blob: c095d79cae73d1acf8d2ae9991661be35be008bd [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>
Philip Sanderson8aeb36e2011-01-20 21:37:28 -060042#include <pwd.h>
43#include <grp.h>
44
Rusty Russellf8466192010-08-27 08:39:48 -060045#include <linux/virtio_config.h>
46#include <linux/virtio_net.h>
47#include <linux/virtio_blk.h>
48#include <linux/virtio_console.h>
49#include <linux/virtio_rng.h>
50#include <linux/virtio_ring.h>
51#include <asm/bootparam.h>
Rusty Russellbc805a02011-05-30 11:14:11 -060052#include "../../../include/linux/lguest_launcher.h"
Rusty Russell2e04ef72009-07-30 16:03:45 -060053/*L:110
Rusty Russell9f542882011-07-22 14:39:50 +093054 * We can ignore the 43 include files we need for this program, but I do want
Rusty Russell2e04ef72009-07-30 16:03:45 -060055 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100056 *
57 * As Linus said, "C is a Spartan language, and so should your naming be." I
58 * like these abbreviations, so we define them here. Note that u64 is always
59 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060060 * use %llu in printf for any u64.
61 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100062typedef unsigned long long u64;
63typedef uint32_t u32;
64typedef uint16_t u16;
65typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070066/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070067
Rusty Russell8ca47e02007-07-19 01:49:29 -070068#define BRIDGE_PFX "bridge:"
69#ifndef SIOCBRADDIF
70#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
71#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100072/* We can have up to 256 pages for devices. */
73#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050074/* This will occupy 3 pages: it must be a power of 2. */
75#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070076
Rusty Russell2e04ef72009-07-30 16:03:45 -060077/*L:120
78 * verbose is both a global flag and a macro. The C preprocessor allows
79 * this, and although I wouldn't recommend it, it works quite nicely here.
80 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070081static bool verbose;
82#define verbose(args...) \
83 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070084/*:*/
85
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100086/* The pointer to the start of guest memory. */
87static void *guest_base;
88/* The maximum guest physical address allowed, and maximum possible. */
89static unsigned long guest_limit, guest_max;
Rusty Russell56739c802009-06-12 22:26:59 -060090/* The /dev/lguest file descriptor. */
91static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070092
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020093/* a per-cpu variable indicating whose vcpu is currently running */
94static unsigned int __thread cpu_id;
95
Rusty Russelldde79782007-07-26 10:41:03 -070096/* This is our list of devices. */
Rusty Russell1842f232009-07-30 16:03:46 -060097struct device_list {
Rusty Russell17cbca22007-10-22 11:24:22 +100098 /* Counter to assign interrupt numbers. */
99 unsigned int next_irq;
100
101 /* Counter to print out convenient device numbers. */
102 unsigned int device_num;
103
Rusty Russelldde79782007-07-26 10:41:03 -0700104 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000105 u8 *descpage;
106
Rusty Russelldde79782007-07-26 10:41:03 -0700107 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700108 struct device *dev;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600109 /* And a pointer to the last device for easy append. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500110 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700111};
112
Rusty Russell17cbca22007-10-22 11:24:22 +1000113/* The list of Guest devices, based on command line arguments. */
114static struct device_list devices;
115
Rusty Russelldde79782007-07-26 10:41:03 -0700116/* The device structure describes a single device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600117struct device {
Rusty Russelldde79782007-07-26 10:41:03 -0700118 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700119 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000120
Rusty Russell713b15b2009-06-12 22:26:58 -0600121 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700122 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000123
Rusty Russell713b15b2009-06-12 22:26:58 -0600124 /* We can't trust desc values once Guest has booted: we use these. */
125 unsigned int feature_len;
126 unsigned int num_vq;
127
Rusty Russell17cbca22007-10-22 11:24:22 +1000128 /* The name of this device, for --verbose. */
129 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700130
Rusty Russell17cbca22007-10-22 11:24:22 +1000131 /* Any queues attached to this device */
132 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700133
Rusty Russell659a0e62009-06-12 22:27:10 -0600134 /* Is it operational */
135 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500136
Rusty Russell8ca47e02007-07-19 01:49:29 -0700137 /* Device-specific data. */
138 void *priv;
139};
140
Rusty Russell17cbca22007-10-22 11:24:22 +1000141/* The virtqueue structure describes a queue attached to a device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600142struct virtqueue {
Rusty Russell17cbca22007-10-22 11:24:22 +1000143 struct virtqueue *next;
144
145 /* Which device owns me. */
146 struct device *dev;
147
148 /* The configuration for this queue. */
149 struct lguest_vqconfig config;
150
151 /* The actual ring of buffers. */
152 struct vring vring;
153
154 /* Last available index we saw. */
155 u16 last_avail_idx;
156
Rusty Russell95c517c2009-06-12 22:27:11 -0600157 /* How many are used since we sent last irq? */
158 unsigned int pending_used;
159
Rusty Russell659a0e62009-06-12 22:27:10 -0600160 /* Eventfd where Guest notifications arrive. */
161 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500162
Rusty Russell659a0e62009-06-12 22:27:10 -0600163 /* Function for the thread which is servicing this virtqueue. */
164 void (*service)(struct virtqueue *vq);
165 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000166};
167
Balaji Raoec04b132007-12-28 14:26:24 +0530168/* Remember the arguments to the program so we can "reboot" */
169static char **main_args;
170
Rusty Russell659a0e62009-06-12 22:27:10 -0600171/* The original tty settings to restore on exit. */
172static struct termios orig_term;
173
Rusty Russell2e04ef72009-07-30 16:03:45 -0600174/*
175 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600176 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600177 * in precise order.
178 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600179#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russellb60da132009-06-12 22:27:12 -0600180#define mb() __asm__ __volatile__("" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000181
Rusty Russell2e04ef72009-07-30 16:03:45 -0600182/*
183 * Convert an iovec element to the given type.
Rusty Russell17cbca22007-10-22 11:24:22 +1000184 *
185 * This is a fairly ugly trick: we need to know the size of the type and
186 * alignment requirement to check the pointer is kosher. It's also nice to
187 * have the name of the type in case we report failure.
188 *
189 * Typing those three things all the time is cumbersome and error prone, so we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600190 * have a macro which sets them all up and passes to the real function.
191 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000192#define convert(iov, type) \
193 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
194
195static void *_convert(struct iovec *iov, size_t size, size_t align,
196 const char *name)
197{
198 if (iov->iov_len != size)
199 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
200 if ((unsigned long)iov->iov_base % align != 0)
201 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
202 return iov->iov_base;
203}
204
Rusty Russellb5111792008-07-29 09:58:34 -0500205/* Wrapper for the last available index. Makes it easier to change. */
206#define lg_last_avail(vq) ((vq)->last_avail_idx)
207
Rusty Russell2e04ef72009-07-30 16:03:45 -0600208/*
209 * The virtio configuration space is defined to be little-endian. x86 is
210 * little-endian too, but it's nice to be explicit so we have these helpers.
211 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000212#define cpu_to_le16(v16) (v16)
213#define cpu_to_le32(v32) (v32)
214#define cpu_to_le64(v64) (v64)
215#define le16_to_cpu(v16) (v16)
216#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500217#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000218
Rusty Russell28fd6d72008-07-29 09:58:33 -0500219/* Is this iovec empty? */
220static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
221{
222 unsigned int i;
223
224 for (i = 0; i < num_iov; i++)
225 if (iov[i].iov_len)
226 return false;
227 return true;
228}
229
230/* Take len bytes from the front of this iovec. */
231static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
232{
233 unsigned int i;
234
235 for (i = 0; i < num_iov; i++) {
236 unsigned int used;
237
238 used = iov[i].iov_len < len ? iov[i].iov_len : len;
239 iov[i].iov_base += used;
240 iov[i].iov_len -= used;
241 len -= used;
242 }
243 assert(len == 0);
244}
245
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500246/* The device virtqueue descriptors are followed by feature bitmasks. */
247static u8 *get_feature_bits(struct device *dev)
248{
249 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600250 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500251}
252
Rusty Russell2e04ef72009-07-30 16:03:45 -0600253/*L:100
254 * The Launcher code itself takes us out into userspace, that scary place where
255 * pointers run wild and free! Unfortunately, like most userspace programs,
256 * it's quite boring (which is why everyone likes to hack on the kernel!).
257 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
258 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000259 *
260 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
261 * memory and stores it in "guest_base". In other words, Guest physical ==
262 * Launcher virtual with an offset.
