~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/lguest/lguest.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*P:100
  2  * This is the Launcher code, a simple program which lays out the "physical"
  3  * memory for the new Guest by mapping the kernel image and the virtual
  4  * devices, then opens /dev/lguest to tell the kernel about the Guest and
  5  * control it.
  6 :*/
  7 #define _LARGEFILE64_SOURCE
  8 #define _GNU_SOURCE
  9 #include <stdio.h>
 10 #include <string.h>
 11 #include <unistd.h>
 12 #include <err.h>
 13 #include <stdint.h>
 14 #include <stdlib.h>
 15 #include <elf.h>
 16 #include <sys/mman.h>
 17 #include <sys/param.h>
 18 #include <sys/types.h>
 19 #include <sys/stat.h>
 20 #include <sys/wait.h>
 21 #include <sys/eventfd.h>
 22 #include <fcntl.h>
 23 #include <stdbool.h>
 24 #include <errno.h>
 25 #include <ctype.h>
 26 #include <sys/socket.h>
 27 #include <sys/ioctl.h>
 28 #include <sys/time.h>
 29 #include <time.h>
 30 #include <netinet/in.h>
 31 #include <net/if.h>
 32 #include <linux/sockios.h>
 33 #include <linux/if_tun.h>
 34 #include <sys/uio.h>
 35 #include <termios.h>
 36 #include <getopt.h>
 37 #include <assert.h>
 38 #include <sched.h>
 39 #include <limits.h>
 40 #include <stddef.h>
 41 #include <signal.h>
 42 #include <pwd.h>
 43 #include <grp.h>
 44 #include <sys/user.h>
 45 #include <linux/pci_regs.h>
 46 
 47 #ifndef VIRTIO_F_ANY_LAYOUT
 48 #define VIRTIO_F_ANY_LAYOUT             27
 49 #endif
 50 
 51 /*L:110
 52  * We can ignore the 43 include files we need for this program, but I do want
 53  * to draw attention to the use of kernel-style types.
 54  *
 55  * As Linus said, "C is a Spartan language, and so should your naming be."  I
 56  * like these abbreviations, so we define them here.  Note that u64 is always
 57  * unsigned long long, which works on all Linux systems: this means that we can
 58  * use %llu in printf for any u64.
 59  */
 60 typedef unsigned long long u64;
 61 typedef uint32_t u32;
 62 typedef uint16_t u16;
 63 typedef uint8_t u8;
 64 /*:*/
 65 
 66 #define VIRTIO_CONFIG_NO_LEGACY
 67 #define VIRTIO_PCI_NO_LEGACY
 68 #define VIRTIO_BLK_NO_LEGACY
 69 #define VIRTIO_NET_NO_LEGACY
 70 
 71 /* Use in-kernel ones, which defines VIRTIO_F_VERSION_1 */
 72 #include "../../include/uapi/linux/virtio_config.h"
 73 #include "../../include/uapi/linux/virtio_net.h"
 74 #include "../../include/uapi/linux/virtio_blk.h"
 75 #include "../../include/uapi/linux/virtio_console.h"
 76 #include "../../include/uapi/linux/virtio_rng.h"
 77 #include <linux/virtio_ring.h>
 78 #include "../../include/uapi/linux/virtio_pci.h"
 79 #include <asm/bootparam.h>
 80 #include "../../include/linux/lguest_launcher.h"
 81 
 82 #define BRIDGE_PFX "bridge:"
 83 #ifndef SIOCBRADDIF
 84 #define SIOCBRADDIF     0x89a2          /* add interface to bridge      */
 85 #endif
 86 /* We can have up to 256 pages for devices. */
 87 #define DEVICE_PAGES 256
 88 /* This will occupy 3 pages: it must be a power of 2. */
 89 #define VIRTQUEUE_NUM 256
 90 
 91 /*L:120
 92  * verbose is both a global flag and a macro.  The C preprocessor allows
 93  * this, and although I wouldn't recommend it, it works quite nicely here.
 94  */
 95 static bool verbose;
 96 #define verbose(args...) \
 97         do { if (verbose) printf(args); } while(0)
 98 /*:*/
 99 
100 /* The pointer to the start of guest memory. */
101 static void *guest_base;
102 /* The maximum guest physical address allowed, and maximum possible. */
103 static unsigned long guest_limit, guest_max, guest_mmio;
104 /* The /dev/lguest file descriptor. */
105 static int lguest_fd;
106 
107 /* a per-cpu variable indicating whose vcpu is currently running */
108 static unsigned int __thread cpu_id;
109 
110 /* 5 bit device number in the PCI_CONFIG_ADDR => 32 only */
111 #define MAX_PCI_DEVICES 32
112 
113 /* This is our list of devices. */
114 struct device_list {
115         /* Counter to assign interrupt numbers. */
116         unsigned int next_irq;
117 
118         /* Counter to print out convenient device numbers. */
119         unsigned int device_num;
120 
121         /* PCI devices. */
122         struct device *pci[MAX_PCI_DEVICES];
123 };
124 
125 /* The list of Guest devices, based on command line arguments. */
126 static struct device_list devices;
127 
128 /*
129  * Just like struct virtio_pci_cfg_cap in uapi/linux/virtio_pci.h,
130  * but uses a u32 explicitly for the data.
131  */
132 struct virtio_pci_cfg_cap_u32 {
133         struct virtio_pci_cap cap;
134         u32 pci_cfg_data; /* Data for BAR access. */
135 };
136 
137 struct virtio_pci_mmio {
138         struct virtio_pci_common_cfg cfg;
139         u16 notify;
140         u8 isr;
141         u8 padding;
142         /* Device-specific configuration follows this. */
143 };
144 
145 /* This is the layout (little-endian) of the PCI config space. */
146 struct pci_config {
147         u16 vendor_id, device_id;
148         u16 command, status;
149         u8 revid, prog_if, subclass, class;
150         u8 cacheline_size, lat_timer, header_type, bist;
151         u32 bar[6];
152         u32 cardbus_cis_ptr;
153         u16 subsystem_vendor_id, subsystem_device_id;
154         u32 expansion_rom_addr;
155         u8 capabilities, reserved1[3];
156         u32 reserved2;
157         u8 irq_line, irq_pin, min_grant, max_latency;
158 
159         /* Now, this is the linked capability list. */
160         struct virtio_pci_cap common;
161         struct virtio_pci_notify_cap notify;
162         struct virtio_pci_cap isr;
163         struct virtio_pci_cap device;
164         struct virtio_pci_cfg_cap_u32 cfg_access;
165 };
166 
167 /* The device structure describes a single device. */
168 struct device {
169         /* The name of this device, for --verbose. */
170         const char *name;
171 
172         /* Any queues attached to this device */
173         struct virtqueue *vq;
174 
175         /* Is it operational */
176         bool running;
177 
178         /* Has it written FEATURES_OK but not re-checked it? */
179         bool wrote_features_ok;
180 
181         /* PCI configuration */
182         union {
183                 struct pci_config config;
184                 u32 config_words[sizeof(struct pci_config) / sizeof(u32)];
185         };
186 
187         /* Features we offer, and those accepted. */
188         u64 features, features_accepted;
189 
190         /* Device-specific config hangs off the end of this. */
191         struct virtio_pci_mmio *mmio;
192 
193         /* PCI MMIO resources (all in BAR0) */
194         size_t mmio_size;
195         u32 mmio_addr;
196 
197         /* Device-specific data. */
198         void *priv;
199 };
200 
201 /* The virtqueue structure describes a queue attached to a device. */
202 struct virtqueue {
203         struct virtqueue *next;
204 
205         /* Which device owns me. */
206         struct device *dev;
207 
208         /* Name for printing errors. */
209         const char *name;
210 
211         /* The actual ring of buffers. */
212         struct vring vring;
213 
214         /* The information about this virtqueue (we only use queue_size on) */
215         struct virtio_pci_common_cfg pci_config;
216 
217         /* Last available index we saw. */
218         u16 last_avail_idx;
219 
220         /* How many are used since we sent last irq? */
221         unsigned int pending_used;
222 
223         /* Eventfd where Guest notifications arrive. */
224         int eventfd;
225 
226         /* Function for the thread which is servicing this virtqueue. */
227         void (*service)(struct virtqueue *vq);
228         pid_t thread;
229 };
230 
231 /* Remember the arguments to the program so we can "reboot" */
232 static char **main_args;
233 
234 /* The original tty settings to restore on exit. */
235 static struct termios orig_term;
236 
237 /*
238  * We have to be careful with barriers: our devices are all run in separate
239  * threads and so we need to make sure that changes visible to the Guest happen
240  * in precise order.
241  */
242 #define wmb() __asm__ __volatile__("" : : : "memory")
243 #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
244 #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
245 
246 /* Wrapper for the last available index.  Makes it easier to change. */
247 #define lg_last_avail(vq)       ((vq)->last_avail_idx)
248 
249 /*
250  * The virtio configuration space is defined to be little-endian.  x86 is
251  * little-endian too, but it's nice to be explicit so we have these helpers.
252  */
253 #define cpu_to_le16(v16) (v16)
254 #define cpu_to_le32(v32) (v32)
255 #define cpu_to_le64(v64) (v64)
256 #define le16_to_cpu(v16) (v16)
257 #define le32_to_cpu(v32) (v32)
258 #define le64_to_cpu(v64) (v64)
259 
260 /*
261  * A real device would ignore weird/non-compliant driver behaviour.  We
262  * stop and flag it, to help debugging Linux problems.
263  */
264 #define bad_driver(d, fmt, ...) \
265         errx(1, "%s: bad driver: " fmt, (d)->name, ## __VA_ARGS__)
266 #define bad_driver_vq(vq, fmt, ...)                            \
267         errx(1, "%s vq %s: bad driver: " fmt, (vq)->dev->name, \
268              vq->name, ## __VA_ARGS__)
269 
270 /* Is this iovec empty? */
271 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
272 {
273         unsigned int i;
274 
275         for (i = 0; i < num_iov; i++)
276                 if (iov[i].iov_len)
277                         return false;
278         return true;
279 }
280 
281 /* Take len bytes from the front of this iovec. */
282 static void iov_consume(struct device *d,
283                         struct iovec iov[], unsigned num_iov,
284                         void *dest, unsigned len)
285 {
286         unsigned int i;
287 
288         for (i = 0; i < num_iov; i++) {
289                 unsigned int used;
290 
291                 used = iov[i].iov_len < len ? iov[i].iov_len : len;
292                 if (dest) {
293                         memcpy(dest, iov[i].iov_base, used);
294                         dest += used;
295                 }
296                 iov[i].iov_base += used;
297                 iov[i].iov_len -= used;
298                 len -= used;
299         }
300         if (len != 0)
301                 bad_driver(d, "iovec too short!");
302 }
303 
304 /*L:100
305  * The Launcher code itself takes us out into userspace, that scary place where
306  * pointers run wild and free!  Unfortunately, like most userspace programs,
307  * it's quite boring (which is why everyone likes to hack on the kernel!).
308  * Perhaps if you make up an Lguest Drinking Game at this point, it will get
309  * you through this section.  Or, maybe not.
310  *
311  * The Launcher sets up a big chunk of memory to be the Guest's "physical"
312  * memory and stores it in "guest_base".  In other words, Guest physical ==
313  * Launcher virtual with an offset.
314  *
315  * This can be tough to get your head around, but usually it just means that we
316  * use these trivial conversion functions when the Guest gives us its
317  * "physical" addresses:
318  */
319 static void *from_guest_phys(unsigned long addr)
320 {
321         return guest_base + addr;
322 }
323 
324 static unsigned long to_guest_phys(const void *addr)
325 {
326         return (addr - guest_base);
327 }
328 
329 /*L:130
330  * Loading the Kernel.
331  *
332  * We start with couple of simple helper routines.  open_or_die() avoids
333  * error-checking code cluttering the callers:
334  */
335 static int open_or_die(const char *name, int flags)
336 {
337         int fd = open(name, flags);
338         if (fd < 0)
339                 err(1, "Failed to open %s", name);
340         return fd;
341 }
342 
343 /* map_zeroed_pages() takes a number of pages. */
344 static void *map_zeroed_pages(unsigned int num)
345 {
346         int fd = open_or_die("/dev/zero", O_RDONLY);
347         void *addr;
348 
349         /*
350          * We use a private mapping (ie. if we write to the page, it will be
351          * copied). We allocate an extra two pages PROT_NONE to act as guard
352          * pages against read/write attempts that exceed allocated space.
353          */
354         addr = mmap(NULL, getpagesize() * (num+2),
355                     PROT_NONE, MAP_PRIVATE, fd, 0);
356 
357         if (addr == MAP_FAILED)
358                 err(1, "Mmapping %u pages of /dev/zero", num);
359 
360         if (mprotect(addr + getpagesize(), getpagesize() * num,
361                      PROT_READ|PROT_WRITE) == -1)
362                 err(1, "mprotect rw %u pages failed", num);
363 
364         /*
365          * One neat mmap feature is that you can close the fd, and it
366          * stays mapped.
367          */
368         close(fd);
369 
370         /* Return address after PROT_NONE page */
371         return addr + getpagesize();
372 }
373 
374 /* Get some bytes which won't be mapped into the guest. */
375 static unsigned long get_mmio_region(size_t size)
376 {
377         unsigned long addr = guest_mmio;
378         size_t i;
379 
380         if (!size)
381                 return addr;
382 
383         /* Size has to be a power of 2 (and multiple of 16) */
384         for (i = 1; i < size; i <<= 1);
385 
386         guest_mmio += i;
387 
388         return addr;
389 }
390 
391 /*
392  * This routine is used to load the kernel or initrd.  It tries mmap, but if
393  * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
394  * it falls back to reading the memory in.
395  */
396 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
397 {
398         ssize_t r;
399 
400         /*
401          * We map writable even though for some segments are marked read-only.
402          * The kernel really wants to be writable: it patches its own
403          * instructions.
404          *
405          * MAP_PRIVATE means that the page won't be copied until a write is
406          * done to it.  This allows us to share untouched memory between
407          * Guests.
408          */
409         if (mmap(addr, len, PROT_READ|PROT_WRITE,
410                  MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
411                 return;
412 
413         /* pread does a seek and a read in one shot: saves a few lines. */
414         r = pread(fd, addr, len, offset);
415         if (r != len)
416                 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
417 }
418 
419 /*
420  * This routine takes an open vmlinux image, which is in ELF, and maps it into
421  * the Guest memory.  ELF = Embedded Linking Format, which is the format used
422  * by all modern binaries on Linux including the kernel.
423  *
424  * The ELF headers give *two* addresses: a physical address, and a virtual
425  * address.  We use the physical address; the Guest will map itself to the
426  * virtual address.
427  *
428  * We return the starting address.
429  */
430 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
431 {
432         Elf32_Phdr phdr[ehdr->e_phnum];
433         unsigned int i;
434 
435         /*
436          * Sanity checks on the main ELF header: an x86 executable with a
437          * reasonable number of correctly-sized program headers.
438          */
439         if (ehdr->e_type != ET_EXEC
440             || ehdr->e_machine != EM_386
441             || ehdr->e_phentsize != sizeof(Elf32_Phdr)
442             || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
443                 errx(1, "Malformed elf header");
444 
445         /*
446          * An ELF executable contains an ELF header and a number of "program"
447          * headers which indicate which parts ("segments") of the program to
448          * load where.
449          */
450 
451         /* We read in all the program headers at once: */
452         if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
453                 err(1, "Seeking to program headers");
454         if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
455                 err(1, "Reading program headers");
456 
457         /*
458          * Try all the headers: there are usually only three.  A read-only one,
459          * a read-write one, and a "note" section which we don't load.
