blob: 6c8a2386cd50d150fb58844324c40fbb8685bb8c [file] [log] [blame]
Rusty Russellf938d2c2007-07-26 10:41:02 -07001/*P:100 This is the Launcher code, a simple program which lays out the
2 * "physical" memory for the new Guest by mapping the kernel image and the
3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10004:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07005#define _LARGEFILE64_SOURCE
6#define _GNU_SOURCE
7#include <stdio.h>
8#include <string.h>
9#include <unistd.h>
10#include <err.h>
11#include <stdint.h>
12#include <stdlib.h>
13#include <elf.h>
14#include <sys/mman.h>
Ronald G. Minnich6649bb72007-08-28 14:35:59 -070015#include <sys/param.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070016#include <sys/types.h>
17#include <sys/stat.h>
18#include <sys/wait.h>
19#include <fcntl.h>
20#include <stdbool.h>
21#include <errno.h>
22#include <ctype.h>
23#include <sys/socket.h>
24#include <sys/ioctl.h>
25#include <sys/time.h>
26#include <time.h>
27#include <netinet/in.h>
28#include <net/if.h>
29#include <linux/sockios.h>
30#include <linux/if_tun.h>
31#include <sys/uio.h>
32#include <termios.h>
33#include <getopt.h>
34#include <zlib.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100035#include <assert.h>
36#include <sched.h>
Rusty Russellb45d8cb2007-10-22 10:56:24 +100037#include "linux/lguest_launcher.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100038#include "linux/virtio_config.h"
39#include "linux/virtio_net.h"
40#include "linux/virtio_blk.h"
41#include "linux/virtio_console.h"
42#include "linux/virtio_ring.h"
Rusty Russell43d33b22007-10-22 11:29:57 +100043#include "asm-x86/bootparam.h"
Rusty Russelldb24e8c2007-10-25 14:09:25 +100044/*L:110 We can ignore the 38 include files we need for this program, but I do
45 * want to draw attention to the use of kernel-style types.
46 *
47 * As Linus said, "C is a Spartan language, and so should your naming be." I
48 * like these abbreviations, so we define them here. Note that u64 is always
49 * unsigned long long, which works on all Linux systems: this means that we can
50 * use %llu in printf for any u64. */
51typedef unsigned long long u64;
52typedef uint32_t u32;
53typedef uint16_t u16;
54typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070055/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070056
57#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
58#define NET_PEERNUM 1
59#define BRIDGE_PFX "bridge:"
60#ifndef SIOCBRADDIF
61#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
62#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100063/* We can have up to 256 pages for devices. */
64#define DEVICE_PAGES 256
Rusty Russell42b36cc2007-11-12 13:39:18 +110065/* This will occupy 2 pages: it must be a power of 2. */
66#define VIRTQUEUE_NUM 128
Rusty Russell8ca47e02007-07-19 01:49:29 -070067
Rusty Russelldde79782007-07-26 10:41:03 -070068/*L:120 verbose is both a global flag and a macro. The C preprocessor allows
69 * this, and although I wouldn't recommend it, it works quite nicely here. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070070static bool verbose;
71#define verbose(args...) \
72 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070073/*:*/
74
75/* The pipe to send commands to the waker process */
Rusty Russell8ca47e02007-07-19 01:49:29 -070076static int waker_fd;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100077/* The pointer to the start of guest memory. */
78static void *guest_base;
79/* The maximum guest physical address allowed, and maximum possible. */
80static unsigned long guest_limit, guest_max;
Rusty Russell8ca47e02007-07-19 01:49:29 -070081
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020082/* a per-cpu variable indicating whose vcpu is currently running */
83static unsigned int __thread cpu_id;
84
Rusty Russelldde79782007-07-26 10:41:03 -070085/* This is our list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070086struct device_list
87{
Rusty Russelldde79782007-07-26 10:41:03 -070088 /* Summary information about the devices in our list: ready to pass to
89 * select() to ask which need servicing.*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070090 fd_set infds;
91 int max_infd;
92
Rusty Russell17cbca22007-10-22 11:24:22 +100093 /* Counter to assign interrupt numbers. */
94 unsigned int next_irq;
95
96 /* Counter to print out convenient device numbers. */
97 unsigned int device_num;
98
Rusty Russelldde79782007-07-26 10:41:03 -070099 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000100 u8 *descpage;
101
102 /* The tail of the last descriptor. */
103 unsigned int desc_used;
Rusty Russelldde79782007-07-26 10:41:03 -0700104
105 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700106 struct device *dev;
Rusty Russelldde79782007-07-26 10:41:03 -0700107 /* ... And an end pointer so we can easily append new devices */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700108 struct device **lastdev;
109};
110
Rusty Russell17cbca22007-10-22 11:24:22 +1000111/* The list of Guest devices, based on command line arguments. */
112static struct device_list devices;
113
Rusty Russelldde79782007-07-26 10:41:03 -0700114/* The device structure describes a single device. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700115struct device
116{
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
120 /* The this 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
123 /* The name of this device, for --verbose. */
124 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700125
Rusty Russelldde79782007-07-26 10:41:03 -0700126 /* If handle_input is set, it wants to be called when this file
127 * descriptor is ready. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700128 int fd;
129 bool (*handle_input)(int fd, struct device *me);
130
Rusty Russell17cbca22007-10-22 11:24:22 +1000131 /* Any queues attached to this device */
132 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700133
134 /* Device-specific data. */
135 void *priv;
136};
137
Rusty Russell17cbca22007-10-22 11:24:22 +1000138/* The virtqueue structure describes a queue attached to a device. */
139struct virtqueue
140{
141 struct virtqueue *next;
142
143 /* Which device owns me. */
144 struct device *dev;
145
146 /* The configuration for this queue. */
147 struct lguest_vqconfig config;
148
149 /* The actual ring of buffers. */
150 struct vring vring;
151
152 /* Last available index we saw. */
153 u16 last_avail_idx;
154
155 /* The routine to call when the Guest pings us. */
156 void (*handle_output)(int fd, struct virtqueue *me);
157};
158
Balaji Raoec04b132007-12-28 14:26:24 +0530159/* Remember the arguments to the program so we can "reboot" */
160static char **main_args;
161
Rusty Russell17cbca22007-10-22 11:24:22 +1000162/* Since guest is UP and we don't run at the same time, we don't need barriers.
163 * But I include them in the code in case others copy it. */
164#define wmb()
165
166/* Convert an iovec element to the given type.
167 *
168 * This is a fairly ugly trick: we need to know the size of the type and
169 * alignment requirement to check the pointer is kosher. It's also nice to
170 * have the name of the type in case we report failure.
171 *
172 * Typing those three things all the time is cumbersome and error prone, so we
173 * have a macro which sets them all up and passes to the real function. */
174#define convert(iov, type) \
175 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
176
177static void *_convert(struct iovec *iov, size_t size, size_t align,
178 const char *name)
179{
180 if (iov->iov_len != size)
181 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
182 if ((unsigned long)iov->iov_base % align != 0)
183 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
184 return iov->iov_base;
185}
186
187/* The virtio configuration space is defined to be little-endian. x86 is
188 * little-endian too, but it's nice to be explicit so we have these helpers. */
189#define cpu_to_le16(v16) (v16)
190#define cpu_to_le32(v32) (v32)
191#define cpu_to_le64(v64) (v64)
192#define le16_to_cpu(v16) (v16)
193#define le32_to_cpu(v32) (v32)
194#define le64_to_cpu(v32) (v64)
195
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000196/*L:100 The Launcher code itself takes us out into userspace, that scary place
197 * where pointers run wild and free! Unfortunately, like most userspace
198 * programs, it's quite boring (which is why everyone likes to hack on the
199 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
200 * will get you through this section. Or, maybe not.
201 *
202 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
203 * memory and stores it in "guest_base". In other words, Guest physical ==
204 * Launcher virtual with an offset.
205 *
206 * This can be tough to get your head around, but usually it just means that we
207 * use these trivial conversion functions when the Guest gives us it's
208 * "physical" addresses: */
209static void *from_guest_phys(unsigned long addr)
210{
211 return guest_base + addr;
212}
213
214static unsigned long to_guest_phys(const void *addr)
215{
216 return (addr - guest_base);
217}
218
Rusty Russelldde79782007-07-26 10:41:03 -0700219/*L:130
220 * Loading the Kernel.
