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
3 * the virtual devices, then opens /dev/lguest to tell the kernel
4 * about the Guest and control it. :*/
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>
39 #include "linux/lguest_launcher.h"
40 #include "linux/virtio_config.h"
41 #include "linux/virtio_net.h"
42 #include "linux/virtio_blk.h"
43 #include "linux/virtio_console.h"
44 #include "linux/virtio_ring.h"
45 #include "asm-x86/bootparam.h"
46 /*L:110 We can ignore the 39 include files we need for this program, but I do
47 * want to draw attention to the use of kernel-style types.
49 * As Linus said, "C is a Spartan language, and so should your naming be." I
50 * like these abbreviations, so we define them here. Note that u64 is always
51 * unsigned long long, which works on all Linux systems: this means that we can
52 * use %llu in printf for any u64. */
53 typedef unsigned long long u64;
59 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
61 #define BRIDGE_PFX "bridge:"
63 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
65 /* We can have up to 256 pages for devices. */
66 #define DEVICE_PAGES 256
67 /* This will occupy 2 pages: it must be a power of 2. */
68 #define VIRTQUEUE_NUM 128
70 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
71 * this, and although I wouldn't recommend it, it works quite nicely here. */
73 #define verbose(args...) \
74 do { if (verbose) printf(args); } while(0)
77 /* The pipe to send commands to the waker process */
79 /* The pointer to the start of guest memory. */
80 static void *guest_base;
81 /* The maximum guest physical address allowed, and maximum possible. */
82 static unsigned long guest_limit, guest_max;
84 /* a per-cpu variable indicating whose vcpu is currently running */
85 static unsigned int __thread cpu_id;
87 /* This is our list of devices. */
90 /* Summary information about the devices in our list: ready to pass to
91 * select() to ask which need servicing.*/
95 /* Counter to assign interrupt numbers. */
96 unsigned int next_irq;
98 /* Counter to print out convenient device numbers. */
99 unsigned int device_num;
101 /* The descriptor page for the devices. */
104 /* A single linked list of devices. */
106 /* And a pointer to the last device for easy append and also for
107 * configuration appending. */
108 struct device *lastdev;
111 /* The list of Guest devices, based on command line arguments. */
112 static struct device_list devices;
114 /* The device structure describes a single device. */
117 /* The linked-list pointer. */
120 /* The this device's descriptor, as mapped into the Guest. */
121 struct lguest_device_desc *desc;
123 /* The name of this device, for --verbose. */
126 /* If handle_input is set, it wants to be called when this file
127 * descriptor is ready. */
129 bool (*handle_input)(int fd, struct device *me);
131 /* Any queues attached to this device */
132 struct virtqueue *vq;
134 /* Handle status being finalized (ie. feature bits stable). */
135 void (*ready)(struct device *me);
137 /* Device-specific data. */
141 /* The virtqueue structure describes a queue attached to a device. */
144 struct virtqueue *next;
146 /* Which device owns me. */
149 /* The configuration for this queue. */
150 struct lguest_vqconfig config;
152 /* The actual ring of buffers. */
155 /* Last available index we saw. */
158 /* The routine to call when the Guest pings us. */
159 void (*handle_output)(int fd, struct virtqueue *me);
161 /* Outstanding buffers */
162 unsigned int inflight;
165 /* Remember the arguments to the program so we can "reboot" */
166 static char **main_args;
168 /* Since guest is UP and we don't run at the same time, we don't need barriers.
169 * But I include them in the code in case others copy it. */
172 /* Convert an iovec element to the given type.
174 * This is a fairly ugly trick: we need to know the size of the type and
175 * alignment requirement to check the pointer is kosher. It's also nice to
176 * have the name of the type in case we report failure.
178 * Typing those three things all the time is cumbersome and error prone, so we
179 * have a macro which sets them all up and passes to the real function. */
180 #define convert(iov, type) \
181 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
183 static void *_convert(struct iovec *iov, size_t size, size_t align,
186 if (iov->iov_len != size)
187 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
188 if ((unsigned long)iov->iov_base % align != 0)
189 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
190 return iov->iov_base;
193 /* The virtio configuration space is defined to be little-endian. x86 is
194 * little-endian too, but it's nice to be explicit so we have these helpers. */
195 #define cpu_to_le16(v16) (v16)
196 #define cpu_to_le32(v32) (v32)
197 #define cpu_to_le64(v64) (v64)
198 #define le16_to_cpu(v16) (v16)
199 #define le32_to_cpu(v32) (v32)
200 #define le64_to_cpu(v64) (v64)
202 /* The device virtqueue descriptors are followed by feature bitmasks. */
203 static u8 *get_feature_bits(struct device *dev)
205 return (u8 *)(dev->desc + 1)
206 + dev->desc->num_vq * sizeof(struct lguest_vqconfig);
209 /*L:100 The Launcher code itself takes us out into userspace, that scary place
210 * where pointers run wild and free! Unfortunately, like most userspace
211 * programs, it's quite boring (which is why everyone likes to hack on the
212 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
213 * will get you through this section. Or, maybe not.
215 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
216 * memory and stores it in "guest_base". In other words, Guest physical ==
217 * Launcher virtual with an offset.
219 * This can be tough to get your head around, but usually it just means that we
220 * use these trivial conversion functions when the Guest gives us it's
221 * "physical" addresses: */
222 static void *from_guest_phys(unsigned long addr)
224 return guest_base + addr;
227 static unsigned long to_guest_phys(const void *addr)
229 return (addr - guest_base);
233 * Loading the Kernel.
235 * We start with couple of simple helper routines. open_or_die() avoids
236 * error-checking code cluttering the callers: */
237 static int open_or_die(const char *name, int flags)
239 int fd = open(name, flags);
241 err(1, "Failed to open %s", name);
245 /* map_zeroed_pages() takes a number of pages. */
246 static void *map_zeroed_pages(unsigned int num)
248 int fd = open_or_die("/dev/zero", O_RDONLY);
251 /* We use a private mapping (ie. if we write to the page, it will be
253 addr = mmap(NULL, getpagesize() * num,
254 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
255 if (addr == MAP_FAILED)
256 err(1, "Mmaping %u pages of /dev/zero", num);
262 /* Get some more pages for a device. */
263 static void *get_pages(unsigned int num)
265 void *addr = from_guest_phys(guest_limit);
267 guest_limit += num * getpagesize();
268 if (guest_limit > guest_max)
269 errx(1, "Not enough memory for devices");
273 /* This routine is used to load the kernel or initrd. It tries mmap, but if
274 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
275 * it falls back to reading the memory in. */
276 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
280 /* We map writable even though for some segments are marked read-only.
