30 16:03:45 -0600821 /*
822 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700823 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600824 * This is such a hack, but works surprisingly well. Each ^C has to
825 * be in a buffer by itself, so they can't be too fast. But we check
826 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600827 * slow.
828 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600829 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700830 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600831 return;
832 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700833
Rusty Russell659a0e62009-06-12 22:27:10 -0600834 abort->count++;
835 if (abort->count == 1)
836 gettimeofday(&abort->start, NULL);
837 else if (abort->count == 3) {
838 struct timeval now;
839 gettimeofday(&now, NULL);
840 /* Kill all Launcher processes with SIGINT, like normal ^C */
841 if (now.tv_sec <= abort->start.tv_sec+1)
842 kill(0, SIGINT);
843 abort->count = 0;
844 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700845}
846
Rusty Russell659a0e62009-06-12 22:27:10 -0600847/* This is the routine which handles console output (ie. stdout). */
848static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700849{
Rusty Russell17cbca22007-10-22 11:24:22 +1000850 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000851 struct iovec iov[vq->vring.num];
852
Rusty Russella91d74a2009-07-30 16:03:45 -0600853 /* We usually wait in here, for the Guest to give us something. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600854 head = wait_for_vq_desc(vq, iov, &out, &in);
855 if (in)
856 errx(1, "Input buffers in console output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600857
858 /* writev can return a partial write, so we loop here. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600859 while (!iov_empty(iov, out)) {
860 int len = writev(STDOUT_FILENO, iov, out);
861 if (len <= 0)
862 err(1, "Write to stdout gave %i", len);
863 iov_consume(iov, out, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000864 }
Rusty Russella91d74a2009-07-30 16:03:45 -0600865
866 /*
867 * We're finished with that buffer: if we're going to sleep,
868 * wait_for_vq_desc() will prod the Guest with an interrupt.
869 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600870 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500871}
872
Rusty Russelle1e72962007-10-25 15:02:50 +1000873/*
874 * The Network
875 *
876 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600877 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500878 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600879struct net_info {
880 int tunfd;
881};
882
883static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700884{
Rusty Russell659a0e62009-06-12 22:27:10 -0600885 struct net_info *net_info = vq->dev->priv;
886 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000887 struct iovec iov[vq->vring.num];
888
Rusty Russella91d74a2009-07-30 16:03:45 -0600889 /* We usually wait in here for the Guest to give us a packet. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600890 head = wait_for_vq_desc(vq, iov, &out, &in);
891 if (in)
892 errx(1, "Input buffers in net output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600893 /*
894 * Send the whole thing through to /dev/net/tun. It expects the exact
895 * same format: what a coincidence!
896 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600897 if (writev(net_info->tunfd, iov, out) < 0)
898 errx(1, "Write to tun failed?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600899
900 /*
901 * Done with that one; wait_for_vq_desc() will send the interrupt if
902 * all packets are processed.
903 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600904 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700905}
906
Rusty Russella91d74a2009-07-30 16:03:45 -0600907/*
908 * Handling network input is a bit trickier, because I've tried to optimize it.
909 *
910 * First we have a helper routine which tells is if from this file descriptor
911 * (ie. the /dev/net/tun device) will block:
912 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600913static bool will_block(int fd)
914{
915 fd_set fdset;
916 struct timeval zero = { 0, 0 };
917 FD_ZERO(&fdset);
918 FD_SET(fd, &fdset);
919 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
920}
921
Rusty Russella91d74a2009-07-30 16:03:45 -0600922/*
923 * This handles packets coming in from the tun device to our Guest. Like all
924 * service routines, it gets called again as soon as it returns, so you don't
925 * see a while(1) loop here.
926 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600927static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700928{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700929 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600930 unsigned int head, out, in;
931 struct iovec iov[vq->vring.num];
932 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700933
Rusty Russella91d74a2009-07-30 16:03:45 -0600934 /*
935 * Get a descriptor to write an incoming packet into. This will also
936 * send an interrupt if they're out of descriptors.
937 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600938 head = wait_for_vq_desc(vq, iov, &out, &in);
939 if (out)
940 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600941
Rusty Russella91d74a2009-07-30 16:03:45 -0600942 /*
943 * If it looks like we'll block reading from the tun device, send them
944 * an interrupt.
945 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600946 if (vq->pending_used && will_block(net_info->tunfd))
947 trigger_irq(vq);
948
Rusty Russella91d74a2009-07-30 16:03:45 -0600949 /*
950 * Read in the packet. This is where we normally wait (when there's no
951 * incoming network traffic).
952 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600953 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700954 if (len <= 0)
Rusty Russell659a0e62009-06-12 22:27:10 -0600955 err(1, "Failed to read from tun.");
Rusty Russella91d74a2009-07-30 16:03:45 -0600956
957 /*
958 * Mark that packet buffer as used, but don't interrupt here. We want
959 * to wait until we've done as much work as we can.