263 *
264 * This can be tough to get your head around, but usually it just means that we
Francis Galieguea33f3222010-04-23 00:08:02 +0200265 * use these trivial conversion functions when the Guest gives us its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600266 * "physical" addresses:
267 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000268static void *from_guest_phys(unsigned long addr)
269{
270 return guest_base + addr;
271}
272
273static unsigned long to_guest_phys(const void *addr)
274{
275 return (addr - guest_base);
276}
277
Rusty Russelldde79782007-07-26 10:41:03 -0700278/*L:130
279 * Loading the Kernel.
280 *
281 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600282 * error-checking code cluttering the callers:
283 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700284static int open_or_die(const char *name, int flags)
285{
286 int fd = open(name, flags);
287 if (fd < 0)
288 err(1, "Failed to open %s", name);
289 return fd;
290}
291
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000292/* map_zeroed_pages() takes a number of pages. */
293static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700294{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000295 int fd = open_or_die("/dev/zero", O_RDONLY);
296 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700297
Rusty Russell2e04ef72009-07-30 16:03:45 -0600298 /*
299 * We use a private mapping (ie. if we write to the page, it will be
Philip Sanderson5230ff02011-01-20 21:37:28 -0600300 * copied). We allocate an extra two pages PROT_NONE to act as guard
301 * pages against read/write attempts that exceed allocated space.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600302 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600303 addr = mmap(NULL, getpagesize() * (num+2),
304 PROT_NONE, MAP_PRIVATE, fd, 0);
305
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000306 if (addr == MAP_FAILED)
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200307 err(1, "Mmapping %u pages of /dev/zero", num);
Rusty Russella91d74a2009-07-30 16:03:45 -0600308
Philip Sanderson5230ff02011-01-20 21:37:28 -0600309 if (mprotect(addr + getpagesize(), getpagesize() * num,
310 PROT_READ|PROT_WRITE) == -1)
311 err(1, "mprotect rw %u pages failed", num);
312
Rusty Russella91d74a2009-07-30 16:03:45 -0600313 /*
314 * One neat mmap feature is that you can close the fd, and it
315 * stays mapped.
316 */
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100317 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700318
Philip Sanderson5230ff02011-01-20 21:37:28 -0600319 /* Return address after PROT_NONE page */
320 return addr + getpagesize();
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000321}
322
323/* Get some more pages for a device. */
324static void *get_pages(unsigned int num)
325{
326 void *addr = from_guest_phys(guest_limit);
327
328 guest_limit += num * getpagesize();
329 if (guest_limit > guest_max)
330 errx(1, "Not enough memory for devices");
331 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700332}
333
Rusty Russell2e04ef72009-07-30 16:03:45 -0600334/*
335 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700336 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600337 * it falls back to reading the memory in.
338 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700339static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
340{
341 ssize_t r;
342
Rusty Russell2e04ef72009-07-30 16:03:45 -0600343 /*
344 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700345 * The kernel really wants to be writable: it patches its own
346 * instructions.
347 *
348 * MAP_PRIVATE means that the page won't be copied until a write is
349 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600350 * Guests.
351 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600352 if (mmap(addr, len, PROT_READ|PROT_WRITE,
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700353 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
354 return;
355
356 /* pread does a seek and a read in one shot: saves a few lines. */
357 r = pread(fd, addr, len, offset);
358 if (r != len)
359 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
360}
361
Rusty Russell2e04ef72009-07-30 16:03:45 -0600362/*
363 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700364 * the Guest memory. ELF = Embedded Linking Format, which is the format used
365 * by all modern binaries on Linux including the kernel.
366 *
367 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000368 * address. We use the physical address; the Guest will map itself to the
369 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700370 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600371 * We return the starting address.
372 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000373static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700374{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700375 Elf32_Phdr phdr[ehdr->e_phnum];
376 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700377
Rusty Russell2e04ef72009-07-30 16:03:45 -0600378 /*
379 * Sanity checks on the main ELF header: an x86 executable with a
380 * reasonable number of correctly-sized program headers.
381 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700382 if (ehdr->e_type != ET_EXEC
383 || ehdr->e_machine != EM_386
384 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
385 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
386 errx(1, "Malformed elf header");
387
Rusty Russell2e04ef72009-07-30 16:03:45 -0600388 /*
389 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700390 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600391 * load where.
392 */
Rusty Russelldde79782007-07-26 10:41:03 -0700393
394 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700395 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
396 err(1, "Seeking to program headers");
397 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
398 err(1, "Reading program headers");
399
Rusty Russell2e04ef72009-07-30 16:03:45 -0600400 /*
401 * Try all the headers: there are usually only three. A read-only one,
402 * a read-write one, and a "note" section which we don't load.
403 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700404 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700405 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700406 if (phdr[i].p_type != PT_LOAD)
407 continue;
408
409 verbose("Section %i: size %i addr %p\n",
410 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
411
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700412 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000413 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700414 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700415 }
416
Rusty Russell814a0e52007-10-22 11:29:44 +1000417 /* The entry point is given in the ELF header. */
418 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700419}
420
Rusty Russell2e04ef72009-07-30 16:03:45 -0600421/*L:150
422 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
423 * to jump into it and it will unpack itself. We used to have to perform some
424 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700425 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000426 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
427 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600428 * the funky header so we know where in the file to load, and away we go!
429 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000430static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700431{
Rusty Russell43d33b22007-10-22 11:29:57 +1000432 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000433 int r;
434 /* Modern bzImages get loaded at 1M. */
435 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700436
Rusty Russell2e04ef72009-07-30 16:03:45 -0600437 /*
438 * Go back to the start of the file and read the header. It should be
Paul Bolle395cf962011-08-15 02:02:26 +0200439 * a Linux boot header (see Documentation/x86/boot.txt)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600440 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000441 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000442 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000443
Rusty Russell43d33b22007-10-22 11:29:57 +1000444 /* Inside the setup_hdr, we expect the magic "HdrS" */
445 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000446 errx(1, "This doesn't look like a bzImage to me");
447
Rusty Russell43d33b22007-10-22 11:29:57 +1000448 /* Skip over the extra sectors of the header. */
449 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000450
451 /* Now read everything into memory. in nice big chunks. */
452 while ((r = read(fd, p, 65536)) > 0)
453 p += r;
454
Rusty Russell43d33b22007-10-22 11:29:57 +1000455 /* Finally, code32_start tells us where to enter the kernel. */
456 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700457}
458
Rusty Russell2e04ef72009-07-30 16:03:45 -0600459/*L:140
460 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000461 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600462 * work, we can load those, too.
463 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000464static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700465{
466 Elf32_Ehdr hdr;
467
Rusty Russelldde79782007-07-26 10:41:03 -0700468 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700469 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
470 err(1, "Reading kernel");
471
Rusty Russelldde79782007-07-26 10:41:03 -0700472 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700473 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000474 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700475
Rusty Russella6bd8e12008-03-28 11:05:53 -0500476 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000477 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700478}
479
Rusty Russell2e04ef72009-07-30 16:03:45 -0600480/*
481 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700482 * it calls getpagesize() twice: "it's dumb code."
483 *
484 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600485 * necessary. I leave this code as a reaction against that.
486 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700487static inline unsigned long page_align(unsigned long addr)
488{
Rusty Russelldde79782007-07-26 10:41:03 -0700489 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700490 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
491}
492
Rusty Russell2e04ef72009-07-30 16:03:45 -0600493/*L:180
494 * An "initial ram disk" is a disk image loaded into memory along with the
495 * kernel which the kernel can use to boot from without needing any drivers.
496 * Most distributions now use this as standard: the initrd contains the code to
497 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700498 *
499 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600500 * kernels. He sent me this (and tells me when I break it).
501 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700502static unsigned long load_initrd(const char *name, unsigned long mem)
503{
504 int ifd;
505 struct stat st;
506 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700507
508 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700509 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700510 if (fstat(ifd, &st) < 0)
511 err(1, "fstat() on initrd '%s'", name);
512
Rusty Russell2e04ef72009-07-30 16:03:45 -0600513 /*
514 * We map the initrd at the top of memory, but mmap wants it to be
515 * page-aligned, so we round the size up for that.
516 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700517 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000518 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600519 /*
520 * Once a file is mapped, you can close the file descriptor. It's a
521 * little odd, but quite useful.
