blob: 32cf2ce15d69bcfca9c24da9ad318fc1a2e84eb2 [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:100
2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5 * control it.
6:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07007#define _LARGEFILE64_SOURCE
8#define _GNU_SOURCE
9#include <stdio.h>
10#include <string.h>
11#include <unistd.h>
12#include <err.h>
13#include <stdint.h>
14#include <stdlib.h>
15#include <elf.h>
16#include <sys/mman.h>
Ronald G. Minnich6649bb72007-08-28 14:35:59 -070017#include <sys/param.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070018#include <sys/types.h>
19#include <sys/stat.h>
20#include <sys/wait.h>
Rusty Russell659a0e62009-06-12 22:27:10 -060021#include <sys/eventfd.h>
Rusty Russell8ca47e02007-07-19 01:49:29 -070022#include <fcntl.h>
23#include <stdbool.h>
24#include <errno.h>
25#include <ctype.h>
26#include <sys/socket.h>
27#include <sys/ioctl.h>
28#include <sys/time.h>
29#include <time.h>
30#include <netinet/in.h>
31#include <net/if.h>
32#include <linux/sockios.h>
33#include <linux/if_tun.h>
34#include <sys/uio.h>
35#include <termios.h>
36#include <getopt.h>
Rusty Russell17cbca22007-10-22 11:24:22 +100037#include <assert.h>
38#include <sched.h>
Rusty Russella586d4f2008-02-04 23:49:56 -050039#include <limits.h>
40#include <stddef.h>
Rusty Russella1618832008-07-29 09:58:35 -050041#include <signal.h>
Philip Sanderson8aeb36e2011-01-20 21:37:28 -060042#include <pwd.h>
43#include <grp.h>
44
Rusty Russell927cfb92013-07-15 10:50:13 +093045#ifndef VIRTIO_F_ANY_LAYOUT
46#define VIRTIO_F_ANY_LAYOUT 27
47#endif
48
Rusty Russell2e04ef72009-07-30 16:03:45 -060049/*L:110
Rusty Russell9f542882011-07-22 14:39:50 +093050 * We can ignore the 43 include files we need for this program, but I do want
Rusty Russell2e04ef72009-07-30 16:03:45 -060051 * to draw attention to the use of kernel-style types.
Rusty Russelldb24e8c2007-10-25 14:09:25 +100052 *
53 * As Linus said, "C is a Spartan language, and so should your naming be." I
54 * like these abbreviations, so we define them here. Note that u64 is always
55 * unsigned long long, which works on all Linux systems: this means that we can
Rusty Russell2e04ef72009-07-30 16:03:45 -060056 * use %llu in printf for any u64.
57 */
Rusty Russelldb24e8c2007-10-25 14:09:25 +100058typedef unsigned long long u64;
59typedef uint32_t u32;
60typedef uint16_t u16;
61typedef uint8_t u8;
Rusty Russelldde79782007-07-26 10:41:03 -070062/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -070063
Rusty Russelle6dc0412013-07-04 11:22:58 +093064#include <linux/virtio_config.h>
65#include <linux/virtio_net.h>
66#include <linux/virtio_blk.h>
67#include <linux/virtio_console.h>
68#include <linux/virtio_rng.h>
69#include <linux/virtio_ring.h>
70#include <asm/bootparam.h>
71#include "../../include/linux/lguest_launcher.h"
72
Rusty Russell8ca47e02007-07-19 01:49:29 -070073#define BRIDGE_PFX "bridge:"
74#ifndef SIOCBRADDIF
75#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
76#endif
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100077/* We can have up to 256 pages for devices. */
78#define DEVICE_PAGES 256
Rusty Russell0f0c4fa2008-07-29 09:58:37 -050079/* This will occupy 3 pages: it must be a power of 2. */
80#define VIRTQUEUE_NUM 256
Rusty Russell8ca47e02007-07-19 01:49:29 -070081
Rusty Russell2e04ef72009-07-30 16:03:45 -060082/*L:120
83 * verbose is both a global flag and a macro. The C preprocessor allows
84 * this, and although I wouldn't recommend it, it works quite nicely here.
85 */
Rusty Russell8ca47e02007-07-19 01:49:29 -070086static bool verbose;
87#define verbose(args...) \
88 do { if (verbose) printf(args); } while(0)
Rusty Russelldde79782007-07-26 10:41:03 -070089/*:*/
90
Rusty Russell3c6b5bf2007-10-22 11:03:26 +100091/* The pointer to the start of guest memory. */
92static void *guest_base;
93/* The maximum guest physical address allowed, and maximum possible. */
94static unsigned long guest_limit, guest_max;
Rusty Russell56739c802009-06-12 22:26:59 -060095/* The /dev/lguest file descriptor. */
96static int lguest_fd;
Rusty Russell8ca47e02007-07-19 01:49:29 -070097
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -020098/* a per-cpu variable indicating whose vcpu is currently running */
99static unsigned int __thread cpu_id;
100
Rusty Russelldde79782007-07-26 10:41:03 -0700101/* This is our list of devices. */
Rusty Russell1842f232009-07-30 16:03:46 -0600102struct device_list {
Rusty Russell17cbca22007-10-22 11:24:22 +1000103 /* Counter to assign interrupt numbers. */
104 unsigned int next_irq;
105
106 /* Counter to print out convenient device numbers. */
107 unsigned int device_num;
108
Rusty Russelldde79782007-07-26 10:41:03 -0700109 /* The descriptor page for the devices. */
Rusty Russell17cbca22007-10-22 11:24:22 +1000110 u8 *descpage;
111
Rusty Russelldde79782007-07-26 10:41:03 -0700112 /* A single linked list of devices. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700113 struct device *dev;
Rusty Russell2e04ef72009-07-30 16:03:45 -0600114 /* And a pointer to the last device for easy append. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500115 struct device *lastdev;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700116};
117
Rusty Russell17cbca22007-10-22 11:24:22 +1000118/* The list of Guest devices, based on command line arguments. */
119static struct device_list devices;
120
Rusty Russelldde79782007-07-26 10:41:03 -0700121/* The device structure describes a single device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600122struct device {
Rusty Russelldde79782007-07-26 10:41:03 -0700123 /* The linked-list pointer. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700124 struct device *next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000125
Rusty Russell713b15b2009-06-12 22:26:58 -0600126 /* The device's descriptor, as mapped into the Guest. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700127 struct lguest_device_desc *desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000128
Rusty Russell713b15b2009-06-12 22:26:58 -0600129 /* We can't trust desc values once Guest has booted: we use these. */
130 unsigned int feature_len;
131 unsigned int num_vq;
132
Rusty Russell17cbca22007-10-22 11:24:22 +1000133 /* The name of this device, for --verbose. */
134 const char *name;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700135
Rusty Russell17cbca22007-10-22 11:24:22 +1000136 /* Any queues attached to this device */
137 struct virtqueue *vq;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700138
Rusty Russell659a0e62009-06-12 22:27:10 -0600139 /* Is it operational */
140 bool running;
Rusty Russella007a752008-05-02 21:50:53 -0500141
Rusty Russell8ca47e02007-07-19 01:49:29 -0700142 /* Device-specific data. */
143 void *priv;
144};
145
Rusty Russell17cbca22007-10-22 11:24:22 +1000146/* The virtqueue structure describes a queue attached to a device. */
Rusty Russell1842f232009-07-30 16:03:46 -0600147struct virtqueue {
Rusty Russell17cbca22007-10-22 11:24:22 +1000148 struct virtqueue *next;
149
150 /* Which device owns me. */
151 struct device *dev;
152
153 /* The configuration for this queue. */
154 struct lguest_vqconfig config;
155
156 /* The actual ring of buffers. */
157 struct vring vring;
158
159 /* Last available index we saw. */
160 u16 last_avail_idx;
161
Rusty Russell95c517c2009-06-12 22:27:11 -0600162 /* How many are used since we sent last irq? */
163 unsigned int pending_used;
164
Rusty Russell659a0e62009-06-12 22:27:10 -0600165 /* Eventfd where Guest notifications arrive. */
166 int eventfd;
Rusty Russell20887612008-05-30 15:09:46 -0500167
Rusty Russell659a0e62009-06-12 22:27:10 -0600168 /* Function for the thread which is servicing this virtqueue. */
169 void (*service)(struct virtqueue *vq);
170 pid_t thread;
Rusty Russell17cbca22007-10-22 11:24:22 +1000171};
172
Balaji Raoec04b132007-12-28 14:26:24 +0530173/* Remember the arguments to the program so we can "reboot" */
174static char **main_args;
175
Rusty Russell659a0e62009-06-12 22:27:10 -0600176/* The original tty settings to restore on exit. */
177static struct termios orig_term;
178
Rusty Russell2e04ef72009-07-30 16:03:45 -0600179/*
180 * We have to be careful with barriers: our devices are all run in separate
Rusty Russellf7027c62009-06-12 22:27:00 -0600181 * threads and so we need to make sure that changes visible to the Guest happen
Rusty Russell2e04ef72009-07-30 16:03:45 -0600182 * in precise order.
183 */
Rusty Russellf7027c62009-06-12 22:27:00 -0600184#define wmb() __asm__ __volatile__("" : : : "memory")
Rusty Russell0d69a652013-07-02 15:35:14 +0930185#define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
186#define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
Rusty Russell17cbca22007-10-22 11:24:22 +1000187
Rusty Russellb5111792008-07-29 09:58:34 -0500188/* Wrapper for the last available index. Makes it easier to change. */
189#define lg_last_avail(vq) ((vq)->last_avail_idx)
190
Rusty Russell2e04ef72009-07-30 16:03:45 -0600191/*
192 * The virtio configuration space is defined to be little-endian. x86 is
193 * little-endian too, but it's nice to be explicit so we have these helpers.
194 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000195#define cpu_to_le16(v16) (v16)
196#define cpu_to_le32(v32) (v32)
197#define cpu_to_le64(v64) (v64)
198#define le16_to_cpu(v16) (v16)
199#define le32_to_cpu(v32) (v32)
Rusty Russella586d4f2008-02-04 23:49:56 -0500200#define le64_to_cpu(v64) (v64)
Rusty Russell17cbca22007-10-22 11:24:22 +1000201
Rusty Russell28fd6d72008-07-29 09:58:33 -0500202/* Is this iovec empty? */
203static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
204{
205 unsigned int i;
206
207 for (i = 0; i < num_iov; i++)
208 if (iov[i].iov_len)
209 return false;
210 return true;
211}
212
213/* Take len bytes from the front of this iovec. */
Rusty Russellc0316a92012-10-16 23:56:13 +1030214static void iov_consume(struct iovec iov[], unsigned num_iov,
215 void *dest, unsigned len)
Rusty Russell28fd6d72008-07-29 09:58:33 -0500216{
217 unsigned int i;
218
219 for (i = 0; i < num_iov; i++) {
220 unsigned int used;
221
222 used = iov[i].iov_len < len ? iov[i].iov_len : len;
Rusty Russellc0316a92012-10-16 23:56:13 +1030223 if (dest) {
224 memcpy(dest, iov[i].iov_base, used);
225 dest += used;
226 }
Rusty Russell28fd6d72008-07-29 09:58:33 -0500227 iov[i].iov_base += used;
228 iov[i].iov_len -= used;
229 len -= used;
230 }
Rusty Russellc0316a92012-10-16 23:56:13 +1030231 if (len != 0)
232 errx(1, "iovec too short!");
Rusty Russell28fd6d72008-07-29 09:58:33 -0500233}
234
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500235/* The device virtqueue descriptors are followed by feature bitmasks. */
236static u8 *get_feature_bits(struct device *dev)
237{
238 return (u8 *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -0600239 + dev->num_vq * sizeof(struct lguest_vqconfig);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500240}
241
Rusty Russell2e04ef72009-07-30 16:03:45 -0600242/*L:100
243 * The Launcher code itself takes us out into userspace, that scary place where
244 * pointers run wild and free! Unfortunately, like most userspace programs,
245 * it's quite boring (which is why everyone likes to hack on the kernel!).
246 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
247 * you through this section. Or, maybe not.
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000248 *
249 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
250 * memory and stores it in "guest_base". In other words, Guest physical ==
251 * Launcher virtual with an offset.
252 *
253 * This can be tough to get your head around, but usually it just means that we
Francis Galieguea33f3222010-04-23 00:08:02 +0200254 * use these trivial conversion functions when the Guest gives us its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600255 * "physical" addresses:
256 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000257static void *from_guest_phys(unsigned long addr)
258{
259 return guest_base + addr;
260}
261
262static unsigned long to_guest_phys(const void *addr)
263{
264 return (addr - guest_base);
265}
266
Rusty Russelldde79782007-07-26 10:41:03 -0700267/*L:130
268 * Loading the Kernel.