960 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600961 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700962}
Rusty Russella91d74a2009-07-30 16:03:45 -0600963/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700964
Rusty Russella91d74a2009-07-30 16:03:45 -0600965/* This is the helper to create threads: run the service routine in a loop. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600966static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000967{
Rusty Russell659a0e62009-06-12 22:27:10 -0600968 struct virtqueue *vq = _vq;
969
970 for (;;)
971 vq->service(vq);
972 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000973}
974
Rusty Russell2e04ef72009-07-30 16:03:45 -0600975/*
976 * When a child dies, we kill our entire process group with SIGTERM. This
977 * also has the side effect that the shell restores the console for us!
978 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600979static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500980{
Rusty Russell659a0e62009-06-12 22:27:10 -0600981 kill(0, SIGTERM);
982}
983
984static void reset_device(struct device *dev)
985{
986 struct virtqueue *vq;
987
988 verbose("Resetting device %s\n", dev->name);
989
990 /* Clear any features they've acked. */
991 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
992
993 /* We're going to be explicitly killing threads, so ignore them. */
994 signal(SIGCHLD, SIG_IGN);
995
996 /* Zero out the virtqueues, get rid of their threads */
997 for (vq = dev->vq; vq; vq = vq->next) {
998 if (vq->thread != (pid_t)-1) {
999 kill(vq->thread, SIGTERM);
1000 waitpid(vq->thread, NULL, 0);
1001 vq->thread = (pid_t)-1;
1002 }
1003 memset(vq->vring.desc, 0,
1004 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1005 lg_last_avail(vq) = 0;
1006 }
1007 dev->running = false;
1008
1009 /* Now we care if threads die. */
1010 signal(SIGCHLD, (void *)kill_launcher);
1011}
1012
Rusty Russella91d74a2009-07-30 16:03:45 -06001013/*L:216
1014 * This actually creates the thread which services the virtqueue for a device.
1015 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001016static void create_thread(struct virtqueue *vq)
1017{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001018 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001019 * Create stack for thread. Since the stack grows upwards, we point
1020 * the stack pointer to the end of this region.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001021 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001022 char *stack = malloc(32768);
1023 unsigned long args[] = { LHREQ_EVENTFD,
1024 vq->config.pfn*getpagesize(), 0 };
1025
1026 /* Create a zero-initialized eventfd. */
1027 vq->eventfd = eventfd(0, 0);
1028 if (vq->eventfd < 0)
1029 err(1, "Creating eventfd");
1030 args[2] = vq->eventfd;
1031
Rusty Russella91d74a2009-07-30 16:03:45 -06001032 /*
1033 * Attach an eventfd to this virtqueue: it will go off when the Guest
1034 * does an LHCALL_NOTIFY for this vq.
1035 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001036 if (write(lguest_fd, &args, sizeof(args)) != 0)
1037 err(1, "Attaching eventfd");
1038
Rusty Russella91d74a2009-07-30 16:03:45 -06001039 /*
1040 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1041 * we get a signal if it dies.
1042 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001043 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1044 if (vq->thread == (pid_t)-1)
1045 err(1, "Creating clone");
Rusty Russella91d74a2009-07-30 16:03:45 -06001046
1047 /* We close our local copy now the child has it. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001048 close(vq->eventfd);
1049}
1050
Rusty Russellca60a422009-09-23 22:26:47 -06001051static bool accepted_feature(struct device *dev, unsigned int bit)
1052{
1053 const u8 *features = get_feature_bits(dev) + dev->feature_len;
1054
1055 if (dev->feature_len < bit / CHAR_BIT)
1056 return false;
1057 return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT));
1058}
1059
Rusty Russell659a0e62009-06-12 22:27:10 -06001060static void start_device(struct device *dev)
1061{
1062 unsigned int i;
1063 struct virtqueue *vq;
1064
1065 verbose("Device %s OK: offered", dev->name);
1066 for (i = 0; i < dev->feature_len; i++)
1067 verbose(" %02x", get_feature_bits(dev)[i]);
1068 verbose(", accepted");
1069 for (i = 0; i < dev->feature_len; i++)
1070 verbose(" %02x", get_feature_bits(dev)
1071 [dev->feature_len+i]);
1072
Rusty Russellca60a422009-09-23 22:26:47 -06001073 dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1074
Rusty Russell659a0e62009-06-12 22:27:10 -06001075 for (vq = dev->vq; vq; vq = vq->next) {
1076 if (vq->service)
1077 create_thread(vq);
1078 }
1079 dev->running = true;
1080}
1081
1082static void cleanup_devices(void)
1083{
1084 struct device *dev;
1085
1086 for (dev = devices.dev; dev; dev = dev->next)
1087 reset_device(dev);
1088
1089 /* If we saved off the original terminal settings, restore them now. */
1090 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1091 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001092}
1093
Rusty Russella007a752008-05-02 21:50:53 -05001094/* When the Guest tells us they updated the status field, we handle it. */
1095static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001096{
Rusty Russell659a0e62009-06-12 22:27:10 -06001097 /* A zero status is a reset, otherwise it's a set of flags. */
1098 if (dev->desc->status == 0)
1099 reset_device(dev);
1100 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001101 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001102 if (dev->running)
1103 reset_device(dev);
Rusty Russella007a752008-05-02 21:50:53 -05001104 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001105 if (!dev->running)
1106 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001107 }
1108}
1109
Rusty Russella91d74a2009-07-30 16:03:45 -06001110/*L:215
1111 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1112 * particular, it's used to notify us of device status changes during boot.