281 * The kernel really wants to be writable: it patches its own
284 * MAP_PRIVATE means that the page won't be copied until a write is
285 * done to it. This allows us to share untouched memory between
287 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
288 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
291 /* pread does a seek and a read in one shot: saves a few lines. */
292 r = pread(fd, addr, len, offset);
294 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
297 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
298 * the Guest memory. ELF = Embedded Linking Format, which is the format used
299 * by all modern binaries on Linux including the kernel.
301 * The ELF headers give *two* addresses: a physical address, and a virtual
302 * address. We use the physical address; the Guest will map itself to the
305 * We return the starting address. */
306 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
308 Elf32_Phdr phdr[ehdr->e_phnum];
311 /* Sanity checks on the main ELF header: an x86 executable with a
312 * reasonable number of correctly-sized program headers. */
313 if (ehdr->e_type != ET_EXEC
314 || ehdr->e_machine != EM_386
315 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
316 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
317 errx(1, "Malformed elf header");
319 /* An ELF executable contains an ELF header and a number of "program"
320 * headers which indicate which parts ("segments") of the program to
323 /* We read in all the program headers at once: */
324 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
325 err(1, "Seeking to program headers");
326 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
327 err(1, "Reading program headers");
329 /* Try all the headers: there are usually only three. A read-only one,
330 * a read-write one, and a "note" section which we don't load. */
331 for (i = 0; i < ehdr->e_phnum; i++) {
332 /* If this isn't a loadable segment, we ignore it */
333 if (phdr[i].p_type != PT_LOAD)
336 verbose("Section %i: size %i addr %p\n",
337 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
339 /* We map this section of the file at its physical address. */
340 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
341 phdr[i].p_offset, phdr[i].p_filesz);
344 /* The entry point is given in the ELF header. */
345 return ehdr->e_entry;
348 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
349 * supposed to jump into it and it will unpack itself. We used to have to
350 * perform some hairy magic because the unpacking code scared me.
352 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
353 * a small patch to jump over the tricky bits in the Guest, so now we just read
354 * the funky header so we know where in the file to load, and away we go! */
355 static unsigned long load_bzimage(int fd)
357 struct boot_params boot;
359 /* Modern bzImages get loaded at 1M. */
360 void *p = from_guest_phys(0x100000);
362 /* Go back to the start of the file and read the header. It should be
363 * a Linux boot header (see Documentation/i386/boot.txt) */
364 lseek(fd, 0, SEEK_SET);
365 read(fd, &boot, sizeof(boot));
367 /* Inside the setup_hdr, we expect the magic "HdrS" */
368 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
369 errx(1, "This doesn't look like a bzImage to me");
371 /* Skip over the extra sectors of the header. */
372 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
374 /* Now read everything into memory. in nice big chunks. */
375 while ((r = read(fd, p, 65536)) > 0)
378 /* Finally, code32_start tells us where to enter the kernel. */
379 return boot.hdr.code32_start;
382 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
383 * come wrapped up in the self-decompressing "bzImage" format. With a little
384 * work, we can load those, too. */
385 static unsigned long load_kernel(int fd)
389 /* Read in the first few bytes. */
390 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
391 err(1, "Reading kernel");
393 /* If it's an ELF file, it starts with "\177ELF" */
394 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
395 return map_elf(fd, &hdr);
397 /* Otherwise we assume it's a bzImage, and try to load it. */
398 return load_bzimage(fd);
401 /* This is a trivial little helper to align pages. Andi Kleen hated it because
402 * it calls getpagesize() twice: "it's dumb code."
404 * Kernel guys get really het up about optimization, even when it's not
405 * necessary. I leave this code as a reaction against that. */
406 static inline unsigned long page_align(unsigned long addr)
408 /* Add upwards and truncate downwards. */
409 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
412 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
413 * the kernel which the kernel can use to boot from without needing any
414 * drivers. Most distributions now use this as standard: the initrd contains
415 * the code to load the appropriate driver modules for the current machine.
417 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
418 * kernels. He sent me this (and tells me when I break it). */
419 static unsigned long load_initrd(const char *name, unsigned long mem)
425 ifd = open_or_die(name, O_RDONLY);
426 /* fstat() is needed to get the file size. */
427 if (fstat(ifd, &st) < 0)
428 err(1, "fstat() on initrd '%s'", name);
430 /* We map the initrd at the top of memory, but mmap wants it to be
431 * page-aligned, so we round the size up for that. */
432 len = page_align(st.st_size);
433 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
434 /* Once a file is mapped, you can close the file descriptor. It's a
435 * little odd, but quite useful. */
437 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
439 /* We return the initrd size. */
443 /* Once we know how much memory we have we can construct simple linear page
444 * tables which set virtual == physical which will get the Guest far enough
445 * into the boot to create its own.
447 * We lay them out of the way, just below the initrd (which is why we need to
448 * know its size here). */
449 static unsigned long setup_pagetables(unsigned long mem,
450 unsigned long initrd_size)
452 unsigned long *pgdir, *linear;
453 unsigned int mapped_pages, i, linear_pages;
454 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
456 mapped_pages = mem/getpagesize();
458 /* Each PTE page can map ptes_per_page pages: how many do we need? */
459 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
461 /* We put the toplevel page directory page at the top of memory. */
462 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
464 /* Now we use the next linear_pages pages as pte pages */
465 linear = (void *)pgdir - linear_pages*getpagesize();
467 /* Linear mapping is easy: put every page's address into the mapping in
468 * order. PAGE_PRESENT contains the flags Present, Writable and
470 for (i = 0; i < mapped_pages; i++)
471 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
473 /* The top level points to the linear page table pages above. */
474 for (i = 0; i < mapped_pages; i += ptes_per_page) {
475 pgdir[i/ptes_per_page]
476 = ((to_guest_phys(linear) + i*sizeof(void *))
480 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
481 mapped_pages, linear_pages, to_guest_phys(linear));
483 /* We return the top level (guest-physical) address: the kernel needs
484 * to know where it is. */
485 return to_guest_phys(pgdir);
489 /* Simple routine to roll all the commandline arguments together with spaces
491 static void concat(char *dst, char *args[])
493 unsigned int i, len = 0;
495 for (i = 0; args[i]; i++) {
497 strcat(dst+len, " ");
500 strcpy(dst+len, args[i]);
501 len += strlen(args[i]);
503 /* In case it's empty. */
507 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
508 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
509 * the base of Guest "physical" memory, the top physical page to allow, the
510 * top level pagetable and the entry point for the Guest. */
511 static int tell_kernel(unsigned long pgdir, unsigned long start)
513 unsigned long args[] = { LHREQ_INITIALIZE,
514 (unsigned long)guest_base,
515 guest_limit / getpagesize(), pgdir, start };
518 verbose("Guest: %p - %p (%#lx)\n",
519 guest_base, guest_base + guest_limit, guest_limit);
520 fd = open_or_die("/dev/lguest", O_RDWR);
521 if (write(fd, args, sizeof(args)) < 0)
522 err(1, "Writing to /dev/lguest");
524 /* We return the /dev/lguest file descriptor to control this Guest */
529 static void add_device_fd(int fd)
531 FD_SET(fd, &devices.infds);
532 if (fd > devices.max_infd)
533 devices.max_infd = fd;
539 * With console, block and network devices, we can have lots of input which we
540 * need to process. We could try to tell the kernel what file descriptors to
541 * watch, but handing a file descriptor mask through to the kernel is fairly
544 * Instead, we fork off a process which watches the file descriptors and writes
545 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
546 * stop running the Guest. This causes the Launcher to return from the
547 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
548 * the LHREQ_BREAK and wake us up again.