269 *
270 * We start with couple of simple helper routines. open_or_die() avoids
Rusty Russell2e04ef72009-07-30 16:03:45 -0600271 * error-checking code cluttering the callers:
272 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700273static int open_or_die(const char *name, int flags)
274{
275 int fd = open(name, flags);
276 if (fd < 0)
277 err(1, "Failed to open %s", name);
278 return fd;
279}
280
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000281/* map_zeroed_pages() takes a number of pages. */
282static void *map_zeroed_pages(unsigned int num)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700283{
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000284 int fd = open_or_die("/dev/zero", O_RDONLY);
285 void *addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700286
Rusty Russell2e04ef72009-07-30 16:03:45 -0600287 /*
288 * We use a private mapping (ie. if we write to the page, it will be
Philip Sanderson5230ff02011-01-20 21:37:28 -0600289 * copied). We allocate an extra two pages PROT_NONE to act as guard
290 * pages against read/write attempts that exceed allocated space.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600291 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600292 addr = mmap(NULL, getpagesize() * (num+2),
293 PROT_NONE, MAP_PRIVATE, fd, 0);
294
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000295 if (addr == MAP_FAILED)
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200296 err(1, "Mmapping %u pages of /dev/zero", num);
Rusty Russella91d74a2009-07-30 16:03:45 -0600297
Philip Sanderson5230ff02011-01-20 21:37:28 -0600298 if (mprotect(addr + getpagesize(), getpagesize() * num,
299 PROT_READ|PROT_WRITE) == -1)
300 err(1, "mprotect rw %u pages failed", num);
301
Rusty Russella91d74a2009-07-30 16:03:45 -0600302 /*
303 * One neat mmap feature is that you can close the fd, and it
304 * stays mapped.
305 */
Mark McLoughlin34bdaab2008-06-13 14:04:58 +0100306 close(fd);
Rusty Russelldde79782007-07-26 10:41:03 -0700307
Philip Sanderson5230ff02011-01-20 21:37:28 -0600308 /* Return address after PROT_NONE page */
309 return addr + getpagesize();
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000310}
311
312/* Get some more pages for a device. */
313static void *get_pages(unsigned int num)
314{
315 void *addr = from_guest_phys(guest_limit);
316
317 guest_limit += num * getpagesize();
318 if (guest_limit > guest_max)
319 errx(1, "Not enough memory for devices");
320 return addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700321}
322
Rusty Russell2e04ef72009-07-30 16:03:45 -0600323/*
324 * This routine is used to load the kernel or initrd. It tries mmap, but if
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700325 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
Rusty Russell2e04ef72009-07-30 16:03:45 -0600326 * it falls back to reading the memory in.
327 */
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700328static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
329{
330 ssize_t r;
331
Rusty Russell2e04ef72009-07-30 16:03:45 -0600332 /*
333 * We map writable even though for some segments are marked read-only.
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700334 * The kernel really wants to be writable: it patches its own
335 * instructions.
336 *
337 * MAP_PRIVATE means that the page won't be copied until a write is
338 * done to it. This allows us to share untouched memory between
Rusty Russell2e04ef72009-07-30 16:03:45 -0600339 * Guests.
340 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600341 if (mmap(addr, len, PROT_READ|PROT_WRITE,
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700342 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
343 return;
344
345 /* pread does a seek and a read in one shot: saves a few lines. */
346 r = pread(fd, addr, len, offset);
347 if (r != len)
348 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
349}
350
Rusty Russell2e04ef72009-07-30 16:03:45 -0600351/*
352 * This routine takes an open vmlinux image, which is in ELF, and maps it into
Rusty Russelldde79782007-07-26 10:41:03 -0700353 * the Guest memory. ELF = Embedded Linking Format, which is the format used
354 * by all modern binaries on Linux including the kernel.
355 *
356 * The ELF headers give *two* addresses: a physical address, and a virtual
Rusty Russell47436aa2007-10-22 11:03:36 +1000357 * address. We use the physical address; the Guest will map itself to the
358 * virtual address.
Rusty Russelldde79782007-07-26 10:41:03 -0700359 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600360 * We return the starting address.
361 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000362static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700363{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700364 Elf32_Phdr phdr[ehdr->e_phnum];
365 unsigned int i;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700366
Rusty Russell2e04ef72009-07-30 16:03:45 -0600367 /*
368 * Sanity checks on the main ELF header: an x86 executable with a
369 * reasonable number of correctly-sized program headers.
370 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700371 if (ehdr->e_type != ET_EXEC
372 || ehdr->e_machine != EM_386
373 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
374 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
375 errx(1, "Malformed elf header");
376
Rusty Russell2e04ef72009-07-30 16:03:45 -0600377 /*
378 * An ELF executable contains an ELF header and a number of "program"
Rusty Russelldde79782007-07-26 10:41:03 -0700379 * headers which indicate which parts ("segments") of the program to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600380 * load where.
381 */
Rusty Russelldde79782007-07-26 10:41:03 -0700382
383 /* We read in all the program headers at once: */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700384 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
385 err(1, "Seeking to program headers");
386 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
387 err(1, "Reading program headers");
388
Rusty Russell2e04ef72009-07-30 16:03:45 -0600389 /*
390 * Try all the headers: there are usually only three. A read-only one,
391 * a read-write one, and a "note" section which we don't load.
392 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700393 for (i = 0; i < ehdr->e_phnum; i++) {
Rusty Russelldde79782007-07-26 10:41:03 -0700394 /* If this isn't a loadable segment, we ignore it */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700395 if (phdr[i].p_type != PT_LOAD)
396 continue;
397
398 verbose("Section %i: size %i addr %p\n",
399 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
400
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700401 /* We map this section of the file at its physical address. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000402 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700403 phdr[i].p_offset, phdr[i].p_filesz);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700404 }
405
Rusty Russell814a0e52007-10-22 11:29:44 +1000406 /* The entry point is given in the ELF header. */
407 return ehdr->e_entry;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700408}
409
Rusty Russell2e04ef72009-07-30 16:03:45 -0600410/*L:150
411 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
412 * to jump into it and it will unpack itself. We used to have to perform some
413 * hairy magic because the unpacking code scared me.
Rusty Russelldde79782007-07-26 10:41:03 -0700414 *
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000415 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
416 * a small patch to jump over the tricky bits in the Guest, so now we just read
Rusty Russell2e04ef72009-07-30 16:03:45 -0600417 * the funky header so we know where in the file to load, and away we go!
418 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000419static unsigned long load_bzimage(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700420{
Rusty Russell43d33b22007-10-22 11:29:57 +1000421 struct boot_params boot;
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000422 int r;
423 /* Modern bzImages get loaded at 1M. */
424 void *p = from_guest_phys(0x100000);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700425
Rusty Russell2e04ef72009-07-30 16:03:45 -0600426 /*
427 * Go back to the start of the file and read the header. It should be
Paul Bolle395cf962011-08-15 02:02:26 +0200428 * a Linux boot header (see Documentation/x86/boot.txt)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600429 */
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000430 lseek(fd, 0, SEEK_SET);
Rusty Russell43d33b22007-10-22 11:29:57 +1000431 read(fd, &boot, sizeof(boot));
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000432
Rusty Russell43d33b22007-10-22 11:29:57 +1000433 /* Inside the setup_hdr, we expect the magic "HdrS" */
434 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000435 errx(1, "This doesn't look like a bzImage to me");
436
Rusty Russell43d33b22007-10-22 11:29:57 +1000437 /* Skip over the extra sectors of the header. */
438 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
Rusty Russell5bbf89f2007-10-22 11:29:56 +1000439
440 /* Now read everything into memory. in nice big chunks. */
441 while ((r = read(fd, p, 65536)) > 0)
442 p += r;
443
Rusty Russell43d33b22007-10-22 11:29:57 +1000444 /* Finally, code32_start tells us where to enter the kernel. */
445 return boot.hdr.code32_start;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700446}
447
Rusty Russell2e04ef72009-07-30 16:03:45 -0600448/*L:140
449 * Loading the kernel is easy when it's a "vmlinux", but most kernels
Rusty Russelle1e72962007-10-25 15:02:50 +1000450 * come wrapped up in the self-decompressing "bzImage" format. With a little
Rusty Russell2e04ef72009-07-30 16:03:45 -0600451 * work, we can load those, too.
452 */
Rusty Russell47436aa2007-10-22 11:03:36 +1000453static unsigned long load_kernel(int fd)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700454{
455 Elf32_Ehdr hdr;
456
Rusty Russelldde79782007-07-26 10:41:03 -0700457 /* Read in the first few bytes. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700458 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
459 err(1, "Reading kernel");
460
Rusty Russelldde79782007-07-26 10:41:03 -0700461 /* If it's an ELF file, it starts with "\177ELF" */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700462 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
Rusty Russell47436aa2007-10-22 11:03:36 +1000463 return map_elf(fd, &hdr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700464
Rusty Russella6bd8e12008-03-28 11:05:53 -0500465 /* Otherwise we assume it's a bzImage, and try to load it. */
Rusty Russell47436aa2007-10-22 11:03:36 +1000466 return load_bzimage(fd);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700467}
468
Rusty Russell2e04ef72009-07-30 16:03:45 -0600469/*
470 * This is a trivial little helper to align pages. Andi Kleen hated it because
Rusty Russelldde79782007-07-26 10:41:03 -0700471 * it calls getpagesize() twice: "it's dumb code."
472 *
473 * Kernel guys get really het up about optimization, even when it's not
Rusty Russell2e04ef72009-07-30 16:03:45 -0600474 * necessary. I leave this code as a reaction against that.
475 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700476static inline unsigned long page_align(unsigned long addr)
477{
Rusty Russelldde79782007-07-26 10:41:03 -0700478 /* Add upwards and truncate downwards. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700479 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
480}
481
Rusty Russell2e04ef72009-07-30 16:03:45 -0600482/*L:180
483 * An "initial ram disk" is a disk image loaded into memory along with the
484 * kernel which the kernel can use to boot from without needing any drivers.
485 * Most distributions now use this as standard: the initrd contains the code to
486 * load the appropriate driver modules for the current machine.
Rusty Russelldde79782007-07-26 10:41:03 -0700487 *
488 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
Rusty Russell2e04ef72009-07-30 16:03:45 -0600489 * kernels. He sent me this (and tells me when I break it).
490 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700491static unsigned long load_initrd(const char *name, unsigned long mem)
492{
493 int ifd;
494 struct stat st;
495 unsigned long len;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700496
497 ifd = open_or_die(name, O_RDONLY);
Rusty Russelldde79782007-07-26 10:41:03 -0700498 /* fstat() is needed to get the file size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700499 if (fstat(ifd, &st) < 0)
500 err(1, "fstat() on initrd '%s'", name);
501
Rusty Russell2e04ef72009-07-30 16:03:45 -0600502 /*
503 * We map the initrd at the top of memory, but mmap wants it to be
504 * page-aligned, so we round the size up for that.
505 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700506 len = page_align(st.st_size);
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000507 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600508 /*
509 * Once a file is mapped, you can close the file descriptor. It's a
510 * little odd, but quite useful.
511 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700512 close(ifd);
Ronald G. Minnich6649bb72007-08-28 14:35:59 -0700513 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
Rusty Russelldde79782007-07-26 10:41:03 -0700514
515 /* We return the initrd size. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700516 return len;
517}
Rusty Russelle1e72962007-10-25 15:02:50 +1000518/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700519
Rusty Russell2e04ef72009-07-30 16:03:45 -0600520/*
521 * Simple routine to roll all the commandline arguments together with spaces
522 * between them.
523 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700524static void concat(char *dst, char *args[])
525{
526 unsigned int i, len = 0;
527
528 for (i = 0; args[i]; i++) {
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100529 if (i) {
530 strcat(dst+len, " ");
531 len++;
532 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700533 strcpy(dst+len, args[i]);
Paul Bolle1ef36fa2008-03-10 16:39:03 +0100534 len += strlen(args[i]);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700535 }
536 /* In case it's empty. */
537 dst[len] = '\0';
538}
539
Rusty Russell2e04ef72009-07-30 16:03:45 -0600540/*L:185
541 * This is where we actually tell the kernel to initialize the Guest. We
Rusty Russelle1e72962007-10-25 15:02:50 +1000542 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300543 * the base of Guest "physical" memory, the top physical page to allow and the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600544 * entry point for the Guest.
545 */
Rusty Russell56739c802009-06-12 22:26:59 -0600546static void tell_kernel(unsigned long start)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700547{
Jes Sorensen511801d2007-10-22 11:03:31 +1000548 unsigned long args[] = { LHREQ_INITIALIZE,
549 (unsigned long)guest_base,
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300550 guest_limit / getpagesize(), start };
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000551 verbose("Guest: %p - %p (%#lx)\n",
552 guest_base, guest_base + guest_limit, guest_limit);
Rusty Russell56739c802009-06-12 22:26:59 -0600553 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
554 if (write(lguest_fd, args, sizeof(args)) < 0)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700555 err(1, "Writing to /dev/lguest");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700556}
Rusty Russelldde79782007-07-26 10:41:03 -0700557/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700558
Rusty Russella91d74a2009-07-30 16:03:45 -0600559/*L:200
Rusty Russelldde79782007-07-26 10:41:03 -0700560 * Device Handling.