1113 */
Rusty Russell56739c82009-06-12 22:26:59 -06001114static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001115{
1116 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001117
Rusty Russell659a0e62009-06-12 22:27:10 -06001118 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001119 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001120 struct virtqueue *vq;
1121
Rusty Russella91d74a2009-07-30 16:03:45 -06001122 /*
1123 * Notifications to device descriptors mean they updated the
1124 * device status.
1125 */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001126 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001127 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001128 return;
1129 }
1130
Rusty Russella91d74a2009-07-30 16:03:45 -06001131 /*
1132 * Devices *can* be used before status is set to DRIVER_OK.
1133 * The original plan was that they would never do this: they
1134 * would always finish setting up their status bits before
1135 * actually touching the virtqueues. In practice, we allowed
1136 * them to, and they do (eg. the disk probes for partition
1137 * tables as part of initialization).
1138 *
1139 * If we see this, we start the device: once it's running, we
1140 * expect the device to catch all the notifications.
1141 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001142 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001143 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001144 continue;
Rusty Russell659a0e62009-06-12 22:27:10 -06001145 if (i->running)
1146 errx(1, "Notification on running %s", i->name);
Rusty Russella91d74a2009-07-30 16:03:45 -06001147 /* This just calls create_thread() for each virtqueue */
Rusty Russell659a0e62009-06-12 22:27:10 -06001148 start_device(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001149 return;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001150 }
1151 }
Rusty Russelldde79782007-07-26 10:41:03 -07001152
Rusty Russell2e04ef72009-07-30 16:03:45 -06001153 /*
1154 * Early console write is done using notify on a nul-terminated string
1155 * in Guest memory. It's also great for hacking debugging messages
1156 * into a Guest.
1157 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001158 if (addr >= guest_limit)
1159 errx(1, "Bad NOTIFY %#lx", addr);
1160
1161 write(STDOUT_FILENO, from_guest_phys(addr),
1162 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001163}
1164
Rusty Russelldde79782007-07-26 10:41:03 -07001165/*L:190
1166 * Device Setup
1167 *
1168 * All devices need a descriptor so the Guest knows it exists, and a "struct
1169 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001170 * routines to allocate and manage them.
1171 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001172
Rusty Russell2e04ef72009-07-30 16:03:45 -06001173/*
1174 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001175 * number of virtqueue descriptors, then two sets of feature bits, then an
1176 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001177 * pointer.
1178 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001179static u8 *device_config(const struct device *dev)
1180{
1181 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001182 + dev->num_vq * sizeof(struct lguest_vqconfig)
1183 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001184}
1185
Rusty Russell2e04ef72009-07-30 16:03:45 -06001186/*
1187 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001188 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001189 * that descriptor.
1190 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001191static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001192{
Rusty Russella586d4f2008-02-04 23:49:56 -05001193 struct lguest_device_desc d = { .type = type };
1194 void *p;
1195
1196 /* Figure out where the next device config is, based on the last one. */
1197 if (devices.lastdev)
1198 p = device_config(devices.lastdev)
1199 + devices.lastdev->desc->config_len;
1200 else
1201 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001202
Rusty Russell17cbca22007-10-22 11:24:22 +10001203 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001204 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001205 errx(1, "Too many devices");
1206
Rusty Russella586d4f2008-02-04 23:49:56 -05001207 /* p might not be aligned, so we memcpy in. */
1208 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001209}
1210
Rusty Russell2e04ef72009-07-30 16:03:45 -06001211/*
1212 * Each device descriptor is followed by the description of its virtqueues. We
1213 * specify how many descriptors the virtqueue is to have.
1214 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001215static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001216 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001217{
1218 unsigned int pages;
1219 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1220 void *p;
1221
Rusty Russella6bd8e12008-03-28 11:05:53 -05001222 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001223 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001224 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001225 p = get_pages(pages);
1226
Rusty Russelld1c856e2007-11-19 11:20:40 -05001227 /* Initialize the virtqueue */
1228 vq->next = NULL;
1229 vq->last_avail_idx = 0;
1230 vq->dev = dev;
Rusty Russella91d74a2009-07-30 16:03:45 -06001231
1232 /*
1233 * This is the routine the service thread will run, and its Process ID
1234 * once it's running.
1235 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001236 vq->service = service;
1237 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001238
Rusty Russell17cbca22007-10-22 11:24:22 +10001239 /* Initialize the configuration. */
1240 vq->config.num = num_descs;
1241 vq->config.irq = devices.next_irq++;
1242 vq->config.pfn = to_guest_phys(p) / getpagesize();
1243
1244 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001245 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001246
Rusty Russell2e04ef72009-07-30 16:03:45 -06001247 /*
1248 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001249 * device_config() to get the end of the device's current virtqueues;
1250 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001251 * yet, otherwise we'd be overwriting them.
1252 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001253 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1254 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001255 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001256 dev->desc->num_vq++;
1257
1258 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001259
Rusty Russell2e04ef72009-07-30 16:03:45 -06001260 /*
1261 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1262 * second.
