Rusty Russell | f938d2c | 2007-07-26 10:41:02 -0700 | [diff] [blame] | 1 | /*P:100 This is the Launcher code, a simple program which lays out the |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 2 | * "physical" memory for the new Guest by mapping the kernel image and |
| 3 | * the virtual devices, then opens /dev/lguest to tell the kernel |
| 4 | * about the Guest and control it. :*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 5 | #define _LARGEFILE64_SOURCE |
| 6 | #define _GNU_SOURCE |
| 7 | #include <stdio.h> |
| 8 | #include <string.h> |
| 9 | #include <unistd.h> |
| 10 | #include <err.h> |
| 11 | #include <stdint.h> |
| 12 | #include <stdlib.h> |
| 13 | #include <elf.h> |
| 14 | #include <sys/mman.h> |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 15 | #include <sys/param.h> |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 16 | #include <sys/types.h> |
| 17 | #include <sys/stat.h> |
| 18 | #include <sys/wait.h> |
| 19 | #include <fcntl.h> |
| 20 | #include <stdbool.h> |
| 21 | #include <errno.h> |
| 22 | #include <ctype.h> |
| 23 | #include <sys/socket.h> |
| 24 | #include <sys/ioctl.h> |
| 25 | #include <sys/time.h> |
| 26 | #include <time.h> |
| 27 | #include <netinet/in.h> |
| 28 | #include <net/if.h> |
| 29 | #include <linux/sockios.h> |
| 30 | #include <linux/if_tun.h> |
| 31 | #include <sys/uio.h> |
| 32 | #include <termios.h> |
| 33 | #include <getopt.h> |
| 34 | #include <zlib.h> |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 35 | #include <assert.h> |
| 36 | #include <sched.h> |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 37 | #include <limits.h> |
| 38 | #include <stddef.h> |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 39 | #include <signal.h> |
Rusty Russell | b45d8cb | 2007-10-22 10:56:24 +1000 | [diff] [blame] | 40 | #include "linux/lguest_launcher.h" |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 41 | #include "linux/virtio_config.h" |
| 42 | #include "linux/virtio_net.h" |
| 43 | #include "linux/virtio_blk.h" |
| 44 | #include "linux/virtio_console.h" |
Rusty Russell | 28fd6d7 | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 45 | #include "linux/virtio_rng.h" |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 46 | #include "linux/virtio_ring.h" |
Rusty Russell | d5d02d6 | 2008-10-31 11:24:25 -0500 | [diff] [blame^] | 47 | #include "asm/bootparam.h" |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 48 | /*L:110 We can ignore the 39 include files we need for this program, but I do |
Rusty Russell | db24e8c | 2007-10-25 14:09:25 +1000 | [diff] [blame] | 49 | * want to draw attention to the use of kernel-style types. |
| 50 | * |
| 51 | * As Linus said, "C is a Spartan language, and so should your naming be." I |
| 52 | * like these abbreviations, so we define them here. Note that u64 is always |
| 53 | * unsigned long long, which works on all Linux systems: this means that we can |
| 54 | * use %llu in printf for any u64. */ |
| 55 | typedef unsigned long long u64; |
| 56 | typedef uint32_t u32; |
| 57 | typedef uint16_t u16; |
| 58 | typedef uint8_t u8; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 59 | /*:*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 60 | |
| 61 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ |
| 62 | #define NET_PEERNUM 1 |
| 63 | #define BRIDGE_PFX "bridge:" |
| 64 | #ifndef SIOCBRADDIF |
| 65 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ |
| 66 | #endif |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 67 | /* We can have up to 256 pages for devices. */ |
| 68 | #define DEVICE_PAGES 256 |
Rusty Russell | 0f0c4fa | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 69 | /* This will occupy 3 pages: it must be a power of 2. */ |
| 70 | #define VIRTQUEUE_NUM 256 |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 71 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 72 | /*L:120 verbose is both a global flag and a macro. The C preprocessor allows |
| 73 | * this, and although I wouldn't recommend it, it works quite nicely here. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 74 | static bool verbose; |
| 75 | #define verbose(args...) \ |
| 76 | do { if (verbose) printf(args); } while(0) |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 77 | /*:*/ |
| 78 | |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 79 | /* File descriptors for the Waker. */ |
| 80 | struct { |
| 81 | int pipe[2]; |
| 82 | int lguest_fd; |
| 83 | } waker_fds; |
| 84 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 85 | /* The pointer to the start of guest memory. */ |
| 86 | static void *guest_base; |
| 87 | /* The maximum guest physical address allowed, and maximum possible. */ |
| 88 | static unsigned long guest_limit, guest_max; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 89 | /* The pipe for signal hander to write to. */ |
| 90 | static int timeoutpipe[2]; |
Rusty Russell | aa12498 | 2008-07-29 09:58:36 -0500 | [diff] [blame] | 91 | static unsigned int timeout_usec = 500; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 92 | |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 93 | /* a per-cpu variable indicating whose vcpu is currently running */ |
| 94 | static unsigned int __thread cpu_id; |
| 95 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 96 | /* This is our list of devices. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 97 | struct device_list |
| 98 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 99 | /* Summary information about the devices in our list: ready to pass to |
| 100 | * select() to ask which need servicing.*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 101 | fd_set infds; |
| 102 | int max_infd; |
| 103 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 104 | /* Counter to assign interrupt numbers. */ |
| 105 | unsigned int next_irq; |
| 106 | |
| 107 | /* Counter to print out convenient device numbers. */ |
| 108 | unsigned int device_num; |
| 109 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 110 | /* The descriptor page for the devices. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 111 | u8 *descpage; |
| 112 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 113 | /* A single linked list of devices. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 114 | struct device *dev; |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 115 | /* And a pointer to the last device for easy append and also for |
| 116 | * configuration appending. */ |
| 117 | struct device *lastdev; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 118 | }; |
| 119 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 120 | /* The list of Guest devices, based on command line arguments. */ |
| 121 | static struct device_list devices; |
| 122 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 123 | /* The device structure describes a single device. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 124 | struct device |
| 125 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 126 | /* The linked-list pointer. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 127 | struct device *next; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 128 | |
| 129 | /* The this device's descriptor, as mapped into the Guest. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 130 | struct lguest_device_desc *desc; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 131 | |
| 132 | /* The name of this device, for --verbose. */ |
| 133 | const char *name; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 134 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 135 | /* If handle_input is set, it wants to be called when this file |
| 136 | * descriptor is ready. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 137 | int fd; |
| 138 | bool (*handle_input)(int fd, struct device *me); |
| 139 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 140 | /* Any queues attached to this device */ |
| 141 | struct virtqueue *vq; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 142 | |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 143 | /* Handle status being finalized (ie. feature bits stable). */ |
| 144 | void (*ready)(struct device *me); |
| 145 | |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 146 | /* Device-specific data. */ |
| 147 | void *priv; |
| 148 | }; |
| 149 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 150 | /* The virtqueue structure describes a queue attached to a device. */ |
| 151 | struct virtqueue |
| 152 | { |
| 153 | struct virtqueue *next; |
| 154 | |
| 155 | /* Which device owns me. */ |
| 156 | struct device *dev; |
| 157 | |
| 158 | /* The configuration for this queue. */ |
| 159 | struct lguest_vqconfig config; |
| 160 | |
| 161 | /* The actual ring of buffers. */ |
| 162 | struct vring vring; |
| 163 | |
| 164 | /* Last available index we saw. */ |
| 165 | u16 last_avail_idx; |
| 166 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 167 | /* The routine to call when the Guest pings us, or timeout. */ |
| 168 | void (*handle_output)(int fd, struct virtqueue *me, bool timeout); |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 169 | |
| 170 | /* Outstanding buffers */ |
| 171 | unsigned int inflight; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 172 | |
| 173 | /* Is this blocked awaiting a timer? */ |
| 174 | bool blocked; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 175 | }; |
| 176 | |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 177 | /* Remember the arguments to the program so we can "reboot" */ |
| 178 | static char **main_args; |
| 179 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 180 | /* Since guest is UP and we don't run at the same time, we don't need barriers. |
| 181 | * But I include them in the code in case others copy it. */ |
| 182 | #define wmb() |
| 183 | |
| 184 | /* Convert an iovec element to the given type. |
| 185 | * |
| 186 | * This is a fairly ugly trick: we need to know the size of the type and |
| 187 | * alignment requirement to check the pointer is kosher. It's also nice to |
| 188 | * have the name of the type in case we report failure. |
| 189 | * |
| 190 | * Typing those three things all the time is cumbersome and error prone, so we |
| 191 | * have a macro which sets them all up and passes to the real function. */ |
| 192 | #define convert(iov, type) \ |
| 193 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) |
| 194 | |
| 195 | static void *_convert(struct iovec *iov, size_t size, size_t align, |
| 196 | const char *name) |
| 197 | { |
| 198 | if (iov->iov_len != size) |
| 199 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); |
| 200 | if ((unsigned long)iov->iov_base % align != 0) |
| 201 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); |
| 202 | return iov->iov_base; |
| 203 | } |
| 204 | |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 205 | /* Wrapper for the last available index. Makes it easier to change. */ |
| 206 | #define lg_last_avail(vq) ((vq)->last_avail_idx) |
| 207 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 208 | /* The virtio configuration space is defined to be little-endian. x86 is |
| 209 | * little-endian too, but it's nice to be explicit so we have these helpers. */ |
| 210 | #define cpu_to_le16(v16) (v16) |
| 211 | #define cpu_to_le32(v32) (v32) |
| 212 | #define cpu_to_le64(v64) (v64) |
| 213 | #define le16_to_cpu(v16) (v16) |
| 214 | #define le32_to_cpu(v32) (v32) |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 215 | #define le64_to_cpu(v64) (v64) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 216 | |
Rusty Russell | 28fd6d7 | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 217 | /* Is this iovec empty? */ |
| 218 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) |
| 219 | { |
| 220 | unsigned int i; |
| 221 | |
| 222 | for (i = 0; i < num_iov; i++) |
| 223 | if (iov[i].iov_len) |
| 224 | return false; |
| 225 | return true; |
| 226 | } |
| 227 | |
| 228 | /* Take len bytes from the front of this iovec. */ |
| 229 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) |
| 230 | { |
| 231 | unsigned int i; |
| 232 | |
| 233 | for (i = 0; i < num_iov; i++) { |
| 234 | unsigned int used; |
| 235 | |
| 236 | used = iov[i].iov_len < len ? iov[i].iov_len : len; |
| 237 | iov[i].iov_base += used; |
| 238 | iov[i].iov_len -= used; |
| 239 | len -= used; |
| 240 | } |
| 241 | assert(len == 0); |
| 242 | } |
| 243 | |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 244 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
| 245 | static u8 *get_feature_bits(struct device *dev) |
| 246 | { |
| 247 | return (u8 *)(dev->desc + 1) |
| 248 | + dev->desc->num_vq * sizeof(struct lguest_vqconfig); |
| 249 | } |
| 250 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 251 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
| 252 | * where pointers run wild and free! Unfortunately, like most userspace |
| 253 | * programs, it's quite boring (which is why everyone likes to hack on the |
| 254 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it |
| 255 | * will get you through this section. Or, maybe not. |
| 256 | * |
| 257 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" |
| 258 | * memory and stores it in "guest_base". In other words, Guest physical == |
| 259 | * Launcher virtual with an offset. |
| 260 | * |
| 261 | * This can be tough to get your head around, but usually it just means that we |
| 262 | * use these trivial conversion functions when the Guest gives us it's |
| 263 | * "physical" addresses: */ |
| 264 | static void *from_guest_phys(unsigned long addr) |
| 265 | { |
| 266 | return guest_base + addr; |
| 267 | } |
| 268 | |
| 269 | static unsigned long to_guest_phys(const void *addr) |
| 270 | { |
| 271 | return (addr - guest_base); |
| 272 | } |
| 273 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 274 | /*L:130 |
| 275 | * Loading the Kernel. |
| 276 | * |
| 277 | * We start with couple of simple helper routines. open_or_die() avoids |
| 278 | * error-checking code cluttering the callers: */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 279 | static int open_or_die(const char *name, int flags) |
| 280 | { |
| 281 | int fd = open(name, flags); |
| 282 | if (fd < 0) |
| 283 | err(1, "Failed to open %s", name); |
| 284 | return fd; |
| 285 | } |
| 286 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 287 | /* map_zeroed_pages() takes a number of pages. */ |
| 288 | static void *map_zeroed_pages(unsigned int num) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 289 | { |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 290 | int fd = open_or_die("/dev/zero", O_RDONLY); |
| 291 | void *addr; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 292 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 293 | /* We use a private mapping (ie. if we write to the page, it will be |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 294 | * copied). */ |
