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-x86/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 /* The pipe to send commands to the waker process */
81 /* The pointer to the start of guest memory. */
82 static void *guest_base;
83 /* The maximum guest physical address allowed, and maximum possible. */
84 static unsigned long guest_limit, guest_max;
85 /* The pipe for signal hander to write to. */
86 static int timeoutpipe[2];
87 static unsigned int timeout_usec = 500;
89 /* a per-cpu variable indicating whose vcpu is currently running */
90 static unsigned int __thread cpu_id;
92 /* This is our list of devices. */
95 /* Summary information about the devices in our list: ready to pass to
96 * select() to ask which need servicing.*/
100 /* Counter to assign interrupt numbers. */
101 unsigned int next_irq;
103 /* Counter to print out convenient device numbers. */
104 unsigned int device_num;
106 /* The descriptor page for the devices. */
109 /* A single linked list of devices. */
111 /* And a pointer to the last device for easy append and also for
112 * configuration appending. */
113 struct device *lastdev;
116 /* The list of Guest devices, based on command line arguments. */
117 static struct device_list devices;
119 /* The device structure describes a single device. */
122 /* The linked-list pointer. */
125 /* The this device's descriptor, as mapped into the Guest. */
126 struct lguest_device_desc *desc;
128 /* The name of this device, for --verbose. */
131 /* If handle_input is set, it wants to be called when this file
132 * descriptor is ready. */
134 bool (*handle_input)(int fd, struct device *me);
136 /* Any queues attached to this device */
137 struct virtqueue *vq;
139 /* Handle status being finalized (ie. feature bits stable). */
140 void (*ready)(struct device *me);
142 /* Device-specific data. */
146 /* The virtqueue structure describes a queue attached to a device. */
149 struct virtqueue *next;
151 /* Which device owns me. */
154 /* The configuration for this queue. */
155 struct lguest_vqconfig config;
157 /* The actual ring of buffers. */
160 /* Last available index we saw. */
163 /* The routine to call when the Guest pings us, or timeout. */
164 void (*handle_output)(int fd, struct virtqueue *me, bool timeout);
166 /* Outstanding buffers */
167 unsigned int inflight;
169 /* Is this blocked awaiting a timer? */
173 /* Remember the arguments to the program so we can "reboot" */
174 static char **main_args;
176 /* Since guest is UP and we don't run at the same time, we don't need barriers.
177 * But I include them in the code in case others copy it. */
180 /* Convert an iovec element to the given type.
182 * This is a fairly ugly trick: we need to know the size of the type and
183 * alignment requirement to check the pointer is kosher. It's also nice to
184 * have the name of the type in case we report failure.
186 * Typing those three things all the time is cumbersome and error prone, so we
187 * have a macro which sets them all up and passes to the real function. */
188 #define convert(iov, type) \
189 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
191 static void *_convert(struct iovec *iov, size_t size, size_t align,
194 if (iov->iov_len != size)
195 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
196 if ((unsigned long)iov->iov_base % align != 0)
197 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
198 return iov->iov_base;
201 /* Wrapper for the last available index. Makes it easier to change. */
202 #define lg_last_avail(vq) ((vq)->last_avail_idx)
204 /* The virtio configuration space is defined to be little-endian. x86 is
205 * little-endian too, but it's nice to be explicit so we have these helpers. */
206 #define cpu_to_le16(v16) (v16)
207 #define cpu_to_le32(v32) (v32)
208 #define cpu_to_le64(v64) (v64)
209 #define le16_to_cpu(v16) (v16)
210 #define le32_to_cpu(v32) (v32)
211 #define le64_to_cpu(v64) (v64)
213 /* Is this iovec empty? */
214 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
218 for (i = 0; i < num_iov; i++)
224 /* Take len bytes from the front of this iovec. */
225 static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
229 for (i = 0; i < num_iov; i++) {
232 used = iov[i].iov_len < len ? iov[i].iov_len : len;
233 iov[i].iov_base += used;
234 iov[i].iov_len -= used;
240 /* The device virtqueue descriptors are followed by feature bitmasks. */
241 static u8 *get_feature_bits(struct device *dev)
243 return (u8 *)(dev->desc + 1)
244 + dev->desc->num_vq * sizeof(struct lguest_vqconfig);
247 /*L:100 The Launcher code itself takes us out into userspace, that scary place
248 * where pointers run wild and free! Unfortunately, like most userspace
249 * programs, it's quite boring (which is why everyone likes to hack on the
250 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
251 * will get you through this section. Or, maybe not.