1263 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001264 for (i = &dev->vq; *i; i = &(*i)->next);
1265 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001266}
1267
Rusty Russell2e04ef72009-07-30 16:03:45 -06001268/*
1269 * The first half of the feature bitmask is for us to advertise features. The
1270 * second half is for the Guest to accept features.
1271 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001272static void add_feature(struct device *dev, unsigned bit)
1273{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001274 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001275
1276 /* We can't extend the feature bits once we've added config bytes */
1277 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1278 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001279 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001280 }
1281
Rusty Russella586d4f2008-02-04 23:49:56 -05001282 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1283}
1284
Rusty Russell2e04ef72009-07-30 16:03:45 -06001285/*
1286 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001287 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001288 * how we use it.
1289 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001290static void set_config(struct device *dev, unsigned len, const void *conf)
1291{
1292 /* Check we haven't overflowed our single page. */
1293 if (device_config(dev) + len > devices.descpage + getpagesize())
1294 errx(1, "Too many devices");
1295
1296 /* Copy in the config information, and store the length. */
1297 memcpy(device_config(dev), conf, len);
1298 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001299
1300 /* Size must fit in config_len field (8 bits)! */
1301 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001302}
1303
Rusty Russell2e04ef72009-07-30 16:03:45 -06001304/*
1305 * This routine does all the creation and setup of a new device, including
Rusty Russella91d74a2009-07-30 16:03:45 -06001306 * calling new_dev_desc() to allocate the descriptor and device memory. We
1307 * don't actually start the service threads until later.
Rusty Russella6bd8e12008-03-28 11:05:53 -05001308 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001309 * See what I mean about userspace being boring?
1310 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001311static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001312{
1313 struct device *dev = malloc(sizeof(*dev));
1314
Rusty Russelldde79782007-07-26 10:41:03 -07001315 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001316 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001317 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001318 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001319 dev->feature_len = 0;
1320 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001321 dev->running = false;
Rusty Russella586d4f2008-02-04 23:49:56 -05001322
Rusty Russell2e04ef72009-07-30 16:03:45 -06001323 /*
1324 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001325 * easier, but the user expects the devices to be arranged on the bus
1326 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001327 * is eth0, the first block device /dev/vda, etc.
1328 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001329 if (devices.lastdev)
1330 devices.lastdev->next = dev;
1331 else
1332 devices.dev = dev;
1333 devices.lastdev = dev;
1334
Rusty Russell8ca47e02007-07-19 01:49:29 -07001335 return dev;
1336}
1337
Rusty Russell2e04ef72009-07-30 16:03:45 -06001338/*
1339 * Our first setup routine is the console. It's a fairly simple device, but
1340 * UNIX tty handling makes it uglier than it could be.
1341 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001342static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001343{
1344 struct device *dev;
1345
Rusty Russelldde79782007-07-26 10:41:03 -07001346 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001347 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1348 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001349 /*
1350 * Then we turn off echo, line buffering and ^C etc: We want a
1351 * raw input stream to the Guest.
1352 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001353 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1354 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001355 }
1356
Rusty Russell659a0e62009-06-12 22:27:10 -06001357 dev = new_device("console", VIRTIO_ID_CONSOLE);
1358
Rusty Russelldde79782007-07-26 10:41:03 -07001359 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001360 dev->priv = malloc(sizeof(struct console_abort));
1361 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001362
Rusty Russell2e04ef72009-07-30 16:03:45 -06001363 /*
1364 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001365 * they put something the input queue, we make sure we're listening to
1366 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001367 * stdout.
1368 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001369 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1370 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001371
Rusty Russell659a0e62009-06-12 22:27:10 -06001372 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001373}
Rusty Russelldde79782007-07-26 10:41:03 -07001374/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001375
Rusty Russell2e04ef72009-07-30 16:03:45 -06001376/*M:010
1377 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001378 * --sharenet=<name> option which opens or creates a named pipe. This can be
1379 * used to send packets to another guest in a 1:1 manner.
1380 *
1381 * More sopisticated is to use one of the tools developed for project like UML
1382 * to do networking.
1383 *
1384 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1385 * completely generic ("here's my vring, attach to your vring") and would work
1386 * for any traffic. Of course, namespace and permissions issues need to be
1387 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1388 * multiple inter-guest channels behind one interface, although it would
1389 * require some manner of hotplugging new virtio channels.
1390 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001391 * Finally, we could implement a virtio network switch in the kernel.
1392:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001393
Rusty Russell8ca47e02007-07-19 01:49:29 -07001394static u32 str2ip(const char *ipaddr)
1395{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001396 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001397
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001398 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1399 errx(1, "Failed to parse IP address '%s'", ipaddr);
1400 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1401}
1402
1403static void str2mac(const char *macaddr, unsigned char mac[6])
1404{
1405 unsigned int m[6];
1406 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1407 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1408 errx(1, "Failed to parse mac address '%s'", macaddr);
1409 mac[0] = m[0];
1410 mac[1] = m[1];
1411 mac[2] = m[2];
1412 mac[3] = m[3];
1413 mac[4] = m[4];
1414 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001415}
1416
Rusty Russell2e04ef72009-07-30 16:03:45 -06001417/*
1418 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001419 * network device to the bridge device specified by the command line.