561 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000562 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
Rusty Russelldde79782007-07-26 10:41:03 -0700563 * We need to make sure it's not trying to reach into the Launcher itself, so
Rusty Russelle1e72962007-10-25 15:02:50 +1000564 * we have a convenient routine which checks it and exits with an error message
Rusty Russelldde79782007-07-26 10:41:03 -0700565 * if something funny is going on:
566 */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700567static void *_check_pointer(unsigned long addr, unsigned int size,
568 unsigned int line)
569{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600570 /*
Philip Sanderson5230ff02011-01-20 21:37:28 -0600571 * Check if the requested address and size exceeds the allocated memory,
572 * or addr + size wraps around.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600573 */
Philip Sanderson5230ff02011-01-20 21:37:28 -0600574 if ((addr + size) > guest_limit || (addr + size) < addr)
Rusty Russell17cbca22007-10-22 11:24:22 +1000575 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
Rusty Russell2e04ef72009-07-30 16:03:45 -0600576 /*
577 * We return a pointer for the caller's convenience, now we know it's
578 * safe to use.
579 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000580 return from_guest_phys(addr);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700581}
Rusty Russelldde79782007-07-26 10:41:03 -0700582/* A macro which transparently hands the line number to the real function. */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700583#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
584
Rusty Russell2e04ef72009-07-30 16:03:45 -0600585/*
586 * Each buffer in the virtqueues is actually a chain of descriptors. This
Rusty Russelle1e72962007-10-25 15:02:50 +1000587 * function returns the next descriptor in the chain, or vq->vring.num if we're
Rusty Russell2e04ef72009-07-30 16:03:45 -0600588 * at the end.
589 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100590static unsigned next_desc(struct vring_desc *desc,
591 unsigned int i, unsigned int max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000592{
593 unsigned int next;
594
595 /* If this descriptor says it doesn't chain, we're done. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100596 if (!(desc[i].flags & VRING_DESC_F_NEXT))
597 return max;
Rusty Russell17cbca22007-10-22 11:24:22 +1000598
599 /* Check they're not leading us off end of descriptors. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100600 next = desc[i].next;
Rusty Russell17cbca22007-10-22 11:24:22 +1000601 /* Make sure compiler knows to grab that: we don't want it changing! */
602 wmb();
603
Mark McLoughlind1f01322009-05-11 18:11:46 +0100604 if (next >= max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000605 errx(1, "Desc next is %u", next);
606
607 return next;
608}
609
Rusty Russella91d74a2009-07-30 16:03:45 -0600610/*
611 * This actually sends the interrupt for this virtqueue, if we've used a
612 * buffer.
613 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600614static void trigger_irq(struct virtqueue *vq)
615{
616 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
617
Rusty Russell95c517c2009-06-12 22:27:11 -0600618 /* Don't inform them if nothing used. */
619 if (!vq->pending_used)
620 return;
621 vq->pending_used = 0;
622
Rusty Russellca60a422009-09-23 22:26:47 -0600623 /* If they don't want an interrupt, don't send one... */
624 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
Rusty Russell990c91f2011-05-30 11:14:12 -0600625 return;
Rusty Russellca60a422009-09-23 22:26:47 -0600626 }
Rusty Russell38bc2b82009-06-12 22:27:11 -0600627
628 /* Send the Guest an interrupt tell them we used something up. */
629 if (write(lguest_fd, buf, sizeof(buf)) != 0)
630 err(1, "Triggering irq %i", vq->config.irq);
631}
632
Rusty Russell2e04ef72009-07-30 16:03:45 -0600633/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600634 * This looks in the virtqueue for the first available buffer, and converts
Rusty Russell17cbca22007-10-22 11:24:22 +1000635 * it to an iovec for convenient access. Since descriptors consist of some
636 * number of output then some number of input descriptors, it's actually two
637 * iovecs, but we pack them into one and note how many of each there were.
638 *
Rusty Russella91d74a2009-07-30 16:03:45 -0600639 * This function waits if necessary, and returns the descriptor number found.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600640 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600641static unsigned wait_for_vq_desc(struct virtqueue *vq,
642 struct iovec iov[],
643 unsigned int *out_num, unsigned int *in_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000644{
Mark McLoughlind1f01322009-05-11 18:11:46 +0100645 unsigned int i, head, max;
646 struct vring_desc *desc;
Rusty Russell659a0e62009-06-12 22:27:10 -0600647 u16 last_avail = lg_last_avail(vq);
648
Rusty Russella91d74a2009-07-30 16:03:45 -0600649 /* There's nothing available? */
Rusty Russell659a0e62009-06-12 22:27:10 -0600650 while (last_avail == vq->vring.avail->idx) {
651 u64 event;
652
Rusty Russella91d74a2009-07-30 16:03:45 -0600653 /*
654 * Since we're about to sleep, now is a good time to tell the
655 * Guest about what we've used up to now.
656 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600657 trigger_irq(vq);
658
Rusty Russellb60da132009-06-12 22:27:12 -0600659 /* OK, now we need to know about added descriptors. */
660 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
661
Rusty Russell2e04ef72009-07-30 16:03:45 -0600662 /*
663 * They could have slipped one in as we were doing that: make
664 * sure it's written, then check again.
665 */
Rusty Russellb60da132009-06-12 22:27:12 -0600666 mb();
667 if (last_avail != vq->vring.avail->idx) {
668 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
669 break;
670 }
671
Rusty Russell659a0e62009-06-12 22:27:10 -0600672 /* Nothing new? Wait for eventfd to tell us they refilled. */
673 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
674 errx(1, "Event read failed?");
Rusty Russellb60da132009-06-12 22:27:12 -0600675
676 /* We don't need to be notified again. */
677 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
Rusty Russell659a0e62009-06-12 22:27:10 -0600678 }
Rusty Russell17cbca22007-10-22 11:24:22 +1000679
680 /* Check it isn't doing very strange things with descriptor numbers. */
Rusty Russellb5111792008-07-29 09:58:34 -0500681 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000682 errx(1, "Guest moved used index from %u to %u",
Rusty Russellb5111792008-07-29 09:58:34 -0500683 last_avail, vq->vring.avail->idx);
Rusty Russell17cbca22007-10-22 11:24:22 +1000684
Rusty Russell8fd9a632013-07-02 15:35:13 +0930685 /*
686 * Make sure we read the descriptor number *after* we read the ring
687 * update; don't let the cpu or compiler change the order.
688 */
689 rmb();
690
Rusty Russell2e04ef72009-07-30 16:03:45 -0600691 /*
692 * Grab the next descriptor number they're advertising, and increment
693 * the index we've seen.
694 */
Rusty Russellb5111792008-07-29 09:58:34 -0500695 head = vq->vring.avail->ring[last_avail % vq->vring.num];
696 lg_last_avail(vq)++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000697
698 /* If their number is silly, that's a fatal mistake. */
699 if (head >= vq->vring.num)
700 errx(1, "Guest says index %u is available", head);
701
702 /* When we start there are none of either input nor output. */
703 *out_num = *in_num = 0;
704
Mark McLoughlind1f01322009-05-11 18:11:46 +0100705 max = vq->vring.num;
706 desc = vq->vring.desc;
Rusty Russell17cbca22007-10-22 11:24:22 +1000707 i = head;
Mark McLoughlind1f01322009-05-11 18:11:46 +0100708
Rusty Russell2e04ef72009-07-30 16:03:45 -0600709 /*
Rusty Russell8fd9a632013-07-02 15:35:13 +0930710 * We have to read the descriptor after we read the descriptor number,
711 * but there's a data dependency there so the CPU shouldn't reorder
712 * that: no rmb() required.
713 */
714
715 /*
Rusty Russell2e04ef72009-07-30 16:03:45 -0600716 * If this is an indirect entry, then this buffer contains a descriptor
717 * table which we handle as if it's any normal descriptor chain.
718 */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100719 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
720 if (desc[i].len % sizeof(struct vring_desc))
721 errx(1, "Invalid size for indirect buffer table");
722
723 max = desc[i].len / sizeof(struct vring_desc);
724 desc = check_pointer(desc[i].addr, desc[i].len);
725 i = 0;
726 }
727
Rusty Russell17cbca22007-10-22 11:24:22 +1000728 do {
729 /* Grab the first descriptor, and check it's OK. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100730 iov[*out_num + *in_num].iov_len = desc[i].len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000731 iov[*out_num + *in_num].iov_base
Mark McLoughlind1f01322009-05-11 18:11:46 +0100732 = check_pointer(desc[i].addr, desc[i].len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000733 /* If this is an input descriptor, increment that count. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100734 if (desc[i].flags & VRING_DESC_F_WRITE)
Rusty Russell17cbca22007-10-22 11:24:22 +1000735 (*in_num)++;
736 else {
Rusty Russell2e04ef72009-07-30 16:03:45 -0600737 /*
738 * If it's an output descriptor, they're all supposed
739 * to come before any input descriptors.
740 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000741 if (*in_num)
742 errx(1, "Descriptor has out after in");
743 (*out_num)++;
744 }
745
746 /* If we've got too many, that implies a descriptor loop. */
Mark McLoughlind1f01322009-05-11 18:11:46 +0100747 if (*out_num + *in_num > max)
Rusty Russell17cbca22007-10-22 11:24:22 +1000748 errx(1, "Looped descriptor");
Mark McLoughlind1f01322009-05-11 18:11:46 +0100749 } while ((i = next_desc(desc, i, max)) != max);
Rusty Russell17cbca22007-10-22 11:24:22 +1000750
751 return head;
752}
753
Rusty Russell2e04ef72009-07-30 16:03:45 -0600754/*
Rusty Russella91d74a2009-07-30 16:03:45 -0600755 * After we've used one of their buffers, we tell the Guest about it. Sometime
756 * later we'll want to send them an interrupt using trigger_irq(); note that
757 * wait_for_vq_desc() does that for us if it has to wait.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600758 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000759static void add_used(struct virtqueue *vq, unsigned int head, int len)
760{
761 struct vring_used_elem *used;
762
Rusty Russell2e04ef72009-07-30 16:03:45 -0600763 /*
764 * The virtqueue contains a ring of used buffers. Get a pointer to the
765 * next entry in that used ring.
766 */
Rusty Russell17cbca22007-10-22 11:24:22 +1000767 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
768 used->id = head;
769 used->len = len;
770 /* Make sure buffer is written before we update index. */
771 wmb();
772 vq->vring.used->idx++;
Rusty Russell95c517c2009-06-12 22:27:11 -0600773 vq->pending_used++;
Rusty Russell17cbca22007-10-22 11:24:22 +1000774}
775
Rusty Russell17cbca22007-10-22 11:24:22 +1000776/* And here's the combo meal deal. Supersize me! */
Rusty Russell56739c802009-06-12 22:26:59 -0600777static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
Rusty Russell17cbca22007-10-22 11:24:22 +1000778{
779 add_used(vq, head, len);
Rusty Russell56739c802009-06-12 22:26:59 -0600780 trigger_irq(vq);
Rusty Russell17cbca22007-10-22 11:24:22 +1000781}
782
Rusty Russelle1e72962007-10-25 15:02:50 +1000783/*
784 * The Console
785 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600786 * We associate some data with the console for our exit hack.
787 */
Rusty Russell1842f232009-07-30 16:03:46 -0600788struct console_abort {
Rusty Russelldde79782007-07-26 10:41:03 -0700789 /* How many times have they hit ^C? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700790 int count;
Rusty Russelldde79782007-07-26 10:41:03 -0700791 /* When did they start? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700792 struct timeval start;
793};
794
Rusty Russelldde79782007-07-26 10:41:03 -0700795/* This is the routine which handles console input (ie. stdin). */
Rusty Russell659a0e62009-06-12 22:27:10 -0600796static void console_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700797{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700798 int len;
Rusty Russell17cbca22007-10-22 11:24:22 +1000799 unsigned int head, in_num, out_num;
Rusty Russell659a0e62009-06-12 22:27:10 -0600800 struct console_abort *abort = vq->dev->priv;
801 struct iovec iov[vq->vring.num];
Rusty Russell8ca47e02007-07-19 01:49:29 -0700802
Rusty Russella91d74a2009-07-30 16:03:45 -0600803 /* Make sure there's a descriptor available. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600804 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell56ae43d2007-10-22 11:24:23 +1000805 if (out_num)
Rusty Russell17cbca22007-10-22 11:24:22 +1000806 errx(1, "Output buffers in console in queue?");
Rusty Russell8ca47e02007-07-19 01:49:29 -0700807
Rusty Russella91d74a2009-07-30 16:03:45 -0600808 /* Read into it. This is where we usually wait. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600809 len = readv(STDIN_FILENO, iov, in_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700810 if (len <= 0) {
Rusty Russell659a0e62009-06-12 22:27:10 -0600811 /* Ran out of input? */
Rusty Russell8ca47e02007-07-19 01:49:29 -0700812 warnx("Failed to get console input, ignoring console.");
Rusty Russell2e04ef72009-07-30 16:03:45 -0600813 /*
814 * For simplicity, dying threads kill the whole Launcher. So
815 * just nap here.
