blob: c91c28ae82904c0eb0304df3bd472c9cde5cb63c [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 Russell17cbca22007-10-22 11:24:22 +100065/* This fits nicely in a single 4096-byte page. */
66#define VIRTQUEUE_NUM 127
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
Rusty Russelldde79782007-07-26 10:41:03 -070082/* This is our list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070083struct device_list
84{
Rusty Russelldde79782007-07-26 10:41:03 -070085 /* Summary information about the devices in our list: ready to pass to
86 * select() to ask which need servicing.*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070087 fd_set infds;
88 int max_infd;
89
Rusty Russell17cbca22007-10-22 11:24:22 +100090 /* Counter to assign interrupt numbers. */
91 unsigned int next_irq;
92
93 /* Counter to print out convenient device numbers. */
94 unsigned int device_num;
95
Rusty Russelldde79782007-07-26 10:41:03 -070096 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +100097 u8 *descpage;
98
99 /* The tail of the last descriptor. */
100 unsigned int desc_used;
Rusty Russelldde79782007-07-26 10:41:03 -0700101
102 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700103 struct device *dev;
Rusty Russelldde79782007-07-26 10:41:03 -0700104 /* ... And an end pointer so we can easily append new devices */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700105 struct device **lastdev;
106};
107
Rusty Russell17cbca22007-10-22 11:24:22 +1000108/* The list of Guest devices, based on command line arguments. */
109static struct device_list devices;
110
Rusty Russelldde79782007-07-26 10:41:03 -0700111/* The device structure describes a single device. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700112struct device
113{
Rusty Russelldde79782007-07-26 10:41:03 -0700114 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700115 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000116
117 /* The this device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700118 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000119
120 /* The name of this device, for --verbose. */
121 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700122
Rusty Russelldde79782007-07-26 10:41:03 -0700123 /* If handle_input is set, it wants to be called when this file
124 * descriptor is ready. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700125 int fd;
126 bool (*handle_input)(int fd, struct device *me);
127
Rusty Russell17cbca22007-10-22 11:24:22 +1000128 /* Any queues attached to this device */
129 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700130
131 /* Device-specific data. */
132 void *priv;
133};
134
Rusty Russell17cbca22007-10-22 11:24:22 +1000135/* The virtqueue structure describes a queue attached to a device. */
136struct virtqueue
137{
138 struct virtqueue *next;
139
140 /* Which device owns me. */
141 struct device *dev;
142
143 /* The configuration for this queue. */
144 struct lguest_vqconfig config;
145
146 /* The actual ring of buffers. */
147 struct vring vring;
148
149 /* Last available index we saw. */
150 u16 last_avail_idx;
151
152 /* The routine to call when the Guest pings us. */
153 void (*handle_output)(int fd, struct virtqueue *me);
154};
155
156/* Since guest is UP and we don't run at the same time, we don't need barriers.
157 * But I include them in the code in case others copy it. */
158#define wmb()
159
160/* Convert an iovec element to the given type.
161 *
162 * This is a fairly ugly trick: we need to know the size of the type and
163 * alignment requirement to check the pointer is kosher. It's also nice to
164 * have the name of the type in case we report failure.
165 *
166 * Typing those three things all the time is cumbersome and error prone, so we
167 * have a macro which sets them all up and passes to the real function. */
168#define convert(iov, type) \
169 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
170
171static void *_convert(struct iovec *iov, size_t size, size_t align,
172 const char *name)
173{
174 if (iov->iov_len != size)
175 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
176 if ((unsigned long)iov->iov_base % align != 0)
177 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
178 return iov->iov_base;
179}
180
181/* The virtio configuration space is defined to be little-endian. x86 is
182 * little-endian too, but it's nice to be explicit so we have these helpers. */
183#define cpu_to_le16(v16) (v16)
184#define cpu_to_le32(v32) (v32)
185#define cpu_to_le64(v64) (v64)
186#define le16_to_cpu(v16) (v16)
187#define le32_to_cpu(v32) (v32)
188#define le64_to_cpu(v32) (v64)
189
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000190/*L:100 The Launcher code itself takes us out into userspace, that scary place
191 * where pointers run wild and free! Unfortunately, like most userspace
192 * programs, it's quite boring (which is why everyone likes to hack on the
193 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
194 * will get you through this section. Or, maybe not.
195 *
196 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
197 * memory and stores it in "guest_base". In other words, Guest physical ==
198 * Launcher virtual with an offset.
199 *
200 * This can be tough to get your head around, but usually it just means that we
201 * use these trivial conversion functions when the Guest gives us it's
202 * "physical" addresses: */
203static void *from_guest_phys(unsigned long addr)
204{
205 return guest_base + addr;
206}
207
208static unsigned long to_guest_phys(const void *addr)
209{
210 return (addr - guest_base);
211}
212
Rusty Russelldde79782007-07-26 10:41:03 -0700213/*L:130
214 * Loading the Kernel.
215 *
216 * We start with couple of simple helper routines. open_or_die() avoids
217 * error-checking code cluttering the callers: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700218static int open_or_die(const char *name, int flags)
219{
220 int fd = open(name, flags);
221 if (fd < 0)
222 err(1, "Failed to open %s", name);
223 return fd;
224}
225
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000226/* map_zeroed_pages() takes a number of pages. */
227static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700228{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000229 int fd = open_or_die("/dev/zero", O_RDONLY);
230 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700231
Rusty Russelldde79782007-07-26 10:41:03 -0700232 /* We use a private mapping (ie. if we write to the page, it will be
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000233 * copied). */
234 addr = mmap(NULL, getpagesize() * num,
235 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
236 if (addr == MAP_FAILED)
237 err(1, "Mmaping %u pages of /dev/zero", num);
Rusty Russelldde79782007-07-26 10:41:03 -0700238
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000239 return addr;
240}
241
242/* Get some more pages for a device. */
243static void *get_pages(unsigned int num)
244{
245 void *addr = from_guest_phys(guest_limit);
246
247 guest_limit += num * getpagesize();
248 if (guest_limit > guest_max)
249 errx(1, "Not enough memory for devices");
250 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700251}
252
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700253/* This routine is used to load the kernel or initrd. It tries mmap, but if
254 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
255 * it falls back to reading the memory in. */
256static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
257{
258 ssize_t r;
259
260 /* We map writable even though for some segments are marked read-only.
261 * The kernel really wants to be writable: it patches its own
262 * instructions.
263 *
264 * MAP_PRIVATE means that the page won't be copied until a write is
265 * done to it. This allows us to share untouched memory between
266 * Guests. */
267 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
268 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
269 return;
270
271 /* pread does a seek and a read in one shot: saves a few lines. */
272 r = pread(fd, addr, len, offset);
273 if (r != len)
274 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
275}
276
Rusty Russelldde79782007-07-26 10:41:03 -0700277/* This routine takes an open vmlinux image, which is in ELF, and maps it into
278 * the Guest memory. ELF = Embedded Linking Format, which is the format used
279 * by all modern binaries on Linux including the kernel.
