aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/lguest/lguest.c
diff options
context:
space:
mode:
Diffstat (limited to 'Documentation/lguest/lguest.c')
-rw-r--r--Documentation/lguest/lguest.c1629
1 files changed, 882 insertions, 747 deletions
diff --git a/Documentation/lguest/lguest.c b/Documentation/lguest/lguest.c
index 103e346c8b6a..5bdc37f81842 100644
--- a/Documentation/lguest/lguest.c
+++ b/Documentation/lguest/lguest.c
@@ -1,10 +1,7 @@
1/*P:100 This is the Launcher code, a simple program which lays out the 1/*P:100 This is the Launcher code, a simple program which lays out the
2 * "physical" memory for the new Guest by mapping the kernel image and the 2 * "physical" memory for the new Guest by mapping the kernel image and the
3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest. 3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest.
4 * 4:*/
5 * The only trick: the Makefile links it at a high address so it will be clear
6 * of the guest memory region. It means that each Guest cannot have more than
7 * about 2.5G of memory on a normally configured Host. :*/
8#define _LARGEFILE64_SOURCE 5#define _LARGEFILE64_SOURCE
9#define _GNU_SOURCE 6#define _GNU_SOURCE
10#include <stdio.h> 7#include <stdio.h>
@@ -15,6 +12,7 @@
15#include <stdlib.h> 12#include <stdlib.h>
16#include <elf.h> 13#include <elf.h>
17#include <sys/mman.h> 14#include <sys/mman.h>
15#include <sys/param.h>
18#include <sys/types.h> 16#include <sys/types.h>
19#include <sys/stat.h> 17#include <sys/stat.h>
20#include <sys/wait.h> 18#include <sys/wait.h>
@@ -34,7 +32,9 @@
34#include <termios.h> 32#include <termios.h>
35#include <getopt.h> 33#include <getopt.h>
36#include <zlib.h> 34#include <zlib.h>
37/*L:110 We can ignore the 28 include files we need for this program, but I do 35#include <assert.h>
36#include <sched.h>
37/*L:110 We can ignore the 30 include files we need for this program, but I do
38 * want to draw attention to the use of kernel-style types. 38 * want to draw attention to the use of kernel-style types.
39 * 39 *
40 * As Linus said, "C is a Spartan language, and so should your naming be." I 40 * As Linus said, "C is a Spartan language, and so should your naming be." I
@@ -45,8 +45,14 @@ typedef unsigned long long u64;
45typedef uint32_t u32; 45typedef uint32_t u32;
46typedef uint16_t u16; 46typedef uint16_t u16;
47typedef uint8_t u8; 47typedef uint8_t u8;
48#include "../../include/linux/lguest_launcher.h" 48#include "linux/lguest_launcher.h"
49#include "../../include/asm-x86/e820_32.h" 49#include "linux/pci_ids.h"
50#include "linux/virtio_config.h"
51#include "linux/virtio_net.h"
52#include "linux/virtio_blk.h"
53#include "linux/virtio_console.h"
54#include "linux/virtio_ring.h"
55#include "asm-x86/bootparam.h"
50/*:*/ 56/*:*/
51 57
52#define PAGE_PRESENT 0x7 /* Present, RW, Execute */ 58#define PAGE_PRESENT 0x7 /* Present, RW, Execute */
@@ -55,6 +61,10 @@ typedef uint8_t u8;
55#ifndef SIOCBRADDIF 61#ifndef SIOCBRADDIF
56#define SIOCBRADDIF 0x89a2 /* add interface to bridge */ 62#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
57#endif 63#endif
64/* We can have up to 256 pages for devices. */
65#define DEVICE_PAGES 256
66/* This fits nicely in a single 4096-byte page. */
67#define VIRTQUEUE_NUM 127
58 68
59/*L:120 verbose is both a global flag and a macro. The C preprocessor allows 69/*L:120 verbose is both a global flag and a macro. The C preprocessor allows
60 * this, and although I wouldn't recommend it, it works quite nicely here. */ 70 * this, and although I wouldn't recommend it, it works quite nicely here. */
@@ -65,8 +75,10 @@ static bool verbose;
65 75
66/* The pipe to send commands to the waker process */ 76/* The pipe to send commands to the waker process */
67static int waker_fd; 77static int waker_fd;
68/* The top of guest physical memory. */ 78/* The pointer to the start of guest memory. */
69static u32 top; 79static void *guest_base;
80/* The maximum guest physical address allowed, and maximum possible. */
81static unsigned long guest_limit, guest_max;
70 82
71/* This is our list of devices. */ 83/* This is our list of devices. */
72struct device_list 84struct device_list
@@ -76,8 +88,17 @@ struct device_list
76 fd_set infds; 88 fd_set infds;
77 int max_infd; 89 int max_infd;
78 90
91 /* Counter to assign interrupt numbers. */
92 unsigned int next_irq;
93
94 /* Counter to print out convenient device numbers. */
95 unsigned int device_num;
96
79 /* The descriptor page for the devices. */ 97 /* The descriptor page for the devices. */
80 struct lguest_device_desc *descs; 98 u8 *descpage;
99
100 /* The tail of the last descriptor. */
101 unsigned int desc_used;
81 102
82 /* A single linked list of devices. */ 103 /* A single linked list of devices. */
83 struct device *dev; 104 struct device *dev;
@@ -85,31 +106,111 @@ struct device_list
85 struct device **lastdev; 106 struct device **lastdev;
86}; 107};
87 108
109/* The list of Guest devices, based on command line arguments. */
110static struct device_list devices;
111
88/* The device structure describes a single device. */ 112/* The device structure describes a single device. */
89struct device 113struct device
90{ 114{
91 /* The linked-list pointer. */ 115 /* The linked-list pointer. */
92 struct device *next; 116 struct device *next;
93 /* The descriptor for this device, as mapped into the Guest. */ 117
118 /* The this device's descriptor, as mapped into the Guest. */
94 struct lguest_device_desc *desc; 119 struct lguest_device_desc *desc;
95 /* The memory page(s) of this device, if any. Also mapped in Guest. */ 120
96 void *mem; 121 /* The name of this device, for --verbose. */
122 const char *name;
97 123
98 /* If handle_input is set, it wants to be called when this file 124 /* If handle_input is set, it wants to be called when this file
99 * descriptor is ready. */ 125 * descriptor is ready. */
100 int fd; 126 int fd;
101 bool (*handle_input)(int fd, struct device *me); 127 bool (*handle_input)(int fd, struct device *me);
102 128
103 /* If handle_output is set, it wants to be called when the Guest sends 129 /* Any queues attached to this device */
104 * DMA to this key. */ 130 struct virtqueue *vq;
105 unsigned long watch_key;
106 u32 (*handle_output)(int fd, const struct iovec *iov,
107 unsigned int num, struct device *me);
108 131
109 /* Device-specific data. */ 132 /* Device-specific data. */
110 void *priv; 133 void *priv;
111}; 134};
112 135
136/* The virtqueue structure describes a queue attached to a device. */
137struct virtqueue
138{
139 struct virtqueue *next;
140
141 /* Which device owns me. */
142 struct device *dev;
143
144 /* The configuration for this queue. */
145 struct lguest_vqconfig config;
146
147 /* The actual ring of buffers. */
148 struct vring vring;
149
150 /* Last available index we saw. */
151 u16 last_avail_idx;
152
153 /* The routine to call when the Guest pings us. */
154 void (*handle_output)(int fd, struct virtqueue *me);
155};
156
157/* Since guest is UP and we don't run at the same time, we don't need barriers.
158 * But I include them in the code in case others copy it. */
159#define wmb()
160
161/* Convert an iovec element to the given type.
162 *
163 * This is a fairly ugly trick: we need to know the size of the type and
164 * alignment requirement to check the pointer is kosher. It's also nice to
165 * have the name of the type in case we report failure.
166 *
167 * Typing those three things all the time is cumbersome and error prone, so we
168 * have a macro which sets them all up and passes to the real function. */
169#define convert(iov, type) \
170 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
171
172static void *_convert(struct iovec *iov, size_t size, size_t align,
173 const char *name)
174{
175 if (iov->iov_len != size)
176 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
177 if ((unsigned long)iov->iov_base % align != 0)
178 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
179 return iov->iov_base;
180}
181
182/* The virtio configuration space is defined to be little-endian. x86 is
183 * little-endian too, but it's nice to be explicit so we have these helpers. */
184#define cpu_to_le16(v16) (v16)
185#define cpu_to_le32(v32) (v32)
186#define cpu_to_le64(v64) (v64)
187#define le16_to_cpu(v16) (v16)
188#define le32_to_cpu(v32) (v32)
189#define le64_to_cpu(v32) (v64)
190
191/*L:100 The Launcher code itself takes us out into userspace, that scary place
192 * where pointers run wild and free! Unfortunately, like most userspace
193 * programs, it's quite boring (which is why everyone likes to hack on the
194 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
195 * will get you through this section. Or, maybe not.
196 *
197 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
198 * memory and stores it in "guest_base". In other words, Guest physical ==
199 * Launcher virtual with an offset.
200 *
201 * This can be tough to get your head around, but usually it just means that we
202 * use these trivial conversion functions when the Guest gives us it's
203 * "physical" addresses: */
204static void *from_guest_phys(unsigned long addr)
205{
206 return guest_base + addr;
207}
208
209static unsigned long to_guest_phys(const void *addr)
210{
211 return (addr - guest_base);
212}
213
113/*L:130 214/*L:130
114 * Loading the Kernel. 215 * Loading the Kernel.
115 * 216 *
@@ -123,43 +224,55 @@ static int open_or_die(const char *name, int flags)
123 return fd; 224 return fd;
124} 225}
125 226
126/* map_zeroed_pages() takes a (page-aligned) address and a number of pages. */ 227/* map_zeroed_pages() takes a number of pages. */
127static void *map_zeroed_pages(unsigned long addr, unsigned int num) 228static void *map_zeroed_pages(unsigned int num)
128{ 229{
129 /* We cache the /dev/zero file-descriptor so we only open it once. */ 230 int fd = open_or_die("/dev/zero", O_RDONLY);
130 static int fd = -1; 231 void *addr;
131
132 if (fd == -1)
133 fd = open_or_die("/dev/zero", O_RDONLY);
134 232
135 /* We use a private mapping (ie. if we write to the page, it will be 233 /* We use a private mapping (ie. if we write to the page, it will be
136 * copied), and obviously we insist that it be mapped where we ask. */ 234 * copied). */
137 if (mmap((void *)addr, getpagesize() * num, 235 addr = mmap(NULL, getpagesize() * num,
138 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_FIXED|MAP_PRIVATE, fd, 0) 236 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
139 != (void *)addr) 237 if (addr == MAP_FAILED)
140 err(1, "Mmaping %u pages of /dev/zero @%p", num, (void *)addr); 238 err(1, "Mmaping %u pages of /dev/zero", num);
141 239
142 /* Returning the address is just a courtesy: can simplify callers. */ 240 return addr;
143 return (void *)addr;
144} 241}
145 242
146/* To find out where to start we look for the magic Guest string, which marks 243/* Get some more pages for a device. */
147 * the code we see in lguest_asm.S. This is a hack which we are currently 244static void *get_pages(unsigned int num)
148 * plotting to replace with the normal Linux entry point. */
149static unsigned long entry_point(void *start, void *end,
150 unsigned long page_offset)
151{ 245{
152 void *p; 246 void *addr = from_guest_phys(guest_limit);
153 247
154 /* The scan gives us the physical starting address. We want the 248 guest_limit += num * getpagesize();
155 * virtual address in this case, and fortunately, we already figured 249 if (guest_limit > guest_max)
156 * out the physical-virtual difference and passed it here in 250 errx(1, "Not enough memory for devices");
157 * "page_offset". */ 251 return addr;
158 for (p = start; p < end; p++) 252}
159 if (memcmp(p, "GenuineLguest", strlen("GenuineLguest")) == 0)
160 return (long)p + strlen("GenuineLguest") + page_offset;
161 253
162 err(1, "Is this image a genuine lguest?"); 254/* This routine is used to load the kernel or initrd. It tries mmap, but if
255 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
256 * it falls back to reading the memory in. */
257static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
258{
259 ssize_t r;
260
261 /* We map writable even though for some segments are marked read-only.
262 * The kernel really wants to be writable: it patches its own
263 * instructions.