253 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
254 * memory and stores it in "guest_base". In other words, Guest physical ==
255 * Launcher virtual with an offset.
257 * This can be tough to get your head around, but usually it just means that we
258 * use these trivial conversion functions when the Guest gives us it's
259 * "physical" addresses: */
260 static void *from_guest_phys(unsigned long addr)
262 return guest_base + addr;
265 static unsigned long to_guest_phys(const void *addr)
267 return (addr - guest_base);
271 * Loading the Kernel.
273 * We start with couple of simple helper routines. open_or_die() avoids
274 * error-checking code cluttering the callers: */
275 static int open_or_die(const char *name, int flags)
277 int fd = open(name, flags);
279 err(1, "Failed to open %s", name);
283 /* map_zeroed_pages() takes a number of pages. */
284 static void *map_zeroed_pages(unsigned int num)
286 int fd = open_or_die("/dev/zero", O_RDONLY);
289 /* We use a private mapping (ie. if we write to the page, it will be
291 addr = mmap(NULL, getpagesize() * num,
292 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
293 if (addr == MAP_FAILED)
294 err(1, "Mmaping %u pages of /dev/zero", num);
300 /* Get some more pages for a device. */
301 static void *get_pages(unsigned int num)
303 void *addr = from_guest_phys(guest_limit);
305 guest_limit += num * getpagesize();
306 if (guest_limit > guest_max)
307 errx(1, "Not enough memory for devices");
311 /* This routine is used to load the kernel or initrd. It tries mmap, but if
312 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
313 * it falls back to reading the memory in. */
314 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
318 /* We map writable even though for some segments are marked read-only.
319 * The kernel really wants to be writable: it patches its own
322 * MAP_PRIVATE means that the page won't be copied until a write is
323 * done to it. This allows us to share untouched memory between
325 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
326 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
329 /* pread does a seek and a read in one shot: saves a few lines. */
330 r = pread(fd, addr, len, offset);
332 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
335 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
336 * the Guest memory. ELF = Embedded Linking Format, which is the format used
337 * by all modern binaries on Linux including the kernel.
339 * The ELF headers give *two* addresses: a physical address, and a virtual
340 * address. We use the physical address; the Guest will map itself to the
343 * We return the starting address. */
344 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
346 Elf32_Phdr phdr[ehdr->e_phnum];
349 /* Sanity checks on the main ELF header: an x86 executable with a
350 * reasonable number of correctly-sized program headers. */
351 if (ehdr->e_type != ET_EXEC
352 || ehdr->e_machine != EM_386
353 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
354 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
355 errx(1, "Malformed elf header");
357 /* An ELF executable contains an ELF header and a number of "program"
358 * headers which indicate which parts ("segments") of the program to
361 /* We read in all the program headers at once: */
362 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
363 err(1, "Seeking to program headers");
364 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
365 err(1, "Reading program headers");
367 /* Try all the headers: there are usually only three. A read-only one,
368 * a read-write one, and a "note" section which we don't load. */
369 for (i = 0; i < ehdr->e_phnum; i++) {
370 /* If this isn't a loadable segment, we ignore it */
371 if (phdr[i].p_type != PT_LOAD)
374 verbose("Section %i: size %i addr %p\n",
375 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
377 /* We map this section of the file at its physical address. */
378 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
379 phdr[i].p_offset, phdr[i].p_filesz);
382 /* The entry point is given in the ELF header. */
383 return ehdr->e_entry;
386 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
387 * supposed to jump into it and it will unpack itself. We used to have to
388 * perform some hairy magic because the unpacking code scared me.
390 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
391 * a small patch to jump over the tricky bits in the Guest, so now we just read
392 * the funky header so we know where in the file to load, and away we go! */
393 static unsigned long load_bzimage(int fd)
395 struct boot_params boot;
397 /* Modern bzImages get loaded at 1M. */
398 void *p = from_guest_phys(0x100000);
400 /* Go back to the start of the file and read the header. It should be
401 * a Linux boot header (see Documentation/i386/boot.txt) */
402 lseek(fd, 0, SEEK_SET);
403 read(fd, &boot, sizeof(boot));
405 /* Inside the setup_hdr, we expect the magic "HdrS" */
406 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
407 errx(1, "This doesn't look like a bzImage to me");
409 /* Skip over the extra sectors of the header. */
410 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
412 /* Now read everything into memory. in nice big chunks. */
413 while ((r = read(fd, p, 65536)) > 0)
416 /* Finally, code32_start tells us where to enter the kernel. */
417 return boot.hdr.code32_start;
420 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
421 * come wrapped up in the self-decompressing "bzImage" format. With a little
422 * work, we can load those, too. */
423 static unsigned long load_kernel(int fd)
427 /* Read in the first few bytes. */
428 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
429 err(1, "Reading kernel");
431 /* If it's an ELF file, it starts with "\177ELF" */
432 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
433 return map_elf(fd, &hdr);
435 /* Otherwise we assume it's a bzImage, and try to load it. */
436 return load_bzimage(fd);
439 /* This is a trivial little helper to align pages. Andi Kleen hated it because
440 * it calls getpagesize() twice: "it's dumb code."
442 * Kernel guys get really het up about optimization, even when it's not
443 * necessary. I leave this code as a reaction against that. */
444 static inline unsigned long page_align(unsigned long addr)
446 /* Add upwards and truncate downwards. */
447 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
450 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
451 * the kernel which the kernel can use to boot from without needing any
452 * drivers. Most distributions now use this as standard: the initrd contains
453 * the code to load the appropriate driver modules for the current machine.
455 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
456 * kernels. He sent me this (and tells me when I break it). */
457 static unsigned long load_initrd(const char *name, unsigned long mem)
463 ifd = open_or_die(name, O_RDONLY);
464 /* fstat() is needed to get the file size. */
465 if (fstat(ifd, &st) < 0)
466 err(1, "fstat() on initrd '%s'", name);
468 /* We map the initrd at the top of memory, but mmap wants it to be
469 * page-aligned, so we round the size up for that. */
470 len = page_align(st.st_size);
471 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
472 /* Once a file is mapped, you can close the file descriptor. It's a
473 * little odd, but quite useful. */
475 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
477 /* We return the initrd size. */
481 /* Once we know how much memory we have we can construct simple linear page
482 * tables which set virtual == physical which will get the Guest far enough
483 * into the boot to create its own.