1420 *
1421 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001422 * dislike bridging), and I just try not to break it.
1423 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001424static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1425{
1426 int ifidx;
1427 struct ifreq ifr;
1428
1429 if (!*br_name)
1430 errx(1, "must specify bridge name");
1431
1432 ifidx = if_nametoindex(if_name);
1433 if (!ifidx)
1434 errx(1, "interface %s does not exist!", if_name);
1435
1436 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001437 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001438 ifr.ifr_ifindex = ifidx;
1439 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1440 err(1, "can't add %s to bridge %s", if_name, br_name);
1441}
1442
Rusty Russell2e04ef72009-07-30 16:03:45 -06001443/*
1444 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001445 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001446 * pointer.
1447 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001448static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001449{
1450 struct ifreq ifr;
1451 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1452
1453 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001454 strcpy(ifr.ifr_name, tapif);
1455
1456 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001457 sin->sin_family = AF_INET;
1458 sin->sin_addr.s_addr = htonl(ipaddr);
1459 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001460 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001461 ifr.ifr_flags = IFF_UP;
1462 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001463 err(1, "Bringing interface %s up", tapif);
1464}
1465
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001466static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001467{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001468 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001469 int netfd;
1470
1471 /* Start with this zeroed. Messy but sure. */
1472 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001473
Rusty Russell2e04ef72009-07-30 16:03:45 -06001474 /*
1475 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001476 * tap device is like a tun device, only somehow different. To tell
1477 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001478 * works now!
1479 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001480 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001481 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001482 strcpy(ifr.ifr_name, "tap%d");
1483 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1484 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001485
Rusty Russell398f1872008-07-29 09:58:37 -05001486 if (ioctl(netfd, TUNSETOFFLOAD,
1487 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1488 err(1, "Could not set features for tun device");
1489
Rusty Russell2e04ef72009-07-30 16:03:45 -06001490 /*
1491 * We don't need checksums calculated for packets coming in this
1492 * device: trust us!
1493 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001494 ioctl(netfd, TUNSETNOCSUM, 1);
1495
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001496 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1497 return netfd;
1498}
1499
Rusty Russell2e04ef72009-07-30 16:03:45 -06001500/*L:195
1501 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001502 * routing, but the principle is the same: it uses the "tun" device to inject
1503 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001504 * just shunt packets between the Guest and the tun device.
1505 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001506static void setup_tun_net(char *arg)
1507{
1508 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001509 struct net_info *net_info = malloc(sizeof(*net_info));
1510 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001511 u32 ip = INADDR_ANY;
1512 bool bridging = false;
1513 char tapif[IFNAMSIZ], *p;
1514 struct virtio_net_config conf;
1515
Rusty Russell659a0e62009-06-12 22:27:10 -06001516 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001517
Rusty Russell17cbca22007-10-22 11:24:22 +10001518 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001519 dev = new_device("net", VIRTIO_ID_NET);
1520 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001521
Rusty Russell2e04ef72009-07-30 16:03:45 -06001522 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001523 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1524 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001525
Rusty Russell2e04ef72009-07-30 16:03:45 -06001526 /*
1527 * We need a socket to perform the magic network ioctls to bring up the
1528 * tap interface, connect to the bridge etc. Any socket will do!
1529 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001530 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1531 if (ipfd < 0)
1532 err(1, "opening IP socket");
1533
Rusty Russelldde79782007-07-26 10:41:03 -07001534 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001535 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001536 arg += strlen(BRIDGE_PFX);
1537 bridging = true;
1538 }
1539
1540 /* A mac address may follow the bridge name or IP address */
1541 p = strchr(arg, ':');
1542 if (p) {
1543 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001544 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001545 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001546 }
1547
1548 /* arg is now either an IP address or a bridge name */
1549 if (bridging)
1550 add_to_bridge(ipfd, tapif, arg);
1551 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001552 ip = str2ip(arg);
1553
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001554 /* Set up the tun device. */
1555 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001556
Rusty Russell20887612008-05-30 15:09:46 -05001557 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
Rusty Russell398f1872008-07-29 09:58:37 -05001558 /* Expect Guest to handle everything except UFO */
1559 add_feature(dev, VIRTIO_NET_F_CSUM);
1560 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001561 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1562 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1563 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1564 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1565 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1566 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001567 /* We handle indirect ring entries */
1568 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russella586d4f2008-02-04 23:49:56 -05001569 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001570
Rusty Russella586d4f2008-02-04 23:49:56 -05001571 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001572 close(ipfd);
1573
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001574 devices.device_num++;
1575
1576 if (bridging)
1577 verbose("device %u: tun %s attached to bridge: %s\n",
1578 devices.device_num, tapif, arg);
1579 else
1580 verbose("device %u: tun %s: %s\n",
1581 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001582}
Rusty Russella91d74a2009-07-30 16:03:45 -06001583/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001584
Rusty Russelle1e72962007-10-25 15:02:50 +10001585/* This hangs off device->priv. */
Rusty Russell1842f232009-07-30 16:03:46 -06001586struct vblk_info {
Rusty Russell17cbca22007-10-22 11:24:22 +10001587 /* The size of the file. */
1588 off64_t len;
1589
1590 /* The file descriptor for the file. */
1591 int fd;
1592
Rusty Russell17cbca22007-10-22 11:24:22 +10001593};
1594
Rusty Russelle1e72962007-10-25 15:02:50 +10001595/*L:210
1596 * The Disk
1597 *
Rusty Russella91d74a2009-07-30 16:03:45 -06001598 * The disk only has one virtqueue, so it only has one thread. It is really
1599 * simple: the Guest asks for a block number and we read or write that position
1600 * in the file.