280 *
281 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000282 * address. We use the physical address; the Guest will map itself to the
283 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700284 *
285 * We return the starting address. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000286static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700287{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700288 Elf32_Phdr phdr[ehdr->e_phnum];
289 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700290
Rusty Russelldde79782007-07-26 10:41:03 -0700291 /* Sanity checks on the main ELF header: an x86 executable with a
292 * reasonable number of correctly-sized program headers. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700293 if (ehdr->e_type != ET_EXEC
294 || ehdr->e_machine != EM_386
295 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
296 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
297 errx(1, "Malformed elf header");
298
Rusty Russelldde79782007-07-26 10:41:03 -0700299 /* An ELF executable contains an ELF header and a number of "program"
300 * headers which indicate which parts ("segments") of the program to
301 * load where. */
302
303 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700304 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
305 err(1, "Seeking to program headers");
306 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
307 err(1, "Reading program headers");
308
Rusty Russelldde79782007-07-26 10:41:03 -0700309 /* Try all the headers: there are usually only three. A read-only one,
310 * a read-write one, and a "note" section which isn't loadable. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700311 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700312 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700313 if (phdr[i].p_type != PT_LOAD)
314 continue;
315
316 verbose("Section %i: size %i addr %p\n",
317 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
318
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700319 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000320 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700321 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700322 }
323
Rusty Russell814a0e52007-10-22 11:29:44 +1000324 /* The entry point is given in the ELF header. */
325 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700326}
327
Rusty Russelldde79782007-07-26 10:41:03 -0700328/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000329 * supposed to jump into it and it will unpack itself. We used to have to
330 * perform some hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700331 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000332 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
333 * a small patch to jump over the tricky bits in the Guest, so now we just read
334 * the funky header so we know where in the file to load, and away we go! */
Rusty Russell47436aa2007-10-22 11:03:36 +1000335static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700336{
Rusty Russell43d33b22007-10-22 11:29:57 +1000337 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000338 int r;
339 /* Modern bzImages get loaded at 1M. */
340 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700341
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000342 /* Go back to the start of the file and read the header. It should be
343 * a Linux boot header (see Documentation/i386/boot.txt) */
344 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000345 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000346
Rusty Russell43d33b22007-10-22 11:29:57 +1000347 /* Inside the setup_hdr, we expect the magic "HdrS" */
348 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000349 errx(1, "This doesn't look like a bzImage to me");
350
Rusty Russell43d33b22007-10-22 11:29:57 +1000351 /* Skip over the extra sectors of the header. */
352 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000353
354 /* Now read everything into memory. in nice big chunks. */
355 while ((r = read(fd, p, 65536)) > 0)
356 p += r;
357
Rusty Russell43d33b22007-10-22 11:29:57 +1000358 /* Finally, code32_start tells us where to enter the kernel. */
359 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700360}
361
Rusty Russelldde79782007-07-26 10:41:03 -0700362/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
363 * come wrapped up in the self-decompressing "bzImage" format. With some funky
364 * coding, we can load those, too. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000365static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700366{
367 Elf32_Ehdr hdr;
368
Rusty Russelldde79782007-07-26 10:41:03 -0700369 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700370 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
371 err(1, "Reading kernel");
372
Rusty Russelldde79782007-07-26 10:41:03 -0700373 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700374 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000375 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700376
Rusty Russelldde79782007-07-26 10:41:03 -0700377 /* Otherwise we assume it's a bzImage, and try to unpack it */
Rusty Russell47436aa2007-10-22 11:03:36 +1000378 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700379}
380
Rusty Russelldde79782007-07-26 10:41:03 -0700381/* This is a trivial little helper to align pages. Andi Kleen hated it because
382 * it calls getpagesize() twice: "it's dumb code."
383 *
384 * Kernel guys get really het up about optimization, even when it's not
385 * necessary. I leave this code as a reaction against that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700386static inline unsigned long page_align(unsigned long addr)
387{
Rusty Russelldde79782007-07-26 10:41:03 -0700388 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700389 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
390}
391
Rusty Russelldde79782007-07-26 10:41:03 -0700392/*L:180 An "initial ram disk" is a disk image loaded into memory along with
393 * the kernel which the kernel can use to boot from without needing any
394 * drivers. Most distributions now use this as standard: the initrd contains
395 * the code to load the appropriate driver modules for the current machine.
396 *
397 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
398 * kernels. He sent me this (and tells me when I break it). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700399static unsigned long load_initrd(const char *name, unsigned long mem)
400{
401 int ifd;
402 struct stat st;
403 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700404
405 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700406 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700407 if (fstat(ifd, &st) < 0)
408 err(1, "fstat() on initrd '%s'", name);
409
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700410 /* We map the initrd at the top of memory, but mmap wants it to be
411 * page-aligned, so we round the size up for that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700412 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000413 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russelldde79782007-07-26 10:41:03 -0700414 /* Once a file is mapped, you can close the file descriptor. It's a
415 * little odd, but quite useful. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700416 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700417 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700418
419 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700420 return len;
421}
422
Rusty Russell47436aa2007-10-22 11:03:36 +1000423/* Once we know how much memory we have, we can construct simple linear page
424 * tables which set virtual == physical which will get the Guest far enough
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000425 * into the boot to create its own.
Rusty Russelldde79782007-07-26 10:41:03 -0700426 *
427 * We lay them out of the way, just below the initrd (which is why we need to
428 * know its size). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700429static unsigned long setup_pagetables(unsigned long mem,
Rusty Russell47436aa2007-10-22 11:03:36 +1000430 unsigned long initrd_size)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700431{
Jes Sorensen511801d2007-10-22 11:03:31 +1000432 unsigned long *pgdir, *linear;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700433 unsigned int mapped_pages, i, linear_pages;
Jes Sorensen511801d2007-10-22 11:03:31 +1000434 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700435
Rusty Russell47436aa2007-10-22 11:03:36 +1000436 mapped_pages = mem/getpagesize();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700437
Rusty Russelldde79782007-07-26 10:41:03 -0700438 /* Each PTE page can map ptes_per_page pages: how many do we need? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700439 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
440
Rusty Russelldde79782007-07-26 10:41:03 -0700441 /* We put the toplevel page directory page at the top of memory. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000442 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
Rusty Russelldde79782007-07-26 10:41:03 -0700443
444 /* Now we use the next linear_pages pages as pte pages */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700445 linear = (void *)pgdir - linear_pages*getpagesize();
446
Rusty Russelldde79782007-07-26 10:41:03 -0700447 /* Linear mapping is easy: put every page's address into the mapping in
448 * order. PAGE_PRESENT contains the flags Present, Writable and
449 * Executable. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700450 for (i = 0; i < mapped_pages; i++)
451 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
452
Rusty Russell47436aa2007-10-22 11:03:36 +1000453 /* The top level points to the linear page table pages above. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700454 for (i = 0; i < mapped_pages; i += ptes_per_page) {
Rusty Russell47436aa2007-10-22 11:03:36 +1000455 pgdir[i/ptes_per_page]
Jes Sorensen511801d2007-10-22 11:03:31 +1000456 = ((to_guest_phys(linear) + i*sizeof(void *))
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000457 | PAGE_PRESENT);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700458 }
459
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000460 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
461 mapped_pages, linear_pages, to_guest_phys(linear));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700462
Rusty Russelldde79782007-07-26 10:41:03 -0700463 /* We return the top level (guest-physical) address: the kernel needs
464 * to know where it is. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000465 return to_guest_phys(pgdir);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700466}
467
Rusty Russelldde79782007-07-26 10:41:03 -0700468/* Simple routine to roll all the commandline arguments together with spaces
469 * between them. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700470static void concat(char *dst, char *args[])
471{
472 unsigned int i, len = 0;
473
474 for (i = 0; args[i]; i++) {
475 strcpy(dst+len, args[i]);
476 strcat(dst+len, " ");
477 len += strlen(args[i]) + 1;
478 }
479 /* In case it's empty. */
480 dst[len] = '\0';
481}
482
Rusty Russelldde79782007-07-26 10:41:03 -0700483/* This is where we actually tell the kernel to initialize the Guest. We saw
484 * the arguments it expects when we looked at initialize() in lguest_user.c:
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000485 * the base of guest "physical" memory, the top physical page to allow, the
Rusty Russell47436aa2007-10-22 11:03:36 +1000486 * top level pagetable and the entry point for the Guest. */
487static int tell_kernel(unsigned long pgdir, unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700488{
Jes Sorensen511801d2007-10-22 11:03:31 +1000489 unsigned long args[] = { LHREQ_INITIALIZE,
490 (unsigned long)guest_base,
Rusty Russell47436aa2007-10-22 11:03:36 +1000491 guest_limit / getpagesize(), pgdir, start };
Rusty Russell8ca47e02007-07-19 01:49:29 -0700492 int fd;
493
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000494 verbose("Guest: %p - %p (%#lx)\n",
495 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700496 fd = open_or_die("/dev/lguest", O_RDWR);
497 if (write(fd, args, sizeof(args)) < 0)
498 err(1, "Writing to /dev/lguest");
Rusty Russelldde79782007-07-26 10:41:03 -0700499
500 /* We return the /dev/lguest file descriptor to control this Guest */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700501 return fd;
502}
Rusty Russelldde79782007-07-26 10:41:03 -0700503/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700504
Rusty Russell17cbca22007-10-22 11:24:22 +1000505static void add_device_fd(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700506{
Rusty Russell17cbca22007-10-22 11:24:22 +1000507 FD_SET(fd, &devices.infds);
508 if (fd > devices.max_infd)
509 devices.max_infd = fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700510}
511
Rusty Russelldde79782007-07-26 10:41:03 -0700512/*L:200
513 * The Waker.
