1 /*P:100 This is the Launcher code, a simple program which lays out the
2 * "physical" memory for the new Guest by mapping the kernel image and the
3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest.
5 #define _LARGEFILE64_SOURCE
15 #include <sys/param.h>
16 #include <sys/types.h>
23 #include <sys/socket.h>
24 #include <sys/ioctl.h>
27 #include <netinet/in.h>
29 #include <linux/sockios.h>
30 #include <linux/if_tun.h>
37 #include "linux/lguest_launcher.h"
38 #include "linux/virtio_config.h"
39 #include "linux/virtio_net.h"
40 #include "linux/virtio_blk.h"
41 #include "linux/virtio_console.h"
42 #include "linux/virtio_ring.h"
43 #include "asm-x86/bootparam.h"
44 /*L:110 We can ignore the 38 include files we need for this program, but I do
45 * want to draw attention to the use of kernel-style types.
47 * As Linus said, "C is a Spartan language, and so should your naming be." I
48 * like these abbreviations, so we define them here. Note that u64 is always
49 * unsigned long long, which works on all Linux systems: this means that we can
50 * use %llu in printf for any u64. */
51 typedef unsigned long long u64;
57 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
59 #define BRIDGE_PFX "bridge:"
61 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
63 /* We can have up to 256 pages for devices. */
64 #define DEVICE_PAGES 256
65 /* This will occupy 2 pages: it must be a power of 2. */
66 #define VIRTQUEUE_NUM 128
68 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
69 * this, and although I wouldn't recommend it, it works quite nicely here. */
71 #define verbose(args...) \
72 do { if (verbose) printf(args); } while(0)
75 /* The pipe to send commands to the waker process */
77 /* The pointer to the start of guest memory. */
78 static void *guest_base;
79 /* The maximum guest physical address allowed, and maximum possible. */
80 static unsigned long guest_limit, guest_max;
82 /* This is our list of devices. */
85 /* Summary information about the devices in our list: ready to pass to
86 * select() to ask which need servicing.*/
90 /* Counter to assign interrupt numbers. */
91 unsigned int next_irq;
93 /* Counter to print out convenient device numbers. */
94 unsigned int device_num;
96 /* The descriptor page for the devices. */
99 /* The tail of the last descriptor. */
100 unsigned int desc_used;
102 /* A single linked list of devices. */
104 /* ... And an end pointer so we can easily append new devices */
105 struct device **lastdev;
108 /* The list of Guest devices, based on command line arguments. */
109 static struct device_list devices;
111 /* The device structure describes a single device. */
114 /* The linked-list pointer. */
117 /* The this device's descriptor, as mapped into the Guest. */
118 struct lguest_device_desc *desc;
120 /* The name of this device, for --verbose. */
123 /* If handle_input is set, it wants to be called when this file
124 * descriptor is ready. */
126 bool (*handle_input)(int fd, struct device *me);
128 /* Any queues attached to this device */
129 struct virtqueue *vq;
131 /* Device-specific data. */
135 /* The virtqueue structure describes a queue attached to a device. */
138 struct virtqueue *next;
140 /* Which device owns me. */
143 /* The configuration for this queue. */
144 struct lguest_vqconfig config;
146 /* The actual ring of buffers. */
149 /* Last available index we saw. */
152 /* The routine to call when the Guest pings us. */
153 void (*handle_output)(int fd, struct virtqueue *me);
156 /* Remember the arguments to the program so we can "reboot" */
157 static char **main_args;
159 /* Since guest is UP and we don't run at the same time, we don't need barriers.
160 * But I include them in the code in case others copy it. */
163 /* Convert an iovec element to the given type.
165 * This is a fairly ugly trick: we need to know the size of the type and
166 * alignment requirement to check the pointer is kosher. It's also nice to
167 * have the name of the type in case we report failure.
169 * Typing those three things all the time is cumbersome and error prone, so we
170 * have a macro which sets them all up and passes to the real function. */
171 #define convert(iov, type) \
172 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
174 static void *_convert(struct iovec *iov, size_t size, size_t align,
177 if (iov->iov_len != size)
178 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
179 if ((unsigned long)iov->iov_base % align != 0)
180 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
181 return iov->iov_base;
184 /* The virtio configuration space is defined to be little-endian. x86 is
185 * little-endian too, but it's nice to be explicit so we have these helpers. */
186 #define cpu_to_le16(v16) (v16)
187 #define cpu_to_le32(v32) (v32)
188 #define cpu_to_le64(v64) (v64)
189 #define le16_to_cpu(v16) (v16)
190 #define le32_to_cpu(v32) (v32)
191 #define le64_to_cpu(v32) (v64)
193 /*L:100 The Launcher code itself takes us out into userspace, that scary place
194 * where pointers run wild and free! Unfortunately, like most userspace
195 * programs, it's quite boring (which is why everyone likes to hack on the
196 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
197 * will get you through this section. Or, maybe not.
199 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
200 * memory and stores it in "guest_base". In other words, Guest physical ==
201 * Launcher virtual with an offset.
203 * This can be tough to get your head around, but usually it just means that we
204 * use these trivial conversion functions when the Guest gives us it's
205 * "physical" addresses: */
206 static void *from_guest_phys(unsigned long addr)
208 return guest_base + addr;
211 static unsigned long to_guest_phys(const void *addr)
213 return (addr - guest_base);
217 * Loading the Kernel.
219 * We start with couple of simple helper routines. open_or_die() avoids
220 * error-checking code cluttering the callers: */
221 static int open_or_die(const char *name, int flags)
223 int fd = open(name, flags);
225 err(1, "Failed to open %s", name);
229 /* map_zeroed_pages() takes a number of pages. */
230 static void *map_zeroed_pages(unsigned int num)
232 int fd = open_or_die("/dev/zero", O_RDONLY);
235 /* We use a private mapping (ie. if we write to the page, it will be
237 addr = mmap(NULL, getpagesize() * num,
238 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
239 if (addr == MAP_FAILED)
240 err(1, "Mmaping %u pages of /dev/zero", num);
245 /* Get some more pages for a device. */
246 static void *get_pages(unsigned int num)
248 void *addr = from_guest_phys(guest_limit);
250 guest_limit += num * getpagesize();
251 if (guest_limit > guest_max)
252 errx(1, "Not enough memory for devices");
256 /* This routine is used to load the kernel or initrd. It tries mmap, but if
257 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
258 * it falls back to reading the memory in. */
259 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
263 /* We map writable even though for some segments are marked read-only.
264 * The kernel really wants to be writable: it patches its own
267 * MAP_PRIVATE means that the page won't be copied until a write is
268 * done to it. This allows us to share untouched memory between
270 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
271 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
274 /* pread does a seek and a read in one shot: saves a few lines. */
275 r = pread(fd, addr, len, offset);
277 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
280 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
281 * the Guest memory. ELF = Embedded Linking Format, which is the format used
282 * by all modern binaries on Linux including the kernel.
284 * The ELF headers give *two* addresses: a physical address, and a virtual
285 * address. We use the physical address; the Guest will map itself to the
288 * We return the starting address. */
289 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
291 Elf32_Phdr phdr[ehdr->e_phnum];
294 /* Sanity checks on the main ELF header: an x86 executable with a
295 * reasonable number of correctly-sized program headers. */
296 if (ehdr->e_type != ET_EXEC
297 || ehdr->e_machine != EM_386
298 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
299 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
300 errx(1, "Malformed elf header");
302 /* An ELF executable contains an ELF header and a number of "program"
303 * headers which indicate which parts ("segments") of the program to
306 /* We read in all the program headers at once: */
307 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
308 err(1, "Seeking to program headers");
309 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
310 err(1, "Reading program headers");
312 /* Try all the headers: there are usually only three. A read-only one,
313 * a read-write one, and a "note" section which isn't loadable. */
314 for (i = 0; i < ehdr->e_phnum; i++) {
315 /* If this isn't a loadable segment, we ignore it */
316 if (phdr[i].p_type != PT_LOAD)
319 verbose("Section %i: size %i addr %p\n",
320 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
322 /* We map this section of the file at its physical address. */
323 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
324 phdr[i].p_offset, phdr[i].p_filesz);
327 /* The entry point is given in the ELF header. */
328 return ehdr->e_entry;
331 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
332 * supposed to jump into it and it will unpack itself. We used to have to
333 * perform some hairy magic because the unpacking code scared me.
335 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
336 * a small patch to jump over the tricky bits in the Guest, so now we just read
337 * the funky header so we know where in the file to load, and away we go! */
338 static unsigned long load_bzimage(int fd)
340 struct boot_params boot;
342 /* Modern bzImages get loaded at 1M. */
343 void *p = from_guest_phys(0x100000);
345 /* Go back to the start of the file and read the header. It should be
346 * a Linux boot header (see Documentation/i386/boot.txt) */
347 lseek(fd, 0, SEEK_SET);
348 read(fd, &boot, sizeof(boot));
350 /* Inside the setup_hdr, we expect the magic "HdrS" */
351 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
352 errx(1, "This doesn't look like a bzImage to me");
354 /* Skip over the extra sectors of the header. */
355 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
357 /* Now read everything into memory. in nice big chunks. */
358 while ((r = read(fd, p, 65536)) > 0)
361 /* Finally, code32_start tells us where to enter the kernel. */
362 return boot.hdr.code32_start;
365 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
366 * come wrapped up in the self-decompressing "bzImage" format. With a little
367 * work, we can load those, too. */
368 static unsigned long load_kernel(int fd)
372 /* Read in the first few bytes. */
373 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
374 err(1, "Reading kernel");
376 /* If it's an ELF file, it starts with "\177ELF" */
377 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
378 return map_elf(fd, &hdr);
380 /* Otherwise we assume it's a bzImage, and try to unpack it */
381 return load_bzimage(fd);
384 /* This is a trivial little helper to align pages. Andi Kleen hated it because
385 * it calls getpagesize() twice: "it's dumb code."
387 * Kernel guys get really het up about optimization, even when it's not
388 * necessary. I leave this code as a reaction against that. */
389 static inline unsigned long page_align(unsigned long addr)
391 /* Add upwards and truncate downwards. */
392 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
395 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
396 * the kernel which the kernel can use to boot from without needing any
397 * drivers. Most distributions now use this as standard: the initrd contains
398 * the code to load the appropriate driver modules for the current machine.
400 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
401 * kernels. He sent me this (and tells me when I break it). */
402 static unsigned long load_initrd(const char *name, unsigned long mem)
408 ifd = open_or_die(name, O_RDONLY);
409 /* fstat() is needed to get the file size. */
410 if (fstat(ifd, &st) < 0)
411 err(1, "fstat() on initrd '%s'", name);
413 /* We map the initrd at the top of memory, but mmap wants it to be
414 * page-aligned, so we round the size up for that. */
415 len = page_align(st.st_size);
416 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
417 /* Once a file is mapped, you can close the file descriptor. It's a
418 * little odd, but quite useful. */
420 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
422 /* We return the initrd size. */
426 /* Once we know how much memory we have, we can construct simple linear page
427 * tables which set virtual == physical which will get the Guest far enough
428 * into the boot to create its own.
430 * We lay them out of the way, just below the initrd (which is why we need to
432 static unsigned long setup_pagetables(unsigned long mem,
433 unsigned long initrd_size)
435 unsigned long *pgdir, *linear;
436 unsigned int mapped_pages, i, linear_pages;
437 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
439 mapped_pages = mem/getpagesize();
441 /* Each PTE page can map ptes_per_page pages: how many do we need? */
442 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
444 /* We put the toplevel page directory page at the top of memory. */
445 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
447 /* Now we use the next linear_pages pages as pte pages */
448 linear = (void *)pgdir - linear_pages*getpagesize();
450 /* Linear mapping is easy: put every page's address into the mapping in
451 * order. PAGE_PRESENT contains the flags Present, Writable and
453 for (i = 0; i < mapped_pages; i++)
454 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
456 /* The top level points to the linear page table pages above. */
457 for (i = 0; i < mapped_pages; i += ptes_per_page) {
458 pgdir[i/ptes_per_page]
459 = ((to_guest_phys(linear) + i*sizeof(void *))
463 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
464 mapped_pages, linear_pages, to_guest_phys(linear));
466 /* We return the top level (guest-physical) address: the kernel needs
467 * to know where it is. */
468 return to_guest_phys(pgdir);
472 /* Simple routine to roll all the commandline arguments together with spaces
474 static void concat(char *dst, char *args[])
476 unsigned int i, len = 0;
478 for (i = 0; args[i]; i++) {
479 strcpy(dst+len, args[i]);
480 strcat(dst+len, " ");
481 len += strlen(args[i]) + 1;
483 /* In case it's empty. */
487 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
488 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
489 * the base of Guest "physical" memory, the top physical page to allow, the
490 * top level pagetable and the entry point for the Guest. */
491 static int tell_kernel(unsigned long pgdir, unsigned long start)
493 unsigned long args[] = { LHREQ_INITIALIZE,
494 (unsigned long)guest_base,
495 guest_limit / getpagesize(), pgdir, start };
498 verbose("Guest: %p - %p (%#lx)\n",
499 guest_base, guest_base + guest_limit, guest_limit);
500 fd = open_or_die("/dev/lguest", O_RDWR);
501 if (write(fd, args, sizeof(args)) < 0)
502 err(1, "Writing to /dev/lguest");
504 /* We return the /dev/lguest file descriptor to control this Guest */
509 static void add_device_fd(int fd)
511 FD_SET(fd, &devices.infds);
512 if (fd > devices.max_infd)
513 devices.max_infd = fd;
519 * With console, block and network devices, we can have lots of input which we
520 * need to process. We could try to tell the kernel what file descriptors to
521 * watch, but handing a file descriptor mask through to the kernel is fairly
524 * Instead, we fork off a process which watches the file descriptors and writes
525 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
526 * stop running the Guest. This causes the Launcher to return from the
527 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
528 * the LHREQ_BREAK and wake us up again.
530 * This, of course, is merely a different *kind* of icky.
532 static void wake_parent(int pipefd, int lguest_fd)
534 /* Add the pipe from the Launcher to the fdset in the device_list, so
535 * we watch it, too. */
536 add_device_fd(pipefd);
539 fd_set rfds = devices.infds;
540 unsigned long args[] = { LHREQ_BREAK, 1 };
542 /* Wait until input is ready from one of the devices. */
543 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
544 /* Is it a message from the Launcher? */
545 if (FD_ISSET(pipefd, &rfds)) {
547 /* If read() returns 0, it means the Launcher has
548 * exited. We silently follow. */
549 if (read(pipefd, &fd, sizeof(fd)) == 0)
551 /* Otherwise it's telling us to change what file
552 * descriptors we're to listen to. Positive means
553 * listen to a new one, negative means stop
556 FD_SET(fd, &devices.infds);
558 FD_CLR(-fd - 1, &devices.infds);
559 } else /* Send LHREQ_BREAK command. */
560 write(lguest_fd, args, sizeof(args));
564 /* This routine just sets up a pipe to the Waker process. */
565 static int setup_waker(int lguest_fd)
567 int pipefd[2], child;
569 /* We create a pipe to talk to the Waker, and also so it knows when the
570 * Launcher dies (and closes pipe). */
577 /* We are the Waker: close the "writing" end of our copy of the
578 * pipe and start waiting for input. */
580 wake_parent(pipefd[0], lguest_fd);
582 /* Close the reading end of our copy of the pipe. */
585 /* Here is the fd used to talk to the waker. */
592 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
593 * We need to make sure it's not trying to reach into the Launcher itself, so
594 * we have a convenient routine which checks it and exits with an error message
595 * if something funny is going on:
597 static void *_check_pointer(unsigned long addr, unsigned int size,
600 /* We have to separately check addr and addr+size, because size could
601 * be huge and addr + size might wrap around. */
602 if (addr >= guest_limit || addr + size >= guest_limit)
603 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
604 /* We return a pointer for the caller's convenience, now we know it's
606 return from_guest_phys(addr);
608 /* A macro which transparently hands the line number to the real function. */
609 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
611 /* Each buffer in the virtqueues is actually a chain of descriptors. This
612 * function returns the next descriptor in the chain, or vq->vring.num if we're
614 static unsigned next_desc(struct virtqueue *vq, unsigned int i)
618 /* If this descriptor says it doesn't chain, we're done. */
619 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
620 return vq->vring.num;
622 /* Check they're not leading us off end of descriptors. */
623 next = vq->vring.desc[i].next;
624 /* Make sure compiler knows to grab that: we don't want it changing! */
627 if (next >= vq->vring.num)
628 errx(1, "Desc next is %u", next);
633 /* This looks in the virtqueue and for the first available buffer, and converts
634 * it to an iovec for convenient access. Since descriptors consist of some
635 * number of output then some number of input descriptors, it's actually two
636 * iovecs, but we pack them into one and note how many of each there were.
638 * This function returns the descriptor number found, or vq->vring.num (which
639 * is never a valid descriptor number) if none was found. */
640 static unsigned get_vq_desc(struct virtqueue *vq,
642 unsigned int *out_num, unsigned int *in_num)
644 unsigned int i, head;
646 /* Check it isn't doing very strange things with descriptor numbers. */
647 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
648 errx(1, "Guest moved used index from %u to %u",
649 vq->last_avail_idx, vq->vring.avail->idx);
651 /* If there's nothing new since last we looked, return invalid. */
652 if (vq->vring.avail->idx == vq->last_avail_idx)
653 return vq->vring.num;
655 /* Grab the next descriptor number they're advertising, and increment
656 * the index we've seen. */
657 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
659 /* If their number is silly, that's a fatal mistake. */
660 if (head >= vq->vring.num)
661 errx(1, "Guest says index %u is available", head);
663 /* When we start there are none of either input nor output. */
664 *out_num = *in_num = 0;
668 /* Grab the first descriptor, and check it's OK. */
669 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
670 iov[*out_num + *in_num].iov_base
671 = check_pointer(vq->vring.desc[i].addr,
672 vq->vring.desc[i].len);
673 /* If this is an input descriptor, increment that count. */
674 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
677 /* If it's an output descriptor, they're all supposed
678 * to come before any input descriptors. */
680 errx(1, "Descriptor has out after in");
684 /* If we've got too many, that implies a descriptor loop. */
685 if (*out_num + *in_num > vq->vring.num)
686 errx(1, "Looped descriptor");
687 } while ((i = next_desc(vq, i)) != vq->vring.num);
692 /* After we've used one of their buffers, we tell them about it. We'll then
693 * want to send them an interrupt, using trigger_irq(). */
694 static void add_used(struct virtqueue *vq, unsigned int head, int len)
696 struct vring_used_elem *used;
698 /* The virtqueue contains a ring of used buffers. Get a pointer to the
699 * next entry in that used ring. */
700 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
703 /* Make sure buffer is written before we update index. */
705 vq->vring.used->idx++;
708 /* This actually sends the interrupt for this virtqueue */
709 static void trigger_irq(int fd, struct virtqueue *vq)
711 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
713 /* If they don't want an interrupt, don't send one. */
714 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
717 /* Send the Guest an interrupt tell them we used something up. */
718 if (write(fd, buf, sizeof(buf)) != 0)
719 err(1, "Triggering irq %i", vq->config.irq);
722 /* And here's the combo meal deal. Supersize me! */
723 static void add_used_and_trigger(int fd, struct virtqueue *vq,
724 unsigned int head, int len)
726 add_used(vq, head, len);
733 * Here is the input terminal setting we save, and the routine to restore them
734 * on exit so the user gets their terminal back. */
735 static struct termios orig_term;
736 static void restore_term(void)
738 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
741 /* We associate some data with the console for our exit hack. */
744 /* How many times have they hit ^C? */
746 /* When did they start? */
747 struct timeval start;
750 /* This is the routine which handles console input (ie. stdin). */
751 static bool handle_console_input(int fd, struct device *dev)
754 unsigned int head, in_num, out_num;
755 struct iovec iov[dev->vq->vring.num];
756 struct console_abort *abort = dev->priv;
758 /* First we need a console buffer from the Guests's input virtqueue. */
759 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
761 /* If they're not ready for input, stop listening to this file
762 * descriptor. We'll start again once they add an input buffer. */
763 if (head == dev->vq->vring.num)
767 errx(1, "Output buffers in console in queue?");
769 /* This is why we convert to iovecs: the readv() call uses them, and so
770 * it reads straight into the Guest's buffer. */
771 len = readv(dev->fd, iov, in_num);
773 /* This implies that the console is closed, is /dev/null, or
774 * something went terribly wrong. */
775 warnx("Failed to get console input, ignoring console.");
776 /* Put the input terminal back. */
778 /* Remove callback from input vq, so it doesn't restart us. */
779 dev->vq->handle_output = NULL;
780 /* Stop listening to this fd: don't call us again. */
784 /* Tell the Guest about the new input. */
785 add_used_and_trigger(fd, dev->vq, head, len);
787 /* Three ^C within one second? Exit.
789 * This is such a hack, but works surprisingly well. Each ^C has to be
790 * in a buffer by itself, so they can't be too fast. But we check that
791 * we get three within about a second, so they can't be too slow. */
792 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
794 gettimeofday(&abort->start, NULL);
795 else if (abort->count == 3) {
797 gettimeofday(&now, NULL);
798 if (now.tv_sec <= abort->start.tv_sec+1) {
799 unsigned long args[] = { LHREQ_BREAK, 0 };
800 /* Close the fd so Waker will know it has to
803 /* Just in case waker is blocked in BREAK, send
805 write(fd, args, sizeof(args));
811 /* Any other key resets the abort counter. */
814 /* Everything went OK! */
818 /* Handling output for console is simple: we just get all the output buffers
819 * and write them to stdout. */
820 static void handle_console_output(int fd, struct virtqueue *vq)
822 unsigned int head, out, in;
824 struct iovec iov[vq->vring.num];
826 /* Keep getting output buffers from the Guest until we run out. */
827 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
829 errx(1, "Input buffers in output queue?");
830 len = writev(STDOUT_FILENO, iov, out);
831 add_used_and_trigger(fd, vq, head, len);
838 * Handling output for network is also simple: we get all the output buffers
839 * and write them (ignoring the first element) to this device's file descriptor
841 static void handle_net_output(int fd, struct virtqueue *vq)
843 unsigned int head, out, in;
845 struct iovec iov[vq->vring.num];
847 /* Keep getting output buffers from the Guest until we run out. */
848 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
850 errx(1, "Input buffers in output queue?");
851 /* Check header, but otherwise ignore it (we told the Guest we
852 * supported no features, so it shouldn't have anything
854 (void)convert(&iov[0], struct virtio_net_hdr);
855 len = writev(vq->dev->fd, iov+1, out-1);
856 add_used_and_trigger(fd, vq, head, len);
860 /* This is where we handle a packet coming in from the tun device to our
862 static bool handle_tun_input(int fd, struct device *dev)
864 unsigned int head, in_num, out_num;
866 struct iovec iov[dev->vq->vring.num];
867 struct virtio_net_hdr *hdr;
869 /* First we need a network buffer from the Guests's recv virtqueue. */
870 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
871 if (head == dev->vq->vring.num) {
872 /* Now, it's expected that if we try to send a packet too
873 * early, the Guest won't be ready yet. Wait until the device
874 * status says it's ready. */
875 /* FIXME: Actually want DRIVER_ACTIVE here. */
876 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
877 warn("network: no dma buffer!");
878 /* We'll turn this back on if input buffers are registered. */
881 errx(1, "Output buffers in network recv queue?");
883 /* First element is the header: we set it to 0 (no features). */
884 hdr = convert(&iov[0], struct virtio_net_hdr);
886 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
888 /* Read the packet from the device directly into the Guest's buffer. */
889 len = readv(dev->fd, iov+1, in_num-1);
891 err(1, "reading network");
893 /* Tell the Guest about the new packet. */
894 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
896 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
897 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
898 head != dev->vq->vring.num ? "sent" : "discarded");
904 /*L:215 This is the callback attached to the network and console input
905 * virtqueues: it ensures we try again, in case we stopped console or net
906 * delivery because Guest didn't have any buffers. */
907 static void enable_fd(int fd, struct virtqueue *vq)
909 add_device_fd(vq->dev->fd);
910 /* Tell waker to listen to it again */
911 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
914 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
915 static void handle_output(int fd, unsigned long addr)
918 struct virtqueue *vq;
920 /* Check each virtqueue. */
921 for (i = devices.dev; i; i = i->next) {
922 for (vq = i->vq; vq; vq = vq->next) {
923 if (vq->config.pfn == addr/getpagesize()
924 && vq->handle_output) {
925 verbose("Output to %s\n", vq->dev->name);
926 vq->handle_output(fd, vq);
932 /* Early console write is done using notify on a nul-terminated string
933 * in Guest memory. */
934 if (addr >= guest_limit)
935 errx(1, "Bad NOTIFY %#lx", addr);
937 write(STDOUT_FILENO, from_guest_phys(addr),
938 strnlen(from_guest_phys(addr), guest_limit - addr));
941 /* This is called when the Waker wakes us up: check for incoming file
943 static void handle_input(int fd)
945 /* select() wants a zeroed timeval to mean "don't wait". */
946 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
950 fd_set fds = devices.infds;
952 /* If nothing is ready, we're done. */
953 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
956 /* Otherwise, call the device(s) which have readable
957 * file descriptors and a method of handling them. */
958 for (i = devices.dev; i; i = i->next) {
959 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
961 if (i->handle_input(fd, i))
964 /* If handle_input() returns false, it means we
965 * should no longer service it. Networking and
966 * console do this when there's no input
967 * buffers to deliver into. Console also uses
968 * it when it discovers that stdin is
970 FD_CLR(i->fd, &devices.infds);
971 /* Tell waker to ignore it too, by sending a
972 * negative fd number (-1, since 0 is a valid
975 write(waker_fd, &dev_fd, sizeof(dev_fd));
984 * All devices need a descriptor so the Guest knows it exists, and a "struct
985 * device" so the Launcher can keep track of it. We have common helper
986 * routines to allocate them.
988 * This routine allocates a new "struct lguest_device_desc" from descriptor
989 * table just above the Guest's normal memory. It returns a pointer to that
991 static struct lguest_device_desc *new_dev_desc(u16 type)
993 struct lguest_device_desc *d;
995 /* We only have one page for all the descriptors. */
996 if (devices.desc_used + sizeof(*d) > getpagesize())
997 errx(1, "Too many devices");
999 /* We don't need to set config_len or status: page is 0 already. */
1000 d = (void *)devices.descpage + devices.desc_used;
1002 devices.desc_used += sizeof(*d);
1007 /* Each device descriptor is followed by some configuration information.
1008 * Each configuration field looks like: u8 type, u8 len, [... len bytes...].
1010 * This routine adds a new field to an existing device's descriptor. It only
1011 * works for the last device, but that's OK because that's how we use it. */
1012 static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
1014 /* This is the last descriptor, right? */
1015 assert(devices.descpage + devices.desc_used
1016 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
1018 /* We only have one page of device descriptions. */
1019 if (devices.desc_used + 2 + len > getpagesize())
1020 errx(1, "Too many devices");
1022 /* Copy in the new config header: type then length. */
1023 devices.descpage[devices.desc_used++] = type;
1024 devices.descpage[devices.desc_used++] = len;
1025 memcpy(devices.descpage + devices.desc_used, c, len);
1026 devices.desc_used += len;
1028 /* Update the device descriptor length: two byte head then data. */
1029 dev->desc->config_len += 2 + len;
1032 /* This routine adds a virtqueue to a device. We specify how many descriptors
1033 * the virtqueue is to have. */
1034 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1035 void (*handle_output)(int fd, struct virtqueue *me))
1038 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1041 /* First we need some pages for this virtqueue. */
1042 pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1)
1044 p = get_pages(pages);
1046 /* Initialize the virtqueue */
1048 vq->last_avail_idx = 0;
1051 /* Initialize the configuration. */
1052 vq->config.num = num_descs;
1053 vq->config.irq = devices.next_irq++;
1054 vq->config.pfn = to_guest_phys(p) / getpagesize();
1056 /* Initialize the vring. */
1057 vring_init(&vq->vring, num_descs, p, getpagesize());
1059 /* Add the configuration information to this device's descriptor. */
1060 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE,
1061 sizeof(vq->config), &vq->config);
1063 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1065 for (i = &dev->vq; *i; i = &(*i)->next);
1068 /* Set the routine to call when the Guest does something to this
1070 vq->handle_output = handle_output;
1072 /* Set the "Don't Notify Me" flag if we don't have a handler */
1074 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1077 /* This routine does all the creation and setup of a new device, including
1078 * calling new_dev_desc() to allocate the descriptor and device memory. */
1079 static struct device *new_device(const char *name, u16 type, int fd,
1080 bool (*handle_input)(int, struct device *))
1082 struct device *dev = malloc(sizeof(*dev));
1084 /* Append to device list. Prepending to a single-linked list is
1085 * easier, but the user expects the devices to be arranged on the bus
1086 * in command-line order. The first network device on the command line
1087 * is eth0, the first block device /dev/vda, etc. */
1088 *devices.lastdev = dev;
1090 devices.lastdev = &dev->next;
1092 /* Now we populate the fields one at a time. */
1094 /* If we have an input handler for this file descriptor, then we add it
1095 * to the device_list's fdset and maxfd. */
1097 add_device_fd(dev->fd);
1098 dev->desc = new_dev_desc(type);
1099 dev->handle_input = handle_input;
1105 /* Our first setup routine is the console. It's a fairly simple device, but
1106 * UNIX tty handling makes it uglier than it could be. */
1107 static void setup_console(void)
1111 /* If we can save the initial standard input settings... */
1112 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1113 struct termios term = orig_term;
1114 /* Then we turn off echo, line buffering and ^C etc. We want a
1115 * raw input stream to the Guest. */
1116 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1117 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1118 /* If we exit gracefully, the original settings will be
1119 * restored so the user can see what they're typing. */
1120 atexit(restore_term);
1123 dev = new_device("console", VIRTIO_ID_CONSOLE,
1124 STDIN_FILENO, handle_console_input);
1125 /* We store the console state in dev->priv, and initialize it. */
1126 dev->priv = malloc(sizeof(struct console_abort));
1127 ((struct console_abort *)dev->priv)->count = 0;
1129 /* The console needs two virtqueues: the input then the output. When
1130 * they put something the input queue, we make sure we're listening to
1131 * stdin. When they put something in the output queue, we write it to
1133 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1134 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1136 verbose("device %u: console\n", devices.device_num++);
1140 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1141 * --sharenet=<name> option which opens or creates a named pipe. This can be
1142 * used to send packets to another guest in a 1:1 manner.
1144 * More sopisticated is to use one of the tools developed for project like UML
1147 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1148 * completely generic ("here's my vring, attach to your vring") and would work
1149 * for any traffic. Of course, namespace and permissions issues need to be
1150 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1151 * multiple inter-guest channels behind one interface, although it would
1152 * require some manner of hotplugging new virtio channels.
1154 * Finally, we could implement a virtio network switch in the kernel. :*/
1156 static u32 str2ip(const char *ipaddr)
1158 unsigned int byte[4];
1160 sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]);
1161 return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3];
1164 /* This code is "adapted" from libbridge: it attaches the Host end of the
1165 * network device to the bridge device specified by the command line.
1167 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1168 * dislike bridging), and I just try not to break it. */
1169 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1175 errx(1, "must specify bridge name");
1177 ifidx = if_nametoindex(if_name);
1179 errx(1, "interface %s does not exist!", if_name);
1181 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1182 ifr.ifr_ifindex = ifidx;
1183 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1184 err(1, "can't add %s to bridge %s", if_name, br_name);
1187 /* This sets up the Host end of the network device with an IP address, brings
1188 * it up so packets will flow, the copies the MAC address into the hwaddr
1190 static void configure_device(int fd, const char *devname, u32 ipaddr,
1191 unsigned char hwaddr[6])
1194 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1196 /* Don't read these incantations. Just cut & paste them like I did! */
1197 memset(&ifr, 0, sizeof(ifr));
1198 strcpy(ifr.ifr_name, devname);
1199 sin->sin_family = AF_INET;
1200 sin->sin_addr.s_addr = htonl(ipaddr);
1201 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1202 err(1, "Setting %s interface address", devname);
1203 ifr.ifr_flags = IFF_UP;
1204 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1205 err(1, "Bringing interface %s up", devname);
1207 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1208 * above). IF means Interface, and HWADDR is hardware address.
1210 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1211 err(1, "getting hw address for %s", devname);
1212 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1215 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1216 * routing, but the principle is the same: it uses the "tun" device to inject
1217 * packets into the Host as if they came in from a normal network card. We
1218 * just shunt packets between the Guest and the tun device. */
1219 static void setup_tun_net(const char *arg)
1225 const char *br_name = NULL;
1228 /* We open the /dev/net/tun device and tell it we want a tap device. A
1229 * tap device is like a tun device, only somehow different. To tell
1230 * the truth, I completely blundered my way through this code, but it
1232 netfd = open_or_die("/dev/net/tun", O_RDWR);
1233 memset(&ifr, 0, sizeof(ifr));
1234 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1235 strcpy(ifr.ifr_name, "tap%d");
1236 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1237 err(1, "configuring /dev/net/tun");
1238 /* We don't need checksums calculated for packets coming in this
1239 * device: trust us! */
1240 ioctl(netfd, TUNSETNOCSUM, 1);
1242 /* First we create a new network device. */
1243 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1245 /* Network devices need a receive and a send queue, just like
1247 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1248 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1250 /* We need a socket to perform the magic network ioctls to bring up the
1251 * tap interface, connect to the bridge etc. Any socket will do! */
1252 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1254 err(1, "opening IP socket");
1256 /* If the command line was --tunnet=bridge:<name> do bridging. */
1257 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1259 br_name = arg + strlen(BRIDGE_PFX);
1260 add_to_bridge(ipfd, ifr.ifr_name, br_name);
1261 } else /* It is an IP address to set up the device with */
1264 /* Set up the tun device, and get the mac address for the interface. */
1265 configure_device(ipfd, ifr.ifr_name, ip, hwaddr);
1267 /* Tell Guest what MAC address to use. */
1268 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr);
1270 /* We don't seed the socket any more; setup is done. */
1273 verbose("device %u: tun net %u.%u.%u.%u\n",
1274 devices.device_num++,
1275 (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip);
1277 verbose("attached to bridge: %s\n", br_name);
1280 /* Our block (disk) device should be really simple: the Guest asks for a block
1281 * number and we read or write that position in the file. Unfortunately, that
1282 * was amazingly slow: the Guest waits until the read is finished before
1283 * running anything else, even if it could have been doing useful work.
1285 * We could use async I/O, except it's reputed to suck so hard that characters
1286 * actually go missing from your code when you try to use it.
1288 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1290 /* This hangs off device->priv. */
1293 /* The size of the file. */
1296 /* The file descriptor for the file. */
1299 /* IO thread listens on this file descriptor [0]. */
1302 /* IO thread writes to this file descriptor to mark it done, then
1303 * Launcher triggers interrupt to Guest. */
1311 * Remember that the block device is handled by a separate I/O thread. We head
1312 * straight into the core of that thread here:
1314 static bool service_io(struct device *dev)
1316 struct vblk_info *vblk = dev->priv;
1317 unsigned int head, out_num, in_num, wlen;
1319 struct virtio_blk_inhdr *in;
1320 struct virtio_blk_outhdr *out;
1321 struct iovec iov[dev->vq->vring.num];
1324 /* See if there's a request waiting. If not, nothing to do. */
1325 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1326 if (head == dev->vq->vring.num)
1329 /* Every block request should contain at least one output buffer
1330 * (detailing the location on disk and the type of request) and one
1331 * input buffer (to hold the result). */
1332 if (out_num == 0 || in_num == 0)
1333 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1334 head, out_num, in_num);
1336 out = convert(&iov[0], struct virtio_blk_outhdr);
1337 in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr);
1338 off = out->sector * 512;
1340 /* The block device implements "barriers", where the Guest indicates
1341 * that it wants all previous writes to occur before this write. We
1342 * don't have a way of asking our kernel to do a barrier, so we just
1343 * synchronize all the data in the file. Pretty poor, no? */
1344 if (out->type & VIRTIO_BLK_T_BARRIER)
1345 fdatasync(vblk->fd);
1347 /* In general the virtio block driver is allowed to try SCSI commands.
1348 * It'd be nice if we supported eject, for example, but we don't. */
1349 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1350 fprintf(stderr, "Scsi commands unsupported\n");
1351 in->status = VIRTIO_BLK_S_UNSUPP;
1353 } else if (out->type & VIRTIO_BLK_T_OUT) {
1356 /* Move to the right location in the block file. This can fail
1357 * if they try to write past end. */
1358 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1359 err(1, "Bad seek to sector %llu", out->sector);
1361 ret = writev(vblk->fd, iov+1, out_num-1);
1362 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1364 /* Grr... Now we know how long the descriptor they sent was, we
1365 * make sure they didn't try to write over the end of the block
1366 * file (possibly extending it). */
1367 if (ret > 0 && off + ret > vblk->len) {
1368 /* Trim it back to the correct length */
1369 ftruncate64(vblk->fd, vblk->len);
1370 /* Die, bad Guest, die. */
1371 errx(1, "Write past end %llu+%u", off, ret);
1374 in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1378 /* Move to the right location in the block file. This can fail
1379 * if they try to read past end. */
1380 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1381 err(1, "Bad seek to sector %llu", out->sector);
1383 ret = readv(vblk->fd, iov+1, in_num-1);
1384 verbose("READ from sector %llu: %i\n", out->sector, ret);
1386 wlen = sizeof(*in) + ret;
1387 in->status = VIRTIO_BLK_S_OK;
1390 in->status = VIRTIO_BLK_S_IOERR;
1394 /* We can't trigger an IRQ, because we're not the Launcher. It does
1395 * that when we tell it we're done. */
1396 add_used(dev->vq, head, wlen);
1400 /* This is the thread which actually services the I/O. */
1401 static int io_thread(void *_dev)
1403 struct device *dev = _dev;
1404 struct vblk_info *vblk = dev->priv;
1407 /* Close other side of workpipe so we get 0 read when main dies. */
1408 close(vblk->workpipe[1]);
1409 /* Close the other side of the done_fd pipe. */
1412 /* When this read fails, it means Launcher died, so we follow. */
1413 while (read(vblk->workpipe[0], &c, 1) == 1) {
1414 /* We acknowledge each request immediately to reduce latency,
1415 * rather than waiting until we've done them all. I haven't
1416 * measured to see if it makes any difference. */
1417 while (service_io(dev))
1418 write(vblk->done_fd, &c, 1);
1423 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1424 * when the thread tells us it's completed some I/O. */
1425 static bool handle_io_finish(int fd, struct device *dev)
1429 /* If the I/O thread died, presumably it printed the error, so we
1431 if (read(dev->fd, &c, 1) != 1)
1434 /* It did some work, so trigger the irq. */
1435 trigger_irq(fd, dev->vq);
1439 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1440 static void handle_virtblk_output(int fd, struct virtqueue *vq)
1442 struct vblk_info *vblk = vq->dev->priv;
1445 /* Wake up I/O thread and tell it to go to work! */
1446 if (write(vblk->workpipe[1], &c, 1) != 1)
1447 /* Presumably it indicated why it died. */
1451 /*L:198 This actually sets up a virtual block device. */
1452 static void setup_block_file(const char *filename)
1456 struct vblk_info *vblk;
1461 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1464 /* The device responds to return from I/O thread. */
1465 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1467 /* The device has one virtqueue, where the Guest places requests. */
1468 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1470 /* Allocate the room for our own bookkeeping */
1471 vblk = dev->priv = malloc(sizeof(*vblk));
1473 /* First we open the file and store the length. */
1474 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1475 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1477 /* Tell Guest how many sectors this device has. */
1478 cap = cpu_to_le64(vblk->len / 512);
1479 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1481 /* Tell Guest not to put in too many descriptors at once: two are used
1482 * for the in and out elements. */
1483 val = cpu_to_le32(VIRTQUEUE_NUM - 2);
1484 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val);
1486 /* The I/O thread writes to this end of the pipe when done. */
1487 vblk->done_fd = p[1];
1489 /* This is the second pipe, which is how we tell the I/O thread about
1491 pipe(vblk->workpipe);
1493 /* Create stack for thread and run it */
1494 stack = malloc(32768);
1495 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1496 * becoming a zombie. */
1497 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
1498 err(1, "Creating clone");
1500 /* We don't need to keep the I/O thread's end of the pipes open. */
1501 close(vblk->done_fd);
1502 close(vblk->workpipe[0]);
1504 verbose("device %u: virtblock %llu sectors\n",
1505 devices.device_num, cap);
1507 /* That's the end of device setup. :*/
1510 static void __attribute__((noreturn)) restart_guest(void)
1514 /* Closing pipes causes the waker thread and io_threads to die, and
1515 * closing /dev/lguest cleans up the Guest. Since we don't track all
1516 * open fds, we simply close everything beyond stderr. */
1517 for (i = 3; i < FD_SETSIZE; i++)
1519 execv(main_args[0], main_args);
1520 err(1, "Could not exec %s", main_args[0]);
1523 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1524 * its input and output, and finally, lays it to rest. */
1525 static void __attribute__((noreturn)) run_guest(int lguest_fd)
1528 unsigned long args[] = { LHREQ_BREAK, 0 };
1529 unsigned long notify_addr;
1532 /* We read from the /dev/lguest device to run the Guest. */
1533 readval = read(lguest_fd, ¬ify_addr, sizeof(notify_addr));
1535 /* One unsigned long means the Guest did HCALL_NOTIFY */
1536 if (readval == sizeof(notify_addr)) {
1537 verbose("Notify on address %#lx\n", notify_addr);
1538 handle_output(lguest_fd, notify_addr);
1540 /* ENOENT means the Guest died. Reading tells us why. */
1541 } else if (errno == ENOENT) {
1542 char reason[1024] = { 0 };
1543 read(lguest_fd, reason, sizeof(reason)-1);
1544 errx(1, "%s", reason);
1545 /* ERESTART means that we need to reboot the guest */
1546 } else if (errno == ERESTART) {
1548 /* EAGAIN means the Waker wanted us to look at some input.
1549 * Anything else means a bug or incompatible change. */
1550 } else if (errno != EAGAIN)
1551 err(1, "Running guest failed");
1553 /* Service input, then unset the BREAK to release the Waker. */
1554 handle_input(lguest_fd);
1555 if (write(lguest_fd, args, sizeof(args)) < 0)
1556 err(1, "Resetting break");
1560 * This is the end of the Launcher. The good news: we are over halfway
1561 * through! The bad news: the most fiendish part of the code still lies ahead
1564 * Are you ready? Take a deep breath and join me in the core of the Host, in
1568 static struct option opts[] = {
1569 { "verbose", 0, NULL, 'v' },
1570 { "tunnet", 1, NULL, 't' },
1571 { "block", 1, NULL, 'b' },
1572 { "initrd", 1, NULL, 'i' },
1575 static void usage(void)
1577 errx(1, "Usage: lguest [--verbose] "
1578 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1579 "|--block=<filename>|--initrd=<filename>]...\n"
1580 "<mem-in-mb> vmlinux [args...]");
1583 /*L:105 The main routine is where the real work begins: */
1584 int main(int argc, char *argv[])
1586 /* Memory, top-level pagetable, code startpoint and size of the
1587 * (optional) initrd. */
1588 unsigned long mem = 0, pgdir, start, initrd_size = 0;
1589 /* Two temporaries and the /dev/lguest file descriptor. */
1590 int i, c, lguest_fd;
1591 /* The boot information for the Guest. */
1592 struct boot_params *boot;
1593 /* If they specify an initrd file to load. */
1594 const char *initrd_name = NULL;
1596 /* Save the args: we "reboot" by execing ourselves again. */
1598 /* We don't "wait" for the children, so prevent them from becoming
1600 signal(SIGCHLD, SIG_IGN);
1602 /* First we initialize the device list. Since console and network
1603 * device receive input from a file descriptor, we keep an fdset
1604 * (infds) and the maximum fd number (max_infd) with the head of the
1605 * list. We also keep a pointer to the last device, for easy appending
1606 * to the list. Finally, we keep the next interrupt number to hand out
1607 * (1: remember that 0 is used by the timer). */
1608 FD_ZERO(&devices.infds);
1609 devices.max_infd = -1;
1610 devices.lastdev = &devices.dev;
1611 devices.next_irq = 1;
1613 /* We need to know how much memory so we can set up the device
1614 * descriptor and memory pages for the devices as we parse the command
1615 * line. So we quickly look through the arguments to find the amount
1617 for (i = 1; i < argc; i++) {
1618 if (argv[i][0] != '-') {
1619 mem = atoi(argv[i]) * 1024 * 1024;
1620 /* We start by mapping anonymous pages over all of
1621 * guest-physical memory range. This fills it with 0,
1622 * and ensures that the Guest won't be killed when it
1623 * tries to access it. */
1624 guest_base = map_zeroed_pages(mem / getpagesize()
1627 guest_max = mem + DEVICE_PAGES*getpagesize();
1628 devices.descpage = get_pages(1);
1633 /* The options are fairly straight-forward */
1634 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1640 setup_tun_net(optarg);
1643 setup_block_file(optarg);
1646 initrd_name = optarg;
1649 warnx("Unknown argument %s", argv[optind]);
1653 /* After the other arguments we expect memory and kernel image name,
1654 * followed by command line arguments for the kernel. */
1655 if (optind + 2 > argc)
1658 verbose("Guest base is at %p\n", guest_base);
1660 /* We always have a console device */
1663 /* Now we load the kernel */
1664 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1666 /* Boot information is stashed at physical address 0 */
1667 boot = from_guest_phys(0);
1669 /* Map the initrd image if requested (at top of physical memory) */
1671 initrd_size = load_initrd(initrd_name, mem);
1672 /* These are the location in the Linux boot header where the
1673 * start and size of the initrd are expected to be found. */
1674 boot->hdr.ramdisk_image = mem - initrd_size;
1675 boot->hdr.ramdisk_size = initrd_size;
1676 /* The bootloader type 0xFF means "unknown"; that's OK. */
1677 boot->hdr.type_of_loader = 0xFF;
1680 /* Set up the initial linear pagetables, starting below the initrd. */
1681 pgdir = setup_pagetables(mem, initrd_size);
1683 /* The Linux boot header contains an "E820" memory map: ours is a
1684 * simple, single region. */
1685 boot->e820_entries = 1;
1686 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1687 /* The boot header contains a command line pointer: we put the command
1688 * line after the boot header. */
1689 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1690 /* We use a simple helper to copy the arguments separated by spaces. */
1691 concat((char *)(boot + 1), argv+optind+2);
1693 /* Boot protocol version: 2.07 supports the fields for lguest. */
1694 boot->hdr.version = 0x207;
1696 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1697 boot->hdr.hardware_subarch = 1;
1699 /* Tell the entry path not to try to reload segment registers. */
1700 boot->hdr.loadflags |= KEEP_SEGMENTS;
1702 /* We tell the kernel to initialize the Guest: this returns the open
1703 * /dev/lguest file descriptor. */
1704 lguest_fd = tell_kernel(pgdir, start);
1706 /* We fork off a child process, which wakes the Launcher whenever one
1707 * of the input file descriptors needs attention. Otherwise we would
1708 * run the Guest until it tries to output something. */
1709 waker_fd = setup_waker(lguest_fd);
1711 /* Finally, run the Guest. This doesn't return. */
1712 run_guest(lguest_fd);
1717 * Mastery is done: you now know everything I do.
1719 * But surely you have seen code, features and bugs in your wanderings which
1720 * you now yearn to attack? That is the real game, and I look forward to you
1721 * patching and forking lguest into the Your-Name-Here-visor.
1723 * Farewell, and good coding!