460          */
461         for (i = 0; i < ehdr->e_phnum; i++) {
462                 /* If this isn't a loadable segment, we ignore it */
463                 if (phdr[i].p_type != PT_LOAD)
464                         continue;
465 
466                 verbose("Section %i: size %i addr %p\n",
467                         i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
468 
469                 /* We map this section of the file at its physical address. */
470                 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
471                        phdr[i].p_offset, phdr[i].p_filesz);
472         }
473 
474         /* The entry point is given in the ELF header. */
475         return ehdr->e_entry;
476 }
477 
478 /*L:150
479  * A bzImage, unlike an ELF file, is not meant to be loaded.  You're supposed
480  * to jump into it and it will unpack itself.  We used to have to perform some
481  * hairy magic because the unpacking code scared me.
482  *
483  * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
484  * a small patch to jump over the tricky bits in the Guest, so now we just read
485  * the funky header so we know where in the file to load, and away we go!
486  */
487 static unsigned long load_bzimage(int fd)
488 {
489         struct boot_params boot;
490         int r;
491         /* Modern bzImages get loaded at 1M. */
492         void *p = from_guest_phys(0x100000);
493 
494         /*
495          * Go back to the start of the file and read the header.  It should be
496          * a Linux boot header (see Documentation/x86/boot.txt)
497          */
498         lseek(fd, 0, SEEK_SET);
499         read(fd, &boot, sizeof(boot));
500 
501         /* Inside the setup_hdr, we expect the magic "HdrS" */
502         if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
503                 errx(1, "This doesn't look like a bzImage to me");
504 
505         /* Skip over the extra sectors of the header. */
506         lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
507 
508         /* Now read everything into memory. in nice big chunks. */
509         while ((r = read(fd, p, 65536)) > 0)
510                 p += r;
511 
512         /* Finally, code32_start tells us where to enter the kernel. */
513         return boot.hdr.code32_start;
514 }
515 
516 /*L:140
517  * Loading the kernel is easy when it's a "vmlinux", but most kernels
518  * come wrapped up in the self-decompressing "bzImage" format.  With a little
519  * work, we can load those, too.
520  */
521 static unsigned long load_kernel(int fd)
522 {
523         Elf32_Ehdr hdr;
524 
525         /* Read in the first few bytes. */
526         if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
527                 err(1, "Reading kernel");
528 
529         /* If it's an ELF file, it starts with "\177ELF" */
530         if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
531                 return map_elf(fd, &hdr);
532 
533         /* Otherwise we assume it's a bzImage, and try to load it. */
534         return load_bzimage(fd);
535 }
536 
537 /*
538  * This is a trivial little helper to align pages.  Andi Kleen hated it because
539  * it calls getpagesize() twice: "it's dumb code."
540  *
541  * Kernel guys get really het up about optimization, even when it's not
542  * necessary.  I leave this code as a reaction against that.
543  */
544 static inline unsigned long page_align(unsigned long addr)
545 {
546         /* Add upwards and truncate downwards. */
547         return ((addr + getpagesize()-1) & ~(getpagesize()-1));
548 }
549 
550 /*L:180
551  * An "initial ram disk" is a disk image loaded into memory along with the
552  * kernel which the kernel can use to boot from without needing any drivers.
553  * Most distributions now use this as standard: the initrd contains the code to
554  * load the appropriate driver modules for the current machine.
555  *
556  * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
557  * kernels.  He sent me this (and tells me when I break it).
558  */
559 static unsigned long load_initrd(const char *name, unsigned long mem)
560 {
561         int ifd;
562         struct stat st;
563         unsigned long len;
564 
565         ifd = open_or_die(name, O_RDONLY);
566         /* fstat() is needed to get the file size. */
567         if (fstat(ifd, &st) < 0)
568                 err(1, "fstat() on initrd '%s'", name);
569 
570         /*
571          * We map the initrd at the top of memory, but mmap wants it to be
572          * page-aligned, so we round the size up for that.
573          */
574         len = page_align(st.st_size);
575         map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
576         /*
577          * Once a file is mapped, you can close the file descriptor.  It's a
578          * little odd, but quite useful.
579          */
580         close(ifd);
581         verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
582 
583         /* We return the initrd size. */
584         return len;
585 }
586 /*:*/
587 
588 /*
589  * Simple routine to roll all the commandline arguments together with spaces
590  * between them.
591  */
592 static void concat(char *dst, char *args[])
593 {
594         unsigned int i, len = 0;
595 
596         for (i = 0; args[i]; i++) {
597                 if (i) {
598                         strcat(dst+len, " ");
599                         len++;
600                 }
601                 strcpy(dst+len, args[i]);
602                 len += strlen(args[i]);
603         }
604         /* In case it's empty. */
605         dst[len] = '\0';
606 }
607 
608 /*L:185
609  * This is where we actually tell the kernel to initialize the Guest.  We
610  * saw the arguments it expects when we looked at initialize() in lguest_user.c:
611  * the base of Guest "physical" memory, the top physical page to allow and the
612  * entry point for the Guest.
613  */
614 static void tell_kernel(unsigned long start)
615 {
616         unsigned long args[] = { LHREQ_INITIALIZE,
617                                  (unsigned long)guest_base,
618                                  guest_limit / getpagesize(), start,
619                                  (guest_mmio+getpagesize()-1) / getpagesize() };
620         verbose("Guest: %p - %p (%#lx, MMIO %#lx)\n",
621                 guest_base, guest_base + guest_limit,
622                 guest_limit, guest_mmio);
623         lguest_fd = open_or_die("/dev/lguest", O_RDWR);
624         if (write(lguest_fd, args, sizeof(args)) < 0)
625                 err(1, "Writing to /dev/lguest");
626 }
627 /*:*/
628 
629 /*L:200
630  * Device Handling.
631  *
632  * When the Guest gives us a buffer, it sends an array of addresses and sizes.
633  * We need to make sure it's not trying to reach into the Launcher itself, so
634  * we have a convenient routine which checks it and exits with an error message
635  * if something funny is going on:
636  */
637 static void *_check_pointer(struct device *d,
638                             unsigned long addr, unsigned int size,
639                             unsigned int line)
640 {
641         /*
642          * Check if the requested address and size exceeds the allocated memory,
643          * or addr + size wraps around.
644          */
645         if ((addr + size) > guest_limit || (addr + size) < addr)
646                 bad_driver(d, "%s:%i: Invalid address %#lx",
647                            __FILE__, line, addr);
648         /*
649          * We return a pointer for the caller's convenience, now we know it's
650          * safe to use.
651          */
652         return from_guest_phys(addr);
653 }
654 /* A macro which transparently hands the line number to the real function. */
655 #define check_pointer(d,addr,size) _check_pointer(d, addr, size, __LINE__)
656 
657 /*
658  * Each buffer in the virtqueues is actually a chain of descriptors.  This
659  * function returns the next descriptor in the chain, or vq->vring.num if we're
660  * at the end.
661  */
662 static unsigned next_desc(struct device *d, struct vring_desc *desc,
663                           unsigned int i, unsigned int max)
664 {
665         unsigned int next;
666 
667         /* If this descriptor says it doesn't chain, we're done. */
668         if (!(desc[i].flags & VRING_DESC_F_NEXT))
669                 return max;
670 
671         /* Check they're not leading us off end of descriptors. */
672         next = desc[i].next;
673         /* Make sure compiler knows to grab that: we don't want it changing! */
674         wmb();
675 
676         if (next >= max)
677                 bad_driver(d, "Desc next is %u", next);
678 
679         return next;
680 }
681 
682 /*
683  * This actually sends the interrupt for this virtqueue, if we've used a
684  * buffer.
685  */
686 static void trigger_irq(struct virtqueue *vq)
687 {
688         unsigned long buf[] = { LHREQ_IRQ, vq->dev->config.irq_line };
689 
690         /* Don't inform them if nothing used. */
691         if (!vq->pending_used)
692                 return;
693         vq->pending_used = 0;
694 
695         /*
696          * 2.4.7.1:
697          *
698          *  If the VIRTIO_F_EVENT_IDX feature bit is not negotiated:
699          *    The driver MUST set flags to 0 or 1. 
700          */
701         if (vq->vring.avail->flags > 1)
702                 bad_driver_vq(vq, "avail->flags = %u\n", vq->vring.avail->flags);
703 
704         /*
705          * 2.4.7.2:
706          *
707          *  If the VIRTIO_F_EVENT_IDX feature bit is not negotiated:
708          *
709          *     - The device MUST ignore the used_event value.
710          *     - After the device writes a descriptor index into the used ring:
711          *         - If flags is 1, the device SHOULD NOT send an interrupt.
712          *         - If flags is 0, the device MUST send an interrupt.
713          */
714         if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
715                 return;
716         }
717 
718         /*
719          * 4.1.4.5.1:
720          *
721          *  If MSI-X capability is disabled, the device MUST set the Queue
722          *  Interrupt bit in ISR status before sending a virtqueue notification
723          *  to the driver.
724          */
725         vq->dev->mmio->isr = 0x1;
726 
727         /* Send the Guest an interrupt tell them we used something up. */
728         if (write(lguest_fd, buf, sizeof(buf)) != 0)
729                 err(1, "Triggering irq %i", vq->dev->config.irq_line);
730 }
731 
732 /*
733  * This looks in the virtqueue for the first available buffer, and converts
734  * it to an iovec for convenient access.  Since descriptors consist of some
735  * number of output then some number of input descriptors, it's actually two
736  * iovecs, but we pack them into one and note how many of each there were.
737  *
738  * This function waits if necessary, and returns the descriptor number found.
739  */
740 static unsigned wait_for_vq_desc(struct virtqueue *vq,
741                                  struct iovec iov[],
742                                  unsigned int *out_num, unsigned int *in_num)
743 {
744         unsigned int i, head, max;
745         struct vring_desc *desc;
746         u16 last_avail = lg_last_avail(vq);
747 
748         /*
749          * 2.4.7.1:
750          *
751          *   The driver MUST handle spurious interrupts from the device.
752          *
753          * That's why this is a while loop.
754          */
755 
756         /* There's nothing available? */
757         while (last_avail == vq->vring.avail->idx) {
758                 u64 event;
759 
760                 /*
761                  * Since we're about to sleep, now is a good time to tell the
762                  * Guest about what we've used up to now.
763                  */
764                 trigger_irq(vq);
765 
766                 /* OK, now we need to know about added descriptors. */
767                 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
768 
769                 /*
770                  * They could have slipped one in as we were doing that: make
771                  * sure it's written, then check again.
772                  */
773                 mb();
774                 if (last_avail != vq->vring.avail->idx) {
775                         vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
776                         break;
777                 }
778 
779                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
780                 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
781                         errx(1, "Event read failed?");
782 
783                 /* We don't need to be notified again. */
784                 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
785         }
786 
787         /* Check it isn't doing very strange things with descriptor numbers. */
788         if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
789                 bad_driver_vq(vq, "Guest moved used index from %u to %u",
790                               last_avail, vq->vring.avail->idx);
791 
792         /* 
793          * Make sure we read the descriptor number *after* we read the ring
794          * update; don't let the cpu or compiler change the order.
795          */
796         rmb();
797 
798         /*
799          * Grab the next descriptor number they're advertising, and increment
800          * the index we've seen.
801          */
802         head = vq->vring.avail->ring[last_avail % vq->vring.num];
803         lg_last_avail(vq)++;
804 
805         /* If their number is silly, that's a fatal mistake. */
806         if (head >= vq->vring.num)
807                 bad_driver_vq(vq, "Guest says index %u is available", head);
808 
809         /* When we start there are none of either input nor output. */
810         *out_num = *in_num = 0;
811 
812         max = vq->vring.num;
813         desc = vq->vring.desc;
814         i = head;
815 
816         /*
817          * We have to read the descriptor after we read the descriptor number,
818          * but there's a data dependency there so the CPU shouldn't reorder
819          * that: no rmb() required.
820          */
821 
822         do {
823                 /*
824                  * If this is an indirect entry, then this buffer contains a
825                  * descriptor table which we handle as if it's any normal
826                  * descriptor chain.
827                  */
828                 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
829                         /* 2.4.5.3.1:
830                          *
831                          *  The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT
832                          *  flag unless the VIRTIO_F_INDIRECT_DESC feature was
833                          *  negotiated.
834                          */
835                         if (!(vq->dev->features_accepted &
836                               (1<<VIRTIO_RING_F_INDIRECT_DESC)))
837                                 bad_driver_vq(vq, "vq indirect not negotiated");
838 
839                         /*
840                          * 2.4.5.3.1:
841                          *
842                          *   The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT
843                          *   flag within an indirect descriptor (ie. only one
844                          *   table per descriptor).
845                          */
846                         if (desc != vq->vring.desc)
847                                 bad_driver_vq(vq, "Indirect within indirect");
848 
849                         /*
850                          * Proposed update VIRTIO-134 spells this out:
851                          *
852                          *   A driver MUST NOT set both VIRTQ_DESC_F_INDIRECT
853                          *   and VIRTQ_DESC_F_NEXT in flags.
854                          */
855                         if (desc[i].flags & VRING_DESC_F_NEXT)
856                                 bad_driver_vq(vq, "indirect and next together");
857 
858                         if (desc[i].len % sizeof(struct vring_desc))
859                                 bad_driver_vq(vq,
860                                               "Invalid size for indirect table");
861                         /*
862                          * 2.4.5.3.2:
863                          *
864                          *  The device MUST ignore the write-only flag
865                          *  (flags&VIRTQ_DESC_F_WRITE) in the descriptor that
866                          *  refers to an indirect table.
867                          *
868                          * We ignore it here: :)
869                          */
870 
871                         max = desc[i].len / sizeof(struct vring_desc);
872                         desc = check_pointer(vq->dev, desc[i].addr, desc[i].len);
873                         i = 0;
874 
875                         /* 2.4.5.3.1:
876                          *
877                          *  A driver MUST NOT create a descriptor chain longer
878                          *  than the Queue Size of the device.
879                          */
880                         if (max > vq->pci_config.queue_size)
881                                 bad_driver_vq(vq,
882                                               "indirect has too many entries");
883                 }
884 
885                 /* Grab the first descriptor, and check it's OK. */
886                 iov[*out_num + *in_num].iov_len = desc[i].len;
887                 iov[*out_num + *in_num].iov_base
888                         = check_pointer(vq->dev, desc[i].addr, desc[i].len);
889                 /* If this is an input descriptor, increment that count. */
890                 if (desc[i].flags & VRING_DESC_F_WRITE)
891                         (*in_num)++;
892                 else {
893                         /*
894                          * If it's an output descriptor, they're all supposed
895                          * to come before any input descriptors.
896                          */
897                         if (*in_num)
898                                 bad_driver_vq(vq,
899                                               "Descriptor has out after in");
900                         (*out_num)++;
901                 }
902 
903                 /* If we've got too many, that implies a descriptor loop. */
904                 if (*out_num + *in_num > max)
905                         bad_driver_vq(vq, "Looped descriptor");
906         } while ((i = next_desc(vq->dev, desc, i, max)) != max);
907 
908         return head;
909 }
910 
911 /*
912  * After we've used one of their buffers, we tell the Guest about it.  Sometime
913  * later we'll want to send them an interrupt using trigger_irq(); note that
914  * wait_for_vq_desc() does that for us if it has to wait.
915  */
916 static void add_used(struct virtqueue *vq, unsigned int head, int len)
917 {
918         struct vring_used_elem *used;
919 
920         /*
921          * The virtqueue contains a ring of used buffers.  Get a pointer to the
922          * next entry in that used ring.