550 * This, of course, is merely a different *kind* of icky.
552 static void wake_parent(int pipefd, int lguest_fd)
554 /* Add the pipe from the Launcher to the fdset in the device_list, so
555 * we watch it, too. */
556 add_device_fd(pipefd);
559 fd_set rfds = devices.infds;
560 unsigned long args[] = { LHREQ_BREAK, 1 };
562 /* Wait until input is ready from one of the devices. */
563 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
564 /* Is it a message from the Launcher? */
565 if (FD_ISSET(pipefd, &rfds)) {
567 /* If read() returns 0, it means the Launcher has
568 * exited. We silently follow. */
569 if (read(pipefd, &fd, sizeof(fd)) == 0)
571 /* Otherwise it's telling us to change what file
572 * descriptors we're to listen to. Positive means
573 * listen to a new one, negative means stop
576 FD_SET(fd, &devices.infds);
578 FD_CLR(-fd - 1, &devices.infds);
579 } else /* Send LHREQ_BREAK command. */
580 pwrite(lguest_fd, args, sizeof(args), cpu_id);
584 /* This routine just sets up a pipe to the Waker process. */
585 static int setup_waker(int lguest_fd)
587 int pipefd[2], child;
589 /* We create a pipe to talk to the Waker, and also so it knows when the
590 * Launcher dies (and closes pipe). */
597 /* We are the Waker: close the "writing" end of our copy of the
598 * pipe and start waiting for input. */
600 wake_parent(pipefd[0], lguest_fd);
602 /* Close the reading end of our copy of the pipe. */
605 /* Here is the fd used to talk to the waker. */
612 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
613 * We need to make sure it's not trying to reach into the Launcher itself, so
614 * we have a convenient routine which checks it and exits with an error message
615 * if something funny is going on:
617 static void *_check_pointer(unsigned long addr, unsigned int size,
620 /* We have to separately check addr and addr+size, because size could
621 * be huge and addr + size might wrap around. */
622 if (addr >= guest_limit || addr + size >= guest_limit)
623 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
624 /* We return a pointer for the caller's convenience, now we know it's
626 return from_guest_phys(addr);
628 /* A macro which transparently hands the line number to the real function. */
629 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
631 /* Each buffer in the virtqueues is actually a chain of descriptors. This
632 * function returns the next descriptor in the chain, or vq->vring.num if we're
634 static unsigned next_desc(struct virtqueue *vq, unsigned int i)
638 /* If this descriptor says it doesn't chain, we're done. */
639 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
640 return vq->vring.num;
642 /* Check they're not leading us off end of descriptors. */
643 next = vq->vring.desc[i].next;
644 /* Make sure compiler knows to grab that: we don't want it changing! */
647 if (next >= vq->vring.num)
648 errx(1, "Desc next is %u", next);
653 /* This looks in the virtqueue and for the first available buffer, and converts
654 * it to an iovec for convenient access. Since descriptors consist of some
655 * number of output then some number of input descriptors, it's actually two
656 * iovecs, but we pack them into one and note how many of each there were.
658 * This function returns the descriptor number found, or vq->vring.num (which
659 * is never a valid descriptor number) if none was found. */
660 static unsigned get_vq_desc(struct virtqueue *vq,
662 unsigned int *out_num, unsigned int *in_num)
664 unsigned int i, head;
666 /* Check it isn't doing very strange things with descriptor numbers. */
667 if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num)
668 errx(1, "Guest moved used index from %u to %u",
669 vq->last_avail_idx, vq->vring.avail->idx);
671 /* If there's nothing new since last we looked, return invalid. */
672 if (vq->vring.avail->idx == vq->last_avail_idx)
673 return vq->vring.num;
675 /* Grab the next descriptor number they're advertising, and increment
676 * the index we've seen. */
677 head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num];
679 /* If their number is silly, that's a fatal mistake. */
680 if (head >= vq->vring.num)
681 errx(1, "Guest says index %u is available", head);
683 /* When we start there are none of either input nor output. */
684 *out_num = *in_num = 0;
688 /* Grab the first descriptor, and check it's OK. */
689 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
690 iov[*out_num + *in_num].iov_base
691 = check_pointer(vq->vring.desc[i].addr,
692 vq->vring.desc[i].len);
693 /* If this is an input descriptor, increment that count. */
694 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
697 /* If it's an output descriptor, they're all supposed
698 * to come before any input descriptors. */
700 errx(1, "Descriptor has out after in");
704 /* If we've got too many, that implies a descriptor loop. */
705 if (*out_num + *in_num > vq->vring.num)
706 errx(1, "Looped descriptor");
707 } while ((i = next_desc(vq, i)) != vq->vring.num);
713 /* After we've used one of their buffers, we tell them about it. We'll then
714 * want to send them an interrupt, using trigger_irq(). */
715 static void add_used(struct virtqueue *vq, unsigned int head, int len)
717 struct vring_used_elem *used;
719 /* The virtqueue contains a ring of used buffers. Get a pointer to the
720 * next entry in that used ring. */
721 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
724 /* Make sure buffer is written before we update index. */
726 vq->vring.used->idx++;
730 /* This actually sends the interrupt for this virtqueue */
731 static void trigger_irq(int fd, struct virtqueue *vq)
733 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
735 /* If they don't want an interrupt, don't send one, unless empty. */
736 if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
740 /* Send the Guest an interrupt tell them we used something up. */
741 if (write(fd, buf, sizeof(buf)) != 0)
742 err(1, "Triggering irq %i", vq->config.irq);
745 /* And here's the combo meal deal. Supersize me! */
746 static void add_used_and_trigger(int fd, struct virtqueue *vq,
747 unsigned int head, int len)
749 add_used(vq, head, len);
756 * Here is the input terminal setting we save, and the routine to restore them
757 * on exit so the user gets their terminal back. */
758 static struct termios orig_term;
759 static void restore_term(void)
761 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
764 /* We associate some data with the console for our exit hack. */
767 /* How many times have they hit ^C? */
769 /* When did they start? */
770 struct timeval start;
773 /* This is the routine which handles console input (ie. stdin). */
774 static bool handle_console_input(int fd, struct device *dev)
777 unsigned int head, in_num, out_num;
778 struct iovec iov[dev->vq->vring.num];
779 struct console_abort *abort = dev->priv;
781 /* First we need a console buffer from the Guests's input virtqueue. */
782 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
784 /* If they're not ready for input, stop listening to this file
785 * descriptor. We'll start again once they add an input buffer. */
786 if (head == dev->vq->vring.num)
790 errx(1, "Output buffers in console in queue?");
792 /* This is why we convert to iovecs: the readv() call uses them, and so
793 * it reads straight into the Guest's buffer. */
794 len = readv(dev->fd, iov, in_num);
796 /* This implies that the console is closed, is /dev/null, or
797 * something went terribly wrong. */
798 warnx("Failed to get console input, ignoring console.");
799 /* Put the input terminal back. */
801 /* Remove callback from input vq, so it doesn't restart us. */
802 dev->vq->handle_output = NULL;
803 /* Stop listening to this fd: don't call us again. */
807 /* Tell the Guest about the new input. */
808 add_used_and_trigger(fd, dev->vq, head, len);
810 /* Three ^C within one second? Exit.