522 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700523 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700524 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700525
526 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700527 return len;
528}
Rusty Russelle1e72962007-10-25 15:02:50 +1000529/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700530
Rusty Russell2e04ef72009-07-30 16:03:45 -0600531/*
532 * Simple routine to roll all the commandline arguments together with spaces
533 * between them.
534 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700535static void concat(char *dst, char *args[])
536{
537 unsigned int i, len = 0;
538
539 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100540 if (i) {
541 strcat(dst+len, " ");
542 len++;
543 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700544 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100545 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700546 }
547 /* In case it's empty. */
548 dst[len] = '\0';
549}
550
Rusty Russell2e04ef72009-07-30 16:03:45 -0600551/*L:185
552 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000553 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300554 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600555 * entry point for the Guest.
556 */
Rusty Russell56739c802009-06-12 22:26:59 -0600557static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700558{
Jes Sorensen511801d2007-10-22 11:03:31 +1000559 unsigned long args[] = { LHREQ_INITIALIZE,
560 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300561 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000562 verbose("Guest: %p - %p (%#lx)\n",
563 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c802009-06-12 22:26:59 -0600564 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
565 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700566 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700567}
Rusty Russelldde79782007-07-26 10:41:03 -0700568/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700569
Rusty Russella91d74a2009-07-30 16:03:45 -0600570/*L:200
Rusty Russelldde79782007-07-26 10:41:03 -0700571 * Device Handling.
572 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000573 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700574 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000575 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700576 * if something funny is going on:
577 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700578static void *_check_pointer(unsigned long addr, unsigned int size,
579 unsigned int line)
580{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600581 /*
Philip Sanderson5230ff02011-01-20 21:37:28 -0600582 * Check if the requested address and size exceeds the allocated memory,
583 * or addr + size wraps around.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600584 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600585 if ((addr + size) > guest_limit || (addr + size) < addr)
Rusty Russell17cbca22007-10-22 11:24:22 +1000586 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600587 /*
588 * We return a pointer for the caller's convenience, now we know it's
589 * safe to use.
590 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000591 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700592}
Rusty Russelldde79782007-07-26 10:41:03 -0700593/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700594#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
595
Rusty Russell2e04ef72009-07-30 16:03:45 -0600596/*
597 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000598 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600599 * at the end.
600 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100601static unsigned next_desc(struct vring_desc *desc,
602 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000603{
604 unsigned int next;
605
606 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100607 if (!(desc[i].flags & VRING_DESC_F_NEXT))
608 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000609
610 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100611 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000612 /* Make sure compiler knows to grab that: we don't want it changing! */
613 wmb();
614
Mark McLoughlind1f01322009-05-11 18:11:46 +0100615 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000616 errx(1, "Desc next is %u", next);
617
618 return next;
619}
620
Rusty Russella91d74a2009-07-30 16:03:45 -0600621/*
622 * This actually sends the interrupt for this virtqueue, if we've used a
623 * buffer.
624 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600625static void trigger_irq(struct virtqueue *vq)
626{
627 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
628
Rusty Russell95c517c2009-06-12 22:27:11 -0600629 /* Don't inform them if nothing used. */
630 if (!vq->pending_used)
631 return;
632 vq->pending_used = 0;
633
Rusty Russellca60a422009-09-23 22:26:47 -0600634 /* If they don't want an interrupt, don't send one... */
635 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
Rusty Russell990c91f2011-05-30 11:14:12 -0600636 return;
Rusty Russellca60a422009-09-23 22:26:47 -0600637 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600638
639 /* Send the Guest an interrupt tell them we used something up. */
640 if (write(lguest_fd, buf, sizeof(buf)) != 0)
641 err(1, "Triggering irq %i", vq->config.irq);
642}
643
Rusty Russell2e04ef72009-07-30 16:03:45 -0600644/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600645 * This looks in the virtqueue for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000646 * it to an iovec for convenient access. Since descriptors consist of some
647 * number of output then some number of input descriptors, it's actually two
648 * iovecs, but we pack them into one and note how many of each there were.
649 *
Rusty Russella91d74a2009-07-30 16:03:45 -0600650 * This function waits if necessary, and returns the descriptor number found.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600651 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600652static unsigned wait_for_vq_desc(struct virtqueue *vq,
653 struct iovec iov[],
654 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000655{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100656 unsigned int i, head, max;
657 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600658 u16 last_avail = lg_last_avail(vq);
659
Rusty Russella91d74a2009-07-30 16:03:45 -0600660 /* There's nothing available? */
Rusty Russell659a0e62009-06-12 22:27:10 -0600661 while (last_avail == vq->vring.avail->idx) {
662 u64 event;
663
Rusty Russella91d74a2009-07-30 16:03:45 -0600664 /*
665 * Since we're about to sleep, now is a good time to tell the
666 * Guest about what we've used up to now.
667 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600668 trigger_irq(vq);
669
Rusty Russellb60da132009-06-12 22:27:12 -0600670 /* OK, now we need to know about added descriptors. */
671 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
672
Rusty Russell2e04ef72009-07-30 16:03:45 -0600673 /*
674 * They could have slipped one in as we were doing that: make
675 * sure it's written, then check again.
676 */
Rusty Russellb60da132009-06-12 22:27:12 -0600677 mb();
678 if (last_avail != vq->vring.avail->idx) {
679 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
680 break;
681 }
682
Rusty Russell659a0e62009-06-12 22:27:10 -0600683 /* Nothing new? Wait for eventfd to tell us they refilled. */
684 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
685 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600686
687 /* We don't need to be notified again. */
688 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600689 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000690
691 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500692 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000693 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500694 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000695
Rusty Russell2e04ef72009-07-30 16:03:45 -0600696 /*
697 * Grab the next descriptor number they're advertising, and increment
698 * the index we've seen.
699 */
Rusty Russellb5111792008-07-29 09:58:34 -0500700 head = vq->vring.avail->ring[last_avail % vq->vring.num];
701 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000702
703 /* If their number is silly, that's a fatal mistake. */
704 if (head >= vq->vring.num)
705 errx(1, "Guest says index %u is available", head);
706
707 /* When we start there are none of either input nor output. */
708 *out_num = *in_num = 0;
709
Mark McLoughlind1f01322009-05-11 18:11:46 +0100710 max = vq->vring.num;
711 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000712 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100713
Rusty Russell2e04ef72009-07-30 16:03:45 -0600714 /*
715 * If this is an indirect entry, then this buffer contains a descriptor
716 * table which we handle as if it's any normal descriptor chain.
717 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100718 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
719 if (desc[i].len % sizeof(struct vring_desc))
720 errx(1, "Invalid size for indirect buffer table");
721
722 max = desc[i].len / sizeof(struct vring_desc);
723 desc = check_pointer(desc[i].addr, desc[i].len);
724 i = 0;
725 }
726
Rusty Russell17cbca22007-10-22 11:24:22 +1000727 do {
728 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100729 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000730 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100731 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000732 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100733 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000734 (*in_num)++;
735 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600736 /*
737 * If it's an output descriptor, they're all supposed
738 * to come before any input descriptors.
739 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000740 if (*in_num)
741 errx(1, "Descriptor has out after in");
742 (*out_num)++;
743 }
744
745 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100746 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000747 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100748 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000749
750 return head;
751}
752
Rusty Russell2e04ef72009-07-30 16:03:45 -0600753/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600754 * After we've used one of their buffers, we tell the Guest about it. Sometime
755 * later we'll want to send them an interrupt using trigger_irq(); note that
756 * wait_for_vq_desc() does that for us if it has to wait.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600757 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000758static void add_used(struct virtqueue *vq, unsigned int head, int len)
759{
760 struct vring_used_elem *used;
761
Rusty Russell2e04ef72009-07-30 16:03:45 -0600762 /*
763 * The virtqueue contains a ring of used buffers. Get a pointer to the
764 * next entry in that used ring.
765 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000766 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
767 used->id = head;
768 used->len = len;
769 /* Make sure buffer is written before we update index. */
770 wmb();
771 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600772 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000773}
774
Rusty Russell17cbca22007-10-22 11:24:22 +1000775/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c802009-06-12 22:26:59 -0600776static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000777{
778 add_used(vq, head, len);
Rusty Russell56739c802009-06-12 22:26:59 -0600779 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000780}
781
Rusty Russelle1e72962007-10-25 15:02:50 +1000782/*
783 * The Console
784 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600785 * We associate some data with the console for our exit hack.