923          */
924         used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
925         used->id = head;
926         used->len = len;
927         /* Make sure buffer is written before we update index. */
928         wmb();
929         vq->vring.used->idx++;
930         vq->pending_used++;
931 }
932 
933 /* And here's the combo meal deal.  Supersize me! */
934 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
935 {
936         add_used(vq, head, len);
937         trigger_irq(vq);
938 }
939 
940 /*
941  * The Console
942  *
943  * We associate some data with the console for our exit hack.
944  */
945 struct console_abort {
946         /* How many times have they hit ^C? */
947         int count;
948         /* When did they start? */
949         struct timeval start;
950 };
951 
952 /* This is the routine which handles console input (ie. stdin). */
953 static void console_input(struct virtqueue *vq)
954 {
955         int len;
956         unsigned int head, in_num, out_num;
957         struct console_abort *abort = vq->dev->priv;
958         struct iovec iov[vq->vring.num];
959 
960         /* Make sure there's a descriptor available. */
961         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
962         if (out_num)
963                 bad_driver_vq(vq, "Output buffers in console in queue?");
964 
965         /* Read into it.  This is where we usually wait. */
966         len = readv(STDIN_FILENO, iov, in_num);
967         if (len <= 0) {
968                 /* Ran out of input? */
969                 warnx("Failed to get console input, ignoring console.");
970                 /*
971                  * For simplicity, dying threads kill the whole Launcher.  So
972                  * just nap here.
973                  */
974                 for (;;)
975                         pause();
976         }
977 
978         /* Tell the Guest we used a buffer. */
979         add_used_and_trigger(vq, head, len);
980 
981         /*
982          * Three ^C within one second?  Exit.
983          *
984          * This is such a hack, but works surprisingly well.  Each ^C has to
985          * be in a buffer by itself, so they can't be too fast.  But we check
986          * that we get three within about a second, so they can't be too
987          * slow.
988          */
989         if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
990                 abort->count = 0;
991                 return;
992         }
993 
994         abort->count++;
995         if (abort->count == 1)
996                 gettimeofday(&abort->start, NULL);
997         else if (abort->count == 3) {
998                 struct timeval now;
999                 gettimeofday(&now, NULL);
1000                 /* Kill all Launcher processes with SIGINT, like normal ^C */
1001                 if (now.tv_sec <= abort->start.tv_sec+1)
1002                         kill(0, SIGINT);
1003                 abort->count = 0;
1004         }
1005 }
1006 
1007 /* This is the routine which handles console output (ie. stdout). */
1008 static void console_output(struct virtqueue *vq)
1009 {
1010         unsigned int head, out, in;
1011         struct iovec iov[vq->vring.num];
1012 
1013         /* We usually wait in here, for the Guest to give us something. */
1014         head = wait_for_vq_desc(vq, iov, &out, &in);
1015         if (in)
1016                 bad_driver_vq(vq, "Input buffers in console output queue?");
1017 
1018         /* writev can return a partial write, so we loop here. */
1019         while (!iov_empty(iov, out)) {
1020                 int len = writev(STDOUT_FILENO, iov, out);
1021                 if (len <= 0) {
1022                         warn("Write to stdout gave %i (%d)", len, errno);
1023                         break;
1024                 }
1025                 iov_consume(vq->dev, iov, out, NULL, len);
1026         }
1027 
1028         /*
1029          * We're finished with that buffer: if we're going to sleep,
1030          * wait_for_vq_desc() will prod the Guest with an interrupt.
1031          */
1032         add_used(vq, head, 0);
1033 }
1034 
1035 /*
1036  * The Network
1037  *
1038  * Handling output for network is also simple: we get all the output buffers
1039  * and write them to /dev/net/tun.
1040  */
1041 struct net_info {
1042         int tunfd;
1043 };
1044 
1045 static void net_output(struct virtqueue *vq)
1046 {
1047         struct net_info *net_info = vq->dev->priv;
1048         unsigned int head, out, in;
1049         struct iovec iov[vq->vring.num];
1050 
1051         /* We usually wait in here for the Guest to give us a packet. */
1052         head = wait_for_vq_desc(vq, iov, &out, &in);
1053         if (in)
1054                 bad_driver_vq(vq, "Input buffers in net output queue?");
1055         /*
1056          * Send the whole thing through to /dev/net/tun.  It expects the exact
1057          * same format: what a coincidence!
1058          */
1059         if (writev(net_info->tunfd, iov, out) < 0)
1060                 warnx("Write to tun failed (%d)?", errno);
1061 
1062         /*
1063          * Done with that one; wait_for_vq_desc() will send the interrupt if
1064          * all packets are processed.
1065          */
1066         add_used(vq, head, 0);
1067 }
1068 
1069 /*
1070  * Handling network input is a bit trickier, because I've tried to optimize it.
1071  *
1072  * First we have a helper routine which tells is if from this file descriptor
1073  * (ie. the /dev/net/tun device) will block:
1074  */
1075 static bool will_block(int fd)
1076 {
1077         fd_set fdset;
1078         struct timeval zero = { 0, 0 };
1079         FD_ZERO(&fdset);
1080         FD_SET(fd, &fdset);
1081         return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
1082 }
1083 
1084 /*
1085  * This handles packets coming in from the tun device to our Guest.  Like all
1086  * service routines, it gets called again as soon as it returns, so you don't
1087  * see a while(1) loop here.
1088  */
1089 static void net_input(struct virtqueue *vq)
1090 {
1091         int len;
1092         unsigned int head, out, in;
1093         struct iovec iov[vq->vring.num];
1094         struct net_info *net_info = vq->dev->priv;
1095 
1096         /*
1097          * Get a descriptor to write an incoming packet into.  This will also
1098          * send an interrupt if they're out of descriptors.
1099          */
1100         head = wait_for_vq_desc(vq, iov, &out, &in);
1101         if (out)
1102                 bad_driver_vq(vq, "Output buffers in net input queue?");
1103 
1104         /*
1105          * If it looks like we'll block reading from the tun device, send them
1106          * an interrupt.
1107          */
1108         if (vq->pending_used && will_block(net_info->tunfd))
1109                 trigger_irq(vq);
1110 
1111         /*
1112          * Read in the packet.  This is where we normally wait (when there's no
1113          * incoming network traffic).
1114          */
1115         len = readv(net_info->tunfd, iov, in);
1116         if (len <= 0)
1117                 warn("Failed to read from tun (%d).", errno);
1118 
1119         /*
1120          * Mark that packet buffer as used, but don't interrupt here.  We want
1121          * to wait until we've done as much work as we can.
1122          */
1123         add_used(vq, head, len);
1124 }
1125 /*:*/
1126 
1127 /* This is the helper to create threads: run the service routine in a loop. */
1128 static int do_thread(void *_vq)
1129 {
1130         struct virtqueue *vq = _vq;
1131 
1132         for (;;)
1133                 vq->service(vq);
1134         return 0;
1135 }
1136 
1137 /*
1138  * When a child dies, we kill our entire process group with SIGTERM.  This
1139  * also has the side effect that the shell restores the console for us!
1140  */
1141 static void kill_launcher(int signal)
1142 {
1143         kill(0, SIGTERM);
1144 }
1145 
1146 static void reset_vq_pci_config(struct virtqueue *vq)
1147 {
1148         vq->pci_config.queue_size = VIRTQUEUE_NUM;
1149         vq->pci_config.queue_enable = 0;
1150 }
1151 
1152 static void reset_device(struct device *dev)
1153 {
1154         struct virtqueue *vq;
1155 
1156         verbose("Resetting device %s\n", dev->name);
1157 
1158         /* Clear any features they've acked. */
1159         dev->features_accepted = 0;
1160 
1161         /* We're going to be explicitly killing threads, so ignore them. */
1162         signal(SIGCHLD, SIG_IGN);
1163 
1164         /*
1165          * 4.1.4.3.1:
1166          *
1167          *   The device MUST present a 0 in queue_enable on reset. 
1168          *
1169          * This means we set it here, and reset the saved ones in every vq.
1170          */
1171         dev->mmio->cfg.queue_enable = 0;
1172 
1173         /* Get rid of the virtqueue threads */
1174         for (vq = dev->vq; vq; vq = vq->next) {
1175                 vq->last_avail_idx = 0;
1176                 reset_vq_pci_config(vq);
1177                 if (vq->thread != (pid_t)-1) {
1178                         kill(vq->thread, SIGTERM);
1179                         waitpid(vq->thread, NULL, 0);
1180                         vq->thread = (pid_t)-1;
1181                 }
1182         }
1183         dev->running = false;
1184         dev->wrote_features_ok = false;
1185 
1186         /* Now we care if threads die. */
1187         signal(SIGCHLD, (void *)kill_launcher);
1188 }
1189 
1190 static void cleanup_devices(void)
1191 {
1192         unsigned int i;
1193 
1194         for (i = 1; i < MAX_PCI_DEVICES; i++) {
1195                 struct device *d = devices.pci[i];
1196                 if (!d)
1197                         continue;
1198                 reset_device(d);
1199         }
1200 
1201         /* If we saved off the original terminal settings, restore them now. */
1202         if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1203                 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1204 }
1205 
1206 /*L:217
1207  * We do PCI.  This is mainly done to let us test the kernel virtio PCI
1208  * code.
1209  */
1210 
1211 /* Linux expects a PCI host bridge: ours is a dummy, and first on the bus. */
1212 static struct device pci_host_bridge;
1213 
1214 static void init_pci_host_bridge(void)
1215 {
1216         pci_host_bridge.name = "PCI Host Bridge";
1217         pci_host_bridge.config.class = 0x06; /* bridge */
1218         pci_host_bridge.config.subclass = 0; /* host bridge */
1219         devices.pci[0] = &pci_host_bridge;
1220 }
1221 
1222 /* The IO ports used to read the PCI config space. */
1223 #define PCI_CONFIG_ADDR 0xCF8
1224 #define PCI_CONFIG_DATA 0xCFC
1225 
1226 /*
1227  * Not really portable, but does help readability: this is what the Guest
1228  * writes to the PCI_CONFIG_ADDR IO port.
1229  */
1230 union pci_config_addr {
1231         struct {
1232                 unsigned mbz: 2;
1233                 unsigned offset: 6;
1234                 unsigned funcnum: 3;
1235                 unsigned devnum: 5;
1236                 unsigned busnum: 8;
1237                 unsigned reserved: 7;
1238                 unsigned enabled : 1;
1239         } bits;
1240         u32 val;
1241 };
1242 
1243 /*
1244  * We cache what they wrote to the address port, so we know what they're
1245  * talking about when they access the data port.
1246  */
1247 static union pci_config_addr pci_config_addr;
1248 
1249 static struct device *find_pci_device(unsigned int index)
1250 {
1251         return devices.pci[index];
1252 }
1253 
1254 /* PCI can do 1, 2 and 4 byte reads; we handle that here. */
1255 static void ioread(u16 off, u32 v, u32 mask, u32 *val)
1256 {
1257         assert(off < 4);
1258         assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF);
1259         *val = (v >> (off * 8)) & mask;
1260 }
1261 
1262 /* PCI can do 1, 2 and 4 byte writes; we handle that here. */
1263 static void iowrite(u16 off, u32 v, u32 mask, u32 *dst)
1264 {
1265         assert(off < 4);
1266         assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF);
1267         *dst &= ~(mask << (off * 8));
1268         *dst |= (v & mask) << (off * 8);
1269 }
1270 
1271 /*
1272  * Where PCI_CONFIG_DATA accesses depends on the previous write to
1273  * PCI_CONFIG_ADDR.
1274  */
1275 static struct device *dev_and_reg(u32 *reg)
1276 {
1277         if (!pci_config_addr.bits.enabled)
1278                 return NULL;
1279 
1280         if (pci_config_addr.bits.funcnum != 0)
1281                 return NULL;
1282 
1283         if (pci_config_addr.bits.busnum != 0)
1284                 return NULL;
1285 
1286         if (pci_config_addr.bits.offset * 4 >= sizeof(struct pci_config))
1287                 return NULL;
1288 
1289         *reg = pci_config_addr.bits.offset;
1290         return find_pci_device(pci_config_addr.bits.devnum);
1291 }
1292 
1293 /*
1294  * We can get invalid combinations of values while they're writing, so we
1295  * only fault if they try to write with some invalid bar/offset/length.
1296  */
1297 static bool valid_bar_access(struct device *d,
1298                              struct virtio_pci_cfg_cap_u32 *cfg_access)
1299 {
1300         /* We only have 1 bar (BAR0) */
1301         if (cfg_access->cap.bar != 0)
1302                 return false;
1303 
1304         /* Check it's within BAR0. */
1305         if (cfg_access->cap.offset >= d->mmio_size
1306             || cfg_access->cap.offset + cfg_access->cap.length > d->mmio_size)
1307                 return false;
1308 
1309         /* Check length is 1, 2 or 4. */
1310         if (cfg_access->cap.length != 1
1311             && cfg_access->cap.length != 2
1312             && cfg_access->cap.length != 4)
1313                 return false;
1314 
1315         /*
1316          * 4.1.4.7.2:
1317          *
1318          *  The driver MUST NOT write a cap.offset which is not a multiple of
1319          *  cap.length (ie. all accesses MUST be aligned).
1320          */
1321         if (cfg_access->cap.offset % cfg_access->cap.length != 0)
1322                 return false;
1323 
1324         /* Return pointer into word in BAR0. */
1325         return true;
1326 }
1327 
1328 /* Is this accessing the PCI config address port?. */
1329 static bool is_pci_addr_port(u16 port)
1330 {
1331         return port >= PCI_CONFIG_ADDR && port < PCI_CONFIG_ADDR + 4;
1332 }
1333 
1334 static bool pci_addr_iowrite(u16 port, u32 mask, u32 val)
1335 {
1336         iowrite(port - PCI_CONFIG_ADDR, val, mask,
1337                 &pci_config_addr.val);
1338         verbose("PCI%s: %#x/%x: bus %u dev %u func %u reg %u\n",
1339                 pci_config_addr.bits.enabled ? "" : " DISABLED",
1340                 val, mask,
1341                 pci_config_addr.bits.busnum,
1342                 pci_config_addr.bits.devnum,
1343                 pci_config_addr.bits.funcnum,
1344                 pci_config_addr.bits.offset);
1345         return true;
1346 }
1347 
1348 static void pci_addr_ioread(u16 port, u32 mask, u32 *val)
1349 {
1350         ioread(port - PCI_CONFIG_ADDR, pci_config_addr.val, mask, val);
1351 }
1352 
1353 /* Is this accessing the PCI config data port?. */
1354 static bool is_pci_data_port(u16 port)
1355 {
1356         return port >= PCI_CONFIG_DATA && port < PCI_CONFIG_DATA + 4;
1357 }
1358 
1359 static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask);
1360 
1361 static bool pci_data_iowrite(u16 port, u32 mask, u32 val)
1362 {
1363         u32 reg, portoff;
1364         struct device *d = dev_and_reg(&reg);
1365 
1366         /* Complain if they don't belong to a device. */
1367         if (!d)
1368                 return false;
1369 
1370         /* They can do 1 byte writes, etc. */
1371         portoff = port - PCI_CONFIG_DATA;
1372 
1373         /*
1374          * PCI uses a weird way to determine the BAR size: the OS
1375          * writes all 1's, and sees which ones stick.