514 *
515 * With a console and network devices, we can have lots of input which we need
516 * to process. We could try to tell the kernel what file descriptors to watch,
517 * but handing a file descriptor mask through to the kernel is fairly icky.
518 *
519 * Instead, we fork off a process which watches the file descriptors and writes
520 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
521 * loop to stop running the Guest. This causes it to return from the
522 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
523 * the LHREQ_BREAK and wake us up again.
524 *
525 * This, of course, is merely a different *kind* of icky.
526 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000527static void wake_parent(int pipefd, int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700528{
Rusty Russelldde79782007-07-26 10:41:03 -0700529 /* Add the pipe from the Launcher to the fdset in the device_list, so
530 * we watch it, too. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000531 add_device_fd(pipefd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700532
533 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +1000534 fd_set rfds = devices.infds;
Jes Sorensen511801d2007-10-22 11:03:31 +1000535 unsigned long args[] = { LHREQ_BREAK, 1 };
Rusty Russell8ca47e02007-07-19 01:49:29 -0700536
Rusty Russelldde79782007-07-26 10:41:03 -0700537 /* Wait until input is ready from one of the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000538 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
Rusty Russelldde79782007-07-26 10:41:03 -0700539 /* Is it a message from the Launcher? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700540 if (FD_ISSET(pipefd, &rfds)) {
Rusty Russell56ae43d2007-10-22 11:24:23 +1000541 int fd;
Rusty Russelldde79782007-07-26 10:41:03 -0700542 /* If read() returns 0, it means the Launcher has
543 * exited. We silently follow. */
Rusty Russell56ae43d2007-10-22 11:24:23 +1000544 if (read(pipefd, &fd, sizeof(fd)) == 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700545 exit(0);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000546 /* Otherwise it's telling us to change what file
547 * descriptors we're to listen to. */
548 if (fd >= 0)
549 FD_SET(fd, &devices.infds);
550 else
551 FD_CLR(-fd - 1, &devices.infds);
Rusty Russelldde79782007-07-26 10:41:03 -0700552 } else /* Send LHREQ_BREAK command. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700553 write(lguest_fd, args, sizeof(args));
554 }
555}
556
Rusty Russelldde79782007-07-26 10:41:03 -0700557/* This routine just sets up a pipe to the Waker process. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000558static int setup_waker(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700559{
560 int pipefd[2], child;
561
Rusty Russelldde79782007-07-26 10:41:03 -0700562 /* We create a pipe to talk to the waker, and also so it knows when the
563 * Launcher dies (and closes pipe). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700564 pipe(pipefd);
565 child = fork();
566 if (child == -1)
567 err(1, "forking");
568
569 if (child == 0) {
Rusty Russelldde79782007-07-26 10:41:03 -0700570 /* Close the "writing" end of our copy of the pipe */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700571 close(pipefd[1]);
Rusty Russell17cbca22007-10-22 11:24:22 +1000572 wake_parent(pipefd[0], lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700573 }
Rusty Russelldde79782007-07-26 10:41:03 -0700574 /* Close the reading end of our copy of the pipe. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700575 close(pipefd[0]);
576
Rusty Russelldde79782007-07-26 10:41:03 -0700577 /* Here is the fd used to talk to the waker. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700578 return pipefd[1];
579}
580
Rusty Russelldde79782007-07-26 10:41:03 -0700581/*L:210
582 * Device Handling.
583 *
584 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
585 * We need to make sure it's not trying to reach into the Launcher itself, so
586 * we have a convenient routine which check it and exits with an error message
587 * if something funny is going on:
588 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700589static void *_check_pointer(unsigned long addr, unsigned int size,
590 unsigned int line)
591{
Rusty Russelldde79782007-07-26 10:41:03 -0700592 /* We have to separately check addr and addr+size, because size could
593 * be huge and addr + size might wrap around. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000594 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000595 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russelldde79782007-07-26 10:41:03 -0700596 /* We return a pointer for the caller's convenience, now we know it's
597 * safe to use. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000598 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700599}
Rusty Russelldde79782007-07-26 10:41:03 -0700600/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700601#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
602
Rusty Russell17cbca22007-10-22 11:24:22 +1000603/* This function returns the next descriptor in the chain, or vq->vring.num. */
604static unsigned next_desc(struct virtqueue *vq, unsigned int i)
605{
606 unsigned int next;
607
608 /* If this descriptor says it doesn't chain, we're done. */
609 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
610 return vq->vring.num;
611
612 /* Check they're not leading us off end of descriptors. */
613 next = vq->vring.desc[i].next;
614 /* Make sure compiler knows to grab that: we don't want it changing! */
615 wmb();
616
617 if (next >= vq->vring.num)
618 errx(1, "Desc next is %u", next);
619
620 return next;
621}
622
623/* This looks in the virtqueue and for the first available buffer, and converts
624 * it to an iovec for convenient access. Since descriptors consist of some
625 * number of output then some number of input descriptors, it's actually two
626 * iovecs, but we pack them into one and note how many of each there were.
