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

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

Version: ~ [ linux-5.6-rc7 ] ~ [ linux-5.5.11 ] ~ [ linux-5.4.27 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.112 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ 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-3.16.82 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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 
 45 #ifndef VIRTIO_F_ANY_LAYOUT
 46 #define VIRTIO_F_ANY_LAYOUT             27
 47 #endif
 48 
 49 /*L:110
 50  * We can ignore the 43 include files we need for this program, but I do want
 51  * to draw attention to the use of kernel-style types.
 52  *
 53  * As Linus said, "C is a Spartan language, and so should your naming be."  I
 54  * like these abbreviations, so we define them here.  Note that u64 is always
 55  * unsigned long long, which works on all Linux systems: this means that we can
 56  * use %llu in printf for any u64.
 57  */
 58 typedef unsigned long long u64;
 59 typedef uint32_t u32;
 60 typedef uint16_t u16;
 61 typedef uint8_t u8;
 62 /*:*/
 63 
 64 #include <linux/virtio_config.h>
 65 #include <linux/virtio_net.h>
 66 #include <linux/virtio_blk.h>
 67 #include <linux/virtio_console.h>
 68 #include <linux/virtio_rng.h>
 69 #include <linux/virtio_ring.h>
 70 #include <asm/bootparam.h>
 71 #include "../../include/linux/lguest_launcher.h"
 72 
 73 #define BRIDGE_PFX "bridge:"
 74 #ifndef SIOCBRADDIF
 75 #define SIOCBRADDIF     0x89a2          /* add interface to bridge      */
 76 #endif
 77 /* We can have up to 256 pages for devices. */
 78 #define DEVICE_PAGES 256
 79 /* This will occupy 3 pages: it must be a power of 2. */
 80 #define VIRTQUEUE_NUM 256
 81 
 82 /*L:120
 83  * verbose is both a global flag and a macro.  The C preprocessor allows
 84  * this, and although I wouldn't recommend it, it works quite nicely here.
 85  */
 86 static bool verbose;
 87 #define verbose(args...) \
 88         do { if (verbose) printf(args); } while(0)
 89 /*:*/
 90 
 91 /* The pointer to the start of guest memory. */
 92 static void *guest_base;
 93 /* The maximum guest physical address allowed, and maximum possible. */
 94 static unsigned long guest_limit, guest_max;
 95 /* The /dev/lguest file descriptor. */
 96 static int lguest_fd;
 97 
 98 /* a per-cpu variable indicating whose vcpu is currently running */
 99 static unsigned int __thread cpu_id;
100 
101 /* This is our list of devices. */
102 struct device_list {
103         /* Counter to assign interrupt numbers. */
104         unsigned int next_irq;
105 
106         /* Counter to print out convenient device numbers. */
107         unsigned int device_num;
108 
109         /* The descriptor page for the devices. */
110         u8 *descpage;
111 
112         /* A single linked list of devices. */
113         struct device *dev;
114         /* And a pointer to the last device for easy append. */
115         struct device *lastdev;
116 };
117 
118 /* The list of Guest devices, based on command line arguments. */
119 static struct device_list devices;
120 
121 /* The device structure describes a single device. */
122 struct device {
123         /* The linked-list pointer. */
124         struct device *next;
125 
126         /* The device's descriptor, as mapped into the Guest. */
127         struct lguest_device_desc *desc;
128 
129         /* We can't trust desc values once Guest has booted: we use these. */
130         unsigned int feature_len;
131         unsigned int num_vq;
132 
133         /* The name of this device, for --verbose. */
134         const char *name;
135 
136         /* Any queues attached to this device */
137         struct virtqueue *vq;
138 
139         /* Is it operational */
140         bool running;
141 
142         /* Device-specific data. */
143         void *priv;
144 };
145 
146 /* The virtqueue structure describes a queue attached to a device. */
147 struct virtqueue {
148         struct virtqueue *next;
149 
150         /* Which device owns me. */
151         struct device *dev;
152 
153         /* The configuration for this queue. */
154         struct lguest_vqconfig config;
155 
156         /* The actual ring of buffers. */
157         struct vring vring;
158 
159         /* Last available index we saw. */
160         u16 last_avail_idx;
161 
162         /* How many are used since we sent last irq? */
163         unsigned int pending_used;
164 
165         /* Eventfd where Guest notifications arrive. */
166         int eventfd;
167 
168         /* Function for the thread which is servicing this virtqueue. */
169         void (*service)(struct virtqueue *vq);
170         pid_t thread;
171 };
172 
173 /* Remember the arguments to the program so we can "reboot" */
174 static char **main_args;
175 
176 /* The original tty settings to restore on exit. */
177 static struct termios orig_term;
178 
179 /*
180  * We have to be careful with barriers: our devices are all run in separate
181  * threads and so we need to make sure that changes visible to the Guest happen
182  * in precise order.
183  */
184 #define wmb() __asm__ __volatile__("" : : : "memory")
185 #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
186 #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
187 
188 /* Wrapper for the last available index.  Makes it easier to change. */
189 #define lg_last_avail(vq)       ((vq)->last_avail_idx)
190 
191 /*
192  * The virtio configuration space is defined to be little-endian.  x86 is
193  * little-endian too, but it's nice to be explicit so we have these helpers.
194  */
195 #define cpu_to_le16(v16) (v16)
196 #define cpu_to_le32(v32) (v32)
197 #define cpu_to_le64(v64) (v64)
198 #define le16_to_cpu(v16) (v16)
199 #define le32_to_cpu(v32) (v32)
200 #define le64_to_cpu(v64) (v64)
201 
202 /* Is this iovec empty? */
203 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
204 {
205         unsigned int i;
206 
207         for (i = 0; i < num_iov; i++)
208                 if (iov[i].iov_len)
209                         return false;
210         return true;
211 }
212 
213 /* Take len bytes from the front of this iovec. */
214 static void iov_consume(struct iovec iov[], unsigned num_iov,
215                         void *dest, unsigned len)
216 {
217         unsigned int i;
218 
219         for (i = 0; i < num_iov; i++) {
220                 unsigned int used;
221 
222                 used = iov[i].iov_len < len ? iov[i].iov_len : len;
223                 if (dest) {
224                         memcpy(dest, iov[i].iov_base, used);
225                         dest += used;
226                 }
227                 iov[i].iov_base += used;
228                 iov[i].iov_len -= used;
229                 len -= used;
230         }
231         if (len != 0)
232                 errx(1, "iovec too short!");
233 }
234 
235 /* The device virtqueue descriptors are followed by feature bitmasks. */
236 static u8 *get_feature_bits(struct device *dev)
237 {
238         return (u8 *)(dev->desc + 1)
239                 + dev->num_vq * sizeof(struct lguest_vqconfig);
240 }
241 
242 /*L:100
243  * The Launcher code itself takes us out into userspace, that scary place where
244  * pointers run wild and free!  Unfortunately, like most userspace programs,
245  * it's quite boring (which is why everyone likes to hack on the kernel!).
246  * Perhaps if you make up an Lguest Drinking Game at this point, it will get
247  * you through this section.  Or, maybe not.
248  *
249  * The Launcher sets up a big chunk of memory to be the Guest's "physical"
250  * memory and stores it in "guest_base".  In other words, Guest physical ==
251  * Launcher virtual with an offset.
252  *
253  * This can be tough to get your head around, but usually it just means that we
254  * use these trivial conversion functions when the Guest gives us its
255  * "physical" addresses:
256  */
257 static void *from_guest_phys(unsigned long addr)
258 {
259         return guest_base + addr;
260 }
261 
262 static unsigned long to_guest_phys(const void *addr)
263 {
264         return (addr - guest_base);
265 }
266 
267 /*L:130
268  * Loading the Kernel.
269  *
270  * We start with couple of simple helper routines.  open_or_die() avoids
271  * error-checking code cluttering the callers:
272  */
273 static int open_or_die(const char *name, int flags)
274 {
275         int fd = open(name, flags);
276         if (fd < 0)
277                 err(1, "Failed to open %s", name);
278         return fd;
279 }
280 
281 /* map_zeroed_pages() takes a number of pages. */
282 static void *map_zeroed_pages(unsigned int num)
283 {
284         int fd = open_or_die("/dev/zero", O_RDONLY);
285         void *addr;
286 
287         /*
288          * We use a private mapping (ie. if we write to the page, it will be
289          * copied). We allocate an extra two pages PROT_NONE to act as guard
290          * pages against read/write attempts that exceed allocated space.