786 */
Rusty Russell1842f232009-07-30 16:03:46 -0600787struct console_abort {
Rusty Russelldde79782007-07-26 10:41:03 -0700788 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700789 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700790 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700791 struct timeval start;
792};
793
Rusty Russelldde79782007-07-26 10:41:03 -0700794/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600795static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700796{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700797 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000798 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600799 struct console_abort *abort = vq->dev->priv;
800 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700801
Rusty Russella91d74a2009-07-30 16:03:45 -0600802 /* Make sure there's a descriptor available. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600803 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000804 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000805 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700806
Rusty Russella91d74a2009-07-30 16:03:45 -0600807 /* Read into it. This is where we usually wait. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600808 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700809 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600810 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700811 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600812 /*
813 * For simplicity, dying threads kill the whole Launcher. So
814 * just nap here.
815 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600816 for (;;)
817 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700818 }
819
Rusty Russella91d74a2009-07-30 16:03:45 -0600820 /* Tell the Guest we used a buffer. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600821 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700822
Rusty Russell2e04ef72009-07-30 16:03:45 -0600823 /*
824 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700825 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600826 * This is such a hack, but works surprisingly well. Each ^C has to
827 * be in a buffer by itself, so they can't be too fast. But we check
828 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600829 * slow.
830 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600831 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700832 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600833 return;
834 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700835
Rusty Russell659a0e62009-06-12 22:27:10 -0600836 abort->count++;
837 if (abort->count == 1)
838 gettimeofday(&abort->start, NULL);
839 else if (abort->count == 3) {
840 struct timeval now;
841 gettimeofday(&now, NULL);
842 /* Kill all Launcher processes with SIGINT, like normal ^C */
843 if (now.tv_sec <= abort->start.tv_sec+1)
844 kill(0, SIGINT);
845 abort->count = 0;
846 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700847}
848
Rusty Russell659a0e62009-06-12 22:27:10 -0600849/* This is the routine which handles console output (ie. stdout). */
850static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700851{
Rusty Russell17cbca22007-10-22 11:24:22 +1000852 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000853 struct iovec iov[vq->vring.num];
854
Rusty Russella91d74a2009-07-30 16:03:45 -0600855 /* We usually wait in here, for the Guest to give us something. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600856 head = wait_for_vq_desc(vq, iov, &out, &in);
857 if (in)
858 errx(1, "Input buffers in console output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600859
860 /* writev can return a partial write, so we loop here. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600861 while (!iov_empty(iov, out)) {
862 int len = writev(STDOUT_FILENO, iov, out);
Sakari Ailuse0377e22011-06-26 19:36:46 +0300863 if (len <= 0) {
864 warn("Write to stdout gave %i (%d)", len, errno);
865 break;
866 }
Rusty Russell659a0e62009-06-12 22:27:10 -0600867 iov_consume(iov, out, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000868 }
Rusty Russella91d74a2009-07-30 16:03:45 -0600869
870 /*
871 * We're finished with that buffer: if we're going to sleep,
872 * wait_for_vq_desc() will prod the Guest with an interrupt.
873 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600874 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500875}
876
Rusty Russelle1e72962007-10-25 15:02:50 +1000877/*
878 * The Network
879 *
880 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600881 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500882 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600883struct net_info {
884 int tunfd;
885};
886
887static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700888{
Rusty Russell659a0e62009-06-12 22:27:10 -0600889 struct net_info *net_info = vq->dev->priv;
890 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000891 struct iovec iov[vq->vring.num];
892
Rusty Russella91d74a2009-07-30 16:03:45 -0600893 /* We usually wait in here for the Guest to give us a packet. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600894 head = wait_for_vq_desc(vq, iov, &out, &in);
895 if (in)
896 errx(1, "Input buffers in net output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600897 /*
898 * Send the whole thing through to /dev/net/tun. It expects the exact
899 * same format: what a coincidence!
900 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600901 if (writev(net_info->tunfd, iov, out) < 0)
Sakari Ailuse0377e22011-06-26 19:36:46 +0300902 warnx("Write to tun failed (%d)?", errno);
Rusty Russella91d74a2009-07-30 16:03:45 -0600903
904 /*
905 * Done with that one; wait_for_vq_desc() will send the interrupt if
906 * all packets are processed.
907 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600908 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700909}
910
Rusty Russella91d74a2009-07-30 16:03:45 -0600911/*
912 * Handling network input is a bit trickier, because I've tried to optimize it.
913 *
914 * First we have a helper routine which tells is if from this file descriptor
915 * (ie. the /dev/net/tun device) will block:
916 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600917static bool will_block(int fd)
918{
919 fd_set fdset;
920 struct timeval zero = { 0, 0 };
921 FD_ZERO(&fdset);
922 FD_SET(fd, &fdset);
923 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
924}
925
Rusty Russella91d74a2009-07-30 16:03:45 -0600926/*
927 * This handles packets coming in from the tun device to our Guest. Like all
928 * service routines, it gets called again as soon as it returns, so you don't
929 * see a while(1) loop here.
930 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600931static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700932{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700933 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600934 unsigned int head, out, in;
935 struct iovec iov[vq->vring.num];
936 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700937
Rusty Russella91d74a2009-07-30 16:03:45 -0600938 /*
939 * Get a descriptor to write an incoming packet into. This will also
940 * send an interrupt if they're out of descriptors.
941 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600942 head = wait_for_vq_desc(vq, iov, &out, &in);
943 if (out)
944 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600945
Rusty Russella91d74a2009-07-30 16:03:45 -0600946 /*
947 * If it looks like we'll block reading from the tun device, send them
948 * an interrupt.
949 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600950 if (vq->pending_used && will_block(net_info->tunfd))
951 trigger_irq(vq);
952
Rusty Russella91d74a2009-07-30 16:03:45 -0600953 /*
954 * Read in the packet. This is where we normally wait (when there's no
955 * incoming network traffic).
956 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600957 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700958 if (len <= 0)
Sakari Ailuse0377e22011-06-26 19:36:46 +0300959 warn("Failed to read from tun (%d).", errno);
Rusty Russella91d74a2009-07-30 16:03:45 -0600960
961 /*
962 * Mark that packet buffer as used, but don't interrupt here. We want
963 * to wait until we've done as much work as we can.
964 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600965 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700966}
Rusty Russella91d74a2009-07-30 16:03:45 -0600967/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700968
Rusty Russella91d74a2009-07-30 16:03:45 -0600969/* This is the helper to create threads: run the service routine in a loop. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600970static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000971{
Rusty Russell659a0e62009-06-12 22:27:10 -0600972 struct virtqueue *vq = _vq;
973
974 for (;;)
975 vq->service(vq);
976 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000977}
978
Rusty Russell2e04ef72009-07-30 16:03:45 -0600979/*
980 * When a child dies, we kill our entire process group with SIGTERM. This
981 * also has the side effect that the shell restores the console for us!
982 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600983static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500984{
Rusty Russell659a0e62009-06-12 22:27:10 -0600985 kill(0, SIGTERM);
986}
987
988static void reset_device(struct device *dev)
989{
990 struct virtqueue *vq;
991
992 verbose("Resetting device %s\n", dev->name);
993
994 /* Clear any features they've acked. */
995 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
996
997 /* We're going to be explicitly killing threads, so ignore them. */
998 signal(SIGCHLD, SIG_IGN);
999
1000 /* Zero out the virtqueues, get rid of their threads */
1001 for (vq = dev->vq; vq; vq = vq->next) {
1002 if (vq->thread != (pid_t)-1) {
1003 kill(vq->thread, SIGTERM);
1004 waitpid(vq->thread, NULL, 0);
1005 vq->thread = (pid_t)-1;
1006 }
1007 memset(vq->vring.desc, 0,
1008 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1009 lg_last_avail(vq) = 0;
1010 }
1011 dev->running = false;
1012
1013 /* Now we care if threads die. */
1014 signal(SIGCHLD, (void *)kill_launcher);
1015}
1016
Rusty Russella91d74a2009-07-30 16:03:45 -06001017/*L:216
1018 * This actually creates the thread which services the virtqueue for a device.