816 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600817 for (;;)
818 pause();
Rusty Russell8ca47e02007-07-19 01:49:29 -0700819 }
820
Rusty Russella91d74a2009-07-30 16:03:45 -0600821 /* Tell the Guest we used a buffer. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600822 add_used_and_trigger(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700823
Rusty Russell2e04ef72009-07-30 16:03:45 -0600824 /*
825 * Three ^C within one second? Exit.
Rusty Russelldde79782007-07-26 10:41:03 -0700826 *
Rusty Russell659a0e62009-06-12 22:27:10 -0600827 * This is such a hack, but works surprisingly well. Each ^C has to
828 * be in a buffer by itself, so they can't be too fast. But we check
829 * that we get three within about a second, so they can't be too
Rusty Russell2e04ef72009-07-30 16:03:45 -0600830 * slow.
831 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600832 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
Rusty Russell8ca47e02007-07-19 01:49:29 -0700833 abort->count = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -0600834 return;
835 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700836
Rusty Russell659a0e62009-06-12 22:27:10 -0600837 abort->count++;
838 if (abort->count == 1)
839 gettimeofday(&abort->start, NULL);
840 else if (abort->count == 3) {
841 struct timeval now;
842 gettimeofday(&now, NULL);
843 /* Kill all Launcher processes with SIGINT, like normal ^C */
844 if (now.tv_sec <= abort->start.tv_sec+1)
845 kill(0, SIGINT);
846 abort->count = 0;
847 }
Rusty Russell8ca47e02007-07-19 01:49:29 -0700848}
849
Rusty Russell659a0e62009-06-12 22:27:10 -0600850/* This is the routine which handles console output (ie. stdout). */
851static void console_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700852{
Rusty Russell17cbca22007-10-22 11:24:22 +1000853 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000854 struct iovec iov[vq->vring.num];
855
Rusty Russella91d74a2009-07-30 16:03:45 -0600856 /* We usually wait in here, for the Guest to give us something. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600857 head = wait_for_vq_desc(vq, iov, &out, &in);
858 if (in)
859 errx(1, "Input buffers in console output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600860
861 /* writev can return a partial write, so we loop here. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600862 while (!iov_empty(iov, out)) {
863 int len = writev(STDOUT_FILENO, iov, out);
Sakari Ailuse0377e22011-06-26 19:36:46 +0300864 if (len <= 0) {
865 warn("Write to stdout gave %i (%d)", len, errno);
866 break;
867 }
Rusty Russellc0316a92012-10-16 23:56:13 +1030868 iov_consume(iov, out, NULL, len);
Rusty Russell17cbca22007-10-22 11:24:22 +1000869 }
Rusty Russella91d74a2009-07-30 16:03:45 -0600870
871 /*
872 * We're finished with that buffer: if we're going to sleep,
873 * wait_for_vq_desc() will prod the Guest with an interrupt.
874 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600875 add_used(vq, head, 0);
Rusty Russella1618832008-07-29 09:58:35 -0500876}
877
Rusty Russelle1e72962007-10-25 15:02:50 +1000878/*
879 * The Network
880 *
881 * Handling output for network is also simple: we get all the output buffers
Rusty Russell659a0e62009-06-12 22:27:10 -0600882 * and write them to /dev/net/tun.
Rusty Russella6bd8e12008-03-28 11:05:53 -0500883 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600884struct net_info {
885 int tunfd;
886};
887
888static void net_output(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700889{
Rusty Russell659a0e62009-06-12 22:27:10 -0600890 struct net_info *net_info = vq->dev->priv;
891 unsigned int head, out, in;
Rusty Russell17cbca22007-10-22 11:24:22 +1000892 struct iovec iov[vq->vring.num];
893
Rusty Russella91d74a2009-07-30 16:03:45 -0600894 /* We usually wait in here for the Guest to give us a packet. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600895 head = wait_for_vq_desc(vq, iov, &out, &in);
896 if (in)
897 errx(1, "Input buffers in net output queue?");
Rusty Russella91d74a2009-07-30 16:03:45 -0600898 /*
899 * Send the whole thing through to /dev/net/tun. It expects the exact
900 * same format: what a coincidence!
901 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600902 if (writev(net_info->tunfd, iov, out) < 0)
Sakari Ailuse0377e22011-06-26 19:36:46 +0300903 warnx("Write to tun failed (%d)?", errno);
Rusty Russella91d74a2009-07-30 16:03:45 -0600904
905 /*
906 * Done with that one; wait_for_vq_desc() will send the interrupt if
907 * all packets are processed.
908 */
Rusty Russell38bc2b82009-06-12 22:27:11 -0600909 add_used(vq, head, 0);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700910}
911
Rusty Russella91d74a2009-07-30 16:03:45 -0600912/*
913 * Handling network input is a bit trickier, because I've tried to optimize it.
914 *
915 * First we have a helper routine which tells is if from this file descriptor
916 * (ie. the /dev/net/tun device) will block:
917 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600918static bool will_block(int fd)
919{
920 fd_set fdset;
921 struct timeval zero = { 0, 0 };
922 FD_ZERO(&fdset);
923 FD_SET(fd, &fdset);
924 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
925}
926
Rusty Russella91d74a2009-07-30 16:03:45 -0600927/*
928 * This handles packets coming in from the tun device to our Guest. Like all
929 * service routines, it gets called again as soon as it returns, so you don't
930 * see a while(1) loop here.
931 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600932static void net_input(struct virtqueue *vq)
Rusty Russell8ca47e02007-07-19 01:49:29 -0700933{
Rusty Russell8ca47e02007-07-19 01:49:29 -0700934 int len;
Rusty Russell659a0e62009-06-12 22:27:10 -0600935 unsigned int head, out, in;
936 struct iovec iov[vq->vring.num];
937 struct net_info *net_info = vq->dev->priv;
Rusty Russell8ca47e02007-07-19 01:49:29 -0700938
Rusty Russella91d74a2009-07-30 16:03:45 -0600939 /*
940 * Get a descriptor to write an incoming packet into. This will also
941 * send an interrupt if they're out of descriptors.
942 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600943 head = wait_for_vq_desc(vq, iov, &out, &in);
944 if (out)
945 errx(1, "Output buffers in net input queue?");
Rusty Russell4a8962e2009-06-12 22:27:12 -0600946
Rusty Russella91d74a2009-07-30 16:03:45 -0600947 /*
948 * If it looks like we'll block reading from the tun device, send them
949 * an interrupt.
950 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600951 if (vq->pending_used && will_block(net_info->tunfd))
952 trigger_irq(vq);
953
Rusty Russella91d74a2009-07-30 16:03:45 -0600954 /*
955 * Read in the packet. This is where we normally wait (when there's no
956 * incoming network traffic).
957 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600958 len = readv(net_info->tunfd, iov, in);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700959 if (len <= 0)
Sakari Ailuse0377e22011-06-26 19:36:46 +0300960 warn("Failed to read from tun (%d).", errno);
Rusty Russella91d74a2009-07-30 16:03:45 -0600961
962 /*
963 * Mark that packet buffer as used, but don't interrupt here. We want
964 * to wait until we've done as much work as we can.
965 */
Rusty Russell4a8962e2009-06-12 22:27:12 -0600966 add_used(vq, head, len);
Rusty Russell8ca47e02007-07-19 01:49:29 -0700967}
Rusty Russella91d74a2009-07-30 16:03:45 -0600968/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -0700969
Rusty Russella91d74a2009-07-30 16:03:45 -0600970/* This is the helper to create threads: run the service routine in a loop. */
Rusty Russell659a0e62009-06-12 22:27:10 -0600971static int do_thread(void *_vq)
Rusty Russell56ae43d2007-10-22 11:24:23 +1000972{
Rusty Russell659a0e62009-06-12 22:27:10 -0600973 struct virtqueue *vq = _vq;
974
975 for (;;)
976 vq->service(vq);
977 return 0;
Rusty Russell56ae43d2007-10-22 11:24:23 +1000978}
979
Rusty Russell2e04ef72009-07-30 16:03:45 -0600980/*
981 * When a child dies, we kill our entire process group with SIGTERM. This
982 * also has the side effect that the shell restores the console for us!
983 */
Rusty Russell659a0e62009-06-12 22:27:10 -0600984static void kill_launcher(int signal)
Rusty Russell5dae7852008-07-29 09:58:35 -0500985{
Rusty Russell659a0e62009-06-12 22:27:10 -0600986 kill(0, SIGTERM);
987}
988
989static void reset_device(struct device *dev)
990{
991 struct virtqueue *vq;
992
993 verbose("Resetting device %s\n", dev->name);
994
995 /* Clear any features they've acked. */
996 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
997
998 /* We're going to be explicitly killing threads, so ignore them. */
999 signal(SIGCHLD, SIG_IGN);
1000
1001 /* Zero out the virtqueues, get rid of their threads */
1002 for (vq = dev->vq; vq; vq = vq->next) {
1003 if (vq->thread != (pid_t)-1) {
1004 kill(vq->thread, SIGTERM);
1005 waitpid(vq->thread, NULL, 0);
1006 vq->thread = (pid_t)-1;
1007 }
1008 memset(vq->vring.desc, 0,
1009 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1010 lg_last_avail(vq) = 0;
1011 }
1012 dev->running = false;
1013
1014 /* Now we care if threads die. */
1015 signal(SIGCHLD, (void *)kill_launcher);
1016}
1017
Rusty Russella91d74a2009-07-30 16:03:45 -06001018/*L:216
1019 * This actually creates the thread which services the virtqueue for a device.
1020 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001021static void create_thread(struct virtqueue *vq)
1022{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001023 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001024 * Create stack for thread. Since the stack grows upwards, we point
1025 * the stack pointer to the end of this region.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001026 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001027 char *stack = malloc(32768);
1028 unsigned long args[] = { LHREQ_EVENTFD,
1029 vq->config.pfn*getpagesize(), 0 };
1030
1031 /* Create a zero-initialized eventfd. */
1032 vq->eventfd = eventfd(0, 0);
1033 if (vq->eventfd < 0)
1034 err(1, "Creating eventfd");
1035 args[2] = vq->eventfd;
1036
Rusty Russella91d74a2009-07-30 16:03:45 -06001037 /*
1038 * Attach an eventfd to this virtqueue: it will go off when the Guest
1039 * does an LHCALL_NOTIFY for this vq.
1040 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001041 if (write(lguest_fd, &args, sizeof(args)) != 0)
1042 err(1, "Attaching eventfd");
1043
Rusty Russella91d74a2009-07-30 16:03:45 -06001044 /*
1045 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1046 * we get a signal if it dies.
1047 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001048 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1049 if (vq->thread == (pid_t)-1)
1050 err(1, "Creating clone");
Rusty Russella91d74a2009-07-30 16:03:45 -06001051
1052 /* We close our local copy now the child has it. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001053 close(vq->eventfd);
1054}
1055
1056static void start_device(struct device *dev)
1057{
1058 unsigned int i;
1059 struct virtqueue *vq;
1060
1061 verbose("Device %s OK: offered", dev->name);
1062 for (i = 0; i < dev->feature_len; i++)
1063 verbose(" %02x", get_feature_bits(dev)[i]);
1064 verbose(", accepted");
1065 for (i = 0; i < dev->feature_len; i++)
1066 verbose(" %02x", get_feature_bits(dev)
1067 [dev->feature_len+i]);
1068
1069 for (vq = dev->vq; vq; vq = vq->next) {
1070 if (vq->service)
1071 create_thread(vq);
1072 }
1073 dev->running = true;
1074}
1075
1076static void cleanup_devices(void)
1077{
1078 struct device *dev;
1079
1080 for (dev = devices.dev; dev; dev = dev->next)
1081 reset_device(dev);
1082
1083 /* If we saved off the original terminal settings, restore them now. */
1084 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1085 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
Rusty Russell5dae7852008-07-29 09:58:35 -05001086}
1087
Rusty Russella007a752008-05-02 21:50:53 -05001088/* When the Guest tells us they updated the status field, we handle it. */
1089static void update_device_status(struct device *dev)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001090{
Rusty Russell659a0e62009-06-12 22:27:10 -06001091 /* A zero status is a reset, otherwise it's a set of flags. */
1092 if (dev->desc->status == 0)
1093 reset_device(dev);
1094 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
Rusty Russella007a752008-05-02 21:50:53 -05001095 warnx("Device %s configuration FAILED", dev->name);
Rusty Russell659a0e62009-06-12 22:27:10 -06001096 if (dev->running)
1097 reset_device(dev);
Rusty Russell3c3ed482011-07-22 14:39:49 +09301098 } else {
1099 if (dev->running)
1100 err(1, "Device %s features finalized twice", dev->name);
1101 start_device(dev);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001102 }
1103}
1104
Rusty Russella91d74a2009-07-30 16:03:45 -06001105/*L:215
1106 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1107 * particular, it's used to notify us of device status changes during boot.