1376          */
1377         if (&d->config_words[reg] == &d->config.bar[0]) {
1378                 int i;
1379 
1380                 iowrite(portoff, val, mask, &d->config.bar[0]);
1381                 for (i = 0; (1 << i) < d->mmio_size; i++)
1382                         d->config.bar[0] &= ~(1 << i);
1383                 return true;
1384         } else if ((&d->config_words[reg] > &d->config.bar[0]
1385                     && &d->config_words[reg] <= &d->config.bar[6])
1386                    || &d->config_words[reg] == &d->config.expansion_rom_addr) {
1387                 /* Allow writing to any other BAR, or expansion ROM */
1388                 iowrite(portoff, val, mask, &d->config_words[reg]);
1389                 return true;
1390                 /* We let them override latency timer and cacheline size */
1391         } else if (&d->config_words[reg] == (void *)&d->config.cacheline_size) {
1392                 /* Only let them change the first two fields. */
1393                 if (mask == 0xFFFFFFFF)
1394                         mask = 0xFFFF;
1395                 iowrite(portoff, val, mask, &d->config_words[reg]);
1396                 return true;
1397         } else if (&d->config_words[reg] == (void *)&d->config.command
1398                    && mask == 0xFFFF) {
1399                 /* Ignore command writes. */
1400                 return true;
1401         } else if (&d->config_words[reg]
1402                    == (void *)&d->config.cfg_access.cap.bar
1403                    || &d->config_words[reg]
1404                    == &d->config.cfg_access.cap.length
1405                    || &d->config_words[reg]
1406                    == &d->config.cfg_access.cap.offset) {
1407 
1408                 /*
1409                  * The VIRTIO_PCI_CAP_PCI_CFG capability
1410                  * provides a backdoor to access the MMIO
1411                  * regions without mapping them.  Weird, but
1412                  * useful.
1413                  */
1414                 iowrite(portoff, val, mask, &d->config_words[reg]);
1415                 return true;
1416         } else if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) {
1417                 u32 write_mask;
1418 
1419                 /*
1420                  * 4.1.4.7.1:
1421                  *
1422                  *  Upon detecting driver write access to pci_cfg_data, the
1423                  *  device MUST execute a write access at offset cap.offset at
1424                  *  BAR selected by cap.bar using the first cap.length bytes
1425                  *  from pci_cfg_data.
1426                  */
1427 
1428                 /* Must be bar 0 */
1429                 if (!valid_bar_access(d, &d->config.cfg_access))
1430                         return false;
1431 
1432                 iowrite(portoff, val, mask, &d->config.cfg_access.pci_cfg_data);
1433 
1434                 /*
1435                  * Now emulate a write.  The mask we use is set by
1436                  * len, *not* this write!
1437                  */
1438                 write_mask = (1ULL<<(8*d->config.cfg_access.cap.length)) - 1;
1439                 verbose("Window writing %#x/%#x to bar %u, offset %u len %u\n",
1440                         d->config.cfg_access.pci_cfg_data, write_mask,
1441                         d->config.cfg_access.cap.bar,
1442                         d->config.cfg_access.cap.offset,
1443                         d->config.cfg_access.cap.length);
1444 
1445                 emulate_mmio_write(d, d->config.cfg_access.cap.offset,
1446                                    d->config.cfg_access.pci_cfg_data,
1447                                    write_mask);
1448                 return true;
1449         }
1450 
1451         /*
1452          * 4.1.4.1:
1453          *
1454          *  The driver MUST NOT write into any field of the capability
1455          *  structure, with the exception of those with cap_type
1456          *  VIRTIO_PCI_CAP_PCI_CFG...
1457          */
1458         return false;
1459 }
1460 
1461 static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask);
1462 
1463 static void pci_data_ioread(u16 port, u32 mask, u32 *val)
1464 {
1465         u32 reg;
1466         struct device *d = dev_and_reg(&reg);
1467 
1468         if (!d)
1469                 return;
1470 
1471         /* Read through the PCI MMIO access window is special */
1472         if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) {
1473                 u32 read_mask;
1474 
1475                 /*
1476                  * 4.1.4.7.1:
1477                  *
1478                  *  Upon detecting driver read access to pci_cfg_data, the
1479                  *  device MUST execute a read access of length cap.length at
1480                  *  offset cap.offset at BAR selected by cap.bar and store the
1481                  *  first cap.length bytes in pci_cfg_data.
1482                  */
1483                 /* Must be bar 0 */
1484                 if (!valid_bar_access(d, &d->config.cfg_access))
1485                         bad_driver(d,
1486                              "Invalid cfg_access to bar%u, offset %u len %u",
1487                              d->config.cfg_access.cap.bar,
1488                              d->config.cfg_access.cap.offset,
1489                              d->config.cfg_access.cap.length);
1490 
1491                 /*
1492                  * Read into the window.  The mask we use is set by
1493                  * len, *not* this read!
1494                  */
1495                 read_mask = (1ULL<<(8*d->config.cfg_access.cap.length))-1;
1496                 d->config.cfg_access.pci_cfg_data
1497                         = emulate_mmio_read(d,
1498                                             d->config.cfg_access.cap.offset,
1499                                             read_mask);
1500                 verbose("Window read %#x/%#x from bar %u, offset %u len %u\n",
1501                         d->config.cfg_access.pci_cfg_data, read_mask,
1502                         d->config.cfg_access.cap.bar,
1503                         d->config.cfg_access.cap.offset,
1504                         d->config.cfg_access.cap.length);
1505         }
1506         ioread(port - PCI_CONFIG_DATA, d->config_words[reg], mask, val);
1507 }
1508 
1509 /*L:216
1510  * This is where we emulate a handful of Guest instructions.  It's ugly
1511  * and we used to do it in the kernel but it grew over time.
1512  */
1513 
1514 /*
1515  * We use the ptrace syscall's pt_regs struct to talk about registers
1516  * to lguest: these macros convert the names to the offsets.
1517  */
1518 #define getreg(name) getreg_off(offsetof(struct user_regs_struct, name))
1519 #define setreg(name, val) \
1520         setreg_off(offsetof(struct user_regs_struct, name), (val))
1521 
1522 static u32 getreg_off(size_t offset)
1523 {
1524         u32 r;
1525         unsigned long args[] = { LHREQ_GETREG, offset };
1526 
1527         if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1528                 err(1, "Getting register %u", offset);
1529         if (pread(lguest_fd, &r, sizeof(r), cpu_id) != sizeof(r))
1530                 err(1, "Reading register %u", offset);
1531 
1532         return r;
1533 }
1534 
1535 static void setreg_off(size_t offset, u32 val)
1536 {
1537         unsigned long args[] = { LHREQ_SETREG, offset, val };
1538 
1539         if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1540                 err(1, "Setting register %u", offset);
1541 }
1542 
1543 /* Get register by instruction encoding */
1544 static u32 getreg_num(unsigned regnum, u32 mask)
1545 {
1546         /* 8 bit ops use regnums 4-7 for high parts of word */
1547         if (mask == 0xFF && (regnum & 0x4))
1548                 return getreg_num(regnum & 0x3, 0xFFFF) >> 8;
1549 
1550         switch (regnum) {
1551         case 0: return getreg(eax) & mask;
1552         case 1: return getreg(ecx) & mask;
1553         case 2: return getreg(edx) & mask;
1554         case 3: return getreg(ebx) & mask;
1555         case 4: return getreg(esp) & mask;
1556         case 5: return getreg(ebp) & mask;
1557         case 6: return getreg(esi) & mask;
1558         case 7: return getreg(edi) & mask;
1559         }
1560         abort();
1561 }
1562 
1563 /* Set register by instruction encoding */
1564 static void setreg_num(unsigned regnum, u32 val, u32 mask)
1565 {
1566         /* Don't try to set bits out of range */
1567         assert(~(val & ~mask));
1568 
1569         /* 8 bit ops use regnums 4-7 for high parts of word */
1570         if (mask == 0xFF && (regnum & 0x4)) {
1571                 /* Construct the 16 bits we want. */
1572                 val = (val << 8) | getreg_num(regnum & 0x3, 0xFF);
1573                 setreg_num(regnum & 0x3, val, 0xFFFF);
1574                 return;
1575         }
1576 
1577         switch (regnum) {
1578         case 0: setreg(eax, val | (getreg(eax) & ~mask)); return;
1579         case 1: setreg(ecx, val | (getreg(ecx) & ~mask)); return;
1580         case 2: setreg(edx, val | (getreg(edx) & ~mask)); return;
1581         case 3: setreg(ebx, val | (getreg(ebx) & ~mask)); return;
1582         case 4: setreg(esp, val | (getreg(esp) & ~mask)); return;
1583         case 5: setreg(ebp, val | (getreg(ebp) & ~mask)); return;
1584         case 6: setreg(esi, val | (getreg(esi) & ~mask)); return;
1585         case 7: setreg(edi, val | (getreg(edi) & ~mask)); return;
1586         }
1587         abort();
1588 }
1589 
1590 /* Get bytes of displacement appended to instruction, from r/m encoding */
1591 static u32 insn_displacement_len(u8 mod_reg_rm)
1592 {
1593         /* Switch on the mod bits */
1594         switch (mod_reg_rm >> 6) {
1595         case 0:
1596                 /* If mod == 0, and r/m == 101, 16-bit displacement follows */
1597                 if ((mod_reg_rm & 0x7) == 0x5)
1598                         return 2;
1599                 /* Normally, mod == 0 means no literal displacement */
1600                 return 0;
1601         case 1:
1602                 /* One byte displacement */
1603                 return 1;
1604         case 2:
1605                 /* Four byte displacement */
1606                 return 4;
1607         case 3:
1608                 /* Register mode */
1609                 return 0;
1610         }
1611         abort();
1612 }
1613 
1614 static void emulate_insn(const u8 insn[])
1615 {
1616         unsigned long args[] = { LHREQ_TRAP, 13 };
1617         unsigned int insnlen = 0, in = 0, small_operand = 0, byte_access;
1618         unsigned int eax, port, mask;
1619         /*
1620          * Default is to return all-ones on IO port reads, which traditionally
1621          * means "there's nothing there".
1622          */
1623         u32 val = 0xFFFFFFFF;
1624 
1625         /*
1626          * This must be the Guest kernel trying to do something, not userspace!
1627          * The bottom two bits of the CS segment register are the privilege
1628          * level.
1629          */
1630         if ((getreg(xcs) & 3) != 0x1)
1631                 goto no_emulate;
1632 
1633         /* Decoding x86 instructions is icky. */
1634 
1635         /*
1636          * Around 2.6.33, the kernel started using an emulation for the
1637          * cmpxchg8b instruction in early boot on many configurations.  This
1638          * code isn't paravirtualized, and it tries to disable interrupts.
1639          * Ignore it, which will Mostly Work.
1640          */
1641         if (insn[insnlen] == 0xfa) {
1642                 /* "cli", or Clear Interrupt Enable instruction.  Skip it. */
1643                 insnlen = 1;
1644                 goto skip_insn;
1645         }
1646 
1647         /*
1648          * 0x66 is an "operand prefix".  It means a 16, not 32 bit in/out.
1649          */
1650         if (insn[insnlen] == 0x66) {
1651                 small_operand = 1;
1652                 /* The instruction is 1 byte so far, read the next byte. */
1653                 insnlen = 1;
1654         }
1655 
1656         /* If the lower bit isn't set, it's a single byte access */
1657         byte_access = !(insn[insnlen] & 1);
1658 
1659         /*
1660          * Now we can ignore the lower bit and decode the 4 opcodes
1661          * we need to emulate.
1662          */
1663         switch (insn[insnlen] & 0xFE) {
1664         case 0xE4: /* in     <next byte>,%al */
1665                 port = insn[insnlen+1];
1666                 insnlen += 2;
1667                 in = 1;
1668                 break;
1669         case 0xEC: /* in     (%dx),%al */
1670                 port = getreg(edx) & 0xFFFF;
1671                 insnlen += 1;
1672                 in = 1;
1673                 break;
1674         case 0xE6: /* out    %al,<next byte> */
1675                 port = insn[insnlen+1];
1676                 insnlen += 2;
1677                 break;
1678         case 0xEE: /* out    %al,(%dx) */
1679                 port = getreg(edx) & 0xFFFF;
1680                 insnlen += 1;
1681                 break;
1682         default:
1683                 /* OK, we don't know what this is, can't emulate. */
1684                 goto no_emulate;
1685         }
1686 
1687         /* Set a mask of the 1, 2 or 4 bytes, depending on size of IO */
1688         if (byte_access)
1689                 mask = 0xFF;
1690         else if (small_operand)
1691                 mask = 0xFFFF;
1692         else
1693                 mask = 0xFFFFFFFF;
1694 
1695         /*
1696          * If it was an "IN" instruction, they expect the result to be read
1697          * into %eax, so we change %eax.
1698          */
1699         eax = getreg(eax);
1700 
1701         if (in) {
1702                 /* This is the PS/2 keyboard status; 1 means ready for output */
1703                 if (port == 0x64)
1704                         val = 1;
1705                 else if (is_pci_addr_port(port))
1706                         pci_addr_ioread(port, mask, &val);
1707                 else if (is_pci_data_port(port))
1708                         pci_data_ioread(port, mask, &val);
1709 
1710                 /* Clear the bits we're about to read */
1711                 eax &= ~mask;
1712                 /* Copy bits in from val. */
1713                 eax |= val & mask;
1714                 /* Now update the register. */
1715                 setreg(eax, eax);
1716         } else {
1717                 if (is_pci_addr_port(port)) {
1718                         if (!pci_addr_iowrite(port, mask, eax))
1719                                 goto bad_io;
1720                 } else if (is_pci_data_port(port)) {
1721                         if (!pci_data_iowrite(port, mask, eax))
1722                                 goto bad_io;
1723                 }
1724                 /* There are many other ports, eg. CMOS clock, serial
1725                  * and parallel ports, so we ignore them all. */
1726         }
1727 
1728         verbose("IO %s of %x to %u: %#08x\n",
1729                 in ? "IN" : "OUT", mask, port, eax);
1730 skip_insn:
1731         /* Finally, we've "done" the instruction, so move past it. */
1732         setreg(eip, getreg(eip) + insnlen);
1733         return;
1734 
1735 bad_io:
1736         warnx("Attempt to %s port %u (%#x mask)",
1737               in ? "read from" : "write to", port, mask);
1738 
1739 no_emulate:
1740         /* Inject trap into Guest. */
1741         if (write(lguest_fd, args, sizeof(args)) < 0)
1742                 err(1, "Reinjecting trap 13 for fault at %#x", getreg(eip));
1743 }
1744 
1745 static struct device *find_mmio_region(unsigned long paddr, u32 *off)
1746 {
1747         unsigned int i;
1748 
1749         for (i = 1; i < MAX_PCI_DEVICES; i++) {
1750                 struct device *d = devices.pci[i];
1751 
1752                 if (!d)
1753                         continue;
1754                 if (paddr < d->mmio_addr)
1755                         continue;
1756                 if (paddr >= d->mmio_addr + d->mmio_size)
1757                         continue;
1758                 *off = paddr - d->mmio_addr;
1759                 return d;
1760         }
1761         return NULL;
1762 }
1763 
1764 /* FIXME: Use vq array. */
1765 static struct virtqueue *vq_by_num(struct device *d, u32 num)
1766 {
1767         struct virtqueue *vq = d->vq;
1768 
1769         while (num-- && vq)
1770                 vq = vq->next;
1771 
1772         return vq;
1773 }
1774 
1775 static void save_vq_config(const struct virtio_pci_common_cfg *cfg,
1776                            struct virtqueue *vq)
1777 {
1778         vq->pci_config = *cfg;
1779 }
1780 
1781 static void restore_vq_config(struct virtio_pci_common_cfg *cfg,
1782                               struct virtqueue *vq)
1783 {
1784         /* Only restore the per-vq part */
1785         size_t off = offsetof(struct virtio_pci_common_cfg, queue_size);
1786 
1787         memcpy((void *)cfg + off, (void *)&vq->pci_config + off,
1788                sizeof(*cfg) - off);
1789 }
1790 
1791 /*
1792  * 4.1.4.3.2:
1793  *
1794  *  The driver MUST configure the other virtqueue fields before
1795  *  enabling the virtqueue with queue_enable.