221 *
222 * We start with couple of simple helper routines. open_or_die() avoids
223 * error-checking code cluttering the callers: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700224static int open_or_die(const char *name, int flags)
225{
226 int fd = open(name, flags);
227 if (fd < 0)
228 err(1, "Failed to open %s", name);
229 return fd;
230}
231
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000232/* map_zeroed_pages() takes a number of pages. */
233static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700234{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000235 int fd = open_or_die("/dev/zero", O_RDONLY);
236 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700237
Rusty Russelldde79782007-07-26 10:41:03 -0700238 /* We use a private mapping (ie. if we write to the page, it will be
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000239 * copied). */
240 addr = mmap(NULL, getpagesize() * num,
241 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
242 if (addr == MAP_FAILED)
243 err(1, "Mmaping %u pages of /dev/zero", num);
Rusty Russelldde79782007-07-26 10:41:03 -0700244
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000245 return addr;
246}
247
248/* Get some more pages for a device. */
249static void *get_pages(unsigned int num)
250{
251 void *addr = from_guest_phys(guest_limit);
252
253 guest_limit += num * getpagesize();
254 if (guest_limit > guest_max)
255 errx(1, "Not enough memory for devices");
256 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700257}
258
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700259/* This routine is used to load the kernel or initrd. It tries mmap, but if
260 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
261 * it falls back to reading the memory in. */
262static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
263{
264 ssize_t r;
265
266 /* We map writable even though for some segments are marked read-only.
267 * The kernel really wants to be writable: it patches its own
268 * instructions.
269 *
270 * MAP_PRIVATE means that the page won't be copied until a write is
271 * done to it. This allows us to share untouched memory between
272 * Guests. */
273 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
274 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
275 return;
276
277 /* pread does a seek and a read in one shot: saves a few lines. */
278 r = pread(fd, addr, len, offset);
279 if (r != len)
280 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
281}
282
Rusty Russelldde79782007-07-26 10:41:03 -0700283/* This routine takes an open vmlinux image, which is in ELF, and maps it into
284 * the Guest memory. ELF = Embedded Linking Format, which is the format used
285 * by all modern binaries on Linux including the kernel.
286 *
287 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000288 * address. We use the physical address; the Guest will map itself to the
289 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700290 *
291 * We return the starting address. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000292static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700293{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700294 Elf32_Phdr phdr[ehdr->e_phnum];
295 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700296
Rusty Russelldde79782007-07-26 10:41:03 -0700297 /* Sanity checks on the main ELF header: an x86 executable with a
298 * reasonable number of correctly-sized program headers. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700299 if (ehdr->e_type != ET_EXEC
300 || ehdr->e_machine != EM_386
301 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
302 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
303 errx(1, "Malformed elf header");
304
Rusty Russelldde79782007-07-26 10:41:03 -0700305 /* An ELF executable contains an ELF header and a number of "program"
306 * headers which indicate which parts ("segments") of the program to
307 * load where. */
308
309 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700310 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
311 err(1, "Seeking to program headers");
312 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
313 err(1, "Reading program headers");
314
Rusty Russelldde79782007-07-26 10:41:03 -0700315 /* Try all the headers: there are usually only three. A read-only one,
316 * a read-write one, and a "note" section which isn't loadable. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700317 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700318 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700319 if (phdr[i].p_type != PT_LOAD)
320 continue;
321
322 verbose("Section %i: size %i addr %p\n",
323 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
324
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700325 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000326 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700327 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700328 }
329
Rusty Russell814a0e52007-10-22 11:29:44 +1000330 /* The entry point is given in the ELF header. */
331 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700332}
333
Rusty Russelldde79782007-07-26 10:41:03 -0700334/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000335 * supposed to jump into it and it will unpack itself. We used to have to
336 * perform some hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700337 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000338 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
339 * a small patch to jump over the tricky bits in the Guest, so now we just read
340 * the funky header so we know where in the file to load, and away we go! */
Rusty Russell47436aa2007-10-22 11:03:36 +1000341static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700342{
Rusty Russell43d33b22007-10-22 11:29:57 +1000343 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000344 int r;
345 /* Modern bzImages get loaded at 1M. */
346 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700347
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000348 /* Go back to the start of the file and read the header. It should be
349 * a Linux boot header (see Documentation/i386/boot.txt) */
350 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000351 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000352
Rusty Russell43d33b22007-10-22 11:29:57 +1000353 /* Inside the setup_hdr, we expect the magic "HdrS" */
354 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000355 errx(1, "This doesn't look like a bzImage to me");
356
Rusty Russell43d33b22007-10-22 11:29:57 +1000357 /* Skip over the extra sectors of the header. */
358 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000359
360 /* Now read everything into memory. in nice big chunks. */
361 while ((r = read(fd, p, 65536)) > 0)
362 p += r;
363
Rusty Russell43d33b22007-10-22 11:29:57 +1000364 /* Finally, code32_start tells us where to enter the kernel. */
365 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700366}
367
Rusty Russelldde79782007-07-26 10:41:03 -0700368/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000369 * come wrapped up in the self-decompressing "bzImage" format. With a little
370 * work, we can load those, too. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000371static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700372{
373 Elf32_Ehdr hdr;
374
Rusty Russelldde79782007-07-26 10:41:03 -0700375 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700376 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
377 err(1, "Reading kernel");
378
Rusty Russelldde79782007-07-26 10:41:03 -0700379 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700380 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000381 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700382
Rusty Russelldde79782007-07-26 10:41:03 -0700383 /* Otherwise we assume it's a bzImage, and try to unpack it */
Rusty Russell47436aa2007-10-22 11:03:36 +1000384 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700385}
386
Rusty Russelldde79782007-07-26 10:41:03 -0700387/* This is a trivial little helper to align pages. Andi Kleen hated it because
388 * it calls getpagesize() twice: "it's dumb code."
389 *
390 * Kernel guys get really het up about optimization, even when it's not
391 * necessary. I leave this code as a reaction against that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700392static inline unsigned long page_align(unsigned long addr)
393{
Rusty Russelldde79782007-07-26 10:41:03 -0700394 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700395 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
396}
397
Rusty Russelldde79782007-07-26 10:41:03 -0700398/*L:180 An "initial ram disk" is a disk image loaded into memory along with
399 * the kernel which the kernel can use to boot from without needing any
400 * drivers. Most distributions now use this as standard: the initrd contains
401 * the code to load the appropriate driver modules for the current machine.
402 *
403 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
404 * kernels. He sent me this (and tells me when I break it). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700405static unsigned long load_initrd(const char *name, unsigned long mem)
406{
407 int ifd;
408 struct stat st;
409 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700410
411 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700412 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700413 if (fstat(ifd, &st) < 0)
414 err(1, "fstat() on initrd '%s'", name);
415
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700416 /* We map the initrd at the top of memory, but mmap wants it to be
417 * page-aligned, so we round the size up for that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700418 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000419 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russelldde79782007-07-26 10:41:03 -0700420 /* Once a file is mapped, you can close the file descriptor. It's a
421 * little odd, but quite useful. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700422 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700423 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700424
425 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700426 return len;
427}
428
Rusty Russell47436aa2007-10-22 11:03:36 +1000429/* Once we know how much memory we have, we can construct simple linear page
430 * tables which set virtual == physical which will get the Guest far enough
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000431 * into the boot to create its own.
Rusty Russelldde79782007-07-26 10:41:03 -0700432 *
433 * We lay them out of the way, just below the initrd (which is why we need to
434 * know its size). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700435static unsigned long setup_pagetables(unsigned long mem,
Rusty Russell47436aa2007-10-22 11:03:36 +1000436 unsigned long initrd_size)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700437{
Jes Sorensen511801d2007-10-22 11:03:31 +1000438 unsigned long *pgdir, *linear;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700439 unsigned int mapped_pages, i, linear_pages;
Jes Sorensen511801d2007-10-22 11:03:31 +1000440 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700441
Rusty Russell47436aa2007-10-22 11:03:36 +1000442 mapped_pages = mem/getpagesize();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700443
Rusty Russelldde79782007-07-26 10:41:03 -0700444 /* Each PTE page can map ptes_per_page pages: how many do we need? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700445 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
446
Rusty Russelldde79782007-07-26 10:41:03 -0700447 /* We put the toplevel page directory page at the top of memory. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000448 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
Rusty Russelldde79782007-07-26 10:41:03 -0700449
450 /* Now we use the next linear_pages pages as pte pages */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700451 linear = (void *)pgdir - linear_pages*getpagesize();
452
Rusty Russelldde79782007-07-26 10:41:03 -0700453 /* Linear mapping is easy: put every page's address into the mapping in
454 * order. PAGE_PRESENT contains the flags Present, Writable and
455 * Executable. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700456 for (i = 0; i < mapped_pages; i++)
457 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
458
Rusty Russell47436aa2007-10-22 11:03:36 +1000459 /* The top level points to the linear page table pages above. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700460 for (i = 0; i < mapped_pages; i += ptes_per_page) {
Rusty Russell47436aa2007-10-22 11:03:36 +1000461 pgdir[i/ptes_per_page]
Jes Sorensen511801d2007-10-22 11:03:31 +1000462 = ((to_guest_phys(linear) + i*sizeof(void *))
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000463 | PAGE_PRESENT);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700464 }
465
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000466 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
467 mapped_pages, linear_pages, to_guest_phys(linear));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700468
Rusty Russelldde79782007-07-26 10:41:03 -0700469 /* We return the top level (guest-physical) address: the kernel needs
470 * to know where it is. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000471 return to_guest_phys(pgdir);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700472}
Rusty Russelle1e72962007-10-25 15:02:50 +1000473/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700474
Rusty Russelldde79782007-07-26 10:41:03 -0700475/* Simple routine to roll all the commandline arguments together with spaces
476 * between them. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700477static void concat(char *dst, char *args[])
478{
479 unsigned int i, len = 0;
480
481 for (i = 0; args[i]; i++) {
482 strcpy(dst+len, args[i]);
483 strcat(dst+len, " ");
484 len += strlen(args[i]) + 1;
485 }
486 /* In case it's empty. */
487 dst[len] = '\0';
488}
489
Rusty Russelle1e72962007-10-25 15:02:50 +1000490/*L:185 This is where we actually tell the kernel to initialize the Guest. We
491 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
492 * the base of Guest "physical" memory, the top physical page to allow, the
Rusty Russell47436aa2007-10-22 11:03:36 +1000493 * top level pagetable and the entry point for the Guest. */
494static int tell_kernel(unsigned long pgdir, unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700495{
Jes Sorensen511801d2007-10-22 11:03:31 +1000496 unsigned long args[] = { LHREQ_INITIALIZE,
497 (unsigned long)guest_base,
Rusty Russell47436aa2007-10-22 11:03:36 +1000498 guest_limit / getpagesize(), pgdir, start };
Rusty Russell8ca47e02007-07-19 01:49:29 -0700499 int fd;
500
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000501 verbose("Guest: %p - %p (%#lx)\n",
502 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700503 fd = open_or_die("/dev/lguest", O_RDWR);
504 if (write(fd, args, sizeof(args)) < 0)
505 err(1, "Writing to /dev/lguest");
Rusty Russelldde79782007-07-26 10:41:03 -0700506
507 /* We return the /dev/lguest file descriptor to control this Guest */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700508 return fd;
509}
Rusty Russelldde79782007-07-26 10:41:03 -0700510/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700511
Rusty Russell17cbca22007-10-22 11:24:22 +1000512static void add_device_fd(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700513{
Rusty Russell17cbca22007-10-22 11:24:22 +1000514 FD_SET(fd, &devices.infds);
515 if (fd > devices.max_infd)
516 devices.max_infd = fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700517}
518
Rusty Russelldde79782007-07-26 10:41:03 -0700519/*L:200
520 * The Waker.