627 *
628 * This function returns the descriptor number found, or vq->vring.num (which
629 * is never a valid descriptor number) if none was found. */
630static unsigned get_vq_desc(struct virtqueue *vq,
631 struct iovec iov[],
632 unsigned int *out_num, unsigned int *in_num)
633{
634 unsigned int i, head;
635
636 /* Check it isn't doing very strange things with descriptor numbers. */
637 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
638 errx(1, "Guest moved used index from %u to %u",
639 vq->last_avail_idx, vq->vring.avail->idx);
640
641 /* If there's nothing new since last we looked, return invalid. */
642 if (vq->vring.avail->idx == vq->last_avail_idx)
643 return vq->vring.num;
644
645 /* Grab the next descriptor number they're advertising, and increment
646 * the index we've seen. */
647 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
648
649 /* If their number is silly, that's a fatal mistake. */
650 if (head >= vq->vring.num)
651 errx(1, "Guest says index %u is available", head);
652
653 /* When we start there are none of either input nor output. */
654 *out_num = *in_num = 0;
655
656 i = head;
657 do {
658 /* Grab the first descriptor, and check it's OK. */
659 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
660 iov[*out_num + *in_num].iov_base
661 = check_pointer(vq->vring.desc[i].addr,
662 vq->vring.desc[i].len);
663 /* If this is an input descriptor, increment that count. */
664 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
665 (*in_num)++;
666 else {
667 /* If it's an output descriptor, they're all supposed
668 * to come before any input descriptors. */
669 if (*in_num)
670 errx(1, "Descriptor has out after in");
671 (*out_num)++;
672 }
673
674 /* If we've got too many, that implies a descriptor loop. */
675 if (*out_num + *in_num > vq->vring.num)
676 errx(1, "Looped descriptor");
677 } while ((i = next_desc(vq, i)) != vq->vring.num);
678
679 return head;
680}
681
682/* Once we've used one of their buffers, we tell them about it. We'll then
683 * want to send them an interrupt, using trigger_irq(). */
684static void add_used(struct virtqueue *vq, unsigned int head, int len)
685{
686 struct vring_used_elem *used;
687
688 /* Get a pointer to the next entry in the used ring. */
689 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
690 used->id = head;
691 used->len = len;
692 /* Make sure buffer is written before we update index. */
693 wmb();
694 vq->vring.used->idx++;
695}
696
697/* This actually sends the interrupt for this virtqueue */
698static void trigger_irq(int fd, struct virtqueue *vq)
699{
700 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
701
702 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
703 return;
704
705 /* Send the Guest an interrupt tell them we used something up. */
706 if (write(fd, buf, sizeof(buf)) != 0)
707 err(1, "Triggering irq %i", vq->config.irq);
708}
709
710/* And here's the combo meal deal. Supersize me! */
711static void add_used_and_trigger(int fd, struct virtqueue *vq,
712 unsigned int head, int len)
713{
714 add_used(vq, head, len);
715 trigger_irq(fd, vq);
716}
717
Rusty Russelldde79782007-07-26 10:41:03 -0700718/* Here is the input terminal setting we save, and the routine to restore them
719 * on exit so the user can see what they type next. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700720static struct termios orig_term;
721static void restore_term(void)
722{
723 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
724}
725
Rusty Russelldde79782007-07-26 10:41:03 -0700726/* We associate some data with the console for our exit hack. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700727struct console_abort
728{
Rusty Russelldde79782007-07-26 10:41:03 -0700729 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700730 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700731 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700732 struct timeval start;
733};
734
Rusty Russelldde79782007-07-26 10:41:03 -0700735/* This is the routine which handles console input (ie. stdin). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700736static bool handle_console_input(int fd, struct device *dev)
737{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700738 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000739 unsigned int head, in_num, out_num;
740 struct iovec iov[dev->vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700741 struct console_abort *abort = dev->priv;
742
Rusty Russell17cbca22007-10-22 11:24:22 +1000743 /* First we need a console buffer from the Guests's input virtqueue. */
744 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000745
746 /* If they're not ready for input, stop listening to this file
747 * descriptor. We'll start again once they add an input buffer. */
748 if (head == dev->vq->vring.num)
749 return false;
750
751 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000752 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700753
Rusty Russelldde79782007-07-26 10:41:03 -0700754 /* This is why we convert to iovecs: the readv() call uses them, and so
755 * it reads straight into the Guest's buffer. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000756 len = readv(dev->fd, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700757 if (len <= 0) {
Rusty Russelldde79782007-07-26 10:41:03 -0700758 /* This implies that the console is closed, is /dev/null, or
Rusty Russell17cbca22007-10-22 11:24:22 +1000759 * something went terribly wrong. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700760 warnx("Failed to get console input, ignoring console.");
Rusty Russell56ae43d2007-10-22 11:24:23 +1000761 /* Put the input terminal back. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000762 restore_term();
Rusty Russell56ae43d2007-10-22 11:24:23 +1000763 /* Remove callback from input vq, so it doesn't restart us. */
764 dev->vq->handle_output = NULL;
765 /* Stop listening to this fd: don't call us again. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000766 return false;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700767 }
768
Rusty Russell56ae43d2007-10-22 11:24:23 +1000769 /* Tell the Guest about the new input. */
770 add_used_and_trigger(fd, dev->vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700771
Rusty Russelldde79782007-07-26 10:41:03 -0700772 /* Three ^C within one second? Exit.
773 *
774 * This is such a hack, but works surprisingly well. Each ^C has to be
775 * in a buffer by itself, so they can't be too fast. But we check that
776 * we get three within about a second, so they can't be too slow. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700777 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
778 if (!abort->count++)
779 gettimeofday(&abort->start, NULL);
780 else if (abort->count == 3) {
781 struct timeval now;
782 gettimeofday(&now, NULL);
783 if (now.tv_sec <= abort->start.tv_sec+1) {
Jes Sorensen511801d2007-10-22 11:03:31 +1000784 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russelldde79782007-07-26 10:41:03 -0700785 /* Close the fd so Waker will know it has to
786 * exit. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700787 close(waker_fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700788 /* Just in case waker is blocked in BREAK, send
789 * unbreak now. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700790 write(fd, args, sizeof(args));
791 exit(2);
792 }
793 abort->count = 0;
794 }
795 } else
Rusty Russelldde79782007-07-26 10:41:03 -0700796 /* Any other key resets the abort counter. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700797 abort->count = 0;
798
Rusty Russelldde79782007-07-26 10:41:03 -0700799 /* Everything went OK! */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700800 return true;
801}
802
Rusty Russell17cbca22007-10-22 11:24:22 +1000803/* Handling output for console is simple: we just get all the output buffers
804 * and write them to stdout. */
805static void handle_console_output(int fd, struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700806{
Rusty Russell17cbca22007-10-22 11:24:22 +1000807 unsigned int head, out, in;
808 int len;
809 struct iovec iov[vq->vring.num];
810
811 /* Keep getting output buffers from the Guest until we run out. */
812 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
813 if (in)
814 errx(1, "Input buffers in output queue?");
815 len = writev(STDOUT_FILENO, iov, out);
816 add_used_and_trigger(fd, vq, head, len);
817 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700818}
819
Rusty Russell17cbca22007-10-22 11:24:22 +1000820/* Handling output for network is also simple: we get all the output buffers
821 * and write them (ignoring the first element) to this device's file descriptor
822 * (stdout). */
823static void handle_net_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 /* Check header, but otherwise ignore it (we said we supported
834 * no features). */
835 (void)convert(&iov[0], struct virtio_net_hdr);
836 len = writev(vq->dev->fd, iov+1, out-1);
837 add_used_and_trigger(fd, vq, head, len);
838 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700839}
840
Rusty Russell17cbca22007-10-22 11:24:22 +1000841/* This is where we handle a packet coming in from the tun device to our
842 * Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700843static bool handle_tun_input(int fd, struct device *dev)
844{
Rusty Russell17cbca22007-10-22 11:24:22 +1000845 unsigned int head, in_num, out_num;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700846 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000847 struct iovec iov[dev->vq->vring.num];
848 struct virtio_net_hdr *hdr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700849
Rusty Russell17cbca22007-10-22 11:24:22 +1000850 /* First we need a network buffer from the Guests's recv virtqueue. */
851 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
852 if (head == dev->vq->vring.num) {
Rusty Russelldde79782007-07-26 10:41:03 -0700853 /* Now, it's expected that if we try to send a packet too
Rusty Russell17cbca22007-10-22 11:24:22 +1000854 * early, the Guest won't be ready yet. Wait until the device
855 * status says it's ready. */
856 /* FIXME: Actually want DRIVER_ACTIVE here. */
857 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700858 warn("network: no dma buffer!");
Rusty Russell56ae43d2007-10-22 11:24:23 +1000859 /* We'll turn this back on if input buffers are registered. */
860 return false;
Rusty Russell17cbca22007-10-22 11:24:22 +1000861 } else if (out_num)
862 errx(1, "Output buffers in network recv queue?");
863
864 /* First element is the header: we set it to 0 (no features). */
865 hdr = convert(&iov[0], struct virtio_net_hdr);
866 hdr->flags = 0;
867 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700868
Rusty Russelldde79782007-07-26 10:41:03 -0700869 /* Read the packet from the device directly into the Guest's buffer. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000870 len = readv(dev->fd, iov+1, in_num-1);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700871 if (len <= 0)
872 err(1, "reading network");
Rusty Russelldde79782007-07-26 10:41:03 -0700873
Rusty Russell56ae43d2007-10-22 11:24:23 +1000874 /* Tell the Guest about the new packet. */
875 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000876
Rusty Russell8ca47e02007-07-19 01:49:29 -0700877 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
Rusty Russell17cbca22007-10-22 11:24:22 +1000878 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