| 295 | addr = mmap(NULL, getpagesize() * num, |
| 296 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); |
| 297 | if (addr == MAP_FAILED) |
| 298 | err(1, "Mmaping %u pages of /dev/zero", num); |
Mark McLoughlin | 34bdaab | 2008-06-13 14:04:58 +0100 | [diff] [blame] | 299 | close(fd); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 300 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 301 | return addr; |
| 302 | } |
| 303 | |
| 304 | /* Get some more pages for a device. */ |
| 305 | static void *get_pages(unsigned int num) |
| 306 | { |
| 307 | void *addr = from_guest_phys(guest_limit); |
| 308 | |
| 309 | guest_limit += num * getpagesize(); |
| 310 | if (guest_limit > guest_max) |
| 311 | errx(1, "Not enough memory for devices"); |
| 312 | return addr; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 313 | } |
| 314 | |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 315 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
| 316 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
| 317 | * it falls back to reading the memory in. */ |
| 318 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
| 319 | { |
| 320 | ssize_t r; |
| 321 | |
| 322 | /* We map writable even though for some segments are marked read-only. |
| 323 | * The kernel really wants to be writable: it patches its own |
| 324 | * instructions. |
| 325 | * |
| 326 | * MAP_PRIVATE means that the page won't be copied until a write is |
| 327 | * done to it. This allows us to share untouched memory between |
| 328 | * Guests. */ |
| 329 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, |
| 330 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) |
| 331 | return; |
| 332 | |
| 333 | /* pread does a seek and a read in one shot: saves a few lines. */ |
| 334 | r = pread(fd, addr, len, offset); |
| 335 | if (r != len) |
| 336 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); |
| 337 | } |
| 338 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 339 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
| 340 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
| 341 | * by all modern binaries on Linux including the kernel. |
| 342 | * |
| 343 | * The ELF headers give *two* addresses: a physical address, and a virtual |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 344 | * address. We use the physical address; the Guest will map itself to the |
| 345 | * virtual address. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 346 | * |
| 347 | * We return the starting address. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 348 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 349 | { |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 350 | Elf32_Phdr phdr[ehdr->e_phnum]; |
| 351 | unsigned int i; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 352 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 353 | /* Sanity checks on the main ELF header: an x86 executable with a |
| 354 | * reasonable number of correctly-sized program headers. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 355 | if (ehdr->e_type != ET_EXEC |
| 356 | || ehdr->e_machine != EM_386 |
| 357 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) |
| 358 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) |
| 359 | errx(1, "Malformed elf header"); |
| 360 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 361 | /* An ELF executable contains an ELF header and a number of "program" |
| 362 | * headers which indicate which parts ("segments") of the program to |
| 363 | * load where. */ |
| 364 | |
| 365 | /* We read in all the program headers at once: */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 366 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
| 367 | err(1, "Seeking to program headers"); |
| 368 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) |
| 369 | err(1, "Reading program headers"); |
| 370 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 371 | /* Try all the headers: there are usually only three. A read-only one, |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 372 | * a read-write one, and a "note" section which we don't load. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 373 | for (i = 0; i < ehdr->e_phnum; i++) { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 374 | /* If this isn't a loadable segment, we ignore it */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 375 | if (phdr[i].p_type != PT_LOAD) |
| 376 | continue; |
| 377 | |
| 378 | verbose("Section %i: size %i addr %p\n", |
| 379 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); |
| 380 | |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 381 | /* We map this section of the file at its physical address. */ |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 382 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 383 | phdr[i].p_offset, phdr[i].p_filesz); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 384 | } |
| 385 | |
Rusty Russell | 814a0e5 | 2007-10-22 11:29:44 +1000 | [diff] [blame] | 386 | /* The entry point is given in the ELF header. */ |
| 387 | return ehdr->e_entry; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 388 | } |
| 389 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 390 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 391 | * supposed to jump into it and it will unpack itself. We used to have to |
| 392 | * perform some hairy magic because the unpacking code scared me. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 393 | * |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 394 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
| 395 | * a small patch to jump over the tricky bits in the Guest, so now we just read |
| 396 | * the funky header so we know where in the file to load, and away we go! */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 397 | static unsigned long load_bzimage(int fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 398 | { |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 399 | struct boot_params boot; |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 400 | int r; |
| 401 | /* Modern bzImages get loaded at 1M. */ |
| 402 | void *p = from_guest_phys(0x100000); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 403 | |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 404 | /* Go back to the start of the file and read the header. It should be |
Uwe Hermann | 71cced6 | 2008-10-20 09:32:21 -0700 | [diff] [blame] | 405 | * a Linux boot header (see Documentation/x86/i386/boot.txt) */ |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 406 | lseek(fd, 0, SEEK_SET); |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 407 | read(fd, &boot, sizeof(boot)); |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 408 | |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 409 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
| 410 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 411 | errx(1, "This doesn't look like a bzImage to me"); |
| 412 | |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 413 | /* Skip over the extra sectors of the header. */ |
| 414 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); |
Rusty Russell | 5bbf89f | 2007-10-22 11:29:56 +1000 | [diff] [blame] | 415 | |
| 416 | /* Now read everything into memory. in nice big chunks. */ |
| 417 | while ((r = read(fd, p, 65536)) > 0) |
| 418 | p += r; |
| 419 | |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 420 | /* Finally, code32_start tells us where to enter the kernel. */ |
| 421 | return boot.hdr.code32_start; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 422 | } |
| 423 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 424 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 425 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
| 426 | * work, we can load those, too. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 427 | static unsigned long load_kernel(int fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 428 | { |
| 429 | Elf32_Ehdr hdr; |
| 430 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 431 | /* Read in the first few bytes. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 432 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
| 433 | err(1, "Reading kernel"); |
| 434 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 435 | /* If it's an ELF file, it starts with "\177ELF" */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 436 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 437 | return map_elf(fd, &hdr); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 438 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 439 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 440 | return load_bzimage(fd); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 441 | } |
| 442 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 443 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
| 444 | * it calls getpagesize() twice: "it's dumb code." |
| 445 | * |
| 446 | * Kernel guys get really het up about optimization, even when it's not |
| 447 | * necessary. I leave this code as a reaction against that. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 448 | static inline unsigned long page_align(unsigned long addr) |
| 449 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 450 | /* Add upwards and truncate downwards. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 451 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
| 452 | } |
| 453 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 454 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
| 455 | * the kernel which the kernel can use to boot from without needing any |
| 456 | * drivers. Most distributions now use this as standard: the initrd contains |
| 457 | * the code to load the appropriate driver modules for the current machine. |
| 458 | * |
| 459 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its |
| 460 | * kernels. He sent me this (and tells me when I break it). */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 461 | static unsigned long load_initrd(const char *name, unsigned long mem) |
| 462 | { |
| 463 | int ifd; |
| 464 | struct stat st; |
| 465 | unsigned long len; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 466 | |
| 467 | ifd = open_or_die(name, O_RDONLY); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 468 | /* fstat() is needed to get the file size. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 469 | if (fstat(ifd, &st) < 0) |
| 470 | err(1, "fstat() on initrd '%s'", name); |
| 471 | |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 472 | /* We map the initrd at the top of memory, but mmap wants it to be |
| 473 | * page-aligned, so we round the size up for that. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 474 | len = page_align(st.st_size); |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 475 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 476 | /* Once a file is mapped, you can close the file descriptor. It's a |
| 477 | * little odd, but quite useful. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 478 | close(ifd); |
Ronald G. Minnich | 6649bb7 | 2007-08-28 14:35:59 -0700 | [diff] [blame] | 479 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 480 | |
| 481 | /* We return the initrd size. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 482 | return len; |
| 483 | } |
| 484 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 485 | /* Once we know how much memory we have we can construct simple linear page |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 486 | * tables which set virtual == physical which will get the Guest far enough |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 487 | * into the boot to create its own. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 488 | * |
| 489 | * We lay them out of the way, just below the initrd (which is why we need to |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 490 | * know its size here). */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 491 | static unsigned long setup_pagetables(unsigned long mem, |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 492 | unsigned long initrd_size) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 493 | { |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 494 | unsigned long *pgdir, *linear; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 495 | unsigned int mapped_pages, i, linear_pages; |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 496 | unsigned int ptes_per_page = getpagesize()/sizeof(void *); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 497 | |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 498 | mapped_pages = mem/getpagesize(); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 499 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 500 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 501 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
| 502 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 503 | /* We put the toplevel page directory page at the top of memory. */ |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 504 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 505 | |
| 506 | /* Now we use the next linear_pages pages as pte pages */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 507 | linear = (void *)pgdir - linear_pages*getpagesize(); |
| 508 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 509 | /* Linear mapping is easy: put every page's address into the mapping in |
| 510 | * order. PAGE_PRESENT contains the flags Present, Writable and |
| 511 | * Executable. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 512 | for (i = 0; i < mapped_pages; i++) |
| 513 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); |
| 514 | |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 515 | /* The top level points to the linear page table pages above. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 516 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 517 | pgdir[i/ptes_per_page] |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 518 | = ((to_guest_phys(linear) + i*sizeof(void *)) |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 519 | | PAGE_PRESENT); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 520 | } |
| 521 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 522 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
| 523 | mapped_pages, linear_pages, to_guest_phys(linear)); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 524 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 525 | /* We return the top level (guest-physical) address: the kernel needs |