1108 */
Rusty Russell56739c802009-06-12 22:26:59 -06001109static void handle_output(unsigned long addr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001110{
1111 struct device *i;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001112
Rusty Russell659a0e62009-06-12 22:27:10 -06001113 /* Check each device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001114 for (i = devices.dev; i; i = i->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001115 struct virtqueue *vq;
1116
Rusty Russella91d74a2009-07-30 16:03:45 -06001117 /*
1118 * Notifications to device descriptors mean they updated the
1119 * device status.
1120 */
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001121 if (from_guest_phys(addr) == i->desc) {
Rusty Russella007a752008-05-02 21:50:53 -05001122 update_device_status(i);
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001123 return;
1124 }
1125
Rusty Russell3c3ed482011-07-22 14:39:49 +09301126 /* Devices should not be used before features are finalized. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001127 for (vq = i->vq; vq; vq = vq->next) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001128 if (addr != vq->config.pfn*getpagesize())
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001129 continue;
Rusty Russell3c3ed482011-07-22 14:39:49 +09301130 errx(1, "Notification on %s before setup!", i->name);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001131 }
1132 }
Rusty Russelldde79782007-07-26 10:41:03 -07001133
Rusty Russell2e04ef72009-07-30 16:03:45 -06001134 /*
1135 * Early console write is done using notify on a nul-terminated string
1136 * in Guest memory. It's also great for hacking debugging messages
1137 * into a Guest.
1138 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001139 if (addr >= guest_limit)
1140 errx(1, "Bad NOTIFY %#lx", addr);
1141
1142 write(STDOUT_FILENO, from_guest_phys(addr),
1143 strnlen(from_guest_phys(addr), guest_limit - addr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001144}
1145
Rusty Russelldde79782007-07-26 10:41:03 -07001146/*L:190
1147 * Device Setup
1148 *
1149 * All devices need a descriptor so the Guest knows it exists, and a "struct
1150 * device" so the Launcher can keep track of it. We have common helper
Rusty Russella6bd8e12008-03-28 11:05:53 -05001151 * routines to allocate and manage them.
1152 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001153
Rusty Russell2e04ef72009-07-30 16:03:45 -06001154/*
1155 * The layout of the device page is a "struct lguest_device_desc" followed by a
Rusty Russella586d4f2008-02-04 23:49:56 -05001156 * number of virtqueue descriptors, then two sets of feature bits, then an
1157 * array of configuration bytes. This routine returns the configuration
Rusty Russell2e04ef72009-07-30 16:03:45 -06001158 * pointer.
1159 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001160static u8 *device_config(const struct device *dev)
1161{
1162 return (void *)(dev->desc + 1)
Rusty Russell713b15b2009-06-12 22:26:58 -06001163 + dev->num_vq * sizeof(struct lguest_vqconfig)
1164 + dev->feature_len * 2;
Rusty Russella586d4f2008-02-04 23:49:56 -05001165}
1166
Rusty Russell2e04ef72009-07-30 16:03:45 -06001167/*
1168 * This routine allocates a new "struct lguest_device_desc" from descriptor
Rusty Russella586d4f2008-02-04 23:49:56 -05001169 * table page just above the Guest's normal memory. It returns a pointer to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001170 * that descriptor.
1171 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001172static struct lguest_device_desc *new_dev_desc(u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001173{
Rusty Russella586d4f2008-02-04 23:49:56 -05001174 struct lguest_device_desc d = { .type = type };
1175 void *p;
1176
1177 /* Figure out where the next device config is, based on the last one. */
1178 if (devices.lastdev)
1179 p = device_config(devices.lastdev)
1180 + devices.lastdev->desc->config_len;
1181 else
1182 p = devices.descpage;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001183
Rusty Russell17cbca22007-10-22 11:24:22 +10001184 /* We only have one page for all the descriptors. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001185 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
Rusty Russell17cbca22007-10-22 11:24:22 +10001186 errx(1, "Too many devices");
1187
Rusty Russella586d4f2008-02-04 23:49:56 -05001188 /* p might not be aligned, so we memcpy in. */
1189 return memcpy(p, &d, sizeof(d));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001190}
1191
Rusty Russell2e04ef72009-07-30 16:03:45 -06001192/*
1193 * Each device descriptor is followed by the description of its virtqueues. We
1194 * specify how many descriptors the virtqueue is to have.
1195 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001196static void add_virtqueue(struct device *dev, unsigned int num_descs,
Rusty Russell659a0e62009-06-12 22:27:10 -06001197 void (*service)(struct virtqueue *))
Rusty Russell17cbca22007-10-22 11:24:22 +10001198{
1199 unsigned int pages;
1200 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1201 void *p;
1202
Rusty Russella6bd8e12008-03-28 11:05:53 -05001203 /* First we need some memory for this virtqueue. */
Rusty Russell2966af72008-12-30 09:25:58 -06001204 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
Rusty Russell42b36cc2007-11-12 13:39:18 +11001205 / getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001206 p = get_pages(pages);
1207
Rusty Russelld1c856e2007-11-19 11:20:40 -05001208 /* Initialize the virtqueue */
1209 vq->next = NULL;
1210 vq->last_avail_idx = 0;
1211 vq->dev = dev;
Rusty Russella91d74a2009-07-30 16:03:45 -06001212
1213 /*
1214 * This is the routine the service thread will run, and its Process ID
1215 * once it's running.
1216 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001217 vq->service = service;
1218 vq->thread = (pid_t)-1;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001219
Rusty Russell17cbca22007-10-22 11:24:22 +10001220 /* Initialize the configuration. */
1221 vq->config.num = num_descs;
1222 vq->config.irq = devices.next_irq++;
1223 vq->config.pfn = to_guest_phys(p) / getpagesize();
1224
1225 /* Initialize the vring. */
Rusty Russell2966af72008-12-30 09:25:58 -06001226 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
Rusty Russell17cbca22007-10-22 11:24:22 +10001227
Rusty Russell2e04ef72009-07-30 16:03:45 -06001228 /*
1229 * Append virtqueue to this device's descriptor. We use
Rusty Russella586d4f2008-02-04 23:49:56 -05001230 * device_config() to get the end of the device's current virtqueues;
1231 * we check that we haven't added any config or feature information
Rusty Russell2e04ef72009-07-30 16:03:45 -06001232 * yet, otherwise we'd be overwriting them.
1233 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001234 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1235 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
Rusty Russell713b15b2009-06-12 22:26:58 -06001236 dev->num_vq++;
Rusty Russella586d4f2008-02-04 23:49:56 -05001237 dev->desc->num_vq++;
1238
1239 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
Rusty Russell17cbca22007-10-22 11:24:22 +10001240
Rusty Russell2e04ef72009-07-30 16:03:45 -06001241 /*
1242 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1243 * second.
1244 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001245 for (i = &dev->vq; *i; i = &(*i)->next);
1246 *i = vq;
Rusty Russell17cbca22007-10-22 11:24:22 +10001247}
1248
Rusty Russell2e04ef72009-07-30 16:03:45 -06001249/*
1250 * The first half of the feature bitmask is for us to advertise features. The
1251 * second half is for the Guest to accept features.
1252 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001253static void add_feature(struct device *dev, unsigned bit)
1254{
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001255 u8 *features = get_feature_bits(dev);
Rusty Russella586d4f2008-02-04 23:49:56 -05001256
1257 /* We can't extend the feature bits once we've added config bytes */
1258 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1259 assert(dev->desc->config_len == 0);
Rusty Russell713b15b2009-06-12 22:26:58 -06001260 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
Rusty Russella586d4f2008-02-04 23:49:56 -05001261 }
1262
Rusty Russella586d4f2008-02-04 23:49:56 -05001263 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1264}
1265
Rusty Russell2e04ef72009-07-30 16:03:45 -06001266/*
1267 * This routine sets the configuration fields for an existing device's
Rusty Russella586d4f2008-02-04 23:49:56 -05001268 * descriptor. It only works for the last device, but that's OK because that's
Rusty Russell2e04ef72009-07-30 16:03:45 -06001269 * how we use it.
1270 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001271static void set_config(struct device *dev, unsigned len, const void *conf)
1272{
1273 /* Check we haven't overflowed our single page. */
1274 if (device_config(dev) + len > devices.descpage + getpagesize())
1275 errx(1, "Too many devices");
1276
1277 /* Copy in the config information, and store the length. */
1278 memcpy(device_config(dev), conf, len);
1279 dev->desc->config_len = len;
Rusty Russell8ef562d2009-07-30 16:03:43 -06001280
1281 /* Size must fit in config_len field (8 bits)! */
1282 assert(dev->desc->config_len == len);
Rusty Russella586d4f2008-02-04 23:49:56 -05001283}
1284
Rusty Russell2e04ef72009-07-30 16:03:45 -06001285/*
1286 * This routine does all the creation and setup of a new device, including
Rusty Russella91d74a2009-07-30 16:03:45 -06001287 * calling new_dev_desc() to allocate the descriptor and device memory. We
1288 * don't actually start the service threads until later.
Rusty Russella6bd8e12008-03-28 11:05:53 -05001289 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001290 * See what I mean about userspace being boring?
1291 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001292static struct device *new_device(const char *name, u16 type)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001293{
1294 struct device *dev = malloc(sizeof(*dev));
1295
Rusty Russelldde79782007-07-26 10:41:03 -07001296 /* Now we populate the fields one at a time. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001297 dev->desc = new_dev_desc(type);
Rusty Russell17cbca22007-10-22 11:24:22 +10001298 dev->name = name;
Rusty Russelld1c856e2007-11-19 11:20:40 -05001299 dev->vq = NULL;
Rusty Russell713b15b2009-06-12 22:26:58 -06001300 dev->feature_len = 0;
1301 dev->num_vq = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001302 dev->running = false;
Rusty Russellca16f582012-10-04 12:03:25 +09301303 dev->next = NULL;
Rusty Russella586d4f2008-02-04 23:49:56 -05001304
Rusty Russell2e04ef72009-07-30 16:03:45 -06001305 /*
1306 * Append to device list. Prepending to a single-linked list is
Rusty Russella586d4f2008-02-04 23:49:56 -05001307 * easier, but the user expects the devices to be arranged on the bus
1308 * in command-line order. The first network device on the command line
Rusty Russell2e04ef72009-07-30 16:03:45 -06001309 * is eth0, the first block device /dev/vda, etc.
1310 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001311 if (devices.lastdev)
1312 devices.lastdev->next = dev;
1313 else
1314 devices.dev = dev;
1315 devices.lastdev = dev;
1316
Rusty Russell8ca47e02007-07-19 01:49:29 -07001317 return dev;
1318}
1319
Rusty Russell2e04ef72009-07-30 16:03:45 -06001320/*
1321 * Our first setup routine is the console. It's a fairly simple device, but
1322 * UNIX tty handling makes it uglier than it could be.
1323 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001324static void setup_console(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001325{
1326 struct device *dev;
1327
Rusty Russelldde79782007-07-26 10:41:03 -07001328 /* If we can save the initial standard input settings... */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001329 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1330 struct termios term = orig_term;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001331 /*
1332 * Then we turn off echo, line buffering and ^C etc: We want a
1333 * raw input stream to the Guest.
1334 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001335 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1336 tcsetattr(STDIN_FILENO, TCSANOW, &term);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001337 }
1338
Rusty Russell659a0e62009-06-12 22:27:10 -06001339 dev = new_device("console", VIRTIO_ID_CONSOLE);
1340
Rusty Russelldde79782007-07-26 10:41:03 -07001341 /* We store the console state in dev->priv, and initialize it. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001342 dev->priv = malloc(sizeof(struct console_abort));
1343 ((struct console_abort *)dev->priv)->count = 0;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001344
Rusty Russell2e04ef72009-07-30 16:03:45 -06001345 /*
1346 * The console needs two virtqueues: the input then the output. When
Rusty Russell56ae43d2007-10-22 11:24:23 +10001347 * they put something the input queue, we make sure we're listening to
1348 * stdin. When they put something in the output queue, we write it to
Rusty Russell2e04ef72009-07-30 16:03:45 -06001349 * stdout.