1796  *
1797  * When they enable the virtqueue, we check that their setup is valid.
1798  */
1799 static void check_virtqueue(struct device *d, struct virtqueue *vq)
1800 {
1801         /* Because lguest is 32 bit, all the descriptor high bits must be 0 */
1802         if (vq->pci_config.queue_desc_hi
1803             || vq->pci_config.queue_avail_hi
1804             || vq->pci_config.queue_used_hi)
1805                 bad_driver_vq(vq, "invalid 64-bit queue address");
1806 
1807         /*
1808          * 2.4.1:
1809          *
1810          *  The driver MUST ensure that the physical address of the first byte
1811          *  of each virtqueue part is a multiple of the specified alignment
1812          *  value in the above table.
1813          */
1814         if (vq->pci_config.queue_desc_lo % 16
1815             || vq->pci_config.queue_avail_lo % 2
1816             || vq->pci_config.queue_used_lo % 4)
1817                 bad_driver_vq(vq, "invalid alignment in queue addresses");
1818 
1819         /* Initialize the virtqueue and check they're all in range. */
1820         vq->vring.num = vq->pci_config.queue_size;
1821         vq->vring.desc = check_pointer(vq->dev,
1822                                        vq->pci_config.queue_desc_lo,
1823                                        sizeof(*vq->vring.desc) * vq->vring.num);
1824         vq->vring.avail = check_pointer(vq->dev,
1825                                         vq->pci_config.queue_avail_lo,
1826                                         sizeof(*vq->vring.avail)
1827                                         + (sizeof(vq->vring.avail->ring[0])
1828                                            * vq->vring.num));
1829         vq->vring.used = check_pointer(vq->dev,
1830                                        vq->pci_config.queue_used_lo,
1831                                        sizeof(*vq->vring.used)
1832                                        + (sizeof(vq->vring.used->ring[0])
1833                                           * vq->vring.num));
1834 
1835         /*
1836          * 2.4.9.1:
1837          *
1838          *   The driver MUST initialize flags in the used ring to 0
1839          *   when allocating the used ring.
1840          */
1841         if (vq->vring.used->flags != 0)
1842                 bad_driver_vq(vq, "invalid initial used.flags %#x",
1843                               vq->vring.used->flags);
1844 }
1845 
1846 static void start_virtqueue(struct virtqueue *vq)
1847 {
1848         /*
1849          * Create stack for thread.  Since the stack grows upwards, we point
1850          * the stack pointer to the end of this region.
1851          */
1852         char *stack = malloc(32768);
1853 
1854         /* Create a zero-initialized eventfd. */
1855         vq->eventfd = eventfd(0, 0);
1856         if (vq->eventfd < 0)
1857                 err(1, "Creating eventfd");
1858 
1859         /*
1860          * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1861          * we get a signal if it dies.
1862          */
1863         vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1864         if (vq->thread == (pid_t)-1)
1865                 err(1, "Creating clone");
1866 }
1867 
1868 static void start_virtqueues(struct device *d)
1869 {
1870         struct virtqueue *vq;
1871 
1872         for (vq = d->vq; vq; vq = vq->next) {
1873                 if (vq->pci_config.queue_enable)
1874                         start_virtqueue(vq);
1875         }
1876 }
1877 
1878 static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask)
1879 {
1880         struct virtqueue *vq;
1881 
1882         switch (off) {
1883         case offsetof(struct virtio_pci_mmio, cfg.device_feature_select):
1884                 /*
1885                  * 4.1.4.3.1:
1886                  *
1887                  * The device MUST present the feature bits it is offering in
1888                  * device_feature, starting at bit device_feature_select ∗ 32
1889                  * for any device_feature_select written by the driver
1890                  */
1891                 if (val == 0)
1892                         d->mmio->cfg.device_feature = d->features;
1893                 else if (val == 1)
1894                         d->mmio->cfg.device_feature = (d->features >> 32);
1895                 else
1896                         d->mmio->cfg.device_feature = 0;
1897                 goto feature_write_through32;
1898         case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select):
1899                 if (val > 1)
1900                         bad_driver(d, "Unexpected driver select %u", val);
1901                 goto feature_write_through32;
1902         case offsetof(struct virtio_pci_mmio, cfg.guest_feature):
1903                 if (d->mmio->cfg.guest_feature_select == 0) {
1904                         d->features_accepted &= ~((u64)0xFFFFFFFF);
1905                         d->features_accepted |= val;
1906                 } else {
1907                         assert(d->mmio->cfg.guest_feature_select == 1);
1908                         d->features_accepted &= 0xFFFFFFFF;
1909                         d->features_accepted |= ((u64)val) << 32;
1910                 }
1911                 /*
1912                  * 2.2.1:
1913                  *
1914                  *   The driver MUST NOT accept a feature which the device did
1915                  *   not offer
1916                  */
1917                 if (d->features_accepted & ~d->features)
1918                         bad_driver(d, "over-accepted features %#llx of %#llx",
1919                                    d->features_accepted, d->features);
1920                 goto feature_write_through32;
1921         case offsetof(struct virtio_pci_mmio, cfg.device_status): {
1922                 u8 prev;
1923 
1924                 verbose("%s: device status -> %#x\n", d->name, val);
1925                 /*
1926                  * 4.1.4.3.1:
1927                  * 
1928                  *  The device MUST reset when 0 is written to device_status,
1929                  *  and present a 0 in device_status once that is done.
1930                  */
1931                 if (val == 0) {
1932                         reset_device(d);
1933                         goto write_through8;
1934                 }
1935 
1936                 /* 2.1.1: The driver MUST NOT clear a device status bit. */
1937                 if (d->mmio->cfg.device_status & ~val)
1938                         bad_driver(d, "unset of device status bit %#x -> %#x",
1939                                    d->mmio->cfg.device_status, val);
1940 
1941                 /*
1942                  * 2.1.2:
1943                  *
1944                  *  The device MUST NOT consume buffers or notify the driver
1945                  *  before DRIVER_OK.
1946                  */
1947                 if (val & VIRTIO_CONFIG_S_DRIVER_OK
1948                     && !(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK))
1949                         start_virtqueues(d);
1950 
1951                 /*
1952                  * 3.1.1:
1953                  *
1954                  *   The driver MUST follow this sequence to initialize a device:
1955                  *   - Reset the device.
1956                  *   - Set the ACKNOWLEDGE status bit: the guest OS has
1957                  *     notice the device.
1958                  *   - Set the DRIVER status bit: the guest OS knows how
1959                  *     to drive the device.
1960                  *   - Read device feature bits, and write the subset
1961                  *     of feature bits understood by the OS and driver
1962                  *     to the device. During this step the driver MAY
1963                  *     read (but MUST NOT write) the device-specific
1964                  *     configuration fields to check that it can
1965                  *     support the device before accepting it.
1966                  *   - Set the FEATURES_OK status bit.  The driver
1967                  *     MUST not accept new feature bits after this
1968                  *     step.
1969                  *   - Re-read device status to ensure the FEATURES_OK
1970                  *     bit is still set: otherwise, the device does
1971                  *     not support our subset of features and the
1972                  *     device is unusable.
1973                  *   - Perform device-specific setup, including
1974                  *     discovery of virtqueues for the device,
1975                  *     optional per-bus setup, reading and possibly
1976                  *     writing the device’s virtio configuration
1977                  *     space, and population of virtqueues.
1978                  *   - Set the DRIVER_OK status bit. At this point the
1979                  *     device is “live”.
1980                  */
1981                 prev = 0;
1982                 switch (val & ~d->mmio->cfg.device_status) {
1983                 case VIRTIO_CONFIG_S_DRIVER_OK:
1984                         prev |= VIRTIO_CONFIG_S_FEATURES_OK; /* fall thru */
1985                 case VIRTIO_CONFIG_S_FEATURES_OK:
1986                         prev |= VIRTIO_CONFIG_S_DRIVER; /* fall thru */
1987                 case VIRTIO_CONFIG_S_DRIVER:
1988                         prev |= VIRTIO_CONFIG_S_ACKNOWLEDGE; /* fall thru */
1989                 case VIRTIO_CONFIG_S_ACKNOWLEDGE:
1990                         break;
1991                 default:
1992                         bad_driver(d, "unknown device status bit %#x -> %#x",
1993                                    d->mmio->cfg.device_status, val);
1994                 }
1995                 if (d->mmio->cfg.device_status != prev)
1996                         bad_driver(d, "unexpected status transition %#x -> %#x",
1997                                    d->mmio->cfg.device_status, val);
1998 
1999                 /* If they just wrote FEATURES_OK, we make sure they read */
2000                 switch (val & ~d->mmio->cfg.device_status) {
2001                 case VIRTIO_CONFIG_S_FEATURES_OK:
2002                         d->wrote_features_ok = true;
2003                         break;
2004                 case VIRTIO_CONFIG_S_DRIVER_OK:
2005                         if (d->wrote_features_ok)
2006                                 bad_driver(d, "did not re-read FEATURES_OK");
2007                         break;
2008                 }
2009                 goto write_through8;
2010         }
2011         case offsetof(struct virtio_pci_mmio, cfg.queue_select):
2012                 vq = vq_by_num(d, val);
2013                 /*
2014                  * 4.1.4.3.1:
2015                  *
2016                  *  The device MUST present a 0 in queue_size if the virtqueue
2017                  *  corresponding to the current queue_select is unavailable.
2018                  */
2019                 if (!vq) {
2020                         d->mmio->cfg.queue_size = 0;
2021                         goto write_through16;
2022                 }
2023                 /* Save registers for old vq, if it was a valid vq */
2024                 if (d->mmio->cfg.queue_size)
2025                         save_vq_config(&d->mmio->cfg,
2026                                        vq_by_num(d, d->mmio->cfg.queue_select));
2027                 /* Restore the registers for the queue they asked for */
2028                 restore_vq_config(&d->mmio->cfg, vq);
2029                 goto write_through16;
2030         case offsetof(struct virtio_pci_mmio, cfg.queue_size):
2031                 /*
2032                  * 4.1.4.3.2:
2033                  *
2034                  *  The driver MUST NOT write a value which is not a power of 2
2035                  *  to queue_size.
2036                  */
2037                 if (val & (val-1))
2038                         bad_driver(d, "invalid queue size %u", val);
2039                 if (d->mmio->cfg.queue_enable)
2040                         bad_driver(d, "changing queue size on live device");
2041                 goto write_through16;
2042         case offsetof(struct virtio_pci_mmio, cfg.queue_msix_vector):
2043                 bad_driver(d, "attempt to set MSIX vector to %u", val);
2044         case offsetof(struct virtio_pci_mmio, cfg.queue_enable): {
2045                 struct virtqueue *vq = vq_by_num(d, d->mmio->cfg.queue_select);
2046 
2047                 /*
2048                  * 4.1.4.3.2:
2049                  *
2050                  *  The driver MUST NOT write a 0 to queue_enable.
2051                  */
2052                 if (val != 1)
2053                         bad_driver(d, "setting queue_enable to %u", val);
2054 
2055                 /*
2056                  * 3.1.1:
2057                  *
2058                  *  7. Perform device-specific setup, including discovery of
2059                  *     virtqueues for the device, optional per-bus setup,
2060                  *     reading and possibly writing the device’s virtio
2061                  *     configuration space, and population of virtqueues.
2062                  *  8. Set the DRIVER_OK status bit.
2063                  *
2064                  * All our devices require all virtqueues to be enabled, so
2065                  * they should have done that before setting DRIVER_OK.
2066                  */
2067                 if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK)
2068                         bad_driver(d, "enabling vq after DRIVER_OK");
2069 
2070                 d->mmio->cfg.queue_enable = val;
2071                 save_vq_config(&d->mmio->cfg, vq);
2072                 check_virtqueue(d, vq);
2073                 goto write_through16;
2074         }
2075         case offsetof(struct virtio_pci_mmio, cfg.queue_notify_off):
2076                 bad_driver(d, "attempt to write to queue_notify_off");
2077         case offsetof(struct virtio_pci_mmio, cfg.queue_desc_lo):
2078         case offsetof(struct virtio_pci_mmio, cfg.queue_desc_hi):
2079         case offsetof(struct virtio_pci_mmio, cfg.queue_avail_lo):
2080         case offsetof(struct virtio_pci_mmio, cfg.queue_avail_hi):
2081         case offsetof(struct virtio_pci_mmio, cfg.queue_used_lo):
2082         case offsetof(struct virtio_pci_mmio, cfg.queue_used_hi):
2083                 /*
2084                  * 4.1.4.3.2:
2085                  *
2086                  *  The driver MUST configure the other virtqueue fields before
2087                  *  enabling the virtqueue with queue_enable.
2088                  */
2089                 if (d->mmio->cfg.queue_enable)
2090                         bad_driver(d, "changing queue on live device");
2091 
2092                 /*
2093                  * 3.1.1:
2094                  *
2095                  *  The driver MUST follow this sequence to initialize a device:
2096                  *...
2097                  *  5. Set the FEATURES_OK status bit. The driver MUST not
2098                  *  accept new feature bits after this step.
2099                  */
2100                 if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK))
2101                         bad_driver(d, "setting up vq before FEATURES_OK");
2102 
2103                 /*
2104                  *  6. Re-read device status to ensure the FEATURES_OK bit is
2105                  *     still set...
2106                  */
2107                 if (d->wrote_features_ok)
2108                         bad_driver(d, "didn't re-read FEATURES_OK before setup");
2109 
2110                 goto write_through32;
2111         case offsetof(struct virtio_pci_mmio, notify):
2112                 vq = vq_by_num(d, val);
2113                 if (!vq)
2114                         bad_driver(d, "Invalid vq notification on %u", val);
2115                 /* Notify the process handling this vq by adding 1 to eventfd */
2116                 write(vq->eventfd, "\1\0\0\0\0\0\0\0", 8);
2117                 goto write_through16;
2118         case offsetof(struct virtio_pci_mmio, isr):
2119                 bad_driver(d, "Unexpected write to isr");
2120         /* Weird corner case: write to emerg_wr of console */
2121         case sizeof(struct virtio_pci_mmio)
2122                 + offsetof(struct virtio_console_config, emerg_wr):
2123                 if (strcmp(d->name, "console") == 0) {
2124                         char c = val;
2125                         write(STDOUT_FILENO, &c, 1);
2126                         goto write_through32;
2127                 }
2128                 /* Fall through... */
2129         default:
2130                 /*
2131                  * 4.1.4.3.2:
2132                  *
2133                  *   The driver MUST NOT write to device_feature, num_queues,
2134                  *   config_generation or queue_notify_off.
2135                  */
2136                 bad_driver(d, "Unexpected write to offset %u", off);
2137         }
2138 
2139 feature_write_through32:
2140         /*
2141          * 3.1.1:
2142          *
2143          *   The driver MUST follow this sequence to initialize a device:
2144          *...
2145          *   - Set the DRIVER status bit: the guest OS knows how
2146          *     to drive the device.
2147          *   - Read device feature bits, and write the subset
2148          *     of feature bits understood by the OS and driver
2149          *     to the device.
2150          *...
2151          *   - Set the FEATURES_OK status bit. The driver MUST not
2152          *     accept new feature bits after this step.
2153          */
2154         if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER))
2155                 bad_driver(d, "feature write before VIRTIO_CONFIG_S_DRIVER");
2156         if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK)
2157                 bad_driver(d, "feature write after VIRTIO_CONFIG_S_FEATURES_OK");
2158 
2159         /*
2160          * 4.1.3.1:
2161          *
2162          *  The driver MUST access each field using the “natural” access
2163          *  method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for
2164          *  16-bit fields and 8-bit accesses for 8-bit fields.