879 head != dev->vq->vring.num ? "sent" : "discarded");
880
Rusty Russelldde79782007-07-26 10:41:03 -0700881 /* All good. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700882 return true;
883}
884
Rusty Russell56ae43d2007-10-22 11:24:23 +1000885/* This callback ensures we try again, in case we stopped console or net
886 * delivery because Guest didn't have any buffers. */
887static void enable_fd(int fd, struct virtqueue *vq)
888{
889 add_device_fd(vq->dev->fd);
890 /* Tell waker to listen to it again */
891 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
892}
893
Rusty Russell17cbca22007-10-22 11:24:22 +1000894/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
895static void handle_output(int fd, unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700896{
897 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +1000898 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700899
Rusty Russell17cbca22007-10-22 11:24:22 +1000900 /* Check each virtqueue. */
901 for (i = devices.dev; i; i = i->next) {
902 for (vq = i->vq; vq; vq = vq->next) {
903 if (vq->config.pfn == addr/getpagesize()
904 && vq->handle_output) {
905 verbose("Output to %s\n", vq->dev->name);
906 vq->handle_output(fd, vq);
907 return;
908 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700909 }
910 }
Rusty Russelldde79782007-07-26 10:41:03 -0700911
Rusty Russell17cbca22007-10-22 11:24:22 +1000912 /* Early console write is done using notify on a nul-terminated string
913 * in Guest memory. */
914 if (addr >= guest_limit)
915 errx(1, "Bad NOTIFY %#lx", addr);
916
917 write(STDOUT_FILENO, from_guest_phys(addr),
918 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700919}
920
Rusty Russelldde79782007-07-26 10:41:03 -0700921/* This is called when the waker wakes us up: check for incoming file
922 * descriptors. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000923static void handle_input(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700924{
Rusty Russelldde79782007-07-26 10:41:03 -0700925 /* select() wants a zeroed timeval to mean "don't wait". */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700926 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
927
928 for (;;) {
929 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +1000930 fd_set fds = devices.infds;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700931
Rusty Russelldde79782007-07-26 10:41:03 -0700932 /* If nothing is ready, we're done. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000933 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700934 break;
935
Rusty Russelldde79782007-07-26 10:41:03 -0700936 /* Otherwise, call the device(s) which have readable
937 * file descriptors and a method of handling them. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000938 for (i = devices.dev; i; i = i->next) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700939 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
Rusty Russell56ae43d2007-10-22 11:24:23 +1000940 int dev_fd;
941 if (i->handle_input(fd, i))
942 continue;
943
Rusty Russelldde79782007-07-26 10:41:03 -0700944 /* If handle_input() returns false, it means we
Rusty Russell56ae43d2007-10-22 11:24:23 +1000945 * should no longer service it. Networking and
946 * console do this when there's no input
947 * buffers to deliver into. Console also uses
948 * it when it discovers that stdin is
949 * closed. */
950 FD_CLR(i->fd, &devices.infds);
951 /* Tell waker to ignore it too, by sending a
952 * negative fd number (-1, since 0 is a valid
953 * FD number). */
954 dev_fd = -i->fd - 1;
955 write(waker_fd, &dev_fd, sizeof(dev_fd));
Rusty Russell8ca47e02007-07-19 01:49:29 -0700956 }
957 }
958 }
959}
960
Rusty Russelldde79782007-07-26 10:41:03 -0700961/*L:190
962 * Device Setup
963 *
964 * All devices need a descriptor so the Guest knows it exists, and a "struct
965 * device" so the Launcher can keep track of it. We have common helper
966 * routines to allocate them.
967 *
968 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russell17cbca22007-10-22 11:24:22 +1000969 * table just above the Guest's normal memory. It returns a pointer to that
970 * descriptor. */
971static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700972{
Rusty Russell17cbca22007-10-22 11:24:22 +1000973 struct lguest_device_desc *d;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700974
Rusty Russell17cbca22007-10-22 11:24:22 +1000975 /* We only have one page for all the descriptors. */
976 if (devices.desc_used + sizeof(*d) > getpagesize())
977 errx(1, "Too many devices");
978
979 /* We don't need to set config_len or status: page is 0 already. */
980 d = (void *)devices.descpage + devices.desc_used;
981 d->type = type;
982 devices.desc_used += sizeof(*d);
983
984 return d;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700985}
986
Rusty Russell17cbca22007-10-22 11:24:22 +1000987/* Each device descriptor is followed by some configuration information.
988 * The first byte is a "status" byte for the Guest to report what's happening.
989 * After that are fields: u8 type, u8 len, [... len bytes...].
990 *
991 * This routine adds a new field to an existing device's descriptor. It only
992 * works for the last device, but that's OK because that's how we use it. */
993static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
994{
995 /* This is the last descriptor, right? */
996 assert(devices.descpage + devices.desc_used
997 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
998
999 /* We only have one page of device descriptions. */
1000 if (devices.desc_used + 2 + len > getpagesize())
1001 errx(1, "Too many devices");
1002
1003 /* Copy in the new config header: type then length. */
1004 devices.descpage[devices.desc_used++] = type;
1005 devices.descpage[devices.desc_used++] = len;
1006 memcpy(devices.descpage + devices.desc_used, c, len);
1007 devices.desc_used += len;
1008
1009 /* Update the device descriptor length: two byte head then data. */
1010 dev->desc->config_len += 2 + len;
1011}
1012
1013/* This routine adds a virtqueue to a device. We specify how many descriptors
1014 * the virtqueue is to have. */
1015static void add_virtqueue(struct device *dev, unsigned int num_descs,
1016 void (*handle_output)(int fd, struct virtqueue *me))
1017{
1018 unsigned int pages;
1019 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1020 void *p;
1021
1022 /* First we need some pages for this virtqueue. */
1023 pages = (vring_size(num_descs) + getpagesize() - 1) / getpagesize();
1024 p = get_pages(pages);
1025
1026 /* Initialize the configuration. */
1027 vq->config.num = num_descs;
1028 vq->config.irq = devices.next_irq++;
1029 vq->config.pfn = to_guest_phys(p) / getpagesize();
1030
1031 /* Initialize the vring. */
1032 vring_init(&vq->vring, num_descs, p);
1033
1034 /* Add the configuration information to this device's descriptor. */
1035 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE,
1036 sizeof(vq->config), &vq->config);
1037
1038 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1039 * second. */
1040 for (i = &dev->vq; *i; i = &(*i)->next);
1041 *i = vq;
1042
1043 /* Link virtqueue back to device. */
1044 vq->dev = dev;
1045
1046 /* Set up handler. */
1047 vq->handle_output = handle_output;
1048 if (!handle_output)
1049 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1050}
1051
1052/* This routine does all the creation and setup of a new device, including
1053 * caling new_dev_desc() to allocate the descriptor and device memory. */
1054static struct device *new_device(const char *name, u16 type, int fd,
1055 bool (*handle_input)(int, struct device *))
Rusty Russell8ca47e02007-07-19 01:49:29 -07001056{
1057 struct device *dev = malloc(sizeof(*dev));
1058
Rusty Russelldde79782007-07-26 10:41:03 -07001059 /* Append to device list. Prepending to a single-linked list is
1060 * easier, but the user expects the devices to be arranged on the bus
1061 * in command-line order. The first network device on the command line
1062 * is eth0, the first block device /dev/lgba, etc. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001063 *devices.lastdev = dev;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001064 dev->next = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001065 devices.lastdev = &dev->next;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001066
Rusty Russelldde79782007-07-26 10:41:03 -07001067 /* Now we populate the fields one at a time. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001068 dev->fd = fd;
Rusty Russelldde79782007-07-26 10:41:03 -07001069 /* If we have an input handler for this file descriptor, then we add it
1070 * to the device_list's fdset and maxfd. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001071 if (handle_input)
Rusty Russell17cbca22007-10-22 11:24:22 +10001072 add_device_fd(dev->fd);
1073 dev->desc = new_dev_desc(type);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001074 dev->handle_input = handle_input;
Rusty Russell17cbca22007-10-22 11:24:22 +10001075 dev->name = name;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001076 return dev;
1077}
1078
Rusty Russelldde79782007-07-26 10:41:03 -07001079/* Our first setup routine is the console. It's a fairly simple device, but
1080 * UNIX tty handling makes it uglier than it could be. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001081static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001082{
1083 struct device *dev;
1084
Rusty Russelldde79782007-07-26 10:41:03 -07001085 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001086 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1087 struct termios term = orig_term;
Rusty Russelldde79782007-07-26 10:41:03 -07001088 /* Then we turn off echo, line buffering and ^C etc. We want a
1089 * raw input stream to the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001090 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1091 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russelldde79782007-07-26 10:41:03 -07001092 /* If we exit gracefully, the original settings will be
1093 * restored so the user can see what they're typing. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001094 atexit(restore_term);
1095 }
1096
Rusty Russell17cbca22007-10-22 11:24:22 +10001097 dev = new_device("console", VIRTIO_ID_CONSOLE,
1098 STDIN_FILENO, handle_console_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001099 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001100 dev->priv = malloc(sizeof(struct console_abort));
1101 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001102
Rusty Russell56ae43d2007-10-22 11:24:23 +10001103 /* The console needs two virtqueues: the input then the output. When
1104 * they put something the input queue, we make sure we're listening to
1105 * stdin. When they put something in the output queue, we write it to
1106 * stdout. */
1107 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001108 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001109
Rusty Russell17cbca22007-10-22 11:24:22 +10001110 verbose("device %u: console\n", devices.device_num++);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001111}
Rusty Russelldde79782007-07-26 10:41:03 -07001112/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001113
Rusty Russell17cbca22007-10-22 11:24:22 +10001114/*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1115 * --sharenet=<name> option which opens or creates a named pipe. This can be
1116 * used to send packets to another guest in a 1:1 manner.
