blob: e9ce3c5545145f9b2c683c26690c0159b6c437fc [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:100
2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5 * control it.
6:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07007#define _LARGEFILE64_SOURCE
8#define _GNU_SOURCE
9#include <stdio.h>
10#include <string.h>
11#include <unistd.h>
12#include <err.h>
13#include <stdint.h>
14#include <stdlib.h>
15#include <elf.h>
16#include <sys/mman.h>
Ronald G. Minnich6649bb72007-08-28 14:35:59 -070017#include <sys/param.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070018#include <sys/types.h>
19#include <sys/stat.h>
20#include <sys/wait.h>
Rusty Russell659a0e62009-06-12 22:27:10 -060021#include <sys/eventfd.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070022#include <fcntl.h>
23#include <stdbool.h>
24#include <errno.h>
25#include <ctype.h>
26#include <sys/socket.h>
27#include <sys/ioctl.h>
28#include <sys/time.h>
29#include <time.h>
30#include <netinet/in.h>
31#include <net/if.h>
32#include <linux/sockios.h>
33#include <linux/if_tun.h>
34#include <sys/uio.h>
35#include <termios.h>
36#include <getopt.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100037#include <assert.h>
38#include <sched.h>
Rusty Russella586d4f2008-02-04 23:49:56 -050039#include <limits.h>
40#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050041#include <signal.h>
Rusty Russellb45d8cb2007-10-22 10:56:24 +100042#include "linux/lguest_launcher.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100043#include "linux/virtio_config.h"
44#include "linux/virtio_net.h"
45#include "linux/virtio_blk.h"
46#include "linux/virtio_console.h"
Rusty Russell28fd6d72008-07-29 09:58:33 -050047#include "linux/virtio_rng.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100048#include "linux/virtio_ring.h"
Rusty Russelld5d02d62008-10-31 11:24:25 -050049#include "asm/bootparam.h"
Rusty Russell2e04ef72009-07-30 16:03:45 -060050/*L:110
Rusty Russella91d74a2009-07-30 16:03:45 -060051 * We can ignore the 42 include files we need for this program, but I do want
Rusty Russell2e04ef72009-07-30 16:03:45 -060052 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100053 *
54 * As Linus said, "C is a Spartan language, and so should your naming be." I
55 * like these abbreviations, so we define them here. Note that u64 is always
56 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060057 * use %llu in printf for any u64.
58 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100059typedef unsigned long long u64;
60typedef uint32_t u32;
61typedef uint16_t u16;
62typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070063/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070064
65#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
Rusty Russell8ca47e02007-07-19 01:49:29 -070066#define BRIDGE_PFX "bridge:"
67#ifndef SIOCBRADDIF
68#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
69#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100070/* We can have up to 256 pages for devices. */
71#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050072/* This will occupy 3 pages: it must be a power of 2. */
73#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070074
Rusty Russell2e04ef72009-07-30 16:03:45 -060075/*L:120
76 * verbose is both a global flag and a macro. The C preprocessor allows
77 * this, and although I wouldn't recommend it, it works quite nicely here.
78 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070079static bool verbose;
80#define verbose(args...) \
81 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070082/*:*/
83
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100084/* The pointer to the start of guest memory. */
85static void *guest_base;
86/* The maximum guest physical address allowed, and maximum possible. */
87static unsigned long guest_limit, guest_max;
Rusty Russell56739c802009-06-12 22:26:59 -060088/* The /dev/lguest file descriptor. */
89static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070090
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020091/* a per-cpu variable indicating whose vcpu is currently running */
92static unsigned int __thread cpu_id;
93
Rusty Russelldde79782007-07-26 10:41:03 -070094/* This is our list of devices. */
Rusty Russell1842f232009-07-30 16:03:46 -060095struct device_list {
Rusty Russell17cbca22007-10-22 11:24:22 +100096 /* Counter to assign interrupt numbers. */
97 unsigned int next_irq;
98
99 /* Counter to print out convenient device numbers. */
100 unsigned int device_num;
101
Rusty Russelldde79782007-07-26 10:41:03 -0700102 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000103 u8 *descpage;
104
Rusty Russelldde79782007-07-26 10:41:03 -0700105 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700106 struct device *dev;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600107 /* And a pointer to the last device for easy append. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500108 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700109};
110
Rusty Russell17cbca22007-10-22 11:24:22 +1000111/* The list of Guest devices, based on command line arguments. */
112static struct device_list devices;
113
Rusty Russelldde79782007-07-26 10:41:03 -0700114/* The device structure describes a single device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600115struct device {
Rusty Russelldde79782007-07-26 10:41:03 -0700116 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700117 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000118
Rusty Russell713b15b2009-06-12 22:26:58 -0600119 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700120 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000121
Rusty Russell713b15b2009-06-12 22:26:58 -0600122 /* We can't trust desc values once Guest has booted: we use these. */
123 unsigned int feature_len;
124 unsigned int num_vq;
125
Rusty Russell17cbca22007-10-22 11:24:22 +1000126 /* The name of this device, for --verbose. */
127 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700128
Rusty Russell17cbca22007-10-22 11:24:22 +1000129 /* Any queues attached to this device */
130 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700131
Rusty Russell659a0e62009-06-12 22:27:10 -0600132 /* Is it operational */
133 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500134
Rusty Russellca60a422009-09-23 22:26:47 -0600135 /* Does Guest want an intrrupt on empty? */
136 bool irq_on_empty;
137
Rusty Russell8ca47e02007-07-19 01:49:29 -0700138 /* Device-specific data. */
139 void *priv;
140};
141
Rusty Russell17cbca22007-10-22 11:24:22 +1000142/* The virtqueue structure describes a queue attached to a device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600143struct virtqueue {
Rusty Russell17cbca22007-10-22 11:24:22 +1000144 struct virtqueue *next;
145
146 /* Which device owns me. */
147 struct device *dev;
148
149 /* The configuration for this queue. */
150 struct lguest_vqconfig config;
151
152 /* The actual ring of buffers. */
153 struct vring vring;
154
155 /* Last available index we saw. */
156 u16 last_avail_idx;
157
Rusty Russell95c517c2009-06-12 22:27:11 -0600158 /* How many are used since we sent last irq? */
159 unsigned int pending_used;
160
Rusty Russell659a0e62009-06-12 22:27:10 -0600161 /* Eventfd where Guest notifications arrive. */
162 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500163
Rusty Russell659a0e62009-06-12 22:27:10 -0600164 /* Function for the thread which is servicing this virtqueue. */
165 void (*service)(struct virtqueue *vq);
166 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000167};
168
Balaji Raoec04b132007-12-28 14:26:24 +0530169/* Remember the arguments to the program so we can "reboot" */
170static char **main_args;
171
Rusty Russell659a0e62009-06-12 22:27:10 -0600172/* The original tty settings to restore on exit. */
173static struct termios orig_term;
174
Rusty Russell2e04ef72009-07-30 16:03:45 -0600175/*
176 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600177 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600178 * in precise order.
179 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600180#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russellb60da132009-06-12 22:27:12 -0600181#define mb() __asm__ __volatile__("" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000182
Rusty Russell2e04ef72009-07-30 16:03:45 -0600183/*
184 * Convert an iovec element to the given type.
Rusty Russell17cbca22007-10-22 11:24:22 +1000185 *
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
Rusty Russell2e04ef72009-07-30 16:03:45 -0600191 * have a macro which sets them all up and passes to the real function.
192 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000193#define convert(iov, type) \
194 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
195
196static void *_convert(struct iovec *iov, size_t size, size_t align,
197 const char *name)
198{
199 if (iov->iov_len != size)
200 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
201 if ((unsigned long)iov->iov_base % align != 0)
202 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
203 return iov->iov_base;
204}
205
Rusty Russellb5111792008-07-29 09:58:34 -0500206/* Wrapper for the last available index. Makes it easier to change. */
207#define lg_last_avail(vq) ((vq)->last_avail_idx)
208
Rusty Russell2e04ef72009-07-30 16:03:45 -0600209/*
210 * The virtio configuration space is defined to be little-endian. x86 is
211 * little-endian too, but it's nice to be explicit so we have these helpers.
212 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000213#define cpu_to_le16(v16) (v16)
214#define cpu_to_le32(v32) (v32)
215#define cpu_to_le64(v64) (v64)
216#define le16_to_cpu(v16) (v16)
217#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500218#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000219
Rusty Russell28fd6d72008-07-29 09:58:33 -0500220/* Is this iovec empty? */
221static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
222{
223 unsigned int i;
224
225 for (i = 0; i < num_iov; i++)
226 if (iov[i].iov_len)
227 return false;
228 return true;
229}
230
231/* Take len bytes from the front of this iovec. */
232static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
233{
234 unsigned int i;
235
236 for (i = 0; i < num_iov; i++) {
237 unsigned int used;
238
239 used = iov[i].iov_len < len ? iov[i].iov_len : len;
240 iov[i].iov_base += used;
241 iov[i].iov_len -= used;
242 len -= used;
243 }
244 assert(len == 0);
245}
246
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500247/* The device virtqueue descriptors are followed by feature bitmasks. */
248static u8 *get_feature_bits(struct device *dev)
249{
250 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600251 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500252}
253
Rusty Russell2e04ef72009-07-30 16:03:45 -0600254/*L:100
255 * The Launcher code itself takes us out into userspace, that scary place where
256 * pointers run wild and free! Unfortunately, like most userspace programs,
257 * it's quite boring (which is why everyone likes to hack on the kernel!).
258 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
259 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000260 *
261 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
262 * memory and stores it in "guest_base". In other words, Guest physical ==
263 * Launcher virtual with an offset.
264 *
265 * This can be tough to get your head around, but usually it just means that we
Francis Galieguea33f3222010-04-23 00:08:02 +0200266 * use these trivial conversion functions when the Guest gives us its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600267 * "physical" addresses:
268 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000269static void *from_guest_phys(unsigned long addr)
270{
271 return guest_base + addr;
272}
273
274static unsigned long to_guest_phys(const void *addr)
275{
276 return (addr - guest_base);
277}
278
Rusty Russelldde79782007-07-26 10:41:03 -0700279/*L:130
280 * Loading the Kernel.
281 *
282 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600283 * error-checking code cluttering the callers:
284 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700285static int open_or_die(const char *name, int flags)
286{
287 int fd = open(name, flags);
288 if (fd < 0)
289 err(1, "Failed to open %s", name);
290 return fd;
291}
292
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000293/* map_zeroed_pages() takes a number of pages. */
294static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700295{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000296 int fd = open_or_die("/dev/zero", O_RDONLY);
297 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700298
Rusty Russell2e04ef72009-07-30 16:03:45 -0600299 /*
300 * We use a private mapping (ie. if we write to the page, it will be
301 * copied).
302 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000303 addr = mmap(NULL, getpagesize() * num,
304 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
305 if (addr == MAP_FAILED)
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200306 err(1, "Mmapping %u pages of /dev/zero", num);
Rusty Russella91d74a2009-07-30 16:03:45 -0600307
308 /*
309 * One neat mmap feature is that you can close the fd, and it
310 * stays mapped.
311 */
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100312 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700313
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000314 return addr;
315}
316
317/* Get some more pages for a device. */
318static void *get_pages(unsigned int num)
319{
320 void *addr = from_guest_phys(guest_limit);
321
322 guest_limit += num * getpagesize();
323 if (guest_limit > guest_max)
324 errx(1, "Not enough memory for devices");
325 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700326}
327
Rusty Russell2e04ef72009-07-30 16:03:45 -0600328/*
329 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700330 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600331 * it falls back to reading the memory in.
332 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700333static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
334{
335 ssize_t r;
336
Rusty Russell2e04ef72009-07-30 16:03:45 -0600337 /*
338 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700339 * The kernel really wants to be writable: it patches its own
340 * instructions.
341 *
342 * MAP_PRIVATE means that the page won't be copied until a write is
343 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600344 * Guests.
345 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700346 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
347 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
348 return;
349
350 /* pread does a seek and a read in one shot: saves a few lines. */
351 r = pread(fd, addr, len, offset);
352 if (r != len)
353 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
354}
355
Rusty Russell2e04ef72009-07-30 16:03:45 -0600356/*
357 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700358 * the Guest memory. ELF = Embedded Linking Format, which is the format used
359 * by all modern binaries on Linux including the kernel.
360 *
361 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000362 * address. We use the physical address; the Guest will map itself to the
363 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700364 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600365 * We return the starting address.
366 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000367static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700368{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700369 Elf32_Phdr phdr[ehdr->e_phnum];
370 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700371
Rusty Russell2e04ef72009-07-30 16:03:45 -0600372 /*
373 * Sanity checks on the main ELF header: an x86 executable with a
374 * reasonable number of correctly-sized program headers.
375 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700376 if (ehdr->e_type != ET_EXEC
377 || ehdr->e_machine != EM_386
378 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
379 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
380 errx(1, "Malformed elf header");
381
Rusty Russell2e04ef72009-07-30 16:03:45 -0600382 /*
383 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700384 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600385 * load where.
386 */
Rusty Russelldde79782007-07-26 10:41:03 -0700387
388 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700389 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
390 err(1, "Seeking to program headers");
391 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
392 err(1, "Reading program headers");
393
Rusty Russell2e04ef72009-07-30 16:03:45 -0600394 /*
395 * Try all the headers: there are usually only three. A read-only one,
396 * a read-write one, and a "note" section which we don't load.
397 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700398 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700399 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700400 if (phdr[i].p_type != PT_LOAD)
401 continue;
402
403 verbose("Section %i: size %i addr %p\n",
404 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
405
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700406 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000407 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700408 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700409 }
410
Rusty Russell814a0e52007-10-22 11:29:44 +1000411 /* The entry point is given in the ELF header. */
412 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700413}
414
Rusty Russell2e04ef72009-07-30 16:03:45 -0600415/*L:150
416 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
417 * to jump into it and it will unpack itself. We used to have to perform some
418 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700419 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000420 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
421 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600422 * the funky header so we know where in the file to load, and away we go!
423 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000424static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700425{
Rusty Russell43d33b22007-10-22 11:29:57 +1000426 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000427 int r;
428 /* Modern bzImages get loaded at 1M. */
429 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700430
Rusty Russell2e04ef72009-07-30 16:03:45 -0600431 /*
432 * Go back to the start of the file and read the header. It should be
433 * a Linux boot header (see Documentation/x86/i386/boot.txt)
434 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000435 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000436 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000437
Rusty Russell43d33b22007-10-22 11:29:57 +1000438 /* Inside the setup_hdr, we expect the magic "HdrS" */
439 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000440 errx(1, "This doesn't look like a bzImage to me");
441
Rusty Russell43d33b22007-10-22 11:29:57 +1000442 /* Skip over the extra sectors of the header. */
443 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000444
445 /* Now read everything into memory. in nice big chunks. */
446 while ((r = read(fd, p, 65536)) > 0)
447 p += r;
448
Rusty Russell43d33b22007-10-22 11:29:57 +1000449 /* Finally, code32_start tells us where to enter the kernel. */
450 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700451}
452
Rusty Russell2e04ef72009-07-30 16:03:45 -0600453/*L:140
454 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000455 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600456 * work, we can load those, too.
457 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000458static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700459{
460 Elf32_Ehdr hdr;
461
Rusty Russelldde79782007-07-26 10:41:03 -0700462 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700463 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
464 err(1, "Reading kernel");
465
Rusty Russelldde79782007-07-26 10:41:03 -0700466 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700467 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000468 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700469
Rusty Russella6bd8e12008-03-28 11:05:53 -0500470 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000471 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700472}
473
Rusty Russell2e04ef72009-07-30 16:03:45 -0600474/*
475 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700476 * it calls getpagesize() twice: "it's dumb code."
477 *
478 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600479 * necessary. I leave this code as a reaction against that.
480 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700481static inline unsigned long page_align(unsigned long addr)
482{
Rusty Russelldde79782007-07-26 10:41:03 -0700483 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700484 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
485}
486
Rusty Russell2e04ef72009-07-30 16:03:45 -0600487/*L:180
488 * An "initial ram disk" is a disk image loaded into memory along with the
489 * kernel which the kernel can use to boot from without needing any drivers.
490 * Most distributions now use this as standard: the initrd contains the code to
491 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700492 *
493 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600494 * kernels. He sent me this (and tells me when I break it).