| 526 | * to know where it is. */ |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 527 | return to_guest_phys(pgdir); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 528 | } |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 529 | /*:*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 530 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 531 | /* Simple routine to roll all the commandline arguments together with spaces |
| 532 | * between them. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 533 | static void concat(char *dst, char *args[]) |
| 534 | { |
| 535 | unsigned int i, len = 0; |
| 536 | |
| 537 | for (i = 0; args[i]; i++) { |
Paul Bolle | 1ef36fa | 2008-03-10 16:39:03 +0100 | [diff] [blame] | 538 | if (i) { |
| 539 | strcat(dst+len, " "); |
| 540 | len++; |
| 541 | } |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 542 | strcpy(dst+len, args[i]); |
Paul Bolle | 1ef36fa | 2008-03-10 16:39:03 +0100 | [diff] [blame] | 543 | len += strlen(args[i]); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 544 | } |
| 545 | /* In case it's empty. */ |
| 546 | dst[len] = '\0'; |
| 547 | } |
| 548 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 549 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
| 550 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
| 551 | * the base of Guest "physical" memory, the top physical page to allow, the |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 552 | * top level pagetable and the entry point for the Guest. */ |
| 553 | static int tell_kernel(unsigned long pgdir, unsigned long start) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 554 | { |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 555 | unsigned long args[] = { LHREQ_INITIALIZE, |
| 556 | (unsigned long)guest_base, |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 557 | guest_limit / getpagesize(), pgdir, start }; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 558 | int fd; |
| 559 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 560 | verbose("Guest: %p - %p (%#lx)\n", |
| 561 | guest_base, guest_base + guest_limit, guest_limit); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 562 | fd = open_or_die("/dev/lguest", O_RDWR); |
| 563 | if (write(fd, args, sizeof(args)) < 0) |
| 564 | err(1, "Writing to /dev/lguest"); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 565 | |
| 566 | /* We return the /dev/lguest file descriptor to control this Guest */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 567 | return fd; |
| 568 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 569 | /*:*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 570 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 571 | static void add_device_fd(int fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 572 | { |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 573 | FD_SET(fd, &devices.infds); |
| 574 | if (fd > devices.max_infd) |
| 575 | devices.max_infd = fd; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 576 | } |
| 577 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 578 | /*L:200 |
| 579 | * The Waker. |
| 580 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 581 | * With console, block and network devices, we can have lots of input which we |
| 582 | * need to process. We could try to tell the kernel what file descriptors to |
| 583 | * watch, but handing a file descriptor mask through to the kernel is fairly |
| 584 | * icky. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 585 | * |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 586 | * Instead, we clone off a thread which watches the file descriptors and writes |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 587 | * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host |
| 588 | * stop running the Guest. This causes the Launcher to return from the |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 589 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset |
| 590 | * the LHREQ_BREAK and wake us up again. |
| 591 | * |
| 592 | * This, of course, is merely a different *kind* of icky. |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 593 | * |
| 594 | * Given my well-known antipathy to threads, I'd prefer to use processes. But |
| 595 | * it's easier to share Guest memory with threads, and trivial to share the |
| 596 | * devices.infds as the Launcher changes it. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 597 | */ |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 598 | static int waker(void *unused) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 599 | { |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 600 | /* Close the write end of the pipe: only the Launcher has it open. */ |
| 601 | close(waker_fds.pipe[1]); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 602 | |
| 603 | for (;;) { |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 604 | fd_set rfds = devices.infds; |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 605 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 606 | unsigned int maxfd = devices.max_infd; |
| 607 | |
| 608 | /* We also listen to the pipe from the Launcher. */ |
| 609 | FD_SET(waker_fds.pipe[0], &rfds); |
| 610 | if (waker_fds.pipe[0] > maxfd) |
| 611 | maxfd = waker_fds.pipe[0]; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 612 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 613 | /* Wait until input is ready from one of the devices. */ |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 614 | select(maxfd+1, &rfds, NULL, NULL, NULL); |
| 615 | |
| 616 | /* Message from Launcher? */ |
| 617 | if (FD_ISSET(waker_fds.pipe[0], &rfds)) { |
| 618 | char c; |
| 619 | /* If this fails, then assume Launcher has exited. |
| 620 | * Don't do anything on exit: we're just a thread! */ |
| 621 | if (read(waker_fds.pipe[0], &c, 1) != 1) |
| 622 | _exit(0); |
| 623 | continue; |
| 624 | } |
| 625 | |
| 626 | /* Send LHREQ_BREAK command to snap the Launcher out of it. */ |
| 627 | pwrite(waker_fds.lguest_fd, args, sizeof(args), cpu_id); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 628 | } |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 629 | return 0; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 630 | } |
| 631 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 632 | /* This routine just sets up a pipe to the Waker process. */ |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 633 | static void setup_waker(int lguest_fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 634 | { |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 635 | /* This pipe is closed when Launcher dies, telling Waker. */ |
| 636 | if (pipe(waker_fds.pipe) != 0) |
| 637 | err(1, "Creating pipe for Waker"); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 638 | |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 639 | /* Waker also needs to know the lguest fd */ |
| 640 | waker_fds.lguest_fd = lguest_fd; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 641 | |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 642 | if (clone(waker, malloc(4096) + 4096, CLONE_VM | SIGCHLD, NULL) == -1) |
| 643 | err(1, "Creating Waker"); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 644 | } |
| 645 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 646 | /* |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 647 | * Device Handling. |
| 648 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 649 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 650 | * We need to make sure it's not trying to reach into the Launcher itself, so |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 651 | * we have a convenient routine which checks it and exits with an error message |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 652 | * if something funny is going on: |
| 653 | */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 654 | static void *_check_pointer(unsigned long addr, unsigned int size, |
| 655 | unsigned int line) |
| 656 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 657 | /* We have to separately check addr and addr+size, because size could |
| 658 | * be huge and addr + size might wrap around. */ |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 659 | if (addr >= guest_limit || addr + size >= guest_limit) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 660 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 661 | /* We return a pointer for the caller's convenience, now we know it's |
| 662 | * safe to use. */ |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 663 | return from_guest_phys(addr); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 664 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 665 | /* A macro which transparently hands the line number to the real function. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 666 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
| 667 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 668 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
| 669 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
| 670 | * at the end. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 671 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) |
| 672 | { |
| 673 | unsigned int next; |
| 674 | |
| 675 | /* If this descriptor says it doesn't chain, we're done. */ |
| 676 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) |
| 677 | return vq->vring.num; |
| 678 | |
| 679 | /* Check they're not leading us off end of descriptors. */ |
| 680 | next = vq->vring.desc[i].next; |
| 681 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
| 682 | wmb(); |
| 683 | |
| 684 | if (next >= vq->vring.num) |
| 685 | errx(1, "Desc next is %u", next); |
| 686 | |
| 687 | return next; |
| 688 | } |
| 689 | |
| 690 | /* This looks in the virtqueue and for the first available buffer, and converts |
| 691 | * it to an iovec for convenient access. Since descriptors consist of some |
| 692 | * number of output then some number of input descriptors, it's actually two |
| 693 | * iovecs, but we pack them into one and note how many of each there were. |
| 694 | * |
| 695 | * This function returns the descriptor number found, or vq->vring.num (which |
| 696 | * is never a valid descriptor number) if none was found. */ |
| 697 | static unsigned get_vq_desc(struct virtqueue *vq, |
| 698 | struct iovec iov[], |
| 699 | unsigned int *out_num, unsigned int *in_num) |
| 700 | { |
| 701 | unsigned int i, head; |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 702 | u16 last_avail; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 703 | |
| 704 | /* Check it isn't doing very strange things with descriptor numbers. */ |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 705 | last_avail = lg_last_avail(vq); |
| 706 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 707 | errx(1, "Guest moved used index from %u to %u", |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 708 | last_avail, vq->vring.avail->idx); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 709 | |
| 710 | /* If there's nothing new since last we looked, return invalid. */ |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 711 | if (vq->vring.avail->idx == last_avail) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 712 | return vq->vring.num; |
| 713 | |
| 714 | /* Grab the next descriptor number they're advertising, and increment |
| 715 | * the index we've seen. */ |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 716 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
| 717 | lg_last_avail(vq)++; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 718 | |
| 719 | /* If their number is silly, that's a fatal mistake. */ |
| 720 | if (head >= vq->vring.num) |
| 721 | errx(1, "Guest says index %u is available", head); |
| 722 | |
| 723 | /* When we start there are none of either input nor output. */ |
| 724 | *out_num = *in_num = 0; |
| 725 | |
| 726 | i = head; |
| 727 | do { |
| 728 | /* Grab the first descriptor, and check it's OK. */ |
| 729 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; |
| 730 | iov[*out_num + *in_num].iov_base |
| 731 | = check_pointer(vq->vring.desc[i].addr, |
| 732 | vq->vring.desc[i].len); |
| 733 | /* If this is an input descriptor, increment that count. */ |
| 734 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) |
| 735 | (*in_num)++; |
| 736 | else { |
| 737 | /* If it's an output descriptor, they're all supposed |
| 738 | * to come before any input descriptors. */ |
| 739 | if (*in_num) |
| 740 | errx(1, "Descriptor has out after in"); |
| 741 | (*out_num)++; |
| 742 | } |
| 743 | |
| 744 | /* If we've got too many, that implies a descriptor loop. */ |
| 745 | if (*out_num + *in_num > vq->vring.num) |
| 746 | errx(1, "Looped descriptor"); |
| 747 | } while ((i = next_desc(vq, i)) != vq->vring.num); |
| 748 | |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 749 | vq->inflight++; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 750 | return head; |
| 751 | } |
| 752 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 753 | /* After we've used one of their buffers, we tell them about it. We'll then |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 754 | * want to send them an interrupt, using trigger_irq(). */ |
| 755 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
| 756 | { |
| 757 | struct vring_used_elem *used; |
| 758 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 759 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
| 760 | * next entry in that used ring. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 761 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
| 762 | used->id = head; |
| 763 | used->len = len; |
| 764 | /* Make sure buffer is written before we update index. */ |
| 765 | wmb(); |
| 766 | vq->vring.used->idx++; |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 767 | vq->inflight--; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 768 | } |
| 769 | |
| 770 | /* This actually sends the interrupt for this virtqueue */ |
| 771 | static void trigger_irq(int fd, struct virtqueue *vq) |
| 772 | { |
| 773 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
| 774 | |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 775 | /* If they don't want an interrupt, don't send one, unless empty. */ |
| 776 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) |
| 777 | && vq->inflight) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 778 | return; |
| 779 | |
| 780 | /* Send the Guest an interrupt tell them we used something up. */ |
| 781 | if (write(fd, buf, sizeof(buf)) != 0) |
| 782 | err(1, "Triggering irq %i", vq->config.irq); |
| 783 | } |
| 784 | |
| 785 | /* And here's the combo meal deal. Supersize me! */ |
| 786 | static void add_used_and_trigger(int fd, struct virtqueue *vq, |
| 787 | unsigned int head, int len) |
| 788 | { |
| 789 | add_used(vq, head, len); |
| 790 | trigger_irq(fd, vq); |
| 791 | } |
| 792 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 793 | /* |
