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>
40 #include "linux/lguest_launcher.h"
41 #include "linux/virtio_config.h"
42 #include "linux/virtio_net.h"
43 #include "linux/virtio_blk.h"
44 #include "linux/virtio_console.h"
45 #include "linux/virtio_rng.h"
46 #include "linux/virtio_ring.h"
47 #include "asm/bootparam.h"
48 /*L:110 We can ignore the 39 include files we need for this program, but I do
49 * want to draw attention to the use of kernel-style types.
51 * As Linus said, "C is a Spartan language, and so should your naming be." I
52 * like these abbreviations, so we define them here. Note that u64 is always
53 * unsigned long long, which works on all Linux systems: this means that we can
54 * use %llu in printf for any u64. */
55 typedef unsigned long long u64;
61 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
63 #define BRIDGE_PFX "bridge:"
65 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
67 /* We can have up to 256 pages for devices. */
68 #define DEVICE_PAGES 256
69 /* This will occupy 3 pages: it must be a power of 2. */
70 #define VIRTQUEUE_NUM 256
72 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
73 * this, and although I wouldn't recommend it, it works quite nicely here. */
75 #define verbose(args...) \
76 do { if (verbose) printf(args); } while(0)
79 /* File descriptors for the Waker. */
84 /* The pointer to the start of guest memory. */
85 static void *guest_base;
86 /* The maximum guest physical address allowed, and maximum possible. */
87 static unsigned long guest_limit, guest_max;
88 /* The pipe for signal hander to write to. */
89 static int timeoutpipe[2];
90 static unsigned int timeout_usec = 500;
91 /* The /dev/lguest file descriptor. */
94 /* a per-cpu variable indicating whose vcpu is currently running */
95 static unsigned int __thread cpu_id;
97 /* This is our list of devices. */
100 /* Summary information about the devices in our list: ready to pass to
101 * select() to ask which need servicing.*/
105 /* Counter to assign interrupt numbers. */
106 unsigned int next_irq;
108 /* Counter to print out convenient device numbers. */
109 unsigned int device_num;
111 /* The descriptor page for the devices. */
114 /* A single linked list of devices. */
116 /* And a pointer to the last device for easy append and also for
117 * configuration appending. */
118 struct device *lastdev;
121 /* The list of Guest devices, based on command line arguments. */
122 static struct device_list devices;
124 /* The device structure describes a single device. */
127 /* The linked-list pointer. */
130 /* The device's descriptor, as mapped into the Guest. */
131 struct lguest_device_desc *desc;
133 /* We can't trust desc values once Guest has booted: we use these. */
134 unsigned int feature_len;
137 /* The name of this device, for --verbose. */
140 /* If handle_input is set, it wants to be called when this file
141 * descriptor is ready. */
143 bool (*handle_input)(struct device *me);
145 /* Any queues attached to this device */
146 struct virtqueue *vq;
148 /* Handle status being finalized (ie. feature bits stable). */
149 void (*ready)(struct device *me);
151 /* Device-specific data. */
155 /* The virtqueue structure describes a queue attached to a device. */
158 struct virtqueue *next;
160 /* Which device owns me. */
163 /* The configuration for this queue. */
164 struct lguest_vqconfig config;
166 /* The actual ring of buffers. */
169 /* Last available index we saw. */
172 /* The routine to call when the Guest pings us, or timeout. */
173 void (*handle_output)(struct virtqueue *me, bool timeout);
175 /* Is this blocked awaiting a timer? */
179 /* Remember the arguments to the program so we can "reboot" */
180 static char **main_args;
182 /* We have to be careful with barriers: our devices are all run in separate
183 * threads and so we need to make sure that changes visible to the Guest happen
184 * in precise order. */
185 #define wmb() __asm__ __volatile__("" : : : "memory")
187 /* Convert an iovec element to the given type.
189 * This is a fairly ugly trick: we need to know the size of the type and
190 * alignment requirement to check the pointer is kosher. It's also nice to
191 * have the name of the type in case we report failure.
193 * Typing those three things all the time is cumbersome and error prone, so we
194 * have a macro which sets them all up and passes to the real function. */
195 #define convert(iov, type) \
196 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
198 static void *_convert(struct iovec *iov, size_t size, size_t align,
201 if (iov->iov_len != size)
202 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
203 if ((unsigned long)iov->iov_base % align != 0)
204 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
205 return iov->iov_base;
208 /* Wrapper for the last available index. Makes it easier to change. */
209 #define lg_last_avail(vq) ((vq)->last_avail_idx)
211 /* The virtio configuration space is defined to be little-endian. x86 is
212 * little-endian too, but it's nice to be explicit so we have these helpers. */
213 #define cpu_to_le16(v16) (v16)
214 #define cpu_to_le32(v32) (v32)
215 #define cpu_to_le64(v64) (v64)
216 #define le16_to_cpu(v16) (v16)
217 #define le32_to_cpu(v32) (v32)
218 #define le64_to_cpu(v64) (v64)
220 /* Is this iovec empty? */
221 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
225 for (i = 0; i < num_iov; i++)
231 /* Take len bytes from the front of this iovec. */
232 static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
236 for (i = 0; i < num_iov; i++) {
239 used = iov[i].iov_len < len ? iov[i].iov_len : len;
240 iov[i].iov_base += used;
241 iov[i].iov_len -= used;
247 /* The device virtqueue descriptors are followed by feature bitmasks. */
248 static u8 *get_feature_bits(struct device *dev)
250 return (u8 *)(dev->desc + 1)
251 + dev->num_vq * sizeof(struct lguest_vqconfig);
254 /*L:100 The Launcher code itself takes us out into userspace, that scary place
255 * where pointers run wild and free! Unfortunately, like most userspace
256 * programs, it's quite boring (which is why everyone likes to hack on the
257 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
258 * will get you through this section. Or, maybe not.
260 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
261 * memory and stores it in "guest_base". In other words, Guest physical ==
262 * Launcher virtual with an offset.
264 * This can be tough to get your head around, but usually it just means that we
265 * use these trivial conversion functions when the Guest gives us it's
266 * "physical" addresses: */
267 static void *from_guest_phys(unsigned long addr)
269 return guest_base + addr;
272 static unsigned long to_guest_phys(const void *addr)
274 return (addr - guest_base);
278 * Loading the Kernel.
280 * We start with couple of simple helper routines. open_or_die() avoids
281 * error-checking code cluttering the callers: */
282 static int open_or_die(const char *name, int flags)
284 int fd = open(name, flags);
286 err(1, "Failed to open %s", name);
290 /* map_zeroed_pages() takes a number of pages. */
291 static void *map_zeroed_pages(unsigned int num)
293 int fd = open_or_die("/dev/zero", O_RDONLY);
296 /* We use a private mapping (ie. if we write to the page, it will be
298 addr = mmap(NULL, getpagesize() * num,
299 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
300 if (addr == MAP_FAILED)
301 err(1, "Mmaping %u pages of /dev/zero", num);
307 /* Get some more pages for a device. */
308 static void *get_pages(unsigned int num)
310 void *addr = from_guest_phys(guest_limit);
312 guest_limit += num * getpagesize();
313 if (guest_limit > guest_max)
314 errx(1, "Not enough memory for devices");
318 /* This routine is used to load the kernel or initrd. It tries mmap, but if
319 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
320 * it falls back to reading the memory in. */
321 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
325 /* We map writable even though for some segments are marked read-only.
326 * The kernel really wants to be writable: it patches its own
329 * MAP_PRIVATE means that the page won't be copied until a write is
330 * done to it. This allows us to share untouched memory between
332 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
333 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
336 /* pread does a seek and a read in one shot: saves a few lines. */
337 r = pread(fd, addr, len, offset);
339 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
342 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
343 * the Guest memory. ELF = Embedded Linking Format, which is the format used
344 * by all modern binaries on Linux including the kernel.
346 * The ELF headers give *two* addresses: a physical address, and a virtual
347 * address. We use the physical address; the Guest will map itself to the
350 * We return the starting address. */
351 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
353 Elf32_Phdr phdr[ehdr->e_phnum];
356 /* Sanity checks on the main ELF header: an x86 executable with a
357 * reasonable number of correctly-sized program headers. */
358 if (ehdr->e_type != ET_EXEC
359 || ehdr->e_machine != EM_386
360 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
361 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
362 errx(1, "Malformed elf header");
364 /* An ELF executable contains an ELF header and a number of "program"
365 * headers which indicate which parts ("segments") of the program to
368 /* We read in all the program headers at once: */
369 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
370 err(1, "Seeking to program headers");
371 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
372 err(1, "Reading program headers");
374 /* Try all the headers: there are usually only three. A read-only one,
375 * a read-write one, and a "note" section which we don't load. */
376 for (i = 0; i < ehdr->e_phnum; i++) {
377 /* If this isn't a loadable segment, we ignore it */
378 if (phdr[i].p_type != PT_LOAD)
381 verbose("Section %i: size %i addr %p\n",
382 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
384 /* We map this section of the file at its physical address. */
385 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
386 phdr[i].p_offset, phdr[i].p_filesz);
389 /* The entry point is given in the ELF header. */
390 return ehdr->e_entry;
393 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
394 * supposed to jump into it and it will unpack itself. We used to have to
395 * perform some hairy magic because the unpacking code scared me.
397 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
398 * a small patch to jump over the tricky bits in the Guest, so now we just read
399 * the funky header so we know where in the file to load, and away we go! */
400 static unsigned long load_bzimage(int fd)
402 struct boot_params boot;
404 /* Modern bzImages get loaded at 1M. */
405 void *p = from_guest_phys(0x100000);
407 /* Go back to the start of the file and read the header. It should be
408 * a Linux boot header (see Documentation/x86/i386/boot.txt) */
409 lseek(fd, 0, SEEK_SET);
410 read(fd, &boot, sizeof(boot));
412 /* Inside the setup_hdr, we expect the magic "HdrS" */
413 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
414 errx(1, "This doesn't look like a bzImage to me");
416 /* Skip over the extra sectors of the header. */
417 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
419 /* Now read everything into memory. in nice big chunks. */
420 while ((r = read(fd, p, 65536)) > 0)
423 /* Finally, code32_start tells us where to enter the kernel. */
424 return boot.hdr.code32_start;
427 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
428 * come wrapped up in the self-decompressing "bzImage" format. With a little
429 * work, we can load those, too. */
430 static unsigned long load_kernel(int fd)
434 /* Read in the first few bytes. */
435 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
436 err(1, "Reading kernel");
438 /* If it's an ELF file, it starts with "\177ELF" */
439 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
440 return map_elf(fd, &hdr);
442 /* Otherwise we assume it's a bzImage, and try to load it. */
443 return load_bzimage(fd);
446 /* This is a trivial little helper to align pages. Andi Kleen hated it because
447 * it calls getpagesize() twice: "it's dumb code."
449 * Kernel guys get really het up about optimization, even when it's not
450 * necessary. I leave this code as a reaction against that. */
451 static inline unsigned long page_align(unsigned long addr)
453 /* Add upwards and truncate downwards. */
454 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
457 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
458 * the kernel which the kernel can use to boot from without needing any
459 * drivers. Most distributions now use this as standard: the initrd contains
460 * the code to load the appropriate driver modules for the current machine.
462 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
463 * kernels. He sent me this (and tells me when I break it). */
464 static unsigned long load_initrd(const char *name, unsigned long mem)
470 ifd = open_or_die(name, O_RDONLY);
471 /* fstat() is needed to get the file size. */
472 if (fstat(ifd, &st) < 0)
473 err(1, "fstat() on initrd '%s'", name);
475 /* We map the initrd at the top of memory, but mmap wants it to be
476 * page-aligned, so we round the size up for that. */
477 len = page_align(st.st_size);
478 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
479 /* Once a file is mapped, you can close the file descriptor. It's a
480 * little odd, but quite useful. */
482 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
484 /* We return the initrd size. */
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 and the
510 * entry point for the Guest. */
511 static void tell_kernel(unsigned long start)
513 unsigned long args[] = { LHREQ_INITIALIZE,
514 (unsigned long)guest_base,
515 guest_limit / getpagesize(), start };
516 verbose("Guest: %p - %p (%#lx)\n",
517 guest_base, guest_base + guest_limit, guest_limit);
518 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
519 if (write(lguest_fd, args, sizeof(args)) < 0)
520 err(1, "Writing to /dev/lguest");
524 static void add_device_fd(int fd)
526 FD_SET(fd, &devices.infds);
527 if (fd > devices.max_infd)
528 devices.max_infd = fd;
534 * With console, block and network devices, we can have lots of input which we
535 * need to process. We could try to tell the kernel what file descriptors to
536 * watch, but handing a file descriptor mask through to the kernel is fairly
539 * Instead, we clone off a thread which watches the file descriptors and writes
540 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
541 * stop running the Guest. This causes the Launcher to return from the
542 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
543 * the LHREQ_BREAK and wake us up again.
545 * This, of course, is merely a different *kind* of icky.
547 * Given my well-known antipathy to threads, I'd prefer to use processes. But
548 * it's easier to share Guest memory with threads, and trivial to share the
549 * devices.infds as the Launcher changes it.
551 static int waker(void *unused)
553 /* Close the write end of the pipe: only the Launcher has it open. */
554 close(waker_fds.pipe[1]);
557 fd_set rfds = devices.infds;
558 unsigned long args[] = { LHREQ_BREAK, 1 };
559 unsigned int maxfd = devices.max_infd;
561 /* We also listen to the pipe from the Launcher. */
562 FD_SET(waker_fds.pipe[0], &rfds);
563 if (waker_fds.pipe[0] > maxfd)
564 maxfd = waker_fds.pipe[0];
566 /* Wait until input is ready from one of the devices. */
567 select(maxfd+1, &rfds, NULL, NULL, NULL);
569 /* Message from Launcher? */
570 if (FD_ISSET(waker_fds.pipe[0], &rfds)) {
572 /* If this fails, then assume Launcher has exited.
573 * Don't do anything on exit: we're just a thread! */
574 if (read(waker_fds.pipe[0], &c, 1) != 1)
579 /* Send LHREQ_BREAK command to snap the Launcher out of it. */
580 pwrite(lguest_fd, args, sizeof(args), cpu_id);
585 /* This routine just sets up a pipe to the Waker process. */
586 static void setup_waker(void)
588 /* This pipe is closed when Launcher dies, telling Waker. */
589 if (pipe(waker_fds.pipe) != 0)
590 err(1, "Creating pipe for Waker");
592 if (clone(waker, malloc(4096) + 4096, CLONE_VM | SIGCHLD, NULL) == -1)
593 err(1, "Creating Waker");
599 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
600 * We need to make sure it's not trying to reach into the Launcher itself, so
601 * we have a convenient routine which checks it and exits with an error message
602 * if something funny is going on:
604 static void *_check_pointer(unsigned long addr, unsigned int size,
607 /* We have to separately check addr and addr+size, because size could
608 * be huge and addr + size might wrap around. */
609 if (addr >= guest_limit || addr + size >= guest_limit)
610 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
611 /* We return a pointer for the caller's convenience, now we know it's
613 return from_guest_phys(addr);
615 /* A macro which transparently hands the line number to the real function. */
616 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
618 /* Each buffer in the virtqueues is actually a chain of descriptors. This
619 * function returns the next descriptor in the chain, or vq->vring.num if we're
621 static unsigned next_desc(struct virtqueue *vq, unsigned int i)
625 /* If this descriptor says it doesn't chain, we're done. */
626 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
627 return vq->vring.num;
629 /* Check they're not leading us off end of descriptors. */
630 next = vq->vring.desc[i].next;
631 /* Make sure compiler knows to grab that: we don't want it changing! */
634 if (next >= vq->vring.num)
635 errx(1, "Desc next is %u", next);
640 /* This looks in the virtqueue and for the first available buffer, and converts
641 * it to an iovec for convenient access. Since descriptors consist of some
642 * number of output then some number of input descriptors, it's actually two
643 * iovecs, but we pack them into one and note how many of each there were.
645 * This function returns the descriptor number found, or vq->vring.num (which
646 * is never a valid descriptor number) if none was found. */
647 static unsigned get_vq_desc(struct virtqueue *vq,
649 unsigned int *out_num, unsigned int *in_num)
651 unsigned int i, head;
654 /* Check it isn't doing very strange things with descriptor numbers. */
655 last_avail = lg_last_avail(vq);
656 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
657 errx(1, "Guest moved used index from %u to %u",
658 last_avail, vq->vring.avail->idx);
660 /* If there's nothing new since last we looked, return invalid. */
661 if (vq->vring.avail->idx == last_avail)
662 return vq->vring.num;
664 /* Grab the next descriptor number they're advertising, and increment
665 * the index we've seen. */
666 head = vq->vring.avail->ring[last_avail % vq->vring.num];
669 /* If their number is silly, that's a fatal mistake. */
670 if (head >= vq->vring.num)
671 errx(1, "Guest says index %u is available", head);
673 /* When we start there are none of either input nor output. */
674 *out_num = *in_num = 0;
678 /* Grab the first descriptor, and check it's OK. */
679 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
680 iov[*out_num + *in_num].iov_base
681 = check_pointer(vq->vring.desc[i].addr,
682 vq->vring.desc[i].len);
683 /* If this is an input descriptor, increment that count. */
684 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
687 /* If it's an output descriptor, they're all supposed
688 * to come before any input descriptors. */
690 errx(1, "Descriptor has out after in");
694 /* If we've got too many, that implies a descriptor loop. */
695 if (*out_num + *in_num > vq->vring.num)
696 errx(1, "Looped descriptor");
697 } while ((i = next_desc(vq, i)) != vq->vring.num);
702 /* After we've used one of their buffers, we tell them about it. We'll then
703 * want to send them an interrupt, using trigger_irq(). */
704 static void add_used(struct virtqueue *vq, unsigned int head, int len)
706 struct vring_used_elem *used;
708 /* The virtqueue contains a ring of used buffers. Get a pointer to the
709 * next entry in that used ring. */
710 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
713 /* Make sure buffer is written before we update index. */
715 vq->vring.used->idx++;
718 /* This actually sends the interrupt for this virtqueue */
719 static void trigger_irq(struct virtqueue *vq)
721 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
723 /* If they don't want an interrupt, don't send one, unless empty. */
724 if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
725 && lg_last_avail(vq) != vq->vring.avail->idx)
728 /* Send the Guest an interrupt tell them we used something up. */
729 if (write(lguest_fd, buf, sizeof(buf)) != 0)
730 err(1, "Triggering irq %i", vq->config.irq);
733 /* And here's the combo meal deal. Supersize me! */
734 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
736 add_used(vq, head, len);
743 * Here is the input terminal setting we save, and the routine to restore them
744 * on exit so the user gets their terminal back. */
745 static struct termios orig_term;
746 static void restore_term(void)
748 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
751 /* We associate some data with the console for our exit hack. */
754 /* How many times have they hit ^C? */
756 /* When did they start? */
757 struct timeval start;
760 /* This is the routine which handles console input (ie. stdin). */
761 static bool handle_console_input(struct device *dev)
764 unsigned int head, in_num, out_num;
765 struct iovec iov[dev->vq->vring.num];
766 struct console_abort *abort = dev->priv;
768 /* First we need a console buffer from the Guests's input virtqueue. */
769 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
771 /* If they're not ready for input, stop listening to this file
772 * descriptor. We'll start again once they add an input buffer. */
773 if (head == dev->vq->vring.num)
777 errx(1, "Output buffers in console in queue?");
779 /* This is why we convert to iovecs: the readv() call uses them, and so
780 * it reads straight into the Guest's buffer. */
781 len = readv(dev->fd, iov, in_num);
783 /* This implies that the console is closed, is /dev/null, or
784 * something went terribly wrong. */
785 warnx("Failed to get console input, ignoring console.");
786 /* Put the input terminal back. */
788 /* Remove callback from input vq, so it doesn't restart us. */
789 dev->vq->handle_output = NULL;
790 /* Stop listening to this fd: don't call us again. */
794 /* Tell the Guest about the new input. */
795 add_used_and_trigger(dev->vq, head, len);
797 /* Three ^C within one second? Exit.
799 * This is such a hack, but works surprisingly well. Each ^C has to be
800 * in a buffer by itself, so they can't be too fast. But we check that
801 * we get three within about a second, so they can't be too slow. */
802 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
804 gettimeofday(&abort->start, NULL);
805 else if (abort->count == 3) {
807 gettimeofday(&now, NULL);
808 if (now.tv_sec <= abort->start.tv_sec+1) {
809 unsigned long args[] = { LHREQ_BREAK, 0 };
810 /* Close the fd so Waker will know it has to
812 close(waker_fds.pipe[1]);
813 /* Just in case Waker is blocked in BREAK, send
815 write(lguest_fd, args, sizeof(args));
821 /* Any other key resets the abort counter. */
824 /* Everything went OK! */
828 /* Handling output for console is simple: we just get all the output buffers
829 * and write them to stdout. */
830 static void handle_console_output(struct virtqueue *vq, bool timeout)
832 unsigned int head, out, in;
834 struct iovec iov[vq->vring.num];
836 /* Keep getting output buffers from the Guest until we run out. */
837 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
839 errx(1, "Input buffers in output queue?");
840 len = writev(STDOUT_FILENO, iov, out);
841 add_used_and_trigger(vq, head, len);
845 /* This is called when we no longer want to hear about Guest changes to a
846 * virtqueue. This is more efficient in high-traffic cases, but it means we
847 * have to set a timer to check if any more changes have occurred. */
848 static void block_vq(struct virtqueue *vq)
850 struct itimerval itm;
852 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
855 itm.it_interval.tv_sec = 0;
856 itm.it_interval.tv_usec = 0;
857 itm.it_value.tv_sec = 0;
858 itm.it_value.tv_usec = timeout_usec;
860 setitimer(ITIMER_REAL, &itm, NULL);
866 * Handling output for network is also simple: we get all the output buffers
867 * and write them (ignoring the first element) to this device's file descriptor
870 static void handle_net_output(struct virtqueue *vq, bool timeout)
872 unsigned int head, out, in, num = 0;
874 struct iovec iov[vq->vring.num];
875 static int last_timeout_num;
877 /* Keep getting output buffers from the Guest until we run out. */
878 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
880 errx(1, "Input buffers in output queue?");
881 len = writev(vq->dev->fd, iov, out);
883 err(1, "Writing network packet to tun");
884 add_used_and_trigger(vq, head, len);
888 /* Block further kicks and set up a timer if we saw anything. */
892 /* We never quite know how long should we wait before we check the
893 * queue again for more packets. We start at 500 microseconds, and if
894 * we get fewer packets than last time, we assume we made the timeout
895 * too small and increase it by 10 microseconds. Otherwise, we drop it
896 * by one microsecond every time. It seems to work well enough. */
898 if (num < last_timeout_num)
900 else if (timeout_usec > 1)
902 last_timeout_num = num;
906 /* This is where we handle a packet coming in from the tun device to our
908 static bool handle_tun_input(struct device *dev)
910 unsigned int head, in_num, out_num;
912 struct iovec iov[dev->vq->vring.num];
914 /* First we need a network buffer from the Guests's recv virtqueue. */
915 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
916 if (head == dev->vq->vring.num) {
917 /* Now, it's expected that if we try to send a packet too
918 * early, the Guest won't be ready yet. Wait until the device
919 * status says it's ready. */
920 /* FIXME: Actually want DRIVER_ACTIVE here. */
922 /* Now tell it we want to know if new things appear. */
923 dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
926 /* We'll turn this back on if input buffers are registered. */
929 errx(1, "Output buffers in network recv queue?");
931 /* Read the packet from the device directly into the Guest's buffer. */
932 len = readv(dev->fd, iov, in_num);
934 err(1, "reading network");
936 /* Tell the Guest about the new packet. */
937 add_used_and_trigger(dev->vq, head, len);
939 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
940 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
941 head != dev->vq->vring.num ? "sent" : "discarded");
947 /*L:215 This is the callback attached to the network and console input
948 * virtqueues: it ensures we try again, in case we stopped console or net
949 * delivery because Guest didn't have any buffers. */
950 static void enable_fd(struct virtqueue *vq, bool timeout)
952 add_device_fd(vq->dev->fd);
953 /* Snap the Waker out of its select loop. */
954 write(waker_fds.pipe[1], "", 1);
957 static void net_enable_fd(struct virtqueue *vq, bool timeout)
959 /* We don't need to know again when Guest refills receive buffer. */
960 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
961 enable_fd(vq, timeout);
964 /* When the Guest tells us they updated the status field, we handle it. */
965 static void update_device_status(struct device *dev)
967 struct virtqueue *vq;
969 /* This is a reset. */
970 if (dev->desc->status == 0) {
971 verbose("Resetting device %s\n", dev->name);
973 /* Clear any features they've acked. */
974 memset(get_feature_bits(dev) + dev->feature_len, 0,
977 /* Zero out the virtqueues. */
978 for (vq = dev->vq; vq; vq = vq->next) {
979 memset(vq->vring.desc, 0,
980 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
981 lg_last_avail(vq) = 0;
983 } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
984 warnx("Device %s configuration FAILED", dev->name);
985 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
988 verbose("Device %s OK: offered", dev->name);
989 for (i = 0; i < dev->feature_len; i++)
990 verbose(" %02x", get_feature_bits(dev)[i]);
991 verbose(", accepted");
992 for (i = 0; i < dev->feature_len; i++)
993 verbose(" %02x", get_feature_bits(dev)
994 [dev->feature_len+i]);
1001 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1002 static void handle_output(unsigned long addr)
1005 struct virtqueue *vq;
1007 /* Check each device and virtqueue. */
1008 for (i = devices.dev; i; i = i->next) {
1009 /* Notifications to device descriptors update device status. */
1010 if (from_guest_phys(addr) == i->desc) {
1011 update_device_status(i);
1015 /* Notifications to virtqueues mean output has occurred. */
1016 for (vq = i->vq; vq; vq = vq->next) {
1017 if (vq->config.pfn != addr/getpagesize())
1020 /* Guest should acknowledge (and set features!) before
1021 * using the device. */
1022 if (i->desc->status == 0) {
1023 warnx("%s gave early output", i->name);
1027 if (strcmp(vq->dev->name, "console") != 0)
1028 verbose("Output to %s\n", vq->dev->name);
1029 if (vq->handle_output)
1030 vq->handle_output(vq, false);
1035 /* Early console write is done using notify on a nul-terminated string
1036 * in Guest memory. */
1037 if (addr >= guest_limit)
1038 errx(1, "Bad NOTIFY %#lx", addr);
1040 write(STDOUT_FILENO, from_guest_phys(addr),
1041 strnlen(from_guest_phys(addr), guest_limit - addr));
1044 static void handle_timeout(void)
1048 struct virtqueue *vq;
1050 /* Clear the pipe */
1051 read(timeoutpipe[0], buf, sizeof(buf));
1053 /* Check each device and virtqueue: flush blocked ones. */
1054 for (i = devices.dev; i; i = i->next) {
1055 for (vq = i->vq; vq; vq = vq->next) {
1059 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
1060 vq->blocked = false;
1061 if (vq->handle_output)
1062 vq->handle_output(vq, true);
1067 /* This is called when the Waker wakes us up: check for incoming file
1069 static void handle_input(void)
1071 /* select() wants a zeroed timeval to mean "don't wait". */
1072 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
1076 fd_set fds = devices.infds;
1079 num = select(devices.max_infd+1, &fds, NULL, NULL, &poll);
1080 /* Could get interrupted */
1083 /* If nothing is ready, we're done. */
1087 /* Otherwise, call the device(s) which have readable file
1088 * descriptors and a method of handling them. */
1089 for (i = devices.dev; i; i = i->next) {
1090 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
1091 if (i->handle_input(i))
1094 /* If handle_input() returns false, it means we
1095 * should no longer service it. Networking and
1096 * console do this when there's no input
1097 * buffers to deliver into. Console also uses
1098 * it when it discovers that stdin is closed. */
1099 FD_CLR(i->fd, &devices.infds);
1103 /* Is this the timeout fd? */
1104 if (FD_ISSET(timeoutpipe[0], &fds))
1112 * All devices need a descriptor so the Guest knows it exists, and a "struct
1113 * device" so the Launcher can keep track of it. We have common helper
1114 * routines to allocate and manage them.
1117 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1118 * number of virtqueue descriptors, then two sets of feature bits, then an
1119 * array of configuration bytes. This routine returns the configuration
1121 static u8 *device_config(const struct device *dev)
1123 return (void *)(dev->desc + 1)
1124 + dev->num_vq * sizeof(struct lguest_vqconfig)
1125 + dev->feature_len * 2;
1128 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1129 * table page just above the Guest's normal memory. It returns a pointer to
1130 * that descriptor. */
1131 static struct lguest_device_desc *new_dev_desc(u16 type)
1133 struct lguest_device_desc d = { .type = type };
1136 /* Figure out where the next device config is, based on the last one. */
1137 if (devices.lastdev)
1138 p = device_config(devices.lastdev)
1139 + devices.lastdev->desc->config_len;
1141 p = devices.descpage;
1143 /* We only have one page for all the descriptors. */
1144 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1145 errx(1, "Too many devices");
1147 /* p might not be aligned, so we memcpy in. */
1148 return memcpy(p, &d, sizeof(d));
1151 /* Each device descriptor is followed by the description of its virtqueues. We
1152 * specify how many descriptors the virtqueue is to have. */
1153 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1154 void (*handle_output)(struct virtqueue *, bool))
1157 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1160 /* First we need some memory for this virtqueue. */
1161 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1163 p = get_pages(pages);
1165 /* Initialize the virtqueue */
1167 vq->last_avail_idx = 0;
1169 vq->blocked = false;
1171 /* Initialize the configuration. */
1172 vq->config.num = num_descs;
1173 vq->config.irq = devices.next_irq++;
1174 vq->config.pfn = to_guest_phys(p) / getpagesize();
1176 /* Initialize the vring. */
1177 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1179 /* Append virtqueue to this device's descriptor. We use
1180 * device_config() to get the end of the device's current virtqueues;
1181 * we check that we haven't added any config or feature information
1182 * yet, otherwise we'd be overwriting them. */
1183 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1184 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1186 dev->desc->num_vq++;
1188 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1190 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1192 for (i = &dev->vq; *i; i = &(*i)->next);
1195 /* Set the routine to call when the Guest does something to this
1197 vq->handle_output = handle_output;
1199 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1200 * don't have a handler */
1202 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1205 /* The first half of the feature bitmask is for us to advertise features. The
1206 * second half is for the Guest to accept features. */
1207 static void add_feature(struct device *dev, unsigned bit)
1209 u8 *features = get_feature_bits(dev);
1211 /* We can't extend the feature bits once we've added config bytes */
1212 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1213 assert(dev->desc->config_len == 0);
1214 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1217 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1220 /* This routine sets the configuration fields for an existing device's
1221 * descriptor. It only works for the last device, but that's OK because that's
1223 static void set_config(struct device *dev, unsigned len, const void *conf)
1225 /* Check we haven't overflowed our single page. */
1226 if (device_config(dev) + len > devices.descpage + getpagesize())
1227 errx(1, "Too many devices");
1229 /* Copy in the config information, and store the length. */
1230 memcpy(device_config(dev), conf, len);
1231 dev->desc->config_len = len;
1234 /* This routine does all the creation and setup of a new device, including
1235 * calling new_dev_desc() to allocate the descriptor and device memory.
1237 * See what I mean about userspace being boring? */
1238 static struct device *new_device(const char *name, u16 type, int fd,
1239 bool (*handle_input)(struct device *))
1241 struct device *dev = malloc(sizeof(*dev));
1243 /* Now we populate the fields one at a time. */
1245 /* If we have an input handler for this file descriptor, then we add it
1246 * to the device_list's fdset and maxfd. */
1248 add_device_fd(dev->fd);
1249 dev->desc = new_dev_desc(type);
1250 dev->handle_input = handle_input;
1254 dev->feature_len = 0;
1257 /* Append to device list. Prepending to a single-linked list is
1258 * easier, but the user expects the devices to be arranged on the bus
1259 * in command-line order. The first network device on the command line
1260 * is eth0, the first block device /dev/vda, etc. */
1261 if (devices.lastdev)
1262 devices.lastdev->next = dev;
1265 devices.lastdev = dev;
1270 /* Our first setup routine is the console. It's a fairly simple device, but
1271 * UNIX tty handling makes it uglier than it could be. */
1272 static void setup_console(void)
1276 /* If we can save the initial standard input settings... */
1277 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1278 struct termios term = orig_term;
1279 /* Then we turn off echo, line buffering and ^C etc. We want a
1280 * raw input stream to the Guest. */
1281 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1282 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1283 /* If we exit gracefully, the original settings will be
1284 * restored so the user can see what they're typing. */
1285 atexit(restore_term);
1288 dev = new_device("console", VIRTIO_ID_CONSOLE,
1289 STDIN_FILENO, handle_console_input);
1290 /* We store the console state in dev->priv, and initialize it. */
1291 dev->priv = malloc(sizeof(struct console_abort));
1292 ((struct console_abort *)dev->priv)->count = 0;
1294 /* The console needs two virtqueues: the input then the output. When
1295 * they put something the input queue, we make sure we're listening to
1296 * stdin. When they put something in the output queue, we write it to
1298 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1299 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1301 verbose("device %u: console\n", devices.device_num++);
1305 static void timeout_alarm(int sig)
1307 write(timeoutpipe[1], "", 1);
1310 static void setup_timeout(void)
1312 if (pipe(timeoutpipe) != 0)
1313 err(1, "Creating timeout pipe");
1315 if (fcntl(timeoutpipe[1], F_SETFL,
1316 fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0)
1317 err(1, "Making timeout pipe nonblocking");
1319 add_device_fd(timeoutpipe[0]);
1320 signal(SIGALRM, timeout_alarm);
1323 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1324 * --sharenet=<name> option which opens or creates a named pipe. This can be
1325 * used to send packets to another guest in a 1:1 manner.
1327 * More sopisticated is to use one of the tools developed for project like UML
1330 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1331 * completely generic ("here's my vring, attach to your vring") and would work
1332 * for any traffic. Of course, namespace and permissions issues need to be
1333 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1334 * multiple inter-guest channels behind one interface, although it would
1335 * require some manner of hotplugging new virtio channels.
1337 * Finally, we could implement a virtio network switch in the kernel. :*/
1339 static u32 str2ip(const char *ipaddr)
1343 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1344 errx(1, "Failed to parse IP address '%s'", ipaddr);
1345 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1348 static void str2mac(const char *macaddr, unsigned char mac[6])
1351 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1352 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1353 errx(1, "Failed to parse mac address '%s'", macaddr);
1362 /* This code is "adapted" from libbridge: it attaches the Host end of the
1363 * network device to the bridge device specified by the command line.
1365 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1366 * dislike bridging), and I just try not to break it. */
1367 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1373 errx(1, "must specify bridge name");
1375 ifidx = if_nametoindex(if_name);
1377 errx(1, "interface %s does not exist!", if_name);
1379 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1380 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1381 ifr.ifr_ifindex = ifidx;
1382 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1383 err(1, "can't add %s to bridge %s", if_name, br_name);
1386 /* This sets up the Host end of the network device with an IP address, brings
1387 * it up so packets will flow, the copies the MAC address into the hwaddr
1389 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1392 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1394 memset(&ifr, 0, sizeof(ifr));
1395 strcpy(ifr.ifr_name, tapif);
1397 /* Don't read these incantations. Just cut & paste them like I did! */
1398 sin->sin_family = AF_INET;
1399 sin->sin_addr.s_addr = htonl(ipaddr);
1400 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1401 err(1, "Setting %s interface address", tapif);
1402 ifr.ifr_flags = IFF_UP;
1403 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1404 err(1, "Bringing interface %s up", tapif);
1407 static int get_tun_device(char tapif[IFNAMSIZ])
1412 /* Start with this zeroed. Messy but sure. */
1413 memset(&ifr, 0, sizeof(ifr));
1415 /* We open the /dev/net/tun device and tell it we want a tap device. A
1416 * tap device is like a tun device, only somehow different. To tell
1417 * the truth, I completely blundered my way through this code, but it
1419 netfd = open_or_die("/dev/net/tun", O_RDWR);
1420 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1421 strcpy(ifr.ifr_name, "tap%d");
1422 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1423 err(1, "configuring /dev/net/tun");
1425 if (ioctl(netfd, TUNSETOFFLOAD,
1426 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1427 err(1, "Could not set features for tun device");
1429 /* We don't need checksums calculated for packets coming in this
1430 * device: trust us! */
1431 ioctl(netfd, TUNSETNOCSUM, 1);
1433 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1437 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1438 * routing, but the principle is the same: it uses the "tun" device to inject
1439 * packets into the Host as if they came in from a normal network card. We
1440 * just shunt packets between the Guest and the tun device. */
1441 static void setup_tun_net(char *arg)
1445 u32 ip = INADDR_ANY;
1446 bool bridging = false;
1447 char tapif[IFNAMSIZ], *p;
1448 struct virtio_net_config conf;
1450 netfd = get_tun_device(tapif);
1452 /* First we create a new network device. */
1453 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1455 /* Network devices need a receive and a send queue, just like
1457 add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd);
1458 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1460 /* We need a socket to perform the magic network ioctls to bring up the
1461 * tap interface, connect to the bridge etc. Any socket will do! */
1462 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1464 err(1, "opening IP socket");
1466 /* If the command line was --tunnet=bridge:<name> do bridging. */
1467 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1468 arg += strlen(BRIDGE_PFX);
1472 /* A mac address may follow the bridge name or IP address */
1473 p = strchr(arg, ':');
1475 str2mac(p+1, conf.mac);
1476 add_feature(dev, VIRTIO_NET_F_MAC);
1480 /* arg is now either an IP address or a bridge name */
1482 add_to_bridge(ipfd, tapif, arg);
1486 /* Set up the tun device. */
1487 configure_device(ipfd, tapif, ip);
1489 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1490 /* Expect Guest to handle everything except UFO */
1491 add_feature(dev, VIRTIO_NET_F_CSUM);
1492 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1493 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1494 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1495 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1496 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1497 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1498 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1499 set_config(dev, sizeof(conf), &conf);
1501 /* We don't need the socket any more; setup is done. */
1504 devices.device_num++;
1507 verbose("device %u: tun %s attached to bridge: %s\n",
1508 devices.device_num, tapif, arg);
1510 verbose("device %u: tun %s: %s\n",
1511 devices.device_num, tapif, arg);
1514 /* Our block (disk) device should be really simple: the Guest asks for a block
1515 * number and we read or write that position in the file. Unfortunately, that
1516 * was amazingly slow: the Guest waits until the read is finished before
1517 * running anything else, even if it could have been doing useful work.
1519 * We could use async I/O, except it's reputed to suck so hard that characters
1520 * actually go missing from your code when you try to use it.
1522 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1524 /* This hangs off device->priv. */
1527 /* The size of the file. */
1530 /* The file descriptor for the file. */
1533 /* IO thread listens on this file descriptor [0]. */
1536 /* IO thread writes to this file descriptor to mark it done, then
1537 * Launcher triggers interrupt to Guest. */
1544 * Remember that the block device is handled by a separate I/O thread. We head
1545 * straight into the core of that thread here:
1547 static bool service_io(struct device *dev)
1549 struct vblk_info *vblk = dev->priv;
1550 unsigned int head, out_num, in_num, wlen;
1553 struct virtio_blk_outhdr *out;
1554 struct iovec iov[dev->vq->vring.num];
1557 /* See if there's a request waiting. If not, nothing to do. */
1558 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1559 if (head == dev->vq->vring.num)
1562 /* Every block request should contain at least one output buffer
1563 * (detailing the location on disk and the type of request) and one
1564 * input buffer (to hold the result). */
1565 if (out_num == 0 || in_num == 0)
1566 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1567 head, out_num, in_num);
1569 out = convert(&iov[0], struct virtio_blk_outhdr);
1570 in = convert(&iov[out_num+in_num-1], u8);
1571 off = out->sector * 512;
1573 /* The block device implements "barriers", where the Guest indicates
1574 * that it wants all previous writes to occur before this write. We
1575 * don't have a way of asking our kernel to do a barrier, so we just
1576 * synchronize all the data in the file. Pretty poor, no? */
1577 if (out->type & VIRTIO_BLK_T_BARRIER)
1578 fdatasync(vblk->fd);
1580 /* In general the virtio block driver is allowed to try SCSI commands.
1581 * It'd be nice if we supported eject, for example, but we don't. */
1582 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1583 fprintf(stderr, "Scsi commands unsupported\n");
1584 *in = VIRTIO_BLK_S_UNSUPP;
1586 } else if (out->type & VIRTIO_BLK_T_OUT) {
1589 /* Move to the right location in the block file. This can fail
1590 * if they try to write past end. */
1591 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1592 err(1, "Bad seek to sector %llu", out->sector);
1594 ret = writev(vblk->fd, iov+1, out_num-1);
1595 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1597 /* Grr... Now we know how long the descriptor they sent was, we
1598 * make sure they didn't try to write over the end of the block
1599 * file (possibly extending it). */
1600 if (ret > 0 && off + ret > vblk->len) {
1601 /* Trim it back to the correct length */
1602 ftruncate64(vblk->fd, vblk->len);
1603 /* Die, bad Guest, die. */
1604 errx(1, "Write past end %llu+%u", off, ret);
1607 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1611 /* Move to the right location in the block file. This can fail
1612 * if they try to read past end. */
1613 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1614 err(1, "Bad seek to sector %llu", out->sector);
1616 ret = readv(vblk->fd, iov+1, in_num-1);
1617 verbose("READ from sector %llu: %i\n", out->sector, ret);
1619 wlen = sizeof(*in) + ret;
1620 *in = VIRTIO_BLK_S_OK;
1623 *in = VIRTIO_BLK_S_IOERR;
1627 /* OK, so we noted that it was pretty poor to use an fdatasync as a
1628 * barrier. But Christoph Hellwig points out that we need a sync
1629 * *afterwards* as well: "Barriers specify no reordering to the front
1630 * or the back." And Jens Axboe confirmed it, so here we are: */
1631 if (out->type & VIRTIO_BLK_T_BARRIER)
1632 fdatasync(vblk->fd);
1634 /* We can't trigger an IRQ, because we're not the Launcher. It does
1635 * that when we tell it we're done. */
1636 add_used(dev->vq, head, wlen);
1640 /* This is the thread which actually services the I/O. */
1641 static int io_thread(void *_dev)
1643 struct device *dev = _dev;
1644 struct vblk_info *vblk = dev->priv;
1647 /* Close other side of workpipe so we get 0 read when main dies. */
1648 close(vblk->workpipe[1]);
1649 /* Close the other side of the done_fd pipe. */
1652 /* When this read fails, it means Launcher died, so we follow. */
1653 while (read(vblk->workpipe[0], &c, 1) == 1) {
1654 /* We acknowledge each request immediately to reduce latency,
1655 * rather than waiting until we've done them all. I haven't
1656 * measured to see if it makes any difference.
1658 * That would be an interesting test, wouldn't it? You could
1659 * also try having more than one I/O thread. */
1660 while (service_io(dev))
1661 write(vblk->done_fd, &c, 1);
1666 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1667 * when that thread tells us it's completed some I/O. */
1668 static bool handle_io_finish(struct device *dev)
1672 /* If the I/O thread died, presumably it printed the error, so we
1674 if (read(dev->fd, &c, 1) != 1)
1677 /* It did some work, so trigger the irq. */
1678 trigger_irq(dev->vq);
1682 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1683 static void handle_virtblk_output(struct virtqueue *vq, bool timeout)
1685 struct vblk_info *vblk = vq->dev->priv;
1688 /* Wake up I/O thread and tell it to go to work! */
1689 if (write(vblk->workpipe[1], &c, 1) != 1)
1690 /* Presumably it indicated why it died. */
1694 /*L:198 This actually sets up a virtual block device. */
1695 static void setup_block_file(const char *filename)
1699 struct vblk_info *vblk;
1701 struct virtio_blk_config conf;
1703 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1706 /* The device responds to return from I/O thread. */
1707 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1709 /* The device has one virtqueue, where the Guest places requests. */
1710 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1712 /* Allocate the room for our own bookkeeping */
1713 vblk = dev->priv = malloc(sizeof(*vblk));
1715 /* First we open the file and store the length. */
1716 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1717 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1719 /* We support barriers. */
1720 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1722 /* Tell Guest how many sectors this device has. */
1723 conf.capacity = cpu_to_le64(vblk->len / 512);
1725 /* Tell Guest not to put in too many descriptors at once: two are used
1726 * for the in and out elements. */
1727 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1728 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1730 set_config(dev, sizeof(conf), &conf);
1732 /* The I/O thread writes to this end of the pipe when done. */
1733 vblk->done_fd = p[1];
1735 /* This is the second pipe, which is how we tell the I/O thread about
1737 pipe(vblk->workpipe);
1739 /* Create stack for thread and run it. Since stack grows upwards, we
1740 * point the stack pointer to the end of this region. */
1741 stack = malloc(32768);
1742 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1743 * becoming a zombie. */
1744 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
1745 err(1, "Creating clone");
1747 /* We don't need to keep the I/O thread's end of the pipes open. */
1748 close(vblk->done_fd);
1749 close(vblk->workpipe[0]);
1751 verbose("device %u: virtblock %llu sectors\n",
1752 devices.device_num, le64_to_cpu(conf.capacity));
1755 /* Our random number generator device reads from /dev/random into the Guest's
1756 * input buffers. The usual case is that the Guest doesn't want random numbers
1757 * and so has no buffers although /dev/random is still readable, whereas
1758 * console is the reverse.
1760 * The same logic applies, however. */
1761 static bool handle_rng_input(struct device *dev)
1764 unsigned int head, in_num, out_num, totlen = 0;
1765 struct iovec iov[dev->vq->vring.num];
1767 /* First we need a buffer from the Guests's virtqueue. */
1768 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1770 /* If they're not ready for input, stop listening to this file
1771 * descriptor. We'll start again once they add an input buffer. */
1772 if (head == dev->vq->vring.num)
1776 errx(1, "Output buffers in rng?");
1778 /* This is why we convert to iovecs: the readv() call uses them, and so
1779 * it reads straight into the Guest's buffer. We loop to make sure we
1781 while (!iov_empty(iov, in_num)) {
1782 len = readv(dev->fd, iov, in_num);
1784 err(1, "Read from /dev/random gave %i", len);
1785 iov_consume(iov, in_num, len);
1789 /* Tell the Guest about the new input. */
1790 add_used_and_trigger(dev->vq, head, totlen);
1792 /* Everything went OK! */
1796 /* And this creates a "hardware" random number device for the Guest. */
1797 static void setup_rng(void)
1802 fd = open_or_die("/dev/random", O_RDONLY);
1804 /* The device responds to return from I/O thread. */
1805 dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input);
1807 /* The device has one virtqueue, where the Guest places inbufs. */
1808 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1810 verbose("device %u: rng\n", devices.device_num++);
1812 /* That's the end of device setup. */
1814 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1815 static void __attribute__((noreturn)) restart_guest(void)
1819 /* Since we don't track all open fds, we simply close everything beyond
1821 for (i = 3; i < FD_SETSIZE; i++)
1824 /* The exec automatically gets rid of the I/O and Waker threads. */
1825 execv(main_args[0], main_args);
1826 err(1, "Could not exec %s", main_args[0]);
1829 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1830 * its input and output, and finally, lays it to rest. */
1831 static void __attribute__((noreturn)) run_guest(void)
1834 unsigned long args[] = { LHREQ_BREAK, 0 };
1835 unsigned long notify_addr;
1838 /* We read from the /dev/lguest device to run the Guest. */
1839 readval = pread(lguest_fd, ¬ify_addr,
1840 sizeof(notify_addr), cpu_id);
1842 /* One unsigned long means the Guest did HCALL_NOTIFY */
1843 if (readval == sizeof(notify_addr)) {
1844 verbose("Notify on address %#lx\n", notify_addr);
1845 handle_output(notify_addr);
1847 /* ENOENT means the Guest died. Reading tells us why. */
1848 } else if (errno == ENOENT) {
1849 char reason[1024] = { 0 };
1850 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1851 errx(1, "%s", reason);
1852 /* ERESTART means that we need to reboot the guest */
1853 } else if (errno == ERESTART) {
1855 /* EAGAIN means a signal (timeout).
1856 * Anything else means a bug or incompatible change. */
1857 } else if (errno != EAGAIN)
1858 err(1, "Running guest failed");
1860 /* Only service input on thread for CPU 0. */
1864 /* Service input, then unset the BREAK to release the Waker. */
1866 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1867 err(1, "Resetting break");
1871 * This is the end of the Launcher. The good news: we are over halfway
1872 * through! The bad news: the most fiendish part of the code still lies ahead
1875 * Are you ready? Take a deep breath and join me in the core of the Host, in
1879 static struct option opts[] = {
1880 { "verbose", 0, NULL, 'v' },
1881 { "tunnet", 1, NULL, 't' },
1882 { "block", 1, NULL, 'b' },
1883 { "rng", 0, NULL, 'r' },
1884 { "initrd", 1, NULL, 'i' },
1887 static void usage(void)
1889 errx(1, "Usage: lguest [--verbose] "
1890 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1891 "|--block=<filename>|--initrd=<filename>]...\n"
1892 "<mem-in-mb> vmlinux [args...]");
1895 /*L:105 The main routine is where the real work begins: */
1896 int main(int argc, char *argv[])
1898 /* Memory, top-level pagetable, code startpoint and size of the
1899 * (optional) initrd. */
1900 unsigned long mem = 0, start, initrd_size = 0;
1901 /* Two temporaries. */
1903 /* The boot information for the Guest. */
1904 struct boot_params *boot;
1905 /* If they specify an initrd file to load. */
1906 const char *initrd_name = NULL;
1908 /* Save the args: we "reboot" by execing ourselves again. */
1910 /* We don't "wait" for the children, so prevent them from becoming
1912 signal(SIGCHLD, SIG_IGN);
1914 /* First we initialize the device list. Since console and network
1915 * device receive input from a file descriptor, we keep an fdset
1916 * (infds) and the maximum fd number (max_infd) with the head of the
1917 * list. We also keep a pointer to the last device. Finally, we keep
1918 * the next interrupt number to use for devices (1: remember that 0 is
1919 * used by the timer). */
1920 FD_ZERO(&devices.infds);
1921 devices.max_infd = -1;
1922 devices.lastdev = NULL;
1923 devices.next_irq = 1;
1926 /* We need to know how much memory so we can set up the device
1927 * descriptor and memory pages for the devices as we parse the command
1928 * line. So we quickly look through the arguments to find the amount
1930 for (i = 1; i < argc; i++) {
1931 if (argv[i][0] != '-') {
1932 mem = atoi(argv[i]) * 1024 * 1024;
1933 /* We start by mapping anonymous pages over all of
1934 * guest-physical memory range. This fills it with 0,
1935 * and ensures that the Guest won't be killed when it
1936 * tries to access it. */
1937 guest_base = map_zeroed_pages(mem / getpagesize()
1940 guest_max = mem + DEVICE_PAGES*getpagesize();
1941 devices.descpage = get_pages(1);
1946 /* The options are fairly straight-forward */
1947 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1953 setup_tun_net(optarg);
1956 setup_block_file(optarg);
1962 initrd_name = optarg;
1965 warnx("Unknown argument %s", argv[optind]);
1969 /* After the other arguments we expect memory and kernel image name,
1970 * followed by command line arguments for the kernel. */
1971 if (optind + 2 > argc)
1974 verbose("Guest base is at %p\n", guest_base);
1976 /* We always have a console device */
1979 /* We can timeout waiting for Guest network transmit. */
1982 /* Now we load the kernel */
1983 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1985 /* Boot information is stashed at physical address 0 */
1986 boot = from_guest_phys(0);
1988 /* Map the initrd image if requested (at top of physical memory) */
1990 initrd_size = load_initrd(initrd_name, mem);
1991 /* These are the location in the Linux boot header where the
1992 * start and size of the initrd are expected to be found. */
1993 boot->hdr.ramdisk_image = mem - initrd_size;
1994 boot->hdr.ramdisk_size = initrd_size;
1995 /* The bootloader type 0xFF means "unknown"; that's OK. */
1996 boot->hdr.type_of_loader = 0xFF;
1999 /* The Linux boot header contains an "E820" memory map: ours is a
2000 * simple, single region. */
2001 boot->e820_entries = 1;
2002 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
2003 /* The boot header contains a command line pointer: we put the command
2004 * line after the boot header. */
2005 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2006 /* We use a simple helper to copy the arguments separated by spaces. */
2007 concat((char *)(boot + 1), argv+optind+2);
2009 /* Boot protocol version: 2.07 supports the fields for lguest. */
2010 boot->hdr.version = 0x207;
2012 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2013 boot->hdr.hardware_subarch = 1;
2015 /* Tell the entry path not to try to reload segment registers. */
2016 boot->hdr.loadflags |= KEEP_SEGMENTS;
2018 /* We tell the kernel to initialize the Guest: this returns the open
2019 * /dev/lguest file descriptor. */
2022 /* We clone off a thread, which wakes the Launcher whenever one of the
2023 * input file descriptors needs attention. We call this the Waker, and
2024 * we'll cover it in a moment. */
2027 /* Finally, run the Guest. This doesn't return. */
2033 * Mastery is done: you now know everything I do.
2035 * But surely you have seen code, features and bugs in your wanderings which
2036 * you now yearn to attack? That is the real game, and I look forward to you
2037 * patching and forking lguest into the Your-Name-Here-visor.
2039 * Farewell, and good coding!