495 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700496static unsigned long load_initrd(const char *name, unsigned long mem)
497{
498 int ifd;
499 struct stat st;
500 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700501
502 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700503 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700504 if (fstat(ifd, &st) < 0)
505 err(1, "fstat() on initrd '%s'", name);
506
Rusty Russell2e04ef72009-07-30 16:03:45 -0600507 /*
508 * We map the initrd at the top of memory, but mmap wants it to be
509 * page-aligned, so we round the size up for that.
510 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700511 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000512 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600513 /*
514 * Once a file is mapped, you can close the file descriptor. It's a
515 * little odd, but quite useful.
516 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700517 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700518 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700519
520 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700521 return len;
522}
Rusty Russelle1e72962007-10-25 15:02:50 +1000523/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700524
Rusty Russell2e04ef72009-07-30 16:03:45 -0600525/*
526 * Simple routine to roll all the commandline arguments together with spaces
527 * between them.
528 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700529static void concat(char *dst, char *args[])
530{
531 unsigned int i, len = 0;
532
533 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100534 if (i) {
535 strcat(dst+len, " ");
536 len++;
537 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700538 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100539 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700540 }
541 /* In case it's empty. */
542 dst[len] = '\0';
543}
544
Rusty Russell2e04ef72009-07-30 16:03:45 -0600545/*L:185
546 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000547 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300548 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600549 * entry point for the Guest.
550 */
Rusty Russell56739c802009-06-12 22:26:59 -0600551static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700552{
Jes Sorensen511801d2007-10-22 11:03:31 +1000553 unsigned long args[] = { LHREQ_INITIALIZE,
554 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300555 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000556 verbose("Guest: %p - %p (%#lx)\n",
557 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c802009-06-12 22:26:59 -0600558 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
559 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700560 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700561}
Rusty Russelldde79782007-07-26 10:41:03 -0700562/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700563
Rusty Russella91d74a2009-07-30 16:03:45 -0600564/*L:200
Rusty Russelldde79782007-07-26 10:41:03 -0700565 * Device Handling.
566 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000567 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700568 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000569 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700570 * if something funny is going on:
571 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700572static void *_check_pointer(unsigned long addr, unsigned int size,
573 unsigned int line)
574{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600575 /*
576 * We have to separately check addr and addr+size, because size could
577 * be huge and addr + size might wrap around.
578 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000579 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000580 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600581 /*
582 * We return a pointer for the caller's convenience, now we know it's
583 * safe to use.
584 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000585 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700586}
Rusty Russelldde79782007-07-26 10:41:03 -0700587/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700588#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
589
Rusty Russell2e04ef72009-07-30 16:03:45 -0600590/*
591 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000592 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600593 * at the end.
594 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100595static unsigned next_desc(struct vring_desc *desc,
596 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000597{
598 unsigned int next;
599
600 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100601 if (!(desc[i].flags & VRING_DESC_F_NEXT))
602 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000603
604 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100605 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000606 /* Make sure compiler knows to grab that: we don't want it changing! */
607 wmb();
608
Mark McLoughlind1f01322009-05-11 18:11:46 +0100609 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000610 errx(1, "Desc next is %u", next);
611
612 return next;
613}
614
Rusty Russella91d74a2009-07-30 16:03:45 -0600615/*
616 * This actually sends the interrupt for this virtqueue, if we've used a
617 * buffer.
618 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600619static void trigger_irq(struct virtqueue *vq)
620{
621 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
622
Rusty Russell95c517c2009-06-12 22:27:11 -0600623 /* Don't inform them if nothing used. */
624 if (!vq->pending_used)
625 return;
626 vq->pending_used = 0;
627
Rusty Russellca60a422009-09-23 22:26:47 -0600628 /* If they don't want an interrupt, don't send one... */
629 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
630 /* ... unless they've asked us to force one on empty. */
631 if (!vq->dev->irq_on_empty
632 || lg_last_avail(vq) != vq->vring.avail->idx)
633 return;
634 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600635
636 /* Send the Guest an interrupt tell them we used something up. */
637 if (write(lguest_fd, buf, sizeof(buf)) != 0)
638 err(1, "Triggering irq %i", vq->config.irq);
639}
640
Rusty Russell2e04ef72009-07-30 16:03:45 -0600641/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600642 * This looks in the virtqueue for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000643 * it to an iovec for convenient access. Since descriptors consist of some
644 * number of output then some number of input descriptors, it's actually two
645 * iovecs, but we pack them into one and note how many of each there were.
646 *
Rusty Russella91d74a2009-07-30 16:03:45 -0600647 * This function waits if necessary, and returns the descriptor number found.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600648 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600649static unsigned wait_for_vq_desc(struct virtqueue *vq,
650 struct iovec iov[],
651 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000652{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100653 unsigned int i, head, max;
654 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600655 u16 last_avail = lg_last_avail(vq);
656
Rusty Russella91d74a2009-07-30 16:03:45 -0600657 /* There's nothing available? */
Rusty Russell659a0e62009-06-12 22:27:10 -0600658 while (last_avail == vq->vring.avail->idx) {
659 u64 event;
660
Rusty Russella91d74a2009-07-30 16:03:45 -0600661 /*
662 * Since we're about to sleep, now is a good time to tell the
663 * Guest about what we've used up to now.
664 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600665 trigger_irq(vq);
666
Rusty Russellb60da132009-06-12 22:27:12 -0600667 /* OK, now we need to know about added descriptors. */
668 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
669
Rusty Russell2e04ef72009-07-30 16:03:45 -0600670 /*
671 * They could have slipped one in as we were doing that: make
672 * sure it's written, then check again.
673 */
Rusty Russellb60da132009-06-12 22:27:12 -0600674 mb();
675 if (last_avail != vq->vring.avail->idx) {
676 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
677 break;
678 }
679
Rusty Russell659a0e62009-06-12 22:27:10 -0600680 /* Nothing new? Wait for eventfd to tell us they refilled. */
681 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
682 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600683
684 /* We don't need to be notified again. */
685 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600686 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000687
688 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500689 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000690 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500691 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000692
Rusty Russell2e04ef72009-07-30 16:03:45 -0600693 /*
694 * Grab the next descriptor number they're advertising, and increment
695 * the index we've seen.
696 */
Rusty Russellb5111792008-07-29 09:58:34 -0500697 head = vq->vring.avail->ring[last_avail % vq->vring.num];
698 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000699
700 /* If their number is silly, that's a fatal mistake. */
701 if (head >= vq->vring.num)
702 errx(1, "Guest says index %u is available", head);
703
704 /* When we start there are none of either input nor output. */
705 *out_num = *in_num = 0;
706
Mark McLoughlind1f01322009-05-11 18:11:46 +0100707 max = vq->vring.num;
708 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000709 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100710
Rusty Russell2e04ef72009-07-30 16:03:45 -0600711 /*
712 * If this is an indirect entry, then this buffer contains a descriptor
713 * table which we handle as if it's any normal descriptor chain.
714 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100715 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
716 if (desc[i].len % sizeof(struct vring_desc))
717 errx(1, "Invalid size for indirect buffer table");
718
719 max = desc[i].len / sizeof(struct vring_desc);
720 desc = check_pointer(desc[i].addr, desc[i].len);
721 i = 0;
722 }
723
Rusty Russell17cbca22007-10-22 11:24:22 +1000724 do {
725 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100726 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000727 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100728 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000729 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100730 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000731 (*in_num)++;
732 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600733 /*
734 * If it's an output descriptor, they're all supposed
735 * to come before any input descriptors.
736 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000737 if (*in_num)
738 errx(1, "Descriptor has out after in");
739 (*out_num)++;
740 }
741
742 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100743 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000744 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100745 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000746
747 return head;
748}
749
Rusty Russell2e04ef72009-07-30 16:03:45 -0600750/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600751 * After we've used one of their buffers, we tell the Guest about it. Sometime
752 * later we'll want to send them an interrupt using trigger_irq(); note that
753 * wait_for_vq_desc() does that for us if it has to wait.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600754 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000755static void add_used(struct virtqueue *vq, unsigned int head, int len)
756{
757 struct vring_used_elem *used;
758
Rusty Russell2e04ef72009-07-30 16:03:45 -0600759 /*
760 * The virtqueue contains a ring of used buffers. Get a pointer to the
761 * next entry in that used ring.
762 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000763 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
764 used->id = head;
765 used->len = len;
766 /* Make sure buffer is written before we update index. */
767 wmb();
768 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600769 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000770}
771
Rusty Russell17cbca22007-10-22 11:24:22 +1000772/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c802009-06-12 22:26:59 -0600773static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000774{
775 add_used(vq, head, len);
Rusty Russell56739c802009-06-12 22:26:59 -0600776 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000777}
778
Rusty Russelle1e72962007-10-25 15:02:50 +1000779/*
780 * The Console
781 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600782 * We associate some data with the console for our exit hack.
783 */
Rusty Russell1842f232009-07-30 16:03:46 -0600784struct console_abort {
Rusty Russelldde79782007-07-26 10:41:03 -0700785 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700786 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700787 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700788 struct timeval start;
789};
790
Rusty Russelldde79782007-07-26 10:41:03 -0700791/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600792static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700793{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700794 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000795 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600796 struct console_abort *abort = vq->dev->priv;
797 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700798
Rusty Russella91d74a2009-07-30 16:03:45 -0600799 /* Make sure there's a descriptor available. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600800 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000801 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000802 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700803
Rusty Russella91d74a2009-07-30 16:03:45 -0600804 /* Read into it. This is where we usually wait. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600805 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700806 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600807 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700808 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600809 /*
810 * For simplicity, dying threads kill the whole Launcher. So
811 * just nap here.
812 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600813 for (;;)
814 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700815 }
816
Rusty Russella91d74a2009-07-30 16:03:45 -0600817 /* Tell the Guest we used a buffer. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600818 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700819
Rusty Russell2e04ef72009-07-30 16:03:45 -0600820 /*
821 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700822 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600823 * This is such a hack, but works surprisingly well. Each ^C has to
824 * be in a buffer by itself, so they can't be too fast. But we check
825 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600826 * slow.
827 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600828 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700829 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600830 return;
831 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700832
Rusty Russell659a0e62009-06-12 22:27:10 -0600833 abort->count++;
834 if (abort->count == 1)
835 gettimeofday(&abort->start, NULL);
836 else if (abort->count == 3) {
837 struct timeval now;
838 gettimeofday(&now, NULL);
839 /* Kill all Launcher processes with SIGINT, like normal ^C */
840 if (now.tv_sec <= abort->start.tv_sec+1)
841 kill(0, SIGINT);
842 abort->count = 0;
843 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700844}
845
Rusty Russell659a0e62009-06-12 22:27:10 -0600846/* This is the routine which handles console output (ie. stdout). */
847static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700848{
Rusty Russell17cbca22007-10-22 11:24:22 +1000849 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000850 struct iovec iov[vq->vring.num];
851
Rusty Russella91d74a2009-07-30 16:03:45 -0600852 /* We usually wait in here, for the Guest to give us something. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600853 head = wait_for_vq_desc(vq, iov, &out, &in);
854 if (in)
855 errx(1, "Input buffers in console output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600856
857 /* writev can return a partial write, so we loop here. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600858 while (!iov_empty(iov, out)) {
859 int len = writev(STDOUT_FILENO, iov, out);
860 if (len <= 0)
861 err(1, "Write to stdout gave %i", len);
862 iov_consume(iov, out, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000863 }
Rusty Russella91d74a2009-07-30 16:03:45 -0600864
865 /*
866 * We're finished with that buffer: if we're going to sleep,
867 * wait_for_vq_desc() will prod the Guest with an interrupt.
868 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600869 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500870}
871
Rusty Russelle1e72962007-10-25 15:02:50 +1000872/*
873 * The Network
874 *
875 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600876 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500877 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600878struct net_info {
879 int tunfd;
880};
881
882static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700883{
Rusty Russell659a0e62009-06-12 22:27:10 -0600884 struct net_info *net_info = vq->dev->priv;
885 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000886 struct iovec iov[vq->vring.num];
887
Rusty Russella91d74a2009-07-30 16:03:45 -0600888 /* We usually wait in here for the Guest to give us a packet. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600889 head = wait_for_vq_desc(vq, iov, &out, &in);
890 if (in)
891 errx(1, "Input buffers in net output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600892 /*
893 * Send the whole thing through to /dev/net/tun. It expects the exact
894 * same format: what a coincidence!
895 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600896 if (writev(net_info->tunfd, iov, out) < 0)
897 errx(1, "Write to tun failed?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600898
899 /*
900 * Done with that one; wait_for_vq_desc() will send the interrupt if
901 * all packets are processed.
902 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600903 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700904}
905
Rusty Russella91d74a2009-07-30 16:03:45 -0600906/*
907 * Handling network input is a bit trickier, because I've tried to optimize it.
908 *
909 * First we have a helper routine which tells is if from this file descriptor
910 * (ie. the /dev/net/tun device) will block:
911 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600912static bool will_block(int fd)
913{
914 fd_set fdset;
915 struct timeval zero = { 0, 0 };
916 FD_ZERO(&fdset);
917 FD_SET(fd, &fdset);
918 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
919}
920
Rusty Russella91d74a2009-07-30 16:03:45 -0600921/*
922 * This handles packets coming in from the tun device to our Guest. Like all
923 * service routines, it gets called again as soon as it returns, so you don't
924 * see a while(1) loop here.
925 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600926static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700927{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700928 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600929 unsigned int head, out, in;
930 struct iovec iov[vq->vring.num];
931 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700932
Rusty Russella91d74a2009-07-30 16:03:45 -0600933 /*
934 * Get a descriptor to write an incoming packet into. This will also
935 * send an interrupt if they're out of descriptors.
936 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600937 head = wait_for_vq_desc(vq, iov, &out, &in);
938 if (out)
939 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600940
Rusty Russella91d74a2009-07-30 16:03:45 -0600941 /*
942 * If it looks like we'll block reading from the tun device, send them
943 * an interrupt.
944 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600945 if (vq->pending_used && will_block(net_info->tunfd))
946 trigger_irq(vq);
947
Rusty Russella91d74a2009-07-30 16:03:45 -0600948 /*
949 * Read in the packet. This is where we normally wait (when there's no
950 * incoming network traffic).
951 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600952 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700953 if (len <= 0)
Rusty Russell659a0e62009-06-12 22:27:10 -0600954 err(1, "Failed to read from tun.");
Rusty Russella91d74a2009-07-30 16:03:45 -0600955
956 /*
957 * Mark that packet buffer as used, but don't interrupt here. We want
958 * to wait until we've done as much work as we can.
959 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600960 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700961}
Rusty Russella91d74a2009-07-30 16:03:45 -0600962/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700963
Rusty Russella91d74a2009-07-30 16:03:45 -0600964/* This is the helper to create threads: run the service routine in a loop. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600965static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000966{
Rusty Russell659a0e62009-06-12 22:27:10 -0600967 struct virtqueue *vq = _vq;
968
969 for (;;)
970 vq->service(vq);
971 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000972}
973
Rusty Russell2e04ef72009-07-30 16:03:45 -0600974/*
975 * When a child dies, we kill our entire process group with SIGTERM. This
976 * also has the side effect that the shell restores the console for us!
977 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600978static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500979{
Rusty Russell659a0e62009-06-12 22:27:10 -0600980 kill(0, SIGTERM);
981}
982
983static void reset_device(struct device *dev)
984{
985 struct virtqueue *vq;
986
987 verbose("Resetting device %s\n", dev->name);
988
989 /* Clear any features they've acked. */
990 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
991
992 /* We're going to be explicitly killing threads, so ignore them. */
993 signal(SIGCHLD, SIG_IGN);
994
995 /* Zero out the virtqueues, get rid of their threads */
996 for (vq = dev->vq; vq; vq = vq->next) {
997 if (vq->thread != (pid_t)-1) {
998 kill(vq->thread, SIGTERM);
999 waitpid(vq->thread, NULL, 0);
1000 vq->thread = (pid_t)-1;
1001 }
1002 memset(vq->vring.desc, 0,
1003 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1004 lg_last_avail(vq) = 0;
1005 }
1006 dev->running = false;
1007
1008 /* Now we care if threads die. */
1009 signal(SIGCHLD, (void *)kill_launcher);
1010}
1011
Rusty Russella91d74a2009-07-30 16:03:45 -06001012/*L:216
1013 * This actually creates the thread which services the virtqueue for a device.