521 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000522 * With console, block and network devices, we can have lots of input which we
523 * need to process. We could try to tell the kernel what file descriptors to
524 * watch, but handing a file descriptor mask through to the kernel is fairly
525 * icky.
Rusty Russelldde79782007-07-26 10:41:03 -0700526 *
527 * Instead, we fork off a process which watches the file descriptors and writes
Rusty Russelle1e72962007-10-25 15:02:50 +1000528 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
529 * stop running the Guest. This causes the Launcher to return from the
Rusty Russelldde79782007-07-26 10:41:03 -0700530 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
531 * the LHREQ_BREAK and wake us up again.
532 *
533 * This, of course, is merely a different *kind* of icky.
534 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000535static void wake_parent(int pipefd, int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700536{
Rusty Russelldde79782007-07-26 10:41:03 -0700537 /* Add the pipe from the Launcher to the fdset in the device_list, so
538 * we watch it, too. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000539 add_device_fd(pipefd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700540
541 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +1000542 fd_set rfds = devices.infds;
Jes Sorensen511801d2007-10-22 11:03:31 +1000543 unsigned long args[] = { LHREQ_BREAK, 1 };
Rusty Russell8ca47e02007-07-19 01:49:29 -0700544
Rusty Russelldde79782007-07-26 10:41:03 -0700545 /* Wait until input is ready from one of the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000546 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
Rusty Russelldde79782007-07-26 10:41:03 -0700547 /* Is it a message from the Launcher? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700548 if (FD_ISSET(pipefd, &rfds)) {
Rusty Russell56ae43d2007-10-22 11:24:23 +1000549 int fd;
Rusty Russelldde79782007-07-26 10:41:03 -0700550 /* If read() returns 0, it means the Launcher has
551 * exited. We silently follow. */
Rusty Russell56ae43d2007-10-22 11:24:23 +1000552 if (read(pipefd, &fd, sizeof(fd)) == 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700553 exit(0);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000554 /* Otherwise it's telling us to change what file
Rusty Russelle1e72962007-10-25 15:02:50 +1000555 * descriptors we're to listen to. Positive means
556 * listen to a new one, negative means stop
557 * listening. */
Rusty Russell56ae43d2007-10-22 11:24:23 +1000558 if (fd >= 0)
559 FD_SET(fd, &devices.infds);
560 else
561 FD_CLR(-fd - 1, &devices.infds);
Rusty Russelldde79782007-07-26 10:41:03 -0700562 } else /* Send LHREQ_BREAK command. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -0200563 pwrite(lguest_fd, args, sizeof(args), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700564 }
565}
566
Rusty Russelldde79782007-07-26 10:41:03 -0700567/* This routine just sets up a pipe to the Waker process. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000568static int setup_waker(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700569{
570 int pipefd[2], child;
571
Rusty Russelle1e72962007-10-25 15:02:50 +1000572 /* We create a pipe to talk to the Waker, and also so it knows when the
Rusty Russelldde79782007-07-26 10:41:03 -0700573 * Launcher dies (and closes pipe). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700574 pipe(pipefd);
575 child = fork();
576 if (child == -1)
577 err(1, "forking");
578
579 if (child == 0) {
Rusty Russelle1e72962007-10-25 15:02:50 +1000580 /* We are the Waker: close the "writing" end of our copy of the
581 * pipe and start waiting for input. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700582 close(pipefd[1]);
Rusty Russell17cbca22007-10-22 11:24:22 +1000583 wake_parent(pipefd[0], lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700584 }
Rusty Russelldde79782007-07-26 10:41:03 -0700585 /* Close the reading end of our copy of the pipe. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700586 close(pipefd[0]);
587
Rusty Russelldde79782007-07-26 10:41:03 -0700588 /* Here is the fd used to talk to the waker. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700589 return pipefd[1];
590}
591
Rusty Russelle1e72962007-10-25 15:02:50 +1000592/*
Rusty Russelldde79782007-07-26 10:41:03 -0700593 * Device Handling.
594 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000595 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700596 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000597 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700598 * if something funny is going on:
599 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700600static void *_check_pointer(unsigned long addr, unsigned int size,
601 unsigned int line)
602{
Rusty Russelldde79782007-07-26 10:41:03 -0700603 /* We have to separately check addr and addr+size, because size could
604 * be huge and addr + size might wrap around. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000605 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000606 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russelldde79782007-07-26 10:41:03 -0700607 /* We return a pointer for the caller's convenience, now we know it's
608 * safe to use. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000609 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700610}
Rusty Russelldde79782007-07-26 10:41:03 -0700611/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700612#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
613
Rusty Russelle1e72962007-10-25 15:02:50 +1000614/* Each buffer in the virtqueues is actually a chain of descriptors. This
615 * function returns the next descriptor in the chain, or vq->vring.num if we're
616 * at the end. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000617static unsigned next_desc(struct virtqueue *vq, unsigned int i)
618{
619 unsigned int next;
620
621 /* If this descriptor says it doesn't chain, we're done. */
622 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
623 return vq->vring.num;
624
625 /* Check they're not leading us off end of descriptors. */
626 next = vq->vring.desc[i].next;
627 /* Make sure compiler knows to grab that: we don't want it changing! */
628 wmb();
629
630 if (next >= vq->vring.num)
631 errx(1, "Desc next is %u", next);
632
633 return next;
634}
635
636/* This looks in the virtqueue and for the first available buffer, and converts
637 * it to an iovec for convenient access. Since descriptors consist of some
638 * number of output then some number of input descriptors, it's actually two
639 * iovecs, but we pack them into one and note how many of each there were.