1350 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001351 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1352 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001353
Rusty Russell659a0e62009-06-12 22:27:10 -06001354 verbose("device %u: console\n", ++devices.device_num);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001355}
Rusty Russelldde79782007-07-26 10:41:03 -07001356/*:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001357
Rusty Russell2e04ef72009-07-30 16:03:45 -06001358/*M:010
1359 * Inter-guest networking is an interesting area. Simplest is to have a
Rusty Russell17cbca22007-10-22 11:24:22 +10001360 * --sharenet=<name> option which opens or creates a named pipe. This can be
1361 * used to send packets to another guest in a 1:1 manner.
1362 *
Rusty Russell9f542882011-07-22 14:39:50 +09301363 * More sophisticated is to use one of the tools developed for project like UML
Rusty Russell17cbca22007-10-22 11:24:22 +10001364 * to do networking.
1365 *
1366 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1367 * completely generic ("here's my vring, attach to your vring") and would work
1368 * for any traffic. Of course, namespace and permissions issues need to be
1369 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1370 * multiple inter-guest channels behind one interface, although it would
1371 * require some manner of hotplugging new virtio channels.
1372 *
Rusty Russell9f542882011-07-22 14:39:50 +09301373 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001374:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001375
Rusty Russell8ca47e02007-07-19 01:49:29 -07001376static u32 str2ip(const char *ipaddr)
1377{
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001378 unsigned int b[4];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001379
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001380 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1381 errx(1, "Failed to parse IP address '%s'", ipaddr);
1382 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1383}
1384
1385static void str2mac(const char *macaddr, unsigned char mac[6])
1386{
1387 unsigned int m[6];
1388 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1389 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1390 errx(1, "Failed to parse mac address '%s'", macaddr);
1391 mac[0] = m[0];
1392 mac[1] = m[1];
1393 mac[2] = m[2];
1394 mac[3] = m[3];
1395 mac[4] = m[4];
1396 mac[5] = m[5];
Rusty Russell8ca47e02007-07-19 01:49:29 -07001397}
1398
Rusty Russell2e04ef72009-07-30 16:03:45 -06001399/*
1400 * This code is "adapted" from libbridge: it attaches the Host end of the
Rusty Russelldde79782007-07-26 10:41:03 -07001401 * network device to the bridge device specified by the command line.
1402 *
1403 * This is yet another James Morris contribution (I'm an IP-level guy, so I
Rusty Russell2e04ef72009-07-30 16:03:45 -06001404 * dislike bridging), and I just try not to break it.
1405 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001406static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1407{
1408 int ifidx;
1409 struct ifreq ifr;
1410
1411 if (!*br_name)
1412 errx(1, "must specify bridge name");
1413
1414 ifidx = if_nametoindex(if_name);
1415 if (!ifidx)
1416 errx(1, "interface %s does not exist!", if_name);
1417
1418 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001419 ifr.ifr_name[IFNAMSIZ-1] = '\0';
Rusty Russell8ca47e02007-07-19 01:49:29 -07001420 ifr.ifr_ifindex = ifidx;
1421 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1422 err(1, "can't add %s to bridge %s", if_name, br_name);
1423}
1424
Rusty Russell2e04ef72009-07-30 16:03:45 -06001425/*
1426 * This sets up the Host end of the network device with an IP address, brings
Rusty Russelldde79782007-07-26 10:41:03 -07001427 * it up so packets will flow, the copies the MAC address into the hwaddr
Rusty Russell2e04ef72009-07-30 16:03:45 -06001428 * pointer.
1429 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001430static void configure_device(int fd, const char *tapif, u32 ipaddr)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001431{
1432 struct ifreq ifr;
Rusty Russellf8466192010-08-27 08:39:48 -06001433 struct sockaddr_in sin;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001434
1435 memset(&ifr, 0, sizeof(ifr));
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001436 strcpy(ifr.ifr_name, tapif);
1437
1438 /* Don't read these incantations. Just cut & paste them like I did! */
Rusty Russellf8466192010-08-27 08:39:48 -06001439 sin.sin_family = AF_INET;
1440 sin.sin_addr.s_addr = htonl(ipaddr);
1441 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001442 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001443 err(1, "Setting %s interface address", tapif);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001444 ifr.ifr_flags = IFF_UP;
1445 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001446 err(1, "Bringing interface %s up", tapif);
1447}
1448
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001449static int get_tun_device(char tapif[IFNAMSIZ])
Rusty Russell8ca47e02007-07-19 01:49:29 -07001450{
Rusty Russell8ca47e02007-07-19 01:49:29 -07001451 struct ifreq ifr;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001452 int netfd;
1453
1454 /* Start with this zeroed. Messy but sure. */
1455 memset(&ifr, 0, sizeof(ifr));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001456
Rusty Russell2e04ef72009-07-30 16:03:45 -06001457 /*
1458 * We open the /dev/net/tun device and tell it we want a tap device. A
Rusty Russelldde79782007-07-26 10:41:03 -07001459 * tap device is like a tun device, only somehow different. To tell
1460 * the truth, I completely blundered my way through this code, but it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001461 * works now!
1462 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001463 netfd = open_or_die("/dev/net/tun", O_RDWR);
Rusty Russell398f1872008-07-29 09:58:37 -05001464 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001465 strcpy(ifr.ifr_name, "tap%d");
1466 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1467 err(1, "configuring /dev/net/tun");
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001468
Rusty Russell398f1872008-07-29 09:58:37 -05001469 if (ioctl(netfd, TUNSETOFFLOAD,
1470 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1471 err(1, "Could not set features for tun device");
1472
Rusty Russell2e04ef72009-07-30 16:03:45 -06001473 /*
1474 * We don't need checksums calculated for packets coming in this
1475 * device: trust us!
1476 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001477 ioctl(netfd, TUNSETNOCSUM, 1);
1478
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001479 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1480 return netfd;
1481}
1482
Rusty Russell2e04ef72009-07-30 16:03:45 -06001483/*L:195
1484 * Our network is a Host<->Guest network. This can either use bridging or
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001485 * routing, but the principle is the same: it uses the "tun" device to inject
1486 * packets into the Host as if they came in from a normal network card. We
Rusty Russell2e04ef72009-07-30 16:03:45 -06001487 * just shunt packets between the Guest and the tun device.
1488 */
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001489static void setup_tun_net(char *arg)
1490{
1491 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001492 struct net_info *net_info = malloc(sizeof(*net_info));
1493 int ipfd;
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001494 u32 ip = INADDR_ANY;
1495 bool bridging = false;
1496 char tapif[IFNAMSIZ], *p;
1497 struct virtio_net_config conf;
1498
Rusty Russell659a0e62009-06-12 22:27:10 -06001499 net_info->tunfd = get_tun_device(tapif);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001500
Rusty Russell17cbca22007-10-22 11:24:22 +10001501 /* First we create a new network device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001502 dev = new_device("net", VIRTIO_ID_NET);
1503 dev->priv = net_info;
Rusty Russelldde79782007-07-26 10:41:03 -07001504
Rusty Russell2e04ef72009-07-30 16:03:45 -06001505 /* Network devices need a recv and a send queue, just like console. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001506 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1507 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001508
Rusty Russell2e04ef72009-07-30 16:03:45 -06001509 /*
1510 * We need a socket to perform the magic network ioctls to bring up the
1511 * tap interface, connect to the bridge etc. Any socket will do!
1512 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001513 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1514 if (ipfd < 0)
1515 err(1, "opening IP socket");
1516
Rusty Russelldde79782007-07-26 10:41:03 -07001517 /* If the command line was --tunnet=bridge:<name> do bridging. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001518 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001519 arg += strlen(BRIDGE_PFX);
1520 bridging = true;
1521 }
1522
1523 /* A mac address may follow the bridge name or IP address */
1524 p = strchr(arg, ':');
1525 if (p) {
1526 str2mac(p+1, conf.mac);
Rusty Russell40c42072008-08-12 17:52:51 -05001527 add_feature(dev, VIRTIO_NET_F_MAC);
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001528 *p = '\0';
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001529 }
1530
1531 /* arg is now either an IP address or a bridge name */
1532 if (bridging)
1533 add_to_bridge(ipfd, tapif, arg);
1534 else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001535 ip = str2ip(arg);
1536
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001537 /* Set up the tun device. */
1538 configure_device(ipfd, tapif, ip);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001539
Rusty Russell398f1872008-07-29 09:58:37 -05001540 /* Expect Guest to handle everything except UFO */
1541 add_feature(dev, VIRTIO_NET_F_CSUM);
1542 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
Rusty Russell398f1872008-07-29 09:58:37 -05001543 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1544 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1545 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1546 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1547 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1548 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
Mark McLoughlind1f01322009-05-11 18:11:46 +01001549 /* We handle indirect ring entries */
1550 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
Rusty Russell927cfb92013-07-15 10:50:13 +09301551 /* We're compliant with the damn spec. */
1552 add_feature(dev, VIRTIO_F_ANY_LAYOUT);
Rusty Russella586d4f2008-02-04 23:49:56 -05001553 set_config(dev, sizeof(conf), &conf);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001554
Rusty Russella586d4f2008-02-04 23:49:56 -05001555 /* We don't need the socket any more; setup is done. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001556 close(ipfd);
1557
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001558 devices.device_num++;
1559
1560 if (bridging)
1561 verbose("device %u: tun %s attached to bridge: %s\n",
1562 devices.device_num, tapif, arg);
1563 else
1564 verbose("device %u: tun %s: %s\n",
1565 devices.device_num, tapif, arg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001566}
Rusty Russella91d74a2009-07-30 16:03:45 -06001567/*:*/
Rusty Russell17cbca22007-10-22 11:24:22 +10001568
Rusty Russelle1e72962007-10-25 15:02:50 +10001569/* This hangs off device->priv. */
Rusty Russell1842f232009-07-30 16:03:46 -06001570struct vblk_info {
Rusty Russell17cbca22007-10-22 11:24:22 +10001571 /* The size of the file. */
1572 off64_t len;
1573
1574 /* The file descriptor for the file. */
1575 int fd;
1576
Rusty Russell17cbca22007-10-22 11:24:22 +10001577};
1578
Rusty Russelle1e72962007-10-25 15:02:50 +10001579/*L:210
1580 * The Disk
1581 *
Rusty Russella91d74a2009-07-30 16:03:45 -06001582 * The disk only has one virtqueue, so it only has one thread. It is really
1583 * simple: the Guest asks for a block number and we read or write that position
1584 * in the file.
1585 *
1586 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1587 * slow: the Guest waits until the read is finished before running anything
1588 * else, even if it could have been doing useful work.
1589 *
1590 * We could have used async I/O, except it's reputed to suck so hard that
1591 * characters actually go missing from your code when you try to use it.
Rusty Russelle1e72962007-10-25 15:02:50 +10001592 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001593static void blk_request(struct virtqueue *vq)
Rusty Russell17cbca22007-10-22 11:24:22 +10001594{
Rusty Russell659a0e62009-06-12 22:27:10 -06001595 struct vblk_info *vblk = vq->dev->priv;
Rusty Russell17cbca22007-10-22 11:24:22 +10001596 unsigned int head, out_num, in_num, wlen;
Rusty Russellc0316a92012-10-16 23:56:13 +10301597 int ret, i;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001598 u8 *in;
Rusty Russellc0316a92012-10-16 23:56:13 +10301599 struct virtio_blk_outhdr out;
Rusty Russell659a0e62009-06-12 22:27:10 -06001600 struct iovec iov[vq->vring.num];
Rusty Russell17cbca22007-10-22 11:24:22 +10001601 off64_t off;
1602
Rusty Russella91d74a2009-07-30 16:03:45 -06001603 /*
1604 * Get the next request, where we normally wait. It triggers the
1605 * interrupt to acknowledge previously serviced requests (if any).
1606 */
Rusty Russell659a0e62009-06-12 22:27:10 -06001607 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001608
Rusty Russellc0316a92012-10-16 23:56:13 +10301609 /* Copy the output header from the front of the iov (adjusts iov) */
1610 iov_consume(iov, out_num, &out, sizeof(out));
Rusty Russell17cbca22007-10-22 11:24:22 +10001611
Rusty Russellc0316a92012-10-16 23:56:13 +10301612 /* Find and trim end of iov input array, for our status byte. */
1613 in = NULL;
1614 for (i = out_num + in_num - 1; i >= out_num; i--) {
1615 if (iov[i].iov_len > 0) {
1616 in = iov[i].iov_base + iov[i].iov_len - 1;
1617 iov[i].iov_len--;
1618 break;
1619 }
1620 }
1621 if (!in)
1622 errx(1, "Bad virtblk cmd with no room for status");
1623
Rusty Russella91d74a2009-07-30 16:03:45 -06001624 /*
1625 * For historical reasons, block operations are expressed in 512 byte
1626 * "sectors".