| 794 | * The Console |
| 795 | * |
| 796 | * Here is the input terminal setting we save, and the routine to restore them |
| 797 | * on exit so the user gets their terminal back. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 798 | static struct termios orig_term; |
| 799 | static void restore_term(void) |
| 800 | { |
| 801 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); |
| 802 | } |
| 803 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 804 | /* We associate some data with the console for our exit hack. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 805 | struct console_abort |
| 806 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 807 | /* How many times have they hit ^C? */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 808 | int count; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 809 | /* When did they start? */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 810 | struct timeval start; |
| 811 | }; |
| 812 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 813 | /* This is the routine which handles console input (ie. stdin). */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 814 | static bool handle_console_input(int fd, struct device *dev) |
| 815 | { |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 816 | int len; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 817 | unsigned int head, in_num, out_num; |
| 818 | struct iovec iov[dev->vq->vring.num]; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 819 | struct console_abort *abort = dev->priv; |
| 820 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 821 | /* First we need a console buffer from the Guests's input virtqueue. */ |
| 822 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 823 | |
| 824 | /* If they're not ready for input, stop listening to this file |
| 825 | * descriptor. We'll start again once they add an input buffer. */ |
| 826 | if (head == dev->vq->vring.num) |
| 827 | return false; |
| 828 | |
| 829 | if (out_num) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 830 | errx(1, "Output buffers in console in queue?"); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 831 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 832 | /* This is why we convert to iovecs: the readv() call uses them, and so |
| 833 | * it reads straight into the Guest's buffer. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 834 | len = readv(dev->fd, iov, in_num); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 835 | if (len <= 0) { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 836 | /* This implies that the console is closed, is /dev/null, or |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 837 | * something went terribly wrong. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 838 | warnx("Failed to get console input, ignoring console."); |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 839 | /* Put the input terminal back. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 840 | restore_term(); |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 841 | /* Remove callback from input vq, so it doesn't restart us. */ |
| 842 | dev->vq->handle_output = NULL; |
| 843 | /* Stop listening to this fd: don't call us again. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 844 | return false; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 845 | } |
| 846 | |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 847 | /* Tell the Guest about the new input. */ |
| 848 | add_used_and_trigger(fd, dev->vq, head, len); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 849 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 850 | /* Three ^C within one second? Exit. |
| 851 | * |
| 852 | * This is such a hack, but works surprisingly well. Each ^C has to be |
| 853 | * in a buffer by itself, so they can't be too fast. But we check that |
| 854 | * we get three within about a second, so they can't be too slow. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 855 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
| 856 | if (!abort->count++) |
| 857 | gettimeofday(&abort->start, NULL); |
| 858 | else if (abort->count == 3) { |
| 859 | struct timeval now; |
| 860 | gettimeofday(&now, NULL); |
| 861 | if (now.tv_sec <= abort->start.tv_sec+1) { |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 862 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 863 | /* Close the fd so Waker will know it has to |
| 864 | * exit. */ |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 865 | close(waker_fds.pipe[1]); |
| 866 | /* Just in case Waker is blocked in BREAK, send |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 867 | * unbreak now. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 868 | write(fd, args, sizeof(args)); |
| 869 | exit(2); |
| 870 | } |
| 871 | abort->count = 0; |
| 872 | } |
| 873 | } else |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 874 | /* Any other key resets the abort counter. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 875 | abort->count = 0; |
| 876 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 877 | /* Everything went OK! */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 878 | return true; |
| 879 | } |
| 880 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 881 | /* Handling output for console is simple: we just get all the output buffers |
| 882 | * and write them to stdout. */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 883 | static void handle_console_output(int fd, struct virtqueue *vq, bool timeout) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 884 | { |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 885 | unsigned int head, out, in; |
| 886 | int len; |
| 887 | struct iovec iov[vq->vring.num]; |
| 888 | |
| 889 | /* Keep getting output buffers from the Guest until we run out. */ |
| 890 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { |
| 891 | if (in) |
| 892 | errx(1, "Input buffers in output queue?"); |
| 893 | len = writev(STDOUT_FILENO, iov, out); |
| 894 | add_used_and_trigger(fd, vq, head, len); |
| 895 | } |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 896 | } |
| 897 | |
Rusty Russell | 1dc3e3b | 2008-08-26 00:19:27 -0500 | [diff] [blame] | 898 | /* This is called when we no longer want to hear about Guest changes to a |
| 899 | * virtqueue. This is more efficient in high-traffic cases, but it means we |
| 900 | * have to set a timer to check if any more changes have occurred. */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 901 | static void block_vq(struct virtqueue *vq) |
| 902 | { |
| 903 | struct itimerval itm; |
| 904 | |
| 905 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; |
| 906 | vq->blocked = true; |
| 907 | |
| 908 | itm.it_interval.tv_sec = 0; |
| 909 | itm.it_interval.tv_usec = 0; |
| 910 | itm.it_value.tv_sec = 0; |
Rusty Russell | aa12498 | 2008-07-29 09:58:36 -0500 | [diff] [blame] | 911 | itm.it_value.tv_usec = timeout_usec; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 912 | |
| 913 | setitimer(ITIMER_REAL, &itm, NULL); |
| 914 | } |
| 915 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 916 | /* |
| 917 | * The Network |
| 918 | * |
| 919 | * Handling output for network is also simple: we get all the output buffers |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 920 | * and write them (ignoring the first element) to this device's file descriptor |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 921 | * (/dev/net/tun). |
| 922 | */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 923 | static void handle_net_output(int fd, struct virtqueue *vq, bool timeout) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 924 | { |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 925 | unsigned int head, out, in, num = 0; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 926 | int len; |
| 927 | struct iovec iov[vq->vring.num]; |
Rusty Russell | aa12498 | 2008-07-29 09:58:36 -0500 | [diff] [blame] | 928 | static int last_timeout_num; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 929 | |
| 930 | /* Keep getting output buffers from the Guest until we run out. */ |
| 931 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { |
| 932 | if (in) |
| 933 | errx(1, "Input buffers in output queue?"); |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 934 | len = writev(vq->dev->fd, iov, out); |
| 935 | if (len < 0) |
| 936 | err(1, "Writing network packet to tun"); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 937 | add_used_and_trigger(fd, vq, head, len); |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 938 | num++; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 939 | } |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 940 | |
| 941 | /* Block further kicks and set up a timer if we saw anything. */ |
| 942 | if (!timeout && num) |
| 943 | block_vq(vq); |
Rusty Russell | aa12498 | 2008-07-29 09:58:36 -0500 | [diff] [blame] | 944 | |
Rusty Russell | 1dc3e3b | 2008-08-26 00:19:27 -0500 | [diff] [blame] | 945 | /* We never quite know how long should we wait before we check the |
| 946 | * queue again for more packets. We start at 500 microseconds, and if |
| 947 | * we get fewer packets than last time, we assume we made the timeout |
| 948 | * too small and increase it by 10 microseconds. Otherwise, we drop it |
| 949 | * by one microsecond every time. It seems to work well enough. */ |
Rusty Russell | aa12498 | 2008-07-29 09:58:36 -0500 | [diff] [blame] | 950 | if (timeout) { |
| 951 | if (num < last_timeout_num) |
| 952 | timeout_usec += 10; |
| 953 | else if (timeout_usec > 1) |
| 954 | timeout_usec--; |
| 955 | last_timeout_num = num; |
| 956 | } |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 957 | } |
| 958 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 959 | /* This is where we handle a packet coming in from the tun device to our |
| 960 | * Guest. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 961 | static bool handle_tun_input(int fd, struct device *dev) |
| 962 | { |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 963 | unsigned int head, in_num, out_num; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 964 | int len; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 965 | struct iovec iov[dev->vq->vring.num]; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 966 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 967 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
| 968 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
| 969 | if (head == dev->vq->vring.num) { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 970 | /* Now, it's expected that if we try to send a packet too |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 971 | * early, the Guest won't be ready yet. Wait until the device |
| 972 | * status says it's ready. */ |
| 973 | /* FIXME: Actually want DRIVER_ACTIVE here. */ |
Rusty Russell | 5dae785 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 974 | |
| 975 | /* Now tell it we want to know if new things appear. */ |
| 976 | dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; |
| 977 | wmb(); |
| 978 | |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 979 | /* We'll turn this back on if input buffers are registered. */ |
| 980 | return false; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 981 | } else if (out_num) |
| 982 | errx(1, "Output buffers in network recv queue?"); |
| 983 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 984 | /* Read the packet from the device directly into the Guest's buffer. */ |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 985 | len = readv(dev->fd, iov, in_num); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 986 | if (len <= 0) |
| 987 | err(1, "reading network"); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 988 | |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 989 | /* Tell the Guest about the new packet. */ |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 990 | add_used_and_trigger(fd, dev->vq, head, len); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 991 | |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 992 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 993 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
| 994 | head != dev->vq->vring.num ? "sent" : "discarded"); |
| 995 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 996 | /* All good. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 997 | return true; |
| 998 | } |
| 999 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1000 | /*L:215 This is the callback attached to the network and console input |
| 1001 | * virtqueues: it ensures we try again, in case we stopped console or net |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1002 | * delivery because Guest didn't have any buffers. */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1003 | static void enable_fd(int fd, struct virtqueue *vq, bool timeout) |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1004 | { |
| 1005 | add_device_fd(vq->dev->fd); |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 1006 | /* Snap the Waker out of its select loop. */ |
| 1007 | write(waker_fds.pipe[1], "", 1); |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1008 | } |
| 1009 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1010 | static void net_enable_fd(int fd, struct virtqueue *vq, bool timeout) |
Rusty Russell | 5dae785 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1011 | { |
| 1012 | /* We don't need to know again when Guest refills receive buffer. */ |
| 1013 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1014 | enable_fd(fd, vq, timeout); |
Rusty Russell | 5dae785 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1015 | } |
| 1016 | |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1017 | /* When the Guest tells us they updated the status field, we handle it. */ |
| 1018 | static void update_device_status(struct device *dev) |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1019 | { |
| 1020 | struct virtqueue *vq; |
| 1021 | |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1022 | /* This is a reset. */ |
| 1023 | if (dev->desc->status == 0) { |
| 1024 | verbose("Resetting device %s\n", dev->name); |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1025 | |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1026 | /* Clear any features they've acked. */ |
| 1027 | memset(get_feature_bits(dev) + dev->desc->feature_len, 0, |
| 1028 | dev->desc->feature_len); |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1029 | |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1030 | /* Zero out the virtqueues. */ |
| 1031 | for (vq = dev->vq; vq; vq = vq->next) { |
| 1032 | memset(vq->vring.desc, 0, |
| 1033 | vring_size(vq->config.num, getpagesize())); |