1019 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001020static void create_thread(struct virtqueue *vq)
1021{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001022 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001023 * Create stack for thread. Since the stack grows upwards, we point
1024 * the stack pointer to the end of this region.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001025 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001026 char *stack = malloc(32768);
1027 unsigned long args[] = { LHREQ_EVENTFD,
1028 vq->config.pfn*getpagesize(), 0 };
1029
1030 /* Create a zero-initialized eventfd. */
1031 vq->eventfd = eventfd(0, 0);
1032 if (vq->eventfd < 0)
1033 err(1, "Creating eventfd");
1034 args[2] = vq->eventfd;
1035
Rusty Russella91d74a2009-07-30 16:03:45 -06001036 /*
1037 * Attach an eventfd to this virtqueue: it will go off when the Guest
1038 * does an LHCALL_NOTIFY for this vq.
1039 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001040 if (write(lguest_fd, &args, sizeof(args)) != 0)
1041 err(1, "Attaching eventfd");
1042
Rusty Russella91d74a2009-07-30 16:03:45 -06001043 /*
1044 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1045 * we get a signal if it dies.
1046 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001047 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1048 if (vq->thread == (pid_t)-1)
1049 err(1, "Creating clone");
Rusty Russella91d74a2009-07-30 16:03:45 -06001050
1051 /* We close our local copy now the child has it. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001052 close(vq->eventfd);
1053}
1054
1055static void start_device(struct device *dev)
1056{
1057 unsigned int i;
1058 struct virtqueue *vq;
1059
1060 verbose("Device %s OK: offered", dev->name);
1061 for (i = 0; i < dev->feature_len; i++)
1062 verbose(" %02x", get_feature_bits(dev)[i]);
1063 verbose(", accepted");
1064 for (i = 0; i < dev->feature_len; i++)
1065 verbose(" %02x", get_feature_bits(dev)
1066 [dev->feature_len+i]);
1067
1068 for (vq = dev->vq; vq; vq = vq->next) {
1069 if (vq->service)
1070 create_thread(vq);
1071 }
1072 dev->running = true;
1073}
1074
1075static void cleanup_devices(void)
1076{
1077 struct device *dev;
1078
1079 for (dev = devices.dev; dev; dev = dev->next)
1080 reset_device(dev);
1081
1082 /* If we saved off the original terminal settings, restore them now. */
1083 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1084 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001085}
1086
Rusty Russella007a752008-05-02 21:50:53 -05001087/* When the Guest tells us they updated the status field, we handle it. */
1088static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001089{
Rusty Russell659a0e62009-06-12 22:27:10 -06001090 /* A zero status is a reset, otherwise it's a set of flags. */
1091 if (dev->desc->status == 0)
1092 reset_device(dev);
1093 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001094 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001095 if (dev->running)
1096 reset_device(dev);
Rusty Russell3c3ed482011-07-22 14:39:49 +09301097 } else {
1098 if (dev->running)
1099 err(1, "Device %s features finalized twice", dev->name);
1100 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001101 }
1102}
1103
Rusty Russella91d74a2009-07-30 16:03:45 -06001104/*L:215
1105 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1106 * particular, it's used to notify us of device status changes during boot.
1107 */
Rusty Russell56739c802009-06-12 22:26:59 -06001108static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001109{
1110 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001111
Rusty Russell659a0e62009-06-12 22:27:10 -06001112 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001113 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001114 struct virtqueue *vq;
1115
Rusty Russella91d74a2009-07-30 16:03:45 -06001116 /*
1117 * Notifications to device descriptors mean they updated the
1118 * device status.
1119 */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001120 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001121 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001122 return;
1123 }
1124
Rusty Russell3c3ed482011-07-22 14:39:49 +09301125 /* Devices should not be used before features are finalized. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001126 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001127 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001128 continue;
Rusty Russell3c3ed482011-07-22 14:39:49 +09301129 errx(1, "Notification on %s before setup!", i->name);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001130 }
1131 }
Rusty Russelldde79782007-07-26 10:41:03 -07001132
Rusty Russell2e04ef72009-07-30 16:03:45 -06001133 /*
1134 * Early console write is done using notify on a nul-terminated string
1135 * in Guest memory. It's also great for hacking debugging messages
1136 * into a Guest.
1137 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001138 if (addr >= guest_limit)
1139 errx(1, "Bad NOTIFY %#lx", addr);
1140
1141 write(STDOUT_FILENO, from_guest_phys(addr),
1142 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001143}
1144
Rusty Russelldde79782007-07-26 10:41:03 -07001145/*L:190
1146 * Device Setup
1147 *
1148 * All devices need a descriptor so the Guest knows it exists, and a "struct
1149 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001150 * routines to allocate and manage them.
1151 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001152
Rusty Russell2e04ef72009-07-30 16:03:45 -06001153/*
1154 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001155 * number of virtqueue descriptors, then two sets of feature bits, then an
1156 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001157 * pointer.
1158 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001159static u8 *device_config(const struct device *dev)
1160{
1161 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001162 + dev->num_vq * sizeof(struct lguest_vqconfig)
1163 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001164}
1165
Rusty Russell2e04ef72009-07-30 16:03:45 -06001166/*
1167 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001168 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001169 * that descriptor.
1170 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001171static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001172{
Rusty Russella586d4f2008-02-04 23:49:56 -05001173 struct lguest_device_desc d = { .type = type };
1174 void *p;
1175
1176 /* Figure out where the next device config is, based on the last one. */
1177 if (devices.lastdev)
1178 p = device_config(devices.lastdev)
1179 + devices.lastdev->desc->config_len;
1180 else
1181 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001182
Rusty Russell17cbca22007-10-22 11:24:22 +10001183 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001184 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001185 errx(1, "Too many devices");
1186
Rusty Russella586d4f2008-02-04 23:49:56 -05001187 /* p might not be aligned, so we memcpy in. */
1188 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001189}
1190
Rusty Russell2e04ef72009-07-30 16:03:45 -06001191/*
1192 * Each device descriptor is followed by the description of its virtqueues. We
1193 * specify how many descriptors the virtqueue is to have.
1194 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001195static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001196 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001197{
1198 unsigned int pages;
1199 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1200 void *p;
1201
Rusty Russella6bd8e12008-03-28 11:05:53 -05001202 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001203 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001204 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001205 p = get_pages(pages);
1206
Rusty Russelld1c856e2007-11-19 11:20:40 -05001207 /* Initialize the virtqueue */
1208 vq->next = NULL;
1209 vq->last_avail_idx = 0;
1210 vq->dev = dev;
Rusty Russella91d74a2009-07-30 16:03:45 -06001211
1212 /*
1213 * This is the routine the service thread will run, and its Process ID
1214 * once it's running.
1215 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001216 vq->service = service;
1217 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001218
Rusty Russell17cbca22007-10-22 11:24:22 +10001219 /* Initialize the configuration. */
1220 vq->config.num = num_descs;
1221 vq->config.irq = devices.next_irq++;
1222 vq->config.pfn = to_guest_phys(p) / getpagesize();
1223
1224 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001225 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001226
Rusty Russell2e04ef72009-07-30 16:03:45 -06001227 /*
1228 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001229 * device_config() to get the end of the device's current virtqueues;
1230 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001231 * yet, otherwise we'd be overwriting them.
1232 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001233 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1234 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001235 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001236 dev->desc->num_vq++;
1237
1238 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001239
Rusty Russell2e04ef72009-07-30 16:03:45 -06001240 /*
1241 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1242 * second.
1243 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001244 for (i = &dev->vq; *i; i = &(*i)->next);
1245 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001246}
1247
Rusty Russell2e04ef72009-07-30 16:03:45 -06001248/*
1249 * The first half of the feature bitmask is for us to advertise features. The
1250 * second half is for the Guest to accept features.
1251 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001252static void add_feature(struct device *dev, unsigned bit)
1253{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001254 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001255
1256 /* We can't extend the feature bits once we've added config bytes */
1257 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1258 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001259 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001260 }
1261
Rusty Russella586d4f2008-02-04 23:49:56 -05001262 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1263}
1264
Rusty Russell2e04ef72009-07-30 16:03:45 -06001265/*
1266 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001267 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001268 * how we use it.
1269 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001270static void set_config(struct device *dev, unsigned len, const void *conf)
1271{
1272 /* Check we haven't overflowed our single page. */
1273 if (device_config(dev) + len > devices.descpage + getpagesize())
1274 errx(1, "Too many devices");
1275
1276 /* Copy in the config information, and store the length. */
1277 memcpy(device_config(dev), conf, len);
1278 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001279
1280 /* Size must fit in config_len field (8 bits)! */
1281 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001282}
1283
Rusty Russell2e04ef72009-07-30 16:03:45 -06001284/*
1285 * This routine does all the creation and setup of a new device, including
Rusty Russella91d74a2009-07-30 16:03:45 -06001286 * calling new_dev_desc() to allocate the descriptor and device memory. We
1287 * don't actually start the service threads until later.