1627 */
Rusty Russellc0316a92012-10-16 23:56:13 +10301628 off = out.sector * 512;
Rusty Russell17cbca22007-10-22 11:24:22 +10001629
Rusty Russell2e04ef72009-07-30 16:03:45 -06001630 /*
Rusty Russell2e04ef72009-07-30 16:03:45 -06001631 * In general the virtio block driver is allowed to try SCSI commands.
1632 * It'd be nice if we supported eject, for example, but we don't.
1633 */
Rusty Russellc0316a92012-10-16 23:56:13 +10301634 if (out.type & VIRTIO_BLK_T_SCSI_CMD) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001635 fprintf(stderr, "Scsi commands unsupported\n");
Rusty Russellcb38fa22008-05-02 21:50:45 -05001636 *in = VIRTIO_BLK_S_UNSUPP;
Anthony Liguori1200e642007-11-08 21:13:44 -06001637 wlen = sizeof(*in);
Rusty Russellc0316a92012-10-16 23:56:13 +10301638 } else if (out.type & VIRTIO_BLK_T_OUT) {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001639 /*
1640 * Write
1641 *
1642 * Move to the right location in the block file. This can fail
1643 * if they try to write past end.
1644 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001645 if (lseek64(vblk->fd, off, SEEK_SET) != off)
Rusty Russellc0316a92012-10-16 23:56:13 +10301646 err(1, "Bad seek to sector %llu", out.sector);
Rusty Russell17cbca22007-10-22 11:24:22 +10001647
Rusty Russellc0316a92012-10-16 23:56:13 +10301648 ret = writev(vblk->fd, iov, out_num);
1649 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
Rusty Russell17cbca22007-10-22 11:24:22 +10001650
Rusty Russell2e04ef72009-07-30 16:03:45 -06001651 /*
1652 * Grr... Now we know how long the descriptor they sent was, we
Rusty Russell17cbca22007-10-22 11:24:22 +10001653 * make sure they didn't try to write over the end of the block
Rusty Russell2e04ef72009-07-30 16:03:45 -06001654 * file (possibly extending it).
1655 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001656 if (ret > 0 && off + ret > vblk->len) {
1657 /* Trim it back to the correct length */
1658 ftruncate64(vblk->fd, vblk->len);
1659 /* Die, bad Guest, die. */
1660 errx(1, "Write past end %llu+%u", off, ret);
1661 }
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001662
1663 wlen = sizeof(*in);
1664 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russellc0316a92012-10-16 23:56:13 +10301665 } else if (out.type & VIRTIO_BLK_T_FLUSH) {
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001666 /* Flush */
1667 ret = fdatasync(vblk->fd);
1668 verbose("FLUSH fdatasync: %i\n", ret);
Anthony Liguori1200e642007-11-08 21:13:44 -06001669 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001670 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
Rusty Russell17cbca22007-10-22 11:24:22 +10001671 } else {
Rusty Russell2e04ef72009-07-30 16:03:45 -06001672 /*
1673 * Read
1674 *
1675 * Move to the right location in the block file. This can fail
1676 * if they try to read past end.
1677 */
Rusty Russell17cbca22007-10-22 11:24:22 +10001678 if (lseek64(vblk->fd, off, SEEK_SET) != off)
Rusty Russellc0316a92012-10-16 23:56:13 +10301679 err(1, "Bad seek to sector %llu", out.sector);
Rusty Russell17cbca22007-10-22 11:24:22 +10001680
Rusty Russellc0316a92012-10-16 23:56:13 +10301681 ret = readv(vblk->fd, iov + out_num, in_num);
Rusty Russell17cbca22007-10-22 11:24:22 +10001682 if (ret >= 0) {
Anthony Liguori1200e642007-11-08 21:13:44 -06001683 wlen = sizeof(*in) + ret;
Rusty Russellcb38fa22008-05-02 21:50:45 -05001684 *in = VIRTIO_BLK_S_OK;
Rusty Russell17cbca22007-10-22 11:24:22 +10001685 } else {
Anthony Liguori1200e642007-11-08 21:13:44 -06001686 wlen = sizeof(*in);
Rusty Russellcb38fa22008-05-02 21:50:45 -05001687 *in = VIRTIO_BLK_S_IOERR;
Rusty Russell17cbca22007-10-22 11:24:22 +10001688 }
1689 }
1690
Rusty Russella91d74a2009-07-30 16:03:45 -06001691 /* Finished that request. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001692 add_used(vq, head, wlen);
Rusty Russell17cbca22007-10-22 11:24:22 +10001693}
1694
Rusty Russelle1e72962007-10-25 15:02:50 +10001695/*L:198 This actually sets up a virtual block device. */
Rusty Russell17cbca22007-10-22 11:24:22 +10001696static void setup_block_file(const char *filename)
1697{
Rusty Russell17cbca22007-10-22 11:24:22 +10001698 struct device *dev;
1699 struct vblk_info *vblk;
Rusty Russella586d4f2008-02-04 23:49:56 -05001700 struct virtio_blk_config conf;
Rusty Russell17cbca22007-10-22 11:24:22 +10001701
Rusty Russell2e04ef72009-07-30 16:03:45 -06001702 /* Creat the device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001703 dev = new_device("block", VIRTIO_ID_BLOCK);
Rusty Russell17cbca22007-10-22 11:24:22 +10001704
Rusty Russelle1e72962007-10-25 15:02:50 +10001705 /* The device has one virtqueue, where the Guest places requests. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001706 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
Rusty Russell17cbca22007-10-22 11:24:22 +10001707
1708 /* Allocate the room for our own bookkeeping */
1709 vblk = dev->priv = malloc(sizeof(*vblk));
1710
1711 /* First we open the file and store the length. */
1712 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1713 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1714
Tejun Heo7bc9fdd2010-09-03 11:56:18 +02001715 /* We support FLUSH. */
1716 add_feature(dev, VIRTIO_BLK_F_FLUSH);
Rusty Russella586d4f2008-02-04 23:49:56 -05001717
Rusty Russell17cbca22007-10-22 11:24:22 +10001718 /* Tell Guest how many sectors this device has. */
Rusty Russella586d4f2008-02-04 23:49:56 -05001719 conf.capacity = cpu_to_le64(vblk->len / 512);
Rusty Russell17cbca22007-10-22 11:24:22 +10001720
Rusty Russell2e04ef72009-07-30 16:03:45 -06001721 /*
1722 * Tell Guest not to put in too many descriptors at once: two are used
1723 * for the in and out elements.
1724 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001725 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1726 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1727
Rusty Russell8ef562d2009-07-30 16:03:43 -06001728 /* Don't try to put whole struct: we have 8 bit limit. */
1729 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
Rusty Russell17cbca22007-10-22 11:24:22 +10001730
Rusty Russell17cbca22007-10-22 11:24:22 +10001731 verbose("device %u: virtblock %llu sectors\n",
Rusty Russell659a0e62009-06-12 22:27:10 -06001732 ++devices.device_num, le64_to_cpu(conf.capacity));
Rusty Russell17cbca22007-10-22 11:24:22 +10001733}
Rusty Russell28fd6d72008-07-29 09:58:33 -05001734
Rusty Russell2e04ef72009-07-30 16:03:45 -06001735/*L:211
1736 * Our random number generator device reads from /dev/random into the Guest's
Rusty Russell28fd6d72008-07-29 09:58:33 -05001737 * input buffers. The usual case is that the Guest doesn't want random numbers
1738 * and so has no buffers although /dev/random is still readable, whereas
1739 * console is the reverse.
1740 *
Rusty Russell2e04ef72009-07-30 16:03:45 -06001741 * The same logic applies, however.
1742 */
1743struct rng_info {
1744 int rfd;
1745};
1746
Rusty Russell659a0e62009-06-12 22:27:10 -06001747static void rng_input(struct virtqueue *vq)
Rusty Russell28fd6d72008-07-29 09:58:33 -05001748{
1749 int len;
1750 unsigned int head, in_num, out_num, totlen = 0;
Rusty Russell659a0e62009-06-12 22:27:10 -06001751 struct rng_info *rng_info = vq->dev->priv;
1752 struct iovec iov[vq->vring.num];
Rusty Russell28fd6d72008-07-29 09:58:33 -05001753
1754 /* First we need a buffer from the Guests's virtqueue. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001755 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001756 if (out_num)
1757 errx(1, "Output buffers in rng?");
1758
Rusty Russell2e04ef72009-07-30 16:03:45 -06001759 /*
Rusty Russella91d74a2009-07-30 16:03:45 -06001760 * Just like the console write, we loop to cover the whole iovec.
1761 * In this case, short reads actually happen quite a bit.
Rusty Russell2e04ef72009-07-30 16:03:45 -06001762 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001763 while (!iov_empty(iov, in_num)) {
Rusty Russell659a0e62009-06-12 22:27:10 -06001764 len = readv(rng_info->rfd, iov, in_num);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001765 if (len <= 0)
1766 err(1, "Read from /dev/random gave %i", len);
Rusty Russellc0316a92012-10-16 23:56:13 +10301767 iov_consume(iov, in_num, NULL, len);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001768 totlen += len;
1769 }
1770
1771 /* Tell the Guest about the new input. */
Rusty Russell38bc2b82009-06-12 22:27:11 -06001772 add_used(vq, head, totlen);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001773}
1774
Rusty Russell2e04ef72009-07-30 16:03:45 -06001775/*L:199
1776 * This creates a "hardware" random number device for the Guest.
1777 */
Rusty Russell28fd6d72008-07-29 09:58:33 -05001778static void setup_rng(void)
1779{
1780 struct device *dev;
Rusty Russell659a0e62009-06-12 22:27:10 -06001781 struct rng_info *rng_info = malloc(sizeof(*rng_info));
Rusty Russell28fd6d72008-07-29 09:58:33 -05001782
Rusty Russell2e04ef72009-07-30 16:03:45 -06001783 /* Our device's privat info simply contains the /dev/random fd. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001784 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001785
Rusty Russell2e04ef72009-07-30 16:03:45 -06001786 /* Create the new device. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001787 dev = new_device("rng", VIRTIO_ID_RNG);
1788 dev->priv = rng_info;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001789
1790 /* The device has one virtqueue, where the Guest places inbufs. */
Rusty Russell659a0e62009-06-12 22:27:10 -06001791 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
Rusty Russell28fd6d72008-07-29 09:58:33 -05001792
1793 verbose("device %u: rng\n", devices.device_num++);
1794}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001795/* That's the end of device setup. */
Balaji Raoec04b132007-12-28 14:26:24 +05301796
Rusty Russella6bd8e12008-03-28 11:05:53 -05001797/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
Balaji Raoec04b132007-12-28 14:26:24 +05301798static void __attribute__((noreturn)) restart_guest(void)
1799{
1800 unsigned int i;
1801
Rusty Russell2e04ef72009-07-30 16:03:45 -06001802 /*
1803 * Since we don't track all open fds, we simply close everything beyond
1804 * stderr.
1805 */
Balaji Raoec04b132007-12-28 14:26:24 +05301806 for (i = 3; i < FD_SETSIZE; i++)
1807 close(i);
Rusty Russell8c798732008-07-29 09:58:38 -05001808
Rusty Russell659a0e62009-06-12 22:27:10 -06001809 /* Reset all the devices (kills all threads). */
1810 cleanup_devices();
1811
Balaji Raoec04b132007-12-28 14:26:24 +05301812 execv(main_args[0], main_args);
1813 err(1, "Could not exec %s", main_args[0]);
1814}
Rusty Russell8ca47e02007-07-19 01:49:29 -07001815
Rusty Russell2e04ef72009-07-30 16:03:45 -06001816/*L:220
1817 * Finally we reach the core of the Launcher which runs the Guest, serves
1818 * its input and output, and finally, lays it to rest.
1819 */
Rusty Russell56739c802009-06-12 22:26:59 -06001820static void __attribute__((noreturn)) run_guest(void)
Rusty Russell8ca47e02007-07-19 01:49:29 -07001821{
1822 for (;;) {
Rusty Russell17cbca22007-10-22 11:24:22 +10001823 unsigned long notify_addr;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001824 int readval;
1825
1826 /* We read from the /dev/lguest device to run the Guest. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001827 readval = pread(lguest_fd, &notify_addr,
1828 sizeof(notify_addr), cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001829
Rusty Russell17cbca22007-10-22 11:24:22 +10001830 /* One unsigned long means the Guest did HCALL_NOTIFY */
1831 if (readval == sizeof(notify_addr)) {
1832 verbose("Notify on address %#lx\n", notify_addr);
Rusty Russell56739c802009-06-12 22:26:59 -06001833 handle_output(notify_addr);
Rusty Russelldde79782007-07-26 10:41:03 -07001834 /* ENOENT means the Guest died. Reading tells us why. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001835 } else if (errno == ENOENT) {
1836 char reason[1024] = { 0 };
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001837 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001838 errx(1, "%s", reason);
Balaji Raoec04b132007-12-28 14:26:24 +05301839 /* ERESTART means that we need to reboot the guest */
1840 } else if (errno == ERESTART) {
1841 restart_guest();
Rusty Russell659a0e62009-06-12 22:27:10 -06001842 /* Anything else means a bug or incompatible change. */
1843 } else
Rusty Russell8ca47e02007-07-19 01:49:29 -07001844 err(1, "Running guest failed");
Rusty Russell8ca47e02007-07-19 01:49:29 -07001845 }
1846}
Rusty Russella6bd8e12008-03-28 11:05:53 -05001847/*L:240
Rusty Russelle1e72962007-10-25 15:02:50 +10001848 * This is the end of the Launcher. The good news: we are over halfway
1849 * through! The bad news: the most fiendish part of the code still lies ahead
1850 * of us.