1014 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001015static void create_thread(struct virtqueue *vq)
1016{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001017 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001018 * Create stack for thread. Since the stack grows upwards, we point
1019 * the stack pointer to the end of this region.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001020 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001021 char *stack = malloc(32768);
1022 unsigned long args[] = { LHREQ_EVENTFD,
1023 vq->config.pfn*getpagesize(), 0 };
1024
1025 /* Create a zero-initialized eventfd. */
1026 vq->eventfd = eventfd(0, 0);
1027 if (vq->eventfd < 0)
1028 err(1, "Creating eventfd");
1029 args[2] = vq->eventfd;
1030
Rusty Russella91d74a2009-07-30 16:03:45 -06001031 /*
1032 * Attach an eventfd to this virtqueue: it will go off when the Guest
1033 * does an LHCALL_NOTIFY for this vq.
1034 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001035 if (write(lguest_fd, &args, sizeof(args)) != 0)
1036 err(1, "Attaching eventfd");
1037
Rusty Russella91d74a2009-07-30 16:03:45 -06001038 /*
1039 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1040 * we get a signal if it dies.
1041 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001042 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1043 if (vq->thread == (pid_t)-1)
1044 err(1, "Creating clone");
Rusty Russella91d74a2009-07-30 16:03:45 -06001045
1046 /* We close our local copy now the child has it. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001047 close(vq->eventfd);
1048}
1049
Rusty Russellca60a422009-09-23 22:26:47 -06001050static bool accepted_feature(struct device *dev, unsigned int bit)
1051{
1052 const u8 *features = get_feature_bits(dev) + dev->feature_len;
1053
1054 if (dev->feature_len < bit / CHAR_BIT)
1055 return false;
1056 return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT));
1057}
1058
Rusty Russell659a0e62009-06-12 22:27:10 -06001059static void start_device(struct device *dev)
1060{
1061 unsigned int i;
1062 struct virtqueue *vq;
1063
1064 verbose("Device %s OK: offered", dev->name);
1065 for (i = 0; i < dev->feature_len; i++)
1066 verbose(" %02x", get_feature_bits(dev)[i]);
1067 verbose(", accepted");
1068 for (i = 0; i < dev->feature_len; i++)
1069 verbose(" %02x", get_feature_bits(dev)
1070 [dev->feature_len+i]);
1071
Rusty Russellca60a422009-09-23 22:26:47 -06001072 dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1073
Rusty Russell659a0e62009-06-12 22:27:10 -06001074 for (vq = dev->vq; vq; vq = vq->next) {
1075 if (vq->service)
1076 create_thread(vq);
1077 }
1078 dev->running = true;
1079}
1080
1081static void cleanup_devices(void)
1082{
1083 struct device *dev;
1084
1085 for (dev = devices.dev; dev; dev = dev->next)
1086 reset_device(dev);
1087
1088 /* If we saved off the original terminal settings, restore them now. */
1089 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1090 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001091}
1092
Rusty Russella007a752008-05-02 21:50:53 -05001093/* When the Guest tells us they updated the status field, we handle it. */
1094static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001095{
Rusty Russell659a0e62009-06-12 22:27:10 -06001096 /* A zero status is a reset, otherwise it's a set of flags. */
1097 if (dev->desc->status == 0)
1098 reset_device(dev);
1099 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001100 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001101 if (dev->running)
1102 reset_device(dev);
Rusty Russella007a752008-05-02 21:50:53 -05001103 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001104 if (!dev->running)
1105 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001106 }
1107}
1108
Rusty Russella91d74a2009-07-30 16:03:45 -06001109/*L:215
1110 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1111 * particular, it's used to notify us of device status changes during boot.
1112 */
Rusty Russell56739c802009-06-12 22:26:59 -06001113static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001114{
1115 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001116
Rusty Russell659a0e62009-06-12 22:27:10 -06001117 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001118 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001119 struct virtqueue *vq;
1120
Rusty Russella91d74a2009-07-30 16:03:45 -06001121 /*
1122 * Notifications to device descriptors mean they updated the
1123 * device status.
1124 */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001125 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001126 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001127 return;
1128 }
1129
Rusty Russella91d74a2009-07-30 16:03:45 -06001130 /*
1131 * Devices *can* be used before status is set to DRIVER_OK.
1132 * The original plan was that they would never do this: they
1133 * would always finish setting up their status bits before
1134 * actually touching the virtqueues. In practice, we allowed
1135 * them to, and they do (eg. the disk probes for partition
1136 * tables as part of initialization).
1137 *
1138 * If we see this, we start the device: once it's running, we
1139 * expect the device to catch all the notifications.
1140 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001141 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001142 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001143 continue;
Rusty Russell659a0e62009-06-12 22:27:10 -06001144 if (i->running)
1145 errx(1, "Notification on running %s", i->name);
Rusty Russella91d74a2009-07-30 16:03:45 -06001146 /* This just calls create_thread() for each virtqueue */
Rusty Russell659a0e62009-06-12 22:27:10 -06001147 start_device(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001148 return;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001149 }
1150 }
Rusty Russelldde79782007-07-26 10:41:03 -07001151
Rusty Russell2e04ef72009-07-30 16:03:45 -06001152 /*
1153 * Early console write is done using notify on a nul-terminated string
1154 * in Guest memory. It's also great for hacking debugging messages
1155 * into a Guest.
1156 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001157 if (addr >= guest_limit)
1158 errx(1, "Bad NOTIFY %#lx", addr);
1159
1160 write(STDOUT_FILENO, from_guest_phys(addr),
1161 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001162}
1163
Rusty Russelldde79782007-07-26 10:41:03 -07001164/*L:190
1165 * Device Setup
1166 *
1167 * All devices need a descriptor so the Guest knows it exists, and a "struct
1168 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001169 * routines to allocate and manage them.
1170 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001171
Rusty Russell2e04ef72009-07-30 16:03:45 -06001172/*
1173 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001174 * number of virtqueue descriptors, then two sets of feature bits, then an
1175 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001176 * pointer.
1177 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001178static u8 *device_config(const struct device *dev)
1179{
1180 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001181 + dev->num_vq * sizeof(struct lguest_vqconfig)
1182 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001183}
1184
Rusty Russell2e04ef72009-07-30 16:03:45 -06001185/*
1186 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001187 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001188 * that descriptor.
1189 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001190static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001191{
Rusty Russella586d4f2008-02-04 23:49:56 -05001192 struct lguest_device_desc d = { .type = type };
1193 void *p;
1194
1195 /* Figure out where the next device config is, based on the last one. */
1196 if (devices.lastdev)
1197 p = device_config(devices.lastdev)
1198 + devices.lastdev->desc->config_len;
1199 else
1200 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001201
Rusty Russell17cbca22007-10-22 11:24:22 +10001202 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001203 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001204 errx(1, "Too many devices");
1205
Rusty Russella586d4f2008-02-04 23:49:56 -05001206 /* p might not be aligned, so we memcpy in. */
1207 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001208}
1209
Rusty Russell2e04ef72009-07-30 16:03:45 -06001210/*
1211 * Each device descriptor is followed by the description of its virtqueues. We
1212 * specify how many descriptors the virtqueue is to have.
1213 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001214static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001215 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001216{
1217 unsigned int pages;
1218 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1219 void *p;
1220
Rusty Russella6bd8e12008-03-28 11:05:53 -05001221 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001222 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001223 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001224 p = get_pages(pages);
1225
Rusty Russelld1c856e2007-11-19 11:20:40 -05001226 /* Initialize the virtqueue */
1227 vq->next = NULL;
1228 vq->last_avail_idx = 0;
1229 vq->dev = dev;
Rusty Russella91d74a2009-07-30 16:03:45 -06001230
1231 /*
1232 * This is the routine the service thread will run, and its Process ID
1233 * once it's running.
1234 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001235 vq->service = service;
1236 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001237
Rusty Russell17cbca22007-10-22 11:24:22 +10001238 /* Initialize the configuration. */
1239 vq->config.num = num_descs;
1240 vq->config.irq = devices.next_irq++;
1241 vq->config.pfn = to_guest_phys(p) / getpagesize();
1242
1243 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001244 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001245
Rusty Russell2e04ef72009-07-30 16:03:45 -06001246 /*
1247 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001248 * device_config() to get the end of the device's current virtqueues;
1249 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001250 * yet, otherwise we'd be overwriting them.