812 * This is such a hack, but works surprisingly well. Each ^C has to be
813 * in a buffer by itself, so they can't be too fast. But we check that
814 * we get three within about a second, so they can't be too slow. */
815 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
817 gettimeofday(&abort->start, NULL);
818 else if (abort->count == 3) {
820 gettimeofday(&now, NULL);
821 if (now.tv_sec <= abort->start.tv_sec+1) {
822 unsigned long args[] = { LHREQ_BREAK, 0 };
823 /* Close the fd so Waker will know it has to
826 /* Just in case waker is blocked in BREAK, send
828 write(fd, args, sizeof(args));
834 /* Any other key resets the abort counter. */
837 /* Everything went OK! */
841 /* Handling output for console is simple: we just get all the output buffers
842 * and write them to stdout. */
843 static void handle_console_output(int fd, struct virtqueue *vq)
845 unsigned int head, out, in;
847 struct iovec iov[vq->vring.num];
849 /* Keep getting output buffers from the Guest until we run out. */
850 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
852 errx(1, "Input buffers in output queue?");
853 len = writev(STDOUT_FILENO, iov, out);
854 add_used_and_trigger(fd, vq, head, len);
861 * Handling output for network is also simple: we get all the output buffers
862 * and write them (ignoring the first element) to this device's file descriptor
865 static void handle_net_output(int fd, struct virtqueue *vq)
867 unsigned int head, out, in;
869 struct iovec iov[vq->vring.num];
871 /* Keep getting output buffers from the Guest until we run out. */
872 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
874 errx(1, "Input buffers in output queue?");
875 /* Check header, but otherwise ignore it (we told the Guest we
876 * supported no features, so it shouldn't have anything
878 (void)convert(&iov[0], struct virtio_net_hdr);
879 len = writev(vq->dev->fd, iov+1, out-1);
880 add_used_and_trigger(fd, vq, head, len);
884 /* This is where we handle a packet coming in from the tun device to our
886 static bool handle_tun_input(int fd, struct device *dev)
888 unsigned int head, in_num, out_num;
890 struct iovec iov[dev->vq->vring.num];
891 struct virtio_net_hdr *hdr;
893 /* First we need a network buffer from the Guests's recv virtqueue. */
894 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
895 if (head == dev->vq->vring.num) {
896 /* Now, it's expected that if we try to send a packet too
897 * early, the Guest won't be ready yet. Wait until the device
898 * status says it's ready. */
899 /* FIXME: Actually want DRIVER_ACTIVE here. */
900 if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK)
901 warn("network: no dma buffer!");
902 /* We'll turn this back on if input buffers are registered. */
905 errx(1, "Output buffers in network recv queue?");
907 /* First element is the header: we set it to 0 (no features). */
908 hdr = convert(&iov[0], struct virtio_net_hdr);
910 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
912 /* Read the packet from the device directly into the Guest's buffer. */
913 len = readv(dev->fd, iov+1, in_num-1);
915 err(1, "reading network");
917 /* Tell the Guest about the new packet. */
918 add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len);
920 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
921 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
922 head != dev->vq->vring.num ? "sent" : "discarded");
928 /*L:215 This is the callback attached to the network and console input
929 * virtqueues: it ensures we try again, in case we stopped console or net
930 * delivery because Guest didn't have any buffers. */
931 static void enable_fd(int fd, struct virtqueue *vq)
933 add_device_fd(vq->dev->fd);
934 /* Tell waker to listen to it again */
935 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
938 /* When the Guest tells us they updated the status field, we handle it. */
939 static void update_device_status(struct device *dev)
941 struct virtqueue *vq;
943 /* This is a reset. */
944 if (dev->desc->status == 0) {
945 verbose("Resetting device %s\n", dev->name);
947 /* Clear any features they've acked. */
948 memset(get_feature_bits(dev) + dev->desc->feature_len, 0,
949 dev->desc->feature_len);
951 /* Zero out the virtqueues. */
952 for (vq = dev->vq; vq; vq = vq->next) {
953 memset(vq->vring.desc, 0,
954 vring_size(vq->config.num, getpagesize()));
955 vq->last_avail_idx = 0;
957 } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
958 warnx("Device %s configuration FAILED", dev->name);
959 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
962 verbose("Device %s OK: offered", dev->name);
963 for (i = 0; i < dev->desc->feature_len; i++)
964 verbose(" %02x", get_feature_bits(dev)[i]);
965 verbose(", accepted");
966 for (i = 0; i < dev->desc->feature_len; i++)
967 verbose(" %02x", get_feature_bits(dev)
968 [dev->desc->feature_len+i]);
975 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
976 static void handle_output(int fd, unsigned long addr)
979 struct virtqueue *vq;
981 /* Check each device and virtqueue. */
982 for (i = devices.dev; i; i = i->next) {
983 /* Notifications to device descriptors update device status. */
984 if (from_guest_phys(addr) == i->desc) {
985 update_device_status(i);
989 /* Notifications to virtqueues mean output has occurred. */
990 for (vq = i->vq; vq; vq = vq->next) {
991 if (vq->config.pfn != addr/getpagesize())
994 /* Guest should acknowledge (and set features!) before
995 * using the device. */
996 if (i->desc->status == 0) {
997 warnx("%s gave early output", i->name);
1001 if (strcmp(vq->dev->name, "console") != 0)
1002 verbose("Output to %s\n", vq->dev->name);
1003 if (vq->handle_output)
1004 vq->handle_output(fd, vq);
1009 /* Early console write is done using notify on a nul-terminated string
1010 * in Guest memory. */
1011 if (addr >= guest_limit)
1012 errx(1, "Bad NOTIFY %#lx", addr);
1014 write(STDOUT_FILENO, from_guest_phys(addr),
1015 strnlen(from_guest_phys(addr), guest_limit - addr));
1018 /* This is called when the Waker wakes us up: check for incoming file
1020 static void handle_input(int fd)
1022 /* select() wants a zeroed timeval to mean "don't wait". */
1023 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
1027 fd_set fds = devices.infds;
1029 /* If nothing is ready, we're done. */
1030 if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0)
1033 /* Otherwise, call the device(s) which have readable file
1034 * descriptors and a method of handling them. */
1035 for (i = devices.dev; i; i = i->next) {
1036 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
1038 if (i->handle_input(fd, i))
1041 /* If handle_input() returns false, it means we
1042 * should no longer service it. Networking and
1043 * console do this when there's no input
1044 * buffers to deliver into. Console also uses
1045 * it when it discovers that stdin is closed. */