640 *
641 * This function returns the descriptor number found, or vq->vring.num (which
642 * is never a valid descriptor number) if none was found. */
643static unsigned get_vq_desc(struct virtqueue *vq,
644 struct iovec iov[],
645 unsigned int *out_num, unsigned int *in_num)
646{
647 unsigned int i, head;
648
649 /* Check it isn't doing very strange things with descriptor numbers. */
650 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
651 errx(1, "Guest moved used index from %u to %u",
652 vq->last_avail_idx, vq->vring.avail->idx);
653
654 /* If there's nothing new since last we looked, return invalid. */
655 if (vq->vring.avail->idx == vq->last_avail_idx)
656 return vq->vring.num;
657
658 /* Grab the next descriptor number they're advertising, and increment
659 * the index we've seen. */
660 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
661
662 /* If their number is silly, that's a fatal mistake. */
663 if (head >= vq->vring.num)
664 errx(1, "Guest says index %u is available", head);
665
666 /* When we start there are none of either input nor output. */
667 *out_num = *in_num = 0;
668
669 i = head;
670 do {
671 /* Grab the first descriptor, and check it's OK. */
672 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
673 iov[*out_num + *in_num].iov_base
674 = check_pointer(vq->vring.desc[i].addr,
675 vq->vring.desc[i].len);
676 /* If this is an input descriptor, increment that count. */
677 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
678 (*in_num)++;
679 else {
680 /* If it's an output descriptor, they're all supposed
681 * to come before any input descriptors. */
682 if (*in_num)
683 errx(1, "Descriptor has out after in");
684 (*out_num)++;
685 }
686
687 /* If we've got too many, that implies a descriptor loop. */
688 if (*out_num + *in_num > vq->vring.num)
689 errx(1, "Looped descriptor");
690 } while ((i = next_desc(vq, i)) != vq->vring.num);
691
692 return head;
693}
694
Rusty Russelle1e72962007-10-25 15:02:50 +1000695/* After we've used one of their buffers, we tell them about it. We'll then
Rusty Russell17cbca22007-10-22 11:24:22 +1000696 * want to send them an interrupt, using trigger_irq(). */
697static void add_used(struct virtqueue *vq, unsigned int head, int len)
698{
699 struct vring_used_elem *used;
700
Rusty Russelle1e72962007-10-25 15:02:50 +1000701 /* The virtqueue contains a ring of used buffers. Get a pointer to the
702 * next entry in that used ring. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000703 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
704 used->id = head;
705 used->len = len;
706 /* Make sure buffer is written before we update index. */
707 wmb();
708 vq->vring.used->idx++;
709}
710
711/* This actually sends the interrupt for this virtqueue */
712static void trigger_irq(int fd, struct virtqueue *vq)
713{
714 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
715
Rusty Russelle1e72962007-10-25 15:02:50 +1000716 /* If they don't want an interrupt, don't send one. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000717 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
718 return;
719
720 /* Send the Guest an interrupt tell them we used something up. */
721 if (write(fd, buf, sizeof(buf)) != 0)
722 err(1, "Triggering irq %i", vq->config.irq);
723}
724
725/* And here's the combo meal deal. Supersize me! */
726static void add_used_and_trigger(int fd, struct virtqueue *vq,
727 unsigned int head, int len)
728{
729 add_used(vq, head, len);
730 trigger_irq(fd, vq);
731}
732
Rusty Russelle1e72962007-10-25 15:02:50 +1000733/*
734 * The Console
735 *
736 * Here is the input terminal setting we save, and the routine to restore them
737 * on exit so the user gets their terminal back. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700738static struct termios orig_term;
739static void restore_term(void)
740{
741 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
742}
743
Rusty Russelldde79782007-07-26 10:41:03 -0700744/* We associate some data with the console for our exit hack. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700745struct console_abort
746{
Rusty Russelldde79782007-07-26 10:41:03 -0700747 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700748 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700749 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700750 struct timeval start;
751};
752
Rusty Russelldde79782007-07-26 10:41:03 -0700753/* This is the routine which handles console input (ie. stdin). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700754static bool handle_console_input(int fd, struct device *dev)
755{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700756 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000757 unsigned int head, in_num, out_num;
758 struct iovec iov[dev->vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700759 struct console_abort *abort = dev->priv;
760
Rusty Russell17cbca22007-10-22 11:24:22 +1000761 /* First we need a console buffer from the Guests's input virtqueue. */
762 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000763
764 /* If they're not ready for input, stop listening to this file
765 * descriptor. We'll start again once they add an input buffer. */
766 if (head == dev->vq->vring.num)
767 return false;
768
769 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000770 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700771
Rusty Russelldde79782007-07-26 10:41:03 -0700772 /* This is why we convert to iovecs: the readv() call uses them, and so
773 * it reads straight into the Guest's buffer. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000774 len = readv(dev->fd, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700775 if (len <= 0) {
Rusty Russelldde79782007-07-26 10:41:03 -0700776 /* This implies that the console is closed, is /dev/null, or
Rusty Russell17cbca22007-10-22 11:24:22 +1000777 * something went terribly wrong. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700778 warnx("Failed to get console input, ignoring console.");
Rusty Russell56ae43d2007-10-22 11:24:23 +1000779 /* Put the input terminal back. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000780 restore_term();
Rusty Russell56ae43d2007-10-22 11:24:23 +1000781 /* Remove callback from input vq, so it doesn't restart us. */
782 dev->vq->handle_output = NULL;
783 /* Stop listening to this fd: don't call us again. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000784 return false;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700785 }
786
Rusty Russell56ae43d2007-10-22 11:24:23 +1000787 /* Tell the Guest about the new input. */
788 add_used_and_trigger(fd, dev->vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700789
Rusty Russelldde79782007-07-26 10:41:03 -0700790 /* Three ^C within one second? Exit.
791 *
792 * This is such a hack, but works surprisingly well. Each ^C has to be
793 * in a buffer by itself, so they can't be too fast. But we check that
794 * we get three within about a second, so they can't be too slow. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700795 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
796 if (!abort->count++)
797 gettimeofday(&abort->start, NULL);
798 else if (abort->count == 3) {
799 struct timeval now;
800 gettimeofday(&now, NULL);
801 if (now.tv_sec <= abort->start.tv_sec+1) {
Jes Sorensen511801d2007-10-22 11:03:31 +1000802 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russelldde79782007-07-26 10:41:03 -0700803 /* Close the fd so Waker will know it has to
804 * exit. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700805 close(waker_fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700806 /* Just in case waker is blocked in BREAK, send
807 * unbreak now. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700808 write(fd, args, sizeof(args));
809 exit(2);
810 }
811 abort->count = 0;
812 }
813 } else
Rusty Russelldde79782007-07-26 10:41:03 -0700814 /* Any other key resets the abort counter. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700815 abort->count = 0;
816
Rusty Russelldde79782007-07-26 10:41:03 -0700817 /* Everything went OK! */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700818 return true;
819}
820
Rusty Russell17cbca22007-10-22 11:24:22 +1000821/* Handling output for console is simple: we just get all the output buffers
822 * and write them to stdout. */
823static void handle_console_output(int fd, struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700824{
Rusty Russell17cbca22007-10-22 11:24:22 +1000825 unsigned int head, out, in;
826 int len;
827 struct iovec iov[vq->vring.num];
828
829 /* Keep getting output buffers from the Guest until we run out. */
830 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
831 if (in)
832 errx(1, "Input buffers in output queue?");
833 len = writev(STDOUT_FILENO, iov, out);
834 add_used_and_trigger(fd, vq, head, len);
835 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700836}
837
Rusty Russelle1e72962007-10-25 15:02:50 +1000838/*
839 * The Network
840 *
841 * Handling output for network is also simple: we get all the output buffers
Rusty Russell17cbca22007-10-22 11:24:22 +1000842 * and write them (ignoring the first element) to this device's file descriptor
843 * (stdout). */
844static void handle_net_output(int fd, struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700845{
Rusty Russell17cbca22007-10-22 11:24:22 +1000846 unsigned int head, out, in;
847 int len;
848 struct iovec iov[vq->vring.num];
849
850 /* Keep getting output buffers from the Guest until we run out. */
851 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
852 if (in)
853 errx(1, "Input buffers in output queue?");
Rusty Russelle1e72962007-10-25 15:02:50 +1000854 /* Check header, but otherwise ignore it (we told the Guest we
855 * supported no features, so it shouldn't have anything
856 * interesting). */
Rusty Russell17cbca22007-10-22 11:24:22 +1000857 (void)convert(&iov[0], struct virtio_net_hdr);
858 len = writev(vq->dev->fd, iov+1, out-1);
859 add_used_and_trigger(fd, vq, head, len);
860 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700861}
862
Rusty Russell17cbca22007-10-22 11:24:22 +1000863/* This is where we handle a packet coming in from the tun device to our
864 * Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700865static bool handle_tun_input(int fd, struct device *dev)
866{
Rusty Russell17cbca22007-10-22 11:24:22 +1000867 unsigned int head, in_num, out_num;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700868 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000869 struct iovec iov[dev->vq->vring.num];
870 struct virtio_net_hdr *hdr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700871
Rusty Russell17cbca22007-10-22 11:24:22 +1000872 /* First we need a network buffer from the Guests's recv virtqueue. */
873 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
874 if (head == dev->vq->vring.num) {
Rusty Russelldde79782007-07-26 10:41:03 -0700875 /* Now, it's expected that if we try to send a packet too
Rusty Russell17cbca22007-10-22 11:24:22 +1000876 * early, the Guest won't be ready yet. Wait until the device
877 * status says it's ready. */
878 /* FIXME: Actually want DRIVER_ACTIVE here. */
879 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700880 warn("network: no dma buffer!");
Rusty Russell56ae43d2007-10-22 11:24:23 +1000881 /* We'll turn this back on if input buffers are registered. */
882 return false;
Rusty Russell17cbca22007-10-22 11:24:22 +1000883 } else if (out_num)
884 errx(1, "Output buffers in network recv queue?");
885
886 /* First element is the header: we set it to 0 (no features). */
887 hdr = convert(&iov[0], struct virtio_net_hdr);
888 hdr->flags = 0;
889 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700890
Rusty Russelldde79782007-07-26 10:41:03 -0700891 /* Read the packet from the device directly into the Guest's buffer. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000892 len = readv(dev->fd, iov+1, in_num-1);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700893 if (len <= 0)
894 err(1, "reading network");
Rusty Russelldde79782007-07-26 10:41:03 -0700895
Rusty Russell56ae43d2007-10-22 11:24:23 +1000896 /* Tell the Guest about the new packet. */
897 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000898
Rusty Russell8ca47e02007-07-19 01:49:29 -0700899 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
Rusty Russell17cbca22007-10-22 11:24:22 +1000900 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
901 head != dev->vq->vring.num ? "sent" : "discarded");
902