1251 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001252 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1253 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001254 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001255 dev->desc->num_vq++;
1256
1257 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001258
Rusty Russell2e04ef72009-07-30 16:03:45 -06001259 /*
1260 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1261 * second.
1262 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001263 for (i = &dev->vq; *i; i = &(*i)->next);
1264 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001265}
1266
Rusty Russell2e04ef72009-07-30 16:03:45 -06001267/*
1268 * The first half of the feature bitmask is for us to advertise features. The
1269 * second half is for the Guest to accept features.
1270 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001271static void add_feature(struct device *dev, unsigned bit)
1272{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001273 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001274
1275 /* We can't extend the feature bits once we've added config bytes */
1276 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1277 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001278 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001279 }
1280
Rusty Russella586d4f2008-02-04 23:49:56 -05001281 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1282}
1283
Rusty Russell2e04ef72009-07-30 16:03:45 -06001284/*
1285 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001286 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001287 * how we use it.
1288 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001289static void set_config(struct device *dev, unsigned len, const void *conf)
1290{
1291 /* Check we haven't overflowed our single page. */
1292 if (device_config(dev) + len > devices.descpage + getpagesize())
1293 errx(1, "Too many devices");
1294
1295 /* Copy in the config information, and store the length. */
1296 memcpy(device_config(dev), conf, len);
1297 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001298
1299 /* Size must fit in config_len field (8 bits)! */
1300 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001301}
1302
Rusty Russell2e04ef72009-07-30 16:03:45 -06001303/*
1304 * This routine does all the creation and setup of a new device, including
Rusty Russella91d74a2009-07-30 16:03:45 -06001305 * calling new_dev_desc() to allocate the descriptor and device memory. We
1306 * don't actually start the service threads until later.
Rusty Russella6bd8e12008-03-28 11:05:53 -05001307 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001308 * See what I mean about userspace being boring?
1309 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001310static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001311{
1312 struct device *dev = malloc(sizeof(*dev));
1313
Rusty Russelldde79782007-07-26 10:41:03 -07001314 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001315 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001316 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001317 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001318 dev->feature_len = 0;
1319 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001320 dev->running = false;
Rusty Russella586d4f2008-02-04 23:49:56 -05001321
Rusty Russell2e04ef72009-07-30 16:03:45 -06001322 /*
1323 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001324 * easier, but the user expects the devices to be arranged on the bus
1325 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001326 * is eth0, the first block device /dev/vda, etc.
1327 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001328 if (devices.lastdev)
1329 devices.lastdev->next = dev;
1330 else
1331 devices.dev = dev;
1332 devices.lastdev = dev;
1333
Rusty Russell8ca47e02007-07-19 01:49:29 -07001334 return dev;
1335}
1336
Rusty Russell2e04ef72009-07-30 16:03:45 -06001337/*
1338 * Our first setup routine is the console. It's a fairly simple device, but
1339 * UNIX tty handling makes it uglier than it could be.
1340 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001341static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001342{
1343 struct device *dev;
1344
Rusty Russelldde79782007-07-26 10:41:03 -07001345 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001346 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1347 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001348 /*
1349 * Then we turn off echo, line buffering and ^C etc: We want a
1350 * raw input stream to the Guest.
1351 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001352 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1353 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001354 }
1355
Rusty Russell659a0e62009-06-12 22:27:10 -06001356 dev = new_device("console", VIRTIO_ID_CONSOLE);
1357
Rusty Russelldde79782007-07-26 10:41:03 -07001358 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001359 dev->priv = malloc(sizeof(struct console_abort));
1360 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001361
Rusty Russell2e04ef72009-07-30 16:03:45 -06001362 /*
1363 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001364 * they put something the input queue, we make sure we're listening to
1365 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001366 * stdout.
1367 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001368 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1369 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001370
Rusty Russell659a0e62009-06-12 22:27:10 -06001371 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001372}
Rusty Russelldde79782007-07-26 10:41:03 -07001373/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001374
Rusty Russell2e04ef72009-07-30 16:03:45 -06001375/*M:010
1376 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001377 * --sharenet=<name> option which opens or creates a named pipe. This can be
1378 * used to send packets to another guest in a 1:1 manner.
1379 *
1380 * More sopisticated is to use one of the tools developed for project like UML
1381 * to do networking.
1382 *
1383 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1384 * completely generic ("here's my vring, attach to your vring") and would work
1385 * for any traffic. Of course, namespace and permissions issues need to be
1386 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1387 * multiple inter-guest channels behind one interface, although it would
1388 * require some manner of hotplugging new virtio channels.
1389 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001390 * Finally, we could implement a virtio network switch in the kernel.
1391:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001392
Rusty Russell8ca47e02007-07-19 01:49:29 -07001393static u32 str2ip(const char *ipaddr)
1394{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001395 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001396
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001397 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1398 errx(1, "Failed to parse IP address '%s'", ipaddr);
1399 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1400}
1401
1402static void str2mac(const char *macaddr, unsigned char mac[6])
1403{
1404 unsigned int m[6];
1405 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1406 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1407 errx(1, "Failed to parse mac address '%s'", macaddr);
1408 mac[0] = m[0];
1409 mac[1] = m[1];
1410 mac[2] = m[2];
1411 mac[3] = m[3];
1412 mac[4] = m[4];
1413 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001414}
1415
Rusty Russell2e04ef72009-07-30 16:03:45 -06001416/*
1417 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001418 * network device to the bridge device specified by the command line.
1419 *
1420 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001421 * dislike bridging), and I just try not to break it.
1422 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001423static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1424{
1425 int ifidx;
1426 struct ifreq ifr;
1427
1428 if (!*br_name)
1429 errx(1, "must specify bridge name");
1430
1431 ifidx = if_nametoindex(if_name);
1432 if (!ifidx)
1433 errx(1, "interface %s does not exist!", if_name);
1434
1435 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001436 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001437 ifr.ifr_ifindex = ifidx;
1438 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1439 err(1, "can't add %s to bridge %s", if_name, br_name);
1440}
1441
Rusty Russell2e04ef72009-07-30 16:03:45 -06001442/*
1443 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001444 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001445 * pointer.
1446 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001447static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001448{
1449 struct ifreq ifr;
1450 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1451
1452 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001453 strcpy(ifr.ifr_name, tapif);
1454
1455 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001456 sin->sin_family = AF_INET;
1457 sin->sin_addr.s_addr = htonl(ipaddr);
1458 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001459 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001460 ifr.ifr_flags = IFF_UP;
1461 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001462 err(1, "Bringing interface %s up", tapif);
1463}
1464
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001465static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001466{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001467 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001468 int netfd;
1469
1470 /* Start with this zeroed. Messy but sure. */
1471 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001472
Rusty Russell2e04ef72009-07-30 16:03:45 -06001473 /*
1474 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001475 * tap device is like a tun device, only somehow different. To tell
1476 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001477 * works now!
1478 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001479 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001480 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001481 strcpy(ifr.ifr_name, "tap%d");
1482 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1483 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001484
Rusty Russell398f1872008-07-29 09:58:37 -05001485 if (ioctl(netfd, TUNSETOFFLOAD,
1486 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1487 err(1, "Could not set features for tun device");
1488
Rusty Russell2e04ef72009-07-30 16:03:45 -06001489 /*
1490 * We don't need checksums calculated for packets coming in this
1491 * device: trust us!
1492 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001493 ioctl(netfd, TUNSETNOCSUM, 1);
1494
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001495 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1496 return netfd;
1497}
1498
Rusty Russell2e04ef72009-07-30 16:03:45 -06001499/*L:195
1500 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001501 * routing, but the principle is the same: it uses the "tun" device to inject
1502 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001503 * just shunt packets between the Guest and the tun device.
1504 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001505static void setup_tun_net(char *arg)
1506{
1507 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001508 struct net_info *net_info = malloc(sizeof(*net_info));
1509 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001510 u32 ip = INADDR_ANY;
1511 bool bridging = false;
1512 char tapif[IFNAMSIZ], *p;
1513 struct virtio_net_config conf;
1514
Rusty Russell659a0e62009-06-12 22:27:10 -06001515 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001516
Rusty Russell17cbca22007-10-22 11:24:22 +10001517 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001518 dev = new_device("net", VIRTIO_ID_NET);
1519 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001520
Rusty Russell2e04ef72009-07-30 16:03:45 -06001521 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001522 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1523 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001524
Rusty Russell2e04ef72009-07-30 16:03:45 -06001525 /*
1526 * We need a socket to perform the magic network ioctls to bring up the
1527 * tap interface, connect to the bridge etc. Any socket will do!