291          */
292         addr = mmap(NULL, getpagesize() * (num+2),
293                     PROT_NONE, MAP_PRIVATE, fd, 0);
294 
295         if (addr == MAP_FAILED)
296                 err(1, "Mmapping %u pages of /dev/zero", num);
297 
298         if (mprotect(addr + getpagesize(), getpagesize() * num,
299                      PROT_READ|PROT_WRITE) == -1)
300                 err(1, "mprotect rw %u pages failed", num);
301 
302         /*
303          * One neat mmap feature is that you can close the fd, and it
304          * stays mapped.
305          */
306         close(fd);
307 
308         /* Return address after PROT_NONE page */
309         return addr + getpagesize();
310 }
311 
312 /* Get some more pages for a device. */
313 static void *get_pages(unsigned int num)
314 {
315         void *addr = from_guest_phys(guest_limit);
316 
317         guest_limit += num * getpagesize();
318         if (guest_limit > guest_max)
319                 errx(1, "Not enough memory for devices");
320         return addr;
321 }
322 
323 /*
324  * This routine is used to load the kernel or initrd.  It tries mmap, but if
325  * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
326  * it falls back to reading the memory in.
327  */
328 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
329 {
330         ssize_t r;
331 
332         /*
333          * We map writable even though for some segments are marked read-only.
334          * The kernel really wants to be writable: it patches its own
335          * instructions.
336          *
337          * MAP_PRIVATE means that the page won't be copied until a write is
338          * done to it.  This allows us to share untouched memory between
339          * Guests.
340          */
341         if (mmap(addr, len, PROT_READ|PROT_WRITE,
342                  MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
343                 return;
344 
345         /* pread does a seek and a read in one shot: saves a few lines. */
346         r = pread(fd, addr, len, offset);
347         if (r != len)
348                 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
349 }
350 
351 /*
352  * This routine takes an open vmlinux image, which is in ELF, and maps it into
353  * the Guest memory.  ELF = Embedded Linking Format, which is the format used
354  * by all modern binaries on Linux including the kernel.
355  *
356  * The ELF headers give *two* addresses: a physical address, and a virtual
357  * address.  We use the physical address; the Guest will map itself to the
358  * virtual address.
359  *
360  * We return the starting address.
361  */
362 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
363 {
364         Elf32_Phdr phdr[ehdr->e_phnum];
365         unsigned int i;
366 
367         /*
368          * Sanity checks on the main ELF header: an x86 executable with a
369          * reasonable number of correctly-sized program headers.
370          */
371         if (ehdr->e_type != ET_EXEC
372             || ehdr->e_machine != EM_386
373             || ehdr->e_phentsize != sizeof(Elf32_Phdr)
374             || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
375                 errx(1, "Malformed elf header");
376 
377         /*
378          * An ELF executable contains an ELF header and a number of "program"
379          * headers which indicate which parts ("segments") of the program to
380          * load where.
381          */
382 
383         /* We read in all the program headers at once: */
384         if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
385                 err(1, "Seeking to program headers");
386         if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
387                 err(1, "Reading program headers");
388 
389         /*
390          * Try all the headers: there are usually only three.  A read-only one,
391          * a read-write one, and a "note" section which we don't load.
392          */
393         for (i = 0; i < ehdr->e_phnum; i++) {
394                 /* If this isn't a loadable segment, we ignore it */
395                 if (phdr[i].p_type != PT_LOAD)
396                         continue;
397 
398                 verbose("Section %i: size %i addr %p\n",
399                         i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
400 
401                 /* We map this section of the file at its physical address. */
402                 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
403                        phdr[i].p_offset, phdr[i].p_filesz);
404         }
405 
406         /* The entry point is given in the ELF header. */
407         return ehdr->e_entry;
408 }
409 
410 /*L:150
411  * A bzImage, unlike an ELF file, is not meant to be loaded.  You're supposed
412  * to jump into it and it will unpack itself.  We used to have to perform some
413  * hairy magic because the unpacking code scared me.
414  *
415  * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
416  * a small patch to jump over the tricky bits in the Guest, so now we just read
417  * the funky header so we know where in the file to load, and away we go!
418  */
419 static unsigned long load_bzimage(int fd)
420 {
421         struct boot_params boot;
422         int r;
423         /* Modern bzImages get loaded at 1M. */
424         void *p = from_guest_phys(0x100000);
425 
426         /*
427          * Go back to the start of the file and read the header.  It should be
428          * a Linux boot header (see Documentation/x86/boot.txt)
429          */
430         lseek(fd, 0, SEEK_SET);
431         read(fd, &boot, sizeof(boot));
432 
433         /* Inside the setup_hdr, we expect the magic "HdrS" */
434         if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
435                 errx(1, "This doesn't look like a bzImage to me");
436 
437         /* Skip over the extra sectors of the header. */
438         lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
439 
440         /* Now read everything into memory. in nice big chunks. */
441         while ((r = read(fd, p, 65536)) > 0)
442                 p += r;
443 
444         /* Finally, code32_start tells us where to enter the kernel. */
445         return boot.hdr.code32_start;
446 }
447 
448 /*L:140
449  * Loading the kernel is easy when it's a "vmlinux", but most kernels
450  * come wrapped up in the self-decompressing "bzImage" format.  With a little
451  * work, we can load those, too.
452  */
453 static unsigned long load_kernel(int fd)
454 {
455         Elf32_Ehdr hdr;
456 
457         /* Read in the first few bytes. */
458         if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
459                 err(1, "Reading kernel");
460 
461         /* If it's an ELF file, it starts with "\177ELF" */
462         if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
463                 return map_elf(fd, &hdr);
464 
465         /* Otherwise we assume it's a bzImage, and try to load it. */
466         return load_bzimage(fd);
467 }
468 
469 /*
470  * This is a trivial little helper to align pages.  Andi Kleen hated it because
471  * it calls getpagesize() twice: "it's dumb code."
472  *
473  * Kernel guys get really het up about optimization, even when it's not
474  * necessary.  I leave this code as a reaction against that.
475  */
476 static inline unsigned long page_align(unsigned long addr)
477 {
478         /* Add upwards and truncate downwards. */
479         return ((addr + getpagesize()-1) & ~(getpagesize()-1));
480 }
481 
482 /*L:180
483  * An "initial ram disk" is a disk image loaded into memory along with the
484  * kernel which the kernel can use to boot from without needing any drivers.
485  * Most distributions now use this as standard: the initrd contains the code to
486  * load the appropriate driver modules for the current machine.
487  *
488  * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
489  * kernels.  He sent me this (and tells me when I break it).
490  */
491 static unsigned long load_initrd(const char *name, unsigned long mem)
492 {
493         int ifd;
494         struct stat st;
495         unsigned long len;
496 
497         ifd = open_or_die(name, O_RDONLY);
498         /* fstat() is needed to get the file size. */
499         if (fstat(ifd, &st) < 0)
500                 err(1, "fstat() on initrd '%s'", name);
501 
502         /*
503          * We map the initrd at the top of memory, but mmap wants it to be
504          * page-aligned, so we round the size up for that.
505          */
506         len = page_align(st.st_size);
507         map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
508         /*
509          * Once a file is mapped, you can close the file descriptor.  It's a
510          * little odd, but quite useful.
511          */
512         close(ifd);
513         verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
514 
515         /* We return the initrd size. */
516         return len;
517 }
518 /*:*/
519 
520 /*
521  * Simple routine to roll all the commandline arguments together with spaces
522  * between them.
523  */
524 static void concat(char *dst, char *args[])
525 {
526         unsigned int i, len = 0;
527 
528         for (i = 0; args[i]; i++) {
529                 if (i) {
530                         strcat(dst+len, " ");
531                         len++;
532                 }
533                 strcpy(dst+len, args[i]);
534                 len += strlen(args[i]);
535         }
536         /* In case it's empty. */
537         dst[len] = '\0';
538 }
539 
540 /*L:185
541  * This is where we actually tell the kernel to initialize the Guest.  We
542  * saw the arguments it expects when we looked at initialize() in lguest_user.c:
543  * the base of Guest "physical" memory, the top physical page to allow and the
544  * entry point for the Guest.
545  */
546 static void tell_kernel(unsigned long start)
547 {
548         unsigned long args[] = { LHREQ_INITIALIZE,
549                                  (unsigned long)guest_base,
550                                  guest_limit / getpagesize(), start };
551         verbose("Guest: %p - %p (%#lx)\n",
552                 guest_base, guest_base + guest_limit, guest_limit);
553         lguest_fd = open_or_die("/dev/lguest", O_RDWR);
554         if (write(lguest_fd, args, sizeof(args)) < 0)
555                 err(1, "Writing to /dev/lguest");
556 }
557 /*:*/
558 
559 /*L:200
560  * Device Handling.
561  *
562  * When the Guest gives us a buffer, it sends an array of addresses and sizes.
563  * We need to make sure it's not trying to reach into the Launcher itself, so
564  * we have a convenient routine which checks it and exits with an error message
565  * if something funny is going on:
566  */
567 static void *_check_pointer(unsigned long addr, unsigned int size,
568                             unsigned int line)
569 {
570         /*
571          * Check if the requested address and size exceeds the allocated memory,
572          * or addr + size wraps around.
573          */
574         if ((addr + size) > guest_limit || (addr + size) < addr)
575                 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
576         /*
577          * We return a pointer for the caller's convenience, now we know it's
578          * safe to use.
579          */
580         return from_guest_phys(addr);
581 }
582 /* A macro which transparently hands the line number to the real function. */
583 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
584 
585 /*
586  * Each buffer in the virtqueues is actually a chain of descriptors.  This
587  * function returns the next descriptor in the chain, or vq->vring.num if we're
588  * at the end.
589  */
590 static unsigned next_desc(struct vring_desc *desc,
591                           unsigned int i, unsigned int max)
592 {
593         unsigned int next;
594 
595         /* If this descriptor says it doesn't chain, we're done. */
596         if (!(desc[i].flags & VRING_DESC_F_NEXT))
597                 return max;
598 
599         /* Check they're not leading us off end of descriptors. */
600         next = desc[i].next;
601         /* Make sure compiler knows to grab that: we don't want it changing! */
602         wmb();
603 
604         if (next >= max)
605                 errx(1, "Desc next is %u", next);
606 
607         return next;
608 }
609 
610 /*
611  * This actually sends the interrupt for this virtqueue, if we've used a
612  * buffer.
613  */
614 static void trigger_irq(struct virtqueue *vq)
615 {
616         unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
617 
618         /* Don't inform them if nothing used. */
619         if (!vq->pending_used)
620                 return;
621         vq->pending_used = 0;
622 
623         /* If they don't want an interrupt, don't send one... */
624         if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
625                 return;
626         }
627 
628         /* Send the Guest an interrupt tell them we used something up. */
629         if (write(lguest_fd, buf, sizeof(buf)) != 0)
630                 err(1, "Triggering irq %i", vq->config.irq);
631 }
632 
633 /*
634  * This looks in the virtqueue for the first available buffer, and converts
635  * it to an iovec for convenient access.  Since descriptors consist of some
636  * number of output then some number of input descriptors, it's actually two
637  * iovecs, but we pack them into one and note how many of each there were.
638  *
639  * This function waits if necessary, and returns the descriptor number found.
640  */
641 static unsigned wait_for_vq_desc(struct virtqueue *vq,
642                                  struct iovec iov[],
643                                  unsigned int *out_num, unsigned int *in_num)
644 {
645         unsigned int i, head, max;
646         struct vring_desc *desc;
647         u16 last_avail = lg_last_avail(vq);
648 
649         /* There's nothing available? */
650         while (last_avail == vq->vring.avail->idx) {
651                 u64 event;
652 
653                 /*
654                  * Since we're about to sleep, now is a good time to tell the
655                  * Guest about what we've used up to now.
656                  */
657                 trigger_irq(vq);
658 
659                 /* OK, now we need to know about added descriptors. */
660                 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
661 
662                 /*
663                  * They could have slipped one in as we were doing that: make
664                  * sure it's written, then check again.
665                  */
666                 mb();
667                 if (last_avail != vq->vring.avail->idx) {
668                         vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
669                         break;
670                 }
671 
672                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
673                 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
674                         errx(1, "Event read failed?");
675 
676                 /* We don't need to be notified again. */
677                 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
678         }
679 
680         /* Check it isn't doing very strange things with descriptor numbers. */
681         if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
682                 errx(1, "Guest moved used index from %u to %u",
683                      last_avail, vq->vring.avail->idx);
684 
685         /* 
686          * Make sure we read the descriptor number *after* we read the ring
687          * update; don't let the cpu or compiler change the order.
688          */
689         rmb();
690 
691         /*
692          * Grab the next descriptor number they're advertising, and increment
693          * the index we've seen.
694          */
695         head = vq->vring.avail->ring[last_avail % vq->vring.num];
696         lg_last_avail(vq)++;
697 
698         /* If their number is silly, that's a fatal mistake. */
699         if (head >= vq->vring.num)
700                 errx(1, "Guest says index %u is available", head);
701 
702         /* When we start there are none of either input nor output. */
703         *out_num = *in_num = 0;
704 
705         max = vq->vring.num;
706         desc = vq->vring.desc;
707         i = head;
708 
709         /*
710          * We have to read the descriptor after we read the descriptor number,
711          * but there's a data dependency there so the CPU shouldn't reorder
712          * that: no rmb() required.
713          */
714 
715         /*
716          * If this is an indirect entry, then this buffer contains a descriptor
717          * table which we handle as if it's any normal descriptor chain.
718          */
719         if (desc[i].flags & VRING_DESC_F_INDIRECT) {
720                 if (desc[i].len % sizeof(struct vring_desc))
721                         errx(1, "Invalid size for indirect buffer table");
722 
723                 max = desc[i].len / sizeof(struct vring_desc);
724                 desc = check_pointer(desc[i].addr, desc[i].len);
725                 i = 0;
726         }
727 
728         do {
729                 /* Grab the first descriptor, and check it's OK. */
730                 iov[*out_num + *in_num].iov_len = desc[i].len;
731                 iov[*out_num + *in_num].iov_base
732                         = check_pointer(desc[i].addr, desc[i].len);
733                 /* If this is an input descriptor, increment that count. */
734                 if (desc[i].flags & VRING_DESC_F_WRITE)
735                         (*in_num)++;
736                 else {
737                         /*
738                          * If it's an output descriptor, they're all supposed
739                          * to come before any input descriptors.
740                          */
741                         if (*in_num)
742                                 errx(1, "Descriptor has out after in");
743                         (*out_num)++;
744                 }
745 
746                 /* If we've got too many, that implies a descriptor loop. */
747                 if (*out_num + *in_num > max)
748                         errx(1, "Looped descriptor");
749         } while ((i = next_desc(desc, i, max)) != max);
750 
751         return head;
752 }
753 
754 /*
755  * After we've used one of their buffers, we tell the Guest about it.  Sometime
756  * later we'll want to send them an interrupt using trigger_irq(); note that
757  * wait_for_vq_desc() does that for us if it has to wait.
758  */
759 static void add_used(struct virtqueue *vq, unsigned int head, int len)
760 {
761         struct vring_used_elem *used;
762 
763         /*
764          * The virtqueue contains a ring of used buffers.  Get a pointer to the
765          * next entry in that used ring.
766          */
767         used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
768         used->id = head;
769         used->len = len;
770         /* Make sure buffer is written before we update index. */
771         wmb();
772         vq->vring.used->idx++;
773         vq->pending_used++;
774 }
775 
776 /* And here's the combo meal deal.  Supersize me! */
777 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
778 {
779         add_used(vq, head, len);
780         trigger_irq(vq);
781 }
782 
783 /*
784  * The Console
785  *
786  * We associate some data with the console for our exit hack.
787  */
788 struct console_abort {
789         /* How many times have they hit ^C? */
790         int count;
791         /* When did they start? */
792         struct timeval start;
793 };
794 
795 /* This is the routine which handles console input (ie. stdin). */
796 static void console_input(struct virtqueue *vq)
797 {
798         int len;
799         unsigned int head, in_num, out_num;
800         struct console_abort *abort = vq->dev->priv;
801         struct iovec iov[vq->vring.num];
802 
803         /* Make sure there's a descriptor available. */
804         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
805         if (out_num)
806                 errx(1, "Output buffers in console in queue?");
807 
808         /* Read into it.  This is where we usually wait. */
809         len = readv(STDIN_FILENO, iov, in_num);
810         if (len <= 0) {
811                 /* Ran out of input? */
812                 warnx("Failed to get console input, ignoring console.");
813                 /*
814                  * For simplicity, dying threads kill the whole Launcher.  So
815                  * just nap here.
816                  */
817                 for (;;)
818                         pause();
819         }
820 
821         /* Tell the Guest we used a buffer. */
822         add_used_and_trigger(vq, head, len);
823 
824         /*
825          * Three ^C within one second?  Exit.
826          *
827          * This is such a hack, but works surprisingly well.  Each ^C has to
828          * be in a buffer by itself, so they can't be too fast.  But we check
829          * that we get three within about a second, so they can't be too
830          * slow.
831          */
832         if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
833                 abort->count = 0;
834                 return;
835         }
836 
837         abort->count++;
838         if (abort->count == 1)
839                 gettimeofday(&abort->start, NULL);
840         else if (abort->count == 3) {
841                 struct timeval now;
842                 gettimeofday(&now, NULL);
843                 /* Kill all Launcher processes with SIGINT, like normal ^C */
844                 if (now.tv_sec <= abort->start.tv_sec+1)
845                         kill(0, SIGINT);
846                 abort->count = 0;
847         }
848 }
849 
850 /* This is the routine which handles console output (ie. stdout). */
851 static void console_output(struct virtqueue *vq)
852 {
853         unsigned int head, out, in;
854         struct iovec iov[vq->vring.num];
855 
856         /* We usually wait in here, for the Guest to give us something. */
857         head = wait_for_vq_desc(vq, iov, &out, &in);
858         if (in)
859                 errx(1, "Input buffers in console output queue?");
860 
861         /* writev can return a partial write, so we loop here. */
862         while (!iov_empty(iov, out)) {
863                 int len = writev(STDOUT_FILENO, iov, out);
864                 if (len <= 0) {
865                         warn("Write to stdout gave %i (%d)", len, errno);
866                         break;
867                 }
868                 iov_consume(iov, out, NULL, len);
869         }
870 
871         /*
872          * We're finished with that buffer: if we're going to sleep,
873          * wait_for_vq_desc() will prod the Guest with an interrupt.
874          */
875         add_used(vq, head, 0);
876 }
877 
878 /*
879  * The Network
880  *
881  * Handling output for network is also simple: we get all the output buffers
882  * and write them to /dev/net/tun.
883  */
884 struct net_info {
885         int tunfd;
886 };
887 
888 static void net_output(struct virtqueue *vq)
889 {
890         struct net_info *net_info = vq->dev->priv;
891         unsigned int head, out, in;
892         struct iovec iov[vq->vring.num];
893 
894         /* We usually wait in here for the Guest to give us a packet. */
895         head = wait_for_vq_desc(vq, iov, &out, &in);
896         if (in)
897                 errx(1, "Input buffers in net output queue?");
898         /*
899          * Send the whole thing through to /dev/net/tun.  It expects the exact
900          * same format: what a coincidence!
901          */
902         if (writev(net_info->tunfd, iov, out) < 0)
903                 warnx("Write to tun failed (%d)?", errno);
904 
905         /*
906          * Done with that one; wait_for_vq_desc() will send the interrupt if
907          * all packets are processed.
908          */
909         add_used(vq, head, 0);
910 }
911 
912 /*
913  * Handling network input is a bit trickier, because I've tried to optimize it.
914  *
915  * First we have a helper routine which tells is if from this file descriptor
916  * (ie. the /dev/net/tun device) will block:
917  */
918 static bool will_block(int fd)
919 {
920         fd_set fdset;
921         struct timeval zero = { 0, 0 };
922         FD_ZERO(&fdset);
923         FD_SET(fd, &fdset);
924         return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
925 }
926 
927 /*
928  * This handles packets coming in from the tun device to our Guest.  Like all
929  * service routines, it gets called again as soon as it returns, so you don't
930  * see a while(1) loop here.
931  */
932 static void net_input(struct virtqueue *vq)
933 {
934         int len;
935         unsigned int head, out, in;
936         struct iovec iov[vq->vring.num];
937         struct net_info *net_info = vq->dev->priv;
938 
939         /*
940          * Get a descriptor to write an incoming packet into.  This will also
941          * send an interrupt if they're out of descriptors.
942          */
943         head = wait_for_vq_desc(vq, iov, &out, &in);
944         if (out)
945                 errx(1, "Output buffers in net input queue?");
946 
947         /*
948          * If it looks like we'll block reading from the tun device, send them
949          * an interrupt.
950          */
951         if (vq->pending_used && will_block(net_info->tunfd))
952                 trigger_irq(vq);
953 
954         /*
955          * Read in the packet.  This is where we normally wait (when there's no
956          * incoming network traffic).
957          */
958         len = readv(net_info->tunfd, iov, in);
959         if (len <= 0)
960                 warn("Failed to read from tun (%d).", errno);
961 
962         /*
963          * Mark that packet buffer as used, but don't interrupt here.  We want
964          * to wait until we've done as much work as we can.
965          */
966         add_used(vq, head, len);
967 }
968 /*:*/
969 
970 /* This is the helper to create threads: run the service routine in a loop. */
971 static int do_thread(void *_vq)
972 {
973         struct virtqueue *vq = _vq;
974 
975         for (;;)
976                 vq->service(vq);
977         return 0;
978 }
979 
980 /*
981  * When a child dies, we kill our entire process group with SIGTERM.  This
982  * also has the side effect that the shell restores the console for us!
983  */
984 static void kill_launcher(int signal)
985 {
986         kill(0, SIGTERM);
987 }
988 
989 static void reset_device(struct device *dev)
990 {
991         struct virtqueue *vq;
992 
993         verbose("Resetting device %s\n", dev->name);
994 
995         /* Clear any features they've acked. */
996         memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
997 
998         /* We're going to be explicitly killing threads, so ignore them. */
999         signal(SIGCHLD, SIG_IGN);
1000 
1001         /* Zero out the virtqueues, get rid of their threads */
1002         for (vq = dev->vq; vq; vq = vq->next) {
1003                 if (vq->thread != (pid_t)-1) {
1004                         kill(vq->thread, SIGTERM);
1005                         waitpid(vq->thread, NULL, 0);
1006                         vq->thread = (pid_t)-1;
1007                 }
1008                 memset(vq->vring.desc, 0,
1009                        vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1010                 lg_last_avail(vq) = 0;
1011         }
1012         dev->running = false;
1013 
1014         /* Now we care if threads die. */
1015         signal(SIGCHLD, (void *)kill_launcher);
1016 }
1017 
1018 /*L:216
1019  * This actually creates the thread which services the virtqueue for a device.
1020  */
1021 static void create_thread(struct virtqueue *vq)
1022 {
1023         /*
1024          * Create stack for thread.  Since the stack grows upwards, we point
1025          * the stack pointer to the end of this region.
1026          */
1027         char *stack = malloc(32768);
1028         unsigned long args[] = { LHREQ_EVENTFD,
1029                                  vq->config.pfn*getpagesize(), 0 };
1030 
1031         /* Create a zero-initialized eventfd. */
1032         vq->eventfd = eventfd(0, 0);
1033         if (vq->eventfd < 0)
1034                 err(1, "Creating eventfd");
1035         args[2] = vq->eventfd;
1036 
1037         /*
1038          * Attach an eventfd to this virtqueue: it will go off when the Guest
1039          * does an LHCALL_NOTIFY for this vq.
1040          */
1041         if (write(lguest_fd, &args, sizeof(args)) != 0)
1042                 err(1, "Attaching eventfd");
1043 
1044         /*
1045          * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1046          * we get a signal if it dies.
1047          */
1048         vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1049         if (vq->thread == (pid_t)-1)
1050                 err(1, "Creating clone");
1051 
1052         /* We close our local copy now the child has it. */
1053         close(vq->eventfd);
1054 }
1055 
1056 static void start_device(struct device *dev)
1057 {
1058         unsigned int i;
1059         struct virtqueue *vq;
1060 
1061         verbose("Device %s OK: offered", dev->name);
1062         for (i = 0; i < dev->feature_len; i++)
1063                 verbose(" %02x", get_feature_bits(dev)[i]);
1064         verbose(", accepted");
1065         for (i = 0; i < dev->feature_len; i++)
1066                 verbose(" %02x", get_feature_bits(dev)
1067                         [dev->feature_len+i]);
1068 
1069         for (vq = dev->vq; vq; vq = vq->next) {
1070                 if (vq->service)
1071                         create_thread(vq);
1072         }
1073         dev->running = true;
1074 }
1075 
1076 static void cleanup_devices(void)
1077 {
1078         struct device *dev;
1079 
1080         for (dev = devices.dev; dev; dev = dev->next)
1081                 reset_device(dev);
1082 
1083         /* If we saved off the original terminal settings, restore them now. */
1084         if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1085                 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1086 }
1087 
1088 /* When the Guest tells us they updated the status field, we handle it. */
1089 static void update_device_status(struct device *dev)
1090 {
1091         /* A zero status is a reset, otherwise it's a set of flags. */
1092         if (dev->desc->status == 0)
1093                 reset_device(dev);
1094         else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1095                 warnx("Device %s configuration FAILED", dev->name);
1096                 if (dev->running)
1097                         reset_device(dev);
1098         } else {
1099                 if (dev->running)
1100                         err(1, "Device %s features finalized twice", dev->name);
1101                 start_device(dev);
1102         }
1103 }
1104 
1105 /*L:215
1106  * This is the generic routine we call when the Guest uses LHCALL_NOTIFY.  In
1107  * particular, it's used to notify us of device status changes during boot.
1108  */
1109 static void handle_output(unsigned long addr)
1110 {
1111         struct device *i;
1112 
1113         /* Check each device. */
1114         for (i = devices.dev; i; i = i->next) {
1115                 struct virtqueue *vq;
1116 
1117                 /*
1118                  * Notifications to device descriptors mean they updated the
1119                  * device status.
1120                  */
1121                 if (from_guest_phys(addr) == i->desc) {
1122                         update_device_status(i);
1123                         return;
1124                 }
1125 
1126                 /* Devices should not be used before features are finalized. */
1127                 for (vq = i->vq; vq; vq = vq->next) {
1128                         if (addr != vq->config.pfn*getpagesize())
1129                                 continue;
1130                         errx(1, "Notification on %s before setup!", i->name);
1131                 }
1132         }
1133 
1134         /*
1135          * Early console write is done using notify on a nul-terminated string
1136          * in Guest memory.  It's also great for hacking debugging messages
1137          * into a Guest.
1138          */
1139         if (addr >= guest_limit)
1140                 errx(1, "Bad NOTIFY %#lx", addr);
1141 
1142         write(STDOUT_FILENO, from_guest_phys(addr),
1143               strnlen(from_guest_phys(addr), guest_limit - addr));
1144 }
1145 
1146 /*L:190
1147  * Device Setup
1148  *
1149  * All devices need a descriptor so the Guest knows it exists, and a "struct
1150  * device" so the Launcher can keep track of it.  We have common helper
1151  * routines to allocate and manage them.
1152  */
1153 
1154 /*
1155  * The layout of the device page is a "struct lguest_device_desc" followed by a
1156  * number of virtqueue descriptors, then two sets of feature bits, then an
1157  * array of configuration bytes.  This routine returns the configuration
1158  * pointer.
1159  */
1160 static u8 *device_config(const struct device *dev)
1161 {
1162         return (void *)(dev->desc + 1)
1163                 + dev->num_vq * sizeof(struct lguest_vqconfig)
1164                 + dev->feature_len * 2;
1165 }
1166 
1167 /*
1168  * This routine allocates a new "struct lguest_device_desc" from descriptor
1169  * table page just above the Guest's normal memory.  It returns a pointer to
1170  * that descriptor.
1171  */
1172 static struct lguest_device_desc *new_dev_desc(u16 type)
1173 {
1174         struct lguest_device_desc d = { .type = type };
1175         void *p;
1176 
1177         /* Figure out where the next device config is, based on the last one. */
1178         if (devices.lastdev)
1179                 p = device_config(devices.lastdev)
1180                         + devices.lastdev->desc->config_len;
1181         else
1182                 p = devices.descpage;
1183 
1184         /* We only have one page for all the descriptors. */
1185         if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1186                 errx(1, "Too many devices");
1187 
1188         /* p might not be aligned, so we memcpy in. */
1189         return memcpy(p, &d, sizeof(d));
1190 }
1191 
1192 /*
1193  * Each device descriptor is followed by the description of its virtqueues.  We
1194  * specify how many descriptors the virtqueue is to have.
1195  */
1196 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1197                           void (*service)(struct virtqueue *))
1198 {
1199         unsigned int pages;
1200         struct virtqueue **i, *vq = malloc(sizeof(*vq));
1201         void *p;
1202 
1203         /* First we need some memory for this virtqueue. */
1204         pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1205                 / getpagesize();
1206         p = get_pages(pages);
1207 
1208         /* Initialize the virtqueue */
1209         vq->next = NULL;
1210         vq->last_avail_idx = 0;
1211         vq->dev = dev;
1212 
1213         /*
1214          * This is the routine the service thread will run, and its Process ID
1215          * once it's running.
1216          */
1217         vq->service = service;
1218         vq->thread = (pid_t)-1;
1219 
1220         /* Initialize the configuration. */
1221         vq->config.num = num_descs;
1222         vq->config.irq = devices.next_irq++;
1223         vq->config.pfn = to_guest_phys(p) / getpagesize();
1224 
1225         /* Initialize the vring. */
1226         vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1227 
1228         /*
1229          * Append virtqueue to this device's descriptor.  We use
1230          * device_config() to get the end of the device's current virtqueues;
1231          * we check that we haven't added any config or feature information
1232          * yet, otherwise we'd be overwriting them.
1233          */
1234         assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1235         memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1236         dev->num_vq++;
1237         dev->desc->num_vq++;
1238 
1239         verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1240 
1241         /*
1242          * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1243          * second.
1244          */
1245         for (i = &dev->vq; *i; i = &(*i)->next);
1246         *i = vq;
1247 }
1248 
1249 /*
1250  * The first half of the feature bitmask is for us to advertise features.  The
1251  * second half is for the Guest to accept features.
1252  */
1253 static void add_feature(struct device *dev, unsigned bit)
1254 {
1255         u8 *features = get_feature_bits(dev);
1256 
1257         /* We can't extend the feature bits once we've added config bytes */
1258         if (dev->desc->feature_len <= bit / CHAR_BIT) {
1259                 assert(dev->desc->config_len == 0);
1260                 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1261         }
1262 
1263         features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1264 }
1265 
1266 /*
1267  * This routine sets the configuration fields for an existing device's
1268  * descriptor.  It only works for the last device, but that's OK because that's
1269  * how we use it.
1270  */
1271 static void set_config(struct device *dev, unsigned len, const void *conf)
1272 {
1273         /* Check we haven't overflowed our single page. */
1274         if (device_config(dev) + len > devices.descpage + getpagesize())
1275                 errx(1, "Too many devices");
1276 
1277         /* Copy in the config information, and store the length. */
1278         memcpy(device_config(dev), conf, len);
1279         dev->desc->config_len = len;
1280 
1281         /* Size must fit in config_len field (8 bits)! */
1282         assert(dev->desc->config_len == len);
1283 }
1284 
1285 /*
1286  * This routine does all the creation and setup of a new device, including
1287  * calling new_dev_desc() to allocate the descriptor and device memory.  We
1288  * don't actually start the service threads until later.
1289  *
1290  * See what I mean about userspace being boring?
1291  */
1292 static struct device *new_device(const char *name, u16 type)
1293 {
1294         struct device *dev = malloc(sizeof(*dev));
1295 
1296         /* Now we populate the fields one at a time. */
1297         dev->desc = new_dev_desc(type);
1298         dev->name = name;
1299         dev->vq = NULL;
1300         dev->feature_len = 0;
1301         dev->num_vq = 0;
1302         dev->running = false;
1303         dev->next = NULL;
1304 
1305         /*
1306          * Append to device list.  Prepending to a single-linked list is
1307          * easier, but the user expects the devices to be arranged on the bus
1308          * in command-line order.  The first network device on the command line
1309          * is eth0, the first block device /dev/vda, etc.
1310          */
1311         if (devices.lastdev)
1312                 devices.lastdev->next = dev;
1313         else
1314                 devices.dev = dev;
1315         devices.lastdev = dev;
1316 
1317         return dev;
1318 }
1319 
1320 /*
1321  * Our first setup routine is the console.  It's a fairly simple device, but
1322  * UNIX tty handling makes it uglier than it could be.
1323  */
1324 static void setup_console(void)
1325 {
1326         struct device *dev;
1327 
1328         /* If we can save the initial standard input settings... */
1329         if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1330                 struct termios term = orig_term;
1331                 /*
1332                  * Then we turn off echo, line buffering and ^C etc: We want a
1333                  * raw input stream to the Guest.
1334                  */
1335                 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1336                 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1337         }
1338 
1339         dev = new_device("console", VIRTIO_ID_CONSOLE);
1340 
1341         /* We store the console state in dev->priv, and initialize it. */
1342         dev->priv = malloc(sizeof(struct console_abort));
1343         ((struct console_abort *)dev->priv)->count = 0;
1344 
1345         /*
1346          * The console needs two virtqueues: the input then the output.  When
1347          * they put something the input queue, we make sure we're listening to
1348          * stdin.  When they put something in the output queue, we write it to
1349          * stdout.
1350          */
1351         add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1352         add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1353 
1354         verbose("device %u: console\n", ++devices.device_num);
1355 }
1356 /*:*/
1357 
1358 /*M:010
1359  * Inter-guest networking is an interesting area.  Simplest is to have a
1360  * --sharenet=<name> option which opens or creates a named pipe.  This can be
1361  * used to send packets to another guest in a 1:1 manner.
1362  *
1363  * More sophisticated is to use one of the tools developed for project like UML
1364  * to do networking.
1365  *
1366  * Faster is to do virtio bonding in kernel.  Doing this 1:1 would be
1367  * completely generic ("here's my vring, attach to your vring") and would work
1368  * for any traffic.  Of course, namespace and permissions issues need to be
1369  * dealt with.  A more sophisticated "multi-channel" virtio_net.c could hide
1370  * multiple inter-guest channels behind one interface, although it would
1371  * require some manner of hotplugging new virtio channels.
1372  *
1373  * Finally, we could use a virtio network switch in the kernel, ie. vhost.
1374 :*/
1375 
1376 static u32 str2ip(const char *ipaddr)
1377 {
1378         unsigned int b[4];
1379 
1380         if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1381                 errx(1, "Failed to parse IP address '%s'", ipaddr);
1382         return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1383 }
1384 
1385 static void str2mac(const char *macaddr, unsigned char mac[6])
1386 {
1387         unsigned int m[6];
1388         if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1389                    &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1390                 errx(1, "Failed to parse mac address '%s'", macaddr);
1391         mac[0] = m[0];
1392         mac[1] = m[1];
1393         mac[2] = m[2];
1394         mac[3] = m[3];
1395         mac[4] = m[4];
1396         mac[5] = m[5];
1397 }
1398 
1399 /*
1400  * This code is "adapted" from libbridge: it attaches the Host end of the
1401  * network device to the bridge device specified by the command line.
1402  *
1403  * This is yet another James Morris contribution (I'm an IP-level guy, so I
1404  * dislike bridging), and I just try not to break it.
1405  */
1406 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1407 {
1408         int ifidx;
1409         struct ifreq ifr;
1410 
1411         if (!*br_name)
1412                 errx(1, "must specify bridge name");
1413 
1414         ifidx = if_nametoindex(if_name);
1415         if (!ifidx)
1416                 errx(1, "interface %s does not exist!", if_name);
1417 
1418         strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1419         ifr.ifr_name[IFNAMSIZ-1] = '\0';
1420         ifr.ifr_ifindex = ifidx;
1421         if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1422                 err(1, "can't add %s to bridge %s", if_name, br_name);
1423 }
1424 
1425 /*
1426  * This sets up the Host end of the network device with an IP address, brings
1427  * it up so packets will flow, the copies the MAC address into the hwaddr
1428  * pointer.
1429  */
1430 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1431 {
1432         struct ifreq ifr;
1433         struct sockaddr_in sin;
1434 
1435         memset(&ifr, 0, sizeof(ifr));
1436         strcpy(ifr.ifr_name, tapif);
1437 
1438         /* Don't read these incantations.  Just cut & paste them like I did! */
1439         sin.sin_family = AF_INET;
1440         sin.sin_addr.s_addr = htonl(ipaddr);
1441         memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1442         if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1443                 err(1, "Setting %s interface address", tapif);
1444         ifr.ifr_flags = IFF_UP;
1445         if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1446                 err(1, "Bringing interface %s up", tapif);
1447 }
1448 
1449 static int get_tun_device(char tapif[IFNAMSIZ])
1450 {
1451         struct ifreq ifr;
1452         int netfd;
1453 
1454         /* Start with this zeroed.  Messy but sure. */
1455         memset(&ifr, 0, sizeof(ifr));
1456 
1457         /*
1458          * We open the /dev/net/tun device and tell it we want a tap device.  A
1459          * tap device is like a tun device, only somehow different.  To tell
1460          * the truth, I completely blundered my way through this code, but it
1461          * works now!
1462          */
1463         netfd = open_or_die("/dev/net/tun", O_RDWR);
1464         ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1465         strcpy(ifr.ifr_name, "tap%d");
1466         if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1467                 err(1, "configuring /dev/net/tun");
1468 
1469         if (ioctl(netfd, TUNSETOFFLOAD,
1470                   TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1471                 err(1, "Could not set features for tun device");
1472 
1473         /*
1474          * We don't need checksums calculated for packets coming in this
1475          * device: trust us!
1476          */
1477         ioctl(netfd, TUNSETNOCSUM, 1);
1478 
1479         memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1480         return netfd;
1481 }
1482 
1483 /*L:195
1484  * Our network is a Host<->Guest network.  This can either use bridging or
1485  * routing, but the principle is the same: it uses the "tun" device to inject
1486  * packets into the Host as if they came in from a normal network card.  We
1487  * just shunt packets between the Guest and the tun device.
1488  */
1489 static void setup_tun_net(char *arg)
1490 {
1491         struct device *dev;
1492         struct net_info *net_info = malloc(sizeof(*net_info));
1493         int ipfd;
1494         u32 ip = INADDR_ANY;
1495         bool bridging = false;
1496         char tapif[IFNAMSIZ], *p;
1497         struct virtio_net_config conf;
1498 
1499         net_info->tunfd = get_tun_device(tapif);
1500 
1501         /* First we create a new network device. */
1502         dev = new_device("net", VIRTIO_ID_NET);
1503         dev->priv = net_info;
1504 
1505         /* Network devices need a recv and a send queue, just like console. */
1506         add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1507         add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1508 
1509         /*
1510          * We need a socket to perform the magic network ioctls to bring up the
1511          * tap interface, connect to the bridge etc.  Any socket will do!
1512          */
1513         ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1514         if (ipfd < 0)
1515                 err(1, "opening IP socket");
1516 
1517         /* If the command line was --tunnet=bridge:<name> do bridging. */
1518         if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1519                 arg += strlen(BRIDGE_PFX);
1520                 bridging = true;
1521         }
1522 
1523         /* A mac address may follow the bridge name or IP address */
1524         p = strchr(arg, ':');
1525         if (p) {
1526                 str2mac(p+1, conf.mac);
1527                 add_feature(dev, VIRTIO_NET_F_MAC);
1528                 *p = '\0';
1529         }
1530 
1531         /* arg is now either an IP address or a bridge name */
1532         if (bridging)
1533                 add_to_bridge(ipfd, tapif, arg);
1534         else
1535                 ip = str2ip(arg);
1536 
1537         /* Set up the tun device. */
1538         configure_device(ipfd, tapif, ip);
1539 
1540         /* Expect Guest to handle everything except UFO */
1541         add_feature(dev, VIRTIO_NET_F_CSUM);
1542         add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1543         add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1544         add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1545         add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1546         add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1547         add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1548         add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1549         /* We handle indirect ring entries */
1550         add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1551         /* We're compliant with the damn spec. */
1552         add_feature(dev, VIRTIO_F_ANY_LAYOUT);
1553         set_config(dev, sizeof(conf), &conf);
1554 
1555         /* We don't need the socket any more; setup is done. */
1556         close(ipfd);
1557 
1558         devices.device_num++;
1559 
1560         if (bridging)
1561                 verbose("device %u: tun %s attached to bridge: %s\n",
1562                         devices.device_num, tapif, arg);
1563         else
1564                 verbose("device %u: tun %s: %s\n",
1565                         devices.device_num, tapif, arg);
1566 }
1567 /*:*/
1568 
1569 /* This hangs off device->priv. */
1570 struct vblk_info {
1571         /* The size of the file. */
1572         off64_t len;
1573 
1574         /* The file descriptor for the file. */
1575         int fd;
1576 
1577 };
1578 
1579 /*L:210
1580  * The Disk
1581  *
1582  * The disk only has one virtqueue, so it only has one thread.  It is really
1583  * simple: the Guest asks for a block number and we read or write that position
1584  * in the file.
1585  *
1586  * Before we serviced each virtqueue in a separate thread, that was unacceptably
1587  * slow: the Guest waits until the read is finished before running anything
1588  * else, even if it could have been doing useful work.
1589  *
1590  * We could have used async I/O, except it's reputed to suck so hard that
1591  * characters actually go missing from your code when you try to use it.
1592  */
1593 static void blk_request(struct virtqueue *vq)
1594 {
1595         struct vblk_info *vblk = vq->dev->priv;
1596         unsigned int head, out_num, in_num, wlen;
1597         int ret, i;
1598         u8 *in;
1599         struct virtio_blk_outhdr out;
1600         struct iovec iov[vq->vring.num];
1601         off64_t off;
1602 
1603         /*
1604          * Get the next request, where we normally wait.  It triggers the
1605          * interrupt to acknowledge previously serviced requests (if any).
1606          */
1607         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1608 
1609         /* Copy the output header from the front of the iov (adjusts iov) */
1610         iov_consume(iov, out_num, &out, sizeof(out));
1611 
1612         /* Find and trim end of iov input array, for our status byte. */
1613         in = NULL;
1614         for (i = out_num + in_num - 1; i >= out_num; i--) {
1615                 if (iov[i].iov_len > 0) {
1616                         in = iov[i].iov_base + iov[i].iov_len - 1;
1617                         iov[i].iov_len--;
1618                         break;
1619                 }
1620         }
1621         if (!in)
1622                 errx(1, "Bad virtblk cmd with no room for status");
1623 
1624         /*
1625          * For historical reasons, block operations are expressed in 512 byte
1626          * "sectors".
1627          */
1628         off = out.sector * 512;
1629 
1630         /*
1631          * In general the virtio block driver is allowed to try SCSI commands.
1632          * It'd be nice if we supported eject, for example, but we don't.
1633          */
1634         if (out.type & VIRTIO_BLK_T_SCSI_CMD) {
1635                 fprintf(stderr, "Scsi commands unsupported\n");
1636                 *in = VIRTIO_BLK_S_UNSUPP;
1637                 wlen = sizeof(*in);
1638         } else if (out.type & VIRTIO_BLK_T_OUT) {
1639                 /*
1640                  * Write
1641                  *
1642                  * Move to the right location in the block file.  This can fail
1643                  * if they try to write past end.
1644                  */
1645                 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1646                         err(1, "Bad seek to sector %llu", out.sector);
1647 
1648                 ret = writev(vblk->fd, iov, out_num);
1649                 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
1650 
1651                 /*
1652                  * Grr... Now we know how long the descriptor they sent was, we
1653                  * make sure they didn't try to write over the end of the block
1654                  * file (possibly extending it).
1655                  */
1656                 if (ret > 0 && off + ret > vblk->len) {
1657                         /* Trim it back to the correct length */
1658                         ftruncate64(vblk->fd, vblk->len);
1659                         /* Die, bad Guest, die. */
1660                         errx(1, "Write past end %llu+%u", off, ret);
1661                 }
1662 
1663                 wlen = sizeof(*in);
1664                 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1665         } else if (out.type & VIRTIO_BLK_T_FLUSH) {
1666                 /* Flush */
1667                 ret = fdatasync(vblk->fd);
1668                 verbose("FLUSH fdatasync: %i\n", ret);
1669                 wlen = sizeof(*in);
1670                 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1671         } else {
1672                 /*
1673                  * Read
1674                  *
1675                  * Move to the right location in the block file.  This can fail
1676                  * if they try to read past end.
1677                  */
1678                 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1679                         err(1, "Bad seek to sector %llu", out.sector);
1680 
1681                 ret = readv(vblk->fd, iov + out_num, in_num);
1682                 if (ret >= 0) {
1683                         wlen = sizeof(*in) + ret;
1684                         *in = VIRTIO_BLK_S_OK;
1685                 } else {
1686                         wlen = sizeof(*in);
1687                         *in = VIRTIO_BLK_S_IOERR;
1688                 }
1689         }
1690 
1691         /* Finished that request. */
1692         add_used(vq, head, wlen);
1693 }
1694 
1695 /*L:198 This actually sets up a virtual block device. */
1696 static void setup_block_file(const char *filename)
1697 {
1698         struct device *dev;
1699         struct vblk_info *vblk;
1700         struct virtio_blk_config conf;
1701 
1702         /* Creat the device. */
1703         dev = new_device("block", VIRTIO_ID_BLOCK);
1704 
1705         /* The device has one virtqueue, where the Guest places requests. */
1706         add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1707 
1708         /* Allocate the room for our own bookkeeping */
1709         vblk = dev->priv = malloc(sizeof(*vblk));
1710 
1711         /* First we open the file and store the length. */
1712         vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1713         vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1714 
1715         /* We support FLUSH. */
1716         add_feature(dev, VIRTIO_BLK_F_FLUSH);
1717 
1718         /* Tell Guest how many sectors this device has. */
1719         conf.capacity = cpu_to_le64(vblk->len / 512);
1720 
1721         /*
1722          * Tell Guest not to put in too many descriptors at once: two are used
1723          * for the in and out elements.
1724          */
1725         add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1726         conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1727 
1728         /* Don't try to put whole struct: we have 8 bit limit. */
1729         set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1730 
1731         verbose("device %u: virtblock %llu sectors\n",
1732                 ++devices.device_num, le64_to_cpu(conf.capacity));
1733 }
1734 
1735 /*L:211
1736  * Our random number generator device reads from /dev/random into the Guest's
1737  * input buffers.  The usual case is that the Guest doesn't want random numbers
1738  * and so has no buffers although /dev/random is still readable, whereas
1739  * console is the reverse.
1740  *
1741  * The same logic applies, however.
1742  */
1743 struct rng_info {
1744         int rfd;
1745 };
1746 
1747 static void rng_input(struct virtqueue *vq)
1748 {
1749         int len;
1750         unsigned int head, in_num, out_num, totlen = 0;
1751         struct rng_info *rng_info = vq->dev->priv;
1752         struct iovec iov[vq->vring.num];
1753 
1754         /* First we need a buffer from the Guests's virtqueue. */
1755         head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1756         if (out_num)
1757                 errx(1, "Output buffers in rng?");
1758 
1759         /*
1760          * Just like the console write, we loop to cover the whole iovec.
1761          * In this case, short reads actually happen quite a bit.
1762          */
1763         while (!iov_empty(iov, in_num)) {
1764                 len = readv(rng_info->rfd, iov, in_num);
1765                 if (len <= 0)
1766                         err(1, "Read from /dev/random gave %i", len);
1767                 iov_consume(iov, in_num, NULL, len);
1768                 totlen += len;
1769         }
1770 
1771         /* Tell the Guest about the new input. */
1772         add_used(vq, head, totlen);
1773 }
1774 
1775 /*L:199
1776  * This creates a "hardware" random number device for the Guest.
1777  */
1778 static void setup_rng(void)
1779 {
1780         struct device *dev;
1781         struct rng_info *rng_info = malloc(sizeof(*rng_info));
1782 
1783         /* Our device's privat info simply contains the /dev/random fd. */
1784         rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1785 
1786         /* Create the new device. */
1787         dev = new_device("rng", VIRTIO_ID_RNG);
1788         dev->priv = rng_info;
1789 
1790         /* The device has one virtqueue, where the Guest places inbufs. */
1791         add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1792 
1793         verbose("device %u: rng\n", devices.device_num++);
1794 }
1795 /* That's the end of device setup. */
1796 
1797 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1798 static void __attribute__((noreturn)) restart_guest(void)
1799 {
1800         unsigned int i;
1801 
1802         /*
1803          * Since we don't track all open fds, we simply close everything beyond
1804          * stderr.
1805          */
1806         for (i = 3; i < FD_SETSIZE; i++)
1807                 close(i);
1808 
1809         /* Reset all the devices (kills all threads). */
1810         cleanup_devices();
1811 
1812         execv(main_args[0], main_args);
1813         err(1, "Could not exec %s", main_args[0]);
1814 }
1815 
1816 /*L:220
1817  * Finally we reach the core of the Launcher which runs the Guest, serves
1818  * its input and output, and finally, lays it to rest.
1819  */
1820 static void __attribute__((noreturn)) run_guest(void)
1821 {
1822         for (;;) {
1823                 unsigned long notify_addr;
1824                 int readval;
1825 
1826                 /* We read from the /dev/lguest device to run the Guest. */
1827                 readval = pread(lguest_fd, &notify_addr,
1828                                 sizeof(notify_addr), cpu_id);
1829 
1830                 /* One unsigned long means the Guest did HCALL_NOTIFY */
1831                 if (readval == sizeof(notify_addr)) {
1832                         verbose("Notify on address %#lx\n", notify_addr);
1833                         handle_output(notify_addr);
1834                 /* ENOENT means the Guest died.  Reading tells us why. */
1835                 } else if (errno == ENOENT) {
1836                         char reason[1024] = { 0 };
1837                         pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1838                         errx(1, "%s", reason);
1839                 /* ERESTART means that we need to reboot the guest */
1840                 } else if (errno == ERESTART) {
1841                         restart_guest();
1842                 /* Anything else means a bug or incompatible change. */
1843                 } else
1844                         err(1, "Running guest failed");
1845         }
1846 }
1847 /*L:240
1848  * This is the end of the Launcher.  The good news: we are over halfway
1849  * through!  The bad news: the most fiendish part of the code still lies ahead
1850  * of us.
1851  *
1852  * Are you ready?  Take a deep breath and join me in the core of the Host, in
1853  * "make Host".
1854 :*/
1855 
1856 static struct option opts[] = {
1857         { "verbose", 0, NULL, 'v' },
1858         { "tunnet", 1, NULL, 't' },
1859         { "block", 1, NULL, 'b' },
1860         { "rng", 0, NULL, 'r' },
1861         { "initrd", 1, NULL, 'i' },
1862         { "username", 1, NULL, 'u' },
1863         { "chroot", 1, NULL, 'c' },
1864         { NULL },
1865 };
1866 static void usage(void)
1867 {
1868         errx(1, "Usage: lguest [--verbose] "
1869              "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1870              "|--block=<filename>|--initrd=<filename>]...\n"
1871              "<mem-in-mb> vmlinux [args...]");
1872 }
1873 
1874 /*L:105 The main routine is where the real work begins: */
1875 int main(int argc, char *argv[])
1876 {
1877         /* Memory, code startpoint and size of the (optional) initrd. */
1878         unsigned long mem = 0, start, initrd_size = 0;
1879         /* Two temporaries. */
1880         int i, c;
1881         /* The boot information for the Guest. */
1882         struct boot_params *boot;
1883         /* If they specify an initrd file to load. */
1884         const char *initrd_name = NULL;
1885 
1886         /* Password structure for initgroups/setres[gu]id */
1887         struct passwd *user_details = NULL;
1888 
1889         /* Directory to chroot to */
1890         char *chroot_path = NULL;
1891 
1892         /* Save the args: we "reboot" by execing ourselves again. */
1893         main_args = argv;
1894 
1895         /*
1896          * First we initialize the device list.  We keep a pointer to the last
1897          * device, and the next interrupt number to use for devices (1:
1898          * remember that 0 is used by the timer).
1899          */
1900         devices.lastdev = NULL;
1901         devices.next_irq = 1;
1902 
1903         /* We're CPU 0.  In fact, that's the only CPU possible right now. */
1904         cpu_id = 0;
1905 
1906         /*
1907          * We need to know how much memory so we can set up the device
1908          * descriptor and memory pages for the devices as we parse the command
1909          * line.  So we quickly look through the arguments to find the amount
1910          * of memory now.
1911          */
1912         for (i = 1; i < argc; i++) {
1913                 if (argv[i][0] != '-') {
1914                         mem = atoi(argv[i]) * 1024 * 1024;
1915                         /*
1916                          * We start by mapping anonymous pages over all of
1917                          * guest-physical memory range.  This fills it with 0,
1918                          * and ensures that the Guest won't be killed when it
1919                          * tries to access it.
1920                          */
1921                         guest_base = map_zeroed_pages(mem / getpagesize()
1922                                                       + DEVICE_PAGES);
1923                         guest_limit = mem;
1924                         guest_max = mem + DEVICE_PAGES*getpagesize();
1925                         devices.descpage = get_pages(1);
1926                         break;
1927                 }
1928         }
1929 
1930         /* The options are fairly straight-forward */
1931         while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1932                 switch (c) {
1933                 case 'v':
1934                         verbose = true;
1935                         break;
1936                 case 't':
1937                         setup_tun_net(optarg);
1938                         break;
1939                 case 'b':
1940                         setup_block_file(optarg);
1941                         break;
1942                 case 'r':
1943                         setup_rng();
1944                         break;
1945                 case 'i':
1946                         initrd_name = optarg;
1947                         break;
1948                 case 'u':
1949                         user_details = getpwnam(optarg);
1950                         if (!user_details)
1951                                 err(1, "getpwnam failed, incorrect username?");
1952                         break;
1953                 case 'c':
1954                         chroot_path = optarg;
1955                         break;
1956                 default:
1957                         warnx("Unknown argument %s", argv[optind]);
1958                         usage();
1959                 }
1960         }
1961         /*
1962          * After the other arguments we expect memory and kernel image name,
1963          * followed by command line arguments for the kernel.
1964          */
1965         if (optind + 2 > argc)
1966                 usage();
1967 
1968         verbose("Guest base is at %p\n", guest_base);
1969 
1970         /* We always have a console device */
1971         setup_console();
1972 
1973         /* Now we load the kernel */
1974         start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1975 
1976         /* Boot information is stashed at physical address 0 */
1977         boot = from_guest_phys(0);
1978 
1979         /* Map the initrd image if requested (at top of physical memory) */
1980         if (initrd_name) {
1981                 initrd_size = load_initrd(initrd_name, mem);
1982                 /*
1983                  * These are the location in the Linux boot header where the
1984                  * start and size of the initrd are expected to be found.
1985                  */
1986                 boot->hdr.ramdisk_image = mem - initrd_size;
1987                 boot->hdr.ramdisk_size = initrd_size;
1988                 /* The bootloader type 0xFF means "unknown"; that's OK. */
1989                 boot->hdr.type_of_loader = 0xFF;
1990         }
1991 
1992         /*
1993          * The Linux boot header contains an "E820" memory map: ours is a
1994          * simple, single region.
1995          */
1996         boot->e820_entries = 1;
1997         boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1998         /*
1999          * The boot header contains a command line pointer: we put the command
2000          * line after the boot header.
2001          */
2002         boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2003         /* We use a simple helper to copy the arguments separated by spaces. */
2004         concat((char *)(boot + 1), argv+optind+2);
2005 
2006         /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
2007         boot->hdr.kernel_alignment = 0x1000000;
2008 
2009         /* Boot protocol version: 2.07 supports the fields for lguest. */
2010         boot->hdr.version = 0x207;
2011 
2012         /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2013         boot->hdr.hardware_subarch = 1;
2014 
2015         /* Tell the entry path not to try to reload segment registers. */
2016         boot->hdr.loadflags |= KEEP_SEGMENTS;
2017 
2018         /* We tell the kernel to initialize the Guest. */
2019         tell_kernel(start);
2020 
2021         /* Ensure that we terminate if a device-servicing child dies. */
2022         signal(SIGCHLD, kill_launcher);
2023 
2024         /* If we exit via err(), this kills all the threads, restores tty. */
2025         atexit(cleanup_devices);
2026 
2027         /* If requested, chroot to a directory */
2028         if (chroot_path) {
2029                 if (chroot(chroot_path) != 0)
2030                         err(1, "chroot(\"%s\") failed", chroot_path);
2031 
2032                 if (chdir("/") != 0)
2033                         err(1, "chdir(\"/\") failed");
2034 
2035                 verbose("chroot done\n");
2036         }
2037 
2038         /* If requested, drop privileges */
2039         if (user_details) {
2040                 uid_t u;
2041                 gid_t g;
2042 
2043                 u = user_details->pw_uid;
2044                 g = user_details->pw_gid;
2045 
2046                 if (initgroups(user_details->pw_name, g) != 0)
2047                         err(1, "initgroups failed");
2048 
2049                 if (setresgid(g, g, g) != 0)
2050                         err(1, "setresgid failed");
2051 
2052                 if (setresuid(u, u, u) != 0)
2053                         err(1, "setresuid failed");
2054 
2055                 verbose("Dropping privileges completed\n");
2056         }
2057 
2058         /* Finally, run the Guest.  This doesn't return. */
2059         run_guest();
2060 }
2061 /*:*/
2062 
2063 /*M:999
2064  * Mastery is done: you now know everything I do.
2065  *
2066  * But surely you have seen code, features and bugs in your wanderings which
2067  * you now yearn to attack?  That is the real game, and I look forward to you
2068  * patching and forking lguest into the Your-Name-Here-visor.
2069  *
2070  * Farewell, and good coding!
2071  * Rusty Russell.
2072  */
2073 

~ [ 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