Rusty Russelldde79782007-07-26 10:41:03 -0700903 /* All good. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700904 return true;
905}
906
Rusty Russelle1e72962007-10-25 15:02:50 +1000907/*L:215 This is the callback attached to the network and console input
908 * virtqueues: it ensures we try again, in case we stopped console or net
Rusty Russell56ae43d2007-10-22 11:24:23 +1000909 * delivery because Guest didn't have any buffers. */
910static void enable_fd(int fd, struct virtqueue *vq)
911{
912 add_device_fd(vq->dev->fd);
913 /* Tell waker to listen to it again */
914 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
915}
916
Rusty Russell17cbca22007-10-22 11:24:22 +1000917/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
918static void handle_output(int fd, unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700919{
920 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +1000921 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700922
Rusty Russell17cbca22007-10-22 11:24:22 +1000923 /* Check each virtqueue. */
924 for (i = devices.dev; i; i = i->next) {
925 for (vq = i->vq; vq; vq = vq->next) {
926 if (vq->config.pfn == addr/getpagesize()
927 && vq->handle_output) {
928 verbose("Output to %s\n", vq->dev->name);
929 vq->handle_output(fd, vq);
930 return;
931 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700932 }
933 }
Rusty Russelldde79782007-07-26 10:41:03 -0700934
Rusty Russell17cbca22007-10-22 11:24:22 +1000935 /* Early console write is done using notify on a nul-terminated string
936 * in Guest memory. */
937 if (addr >= guest_limit)
938 errx(1, "Bad NOTIFY %#lx", addr);
939
940 write(STDOUT_FILENO, from_guest_phys(addr),
941 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700942}
943
Rusty Russelle1e72962007-10-25 15:02:50 +1000944/* This is called when the Waker wakes us up: check for incoming file
Rusty Russelldde79782007-07-26 10:41:03 -0700945 * descriptors. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000946static void handle_input(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700947{
Rusty Russelldde79782007-07-26 10:41:03 -0700948 /* select() wants a zeroed timeval to mean "don't wait". */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700949 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
950
951 for (;;) {
952 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +1000953 fd_set fds = devices.infds;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700954
Rusty Russelldde79782007-07-26 10:41:03 -0700955 /* If nothing is ready, we're done. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000956 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700957 break;
958
Rusty Russelldde79782007-07-26 10:41:03 -0700959 /* Otherwise, call the device(s) which have readable
960 * file descriptors and a method of handling them. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000961 for (i = devices.dev; i; i = i->next) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700962 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
Rusty Russell56ae43d2007-10-22 11:24:23 +1000963 int dev_fd;
964 if (i->handle_input(fd, i))
965 continue;
966
Rusty Russelldde79782007-07-26 10:41:03 -0700967 /* If handle_input() returns false, it means we
Rusty Russell56ae43d2007-10-22 11:24:23 +1000968 * should no longer service it. Networking and
969 * console do this when there's no input
970 * buffers to deliver into. Console also uses
971 * it when it discovers that stdin is
972 * closed. */
973 FD_CLR(i->fd, &devices.infds);
974 /* Tell waker to ignore it too, by sending a
975 * negative fd number (-1, since 0 is a valid
976 * FD number). */
977 dev_fd = -i->fd - 1;
978 write(waker_fd, &dev_fd, sizeof(dev_fd));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700979 }
980 }
981 }
982}
983
Rusty Russelldde79782007-07-26 10:41:03 -0700984/*L:190
985 * Device Setup
986 *
987 * All devices need a descriptor so the Guest knows it exists, and a "struct
988 * device" so the Launcher can keep track of it. We have common helper
989 * routines to allocate them.
990 *
991 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russell17cbca22007-10-22 11:24:22 +1000992 * table just above the Guest's normal memory. It returns a pointer to that
993 * descriptor. */
994static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700995{
Rusty Russell17cbca22007-10-22 11:24:22 +1000996 struct lguest_device_desc *d;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700997
Rusty Russell17cbca22007-10-22 11:24:22 +1000998 /* We only have one page for all the descriptors. */
999 if (devices.desc_used + sizeof(*d) > getpagesize())
1000 errx(1, "Too many devices");
1001
1002 /* We don't need to set config_len or status: page is 0 already. */
1003 d = (void *)devices.descpage + devices.desc_used;
1004 d->type = type;
1005 devices.desc_used += sizeof(*d);
1006
1007 return d;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001008}
1009
Rusty Russell17cbca22007-10-22 11:24:22 +10001010/* Each device descriptor is followed by some configuration information.
Rusty Russelle1e72962007-10-25 15:02:50 +10001011 * Each configuration field looks like: u8 type, u8 len, [... len bytes...].
Rusty Russell17cbca22007-10-22 11:24:22 +10001012 *
1013 * This routine adds a new field to an existing device's descriptor. It only
1014 * works for the last device, but that's OK because that's how we use it. */
1015static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
1016{
1017 /* This is the last descriptor, right? */
1018 assert(devices.descpage + devices.desc_used
1019 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
1020
1021 /* We only have one page of device descriptions. */
1022 if (devices.desc_used + 2 + len > getpagesize())
1023 errx(1, "Too many devices");
1024
1025 /* Copy in the new config header: type then length. */
1026 devices.descpage[devices.desc_used++] = type;
1027 devices.descpage[devices.desc_used++] = len;
1028 memcpy(devices.descpage + devices.desc_used, c, len);
1029 devices.desc_used += len;
1030
1031 /* Update the device descriptor length: two byte head then data. */
1032 dev->desc->config_len += 2 + len;
1033}
1034
1035/* This routine adds a virtqueue to a device. We specify how many descriptors
1036 * the virtqueue is to have. */
1037static void add_virtqueue(struct device *dev, unsigned int num_descs,
1038 void (*handle_output)(int fd, struct virtqueue *me))
1039{
1040 unsigned int pages;
1041 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1042 void *p;
1043
1044 /* First we need some pages for this virtqueue. */
Rusty Russell42b36cc2007-11-12 13:39:18 +11001045 pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1)
1046 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001047 p = get_pages(pages);
1048
Rusty Russelld1c856e2007-11-19 11:20:40 -05001049 /* Initialize the virtqueue */
1050 vq->next = NULL;
1051 vq->last_avail_idx = 0;
1052 vq->dev = dev;
1053
Rusty Russell17cbca22007-10-22 11:24:22 +10001054 /* Initialize the configuration. */
1055 vq->config.num = num_descs;
1056 vq->config.irq = devices.next_irq++;
1057 vq->config.pfn = to_guest_phys(p) / getpagesize();
1058
1059 /* Initialize the vring. */
Rusty Russell42b36cc2007-11-12 13:39:18 +11001060 vring_init(&vq->vring, num_descs, p, getpagesize());
Rusty Russell17cbca22007-10-22 11:24:22 +10001061
1062 /* Add the configuration information to this device's descriptor. */
1063 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE,
1064 sizeof(vq->config), &vq->config);
1065
1066 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1067 * second. */
1068 for (i = &dev->vq; *i; i = &(*i)->next);
1069 *i = vq;
1070
Rusty Russelle1e72962007-10-25 15:02:50 +10001071 /* Set the routine to call when the Guest does something to this
1072 * virtqueue. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001073 vq->handle_output = handle_output;
Rusty Russelle1e72962007-10-25 15:02:50 +10001074
1075 /* Set the "Don't Notify Me" flag if we don't have a handler */
Rusty Russell17cbca22007-10-22 11:24:22 +10001076 if (!handle_output)
1077 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1078}
1079
1080/* This routine does all the creation and setup of a new device, including
Rusty Russelle1e72962007-10-25 15:02:50 +10001081 * calling new_dev_desc() to allocate the descriptor and device memory. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001082static struct device *new_device(const char *name, u16 type, int fd,
1083 bool (*handle_input)(int, struct device *))
Rusty Russell8ca47e02007-07-19 01:49:29 -07001084{
1085 struct device *dev = malloc(sizeof(*dev));
1086
Rusty Russelldde79782007-07-26 10:41:03 -07001087 /* Append to device list. Prepending to a single-linked list is
1088 * easier, but the user expects the devices to be arranged on the bus
1089 * in command-line order. The first network device on the command line
Rusty Russelle1e72962007-10-25 15:02:50 +10001090 * is eth0, the first block device /dev/vda, etc. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001091 *devices.lastdev = dev;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001092 dev->next = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001093 devices.lastdev = &dev->next;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001094
Rusty Russelldde79782007-07-26 10:41:03 -07001095 /* Now we populate the fields one at a time. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001096 dev->fd = fd;
Rusty Russelldde79782007-07-26 10:41:03 -07001097 /* If we have an input handler for this file descriptor, then we add it
1098 * to the device_list's fdset and maxfd. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001099 if (handle_input)
Rusty Russell17cbca22007-10-22 11:24:22 +10001100 add_device_fd(dev->fd);
1101 dev->desc = new_dev_desc(type);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001102 dev->handle_input = handle_input;
Rusty Russell17cbca22007-10-22 11:24:22 +10001103 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001104 dev->vq = NULL;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001105 return dev;
1106}
1107
Rusty Russelldde79782007-07-26 10:41:03 -07001108/* Our first setup routine is the console. It's a fairly simple device, but
1109 * UNIX tty handling makes it uglier than it could be. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001110static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001111{
1112 struct device *dev;
1113
Rusty Russelldde79782007-07-26 10:41:03 -07001114 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001115 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1116 struct termios term = orig_term;
Rusty Russelldde79782007-07-26 10:41:03 -07001117 /* Then we turn off echo, line buffering and ^C etc. We want a
1118 * raw input stream to the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001119 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1120 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russelldde79782007-07-26 10:41:03 -07001121 /* If we exit gracefully, the original settings will be
1122 * restored so the user can see what they're typing. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001123 atexit(restore_term);
1124 }
1125
Rusty Russell17cbca22007-10-22 11:24:22 +10001126 dev = new_device("console", VIRTIO_ID_CONSOLE,
1127 STDIN_FILENO, handle_console_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001128 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001129 dev->priv = malloc(sizeof(struct console_abort));
1130 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001131
Rusty Russell56ae43d2007-10-22 11:24:23 +10001132 /* The console needs two virtqueues: the input then the output. When
1133 * they put something the input queue, we make sure we're listening to
1134 * stdin. When they put something in the output queue, we write it to
Rusty Russelle1e72962007-10-25 15:02:50 +10001135 * stdout. */
Rusty Russell56ae43d2007-10-22 11:24:23 +10001136 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001137 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001138
Rusty Russell17cbca22007-10-22 11:24:22 +10001139 verbose("device %u: console\n", devices.device_num++);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001140}
Rusty Russelldde79782007-07-26 10:41:03 -07001141/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001142
Rusty Russell17cbca22007-10-22 11:24:22 +10001143/*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1144 * --sharenet=<name> option which opens or creates a named pipe. This can be
1145 * used to send packets to another guest in a 1:1 manner.