1528 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001529 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1530 if (ipfd < 0)
1531 err(1, "opening IP socket");
1532
Rusty Russelldde79782007-07-26 10:41:03 -07001533 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001534 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001535 arg += strlen(BRIDGE_PFX);
1536 bridging = true;
1537 }
1538
1539 /* A mac address may follow the bridge name or IP address */
1540 p = strchr(arg, ':');
1541 if (p) {
1542 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001543 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001544 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001545 }
1546
1547 /* arg is now either an IP address or a bridge name */
1548 if (bridging)
1549 add_to_bridge(ipfd, tapif, arg);
1550 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001551 ip = str2ip(arg);
1552
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001553 /* Set up the tun device. */
1554 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001555
Rusty Russell20887612008-05-30 15:09:46 -05001556 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
Rusty Russell398f1872008-07-29 09:58:37 -05001557 /* Expect Guest to handle everything except UFO */
1558 add_feature(dev, VIRTIO_NET_F_CSUM);
1559 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001560 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1561 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1562 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1563 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1564 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1565 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001566 /* We handle indirect ring entries */
1567 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russella586d4f2008-02-04 23:49:56 -05001568 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001569
Rusty Russella586d4f2008-02-04 23:49:56 -05001570 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001571 close(ipfd);
1572
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001573 devices.device_num++;
1574
1575 if (bridging)
1576 verbose("device %u: tun %s attached to bridge: %s\n",
1577 devices.device_num, tapif, arg);
1578 else
1579 verbose("device %u: tun %s: %s\n",
1580 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001581}
Rusty Russella91d74a2009-07-30 16:03:45 -06001582/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001583
Rusty Russelle1e72962007-10-25 15:02:50 +10001584/* This hangs off device->priv. */
Rusty Russell1842f232009-07-30 16:03:46 -06001585struct vblk_info {
Rusty Russell17cbca22007-10-22 11:24:22 +10001586 /* The size of the file. */
1587 off64_t len;
1588
1589 /* The file descriptor for the file. */
1590 int fd;
1591
Rusty Russell17cbca22007-10-22 11:24:22 +10001592};
1593
Rusty Russelle1e72962007-10-25 15:02:50 +10001594/*L:210
1595 * The Disk
1596 *
Rusty Russella91d74a2009-07-30 16:03:45 -06001597 * The disk only has one virtqueue, so it only has one thread. It is really
1598 * simple: the Guest asks for a block number and we read or write that position
1599 * in the file.
1600 *
1601 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1602 * slow: the Guest waits until the read is finished before running anything
1603 * else, even if it could have been doing useful work.
1604 *
1605 * We could have used async I/O, except it's reputed to suck so hard that
1606 * characters actually go missing from your code when you try to use it.
Rusty Russelle1e72962007-10-25 15:02:50 +10001607 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001608static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001609{
Rusty Russell659a0e62009-06-12 22:27:10 -06001610 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001611 unsigned int head, out_num, in_num, wlen;
1612 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001613 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001614 struct virtio_blk_outhdr *out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001615 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001616 off64_t off;
1617
Rusty Russella91d74a2009-07-30 16:03:45 -06001618 /*
1619 * Get the next request, where we normally wait. It triggers the
1620 * interrupt to acknowledge previously serviced requests (if any).
1621 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001622 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001623
Rusty Russell2e04ef72009-07-30 16:03:45 -06001624 /*
1625 * Every block request should contain at least one output buffer
Rusty Russelle1e72962007-10-25 15:02:50 +10001626 * (detailing the location on disk and the type of request) and one
Rusty Russell2e04ef72009-07-30 16:03:45 -06001627 * input buffer (to hold the result).
1628 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001629 if (out_num == 0 || in_num == 0)
1630 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1631 head, out_num, in_num);
1632
1633 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001634 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russella91d74a2009-07-30 16:03:45 -06001635 /*
1636 * For historical reasons, block operations are expressed in 512 byte
1637 * "sectors".
1638 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001639 off = out->sector * 512;
1640
Rusty Russell2e04ef72009-07-30 16:03:45 -06001641 /*
1642 * The block device implements "barriers", where the Guest indicates
Rusty Russelle1e72962007-10-25 15:02:50 +10001643 * that it wants all previous writes to occur before this write. We
1644 * don't have a way of asking our kernel to do a barrier, so we just
Rusty Russell2e04ef72009-07-30 16:03:45 -06001645 * synchronize all the data in the file. Pretty poor, no?
1646 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001647 if (out->type & VIRTIO_BLK_T_BARRIER)
1648 fdatasync(vblk->fd);
1649
Rusty Russell2e04ef72009-07-30 16:03:45 -06001650 /*
1651 * In general the virtio block driver is allowed to try SCSI commands.
1652 * It'd be nice if we supported eject, for example, but we don't.
1653 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001654 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1655 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001656 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001657 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001658 } else if (out->type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001659 /*
1660 * Write
1661 *
1662 * Move to the right location in the block file. This can fail
1663 * if they try to write past end.
1664 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001665 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1666 err(1, "Bad seek to sector %llu", out->sector);
1667
1668 ret = writev(vblk->fd, iov+1, out_num-1);
1669 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1670
Rusty Russell2e04ef72009-07-30 16:03:45 -06001671 /*
1672 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001673 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001674 * file (possibly extending it).
1675 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001676 if (ret > 0 && off + ret > vblk->len) {
1677 /* Trim it back to the correct length */
1678 ftruncate64(vblk->fd, vblk->len);
1679 /* Die, bad Guest, die. */
1680 errx(1, "Write past end %llu+%u", off, ret);
1681 }
Anthony Liguori1200e642007-11-08 21:13:44 -06001682 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001683 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001684 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001685 /*
1686 * Read
1687 *
1688 * Move to the right location in the block file. This can fail
1689 * if they try to read past end.
1690 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001691 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1692 err(1, "Bad seek to sector %llu", out->sector);
1693
1694 ret = readv(vblk->fd, iov+1, in_num-1);
1695 verbose("READ from sector %llu: %i\n", out->sector, ret);
1696 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001697 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001698 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001699 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001700 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001701 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001702 }
1703 }
1704
Rusty Russell2e04ef72009-07-30 16:03:45 -06001705 /*
1706 * OK, so we noted that it was pretty poor to use an fdatasync as a
Rusty Russelld1881d32009-03-30 21:55:25 -06001707 * barrier. But Christoph Hellwig points out that we need a sync
1708 * *afterwards* as well: "Barriers specify no reordering to the front
Rusty Russell2e04ef72009-07-30 16:03:45 -06001709 * or the back." And Jens Axboe confirmed it, so here we are:
1710 */
Rusty Russelld1881d32009-03-30 21:55:25 -06001711 if (out->type & VIRTIO_BLK_T_BARRIER)
1712 fdatasync(vblk->fd);
1713
Rusty Russella91d74a2009-07-30 16:03:45 -06001714 /* Finished that request. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001715 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001716}
1717
Rusty Russelle1e72962007-10-25 15:02:50 +10001718/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001719static void setup_block_file(const char *filename)
1720{
Rusty Russell17cbca22007-10-22 11:24:22 +10001721 struct device *dev;
1722 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001723 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001724
Rusty Russell2e04ef72009-07-30 16:03:45 -06001725 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001726 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001727
Rusty Russelle1e72962007-10-25 15:02:50 +10001728 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001729 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001730
1731 /* Allocate the room for our own bookkeeping */
1732 vblk = dev->priv = malloc(sizeof(*vblk));
1733
1734 /* First we open the file and store the length. */
1735 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1736 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1737
Rusty Russella586d4f2008-02-04 23:49:56 -05001738 /* We support barriers. */
1739 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1740
Rusty Russell17cbca22007-10-22 11:24:22 +10001741 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001742 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001743
Rusty Russell2e04ef72009-07-30 16:03:45 -06001744 /*
1745 * Tell Guest not to put in too many descriptors at once: two are used
1746 * for the in and out elements.