2165          */
2166 write_through32:
2167         if (mask != 0xFFFFFFFF) {
2168                 bad_driver(d, "non-32-bit write to offset %u (%#x)",
2169                            off, getreg(eip));
2170                 return;
2171         }
2172         memcpy((char *)d->mmio + off, &val, 4);
2173         return;
2174 
2175 write_through16:
2176         if (mask != 0xFFFF)
2177                 bad_driver(d, "non-16-bit write to offset %u (%#x)",
2178                            off, getreg(eip));
2179         memcpy((char *)d->mmio + off, &val, 2);
2180         return;
2181 
2182 write_through8:
2183         if (mask != 0xFF)
2184                 bad_driver(d, "non-8-bit write to offset %u (%#x)",
2185                            off, getreg(eip));
2186         memcpy((char *)d->mmio + off, &val, 1);
2187         return;
2188 }
2189 
2190 static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask)
2191 {
2192         u8 isr;
2193         u32 val = 0;
2194 
2195         switch (off) {
2196         case offsetof(struct virtio_pci_mmio, cfg.device_feature_select):
2197         case offsetof(struct virtio_pci_mmio, cfg.device_feature):
2198         case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select):
2199         case offsetof(struct virtio_pci_mmio, cfg.guest_feature):
2200                 /*
2201                  * 3.1.1:
2202                  *
2203                  *   The driver MUST follow this sequence to initialize a device:
2204                  *...
2205                  *   - Set the DRIVER status bit: the guest OS knows how
2206                  *     to drive the device.
2207                  *   - Read device feature bits, and write the subset
2208                  *     of feature bits understood by the OS and driver
2209                  *     to the device.
2210                  */
2211                 if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER))
2212                         bad_driver(d,
2213                                    "feature read before VIRTIO_CONFIG_S_DRIVER");
2214                 goto read_through32;
2215         case offsetof(struct virtio_pci_mmio, cfg.msix_config):
2216                 bad_driver(d, "read of msix_config");
2217         case offsetof(struct virtio_pci_mmio, cfg.num_queues):
2218                 goto read_through16;
2219         case offsetof(struct virtio_pci_mmio, cfg.device_status):
2220                 /* As they did read, any write of FEATURES_OK is now fine. */
2221                 d->wrote_features_ok = false;
2222                 goto read_through8;
2223         case offsetof(struct virtio_pci_mmio, cfg.config_generation):
2224                 /*
2225                  * 4.1.4.3.1:
2226                  *
2227                  *  The device MUST present a changed config_generation after
2228                  *  the driver has read a device-specific configuration value
2229                  *  which has changed since any part of the device-specific
2230                  *  configuration was last read.
2231                  *
2232                  * This is simple: none of our devices change config, so this
2233                  * is always 0.
2234                  */
2235                 goto read_through8;
2236         case offsetof(struct virtio_pci_mmio, notify):
2237                 /*
2238                  * 3.1.1:
2239                  *
2240                  *   The driver MUST NOT notify the device before setting
2241                  *   DRIVER_OK.
2242                  */
2243                 if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK))
2244                         bad_driver(d, "notify before VIRTIO_CONFIG_S_DRIVER_OK");
2245                 goto read_through16;
2246         case offsetof(struct virtio_pci_mmio, isr):
2247                 if (mask != 0xFF)
2248                         bad_driver(d, "non-8-bit read from offset %u (%#x)",
2249                                    off, getreg(eip));
2250                 isr = d->mmio->isr;
2251                 /*
2252                  * 4.1.4.5.1:
2253                  *
2254                  *  The device MUST reset ISR status to 0 on driver read. 
2255                  */
2256                 d->mmio->isr = 0;
2257                 return isr;
2258         case offsetof(struct virtio_pci_mmio, padding):
2259                 bad_driver(d, "read from padding (%#x)", getreg(eip));
2260         default:
2261                 /* Read from device config space, beware unaligned overflow */
2262                 if (off > d->mmio_size - 4)
2263                         bad_driver(d, "read past end (%#x)", getreg(eip));
2264 
2265                 /*
2266                  * 3.1.1:
2267                  *  The driver MUST follow this sequence to initialize a device:
2268                  *...
2269                  *  3. Set the DRIVER status bit: the guest OS knows how to
2270                  *  drive the device.
2271                  *  4. Read device feature bits, and write the subset of
2272                  *  feature bits understood by the OS and driver to the
2273                  *  device. During this step the driver MAY read (but MUST NOT
2274                  *  write) the device-specific configuration fields to check
2275                  *  that it can support the device before accepting it.
2276                  */
2277                 if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER))
2278                         bad_driver(d,
2279                                    "config read before VIRTIO_CONFIG_S_DRIVER");
2280 
2281                 if (mask == 0xFFFFFFFF)
2282                         goto read_through32;
2283                 else if (mask == 0xFFFF)
2284                         goto read_through16;
2285                 else
2286                         goto read_through8;
2287         }
2288 
2289         /*
2290          * 4.1.3.1:
2291          *
2292          *  The driver MUST access each field using the “natural” access
2293          *  method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for
2294          *  16-bit fields and 8-bit accesses for 8-bit fields.
2295          */
2296 read_through32:
2297         if (mask != 0xFFFFFFFF)
2298                 bad_driver(d, "non-32-bit read to offset %u (%#x)",
2299                            off, getreg(eip));
2300         memcpy(&val, (char *)d->mmio + off, 4);
2301         return val;
2302 
2303 read_through16:
2304         if (mask != 0xFFFF)
2305                 bad_driver(d, "non-16-bit read to offset %u (%#x)",
2306                            off, getreg(eip));
2307         memcpy(&val, (char *)d->mmio + off, 2);
2308         return val;
2309 
2310 read_through8:
2311         if (mask != 0xFF)
2312                 bad_driver(d, "non-8-bit read to offset %u (%#x)",
2313                            off, getreg(eip));
2314         memcpy(&val, (char *)d->mmio + off, 1);
2315         return val;
2316 }
2317 
2318 static void emulate_mmio(unsigned long paddr, const u8 *insn)
2319 {
2320         u32 val, off, mask = 0xFFFFFFFF, insnlen = 0;
2321         struct device *d = find_mmio_region(paddr, &off);
2322         unsigned long args[] = { LHREQ_TRAP, 14 };
2323 
2324         if (!d) {
2325                 warnx("MMIO touching %#08lx (not a device)", paddr);
2326                 goto reinject;
2327         }
2328 
2329         /* Prefix makes it a 16 bit op */
2330         if (insn[0] == 0x66) {
2331                 mask = 0xFFFF;
2332                 insnlen++;
2333         }
2334 
2335         /* iowrite */
2336         if (insn[insnlen] == 0x89) {
2337                 /* Next byte is r/m byte: bits 3-5 are register. */
2338                 val = getreg_num((insn[insnlen+1] >> 3) & 0x7, mask);
2339                 emulate_mmio_write(d, off, val, mask);
2340                 insnlen += 2 + insn_displacement_len(insn[insnlen+1]);
2341         } else if (insn[insnlen] == 0x8b) { /* ioread */
2342                 /* Next byte is r/m byte: bits 3-5 are register. */
2343                 val = emulate_mmio_read(d, off, mask);
2344                 setreg_num((insn[insnlen+1] >> 3) & 0x7, val, mask);
2345                 insnlen += 2 + insn_displacement_len(insn[insnlen+1]);
2346         } else if (insn[0] == 0x88) { /* 8-bit iowrite */
2347                 mask = 0xff;
2348                 /* Next byte is r/m byte: bits 3-5 are register. */
2349                 val = getreg_num((insn[1] >> 3) & 0x7, mask);
2350                 emulate_mmio_write(d, off, val, mask);
2351                 insnlen = 2 + insn_displacement_len(insn[1]);
2352         } else if (insn[0] == 0x8a) { /* 8-bit ioread */
2353                 mask = 0xff;
2354                 val = emulate_mmio_read(d, off, mask);
2355                 setreg_num((insn[1] >> 3) & 0x7, val, mask);
2356                 insnlen = 2 + insn_displacement_len(insn[1]);
2357         } else {
2358                 warnx("Unknown MMIO instruction touching %#08lx:"
2359                      " %02x %02x %02x %02x at %u",
2360                      paddr, insn[0], insn[1], insn[2], insn[3], getreg(eip));
2361         reinject:
2362                 /* Inject trap into Guest. */
2363                 if (write(lguest_fd, args, sizeof(args)) < 0)
2364                         err(1, "Reinjecting trap 14 for fault at %#x",
2365                             getreg(eip));
2366                 return;
2367         }
2368 
2369         /* Finally, we've "done" the instruction, so move past it. */
2370         setreg(eip, getreg(eip) + insnlen);
2371 }
2372 
2373 /*L:190
2374  * Device Setup
2375  *
2376  * All devices need a descriptor so the Guest knows it exists, and a "struct
2377  * device" so the Launcher can keep track of it.  We have common helper
2378  * routines to allocate and manage them.
2379  */
2380 static void add_pci_virtqueue(struct device *dev,
2381                               void (*service)(struct virtqueue *),
2382                               const char *name)
2383 {
2384         struct virtqueue **i, *vq = malloc(sizeof(*vq));
2385 
2386         /* Initialize the virtqueue */
2387         vq->next = NULL;
2388         vq->last_avail_idx = 0;
2389         vq->dev = dev;
2390         vq->name = name;
2391 
2392         /*
2393          * This is the routine the service thread will run, and its Process ID
2394          * once it's running.
2395          */
2396         vq->service = service;
2397         vq->thread = (pid_t)-1;
2398 
2399         /* Initialize the configuration. */
2400         reset_vq_pci_config(vq);
2401         vq->pci_config.queue_notify_off = 0;
2402 
2403         /* Add one to the number of queues */
2404         vq->dev->mmio->cfg.num_queues++;
2405 
2406         /*
2407          * Add to tail of list, so dev->vq is first vq, dev->vq->next is
2408          * second.
2409          */
2410         for (i = &dev->vq; *i; i = &(*i)->next);
2411         *i = vq;
2412 }
2413 
2414 /* The Guest accesses the feature bits via the PCI common config MMIO region */
2415 static void add_pci_feature(struct device *dev, unsigned bit)
2416 {
2417         dev->features |= (1ULL << bit);
2418 }
2419 
2420 /* For devices with no config. */
2421 static void no_device_config(struct device *dev)
2422 {
2423         dev->mmio_addr = get_mmio_region(dev->mmio_size);
2424 
2425         dev->config.bar[0] = dev->mmio_addr;
2426         /* Bottom 4 bits must be zero */
2427         assert(~(dev->config.bar[0] & 0xF));
2428 }
2429 
2430 /* This puts the device config into BAR0 */
2431 static void set_device_config(struct device *dev, const void *conf, size_t len)
2432 {
2433         /* Set up BAR 0 */
2434         dev->mmio_size += len;
2435         dev->mmio = realloc(dev->mmio, dev->mmio_size);
2436         memcpy(dev->mmio + 1, conf, len);
2437 
2438         /*
2439          * 4.1.4.6:
2440          *
2441          *  The device MUST present at least one VIRTIO_PCI_CAP_DEVICE_CFG
2442          *  capability for any device type which has a device-specific
2443          *  configuration.
2444          */
2445         /* Hook up device cfg */
2446         dev->config.cfg_access.cap.cap_next
2447                 = offsetof(struct pci_config, device);
2448 
2449         /*
2450          * 4.1.4.6.1:
2451          *
2452          *  The offset for the device-specific configuration MUST be 4-byte
2453          *  aligned.
2454          */
2455         assert(dev->config.cfg_access.cap.cap_next % 4 == 0);
2456 
2457         /* Fix up device cfg field length. */
2458         dev->config.device.length = len;
2459 
2460         /* The rest is the same as the no-config case */
2461         no_device_config(dev);
2462 }
2463 
2464 static void init_cap(struct virtio_pci_cap *cap, size_t caplen, int type,
2465                      size_t bar_offset, size_t bar_bytes, u8 next)
2466 {
2467         cap->cap_vndr = PCI_CAP_ID_VNDR;
2468         cap->cap_next = next;
2469         cap->cap_len = caplen;
2470         cap->cfg_type = type;
2471         cap->bar = 0;
2472         memset(cap->padding, 0, sizeof(cap->padding));
2473         cap->offset = bar_offset;
2474         cap->length = bar_bytes;
2475 }
2476 
2477 /*
2478  * This sets up the pci_config structure, as defined in the virtio 1.0
2479  * standard (and PCI standard).
2480  */
2481 static void init_pci_config(struct pci_config *pci, u16 type,
2482                             u8 class, u8 subclass)
2483 {
2484         size_t bar_offset, bar_len;
2485 
2486         /*
2487          * 4.1.4.4.1:
2488          *
2489          *  The device MUST either present notify_off_multiplier as an even
2490          *  power of 2, or present notify_off_multiplier as 0.
2491          *
2492          * 2.1.2:
2493          *
2494          *   The device MUST initialize device status to 0 upon reset. 
2495          */
2496         memset(pci, 0, sizeof(*pci));
2497 
2498         /* 4.1.2.1: Devices MUST have the PCI Vendor ID 0x1AF4 */
2499         pci->vendor_id = 0x1AF4;
2500         /* 4.1.2.1: ... PCI Device ID calculated by adding 0x1040 ... */
2501         pci->device_id = 0x1040 + type;
2502 
2503         /*
2504          * PCI have specific codes for different types of devices.
2505          * Linux doesn't care, but it's a good clue for people looking
2506          * at the device.
2507          */
2508         pci->class = class;
2509         pci->subclass = subclass;
2510 
2511         /*
2512          * 4.1.2.1:
2513          *
2514          *  Non-transitional devices SHOULD have a PCI Revision ID of 1 or
2515          *  higher
2516          */
2517         pci->revid = 1;
2518 
2519         /*
2520          * 4.1.2.1:
2521          *
2522          *  Non-transitional devices SHOULD have a PCI Subsystem Device ID of
2523          *  0x40 or higher.
2524          */
2525         pci->subsystem_device_id = 0x40;
2526 
2527         /* We use our dummy interrupt controller, and irq_line is the irq */
2528         pci->irq_line = devices.next_irq++;
2529         pci->irq_pin = 0;
2530 
2531         /* Support for extended capabilities. */
2532         pci->status = (1 << 4);
2533 
2534         /* Link them in. */
2535         /*
2536          * 4.1.4.3.1:
2537          *
2538          *  The device MUST present at least one common configuration
2539          *  capability.
2540          */
2541         pci->capabilities = offsetof(struct pci_config, common);
2542 
2543         /* 4.1.4.3.1 ... offset MUST be 4-byte aligned. */
2544         assert(pci->capabilities % 4 == 0);
2545 
2546         bar_offset = offsetof(struct virtio_pci_mmio, cfg);
2547         bar_len = sizeof(((struct virtio_pci_mmio *)0)->cfg);
2548         init_cap(&pci->common, sizeof(pci->common), VIRTIO_PCI_CAP_COMMON_CFG,
2549                  bar_offset, bar_len,
2550                  offsetof(struct pci_config, notify));
2551 
2552         /*
2553          * 4.1.4.4.1:
2554          *
2555          *  The device MUST present at least one notification capability.
2556          */
2557         bar_offset += bar_len;
2558         bar_len = sizeof(((struct virtio_pci_mmio *)0)->notify);
2559 
2560         /*
2561          * 4.1.4.4.1:
2562          *
2563          *  The cap.offset MUST be 2-byte aligned.
2564          */
2565         assert(pci->common.cap_next % 2 == 0);
2566 
2567         /* FIXME: Use a non-zero notify_off, for per-queue notification? */
2568         /*
2569          * 4.1.4.4.1:
2570          *
2571          *  The value cap.length presented by the device MUST be at least 2 and
2572          *  MUST be large enough to support queue notification offsets for all
2573          *  supported queues in all possible configurations.
