blob: aa66a52b73e9589ad031310cd7f813bbb241f641 [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>
37#include <zlib.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100038#include <assert.h>
39#include <sched.h>
Rusty Russella586d4f2008-02-04 23:49:56 -050040#include <limits.h>
41#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050042#include <signal.h>
Rusty Russellb45d8cb2007-10-22 10:56:24 +100043#include "linux/lguest_launcher.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100044#include "linux/virtio_config.h"
45#include "linux/virtio_net.h"
46#include "linux/virtio_blk.h"
47#include "linux/virtio_console.h"
Rusty Russell28fd6d72008-07-29 09:58:33 -050048#include "linux/virtio_rng.h"
Rusty Russell17cbca22007-10-22 11:24:22 +100049#include "linux/virtio_ring.h"
Rusty Russelld5d02d62008-10-31 11:24:25 -050050#include "asm/bootparam.h"
Rusty Russell2e04ef72009-07-30 16:03:45 -060051/*L:110
52 * We can ignore the 39 include files we need for this program, but I do want
53 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100054 *
55 * As Linus said, "C is a Spartan language, and so should your naming be." I
56 * like these abbreviations, so we define them here. Note that u64 is always
57 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060058 * use %llu in printf for any u64.
59 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100060typedef unsigned long long u64;
61typedef uint32_t u32;
62typedef uint16_t u16;
63typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070064/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070065
66#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
Rusty Russell8ca47e02007-07-19 01:49:29 -070067#define BRIDGE_PFX "bridge:"
68#ifndef SIOCBRADDIF
69#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
70#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100071/* We can have up to 256 pages for devices. */
72#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050073/* This will occupy 3 pages: it must be a power of 2. */
74#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070075
Rusty Russell2e04ef72009-07-30 16:03:45 -060076/*L:120
77 * verbose is both a global flag and a macro. The C preprocessor allows
78 * this, and although I wouldn't recommend it, it works quite nicely here.
79 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070080static bool verbose;
81#define verbose(args...) \
82 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070083/*:*/
84
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100085/* The pointer to the start of guest memory. */
86static void *guest_base;
87/* The maximum guest physical address allowed, and maximum possible. */
88static unsigned long guest_limit, guest_max;
Rusty Russell56739c802009-06-12 22:26:59 -060089/* The /dev/lguest file descriptor. */
90static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070091
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020092/* a per-cpu variable indicating whose vcpu is currently running */
93static unsigned int __thread cpu_id;
94
Rusty Russelldde79782007-07-26 10:41:03 -070095/* This is our list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -070096struct device_list
97{
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 Russell8ca47e02007-07-19 01:49:29 -0700117struct device
118{
Rusty Russelldde79782007-07-26 10:41:03 -0700119 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700120 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000121
Rusty Russell713b15b2009-06-12 22:26:58 -0600122 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700123 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000124
Rusty Russell713b15b2009-06-12 22:26:58 -0600125 /* We can't trust desc values once Guest has booted: we use these. */
126 unsigned int feature_len;
127 unsigned int num_vq;
128
Rusty Russell17cbca22007-10-22 11:24:22 +1000129 /* The name of this device, for --verbose. */
130 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700131
Rusty Russell17cbca22007-10-22 11:24:22 +1000132 /* Any queues attached to this device */
133 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700134
Rusty Russell659a0e62009-06-12 22:27:10 -0600135 /* Is it operational */
136 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500137
Rusty Russell8ca47e02007-07-19 01:49:29 -0700138 /* Device-specific data. */
139 void *priv;
140};
141
Rusty Russell17cbca22007-10-22 11:24:22 +1000142/* The virtqueue structure describes a queue attached to a device. */
143struct virtqueue
144{
145 struct virtqueue *next;
146
147 /* Which device owns me. */
148 struct device *dev;
149
150 /* The configuration for this queue. */
151 struct lguest_vqconfig config;
152
153 /* The actual ring of buffers. */
154 struct vring vring;
155
156 /* Last available index we saw. */
157 u16 last_avail_idx;
158
Rusty Russell95c517c2009-06-12 22:27:11 -0600159 /* How many are used since we sent last irq? */
160 unsigned int pending_used;
161
Rusty Russell659a0e62009-06-12 22:27:10 -0600162 /* Eventfd where Guest notifications arrive. */
163 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500164
Rusty Russell659a0e62009-06-12 22:27:10 -0600165 /* Function for the thread which is servicing this virtqueue. */
166 void (*service)(struct virtqueue *vq);
167 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000168};
169
Balaji Raoec04b132007-12-28 14:26:24 +0530170/* Remember the arguments to the program so we can "reboot" */
171static char **main_args;
172
Rusty Russell659a0e62009-06-12 22:27:10 -0600173/* The original tty settings to restore on exit. */
174static struct termios orig_term;
175
Rusty Russell2e04ef72009-07-30 16:03:45 -0600176/*
177 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600178 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600179 * in precise order.
180 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600181#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russellb60da132009-06-12 22:27:12 -0600182#define mb() __asm__ __volatile__("" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000183
Rusty Russell2e04ef72009-07-30 16:03:45 -0600184/*
185 * Convert an iovec element to the given type.
Rusty Russell17cbca22007-10-22 11:24:22 +1000186 *
187 * This is a fairly ugly trick: we need to know the size of the type and
188 * alignment requirement to check the pointer is kosher. It's also nice to
189 * have the name of the type in case we report failure.
190 *
191 * Typing those three things all the time is cumbersome and error prone, so we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600192 * have a macro which sets them all up and passes to the real function.
193 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000194#define convert(iov, type) \
195 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
196
197static void *_convert(struct iovec *iov, size_t size, size_t align,
198 const char *name)
199{
200 if (iov->iov_len != size)
201 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
202 if ((unsigned long)iov->iov_base % align != 0)
203 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
204 return iov->iov_base;
205}
206
Rusty Russellb5111792008-07-29 09:58:34 -0500207/* Wrapper for the last available index. Makes it easier to change. */
208#define lg_last_avail(vq) ((vq)->last_avail_idx)
209
Rusty Russell2e04ef72009-07-30 16:03:45 -0600210/*
211 * The virtio configuration space is defined to be little-endian. x86 is
212 * little-endian too, but it's nice to be explicit so we have these helpers.
213 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000214#define cpu_to_le16(v16) (v16)
215#define cpu_to_le32(v32) (v32)
216#define cpu_to_le64(v64) (v64)
217#define le16_to_cpu(v16) (v16)
218#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500219#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000220
Rusty Russell28fd6d72008-07-29 09:58:33 -0500221/* Is this iovec empty? */
222static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
223{
224 unsigned int i;
225
226 for (i = 0; i < num_iov; i++)
227 if (iov[i].iov_len)
228 return false;
229 return true;
230}
231
232/* Take len bytes from the front of this iovec. */
233static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
234{
235 unsigned int i;
236
237 for (i = 0; i < num_iov; i++) {
238 unsigned int used;
239
240 used = iov[i].iov_len < len ? iov[i].iov_len : len;
241 iov[i].iov_base += used;
242 iov[i].iov_len -= used;
243 len -= used;
244 }
245 assert(len == 0);
246}
247
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500248/* The device virtqueue descriptors are followed by feature bitmasks. */
249static u8 *get_feature_bits(struct device *dev)
250{
251 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600252 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500253}
254
Rusty Russell2e04ef72009-07-30 16:03:45 -0600255/*L:100
256 * The Launcher code itself takes us out into userspace, that scary place where
257 * pointers run wild and free! Unfortunately, like most userspace programs,
258 * it's quite boring (which is why everyone likes to hack on the kernel!).
259 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
260 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000261 *
262 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
263 * memory and stores it in "guest_base". In other words, Guest physical ==
264 * Launcher virtual with an offset.