1747 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001748 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1749 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1750
Rusty Russell8ef562d2009-07-30 16:03:43 -06001751 /* Don't try to put whole struct: we have 8 bit limit. */
1752 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001753
Rusty Russell17cbca22007-10-22 11:24:22 +10001754 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001755 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001756}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001757
Rusty Russell2e04ef72009-07-30 16:03:45 -06001758/*L:211
1759 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001760 * input buffers. The usual case is that the Guest doesn't want random numbers
1761 * and so has no buffers although /dev/random is still readable, whereas
1762 * console is the reverse.
1763 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001764 * The same logic applies, however.
1765 */
1766struct rng_info {
1767 int rfd;
1768};
1769
Rusty Russell659a0e62009-06-12 22:27:10 -06001770static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001771{
1772 int len;
1773 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001774 struct rng_info *rng_info = vq->dev->priv;
1775 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001776
1777 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001778 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001779 if (out_num)
1780 errx(1, "Output buffers in rng?");
1781
Rusty Russell2e04ef72009-07-30 16:03:45 -06001782 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001783 * Just like the console write, we loop to cover the whole iovec.
1784 * In this case, short reads actually happen quite a bit.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001785 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001786 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001787 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001788 if (len <= 0)
1789 err(1, "Read from /dev/random gave %i", len);
1790 iov_consume(iov, in_num, len);
1791 totlen += len;
1792 }
1793
1794 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001795 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001796}
1797
Rusty Russell2e04ef72009-07-30 16:03:45 -06001798/*L:199
1799 * This creates a "hardware" random number device for the Guest.
1800 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001801static void setup_rng(void)
1802{
1803 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001804 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001805
Rusty Russell2e04ef72009-07-30 16:03:45 -06001806 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001807 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001808
Rusty Russell2e04ef72009-07-30 16:03:45 -06001809 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001810 dev = new_device("rng", VIRTIO_ID_RNG);
1811 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001812
1813 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001814 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001815
1816 verbose("device %u: rng\n", devices.device_num++);
1817}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001818/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301819
Rusty Russella6bd8e12008-03-28 11:05:53 -05001820/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301821static void __attribute__((noreturn)) restart_guest(void)
1822{
1823 unsigned int i;
1824
Rusty Russell2e04ef72009-07-30 16:03:45 -06001825 /*
1826 * Since we don't track all open fds, we simply close everything beyond
1827 * stderr.
1828 */
Balaji Raoec04b132007-12-28 14:26:24 +05301829 for (i = 3; i < FD_SETSIZE; i++)
1830 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001831
Rusty Russell659a0e62009-06-12 22:27:10 -06001832 /* Reset all the devices (kills all threads). */
1833 cleanup_devices();
1834
Balaji Raoec04b132007-12-28 14:26:24 +05301835 execv(main_args[0], main_args);
1836 err(1, "Could not exec %s", main_args[0]);
1837}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001838
Rusty Russell2e04ef72009-07-30 16:03:45 -06001839/*L:220
1840 * Finally we reach the core of the Launcher which runs the Guest, serves
1841 * its input and output, and finally, lays it to rest.
1842 */
Rusty Russell56739c802009-06-12 22:26:59 -06001843static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001844{
1845 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001846 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001847 int readval;
1848
1849 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001850 readval = pread(lguest_fd, &notify_addr,
1851 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001852
Rusty Russell17cbca22007-10-22 11:24:22 +10001853 /* One unsigned long means the Guest did HCALL_NOTIFY */
1854 if (readval == sizeof(notify_addr)) {
1855 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c802009-06-12 22:26:59 -06001856 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001857 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001858 } else if (errno == ENOENT) {
1859 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001860 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001861 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301862 /* ERESTART means that we need to reboot the guest */
1863 } else if (errno == ERESTART) {
1864 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001865 /* Anything else means a bug or incompatible change. */
1866 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001867 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001868 }
1869}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001870/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001871 * This is the end of the Launcher. The good news: we are over halfway
1872 * through! The bad news: the most fiendish part of the code still lies ahead
1873 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001874 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001875 * Are you ready? Take a deep breath and join me in the core of the Host, in
1876 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001877:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001878
1879static struct option opts[] = {
1880 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001881 { "tunnet", 1, NULL, 't' },
1882 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001883 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001884 { "initrd", 1, NULL, 'i' },
1885 { NULL },
1886};
1887static void usage(void)
1888{
1889 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001890 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001891 "|--block=<filename>|--initrd=<filename>]...\n"
1892 "<mem-in-mb> vmlinux [args...]");
1893}
1894
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001895/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001896int main(int argc, char *argv[])
1897{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001898 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001899 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c802009-06-12 22:26:59 -06001900 /* Two temporaries. */
1901 int i, c;
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;
Balaji Raoec04b132007-12-28 14:26:24 +05301909
Rusty Russell2e04ef72009-07-30 16:03:45 -06001910 /*
1911 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001912 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001913 * remember that 0 is used by the timer).
1914 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001915 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001916 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001917
Rusty Russella91d74a2009-07-30 16:03:45 -06001918 /* We're CPU 0. In fact, that's the only CPU possible right now. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001919 cpu_id = 0;
Rusty Russella91d74a2009-07-30 16:03:45 -06001920
Rusty Russell2e04ef72009-07-30 16:03:45 -06001921 /*
1922 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001923 * descriptor and memory pages for the devices as we parse the command
1924 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001925 * of memory now.
1926 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001927 for (i = 1; i < argc; i++) {
1928 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001929 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001930 /*
1931 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001932 * guest-physical memory range. This fills it with 0,
1933 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001934 * tries to access it.
1935 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001936 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 Russell2e04ef72009-07-30 16:03:45 -06001968 /*
1969 * After the other arguments we expect memory and kernel image name,
1970 * followed by command line arguments for the kernel.
1971 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001972 if (optind + 2 > argc)
1973 usage();
1974
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001975 verbose("Guest base is at %p\n", guest_base);
1976
Rusty Russelldde79782007-07-26 10:41:03 -07001977 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001978 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001979
Rusty Russell8ca47e02007-07-19 01:49:29 -07001980 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001981 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001982
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001983 /* Boot information is stashed at physical address 0 */
1984 boot = from_guest_phys(0);
1985
Rusty Russelldde79782007-07-26 10:41:03 -07001986 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001987 if (initrd_name) {
1988 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06001989 /*
1990 * These are the location in the Linux boot header where the
1991 * start and size of the initrd are expected to be found.
1992 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001993 boot->hdr.ramdisk_image = mem - initrd_size;
1994 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001995 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001996 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001997 }
1998
Rusty Russell2e04ef72009-07-30 16:03:45 -06001999 /*
2000 * The Linux boot header contains an "E820" memory map: ours is a
2001 * simple, single region.
2002 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002003 boot->e820_entries = 1;
2004 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06002005 /*
2006 * The boot header contains a command line pointer: we put the command
2007 * line after the boot header.
2008 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002009 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10002010 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002011 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07002012
Rusty Russell814a0e52007-10-22 11:29:44 +10002013 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002014 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002015
2016 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002017 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002018
Rusty Russell43d33b22007-10-22 11:29:57 +10002019 /* Tell the entry path not to try to reload segment registers. */
2020 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002021
Rusty Russell2e04ef72009-07-30 16:03:45 -06002022 /*
2023 * We tell the kernel to initialize the Guest: this returns the open
2024 * /dev/lguest file descriptor.
2025 */
Rusty Russell56739c802009-06-12 22:26:59 -06002026 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002027
Rusty Russella91d74a2009-07-30 16:03:45 -06002028 /* Ensure that we terminate if a device-servicing child dies. */
Rusty Russell659a0e62009-06-12 22:27:10 -06002029 signal(SIGCHLD, kill_launcher);
2030
2031 /* If we exit via err(), this kills all the threads, restores tty. */
2032 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002033
Rusty Russelldde79782007-07-26 10:41:03 -07002034 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c802009-06-12 22:26:59 -06002035 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07002036}
Rusty Russellf56a3842007-07-26 10:41:05 -07002037/*:*/
2038
2039/*M:999
2040 * Mastery is done: you now know everything I do.
2041 *
2042 * But surely you have seen code, features and bugs in your wanderings which
2043 * you now yearn to attack? That is the real game, and I look forward to you
2044 * patching and forking lguest into the Your-Name-Here-visor.
2045 *
2046 * Farewell, and good coding!
2047 * Rusty Russell.
2048 */