2574          */
2575         assert(bar_len >= 2);
2576 
2577         init_cap(&pci->notify.cap, sizeof(pci->notify),
2578                  VIRTIO_PCI_CAP_NOTIFY_CFG,
2579                  bar_offset, bar_len,
2580                  offsetof(struct pci_config, isr));
2581 
2582         bar_offset += bar_len;
2583         bar_len = sizeof(((struct virtio_pci_mmio *)0)->isr);
2584         /*
2585          * 4.1.4.5.1:
2586          *
2587          *  The device MUST present at least one VIRTIO_PCI_CAP_ISR_CFG
2588          *  capability.
2589          */
2590         init_cap(&pci->isr, sizeof(pci->isr),
2591                  VIRTIO_PCI_CAP_ISR_CFG,
2592                  bar_offset, bar_len,
2593                  offsetof(struct pci_config, cfg_access));
2594 
2595         /*
2596          * 4.1.4.7.1:
2597          *
2598          * The device MUST present at least one VIRTIO_PCI_CAP_PCI_CFG
2599          * capability.
2600          */
2601         /* This doesn't have any presence in the BAR */
2602         init_cap(&pci->cfg_access.cap, sizeof(pci->cfg_access),
2603                  VIRTIO_PCI_CAP_PCI_CFG,
2604                  0, 0, 0);
2605 
2606         bar_offset += bar_len + sizeof(((struct virtio_pci_mmio *)0)->padding);
2607         assert(bar_offset == sizeof(struct virtio_pci_mmio));
2608 
2609         /*
2610          * This gets sewn in and length set in set_device_config().
2611          * Some devices don't have a device configuration interface, so
2612          * we never expose this if we don't call set_device_config().
2613          */
2614         init_cap(&pci->device, sizeof(pci->device), VIRTIO_PCI_CAP_DEVICE_CFG,
2615                  bar_offset, 0, 0);
2616 }
2617 
2618 /*
2619  * This routine does all the creation and setup of a new device, but we don't
2620  * actually place the MMIO region until we know the size (if any) of the
2621  * device-specific config.  And we don't actually start the service threads
2622  * until later.
2623  *
2624  * See what I mean about userspace being boring?
2625  */
2626 static struct device *new_pci_device(const char *name, u16 type,
2627                                      u8 class, u8 subclass)
2628 {
2629         struct device *dev = malloc(sizeof(*dev));
2630 
2631         /* Now we populate the fields one at a time. */
2632         dev->name = name;
2633         dev->vq = NULL;
2634         dev->running = false;
2635         dev->wrote_features_ok = false;
2636         dev->mmio_size = sizeof(struct virtio_pci_mmio);
2637         dev->mmio = calloc(1, dev->mmio_size);
2638         dev->features = (u64)1 << VIRTIO_F_VERSION_1;
2639         dev->features_accepted = 0;
2640 
2641         if (devices.device_num + 1 >= MAX_PCI_DEVICES)
2642                 errx(1, "Can only handle 31 PCI devices");
2643 
2644         init_pci_config(&dev->config, type, class, subclass);
2645         assert(!devices.pci[devices.device_num+1]);
2646         devices.pci[++devices.device_num] = dev;
2647 
2648         return dev;
2649 }
2650 
2651 /*
2652  * Our first setup routine is the console.  It's a fairly simple device, but
2653  * UNIX tty handling makes it uglier than it could be.
2654  */
2655 static void setup_console(void)
2656 {
2657         struct device *dev;
2658         struct virtio_console_config conf;
2659 
2660         /* If we can save the initial standard input settings... */
2661         if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
2662                 struct termios term = orig_term;
2663                 /*
2664                  * Then we turn off echo, line buffering and ^C etc: We want a
2665                  * raw input stream to the Guest.
2666                  */
2667                 term.c_lflag &= ~(ISIG|ICANON|ECHO);
2668                 tcsetattr(STDIN_FILENO, TCSANOW, &term);
2669         }
2670 
2671         dev = new_pci_device("console", VIRTIO_ID_CONSOLE, 0x07, 0x00);
2672 
2673         /* We store the console state in dev->priv, and initialize it. */
2674         dev->priv = malloc(sizeof(struct console_abort));
2675         ((struct console_abort *)dev->priv)->count = 0;
2676 
2677         /*
2678          * The console needs two virtqueues: the input then the output.  When
2679          * they put something the input queue, we make sure we're listening to
2680          * stdin.  When they put something in the output queue, we write it to
2681          * stdout.
2682          */
2683         add_pci_virtqueue(dev, console_input, "input");
2684         add_pci_virtqueue(dev, console_output, "output");
2685 
2686         /* We need a configuration area for the emerg_wr early writes. */
2687         add_pci_feature(dev, VIRTIO_CONSOLE_F_EMERG_WRITE);
2688         set_device_config(dev, &conf, sizeof(conf));
2689 
2690         verbose("device %u: console\n", devices.device_num);
2691 }
2692 /*:*/
2693 
2694 /*M:010
2695  * Inter-guest networking is an interesting area.  Simplest is to have a
2696  * --sharenet=<name> option which opens or creates a named pipe.  This can be
2697  * used to send packets to another guest in a 1:1 manner.
2698  *
2699  * More sophisticated is to use one of the tools developed for project like UML
2700  * to do networking.
2701  *
2702  * Faster is to do virtio bonding in kernel.  Doing this 1:1 would be
2703  * completely generic ("here's my vring, attach to your vring") and would work
2704  * for any traffic.  Of course, namespace and permissions issues need to be
2705  * dealt with.  A more sophisticated "multi-channel" virtio_net.c could hide
2706  * multiple inter-guest channels behind one interface, although it would
2707  * require some manner of hotplugging new virtio channels.
2708  *
2709  * Finally, we could use a virtio network switch in the kernel, ie. vhost.
2710 :*/
2711 
2712 static u32 str2ip(const char *ipaddr)
2713 {
2714         unsigned int b[4];
2715 
2716         if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
2717                 errx(1, "Failed to parse IP address '%s'", ipaddr);
2718         return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
2719 }
2720 
2721 static void str2mac(const char *macaddr, unsigned char mac[6])
2722 {
2723         unsigned int m[6];
2724         if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
2725                    &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
2726                 errx(1, "Failed to parse mac address '%s'", macaddr);
2727         mac[0] = m[0];
2728         mac[1] = m[1];
2729         mac[2] = m[2];
2730         mac[3] = m[3];
2731         mac[4] = m[4];
2732         mac[5] = m[5];
2733 }
2734 
2735 /*
2736  * This code is "adapted" from libbridge: it attaches the Host end of the
2737  * network device to the bridge device specified by the command line.
2738  *
2739  * This is yet another James Morris contribution (I'm an IP-level guy, so I
2740  * dislike bridging), and I just try not to break it.
2741  */
2742 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
2743 {
2744         int ifidx;
2745         struct ifreq ifr;
2746 
2747         if (!*br_name)
2748                 errx(1, "must specify bridge name");
2749 
2750         ifidx = if_nametoindex(if_name);
2751         if (!ifidx)
2752                 errx(1, "interface %s does not exist!", if_name);
2753 
2754         strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
2755         ifr.ifr_name[IFNAMSIZ-1] = '\0';
2756         ifr.ifr_ifindex = ifidx;
2757         if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
2758                 err(1, "can't add %s to bridge %s", if_name, br_name);
2759 }
2760 
2761 /*
2762  * This sets up the Host end of the network device with an IP address, brings
2763  * it up so packets will flow, the copies the MAC address into the hwaddr
2764  * pointer.
2765  */
2766 static void configure_device(int fd, const char *tapif, u32 ipaddr)
2767 {
2768         struct ifreq ifr;
2769         struct sockaddr_in sin;
2770 
2771         memset(&ifr, 0, sizeof(ifr));
2772         strcpy(ifr.ifr_name, tapif);
2773 
2774         /* Don't read these incantations.  Just cut & paste them like I did! */
2775         sin.sin_family = AF_INET;
2776         sin.sin_addr.s_addr = htonl(ipaddr);
2777         memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
2778         if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
2779                 err(1, "Setting %s interface address", tapif);
2780         ifr.ifr_flags = IFF_UP;
2781         if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
2782                 err(1, "Bringing interface %s up", tapif);
2783 }
2784 
2785 static int get_tun_device(char tapif[IFNAMSIZ])
2786 {
2787         struct ifreq ifr;
2788         int vnet_hdr_sz;
2789         int netfd;
2790 
2791         /* Start with this zeroed.  Messy but sure. */
2792         memset(&ifr, 0, sizeof(ifr));
2793 
2794         /*
2795          * We open the /dev/net/tun device and tell it we want a tap device.  A
2796          * tap device is like a tun device, only somehow different.  To tell
2797          * the truth, I completely blundered my way through this code, but it
2798          * works now!
2799          */
2800         netfd = open_or_die("/dev/net/tun", O_RDWR);
2801         ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
2802         strcpy(ifr.ifr_name, "tap%d");
2803         if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
2804                 err(1, "configuring /dev/net/tun");
2805 
2806         if (ioctl(netfd, TUNSETOFFLOAD,
2807                   TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
2808                 err(1, "Could not set features for tun device");
2809 
2810         /*
2811          * We don't need checksums calculated for packets coming in this
2812          * device: trust us!
2813          */
2814         ioctl(netfd, TUNSETNOCSUM, 1);
2815 
2816         /*
2817          * In virtio before 1.0 (aka legacy virtio), we added a 16-bit
2818          * field at the end of the network header iff
2819          * VIRTIO_NET_F_MRG_RXBUF was negotiated.  For virtio 1.0,
2820          * that became the norm, but we need to tell the tun device
2821          * about our expanded header (which is called
2822          * virtio_net_hdr_mrg_rxbuf in the legacy system).
2823          */
2824         vnet_hdr_sz = sizeof(struct virtio_net_hdr_v1);
2825         if (ioctl(netfd, TUNSETVNETHDRSZ, &vnet_hdr_sz) != 0)
2826                 err(1, "Setting tun header size to %u", vnet_hdr_sz);
2827 
2828         memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
2829         return netfd;
2830 }
2831 
2832 /*L:195
2833  * Our network is a Host<->Guest network.  This can either use bridging or
2834  * routing, but the principle is the same: it uses the "tun" device to inject
2835  * packets into the Host as if they came in from a normal network card.  We
2836  * just shunt packets between the Guest and the tun device.
2837  */
2838 static void setup_tun_net(char *arg)
2839 {
2840         struct device *dev;
2841         struct net_info *net_info = malloc(sizeof(*net_info));
2842         int ipfd;
2843         u32 ip = INADDR_ANY;
2844         bool bridging = false;
2845         char tapif[IFNAMSIZ], *p;
2846         struct virtio_net_config conf;
2847 
2848         net_info->tunfd = get_tun_device(tapif);
2849 
2850         /* First we create a new network device. */
2851         dev = new_pci_device("net", VIRTIO_ID_NET, 0x02, 0x00);
2852         dev->priv = net_info;
2853 
2854         /* Network devices need a recv and a send queue, just like console. */
2855         add_pci_virtqueue(dev, net_input, "rx");
2856         add_pci_virtqueue(dev, net_output, "tx");
2857 
2858         /*
2859          * We need a socket to perform the magic network ioctls to bring up the
2860          * tap interface, connect to the bridge etc.  Any socket will do!
2861          */
2862         ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
2863         if (ipfd < 0)
2864                 err(1, "opening IP socket");
2865 
2866         /* If the command line was --tunnet=bridge:<name> do bridging. */
2867         if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
2868                 arg += strlen(BRIDGE_PFX);
2869                 bridging = true;
2870         }
2871 
2872         /* A mac address may follow the bridge name or IP address */
2873         p = strchr(arg, ':');
2874         if (p) {
2875                 str2mac(p+1, conf.mac);
2876                 add_pci_feature(dev, VIRTIO_NET_F_MAC);
2877                 *p = '\0';
2878         }
2879 
2880         /* arg is now either an IP address or a bridge name */
2881         if (bridging)
2882                 add_to_bridge(ipfd, tapif, arg);
2883         else
2884                 ip = str2ip(arg);
2885 
2886         /* Set up the tun device. */
2887         configure_device(ipfd, tapif, ip);
2888 
2889         /* Expect Guest to handle everything except UFO */
2890         add_pci_feature(dev, VIRTIO_NET_F_CSUM);
2891         add_pci_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
2892         add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
2893         add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
2894         add_pci_feature(dev, VIRTIO_NET_F_GUEST_ECN);
2895         add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO4);
2896         add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO6);
2897         add_pci_feature(dev, VIRTIO_NET_F_HOST_ECN);
2898         /* We handle indirect ring entries */
2899         add_pci_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
2900         set_device_config(dev, &conf, sizeof(conf));
2901 
2902         /* We don't need the socket any more; setup is done. */
2903         close(ipfd);
2904 
2905         if (bridging)
2906                 verbose("device %u: tun %s attached to bridge: %s\n",
2907                         devices.device_num, tapif, arg);
2908         else
2909                 verbose("device %u: tun %s: %s\n",
2910                         devices.device_num, tapif, arg);
2911 }
2912 /*:*/
2913 
2914 /* This hangs off device->priv. */
2915 struct vblk_info {
2916         /* The size of the file. */
2917         off64_t len;
2918 
2919         /* The file descriptor for the file. */
2920         int fd;
2921 
2922 };
2923 
2924 /*L:210
2925  * The Disk
2926  *
2927  * The disk only has one virtqueue, so it only has one thread.  It is really
2928  * simple: the Guest asks for a block number and we read or write that position
2929  * in the file.
2930  *
2931  * Before we serviced each virtqueue in a separate thread, that was unacceptably
2932  * slow: the Guest waits until the read is finished before running anything
2933  * else, even if it could have been doing useful work.
2934  *
2935  * We could have used async I/O, except it's reputed to suck so hard that
2936  * characters actually go missing from your code when you try to use it.
2937  */
2938 static void blk_request(struct virtqueue *vq)
2939 {
2940         struct vblk_info *vblk = vq->dev->priv;
2941         unsigned int head, out_num, in_num, wlen;
2942         int ret, i;
2943         u8 *in;
2944         struct virtio_blk_outhdr out;
2945         struct iovec iov[vq->vring.num];
2946         off64_t off;
2947 
2948         /*
2949          * Get the next request, where we normally wait.  It triggers the
2950          * interrupt to acknowledge previously serviced requests (if any).
2951          */
2952         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
2953 
2954         /* Copy the output header from the front of the iov (adjusts iov) */
2955         iov_consume(vq->dev, iov, out_num, &out, sizeof(out));
2956 
2957         /* Find and trim end of iov input array, for our status byte. */
2958         in = NULL;
2959         for (i = out_num + in_num - 1; i >= out_num; i--) {
2960                 if (iov[i].iov_len > 0) {
2961                         in = iov[i].iov_base + iov[i].iov_len - 1;
2962                         iov[i].iov_len--;
2963                         break;
2964                 }
2965         }
2966         if (!in)
2967                 bad_driver_vq(vq, "Bad virtblk cmd with no room for status");
2968 
2969         /*
2970          * For historical reasons, block operations are expressed in 512 byte
2971          * "sectors".
2972          */
2973         off = out.sector * 512;
2974 
2975         if (out.type & VIRTIO_BLK_T_OUT) {
2976                 /*
2977                  * Write
2978                  *
2979                  * Move to the right location in the block file.  This can fail
2980                  * if they try to write past end.