265 *
266 * This can be tough to get your head around, but usually it just means that we
267 * use these trivial conversion functions when the Guest gives us it's
Rusty Russell2e04ef72009-07-30 16:03:45 -0600268 * "physical" addresses:
269 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000270static void *from_guest_phys(unsigned long addr)
271{
272 return guest_base + addr;
273}
274
275static unsigned long to_guest_phys(const void *addr)
276{
277 return (addr - guest_base);
278}
279
Rusty Russelldde79782007-07-26 10:41:03 -0700280/*L:130
281 * Loading the Kernel.
282 *
283 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600284 * error-checking code cluttering the callers:
285 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700286static int open_or_die(const char *name, int flags)
287{
288 int fd = open(name, flags);
289 if (fd < 0)
290 err(1, "Failed to open %s", name);
291 return fd;
292}
293
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000294/* map_zeroed_pages() takes a number of pages. */
295static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700296{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000297 int fd = open_or_die("/dev/zero", O_RDONLY);
298 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700299
Rusty Russell2e04ef72009-07-30 16:03:45 -0600300 /*
301 * We use a private mapping (ie. if we write to the page, it will be
302 * copied).
303 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000304 addr = mmap(NULL, getpagesize() * num,
305 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
306 if (addr == MAP_FAILED)
307 err(1, "Mmaping %u pages of /dev/zero", num);
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100308 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700309
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000310 return addr;
311}
312
313/* Get some more pages for a device. */
314static void *get_pages(unsigned int num)
315{
316 void *addr = from_guest_phys(guest_limit);
317
318 guest_limit += num * getpagesize();
319 if (guest_limit > guest_max)
320 errx(1, "Not enough memory for devices");
321 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700322}
323
Rusty Russell2e04ef72009-07-30 16:03:45 -0600324/*
325 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700326 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600327 * it falls back to reading the memory in.
328 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700329static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
330{
331 ssize_t r;
332
Rusty Russell2e04ef72009-07-30 16:03:45 -0600333 /*
334 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700335 * The kernel really wants to be writable: it patches its own
336 * instructions.
337 *
338 * MAP_PRIVATE means that the page won't be copied until a write is
339 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600340 * Guests.
341 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700342 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
343 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
344 return;
345
346 /* pread does a seek and a read in one shot: saves a few lines. */
347 r = pread(fd, addr, len, offset);
348 if (r != len)
349 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
350}
351
Rusty Russell2e04ef72009-07-30 16:03:45 -0600352/*
353 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700354 * the Guest memory. ELF = Embedded Linking Format, which is the format used
355 * by all modern binaries on Linux including the kernel.
356 *
357 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000358 * address. We use the physical address; the Guest will map itself to the
359 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700360 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600361 * We return the starting address.
362 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000363static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700364{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700365 Elf32_Phdr phdr[ehdr->e_phnum];
366 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700367
Rusty Russell2e04ef72009-07-30 16:03:45 -0600368 /*
369 * Sanity checks on the main ELF header: an x86 executable with a
370 * reasonable number of correctly-sized program headers.
371 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700372 if (ehdr->e_type != ET_EXEC
373 || ehdr->e_machine != EM_386
374 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
375 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
376 errx(1, "Malformed elf header");
377
Rusty Russell2e04ef72009-07-30 16:03:45 -0600378 /*
379 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700380 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600381 * load where.
382 */
Rusty Russelldde79782007-07-26 10:41:03 -0700383
384 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700385 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
386 err(1, "Seeking to program headers");
387 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
388 err(1, "Reading program headers");
389
Rusty Russell2e04ef72009-07-30 16:03:45 -0600390 /*
391 * Try all the headers: there are usually only three. A read-only one,
392 * a read-write one, and a "note" section which we don't load.
393 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700394 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700395 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700396 if (phdr[i].p_type != PT_LOAD)
397 continue;
398
399 verbose("Section %i: size %i addr %p\n",
400 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
401
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700402 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000403 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700404 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700405 }
406
Rusty Russell814a0e52007-10-22 11:29:44 +1000407 /* The entry point is given in the ELF header. */
408 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700409}
410
Rusty Russell2e04ef72009-07-30 16:03:45 -0600411/*L:150
412 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
413 * to jump into it and it will unpack itself. We used to have to perform some
414 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700415 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000416 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
417 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600418 * the funky header so we know where in the file to load, and away we go!
419 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000420static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700421{
Rusty Russell43d33b22007-10-22 11:29:57 +1000422 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000423 int r;
424 /* Modern bzImages get loaded at 1M. */
425 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700426
Rusty Russell2e04ef72009-07-30 16:03:45 -0600427 /*
428 * Go back to the start of the file and read the header. It should be
429 * a Linux boot header (see Documentation/x86/i386/boot.txt)
430 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000431 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000432 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000433
Rusty Russell43d33b22007-10-22 11:29:57 +1000434 /* Inside the setup_hdr, we expect the magic "HdrS" */
435 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000436 errx(1, "This doesn't look like a bzImage to me");
437
Rusty Russell43d33b22007-10-22 11:29:57 +1000438 /* Skip over the extra sectors of the header. */
439 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000440
441 /* Now read everything into memory. in nice big chunks. */
442 while ((r = read(fd, p, 65536)) > 0)
443 p += r;
444
Rusty Russell43d33b22007-10-22 11:29:57 +1000445 /* Finally, code32_start tells us where to enter the kernel. */
446 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700447}
448
Rusty Russell2e04ef72009-07-30 16:03:45 -0600449/*L:140
450 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000451 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600452 * work, we can load those, too.
453 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000454static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700455{
456 Elf32_Ehdr hdr;
457
Rusty Russelldde79782007-07-26 10:41:03 -0700458 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700459 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
460 err(1, "Reading kernel");
461
Rusty Russelldde79782007-07-26 10:41:03 -0700462 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700463 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000464 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700465
Rusty Russella6bd8e12008-03-28 11:05:53 -0500466 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000467 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700468}
469
Rusty Russell2e04ef72009-07-30 16:03:45 -0600470/*
471 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700472 * it calls getpagesize() twice: "it's dumb code."
473 *
474 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600475 * necessary. I leave this code as a reaction against that.
476 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700477static inline unsigned long page_align(unsigned long addr)
478{
Rusty Russelldde79782007-07-26 10:41:03 -0700479 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700480 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
481}
482
Rusty Russell2e04ef72009-07-30 16:03:45 -0600483/*L:180
484 * An "initial ram disk" is a disk image loaded into memory along with the
485 * kernel which the kernel can use to boot from without needing any drivers.
486 * Most distributions now use this as standard: the initrd contains the code to
487 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700488 *
489 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600490 * kernels. He sent me this (and tells me when I break it).
491 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700492static unsigned long load_initrd(const char *name, unsigned long mem)
493{
494 int ifd;
495 struct stat st;
496 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700497
498 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700499 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700500 if (fstat(ifd, &st) < 0)
501 err(1, "fstat() on initrd '%s'", name);
502
Rusty Russell2e04ef72009-07-30 16:03:45 -0600503 /*
504 * We map the initrd at the top of memory, but mmap wants it to be
505 * page-aligned, so we round the size up for that.
506 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700507 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000508 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600509 /*
510 * Once a file is mapped, you can close the file descriptor. It's a
511 * little odd, but quite useful.
512 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700513 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700514 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700515
516 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700517 return len;
518}
Rusty Russelle1e72962007-10-25 15:02:50 +1000519/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700520
Rusty Russell2e04ef72009-07-30 16:03:45 -0600521/*
522 * Simple routine to roll all the commandline arguments together with spaces
523 * between them.
524 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700525static void concat(char *dst, char *args[])
526{
527 unsigned int i, len = 0;
528
529 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100530 if (i) {
531 strcat(dst+len, " ");
532 len++;
533 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700534 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100535 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700536 }
537 /* In case it's empty. */
538 dst[len] = '\0';
539}
540
Rusty Russell2e04ef72009-07-30 16:03:45 -0600541/*L:185
542 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000543 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300544 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600545 * entry point for the Guest.
546 */
Rusty Russell56739c802009-06-12 22:26:59 -0600547static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700548{
Jes Sorensen511801d2007-10-22 11:03:31 +1000549 unsigned long args[] = { LHREQ_INITIALIZE,
550 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300551 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000552 verbose("Guest: %p - %p (%#lx)\n",
553 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c802009-06-12 22:26:59 -0600554 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
555 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700556 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700557}
Rusty Russelldde79782007-07-26 10:41:03 -0700558/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700559
Rusty Russelle1e72962007-10-25 15:02:50 +1000560/*
Rusty Russelldde79782007-07-26 10:41:03 -0700561 * Device Handling.
562 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000563 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700564 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000565 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700566 * if something funny is going on:
567 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700568static void *_check_pointer(unsigned long addr, unsigned int size,
569 unsigned int line)
570{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600571 /*
572 * We have to separately check addr and addr+size, because size could
573 * be huge and addr + size might wrap around.
574 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000575 if (addr >= guest_limit || addr + size >= guest_limit)
Rusty Russell17cbca22007-10-22 11:24:22 +1000576 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600577 /*
578 * We return a pointer for the caller's convenience, now we know it's
579 * safe to use.
580 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000581 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700582}
Rusty Russelldde79782007-07-26 10:41:03 -0700583/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700584#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
585
Rusty Russell2e04ef72009-07-30 16:03:45 -0600586/*
587 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000588 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600589 * at the end.
590 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100591static unsigned next_desc(struct vring_desc *desc,
592 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000593{
594 unsigned int next;
595
596 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100597 if (!(desc[i].flags & VRING_DESC_F_NEXT))
598 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000599
600 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100601 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000602 /* Make sure compiler knows to grab that: we don't want it changing! */
603 wmb();
604
Mark McLoughlind1f01322009-05-11 18:11:46 +0100605 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000606 errx(1, "Desc next is %u", next);
607
608 return next;
609}
610
Rusty Russell38bc2b82009-06-12 22:27:11 -0600611/* This actually sends the interrupt for this virtqueue */
612static void trigger_irq(struct virtqueue *vq)
613{
614 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
615
Rusty Russell95c517c2009-06-12 22:27:11 -0600616 /* Don't inform them if nothing used. */
617 if (!vq->pending_used)
618 return;
619 vq->pending_used = 0;
620
Rusty Russell38bc2b82009-06-12 22:27:11 -0600621 /* If they don't want an interrupt, don't send one, unless empty. */
622 if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
623 && lg_last_avail(vq) != vq->vring.avail->idx)
624 return;
625
626 /* Send the Guest an interrupt tell them we used something up. */
627 if (write(lguest_fd, buf, sizeof(buf)) != 0)
628 err(1, "Triggering irq %i", vq->config.irq);
629}
630
Rusty Russell2e04ef72009-07-30 16:03:45 -0600631/*
632 * This looks in the virtqueue and for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000633 * it to an iovec for convenient access. Since descriptors consist of some
634 * number of output then some number of input descriptors, it's actually two
635 * iovecs, but we pack them into one and note how many of each there were.
636 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600637 * This function returns the descriptor number found.
638 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600639static unsigned wait_for_vq_desc(struct virtqueue *vq,
640 struct iovec iov[],
641 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000642{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100643 unsigned int i, head, max;
644 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600645 u16 last_avail = lg_last_avail(vq);
646
647 while (last_avail == vq->vring.avail->idx) {
648 u64 event;
649
Rusty Russell38bc2b82009-06-12 22:27:11 -0600650 /* OK, tell Guest about progress up to now. */
651 trigger_irq(vq);
652
Rusty Russellb60da132009-06-12 22:27:12 -0600653 /* OK, now we need to know about added descriptors. */
654 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
655
Rusty Russell2e04ef72009-07-30 16:03:45 -0600656 /*
657 * They could have slipped one in as we were doing that: make
658 * sure it's written, then check again.
659 */
Rusty Russellb60da132009-06-12 22:27:12 -0600660 mb();
661 if (last_avail != vq->vring.avail->idx) {
662 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
663 break;
664 }
665
Rusty Russell659a0e62009-06-12 22:27:10 -0600666 /* Nothing new? Wait for eventfd to tell us they refilled. */
667 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
668 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600669
670 /* We don't need to be notified again. */
671 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600672 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000673
674 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500675 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000676 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500677 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000678
Rusty Russell2e04ef72009-07-30 16:03:45 -0600679 /*
680 * Grab the next descriptor number they're advertising, and increment
681 * the index we've seen.
682 */
Rusty Russellb5111792008-07-29 09:58:34 -0500683 head = vq->vring.avail->ring[last_avail % vq->vring.num];
684 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000685
686 /* If their number is silly, that's a fatal mistake. */
687 if (head >= vq->vring.num)
688 errx(1, "Guest says index %u is available", head);
689
690 /* When we start there are none of either input nor output. */
691 *out_num = *in_num = 0;
692
Mark McLoughlind1f01322009-05-11 18:11:46 +0100693 max = vq->vring.num;
694 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000695 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100696
Rusty Russell2e04ef72009-07-30 16:03:45 -0600697 /*
698 * If this is an indirect entry, then this buffer contains a descriptor
699 * table which we handle as if it's any normal descriptor chain.
700 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100701 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
702 if (desc[i].len % sizeof(struct vring_desc))
703 errx(1, "Invalid size for indirect buffer table");
704
705 max = desc[i].len / sizeof(struct vring_desc);
706 desc = check_pointer(desc[i].addr, desc[i].len);
707 i = 0;
708 }
709
Rusty Russell17cbca22007-10-22 11:24:22 +1000710 do {
711 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100712 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000713 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100714 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000715 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100716 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000717 (*in_num)++;
718 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600719 /*
720 * If it's an output descriptor, they're all supposed
721 * to come before any input descriptors.
722 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000723 if (*in_num)
724 errx(1, "Descriptor has out after in");
725 (*out_num)++;
726 }
727
728 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100729 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000730 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100731 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000732
733 return head;
734}
735
Rusty Russell2e04ef72009-07-30 16:03:45 -0600736/*
737 * After we've used one of their buffers, we tell them about it. We'll then
738 * want to send them an interrupt, using trigger_irq().
739 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000740static void add_used(struct virtqueue *vq, unsigned int head, int len)
741{
742 struct vring_used_elem *used;
743
Rusty Russell2e04ef72009-07-30 16:03:45 -0600744 /*
745 * The virtqueue contains a ring of used buffers. Get a pointer to the
746 * next entry in that used ring.
747 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000748 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
749 used->id = head;
750 used->len = len;
751 /* Make sure buffer is written before we update index. */
752 wmb();
753 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600754 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000755}
756
Rusty Russell17cbca22007-10-22 11:24:22 +1000757/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c802009-06-12 22:26:59 -0600758static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000759{
760 add_used(vq, head, len);
Rusty Russell56739c802009-06-12 22:26:59 -0600761 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000762}
763
Rusty Russelle1e72962007-10-25 15:02:50 +1000764/*
765 * The Console
766 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600767 * We associate some data with the console for our exit hack.