485 * We lay them out of the way, just below the initrd (which is why we need to
486 * know its size here). */
487 static unsigned long setup_pagetables(unsigned long mem,
488 unsigned long initrd_size)
490 unsigned long *pgdir, *linear;
491 unsigned int mapped_pages, i, linear_pages;
492 unsigned int ptes_per_page = getpagesize()/sizeof(void *);
494 mapped_pages = mem/getpagesize();
496 /* Each PTE page can map ptes_per_page pages: how many do we need? */
497 linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page;
499 /* We put the toplevel page directory page at the top of memory. */
500 pgdir = from_guest_phys(mem) - initrd_size - getpagesize();
502 /* Now we use the next linear_pages pages as pte pages */
503 linear = (void *)pgdir - linear_pages*getpagesize();
505 /* Linear mapping is easy: put every page's address into the mapping in
506 * order. PAGE_PRESENT contains the flags Present, Writable and
508 for (i = 0; i < mapped_pages; i++)
509 linear[i] = ((i * getpagesize()) | PAGE_PRESENT);
511 /* The top level points to the linear page table pages above. */
512 for (i = 0; i < mapped_pages; i += ptes_per_page) {
513 pgdir[i/ptes_per_page]
514 = ((to_guest_phys(linear) + i*sizeof(void *))
518 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
519 mapped_pages, linear_pages, to_guest_phys(linear));
521 /* We return the top level (guest-physical) address: the kernel needs
522 * to know where it is. */
523 return to_guest_phys(pgdir);
527 /* Simple routine to roll all the commandline arguments together with spaces
529 static void concat(char *dst, char *args[])
531 unsigned int i, len = 0;
533 for (i = 0; args[i]; i++) {
535 strcat(dst+len, " ");
538 strcpy(dst+len, args[i]);
539 len += strlen(args[i]);
541 /* In case it's empty. */
545 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
546 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
547 * the base of Guest "physical" memory, the top physical page to allow, the
548 * top level pagetable and the entry point for the Guest. */
549 static int tell_kernel(unsigned long pgdir, unsigned long start)
551 unsigned long args[] = { LHREQ_INITIALIZE,
552 (unsigned long)guest_base,
553 guest_limit / getpagesize(), pgdir, start };
556 verbose("Guest: %p - %p (%#lx)\n",
557 guest_base, guest_base + guest_limit, guest_limit);
558 fd = open_or_die("/dev/lguest", O_RDWR);
559 if (write(fd, args, sizeof(args)) < 0)
560 err(1, "Writing to /dev/lguest");
562 /* We return the /dev/lguest file descriptor to control this Guest */
567 static void add_device_fd(int fd)
569 FD_SET(fd, &devices.infds);
570 if (fd > devices.max_infd)
571 devices.max_infd = fd;
577 * With console, block and network devices, we can have lots of input which we
578 * need to process. We could try to tell the kernel what file descriptors to
579 * watch, but handing a file descriptor mask through to the kernel is fairly
582 * Instead, we fork off a process which watches the file descriptors and writes
583 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
584 * stop running the Guest. This causes the Launcher to return from the
585 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
586 * the LHREQ_BREAK and wake us up again.
588 * This, of course, is merely a different *kind* of icky.
590 static void wake_parent(int pipefd, int lguest_fd)
592 /* Add the pipe from the Launcher to the fdset in the device_list, so
593 * we watch it, too. */
594 add_device_fd(pipefd);
597 fd_set rfds = devices.infds;
598 unsigned long args[] = { LHREQ_BREAK, 1 };
600 /* Wait until input is ready from one of the devices. */
601 select(devices.max_infd+1, &rfds, NULL, NULL, NULL);
602 /* Is it a message from the Launcher? */
603 if (FD_ISSET(pipefd, &rfds)) {
605 /* If read() returns 0, it means the Launcher has
606 * exited. We silently follow. */
607 if (read(pipefd, &fd, sizeof(fd)) == 0)
609 /* Otherwise it's telling us to change what file
610 * descriptors we're to listen to. Positive means
611 * listen to a new one, negative means stop
614 FD_SET(fd, &devices.infds);
616 FD_CLR(-fd - 1, &devices.infds);
617 } else /* Send LHREQ_BREAK command. */
618 pwrite(lguest_fd, args, sizeof(args), cpu_id);
622 /* This routine just sets up a pipe to the Waker process. */
623 static int setup_waker(int lguest_fd)
625 int pipefd[2], child;
627 /* We create a pipe to talk to the Waker, and also so it knows when the
628 * Launcher dies (and closes pipe). */
635 /* We are the Waker: close the "writing" end of our copy of the
636 * pipe and start waiting for input. */
638 wake_parent(pipefd[0], lguest_fd);
640 /* Close the reading end of our copy of the pipe. */
643 /* Here is the fd used to talk to the waker. */
650 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
651 * We need to make sure it's not trying to reach into the Launcher itself, so
652 * we have a convenient routine which checks it and exits with an error message
653 * if something funny is going on:
655 static void *_check_pointer(unsigned long addr, unsigned int size,
658 /* We have to separately check addr and addr+size, because size could
659 * be huge and addr + size might wrap around. */
660 if (addr >= guest_limit || addr + size >= guest_limit)
661 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
662 /* We return a pointer for the caller's convenience, now we know it's
664 return from_guest_phys(addr);
666 /* A macro which transparently hands the line number to the real function. */
667 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
669 /* Each buffer in the virtqueues is actually a chain of descriptors. This
670 * function returns the next descriptor in the chain, or vq->vring.num if we're
672 static unsigned next_desc(struct virtqueue *vq, unsigned int i)
676 /* If this descriptor says it doesn't chain, we're done. */
677 if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT))
678 return vq->vring.num;
680 /* Check they're not leading us off end of descriptors. */
681 next = vq->vring.desc[i].next;
682 /* Make sure compiler knows to grab that: we don't want it changing! */
685 if (next >= vq->vring.num)
686 errx(1, "Desc next is %u", next);
691 /* This looks in the virtqueue and for the first available buffer, and converts
692 * it to an iovec for convenient access. Since descriptors consist of some
693 * number of output then some number of input descriptors, it's actually two
694 * iovecs, but we pack them into one and note how many of each there were.
696 * This function returns the descriptor number found, or vq->vring.num (which
697 * is never a valid descriptor number) if none was found. */
698 static unsigned get_vq_desc(struct virtqueue *vq,
700 unsigned int *out_num, unsigned int *in_num)
702 unsigned int i, head;
705 /* Check it isn't doing very strange things with descriptor numbers. */
706 last_avail = lg_last_avail(vq);
707 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
708 errx(1, "Guest moved used index from %u to %u",
709 last_avail, vq->vring.avail->idx);
711 /* If there's nothing new since last we looked, return invalid. */
712 if (vq->vring.avail->idx == last_avail)
713 return vq->vring.num;
715 /* Grab the next descriptor number they're advertising, and increment
716 * the index we've seen. */
717 head = vq->vring.avail->ring[last_avail % vq->vring.num];
720 /* If their number is silly, that's a fatal mistake. */
721 if (head >= vq->vring.num)
722 errx(1, "Guest says index %u is available", head);
724 /* When we start there are none of either input nor output. */
725 *out_num = *in_num = 0;
729 /* Grab the first descriptor, and check it's OK. */
730 iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len;
731 iov[*out_num + *in_num].iov_base
732 = check_pointer(vq->vring.desc[i].addr,
733 vq->vring.desc[i].len);
734 /* If this is an input descriptor, increment that count. */
735 if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE)
738 /* If it's an output descriptor, they're all supposed
739 * to come before any input descriptors. */
741 errx(1, "Descriptor has out after in");
745 /* If we've got too many, that implies a descriptor loop. */
746 if (*out_num + *in_num > vq->vring.num)
747 errx(1, "Looped descriptor");
748 } while ((i = next_desc(vq, i)) != vq->vring.num);
754 /* After we've used one of their buffers, we tell them about it. We'll then
755 * want to send them an interrupt, using trigger_irq(). */
756 static void add_used(struct virtqueue *vq, unsigned int head, int len)
758 struct vring_used_elem *used;
760 /* The virtqueue contains a ring of used buffers. Get a pointer to the
761 * next entry in that used ring. */
762 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
765 /* Make sure buffer is written before we update index. */
767 vq->vring.used->idx++;
771 /* This actually sends the interrupt for this virtqueue */
772 static void trigger_irq(int fd, struct virtqueue *vq)
774 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
776 /* If they don't want an interrupt, don't send one, unless empty. */
777 if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
781 /* Send the Guest an interrupt tell them we used something up. */
782 if (write(fd, buf, sizeof(buf)) != 0)
783 err(1, "Triggering irq %i", vq->config.irq);
786 /* And here's the combo meal deal. Supersize me! */
787 static void add_used_and_trigger(int fd, struct virtqueue *vq,
788 unsigned int head, int len)
790 add_used(vq, head, len);
797 * Here is the input terminal setting we save, and the routine to restore them
798 * on exit so the user gets their terminal back. */
799 static struct termios orig_term;
800 static void restore_term(void)
802 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
805 /* We associate some data with the console for our exit hack. */
808 /* How many times have they hit ^C? */
810 /* When did they start? */
811 struct timeval start;
814 /* This is the routine which handles console input (ie. stdin). */
815 static bool handle_console_input(int fd, struct device *dev)
818 unsigned int head, in_num, out_num;
819 struct iovec iov[dev->vq->vring.num];
820 struct console_abort *abort = dev->priv;
822 /* First we need a console buffer from the Guests's input virtqueue. */
823 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
825 /* If they're not ready for input, stop listening to this file
826 * descriptor. We'll start again once they add an input buffer. */
827 if (head == dev->vq->vring.num)
831 errx(1, "Output buffers in console in queue?");
833 /* This is why we convert to iovecs: the readv() call uses them, and so
834 * it reads straight into the Guest's buffer. */
835 len = readv(dev->fd, iov, in_num);
837 /* This implies that the console is closed, is /dev/null, or
838 * something went terribly wrong. */
839 warnx("Failed to get console input, ignoring console.");
840 /* Put the input terminal back. */
842 /* Remove callback from input vq, so it doesn't restart us. */
843 dev->vq->handle_output = NULL;
844 /* Stop listening to this fd: don't call us again. */
848 /* Tell the Guest about the new input. */
849 add_used_and_trigger(fd, dev->vq, head, len);
851 /* Three ^C within one second? Exit.
853 * This is such a hack, but works surprisingly well. Each ^C has to be
854 * in a buffer by itself, so they can't be too fast. But we check that
855 * we get three within about a second, so they can't be too slow. */
856 if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) {
858 gettimeofday(&abort->start, NULL);
859 else if (abort->count == 3) {
861 gettimeofday(&now, NULL);
862 if (now.tv_sec <= abort->start.tv_sec+1) {
863 unsigned long args[] = { LHREQ_BREAK, 0 };
864 /* Close the fd so Waker will know it has to
867 /* Just in case waker is blocked in BREAK, send
869 write(fd, args, sizeof(args));
875 /* Any other key resets the abort counter. */
878 /* Everything went OK! */
882 /* Handling output for console is simple: we just get all the output buffers
883 * and write them to stdout. */
884 static void handle_console_output(int fd, struct virtqueue *vq, bool timeout)
886 unsigned int head, out, in;
888 struct iovec iov[vq->vring.num];
890 /* Keep getting output buffers from the Guest until we run out. */
891 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
893 errx(1, "Input buffers in output queue?");
894 len = writev(STDOUT_FILENO, iov, out);
895 add_used_and_trigger(fd, vq, head, len);
899 static void block_vq(struct virtqueue *vq)
901 struct itimerval itm;
903 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
906 itm.it_interval.tv_sec = 0;
907 itm.it_interval.tv_usec = 0;
908 itm.it_value.tv_sec = 0;
909 itm.it_value.tv_usec = timeout_usec;
911 setitimer(ITIMER_REAL, &itm, NULL);
917 * Handling output for network is also simple: we get all the output buffers
918 * and write them (ignoring the first element) to this device's file descriptor
921 static void handle_net_output(int fd, struct virtqueue *vq, bool timeout)
923 unsigned int head, out, in, num = 0;
925 struct iovec iov[vq->vring.num];
926 static int last_timeout_num;
928 /* Keep getting output buffers from the Guest until we run out. */
929 while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) {
931 errx(1, "Input buffers in output queue?");
932 len = writev(vq->dev->fd, iov, out);
934 err(1, "Writing network packet to tun");
935 add_used_and_trigger(fd, vq, head, len);
939 /* Block further kicks and set up a timer if we saw anything. */
944 if (num < last_timeout_num)
946 else if (timeout_usec > 1)
948 last_timeout_num = num;
952 /* This is where we handle a packet coming in from the tun device to our
954 static bool handle_tun_input(int fd, struct device *dev)
956 unsigned int head, in_num, out_num;
958 struct iovec iov[dev->vq->vring.num];
960 /* First we need a network buffer from the Guests's recv virtqueue. */
961 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
962 if (head == dev->vq->vring.num) {
963 /* Now, it's expected that if we try to send a packet too
964 * early, the Guest won't be ready yet. Wait until the device
965 * status says it's ready. */
966 /* FIXME: Actually want DRIVER_ACTIVE here. */
968 /* Now tell it we want to know if new things appear. */
969 dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
972 /* We'll turn this back on if input buffers are registered. */
975 errx(1, "Output buffers in network recv queue?");
977 /* Read the packet from the device directly into the Guest's buffer. */
978 len = readv(dev->fd, iov, in_num);
980 err(1, "reading network");
982 /* Tell the Guest about the new packet. */
983 add_used_and_trigger(fd, dev->vq, head, len);
985 verbose("tun input packet len %i [%02x %02x] (%s)\n", len,
986 ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1],
987 head != dev->vq->vring.num ? "sent" : "discarded");
993 /*L:215 This is the callback attached to the network and console input
994 * virtqueues: it ensures we try again, in case we stopped console or net
995 * delivery because Guest didn't have any buffers. */
996 static void enable_fd(int fd, struct virtqueue *vq, bool timeout)
998 add_device_fd(vq->dev->fd);
999 /* Tell waker to listen to it again */
1000 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
1003 static void net_enable_fd(int fd, struct virtqueue *vq, bool timeout)
1005 /* We don't need to know again when Guest refills receive buffer. */
1006 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
1007 enable_fd(fd, vq, timeout);
1010 /* When the Guest tells us they updated the status field, we handle it. */
1011 static void update_device_status(struct device *dev)
1013 struct virtqueue *vq;
1015 /* This is a reset. */
1016 if (dev->desc->status == 0) {
1017 verbose("Resetting device %s\n", dev->name);
1019 /* Clear any features they've acked. */
1020 memset(get_feature_bits(dev) + dev->desc->feature_len, 0,
1021 dev->desc->feature_len);
1023 /* Zero out the virtqueues. */
1024 for (vq = dev->vq; vq; vq = vq->next) {
1025 memset(vq->vring.desc, 0,
1026 vring_size(vq->config.num, getpagesize()));
1027 lg_last_avail(vq) = 0;
1029 } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1030 warnx("Device %s configuration FAILED", dev->name);
1031 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
1034 verbose("Device %s OK: offered", dev->name);
1035 for (i = 0; i < dev->desc->feature_len; i++)
1036 verbose(" %02x", get_feature_bits(dev)[i]);
1037 verbose(", accepted");
1038 for (i = 0; i < dev->desc->feature_len; i++)
1039 verbose(" %02x", get_feature_bits(dev)
1040 [dev->desc->feature_len+i]);
1047 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1048 static void handle_output(int fd, unsigned long addr)
1051 struct virtqueue *vq;
1053 /* Check each device and virtqueue. */
1054 for (i = devices.dev; i; i = i->next) {
1055 /* Notifications to device descriptors update device status. */