1146 *
1147 * More sopisticated is to use one of the tools developed for project like UML
1148 * to do networking.
1149 *
1150 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1151 * completely generic ("here's my vring, attach to your vring") and would work
1152 * for any traffic. Of course, namespace and permissions issues need to be
1153 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1154 * multiple inter-guest channels behind one interface, although it would
1155 * require some manner of hotplugging new virtio channels.
1156 *
1157 * Finally, we could implement a virtio network switch in the kernel. :*/
1158
Rusty Russell8ca47e02007-07-19 01:49:29 -07001159static u32 str2ip(const char *ipaddr)
1160{
1161 unsigned int byte[4];
1162
1163 sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]);
1164 return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3];
1165}
1166
Rusty Russelldde79782007-07-26 10:41:03 -07001167/* This code is "adapted" from libbridge: it attaches the Host end of the
1168 * network device to the bridge device specified by the command line.
1169 *
1170 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1171 * dislike bridging), and I just try not to break it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001172static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1173{
1174 int ifidx;
1175 struct ifreq ifr;
1176
1177 if (!*br_name)
1178 errx(1, "must specify bridge name");
1179
1180 ifidx = if_nametoindex(if_name);
1181 if (!ifidx)
1182 errx(1, "interface %s does not exist!", if_name);
1183
1184 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1185 ifr.ifr_ifindex = ifidx;
1186 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1187 err(1, "can't add %s to bridge %s", if_name, br_name);
1188}
1189
Rusty Russelldde79782007-07-26 10:41:03 -07001190/* This sets up the Host end of the network device with an IP address, brings
1191 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell17cbca22007-10-22 11:24:22 +10001192 * pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001193static void configure_device(int fd, const char *devname, u32 ipaddr,
1194 unsigned char hwaddr[6])
1195{
1196 struct ifreq ifr;
1197 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1198
Rusty Russelldde79782007-07-26 10:41:03 -07001199 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001200 memset(&ifr, 0, sizeof(ifr));
1201 strcpy(ifr.ifr_name, devname);
1202 sin->sin_family = AF_INET;
1203 sin->sin_addr.s_addr = htonl(ipaddr);
1204 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1205 err(1, "Setting %s interface address", devname);
1206 ifr.ifr_flags = IFF_UP;
1207 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1208 err(1, "Bringing interface %s up", devname);
1209
Rusty Russelldde79782007-07-26 10:41:03 -07001210 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1211 * above). IF means Interface, and HWADDR is hardware address.
1212 * Simple! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001213 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1214 err(1, "getting hw address for %s", devname);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001215 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1216}
1217
Rusty Russell17cbca22007-10-22 11:24:22 +10001218/*L:195 Our network is a Host<->Guest network. This can either use bridging or
1219 * routing, but the principle is the same: it uses the "tun" device to inject
1220 * packets into the Host as if they came in from a normal network card. We
1221 * just shunt packets between the Guest and the tun device. */
1222static void setup_tun_net(const char *arg)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001223{
1224 struct device *dev;
1225 struct ifreq ifr;
1226 int netfd, ipfd;
1227 u32 ip;
1228 const char *br_name = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001229 u8 hwaddr[6];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001230
Rusty Russelldde79782007-07-26 10:41:03 -07001231 /* We open the /dev/net/tun device and tell it we want a tap device. A
1232 * tap device is like a tun device, only somehow different. To tell
1233 * the truth, I completely blundered my way through this code, but it
1234 * works now! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001235 netfd = open_or_die("/dev/net/tun", O_RDWR);
1236 memset(&ifr, 0, sizeof(ifr));
1237 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1238 strcpy(ifr.ifr_name, "tap%d");
1239 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1240 err(1, "configuring /dev/net/tun");
Rusty Russelldde79782007-07-26 10:41:03 -07001241 /* We don't need checksums calculated for packets coming in this
1242 * device: trust us! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001243 ioctl(netfd, TUNSETNOCSUM, 1);
1244
Rusty Russell17cbca22007-10-22 11:24:22 +10001245 /* First we create a new network device. */
1246 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001247
Rusty Russell56ae43d2007-10-22 11:24:23 +10001248 /* Network devices need a receive and a send queue, just like
1249 * console. */
1250 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001251 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001252
Rusty Russelldde79782007-07-26 10:41:03 -07001253 /* We need a socket to perform the magic network ioctls to bring up the
1254 * tap interface, connect to the bridge etc. Any socket will do! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001255 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1256 if (ipfd < 0)
1257 err(1, "opening IP socket");
1258
Rusty Russelldde79782007-07-26 10:41:03 -07001259 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001260 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1261 ip = INADDR_ANY;
1262 br_name = arg + strlen(BRIDGE_PFX);
1263 add_to_bridge(ipfd, ifr.ifr_name, br_name);
Rusty Russelldde79782007-07-26 10:41:03 -07001264 } else /* It is an IP address to set up the device with */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001265 ip = str2ip(arg);
1266
Rusty Russell17cbca22007-10-22 11:24:22 +10001267 /* Set up the tun device, and get the mac address for the interface. */
1268 configure_device(ipfd, ifr.ifr_name, ip, hwaddr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001269
Rusty Russell17cbca22007-10-22 11:24:22 +10001270 /* Tell Guest what MAC address to use. */
1271 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001272
Rusty Russell17cbca22007-10-22 11:24:22 +10001273 /* We don't seed the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001274 close(ipfd);
1275
Rusty Russell17cbca22007-10-22 11:24:22 +10001276 verbose("device %u: tun net %u.%u.%u.%u\n",
1277 devices.device_num++,
1278 (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001279 if (br_name)
1280 verbose("attached to bridge: %s\n", br_name);
1281}
Rusty Russell17cbca22007-10-22 11:24:22 +10001282
Rusty Russelle1e72962007-10-25 15:02:50 +10001283/* Our block (disk) device should be really simple: the Guest asks for a block
1284 * number and we read or write that position in the file. Unfortunately, that
1285 * was amazingly slow: the Guest waits until the read is finished before
1286 * running anything else, even if it could have been doing useful work.
Rusty Russell17cbca22007-10-22 11:24:22 +10001287 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001288 * We could use async I/O, except it's reputed to suck so hard that characters
1289 * actually go missing from your code when you try to use it.