768 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700769struct console_abort
770{
Rusty Russelldde79782007-07-26 10:41:03 -0700771 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700772 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700773 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700774 struct timeval start;
775};
776
Rusty Russelldde79782007-07-26 10:41:03 -0700777/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600778static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700779{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700780 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000781 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600782 struct console_abort *abort = vq->dev->priv;
783 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700784
Rusty Russell659a0e62009-06-12 22:27:10 -0600785 /* Make sure there's a descriptor waiting. */
786 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000787 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000788 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700789
Rusty Russell659a0e62009-06-12 22:27:10 -0600790 /* Read it in. */
791 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700792 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600793 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700794 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600795 /*
796 * For simplicity, dying threads kill the whole Launcher. So
797 * just nap here.
798 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600799 for (;;)
800 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700801 }
802
Rusty Russell659a0e62009-06-12 22:27:10 -0600803 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700804
Rusty Russell2e04ef72009-07-30 16:03:45 -0600805 /*
806 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700807 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600808 * This is such a hack, but works surprisingly well. Each ^C has to
809 * be in a buffer by itself, so they can't be too fast. But we check
810 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600811 * slow.
812 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600813 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700814 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600815 return;
816 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700817
Rusty Russell659a0e62009-06-12 22:27:10 -0600818 abort->count++;
819 if (abort->count == 1)
820 gettimeofday(&abort->start, NULL);
821 else if (abort->count == 3) {
822 struct timeval now;
823 gettimeofday(&now, NULL);
824 /* Kill all Launcher processes with SIGINT, like normal ^C */
825 if (now.tv_sec <= abort->start.tv_sec+1)
826 kill(0, SIGINT);
827 abort->count = 0;
828 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700829}
830
Rusty Russell659a0e62009-06-12 22:27:10 -0600831/* This is the routine which handles console output (ie. stdout). */
832static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700833{
Rusty Russell17cbca22007-10-22 11:24:22 +1000834 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000835 struct iovec iov[vq->vring.num];
836
Rusty Russell659a0e62009-06-12 22:27:10 -0600837 head = wait_for_vq_desc(vq, iov, &out, &in);
838 if (in)
839 errx(1, "Input buffers in console output queue?");
840 while (!iov_empty(iov, out)) {
841 int len = writev(STDOUT_FILENO, iov, out);
842 if (len <= 0)
843 err(1, "Write to stdout gave %i", len);
844 iov_consume(iov, out, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000845 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600846 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500847}
848
Rusty Russelle1e72962007-10-25 15:02:50 +1000849/*
850 * The Network
851 *
852 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600853 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500854 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600855struct net_info {
856 int tunfd;
857};
858
859static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700860{
Rusty Russell659a0e62009-06-12 22:27:10 -0600861 struct net_info *net_info = vq->dev->priv;
862 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000863 struct iovec iov[vq->vring.num];
864
Rusty Russell659a0e62009-06-12 22:27:10 -0600865 head = wait_for_vq_desc(vq, iov, &out, &in);
866 if (in)
867 errx(1, "Input buffers in net output queue?");
868 if (writev(net_info->tunfd, iov, out) < 0)
869 errx(1, "Write to tun failed?");
Rusty Russell38bc2b82009-06-12 22:27:11 -0600870 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700871}
872
Rusty Russell4a8962e2009-06-12 22:27:12 -0600873/* Will reading from this file descriptor block? */
874static bool will_block(int fd)
875{
876 fd_set fdset;
877 struct timeval zero = { 0, 0 };
878 FD_ZERO(&fdset);
879 FD_SET(fd, &fdset);
880 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
881}
882
Rusty Russell2e04ef72009-07-30 16:03:45 -0600883/* This handles packets coming in from the tun device to our Guest. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600884static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700885{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700886 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600887 unsigned int head, out, in;
888 struct iovec iov[vq->vring.num];
889 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700890
Rusty Russell659a0e62009-06-12 22:27:10 -0600891 head = wait_for_vq_desc(vq, iov, &out, &in);
892 if (out)
893 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600894
895 /* Deliver interrupt now, since we're about to sleep. */
896 if (vq->pending_used && will_block(net_info->tunfd))
897 trigger_irq(vq);
898
Rusty Russell659a0e62009-06-12 22:27:10 -0600899 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700900 if (len <= 0)
Rusty Russell659a0e62009-06-12 22:27:10 -0600901 err(1, "Failed to read from tun.");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600902 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700903}
904
Rusty Russell659a0e62009-06-12 22:27:10 -0600905/* This is the helper to create threads. */
906static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000907{
Rusty Russell659a0e62009-06-12 22:27:10 -0600908 struct virtqueue *vq = _vq;
909
910 for (;;)
911 vq->service(vq);
912 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000913}
914
Rusty Russell2e04ef72009-07-30 16:03:45 -0600915/*
916 * When a child dies, we kill our entire process group with SIGTERM. This
917 * also has the side effect that the shell restores the console for us!
918 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600919static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500920{
Rusty Russell659a0e62009-06-12 22:27:10 -0600921 kill(0, SIGTERM);
922}
923
924static void reset_device(struct device *dev)
925{
926 struct virtqueue *vq;
927
928 verbose("Resetting device %s\n", dev->name);
929
930 /* Clear any features they've acked. */
931 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
932
933 /* We're going to be explicitly killing threads, so ignore them. */
934 signal(SIGCHLD, SIG_IGN);
935
936 /* Zero out the virtqueues, get rid of their threads */
937 for (vq = dev->vq; vq; vq = vq->next) {
938 if (vq->thread != (pid_t)-1) {
939 kill(vq->thread, SIGTERM);
940 waitpid(vq->thread, NULL, 0);
941 vq->thread = (pid_t)-1;
942 }
943 memset(vq->vring.desc, 0,
944 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
945 lg_last_avail(vq) = 0;
946 }
947 dev->running = false;
948
949 /* Now we care if threads die. */
950 signal(SIGCHLD, (void *)kill_launcher);
951}
952
953static void create_thread(struct virtqueue *vq)
954{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600955 /*
956 * Create stack for thread and run it. Since the stack grows upwards,
957 * we point the stack pointer to the end of this region.