264 *
265 * MAP_PRIVATE means that the page won't be copied until a write is
266 * done to it. This allows us to share untouched memory between
267 * Guests. */
268 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
269 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
270 return;
271
272 /* pread does a seek and a read in one shot: saves a few lines. */
273 r = pread(fd, addr, len, offset);
274 if (r != len)
275 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
163} 276}
164 277
165/* This routine takes an open vmlinux image, which is in ELF, and maps it into 278/* This routine takes an open vmlinux image, which is in ELF, and maps it into
@@ -167,19 +280,14 @@ static unsigned long entry_point(void *start, void *end,
167 * by all modern binaries on Linux including the kernel. 280 * by all modern binaries on Linux including the kernel.
168 * 281 *
169 * The ELF headers give *two* addresses: a physical address, and a virtual 282 * The ELF headers give *two* addresses: a physical address, and a virtual
170 * address. The Guest kernel expects to be placed in memory at the physical 283 * address. We use the physical address; the Guest will map itself to the
171 * address, and the page tables set up so it will correspond to that virtual 284 * virtual address.
172 * address. We return the difference between the virtual and physical
173 * addresses in the "page_offset" pointer.
174 * 285 *
175 * We return the starting address. */ 286 * We return the starting address. */
176static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr, 287static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
177 unsigned long *page_offset)
178{ 288{
179 void *addr;
180 Elf32_Phdr phdr[ehdr->e_phnum]; 289 Elf32_Phdr phdr[ehdr->e_phnum];
181 unsigned int i; 290 unsigned int i;
182 unsigned long start = -1UL, end = 0;
183 291
184 /* Sanity checks on the main ELF header: an x86 executable with a 292 /* Sanity checks on the main ELF header: an x86 executable with a
185 * reasonable number of correctly-sized program headers. */ 293 * reasonable number of correctly-sized program headers. */
@@ -199,9 +307,6 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr,
199 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) 307 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
200 err(1, "Reading program headers"); 308 err(1, "Reading program headers");
201 309
202 /* We don't know page_offset yet. */
203 *page_offset = 0;
204
205 /* Try all the headers: there are usually only three. A read-only one, 310 /* Try all the headers: there are usually only three. A read-only one,
206 * a read-write one, and a "note" section which isn't loadable. */ 311 * a read-write one, and a "note" section which isn't loadable. */
207 for (i = 0; i < ehdr->e_phnum; i++) { 312 for (i = 0; i < ehdr->e_phnum; i++) {
@@ -212,158 +317,53 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr,
212 verbose("Section %i: size %i addr %p\n", 317 verbose("Section %i: size %i addr %p\n",
213 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); 318 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
214 319
215 /* We expect a simple linear address space: every segment must 320 /* We map this section of the file at its physical address. */
216 * have the same difference between virtual (p_vaddr) and 321 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
217 * physical (p_paddr) address. */ 322 phdr[i].p_offset, phdr[i].p_filesz);
218 if (!*page_offset)
219 *page_offset = phdr[i].p_vaddr - phdr[i].p_paddr;
220 else if (*page_offset != phdr[i].p_vaddr - phdr[i].p_paddr)
221 errx(1, "Page offset of section %i different", i);
222
223 /* We track the first and last address we mapped, so we can
224 * tell entry_point() where to scan. */
225 if (phdr[i].p_paddr < start)
226 start = phdr[i].p_paddr;
227 if (phdr[i].p_paddr + phdr[i].p_filesz > end)
228 end = phdr[i].p_paddr + phdr[i].p_filesz;
229
230 /* We map this section of the file at its physical address. We
231 * map it read & write even if the header says this segment is
232 * read-only. The kernel really wants to be writable: it
233 * patches its own instructions which would normally be
234 * read-only.
235 *
236 * MAP_PRIVATE means that the page won't be copied until a
237 * write is done to it. This allows us to share much of the
238 * kernel memory between Guests. */
239 addr = mmap((void *)phdr[i].p_paddr,
240 phdr[i].p_filesz,
241 PROT_READ|PROT_WRITE|PROT_EXEC,
242 MAP_FIXED|MAP_PRIVATE,
243 elf_fd, phdr[i].p_offset);
244 if (addr != (void *)phdr[i].p_paddr)
245 err(1, "Mmaping vmlinux seg %i gave %p not %p",
246 i, addr, (void *)phdr[i].p_paddr);
247 } 323 }
248 324
249 return entry_point((void *)start, (void *)end, *page_offset); 325 /* The entry point is given in the ELF header. */
326 return ehdr->e_entry;
250} 327}
251 328
252/*L:170 Prepare to be SHOCKED and AMAZED. And possibly a trifle nauseated. 329/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
253 * 330 * supposed to jump into it and it will unpack itself. We used to have to
254 * We know that CONFIG_PAGE_OFFSET sets what virtual address the kernel expects 331 * perform some hairy magic because the unpacking code scared me.
255 * to be. We don't know what that option was, but we can figure it out
256 * approximately by looking at the addresses in the code. I chose the common
257 * case of reading a memory location into the %eax register:
258 *
259 * movl <some-address>, %eax
260 *
261 * This gets encoded as five bytes: "0xA1 <4-byte-address>". For example,
262 * "0xA1 0x18 0x60 0x47 0xC0" reads the address 0xC0476018 into %eax.
263 *
264 * In this example can guess that the kernel was compiled with
265 * CONFIG_PAGE_OFFSET set to 0xC0000000 (it's always a round number). If the
266 * kernel were larger than 16MB, we might see 0xC1 addresses show up, but our
267 * kernel isn't that bloated yet.
268 *
269 * Unfortunately, x86 has variable-length instructions, so finding this
270 * particular instruction properly involves writing a disassembler. Instead,
271 * we rely on statistics. We look for "0xA1" and tally the different bytes
272 * which occur 4 bytes later (the "0xC0" in our example above). When one of
273 * those bytes appears three times, we can be reasonably confident that it
274 * forms the start of CONFIG_PAGE_OFFSET.
275 * 332 *
276 * This is amazingly reliable. */ 333 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
277static unsigned long intuit_page_offset(unsigned char *img, unsigned long len) 334 * a small patch to jump over the tricky bits in the Guest, so now we just read
335 * the funky header so we know where in the file to load, and away we go! */
336static unsigned long load_bzimage(int fd)
278{ 337{
279 unsigned int i, possibilities[256] = { 0 }; 338 struct boot_params boot;
339 int r;
340 /* Modern bzImages get loaded at 1M. */
341 void *p = from_guest_phys(0x100000);
280 342
281 for (i = 0; i + 4 < len; i++) { 343 /* Go back to the start of the file and read the header. It should be
282 /* mov 0xXXXXXXXX,%eax */ 344 * a Linux boot header (see Documentation/i386/boot.txt) */
283 if (img[i] == 0xA1 && ++possibilities[img[i+4]] > 3) 345 lseek(fd, 0, SEEK_SET);
284 return (unsigned long)img[i+4] << 24; 346 read(fd, &boot, sizeof(boot));
285 }
286 errx(1, "could not determine page offset");
287}
288 347
289/*L:160 Unfortunately the entire ELF image isn't compressed: the segments 348 /* Inside the setup_hdr, we expect the magic "HdrS" */
290 * which need loading are extracted and compressed raw. This denies us the 349 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
291 * information we need to make a fully-general loader. */ 350 errx(1, "This doesn't look like a bzImage to me");
292static unsigned long unpack_bzimage(int fd, unsigned long *page_offset)
293{
294 gzFile f;
295 int ret, len = 0;
296 /* A bzImage always gets loaded at physical address 1M. This is
297 * actually configurable as CONFIG_PHYSICAL_START, but as the comment
298 * there says, "Don't change this unless you know what you are doing".
299 * Indeed. */
300 void *img = (void *)0x100000;
301
302 /* gzdopen takes our file descriptor (carefully placed at the start of
303 * the GZIP header we found) and returns a gzFile. */
304 f = gzdopen(fd, "rb");
305 /* We read it into memory in 64k chunks until we hit the end. */
306 while ((ret = gzread(f, img + len, 65536)) > 0)
307 len += ret;
308 if (ret < 0)
309 err(1, "reading image from bzImage");
310
311 verbose("Unpacked size %i addr %p\n", len, img);
312
313 /* Without the ELF header, we can't tell virtual-physical gap. This is
314 * CONFIG_PAGE_OFFSET, and people do actually change it. Fortunately,
315 * I have a clever way of figuring it out from the code itself. */
316 *page_offset = intuit_page_offset(img, len);
317
318 return entry_point(img, img + len, *page_offset);
319}
320 351
321/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're 352 /* Skip over the extra sectors of the header. */
322 * supposed to jump into it and it will unpack itself. We can't do that 353 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
323 * because the Guest can't run the unpacking code, and adding features to 354
324 * lguest kills puppies, so we don't want to. 355 /* Now read everything into memory. in nice big chunks. */
325 * 356 while ((r = read(fd, p, 65536)) > 0)
326 * The bzImage is formed by putting the decompressing code in front of the 357 p += r;
327 * compressed kernel code. So we can simple scan through it looking for the 358
328 * first "gzip" header, and start decompressing from there. */ 359 /* Finally, code32_start tells us where to enter the kernel. */
329static unsigned long load_bzimage(int fd, unsigned long *page_offset) 360 return boot.hdr.code32_start;
330{
331 unsigned char c;
332 int state = 0;
333
334 /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */
335 while (read(fd, &c, 1) == 1) {
336 switch (state) {
337 case 0:
338 if (c == 0x1F)
339 state++;
340 break;
341 case 1:
342 if (c == 0x8B)
343 state++;
344 else
345 state = 0;
346 break;
347 case 2 ... 8:
348 state++;
349 break;
350 case 9:
351 /* Seek back to the start of the gzip header. */
352 lseek(fd, -10, SEEK_CUR);
353 /* One final check: "compressed under UNIX". */
354 if (c != 0x03)
355 state = -1;
356 else
357 return unpack_bzimage(fd, page_offset);
358 }
359 }
360 errx(1, "Could not find kernel in bzImage");
361} 361}
362 362
363/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels 363/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
364 * come wrapped up in the self-decompressing "bzImage" format. With some funky 364 * come wrapped up in the self-decompressing "bzImage" format. With some funky
365 * coding, we can load those, too. */ 365 * coding, we can load those, too. */
366static unsigned long load_kernel(int fd, unsigned long *page_offset) 366static unsigned long load_kernel(int fd)
367{ 367{
368 Elf32_Ehdr hdr; 368 Elf32_Ehdr hdr;
369 369
@@ -373,10 +373,10 @@ static unsigned long load_kernel(int fd, unsigned long *page_offset)
373 373
374 /* If it's an ELF file, it starts with "\177ELF" */ 374 /* If it's an ELF file, it starts with "\177ELF" */
375 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) 375 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
376 return map_elf(fd, &hdr, page_offset); 376 return map_elf(fd, &hdr);
377 377
378 /* Otherwise we assume it's a bzImage, and try to unpack it */ 378 /* Otherwise we assume it's a bzImage, and try to unpack it */
379 return load_bzimage(fd, page_offset); 379 return load_bzimage(fd);
380} 380}
381 381
382/* This is a trivial little helper to align pages. Andi Kleen hated it because 382/* This is a trivial little helper to align pages. Andi Kleen hated it because
@@ -402,59 +402,45 @@ static unsigned long load_initrd(const char *name, unsigned long mem)
402 int ifd; 402 int ifd;
403 struct stat st; 403 struct stat st;
404 unsigned long len; 404 unsigned long len;
405 void *iaddr;
406 405
407 ifd = open_or_die(name, O_RDONLY); 406 ifd = open_or_die(name, O_RDONLY);
408 /* fstat() is needed to get the file size. */ 407 /* fstat() is needed to get the file size. */
409 if (fstat(ifd, &st) < 0) 408 if (fstat(ifd, &st) < 0)
410 err(1, "fstat() on initrd '%s'", name); 409 err(1, "fstat() on initrd '%s'", name);
411 410
412 /* The length needs to be rounded up to a page size: mmap needs the 411 /* We map the initrd at the top of memory, but mmap wants it to be
413 * address to be page aligned. */ 412 * page-aligned, so we round the size up for that. */
414 len = page_align(st.st_size); 413 len = page_align(st.st_size);
415 /* We map the initrd at the top of memory. */ 414 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
416 iaddr = mmap((void *)mem - len, st.st_size,
417 PROT_READ|PROT_EXEC|PROT_WRITE,
418 MAP_FIXED|MAP_PRIVATE, ifd, 0);
419 if (iaddr != (void *)mem - len)
420 err(1, "Mmaping initrd '%s' returned %p not %p",
421 name, iaddr, (void *)mem - len);
422 /* Once a file is mapped, you can close the file descriptor. It's a 415 /* Once a file is mapped, you can close the file descriptor. It's a
423 * little odd, but quite useful. */ 416 * little odd, but quite useful. */
424 close(ifd); 417 close(ifd);
425 verbose("mapped initrd %s size=%lu @ %p\n", name, st.st_size, iaddr); 418 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
426 419
427 /* We return the initrd size. */ 420 /* We return the initrd size. */
428 return len; 421 return len;
429} 422}
430 423
431/* Once we know how much memory we have, and the address the Guest kernel 424/* Once we know how much memory we have, we can construct simple linear page
432 * expects, we can construct simple linear page tables which will get the Guest 425 * tables which set virtual == physical which will get the Guest far enough
433 * far enough into the boot to create its own. 426 * into the boot to create its own.
434 * 427 *
435 * We lay them out of the way, just below the initrd (which is why we need to 428 * We lay them out of the way, just below the initrd (which is why we need to
436 * know its size). */ 429 * know its size). */
437static unsigned long setup_pagetables(unsigned long mem, 430static unsigned long setup_pagetables(unsigned long mem,
438 unsigned long initrd_size, 431 unsigned long initrd_size)
439 unsigned long page_offset)
440{ 432{
441 u32 *pgdir, *linear; 433 unsigned long *pgdir, *linear;
442 unsigned int mapped_pages, i, linear_pages; 434 unsigned int mapped_pages, i, linear_pages;
443 unsigned int ptes_per_page = getpagesize()/sizeof(u32); 435 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
444 436
445 /* Ideally we map all physical memory starting at page_offset. 437 mapped_pages = mem/getpagesize();
446 * However, if page_offset is 0xC0000000 we can only map 1G of physical
447 * (0xC0000000 + 1G overflows). */
448 if (mem <= -page_offset)
449 mapped_pages = mem/getpagesize();
450 else
451 mapped_pages = -page_offset/getpagesize();
452 438
453 /* Each PTE page can map ptes_per_page pages: how many do we need? */ 439 /* Each PTE page can map ptes_per_page pages: how many do we need? */
454 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; 440 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
455 441
456 /* We put the toplevel page directory page at the top of memory. */ 442 /* We put the toplevel page directory page at the top of memory. */
457 pgdir = (void *)mem - initrd_size - getpagesize(); 443 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
458 444
459 /* Now we use the next linear_pages pages as pte pages */ 445 /* Now we use the next linear_pages pages as pte pages */
460 linear = (void *)pgdir - linear_pages*getpagesize(); 446 linear = (void *)pgdir - linear_pages*getpagesize();
@@ -465,20 +451,19 @@ static unsigned long setup_pagetables(unsigned long mem,
465 for (i = 0; i < mapped_pages; i++) 451 for (i = 0; i < mapped_pages; i++)
466 linear[i] = ((i * getpagesize()) | PAGE_PRESENT); 452 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
467 453
468 /* The top level points to the linear page table pages above. The 454 /* The top level points to the linear page table pages above. */
469 * entry representing page_offset points to the first one, and they
470 * continue from there. */
471 for (i = 0; i < mapped_pages; i += ptes_per_page) { 455 for (i = 0; i < mapped_pages; i += ptes_per_page) {
472 pgdir[(i + page_offset/getpagesize())/ptes_per_page] 456 pgdir[i/ptes_per_page]
473 = (((u32)linear + i*sizeof(u32)) | PAGE_PRESENT); 457 = ((to_guest_phys(linear) + i*sizeof(void *))
458 | PAGE_PRESENT);
474 } 459 }
475 460
476 verbose("Linear mapping of %u pages in %u pte pages at %p\n", 461 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
477 mapped_pages, linear_pages, linear); 462 mapped_pages, linear_pages, to_guest_phys(linear));
478 463
479 /* We return the top level (guest-physical) address: the kernel needs 464 /* We return the top level (guest-physical) address: the kernel needs
480 * to know where it is. */ 465 * to know where it is. */
481 return (unsigned long)pgdir; 466 return to_guest_phys(pgdir);
482} 467}
483 468
484/* Simple routine to roll all the commandline arguments together with spaces 469/* Simple routine to roll all the commandline arguments together with spaces
@@ -498,14 +483,17 @@ static void concat(char *dst, char *args[])
498 483
499/* This is where we actually tell the kernel to initialize the Guest. We saw 484/* This is where we actually tell the kernel to initialize the Guest. We saw
500 * the arguments it expects when we looked at initialize() in lguest_user.c: 485 * the arguments it expects when we looked at initialize() in lguest_user.c:
501 * the top physical page to allow, the top level pagetable, the entry point and 486 * the base of guest "physical" memory, the top physical page to allow, the
502 * the page_offset constant for the Guest. */ 487 * top level pagetable and the entry point for the Guest. */
503static int tell_kernel(u32 pgdir, u32 start, u32 page_offset) 488static int tell_kernel(unsigned long pgdir, unsigned long start)
504{ 489{
505 u32 args[] = { LHREQ_INITIALIZE, 490 unsigned long args[] = { LHREQ_INITIALIZE,
506 top/getpagesize(), pgdir, start, page_offset }; 491 (unsigned long)guest_base,
492 guest_limit / getpagesize(), pgdir, start };
507 int fd; 493 int fd;
508 494
495 verbose("Guest: %p - %p (%#lx)\n",
496 guest_base, guest_base + guest_limit, guest_limit);
509 fd = open_or_die("/dev/lguest", O_RDWR); 497 fd = open_or_die("/dev/lguest", O_RDWR);
510 if (write(fd, args, sizeof(args)) < 0) 498 if (write(fd, args, sizeof(args)) < 0)
511 err(1, "Writing to /dev/lguest"); 499 err(1, "Writing to /dev/lguest");
@@ -515,11 +503,11 @@ static int tell_kernel(u32 pgdir, u32 start, u32 page_offset)
515} 503}
516/*:*/ 504/*:*/
517 505
518static void set_fd(int fd, struct device_list *devices) 506static void add_device_fd(int fd)
519{ 507{
520 FD_SET(fd, &devices->infds); 508 FD_SET(fd, &devices.infds);
521 if (fd > devices->max_infd) 509 if (fd > devices.max_infd)
522 devices->max_infd = fd; 510 devices.max_infd = fd;
523} 511}
524 512
525/*L:200 513/*L:200
@@ -537,36 +525,38 @@ static void set_fd(int fd, struct device_list *devices)
537 * 525 *
538 * This, of course, is merely a different *kind* of icky. 526 * This, of course, is merely a different *kind* of icky.
539 */ 527 */
540static void wake_parent(int pipefd, int lguest_fd, struct device_list *devices) 528static void wake_parent(int pipefd, int lguest_fd)
541{ 529{
542 /* Add the pipe from the Launcher to the fdset in the device_list, so 530 /* Add the pipe from the Launcher to the fdset in the device_list, so
543 * we watch it, too. */ 531 * we watch it, too. */
544 set_fd(pipefd, devices); 532 add_device_fd(pipefd);
545 533
546 for (;;) { 534 for (;;) {
547 fd_set rfds = devices->infds; 535 fd_set rfds = devices.infds;
548 u32 args[] = { LHREQ_BREAK, 1 }; 536 unsigned long args[] = { LHREQ_BREAK, 1 };
549 537
550 /* Wait until input is ready from one of the devices. */ 538 /* Wait until input is ready from one of the devices. */
551 select(devices->max_infd+1, &rfds, NULL, NULL, NULL); 539 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
552 /* Is it a message from the Launcher? */ 540 /* Is it a message from the Launcher? */
553 if (FD_ISSET(pipefd, &rfds)) { 541 if (FD_ISSET(pipefd, &rfds)) {
554 int ignorefd; 542 int fd;
555 /* If read() returns 0, it means the Launcher has 543 /* If read() returns 0, it means the Launcher has
556 * exited. We silently follow. */ 544 * exited. We silently follow. */
557 if (read(pipefd, &ignorefd, sizeof(ignorefd)) == 0) 545 if (read(pipefd, &fd, sizeof(fd)) == 0)
558 exit(0); 546 exit(0);
559 /* Otherwise it's telling us there's a problem with one 547 /* Otherwise it's telling us to change what file
560 * of the devices, and we should ignore that file 548 * descriptors we're to listen to. */
561 * descriptor from now on. */ 549 if (fd >= 0)
562 FD_CLR(ignorefd, &devices->infds); 550 FD_SET(fd, &devices.infds);
551 else
552 FD_CLR(-fd - 1, &devices.infds);
563 } else /* Send LHREQ_BREAK command. */ 553 } else /* Send LHREQ_BREAK command. */
564 write(lguest_fd, args, sizeof(args)); 554 write(lguest_fd, args, sizeof(args));
565 } 555 }
566} 556}
567 557
568/* This routine just sets up a pipe to the Waker process. */ 558/* This routine just sets up a pipe to the Waker process. */
569static int setup_waker(int lguest_fd, struct device_list *device_list) 559static int setup_waker(int lguest_fd)
570{ 560{
571 int pipefd[2], child; 561 int pipefd[2], child;
572 562
@@ -580,7 +570,7 @@ static int setup_waker(int lguest_fd, struct device_list *device_list)
580 if (child == 0) { 570 if (child == 0) {
581 /* Close the "writing" end of our copy of the pipe */ 571 /* Close the "writing" end of our copy of the pipe */
582 close(pipefd[1]); 572 close(pipefd[1]);
583 wake_parent(pipefd[0], lguest_fd, device_list); 573 wake_parent(pipefd[0], lguest_fd);
584 } 574 }
585 /* Close the reading end of our copy of the pipe. */ 575 /* Close the reading end of our copy of the pipe. */
586 close(pipefd[0]); 576 close(pipefd[0]);
@@ -602,83 +592,128 @@ static void *_check_pointer(unsigned long addr, unsigned int size,
602{ 592{
603 /* We have to separately check addr and addr+size, because size could 593 /* We have to separately check addr and addr+size, because size could
604 * be huge and addr + size might wrap around. */ 594 * be huge and addr + size might wrap around. */
605 if (addr >= top || addr + size >= top) 595 if (addr >= guest_limit || addr + size >= guest_limit)
606 errx(1, "%s:%i: Invalid address %li", __FILE__, line, addr); 596 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
607 /* We return a pointer for the caller's convenience, now we know it's 597 /* We return a pointer for the caller's convenience, now we know it's
608 * safe to use. */ 598 * safe to use. */
609 return (void *)addr; 599 return from_guest_phys(addr);
610} 600}
611/* A macro which transparently hands the line number to the real function. */ 601/* A macro which transparently hands the line number to the real function. */
612#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) 602#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
613 603
614/* The Guest has given us the address of a "struct lguest_dma". We check it's 604/* This function returns the next descriptor in the chain, or vq->vring.num. */
615 * OK and convert it to an iovec (which is a simple array of ptr/size 605static unsigned next_desc(struct virtqueue *vq, unsigned int i)
616 * pairs). */
617static u32 *dma2iov(unsigned long dma, struct iovec iov[], unsigned *num)
618{ 606{
619 unsigned int i; 607 unsigned int next;
620 struct lguest_dma *udma;
621
622 /* First we make sure that the array memory itself is valid. */
623 udma = check_pointer(dma, sizeof(*udma));
624 /* Now we check each element */
625 for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) {
626 /* A zero length ends the array. */
627 if (!udma->len[i])
628 break;
629 608
630 iov[i].iov_base = check_pointer(udma->addr[i], udma->len[i]); 609 /* If this descriptor says it doesn't chain, we're done. */
631 iov[i].iov_len = udma->len[i]; 610 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
632 } 611 return vq->vring.num;
633 *num = i; 612
613 /* Check they're not leading us off end of descriptors. */
614 next = vq->vring.desc[i].next;
615 /* Make sure compiler knows to grab that: we don't want it changing! */
616 wmb();
634 617
635 /* We return the pointer to where the caller should write the amount of 618 if (next >= vq->vring.num)
636 * the buffer used. */ 619 errx(1, "Desc next is %u", next);
637 return &udma->used_len; 620
621 return next;
622}
623
624/* This looks in the virtqueue and for the first available buffer, and converts
625 * it to an iovec for convenient access. Since descriptors consist of some
626 * number of output then some number of input descriptors, it's actually two
627 * iovecs, but we pack them into one and note how many of each there were.
628 *
629 * This function returns the descriptor number found, or vq->vring.num (which
630 * is never a valid descriptor number) if none was found. */
631static unsigned get_vq_desc(struct virtqueue *vq,
632 struct iovec iov[],
633 unsigned int *out_num, unsigned int *in_num)
634{
635 unsigned int i, head;
636
637 /* Check it isn't doing very strange things with descriptor numbers. */
638 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
639 errx(1, "Guest moved used index from %u to %u",
640 vq->last_avail_idx, vq->vring.avail->idx);
641
642 /* If there's nothing new since last we looked, return invalid. */
643 if (vq->vring.avail->idx == vq->last_avail_idx)
644 return vq->vring.num;
645
646 /* Grab the next descriptor number they're advertising, and increment
647 * the index we've seen. */
648 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
649
650 /* If their number is silly, that's a fatal mistake. */
651 if (head >= vq->vring.num)
652 errx(1, "Guest says index %u is available", head);
653
654 /* When we start there are none of either input nor output. */
655 *out_num = *in_num = 0;
656
657 i = head;
658 do {
659 /* Grab the first descriptor, and check it's OK. */
660 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
661 iov[*out_num + *in_num].iov_base
662 = check_pointer(vq->vring.desc[i].addr,
663 vq->vring.desc[i].len);
664 /* If this is an input descriptor, increment that count. */
665 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
666 (*in_num)++;
667 else {
668 /* If it's an output descriptor, they're all supposed
669 * to come before any input descriptors. */
670 if (*in_num)
671 errx(1, "Descriptor has out after in");
672 (*out_num)++;
673 }
674
675 /* If we've got too many, that implies a descriptor loop. */
676 if (*out_num + *in_num > vq->vring.num)
677 errx(1, "Looped descriptor");
678 } while ((i = next_desc(vq, i)) != vq->vring.num);
679
680 return head;
638} 681}
639 682
640/* This routine gets a DMA buffer from the Guest for a given key, and converts 683/* Once we've used one of their buffers, we tell them about it. We'll then
641 * it to an iovec array. It returns the interrupt the Guest wants when we're 684 * want to send them an interrupt, using trigger_irq(). */
642 * finished, and a pointer to the "used_len" field to fill in. */ 685static void add_used(struct virtqueue *vq, unsigned int head, int len)
643static u32 *get_dma_buffer(int fd, void *key,
644 struct iovec iov[], unsigned int *num, u32 *irq)
645{ 686{
646 u32 buf[] = { LHREQ_GETDMA, (u32)key }; 687 struct vring_used_elem *used;
647 unsigned long udma; 688
648 u32 *res; 689 /* Get a pointer to the next entry in the used ring. */
649 690 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
650 /* Ask the kernel for a DMA buffer corresponding to this key. */ 691 used->id = head;
651 udma = write(fd, buf, sizeof(buf)); 692 used->len = len;
652 /* They haven't registered any, or they're all used? */ 693 /* Make sure buffer is written before we update index. */
653 if (udma == (unsigned long)-1) 694 wmb();
654 return NULL; 695 vq->vring.used->idx++;
655
656 /* Convert it into our iovec array */
657 res = dma2iov(udma, iov, num);
658 /* The kernel stashes irq in ->used_len to get it out to us. */
659 *irq = *res;
660 /* Return a pointer to ((struct lguest_dma *)udma)->used_len. */
661 return res;
662} 696}
663 697
664/* This is a convenient routine to send the Guest an interrupt. */ 698/* This actually sends the interrupt for this virtqueue */
665static void trigger_irq(int fd, u32 irq) 699static void trigger_irq(int fd, struct virtqueue *vq)
666{ 700{
667 u32 buf[] = { LHREQ_IRQ, irq }; 701 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
702
703 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
704 return;
705
706 /* Send the Guest an interrupt tell them we used something up. */
668 if (write(fd, buf, sizeof(buf)) != 0) 707 if (write(fd, buf, sizeof(buf)) != 0)
669 err(1, "Triggering irq %i", irq); 708 err(1, "Triggering irq %i", vq->config.irq);
670} 709}
671 710
672/* This simply sets up an iovec array where we can put data to be discarded. 711/* And here's the combo meal deal. Supersize me! */
673 * This happens when the Guest doesn't want or can't handle the input: we have 712static void add_used_and_trigger(int fd, struct virtqueue *vq,
674 * to get rid of it somewhere, and if we bury it in the ceiling space it will 713 unsigned int head, int len)
675 * start to smell after a week. */
676static void discard_iovec(struct iovec *iov, unsigned int *num)
677{ 714{
678 static char discard_buf[1024]; 715 add_used(vq, head, len);
679 *num = 1; 716 trigger_irq(fd, vq);
680 iov->iov_base = discard_buf;
681 iov->iov_len = sizeof(discard_buf);
682} 717}
683 718
684/* Here is the input terminal setting we save, and the routine to restore them 719/* Here is the input terminal setting we save, and the routine to restore them
@@ -701,38 +736,39 @@ struct console_abort
701/* This is the routine which handles console input (ie. stdin). */ 736/* This is the routine which handles console input (ie. stdin). */
702static bool handle_console_input(int fd, struct device *dev) 737static bool handle_console_input(int fd, struct device *dev)
703{ 738{
704 u32 irq = 0, *lenp;
705 int len; 739 int len;
706 unsigned int num; 740 unsigned int head, in_num, out_num;
707 struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; 741 struct iovec iov[dev->vq->vring.num];
708 struct console_abort *abort = dev->priv; 742 struct console_abort *abort = dev->priv;
709 743
710 /* First we get the console buffer from the Guest. The key is dev->mem 744 /* First we need a console buffer from the Guests's input virtqueue. */
711 * which was set to 0 in setup_console(). */ 745 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
712 lenp = get_dma_buffer(fd, dev->mem, iov, &num, &irq); 746
713 if (!lenp) { 747 /* If they're not ready for input, stop listening to this file
714 /* If it's not ready for input, warn and set up to discard. */ 748 * descriptor. We'll start again once they add an input buffer. */
715 warn("console: no dma buffer!"); 749 if (head == dev->vq->vring.num)
716 discard_iovec(iov, &num); 750 return false;
717 } 751
752 if (out_num)
753 errx(1, "Output buffers in console in queue?");
718 754
719 /* This is why we convert to iovecs: the readv() call uses them, and so 755 /* This is why we convert to iovecs: the readv() call uses them, and so
720 * it reads straight into the Guest's buffer. */ 756 * it reads straight into the Guest's buffer. */
721 len = readv(dev->fd, iov, num); 757 len = readv(dev->fd, iov, in_num);
722 if (len <= 0) { 758 if (len <= 0) {
723 /* This implies that the console is closed, is /dev/null, or 759 /* This implies that the console is closed, is /dev/null, or
724 * something went terribly wrong. We still go through the rest 760 * something went terribly wrong. */
725 * of the logic, though, especially the exit handling below. */
726 warnx("Failed to get console input, ignoring console."); 761 warnx("Failed to get console input, ignoring console.");
727 len = 0; 762 /* Put the input terminal back. */
763 restore_term();
764 /* Remove callback from input vq, so it doesn't restart us. */
765 dev->vq->handle_output = NULL;
766 /* Stop listening to this fd: don't call us again. */
767 return false;
728 } 768 }
729 769
730 /* If we read the data into the Guest, fill in the length and send the 770 /* Tell the Guest about the new input. */
731 * interrupt. */ 771 add_used_and_trigger(fd, dev->vq, head, len);
732 if (lenp) {
733 *lenp = len;
734 trigger_irq(fd, irq);
735 }
736 772
737 /* Three ^C within one second? Exit. 773 /* Three ^C within one second? Exit.
738 * 774 *
@@ -746,7 +782,7 @@ static bool handle_console_input(int fd, struct device *dev)
746 struct timeval now; 782 struct timeval now;
747 gettimeofday(&now, NULL); 783 gettimeofday(&now, NULL);
748 if (now.tv_sec <= abort->start.tv_sec+1) { 784 if (now.tv_sec <= abort->start.tv_sec+1) {
749 u32 args[] = { LHREQ_BREAK, 0 }; 785 unsigned long args[] = { LHREQ_BREAK, 0 };
750 /* Close the fd so Waker will know it has to 786 /* Close the fd so Waker will know it has to
751 * exit. */ 787 * exit. */
752 close(waker_fd); 788 close(waker_fd);
@@ -761,214 +797,163 @@ static bool handle_console_input(int fd, struct device *dev)
761 /* Any other key resets the abort counter. */ 797 /* Any other key resets the abort counter. */
762 abort->count = 0; 798 abort->count = 0;
763 799
764 /* Now, if we didn't read anything, put the input terminal back and
765 * return failure (meaning, don't call us again). */
766 if (!len) {
767 restore_term();
768 return false;
769 }
770 /* Everything went OK! */ 800 /* Everything went OK! */
771 return true; 801 return true;
772} 802}
773 803
774/* Handling console output is much simpler than input. */ 804/* Handling output for console is simple: we just get all the output buffers
775static u32 handle_console_output(int fd, const struct iovec *iov, 805 * and write them to stdout. */
776 unsigned num, struct device*dev) 806static void handle_console_output(int fd, struct virtqueue *vq)
777{ 807{
778 /* Whatever the Guest sends, write it to standard output. Return the 808 unsigned int head, out, in;
779 * number of bytes written. */ 809 int len;
780 return writev(STDOUT_FILENO, iov, num); 810 struct iovec iov[vq->vring.num];
781} 811
782 812 /* Keep getting output buffers from the Guest until we run out. */
783/* Guest->Host network output is also pretty easy. */ 813 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
784static u32 handle_tun_output(int fd, const struct iovec *iov, 814 if (in)
785 unsigned num, struct device *dev) 815 errx(1, "Input buffers in output queue?");
786{ 816 len = writev(STDOUT_FILENO, iov, out);
787 /* We put a flag in the "priv" pointer of the network device, and set 817 add_used_and_trigger(fd, vq, head, len);
788 * it as soon as we see output. We'll see why in handle_tun_input() */ 818 }
789 *(bool *)dev->priv = true;
790 /* Whatever packet the Guest sent us, write it out to the tun
791 * device. */
792 return writev(dev->fd, iov, num);
793} 819}
794 820
795/* This matches the peer_key() in lguest_net.c. The key for any given slot 821/* Handling output for network is also simple: we get all the output buffers
796 * is the address of the network device's page plus 4 * the slot number. */ 822 * and write them (ignoring the first element) to this device's file descriptor
797static unsigned long peer_offset(unsigned int peernum) 823 * (stdout). */
824static void handle_net_output(int fd, struct virtqueue *vq)
798{ 825{
799 return 4 * peernum; 826 unsigned int head, out, in;
827 int len;
828 struct iovec iov[vq->vring.num];
829
830 /* Keep getting output buffers from the Guest until we run out. */
831 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
832 if (in)
833 errx(1, "Input buffers in output queue?");
834 /* Check header, but otherwise ignore it (we said we supported
835 * no features). */
836 (void)convert(&iov[0], struct virtio_net_hdr);
837 len = writev(vq->dev->fd, iov+1, out-1);
838 add_used_and_trigger(fd, vq, head, len);
839 }
800} 840}
801 841
802/* This is where we handle a packet coming in from the tun device */ 842/* This is where we handle a packet coming in from the tun device to our
843 * Guest. */
803static bool handle_tun_input(int fd, struct device *dev) 844static bool handle_tun_input(int fd, struct device *dev)
804{ 845{
805 u32 irq = 0, *lenp; 846 unsigned int head, in_num, out_num;
806 int len; 847 int len;
807 unsigned num; 848 struct iovec iov[dev->vq->vring.num];
808 struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; 849 struct virtio_net_hdr *hdr;
809 850
810 /* First we get a buffer the Guest has bound to its key. */ 851 /* First we need a network buffer from the Guests's recv virtqueue. */
811 lenp = get_dma_buffer(fd, dev->mem+peer_offset(NET_PEERNUM), iov, &num, 852 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
812 &irq); 853 if (head == dev->vq->vring.num) {
813 if (!lenp) {
814 /* Now, it's expected that if we try to send a packet too 854 /* Now, it's expected that if we try to send a packet too
815 * early, the Guest won't be ready yet. This is why we set a 855 * early, the Guest won't be ready yet. Wait until the device
816 * flag when the Guest sends its first packet. If it's sent a 856 * status says it's ready. */
817 * packet we assume it should be ready to receive them. 857 /* FIXME: Actually want DRIVER_ACTIVE here. */
818 * 858 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
819 * Actually, this is what the status bits in the descriptor are
820 * for: we should *use* them. FIXME! */
821 if (*(bool *)dev->priv)
822 warn("network: no dma buffer!"); 859 warn("network: no dma buffer!");
823 discard_iovec(iov, &num); 860 /* We'll turn this back on if input buffers are registered. */
824 } 861 return false;
862 } else if (out_num)
863 errx(1, "Output buffers in network recv queue?");
864
865 /* First element is the header: we set it to 0 (no features). */
866 hdr = convert(&iov[0], struct virtio_net_hdr);
867 hdr->flags = 0;
868 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
825 869
826 /* Read the packet from the device directly into the Guest's buffer. */ 870 /* Read the packet from the device directly into the Guest's buffer. */
827 len = readv(dev->fd, iov, num); 871 len = readv(dev->fd, iov+1, in_num-1);
828 if (len <= 0) 872 if (len <= 0)
829 err(1, "reading network"); 873 err(1, "reading network");
830 874
831 /* Write the used_len, and trigger the interrupt for the Guest */ 875 /* Tell the Guest about the new packet. */
832 if (lenp) { 876 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
833 *lenp = len; 877
834 trigger_irq(fd, irq);
835 }
836 verbose("tun input packet len %i [%02x %02x] (%s)\n", len, 878 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
837 ((u8 *)iov[0].iov_base)[0], ((u8 *)iov[0].iov_base)[1], 879 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
838 lenp ? "sent" : "discarded"); 880 head != dev->vq->vring.num ? "sent" : "discarded");
881
839 /* All good. */ 882 /* All good. */
840 return true; 883 return true;
841} 884}
842 885
843/* The last device handling routine is block output: the Guest has sent a DMA 886/* This callback ensures we try again, in case we stopped console or net
844 * to the block device. It will have placed the command it wants in the 887 * delivery because Guest didn't have any buffers. */
845 * "struct lguest_block_page". */ 888static void enable_fd(int fd, struct virtqueue *vq)
846static u32 handle_block_output(int fd, const struct iovec *iov,
847 unsigned num, struct device *dev)
848{ 889{
849 struct lguest_block_page *p = dev->mem; 890 add_device_fd(vq->dev->fd);
850 u32 irq, *lenp; 891 /* Tell waker to listen to it again */
851 unsigned int len, reply_num; 892 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
852 struct iovec reply[LGUEST_MAX_DMA_SECTIONS];
853 off64_t device_len, off = (off64_t)p->sector * 512;
854
855 /* First we extract the device length from the dev->priv pointer. */
856 device_len = *(off64_t *)dev->priv;
857
858 /* We first check that the read or write is within the length of the
859 * block file. */
860 if (off >= device_len)
861 err(1, "Bad offset %llu vs %llu", off, device_len);
862 /* Move to the right location in the block file. This shouldn't fail,
863 * but best to check. */
864 if (lseek64(dev->fd, off, SEEK_SET) != off)
865 err(1, "Bad seek to sector %i", p->sector);
866
867 verbose("Block: %s at offset %llu\n", p->type ? "WRITE" : "READ", off);
868
869 /* They were supposed to bind a reply buffer at key equal to the start
870 * of the block device memory. We need this to tell them when the
871 * request is finished. */
872 lenp = get_dma_buffer(fd, dev->mem, reply, &reply_num, &irq);
873 if (!lenp)
874 err(1, "Block request didn't give us a dma buffer");
875
876 if (p->type) {
877 /* A write request. The DMA they sent contained the data, so
878 * write it out. */
879 len = writev(dev->fd, iov, num);
880 /* Grr... Now we know how long the "struct lguest_dma" they
881 * sent was, we make sure they didn't try to write over the end
882 * of the block file (possibly extending it). */
883 if (off + len > device_len) {
884 /* Trim it back to the correct length */
885 ftruncate64(dev->fd, device_len);
886 /* Die, bad Guest, die. */
887 errx(1, "Write past end %llu+%u", off, len);
888 }
889 /* The reply length is 0: we just send back an empty DMA to
890 * interrupt them and tell them the write is finished. */
891 *lenp = 0;
892 } else {
893 /* A read request. They sent an empty DMA to start the
894 * request, and we put the read contents into the reply
895 * buffer. */
896 len = readv(dev->fd, reply, reply_num);
897 *lenp = len;
898 }
899
900 /* The result is 1 (done), 2 if there was an error (short read or
901 * write). */
902 p->result = 1 + (p->bytes != len);
903 /* Now tell them we've used their reply buffer. */
904 trigger_irq(fd, irq);
905
906 /* We're supposed to return the number of bytes of the output buffer we
907 * used. But the block device uses the "result" field instead, so we
908 * don't bother. */
909 return 0;
910} 893}
911 894
912/* This is the generic routine we call when the Guest sends some DMA out. */ 895/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
913static void handle_output(int fd, unsigned long dma, unsigned long key, 896static void handle_output(int fd, unsigned long addr)
914 struct device_list *devices)
915{ 897{
916 struct device *i; 898 struct device *i;
917 u32 *lenp; 899 struct virtqueue *vq;
918 struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; 900
919 unsigned num = 0; 901 /* Check each virtqueue. */
920 902 for (i = devices.dev; i; i = i->next) {
921 /* Convert the "struct lguest_dma" they're sending to a "struct 903 for (vq = i->vq; vq; vq = vq->next) {
922 * iovec". */ 904 if (vq->config.pfn == addr/getpagesize()
923 lenp = dma2iov(dma, iov, &num); 905 && vq->handle_output) {
924 906 verbose("Output to %s\n", vq->dev->name);
925 /* Check each device: if they expect output to this key, tell them to 907 vq->handle_output(fd, vq);
926 * handle it. */ 908 return;
927 for (i = devices->dev; i; i = i->next) { 909 }
928 if (i->handle_output && key == i->watch_key) {
929 /* We write the result straight into the used_len field
930 * for them. */
931 *lenp = i->handle_output(fd, iov, num, i);
932 return;
933 } 910 }
934 } 911 }
935 912
936 /* This can happen: the kernel sends any SEND_DMA which doesn't match 913 /* Early console write is done using notify on a nul-terminated string
937 * another Guest to us. It could be that another Guest just left a 914 * in Guest memory. */
938 * network, for example. But it's unusual. */ 915 if (addr >= guest_limit)
939 warnx("Pending dma %p, key %p", (void *)dma, (void *)key); 916 errx(1, "Bad NOTIFY %#lx", addr);
917
918 write(STDOUT_FILENO, from_guest_phys(addr),
919 strnlen(from_guest_phys(addr), guest_limit - addr));
940} 920}
941 921
942/* This is called when the waker wakes us up: check for incoming file 922/* This is called when the waker wakes us up: check for incoming file
943 * descriptors. */ 923 * descriptors. */
944static void handle_input(int fd, struct device_list *devices) 924static void handle_input(int fd)
945{ 925{
946 /* select() wants a zeroed timeval to mean "don't wait". */ 926 /* select() wants a zeroed timeval to mean "don't wait". */
947 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; 927 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
948 928
949 for (;;) { 929 for (;;) {
950 struct device *i; 930 struct device *i;
951 fd_set fds = devices->infds; 931 fd_set fds = devices.infds;
952 932
953 /* If nothing is ready, we're done. */ 933 /* If nothing is ready, we're done. */
954 if (select(devices->max_infd+1, &fds, NULL, NULL, &poll) == 0) 934 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
955 break; 935 break;
956 936
957 /* Otherwise, call the device(s) which have readable 937 /* Otherwise, call the device(s) which have readable
958 * file descriptors and a method of handling them. */ 938 * file descriptors and a method of handling them. */
959 for (i = devices->dev; i; i = i->next) { 939 for (i = devices.dev; i; i = i->next) {
960 if (i->handle_input && FD_ISSET(i->fd, &fds)) { 940 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
941 int dev_fd;
942 if (i->handle_input(fd, i))
943 continue;
944
961 /* If handle_input() returns false, it means we 945 /* If handle_input() returns false, it means we
962 * should no longer service it. 946 * should no longer service it. Networking and
963 * handle_console_input() does this. */ 947 * console do this when there's no input
964 if (!i->handle_input(fd, i)) { 948 * buffers to deliver into. Console also uses
965 /* Clear it from the set of input file 949 * it when it discovers that stdin is
966 * descriptors kept at the head of the 950 * closed. */
967 * device list. */ 951 FD_CLR(i->fd, &devices.infds);
968 FD_CLR(i->fd, &devices->infds); 952 /* Tell waker to ignore it too, by sending a
969 /* Tell waker to ignore it too... */ 953 * negative fd number (-1, since 0 is a valid
970 write(waker_fd, &i->fd, sizeof(i->fd)); 954 * FD number). */
971 } 955 dev_fd = -i->fd - 1;
956 write(waker_fd, &dev_fd, sizeof(dev_fd));
972 } 957 }
973 } 958 }
974 } 959 }
@@ -982,43 +967,93 @@ static void handle_input(int fd, struct device_list *devices)
982 * routines to allocate them. 967 * routines to allocate them.
983 * 968 *
984 * This routine allocates a new "struct lguest_device_desc" from descriptor 969 * This routine allocates a new "struct lguest_device_desc" from descriptor
985 * table in the devices array just above the Guest's normal memory. */ 970 * table just above the Guest's normal memory. It returns a pointer to that
986static struct lguest_device_desc * 971 * descriptor. */
987new_dev_desc(struct lguest_device_desc *descs, 972static struct lguest_device_desc *new_dev_desc(u16 type)
988 u16 type, u16 features, u16 num_pages)
989{ 973{
990 unsigned int i; 974 struct lguest_device_desc *d;
991 975
992 for (i = 0; i < LGUEST_MAX_DEVICES; i++) { 976 /* We only have one page for all the descriptors. */
993 if (!descs[i].type) { 977 if (devices.desc_used + sizeof(*d) > getpagesize())
994 descs[i].type = type; 978 errx(1, "Too many devices");
995 descs[i].features = features; 979
996 descs[i].num_pages = num_pages; 980 /* We don't need to set config_len or status: page is 0 already. */
997 /* If they said the device needs memory, we allocate 981 d = (void *)devices.descpage + devices.desc_used;
998 * that now, bumping up the top of Guest memory. */ 982 d->type = type;
999 if (num_pages) { 983 devices.desc_used += sizeof(*d);
1000 map_zeroed_pages(top, num_pages); 984
1001 descs[i].pfn = top/getpagesize(); 985 return d;
1002 top += num_pages*getpagesize();
1003 }
1004 return &descs[i];
1005 }
1006 }
1007 errx(1, "too many devices");
1008} 986}
1009 987
1010/* This monster routine does all the creation and setup of a new device, 988/* Each device descriptor is followed by some configuration information.
1011 * including caling new_dev_desc() to allocate the descriptor and device 989 * The first byte is a "status" byte for the Guest to report what's happening.
1012 * memory. */ 990 * After that are fields: u8 type, u8 len, [... len bytes...].
1013static struct device *new_device(struct device_list *devices, 991 *
1014 u16 type, u16 num_pages, u16 features, 992 * This routine adds a new field to an existing device's descriptor. It only
1015 int fd, 993 * works for the last device, but that's OK because that's how we use it. */
1016 bool (*handle_input)(int, struct device *), 994static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
1017 unsigned long watch_off, 995{
1018 u32 (*handle_output)(int, 996 /* This is the last descriptor, right? */
1019 const struct iovec *, 997 assert(devices.descpage + devices.desc_used
1020 unsigned, 998 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
1021 struct device *)) 999
1000 /* We only have one page of device descriptions. */
1001 if (devices.desc_used + 2 + len > getpagesize())
1002 errx(1, "Too many devices");
1003
1004 /* Copy in the new config header: type then length. */
1005 devices.descpage[devices.desc_used++] = type;
1006 devices.descpage[devices.desc_used++] = len;
1007 memcpy(devices.descpage + devices.desc_used, c, len);
1008 devices.desc_used += len;
1009
1010 /* Update the device descriptor length: two byte head then data. */
1011 dev->desc->config_len += 2 + len;
1012}
1013
1014/* This routine adds a virtqueue to a device. We specify how many descriptors
1015 * the virtqueue is to have. */
1016static void add_virtqueue(struct device *dev, unsigned int num_descs,
1017 void (*handle_output)(int fd, struct virtqueue *me))
1018{
1019 unsigned int pages;
1020 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1021 void *p;
1022
1023 /* First we need some pages for this virtqueue. */
1024 pages = (vring_size(num_descs) + getpagesize() - 1) / getpagesize();
1025 p = get_pages(pages);
1026
1027 /* Initialize the configuration. */
1028 vq->config.num = num_descs;
1029 vq->config.irq = devices.next_irq++;
1030 vq->config.pfn = to_guest_phys(p) / getpagesize();
1031
1032 /* Initialize the vring. */
1033 vring_init(&vq->vring, num_descs, p);
1034
1035 /* Add the configuration information to this device's descriptor. */
1036 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE,
1037 sizeof(vq->config), &vq->config);
1038
1039 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1040 * second. */
1041 for (i = &dev->vq; *i; i = &(*i)->next);
1042 *i = vq;
1043
1044 /* Link virtqueue back to device. */
1045 vq->dev = dev;
1046
1047 /* Set up handler. */
1048 vq->handle_output = handle_output;
1049 if (!handle_output)
1050 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1051}
1052
1053/* This routine does all the creation and setup of a new device, including
1054 * caling new_dev_desc() to allocate the descriptor and device memory. */
1055static struct device *new_device(const char *name, u16 type, int fd,
1056 bool (*handle_input)(int, struct device *))
1022{ 1057{
1023 struct device *dev = malloc(sizeof(*dev)); 1058 struct device *dev = malloc(sizeof(*dev));
1024 1059
@@ -1026,27 +1061,25 @@ static struct device *new_device(struct device_list *devices,
1026 * easier, but the user expects the devices to be arranged on the bus 1061 * easier, but the user expects the devices to be arranged on the bus
1027 * in command-line order. The first network device on the command line 1062 * in command-line order. The first network device on the command line
1028 * is eth0, the first block device /dev/lgba, etc. */ 1063 * is eth0, the first block device /dev/lgba, etc. */
1029 *devices->lastdev = dev; 1064 *devices.lastdev = dev;
1030 dev->next = NULL; 1065 dev->next = NULL;
1031 devices->lastdev = &dev->next; 1066 devices.lastdev = &dev->next;
1032 1067
1033 /* Now we populate the fields one at a time. */ 1068 /* Now we populate the fields one at a time. */
1034 dev->fd = fd; 1069 dev->fd = fd;
1035 /* If we have an input handler for this file descriptor, then we add it 1070 /* If we have an input handler for this file descriptor, then we add it
1036 * to the device_list's fdset and maxfd. */ 1071 * to the device_list's fdset and maxfd. */
1037 if (handle_input) 1072 if (handle_input)
1038 set_fd(dev->fd, devices); 1073 add_device_fd(dev->fd);
1039 dev->desc = new_dev_desc(devices->descs, type, features, num_pages); 1074 dev->desc = new_dev_desc(type);
1040 dev->mem = (void *)(dev->desc->pfn * getpagesize());
1041 dev->handle_input = handle_input; 1075 dev->handle_input = handle_input;
1042 dev->watch_key = (unsigned long)dev->mem + watch_off; 1076 dev->name = name;
1043 dev->handle_output = handle_output;
1044 return dev; 1077 return dev;
1045} 1078}
1046 1079
1047/* Our first setup routine is the console. It's a fairly simple device, but 1080/* Our first setup routine is the console. It's a fairly simple device, but
1048 * UNIX tty handling makes it uglier than it could be. */ 1081 * UNIX tty handling makes it uglier than it could be. */
1049static void setup_console(struct device_list *devices) 1082static void setup_console(void)
1050{ 1083{
1051 struct device *dev; 1084 struct device *dev;
1052 1085
@@ -1062,127 +1095,38 @@ static void setup_console(struct device_list *devices)
1062 atexit(restore_term); 1095 atexit(restore_term);
1063 } 1096 }
1064 1097
1065 /* We don't currently require any memory for the console, so we ask for 1098 dev = new_device("console", VIRTIO_ID_CONSOLE,
1066 * 0 pages. */ 1099 STDIN_FILENO, handle_console_input);
1067 dev = new_device(devices, LGUEST_DEVICE_T_CONSOLE, 0, 0,
1068 STDIN_FILENO, handle_console_input,
1069 LGUEST_CONSOLE_DMA_KEY, handle_console_output);
1070 /* We store the console state in dev->priv, and initialize it. */ 1100 /* We store the console state in dev->priv, and initialize it. */
1071 dev->priv = malloc(sizeof(struct console_abort)); 1101 dev->priv = malloc(sizeof(struct console_abort));
1072 ((struct console_abort *)dev->priv)->count = 0; 1102 ((struct console_abort *)dev->priv)->count = 0;
1073 verbose("device %p: console\n",
1074 (void *)(dev->desc->pfn * getpagesize()));
1075}
1076 1103
1077/* Setting up a block file is also fairly straightforward. */ 1104 /* The console needs two virtqueues: the input then the output. When
1078static void setup_block_file(const char *filename, struct device_list *devices) 1105 * they put something the input queue, we make sure we're listening to
1079{ 1106 * stdin. When they put something in the output queue, we write it to
1080 int fd; 1107 * stdout. */
1081 struct device *dev; 1108 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1082 off64_t *device_len; 1109 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1083 struct lguest_block_page *p; 1110
1084 1111 verbose("device %u: console\n", devices.device_num++);
1085 /* We open with O_LARGEFILE because otherwise we get stuck at 2G. We
1086 * open with O_DIRECT because otherwise our benchmarks go much too
1087 * fast. */
1088 fd = open_or_die(filename, O_RDWR|O_LARGEFILE|O_DIRECT);
1089
1090 /* We want one page, and have no input handler (the block file never
1091 * has anything interesting to say to us). Our timing will be quite
1092 * random, so it should be a reasonable randomness source. */
1093 dev = new_device(devices, LGUEST_DEVICE_T_BLOCK, 1,
1094 LGUEST_DEVICE_F_RANDOMNESS,
1095 fd, NULL, 0, handle_block_output);
1096
1097 /* We store the device size in the private area */
1098 device_len = dev->priv = malloc(sizeof(*device_len));
1099 /* This is the safe way of establishing the size of our device: it
1100 * might be a normal file or an actual block device like /dev/hdb. */
1101 *device_len = lseek64(fd, 0, SEEK_END);
1102
1103 /* The device memory is a "struct lguest_block_page". It's zeroed
1104 * already, we just need to put in the device size. Block devices
1105 * think in sectors (ie. 512 byte chunks), so we translate here. */
1106 p = dev->mem;
1107 p->num_sectors = *device_len/512;
1108 verbose("device %p: block %i sectors\n",
1109 (void *)(dev->desc->pfn * getpagesize()), p->num_sectors);
1110} 1112}
1113/*:*/
1111 1114
1112/* 1115/*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1113 * Network Devices. 1116 * --sharenet=<name> option which opens or creates a named pipe. This can be
1117 * used to send packets to another guest in a 1:1 manner.
1114 * 1118 *
1115 * Setting up network devices is quite a pain, because we have three types. 1119 * More sopisticated is to use one of the tools developed for project like UML
1116 * First, we have the inter-Guest network. This is a file which is mapped into 1120 * to do networking.
1117 * the address space of the Guests who are on the network. Because it is a
1118 * shared mapping, the same page underlies all the devices, and they can send
1119 * DMA to each other.
1120 * 1121 *
1121 * Remember from our network driver, the Guest is told what slot in the page it 1122 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1122 * is to use. We use exclusive fnctl locks to reserve a slot. If another 1123 * completely generic ("here's my vring, attach to your vring") and would work
1123 * Guest is using a slot, the lock will fail and we try another. Because fnctl 1124 * for any traffic. Of course, namespace and permissions issues need to be
1124 * locks are cleaned up automatically when we die, this cleverly means that our 1125 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1125 * reservation on the slot will vanish if we crash. */ 1126 * multiple inter-guest channels behind one interface, although it would
1126static unsigned int find_slot(int netfd, const char *filename) 1127 * require some manner of hotplugging new virtio channels.
1127{ 1128 *
1128 struct flock fl; 1129 * Finally, we could implement a virtio network switch in the kernel. :*/
1129
1130 fl.l_type = F_WRLCK;
1131 fl.l_whence = SEEK_SET;
1132 fl.l_len = 1;
1133 /* Try a 1 byte lock in each possible position number */
1134 for (fl.l_start = 0;
1135 fl.l_start < getpagesize()/sizeof(struct lguest_net);
1136 fl.l_start++) {
1137 /* If we succeed, return the slot number. */
1138 if (fcntl(netfd, F_SETLK, &fl) == 0)
1139 return fl.l_start;
1140 }
1141 errx(1, "No free slots in network file %s", filename);
1142}
1143
1144/* This function sets up the network file */
1145static void setup_net_file(const char *filename,
1146 struct device_list *devices)
1147{
1148 int netfd;
1149 struct device *dev;
1150
1151 /* We don't use open_or_die() here: for friendliness we create the file
1152 * if it doesn't already exist. */
1153 netfd = open(filename, O_RDWR, 0);
1154 if (netfd < 0) {
1155 if (errno == ENOENT) {
1156 netfd = open(filename, O_RDWR|O_CREAT, 0600);
1157 if (netfd >= 0) {
1158 /* If we succeeded, initialize the file with a
1159 * blank page. */
1160 char page[getpagesize()];
1161 memset(page, 0, sizeof(page));
1162 write(netfd, page, sizeof(page));
1163 }
1164 }
1165 if (netfd < 0)
1166 err(1, "cannot open net file '%s'", filename);
1167 }
1168
1169 /* We need 1 page, and the features indicate the slot to use and that
1170 * no checksum is needed. We never touch this device again; it's
1171 * between the Guests on the network, so we don't register input or
1172 * output handlers. */
1173 dev = new_device(devices, LGUEST_DEVICE_T_NET, 1,
1174 find_slot(netfd, filename)|LGUEST_NET_F_NOCSUM,
1175 -1, NULL, 0, NULL);
1176
1177 /* Map the shared file. */
1178 if (mmap(dev->mem, getpagesize(), PROT_READ|PROT_WRITE,
1179 MAP_FIXED|MAP_SHARED, netfd, 0) != dev->mem)
1180 err(1, "could not mmap '%s'", filename);
1181 verbose("device %p: shared net %s, peer %i\n",
1182 (void *)(dev->desc->pfn * getpagesize()), filename,
1183 dev->desc->features & ~LGUEST_NET_F_NOCSUM);
1184}
1185/*:*/
1186 1130
1187static u32 str2ip(const char *ipaddr) 1131static u32 str2ip(const char *ipaddr)
1188{ 1132{
@@ -1217,7 +1161,7 @@ static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1217 1161
1218/* This sets up the Host end of the network device with an IP address, brings 1162/* This sets up the Host end of the network device with an IP address, brings
1219 * it up so packets will flow, the copies the MAC address into the hwaddr 1163 * it up so packets will flow, the copies the MAC address into the hwaddr
1220 * pointer (in practice, the Host's slot in the network device's memory). */ 1164 * pointer. */
1221static void configure_device(int fd, const char *devname, u32 ipaddr, 1165static void configure_device(int fd, const char *devname, u32 ipaddr,
1222 unsigned char hwaddr[6]) 1166 unsigned char hwaddr[6])
1223{ 1167{
@@ -1243,18 +1187,18 @@ static void configure_device(int fd, const char *devname, u32 ipaddr,
1243 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); 1187 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1244} 1188}
1245 1189
1246/*L:195 The other kind of network is a Host<->Guest network. This can either 1190/*L:195 Our network is a Host<->Guest network. This can either use bridging or
1247 * use briding or routing, but the principle is the same: it uses the "tun" 1191 * routing, but the principle is the same: it uses the "tun" device to inject
1248 * device to inject packets into the Host as if they came in from a normal 1192 * packets into the Host as if they came in from a normal network card. We
1249 * network card. We just shunt packets between the Guest and the tun 1193 * just shunt packets between the Guest and the tun device. */
1250 * device. */ 1194static void setup_tun_net(const char *arg)
1251static void setup_tun_net(const char *arg, struct device_list *devices)
1252{ 1195{
1253 struct device *dev; 1196 struct device *dev;
1254 struct ifreq ifr; 1197 struct ifreq ifr;
1255 int netfd, ipfd; 1198 int netfd, ipfd;
1256 u32 ip; 1199 u32 ip;
1257 const char *br_name = NULL; 1200 const char *br_name = NULL;
1201 u8 hwaddr[6];
1258 1202
1259 /* We open the /dev/net/tun device and tell it we want a tap device. A 1203 /* We open the /dev/net/tun device and tell it we want a tap device. A
1260 * tap device is like a tun device, only somehow different. To tell 1204 * tap device is like a tun device, only somehow different. To tell
@@ -1270,21 +1214,13 @@ static void setup_tun_net(const char *arg, struct device_list *devices)
1270 * device: trust us! */ 1214 * device: trust us! */
1271 ioctl(netfd, TUNSETNOCSUM, 1); 1215 ioctl(netfd, TUNSETNOCSUM, 1);
1272 1216
1273 /* We create the net device with 1 page, using the features field of 1217 /* First we create a new network device. */
1274 * the descriptor to tell the Guest it is in slot 1 (NET_PEERNUM), and 1218 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1275 * that the device has fairly random timing. We do *not* specify
1276 * LGUEST_NET_F_NOCSUM: these packets can reach the real world.
1277 *
1278 * We will put our MAC address is slot 0 for the Guest to see, so
1279 * it will send packets to us using the key "peer_offset(0)": */
1280 dev = new_device(devices, LGUEST_DEVICE_T_NET, 1,
1281 NET_PEERNUM|LGUEST_DEVICE_F_RANDOMNESS, netfd,
1282 handle_tun_input, peer_offset(0), handle_tun_output);
1283 1219
1284 /* We keep a flag which says whether we've seen packets come out from 1220 /* Network devices need a receive and a send queue, just like
1285 * this network device. */ 1221 * console. */
1286 dev->priv = malloc(sizeof(bool)); 1222 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1287 *(bool *)dev->priv = false; 1223 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1288 1224
1289 /* We need a socket to perform the magic network ioctls to bring up the 1225 /* We need a socket to perform the magic network ioctls to bring up the
1290 * tap interface, connect to the bridge etc. Any socket will do! */ 1226 * tap interface, connect to the bridge etc. Any socket will do! */
@@ -1300,44 +1236,251 @@ static void setup_tun_net(const char *arg, struct device_list *devices)
1300 } else /* It is an IP address to set up the device with */ 1236 } else /* It is an IP address to set up the device with */
1301 ip = str2ip(arg); 1237 ip = str2ip(arg);
1302 1238
1303 /* We are peer 0, ie. first slot, so we hand dev->mem to this routine 1239 /* Set up the tun device, and get the mac address for the interface. */
1304 * to write the MAC address at the start of the device memory. */ 1240 configure_device(ipfd, ifr.ifr_name, ip, hwaddr);
1305 configure_device(ipfd, ifr.ifr_name, ip, dev->mem);
1306 1241
1307 /* Set "promisc" bit: we want every single packet if we're going to 1242 /* Tell Guest what MAC address to use. */
1308 * bridge to other machines (and otherwise it doesn't matter). */ 1243 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr);
1309 *((u8 *)dev->mem) |= 0x1;
1310 1244
1245 /* We don't seed the socket any more; setup is done. */
1311 close(ipfd); 1246 close(ipfd);
1312 1247
1313 verbose("device %p: tun net %u.%u.%u.%u\n", 1248 verbose("device %u: tun net %u.%u.%u.%u\n",
1314 (void *)(dev->desc->pfn * getpagesize()), 1249 devices.device_num++,
1315 (u8)(ip>>24), (u8)(ip>>16), (u8)(ip>>8), (u8)ip); 1250 (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip);
1316 if (br_name) 1251 if (br_name)
1317 verbose("attached to bridge: %s\n", br_name); 1252 verbose("attached to bridge: %s\n", br_name);
1318} 1253}
1254
1255
1256/*
1257 * Block device.
1258 *
1259 * Serving a block device is really easy: the Guest asks for a block number and
1260 * we read or write that position in the file.
1261 *
1262 * Unfortunately, this is amazingly slow: the Guest waits until the read is
1263 * finished before running anything else, even if it could be doing useful
1264 * work. We could use async I/O, except it's reputed to suck so hard that
1265 * characters actually go missing from your code when you try to use it.
1266 *
1267 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1268
1269/* This hangs off device->priv, with the data. */
1270struct vblk_info
1271{
1272 /* The size of the file. */
1273 off64_t len;
1274
1275 /* The file descriptor for the file. */
1276 int fd;
1277
1278 /* IO thread listens on this file descriptor [0]. */
1279 int workpipe[2];
1280
1281 /* IO thread writes to this file descriptor to mark it done, then
1282 * Launcher triggers interrupt to Guest. */
1283 int done_fd;
1284};
1285
1286/* This is the core of the I/O thread. It returns true if it did something. */
1287static bool service_io(struct device *dev)
1288{
1289 struct vblk_info *vblk = dev->priv;
1290 unsigned int head, out_num, in_num, wlen;
1291 int ret;
1292 struct virtio_blk_inhdr *in;
1293 struct virtio_blk_outhdr *out;
1294 struct iovec iov[dev->vq->vring.num];
1295 off64_t off;
1296
1297 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1298 if (head == dev->vq->vring.num)
1299 return false;
1300
1301 if (out_num == 0 || in_num == 0)
1302 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1303 head, out_num, in_num);
1304
1305 out = convert(&iov[0], struct virtio_blk_outhdr);
1306 in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr);
1307 off = out->sector * 512;
1308
1309 /* This is how we implement barriers. Pretty poor, no? */
1310 if (out->type & VIRTIO_BLK_T_BARRIER)
1311 fdatasync(vblk->fd);
1312
1313 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1314 fprintf(stderr, "Scsi commands unsupported\n");
1315 in->status = VIRTIO_BLK_S_UNSUPP;
1316 wlen = sizeof(in);
1317 } else if (out->type & VIRTIO_BLK_T_OUT) {
1318 /* Write */
1319
1320 /* Move to the right location in the block file. This can fail
1321 * if they try to write past end. */
1322 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1323 err(1, "Bad seek to sector %llu", out->sector);
1324
1325 ret = writev(vblk->fd, iov+1, out_num-1);
1326 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1327
1328 /* Grr... Now we know how long the descriptor they sent was, we
1329 * make sure they didn't try to write over the end of the block
1330 * file (possibly extending it). */
1331 if (ret > 0 && off + ret > vblk->len) {
1332 /* Trim it back to the correct length */
1333 ftruncate64(vblk->fd, vblk->len);
1334 /* Die, bad Guest, die. */
1335 errx(1, "Write past end %llu+%u", off, ret);
1336 }
1337 wlen = sizeof(in);
1338 in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1339 } else {
1340 /* Read */
1341
1342 /* Move to the right location in the block file. This can fail
1343 * if they try to read past end. */
1344 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1345 err(1, "Bad seek to sector %llu", out->sector);
1346
1347 ret = readv(vblk->fd, iov+1, in_num-1);
1348 verbose("READ from sector %llu: %i\n", out->sector, ret);
1349 if (ret >= 0) {
1350 wlen = sizeof(in) + ret;
1351 in->status = VIRTIO_BLK_S_OK;
1352 } else {
1353 wlen = sizeof(in);
1354 in->status = VIRTIO_BLK_S_IOERR;
1355 }
1356 }
1357
1358 /* We can't trigger an IRQ, because we're not the Launcher. It does
1359 * that when we tell it we're done. */
1360 add_used(dev->vq, head, wlen);
1361 return true;
1362}
1363
1364/* This is the thread which actually services the I/O. */
1365static int io_thread(void *_dev)
1366{
1367 struct device *dev = _dev;
1368 struct vblk_info *vblk = dev->priv;
1369 char c;
1370
1371 /* Close other side of workpipe so we get 0 read when main dies. */
1372 close(vblk->workpipe[1]);
1373 /* Close the other side of the done_fd pipe. */
1374 close(dev->fd);
1375
1376 /* When this read fails, it means Launcher died, so we follow. */
1377 while (read(vblk->workpipe[0], &c, 1) == 1) {
1378 /* We acknowledge each request immediately, to reduce latency,
1379 * rather than waiting until we've done them all. I haven't
1380 * measured to see if it makes any difference. */
1381 while (service_io(dev))
1382 write(vblk->done_fd, &c, 1);
1383 }
1384 return 0;
1385}
1386
1387/* When the thread says some I/O is done, we interrupt the Guest. */
1388static bool handle_io_finish(int fd, struct device *dev)
1389{
1390 char c;
1391
1392 /* If child died, presumably it printed message. */
1393 if (read(dev->fd, &c, 1) != 1)
1394 exit(1);
1395
1396 /* It did some work, so trigger the irq. */
1397 trigger_irq(fd, dev->vq);
1398 return true;
1399}
1400
1401/* When the Guest submits some I/O, we wake the I/O thread. */
1402static void handle_virtblk_output(int fd, struct virtqueue *vq)
1403{
1404 struct vblk_info *vblk = vq->dev->priv;
1405 char c = 0;
1406
1407 /* Wake up I/O thread and tell it to go to work! */
1408 if (write(vblk->workpipe[1], &c, 1) != 1)
1409 /* Presumably it indicated why it died. */
1410 exit(1);
1411}
1412
1413/* This creates a virtual block device. */
1414static void setup_block_file(const char *filename)
1415{
1416 int p[2];
1417 struct device *dev;
1418 struct vblk_info *vblk;
1419 void *stack;
1420 u64 cap;
1421 unsigned int val;
1422
1423 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1424 pipe(p);
1425
1426 /* The device responds to return from I/O thread. */
1427 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1428
1429 /* The device has a virtqueue. */
1430 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1431
1432 /* Allocate the room for our own bookkeeping */
1433 vblk = dev->priv = malloc(sizeof(*vblk));
1434
1435 /* First we open the file and store the length. */
1436 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1437 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1438
1439 /* Tell Guest how many sectors this device has. */
1440 cap = cpu_to_le64(vblk->len / 512);
1441 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1442
1443 /* Tell Guest not to put in too many descriptors at once: two are used
1444 * for the in and out elements. */
1445 val = cpu_to_le32(VIRTQUEUE_NUM - 2);
1446 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val);
1447
1448 /* The I/O thread writes to this end of the pipe when done. */
1449 vblk->done_fd = p[1];
1450
1451 /* This is how we tell the I/O thread about more work. */
1452 pipe(vblk->workpipe);
1453
1454 /* Create stack for thread and run it */
1455 stack = malloc(32768);
1456 if (clone(io_thread, stack + 32768, CLONE_VM, dev) == -1)
1457 err(1, "Creating clone");
1458
1459 /* We don't need to keep the I/O thread's end of the pipes open. */
1460 close(vblk->done_fd);
1461 close(vblk->workpipe[0]);
1462
1463 verbose("device %u: virtblock %llu sectors\n",
1464 devices.device_num, cap);
1465}
1319/* That's the end of device setup. */ 1466/* That's the end of device setup. */
1320 1467
1321/*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves 1468/*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1322 * its input and output, and finally, lays it to rest. */ 1469 * its input and output, and finally, lays it to rest. */
1323static void __attribute__((noreturn)) 1470static void __attribute__((noreturn)) run_guest(int lguest_fd)
1324run_guest(int lguest_fd, struct device_list *device_list)
1325{ 1471{
1326 for (;;) { 1472 for (;;) {
1327 u32 args[] = { LHREQ_BREAK, 0 }; 1473 unsigned long args[] = { LHREQ_BREAK, 0 };
1328 unsigned long arr[2]; 1474 unsigned long notify_addr;
1329 int readval; 1475 int readval;
1330 1476
1331 /* We read from the /dev/lguest device to run the Guest. */ 1477 /* We read from the /dev/lguest device to run the Guest. */
1332 readval = read(lguest_fd, arr, sizeof(arr)); 1478 readval = read(lguest_fd, &notify_addr, sizeof(notify_addr));
1333
1334 /* The read can only really return sizeof(arr) (the Guest did a
1335 * SEND_DMA to us), or an error. */
1336 1479
1337 /* For a successful read, arr[0] is the address of the "struct 1480 /* One unsigned long means the Guest did HCALL_NOTIFY */
1338 * lguest_dma", and arr[1] is the key the Guest sent to. */ 1481 if (readval == sizeof(notify_addr)) {
1339 if (readval == sizeof(arr)) { 1482 verbose("Notify on address %#lx\n", notify_addr);
1340 handle_output(lguest_fd, arr[0], arr[1], device_list); 1483 handle_output(lguest_fd, notify_addr);
1341 continue; 1484 continue;
1342 /* ENOENT means the Guest died. Reading tells us why. */ 1485 /* ENOENT means the Guest died. Reading tells us why. */
1343 } else if (errno == ENOENT) { 1486 } else if (errno == ENOENT) {
@@ -1351,7 +1494,7 @@ run_guest(int lguest_fd, struct device_list *device_list)
1351 1494
1352 /* Service input, then unset the BREAK which releases 1495 /* Service input, then unset the BREAK which releases
1353 * the Waker. */ 1496 * the Waker. */
1354 handle_input(lguest_fd, device_list); 1497 handle_input(lguest_fd);
1355 if (write(lguest_fd, args, sizeof(args)) < 0) 1498 if (write(lguest_fd, args, sizeof(args)) < 0)
1356 err(1, "Resetting break"); 1499 err(1, "Resetting break");
1357 } 1500 }
@@ -1365,7 +1508,6 @@ run_guest(int lguest_fd, struct device_list *device_list)
1365 1508
1366static struct option opts[] = { 1509static struct option opts[] = {
1367 { "verbose", 0, NULL, 'v' }, 1510 { "verbose", 0, NULL, 'v' },
1368 { "sharenet", 1, NULL, 's' },
1369 { "tunnet", 1, NULL, 't' }, 1511 { "tunnet", 1, NULL, 't' },
1370 { "block", 1, NULL, 'b' }, 1512 { "block", 1, NULL, 'b' },
1371 { "initrd", 1, NULL, 'i' }, 1513 { "initrd", 1, NULL, 'i' },
@@ -1374,37 +1516,21 @@ static struct option opts[] = {
1374static void usage(void) 1516static void usage(void)
1375{ 1517{
1376 errx(1, "Usage: lguest [--verbose] " 1518 errx(1, "Usage: lguest [--verbose] "
1377 "[--sharenet=<filename>|--tunnet=(<ipaddr>|bridge:<bridgename>)\n" 1519 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1378 "|--block=<filename>|--initrd=<filename>]...\n" 1520 "|--block=<filename>|--initrd=<filename>]...\n"
1379 "<mem-in-mb> vmlinux [args...]"); 1521 "<mem-in-mb> vmlinux [args...]");
1380} 1522}
1381 1523
1382/*L:100 The Launcher code itself takes us out into userspace, that scary place 1524/*L:105 The main routine is where the real work begins: */
1383 * where pointers run wild and free! Unfortunately, like most userspace
1384 * programs, it's quite boring (which is why everyone like to hack on the
1385 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
1386 * will get you through this section. Or, maybe not.
1387 *
1388 * The Launcher binary sits up high, usually starting at address 0xB8000000.
1389 * Everything below this is the "physical" memory for the Guest. For example,
1390 * if the Guest were to write a "1" at physical address 0, we would see a "1"
1391 * in the Launcher at "(int *)0". Guest physical == Launcher virtual.
1392 *
1393 * This can be tough to get your head around, but usually it just means that we
1394 * don't need to do any conversion when the Guest gives us it's "physical"
1395 * addresses.
1396 */
1397int main(int argc, char *argv[]) 1525int main(int argc, char *argv[])
1398{ 1526{
1399 /* Memory, top-level pagetable, code startpoint, PAGE_OFFSET and size 1527 /* Memory, top-level pagetable, code startpoint and size of the
1400 * of the (optional) initrd. */ 1528 * (optional) initrd. */
1401 unsigned long mem = 0, pgdir, start, page_offset, initrd_size = 0; 1529 unsigned long mem = 0, pgdir, start, initrd_size = 0;
1402 /* A temporary and the /dev/lguest file descriptor. */ 1530 /* A temporary and the /dev/lguest file descriptor. */
1403 int i, c, lguest_fd; 1531 int i, c, lguest_fd;
1404 /* The list of Guest devices, based on command line arguments. */ 1532 /* The boot information for the Guest. */
1405 struct device_list device_list; 1533 struct boot_params *boot;
1406 /* The boot information for the Guest: at guest-physical address 0. */
1407 void *boot = (void *)0;
1408 /* If they specify an initrd file to load. */ 1534 /* If they specify an initrd file to load. */
1409 const char *initrd_name = NULL; 1535 const char *initrd_name = NULL;
1410 1536
@@ -1412,11 +1538,12 @@ int main(int argc, char *argv[])
1412 * device receive input from a file descriptor, we keep an fdset 1538 * device receive input from a file descriptor, we keep an fdset
1413 * (infds) and the maximum fd number (max_infd) with the head of the 1539 * (infds) and the maximum fd number (max_infd) with the head of the
1414 * list. We also keep a pointer to the last device, for easy appending 1540 * list. We also keep a pointer to the last device, for easy appending
1415 * to the list. */ 1541 * to the list. Finally, we keep the next interrupt number to hand out
1416 device_list.max_infd = -1; 1542 * (1: remember that 0 is used by the timer). */
1417 device_list.dev = NULL; 1543 FD_ZERO(&devices.infds);
1418 device_list.lastdev = &device_list.dev; 1544 devices.max_infd = -1;
1419 FD_ZERO(&device_list.infds); 1545 devices.lastdev = &devices.dev;
1546 devices.next_irq = 1;
1420 1547
1421 /* We need to know how much memory so we can set up the device 1548 /* We need to know how much memory so we can set up the device
1422 * descriptor and memory pages for the devices as we parse the command 1549 * descriptor and memory pages for the devices as we parse the command
@@ -1424,9 +1551,16 @@ int main(int argc, char *argv[])
1424 * of memory now. */ 1551 * of memory now. */
1425 for (i = 1; i < argc; i++) { 1552 for (i = 1; i < argc; i++) {
1426 if (argv[i][0] != '-') { 1553 if (argv[i][0] != '-') {
1427 mem = top = atoi(argv[i]) * 1024 * 1024; 1554 mem = atoi(argv[i]) * 1024 * 1024;
1428 device_list.descs = map_zeroed_pages(top, 1); 1555 /* We start by mapping anonymous pages over all of
1429 top += getpagesize(); 1556 * guest-physical memory range. This fills it with 0,
1557 * and ensures that the Guest won't be killed when it
1558 * tries to access it. */
1559 guest_base = map_zeroed_pages(mem / getpagesize()
1560 + DEVICE_PAGES);
1561 guest_limit = mem;
1562 guest_max = mem + DEVICE_PAGES*getpagesize();
1563 devices.descpage = get_pages(1);
1430 break; 1564 break;
1431 } 1565 }
1432 } 1566 }
@@ -1437,14 +1571,11 @@ int main(int argc, char *argv[])
1437 case 'v': 1571 case 'v':
1438 verbose = true; 1572 verbose = true;
1439 break; 1573 break;
1440 case 's':
1441 setup_net_file(optarg, &device_list);
1442 break;
1443 case 't': 1574 case 't':
1444 setup_tun_net(optarg, &device_list); 1575 setup_tun_net(optarg);
1445 break; 1576 break;
1446 case 'b': 1577 case 'b':
1447 setup_block_file(optarg, &device_list); 1578 setup_block_file(optarg);
1448 break; 1579 break;
1449 case 'i': 1580 case 'i':
1450 initrd_name = optarg; 1581 initrd_name = optarg;
@@ -1459,56 +1590,60 @@ int main(int argc, char *argv[])
1459 if (optind + 2 > argc) 1590 if (optind + 2 > argc)
1460 usage(); 1591 usage();
1461 1592
1462 /* We always have a console device */ 1593 verbose("Guest base is at %p\n", guest_base);
1463 setup_console(&device_list);
1464 1594
1465 /* We start by mapping anonymous pages over all of guest-physical 1595 /* We always have a console device */
1466 * memory range. This fills it with 0, and ensures that the Guest 1596 setup_console();
1467 * won't be killed when it tries to access it. */
1468 map_zeroed_pages(0, mem / getpagesize());
1469 1597
1470 /* Now we load the kernel */ 1598 /* Now we load the kernel */
1471 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY), 1599 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1472 &page_offset); 1600
1601 /* Boot information is stashed at physical address 0 */
1602 boot = from_guest_phys(0);
1473 1603
1474 /* Map the initrd image if requested (at top of physical memory) */ 1604 /* Map the initrd image if requested (at top of physical memory) */
1475 if (initrd_name) { 1605 if (initrd_name) {
1476 initrd_size = load_initrd(initrd_name, mem); 1606 initrd_size = load_initrd(initrd_name, mem);
1477 /* These are the location in the Linux boot header where the 1607 /* These are the location in the Linux boot header where the
1478 * start and size of the initrd are expected to be found. */ 1608 * start and size of the initrd are expected to be found. */
1479 *(unsigned long *)(boot+0x218) = mem - initrd_size; 1609 boot->hdr.ramdisk_image = mem - initrd_size;
1480 *(unsigned long *)(boot+0x21c) = initrd_size; 1610 boot->hdr.ramdisk_size = initrd_size;
1481 /* The bootloader type 0xFF means "unknown"; that's OK. */ 1611 /* The bootloader type 0xFF means "unknown"; that's OK. */
1482 *(unsigned char *)(boot+0x210) = 0xFF; 1612 boot->hdr.type_of_loader = 0xFF;
1483 } 1613 }
1484 1614
1485 /* Set up the initial linear pagetables, starting below the initrd. */ 1615 /* Set up the initial linear pagetables, starting below the initrd. */
1486 pgdir = setup_pagetables(mem, initrd_size, page_offset); 1616 pgdir = setup_pagetables(mem, initrd_size);
1487 1617
1488 /* The Linux boot header contains an "E820" memory map: ours is a 1618 /* The Linux boot header contains an "E820" memory map: ours is a
1489 * simple, single region. */ 1619 * simple, single region. */
1490 *(char*)(boot+E820NR) = 1; 1620 boot->e820_entries = 1;
1491 *((struct e820entry *)(boot+E820MAP)) 1621 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1492 = ((struct e820entry) { 0, mem, E820_RAM });
1493 /* The boot header contains a command line pointer: we put the command 1622 /* The boot header contains a command line pointer: we put the command
1494 * line after the boot header (at address 4096) */ 1623 * line after the boot header. */
1495 *(void **)(boot + 0x228) = boot + 4096; 1624 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1496 concat(boot + 4096, argv+optind+2); 1625 concat((char *)(boot + 1), argv+optind+2);
1626
1627 /* Boot protocol version: 2.07 supports the fields for lguest. */
1628 boot->hdr.version = 0x207;
1629
1630 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1631 boot->hdr.hardware_subarch = 1;
1497 1632
1498 /* The guest type value of "1" tells the Guest it's under lguest. */ 1633 /* Tell the entry path not to try to reload segment registers. */
1499 *(int *)(boot + 0x23c) = 1; 1634 boot->hdr.loadflags |= KEEP_SEGMENTS;
1500 1635
1501 /* We tell the kernel to initialize the Guest: this returns the open 1636 /* We tell the kernel to initialize the Guest: this returns the open
1502 * /dev/lguest file descriptor. */ 1637 * /dev/lguest file descriptor. */
1503 lguest_fd = tell_kernel(pgdir, start, page_offset); 1638 lguest_fd = tell_kernel(pgdir, start);
1504 1639
1505 /* We fork off a child process, which wakes the Launcher whenever one 1640 /* We fork off a child process, which wakes the Launcher whenever one
1506 * of the input file descriptors needs attention. Otherwise we would 1641 * of the input file descriptors needs attention. Otherwise we would
1507 * run the Guest until it tries to output something. */ 1642 * run the Guest until it tries to output something. */
1508 waker_fd = setup_waker(lguest_fd, &device_list); 1643 waker_fd = setup_waker(lguest_fd);
1509 1644
1510 /* Finally, run the Guest. This doesn't return. */ 1645 /* Finally, run the Guest. This doesn't return. */
1511 run_guest(lguest_fd, &device_list); 1646 run_guest(lguest_fd);
1512} 1647}
1513/*:*/ 1648/*:*/
1514 1649
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-01 00:57:02 -0400