Rusty Russell | b511179 | 2008-07-29 09:58:34 -0500 | [diff] [blame] | 1034 | lg_last_avail(vq) = 0; |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1035 | } |
| 1036 | } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { |
| 1037 | warnx("Device %s configuration FAILED", dev->name); |
| 1038 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { |
| 1039 | unsigned int i; |
| 1040 | |
| 1041 | verbose("Device %s OK: offered", dev->name); |
| 1042 | for (i = 0; i < dev->desc->feature_len; i++) |
Rusty Russell | 32c68e5 | 2008-07-29 09:58:32 -0500 | [diff] [blame] | 1043 | verbose(" %02x", get_feature_bits(dev)[i]); |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1044 | verbose(", accepted"); |
| 1045 | for (i = 0; i < dev->desc->feature_len; i++) |
Rusty Russell | 32c68e5 | 2008-07-29 09:58:32 -0500 | [diff] [blame] | 1046 | verbose(" %02x", get_feature_bits(dev) |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1047 | [dev->desc->feature_len+i]); |
| 1048 | |
| 1049 | if (dev->ready) |
| 1050 | dev->ready(dev); |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1051 | } |
| 1052 | } |
| 1053 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1054 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
| 1055 | static void handle_output(int fd, unsigned long addr) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1056 | { |
| 1057 | struct device *i; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1058 | struct virtqueue *vq; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1059 | |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1060 | /* Check each device and virtqueue. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1061 | for (i = devices.dev; i; i = i->next) { |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1062 | /* Notifications to device descriptors update device status. */ |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1063 | if (from_guest_phys(addr) == i->desc) { |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1064 | update_device_status(i); |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1065 | return; |
| 1066 | } |
| 1067 | |
| 1068 | /* Notifications to virtqueues mean output has occurred. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1069 | for (vq = i->vq; vq; vq = vq->next) { |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1070 | if (vq->config.pfn != addr/getpagesize()) |
| 1071 | continue; |
| 1072 | |
| 1073 | /* Guest should acknowledge (and set features!) before |
| 1074 | * using the device. */ |
| 1075 | if (i->desc->status == 0) { |
| 1076 | warnx("%s gave early output", i->name); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1077 | return; |
| 1078 | } |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1079 | |
| 1080 | if (strcmp(vq->dev->name, "console") != 0) |
| 1081 | verbose("Output to %s\n", vq->dev->name); |
| 1082 | if (vq->handle_output) |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1083 | vq->handle_output(fd, vq, false); |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1084 | return; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1085 | } |
| 1086 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1087 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1088 | /* Early console write is done using notify on a nul-terminated string |
| 1089 | * in Guest memory. */ |
| 1090 | if (addr >= guest_limit) |
| 1091 | errx(1, "Bad NOTIFY %#lx", addr); |
| 1092 | |
| 1093 | write(STDOUT_FILENO, from_guest_phys(addr), |
| 1094 | strnlen(from_guest_phys(addr), guest_limit - addr)); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1095 | } |
| 1096 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1097 | static void handle_timeout(int fd) |
| 1098 | { |
| 1099 | char buf[32]; |
| 1100 | struct device *i; |
| 1101 | struct virtqueue *vq; |
| 1102 | |
| 1103 | /* Clear the pipe */ |
| 1104 | read(timeoutpipe[0], buf, sizeof(buf)); |
| 1105 | |
| 1106 | /* Check each device and virtqueue: flush blocked ones. */ |
| 1107 | for (i = devices.dev; i; i = i->next) { |
| 1108 | for (vq = i->vq; vq; vq = vq->next) { |
| 1109 | if (!vq->blocked) |
| 1110 | continue; |
| 1111 | |
| 1112 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; |
| 1113 | vq->blocked = false; |
| 1114 | if (vq->handle_output) |
| 1115 | vq->handle_output(fd, vq, true); |
| 1116 | } |
| 1117 | } |
| 1118 | } |
| 1119 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1120 | /* This is called when the Waker wakes us up: check for incoming file |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1121 | * descriptors. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1122 | static void handle_input(int fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1123 | { |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1124 | /* select() wants a zeroed timeval to mean "don't wait". */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1125 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
| 1126 | |
| 1127 | for (;;) { |
| 1128 | struct device *i; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1129 | fd_set fds = devices.infds; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1130 | int num; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1131 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1132 | num = select(devices.max_infd+1, &fds, NULL, NULL, &poll); |
| 1133 | /* Could get interrupted */ |
| 1134 | if (num < 0) |
| 1135 | continue; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1136 | /* If nothing is ready, we're done. */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1137 | if (num == 0) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1138 | break; |
| 1139 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1140 | /* Otherwise, call the device(s) which have readable file |
| 1141 | * descriptors and a method of handling them. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1142 | for (i = devices.dev; i; i = i->next) { |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1143 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1144 | if (i->handle_input(fd, i)) |
| 1145 | continue; |
| 1146 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1147 | /* If handle_input() returns false, it means we |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1148 | * should no longer service it. Networking and |
| 1149 | * console do this when there's no input |
| 1150 | * buffers to deliver into. Console also uses |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1151 | * it when it discovers that stdin is closed. */ |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1152 | FD_CLR(i->fd, &devices.infds); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1153 | } |
| 1154 | } |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1155 | |
| 1156 | /* Is this the timeout fd? */ |
| 1157 | if (FD_ISSET(timeoutpipe[0], &fds)) |
| 1158 | handle_timeout(fd); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1159 | } |
| 1160 | } |
| 1161 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1162 | /*L:190 |
| 1163 | * Device Setup |
| 1164 | * |
| 1165 | * All devices need a descriptor so the Guest knows it exists, and a "struct |
| 1166 | * device" so the Launcher can keep track of it. We have common helper |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1167 | * routines to allocate and manage them. |
| 1168 | */ |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1169 | |
| 1170 | /* The layout of the device page is a "struct lguest_device_desc" followed by a |
| 1171 | * number of virtqueue descriptors, then two sets of feature bits, then an |
| 1172 | * array of configuration bytes. This routine returns the configuration |
| 1173 | * pointer. */ |
| 1174 | static u8 *device_config(const struct device *dev) |
| 1175 | { |
| 1176 | return (void *)(dev->desc + 1) |
| 1177 | + dev->desc->num_vq * sizeof(struct lguest_vqconfig) |
| 1178 | + dev->desc->feature_len * 2; |
| 1179 | } |
| 1180 | |
| 1181 | /* This routine allocates a new "struct lguest_device_desc" from descriptor |
| 1182 | * table page just above the Guest's normal memory. It returns a pointer to |
| 1183 | * that descriptor. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1184 | static struct lguest_device_desc *new_dev_desc(u16 type) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1185 | { |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1186 | struct lguest_device_desc d = { .type = type }; |
| 1187 | void *p; |
| 1188 | |
| 1189 | /* Figure out where the next device config is, based on the last one. */ |
| 1190 | if (devices.lastdev) |
| 1191 | p = device_config(devices.lastdev) |
| 1192 | + devices.lastdev->desc->config_len; |
| 1193 | else |
| 1194 | p = devices.descpage; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1195 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1196 | /* We only have one page for all the descriptors. */ |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1197 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1198 | errx(1, "Too many devices"); |
| 1199 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1200 | /* p might not be aligned, so we memcpy in. */ |
| 1201 | return memcpy(p, &d, sizeof(d)); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1202 | } |
| 1203 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1204 | /* Each device descriptor is followed by the description of its virtqueues. We |
| 1205 | * specify how many descriptors the virtqueue is to have. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1206 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1207 | void (*handle_output)(int, struct virtqueue *, bool)) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1208 | { |
| 1209 | unsigned int pages; |
| 1210 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); |
| 1211 | void *p; |
| 1212 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1213 | /* First we need some memory for this virtqueue. */ |
Rusty Russell | 42b36cc | 2007-11-12 13:39:18 +1100 | [diff] [blame] | 1214 | pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1) |
| 1215 | / getpagesize(); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1216 | p = get_pages(pages); |
| 1217 | |
Rusty Russell | d1c856e | 2007-11-19 11:20:40 -0500 | [diff] [blame] | 1218 | /* Initialize the virtqueue */ |
| 1219 | vq->next = NULL; |
| 1220 | vq->last_avail_idx = 0; |
| 1221 | vq->dev = dev; |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 1222 | vq->inflight = 0; |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1223 | vq->blocked = false; |
Rusty Russell | d1c856e | 2007-11-19 11:20:40 -0500 | [diff] [blame] | 1224 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1225 | /* Initialize the configuration. */ |
| 1226 | vq->config.num = num_descs; |
| 1227 | vq->config.irq = devices.next_irq++; |
| 1228 | vq->config.pfn = to_guest_phys(p) / getpagesize(); |
| 1229 | |
| 1230 | /* Initialize the vring. */ |
Rusty Russell | 42b36cc | 2007-11-12 13:39:18 +1100 | [diff] [blame] | 1231 | vring_init(&vq->vring, num_descs, p, getpagesize()); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1232 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1233 | /* Append virtqueue to this device's descriptor. We use |
| 1234 | * device_config() to get the end of the device's current virtqueues; |
| 1235 | * we check that we haven't added any config or feature information |
| 1236 | * yet, otherwise we'd be overwriting them. */ |
| 1237 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); |
| 1238 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); |
| 1239 | dev->desc->num_vq++; |
| 1240 | |
| 1241 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1242 | |
| 1243 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is |
| 1244 | * second. */ |
| 1245 | for (i = &dev->vq; *i; i = &(*i)->next); |
| 1246 | *i = vq; |
| 1247 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1248 | /* Set the routine to call when the Guest does something to this |
| 1249 | * virtqueue. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1250 | vq->handle_output = handle_output; |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1251 | |
Rusty Russell | 426e3e0 | 2008-02-04 23:49:59 -0500 | [diff] [blame] | 1252 | /* As an optimization, set the advisory "Don't Notify Me" flag if we |
| 1253 | * don't have a handler */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1254 | if (!handle_output) |
| 1255 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; |
| 1256 | } |
| 1257 | |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1258 | /* The first half of the feature bitmask is for us to advertise features. The |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1259 | * second half is for the Guest to accept features. */ |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1260 | static void add_feature(struct device *dev, unsigned bit) |
| 1261 | { |
Rusty Russell | 6e5aa7e | 2008-02-04 23:50:03 -0500 | [diff] [blame] | 1262 | u8 *features = get_feature_bits(dev); |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1263 | |
| 1264 | /* We can't extend the feature bits once we've added config bytes */ |
| 1265 | if (dev->desc->feature_len <= bit / CHAR_BIT) { |
| 1266 | assert(dev->desc->config_len == 0); |
| 1267 | dev->desc->feature_len = (bit / CHAR_BIT) + 1; |
| 1268 | } |
| 1269 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1270 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
| 1271 | } |
| 1272 | |
| 1273 | /* This routine sets the configuration fields for an existing device's |
| 1274 | * descriptor. It only works for the last device, but that's OK because that's |
| 1275 | * how we use it. */ |
| 1276 | static void set_config(struct device *dev, unsigned len, const void *conf) |
| 1277 | { |
| 1278 | /* Check we haven't overflowed our single page. */ |
| 1279 | if (device_config(dev) + len > devices.descpage + getpagesize()) |
| 1280 | errx(1, "Too many devices"); |
| 1281 | |
| 1282 | /* Copy in the config information, and store the length. */ |
| 1283 | memcpy(device_config(dev), conf, len); |
| 1284 | dev->desc->config_len = len; |
| 1285 | } |
| 1286 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1287 | /* This routine does all the creation and setup of a new device, including |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1288 | * calling new_dev_desc() to allocate the descriptor and device memory. |
| 1289 | * |
| 1290 | * See what I mean about userspace being boring? */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1291 | static struct device *new_device(const char *name, u16 type, int fd, |
| 1292 | bool (*handle_input)(int, struct device *)) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1293 | { |
| 1294 | struct device *dev = malloc(sizeof(*dev)); |