958 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600959 char *stack = malloc(32768);
960 unsigned long args[] = { LHREQ_EVENTFD,
961 vq->config.pfn*getpagesize(), 0 };
962
963 /* Create a zero-initialized eventfd. */
964 vq->eventfd = eventfd(0, 0);
965 if (vq->eventfd < 0)
966 err(1, "Creating eventfd");
967 args[2] = vq->eventfd;
968
969 /* Attach an eventfd to this virtqueue: it will go off
970 * when the Guest does an LHCALL_NOTIFY for this vq. */
971 if (write(lguest_fd, &args, sizeof(args)) != 0)
972 err(1, "Attaching eventfd");
973
974 /* CLONE_VM: because it has to access the Guest memory, and
975 * SIGCHLD so we get a signal if it dies. */
976 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
977 if (vq->thread == (pid_t)-1)
978 err(1, "Creating clone");
979 /* We close our local copy, now the child has it. */
980 close(vq->eventfd);
981}
982
983static void start_device(struct device *dev)
984{
985 unsigned int i;
986 struct virtqueue *vq;
987
988 verbose("Device %s OK: offered", dev->name);
989 for (i = 0; i < dev->feature_len; i++)
990 verbose(" %02x", get_feature_bits(dev)[i]);
991 verbose(", accepted");
992 for (i = 0; i < dev->feature_len; i++)
993 verbose(" %02x", get_feature_bits(dev)
994 [dev->feature_len+i]);
995
996 for (vq = dev->vq; vq; vq = vq->next) {
997 if (vq->service)
998 create_thread(vq);
999 }
1000 dev->running = true;
1001}
1002
1003static void cleanup_devices(void)
1004{
1005 struct device *dev;
1006
1007 for (dev = devices.dev; dev; dev = dev->next)
1008 reset_device(dev);
1009
1010 /* If we saved off the original terminal settings, restore them now. */
1011 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1012 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001013}
1014
Rusty Russella007a752008-05-02 21:50:53 -05001015/* When the Guest tells us they updated the status field, we handle it. */
1016static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001017{
Rusty Russell659a0e62009-06-12 22:27:10 -06001018 /* A zero status is a reset, otherwise it's a set of flags. */
1019 if (dev->desc->status == 0)
1020 reset_device(dev);
1021 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001022 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001023 if (dev->running)
1024 reset_device(dev);
Rusty Russella007a752008-05-02 21:50:53 -05001025 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001026 if (!dev->running)
1027 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001028 }
1029}
1030
Rusty Russell17cbca22007-10-22 11:24:22 +10001031/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
Rusty Russell56739c802009-06-12 22:26:59 -06001032static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001033{
1034 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001035
Rusty Russell659a0e62009-06-12 22:27:10 -06001036 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001037 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001038 struct virtqueue *vq;
1039
Rusty Russella007a752008-05-02 21:50:53 -05001040 /* Notifications to device descriptors update device status. */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001041 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001042 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001043 return;
1044 }
1045
Rusty Russell659a0e62009-06-12 22:27:10 -06001046 /* Devices *can* be used before status is set to DRIVER_OK. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001047 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001048 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001049 continue;
Rusty Russell659a0e62009-06-12 22:27:10 -06001050 if (i->running)
1051 errx(1, "Notification on running %s", i->name);
1052 start_device(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001053 return;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001054 }
1055 }
Rusty Russelldde79782007-07-26 10:41:03 -07001056
Rusty Russell2e04ef72009-07-30 16:03:45 -06001057 /*
1058 * Early console write is done using notify on a nul-terminated string
1059 * in Guest memory. It's also great for hacking debugging messages
1060 * into a Guest.
1061 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001062 if (addr >= guest_limit)
1063 errx(1, "Bad NOTIFY %#lx", addr);
1064
1065 write(STDOUT_FILENO, from_guest_phys(addr),
1066 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001067}
1068
Rusty Russelldde79782007-07-26 10:41:03 -07001069/*L:190
1070 * Device Setup
1071 *
1072 * All devices need a descriptor so the Guest knows it exists, and a "struct
1073 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001074 * routines to allocate and manage them.
1075 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001076
Rusty Russell2e04ef72009-07-30 16:03:45 -06001077/*
1078 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001079 * number of virtqueue descriptors, then two sets of feature bits, then an
1080 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001081 * pointer.
1082 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001083static u8 *device_config(const struct device *dev)
1084{
1085 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001086 + dev->num_vq * sizeof(struct lguest_vqconfig)
1087 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001088}
1089
Rusty Russell2e04ef72009-07-30 16:03:45 -06001090/*
1091 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001092 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001093 * that descriptor.
1094 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001095static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001096{
Rusty Russella586d4f2008-02-04 23:49:56 -05001097 struct lguest_device_desc d = { .type = type };
1098 void *p;
1099
1100 /* Figure out where the next device config is, based on the last one. */
1101 if (devices.lastdev)
1102 p = device_config(devices.lastdev)
1103 + devices.lastdev->desc->config_len;
1104 else
1105 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001106
Rusty Russell17cbca22007-10-22 11:24:22 +10001107 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001108 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001109 errx(1, "Too many devices");
1110
Rusty Russella586d4f2008-02-04 23:49:56 -05001111 /* p might not be aligned, so we memcpy in. */
1112 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001113}
1114
Rusty Russell2e04ef72009-07-30 16:03:45 -06001115/*
1116 * Each device descriptor is followed by the description of its virtqueues. We
1117 * specify how many descriptors the virtqueue is to have.
1118 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001119static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001120 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001121{
1122 unsigned int pages;
1123 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1124 void *p;
1125
Rusty Russella6bd8e12008-03-28 11:05:53 -05001126 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001127 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001128 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001129 p = get_pages(pages);
1130
Rusty Russelld1c856e2007-11-19 11:20:40 -05001131 /* Initialize the virtqueue */
1132 vq->next = NULL;
1133 vq->last_avail_idx = 0;
1134 vq->dev = dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001135 vq->service = service;
1136 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001137
Rusty Russell17cbca22007-10-22 11:24:22 +10001138 /* Initialize the configuration. */
1139 vq->config.num = num_descs;
1140 vq->config.irq = devices.next_irq++;
1141 vq->config.pfn = to_guest_phys(p) / getpagesize();
1142
1143 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001144 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001145
Rusty Russell2e04ef72009-07-30 16:03:45 -06001146 /*
1147 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001148 * device_config() to get the end of the device's current virtqueues;
1149 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001150 * yet, otherwise we'd be overwriting them.
1151 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001152 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1153 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001154 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001155 dev->desc->num_vq++;
1156
1157 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001158
Rusty Russell2e04ef72009-07-30 16:03:45 -06001159 /*
1160 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1161 * second.
1162 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001163 for (i = &dev->vq; *i; i = &(*i)->next);
1164 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001165}
1166
Rusty Russell2e04ef72009-07-30 16:03:45 -06001167/*
1168 * The first half of the feature bitmask is for us to advertise features. The
1169 * second half is for the Guest to accept features.
1170 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001171static void add_feature(struct device *dev, unsigned bit)
1172{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001173 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001174
1175 /* We can't extend the feature bits once we've added config bytes */
1176 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1177 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001178 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001179 }
1180
Rusty Russella586d4f2008-02-04 23:49:56 -05001181 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1182}
1183
Rusty Russell2e04ef72009-07-30 16:03:45 -06001184/*
1185 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001186 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001187 * how we use it.
1188 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001189static void set_config(struct device *dev, unsigned len, const void *conf)
1190{
1191 /* Check we haven't overflowed our single page. */
1192 if (device_config(dev) + len > devices.descpage + getpagesize())
1193 errx(1, "Too many devices");
1194
1195 /* Copy in the config information, and store the length. */
1196 memcpy(device_config(dev), conf, len);
1197 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001198
1199 /* Size must fit in config_len field (8 bits)! */
1200 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001201}
1202
Rusty Russell2e04ef72009-07-30 16:03:45 -06001203/*
1204 * This routine does all the creation and setup of a new device, including
Rusty Russella6bd8e12008-03-28 11:05:53 -05001205 * calling new_dev_desc() to allocate the descriptor and device memory.
1206 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001207 * See what I mean about userspace being boring?
1208 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001209static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001210{
1211 struct device *dev = malloc(sizeof(*dev));
1212
Rusty Russelldde79782007-07-26 10:41:03 -07001213 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001214 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001215 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001216 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001217 dev->feature_len = 0;
1218 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001219 dev->running = false;
Rusty Russella586d4f2008-02-04 23:49:56 -05001220
Rusty Russell2e04ef72009-07-30 16:03:45 -06001221 /*
1222 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001223 * easier, but the user expects the devices to be arranged on the bus
1224 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001225 * is eth0, the first block device /dev/vda, etc.
1226 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001227 if (devices.lastdev)
1228 devices.lastdev->next = dev;
1229 else
1230 devices.dev = dev;
1231 devices.lastdev = dev;
1232
Rusty Russell8ca47e02007-07-19 01:49:29 -07001233 return dev;
1234}
1235
Rusty Russell2e04ef72009-07-30 16:03:45 -06001236/*
1237 * Our first setup routine is the console. It's a fairly simple device, but
1238 * UNIX tty handling makes it uglier than it could be.
1239 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001240static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001241{
1242 struct device *dev;
1243
Rusty Russelldde79782007-07-26 10:41:03 -07001244 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001245 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1246 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001247 /*
1248 * Then we turn off echo, line buffering and ^C etc: We want a
1249 * raw input stream to the Guest.
