blob: f2dbbf3bdeaba7a5c6d58e1ccb3f8a686ade88ed [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
Rusty Russella6bd8e12008-03-28 11:05:53 -05002 * "physical" memory for the new Guest by mapping the kernel image and
3 * the virtual devices, then opens /dev/lguest to tell the kernel
4 * about the Guest and control it. :*/
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 Russella586d4f2008-02-04 23:49:56 -050037#include <limits.h>
38#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050039#include <signal.h>
Rusty Russellb45d8cb2007-10-22 10:56:24 +100040#include "linux/lguest_launcher.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100041#include "linux/virtio_config.h"
42#include "linux/virtio_net.h"
43#include "linux/virtio_blk.h"
44#include "linux/virtio_console.h"
Rusty Russell28fd6d72008-07-29 09:58:33 -050045#include "linux/virtio_rng.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100046#include "linux/virtio_ring.h"
Rusty Russelld5d02d62008-10-31 11:24:25 -050047#include "asm/bootparam.h"
Rusty Russella6bd8e12008-03-28 11:05:53 -050048/*L:110 We can ignore the 39 include files we need for this program, but I do
Rusty Russelldb24e8c2007-10-25 14:09:25 +100049 * want to draw attention to the use of kernel-style types.
50 *
51 * As Linus said, "C is a Spartan language, and so should your naming be." I
52 * like these abbreviations, so we define them here. Note that u64 is always
53 * unsigned long long, which works on all Linux systems: this means that we can
54 * use %llu in printf for any u64. */
55typedef unsigned long long u64;
56typedef uint32_t u32;
57typedef uint16_t u16;
58typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070059/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070060
61#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
62#define NET_PEERNUM 1
63#define BRIDGE_PFX "bridge:"
64#ifndef SIOCBRADDIF
65#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
66#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100067/* We can have up to 256 pages for devices. */
68#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050069/* This will occupy 3 pages: it must be a power of 2. */
70#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070071
Rusty Russelldde79782007-07-26 10:41:03 -070072/*L:120 verbose is both a global flag and a macro. The C preprocessor allows
73 * this, and although I wouldn't recommend it, it works quite nicely here. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070074static bool verbose;
75#define verbose(args...) \
76 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070077/*:*/
78
Rusty Russell8c798732008-07-29 09:58:38 -050079/* File descriptors for the Waker. */
80struct {
81 int pipe[2];
82 int lguest_fd;
83} waker_fds;
84
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100085/* The pointer to the start of guest memory. */
86static void *guest_base;
87/* The maximum guest physical address allowed, and maximum possible. */
88static unsigned long guest_limit, guest_max;
Rusty Russella1618832008-07-29 09:58:35 -050089/* The pipe for signal hander to write to. */
90static int timeoutpipe[2];
Rusty Russellaa124982008-07-29 09:58:36 -050091static unsigned int timeout_usec = 500;
Rusty Russell8ca47e02007-07-19 01:49:29 -070092
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020093/* a per-cpu variable indicating whose vcpu is currently running */
94static unsigned int __thread cpu_id;
95
Rusty Russelldde79782007-07-26 10:41:03 -070096/* This is our list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070097struct device_list
98{
Rusty Russelldde79782007-07-26 10:41:03 -070099 /* Summary information about the devices in our list: ready to pass to
100 * select() to ask which need servicing.*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700101 fd_set infds;
102 int max_infd;
103
Rusty Russell17cbca22007-10-22 11:24:22 +1000104 /* Counter to assign interrupt numbers. */
105 unsigned int next_irq;
106
107 /* Counter to print out convenient device numbers. */
108 unsigned int device_num;
109
Rusty Russelldde79782007-07-26 10:41:03 -0700110 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000111 u8 *descpage;
112
Rusty Russelldde79782007-07-26 10:41:03 -0700113 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700114 struct device *dev;
Rusty Russella586d4f2008-02-04 23:49:56 -0500115 /* And a pointer to the last device for easy append and also for
116 * configuration appending. */
117 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700118};
119
Rusty Russell17cbca22007-10-22 11:24:22 +1000120/* The list of Guest devices, based on command line arguments. */
121static struct device_list devices;
122
Rusty Russelldde79782007-07-26 10:41:03 -0700123/* The device structure describes a single device. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700124struct device
125{
Rusty Russelldde79782007-07-26 10:41:03 -0700126 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700127 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000128
129 /* The this device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700130 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000131
132 /* The name of this device, for --verbose. */
133 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700134
Rusty Russelldde79782007-07-26 10:41:03 -0700135 /* If handle_input is set, it wants to be called when this file
136 * descriptor is ready. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700137 int fd;
138 bool (*handle_input)(int fd, struct device *me);
139
Rusty Russell17cbca22007-10-22 11:24:22 +1000140 /* Any queues attached to this device */
141 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700142
Rusty Russella007a752008-05-02 21:50:53 -0500143 /* Handle status being finalized (ie. feature bits stable). */
144 void (*ready)(struct device *me);
145
Rusty Russell8ca47e02007-07-19 01:49:29 -0700146 /* Device-specific data. */
147 void *priv;
148};
149
Rusty Russell17cbca22007-10-22 11:24:22 +1000150/* The virtqueue structure describes a queue attached to a device. */
151struct virtqueue
152{
153 struct virtqueue *next;
154
155 /* Which device owns me. */
156 struct device *dev;
157
158 /* The configuration for this queue. */
159 struct lguest_vqconfig config;
160
161 /* The actual ring of buffers. */
162 struct vring vring;
163
164 /* Last available index we saw. */
165 u16 last_avail_idx;
166
Rusty Russella1618832008-07-29 09:58:35 -0500167 /* The routine to call when the Guest pings us, or timeout. */
168 void (*handle_output)(int fd, struct virtqueue *me, bool timeout);
Rusty Russell20887612008-05-30 15:09:46 -0500169
170 /* Outstanding buffers */
171 unsigned int inflight;
Rusty Russella1618832008-07-29 09:58:35 -0500172
173 /* Is this blocked awaiting a timer? */
174 bool blocked;
Rusty Russell17cbca22007-10-22 11:24:22 +1000175};
176
Balaji Raoec04b132007-12-28 14:26:24 +0530177/* Remember the arguments to the program so we can "reboot" */
178static char **main_args;
179
Rusty Russell17cbca22007-10-22 11:24:22 +1000180/* Since guest is UP and we don't run at the same time, we don't need barriers.
181 * But I include them in the code in case others copy it. */
182#define wmb()
183
184/* Convert an iovec element to the given type.
185 *
186 * This is a fairly ugly trick: we need to know the size of the type and
187 * alignment requirement to check the pointer is kosher. It's also nice to
188 * have the name of the type in case we report failure.
189 *
190 * Typing those three things all the time is cumbersome and error prone, so we
191 * have a macro which sets them all up and passes to the real function. */
192#define convert(iov, type) \
193 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
194
195static void *_convert(struct iovec *iov, size_t size, size_t align,
196 const char *name)
197{
198 if (iov->iov_len != size)
199 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
200 if ((unsigned long)iov->iov_base % align != 0)
201 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
202 return iov->iov_base;
203}
204
Rusty Russellb5111792008-07-29 09:58:34 -0500205/* Wrapper for the last available index. Makes it easier to change. */
206#define lg_last_avail(vq) ((vq)->last_avail_idx)
207
Rusty Russell17cbca22007-10-22 11:24:22 +1000208/* The virtio configuration space is defined to be little-endian. x86 is
209 * little-endian too, but it's nice to be explicit so we have these helpers. */
210#define cpu_to_le16(v16) (v16)
211#define cpu_to_le32(v32) (v32)
212#define cpu_to_le64(v64) (v64)
213#define le16_to_cpu(v16) (v16)
214#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500215#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000216
Rusty Russell28fd6d72008-07-29 09:58:33 -0500217/* Is this iovec empty? */
218static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
219{
220 unsigned int i;
221
222 for (i = 0; i < num_iov; i++)
223 if (iov[i].iov_len)
224 return false;
225 return true;
226}
227
228/* Take len bytes from the front of this iovec. */
229static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
230{
231 unsigned int i;
232
233 for (i = 0; i < num_iov; i++) {
234 unsigned int used;
235
236 used = iov[i].iov_len < len ? iov[i].iov_len : len;
237 iov[i].iov_base += used;
238 iov[i].iov_len -= used;
239 len -= used;
240 }
241 assert(len == 0);
242}
243
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500244/* The device virtqueue descriptors are followed by feature bitmasks. */
245static u8 *get_feature_bits(struct device *dev)
246{
247 return (u8 *)(dev->desc + 1)
248 + dev->desc->num_vq * sizeof(struct lguest_vqconfig);
249}
250
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000251/*L:100 The Launcher code itself takes us out into userspace, that scary place
252 * where pointers run wild and free! Unfortunately, like most userspace
253 * programs, it's quite boring (which is why everyone likes to hack on the
254 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
255 * will get you through this section. Or, maybe not.
256 *
257 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
258 * memory and stores it in "guest_base". In other words, Guest physical ==
259 * Launcher virtual with an offset.
260 *
261 * This can be tough to get your head around, but usually it just means that we
262 * use these trivial conversion functions when the Guest gives us it's
263 * "physical" addresses: */
264static void *from_guest_phys(unsigned long addr)
265{
266 return guest_base + addr;
267}
268
269static unsigned long to_guest_phys(const void *addr)
270{
271 return (addr - guest_base);
272}
273
Rusty Russelldde79782007-07-26 10:41:03 -0700274/*L:130
275 * Loading the Kernel.
276 *
277 * We start with couple of simple helper routines. open_or_die() avoids
278 * error-checking code cluttering the callers: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700279static int open_or_die(const char *name, int flags)
280{
281 int fd = open(name, flags);
282 if (fd < 0)
283 err(1, "Failed to open %s", name);
284 return fd;
285}
286
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000287/* map_zeroed_pages() takes a number of pages. */
288static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700289{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000290 int fd = open_or_die("/dev/zero", O_RDONLY);
291 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700292
Rusty Russelldde79782007-07-26 10:41:03 -0700293 /* We use a private mapping (ie. if we write to the page, it will be
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000294 * copied). */
295 addr = mmap(NULL, getpagesize() * num,
296 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
297 if (addr == MAP_FAILED)
298 err(1, "Mmaping %u pages of /dev/zero", num);
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100299 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700300
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000301 return addr;
302}
303
304/* Get some more pages for a device. */
305static void *get_pages(unsigned int num)
306{
307 void *addr = from_guest_phys(guest_limit);
308
309 guest_limit += num * getpagesize();
310 if (guest_limit > guest_max)
311 errx(1, "Not enough memory for devices");
312 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700313}
314
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700315/* This routine is used to load the kernel or initrd. It tries mmap, but if
316 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
317 * it falls back to reading the memory in. */
318static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
319{
320 ssize_t r;
321
322 /* We map writable even though for some segments are marked read-only.