Rusty Russell17cbca22007-10-22 11:24:22 +10001290 *
1291 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1292
Rusty Russelle1e72962007-10-25 15:02:50 +10001293/* This hangs off device->priv. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001294struct vblk_info
1295{
1296 /* The size of the file. */
1297 off64_t len;
1298
1299 /* The file descriptor for the file. */
1300 int fd;
1301
1302 /* IO thread listens on this file descriptor [0]. */
1303 int workpipe[2];
1304
1305 /* IO thread writes to this file descriptor to mark it done, then
1306 * Launcher triggers interrupt to Guest. */
1307 int done_fd;
1308};
Rusty Russelle1e72962007-10-25 15:02:50 +10001309/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001310
Rusty Russelle1e72962007-10-25 15:02:50 +10001311/*L:210
1312 * The Disk
1313 *
1314 * Remember that the block device is handled by a separate I/O thread. We head
1315 * straight into the core of that thread here:
1316 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001317static bool service_io(struct device *dev)
1318{
1319 struct vblk_info *vblk = dev->priv;
1320 unsigned int head, out_num, in_num, wlen;
1321 int ret;
1322 struct virtio_blk_inhdr *in;
1323 struct virtio_blk_outhdr *out;
1324 struct iovec iov[dev->vq->vring.num];
1325 off64_t off;
1326
Rusty Russelle1e72962007-10-25 15:02:50 +10001327 /* See if there's a request waiting. If not, nothing to do. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001328 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1329 if (head == dev->vq->vring.num)
1330 return false;
1331
Rusty Russelle1e72962007-10-25 15:02:50 +10001332 /* Every block request should contain at least one output buffer
1333 * (detailing the location on disk and the type of request) and one
1334 * input buffer (to hold the result). */
Rusty Russell17cbca22007-10-22 11:24:22 +10001335 if (out_num == 0 || in_num == 0)
1336 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1337 head, out_num, in_num);
1338
1339 out = convert(&iov[0], struct virtio_blk_outhdr);
1340 in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr);
1341 off = out->sector * 512;
1342
Rusty Russelle1e72962007-10-25 15:02:50 +10001343 /* The block device implements "barriers", where the Guest indicates
1344 * that it wants all previous writes to occur before this write. We
1345 * don't have a way of asking our kernel to do a barrier, so we just
1346 * synchronize all the data in the file. Pretty poor, no? */
Rusty Russell17cbca22007-10-22 11:24:22 +10001347 if (out->type & VIRTIO_BLK_T_BARRIER)
1348 fdatasync(vblk->fd);
1349
Rusty Russelle1e72962007-10-25 15:02:50 +10001350 /* In general the virtio block driver is allowed to try SCSI commands.
1351 * It'd be nice if we supported eject, for example, but we don't. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001352 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1353 fprintf(stderr, "Scsi commands unsupported\n");
1354 in->status = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001355 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001356 } else if (out->type & VIRTIO_BLK_T_OUT) {
1357 /* Write */
1358
1359 /* Move to the right location in the block file. This can fail
1360 * if they try to write past end. */
1361 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1362 err(1, "Bad seek to sector %llu", out->sector);
1363
1364 ret = writev(vblk->fd, iov+1, out_num-1);
1365 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1366
1367 /* Grr... Now we know how long the descriptor they sent was, we
1368 * make sure they didn't try to write over the end of the block
1369 * file (possibly extending it). */
1370 if (ret > 0 && off + ret > vblk->len) {
1371 /* Trim it back to the correct length */
1372 ftruncate64(vblk->fd, vblk->len);
1373 /* Die, bad Guest, die. */
1374 errx(1, "Write past end %llu+%u", off, ret);
1375 }
Anthony Liguori1200e642007-11-08 21:13:44 -06001376 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001377 in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1378 } else {
1379 /* Read */
1380
1381 /* Move to the right location in the block file. This can fail
1382 * if they try to read past end. */
1383 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1384 err(1, "Bad seek to sector %llu", out->sector);
1385
1386 ret = readv(vblk->fd, iov+1, in_num-1);
1387 verbose("READ from sector %llu: %i\n", out->sector, ret);
1388 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001389 wlen = sizeof(*in) + ret;
Rusty Russell17cbca22007-10-22 11:24:22 +10001390 in->status = VIRTIO_BLK_S_OK;
1391 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001392 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001393 in->status = VIRTIO_BLK_S_IOERR;
1394 }
1395 }
1396
1397 /* We can't trigger an IRQ, because we're not the Launcher. It does
1398 * that when we tell it we're done. */
1399 add_used(dev->vq, head, wlen);
1400 return true;
1401}
1402
1403/* This is the thread which actually services the I/O. */
1404static int io_thread(void *_dev)
1405{
1406 struct device *dev = _dev;
1407 struct vblk_info *vblk = dev->priv;
1408 char c;
1409
1410 /* Close other side of workpipe so we get 0 read when main dies. */
1411 close(vblk->workpipe[1]);
1412 /* Close the other side of the done_fd pipe. */
1413 close(dev->fd);
1414
1415 /* When this read fails, it means Launcher died, so we follow. */
1416 while (read(vblk->workpipe[0], &c, 1) == 1) {
Rusty Russelle1e72962007-10-25 15:02:50 +10001417 /* We acknowledge each request immediately to reduce latency,
Rusty Russell17cbca22007-10-22 11:24:22 +10001418 * rather than waiting until we've done them all. I haven't
1419 * measured to see if it makes any difference. */
1420 while (service_io(dev))
1421 write(vblk->done_fd, &c, 1);
1422 }
1423 return 0;
1424}
1425
Rusty Russelle1e72962007-10-25 15:02:50 +10001426/* Now we've seen the I/O thread, we return to the Launcher to see what happens
1427 * when the thread tells us it's completed some I/O. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001428static bool handle_io_finish(int fd, struct device *dev)
1429{
1430 char c;
1431
Rusty Russelle1e72962007-10-25 15:02:50 +10001432 /* If the I/O thread died, presumably it printed the error, so we
1433 * simply exit. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001434 if (read(dev->fd, &c, 1) != 1)
1435 exit(1);
1436
1437 /* It did some work, so trigger the irq. */
1438 trigger_irq(fd, dev->vq);
1439 return true;
1440}
1441
Rusty Russelle1e72962007-10-25 15:02:50 +10001442/* When the Guest submits some I/O, we just need to wake the I/O thread. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001443static void handle_virtblk_output(int fd, struct virtqueue *vq)
1444{
1445 struct vblk_info *vblk = vq->dev->priv;
1446 char c = 0;
1447
1448 /* Wake up I/O thread and tell it to go to work! */
1449 if (write(vblk->workpipe[1], &c, 1) != 1)
1450 /* Presumably it indicated why it died. */
1451 exit(1);
1452}
1453
Rusty Russelle1e72962007-10-25 15:02:50 +10001454/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001455static void setup_block_file(const char *filename)
1456{
1457 int p[2];
1458 struct device *dev;
1459 struct vblk_info *vblk;
1460 void *stack;
1461 u64 cap;
1462 unsigned int val;
1463
1464 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1465 pipe(p);
1466
1467 /* The device responds to return from I/O thread. */
1468 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1469
Rusty Russelle1e72962007-10-25 15:02:50 +10001470 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001471 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1472
1473 /* Allocate the room for our own bookkeeping */
1474 vblk = dev->priv = malloc(sizeof(*vblk));
1475
1476 /* First we open the file and store the length. */
1477 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1478 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1479
1480 /* Tell Guest how many sectors this device has. */
1481 cap = cpu_to_le64(vblk->len / 512);
1482 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1483
1484 /* Tell Guest not to put in too many descriptors at once: two are used
1485 * for the in and out elements. */
1486 val = cpu_to_le32(VIRTQUEUE_NUM - 2);
1487 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val);
1488
1489 /* The I/O thread writes to this end of the pipe when done. */
1490 vblk->done_fd = p[1];
1491
Rusty Russelle1e72962007-10-25 15:02:50 +10001492 /* This is the second pipe, which is how we tell the I/O thread about
1493 * more work. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001494 pipe(vblk->workpipe);
1495
1496 /* Create stack for thread and run it */
1497 stack = malloc(32768);
Balaji Raoec04b132007-12-28 14:26:24 +05301498 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1499 * becoming a zombie. */
1500 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
Rusty Russell17cbca22007-10-22 11:24:22 +10001501 err(1, "Creating clone");
1502
1503 /* We don't need to keep the I/O thread's end of the pipes open. */
1504 close(vblk->done_fd);
1505 close(vblk->workpipe[0]);
1506
1507 verbose("device %u: virtblock %llu sectors\n",
1508 devices.device_num, cap);
1509}
Balaji Raoec04b132007-12-28 14:26:24 +05301510/* That's the end of device setup. :*/
1511
1512/* Reboot */
1513static void __attribute__((noreturn)) restart_guest(void)
1514{
1515 unsigned int i;
1516
1517 /* Closing pipes causes the waker thread and io_threads to die, and
1518 * closing /dev/lguest cleans up the Guest. Since we don't track all
1519 * open fds, we simply close everything beyond stderr. */
1520 for (i = 3; i < FD_SETSIZE; i++)
1521 close(i);
1522 execv(main_args[0], main_args);
1523 err(1, "Could not exec %s", main_args[0]);
1524}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001525
Rusty Russelldde79782007-07-26 10:41:03 -07001526/*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1527 * its input and output, and finally, lays it to rest. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001528static void __attribute__((noreturn)) run_guest(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001529{
1530 for (;;) {
Jes Sorensen511801d2007-10-22 11:03:31 +10001531 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russell17cbca22007-10-22 11:24:22 +10001532 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001533 int readval;
1534
1535 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001536 readval = pread(lguest_fd, &notify_addr,
1537 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001538
Rusty Russell17cbca22007-10-22 11:24:22 +10001539 /* One unsigned long means the Guest did HCALL_NOTIFY */
1540 if (readval == sizeof(notify_addr)) {
1541 verbose("Notify on address %#lx\n", notify_addr);
1542 handle_output(lguest_fd, notify_addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001543 continue;
Rusty Russelldde79782007-07-26 10:41:03 -07001544 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001545 } else if (errno == ENOENT) {
1546 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001547 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001548 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301549 /* ERESTART means that we need to reboot the guest */
1550 } else if (errno == ERESTART) {
1551 restart_guest();
Rusty Russelle1e72962007-10-25 15:02:50 +10001552 /* EAGAIN means the Waker wanted us to look at some input.
Rusty Russelldde79782007-07-26 10:41:03 -07001553 * Anything else means a bug or incompatible change. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001554 } else if (errno != EAGAIN)
1555 err(1, "Running guest failed");
Rusty Russelldde79782007-07-26 10:41:03 -07001556
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001557 /* Only service input on thread for CPU 0. */
1558 if (cpu_id != 0)
1559 continue;
1560
Rusty Russelle1e72962007-10-25 15:02:50 +10001561 /* Service input, then unset the BREAK to release the Waker. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001562 handle_input(lguest_fd);
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001563 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001564 err(1, "Resetting break");
1565 }
1566}
Rusty Russelldde79782007-07-26 10:41:03 -07001567/*
Rusty Russelle1e72962007-10-25 15:02:50 +10001568 * This is the end of the Launcher. The good news: we are over halfway
1569 * through! The bad news: the most fiendish part of the code still lies ahead
1570 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001571 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001572 * Are you ready? Take a deep breath and join me in the core of the Host, in
1573 * "make Host".
1574 :*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001575
1576static struct option opts[] = {
1577 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001578 { "tunnet", 1, NULL, 't' },
1579 { "block", 1, NULL, 'b' },
1580 { "initrd", 1, NULL, 'i' },
1581 { NULL },
1582};
1583static void usage(void)
1584{
1585 errx(1, "Usage: lguest [--verbose] "
Rusty Russell17cbca22007-10-22 11:24:22 +10001586 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001587 "|--block=<filename>|--initrd=<filename>]...\n"
1588 "<mem-in-mb> vmlinux [args...]");
1589}
1590
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001591/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001592int main(int argc, char *argv[])
1593{
Rusty Russell47436aa2007-10-22 11:03:36 +10001594 /* Memory, top-level pagetable, code startpoint and size of the
1595 * (optional) initrd. */
1596 unsigned long mem = 0, pgdir, start, initrd_size = 0;
Rusty Russelle1e72962007-10-25 15:02:50 +10001597 /* Two temporaries and the /dev/lguest file descriptor. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001598 int i, c, lguest_fd;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001599 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001600 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001601 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001602 const char *initrd_name = NULL;
1603
Balaji Raoec04b132007-12-28 14:26:24 +05301604 /* Save the args: we "reboot" by execing ourselves again. */
1605 main_args = argv;
1606 /* We don't "wait" for the children, so prevent them from becoming
1607 * zombies. */
1608 signal(SIGCHLD, SIG_IGN);
1609
Rusty Russelldde79782007-07-26 10:41:03 -07001610 /* First we initialize the device list. Since console and network
1611 * device receive input from a file descriptor, we keep an fdset
1612 * (infds) and the maximum fd number (max_infd) with the head of the
1613 * list. We also keep a pointer to the last device, for easy appending
Rusty Russell17cbca22007-10-22 11:24:22 +10001614 * to the list. Finally, we keep the next interrupt number to hand out
1615 * (1: remember that 0 is used by the timer). */
1616 FD_ZERO(&devices.infds);
1617 devices.max_infd = -1;
1618 devices.lastdev = &devices.dev;
1619 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001620
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001621 cpu_id = 0;
Rusty Russelldde79782007-07-26 10:41:03 -07001622 /* We need to know how much memory so we can set up the device
1623 * descriptor and memory pages for the devices as we parse the command
1624 * line. So we quickly look through the arguments to find the amount
1625 * of memory now. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001626 for (i = 1; i < argc; i++) {
1627 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001628 mem = atoi(argv[i]) * 1024 * 1024;
1629 /* We start by mapping anonymous pages over all of
1630 * guest-physical memory range. This fills it with 0,
1631 * and ensures that the Guest won't be killed when it
1632 * tries to access it. */
1633 guest_base = map_zeroed_pages(mem / getpagesize()
1634 + DEVICE_PAGES);
1635 guest_limit = mem;
1636 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001637 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001638 break;
1639 }
1640 }
Rusty Russelldde79782007-07-26 10:41:03 -07001641
1642 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001643 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1644 switch (c) {
1645 case 'v':
1646 verbose = true;
1647 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001648 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001649 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001650 break;
1651 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001652 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001653 break;
1654 case 'i':
1655 initrd_name = optarg;
1656 break;
1657 default:
1658 warnx("Unknown argument %s", argv[optind]);
1659 usage();
1660 }
1661 }
Rusty Russelldde79782007-07-26 10:41:03 -07001662 /* After the other arguments we expect memory and kernel image name,
1663 * followed by command line arguments for the kernel. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001664 if (optind + 2 > argc)
1665 usage();
1666
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001667 verbose("Guest base is at %p\n", guest_base);
1668
Rusty Russelldde79782007-07-26 10:41:03 -07001669 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001670 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001671
Rusty Russell8ca47e02007-07-19 01:49:29 -07001672 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001673 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001674
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001675 /* Boot information is stashed at physical address 0 */
1676 boot = from_guest_phys(0);
1677
Rusty Russelldde79782007-07-26 10:41:03 -07001678 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001679 if (initrd_name) {
1680 initrd_size = load_initrd(initrd_name, mem);
Rusty Russelldde79782007-07-26 10:41:03 -07001681 /* These are the location in the Linux boot header where the
1682 * start and size of the initrd are expected to be found. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001683 boot->hdr.ramdisk_image = mem - initrd_size;
1684 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001685 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001686 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001687 }
1688
Rusty Russelldde79782007-07-26 10:41:03 -07001689 /* Set up the initial linear pagetables, starting below the initrd. */
Rusty Russell47436aa2007-10-22 11:03:36 +10001690 pgdir = setup_pagetables(mem, initrd_size);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001691
Rusty Russelldde79782007-07-26 10:41:03 -07001692 /* The Linux boot header contains an "E820" memory map: ours is a
1693 * simple, single region. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001694 boot->e820_entries = 1;
1695 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russelldde79782007-07-26 10:41:03 -07001696 /* The boot header contains a command line pointer: we put the command
Rusty Russell43d33b22007-10-22 11:29:57 +10001697 * line after the boot header. */
1698 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10001699 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001700 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07001701
Rusty Russell814a0e52007-10-22 11:29:44 +10001702 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001703 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10001704
1705 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001706 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10001707
Rusty Russell43d33b22007-10-22 11:29:57 +10001708 /* Tell the entry path not to try to reload segment registers. */
1709 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001710
Rusty Russelldde79782007-07-26 10:41:03 -07001711 /* We tell the kernel to initialize the Guest: this returns the open
1712 * /dev/lguest file descriptor. */
Rusty Russell47436aa2007-10-22 11:03:36 +10001713 lguest_fd = tell_kernel(pgdir, start);
Rusty Russelldde79782007-07-26 10:41:03 -07001714
1715 /* We fork off a child process, which wakes the Launcher whenever one
1716 * of the input file descriptors needs attention. Otherwise we would
1717 * run the Guest until it tries to output something. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001718 waker_fd = setup_waker(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001719
Rusty Russelldde79782007-07-26 10:41:03 -07001720 /* Finally, run the Guest. This doesn't return. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001721 run_guest(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001722}
Rusty Russellf56a3842007-07-26 10:41:05 -07001723/*:*/
1724
1725/*M:999
1726 * Mastery is done: you now know everything I do.
1727 *
1728 * But surely you have seen code, features and bugs in your wanderings which
1729 * you now yearn to attack? That is the real game, and I look forward to you
1730 * patching and forking lguest into the Your-Name-Here-visor.
1731 *
1732 * Farewell, and good coding!
1733 * Rusty Russell.
1734 */