Rusty Russella6bd8e12008-03-28 11:05:53 -05001288 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001289 * See what I mean about userspace being boring?
1290 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001291static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001292{
1293 struct device *dev = malloc(sizeof(*dev));
1294
Rusty Russelldde79782007-07-26 10:41:03 -07001295 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001296 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001297 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001298 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001299 dev->feature_len = 0;
1300 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001301 dev->running = false;
Rusty Russella586d4f2008-02-04 23:49:56 -05001302
Rusty Russell2e04ef72009-07-30 16:03:45 -06001303 /*
1304 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001305 * easier, but the user expects the devices to be arranged on the bus
1306 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001307 * is eth0, the first block device /dev/vda, etc.
1308 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001309 if (devices.lastdev)
1310 devices.lastdev->next = dev;
1311 else
1312 devices.dev = dev;
1313 devices.lastdev = dev;
1314
Rusty Russell8ca47e02007-07-19 01:49:29 -07001315 return dev;
1316}
1317
Rusty Russell2e04ef72009-07-30 16:03:45 -06001318/*
1319 * Our first setup routine is the console. It's a fairly simple device, but
1320 * UNIX tty handling makes it uglier than it could be.
1321 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001322static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001323{
1324 struct device *dev;
1325
Rusty Russelldde79782007-07-26 10:41:03 -07001326 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001327 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1328 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001329 /*
1330 * Then we turn off echo, line buffering and ^C etc: We want a
1331 * raw input stream to the Guest.
1332 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001333 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1334 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001335 }
1336
Rusty Russell659a0e62009-06-12 22:27:10 -06001337 dev = new_device("console", VIRTIO_ID_CONSOLE);
1338
Rusty Russelldde79782007-07-26 10:41:03 -07001339 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001340 dev->priv = malloc(sizeof(struct console_abort));
1341 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001342
Rusty Russell2e04ef72009-07-30 16:03:45 -06001343 /*
1344 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001345 * they put something the input queue, we make sure we're listening to
1346 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001347 * stdout.
1348 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001349 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1350 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001351
Rusty Russell659a0e62009-06-12 22:27:10 -06001352 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001353}
Rusty Russelldde79782007-07-26 10:41:03 -07001354/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001355
Rusty Russell2e04ef72009-07-30 16:03:45 -06001356/*M:010
1357 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001358 * --sharenet=<name> option which opens or creates a named pipe. This can be
1359 * used to send packets to another guest in a 1:1 manner.
1360 *
Rusty Russell9f542882011-07-22 14:39:50 +09301361 * More sophisticated is to use one of the tools developed for project like UML
Rusty Russell17cbca22007-10-22 11:24:22 +10001362 * to do networking.
1363 *
1364 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1365 * completely generic ("here's my vring, attach to your vring") and would work
1366 * for any traffic. Of course, namespace and permissions issues need to be
1367 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1368 * multiple inter-guest channels behind one interface, although it would
1369 * require some manner of hotplugging new virtio channels.
1370 *
Rusty Russell9f542882011-07-22 14:39:50 +09301371 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001372:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001373
Rusty Russell8ca47e02007-07-19 01:49:29 -07001374static u32 str2ip(const char *ipaddr)
1375{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001376 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001377
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001378 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1379 errx(1, "Failed to parse IP address '%s'", ipaddr);
1380 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1381}
1382
1383static void str2mac(const char *macaddr, unsigned char mac[6])
1384{
1385 unsigned int m[6];
1386 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1387 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1388 errx(1, "Failed to parse mac address '%s'", macaddr);
1389 mac[0] = m[0];
1390 mac[1] = m[1];
1391 mac[2] = m[2];
1392 mac[3] = m[3];
1393 mac[4] = m[4];
1394 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001395}
1396
Rusty Russell2e04ef72009-07-30 16:03:45 -06001397/*
1398 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001399 * network device to the bridge device specified by the command line.
1400 *
1401 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001402 * dislike bridging), and I just try not to break it.
1403 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001404static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1405{
1406 int ifidx;
1407 struct ifreq ifr;
1408
1409 if (!*br_name)
1410 errx(1, "must specify bridge name");
1411
1412 ifidx = if_nametoindex(if_name);
1413 if (!ifidx)
1414 errx(1, "interface %s does not exist!", if_name);
1415
1416 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001417 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001418 ifr.ifr_ifindex = ifidx;
1419 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1420 err(1, "can't add %s to bridge %s", if_name, br_name);
1421}
1422
Rusty Russell2e04ef72009-07-30 16:03:45 -06001423/*
1424 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001425 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001426 * pointer.
1427 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001428static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001429{
1430 struct ifreq ifr;
Rusty Russellf8466192010-08-27 08:39:48 -06001431 struct sockaddr_in sin;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001432
1433 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001434 strcpy(ifr.ifr_name, tapif);
1435
1436 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russellf8466192010-08-27 08:39:48 -06001437 sin.sin_family = AF_INET;
1438 sin.sin_addr.s_addr = htonl(ipaddr);
1439 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001440 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001441 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001442 ifr.ifr_flags = IFF_UP;
1443 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001444 err(1, "Bringing interface %s up", tapif);
1445}
1446
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001447static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001448{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001449 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001450 int netfd;
1451
1452 /* Start with this zeroed. Messy but sure. */
1453 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001454
Rusty Russell2e04ef72009-07-30 16:03:45 -06001455 /*
1456 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001457 * tap device is like a tun device, only somehow different. To tell
1458 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001459 * works now!
1460 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001461 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001462 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001463 strcpy(ifr.ifr_name, "tap%d");
1464 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1465 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001466
Rusty Russell398f1872008-07-29 09:58:37 -05001467 if (ioctl(netfd, TUNSETOFFLOAD,
1468 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1469 err(1, "Could not set features for tun device");
1470
Rusty Russell2e04ef72009-07-30 16:03:45 -06001471 /*
1472 * We don't need checksums calculated for packets coming in this
1473 * device: trust us!
1474 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001475 ioctl(netfd, TUNSETNOCSUM, 1);
1476
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001477 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1478 return netfd;
1479}
1480
Rusty Russell2e04ef72009-07-30 16:03:45 -06001481/*L:195
1482 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001483 * routing, but the principle is the same: it uses the "tun" device to inject
1484 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001485 * just shunt packets between the Guest and the tun device.
1486 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001487static void setup_tun_net(char *arg)
1488{
1489 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001490 struct net_info *net_info = malloc(sizeof(*net_info));
1491 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001492 u32 ip = INADDR_ANY;
1493 bool bridging = false;
1494 char tapif[IFNAMSIZ], *p;
1495 struct virtio_net_config conf;
1496
Rusty Russell659a0e62009-06-12 22:27:10 -06001497 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001498
Rusty Russell17cbca22007-10-22 11:24:22 +10001499 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001500 dev = new_device("net", VIRTIO_ID_NET);
1501 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001502
Rusty Russell2e04ef72009-07-30 16:03:45 -06001503 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001504 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1505 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001506
Rusty Russell2e04ef72009-07-30 16:03:45 -06001507 /*
1508 * We need a socket to perform the magic network ioctls to bring up the
1509 * tap interface, connect to the bridge etc. Any socket will do!