1250 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001251 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1252 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001253 }
1254
Rusty Russell659a0e62009-06-12 22:27:10 -06001255 dev = new_device("console", VIRTIO_ID_CONSOLE);
1256
Rusty Russelldde79782007-07-26 10:41:03 -07001257 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001258 dev->priv = malloc(sizeof(struct console_abort));
1259 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001260
Rusty Russell2e04ef72009-07-30 16:03:45 -06001261 /*
1262 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001263 * they put something the input queue, we make sure we're listening to
1264 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001265 * stdout.
1266 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001267 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1268 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001269
Rusty Russell659a0e62009-06-12 22:27:10 -06001270 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001271}
Rusty Russelldde79782007-07-26 10:41:03 -07001272/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001273
Rusty Russell2e04ef72009-07-30 16:03:45 -06001274/*M:010
1275 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001276 * --sharenet=<name> option which opens or creates a named pipe. This can be
1277 * used to send packets to another guest in a 1:1 manner.
1278 *
1279 * More sopisticated is to use one of the tools developed for project like UML
1280 * to do networking.
1281 *
1282 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1283 * completely generic ("here's my vring, attach to your vring") and would work
1284 * for any traffic. Of course, namespace and permissions issues need to be
1285 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1286 * multiple inter-guest channels behind one interface, although it would
1287 * require some manner of hotplugging new virtio channels.
1288 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001289 * Finally, we could implement a virtio network switch in the kernel.
1290:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001291
Rusty Russell8ca47e02007-07-19 01:49:29 -07001292static u32 str2ip(const char *ipaddr)
1293{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001294 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001295
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001296 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1297 errx(1, "Failed to parse IP address '%s'", ipaddr);
1298 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1299}
1300
1301static void str2mac(const char *macaddr, unsigned char mac[6])
1302{
1303 unsigned int m[6];
1304 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1305 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1306 errx(1, "Failed to parse mac address '%s'", macaddr);
1307 mac[0] = m[0];
1308 mac[1] = m[1];
1309 mac[2] = m[2];
1310 mac[3] = m[3];
1311 mac[4] = m[4];
1312 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001313}
1314
Rusty Russell2e04ef72009-07-30 16:03:45 -06001315/*
1316 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001317 * network device to the bridge device specified by the command line.
1318 *
1319 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001320 * dislike bridging), and I just try not to break it.
1321 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001322static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1323{
1324 int ifidx;
1325 struct ifreq ifr;
1326
1327 if (!*br_name)
1328 errx(1, "must specify bridge name");
1329
1330 ifidx = if_nametoindex(if_name);
1331 if (!ifidx)
1332 errx(1, "interface %s does not exist!", if_name);
1333
1334 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001335 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001336 ifr.ifr_ifindex = ifidx;
1337 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1338 err(1, "can't add %s to bridge %s", if_name, br_name);
1339}
1340
Rusty Russell2e04ef72009-07-30 16:03:45 -06001341/*
1342 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001343 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001344 * pointer.
1345 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001346static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001347{
1348 struct ifreq ifr;
1349 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1350
1351 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001352 strcpy(ifr.ifr_name, tapif);
1353
1354 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001355 sin->sin_family = AF_INET;
1356 sin->sin_addr.s_addr = htonl(ipaddr);
1357 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001358 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001359 ifr.ifr_flags = IFF_UP;
1360 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001361 err(1, "Bringing interface %s up", tapif);
1362}
1363
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001364static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001365{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001366 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001367 int netfd;
1368
1369 /* Start with this zeroed. Messy but sure. */
1370 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001371
Rusty Russell2e04ef72009-07-30 16:03:45 -06001372 /*
1373 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001374 * tap device is like a tun device, only somehow different. To tell
1375 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001376 * works now!
1377 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001378 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001379 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001380 strcpy(ifr.ifr_name, "tap%d");
1381 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1382 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001383
Rusty Russell398f1872008-07-29 09:58:37 -05001384 if (ioctl(netfd, TUNSETOFFLOAD,
1385 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1386 err(1, "Could not set features for tun device");
1387
Rusty Russell2e04ef72009-07-30 16:03:45 -06001388 /*
1389 * We don't need checksums calculated for packets coming in this
1390 * device: trust us!
1391 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001392 ioctl(netfd, TUNSETNOCSUM, 1);
1393
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001394 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1395 return netfd;
1396}
1397
Rusty Russell2e04ef72009-07-30 16:03:45 -06001398/*L:195
1399 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001400 * routing, but the principle is the same: it uses the "tun" device to inject
1401 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001402 * just shunt packets between the Guest and the tun device.
1403 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001404static void setup_tun_net(char *arg)
1405{
1406 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001407 struct net_info *net_info = malloc(sizeof(*net_info));
1408 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001409 u32 ip = INADDR_ANY;
1410 bool bridging = false;
1411 char tapif[IFNAMSIZ], *p;
1412 struct virtio_net_config conf;
1413
Rusty Russell659a0e62009-06-12 22:27:10 -06001414 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001415
Rusty Russell17cbca22007-10-22 11:24:22 +10001416 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001417 dev = new_device("net", VIRTIO_ID_NET);
1418 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001419
Rusty Russell2e04ef72009-07-30 16:03:45 -06001420 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001421 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1422 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001423
Rusty Russell2e04ef72009-07-30 16:03:45 -06001424 /*
1425 * We need a socket to perform the magic network ioctls to bring up the
1426 * tap interface, connect to the bridge etc. Any socket will do!
1427 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001428 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1429 if (ipfd < 0)
1430 err(1, "opening IP socket");
1431
Rusty Russelldde79782007-07-26 10:41:03 -07001432 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001433 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001434 arg += strlen(BRIDGE_PFX);
1435 bridging = true;
1436 }
1437
1438 /* A mac address may follow the bridge name or IP address */
1439 p = strchr(arg, ':');
1440 if (p) {
1441 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001442 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001443 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001444 }
1445
1446 /* arg is now either an IP address or a bridge name */
1447 if (bridging)
1448 add_to_bridge(ipfd, tapif, arg);
1449 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001450 ip = str2ip(arg);
1451
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001452 /* Set up the tun device. */
1453 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001454
Rusty Russell20887612008-05-30 15:09:46 -05001455 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
Rusty Russell398f1872008-07-29 09:58:37 -05001456 /* Expect Guest to handle everything except UFO */
1457 add_feature(dev, VIRTIO_NET_F_CSUM);
1458 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001459 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1460 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1461 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1462 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1463 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1464 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001465 /* We handle indirect ring entries */
1466 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russella586d4f2008-02-04 23:49:56 -05001467 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001468
Rusty Russella586d4f2008-02-04 23:49:56 -05001469 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001470 close(ipfd);
1471
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001472 devices.device_num++;
1473
1474 if (bridging)
1475 verbose("device %u: tun %s attached to bridge: %s\n",
1476 devices.device_num, tapif, arg);
1477 else
1478 verbose("device %u: tun %s: %s\n",
1479 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001480}
Rusty Russell17cbca22007-10-22 11:24:22 +10001481
Rusty Russell2e04ef72009-07-30 16:03:45 -06001482/*
1483 * Our block (disk) device should be really simple: the Guest asks for a block
Rusty Russelle1e72962007-10-25 15:02:50 +10001484 * number and we read or write that position in the file. Unfortunately, that
1485 * was amazingly slow: the Guest waits until the read is finished before
1486 * running anything else, even if it could have been doing useful work.
Rusty Russell17cbca22007-10-22 11:24:22 +10001487 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001488 * We could use async I/O, except it's reputed to suck so hard that characters
1489 * actually go missing from your code when you try to use it.