| 1295 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1296 | /* Now we populate the fields one at a time. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1297 | dev->fd = fd; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1298 | /* If we have an input handler for this file descriptor, then we add it |
| 1299 | * to the device_list's fdset and maxfd. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1300 | if (handle_input) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1301 | add_device_fd(dev->fd); |
| 1302 | dev->desc = new_dev_desc(type); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1303 | dev->handle_input = handle_input; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1304 | dev->name = name; |
Rusty Russell | d1c856e | 2007-11-19 11:20:40 -0500 | [diff] [blame] | 1305 | dev->vq = NULL; |
Rusty Russell | a007a75 | 2008-05-02 21:50:53 -0500 | [diff] [blame] | 1306 | dev->ready = NULL; |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1307 | |
| 1308 | /* Append to device list. Prepending to a single-linked list is |
| 1309 | * easier, but the user expects the devices to be arranged on the bus |
| 1310 | * in command-line order. The first network device on the command line |
| 1311 | * is eth0, the first block device /dev/vda, etc. */ |
| 1312 | if (devices.lastdev) |
| 1313 | devices.lastdev->next = dev; |
| 1314 | else |
| 1315 | devices.dev = dev; |
| 1316 | devices.lastdev = dev; |
| 1317 | |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1318 | return dev; |
| 1319 | } |
| 1320 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 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. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1323 | static void setup_console(void) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1324 | { |
| 1325 | struct device *dev; |
| 1326 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1327 | /* If we can save the initial standard input settings... */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1328 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
| 1329 | struct termios term = orig_term; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1330 | /* Then we turn off echo, line buffering and ^C etc. We want a |
| 1331 | * raw input stream to the Guest. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1332 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
| 1333 | tcsetattr(STDIN_FILENO, TCSANOW, &term); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1334 | /* If we exit gracefully, the original settings will be |
| 1335 | * restored so the user can see what they're typing. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1336 | atexit(restore_term); |
| 1337 | } |
| 1338 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1339 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
| 1340 | STDIN_FILENO, handle_console_input); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1341 | /* We store the console state in dev->priv, and initialize it. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1342 | dev->priv = malloc(sizeof(struct console_abort)); |
| 1343 | ((struct console_abort *)dev->priv)->count = 0; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1344 | |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1345 | /* The console needs two virtqueues: the input then the output. When |
| 1346 | * they put something the input queue, we make sure we're listening to |
| 1347 | * stdin. When they put something in the output queue, we write it to |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1348 | * stdout. */ |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1349 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1350 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1351 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1352 | verbose("device %u: console\n", devices.device_num++); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1353 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1354 | /*:*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1355 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1356 | static void timeout_alarm(int sig) |
| 1357 | { |
| 1358 | write(timeoutpipe[1], "", 1); |
| 1359 | } |
| 1360 | |
| 1361 | static void setup_timeout(void) |
| 1362 | { |
| 1363 | if (pipe(timeoutpipe) != 0) |
| 1364 | err(1, "Creating timeout pipe"); |
| 1365 | |
| 1366 | if (fcntl(timeoutpipe[1], F_SETFL, |
| 1367 | fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0) |
| 1368 | err(1, "Making timeout pipe nonblocking"); |
| 1369 | |
| 1370 | add_device_fd(timeoutpipe[0]); |
| 1371 | signal(SIGALRM, timeout_alarm); |
| 1372 | } |
| 1373 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1374 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
| 1375 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
| 1376 | * used to send packets to another guest in a 1:1 manner. |
| 1377 | * |
| 1378 | * More sopisticated is to use one of the tools developed for project like UML |
| 1379 | * to do networking. |
| 1380 | * |
| 1381 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
| 1382 | * completely generic ("here's my vring, attach to your vring") and would work |
| 1383 | * for any traffic. Of course, namespace and permissions issues need to be |
| 1384 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide |
| 1385 | * multiple inter-guest channels behind one interface, although it would |
| 1386 | * require some manner of hotplugging new virtio channels. |
| 1387 | * |
| 1388 | * Finally, we could implement a virtio network switch in the kernel. :*/ |
| 1389 | |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1390 | static u32 str2ip(const char *ipaddr) |
| 1391 | { |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1392 | unsigned int b[4]; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1393 | |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1394 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
| 1395 | errx(1, "Failed to parse IP address '%s'", ipaddr); |
| 1396 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; |
| 1397 | } |
| 1398 | |
| 1399 | static void str2mac(const char *macaddr, unsigned char mac[6]) |
| 1400 | { |
| 1401 | unsigned int m[6]; |
| 1402 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", |
| 1403 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) |
| 1404 | errx(1, "Failed to parse mac address '%s'", macaddr); |
| 1405 | mac[0] = m[0]; |
| 1406 | mac[1] = m[1]; |
| 1407 | mac[2] = m[2]; |
| 1408 | mac[3] = m[3]; |
| 1409 | mac[4] = m[4]; |
| 1410 | mac[5] = m[5]; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1411 | } |
| 1412 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1413 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
| 1414 | * network device to the bridge device specified by the command line. |
| 1415 | * |
| 1416 | * This is yet another James Morris contribution (I'm an IP-level guy, so I |
| 1417 | * dislike bridging), and I just try not to break it. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1418 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
| 1419 | { |
| 1420 | int ifidx; |
| 1421 | struct ifreq ifr; |
| 1422 | |
| 1423 | if (!*br_name) |
| 1424 | errx(1, "must specify bridge name"); |
| 1425 | |
| 1426 | ifidx = if_nametoindex(if_name); |
| 1427 | if (!ifidx) |
| 1428 | errx(1, "interface %s does not exist!", if_name); |
| 1429 | |
| 1430 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1431 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1432 | ifr.ifr_ifindex = ifidx; |
| 1433 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) |
| 1434 | err(1, "can't add %s to bridge %s", if_name, br_name); |
| 1435 | } |
| 1436 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1437 | /* This sets up the Host end of the network device with an IP address, brings |
| 1438 | * it up so packets will flow, the copies the MAC address into the hwaddr |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1439 | * pointer. */ |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1440 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1441 | { |
| 1442 | struct ifreq ifr; |
| 1443 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; |
| 1444 | |
| 1445 | memset(&ifr, 0, sizeof(ifr)); |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1446 | strcpy(ifr.ifr_name, tapif); |
| 1447 | |
| 1448 | /* Don't read these incantations. Just cut & paste them like I did! */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1449 | sin->sin_family = AF_INET; |
| 1450 | sin->sin_addr.s_addr = htonl(ipaddr); |
| 1451 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1452 | err(1, "Setting %s interface address", tapif); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1453 | ifr.ifr_flags = IFF_UP; |
| 1454 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1455 | err(1, "Bringing interface %s up", tapif); |
| 1456 | } |
| 1457 | |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1458 | static int get_tun_device(char tapif[IFNAMSIZ]) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1459 | { |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1460 | struct ifreq ifr; |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1461 | int netfd; |
| 1462 | |
| 1463 | /* Start with this zeroed. Messy but sure. */ |
| 1464 | memset(&ifr, 0, sizeof(ifr)); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1465 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1466 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
| 1467 | * tap device is like a tun device, only somehow different. To tell |
| 1468 | * the truth, I completely blundered my way through this code, but it |
| 1469 | * works now! */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1470 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 1471 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1472 | strcpy(ifr.ifr_name, "tap%d"); |
| 1473 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) |
| 1474 | err(1, "configuring /dev/net/tun"); |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1475 | |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 1476 | if (ioctl(netfd, TUNSETOFFLOAD, |
| 1477 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) |
| 1478 | err(1, "Could not set features for tun device"); |
| 1479 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1480 | /* We don't need checksums calculated for packets coming in this |
| 1481 | * device: trust us! */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1482 | ioctl(netfd, TUNSETNOCSUM, 1); |
| 1483 | |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1484 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
| 1485 | return netfd; |
| 1486 | } |
| 1487 | |
| 1488 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or |
| 1489 | * routing, but the principle is the same: it uses the "tun" device to inject |
| 1490 | * packets into the Host as if they came in from a normal network card. We |
| 1491 | * just shunt packets between the Guest and the tun device. */ |
| 1492 | static void setup_tun_net(char *arg) |
| 1493 | { |
| 1494 | struct device *dev; |
| 1495 | int netfd, ipfd; |
| 1496 | u32 ip = INADDR_ANY; |
| 1497 | bool bridging = false; |
| 1498 | char tapif[IFNAMSIZ], *p; |
| 1499 | struct virtio_net_config conf; |
| 1500 | |
| 1501 | netfd = get_tun_device(tapif); |
| 1502 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1503 | /* First we create a new network device. */ |
| 1504 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1505 | |
Rusty Russell | 56ae43d | 2007-10-22 11:24:23 +1000 | [diff] [blame] | 1506 | /* Network devices need a receive and a send queue, just like |
| 1507 | * console. */ |
Rusty Russell | 5dae785 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1508 | add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1509 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1510 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1511 | /* We need a socket to perform the magic network ioctls to bring up the |
| 1512 | * tap interface, connect to the bridge etc. Any socket will do! */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1513 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
| 1514 | if (ipfd < 0) |
| 1515 | err(1, "opening IP socket"); |
| 1516 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1517 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1518 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1519 | 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 Russell | 40c4207 | 2008-08-12 17:52:51 -0500 | [diff] [blame] | 1527 | add_feature(dev, VIRTIO_NET_F_MAC); |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1528 | *p = '\0'; |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1529 | } |
| 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 Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1535 | ip = str2ip(arg); |
| 1536 | |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1537 | /* Set up the tun device. */ |
| 1538 | configure_device(ipfd, tapif, ip); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1539 | |
Rusty Russell | 2088761 | 2008-05-30 15:09:46 -0500 | [diff] [blame] | 1540 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 1541 | /* Expect Guest to handle everything except UFO */ |
| 1542 | add_feature(dev, VIRTIO_NET_F_CSUM); |
| 1543 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); |
Rusty Russell | 398f187 | 2008-07-29 09:58:37 -0500 | [diff] [blame] | 1544 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
| 1545 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); |
| 1546 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); |
| 1547 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); |
| 1548 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); |
| 1549 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1550 | set_config(dev, sizeof(conf), &conf); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1551 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1552 | /* We don't need the socket any more; setup is done. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1553 | close(ipfd); |
| 1554 | |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1555 | devices.device_num++; |
| 1556 | |
| 1557 | if (bridging) |
| 1558 | verbose("device %u: tun %s attached to bridge: %s\n", |
| 1559 | devices.device_num, tapif, arg); |
| 1560 | else |
| 1561 | verbose("device %u: tun %s: %s\n", |
| 1562 | devices.device_num, tapif, arg); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1563 | } |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1564 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1565 | /* Our block (disk) device should be really simple: the Guest asks for a block |
| 1566 | * number and we read or write that position in the file. Unfortunately, that |
| 1567 | * was amazingly slow: the Guest waits until the read is finished before |