1510 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001511 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1512 if (ipfd < 0)
1513 err(1, "opening IP socket");
1514
Rusty Russelldde79782007-07-26 10:41:03 -07001515 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001516 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001517 arg += strlen(BRIDGE_PFX);
1518 bridging = true;
1519 }
1520
1521 /* A mac address may follow the bridge name or IP address */
1522 p = strchr(arg, ':');
1523 if (p) {
1524 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001525 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001526 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001527 }
1528
1529 /* arg is now either an IP address or a bridge name */
1530 if (bridging)
1531 add_to_bridge(ipfd, tapif, arg);
1532 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001533 ip = str2ip(arg);
1534
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001535 /* Set up the tun device. */
1536 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001537
Rusty Russell398f1872008-07-29 09:58:37 -05001538 /* Expect Guest to handle everything except UFO */
1539 add_feature(dev, VIRTIO_NET_F_CSUM);
1540 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001541 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1542 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1543 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1544 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1545 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1546 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001547 /* We handle indirect ring entries */
1548 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russella586d4f2008-02-04 23:49:56 -05001549 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001550
Rusty Russella586d4f2008-02-04 23:49:56 -05001551 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001552 close(ipfd);
1553
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001554 devices.device_num++;
1555
1556 if (bridging)
1557 verbose("device %u: tun %s attached to bridge: %s\n",
1558 devices.device_num, tapif, arg);
1559 else
1560 verbose("device %u: tun %s: %s\n",
1561 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001562}
Rusty Russella91d74a2009-07-30 16:03:45 -06001563/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001564
Rusty Russelle1e72962007-10-25 15:02:50 +10001565/* This hangs off device->priv. */
Rusty Russell1842f232009-07-30 16:03:46 -06001566struct vblk_info {
Rusty Russell17cbca22007-10-22 11:24:22 +10001567 /* The size of the file. */
1568 off64_t len;
1569
1570 /* The file descriptor for the file. */
1571 int fd;
1572
Rusty Russell17cbca22007-10-22 11:24:22 +10001573};
1574
Rusty Russelle1e72962007-10-25 15:02:50 +10001575/*L:210
1576 * The Disk
1577 *
Rusty Russella91d74a2009-07-30 16:03:45 -06001578 * The disk only has one virtqueue, so it only has one thread. It is really
1579 * simple: the Guest asks for a block number and we read or write that position
1580 * in the file.
1581 *
1582 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1583 * slow: the Guest waits until the read is finished before running anything
1584 * else, even if it could have been doing useful work.
1585 *
1586 * We could have used async I/O, except it's reputed to suck so hard that
1587 * characters actually go missing from your code when you try to use it.
Rusty Russelle1e72962007-10-25 15:02:50 +10001588 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001589static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001590{
Rusty Russell659a0e62009-06-12 22:27:10 -06001591 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001592 unsigned int head, out_num, in_num, wlen;
1593 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001594 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001595 struct virtio_blk_outhdr *out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001596 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001597 off64_t off;
1598
Rusty Russella91d74a2009-07-30 16:03:45 -06001599 /*
1600 * Get the next request, where we normally wait. It triggers the
1601 * interrupt to acknowledge previously serviced requests (if any).
1602 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001603 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001604
Rusty Russell2e04ef72009-07-30 16:03:45 -06001605 /*
1606 * Every block request should contain at least one output buffer
Rusty Russelle1e72962007-10-25 15:02:50 +10001607 * (detailing the location on disk and the type of request) and one
Rusty Russell2e04ef72009-07-30 16:03:45 -06001608 * input buffer (to hold the result).
1609 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001610 if (out_num == 0 || in_num == 0)
1611 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1612 head, out_num, in_num);
1613
1614 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001615 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russella91d74a2009-07-30 16:03:45 -06001616 /*
1617 * For historical reasons, block operations are expressed in 512 byte
1618 * "sectors".
1619 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001620 off = out->sector * 512;
1621
Rusty Russell2e04ef72009-07-30 16:03:45 -06001622 /*
Rusty Russell2e04ef72009-07-30 16:03:45 -06001623 * In general the virtio block driver is allowed to try SCSI commands.
1624 * It'd be nice if we supported eject, for example, but we don't.
1625 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001626 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1627 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001628 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001629 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001630 } else if (out->type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001631 /*
1632 * Write
1633 *
1634 * Move to the right location in the block file. This can fail
1635 * if they try to write past end.
1636 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001637 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1638 err(1, "Bad seek to sector %llu", out->sector);
1639
1640 ret = writev(vblk->fd, iov+1, out_num-1);
1641 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1642
Rusty Russell2e04ef72009-07-30 16:03:45 -06001643 /*
1644 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001645 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001646 * file (possibly extending it).
1647 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001648 if (ret > 0 && off + ret > vblk->len) {
1649 /* Trim it back to the correct length */
1650 ftruncate64(vblk->fd, vblk->len);
1651 /* Die, bad Guest, die. */
1652 errx(1, "Write past end %llu+%u", off, ret);
1653 }
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001654
1655 wlen = sizeof(*in);
1656 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1657 } else if (out->type & VIRTIO_BLK_T_FLUSH) {
1658 /* Flush */
1659 ret = fdatasync(vblk->fd);
1660 verbose("FLUSH fdatasync: %i\n", ret);
Anthony Liguori1200e642007-11-08 21:13:44 -06001661 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001662 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001663 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001664 /*
1665 * Read
1666 *
1667 * Move to the right location in the block file. This can fail
1668 * if they try to read past end.
1669 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001670 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1671 err(1, "Bad seek to sector %llu", out->sector);
1672
1673 ret = readv(vblk->fd, iov+1, in_num-1);
1674 verbose("READ from sector %llu: %i\n", out->sector, ret);
1675 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001676 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001677 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001678 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001679 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001680 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001681 }
1682 }
1683
Rusty Russella91d74a2009-07-30 16:03:45 -06001684 /* Finished that request. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001685 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001686}
1687
Rusty Russelle1e72962007-10-25 15:02:50 +10001688/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001689static void setup_block_file(const char *filename)
1690{
Rusty Russell17cbca22007-10-22 11:24:22 +10001691 struct device *dev;
1692 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001693 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001694
Rusty Russell2e04ef72009-07-30 16:03:45 -06001695 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001696 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001697
Rusty Russelle1e72962007-10-25 15:02:50 +10001698 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001699 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001700
1701 /* Allocate the room for our own bookkeeping */
1702 vblk = dev->priv = malloc(sizeof(*vblk));
1703
1704 /* First we open the file and store the length. */
1705 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1706 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1707
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001708 /* We support FLUSH. */
1709 add_feature(dev, VIRTIO_BLK_F_FLUSH);
Rusty Russella586d4f2008-02-04 23:49:56 -05001710
Rusty Russell17cbca22007-10-22 11:24:22 +10001711 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001712 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001713
Rusty Russell2e04ef72009-07-30 16:03:45 -06001714 /*
1715 * Tell Guest not to put in too many descriptors at once: two are used
1716 * for the in and out elements.
1717 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001718 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1719 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1720
Rusty Russell8ef562d2009-07-30 16:03:43 -06001721 /* Don't try to put whole struct: we have 8 bit limit. */
1722 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001723
Rusty Russell17cbca22007-10-22 11:24:22 +10001724 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001725 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001726}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001727
Rusty Russell2e04ef72009-07-30 16:03:45 -06001728/*L:211
1729 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001730 * input buffers. The usual case is that the Guest doesn't want random numbers
1731 * and so has no buffers although /dev/random is still readable, whereas
1732 * console is the reverse.
1733 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001734 * The same logic applies, however.
1735 */
1736struct rng_info {
1737 int rfd;
1738};
1739
Rusty Russell659a0e62009-06-12 22:27:10 -06001740static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001741{
1742 int len;
1743 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001744 struct rng_info *rng_info = vq->dev->priv;
1745 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001746
1747 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001748 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001749 if (out_num)
1750 errx(1, "Output buffers in rng?");
1751
Rusty Russell2e04ef72009-07-30 16:03:45 -06001752 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001753 * Just like the console write, we loop to cover the whole iovec.
1754 * In this case, short reads actually happen quite a bit.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001755 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001756 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001757 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001758 if (len <= 0)
1759 err(1, "Read from /dev/random gave %i", len);
1760 iov_consume(iov, in_num, len);
1761 totlen += len;
1762 }
1763
1764 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001765 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001766}
1767
Rusty Russell2e04ef72009-07-30 16:03:45 -06001768/*L:199
1769 * This creates a "hardware" random number device for the Guest.
1770 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001771static void setup_rng(void)
1772{
1773 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001774 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001775
Rusty Russell2e04ef72009-07-30 16:03:45 -06001776 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001777 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001778
Rusty Russell2e04ef72009-07-30 16:03:45 -06001779 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001780 dev = new_device("rng", VIRTIO_ID_RNG);
1781 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001782
1783 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001784 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001785
1786 verbose("device %u: rng\n", devices.device_num++);
1787}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001788/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301789
Rusty Russella6bd8e12008-03-28 11:05:53 -05001790/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301791static void __attribute__((noreturn)) restart_guest(void)
1792{
1793 unsigned int i;
1794
Rusty Russell2e04ef72009-07-30 16:03:45 -06001795 /*
1796 * Since we don't track all open fds, we simply close everything beyond
1797 * stderr.