1046 FD_CLR(i->fd, &devices.infds);
1047 /* Tell waker to ignore it too, by sending a
1048 * negative fd number (-1, since 0 is a valid
1050 dev_fd = -i->fd - 1;
1051 write(waker_fd, &dev_fd, sizeof(dev_fd));
1060 * All devices need a descriptor so the Guest knows it exists, and a "struct
1061 * device" so the Launcher can keep track of it. We have common helper
1062 * routines to allocate and manage them.
1065 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1066 * number of virtqueue descriptors, then two sets of feature bits, then an
1067 * array of configuration bytes. This routine returns the configuration
1069 static u8 *device_config(const struct device *dev)
1071 return (void *)(dev->desc + 1)
1072 + dev->desc->num_vq * sizeof(struct lguest_vqconfig)
1073 + dev->desc->feature_len * 2;
1076 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1077 * table page just above the Guest's normal memory. It returns a pointer to
1078 * that descriptor. */
1079 static struct lguest_device_desc *new_dev_desc(u16 type)
1081 struct lguest_device_desc d = { .type = type };
1084 /* Figure out where the next device config is, based on the last one. */
1085 if (devices.lastdev)
1086 p = device_config(devices.lastdev)
1087 + devices.lastdev->desc->config_len;
1089 p = devices.descpage;
1091 /* We only have one page for all the descriptors. */
1092 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1093 errx(1, "Too many devices");
1095 /* p might not be aligned, so we memcpy in. */
1096 return memcpy(p, &d, sizeof(d));
1099 /* Each device descriptor is followed by the description of its virtqueues. We
1100 * specify how many descriptors the virtqueue is to have. */
1101 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1102 void (*handle_output)(int fd, struct virtqueue *me))
1105 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1108 /* First we need some memory for this virtqueue. */
1109 pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1)
1111 p = get_pages(pages);
1113 /* Initialize the virtqueue */
1115 vq->last_avail_idx = 0;
1119 /* Initialize the configuration. */
1120 vq->config.num = num_descs;
1121 vq->config.irq = devices.next_irq++;
1122 vq->config.pfn = to_guest_phys(p) / getpagesize();
1124 /* Initialize the vring. */
1125 vring_init(&vq->vring, num_descs, p, getpagesize());
1127 /* Append virtqueue to this device's descriptor. We use
1128 * device_config() to get the end of the device's current virtqueues;
1129 * we check that we haven't added any config or feature information
1130 * yet, otherwise we'd be overwriting them. */
1131 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1132 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1133 dev->desc->num_vq++;
1135 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1137 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1139 for (i = &dev->vq; *i; i = &(*i)->next);
1142 /* Set the routine to call when the Guest does something to this
1144 vq->handle_output = handle_output;
1146 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1147 * don't have a handler */
1149 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1152 /* The first half of the feature bitmask is for us to advertise features. The
1153 * second half is for the Guest to accept features. */
1154 static void add_feature(struct device *dev, unsigned bit)
1156 u8 *features = get_feature_bits(dev);
1158 /* We can't extend the feature bits once we've added config bytes */
1159 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1160 assert(dev->desc->config_len == 0);
1161 dev->desc->feature_len = (bit / CHAR_BIT) + 1;
1164 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1167 /* This routine sets the configuration fields for an existing device's
1168 * descriptor. It only works for the last device, but that's OK because that's
1170 static void set_config(struct device *dev, unsigned len, const void *conf)
1172 /* Check we haven't overflowed our single page. */
1173 if (device_config(dev) + len > devices.descpage + getpagesize())
1174 errx(1, "Too many devices");
1176 /* Copy in the config information, and store the length. */
1177 memcpy(device_config(dev), conf, len);
1178 dev->desc->config_len = len;
1181 /* This routine does all the creation and setup of a new device, including
1182 * calling new_dev_desc() to allocate the descriptor and device memory.
1184 * See what I mean about userspace being boring? */
1185 static struct device *new_device(const char *name, u16 type, int fd,
1186 bool (*handle_input)(int, struct device *))
1188 struct device *dev = malloc(sizeof(*dev));
1190 /* Now we populate the fields one at a time. */
1192 /* If we have an input handler for this file descriptor, then we add it
1193 * to the device_list's fdset and maxfd. */
1195 add_device_fd(dev->fd);
1196 dev->desc = new_dev_desc(type);
1197 dev->handle_input = handle_input;
1202 /* Append to device list. Prepending to a single-linked list is
1203 * easier, but the user expects the devices to be arranged on the bus
1204 * in command-line order. The first network device on the command line
1205 * is eth0, the first block device /dev/vda, etc. */
1206 if (devices.lastdev)
1207 devices.lastdev->next = dev;
1210 devices.lastdev = dev;
1215 /* Our first setup routine is the console. It's a fairly simple device, but
1216 * UNIX tty handling makes it uglier than it could be. */
1217 static void setup_console(void)
1221 /* If we can save the initial standard input settings... */
1222 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1223 struct termios term = orig_term;
1224 /* Then we turn off echo, line buffering and ^C etc. We want a
1225 * raw input stream to the Guest. */
1226 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1227 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1228 /* If we exit gracefully, the original settings will be
1229 * restored so the user can see what they're typing. */
1230 atexit(restore_term);
1233 dev = new_device("console", VIRTIO_ID_CONSOLE,
1234 STDIN_FILENO, handle_console_input);
1235 /* We store the console state in dev->priv, and initialize it. */
1236 dev->priv = malloc(sizeof(struct console_abort));
1237 ((struct console_abort *)dev->priv)->count = 0;
1239 /* The console needs two virtqueues: the input then the output. When
1240 * they put something the input queue, we make sure we're listening to
1241 * stdin. When they put something in the output queue, we write it to
1243 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1244 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1246 verbose("device %u: console\n", devices.device_num++);
1250 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1251 * --sharenet=<name> option which opens or creates a named pipe. This can be
1252 * used to send packets to another guest in a 1:1 manner.