2981                  */
2982                 if (lseek64(vblk->fd, off, SEEK_SET) != off)
2983                         err(1, "Bad seek to sector %llu", out.sector);
2984 
2985                 ret = writev(vblk->fd, iov, out_num);
2986                 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
2987 
2988                 /*
2989                  * Grr... Now we know how long the descriptor they sent was, we
2990                  * make sure they didn't try to write over the end of the block
2991                  * file (possibly extending it).
2992                  */
2993                 if (ret > 0 && off + ret > vblk->len) {
2994                         /* Trim it back to the correct length */
2995                         ftruncate64(vblk->fd, vblk->len);
2996                         /* Die, bad Guest, die. */
2997                         bad_driver_vq(vq, "Write past end %llu+%u", off, ret);
2998                 }
2999 
3000                 wlen = sizeof(*in);
3001                 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
3002         } else if (out.type & VIRTIO_BLK_T_FLUSH) {
3003                 /* Flush */
3004                 ret = fdatasync(vblk->fd);
3005                 verbose("FLUSH fdatasync: %i\n", ret);
3006                 wlen = sizeof(*in);
3007                 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
3008         } else {
3009                 /*
3010                  * Read
3011                  *
3012                  * Move to the right location in the block file.  This can fail
3013                  * if they try to read past end.
3014                  */
3015                 if (lseek64(vblk->fd, off, SEEK_SET) != off)
3016                         err(1, "Bad seek to sector %llu", out.sector);
3017 
3018                 ret = readv(vblk->fd, iov + out_num, in_num);
3019                 if (ret >= 0) {
3020                         wlen = sizeof(*in) + ret;
3021                         *in = VIRTIO_BLK_S_OK;
3022                 } else {
3023                         wlen = sizeof(*in);
3024                         *in = VIRTIO_BLK_S_IOERR;
3025                 }
3026         }
3027 
3028         /* Finished that request. */
3029         add_used(vq, head, wlen);
3030 }
3031 
3032 /*L:198 This actually sets up a virtual block device. */
3033 static void setup_block_file(const char *filename)
3034 {
3035         struct device *dev;
3036         struct vblk_info *vblk;
3037         struct virtio_blk_config conf;
3038 
3039         /* Create the device. */
3040         dev = new_pci_device("block", VIRTIO_ID_BLOCK, 0x01, 0x80);
3041 
3042         /* The device has one virtqueue, where the Guest places requests. */
3043         add_pci_virtqueue(dev, blk_request, "request");
3044 
3045         /* Allocate the room for our own bookkeeping */
3046         vblk = dev->priv = malloc(sizeof(*vblk));
3047 
3048         /* First we open the file and store the length. */
3049         vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
3050         vblk->len = lseek64(vblk->fd, 0, SEEK_END);
3051 
3052         /* Tell Guest how many sectors this device has. */
3053         conf.capacity = cpu_to_le64(vblk->len / 512);
3054 
3055         /*
3056          * Tell Guest not to put in too many descriptors at once: two are used
3057          * for the in and out elements.
3058          */
3059         add_pci_feature(dev, VIRTIO_BLK_F_SEG_MAX);
3060         conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
3061 
3062         set_device_config(dev, &conf, sizeof(struct virtio_blk_config));
3063 
3064         verbose("device %u: virtblock %llu sectors\n",
3065                 devices.device_num, le64_to_cpu(conf.capacity));
3066 }
3067 
3068 /*L:211
3069  * Our random number generator device reads from /dev/urandom into the Guest's
3070  * input buffers.  The usual case is that the Guest doesn't want random numbers
3071  * and so has no buffers although /dev/urandom is still readable, whereas
3072  * console is the reverse.
3073  *
3074  * The same logic applies, however.
3075  */
3076 struct rng_info {
3077         int rfd;
3078 };
3079 
3080 static void rng_input(struct virtqueue *vq)
3081 {
3082         int len;
3083         unsigned int head, in_num, out_num, totlen = 0;
3084         struct rng_info *rng_info = vq->dev->priv;
3085         struct iovec iov[vq->vring.num];
3086 
3087         /* First we need a buffer from the Guests's virtqueue. */
3088         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
3089         if (out_num)
3090                 bad_driver_vq(vq, "Output buffers in rng?");
3091 
3092         /*
3093          * Just like the console write, we loop to cover the whole iovec.
3094          * In this case, short reads actually happen quite a bit.
3095          */
3096         while (!iov_empty(iov, in_num)) {
3097                 len = readv(rng_info->rfd, iov, in_num);
3098                 if (len <= 0)
3099                         err(1, "Read from /dev/urandom gave %i", len);
3100                 iov_consume(vq->dev, iov, in_num, NULL, len);
3101                 totlen += len;
3102         }
3103 
3104         /* Tell the Guest about the new input. */
3105         add_used(vq, head, totlen);
3106 }
3107 
3108 /*L:199
3109  * This creates a "hardware" random number device for the Guest.
3110  */
3111 static void setup_rng(void)
3112 {
3113         struct device *dev;
3114         struct rng_info *rng_info = malloc(sizeof(*rng_info));
3115 
3116         /* Our device's private info simply contains the /dev/urandom fd. */
3117         rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY);
3118 
3119         /* Create the new device. */
3120         dev = new_pci_device("rng", VIRTIO_ID_RNG, 0xff, 0);
3121         dev->priv = rng_info;
3122 
3123         /* The device has one virtqueue, where the Guest places inbufs. */
3124         add_pci_virtqueue(dev, rng_input, "input");
3125 
3126         /* We don't have any configuration space */
3127         no_device_config(dev);
3128 
3129         verbose("device %u: rng\n", devices.device_num);
3130 }
3131 /* That's the end of device setup. */
3132 
3133 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
3134 static void __attribute__((noreturn)) restart_guest(void)
3135 {
3136         unsigned int i;
3137 
3138         /*
3139          * Since we don't track all open fds, we simply close everything beyond
3140          * stderr.
3141          */
3142         for (i = 3; i < FD_SETSIZE; i++)
3143                 close(i);
3144 
3145         /* Reset all the devices (kills all threads). */
3146         cleanup_devices();
3147 
3148         execv(main_args[0], main_args);
3149         err(1, "Could not exec %s", main_args[0]);
3150 }
3151 
3152 /*L:220
3153  * Finally we reach the core of the Launcher which runs the Guest, serves
3154  * its input and output, and finally, lays it to rest.
3155  */
3156 static void __attribute__((noreturn)) run_guest(void)
3157 {
3158         for (;;) {
3159                 struct lguest_pending notify;
3160                 int readval;
3161 
3162                 /* We read from the /dev/lguest device to run the Guest. */
3163                 readval = pread(lguest_fd, &notify, sizeof(notify), cpu_id);
3164                 if (readval == sizeof(notify)) {
3165                         if (notify.trap == 13) {
3166                                 verbose("Emulating instruction at %#x\n",
3167                                         getreg(eip));
3168                                 emulate_insn(notify.insn);
3169                         } else if (notify.trap == 14) {
3170                                 verbose("Emulating MMIO at %#x\n",
3171                                         getreg(eip));
3172                                 emulate_mmio(notify.addr, notify.insn);
3173                         } else
3174                                 errx(1, "Unknown trap %i addr %#08x\n",
3175                                      notify.trap, notify.addr);
3176                 /* ENOENT means the Guest died.  Reading tells us why. */
3177                 } else if (errno == ENOENT) {
3178                         char reason[1024] = { 0 };
3179                         pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
3180                         errx(1, "%s", reason);
3181                 /* ERESTART means that we need to reboot the guest */
3182                 } else if (errno == ERESTART) {
3183                         restart_guest();
3184                 /* Anything else means a bug or incompatible change. */
3185                 } else
3186                         err(1, "Running guest failed");
3187         }
3188 }
3189 /*L:240
3190  * This is the end of the Launcher.  The good news: we are over halfway
3191  * through!  The bad news: the most fiendish part of the code still lies ahead
3192  * of us.
3193  *
3194  * Are you ready?  Take a deep breath and join me in the core of the Host, in
3195  * "make Host".
3196 :*/
3197 
3198 static struct option opts[] = {
3199         { "verbose", 0, NULL, 'v' },
3200         { "tunnet", 1, NULL, 't' },
3201         { "block", 1, NULL, 'b' },
3202         { "rng", 0, NULL, 'r' },
3203         { "initrd", 1, NULL, 'i' },
3204         { "username", 1, NULL, 'u' },
3205         { "chroot", 1, NULL, 'c' },
3206         { NULL },
3207 };
3208 static void usage(void)
3209 {
3210         errx(1, "Usage: lguest [--verbose] "
3211              "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
3212              "|--block=<filename>|--initrd=<filename>]...\n"
3213              "<mem-in-mb> vmlinux [args...]");
3214 }
3215 
3216 /*L:105 The main routine is where the real work begins: */
3217 int main(int argc, char *argv[])
3218 {
3219         /* Memory, code startpoint and size of the (optional) initrd. */
3220         unsigned long mem = 0, start, initrd_size = 0;
3221         /* Two temporaries. */
3222         int i, c;
3223         /* The boot information for the Guest. */
3224         struct boot_params *boot;
3225         /* If they specify an initrd file to load. */
3226         const char *initrd_name = NULL;
3227 
3228         /* Password structure for initgroups/setres[gu]id */
3229         struct passwd *user_details = NULL;
3230 
3231         /* Directory to chroot to */
3232         char *chroot_path = NULL;
3233 
3234         /* Save the args: we "reboot" by execing ourselves again. */
3235         main_args = argv;
3236 
3237         /*
3238          * First we initialize the device list.  We remember next interrupt
3239          * number to use for devices (1: remember that 0 is used by the timer).
3240          */
3241         devices.next_irq = 1;
3242 
3243         /* We're CPU 0.  In fact, that's the only CPU possible right now. */
3244         cpu_id = 0;
3245 
3246         /*
3247          * We need to know how much memory so we can set up the device
3248          * descriptor and memory pages for the devices as we parse the command
3249          * line.  So we quickly look through the arguments to find the amount
3250          * of memory now.
3251          */
3252         for (i = 1; i < argc; i++) {
3253                 if (argv[i][0] != '-') {
3254                         mem = atoi(argv[i]) * 1024 * 1024;
3255                         /*
3256                          * We start by mapping anonymous pages over all of
3257                          * guest-physical memory range.  This fills it with 0,
3258                          * and ensures that the Guest won't be killed when it
3259                          * tries to access it.
3260                          */
3261                         guest_base = map_zeroed_pages(mem / getpagesize()
3262                                                       + DEVICE_PAGES);
3263                         guest_limit = mem;
3264                         guest_max = guest_mmio = mem + DEVICE_PAGES*getpagesize();
3265                         break;
3266                 }
3267         }
3268 
3269         /* If we exit via err(), this kills all the threads, restores tty. */
3270         atexit(cleanup_devices);
3271 
3272         /* We always have a console device, and it's always device 1. */
3273         setup_console();
3274 
3275         /* The options are fairly straight-forward */
3276         while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
3277                 switch (c) {
3278                 case 'v':
3279                         verbose = true;
3280                         break;
3281                 case 't':
3282                         setup_tun_net(optarg);
3283                         break;
3284                 case 'b':
3285                         setup_block_file(optarg);
3286                         break;
3287                 case 'r':
3288                         setup_rng();
3289                         break;
3290                 case 'i':
3291                         initrd_name = optarg;
3292                         break;
3293                 case 'u':
3294                         user_details = getpwnam(optarg);
3295                         if (!user_details)
3296                                 err(1, "getpwnam failed, incorrect username?");
3297                         break;
3298                 case 'c':
3299                         chroot_path = optarg;
3300                         break;
3301                 default:
3302                         warnx("Unknown argument %s", argv[optind]);
3303                         usage();
3304                 }
3305         }
3306         /*
3307          * After the other arguments we expect memory and kernel image name,
3308          * followed by command line arguments for the kernel.
3309          */
3310         if (optind + 2 > argc)
3311                 usage();
3312 
3313         verbose("Guest base is at %p\n", guest_base);
3314 
3315         /* Initialize the (fake) PCI host bridge device. */
3316         init_pci_host_bridge();
3317 
3318         /* Now we load the kernel */
3319         start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
3320 
3321         /* Boot information is stashed at physical address 0 */
3322         boot = from_guest_phys(0);
3323 
3324         /* Map the initrd image if requested (at top of physical memory) */
3325         if (initrd_name) {
3326                 initrd_size = load_initrd(initrd_name, mem);
3327                 /*
3328                  * These are the location in the Linux boot header where the
3329                  * start and size of the initrd are expected to be found.
3330                  */
3331                 boot->hdr.ramdisk_image = mem - initrd_size;
3332                 boot->hdr.ramdisk_size = initrd_size;
3333                 /* The bootloader type 0xFF means "unknown"; that's OK. */
3334                 boot->hdr.type_of_loader = 0xFF;
3335         }
3336 
3337         /*
3338          * The Linux boot header contains an "E820" memory map: ours is a
3339          * simple, single region.
3340          */
3341         boot->e820_entries = 1;
3342         boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
3343         /*
3344          * The boot header contains a command line pointer: we put the command
3345          * line after the boot header.
3346          */
3347         boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
3348         /* We use a simple helper to copy the arguments separated by spaces. */
3349         concat((char *)(boot + 1), argv+optind+2);
3350 
3351         /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
3352         boot->hdr.kernel_alignment = 0x1000000;
3353 
3354         /* Boot protocol version: 2.07 supports the fields for lguest. */
3355         boot->hdr.version = 0x207;
3356 
3357         /* X86_SUBARCH_LGUEST tells the Guest it's an lguest. */
3358         boot->hdr.hardware_subarch = X86_SUBARCH_LGUEST;
3359 
3360         /* Tell the entry path not to try to reload segment registers. */
3361         boot->hdr.loadflags |= KEEP_SEGMENTS;
3362 
3363         /* We don't support tboot: */
3364         boot->tboot_addr = 0;
3365 
3366         /* Ensure this is 0 to prevent APM from loading: */
3367         boot->apm_bios_info.version = 0;
3368 
3369         /* We tell the kernel to initialize the Guest. */
3370         tell_kernel(start);
3371 
3372         /* Ensure that we terminate if a device-servicing child dies. */
3373         signal(SIGCHLD, kill_launcher);
3374 
3375         /* If requested, chroot to a directory */
3376         if (chroot_path) {
3377                 if (chroot(chroot_path) != 0)
3378                         err(1, "chroot(\"%s\") failed", chroot_path);
3379 
3380                 if (chdir("/") != 0)
3381                         err(1, "chdir(\"/\") failed");
3382 
3383                 verbose("chroot done\n");
3384         }
3385 
3386         /* If requested, drop privileges */
3387         if (user_details) {
3388                 uid_t u;
3389                 gid_t g;
3390 
3391                 u = user_details->pw_uid;
3392                 g = user_details->pw_gid;
3393 
3394                 if (initgroups(user_details->pw_name, g) != 0)
3395                         err(1, "initgroups failed");
3396 
3397                 if (setresgid(g, g, g) != 0)
3398                         err(1, "setresgid failed");
3399 
3400                 if (setresuid(u, u, u) != 0)
3401                         err(1, "setresuid failed");
3402 
3403                 verbose("Dropping privileges completed\n");
3404         }
3405 
3406         /* Finally, run the Guest.  This doesn't return. */
3407         run_guest();
3408 }
3409 /*:*/
3410 
3411 /*M:999
3412  * Mastery is done: you now know everything I do.
3413  *
3414  * But surely you have seen code, features and bugs in your wanderings which
3415  * you now yearn to attack?  That is the real game, and I look forward to you
3416  * patching and forking lguest into the Your-Name-Here-visor.
3417  *
3418  * Farewell, and good coding!
3419  * Rusty Russell.
3420  */
3421 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp