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 /*L:110 We can ignore the 30 include files we need for this program, but I do
38 * want to draw attention to the use of kernel-style types.
40 * As Linus said, "C is a Spartan language, and so should your naming be." I
41 * like these abbreviations and the header we need uses them, so we define them
44 typedef unsigned long long u64;
48 #include "linux/lguest_launcher.h"
49 #include "linux/pci_ids.h"
50 #include "linux/virtio_config.h"
51 #include "linux/virtio_net.h"
52 #include "linux/virtio_blk.h"
53 #include "linux/virtio_console.h"
54 #include "linux/virtio_ring.h"
55 #include "asm-x86/e820.h"
58 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
60 #define BRIDGE_PFX "bridge:"
62 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
64 /* We can have up to 256 pages for devices. */
65 #define DEVICE_PAGES 256
66 /* This fits nicely in a single 4096-byte page. */
67 #define VIRTQUEUE_NUM 127
69 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
70 * this, and although I wouldn't recommend it, it works quite nicely here. */
72 #define verbose(args...) \
73 do { if (verbose) printf(args); } while(0)
76 /* The pipe to send commands to the waker process */
78 /* The pointer to the start of guest memory. */
79 static void *guest_base;
80 /* The maximum guest physical address allowed, and maximum possible. */
81 static unsigned long guest_limit, guest_max;
83 /* This is our list of devices. */
86 /* Summary information about the devices in our list: ready to pass to
87 * select() to ask which need servicing.*/
91 /* Counter to assign interrupt numbers. */
92 unsigned int next_irq;
94 /* Counter to print out convenient device numbers. */
95 unsigned int device_num;
97 /* The descriptor page for the devices. */
100 /* The tail of the last descriptor. */
101 unsigned int desc_used;
103 /* A single linked list of devices. */
105 /* ... And an end pointer so we can easily append new devices */
106 struct device **lastdev;
109 /* The list of Guest devices, based on command line arguments. */
110 static struct device_list devices;
112 /* The device structure describes a single device. */
115 /* The linked-list pointer. */
118 /* The this device's descriptor, as mapped into the Guest. */
119 struct lguest_device_desc *desc;
121 /* The name of this device, for --verbose. */
124 /* If handle_input is set, it wants to be called when this file
125 * descriptor is ready. */
127 bool (*handle_input)(int fd, struct device *me);
129 /* Any queues attached to this device */
130 struct virtqueue *vq;
132 /* Device-specific data. */
136 /* The virtqueue structure describes a queue attached to a device. */
139 struct virtqueue *next;
141 /* Which device owns me. */
144 /* The configuration for this queue. */
145 struct lguest_vqconfig config;
147 /* The actual ring of buffers. */
150 /* Last available index we saw. */
153 /* The routine to call when the Guest pings us. */
154 void (*handle_output)(int fd, struct virtqueue *me);
157 /* Since guest is UP and we don't run at the same time, we don't need barriers.
158 * But I include them in the code in case others copy it. */
161 /* Convert an iovec element to the given type.
163 * This is a fairly ugly trick: we need to know the size of the type and
164 * alignment requirement to check the pointer is kosher. It's also nice to
165 * have the name of the type in case we report failure.
167 * Typing those three things all the time is cumbersome and error prone, so we
168 * have a macro which sets them all up and passes to the real function. */
169 #define convert(iov, type) \
170 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
172 static void *_convert(struct iovec *iov, size_t size, size_t align,
175 if (iov->iov_len != size)
176 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
177 if ((unsigned long)iov->iov_base % align != 0)
178 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
179 return iov->iov_base;
182 /* The virtio configuration space is defined to be little-endian. x86 is
183 * little-endian too, but it's nice to be explicit so we have these helpers. */
184 #define cpu_to_le16(v16) (v16)
185 #define cpu_to_le32(v32) (v32)
186 #define cpu_to_le64(v64) (v64)
187 #define le16_to_cpu(v16) (v16)
188 #define le32_to_cpu(v32) (v32)
189 #define le64_to_cpu(v32) (v64)
191 /*L:100 The Launcher code itself takes us out into userspace, that scary place
192 * where pointers run wild and free! Unfortunately, like most userspace
193 * programs, it's quite boring (which is why everyone likes to hack on the
194 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
195 * will get you through this section. Or, maybe not.
197 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
198 * memory and stores it in "guest_base". In other words, Guest physical ==
199 * Launcher virtual with an offset.
201 * This can be tough to get your head around, but usually it just means that we
202 * use these trivial conversion functions when the Guest gives us it's
203 * "physical" addresses: */
204 static void *from_guest_phys(unsigned long addr)
206 return guest_base + addr;
209 static unsigned long to_guest_phys(const void *addr)
211 return (addr - guest_base);
215 * Loading the Kernel.
217 * We start with couple of simple helper routines. open_or_die() avoids
218 * error-checking code cluttering the callers: */
219 static int open_or_die(const char *name, int flags)
221 int fd = open(name, flags);
223 err(1, "Failed to open %s", name);
227 /* map_zeroed_pages() takes a number of pages. */
228 static void *map_zeroed_pages(unsigned int num)
230 int fd = open_or_die("/dev/zero", O_RDONLY);
233 /* We use a private mapping (ie. if we write to the page, it will be
235 addr = mmap(NULL, getpagesize() * num,
236 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
237 if (addr == MAP_FAILED)
238 err(1, "Mmaping %u pages of /dev/zero", num);
243 /* Get some more pages for a device. */
244 static void *get_pages(unsigned int num)
246 void *addr = from_guest_phys(guest_limit);
248 guest_limit += num * getpagesize();
249 if (guest_limit > guest_max)
250 errx(1, "Not enough memory for devices");
254 /* To find out where to start we look for the magic Guest string, which marks
255 * the code we see in lguest_asm.S. This is a hack which we are currently
256 * plotting to replace with the normal Linux entry point. */
257 static unsigned long entry_point(const void *start, const void *end)
261 /* The scan gives us the physical starting address. We boot with
262 * pagetables set up with virtual and physical the same, so that's
264 for (p = start; p < end; p++)
265 if (memcmp(p, "GenuineLguest", strlen("GenuineLguest")) == 0)
266 return to_guest_phys(p + strlen("GenuineLguest"));
268 errx(1, "Is this image a genuine lguest?");
271 /* This routine is used to load the kernel or initrd. It tries mmap, but if
272 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
273 * it falls back to reading the memory in. */
274 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
278 /* We map writable even though for some segments are marked read-only.