Rusty Russell17cbca22007-10-22 11:24:22 +10001490 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001491 * So this was one reason why lguest now does all virtqueue servicing in
1492 * separate threads: it's more efficient and more like a real device.
1493 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001494
Rusty Russelle1e72962007-10-25 15:02:50 +10001495/* This hangs off device->priv. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001496struct vblk_info
1497{
1498 /* The size of the file. */
1499 off64_t len;
1500
1501 /* The file descriptor for the file. */
1502 int fd;
1503
1504 /* IO thread listens on this file descriptor [0]. */
1505 int workpipe[2];
1506
1507 /* IO thread writes to this file descriptor to mark it done, then
1508 * Launcher triggers interrupt to Guest. */
1509 int done_fd;
1510};
1511
Rusty Russelle1e72962007-10-25 15:02:50 +10001512/*L:210
1513 * The Disk
1514 *
1515 * Remember that the block device is handled by a separate I/O thread. We head
1516 * straight into the core of that thread here:
1517 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001518static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001519{
Rusty Russell659a0e62009-06-12 22:27:10 -06001520 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001521 unsigned int head, out_num, in_num, wlen;
1522 int ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001523 u8 *in;
Rusty Russell17cbca22007-10-22 11:24:22 +10001524 struct virtio_blk_outhdr *out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001525 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001526 off64_t off;
1527
Rusty Russell659a0e62009-06-12 22:27:10 -06001528 /* Get the next request. */
1529 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001530
Rusty Russell2e04ef72009-07-30 16:03:45 -06001531 /*
1532 * Every block request should contain at least one output buffer
Rusty Russelle1e72962007-10-25 15:02:50 +10001533 * (detailing the location on disk and the type of request) and one
Rusty Russell2e04ef72009-07-30 16:03:45 -06001534 * input buffer (to hold the result).
1535 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001536 if (out_num == 0 || in_num == 0)
1537 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1538 head, out_num, in_num);
1539
1540 out = convert(&iov[0], struct virtio_blk_outhdr);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001541 in = convert(&iov[out_num+in_num-1], u8);
Rusty Russell17cbca22007-10-22 11:24:22 +10001542 off = out->sector * 512;
1543
Rusty Russell2e04ef72009-07-30 16:03:45 -06001544 /*
1545 * The block device implements "barriers", where the Guest indicates
Rusty Russelle1e72962007-10-25 15:02:50 +10001546 * that it wants all previous writes to occur before this write. We
1547 * don't have a way of asking our kernel to do a barrier, so we just
Rusty Russell2e04ef72009-07-30 16:03:45 -06001548 * synchronize all the data in the file. Pretty poor, no?
1549 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001550 if (out->type & VIRTIO_BLK_T_BARRIER)
1551 fdatasync(vblk->fd);
1552
Rusty Russell2e04ef72009-07-30 16:03:45 -06001553 /*
1554 * In general the virtio block driver is allowed to try SCSI commands.
1555 * It'd be nice if we supported eject, for example, but we don't.
1556 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001557 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1558 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001559 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001560 wlen = sizeof(*in);
Rusty Russell17cbca22007-10-22 11:24:22 +10001561 } else if (out->type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001562 /*
1563 * Write
1564 *
1565 * Move to the right location in the block file. This can fail
1566 * if they try to write past end.
1567 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001568 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1569 err(1, "Bad seek to sector %llu", out->sector);
1570
1571 ret = writev(vblk->fd, iov+1, out_num-1);
1572 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1573
Rusty Russell2e04ef72009-07-30 16:03:45 -06001574 /*
1575 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001576 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001577 * file (possibly extending it).
1578 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001579 if (ret > 0 && off + ret > vblk->len) {
1580 /* Trim it back to the correct length */
1581 ftruncate64(vblk->fd, vblk->len);
1582 /* Die, bad Guest, die. */
1583 errx(1, "Write past end %llu+%u", off, ret);
1584 }
Anthony Liguori1200e642007-11-08 21:13:44 -06001585 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001586 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001587 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001588 /*
1589 * Read
1590 *
1591 * Move to the right location in the block file. This can fail
1592 * if they try to read past end.
1593 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001594 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1595 err(1, "Bad seek to sector %llu", out->sector);
1596
1597 ret = readv(vblk->fd, iov+1, in_num-1);
1598 verbose("READ from sector %llu: %i\n", out->sector, ret);
1599 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001600 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001601 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001602 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001603 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001604 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001605 }
1606 }
1607
Rusty Russell2e04ef72009-07-30 16:03:45 -06001608 /*
1609 * OK, so we noted that it was pretty poor to use an fdatasync as a
Rusty Russelld1881d32009-03-30 21:55:25 -06001610 * barrier. But Christoph Hellwig points out that we need a sync
1611 * *afterwards* as well: "Barriers specify no reordering to the front
Rusty Russell2e04ef72009-07-30 16:03:45 -06001612 * or the back." And Jens Axboe confirmed it, so here we are:
1613 */
Rusty Russelld1881d32009-03-30 21:55:25 -06001614 if (out->type & VIRTIO_BLK_T_BARRIER)
1615 fdatasync(vblk->fd);
1616
Rusty Russell38bc2b82009-06-12 22:27:11 -06001617 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001618}
1619
Rusty Russelle1e72962007-10-25 15:02:50 +10001620/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001621static void setup_block_file(const char *filename)
1622{
Rusty Russell17cbca22007-10-22 11:24:22 +10001623 struct device *dev;
1624 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001625 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001626
Rusty Russell2e04ef72009-07-30 16:03:45 -06001627 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001628 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001629
Rusty Russelle1e72962007-10-25 15:02:50 +10001630 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001631 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001632
1633 /* Allocate the room for our own bookkeeping */
1634 vblk = dev->priv = malloc(sizeof(*vblk));
1635
1636 /* First we open the file and store the length. */
1637 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1638 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1639
Rusty Russella586d4f2008-02-04 23:49:56 -05001640 /* We support barriers. */
1641 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1642
Rusty Russell17cbca22007-10-22 11:24:22 +10001643 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001644 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001645
Rusty Russell2e04ef72009-07-30 16:03:45 -06001646 /*
1647 * Tell Guest not to put in too many descriptors at once: two are used
1648 * for the in and out elements.
1649 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001650 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1651 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1652
Rusty Russell8ef562d2009-07-30 16:03:43 -06001653 /* Don't try to put whole struct: we have 8 bit limit. */
1654 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001655
Rusty Russell17cbca22007-10-22 11:24:22 +10001656 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001657 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001658}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001659
Rusty Russell2e04ef72009-07-30 16:03:45 -06001660/*L:211
1661 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001662 * input buffers. The usual case is that the Guest doesn't want random numbers
1663 * and so has no buffers although /dev/random is still readable, whereas
1664 * console is the reverse.
1665 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001666 * The same logic applies, however.
1667 */
1668struct rng_info {
1669 int rfd;
1670};
1671
Rusty Russell659a0e62009-06-12 22:27:10 -06001672static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001673{
1674 int len;
1675 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001676 struct rng_info *rng_info = vq->dev->priv;
1677 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001678
1679 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001680 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001681 if (out_num)
1682 errx(1, "Output buffers in rng?");
1683
Rusty Russell2e04ef72009-07-30 16:03:45 -06001684 /*
1685 * This is why we convert to iovecs: the readv() call uses them, and so
Rusty Russell28fd6d72008-07-29 09:58:33 -05001686 * it reads straight into the Guest's buffer. We loop to make sure we
Rusty Russell2e04ef72009-07-30 16:03:45 -06001687 * fill it.