1254 * More sopisticated is to use one of the tools developed for project like UML
1257 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1258 * completely generic ("here's my vring, attach to your vring") and would work
1259 * for any traffic. Of course, namespace and permissions issues need to be
1260 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1261 * multiple inter-guest channels behind one interface, although it would
1262 * require some manner of hotplugging new virtio channels.
1264 * Finally, we could implement a virtio network switch in the kernel. :*/
1266 static u32 str2ip(const char *ipaddr)
1270 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1271 errx(1, "Failed to parse IP address '%s'", ipaddr);
1272 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1275 static void str2mac(const char *macaddr, unsigned char mac[6])
1278 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1279 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1280 errx(1, "Failed to parse mac address '%s'", macaddr);
1289 /* This code is "adapted" from libbridge: it attaches the Host end of the
1290 * network device to the bridge device specified by the command line.
1292 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1293 * dislike bridging), and I just try not to break it. */
1294 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1300 errx(1, "must specify bridge name");
1302 ifidx = if_nametoindex(if_name);
1304 errx(1, "interface %s does not exist!", if_name);
1306 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1307 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1308 ifr.ifr_ifindex = ifidx;
1309 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1310 err(1, "can't add %s to bridge %s", if_name, br_name);
1313 /* This sets up the Host end of the network device with an IP address, brings
1314 * it up so packets will flow, the copies the MAC address into the hwaddr
1316 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1319 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1321 memset(&ifr, 0, sizeof(ifr));
1322 strcpy(ifr.ifr_name, tapif);
1324 /* Don't read these incantations. Just cut & paste them like I did! */
1325 sin->sin_family = AF_INET;
1326 sin->sin_addr.s_addr = htonl(ipaddr);
1327 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1328 err(1, "Setting %s interface address", tapif);
1329 ifr.ifr_flags = IFF_UP;
1330 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1331 err(1, "Bringing interface %s up", tapif);
1334 static void get_mac(int fd, const char *tapif, unsigned char hwaddr[6])
1338 memset(&ifr, 0, sizeof(ifr));
1339 strcpy(ifr.ifr_name, tapif);
1341 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1342 * above). IF means Interface, and HWADDR is hardware address.
1344 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1345 err(1, "getting hw address for %s", tapif);
1346 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1349 static int get_tun_device(char tapif[IFNAMSIZ])
1354 /* Start with this zeroed. Messy but sure. */
1355 memset(&ifr, 0, sizeof(ifr));
1357 /* We open the /dev/net/tun device and tell it we want a tap device. A
1358 * tap device is like a tun device, only somehow different. To tell
1359 * the truth, I completely blundered my way through this code, but it
1361 netfd = open_or_die("/dev/net/tun", O_RDWR);
1362 ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
1363 strcpy(ifr.ifr_name, "tap%d");
1364 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1365 err(1, "configuring /dev/net/tun");
1367 /* We don't need checksums calculated for packets coming in this
1368 * device: trust us! */
1369 ioctl(netfd, TUNSETNOCSUM, 1);
1371 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1375 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1376 * routing, but the principle is the same: it uses the "tun" device to inject
1377 * packets into the Host as if they came in from a normal network card. We
1378 * just shunt packets between the Guest and the tun device. */
1379 static void setup_tun_net(char *arg)
1383 u32 ip = INADDR_ANY;
1384 bool bridging = false;
1385 char tapif[IFNAMSIZ], *p;
1386 struct virtio_net_config conf;
1388 netfd = get_tun_device(tapif);
1390 /* First we create a new network device. */
1391 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1393 /* Network devices need a receive and a send queue, just like
1395 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1396 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1398 /* We need a socket to perform the magic network ioctls to bring up the
1399 * tap interface, connect to the bridge etc. Any socket will do! */
1400 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1402 err(1, "opening IP socket");
1404 /* If the command line was --tunnet=bridge:<name> do bridging. */
1405 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1406 arg += strlen(BRIDGE_PFX);
1410 /* A mac address may follow the bridge name or IP address */
1411 p = strchr(arg, ':');
1413 str2mac(p+1, conf.mac);
1416 p = arg + strlen(arg);
1417 /* None supplied; query the randomly assigned mac. */
1418 get_mac(ipfd, tapif, conf.mac);
1421 /* arg is now either an IP address or a bridge name */
1423 add_to_bridge(ipfd, tapif, arg);
1427 /* Set up the tun device. */
1428 configure_device(ipfd, tapif, ip);
1430 /* Tell Guest what MAC address to use. */
1431 add_feature(dev, VIRTIO_NET_F_MAC);
1432 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1433 set_config(dev, sizeof(conf), &conf);
1435 /* We don't need the socket any more; setup is done. */
1438 devices.device_num++;
1441 verbose("device %u: tun %s attached to bridge: %s\n",
1442 devices.device_num, tapif, arg);
1444 verbose("device %u: tun %s: %s\n",
1445 devices.device_num, tapif, arg);
1448 /* Our block (disk) device should be really simple: the Guest asks for a block
1449 * number and we read or write that position in the file. Unfortunately, that
1450 * was amazingly slow: the Guest waits until the read is finished before
1451 * running anything else, even if it could have been doing useful work.
1453 * We could use async I/O, except it's reputed to suck so hard that characters
1454 * actually go missing from your code when you try to use it.