1601 *
1602 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1603 * slow: the Guest waits until the read is finished before running anything
1604 * else, even if it could have been doing useful work.
1605 *
1606 * We could have used async I/O, except it's reputed to suck so hard that
1607 * characters actually go missing from your code when you try to use it.
Rusty Russelle1e72962007-10-25 15:02:50 +10001608 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001609static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001610{
Rusty Russell659a0e62009-06-12 22:27:10 -06001611 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001612 unsigned int head, out_num, in_num, wlen;
1613 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001614 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001615 struct virtio_blk_outhdr *out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001616 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001617 off64_t off;
1618
Rusty Russella91d74a2009-07-30 16:03:45 -06001619 /*
1620 * Get the next request, where we normally wait. It triggers the
1621 * interrupt to acknowledge previously serviced requests (if any).
1622 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001623 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001624
Rusty Russell2e04ef72009-07-30 16:03:45 -06001625 /*
1626 * Every block request should contain at least one output buffer
Rusty Russelle1e72962007-10-25 15:02:50 +10001627 * (detailing the location on disk and the type of request) and one
Rusty Russell2e04ef72009-07-30 16:03:45 -06001628 * input buffer (to hold the result).
1629 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001630 if (out_num == 0 || in_num == 0)
1631 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1632 head, out_num, in_num);
1633
1634 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001635 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russella91d74a2009-07-30 16:03:45 -06001636 /*
1637 * For historical reasons, block operations are expressed in 512 byte
1638 * "sectors".
1639 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001640 off = out->sector * 512;
1641
Rusty Russell2e04ef72009-07-30 16:03:45 -06001642 /*
1643 * The block device implements "barriers", where the Guest indicates
Rusty Russelle1e72962007-10-25 15:02:50 +10001644 * that it wants all previous writes to occur before this write. We
1645 * don't have a way of asking our kernel to do a barrier, so we just
Rusty Russell2e04ef72009-07-30 16:03:45 -06001646 * synchronize all the data in the file. Pretty poor, no?
1647 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001648 if (out->type & VIRTIO_BLK_T_BARRIER)
1649 fdatasync(vblk->fd);
1650
Rusty Russell2e04ef72009-07-30 16:03:45 -06001651 /*
1652 * In general the virtio block driver is allowed to try SCSI commands.
1653 * It'd be nice if we supported eject, for example, but we don't.
1654 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001655 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1656 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001657 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001658 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001659 } else if (out->type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001660 /*
1661 * Write
1662 *
1663 * Move to the right location in the block file. This can fail
1664 * if they try to write past end.
1665 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001666 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1667 err(1, "Bad seek to sector %llu", out->sector);
1668
1669 ret = writev(vblk->fd, iov+1, out_num-1);
1670 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1671
Rusty Russell2e04ef72009-07-30 16:03:45 -06001672 /*
1673 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001674 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001675 * file (possibly extending it).
1676 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001677 if (ret > 0 && off + ret > vblk->len) {
1678 /* Trim it back to the correct length */
1679 ftruncate64(vblk->fd, vblk->len);
1680 /* Die, bad Guest, die. */
1681 errx(1, "Write past end %llu+%u", off, ret);
1682 }
Anthony Liguori1200e642007-11-08 21:13:44 -06001683 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001684 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001685 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001686 /*
1687 * Read
1688 *
1689 * Move to the right location in the block file. This can fail
1690 * if they try to read past end.
1691 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001692 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1693 err(1, "Bad seek to sector %llu", out->sector);
1694
1695 ret = readv(vblk->fd, iov+1, in_num-1);
1696 verbose("READ from sector %llu: %i\n", out->sector, ret);
1697 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001698 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001699 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001700 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001701 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001702 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001703 }
1704 }
1705
Rusty Russell2e04ef72009-07-30 16:03:45 -06001706 /*
1707 * OK, so we noted that it was pretty poor to use an fdatasync as a
Rusty Russelld1881d32009-03-30 21:55:25 -06001708 * barrier. But Christoph Hellwig points out that we need a sync
1709 * *afterwards* as well: "Barriers specify no reordering to the front
Rusty Russell2e04ef72009-07-30 16:03:45 -06001710 * or the back." And Jens Axboe confirmed it, so here we are:
1711 */
Rusty Russelld1881d32009-03-30 21:55:25 -06001712 if (out->type & VIRTIO_BLK_T_BARRIER)
1713 fdatasync(vblk->fd);
1714
Rusty Russella91d74a2009-07-30 16:03:45 -06001715 /* Finished that request. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001716 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001717}
1718
Rusty Russelle1e72962007-10-25 15:02:50 +10001719/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001720static void setup_block_file(const char *filename)
1721{
Rusty Russell17cbca22007-10-22 11:24:22 +10001722 struct device *dev;
1723 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001724 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001725
Rusty Russell2e04ef72009-07-30 16:03:45 -06001726 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001727 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001728
Rusty Russelle1e72962007-10-25 15:02:50 +10001729 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001730 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001731
1732 /* Allocate the room for our own bookkeeping */
1733 vblk = dev->priv = malloc(sizeof(*vblk));
1734
1735 /* First we open the file and store the length. */
1736 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1737 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1738
Rusty Russella586d4f2008-02-04 23:49:56 -05001739 /* We support barriers. */
1740 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1741
Rusty Russell17cbca22007-10-22 11:24:22 +10001742 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001743 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001744
Rusty Russell2e04ef72009-07-30 16:03:45 -06001745 /*
1746 * Tell Guest not to put in too many descriptors at once: two are used
1747 * for the in and out elements.