| 1568 | * running anything else, even if it could have been doing useful work. |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1569 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1570 | * We could use async I/O, except it's reputed to suck so hard that characters |
| 1571 | * actually go missing from your code when you try to use it. |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1572 | * |
| 1573 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ |
| 1574 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1575 | /* This hangs off device->priv. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1576 | struct vblk_info |
| 1577 | { |
| 1578 | /* The size of the file. */ |
| 1579 | off64_t len; |
| 1580 | |
| 1581 | /* The file descriptor for the file. */ |
| 1582 | int fd; |
| 1583 | |
| 1584 | /* IO thread listens on this file descriptor [0]. */ |
| 1585 | int workpipe[2]; |
| 1586 | |
| 1587 | /* IO thread writes to this file descriptor to mark it done, then |
| 1588 | * Launcher triggers interrupt to Guest. */ |
| 1589 | int done_fd; |
| 1590 | }; |
| 1591 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1592 | /*L:210 |
| 1593 | * The Disk |
| 1594 | * |
| 1595 | * Remember that the block device is handled by a separate I/O thread. We head |
| 1596 | * straight into the core of that thread here: |
| 1597 | */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1598 | static bool service_io(struct device *dev) |
| 1599 | { |
| 1600 | struct vblk_info *vblk = dev->priv; |
| 1601 | unsigned int head, out_num, in_num, wlen; |
| 1602 | int ret; |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1603 | u8 *in; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1604 | struct virtio_blk_outhdr *out; |
| 1605 | struct iovec iov[dev->vq->vring.num]; |
| 1606 | off64_t off; |
| 1607 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1608 | /* See if there's a request waiting. If not, nothing to do. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1609 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
| 1610 | if (head == dev->vq->vring.num) |
| 1611 | return false; |
| 1612 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1613 | /* Every block request should contain at least one output buffer |
| 1614 | * (detailing the location on disk and the type of request) and one |
| 1615 | * input buffer (to hold the result). */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1616 | if (out_num == 0 || in_num == 0) |
| 1617 | errx(1, "Bad virtblk cmd %u out=%u in=%u", |
| 1618 | head, out_num, in_num); |
| 1619 | |
| 1620 | out = convert(&iov[0], struct virtio_blk_outhdr); |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1621 | in = convert(&iov[out_num+in_num-1], u8); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1622 | off = out->sector * 512; |
| 1623 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1624 | /* The block device implements "barriers", where the Guest indicates |
| 1625 | * that it wants all previous writes to occur before this write. We |
| 1626 | * don't have a way of asking our kernel to do a barrier, so we just |
| 1627 | * synchronize all the data in the file. Pretty poor, no? */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1628 | if (out->type & VIRTIO_BLK_T_BARRIER) |
| 1629 | fdatasync(vblk->fd); |
| 1630 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1631 | /* 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. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1633 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
| 1634 | fprintf(stderr, "Scsi commands unsupported\n"); |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1635 | *in = VIRTIO_BLK_S_UNSUPP; |
Anthony Liguori | 1200e64 | 2007-11-08 21:13:44 -0600 | [diff] [blame] | 1636 | wlen = sizeof(*in); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1637 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
| 1638 | /* Write */ |
| 1639 | |
| 1640 | /* Move to the right location in the block file. This can fail |
| 1641 | * if they try to write past end. */ |
| 1642 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
| 1643 | err(1, "Bad seek to sector %llu", out->sector); |
| 1644 | |
| 1645 | ret = writev(vblk->fd, iov+1, out_num-1); |
| 1646 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); |
| 1647 | |
| 1648 | /* Grr... Now we know how long the descriptor they sent was, we |
| 1649 | * make sure they didn't try to write over the end of the block |
| 1650 | * file (possibly extending it). */ |
| 1651 | if (ret > 0 && off + ret > vblk->len) { |
| 1652 | /* Trim it back to the correct length */ |
| 1653 | ftruncate64(vblk->fd, vblk->len); |
| 1654 | /* Die, bad Guest, die. */ |
| 1655 | errx(1, "Write past end %llu+%u", off, ret); |
| 1656 | } |
Anthony Liguori | 1200e64 | 2007-11-08 21:13:44 -0600 | [diff] [blame] | 1657 | wlen = sizeof(*in); |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1658 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1659 | } else { |
| 1660 | /* Read */ |
| 1661 | |
| 1662 | /* Move to the right location in the block file. This can fail |
| 1663 | * if they try to read past end. */ |
| 1664 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
| 1665 | err(1, "Bad seek to sector %llu", out->sector); |
| 1666 | |
| 1667 | ret = readv(vblk->fd, iov+1, in_num-1); |
| 1668 | verbose("READ from sector %llu: %i\n", out->sector, ret); |
| 1669 | if (ret >= 0) { |
Anthony Liguori | 1200e64 | 2007-11-08 21:13:44 -0600 | [diff] [blame] | 1670 | wlen = sizeof(*in) + ret; |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1671 | *in = VIRTIO_BLK_S_OK; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1672 | } else { |
Anthony Liguori | 1200e64 | 2007-11-08 21:13:44 -0600 | [diff] [blame] | 1673 | wlen = sizeof(*in); |
Rusty Russell | cb38fa2 | 2008-05-02 21:50:45 -0500 | [diff] [blame] | 1674 | *in = VIRTIO_BLK_S_IOERR; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1675 | } |
| 1676 | } |
| 1677 | |
| 1678 | /* We can't trigger an IRQ, because we're not the Launcher. It does |
| 1679 | * that when we tell it we're done. */ |
| 1680 | add_used(dev->vq, head, wlen); |
| 1681 | return true; |
| 1682 | } |
| 1683 | |
| 1684 | /* This is the thread which actually services the I/O. */ |
| 1685 | static int io_thread(void *_dev) |
| 1686 | { |
| 1687 | struct device *dev = _dev; |
| 1688 | struct vblk_info *vblk = dev->priv; |
| 1689 | char c; |
| 1690 | |
| 1691 | /* Close other side of workpipe so we get 0 read when main dies. */ |
| 1692 | close(vblk->workpipe[1]); |
| 1693 | /* Close the other side of the done_fd pipe. */ |
| 1694 | close(dev->fd); |
| 1695 | |
| 1696 | /* When this read fails, it means Launcher died, so we follow. */ |
| 1697 | while (read(vblk->workpipe[0], &c, 1) == 1) { |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1698 | /* We acknowledge each request immediately to reduce latency, |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1699 | * rather than waiting until we've done them all. I haven't |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1700 | * measured to see if it makes any difference. |
| 1701 | * |
| 1702 | * That would be an interesting test, wouldn't it? You could |
| 1703 | * also try having more than one I/O thread. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1704 | while (service_io(dev)) |
| 1705 | write(vblk->done_fd, &c, 1); |
| 1706 | } |
| 1707 | return 0; |
| 1708 | } |
| 1709 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1710 | /* Now we've seen the I/O thread, we return to the Launcher to see what happens |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1711 | * when that thread tells us it's completed some I/O. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1712 | static bool handle_io_finish(int fd, struct device *dev) |
| 1713 | { |
| 1714 | char c; |
| 1715 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1716 | /* If the I/O thread died, presumably it printed the error, so we |
| 1717 | * simply exit. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1718 | if (read(dev->fd, &c, 1) != 1) |
| 1719 | exit(1); |
| 1720 | |
| 1721 | /* It did some work, so trigger the irq. */ |
| 1722 | trigger_irq(fd, dev->vq); |
| 1723 | return true; |
| 1724 | } |
| 1725 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1726 | /* When the Guest submits some I/O, we just need to wake the I/O thread. */ |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1727 | static void handle_virtblk_output(int fd, struct virtqueue *vq, bool timeout) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1728 | { |
| 1729 | struct vblk_info *vblk = vq->dev->priv; |
| 1730 | char c = 0; |
| 1731 | |
| 1732 | /* Wake up I/O thread and tell it to go to work! */ |
| 1733 | if (write(vblk->workpipe[1], &c, 1) != 1) |
| 1734 | /* Presumably it indicated why it died. */ |
| 1735 | exit(1); |
| 1736 | } |
| 1737 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1738 | /*L:198 This actually sets up a virtual block device. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1739 | static void setup_block_file(const char *filename) |
| 1740 | { |
| 1741 | int p[2]; |
| 1742 | struct device *dev; |
| 1743 | struct vblk_info *vblk; |
| 1744 | void *stack; |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1745 | struct virtio_blk_config conf; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1746 | |
| 1747 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ |
| 1748 | pipe(p); |
| 1749 | |
| 1750 | /* The device responds to return from I/O thread. */ |
| 1751 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); |
| 1752 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1753 | /* The device has one virtqueue, where the Guest places requests. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1754 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); |
| 1755 | |
| 1756 | /* Allocate the room for our own bookkeeping */ |
| 1757 | vblk = dev->priv = malloc(sizeof(*vblk)); |
| 1758 | |
| 1759 | /* First we open the file and store the length. */ |
| 1760 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); |
| 1761 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); |
| 1762 | |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1763 | /* We support barriers. */ |
| 1764 | add_feature(dev, VIRTIO_BLK_F_BARRIER); |
| 1765 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1766 | /* Tell Guest how many sectors this device has. */ |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1767 | conf.capacity = cpu_to_le64(vblk->len / 512); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1768 | |
| 1769 | /* Tell Guest not to put in too many descriptors at once: two are used |
| 1770 | * for the in and out elements. */ |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1771 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
| 1772 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); |
| 1773 | |
| 1774 | set_config(dev, sizeof(conf), &conf); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1775 | |
| 1776 | /* The I/O thread writes to this end of the pipe when done. */ |
| 1777 | vblk->done_fd = p[1]; |
| 1778 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1779 | /* This is the second pipe, which is how we tell the I/O thread about |
| 1780 | * more work. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1781 | pipe(vblk->workpipe); |
| 1782 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1783 | /* Create stack for thread and run it. Since stack grows upwards, we |
| 1784 | * point the stack pointer to the end of this region. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1785 | stack = malloc(32768); |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1786 | /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from |
| 1787 | * becoming a zombie. */ |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1788 | if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1) |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1789 | err(1, "Creating clone"); |
| 1790 | |
| 1791 | /* We don't need to keep the I/O thread's end of the pipes open. */ |
| 1792 | close(vblk->done_fd); |
| 1793 | close(vblk->workpipe[0]); |
| 1794 | |
| 1795 | verbose("device %u: virtblock %llu sectors\n", |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1796 | devices.device_num, le64_to_cpu(conf.capacity)); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1797 | } |
Rusty Russell | 28fd6d7 | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1798 | |
| 1799 | /* Our random number generator device reads from /dev/random into the Guest's |
| 1800 | * input buffers. The usual case is that the Guest doesn't want random numbers |
| 1801 | * and so has no buffers although /dev/random is still readable, whereas |
| 1802 | * console is the reverse. |
| 1803 | * |
| 1804 | * The same logic applies, however. */ |
| 1805 | static bool handle_rng_input(int fd, struct device *dev) |
| 1806 | { |
| 1807 | int len; |
| 1808 | unsigned int head, in_num, out_num, totlen = 0; |
| 1809 | struct iovec iov[dev->vq->vring.num]; |
| 1810 | |
| 1811 | /* First we need a buffer from the Guests's virtqueue. */ |
| 1812 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
| 1813 | |
| 1814 | /* If they're not ready for input, stop listening to this file |
| 1815 | * descriptor. We'll start again once they add an input buffer. */ |
| 1816 | if (head == dev->vq->vring.num) |
| 1817 | return false; |
| 1818 | |
| 1819 | if (out_num) |
| 1820 | errx(1, "Output buffers in rng?"); |
| 1821 | |
| 1822 | /* This is why we convert to iovecs: the readv() call uses them, and so |
| 1823 | * it reads straight into the Guest's buffer. We loop to make sure we |
| 1824 | * fill it. */ |
| 1825 | while (!iov_empty(iov, in_num)) { |
| 1826 | len = readv(dev->fd, iov, in_num); |
| 1827 | if (len <= 0) |
| 1828 | err(1, "Read from /dev/random gave %i", len); |
| 1829 | iov_consume(iov, in_num, len); |
| 1830 | totlen += len; |
| 1831 | } |
| 1832 | |
| 1833 | /* Tell the Guest about the new input. */ |
| 1834 | add_used_and_trigger(fd, dev->vq, head, totlen); |
| 1835 | |
| 1836 | /* Everything went OK! */ |
| 1837 | return true; |
| 1838 | } |
| 1839 | |
| 1840 | /* And this creates a "hardware" random number device for the Guest. */ |
| 1841 | static void setup_rng(void) |
| 1842 | { |
| 1843 | struct device *dev; |
| 1844 | int fd; |
| 1845 | |
| 1846 | fd = open_or_die("/dev/random", O_RDONLY); |
| 1847 | |
| 1848 | /* The device responds to return from I/O thread. */ |
| 1849 | dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input); |
| 1850 | |
| 1851 | /* The device has one virtqueue, where the Guest places inbufs. */ |
| 1852 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
| 1853 | |
| 1854 | verbose("device %u: rng\n", devices.device_num++); |