1798 */
Balaji Raoec04b132007-12-28 14:26:24 +05301799 for (i = 3; i < FD_SETSIZE; i++)
1800 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001801
Rusty Russell659a0e62009-06-12 22:27:10 -06001802 /* Reset all the devices (kills all threads). */
1803 cleanup_devices();
1804
Balaji Raoec04b132007-12-28 14:26:24 +05301805 execv(main_args[0], main_args);
1806 err(1, "Could not exec %s", main_args[0]);
1807}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001808
Rusty Russell2e04ef72009-07-30 16:03:45 -06001809/*L:220
1810 * Finally we reach the core of the Launcher which runs the Guest, serves
1811 * its input and output, and finally, lays it to rest.
1812 */
Rusty Russell56739c802009-06-12 22:26:59 -06001813static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001814{
1815 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001816 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001817 int readval;
1818
1819 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001820 readval = pread(lguest_fd, &notify_addr,
1821 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001822
Rusty Russell17cbca22007-10-22 11:24:22 +10001823 /* One unsigned long means the Guest did HCALL_NOTIFY */
1824 if (readval == sizeof(notify_addr)) {
1825 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c802009-06-12 22:26:59 -06001826 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001827 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001828 } else if (errno == ENOENT) {
1829 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001830 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001831 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301832 /* ERESTART means that we need to reboot the guest */
1833 } else if (errno == ERESTART) {
1834 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001835 /* Anything else means a bug or incompatible change. */
1836 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001837 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001838 }
1839}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001840/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001841 * This is the end of the Launcher. The good news: we are over halfway
1842 * through! The bad news: the most fiendish part of the code still lies ahead
1843 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001844 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001845 * Are you ready? Take a deep breath and join me in the core of the Host, in
1846 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001847:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001848
1849static struct option opts[] = {
1850 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001851 { "tunnet", 1, NULL, 't' },
1852 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001853 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001854 { "initrd", 1, NULL, 'i' },
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001855 { "username", 1, NULL, 'u' },
1856 { "chroot", 1, NULL, 'c' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001857 { NULL },
1858};
1859static void usage(void)
1860{
1861 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001862 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001863 "|--block=<filename>|--initrd=<filename>]...\n"
1864 "<mem-in-mb> vmlinux [args...]");
1865}
1866
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001867/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001868int main(int argc, char *argv[])
1869{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001870 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001871 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c802009-06-12 22:26:59 -06001872 /* Two temporaries. */
1873 int i, c;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001874 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001875 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001876 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001877 const char *initrd_name = NULL;
1878
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001879 /* Password structure for initgroups/setres[gu]id */
1880 struct passwd *user_details = NULL;
1881
1882 /* Directory to chroot to */
1883 char *chroot_path = NULL;
1884
Balaji Raoec04b132007-12-28 14:26:24 +05301885 /* Save the args: we "reboot" by execing ourselves again. */
1886 main_args = argv;
Balaji Raoec04b132007-12-28 14:26:24 +05301887
Rusty Russell2e04ef72009-07-30 16:03:45 -06001888 /*
1889 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001890 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001891 * remember that 0 is used by the timer).
1892 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001893 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001894 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001895
Rusty Russella91d74a2009-07-30 16:03:45 -06001896 /* We're CPU 0. In fact, that's the only CPU possible right now. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001897 cpu_id = 0;
Rusty Russella91d74a2009-07-30 16:03:45 -06001898
Rusty Russell2e04ef72009-07-30 16:03:45 -06001899 /*
1900 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001901 * descriptor and memory pages for the devices as we parse the command
1902 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001903 * of memory now.
1904 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001905 for (i = 1; i < argc; i++) {
1906 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001907 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001908 /*
1909 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001910 * guest-physical memory range. This fills it with 0,
1911 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001912 * tries to access it.
1913 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001914 guest_base = map_zeroed_pages(mem / getpagesize()
1915 + DEVICE_PAGES);
1916 guest_limit = mem;
1917 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001918 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001919 break;
1920 }
1921 }
Rusty Russelldde79782007-07-26 10:41:03 -07001922
1923 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001924 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1925 switch (c) {
1926 case 'v':
1927 verbose = true;
1928 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001929 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001930 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001931 break;
1932 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001933 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001934 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001935 case 'r':
1936 setup_rng();
1937 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001938 case 'i':
1939 initrd_name = optarg;
1940 break;
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001941 case 'u':
1942 user_details = getpwnam(optarg);
1943 if (!user_details)
1944 err(1, "getpwnam failed, incorrect username?");
1945 break;
1946 case 'c':
1947 chroot_path = optarg;
1948 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001949 default:
1950 warnx("Unknown argument %s", argv[optind]);
1951 usage();
1952 }
1953 }
Rusty Russell2e04ef72009-07-30 16:03:45 -06001954 /*
1955 * After the other arguments we expect memory and kernel image name,
1956 * followed by command line arguments for the kernel.
1957 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001958 if (optind + 2 > argc)
1959 usage();
1960
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001961 verbose("Guest base is at %p\n", guest_base);
1962
Rusty Russelldde79782007-07-26 10:41:03 -07001963 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001964 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001965
Rusty Russell8ca47e02007-07-19 01:49:29 -07001966 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001967 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001968
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001969 /* Boot information is stashed at physical address 0 */
1970 boot = from_guest_phys(0);
1971
Rusty Russelldde79782007-07-26 10:41:03 -07001972 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001973 if (initrd_name) {
1974 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06001975 /*
1976 * These are the location in the Linux boot header where the
1977 * start and size of the initrd are expected to be found.
1978 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001979 boot->hdr.ramdisk_image = mem - initrd_size;
1980 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001981 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001982 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001983 }
1984
Rusty Russell2e04ef72009-07-30 16:03:45 -06001985 /*
1986 * The Linux boot header contains an "E820" memory map: ours is a
1987 * simple, single region.
1988 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001989 boot->e820_entries = 1;
1990 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06001991 /*
1992 * The boot header contains a command line pointer: we put the command
1993 * line after the boot header.
1994 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001995 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10001996 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001997 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07001998
Rusty Russelle22a5392011-08-15 10:15:10 +09301999 /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
2000 boot->hdr.kernel_alignment = 0x1000000;
2001
Rusty Russell814a0e52007-10-22 11:29:44 +10002002 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002003 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002004
2005 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002006 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002007
Rusty Russell43d33b22007-10-22 11:29:57 +10002008 /* Tell the entry path not to try to reload segment registers. */
2009 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002010
Rusty Russell9f542882011-07-22 14:39:50 +09302011 /* We tell the kernel to initialize the Guest. */
Rusty Russell56739c802009-06-12 22:26:59 -06002012 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002013
Rusty Russella91d74a2009-07-30 16:03:45 -06002014 /* Ensure that we terminate if a device-servicing child dies. */
Rusty Russell659a0e62009-06-12 22:27:10 -06002015 signal(SIGCHLD, kill_launcher);
2016
2017 /* If we exit via err(), this kills all the threads, restores tty. */
2018 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002019
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06002020 /* If requested, chroot to a directory */
2021 if (chroot_path) {
2022 if (chroot(chroot_path) != 0)
2023 err(1, "chroot(\"%s\") failed", chroot_path);
2024
2025 if (chdir("/") != 0)
2026 err(1, "chdir(\"/\") failed");
2027
2028 verbose("chroot done\n");
2029 }
2030
2031 /* If requested, drop privileges */
2032 if (user_details) {
2033 uid_t u;
2034 gid_t g;
2035
2036 u = user_details->pw_uid;
2037 g = user_details->pw_gid;
2038
2039 if (initgroups(user_details->pw_name, g) != 0)
2040 err(1, "initgroups failed");
2041
2042 if (setresgid(g, g, g) != 0)
2043 err(1, "setresgid failed");
2044
2045 if (setresuid(u, u, u) != 0)
2046 err(1, "setresuid failed");
2047
2048 verbose("Dropping privileges completed\n");
2049 }
2050
Rusty Russelldde79782007-07-26 10:41:03 -07002051 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c802009-06-12 22:26:59 -06002052 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07002053}
Rusty Russellf56a3842007-07-26 10:41:05 -07002054/*:*/
2055
2056/*M:999
2057 * Mastery is done: you now know everything I do.
2058 *
2059 * But surely you have seen code, features and bugs in your wanderings which
2060 * you now yearn to attack? That is the real game, and I look forward to you
2061 * patching and forking lguest into the Your-Name-Here-visor.
2062 *
2063 * Farewell, and good coding!
2064 * Rusty Russell.
2065 */