1117 *
1118 * More sopisticated is to use one of the tools developed for project like UML
1119 * to do networking.
1120 *
1121 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1122 * completely generic ("here's my vring, attach to your vring") and would work
1123 * for any traffic. Of course, namespace and permissions issues need to be
1124 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1125 * multiple inter-guest channels behind one interface, although it would
1126 * require some manner of hotplugging new virtio channels.
1127 *
1128 * Finally, we could implement a virtio network switch in the kernel. :*/
1129
Rusty Russell8ca47e02007-07-19 01:49:29 -07001130static u32 str2ip(const char *ipaddr)
1131{
1132 unsigned int byte[4];
1133
1134 sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]);
1135 return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3];
1136}
1137
Rusty Russelldde79782007-07-26 10:41:03 -07001138/* This code is "adapted" from libbridge: it attaches the Host end of the
1139 * network device to the bridge device specified by the command line.
1140 *
1141 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1142 * dislike bridging), and I just try not to break it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001143static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1144{
1145 int ifidx;
1146 struct ifreq ifr;
1147
1148 if (!*br_name)
1149 errx(1, "must specify bridge name");
1150
1151 ifidx = if_nametoindex(if_name);
1152 if (!ifidx)
1153 errx(1, "interface %s does not exist!", if_name);
1154
1155 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1156 ifr.ifr_ifindex = ifidx;
1157 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1158 err(1, "can't add %s to bridge %s", if_name, br_name);
1159}
1160
Rusty Russelldde79782007-07-26 10:41:03 -07001161/* This sets up the Host end of the network device with an IP address, brings
1162 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell17cbca22007-10-22 11:24:22 +10001163 * pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001164static void configure_device(int fd, const char *devname, u32 ipaddr,
1165 unsigned char hwaddr[6])
1166{
1167 struct ifreq ifr;
1168 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1169
Rusty Russelldde79782007-07-26 10:41:03 -07001170 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001171 memset(&ifr, 0, sizeof(ifr));
1172 strcpy(ifr.ifr_name, devname);
1173 sin->sin_family = AF_INET;
1174 sin->sin_addr.s_addr = htonl(ipaddr);
1175 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1176 err(1, "Setting %s interface address", devname);
1177 ifr.ifr_flags = IFF_UP;
1178 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1179 err(1, "Bringing interface %s up", devname);
1180
Rusty Russelldde79782007-07-26 10:41:03 -07001181 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1182 * above). IF means Interface, and HWADDR is hardware address.
1183 * Simple! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001184 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1185 err(1, "getting hw address for %s", devname);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001186 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1187}
1188
Rusty Russell17cbca22007-10-22 11:24:22 +10001189/*L:195 Our network is a Host<->Guest network. This can either use bridging or
1190 * routing, but the principle is the same: it uses the "tun" device to inject
1191 * packets into the Host as if they came in from a normal network card. We
1192 * just shunt packets between the Guest and the tun device. */
1193static void setup_tun_net(const char *arg)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001194{
1195 struct device *dev;
1196 struct ifreq ifr;
1197 int netfd, ipfd;
1198 u32 ip;
1199 const char *br_name = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001200 u8 hwaddr[6];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001201
Rusty Russelldde79782007-07-26 10:41:03 -07001202 /* We open the /dev/net/tun device and tell it we want a tap device. A
1203 * tap device is like a tun device, only somehow different. To tell
1204 * the truth, I completely blundered my way through this code, but it
1205 * works now! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001206 netfd = open_or_die("/dev/net/tun", O_RDWR);
1207 memset(&ifr, 0, sizeof(ifr));
1208 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1209 strcpy(ifr.ifr_name, "tap%d");
1210 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1211 err(1, "configuring /dev/net/tun");
Rusty Russelldde79782007-07-26 10:41:03 -07001212 /* We don't need checksums calculated for packets coming in this
1213 * device: trust us! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001214 ioctl(netfd, TUNSETNOCSUM, 1);
1215
Rusty Russell17cbca22007-10-22 11:24:22 +10001216 /* First we create a new network device. */
1217 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001218
Rusty Russell56ae43d2007-10-22 11:24:23 +10001219 /* Network devices need a receive and a send queue, just like
1220 * console. */
1221 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001222 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001223
Rusty Russelldde79782007-07-26 10:41:03 -07001224 /* We need a socket to perform the magic network ioctls to bring up the
1225 * tap interface, connect to the bridge etc. Any socket will do! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001226 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1227 if (ipfd < 0)
1228 err(1, "opening IP socket");
1229
Rusty Russelldde79782007-07-26 10:41:03 -07001230 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001231 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1232 ip = INADDR_ANY;
1233 br_name = arg + strlen(BRIDGE_PFX);
1234 add_to_bridge(ipfd, ifr.ifr_name, br_name);
Rusty Russelldde79782007-07-26 10:41:03 -07001235 } else /* It is an IP address to set up the device with */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001236 ip = str2ip(arg);
1237
Rusty Russell17cbca22007-10-22 11:24:22 +10001238 /* Set up the tun device, and get the mac address for the interface. */
1239 configure_device(ipfd, ifr.ifr_name, ip, hwaddr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001240
Rusty Russell17cbca22007-10-22 11:24:22 +10001241 /* Tell Guest what MAC address to use. */
1242 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001243
Rusty Russell17cbca22007-10-22 11:24:22 +10001244 /* We don't seed the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001245 close(ipfd);
1246
Rusty Russell17cbca22007-10-22 11:24:22 +10001247 verbose("device %u: tun net %u.%u.%u.%u\n",
1248 devices.device_num++,
1249 (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001250 if (br_name)
1251 verbose("attached to bridge: %s\n", br_name);
1252}
Rusty Russell17cbca22007-10-22 11:24:22 +10001253
1254
1255/*
1256 * Block device.
1257 *
1258 * Serving a block device is really easy: the Guest asks for a block number and
1259 * we read or write that position in the file.
1260 *
1261 * Unfortunately, this is amazingly slow: the Guest waits until the read is
1262 * finished before running anything else, even if it could be doing useful
1263 * work. We could use async I/O, except it's reputed to suck so hard that
1264 * characters actually go missing from your code when you try to use it.
1265 *
1266 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1267
1268/* This hangs off device->priv, with the data. */
1269struct vblk_info
1270{
1271 /* The size of the file. */
1272 off64_t len;
1273
1274 /* The file descriptor for the file. */
1275 int fd;
1276
1277 /* IO thread listens on this file descriptor [0]. */
1278 int workpipe[2];
1279
1280 /* IO thread writes to this file descriptor to mark it done, then
1281 * Launcher triggers interrupt to Guest. */
1282 int done_fd;
1283};
1284
1285/* This is the core of the I/O thread. It returns true if it did something. */
1286static bool service_io(struct device *dev)
1287{
1288 struct vblk_info *vblk = dev->priv;
1289 unsigned int head, out_num, in_num, wlen;
1290 int ret;
1291 struct virtio_blk_inhdr *in;
1292 struct virtio_blk_outhdr *out;
1293 struct iovec iov[dev->vq->vring.num];
1294 off64_t off;
1295
1296 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1297 if (head == dev->vq->vring.num)
1298 return false;
1299
1300 if (out_num == 0 || in_num == 0)
1301 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1302 head, out_num, in_num);
1303
1304 out = convert(&iov[0], struct virtio_blk_outhdr);
1305 in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr);
1306 off = out->sector * 512;
1307
1308 /* This is how we implement barriers. Pretty poor, no? */
1309 if (out->type & VIRTIO_BLK_T_BARRIER)
1310 fdatasync(vblk->fd);
1311
1312 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1313 fprintf(stderr, "Scsi commands unsupported\n");
1314 in->status = VIRTIO_BLK_S_UNSUPP;
1315 wlen = sizeof(in);
1316 } else if (out->type & VIRTIO_BLK_T_OUT) {
1317 /* Write */
1318
1319 /* Move to the right location in the block file. This can fail
1320 * if they try to write past end. */
1321 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1322 err(1, "Bad seek to sector %llu", out->sector);
1323
1324 ret = writev(vblk->fd, iov+1, out_num-1);
1325 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1326
1327 /* Grr... Now we know how long the descriptor they sent was, we
1328 * make sure they didn't try to write over the end of the block
1329 * file (possibly extending it). */
1330 if (ret > 0 && off + ret > vblk->len) {
1331 /* Trim it back to the correct length */
1332 ftruncate64(vblk->fd, vblk->len);
1333 /* Die, bad Guest, die. */
1334 errx(1, "Write past end %llu+%u", off, ret);
1335 }
1336 wlen = sizeof(in);
1337 in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1338 } else {
1339 /* Read */
1340
1341 /* Move to the right location in the block file. This can fail
1342 * if they try to read past end. */
1343 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1344 err(1, "Bad seek to sector %llu", out->sector);
1345
1346 ret = readv(vblk->fd, iov+1, in_num-1);
1347 verbose("READ from sector %llu: %i\n", out->sector, ret);
1348 if (ret >= 0) {
1349 wlen = sizeof(in) + ret;
1350 in->status = VIRTIO_BLK_S_OK;
1351 } else {
1352 wlen = sizeof(in);
1353 in->status = VIRTIO_BLK_S_IOERR;
1354 }
1355 }
1356
1357 /* We can't trigger an IRQ, because we're not the Launcher. It does
1358 * that when we tell it we're done. */
1359 add_used(dev->vq, head, wlen);
1360 return true;
1361}
1362
1363/* This is the thread which actually services the I/O. */
1364static int io_thread(void *_dev)
1365{
1366 struct device *dev = _dev;
1367 struct vblk_info *vblk = dev->priv;
1368 char c;
1369
1370 /* Close other side of workpipe so we get 0 read when main dies. */
1371 close(vblk->workpipe[1]);
1372 /* Close the other side of the done_fd pipe. */
1373 close(dev->fd);
1374
1375 /* When this read fails, it means Launcher died, so we follow. */
1376 while (read(vblk->workpipe[0], &c, 1) == 1) {
1377 /* We acknowledge each request immediately, to reduce latency,
1378 * rather than waiting until we've done them all. I haven't
1379 * measured to see if it makes any difference. */
1380 while (service_io(dev))
1381 write(vblk->done_fd, &c, 1);
1382 }
1383 return 0;
1384}
1385
1386/* When the thread says some I/O is done, we interrupt the Guest. */
1387static bool handle_io_finish(int fd, struct device *dev)
1388{
1389 char c;
1390
1391 /* If child died, presumably it printed message. */
1392 if (read(dev->fd, &c, 1) != 1)
1393 exit(1);
1394
1395 /* It did some work, so trigger the irq. */
1396 trigger_irq(fd, dev->vq);
1397 return true;
1398}
1399
1400/* When the Guest submits some I/O, we wake the I/O thread. */
1401static void handle_virtblk_output(int fd, struct virtqueue *vq)
1402{
1403 struct vblk_info *vblk = vq->dev->priv;
1404 char c = 0;
1405
1406 /* Wake up I/O thread and tell it to go to work! */
1407 if (write(vblk->workpipe[1], &c, 1) != 1)
1408 /* Presumably it indicated why it died. */
1409 exit(1);
1410}
1411
1412/* This creates a virtual block device. */
1413static void setup_block_file(const char *filename)
1414{
1415 int p[2];
1416 struct device *dev;
1417 struct vblk_info *vblk;
1418 void *stack;
1419 u64 cap;
1420 unsigned int val;
1421
1422 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1423 pipe(p);
1424
1425 /* The device responds to return from I/O thread. */
1426 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1427
1428 /* The device has a virtqueue. */
1429 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1430
1431 /* Allocate the room for our own bookkeeping */
1432 vblk = dev->priv = malloc(sizeof(*vblk));
1433
1434 /* First we open the file and store the length. */
1435 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1436 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1437
1438 /* Tell Guest how many sectors this device has. */
1439 cap = cpu_to_le64(vblk->len / 512);
1440 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1441
1442 /* Tell Guest not to put in too many descriptors at once: two are used
1443 * for the in and out elements. */
1444 val = cpu_to_le32(VIRTQUEUE_NUM - 2);
1445 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val);
1446
1447 /* The I/O thread writes to this end of the pipe when done. */
1448 vblk->done_fd = p[1];
1449
1450 /* This is how we tell the I/O thread about more work. */
1451 pipe(vblk->workpipe);
1452
1453 /* Create stack for thread and run it */
1454 stack = malloc(32768);
1455 if (clone(io_thread, stack + 32768, CLONE_VM, dev) == -1)
1456 err(1, "Creating clone");
1457
1458 /* We don't need to keep the I/O thread's end of the pipes open. */
1459 close(vblk->done_fd);
1460 close(vblk->workpipe[0]);
1461
1462 verbose("device %u: virtblock %llu sectors\n",
1463 devices.device_num, cap);
1464}
Rusty Russelldde79782007-07-26 10:41:03 -07001465/* That's the end of device setup. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001466
Rusty Russelldde79782007-07-26 10:41:03 -07001467/*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1468 * its input and output, and finally, lays it to rest. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001469static void __attribute__((noreturn)) run_guest(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001470{
1471 for (;;) {
Jes Sorensen511801d2007-10-22 11:03:31 +10001472 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russell17cbca22007-10-22 11:24:22 +10001473 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001474 int readval;
1475
1476 /* We read from the /dev/lguest device to run the Guest. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001477 readval = read(lguest_fd, &notify_addr, sizeof(notify_addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001478
Rusty Russell17cbca22007-10-22 11:24:22 +10001479 /* One unsigned long means the Guest did HCALL_NOTIFY */
1480 if (readval == sizeof(notify_addr)) {
1481 verbose("Notify on address %#lx\n", notify_addr);
1482 handle_output(lguest_fd, notify_addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001483 continue;
Rusty Russelldde79782007-07-26 10:41:03 -07001484 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001485 } else if (errno == ENOENT) {
1486 char reason[1024] = { 0 };
1487 read(lguest_fd, reason, sizeof(reason)-1);
1488 errx(1, "%s", reason);
Rusty Russelldde79782007-07-26 10:41:03 -07001489 /* EAGAIN means the waker wanted us to look at some input.
1490 * Anything else means a bug or incompatible change. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001491 } else if (errno != EAGAIN)
1492 err(1, "Running guest failed");
Rusty Russelldde79782007-07-26 10:41:03 -07001493
1494 /* Service input, then unset the BREAK which releases
1495 * the Waker. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001496 handle_input(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001497 if (write(lguest_fd, args, sizeof(args)) < 0)
1498 err(1, "Resetting break");
1499 }
1500}
Rusty Russelldde79782007-07-26 10:41:03 -07001501/*
1502 * This is the end of the Launcher.
1503 *
1504 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1505 * Drivers. If we were to see the Host kernel I/O code, our understanding
1506 * would be complete... :*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001507
1508static struct option opts[] = {
1509 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001510 { "tunnet", 1, NULL, 't' },
1511 { "block", 1, NULL, 'b' },
1512 { "initrd", 1, NULL, 'i' },
1513 { NULL },
1514};
1515static void usage(void)
1516{
1517 errx(1, "Usage: lguest [--verbose] "
Rusty Russell17cbca22007-10-22 11:24:22 +10001518 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001519 "|--block=<filename>|--initrd=<filename>]...\n"
1520 "<mem-in-mb> vmlinux [args...]");
1521}
1522
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001523/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001524int main(int argc, char *argv[])
1525{
Rusty Russell47436aa2007-10-22 11:03:36 +10001526 /* Memory, top-level pagetable, code startpoint and size of the
1527 * (optional) initrd. */
1528 unsigned long mem = 0, pgdir, start, initrd_size = 0;
Rusty Russelldde79782007-07-26 10:41:03 -07001529 /* A temporary and the /dev/lguest file descriptor. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001530 int i, c, lguest_fd;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001531 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001532 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001533 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001534 const char *initrd_name = NULL;
1535
Rusty Russelldde79782007-07-26 10:41:03 -07001536 /* First we initialize the device list. Since console and network
1537 * device receive input from a file descriptor, we keep an fdset
1538 * (infds) and the maximum fd number (max_infd) with the head of the
1539 * list. We also keep a pointer to the last device, for easy appending
Rusty Russell17cbca22007-10-22 11:24:22 +10001540 * to the list. Finally, we keep the next interrupt number to hand out
1541 * (1: remember that 0 is used by the timer). */
1542 FD_ZERO(&devices.infds);
1543 devices.max_infd = -1;
1544 devices.lastdev = &devices.dev;
1545 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001546
Rusty Russelldde79782007-07-26 10:41:03 -07001547 /* We need to know how much memory so we can set up the device
1548 * descriptor and memory pages for the devices as we parse the command
1549 * line. So we quickly look through the arguments to find the amount
1550 * of memory now. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001551 for (i = 1; i < argc; i++) {
1552 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001553 mem = atoi(argv[i]) * 1024 * 1024;
1554 /* We start by mapping anonymous pages over all of
1555 * guest-physical memory range. This fills it with 0,
1556 * and ensures that the Guest won't be killed when it
1557 * tries to access it. */
1558 guest_base = map_zeroed_pages(mem / getpagesize()
1559 + DEVICE_PAGES);
1560 guest_limit = mem;
1561 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001562 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001563 break;
1564 }
1565 }
Rusty Russelldde79782007-07-26 10:41:03 -07001566
1567 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001568 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1569 switch (c) {
1570 case 'v':
1571 verbose = true;
1572 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001573 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001574 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001575 break;
1576 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001577 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001578 break;
1579 case 'i':
1580 initrd_name = optarg;
1581 break;
1582 default:
1583 warnx("Unknown argument %s", argv[optind]);
1584 usage();
1585 }
1586 }
Rusty Russelldde79782007-07-26 10:41:03 -07001587 /* After the other arguments we expect memory and kernel image name,
1588 * followed by command line arguments for the kernel. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001589 if (optind + 2 > argc)
1590 usage();
1591
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001592 verbose("Guest base is at %p\n", guest_base);
1593
Rusty Russelldde79782007-07-26 10:41:03 -07001594 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001595 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001596
Rusty Russell8ca47e02007-07-19 01:49:29 -07001597 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001598 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001599
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001600 /* Boot information is stashed at physical address 0 */
1601 boot = from_guest_phys(0);
1602
Rusty Russelldde79782007-07-26 10:41:03 -07001603 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001604 if (initrd_name) {
1605 initrd_size = load_initrd(initrd_name, mem);
Rusty Russelldde79782007-07-26 10:41:03 -07001606 /* These are the location in the Linux boot header where the
1607 * start and size of the initrd are expected to be found. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001608 boot->hdr.ramdisk_image = mem - initrd_size;
1609 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001610 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001611 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001612 }
1613
Rusty Russelldde79782007-07-26 10:41:03 -07001614 /* Set up the initial linear pagetables, starting below the initrd. */
Rusty Russell47436aa2007-10-22 11:03:36 +10001615 pgdir = setup_pagetables(mem, initrd_size);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001616
Rusty Russelldde79782007-07-26 10:41:03 -07001617 /* The Linux boot header contains an "E820" memory map: ours is a
1618 * simple, single region. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001619 boot->e820_entries = 1;
1620 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russelldde79782007-07-26 10:41:03 -07001621 /* The boot header contains a command line pointer: we put the command
Rusty Russell43d33b22007-10-22 11:29:57 +10001622 * line after the boot header. */
1623 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1624 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07001625
Rusty Russell814a0e52007-10-22 11:29:44 +10001626 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001627 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10001628
1629 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001630 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10001631
Rusty Russell43d33b22007-10-22 11:29:57 +10001632 /* Tell the entry path not to try to reload segment registers. */
1633 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001634
Rusty Russelldde79782007-07-26 10:41:03 -07001635 /* We tell the kernel to initialize the Guest: this returns the open
1636 * /dev/lguest file descriptor. */
Rusty Russell47436aa2007-10-22 11:03:36 +10001637 lguest_fd = tell_kernel(pgdir, start);
Rusty Russelldde79782007-07-26 10:41:03 -07001638
1639 /* We fork off a child process, which wakes the Launcher whenever one
1640 * of the input file descriptors needs attention. Otherwise we would
1641 * run the Guest until it tries to output something. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001642 waker_fd = setup_waker(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001643
Rusty Russelldde79782007-07-26 10:41:03 -07001644 /* Finally, run the Guest. This doesn't return. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001645 run_guest(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001646}
Rusty Russellf56a3842007-07-26 10:41:05 -07001647/*:*/
1648
1649/*M:999
1650 * Mastery is done: you now know everything I do.
1651 *
1652 * But surely you have seen code, features and bugs in your wanderings which
1653 * you now yearn to attack? That is the real game, and I look forward to you
1654 * patching and forking lguest into the Your-Name-Here-visor.
1655 *
1656 * Farewell, and good coding!
1657 * Rusty Russell.
1658 */