| 1855 | } |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1856 | /* That's the end of device setup. */ |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1857 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1858 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1859 | static void __attribute__((noreturn)) restart_guest(void) |
| 1860 | { |
| 1861 | unsigned int i; |
| 1862 | |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 1863 | /* Since we don't track all open fds, we simply close everything beyond |
| 1864 | * stderr. */ |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1865 | for (i = 3; i < FD_SETSIZE; i++) |
| 1866 | close(i); |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 1867 | |
| 1868 | /* The exec automatically gets rid of the I/O and Waker threads. */ |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1869 | execv(main_args[0], main_args); |
| 1870 | err(1, "Could not exec %s", main_args[0]); |
| 1871 | } |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1872 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1873 | /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1874 | * its input and output, and finally, lays it to rest. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1875 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1876 | { |
| 1877 | for (;;) { |
Jes Sorensen | 511801d | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 1878 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1879 | unsigned long notify_addr; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1880 | int readval; |
| 1881 | |
| 1882 | /* We read from the /dev/lguest device to run the Guest. */ |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 1883 | readval = pread(lguest_fd, ¬ify_addr, |
| 1884 | sizeof(notify_addr), cpu_id); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1885 | |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1886 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
| 1887 | if (readval == sizeof(notify_addr)) { |
| 1888 | verbose("Notify on address %#lx\n", notify_addr); |
| 1889 | handle_output(lguest_fd, notify_addr); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1890 | continue; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1891 | /* ENOENT means the Guest died. Reading tells us why. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1892 | } else if (errno == ENOENT) { |
| 1893 | char reason[1024] = { 0 }; |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 1894 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1895 | errx(1, "%s", reason); |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1896 | /* ERESTART means that we need to reboot the guest */ |
| 1897 | } else if (errno == ERESTART) { |
| 1898 | restart_guest(); |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 1899 | /* EAGAIN means a signal (timeout). |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1900 | * Anything else means a bug or incompatible change. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1901 | } else if (errno != EAGAIN) |
| 1902 | err(1, "Running guest failed"); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1903 | |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 1904 | /* Only service input on thread for CPU 0. */ |
| 1905 | if (cpu_id != 0) |
| 1906 | continue; |
| 1907 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1908 | /* Service input, then unset the BREAK to release the Waker. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1909 | handle_input(lguest_fd); |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 1910 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1911 | err(1, "Resetting break"); |
| 1912 | } |
| 1913 | } |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1914 | /*L:240 |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1915 | * This is the end of the Launcher. The good news: we are over halfway |
| 1916 | * through! The bad news: the most fiendish part of the code still lies ahead |
| 1917 | * of us. |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1918 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1919 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
| 1920 | * "make Host". |
| 1921 | :*/ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1922 | |
| 1923 | static struct option opts[] = { |
| 1924 | { "verbose", 0, NULL, 'v' }, |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1925 | { "tunnet", 1, NULL, 't' }, |
| 1926 | { "block", 1, NULL, 'b' }, |
Rusty Russell | 28fd6d7 | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1927 | { "rng", 0, NULL, 'r' }, |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1928 | { "initrd", 1, NULL, 'i' }, |
| 1929 | { NULL }, |
| 1930 | }; |
| 1931 | static void usage(void) |
| 1932 | { |
| 1933 | errx(1, "Usage: lguest [--verbose] " |
Mark McLoughlin | dec6a2b | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 1934 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1935 | "|--block=<filename>|--initrd=<filename>]...\n" |
| 1936 | "<mem-in-mb> vmlinux [args...]"); |
| 1937 | } |
| 1938 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 1939 | /*L:105 The main routine is where the real work begins: */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1940 | int main(int argc, char *argv[]) |
| 1941 | { |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 1942 | /* Memory, top-level pagetable, code startpoint and size of the |
| 1943 | * (optional) initrd. */ |
| 1944 | unsigned long mem = 0, pgdir, start, initrd_size = 0; |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1945 | /* Two temporaries and the /dev/lguest file descriptor. */ |
Rusty Russell | 6570c4599 | 2007-07-23 18:43:56 -0700 | [diff] [blame] | 1946 | int i, c, lguest_fd; |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 1947 | /* The boot information for the Guest. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 1948 | struct boot_params *boot; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1949 | /* If they specify an initrd file to load. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1950 | const char *initrd_name = NULL; |
| 1951 | |
Balaji Rao | ec04b13 | 2007-12-28 14:26:24 +0530 | [diff] [blame] | 1952 | /* Save the args: we "reboot" by execing ourselves again. */ |
| 1953 | main_args = argv; |
| 1954 | /* We don't "wait" for the children, so prevent them from becoming |
| 1955 | * zombies. */ |
| 1956 | signal(SIGCHLD, SIG_IGN); |
| 1957 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1958 | /* First we initialize the device list. Since console and network |
| 1959 | * device receive input from a file descriptor, we keep an fdset |
| 1960 | * (infds) and the maximum fd number (max_infd) with the head of the |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1961 | * list. We also keep a pointer to the last device. Finally, we keep |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1962 | * the next interrupt number to use for devices (1: remember that 0 is |
| 1963 | * used by the timer). */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1964 | FD_ZERO(&devices.infds); |
| 1965 | devices.max_infd = -1; |
Rusty Russell | a586d4f | 2008-02-04 23:49:56 -0500 | [diff] [blame] | 1966 | devices.lastdev = NULL; |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1967 | devices.next_irq = 1; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1968 | |
Glauber de Oliveira Costa | e3283fa | 2008-01-07 11:05:23 -0200 | [diff] [blame] | 1969 | cpu_id = 0; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1970 | /* We need to know how much memory so we can set up the device |
| 1971 | * descriptor and memory pages for the devices as we parse the command |
| 1972 | * line. So we quickly look through the arguments to find the amount |
| 1973 | * of memory now. */ |
Rusty Russell | 6570c4599 | 2007-07-23 18:43:56 -0700 | [diff] [blame] | 1974 | for (i = 1; i < argc; i++) { |
| 1975 | if (argv[i][0] != '-') { |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 1976 | mem = atoi(argv[i]) * 1024 * 1024; |
| 1977 | /* We start by mapping anonymous pages over all of |
| 1978 | * guest-physical memory range. This fills it with 0, |
| 1979 | * and ensures that the Guest won't be killed when it |
| 1980 | * tries to access it. */ |
| 1981 | guest_base = map_zeroed_pages(mem / getpagesize() |
| 1982 | + DEVICE_PAGES); |
| 1983 | guest_limit = mem; |
| 1984 | guest_max = mem + DEVICE_PAGES*getpagesize(); |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1985 | devices.descpage = get_pages(1); |
Rusty Russell | 6570c4599 | 2007-07-23 18:43:56 -0700 | [diff] [blame] | 1986 | break; |
| 1987 | } |
| 1988 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 1989 | |
| 1990 | /* The options are fairly straight-forward */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1991 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
| 1992 | switch (c) { |
| 1993 | case 'v': |
| 1994 | verbose = true; |
| 1995 | break; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1996 | case 't': |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 1997 | setup_tun_net(optarg); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 1998 | break; |
| 1999 | case 'b': |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 2000 | setup_block_file(optarg); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2001 | break; |
Rusty Russell | 28fd6d7 | 2008-07-29 09:58:33 -0500 | [diff] [blame] | 2002 | case 'r': |
| 2003 | setup_rng(); |
| 2004 | break; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2005 | case 'i': |
| 2006 | initrd_name = optarg; |
| 2007 | break; |
| 2008 | default: |
| 2009 | warnx("Unknown argument %s", argv[optind]); |
| 2010 | usage(); |
| 2011 | } |
| 2012 | } |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2013 | /* After the other arguments we expect memory and kernel image name, |
| 2014 | * followed by command line arguments for the kernel. */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2015 | if (optind + 2 > argc) |
| 2016 | usage(); |
| 2017 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 2018 | verbose("Guest base is at %p\n", guest_base); |
| 2019 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2020 | /* We always have a console device */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 2021 | setup_console(); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2022 | |
Rusty Russell | a161883 | 2008-07-29 09:58:35 -0500 | [diff] [blame] | 2023 | /* We can timeout waiting for Guest network transmit. */ |
| 2024 | setup_timeout(); |
| 2025 | |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2026 | /* Now we load the kernel */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 2027 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2028 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 2029 | /* Boot information is stashed at physical address 0 */ |
| 2030 | boot = from_guest_phys(0); |
| 2031 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2032 | /* Map the initrd image if requested (at top of physical memory) */ |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2033 | if (initrd_name) { |
| 2034 | initrd_size = load_initrd(initrd_name, mem); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2035 | /* These are the location in the Linux boot header where the |
| 2036 | * start and size of the initrd are expected to be found. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2037 | boot->hdr.ramdisk_image = mem - initrd_size; |
| 2038 | boot->hdr.ramdisk_size = initrd_size; |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2039 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2040 | boot->hdr.type_of_loader = 0xFF; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2041 | } |
| 2042 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2043 | /* Set up the initial linear pagetables, starting below the initrd. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 2044 | pgdir = setup_pagetables(mem, initrd_size); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2045 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2046 | /* The Linux boot header contains an "E820" memory map: ours is a |
| 2047 | * simple, single region. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2048 | boot->e820_entries = 1; |
| 2049 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2050 | /* The boot header contains a command line pointer: we put the command |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2051 | * line after the boot header. */ |
| 2052 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 2053 | /* We use a simple helper to copy the arguments separated by spaces. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2054 | concat((char *)(boot + 1), argv+optind+2); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2055 | |
Rusty Russell | 814a0e5 | 2007-10-22 11:29:44 +1000 | [diff] [blame] | 2056 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2057 | boot->hdr.version = 0x207; |
Rusty Russell | 814a0e5 | 2007-10-22 11:29:44 +1000 | [diff] [blame] | 2058 | |
| 2059 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2060 | boot->hdr.hardware_subarch = 1; |
Rusty Russell | 814a0e5 | 2007-10-22 11:29:44 +1000 | [diff] [blame] | 2061 | |
Rusty Russell | 43d33b2 | 2007-10-22 11:29:57 +1000 | [diff] [blame] | 2062 | /* Tell the entry path not to try to reload segment registers. */ |
| 2063 | boot->hdr.loadflags |= KEEP_SEGMENTS; |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2064 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2065 | /* We tell the kernel to initialize the Guest: this returns the open |
| 2066 | * /dev/lguest file descriptor. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 2067 | lguest_fd = tell_kernel(pgdir, start); |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2068 | |
Rusty Russell | 8c79873 | 2008-07-29 09:58:38 -0500 | [diff] [blame] | 2069 | /* We clone off a thread, which wakes the Launcher whenever one of the |
| 2070 | * input file descriptors needs attention. We call this the Waker, and |
| 2071 | * we'll cover it in a moment. */ |
| 2072 | setup_waker(lguest_fd); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2073 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 2074 | /* Finally, run the Guest. This doesn't return. */ |
Rusty Russell | 17cbca2 | 2007-10-22 11:24:22 +1000 | [diff] [blame] | 2075 | run_guest(lguest_fd); |
Rusty Russell | 8ca47e0 | 2007-07-19 01:49:29 -0700 | [diff] [blame] | 2076 | } |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 2077 | /*:*/ |
| 2078 | |
| 2079 | /*M:999 |
| 2080 | * Mastery is done: you now know everything I do. |
| 2081 | * |
| 2082 | * But surely you have seen code, features and bugs in your wanderings which |
| 2083 | * you now yearn to attack? That is the real game, and I look forward to you |
| 2084 | * patching and forking lguest into the Your-Name-Here-visor. |
| 2085 | * |
| 2086 | * Farewell, and good coding! |
| 2087 | * Rusty Russell. |
| 2088 | */ |