1456 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1458 /* This hangs off device->priv. */
1461 /* The size of the file. */
1464 /* The file descriptor for the file. */
1467 /* IO thread listens on this file descriptor [0]. */
1470 /* IO thread writes to this file descriptor to mark it done, then
1471 * Launcher triggers interrupt to Guest. */
1478 * Remember that the block device is handled by a separate I/O thread. We head
1479 * straight into the core of that thread here:
1481 static bool service_io(struct device *dev)
1483 struct vblk_info *vblk = dev->priv;
1484 unsigned int head, out_num, in_num, wlen;
1487 struct virtio_blk_outhdr *out;
1488 struct iovec iov[dev->vq->vring.num];
1491 /* See if there's a request waiting. If not, nothing to do. */
1492 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1493 if (head == dev->vq->vring.num)
1496 /* Every block request should contain at least one output buffer
1497 * (detailing the location on disk and the type of request) and one
1498 * input buffer (to hold the result). */
1499 if (out_num == 0 || in_num == 0)
1500 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1501 head, out_num, in_num);
1503 out = convert(&iov[0], struct virtio_blk_outhdr);
1504 in = convert(&iov[out_num+in_num-1], u8);
1505 off = out->sector * 512;
1507 /* The block device implements "barriers", where the Guest indicates
1508 * that it wants all previous writes to occur before this write. We
1509 * don't have a way of asking our kernel to do a barrier, so we just
1510 * synchronize all the data in the file. Pretty poor, no? */
1511 if (out->type & VIRTIO_BLK_T_BARRIER)
1512 fdatasync(vblk->fd);
1514 /* In general the virtio block driver is allowed to try SCSI commands.
1515 * It'd be nice if we supported eject, for example, but we don't. */
1516 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1517 fprintf(stderr, "Scsi commands unsupported\n");
1518 *in = VIRTIO_BLK_S_UNSUPP;
1520 } else if (out->type & VIRTIO_BLK_T_OUT) {
1523 /* Move to the right location in the block file. This can fail
1524 * if they try to write past end. */
1525 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1526 err(1, "Bad seek to sector %llu", out->sector);
1528 ret = writev(vblk->fd, iov+1, out_num-1);
1529 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1531 /* Grr... Now we know how long the descriptor they sent was, we
1532 * make sure they didn't try to write over the end of the block
1533 * file (possibly extending it). */
1534 if (ret > 0 && off + ret > vblk->len) {
1535 /* Trim it back to the correct length */
1536 ftruncate64(vblk->fd, vblk->len);
1537 /* Die, bad Guest, die. */
1538 errx(1, "Write past end %llu+%u", off, ret);
1541 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1545 /* Move to the right location in the block file. This can fail
1546 * if they try to read past end. */
1547 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1548 err(1, "Bad seek to sector %llu", out->sector);
1550 ret = readv(vblk->fd, iov+1, in_num-1);
1551 verbose("READ from sector %llu: %i\n", out->sector, ret);
1553 wlen = sizeof(*in) + ret;
1554 *in = VIRTIO_BLK_S_OK;
1557 *in = VIRTIO_BLK_S_IOERR;
1561 /* We can't trigger an IRQ, because we're not the Launcher. It does
1562 * that when we tell it we're done. */
1563 add_used(dev->vq, head, wlen);
1567 /* This is the thread which actually services the I/O. */
1568 static int io_thread(void *_dev)
1570 struct device *dev = _dev;
1571 struct vblk_info *vblk = dev->priv;
1574 /* Close other side of workpipe so we get 0 read when main dies. */
1575 close(vblk->workpipe[1]);
1576 /* Close the other side of the done_fd pipe. */
1579 /* When this read fails, it means Launcher died, so we follow. */
1580 while (read(vblk->workpipe[0], &c, 1) == 1) {
1581 /* We acknowledge each request immediately to reduce latency,
1582 * rather than waiting until we've done them all. I haven't
1583 * measured to see if it makes any difference.
1585 * That would be an interesting test, wouldn't it? You could
1586 * also try having more than one I/O thread. */
1587 while (service_io(dev))
1588 write(vblk->done_fd, &c, 1);
1593 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1594 * when that thread tells us it's completed some I/O. */
1595 static bool handle_io_finish(int fd, struct device *dev)
1599 /* If the I/O thread died, presumably it printed the error, so we
1601 if (read(dev->fd, &c, 1) != 1)
1604 /* It did some work, so trigger the irq. */
1605 trigger_irq(fd, dev->vq);
1609 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1610 static void handle_virtblk_output(int fd, struct virtqueue *vq)
1612 struct vblk_info *vblk = vq->dev->priv;
1615 /* Wake up I/O thread and tell it to go to work! */
1616 if (write(vblk->workpipe[1], &c, 1) != 1)
1617 /* Presumably it indicated why it died. */
1621 /*L:198 This actually sets up a virtual block device. */
1622 static void setup_block_file(const char *filename)
1626 struct vblk_info *vblk;
1628 struct virtio_blk_config conf;
1630 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1633 /* The device responds to return from I/O thread. */
1634 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1636 /* The device has one virtqueue, where the Guest places requests. */
1637 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1639 /* Allocate the room for our own bookkeeping */
1640 vblk = dev->priv = malloc(sizeof(*vblk));
1642 /* First we open the file and store the length. */
1643 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1644 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1646 /* We support barriers. */
1647 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1649 /* Tell Guest how many sectors this device has. */
1650 conf.capacity = cpu_to_le64(vblk->len / 512);
1652 /* Tell Guest not to put in too many descriptors at once: two are used
1653 * for the in and out elements. */
1654 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1655 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1657 set_config(dev, sizeof(conf), &conf);
1659 /* The I/O thread writes to this end of the pipe when done. */
1660 vblk->done_fd = p[1];
1662 /* This is the second pipe, which is how we tell the I/O thread about
1664 pipe(vblk->workpipe);
1666 /* Create stack for thread and run it. Since stack grows upwards, we
1667 * point the stack pointer to the end of this region. */
1668 stack = malloc(32768);
1669 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1670 * becoming a zombie. */
1671 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
1672 err(1, "Creating clone");
1674 /* We don't need to keep the I/O thread's end of the pipes open. */
1675 close(vblk->done_fd);
1676 close(vblk->workpipe[0]);
1678 verbose("device %u: virtblock %llu sectors\n",
1679 devices.device_num, le64_to_cpu(conf.capacity));
1681 /* That's the end of device setup. */
1683 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1684 static void __attribute__((noreturn)) restart_guest(void)
1688 /* Closing pipes causes the Waker thread and io_threads to die, and
1689 * closing /dev/lguest cleans up the Guest. Since we don't track all
1690 * open fds, we simply close everything beyond stderr. */
1691 for (i = 3; i < FD_SETSIZE; i++)
1693 execv(main_args[0], main_args);
1694 err(1, "Could not exec %s", main_args[0]);
1697 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1698 * its input and output, and finally, lays it to rest. */
1699 static void __attribute__((noreturn)) run_guest(int lguest_fd)
1702 unsigned long args[] = { LHREQ_BREAK, 0 };
1703 unsigned long notify_addr;
1706 /* We read from the /dev/lguest device to run the Guest. */
1707 readval = pread(lguest_fd, ¬ify_addr,
1708 sizeof(notify_addr), cpu_id);
1710 /* One unsigned long means the Guest did HCALL_NOTIFY */
1711 if (readval == sizeof(notify_addr)) {
1712 verbose("Notify on address %#lx\n", notify_addr);
1713 handle_output(lguest_fd, notify_addr);
1715 /* ENOENT means the Guest died. Reading tells us why. */
1716 } else if (errno == ENOENT) {
1717 char reason[1024] = { 0 };
1718 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1719 errx(1, "%s", reason);
1720 /* ERESTART means that we need to reboot the guest */
1721 } else if (errno == ERESTART) {
1723 /* EAGAIN means the Waker wanted us to look at some input.