1748 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001749 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1750 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1751
Rusty Russell8ef562d2009-07-30 16:03:43 -06001752 /* Don't try to put whole struct: we have 8 bit limit. */
1753 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001754
Rusty Russell17cbca22007-10-22 11:24:22 +10001755 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001756 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001757}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001758
Rusty Russell2e04ef72009-07-30 16:03:45 -06001759/*L:211
1760 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001761 * input buffers. The usual case is that the Guest doesn't want random numbers
1762 * and so has no buffers although /dev/random is still readable, whereas
1763 * console is the reverse.
1764 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001765 * The same logic applies, however.
1766 */
1767struct rng_info {
1768 int rfd;
1769};
1770
Rusty Russell659a0e62009-06-12 22:27:10 -06001771static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001772{
1773 int len;
1774 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001775 struct rng_info *rng_info = vq->dev->priv;
1776 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001777
1778 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001779 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001780 if (out_num)
1781 errx(1, "Output buffers in rng?");
1782
Rusty Russell2e04ef72009-07-30 16:03:45 -06001783 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001784 * Just like the console write, we loop to cover the whole iovec.
1785 * In this case, short reads actually happen quite a bit.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001786 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001787 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001788 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001789 if (len <= 0)
1790 err(1, "Read from /dev/random gave %i", len);
1791 iov_consume(iov, in_num, len);
1792 totlen += len;
1793 }
1794
1795 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001796 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001797}
1798
Rusty Russell2e04ef72009-07-30 16:03:45 -06001799/*L:199
1800 * This creates a "hardware" random number device for the Guest.
1801 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001802static void setup_rng(void)
1803{
1804 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001805 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001806
Rusty Russell2e04ef72009-07-30 16:03:45 -06001807 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001808 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001809
Rusty Russell2e04ef72009-07-30 16:03:45 -06001810 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001811 dev = new_device("rng", VIRTIO_ID_RNG);
1812 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001813
1814 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001815 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001816
1817 verbose("device %u: rng\n", devices.device_num++);
1818}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001819/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301820
Rusty Russella6bd8e12008-03-28 11:05:53 -05001821/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301822static void __attribute__((noreturn)) restart_guest(void)
1823{
1824 unsigned int i;
1825
Rusty Russell2e04ef72009-07-30 16:03:45 -06001826 /*
1827 * Since we don't track all open fds, we simply close everything beyond
1828 * stderr.
1829 */
Balaji Raoec04b132007-12-28 14:26:24 +05301830 for (i = 3; i < FD_SETSIZE; i++)
1831 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001832
Rusty Russell659a0e62009-06-12 22:27:10 -06001833 /* Reset all the devices (kills all threads). */
1834 cleanup_devices();
1835
Balaji Raoec04b132007-12-28 14:26:24 +05301836 execv(main_args[0], main_args);
1837 err(1, "Could not exec %s", main_args[0]);
1838}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001839
Rusty Russell2e04ef72009-07-30 16:03:45 -06001840/*L:220
1841 * Finally we reach the core of the Launcher which runs the Guest, serves
1842 * its input and output, and finally, lays it to rest.
1843 */
Rusty Russell56739c82009-06-12 22:26:59 -06001844static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001845{
1846 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001847 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001848 int readval;
1849
1850 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001851 readval = pread(lguest_fd, &notify_addr,
1852 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001853
Rusty Russell17cbca22007-10-22 11:24:22 +10001854 /* One unsigned long means the Guest did HCALL_NOTIFY */
1855 if (readval == sizeof(notify_addr)) {
1856 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c82009-06-12 22:26:59 -06001857 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001858 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001859 } else if (errno == ENOENT) {
1860 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001861 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001862 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301863 /* ERESTART means that we need to reboot the guest */
1864 } else if (errno == ERESTART) {
1865 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001866 /* Anything else means a bug or incompatible change. */
1867 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001868 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001869 }
1870}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001871/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001872 * This is the end of the Launcher. The good news: we are over halfway
1873 * through! The bad news: the most fiendish part of the code still lies ahead
1874 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001875 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001876 * Are you ready? Take a deep breath and join me in the core of the Host, in
1877 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001878:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001879
1880static struct option opts[] = {
1881 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001882 { "tunnet", 1, NULL, 't' },
1883 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001884 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001885 { "initrd", 1, NULL, 'i' },
1886 { NULL },
1887};
1888static void usage(void)
1889{
1890 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001891 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001892 "|--block=<filename>|--initrd=<filename>]...\n"
1893 "<mem-in-mb> vmlinux [args...]");
1894}
1895
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001896/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001897int main(int argc, char *argv[])
1898{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001899 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001900 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c82009-06-12 22:26:59 -06001901 /* Two temporaries. */
1902 int i, c;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001903 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001904 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001905 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001906 const char *initrd_name = NULL;
1907
Balaji Raoec04b132007-12-28 14:26:24 +05301908 /* Save the args: we "reboot" by execing ourselves again. */
1909 main_args = argv;
Balaji Raoec04b132007-12-28 14:26:24 +05301910
Rusty Russell2e04ef72009-07-30 16:03:45 -06001911 /*
1912 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001913 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001914 * remember that 0 is used by the timer).
1915 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001916 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001917 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001918
Rusty Russella91d74a2009-07-30 16:03:45 -06001919 /* We're CPU 0. In fact, that's the only CPU possible right now. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001920 cpu_id = 0;
Rusty Russella91d74a2009-07-30 16:03:45 -06001921
Rusty Russell2e04ef72009-07-30 16:03:45 -06001922 /*
1923 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001924 * descriptor and memory pages for the devices as we parse the command
1925 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001926 * of memory now.
1927 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001928 for (i = 1; i < argc; i++) {
1929 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001930 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001931 /*
1932 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001933 * guest-physical memory range. This fills it with 0,
1934 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001935 * tries to access it.
1936 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001937 guest_base = map_zeroed_pages(mem / getpagesize()
1938 + DEVICE_PAGES);
1939 guest_limit = mem;
1940 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001941 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001942 break;
1943 }
1944 }
Rusty Russelldde79782007-07-26 10:41:03 -07001945
1946 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001947 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1948 switch (c) {
1949 case 'v':
1950 verbose = true;
1951 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001952 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001953 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001954 break;
1955 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001956 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001957 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001958 case 'r':
1959 setup_rng();
1960 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001961 case 'i':
1962 initrd_name = optarg;
1963 break;
1964 default:
1965 warnx("Unknown argument %s", argv[optind]);
1966 usage();
1967 }
1968 }
Rusty Russell2e04ef72009-07-30 16:03:45 -06001969 /*
1970 * After the other arguments we expect memory and kernel image name,
1971 * followed by command line arguments for the kernel.
1972 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001973 if (optind + 2 > argc)
1974 usage();
1975
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001976 verbose("Guest base is at %p\n", guest_base);
1977
Rusty Russelldde79782007-07-26 10:41:03 -07001978 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001979 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001980
Rusty Russell8ca47e02007-07-19 01:49:29 -07001981 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001982 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001983
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001984 /* Boot information is stashed at physical address 0 */
1985 boot = from_guest_phys(0);
1986
Rusty Russelldde79782007-07-26 10:41:03 -07001987 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001988 if (initrd_name) {
1989 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06001990 /*
1991 * These are the location in the Linux boot header where the
1992 * start and size of the initrd are expected to be found.
1993 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001994 boot->hdr.ramdisk_image = mem - initrd_size;
1995 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001996 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001997 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001998 }
1999
Rusty Russell2e04ef72009-07-30 16:03:45 -06002000 /*
2001 * The Linux boot header contains an "E820" memory map: ours is a
2002 * simple, single region.
2003 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002004 boot->e820_entries = 1;
2005 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06002006 /*
2007 * The boot header contains a command line pointer: we put the command
2008 * line after the boot header.
2009 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002010 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10002011 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002012 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07002013
Rusty Russell814a0e52007-10-22 11:29:44 +10002014 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002015 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002016
2017 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002018 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002019
Rusty Russell43d33b22007-10-22 11:29:57 +10002020 /* Tell the entry path not to try to reload segment registers. */
2021 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002022
Rusty Russell2e04ef72009-07-30 16:03:45 -06002023 /*
2024 * We tell the kernel to initialize the Guest: this returns the open
2025 * /dev/lguest file descriptor.
2026 */
Rusty Russell56739c82009-06-12 22:26:59 -06002027 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002028
Rusty Russella91d74a2009-07-30 16:03:45 -06002029 /* Ensure that we terminate if a device-servicing child dies. */
Rusty Russell659a0e62009-06-12 22:27:10 -06002030 signal(SIGCHLD, kill_launcher);
2031
2032 /* If we exit via err(), this kills all the threads, restores tty. */
2033 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002034
Rusty Russelldde79782007-07-26 10:41:03 -07002035 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c82009-06-12 22:26:59 -06002036 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07002037}
Rusty Russellf56a3842007-07-26 10:41:05 -07002038/*:*/
2039
2040/*M:999
2041 * Mastery is done: you now know everything I do.
2042 *
2043 * But surely you have seen code, features and bugs in your wanderings which
2044 * you now yearn to attack? That is the real game, and I look forward to you
2045 * patching and forking lguest into the Your-Name-Here-visor.
2046 *
2047 * Farewell, and good coding!
2048 * Rusty Russell.
2049 */