279 * The kernel really wants to be writable: it patches its own
282 * MAP_PRIVATE means that the page won't be copied until a write is
283 * done to it. This allows us to share untouched memory between
285 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
286 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
289 /* pread does a seek and a read in one shot: saves a few lines. */
290 r = pread(fd, addr, len, offset);
292 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
295 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
296 * the Guest memory. ELF = Embedded Linking Format, which is the format used
297 * by all modern binaries on Linux including the kernel.
299 * The ELF headers give *two* addresses: a physical address, and a virtual
300 * address. We use the physical address; the Guest will map itself to the
303 * We return the starting address. */
304 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
306 void *start = (void *)-1, *end = NULL;
307 Elf32_Phdr phdr[ehdr->e_phnum];
310 /* Sanity checks on the main ELF header: an x86 executable with a
311 * reasonable number of correctly-sized program headers. */
312 if (ehdr->e_type != ET_EXEC
313 || ehdr->e_machine != EM_386
314 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
315 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
316 errx(1, "Malformed elf header");
318 /* An ELF executable contains an ELF header and a number of "program"
319 * headers which indicate which parts ("segments") of the program to
322 /* We read in all the program headers at once: */
323 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
324 err(1, "Seeking to program headers");
325 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
326 err(1, "Reading program headers");
328 /* Try all the headers: there are usually only three. A read-only one,
329 * a read-write one, and a "note" section which isn't loadable. */
330 for (i = 0; i < ehdr->e_phnum; i++) {
331 /* If this isn't a loadable segment, we ignore it */
332 if (phdr[i].p_type != PT_LOAD)
335 verbose("Section %i: size %i addr %p\n",
336 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
338 /* We track the first and last address we mapped, so we can
339 * tell entry_point() where to scan. */
340 if (from_guest_phys(phdr[i].p_paddr) < start)
341 start = from_guest_phys(phdr[i].p_paddr);
342 if (from_guest_phys(phdr[i].p_paddr) + phdr[i].p_filesz > end)
343 end=from_guest_phys(phdr[i].p_paddr)+phdr[i].p_filesz;
345 /* We map this section of the file at its physical address. */
346 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
347 phdr[i].p_offset, phdr[i].p_filesz);
350 return entry_point(start, end);
353 /*L:160 Unfortunately the entire ELF image isn't compressed: the segments
354 * which need loading are extracted and compressed raw. This denies us the
355 * information we need to make a fully-general loader. */
356 static unsigned long unpack_bzimage(int fd)
360 /* A bzImage always gets loaded at physical address 1M. This is
361 * actually configurable as CONFIG_PHYSICAL_START, but as the comment
362 * there says, "Don't change this unless you know what you are doing".
364 void *img = from_guest_phys(0x100000);
366 /* gzdopen takes our file descriptor (carefully placed at the start of
367 * the GZIP header we found) and returns a gzFile. */
368 f = gzdopen(fd, "rb");
369 /* We read it into memory in 64k chunks until we hit the end. */
370 while ((ret = gzread(f, img + len, 65536)) > 0)
373 err(1, "reading image from bzImage");
375 verbose("Unpacked size %i addr %p\n", len, img);
377 return entry_point(img, img + len);
380 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
381 * supposed to jump into it and it will unpack itself. We can't do that
382 * because the Guest can't run the unpacking code, and adding features to
383 * lguest kills puppies, so we don't want to.
385 * The bzImage is formed by putting the decompressing code in front of the
386 * compressed kernel code. So we can simple scan through it looking for the
387 * first "gzip" header, and start decompressing from there. */
388 static unsigned long load_bzimage(int fd)
393 /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */
394 while (read(fd, &c, 1) == 1) {
410 /* Seek back to the start of the gzip header. */
411 lseek(fd, -10, SEEK_CUR);
412 /* One final check: "compressed under UNIX". */
416 return unpack_bzimage(fd);
419 errx(1, "Could not find kernel in bzImage");
422 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
423 * come wrapped up in the self-decompressing "bzImage" format. With some funky
424 * coding, we can load those, too. */
425 static unsigned long load_kernel(int fd)
429 /* Read in the first few bytes. */
430 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
431 err(1, "Reading kernel");
433 /* If it's an ELF file, it starts with "\177ELF" */
434 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
435 return map_elf(fd, &hdr);
437 /* Otherwise we assume it's a bzImage, and try to unpack it */
438 return load_bzimage(fd);
441 /* This is a trivial little helper to align pages. Andi Kleen hated it because
442 * it calls getpagesize() twice: "it's dumb code."
444 * Kernel guys get really het up about optimization, even when it's not
445 * necessary. I leave this code as a reaction against that. */
446 static inline unsigned long page_align(unsigned long addr)
448 /* Add upwards and truncate downwards. */
449 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
452 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
453 * the kernel which the kernel can use to boot from without needing any
454 * drivers. Most distributions now use this as standard: the initrd contains
455 * the code to load the appropriate driver modules for the current machine.
457 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
458 * kernels. He sent me this (and tells me when I break it). */
459 static unsigned long load_initrd(const char *name, unsigned long mem)
465 ifd = open_or_die(name, O_RDONLY);
466 /* fstat() is needed to get the file size. */
467 if (fstat(ifd, &st) < 0)
468 err(1, "fstat() on initrd '%s'", name);
470 /* We map the initrd at the top of memory, but mmap wants it to be
471 * page-aligned, so we round the size up for that. */
472 len = page_align(st.st_size);
473 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
474 /* Once a file is mapped, you can close the file descriptor. It's a
475 * little odd, but quite useful. */
477 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
479 /* We return the initrd size. */
483 /* Once we know how much memory we have, we can construct simple linear page
484 * tables which set virtual == physical which will get the Guest far enough
485 * into the boot to create its own.
487 * We lay them out of the way, just below the initrd (which is why we need to
489 static unsigned long setup_pagetables(unsigned long mem,
490 unsigned long initrd_size)
492 unsigned long *pgdir, *linear;
493 unsigned int mapped_pages, i, linear_pages;
494 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
496 mapped_pages = mem/getpagesize();
498 /* Each PTE page can map ptes_per_page pages: how many do we need? */
499 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
501 /* We put the toplevel page directory page at the top of memory. */
502 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
504 /* Now we use the next linear_pages pages as pte pages */
505 linear = (void *)pgdir - linear_pages*getpagesize();
507 /* Linear mapping is easy: put every page's address into the mapping in
508 * order. PAGE_PRESENT contains the flags Present, Writable and
510 for (i = 0; i < mapped_pages; i++)
511 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
513 /* The top level points to the linear page table pages above. */
514 for (i = 0; i < mapped_pages; i += ptes_per_page) {
515 pgdir[i/ptes_per_page]
516 = ((to_guest_phys(linear) + i*sizeof(void *))
520 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
521 mapped_pages, linear_pages, to_guest_phys(linear));
523 /* We return the top level (guest-physical) address: the kernel needs
524 * to know where it is. */
525 return to_guest_phys(pgdir);
528 /* Simple routine to roll all the commandline arguments together with spaces
530 static void concat(char *dst, char *args[])
532 unsigned int i, len = 0;
534 for (i = 0; args[i]; i++) {
535 strcpy(dst+len, args[i]);
536 strcat(dst+len, " ");
537 len += strlen(args[i]) + 1;
539 /* In case it's empty. */
543 /* This is where we actually tell the kernel to initialize the Guest. We saw
544 * the arguments it expects when we looked at initialize() in lguest_user.c:
545 * the base of guest "physical" memory, the top physical page to allow, the
546 * top level pagetable and the entry point for the Guest. */
547 static int tell_kernel(unsigned long pgdir, unsigned long start)
549 unsigned long args[] = { LHREQ_INITIALIZE,
550 (unsigned long)guest_base,
551 guest_limit / getpagesize(), pgdir, start };
554 verbose("Guest: %p - %p (%#lx)\n",
555 guest_base, guest_base + guest_limit, guest_limit);
556 fd = open_or_die("/dev/lguest", O_RDWR);
557 if (write(fd, args, sizeof(args)) < 0)
558 err(1, "Writing to /dev/lguest");
560 /* We return the /dev/lguest file descriptor to control this Guest */
565 static void add_device_fd(int fd)
567 FD_SET(fd, &devices.infds);
568 if (fd > devices.max_infd)
569 devices.max_infd = fd;
575 * With a console and network devices, we can have lots of input which we need
576 * to process. We could try to tell the kernel what file descriptors to watch,
577 * but handing a file descriptor mask through to the kernel is fairly icky.
579 * Instead, we fork off a process which watches the file descriptors and writes
580 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
581 * loop to stop running the Guest. This causes it to return from the
582 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
583 * the LHREQ_BREAK and wake us up again.
585 * This, of course, is merely a different *kind* of icky.
587 static void wake_parent(int pipefd, int lguest_fd)
589 /* Add the pipe from the Launcher to the fdset in the device_list, so
590 * we watch it, too. */
591 add_device_fd(pipefd);
594 fd_set rfds = devices.infds;
595 unsigned long args[] = { LHREQ_BREAK, 1 };
597 /* Wait until input is ready from one of the devices. */
598 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
599 /* Is it a message from the Launcher? */
600 if (FD_ISSET(pipefd, &rfds)) {
602 /* If read() returns 0, it means the Launcher has
603 * exited. We silently follow. */
604 if (read(pipefd, &ignorefd, sizeof(ignorefd)) == 0)
606 /* Otherwise it's telling us there's a problem with one
607 * of the devices, and we should ignore that file
608 * descriptor from now on. */
609 FD_CLR(ignorefd, &devices.infds);
610 } else /* Send LHREQ_BREAK command. */
611 write(lguest_fd, args, sizeof(args));
615 /* This routine just sets up a pipe to the Waker process. */
616 static int setup_waker(int lguest_fd)
618 int pipefd[2], child;
620 /* We create a pipe to talk to the waker, and also so it knows when the
621 * Launcher dies (and closes pipe). */
628 /* Close the "writing" end of our copy of the pipe */
630 wake_parent(pipefd[0], lguest_fd);
632 /* Close the reading end of our copy of the pipe. */
635 /* Here is the fd used to talk to the waker. */
642 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
643 * We need to make sure it's not trying to reach into the Launcher itself, so
644 * we have a convenient routine which check it and exits with an error message
645 * if something funny is going on:
647 static void *_check_pointer(unsigned long addr, unsigned int size,
650 /* We have to separately check addr and addr+size, because size could
651 * be huge and addr + size might wrap around. */
652 if (addr >= guest_limit || addr + size >= guest_limit)
653 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
654 /* We return a pointer for the caller's convenience, now we know it's
656 return from_guest_phys(addr);
658 /* A macro which transparently hands the line number to the real function. */
659 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
661 /* This simply sets up an iovec array where we can put data to be discarded.
662 * This happens when the Guest doesn't want or can't handle the input: we have
663 * to get rid of it somewhere, and if we bury it in the ceiling space it will
664 * start to smell after a week. */
665 static void discard_iovec(struct iovec *iov, unsigned int *num)
667 static char discard_buf[1024];
669 iov->iov_base = discard_buf;
670 iov->iov_len = sizeof(discard_buf);
673 /* This function returns the next descriptor in the chain, or vq->vring.num. */
674 static unsigned next_desc(struct virtqueue *vq, unsigned int i)
678 /* If this descriptor says it doesn't chain, we're done. */
679 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
680 return vq->vring.num;
682 /* Check they're not leading us off end of descriptors. */
683 next = vq->vring.desc[i].next;
684 /* Make sure compiler knows to grab that: we don't want it changing! */
687 if (next >= vq->vring.num)
688 errx(1, "Desc next is %u", next);
693 /* This looks in the virtqueue and for the first available buffer, and converts
694 * it to an iovec for convenient access. Since descriptors consist of some
695 * number of output then some number of input descriptors, it's actually two
696 * iovecs, but we pack them into one and note how many of each there were.
698 * This function returns the descriptor number found, or vq->vring.num (which
699 * is never a valid descriptor number) if none was found. */
700 static unsigned get_vq_desc(struct virtqueue *vq,
702 unsigned int *out_num, unsigned int *in_num)
704 unsigned int i, head;
706 /* Check it isn't doing very strange things with descriptor numbers. */
707 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
708 errx(1, "Guest moved used index from %u to %u",
709 vq->last_avail_idx, vq->vring.avail->idx);
711 /* If there's nothing new since last we looked, return invalid. */
712 if (vq->vring.avail->idx == vq->last_avail_idx)
713 return vq->vring.num;
715 /* Grab the next descriptor number they're advertising, and increment
716 * the index we've seen. */
717 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
719 /* If their number is silly, that's a fatal mistake. */
720 if (head >= vq->vring.num)
721 errx(1, "Guest says index %u is available", head);
723 /* When we start there are none of either input nor output. */
724 *out_num = *in_num = 0;
728 /* Grab the first descriptor, and check it's OK. */
729 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
730 iov[*out_num + *in_num].iov_base
731 = check_pointer(vq->vring.desc[i].addr,
732 vq->vring.desc[i].len);
733 /* If this is an input descriptor, increment that count. */
734 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
737 /* If it's an output descriptor, they're all supposed
738 * to come before any input descriptors. */
740 errx(1, "Descriptor has out after in");
744 /* If we've got too many, that implies a descriptor loop. */
745 if (*out_num + *in_num > vq->vring.num)
746 errx(1, "Looped descriptor");
747 } while ((i = next_desc(vq, i)) != vq->vring.num);
752 /* Once we've used one of their buffers, we tell them about it. We'll then
753 * want to send them an interrupt, using trigger_irq(). */
754 static void add_used(struct virtqueue *vq, unsigned int head, int len)
756 struct vring_used_elem *used;
758 /* Get a pointer to the next entry in the used ring. */
759 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
762 /* Make sure buffer is written before we update index. */
764 vq->vring.used->idx++;
767 /* This actually sends the interrupt for this virtqueue */
768 static void trigger_irq(int fd, struct virtqueue *vq)
770 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
772 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
775 /* Send the Guest an interrupt tell them we used something up. */
776 if (write(fd, buf, sizeof(buf)) != 0)
777 err(1, "Triggering irq %i", vq->config.irq);
780 /* And here's the combo meal deal. Supersize me! */
781 static void add_used_and_trigger(int fd, struct virtqueue *vq,
782 unsigned int head, int len)
784 add_used(vq, head, len);
788 /* Here is the input terminal setting we save, and the routine to restore them
789 * on exit so the user can see what they type next. */
790 static struct termios orig_term;
791 static void restore_term(void)
793 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
796 /* We associate some data with the console for our exit hack. */
799 /* How many times have they hit ^C? */
801 /* When did they start? */
802 struct timeval start;
805 /* This is the routine which handles console input (ie. stdin). */
806 static bool handle_console_input(int fd, struct device *dev)
809 unsigned int head, in_num, out_num;
810 struct iovec iov[dev->vq->vring.num];
811 struct console_abort *abort = dev->priv;
813 /* First we need a console buffer from the Guests's input virtqueue. */
814 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
815 if (head == dev->vq->vring.num) {
816 /* If they're not ready for input, we warn and set up to
818 warnx("console: no dma buffer!");
819 discard_iovec(iov, &in_num);
821 errx(1, "Output buffers in console in queue?");
823 /* This is why we convert to iovecs: the readv() call uses them, and so
824 * it reads straight into the Guest's buffer. */
825 len = readv(dev->fd, iov, in_num);
827 /* This implies that the console is closed, is /dev/null, or
828 * something went terribly wrong. */
829 warnx("Failed to get console input, ignoring console.");
830 /* Put the input terminal back and return failure (meaning,
831 * don't call us again). */
836 /* If we actually read the data into the Guest, tell them about it. */
837 if (head != dev->vq->vring.num)
838 add_used_and_trigger(fd, dev->vq, head, len);
840 /* Three ^C within one second? Exit.
842 * This is such a hack, but works surprisingly well. Each ^C has to be
843 * in a buffer by itself, so they can't be too fast. But we check that
844 * we get three within about a second, so they can't be too slow. */
845 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
847 gettimeofday(&abort->start, NULL);
848 else if (abort->count == 3) {
850 gettimeofday(&now, NULL);
851 if (now.tv_sec <= abort->start.tv_sec+1) {
852 unsigned long args[] = { LHREQ_BREAK, 0 };
853 /* Close the fd so Waker will know it has to
856 /* Just in case waker is blocked in BREAK, send
858 write(fd, args, sizeof(args));
864 /* Any other key resets the abort counter. */
867 /* Everything went OK! */
871 /* Handling output for console is simple: we just get all the output buffers
872 * and write them to stdout. */
873 static void handle_console_output(int fd, struct virtqueue *vq)
875 unsigned int head, out, in;
877 struct iovec iov[vq->vring.num];
879 /* Keep getting output buffers from the Guest until we run out. */
880 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
882 errx(1, "Input buffers in output queue?");
883 len = writev(STDOUT_FILENO, iov, out);
884 add_used_and_trigger(fd, vq, head, len);
888 /* Handling output for network is also simple: we get all the output buffers
889 * and write them (ignoring the first element) to this device's file descriptor
891 static void handle_net_output(int fd, struct virtqueue *vq)
893 unsigned int head, out, in;
895 struct iovec iov[vq->vring.num];
897 /* Keep getting output buffers from the Guest until we run out. */
898 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
900 errx(1, "Input buffers in output queue?");
901 /* Check header, but otherwise ignore it (we said we supported
903 (void)convert(&iov[0], struct virtio_net_hdr);
904 len = writev(vq->dev->fd, iov+1, out-1);
905 add_used_and_trigger(fd, vq, head, len);
909 /* This is where we handle a packet coming in from the tun device to our
911 static bool handle_tun_input(int fd, struct device *dev)
913 unsigned int head, in_num, out_num;
915 struct iovec iov[dev->vq->vring.num];
916 struct virtio_net_hdr *hdr;
918 /* First we need a network buffer from the Guests's recv virtqueue. */
919 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
920 if (head == dev->vq->vring.num) {
921 /* Now, it's expected that if we try to send a packet too
922 * early, the Guest won't be ready yet. Wait until the device
923 * status says it's ready. */
924 /* FIXME: Actually want DRIVER_ACTIVE here. */
925 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
926 warn("network: no dma buffer!");
927 discard_iovec(iov, &in_num);
929 errx(1, "Output buffers in network recv queue?");
931 /* First element is the header: we set it to 0 (no features). */
932 hdr = convert(&iov[0], struct virtio_net_hdr);
934 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
936 /* Read the packet from the device directly into the Guest's buffer. */
937 len = readv(dev->fd, iov+1, in_num-1);
939 err(1, "reading network");
941 /* If we actually read the data into the Guest, tell them about it. */
942 if (head != dev->vq->vring.num)
943 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
945 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
946 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
947 head != dev->vq->vring.num ? "sent" : "discarded");
953 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
954 static void handle_output(int fd, unsigned long addr)
957 struct virtqueue *vq;
959 /* Check each virtqueue. */
960 for (i = devices.dev; i; i = i->next) {
961 for (vq = i->vq; vq; vq = vq->next) {
962 if (vq->config.pfn == addr/getpagesize()
963 && vq->handle_output) {
964 verbose("Output to %s\n", vq->dev->name);
965 vq->handle_output(fd, vq);
971 /* Early console write is done using notify on a nul-terminated string
972 * in Guest memory. */
973 if (addr >= guest_limit)
974 errx(1, "Bad NOTIFY %#lx", addr);
976 write(STDOUT_FILENO, from_guest_phys(addr),
977 strnlen(from_guest_phys(addr), guest_limit - addr));
980 /* This is called when the waker wakes us up: check for incoming file
982 static void handle_input(int fd)
984 /* select() wants a zeroed timeval to mean "don't wait". */
985 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
989 fd_set fds = devices.infds;
991 /* If nothing is ready, we're done. */
992 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
995 /* Otherwise, call the device(s) which have readable
996 * file descriptors and a method of handling them. */
997 for (i = devices.dev; i; i = i->next) {
998 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
999 /* If handle_input() returns false, it means we
1000 * should no longer service it.
1001 * handle_console_input() does this. */
1002 if (!i->handle_input(fd, i)) {
1003 /* Clear it from the set of input file
1004 * descriptors kept at the head of the
1006 FD_CLR(i->fd, &devices.infds);
1007 /* Tell waker to ignore it too... */
1008 write(waker_fd, &i->fd, sizeof(i->fd));
1018 * All devices need a descriptor so the Guest knows it exists, and a "struct
1019 * device" so the Launcher can keep track of it. We have common helper
1020 * routines to allocate them.
1022 * This routine allocates a new "struct lguest_device_desc" from descriptor
1023 * table just above the Guest's normal memory. It returns a pointer to that
1025 static struct lguest_device_desc *new_dev_desc(u16 type)
1027 struct lguest_device_desc *d;
1029 /* We only have one page for all the descriptors. */
1030 if (devices.desc_used + sizeof(*d) > getpagesize())
1031 errx(1, "Too many devices");
1033 /* We don't need to set config_len or status: page is 0 already. */
1034 d = (void *)devices.descpage + devices.desc_used;
1036 devices.desc_used += sizeof(*d);
1041 /* Each device descriptor is followed by some configuration information.
1042 * The first byte is a "status" byte for the Guest to report what's happening.
1043 * After that are fields: u8 type, u8 len, [... len bytes...].
1045 * This routine adds a new field to an existing device's descriptor. It only
1046 * works for the last device, but that's OK because that's how we use it. */
1047 static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
1049 /* This is the last descriptor, right? */
1050 assert(devices.descpage + devices.desc_used
1051 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
1053 /* We only have one page of device descriptions. */
1054 if (devices.desc_used + 2 + len > getpagesize())
1055 errx(1, "Too many devices");
1057 /* Copy in the new config header: type then length. */
1058 devices.descpage[devices.desc_used++] = type;
1059 devices.descpage[devices.desc_used++] = len;
1060 memcpy(devices.descpage + devices.desc_used, c, len);
1061 devices.desc_used += len;
1063 /* Update the device descriptor length: two byte head then data. */
1064 dev->desc->config_len += 2 + len;
1067 /* This routine adds a virtqueue to a device. We specify how many descriptors
1068 * the virtqueue is to have. */
1069 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1070 void (*handle_output)(int fd, struct virtqueue *me))
1073 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1076 /* First we need some pages for this virtqueue. */
1077 pages = (vring_size(num_descs) + getpagesize() - 1) / getpagesize();
1078 p = get_pages(pages);
1080 /* Initialize the configuration. */
1081 vq->config.num = num_descs;
1082 vq->config.irq = devices.next_irq++;
1083 vq->config.pfn = to_guest_phys(p) / getpagesize();
1085 /* Initialize the vring. */
1086 vring_init(&vq->vring, num_descs, p);
1088 /* Add the configuration information to this device's descriptor. */
1089 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE,
1090 sizeof(vq->config), &vq->config);
1092 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1094 for (i = &dev->vq; *i; i = &(*i)->next);
1097 /* Link virtqueue back to device. */
1100 /* Set up handler. */
1101 vq->handle_output = handle_output;
1103 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1106 /* This routine does all the creation and setup of a new device, including
1107 * caling new_dev_desc() to allocate the descriptor and device memory. */
1108 static struct device *new_device(const char *name, u16 type, int fd,
1109 bool (*handle_input)(int, struct device *))
1111 struct device *dev = malloc(sizeof(*dev));
1113 /* Append to device list. Prepending to a single-linked list is
1114 * easier, but the user expects the devices to be arranged on the bus
1115 * in command-line order. The first network device on the command line
1116 * is eth0, the first block device /dev/lgba, etc. */
1117 *devices.lastdev = dev;
1119 devices.lastdev = &dev->next;
1121 /* Now we populate the fields one at a time. */
1123 /* If we have an input handler for this file descriptor, then we add it
1124 * to the device_list's fdset and maxfd. */
1126 add_device_fd(dev->fd);
1127 dev->desc = new_dev_desc(type);
1128 dev->handle_input = handle_input;
1133 /* Our first setup routine is the console. It's a fairly simple device, but
1134 * UNIX tty handling makes it uglier than it could be. */
1135 static void setup_console(void)
1139 /* If we can save the initial standard input settings... */
1140 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1141 struct termios term = orig_term;
1142 /* Then we turn off echo, line buffering and ^C etc. We want a
1143 * raw input stream to the Guest. */
1144 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1145 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1146 /* If we exit gracefully, the original settings will be
1147 * restored so the user can see what they're typing. */
1148 atexit(restore_term);
1151 dev = new_device("console", VIRTIO_ID_CONSOLE,
1152 STDIN_FILENO, handle_console_input);
1153 /* We store the console state in dev->priv, and initialize it. */
1154 dev->priv = malloc(sizeof(struct console_abort));
1155 ((struct console_abort *)dev->priv)->count = 0;
1157 /* The console needs two virtqueues: the input then the output. We
1158 * don't care when they refill the input queue, since we don't hold
1159 * data waiting for them. That's why the input queue's callback is
1161 add_virtqueue(dev, VIRTQUEUE_NUM, NULL);
1162 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1164 verbose("device %u: console\n", devices.device_num++);
1168 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1169 * --sharenet=<name> option which opens or creates a named pipe. This can be
1170 * used to send packets to another guest in a 1:1 manner.
1172 * More sopisticated is to use one of the tools developed for project like UML
1175 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1176 * completely generic ("here's my vring, attach to your vring") and would work
1177 * for any traffic. Of course, namespace and permissions issues need to be
1178 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1179 * multiple inter-guest channels behind one interface, although it would
1180 * require some manner of hotplugging new virtio channels.
1182 * Finally, we could implement a virtio network switch in the kernel. :*/
1184 static u32 str2ip(const char *ipaddr)
1186 unsigned int byte[4];
1188 sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]);
1189 return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3];
1192 /* This code is "adapted" from libbridge: it attaches the Host end of the
1193 * network device to the bridge device specified by the command line.
1195 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1196 * dislike bridging), and I just try not to break it. */
1197 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1203 errx(1, "must specify bridge name");
1205 ifidx = if_nametoindex(if_name);
1207 errx(1, "interface %s does not exist!", if_name);
1209 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1210 ifr.ifr_ifindex = ifidx;
1211 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1212 err(1, "can't add %s to bridge %s", if_name, br_name);
1215 /* This sets up the Host end of the network device with an IP address, brings
1216 * it up so packets will flow, the copies the MAC address into the hwaddr
1218 static void configure_device(int fd, const char *devname, u32 ipaddr,
1219 unsigned char hwaddr[6])
1222 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1224 /* Don't read these incantations. Just cut & paste them like I did! */
1225 memset(&ifr, 0, sizeof(ifr));
1226 strcpy(ifr.ifr_name, devname);
1227 sin->sin_family = AF_INET;
1228 sin->sin_addr.s_addr = htonl(ipaddr);
1229 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1230 err(1, "Setting %s interface address", devname);
1231 ifr.ifr_flags = IFF_UP;
1232 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1233 err(1, "Bringing interface %s up", devname);
1235 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1236 * above). IF means Interface, and HWADDR is hardware address.
1238 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1239 err(1, "getting hw address for %s", devname);
1240 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1243 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1244 * routing, but the principle is the same: it uses the "tun" device to inject
1245 * packets into the Host as if they came in from a normal network card. We
1246 * just shunt packets between the Guest and the tun device. */
1247 static void setup_tun_net(const char *arg)
1253 const char *br_name = NULL;
1256 /* We open the /dev/net/tun device and tell it we want a tap device. A
1257 * tap device is like a tun device, only somehow different. To tell
1258 * the truth, I completely blundered my way through this code, but it
1260 netfd = open_or_die("/dev/net/tun", O_RDWR);
1261 memset(&ifr, 0, sizeof(ifr));
1262 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1263 strcpy(ifr.ifr_name, "tap%d");
1264 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1265 err(1, "configuring /dev/net/tun");
1266 /* We don't need checksums calculated for packets coming in this
1267 * device: trust us! */
1268 ioctl(netfd, TUNSETNOCSUM, 1);
1270 /* First we create a new network device. */
1271 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1273 /* Network devices need a receive and a send queue. */
1274 add_virtqueue(dev, VIRTQUEUE_NUM, NULL);
1275 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1277 /* We need a socket to perform the magic network ioctls to bring up the
1278 * tap interface, connect to the bridge etc. Any socket will do! */
1279 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1281 err(1, "opening IP socket");
1283 /* If the command line was --tunnet=bridge:<name> do bridging. */
1284 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1286 br_name = arg + strlen(BRIDGE_PFX);
1287 add_to_bridge(ipfd, ifr.ifr_name, br_name);
1288 } else /* It is an IP address to set up the device with */
1291 /* Set up the tun device, and get the mac address for the interface. */
1292 configure_device(ipfd, ifr.ifr_name, ip, hwaddr);
1294 /* Tell Guest what MAC address to use. */
1295 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr);
1297 /* We don't seed the socket any more; setup is done. */
1300 verbose("device %u: tun net %u.%u.%u.%u\n",
1301 devices.device_num++,
1302 (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip);
1304 verbose("attached to bridge: %s\n", br_name);
1311 * Serving a block device is really easy: the Guest asks for a block number and
1312 * we read or write that position in the file.
1314 * Unfortunately, this is amazingly slow: the Guest waits until the read is
1315 * finished before running anything else, even if it could be doing useful
1316 * work. We could use async I/O, except it's reputed to suck so hard that
1317 * characters actually go missing from your code when you try to use it.
1319 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1321 /* This hangs off device->priv, with the data. */
1324 /* The size of the file. */
1327 /* The file descriptor for the file. */
1330 /* IO thread listens on this file descriptor [0]. */
1333 /* IO thread writes to this file descriptor to mark it done, then
1334 * Launcher triggers interrupt to Guest. */
1338 /* This is the core of the I/O thread. It returns true if it did something. */
1339 static bool service_io(struct device *dev)
1341 struct vblk_info *vblk = dev->priv;
1342 unsigned int head, out_num, in_num, wlen;
1344 struct virtio_blk_inhdr *in;
1345 struct virtio_blk_outhdr *out;
1346 struct iovec iov[dev->vq->vring.num];
1349 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1350 if (head == dev->vq->vring.num)
1353 if (out_num == 0 || in_num == 0)
1354 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1355 head, out_num, in_num);
1357 out = convert(&iov[0], struct virtio_blk_outhdr);
1358 in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr);
1359 off = out->sector * 512;
1361 /* This is how we implement barriers. Pretty poor, no? */
1362 if (out->type & VIRTIO_BLK_T_BARRIER)
1363 fdatasync(vblk->fd);
1365 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1366 fprintf(stderr, "Scsi commands unsupported\n");
1367 in->status = VIRTIO_BLK_S_UNSUPP;
1369 } else if (out->type & VIRTIO_BLK_T_OUT) {
1372 /* Move to the right location in the block file. This can fail
1373 * if they try to write past end. */
1374 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1375 err(1, "Bad seek to sector %llu", out->sector);
1377 ret = writev(vblk->fd, iov+1, out_num-1);
1378 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1380 /* Grr... Now we know how long the descriptor they sent was, we
1381 * make sure they didn't try to write over the end of the block
1382 * file (possibly extending it). */
1383 if (ret > 0 && off + ret > vblk->len) {
1384 /* Trim it back to the correct length */
1385 ftruncate64(vblk->fd, vblk->len);
1386 /* Die, bad Guest, die. */
1387 errx(1, "Write past end %llu+%u", off, ret);
1390 in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1394 /* Move to the right location in the block file. This can fail
1395 * if they try to read past end. */
1396 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1397 err(1, "Bad seek to sector %llu", out->sector);
1399 ret = readv(vblk->fd, iov+1, in_num-1);
1400 verbose("READ from sector %llu: %i\n", out->sector, ret);
1402 wlen = sizeof(in) + ret;
1403 in->status = VIRTIO_BLK_S_OK;
1406 in->status = VIRTIO_BLK_S_IOERR;
1410 /* We can't trigger an IRQ, because we're not the Launcher. It does
1411 * that when we tell it we're done. */
1412 add_used(dev->vq, head, wlen);
1416 /* This is the thread which actually services the I/O. */
1417 static int io_thread(void *_dev)
1419 struct device *dev = _dev;
1420 struct vblk_info *vblk = dev->priv;
1423 /* Close other side of workpipe so we get 0 read when main dies. */
1424 close(vblk->workpipe[1]);
1425 /* Close the other side of the done_fd pipe. */
1428 /* When this read fails, it means Launcher died, so we follow. */
1429 while (read(vblk->workpipe[0], &c, 1) == 1) {
1430 /* We acknowledge each request immediately, to reduce latency,
1431 * rather than waiting until we've done them all. I haven't
1432 * measured to see if it makes any difference. */
1433 while (service_io(dev))
1434 write(vblk->done_fd, &c, 1);
1439 /* When the thread says some I/O is done, we interrupt the Guest. */
1440 static bool handle_io_finish(int fd, struct device *dev)
1444 /* If child died, presumably it printed message. */
1445 if (read(dev->fd, &c, 1) != 1)
1448 /* It did some work, so trigger the irq. */
1449 trigger_irq(fd, dev->vq);
1453 /* When the Guest submits some I/O, we wake the I/O thread. */
1454 static void handle_virtblk_output(int fd, struct virtqueue *vq)
1456 struct vblk_info *vblk = vq->dev->priv;
1459 /* Wake up I/O thread and tell it to go to work! */
1460 if (write(vblk->workpipe[1], &c, 1) != 1)
1461 /* Presumably it indicated why it died. */
1465 /* This creates a virtual block device. */
1466 static void setup_block_file(const char *filename)
1470 struct vblk_info *vblk;
1475 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1478 /* The device responds to return from I/O thread. */
1479 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1481 /* The device has a virtqueue. */
1482 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1484 /* Allocate the room for our own bookkeeping */
1485 vblk = dev->priv = malloc(sizeof(*vblk));
1487 /* First we open the file and store the length. */
1488 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1489 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1491 /* Tell Guest how many sectors this device has. */
1492 cap = cpu_to_le64(vblk->len / 512);
1493 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1495 /* Tell Guest not to put in too many descriptors at once: two are used
1496 * for the in and out elements. */
1497 val = cpu_to_le32(VIRTQUEUE_NUM - 2);
1498 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val);
1500 /* The I/O thread writes to this end of the pipe when done. */
1501 vblk->done_fd = p[1];
1503 /* This is how we tell the I/O thread about more work. */
1504 pipe(vblk->workpipe);
1506 /* Create stack for thread and run it */
1507 stack = malloc(32768);
1508 if (clone(io_thread, stack + 32768, CLONE_VM, dev) == -1)
1509 err(1, "Creating clone");
1511 /* We don't need to keep the I/O thread's end of the pipes open. */
1512 close(vblk->done_fd);
1513 close(vblk->workpipe[0]);
1515 verbose("device %u: virtblock %llu sectors\n",
1516 devices.device_num, cap);
1518 /* That's the end of device setup. */
1520 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1521 * its input and output, and finally, lays it to rest. */
1522 static void __attribute__((noreturn)) run_guest(int lguest_fd)
1525 unsigned long args[] = { LHREQ_BREAK, 0 };
1526 unsigned long notify_addr;
1529 /* We read from the /dev/lguest device to run the Guest. */
1530 readval = read(lguest_fd, ¬ify_addr, sizeof(notify_addr));
1532 /* One unsigned long means the Guest did HCALL_NOTIFY */
1533 if (readval == sizeof(notify_addr)) {
1534 verbose("Notify on address %#lx\n", notify_addr);
1535 handle_output(lguest_fd, notify_addr);
1537 /* ENOENT means the Guest died. Reading tells us why. */
1538 } else if (errno == ENOENT) {
1539 char reason[1024] = { 0 };
1540 read(lguest_fd, reason, sizeof(reason)-1);
1541 errx(1, "%s", reason);
1542 /* EAGAIN means the waker wanted us to look at some input.
1543 * Anything else means a bug or incompatible change. */
1544 } else if (errno != EAGAIN)
1545 err(1, "Running guest failed");
1547 /* Service input, then unset the BREAK which releases
1549 handle_input(lguest_fd);
1550 if (write(lguest_fd, args, sizeof(args)) < 0)
1551 err(1, "Resetting break");
1555 * This is the end of the Launcher.
1557 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1558 * Drivers. If we were to see the Host kernel I/O code, our understanding
1559 * would be complete... :*/
1561 static struct option opts[] = {
1562 { "verbose", 0, NULL, 'v' },
1563 { "tunnet", 1, NULL, 't' },
1564 { "block", 1, NULL, 'b' },
1565 { "initrd", 1, NULL, 'i' },
1568 static void usage(void)
1570 errx(1, "Usage: lguest [--verbose] "
1571 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1572 "|--block=<filename>|--initrd=<filename>]...\n"
1573 "<mem-in-mb> vmlinux [args...]");
1576 /*L:105 The main routine is where the real work begins: */
1577 int main(int argc, char *argv[])
1579 /* Memory, top-level pagetable, code startpoint and size of the
1580 * (optional) initrd. */
1581 unsigned long mem = 0, pgdir, start, initrd_size = 0;
1582 /* A temporary and the /dev/lguest file descriptor. */
1583 int i, c, lguest_fd;
1584 /* The boot information for the Guest. */
1586 /* If they specify an initrd file to load. */
1587 const char *initrd_name = NULL;
1589 /* First we initialize the device list. Since console and network
1590 * device receive input from a file descriptor, we keep an fdset
1591 * (infds) and the maximum fd number (max_infd) with the head of the
1592 * list. We also keep a pointer to the last device, for easy appending
1593 * to the list. Finally, we keep the next interrupt number to hand out
1594 * (1: remember that 0 is used by the timer). */
1595 FD_ZERO(&devices.infds);
1596 devices.max_infd = -1;
1597 devices.lastdev = &devices.dev;
1598 devices.next_irq = 1;
1600 /* We need to know how much memory so we can set up the device
1601 * descriptor and memory pages for the devices as we parse the command
1602 * line. So we quickly look through the arguments to find the amount
1604 for (i = 1; i < argc; i++) {
1605 if (argv[i][0] != '-') {
1606 mem = atoi(argv[i]) * 1024 * 1024;
1607 /* We start by mapping anonymous pages over all of
1608 * guest-physical memory range. This fills it with 0,
1609 * and ensures that the Guest won't be killed when it
1610 * tries to access it. */
1611 guest_base = map_zeroed_pages(mem / getpagesize()
1614 guest_max = mem + DEVICE_PAGES*getpagesize();
1615 devices.descpage = get_pages(1);
1620 /* The options are fairly straight-forward */
1621 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1627 setup_tun_net(optarg);
1630 setup_block_file(optarg);
1633 initrd_name = optarg;
1636 warnx("Unknown argument %s", argv[optind]);
1640 /* After the other arguments we expect memory and kernel image name,
1641 * followed by command line arguments for the kernel. */
1642 if (optind + 2 > argc)
1645 verbose("Guest base is at %p\n", guest_base);
1647 /* We always have a console device */
1650 /* Now we load the kernel */
1651 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1653 /* Boot information is stashed at physical address 0 */
1654 boot = from_guest_phys(0);
1656 /* Map the initrd image if requested (at top of physical memory) */
1658 initrd_size = load_initrd(initrd_name, mem);
1659 /* These are the location in the Linux boot header where the
1660 * start and size of the initrd are expected to be found. */
1661 *(unsigned long *)(boot+0x218) = mem - initrd_size;
1662 *(unsigned long *)(boot+0x21c) = initrd_size;
1663 /* The bootloader type 0xFF means "unknown"; that's OK. */
1664 *(unsigned char *)(boot+0x210) = 0xFF;
1667 /* Set up the initial linear pagetables, starting below the initrd. */
1668 pgdir = setup_pagetables(mem, initrd_size);
1670 /* The Linux boot header contains an "E820" memory map: ours is a
1671 * simple, single region. */
1672 *(char*)(boot+E820NR) = 1;
1673 *((struct e820entry *)(boot+E820MAP))
1674 = ((struct e820entry) { 0, mem, E820_RAM });
1675 /* The boot header contains a command line pointer: we put the command
1676 * line after the boot header (at address 4096) */
1677 *(u32 *)(boot + 0x228) = 4096;
1678 concat(boot + 4096, argv+optind+2);
1680 /* The guest type value of "1" tells the Guest it's under lguest. */
1681 *(int *)(boot + 0x23c) = 1;
1683 /* We tell the kernel to initialize the Guest: this returns the open
1684 * /dev/lguest file descriptor. */
1685 lguest_fd = tell_kernel(pgdir, start);
1687 /* We fork off a child process, which wakes the Launcher whenever one
1688 * of the input file descriptors needs attention. Otherwise we would
1689 * run the Guest until it tries to output something. */
1690 waker_fd = setup_waker(lguest_fd);
1692 /* Finally, run the Guest. This doesn't return. */
1693 run_guest(lguest_fd);
1698 * Mastery is done: you now know everything I do.
1700 * But surely you have seen code, features and bugs in your wanderings which
1701 * you now yearn to attack? That is the real game, and I look forward to you
1702 * patching and forking lguest into the Your-Name-Here-visor.
1704 * Farewell, and good coding!