1724 * Anything else means a bug or incompatible change. */
1725 } else if (errno != EAGAIN)
1726 err(1, "Running guest failed");
1728 /* Only service input on thread for CPU 0. */
1732 /* Service input, then unset the BREAK to release the Waker. */
1733 handle_input(lguest_fd);
1734 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1735 err(1, "Resetting break");
1739 * This is the end of the Launcher. The good news: we are over halfway
1740 * through! The bad news: the most fiendish part of the code still lies ahead
1743 * Are you ready? Take a deep breath and join me in the core of the Host, in
1747 static struct option opts[] = {
1748 { "verbose", 0, NULL, 'v' },
1749 { "tunnet", 1, NULL, 't' },
1750 { "block", 1, NULL, 'b' },
1751 { "initrd", 1, NULL, 'i' },
1754 static void usage(void)
1756 errx(1, "Usage: lguest [--verbose] "
1757 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1758 "|--block=<filename>|--initrd=<filename>]...\n"
1759 "<mem-in-mb> vmlinux [args...]");
1762 /*L:105 The main routine is where the real work begins: */
1763 int main(int argc, char *argv[])
1765 /* Memory, top-level pagetable, code startpoint and size of the
1766 * (optional) initrd. */
1767 unsigned long mem = 0, pgdir, start, initrd_size = 0;
1768 /* Two temporaries and the /dev/lguest file descriptor. */
1769 int i, c, lguest_fd;
1770 /* The boot information for the Guest. */
1771 struct boot_params *boot;
1772 /* If they specify an initrd file to load. */
1773 const char *initrd_name = NULL;
1775 /* Save the args: we "reboot" by execing ourselves again. */
1777 /* We don't "wait" for the children, so prevent them from becoming
1779 signal(SIGCHLD, SIG_IGN);
1781 /* First we initialize the device list. Since console and network
1782 * device receive input from a file descriptor, we keep an fdset
1783 * (infds) and the maximum fd number (max_infd) with the head of the
1784 * list. We also keep a pointer to the last device. Finally, we keep
1785 * the next interrupt number to use for devices (1: remember that 0 is
1786 * used by the timer). */
1787 FD_ZERO(&devices.infds);
1788 devices.max_infd = -1;
1789 devices.lastdev = NULL;
1790 devices.next_irq = 1;
1793 /* We need to know how much memory so we can set up the device
1794 * descriptor and memory pages for the devices as we parse the command
1795 * line. So we quickly look through the arguments to find the amount
1797 for (i = 1; i < argc; i++) {
1798 if (argv[i][0] != '-') {
1799 mem = atoi(argv[i]) * 1024 * 1024;
1800 /* We start by mapping anonymous pages over all of
1801 * guest-physical memory range. This fills it with 0,
1802 * and ensures that the Guest won't be killed when it
1803 * tries to access it. */
1804 guest_base = map_zeroed_pages(mem / getpagesize()
1807 guest_max = mem + DEVICE_PAGES*getpagesize();
1808 devices.descpage = get_pages(1);
1813 /* The options are fairly straight-forward */
1814 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1820 setup_tun_net(optarg);
1823 setup_block_file(optarg);
1826 initrd_name = optarg;
1829 warnx("Unknown argument %s", argv[optind]);
1833 /* After the other arguments we expect memory and kernel image name,
1834 * followed by command line arguments for the kernel. */
1835 if (optind + 2 > argc)
1838 verbose("Guest base is at %p\n", guest_base);
1840 /* We always have a console device */
1843 /* Now we load the kernel */
1844 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1846 /* Boot information is stashed at physical address 0 */
1847 boot = from_guest_phys(0);
1849 /* Map the initrd image if requested (at top of physical memory) */
1851 initrd_size = load_initrd(initrd_name, mem);
1852 /* These are the location in the Linux boot header where the
1853 * start and size of the initrd are expected to be found. */
1854 boot->hdr.ramdisk_image = mem - initrd_size;
1855 boot->hdr.ramdisk_size = initrd_size;
1856 /* The bootloader type 0xFF means "unknown"; that's OK. */
1857 boot->hdr.type_of_loader = 0xFF;
1860 /* Set up the initial linear pagetables, starting below the initrd. */
1861 pgdir = setup_pagetables(mem, initrd_size);
1863 /* The Linux boot header contains an "E820" memory map: ours is a
1864 * simple, single region. */
1865 boot->e820_entries = 1;
1866 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1867 /* The boot header contains a command line pointer: we put the command
1868 * line after the boot header. */
1869 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1870 /* We use a simple helper to copy the arguments separated by spaces. */
1871 concat((char *)(boot + 1), argv+optind+2);
1873 /* Boot protocol version: 2.07 supports the fields for lguest. */
1874 boot->hdr.version = 0x207;
1876 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1877 boot->hdr.hardware_subarch = 1;
1879 /* Tell the entry path not to try to reload segment registers. */
1880 boot->hdr.loadflags |= KEEP_SEGMENTS;
1882 /* We tell the kernel to initialize the Guest: this returns the open
1883 * /dev/lguest file descriptor. */
1884 lguest_fd = tell_kernel(pgdir, start);
1886 /* We fork off a child process, which wakes the Launcher whenever one
1887 * of the input file descriptors needs attention. We call this the
1888 * Waker, and we'll cover it in a moment. */
1889 waker_fd = setup_waker(lguest_fd);
1891 /* Finally, run the Guest. This doesn't return. */
1892 run_guest(lguest_fd);
1897 * Mastery is done: you now know everything I do.
1899 * But surely you have seen code, features and bugs in your wanderings which
1900 * you now yearn to attack? That is the real game, and I look forward to you
1901 * patching and forking lguest into the Your-Name-Here-visor.
1903 * Farewell, and good coding!