323 * The kernel really wants to be writable: it patches its own
324 * instructions.
325 *
326 * MAP_PRIVATE means that the page won't be copied until a write is
327 * done to it. This allows us to share untouched memory between
328 * Guests. */
329 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
330 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
331 return;
332
333 /* pread does a seek and a read in one shot: saves a few lines. */
334 r = pread(fd, addr, len, offset);
335 if (r != len)
336 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
337}
338
Rusty Russelldde79782007-07-26 10:41:03 -0700339/* This routine takes an open vmlinux image, which is in ELF, and maps it into
340 * the Guest memory. ELF = Embedded Linking Format, which is the format used
341 * by all modern binaries on Linux including the kernel.
342 *
343 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000344 * address. We use the physical address; the Guest will map itself to the
345 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700346 *
347 * We return the starting address. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000348static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700349{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700350 Elf32_Phdr phdr[ehdr->e_phnum];
351 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700352
Rusty Russelldde79782007-07-26 10:41:03 -0700353 /* Sanity checks on the main ELF header: an x86 executable with a
354 * reasonable number of correctly-sized program headers. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700355 if (ehdr->e_type != ET_EXEC
356 || ehdr->e_machine != EM_386
357 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
358 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
359 errx(1, "Malformed elf header");
360
Rusty Russelldde79782007-07-26 10:41:03 -0700361 /* An ELF executable contains an ELF header and a number of "program"
362 * headers which indicate which parts ("segments") of the program to
363 * load where. */
364
365 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700366 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
367 err(1, "Seeking to program headers");
368 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
369 err(1, "Reading program headers");
370
Rusty Russelldde79782007-07-26 10:41:03 -0700371 /* Try all the headers: there are usually only three. A read-only one,
Rusty Russella6bd8e12008-03-28 11:05:53 -0500372 * a read-write one, and a "note" section which we don't load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700373 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700374 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700375 if (phdr[i].p_type != PT_LOAD)
376 continue;
377
378 verbose("Section %i: size %i addr %p\n",
379 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
380
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700381 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000382 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700383 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700384 }
385
Rusty Russell814a0e52007-10-22 11:29:44 +1000386 /* The entry point is given in the ELF header. */
387 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700388}
389
Rusty Russelldde79782007-07-26 10:41:03 -0700390/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000391 * supposed to jump into it and it will unpack itself. We used to have to
392 * perform some hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700393 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000394 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
395 * a small patch to jump over the tricky bits in the Guest, so now we just read
396 * the funky header so we know where in the file to load, and away we go! */
Rusty Russell47436aa2007-10-22 11:03:36 +1000397static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700398{
Rusty Russell43d33b22007-10-22 11:29:57 +1000399 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000400 int r;
401 /* Modern bzImages get loaded at 1M. */
402 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700403
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000404 /* Go back to the start of the file and read the header. It should be
Uwe Hermann71cced62008-10-20 09:32:21 -0700405 * a Linux boot header (see Documentation/x86/i386/boot.txt) */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000406 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000407 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000408
Rusty Russell43d33b22007-10-22 11:29:57 +1000409 /* Inside the setup_hdr, we expect the magic "HdrS" */
410 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000411 errx(1, "This doesn't look like a bzImage to me");
412
Rusty Russell43d33b22007-10-22 11:29:57 +1000413 /* Skip over the extra sectors of the header. */
414 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000415
416 /* Now read everything into memory. in nice big chunks. */
417 while ((r = read(fd, p, 65536)) > 0)
418 p += r;
419
Rusty Russell43d33b22007-10-22 11:29:57 +1000420 /* Finally, code32_start tells us where to enter the kernel. */
421 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700422}
423
Rusty Russelldde79782007-07-26 10:41:03 -0700424/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000425 * come wrapped up in the self-decompressing "bzImage" format. With a little
426 * work, we can load those, too. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000427static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700428{
429 Elf32_Ehdr hdr;
430
Rusty Russelldde79782007-07-26 10:41:03 -0700431 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700432 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
433 err(1, "Reading kernel");
434
Rusty Russelldde79782007-07-26 10:41:03 -0700435 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700436 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000437 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700438
Rusty Russella6bd8e12008-03-28 11:05:53 -0500439 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000440 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700441}
442
Rusty Russelldde79782007-07-26 10:41:03 -0700443/* This is a trivial little helper to align pages. Andi Kleen hated it because
444 * it calls getpagesize() twice: "it's dumb code."
445 *
446 * Kernel guys get really het up about optimization, even when it's not
447 * necessary. I leave this code as a reaction against that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700448static inline unsigned long page_align(unsigned long addr)
449{
Rusty Russelldde79782007-07-26 10:41:03 -0700450 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700451 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
452}
453
Rusty Russelldde79782007-07-26 10:41:03 -0700454/*L:180 An "initial ram disk" is a disk image loaded into memory along with
455 * the kernel which the kernel can use to boot from without needing any
456 * drivers. Most distributions now use this as standard: the initrd contains
457 * the code to load the appropriate driver modules for the current machine.
458 *
459 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
460 * kernels. He sent me this (and tells me when I break it). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700461static unsigned long load_initrd(const char *name, unsigned long mem)
462{
463 int ifd;
464 struct stat st;
465 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700466
467 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700468 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700469 if (fstat(ifd, &st) < 0)
470 err(1, "fstat() on initrd '%s'", name);
471
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700472 /* We map the initrd at the top of memory, but mmap wants it to be
473 * page-aligned, so we round the size up for that. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700474 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000475 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russelldde79782007-07-26 10:41:03 -0700476 /* Once a file is mapped, you can close the file descriptor. It's a
477 * little odd, but quite useful. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700478 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700479 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700480
481 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700482 return len;
483}
Rusty Russelle1e72962007-10-25 15:02:50 +1000484/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700485
Rusty Russelldde79782007-07-26 10:41:03 -0700486/* Simple routine to roll all the commandline arguments together with spaces
487 * between them. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700488static void concat(char *dst, char *args[])
489{
490 unsigned int i, len = 0;
491
492 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100493 if (i) {
494 strcat(dst+len, " ");
495 len++;
496 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700497 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100498 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700499 }
500 /* In case it's empty. */
501 dst[len] = '\0';
502}
503
Rusty Russelle1e72962007-10-25 15:02:50 +1000504/*L:185 This is where we actually tell the kernel to initialize the Guest. We
505 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300506 * the base of Guest "physical" memory, the top physical page to allow and the
507 * entry point for the Guest. */
508static int tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700509{
Jes Sorensen511801d2007-10-22 11:03:31 +1000510 unsigned long args[] = { LHREQ_INITIALIZE,
511 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300512 guest_limit / getpagesize(), start };
Rusty Russell8ca47e02007-07-19 01:49:29 -0700513 int fd;
514
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000515 verbose("Guest: %p - %p (%#lx)\n",
516 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700517 fd = open_or_die("/dev/lguest", O_RDWR);
518 if (write(fd, args, sizeof(args)) < 0)
519 err(1, "Writing to /dev/lguest");
Rusty Russelldde79782007-07-26 10:41:03 -0700520
521 /* We return the /dev/lguest file descriptor to control this Guest */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700522 return fd;
523}
Rusty Russelldde79782007-07-26 10:41:03 -0700524/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700525
Rusty Russell17cbca22007-10-22 11:24:22 +1000526static void add_device_fd(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700527{
Rusty Russell17cbca22007-10-22 11:24:22 +1000528 FD_SET(fd, &devices.infds);
529 if (fd > devices.max_infd)
530 devices.max_infd = fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700531}
532
Rusty Russelldde79782007-07-26 10:41:03 -0700533/*L:200
534 * The Waker.
535 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000536 * With console, block and network devices, we can have lots of input which we
537 * need to process. We could try to tell the kernel what file descriptors to
538 * watch, but handing a file descriptor mask through to the kernel is fairly
539 * icky.
Rusty Russelldde79782007-07-26 10:41:03 -0700540 *
Rusty Russell8c798732008-07-29 09:58:38 -0500541 * Instead, we clone off a thread which watches the file descriptors and writes
Rusty Russelle1e72962007-10-25 15:02:50 +1000542 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
543 * stop running the Guest. This causes the Launcher to return from the
Rusty Russelldde79782007-07-26 10:41:03 -0700544 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
545 * the LHREQ_BREAK and wake us up again.
546 *
547 * This, of course, is merely a different *kind* of icky.
Rusty Russell8c798732008-07-29 09:58:38 -0500548 *
549 * Given my well-known antipathy to threads, I'd prefer to use processes. But
550 * it's easier to share Guest memory with threads, and trivial to share the
551 * devices.infds as the Launcher changes it.
Rusty Russelldde79782007-07-26 10:41:03 -0700552 */
Rusty Russell8c798732008-07-29 09:58:38 -0500553static int waker(void *unused)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700554{
Rusty Russell8c798732008-07-29 09:58:38 -0500555 /* Close the write end of the pipe: only the Launcher has it open. */
556 close(waker_fds.pipe[1]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700557
558 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +1000559 fd_set rfds = devices.infds;
Jes Sorensen511801d2007-10-22 11:03:31 +1000560 unsigned long args[] = { LHREQ_BREAK, 1 };
Rusty Russell8c798732008-07-29 09:58:38 -0500561 unsigned int maxfd = devices.max_infd;
562
563 /* We also listen to the pipe from the Launcher. */
564 FD_SET(waker_fds.pipe[0], &rfds);
565 if (waker_fds.pipe[0] > maxfd)
566 maxfd = waker_fds.pipe[0];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700567
Rusty Russelldde79782007-07-26 10:41:03 -0700568 /* Wait until input is ready from one of the devices. */
Rusty Russell8c798732008-07-29 09:58:38 -0500569 select(maxfd+1, &rfds, NULL, NULL, NULL);
570
571 /* Message from Launcher? */
572 if (FD_ISSET(waker_fds.pipe[0], &rfds)) {
573 char c;
574 /* If this fails, then assume Launcher has exited.
575 * Don't do anything on exit: we're just a thread! */
576 if (read(waker_fds.pipe[0], &c, 1) != 1)
577 _exit(0);
578 continue;
579 }
580
581 /* Send LHREQ_BREAK command to snap the Launcher out of it. */
582 pwrite(waker_fds.lguest_fd, args, sizeof(args), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700583 }
Rusty Russell8c798732008-07-29 09:58:38 -0500584 return 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700585}
586
Rusty Russelldde79782007-07-26 10:41:03 -0700587/* This routine just sets up a pipe to the Waker process. */
Rusty Russell8c798732008-07-29 09:58:38 -0500588static void setup_waker(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700589{
Rusty Russell8c798732008-07-29 09:58:38 -0500590 /* This pipe is closed when Launcher dies, telling Waker. */
591 if (pipe(waker_fds.pipe) != 0)
592 err(1, "Creating pipe for Waker");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700593
Rusty Russell8c798732008-07-29 09:58:38 -0500594 /* Waker also needs to know the lguest fd */
595 waker_fds.lguest_fd = lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700596
Rusty Russell8c798732008-07-29 09:58:38 -0500597 if (clone(waker, malloc(4096) + 4096, CLONE_VM | SIGCHLD, NULL) == -1)
598 err(1, "Creating Waker");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700599}
600
Rusty Russelle1e72962007-10-25 15:02:50 +1000601/*
Rusty Russelldde79782007-07-26 10:41:03 -0700602 * Device Handling.
603 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000604 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700605 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000606 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700607 * if something funny is going on:
608 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700609static void *_check_pointer(unsigned long addr, unsigned int size,
610 unsigned int line)
611{
Rusty Russelldde79782007-07-26 10:41:03 -0700612 /* We have to separately check addr and addr+size, because size could
613 * be huge and addr + size might wrap around. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000614 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000615 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russelldde79782007-07-26 10:41:03 -0700616 /* We return a pointer for the caller's convenience, now we know it's
617 * safe to use. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000618 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700619}
Rusty Russelldde79782007-07-26 10:41:03 -0700620/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700621#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
622
Rusty Russelle1e72962007-10-25 15:02:50 +1000623/* Each buffer in the virtqueues is actually a chain of descriptors. This
624 * function returns the next descriptor in the chain, or vq->vring.num if we're
625 * at the end. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000626static unsigned next_desc(struct virtqueue *vq, unsigned int i)
627{
628 unsigned int next;
629
630 /* If this descriptor says it doesn't chain, we're done. */
631 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
632 return vq->vring.num;
633
634 /* Check they're not leading us off end of descriptors. */
635 next = vq->vring.desc[i].next;
636 /* Make sure compiler knows to grab that: we don't want it changing! */
637 wmb();
638
639 if (next >= vq->vring.num)
640 errx(1, "Desc next is %u", next);
641
642 return next;
643}
644
645/* This looks in the virtqueue and for the first available buffer, and converts
646 * it to an iovec for convenient access. Since descriptors consist of some
647 * number of output then some number of input descriptors, it's actually two
648 * iovecs, but we pack them into one and note how many of each there were.
649 *
650 * This function returns the descriptor number found, or vq->vring.num (which
651 * is never a valid descriptor number) if none was found. */
652static unsigned get_vq_desc(struct virtqueue *vq,
653 struct iovec iov[],
654 unsigned int *out_num, unsigned int *in_num)
655{
656 unsigned int i, head;
Rusty Russellb5111792008-07-29 09:58:34 -0500657 u16 last_avail;
Rusty Russell17cbca22007-10-22 11:24:22 +1000658
659 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500660 last_avail = lg_last_avail(vq);
661 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000662 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500663 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000664
665 /* If there's nothing new since last we looked, return invalid. */
Rusty Russellb5111792008-07-29 09:58:34 -0500666 if (vq->vring.avail->idx == last_avail)
Rusty Russell17cbca22007-10-22 11:24:22 +1000667 return vq->vring.num;
668
669 /* Grab the next descriptor number they're advertising, and increment
670 * the index we've seen. */
Rusty Russellb5111792008-07-29 09:58:34 -0500671 head = vq->vring.avail->ring[last_avail % vq->vring.num];
672 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000673
674 /* If their number is silly, that's a fatal mistake. */
675 if (head >= vq->vring.num)
676 errx(1, "Guest says index %u is available", head);
677
678 /* When we start there are none of either input nor output. */
679 *out_num = *in_num = 0;
680
681 i = head;
682 do {
683 /* Grab the first descriptor, and check it's OK. */
684 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
685 iov[*out_num + *in_num].iov_base
686 = check_pointer(vq->vring.desc[i].addr,
687 vq->vring.desc[i].len);
688 /* If this is an input descriptor, increment that count. */
689 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
690 (*in_num)++;
691 else {
692 /* If it's an output descriptor, they're all supposed
693 * to come before any input descriptors. */
694 if (*in_num)
695 errx(1, "Descriptor has out after in");
696 (*out_num)++;
697 }
698
699 /* If we've got too many, that implies a descriptor loop. */
700 if (*out_num + *in_num > vq->vring.num)
701 errx(1, "Looped descriptor");
702 } while ((i = next_desc(vq, i)) != vq->vring.num);
703
Rusty Russell20887612008-05-30 15:09:46 -0500704 vq->inflight++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000705 return head;
706}
707
Rusty Russelle1e72962007-10-25 15:02:50 +1000708/* After we've used one of their buffers, we tell them about it. We'll then
Rusty Russell17cbca22007-10-22 11:24:22 +1000709 * want to send them an interrupt, using trigger_irq(). */
710static void add_used(struct virtqueue *vq, unsigned int head, int len)
711{
712 struct vring_used_elem *used;
713
Rusty Russelle1e72962007-10-25 15:02:50 +1000714 /* The virtqueue contains a ring of used buffers. Get a pointer to the
715 * next entry in that used ring. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000716 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
717 used->id = head;
718 used->len = len;
719 /* Make sure buffer is written before we update index. */
720 wmb();
721 vq->vring.used->idx++;
Rusty Russell20887612008-05-30 15:09:46 -0500722 vq->inflight--;
Rusty Russell17cbca22007-10-22 11:24:22 +1000723}
724
725/* This actually sends the interrupt for this virtqueue */
726static void trigger_irq(int fd, struct virtqueue *vq)
727{
728 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
729
Rusty Russell20887612008-05-30 15:09:46 -0500730 /* If they don't want an interrupt, don't send one, unless empty. */
731 if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
732 && vq->inflight)
Rusty Russell17cbca22007-10-22 11:24:22 +1000733 return;
734
735 /* Send the Guest an interrupt tell them we used something up. */
736 if (write(fd, buf, sizeof(buf)) != 0)
737 err(1, "Triggering irq %i", vq->config.irq);
738}
739
740/* And here's the combo meal deal. Supersize me! */
741static void add_used_and_trigger(int fd, struct virtqueue *vq,
742 unsigned int head, int len)
743{
744 add_used(vq, head, len);
745 trigger_irq(fd, vq);
746}
747
Rusty Russelle1e72962007-10-25 15:02:50 +1000748/*
749 * The Console
750 *
751 * Here is the input terminal setting we save, and the routine to restore them
752 * on exit so the user gets their terminal back. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700753static struct termios orig_term;
754static void restore_term(void)
755{
756 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
757}
758
Rusty Russelldde79782007-07-26 10:41:03 -0700759/* We associate some data with the console for our exit hack. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700760struct console_abort
761{
Rusty Russelldde79782007-07-26 10:41:03 -0700762 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700763 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700764 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700765 struct timeval start;
766};
767
Rusty Russelldde79782007-07-26 10:41:03 -0700768/* This is the routine which handles console input (ie. stdin). */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700769static bool handle_console_input(int fd, struct device *dev)
770{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700771 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000772 unsigned int head, in_num, out_num;
773 struct iovec iov[dev->vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700774 struct console_abort *abort = dev->priv;
775
Rusty Russell17cbca22007-10-22 11:24:22 +1000776 /* First we need a console buffer from the Guests's input virtqueue. */
777 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000778
779 /* If they're not ready for input, stop listening to this file
780 * descriptor. We'll start again once they add an input buffer. */
781 if (head == dev->vq->vring.num)
782 return false;
783
784 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000785 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700786
Rusty Russelldde79782007-07-26 10:41:03 -0700787 /* This is why we convert to iovecs: the readv() call uses them, and so
788 * it reads straight into the Guest's buffer. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000789 len = readv(dev->fd, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700790 if (len <= 0) {
Rusty Russelldde79782007-07-26 10:41:03 -0700791 /* This implies that the console is closed, is /dev/null, or
Rusty Russell17cbca22007-10-22 11:24:22 +1000792 * something went terribly wrong. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700793 warnx("Failed to get console input, ignoring console.");
Rusty Russell56ae43d2007-10-22 11:24:23 +1000794 /* Put the input terminal back. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000795 restore_term();
Rusty Russell56ae43d2007-10-22 11:24:23 +1000796 /* Remove callback from input vq, so it doesn't restart us. */
797 dev->vq->handle_output = NULL;
798 /* Stop listening to this fd: don't call us again. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000799 return false;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700800 }
801
Rusty Russell56ae43d2007-10-22 11:24:23 +1000802 /* Tell the Guest about the new input. */
803 add_used_and_trigger(fd, dev->vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700804
Rusty Russelldde79782007-07-26 10:41:03 -0700805 /* Three ^C within one second? Exit.
806 *
807 * This is such a hack, but works surprisingly well. Each ^C has to be
808 * in a buffer by itself, so they can't be too fast. But we check that
809 * we get three within about a second, so they can't be too slow. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700810 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
811 if (!abort->count++)
812 gettimeofday(&abort->start, NULL);
813 else if (abort->count == 3) {
814 struct timeval now;
815 gettimeofday(&now, NULL);
816 if (now.tv_sec <= abort->start.tv_sec+1) {
Jes Sorensen511801d2007-10-22 11:03:31 +1000817 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russelldde79782007-07-26 10:41:03 -0700818 /* Close the fd so Waker will know it has to
819 * exit. */
Rusty Russell8c798732008-07-29 09:58:38 -0500820 close(waker_fds.pipe[1]);
821 /* Just in case Waker is blocked in BREAK, send
Rusty Russelldde79782007-07-26 10:41:03 -0700822 * unbreak now. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700823 write(fd, args, sizeof(args));
824 exit(2);
825 }
826 abort->count = 0;
827 }
828 } else
Rusty Russelldde79782007-07-26 10:41:03 -0700829 /* Any other key resets the abort counter. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700830 abort->count = 0;
831
Rusty Russelldde79782007-07-26 10:41:03 -0700832 /* Everything went OK! */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700833 return true;
834}
835
Rusty Russell17cbca22007-10-22 11:24:22 +1000836/* Handling output for console is simple: we just get all the output buffers
837 * and write them to stdout. */
Rusty Russella1618832008-07-29 09:58:35 -0500838static void handle_console_output(int fd, struct virtqueue *vq, bool timeout)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700839{
Rusty Russell17cbca22007-10-22 11:24:22 +1000840 unsigned int head, out, in;
841 int len;
842 struct iovec iov[vq->vring.num];
843
844 /* Keep getting output buffers from the Guest until we run out. */
845 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
846 if (in)
847 errx(1, "Input buffers in output queue?");
848 len = writev(STDOUT_FILENO, iov, out);
849 add_used_and_trigger(fd, vq, head, len);
850 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700851}
852
Rusty Russell1dc3e3b2008-08-26 00:19:27 -0500853/* This is called when we no longer want to hear about Guest changes to a
854 * virtqueue. This is more efficient in high-traffic cases, but it means we
855 * have to set a timer to check if any more changes have occurred. */
Rusty Russella1618832008-07-29 09:58:35 -0500856static void block_vq(struct virtqueue *vq)
857{
858 struct itimerval itm;
859
860 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
861 vq->blocked = true;
862
863 itm.it_interval.tv_sec = 0;
864 itm.it_interval.tv_usec = 0;
865 itm.it_value.tv_sec = 0;
Rusty Russellaa124982008-07-29 09:58:36 -0500866 itm.it_value.tv_usec = timeout_usec;
Rusty Russella1618832008-07-29 09:58:35 -0500867
868 setitimer(ITIMER_REAL, &itm, NULL);
869}
870
Rusty Russelle1e72962007-10-25 15:02:50 +1000871/*
872 * The Network
873 *
874 * Handling output for network is also simple: we get all the output buffers
Rusty Russell17cbca22007-10-22 11:24:22 +1000875 * and write them (ignoring the first element) to this device's file descriptor
Rusty Russella6bd8e12008-03-28 11:05:53 -0500876 * (/dev/net/tun).
877 */
Rusty Russella1618832008-07-29 09:58:35 -0500878static void handle_net_output(int fd, struct virtqueue *vq, bool timeout)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700879{
Rusty Russella1618832008-07-29 09:58:35 -0500880 unsigned int head, out, in, num = 0;
Rusty Russell17cbca22007-10-22 11:24:22 +1000881 int len;
882 struct iovec iov[vq->vring.num];
Rusty Russellaa124982008-07-29 09:58:36 -0500883 static int last_timeout_num;
Rusty Russell17cbca22007-10-22 11:24:22 +1000884
885 /* Keep getting output buffers from the Guest until we run out. */
886 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
887 if (in)
888 errx(1, "Input buffers in output queue?");
Rusty Russell398f1872008-07-29 09:58:37 -0500889 len = writev(vq->dev->fd, iov, out);
890 if (len < 0)
891 err(1, "Writing network packet to tun");
Rusty Russell17cbca22007-10-22 11:24:22 +1000892 add_used_and_trigger(fd, vq, head, len);
Rusty Russella1618832008-07-29 09:58:35 -0500893 num++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000894 }
Rusty Russella1618832008-07-29 09:58:35 -0500895
896 /* Block further kicks and set up a timer if we saw anything. */
897 if (!timeout && num)
898 block_vq(vq);
Rusty Russellaa124982008-07-29 09:58:36 -0500899
Rusty Russell1dc3e3b2008-08-26 00:19:27 -0500900 /* We never quite know how long should we wait before we check the
901 * queue again for more packets. We start at 500 microseconds, and if
902 * we get fewer packets than last time, we assume we made the timeout
903 * too small and increase it by 10 microseconds. Otherwise, we drop it
904 * by one microsecond every time. It seems to work well enough. */
Rusty Russellaa124982008-07-29 09:58:36 -0500905 if (timeout) {
906 if (num < last_timeout_num)
907 timeout_usec += 10;
908 else if (timeout_usec > 1)
909 timeout_usec--;
910 last_timeout_num = num;
911 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700912}
913
Rusty Russell17cbca22007-10-22 11:24:22 +1000914/* This is where we handle a packet coming in from the tun device to our
915 * Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700916static bool handle_tun_input(int fd, struct device *dev)
917{
Rusty Russell17cbca22007-10-22 11:24:22 +1000918 unsigned int head, in_num, out_num;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700919 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000920 struct iovec iov[dev->vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700921
Rusty Russell17cbca22007-10-22 11:24:22 +1000922 /* First we need a network buffer from the Guests's recv virtqueue. */
923 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
924 if (head == dev->vq->vring.num) {
Rusty Russelldde79782007-07-26 10:41:03 -0700925 /* Now, it's expected that if we try to send a packet too
Rusty Russell17cbca22007-10-22 11:24:22 +1000926 * early, the Guest won't be ready yet. Wait until the device
927 * status says it's ready. */
928 /* FIXME: Actually want DRIVER_ACTIVE here. */
Rusty Russell5dae7852008-07-29 09:58:35 -0500929
930 /* Now tell it we want to know if new things appear. */
931 dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
932 wmb();
933
Rusty Russell56ae43d2007-10-22 11:24:23 +1000934 /* We'll turn this back on if input buffers are registered. */
935 return false;
Rusty Russell17cbca22007-10-22 11:24:22 +1000936 } else if (out_num)
937 errx(1, "Output buffers in network recv queue?");
938
Rusty Russelldde79782007-07-26 10:41:03 -0700939 /* Read the packet from the device directly into the Guest's buffer. */
Rusty Russell398f1872008-07-29 09:58:37 -0500940 len = readv(dev->fd, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700941 if (len <= 0)
942 err(1, "reading network");
Rusty Russelldde79782007-07-26 10:41:03 -0700943
Rusty Russell56ae43d2007-10-22 11:24:23 +1000944 /* Tell the Guest about the new packet. */
Rusty Russell398f1872008-07-29 09:58:37 -0500945 add_used_and_trigger(fd, dev->vq, head, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000946
Rusty Russell8ca47e02007-07-19 01:49:29 -0700947 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
Rusty Russell17cbca22007-10-22 11:24:22 +1000948 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
949 head != dev->vq->vring.num ? "sent" : "discarded");
950
Rusty Russelldde79782007-07-26 10:41:03 -0700951 /* All good. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700952 return true;
953}
954
Rusty Russelle1e72962007-10-25 15:02:50 +1000955/*L:215 This is the callback attached to the network and console input
956 * virtqueues: it ensures we try again, in case we stopped console or net
Rusty Russell56ae43d2007-10-22 11:24:23 +1000957 * delivery because Guest didn't have any buffers. */
Rusty Russella1618832008-07-29 09:58:35 -0500958static void enable_fd(int fd, struct virtqueue *vq, bool timeout)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000959{
960 add_device_fd(vq->dev->fd);
Rusty Russell8c798732008-07-29 09:58:38 -0500961 /* Snap the Waker out of its select loop. */
962 write(waker_fds.pipe[1], "", 1);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000963}
964
Rusty Russella1618832008-07-29 09:58:35 -0500965static void net_enable_fd(int fd, struct virtqueue *vq, bool timeout)
Rusty Russell5dae7852008-07-29 09:58:35 -0500966{
967 /* We don't need to know again when Guest refills receive buffer. */
968 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russella1618832008-07-29 09:58:35 -0500969 enable_fd(fd, vq, timeout);
Rusty Russell5dae7852008-07-29 09:58:35 -0500970}
971
Rusty Russella007a752008-05-02 21:50:53 -0500972/* When the Guest tells us they updated the status field, we handle it. */
973static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500974{
975 struct virtqueue *vq;
976
Rusty Russella007a752008-05-02 21:50:53 -0500977 /* This is a reset. */
978 if (dev->desc->status == 0) {
979 verbose("Resetting device %s\n", dev->name);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500980
Rusty Russella007a752008-05-02 21:50:53 -0500981 /* Clear any features they've acked. */
982 memset(get_feature_bits(dev) + dev->desc->feature_len, 0,
983 dev->desc->feature_len);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500984
Rusty Russella007a752008-05-02 21:50:53 -0500985 /* Zero out the virtqueues. */
986 for (vq = dev->vq; vq; vq = vq->next) {
987 memset(vq->vring.desc, 0,
Rusty Russell2966af72008-12-30 09:25:58 -0600988 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
Rusty Russellb5111792008-07-29 09:58:34 -0500989 lg_last_avail(vq) = 0;
Rusty Russella007a752008-05-02 21:50:53 -0500990 }
991 } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
992 warnx("Device %s configuration FAILED", dev->name);
993 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
994 unsigned int i;
995
996 verbose("Device %s OK: offered", dev->name);
997 for (i = 0; i < dev->desc->feature_len; i++)
Rusty Russell32c68e52008-07-29 09:58:32 -0500998 verbose(" %02x", get_feature_bits(dev)[i]);
Rusty Russella007a752008-05-02 21:50:53 -0500999 verbose(", accepted");
1000 for (i = 0; i < dev->desc->feature_len; i++)
Rusty Russell32c68e52008-07-29 09:58:32 -05001001 verbose(" %02x", get_feature_bits(dev)
Rusty Russella007a752008-05-02 21:50:53 -05001002 [dev->desc->feature_len+i]);
1003
1004 if (dev->ready)
1005 dev->ready(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001006 }
1007}
1008
Rusty Russell17cbca22007-10-22 11:24:22 +10001009/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1010static void handle_output(int fd, unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001011{
1012 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +10001013 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001014
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001015 /* Check each device and virtqueue. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001016 for (i = devices.dev; i; i = i->next) {
Rusty Russella007a752008-05-02 21:50:53 -05001017 /* Notifications to device descriptors update device status. */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001018 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001019 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001020 return;
1021 }
1022
1023 /* Notifications to virtqueues mean output has occurred. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001024 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001025 if (vq->config.pfn != addr/getpagesize())
1026 continue;
1027
1028 /* Guest should acknowledge (and set features!) before
1029 * using the device. */
1030 if (i->desc->status == 0) {
1031 warnx("%s gave early output", i->name);
Rusty Russell17cbca22007-10-22 11:24:22 +10001032 return;
1033 }
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001034
1035 if (strcmp(vq->dev->name, "console") != 0)
1036 verbose("Output to %s\n", vq->dev->name);
1037 if (vq->handle_output)
Rusty Russella1618832008-07-29 09:58:35 -05001038 vq->handle_output(fd, vq, false);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001039 return;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001040 }
1041 }
Rusty Russelldde79782007-07-26 10:41:03 -07001042
Rusty Russell17cbca22007-10-22 11:24:22 +10001043 /* Early console write is done using notify on a nul-terminated string
1044 * in Guest memory. */
1045 if (addr >= guest_limit)
1046 errx(1, "Bad NOTIFY %#lx", addr);
1047
1048 write(STDOUT_FILENO, from_guest_phys(addr),
1049 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001050}
1051
Rusty Russella1618832008-07-29 09:58:35 -05001052static void handle_timeout(int fd)
1053{
1054 char buf[32];
1055 struct device *i;
1056 struct virtqueue *vq;
1057
1058 /* Clear the pipe */
1059 read(timeoutpipe[0], buf, sizeof(buf));
1060
1061 /* Check each device and virtqueue: flush blocked ones. */
1062 for (i = devices.dev; i; i = i->next) {
1063 for (vq = i->vq; vq; vq = vq->next) {
1064 if (!vq->blocked)
1065 continue;
1066
1067 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
1068 vq->blocked = false;
1069 if (vq->handle_output)
1070 vq->handle_output(fd, vq, true);
1071 }
1072 }
1073}
1074
Rusty Russelle1e72962007-10-25 15:02:50 +10001075/* This is called when the Waker wakes us up: check for incoming file
Rusty Russelldde79782007-07-26 10:41:03 -07001076 * descriptors. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001077static void handle_input(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001078{
Rusty Russelldde79782007-07-26 10:41:03 -07001079 /* select() wants a zeroed timeval to mean "don't wait". */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001080 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
1081
1082 for (;;) {
1083 struct device *i;
Rusty Russell17cbca22007-10-22 11:24:22 +10001084 fd_set fds = devices.infds;
Rusty Russella1618832008-07-29 09:58:35 -05001085 int num;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001086
Rusty Russella1618832008-07-29 09:58:35 -05001087 num = select(devices.max_infd+1, &fds, NULL, NULL, &poll);
1088 /* Could get interrupted */
1089 if (num < 0)
1090 continue;
Rusty Russelldde79782007-07-26 10:41:03 -07001091 /* If nothing is ready, we're done. */
Rusty Russella1618832008-07-29 09:58:35 -05001092 if (num == 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001093 break;
1094
Rusty Russella6bd8e12008-03-28 11:05:53 -05001095 /* Otherwise, call the device(s) which have readable file
1096 * descriptors and a method of handling them. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001097 for (i = devices.dev; i; i = i->next) {
Rusty Russell8ca47e02007-07-19 01:49:29 -07001098 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
Rusty Russell56ae43d2007-10-22 11:24:23 +10001099 if (i->handle_input(fd, i))
1100 continue;
1101
Rusty Russelldde79782007-07-26 10:41:03 -07001102 /* If handle_input() returns false, it means we
Rusty Russell56ae43d2007-10-22 11:24:23 +10001103 * should no longer service it. Networking and
1104 * console do this when there's no input
1105 * buffers to deliver into. Console also uses
Rusty Russella6bd8e12008-03-28 11:05:53 -05001106 * it when it discovers that stdin is closed. */
Rusty Russell56ae43d2007-10-22 11:24:23 +10001107 FD_CLR(i->fd, &devices.infds);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001108 }
1109 }
Rusty Russella1618832008-07-29 09:58:35 -05001110
1111 /* Is this the timeout fd? */
1112 if (FD_ISSET(timeoutpipe[0], &fds))
1113 handle_timeout(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001114 }
1115}
1116
Rusty Russelldde79782007-07-26 10:41:03 -07001117/*L:190
1118 * Device Setup
1119 *
1120 * All devices need a descriptor so the Guest knows it exists, and a "struct
1121 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001122 * routines to allocate and manage them.
1123 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001124
1125/* The layout of the device page is a "struct lguest_device_desc" followed by a
1126 * number of virtqueue descriptors, then two sets of feature bits, then an
1127 * array of configuration bytes. This routine returns the configuration
1128 * pointer. */
1129static u8 *device_config(const struct device *dev)
1130{
1131 return (void *)(dev->desc + 1)
1132 + dev->desc->num_vq * sizeof(struct lguest_vqconfig)
1133 + dev->desc->feature_len * 2;
1134}
1135
1136/* This routine allocates a new "struct lguest_device_desc" from descriptor
1137 * table page just above the Guest's normal memory. It returns a pointer to
1138 * that descriptor. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001139static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001140{
Rusty Russella586d4f2008-02-04 23:49:56 -05001141 struct lguest_device_desc d = { .type = type };
1142 void *p;
1143
1144 /* Figure out where the next device config is, based on the last one. */
1145 if (devices.lastdev)
1146 p = device_config(devices.lastdev)
1147 + devices.lastdev->desc->config_len;
1148 else
1149 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001150
Rusty Russell17cbca22007-10-22 11:24:22 +10001151 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001152 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001153 errx(1, "Too many devices");
1154
Rusty Russella586d4f2008-02-04 23:49:56 -05001155 /* p might not be aligned, so we memcpy in. */
1156 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001157}
1158
Rusty Russella586d4f2008-02-04 23:49:56 -05001159/* Each device descriptor is followed by the description of its virtqueues. We
1160 * specify how many descriptors the virtqueue is to have. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001161static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russella1618832008-07-29 09:58:35 -05001162 void (*handle_output)(int, struct virtqueue *, bool))
Rusty Russell17cbca22007-10-22 11:24:22 +10001163{
1164 unsigned int pages;
1165 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1166 void *p;
1167
Rusty Russella6bd8e12008-03-28 11:05:53 -05001168 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001169 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001170 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001171 p = get_pages(pages);
1172
Rusty Russelld1c856e2007-11-19 11:20:40 -05001173 /* Initialize the virtqueue */
1174 vq->next = NULL;
1175 vq->last_avail_idx = 0;
1176 vq->dev = dev;
Rusty Russell20887612008-05-30 15:09:46 -05001177 vq->inflight = 0;
Rusty Russella1618832008-07-29 09:58:35 -05001178 vq->blocked = false;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001179
Rusty Russell17cbca22007-10-22 11:24:22 +10001180 /* Initialize the configuration. */
1181 vq->config.num = num_descs;
1182 vq->config.irq = devices.next_irq++;
1183 vq->config.pfn = to_guest_phys(p) / getpagesize();
1184
1185 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001186 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001187
Rusty Russella586d4f2008-02-04 23:49:56 -05001188 /* Append virtqueue to this device's descriptor. We use
1189 * device_config() to get the end of the device's current virtqueues;
1190 * we check that we haven't added any config or feature information
1191 * yet, otherwise we'd be overwriting them. */
1192 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1193 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1194 dev->desc->num_vq++;
1195
1196 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001197
1198 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1199 * second. */
1200 for (i = &dev->vq; *i; i = &(*i)->next);
1201 *i = vq;
1202
Rusty Russelle1e72962007-10-25 15:02:50 +10001203 /* Set the routine to call when the Guest does something to this
1204 * virtqueue. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001205 vq->handle_output = handle_output;
Rusty Russelle1e72962007-10-25 15:02:50 +10001206
Rusty Russell426e3e02008-02-04 23:49:59 -05001207 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1208 * don't have a handler */
Rusty Russell17cbca22007-10-22 11:24:22 +10001209 if (!handle_output)
1210 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1211}
1212
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001213/* The first half of the feature bitmask is for us to advertise features. The
Rusty Russella6bd8e12008-03-28 11:05:53 -05001214 * second half is for the Guest to accept features. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001215static void add_feature(struct device *dev, unsigned bit)
1216{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001217 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001218
1219 /* We can't extend the feature bits once we've added config bytes */
1220 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1221 assert(dev->desc->config_len == 0);
1222 dev->desc->feature_len = (bit / CHAR_BIT) + 1;
1223 }
1224
Rusty Russella586d4f2008-02-04 23:49:56 -05001225 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1226}
1227
1228/* This routine sets the configuration fields for an existing device's
1229 * descriptor. It only works for the last device, but that's OK because that's
1230 * how we use it. */
1231static void set_config(struct device *dev, unsigned len, const void *conf)
1232{
1233 /* Check we haven't overflowed our single page. */
1234 if (device_config(dev) + len > devices.descpage + getpagesize())
1235 errx(1, "Too many devices");
1236
1237 /* Copy in the config information, and store the length. */
1238 memcpy(device_config(dev), conf, len);
1239 dev->desc->config_len = len;
1240}
1241
Rusty Russell17cbca22007-10-22 11:24:22 +10001242/* This routine does all the creation and setup of a new device, including
Rusty Russella6bd8e12008-03-28 11:05:53 -05001243 * calling new_dev_desc() to allocate the descriptor and device memory.
1244 *
1245 * See what I mean about userspace being boring? */
Rusty Russell17cbca22007-10-22 11:24:22 +10001246static struct device *new_device(const char *name, u16 type, int fd,
1247 bool (*handle_input)(int, struct device *))
Rusty Russell8ca47e02007-07-19 01:49:29 -07001248{
1249 struct device *dev = malloc(sizeof(*dev));
1250
Rusty Russelldde79782007-07-26 10:41:03 -07001251 /* Now we populate the fields one at a time. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001252 dev->fd = fd;
Rusty Russelldde79782007-07-26 10:41:03 -07001253 /* If we have an input handler for this file descriptor, then we add it
1254 * to the device_list's fdset and maxfd. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001255 if (handle_input)
Rusty Russell17cbca22007-10-22 11:24:22 +10001256 add_device_fd(dev->fd);
1257 dev->desc = new_dev_desc(type);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001258 dev->handle_input = handle_input;
Rusty Russell17cbca22007-10-22 11:24:22 +10001259 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001260 dev->vq = NULL;
Rusty Russella007a752008-05-02 21:50:53 -05001261 dev->ready = NULL;
Rusty Russella586d4f2008-02-04 23:49:56 -05001262
1263 /* Append to device list. Prepending to a single-linked list is
1264 * easier, but the user expects the devices to be arranged on the bus
1265 * in command-line order. The first network device on the command line
1266 * is eth0, the first block device /dev/vda, etc. */
1267 if (devices.lastdev)
1268 devices.lastdev->next = dev;
1269 else
1270 devices.dev = dev;
1271 devices.lastdev = dev;
1272
Rusty Russell8ca47e02007-07-19 01:49:29 -07001273 return dev;
1274}
1275
Rusty Russelldde79782007-07-26 10:41:03 -07001276/* Our first setup routine is the console. It's a fairly simple device, but
1277 * UNIX tty handling makes it uglier than it could be. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001278static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001279{
1280 struct device *dev;
1281
Rusty Russelldde79782007-07-26 10:41:03 -07001282 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001283 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1284 struct termios term = orig_term;
Rusty Russelldde79782007-07-26 10:41:03 -07001285 /* Then we turn off echo, line buffering and ^C etc. We want a
1286 * raw input stream to the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001287 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1288 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russelldde79782007-07-26 10:41:03 -07001289 /* If we exit gracefully, the original settings will be
1290 * restored so the user can see what they're typing. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001291 atexit(restore_term);
1292 }
1293
Rusty Russell17cbca22007-10-22 11:24:22 +10001294 dev = new_device("console", VIRTIO_ID_CONSOLE,
1295 STDIN_FILENO, handle_console_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001296 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001297 dev->priv = malloc(sizeof(struct console_abort));
1298 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001299
Rusty Russell56ae43d2007-10-22 11:24:23 +10001300 /* The console needs two virtqueues: the input then the output. When
1301 * they put something the input queue, we make sure we're listening to
1302 * stdin. When they put something in the output queue, we write it to
Rusty Russelle1e72962007-10-25 15:02:50 +10001303 * stdout. */
Rusty Russell56ae43d2007-10-22 11:24:23 +10001304 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001305 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001306
Rusty Russell17cbca22007-10-22 11:24:22 +10001307 verbose("device %u: console\n", devices.device_num++);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001308}
Rusty Russelldde79782007-07-26 10:41:03 -07001309/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001310
Rusty Russella1618832008-07-29 09:58:35 -05001311static void timeout_alarm(int sig)
1312{
1313 write(timeoutpipe[1], "", 1);
1314}
1315
1316static void setup_timeout(void)
1317{
1318 if (pipe(timeoutpipe) != 0)
1319 err(1, "Creating timeout pipe");
1320
1321 if (fcntl(timeoutpipe[1], F_SETFL,
1322 fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0)
1323 err(1, "Making timeout pipe nonblocking");
1324
1325 add_device_fd(timeoutpipe[0]);
1326 signal(SIGALRM, timeout_alarm);
1327}
1328
Rusty Russell17cbca22007-10-22 11:24:22 +10001329/*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1330 * --sharenet=<name> option which opens or creates a named pipe. This can be
1331 * used to send packets to another guest in a 1:1 manner.
1332 *
1333 * More sopisticated is to use one of the tools developed for project like UML
1334 * to do networking.
1335 *
1336 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1337 * completely generic ("here's my vring, attach to your vring") and would work
1338 * for any traffic. Of course, namespace and permissions issues need to be
1339 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1340 * multiple inter-guest channels behind one interface, although it would
1341 * require some manner of hotplugging new virtio channels.
1342 *
1343 * Finally, we could implement a virtio network switch in the kernel. :*/
1344
Rusty Russell8ca47e02007-07-19 01:49:29 -07001345static u32 str2ip(const char *ipaddr)
1346{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001347 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001348
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001349 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1350 errx(1, "Failed to parse IP address '%s'", ipaddr);
1351 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1352}
1353
1354static void str2mac(const char *macaddr, unsigned char mac[6])
1355{
1356 unsigned int m[6];
1357 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1358 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1359 errx(1, "Failed to parse mac address '%s'", macaddr);
1360 mac[0] = m[0];
1361 mac[1] = m[1];
1362 mac[2] = m[2];
1363 mac[3] = m[3];
1364 mac[4] = m[4];
1365 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001366}
1367
Rusty Russelldde79782007-07-26 10:41:03 -07001368/* This code is "adapted" from libbridge: it attaches the Host end of the
1369 * network device to the bridge device specified by the command line.
1370 *
1371 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1372 * dislike bridging), and I just try not to break it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001373static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1374{
1375 int ifidx;
1376 struct ifreq ifr;
1377
1378 if (!*br_name)
1379 errx(1, "must specify bridge name");
1380
1381 ifidx = if_nametoindex(if_name);
1382 if (!ifidx)
1383 errx(1, "interface %s does not exist!", if_name);
1384
1385 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001386 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001387 ifr.ifr_ifindex = ifidx;
1388 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1389 err(1, "can't add %s to bridge %s", if_name, br_name);
1390}
1391
Rusty Russelldde79782007-07-26 10:41:03 -07001392/* This sets up the Host end of the network device with an IP address, brings
1393 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell17cbca22007-10-22 11:24:22 +10001394 * pointer. */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001395static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001396{
1397 struct ifreq ifr;
1398 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1399
1400 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001401 strcpy(ifr.ifr_name, tapif);
1402
1403 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001404 sin->sin_family = AF_INET;
1405 sin->sin_addr.s_addr = htonl(ipaddr);
1406 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001407 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001408 ifr.ifr_flags = IFF_UP;
1409 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001410 err(1, "Bringing interface %s up", tapif);
1411}
1412
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001413static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001414{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001415 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001416 int netfd;
1417
1418 /* Start with this zeroed. Messy but sure. */
1419 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001420
Rusty Russelldde79782007-07-26 10:41:03 -07001421 /* We open the /dev/net/tun device and tell it we want a tap device. A
1422 * tap device is like a tun device, only somehow different. To tell
1423 * the truth, I completely blundered my way through this code, but it
1424 * works now! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001425 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001426 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001427 strcpy(ifr.ifr_name, "tap%d");
1428 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1429 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001430
Rusty Russell398f1872008-07-29 09:58:37 -05001431 if (ioctl(netfd, TUNSETOFFLOAD,
1432 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1433 err(1, "Could not set features for tun device");
1434
Rusty Russelldde79782007-07-26 10:41:03 -07001435 /* We don't need checksums calculated for packets coming in this
1436 * device: trust us! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001437 ioctl(netfd, TUNSETNOCSUM, 1);
1438
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001439 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1440 return netfd;
1441}
1442
1443/*L:195 Our network is a Host<->Guest network. This can either use bridging or
1444 * routing, but the principle is the same: it uses the "tun" device to inject
1445 * packets into the Host as if they came in from a normal network card. We
1446 * just shunt packets between the Guest and the tun device. */
1447static void setup_tun_net(char *arg)
1448{
1449 struct device *dev;
1450 int netfd, ipfd;
1451 u32 ip = INADDR_ANY;
1452 bool bridging = false;
1453 char tapif[IFNAMSIZ], *p;
1454 struct virtio_net_config conf;
1455
1456 netfd = get_tun_device(tapif);
1457
Rusty Russell17cbca22007-10-22 11:24:22 +10001458 /* First we create a new network device. */
1459 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
Rusty Russelldde79782007-07-26 10:41:03 -07001460
Rusty Russell56ae43d2007-10-22 11:24:23 +10001461 /* Network devices need a receive and a send queue, just like
1462 * console. */
Rusty Russell5dae7852008-07-29 09:58:35 -05001463 add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd);
Rusty Russell17cbca22007-10-22 11:24:22 +10001464 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001465
Rusty Russelldde79782007-07-26 10:41:03 -07001466 /* We need a socket to perform the magic network ioctls to bring up the
1467 * tap interface, connect to the bridge etc. Any socket will do! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001468 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1469 if (ipfd < 0)
1470 err(1, "opening IP socket");
1471
Rusty Russelldde79782007-07-26 10:41:03 -07001472 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001473 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001474 arg += strlen(BRIDGE_PFX);
1475 bridging = true;
1476 }
1477
1478 /* A mac address may follow the bridge name or IP address */
1479 p = strchr(arg, ':');
1480 if (p) {
1481 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001482 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001483 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001484 }
1485
1486 /* arg is now either an IP address or a bridge name */
1487 if (bridging)
1488 add_to_bridge(ipfd, tapif, arg);
1489 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001490 ip = str2ip(arg);
1491
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001492 /* Set up the tun device. */
1493 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001494
Rusty Russell20887612008-05-30 15:09:46 -05001495 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
Rusty Russell398f1872008-07-29 09:58:37 -05001496 /* Expect Guest to handle everything except UFO */
1497 add_feature(dev, VIRTIO_NET_F_CSUM);
1498 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001499 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1500 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1501 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1502 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1503 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1504 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Rusty Russella586d4f2008-02-04 23:49:56 -05001505 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001506
Rusty Russella586d4f2008-02-04 23:49:56 -05001507 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001508 close(ipfd);
1509
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001510 devices.device_num++;
1511
1512 if (bridging)
1513 verbose("device %u: tun %s attached to bridge: %s\n",
1514 devices.device_num, tapif, arg);
1515 else
1516 verbose("device %u: tun %s: %s\n",
1517 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001518}
Rusty Russell17cbca22007-10-22 11:24:22 +10001519
Rusty Russelle1e72962007-10-25 15:02:50 +10001520/* Our block (disk) device should be really simple: the Guest asks for a block
1521 * number and we read or write that position in the file. Unfortunately, that
1522 * was amazingly slow: the Guest waits until the read is finished before
1523 * running anything else, even if it could have been doing useful work.
Rusty Russell17cbca22007-10-22 11:24:22 +10001524 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001525 * We could use async I/O, except it's reputed to suck so hard that characters
1526 * actually go missing from your code when you try to use it.
Rusty Russell17cbca22007-10-22 11:24:22 +10001527 *
1528 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1529
Rusty Russelle1e72962007-10-25 15:02:50 +10001530/* This hangs off device->priv. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001531struct vblk_info
1532{
1533 /* The size of the file. */
1534 off64_t len;
1535
1536 /* The file descriptor for the file. */
1537 int fd;
1538
1539 /* IO thread listens on this file descriptor [0]. */
1540 int workpipe[2];
1541
1542 /* IO thread writes to this file descriptor to mark it done, then
1543 * Launcher triggers interrupt to Guest. */
1544 int done_fd;
1545};
1546
Rusty Russelle1e72962007-10-25 15:02:50 +10001547/*L:210
1548 * The Disk
1549 *
1550 * Remember that the block device is handled by a separate I/O thread. We head
1551 * straight into the core of that thread here:
1552 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001553static bool service_io(struct device *dev)
1554{
1555 struct vblk_info *vblk = dev->priv;
1556 unsigned int head, out_num, in_num, wlen;
1557 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001558 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001559 struct virtio_blk_outhdr *out;
1560 struct iovec iov[dev->vq->vring.num];
1561 off64_t off;
1562
Rusty Russelle1e72962007-10-25 15:02:50 +10001563 /* See if there's a request waiting. If not, nothing to do. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001564 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1565 if (head == dev->vq->vring.num)
1566 return false;
1567
Rusty Russelle1e72962007-10-25 15:02:50 +10001568 /* Every block request should contain at least one output buffer
1569 * (detailing the location on disk and the type of request) and one
1570 * input buffer (to hold the result). */
Rusty Russell17cbca22007-10-22 11:24:22 +10001571 if (out_num == 0 || in_num == 0)
1572 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1573 head, out_num, in_num);
1574
1575 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001576 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russell17cbca22007-10-22 11:24:22 +10001577 off = out->sector * 512;
1578
Rusty Russelle1e72962007-10-25 15:02:50 +10001579 /* The block device implements "barriers", where the Guest indicates
1580 * that it wants all previous writes to occur before this write. We
1581 * don't have a way of asking our kernel to do a barrier, so we just
1582 * synchronize all the data in the file. Pretty poor, no? */
Rusty Russell17cbca22007-10-22 11:24:22 +10001583 if (out->type & VIRTIO_BLK_T_BARRIER)
1584 fdatasync(vblk->fd);
1585
Rusty Russelle1e72962007-10-25 15:02:50 +10001586 /* In general the virtio block driver is allowed to try SCSI commands.
1587 * It'd be nice if we supported eject, for example, but we don't. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001588 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1589 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001590 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001591 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001592 } else if (out->type & VIRTIO_BLK_T_OUT) {
1593 /* Write */
1594
1595 /* Move to the right location in the block file. This can fail
1596 * if they try to write past end. */
1597 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1598 err(1, "Bad seek to sector %llu", out->sector);
1599
1600 ret = writev(vblk->fd, iov+1, out_num-1);
1601 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1602
1603 /* Grr... Now we know how long the descriptor they sent was, we
1604 * make sure they didn't try to write over the end of the block
1605 * file (possibly extending it). */
1606 if (ret > 0 && off + ret > vblk->len) {
1607 /* Trim it back to the correct length */
1608 ftruncate64(vblk->fd, vblk->len);
1609 /* Die, bad Guest, die. */
1610 errx(1, "Write past end %llu+%u", off, ret);
1611 }
Anthony Liguori1200e642007-11-08 21:13:44 -06001612 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001613 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001614 } else {
1615 /* Read */
1616
1617 /* Move to the right location in the block file. This can fail
1618 * if they try to read past end. */
1619 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1620 err(1, "Bad seek to sector %llu", out->sector);
1621
1622 ret = readv(vblk->fd, iov+1, in_num-1);
1623 verbose("READ from sector %llu: %i\n", out->sector, ret);
1624 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001625 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001626 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001627 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001628 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001629 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001630 }
1631 }
1632
1633 /* We can't trigger an IRQ, because we're not the Launcher. It does
1634 * that when we tell it we're done. */
1635 add_used(dev->vq, head, wlen);
1636 return true;
1637}
1638
1639/* This is the thread which actually services the I/O. */
1640static int io_thread(void *_dev)
1641{
1642 struct device *dev = _dev;
1643 struct vblk_info *vblk = dev->priv;
1644 char c;
1645
1646 /* Close other side of workpipe so we get 0 read when main dies. */
1647 close(vblk->workpipe[1]);
1648 /* Close the other side of the done_fd pipe. */
1649 close(dev->fd);
1650
1651 /* When this read fails, it means Launcher died, so we follow. */
1652 while (read(vblk->workpipe[0], &c, 1) == 1) {
Rusty Russelle1e72962007-10-25 15:02:50 +10001653 /* We acknowledge each request immediately to reduce latency,
Rusty Russell17cbca22007-10-22 11:24:22 +10001654 * rather than waiting until we've done them all. I haven't
Rusty Russella6bd8e12008-03-28 11:05:53 -05001655 * measured to see if it makes any difference.
1656 *
1657 * That would be an interesting test, wouldn't it? You could
1658 * also try having more than one I/O thread. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001659 while (service_io(dev))
1660 write(vblk->done_fd, &c, 1);
1661 }
1662 return 0;
1663}
1664
Rusty Russelle1e72962007-10-25 15:02:50 +10001665/* Now we've seen the I/O thread, we return to the Launcher to see what happens
Rusty Russella6bd8e12008-03-28 11:05:53 -05001666 * when that thread tells us it's completed some I/O. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001667static bool handle_io_finish(int fd, struct device *dev)
1668{
1669 char c;
1670
Rusty Russelle1e72962007-10-25 15:02:50 +10001671 /* If the I/O thread died, presumably it printed the error, so we
1672 * simply exit. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001673 if (read(dev->fd, &c, 1) != 1)
1674 exit(1);
1675
1676 /* It did some work, so trigger the irq. */
1677 trigger_irq(fd, dev->vq);
1678 return true;
1679}
1680
Rusty Russelle1e72962007-10-25 15:02:50 +10001681/* When the Guest submits some I/O, we just need to wake the I/O thread. */
Rusty Russella1618832008-07-29 09:58:35 -05001682static void handle_virtblk_output(int fd, struct virtqueue *vq, bool timeout)
Rusty Russell17cbca22007-10-22 11:24:22 +10001683{
1684 struct vblk_info *vblk = vq->dev->priv;
1685 char c = 0;
1686
1687 /* Wake up I/O thread and tell it to go to work! */
1688 if (write(vblk->workpipe[1], &c, 1) != 1)
1689 /* Presumably it indicated why it died. */
1690 exit(1);
1691}
1692
Rusty Russelle1e72962007-10-25 15:02:50 +10001693/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001694static void setup_block_file(const char *filename)
1695{
1696 int p[2];
1697 struct device *dev;
1698 struct vblk_info *vblk;
1699 void *stack;
Rusty Russella586d4f2008-02-04 23:49:56 -05001700 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001701
1702 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1703 pipe(p);
1704
1705 /* The device responds to return from I/O thread. */
1706 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1707
Rusty Russelle1e72962007-10-25 15:02:50 +10001708 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001709 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1710
1711 /* Allocate the room for our own bookkeeping */
1712 vblk = dev->priv = malloc(sizeof(*vblk));
1713
1714 /* First we open the file and store the length. */
1715 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1716 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1717
Rusty Russella586d4f2008-02-04 23:49:56 -05001718 /* We support barriers. */
1719 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1720
Rusty Russell17cbca22007-10-22 11:24:22 +10001721 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001722 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001723
1724 /* Tell Guest not to put in too many descriptors at once: two are used
1725 * for the in and out elements. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001726 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1727 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1728
1729 set_config(dev, sizeof(conf), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001730
1731 /* The I/O thread writes to this end of the pipe when done. */
1732 vblk->done_fd = p[1];
1733
Rusty Russelle1e72962007-10-25 15:02:50 +10001734 /* This is the second pipe, which is how we tell the I/O thread about
1735 * more work. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001736 pipe(vblk->workpipe);
1737
Rusty Russella6bd8e12008-03-28 11:05:53 -05001738 /* Create stack for thread and run it. Since stack grows upwards, we
1739 * point the stack pointer to the end of this region. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001740 stack = malloc(32768);
Balaji Raoec04b132007-12-28 14:26:24 +05301741 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1742 * becoming a zombie. */
Rusty Russella6bd8e12008-03-28 11:05:53 -05001743 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
Rusty Russell17cbca22007-10-22 11:24:22 +10001744 err(1, "Creating clone");
1745
1746 /* We don't need to keep the I/O thread's end of the pipes open. */
1747 close(vblk->done_fd);
1748 close(vblk->workpipe[0]);
1749
1750 verbose("device %u: virtblock %llu sectors\n",
Rusty Russella586d4f2008-02-04 23:49:56 -05001751 devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001752}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001753
1754/* Our random number generator device reads from /dev/random into the Guest's
1755 * input buffers. The usual case is that the Guest doesn't want random numbers
1756 * and so has no buffers although /dev/random is still readable, whereas
1757 * console is the reverse.
1758 *
1759 * The same logic applies, however. */
1760static bool handle_rng_input(int fd, struct device *dev)
1761{
1762 int len;
1763 unsigned int head, in_num, out_num, totlen = 0;
1764 struct iovec iov[dev->vq->vring.num];
1765
1766 /* First we need a buffer from the Guests's virtqueue. */
1767 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1768
1769 /* If they're not ready for input, stop listening to this file
1770 * descriptor. We'll start again once they add an input buffer. */
1771 if (head == dev->vq->vring.num)
1772 return false;
1773
1774 if (out_num)
1775 errx(1, "Output buffers in rng?");
1776
1777 /* This is why we convert to iovecs: the readv() call uses them, and so
1778 * it reads straight into the Guest's buffer. We loop to make sure we
1779 * fill it. */
1780 while (!iov_empty(iov, in_num)) {
1781 len = readv(dev->fd, iov, in_num);
1782 if (len <= 0)
1783 err(1, "Read from /dev/random gave %i", len);
1784 iov_consume(iov, in_num, len);
1785 totlen += len;
1786 }
1787
1788 /* Tell the Guest about the new input. */
1789 add_used_and_trigger(fd, dev->vq, head, totlen);
1790
1791 /* Everything went OK! */
1792 return true;
1793}
1794
1795/* And this creates a "hardware" random number device for the Guest. */
1796static void setup_rng(void)
1797{
1798 struct device *dev;
1799 int fd;
1800
1801 fd = open_or_die("/dev/random", O_RDONLY);
1802
1803 /* The device responds to return from I/O thread. */
1804 dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input);
1805
1806 /* The device has one virtqueue, where the Guest places inbufs. */
1807 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1808
1809 verbose("device %u: rng\n", devices.device_num++);
1810}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001811/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301812
Rusty Russella6bd8e12008-03-28 11:05:53 -05001813/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301814static void __attribute__((noreturn)) restart_guest(void)
1815{
1816 unsigned int i;
1817
Rusty Russell8c798732008-07-29 09:58:38 -05001818 /* Since we don't track all open fds, we simply close everything beyond
1819 * stderr. */
Balaji Raoec04b132007-12-28 14:26:24 +05301820 for (i = 3; i < FD_SETSIZE; i++)
1821 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001822
1823 /* The exec automatically gets rid of the I/O and Waker threads. */
Balaji Raoec04b132007-12-28 14:26:24 +05301824 execv(main_args[0], main_args);
1825 err(1, "Could not exec %s", main_args[0]);
1826}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001827
Rusty Russella6bd8e12008-03-28 11:05:53 -05001828/*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
Rusty Russelldde79782007-07-26 10:41:03 -07001829 * its input and output, and finally, lays it to rest. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001830static void __attribute__((noreturn)) run_guest(int lguest_fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001831{
1832 for (;;) {
Jes Sorensen511801d2007-10-22 11:03:31 +10001833 unsigned long args[] = { LHREQ_BREAK, 0 };
Rusty Russell17cbca22007-10-22 11:24:22 +10001834 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001835 int readval;
1836
1837 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001838 readval = pread(lguest_fd, &notify_addr,
1839 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001840
Rusty Russell17cbca22007-10-22 11:24:22 +10001841 /* One unsigned long means the Guest did HCALL_NOTIFY */
1842 if (readval == sizeof(notify_addr)) {
1843 verbose("Notify on address %#lx\n", notify_addr);
1844 handle_output(lguest_fd, notify_addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001845 continue;
Rusty Russelldde79782007-07-26 10:41:03 -07001846 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001847 } else if (errno == ENOENT) {
1848 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001849 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001850 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301851 /* ERESTART means that we need to reboot the guest */
1852 } else if (errno == ERESTART) {
1853 restart_guest();
Rusty Russella1618832008-07-29 09:58:35 -05001854 /* EAGAIN means a signal (timeout).
Rusty Russelldde79782007-07-26 10:41:03 -07001855 * Anything else means a bug or incompatible change. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001856 } else if (errno != EAGAIN)
1857 err(1, "Running guest failed");
Rusty Russelldde79782007-07-26 10:41:03 -07001858
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001859 /* Only service input on thread for CPU 0. */
1860 if (cpu_id != 0)
1861 continue;
1862
Rusty Russelle1e72962007-10-25 15:02:50 +10001863 /* Service input, then unset the BREAK to release the Waker. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001864 handle_input(lguest_fd);
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001865 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001866 err(1, "Resetting break");
1867 }
1868}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001869/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001870 * This is the end of the Launcher. The good news: we are over halfway
1871 * through! The bad news: the most fiendish part of the code still lies ahead
1872 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001873 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001874 * Are you ready? Take a deep breath and join me in the core of the Host, in
1875 * "make Host".
1876 :*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001877
1878static struct option opts[] = {
1879 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001880 { "tunnet", 1, NULL, 't' },
1881 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001882 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001883 { "initrd", 1, NULL, 'i' },
1884 { NULL },
1885};
1886static void usage(void)
1887{
1888 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001889 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001890 "|--block=<filename>|--initrd=<filename>]...\n"
1891 "<mem-in-mb> vmlinux [args...]");
1892}
1893
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001894/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001895int main(int argc, char *argv[])
1896{
Rusty Russell47436aa2007-10-22 11:03:36 +10001897 /* Memory, top-level pagetable, code startpoint and size of the
1898 * (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001899 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russelle1e72962007-10-25 15:02:50 +10001900 /* Two temporaries and the /dev/lguest file descriptor. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001901 int i, c, lguest_fd;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001902 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001903 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001904 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001905 const char *initrd_name = NULL;
1906
Balaji Raoec04b132007-12-28 14:26:24 +05301907 /* Save the args: we "reboot" by execing ourselves again. */
1908 main_args = argv;
1909 /* We don't "wait" for the children, so prevent them from becoming
1910 * zombies. */
1911 signal(SIGCHLD, SIG_IGN);
1912
Rusty Russelldde79782007-07-26 10:41:03 -07001913 /* First we initialize the device list. Since console and network
1914 * device receive input from a file descriptor, we keep an fdset
1915 * (infds) and the maximum fd number (max_infd) with the head of the
Rusty Russella586d4f2008-02-04 23:49:56 -05001916 * list. We also keep a pointer to the last device. Finally, we keep
Rusty Russella6bd8e12008-03-28 11:05:53 -05001917 * the next interrupt number to use for devices (1: remember that 0 is
1918 * used by the timer). */
Rusty Russell17cbca22007-10-22 11:24:22 +10001919 FD_ZERO(&devices.infds);
1920 devices.max_infd = -1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001921 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001922 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001923
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001924 cpu_id = 0;
Rusty Russelldde79782007-07-26 10:41:03 -07001925 /* We need to know how much memory so we can set up the device
1926 * descriptor and memory pages for the devices as we parse the command
1927 * line. So we quickly look through the arguments to find the amount
1928 * of memory now. */
Rusty Russell6570c45992007-07-23 18:43:56 -07001929 for (i = 1; i < argc; i++) {
1930 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001931 mem = atoi(argv[i]) * 1024 * 1024;
1932 /* We start by mapping anonymous pages over all of
1933 * guest-physical memory range. This fills it with 0,
1934 * and ensures that the Guest won't be killed when it
1935 * tries to access it. */
1936 guest_base = map_zeroed_pages(mem / getpagesize()
1937 + DEVICE_PAGES);
1938 guest_limit = mem;
1939 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001940 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001941 break;
1942 }
1943 }
Rusty Russelldde79782007-07-26 10:41:03 -07001944
1945 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001946 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1947 switch (c) {
1948 case 'v':
1949 verbose = true;
1950 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001951 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001952 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001953 break;
1954 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001955 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001956 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001957 case 'r':
1958 setup_rng();
1959 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001960 case 'i':
1961 initrd_name = optarg;
1962 break;
1963 default:
1964 warnx("Unknown argument %s", argv[optind]);
1965 usage();
1966 }
1967 }
Rusty Russelldde79782007-07-26 10:41:03 -07001968 /* After the other arguments we expect memory and kernel image name,
1969 * followed by command line arguments for the kernel. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001970 if (optind + 2 > argc)
1971 usage();
1972
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001973 verbose("Guest base is at %p\n", guest_base);
1974
Rusty Russelldde79782007-07-26 10:41:03 -07001975 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001976 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001977
Rusty Russella1618832008-07-29 09:58:35 -05001978 /* We can timeout waiting for Guest network transmit. */
1979 setup_timeout();
1980
Rusty Russell8ca47e02007-07-19 01:49:29 -07001981 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001982 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001983
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001984 /* Boot information is stashed at physical address 0 */
1985 boot = from_guest_phys(0);
1986
Rusty Russelldde79782007-07-26 10:41:03 -07001987 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001988 if (initrd_name) {
1989 initrd_size = load_initrd(initrd_name, mem);
Rusty Russelldde79782007-07-26 10:41:03 -07001990 /* These are the location in the Linux boot header where the
1991 * start and size of the initrd are expected to be found. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001992 boot->hdr.ramdisk_image = mem - initrd_size;
1993 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001994 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001995 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001996 }
1997
Rusty Russelldde79782007-07-26 10:41:03 -07001998 /* The Linux boot header contains an "E820" memory map: ours is a
1999 * simple, single region. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002000 boot->e820_entries = 1;
2001 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russelldde79782007-07-26 10:41:03 -07002002 /* The boot header contains a command line pointer: we put the command
Rusty Russell43d33b22007-10-22 11:29:57 +10002003 * line after the boot header. */
2004 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10002005 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002006 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07002007
Rusty Russell814a0e52007-10-22 11:29:44 +10002008 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002009 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002010
2011 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002012 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002013
Rusty Russell43d33b22007-10-22 11:29:57 +10002014 /* Tell the entry path not to try to reload segment registers. */
2015 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002016
Rusty Russelldde79782007-07-26 10:41:03 -07002017 /* We tell the kernel to initialize the Guest: this returns the open
2018 * /dev/lguest file descriptor. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03002019 lguest_fd = tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002020
Rusty Russell8c798732008-07-29 09:58:38 -05002021 /* We clone off a thread, which wakes the Launcher whenever one of the
2022 * input file descriptors needs attention. We call this the Waker, and
2023 * we'll cover it in a moment. */
2024 setup_waker(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002025
Rusty Russelldde79782007-07-26 10:41:03 -07002026 /* Finally, run the Guest. This doesn't return. */
Rusty Russell17cbca22007-10-22 11:24:22 +10002027 run_guest(lguest_fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002028}
Rusty Russellf56a3842007-07-26 10:41:05 -07002029/*:*/
2030
2031/*M:999
2032 * Mastery is done: you now know everything I do.
2033 *
2034 * But surely you have seen code, features and bugs in your wanderings which
2035 * you now yearn to attack? That is the real game, and I look forward to you
2036 * patching and forking lguest into the Your-Name-Here-visor.
2037 *
2038 * Farewell, and good coding!
2039 * Rusty Russell.
2040 */