Rusty Russelldde79782007-07-26 10:41:03 -07001851 *
Rusty Russelle1e72962007-10-25 15:02:50 +10001852 * Are you ready? Take a deep breath and join me in the core of the Host, in
1853 * "make Host".
Rusty Russell2e04ef72009-07-30 16:03:45 -06001854:*/
Rusty Russell8ca47e02007-07-19 01:49:29 -07001855
1856static struct option opts[] = {
1857 { "verbose", 0, NULL, 'v' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001858 { "tunnet", 1, NULL, 't' },
1859 { "block", 1, NULL, 'b' },
Rusty Russell28fd6d72008-07-29 09:58:33 -05001860 { "rng", 0, NULL, 'r' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001861 { "initrd", 1, NULL, 'i' },
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001862 { "username", 1, NULL, 'u' },
1863 { "chroot", 1, NULL, 'c' },
Rusty Russell8ca47e02007-07-19 01:49:29 -07001864 { NULL },
1865};
1866static void usage(void)
1867{
1868 errx(1, "Usage: lguest [--verbose] "
Mark McLoughlindec6a2b2008-07-29 09:58:33 -05001869 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
Rusty Russell8ca47e02007-07-19 01:49:29 -07001870 "|--block=<filename>|--initrd=<filename>]...\n"
1871 "<mem-in-mb> vmlinux [args...]");
1872}
1873
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001874/*L:105 The main routine is where the real work begins: */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001875int main(int argc, char *argv[])
1876{
Rusty Russell2e04ef72009-07-30 16:03:45 -06001877 /* Memory, code startpoint and size of the (optional) initrd. */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -03001878 unsigned long mem = 0, start, initrd_size = 0;
Rusty Russell56739c802009-06-12 22:26:59 -06001879 /* Two temporaries. */
1880 int i, c;
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001881 /* The boot information for the Guest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001882 struct boot_params *boot;
Rusty Russelldde79782007-07-26 10:41:03 -07001883 /* If they specify an initrd file to load. */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001884 const char *initrd_name = NULL;
1885
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001886 /* Password structure for initgroups/setres[gu]id */
1887 struct passwd *user_details = NULL;
1888
1889 /* Directory to chroot to */
1890 char *chroot_path = NULL;
1891
Balaji Raoec04b132007-12-28 14:26:24 +05301892 /* Save the args: we "reboot" by execing ourselves again. */
1893 main_args = argv;
Balaji Raoec04b132007-12-28 14:26:24 +05301894
Rusty Russell2e04ef72009-07-30 16:03:45 -06001895 /*
1896 * First we initialize the device list. We keep a pointer to the last
Rusty Russell659a0e62009-06-12 22:27:10 -06001897 * device, and the next interrupt number to use for devices (1:
Rusty Russell2e04ef72009-07-30 16:03:45 -06001898 * remember that 0 is used by the timer).
1899 */
Rusty Russella586d4f2008-02-04 23:49:56 -05001900 devices.lastdev = NULL;
Rusty Russell17cbca22007-10-22 11:24:22 +10001901 devices.next_irq = 1;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001902
Rusty Russella91d74a2009-07-30 16:03:45 -06001903 /* We're CPU 0. In fact, that's the only CPU possible right now. */
Glauber de Oliveira Costae3283fa2008-01-07 11:05:23 -02001904 cpu_id = 0;
Rusty Russella91d74a2009-07-30 16:03:45 -06001905
Rusty Russell2e04ef72009-07-30 16:03:45 -06001906 /*
1907 * We need to know how much memory so we can set up the device
Rusty Russelldde79782007-07-26 10:41:03 -07001908 * descriptor and memory pages for the devices as we parse the command
1909 * line. So we quickly look through the arguments to find the amount
Rusty Russell2e04ef72009-07-30 16:03:45 -06001910 * of memory now.
1911 */
Rusty Russell6570c45992007-07-23 18:43:56 -07001912 for (i = 1; i < argc; i++) {
1913 if (argv[i][0] != '-') {
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001914 mem = atoi(argv[i]) * 1024 * 1024;
Rusty Russell2e04ef72009-07-30 16:03:45 -06001915 /*
1916 * We start by mapping anonymous pages over all of
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001917 * guest-physical memory range. This fills it with 0,
1918 * and ensures that the Guest won't be killed when it
Rusty Russell2e04ef72009-07-30 16:03:45 -06001919 * tries to access it.
1920 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001921 guest_base = map_zeroed_pages(mem / getpagesize()
1922 + DEVICE_PAGES);
1923 guest_limit = mem;
1924 guest_max = mem + DEVICE_PAGES*getpagesize();
Rusty Russell17cbca22007-10-22 11:24:22 +10001925 devices.descpage = get_pages(1);
Rusty Russell6570c45992007-07-23 18:43:56 -07001926 break;
1927 }
1928 }
Rusty Russelldde79782007-07-26 10:41:03 -07001929
1930 /* The options are fairly straight-forward */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001931 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1932 switch (c) {
1933 case 'v':
1934 verbose = true;
1935 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001936 case 't':
Rusty Russell17cbca22007-10-22 11:24:22 +10001937 setup_tun_net(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001938 break;
1939 case 'b':
Rusty Russell17cbca22007-10-22 11:24:22 +10001940 setup_block_file(optarg);
Rusty Russell8ca47e02007-07-19 01:49:29 -07001941 break;
Rusty Russell28fd6d72008-07-29 09:58:33 -05001942 case 'r':
1943 setup_rng();
1944 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001945 case 'i':
1946 initrd_name = optarg;
1947 break;
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06001948 case 'u':
1949 user_details = getpwnam(optarg);
1950 if (!user_details)
1951 err(1, "getpwnam failed, incorrect username?");
1952 break;
1953 case 'c':
1954 chroot_path = optarg;
1955 break;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001956 default:
1957 warnx("Unknown argument %s", argv[optind]);
1958 usage();
1959 }
1960 }
Rusty Russell2e04ef72009-07-30 16:03:45 -06001961 /*
1962 * After the other arguments we expect memory and kernel image name,
1963 * followed by command line arguments for the kernel.
1964 */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001965 if (optind + 2 > argc)
1966 usage();
1967
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001968 verbose("Guest base is at %p\n", guest_base);
1969
Rusty Russelldde79782007-07-26 10:41:03 -07001970 /* We always have a console device */
Rusty Russell17cbca22007-10-22 11:24:22 +10001971 setup_console();
Rusty Russell8ca47e02007-07-19 01:49:29 -07001972
Rusty Russell8ca47e02007-07-19 01:49:29 -07001973 /* Now we load the kernel */
Rusty Russell47436aa2007-10-22 11:03:36 +10001974 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
Rusty Russell8ca47e02007-07-19 01:49:29 -07001975
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10001976 /* Boot information is stashed at physical address 0 */
1977 boot = from_guest_phys(0);
1978
Rusty Russelldde79782007-07-26 10:41:03 -07001979 /* Map the initrd image if requested (at top of physical memory) */
Rusty Russell8ca47e02007-07-19 01:49:29 -07001980 if (initrd_name) {
1981 initrd_size = load_initrd(initrd_name, mem);
Rusty Russell2e04ef72009-07-30 16:03:45 -06001982 /*
1983 * These are the location in the Linux boot header where the
1984 * start and size of the initrd are expected to be found.
1985 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001986 boot->hdr.ramdisk_image = mem - initrd_size;
1987 boot->hdr.ramdisk_size = initrd_size;
Rusty Russelldde79782007-07-26 10:41:03 -07001988 /* The bootloader type 0xFF means "unknown"; that's OK. */
Rusty Russell43d33b22007-10-22 11:29:57 +10001989 boot->hdr.type_of_loader = 0xFF;
Rusty Russell8ca47e02007-07-19 01:49:29 -07001990 }
1991
Rusty Russell2e04ef72009-07-30 16:03:45 -06001992 /*
1993 * The Linux boot header contains an "E820" memory map: ours is a
1994 * simple, single region.
1995 */
Rusty Russell43d33b22007-10-22 11:29:57 +10001996 boot->e820_entries = 1;
1997 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
Rusty Russell2e04ef72009-07-30 16:03:45 -06001998 /*
1999 * The boot header contains a command line pointer: we put the command
2000 * line after the boot header.
2001 */
Rusty Russell43d33b22007-10-22 11:29:57 +10002002 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
Rusty Russelle1e72962007-10-25 15:02:50 +10002003 /* We use a simple helper to copy the arguments separated by spaces. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002004 concat((char *)(boot + 1), argv+optind+2);
Rusty Russelldde79782007-07-26 10:41:03 -07002005
Rusty Russelle22a5392011-08-15 10:15:10 +09302006 /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
2007 boot->hdr.kernel_alignment = 0x1000000;
2008
Rusty Russell814a0e52007-10-22 11:29:44 +10002009 /* Boot protocol version: 2.07 supports the fields for lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002010 boot->hdr.version = 0x207;
Rusty Russell814a0e52007-10-22 11:29:44 +10002011
2012 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
Rusty Russell43d33b22007-10-22 11:29:57 +10002013 boot->hdr.hardware_subarch = 1;
Rusty Russell814a0e52007-10-22 11:29:44 +10002014
Rusty Russell43d33b22007-10-22 11:29:57 +10002015 /* Tell the entry path not to try to reload segment registers. */
2016 boot->hdr.loadflags |= KEEP_SEGMENTS;
Rusty Russell8ca47e02007-07-19 01:49:29 -07002017
Rusty Russell9f542882011-07-22 14:39:50 +09302018 /* We tell the kernel to initialize the Guest. */
Rusty Russell56739c802009-06-12 22:26:59 -06002019 tell_kernel(start);
Rusty Russelldde79782007-07-26 10:41:03 -07002020
Rusty Russella91d74a2009-07-30 16:03:45 -06002021 /* Ensure that we terminate if a device-servicing child dies. */
Rusty Russell659a0e62009-06-12 22:27:10 -06002022 signal(SIGCHLD, kill_launcher);
2023
2024 /* If we exit via err(), this kills all the threads, restores tty. */
2025 atexit(cleanup_devices);
Rusty Russell8ca47e02007-07-19 01:49:29 -07002026
Philip Sanderson8aeb36e2011-01-20 21:37:28 -06002027 /* If requested, chroot to a directory */
2028 if (chroot_path) {
2029 if (chroot(chroot_path) != 0)
2030 err(1, "chroot(\"%s\") failed", chroot_path);
2031
2032 if (chdir("/") != 0)
2033 err(1, "chdir(\"/\") failed");
2034
2035 verbose("chroot done\n");
2036 }
2037
2038 /* If requested, drop privileges */
2039 if (user_details) {
2040 uid_t u;
2041 gid_t g;
2042
2043 u = user_details->pw_uid;
2044 g = user_details->pw_gid;
2045
2046 if (initgroups(user_details->pw_name, g) != 0)
2047 err(1, "initgroups failed");
2048
2049 if (setresgid(g, g, g) != 0)
2050 err(1, "setresgid failed");
2051
2052 if (setresuid(u, u, u) != 0)
2053 err(1, "setresuid failed");
2054
2055 verbose("Dropping privileges completed\n");
2056 }
2057
Rusty Russelldde79782007-07-26 10:41:03 -07002058 /* Finally, run the Guest. This doesn't return. */
Rusty Russell56739c802009-06-12 22:26:59 -06002059 run_guest();
Rusty Russell8ca47e02007-07-19 01:49:29 -07002060}
Rusty Russellf56a3842007-07-26 10:41:05 -07002061/*:*/
2062
2063/*M:999
2064 * Mastery is done: you now know everything I do.
2065 *
2066 * But surely you have seen code, features and bugs in your wanderings which
2067 * you now yearn to attack? That is the real game, and I look forward to you
2068 * patching and forking lguest into the Your-Name-Here-visor.
2069 *
2070 * Farewell, and good coding!
2071 * Rusty Russell.
2072 */