1688 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001689 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001690 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001691 if (len <= 0)
1692 err(1, "Read from /dev/random gave %i", len);
1693 iov_consume(iov, in_num, len);
1694 totlen += len;
1695 }
1696
1697 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001698 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001699}
1700
Rusty Russell2e04ef72009-07-30 16:03:45 -06001701/*L:199
1702 * This creates a "hardware" random number device for the Guest.
1703 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001704static void setup_rng(void)
1705{
1706 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001707 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001708
Rusty Russell2e04ef72009-07-30 16:03:45 -06001709 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001710 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001711
Rusty Russell2e04ef72009-07-30 16:03:45 -06001712 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001713 dev = new_device("rng", VIRTIO_ID_RNG);
1714 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001715
1716 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001717 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001718
1719 verbose("device %u: rng\n", devices.device_num++);
1720}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001721/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301722
Rusty Russella6bd8e12008-03-28 11:05:53 -05001723/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301724static void __attribute__((noreturn)) restart_guest(void)
1725{
1726 unsigned int i;
1727
Rusty Russell2e04ef72009-07-30 16:03:45 -06001728 /*
1729 * Since we don't track all open fds, we simply close everything beyond
1730 * stderr.
1731 */
Balaji Raoec04b132007-12-28 14:26:24 +05301732 for (i = 3; i < FD_SETSIZE; i++)
1733 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001734
Rusty Russell659a0e62009-06-12 22:27:10 -06001735 /* Reset all the devices (kills all threads). */
1736 cleanup_devices();
1737
Balaji Raoec04b132007-12-28 14:26:24 +05301738 execv(main_args[0], main_args);
1739 err(1, "Could not exec %s", main_args[0]);
1740}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001741
Rusty Russell2e04ef72009-07-30 16:03:45 -06001742/*L:220
1743 * Finally we reach the core of the Launcher which runs the Guest, serves
1744 * its input and output, and finally, lays it to rest.
1745 */
Rusty Russell56739c802009-06-12 22:26:59 -06001746static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001747{
1748 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001749 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001750 int readval;
1751
1752 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001753 readval = pread(lguest_fd, &notify_addr,
1754 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001755
Rusty Russell17cbca22007-10-22 11:24:22 +10001756 /* One unsigned long means the Guest did HCALL_NOTIFY */
1757 if (readval == sizeof(notify_addr)) {
1758 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c802009-06-12 22:26:59 -06001759 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001760 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001761 } else if (errno == ENOENT) {
1762 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001763 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001764 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301765 /* ERESTART means that we need to reboot the guest */
1766 } else if (errno == ERESTART) {
1767 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001768 /* Anything else means a bug or incompatible change. */
1769 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001770 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001771 }
1772}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001773/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001774 * This is the end of the Launcher. The good news: we are over halfway
1775 * through! The bad news: the most fiendish part of the code still lies ahead
1776 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001777 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001778 * Are you ready? Take a deep breath and join me in the core of the Host, in
1779 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001780:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001781
1782static struct option opts[] = {
1783 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001784 { "tunnet", 1, NULL, 't' },
1785 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001786 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001787 { "initrd", 1, NULL, 'i' },
1788 { NULL },
1789};
1790static void usage(void)
1791{
1792 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001793 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001794 "|--block=<filename>|--initrd=<filename>]...\n"
1795 "<mem-in-mb> vmlinux [args...]");
1796}
1797
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001798/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001799int main(int argc, char *argv[])
1800{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001801 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001802 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c802009-06-12 22:26:59 -06001803 /* Two temporaries. */
1804 int i, c;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001805 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001806 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001807 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001808 const char *initrd_name = NULL;
1809
Balaji Raoec04b132007-12-28 14:26:24 +05301810 /* Save the args: we "reboot" by execing ourselves again. */
1811 main_args = argv;
Balaji Raoec04b132007-12-28 14:26:24 +05301812
Rusty Russell2e04ef72009-07-30 16:03:45 -06001813 /*
1814 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001815 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001816 * remember that 0 is used by the timer).
1817 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001818 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001819 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001820
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001821 cpu_id = 0;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001822 /*
1823 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001824 * descriptor and memory pages for the devices as we parse the command
1825 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001826 * of memory now.
1827 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001828 for (i = 1; i < argc; i++) {
1829 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001830 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001831 /*
1832 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001833 * guest-physical memory range. This fills it with 0,
1834 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001835 * tries to access it.
1836 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001837 guest_base = map_zeroed_pages(mem / getpagesize()
1838 + DEVICE_PAGES);
1839 guest_limit = mem;
1840 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001841 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001842 break;
1843 }
1844 }
Rusty Russelldde79782007-07-26 10:41:03 -07001845
1846 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001847 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1848 switch (c) {
1849 case 'v':
1850 verbose = true;
1851 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001852 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001853 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001854 break;
1855 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001856 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001857 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001858 case 'r':
1859 setup_rng();
1860 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001861 case 'i':
1862 initrd_name = optarg;
1863 break;
1864 default:
1865 warnx("Unknown argument %s", argv[optind]);
1866 usage();
1867 }
1868 }
Rusty Russell2e04ef72009-07-30 16:03:45 -06001869 /*
1870 * After the other arguments we expect memory and kernel image name,
1871 * followed by command line arguments for the kernel.
1872 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001873 if (optind + 2 > argc)
1874 usage();
1875
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001876 verbose("Guest base is at %p\n", guest_base);
1877
Rusty Russelldde79782007-07-26 10:41:03 -07001878 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001879 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001880
Rusty Russell8ca47e02007-07-19 01:49:29 -07001881 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001882 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001883
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001884 /* Boot information is stashed at physical address 0 */
1885 boot = from_guest_phys(0);
1886
Rusty Russelldde79782007-07-26 10:41:03 -07001887 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001888 if (initrd_name) {
1889 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06001890 /*
1891 * These are the location in the Linux boot header where the
1892 * start and size of the initrd are expected to be found.
1893 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001894 boot->hdr.ramdisk_image = mem - initrd_size;
1895 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001896 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001897 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001898 }
1899
Rusty Russell2e04ef72009-07-30 16:03:45 -06001900 /*
1901 * The Linux boot header contains an "E820" memory map: ours is a
1902 * simple, single region.
1903 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001904 boot->e820_entries = 1;
1905 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06001906 /*
1907 * The boot header contains a command line pointer: we put the command
1908 * line after the boot header.
1909 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001910 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10001911 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001912 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07001913
Rusty Russell814a0e52007-10-22 11:29:44 +10001914 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001915 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10001916
1917 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001918 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10001919
Rusty Russell43d33b22007-10-22 11:29:57 +10001920 /* Tell the entry path not to try to reload segment registers. */
1921 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001922
Rusty Russell2e04ef72009-07-30 16:03:45 -06001923 /*
1924 * We tell the kernel to initialize the Guest: this returns the open
1925 * /dev/lguest file descriptor.
1926 */
Rusty Russell56739c802009-06-12 22:26:59 -06001927 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07001928
Rusty Russell659a0e62009-06-12 22:27:10 -06001929 /* Ensure that we terminate if a child dies. */
1930 signal(SIGCHLD, kill_launcher);
1931
1932 /* If we exit via err(), this kills all the threads, restores tty. */
1933 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001934
Rusty Russelldde79782007-07-26 10:41:03 -07001935 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c802009-06-12 22:26:59 -06001936 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001937}
Rusty Russellf56a3842007-07-26 10:41:05 -07001938/*:*/
1939
1940/*M:999
1941 * Mastery is done: you now know everything I do.
1942 *
1943 * But surely you have seen code, features and bugs in your wanderings which
1944 * you now yearn to attack? That is the real game, and I look forward to you
1945 * patching and forking lguest into the Your-Name-Here-visor.
1946 *
1947 * Farewell, and good coding!
1948 * Rusty Russell.
1949 */