1056 if (from_guest_phys(addr) == i->desc) {
1057 update_device_status(i);
1061 /* Notifications to virtqueues mean output has occurred. */
1062 for (vq = i->vq; vq; vq = vq->next) {
1063 if (vq->config.pfn != addr/getpagesize())
1066 /* Guest should acknowledge (and set features!) before
1067 * using the device. */
1068 if (i->desc->status == 0) {
1069 warnx("%s gave early output", i->name);
1073 if (strcmp(vq->dev->name, "console") != 0)
1074 verbose("Output to %s\n", vq->dev->name);
1075 if (vq->handle_output)
1076 vq->handle_output(fd, vq, false);
1081 /* Early console write is done using notify on a nul-terminated string
1082 * in Guest memory. */
1083 if (addr >= guest_limit)
1084 errx(1, "Bad NOTIFY %#lx", addr);
1086 write(STDOUT_FILENO, from_guest_phys(addr),
1087 strnlen(from_guest_phys(addr), guest_limit - addr));
1090 static void handle_timeout(int fd)
1094 struct virtqueue *vq;
1096 /* Clear the pipe */
1097 read(timeoutpipe[0], buf, sizeof(buf));
1099 /* Check each device and virtqueue: flush blocked ones. */
1100 for (i = devices.dev; i; i = i->next) {
1101 for (vq = i->vq; vq; vq = vq->next) {
1105 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
1106 vq->blocked = false;
1107 if (vq->handle_output)
1108 vq->handle_output(fd, vq, true);
1113 /* This is called when the Waker wakes us up: check for incoming file
1115 static void handle_input(int fd)
1117 /* select() wants a zeroed timeval to mean "don't wait". */
1118 struct timeval poll = { .tv_sec = 0, .tv_usec = 0 };
1122 fd_set fds = devices.infds;
1125 num = select(devices.max_infd+1, &fds, NULL, NULL, &poll);
1126 /* Could get interrupted */
1129 /* If nothing is ready, we're done. */
1133 /* Otherwise, call the device(s) which have readable file
1134 * descriptors and a method of handling them. */
1135 for (i = devices.dev; i; i = i->next) {
1136 if (i->handle_input && FD_ISSET(i->fd, &fds)) {
1138 if (i->handle_input(fd, i))
1141 /* If handle_input() returns false, it means we
1142 * should no longer service it. Networking and
1143 * console do this when there's no input
1144 * buffers to deliver into. Console also uses
1145 * it when it discovers that stdin is closed. */
1146 FD_CLR(i->fd, &devices.infds);
1147 /* Tell waker to ignore it too, by sending a
1148 * negative fd number (-1, since 0 is a valid
1150 dev_fd = -i->fd - 1;
1151 write(waker_fd, &dev_fd, sizeof(dev_fd));
1155 /* Is this the timeout fd? */
1156 if (FD_ISSET(timeoutpipe[0], &fds))
1164 * All devices need a descriptor so the Guest knows it exists, and a "struct
1165 * device" so the Launcher can keep track of it. We have common helper
1166 * routines to allocate and manage them.
1169 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1170 * number of virtqueue descriptors, then two sets of feature bits, then an
1171 * array of configuration bytes. This routine returns the configuration
1173 static u8 *device_config(const struct device *dev)
1175 return (void *)(dev->desc + 1)
1176 + dev->desc->num_vq * sizeof(struct lguest_vqconfig)
1177 + dev->desc->feature_len * 2;
1180 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1181 * table page just above the Guest's normal memory. It returns a pointer to
1182 * that descriptor. */
1183 static struct lguest_device_desc *new_dev_desc(u16 type)
1185 struct lguest_device_desc d = { .type = type };
1188 /* Figure out where the next device config is, based on the last one. */
1189 if (devices.lastdev)
1190 p = device_config(devices.lastdev)
1191 + devices.lastdev->desc->config_len;
1193 p = devices.descpage;
1195 /* We only have one page for all the descriptors. */
1196 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1197 errx(1, "Too many devices");
1199 /* p might not be aligned, so we memcpy in. */
1200 return memcpy(p, &d, sizeof(d));
1203 /* Each device descriptor is followed by the description of its virtqueues. We
1204 * specify how many descriptors the virtqueue is to have. */
1205 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1206 void (*handle_output)(int, struct virtqueue *, bool))
1209 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1212 /* First we need some memory for this virtqueue. */
1213 pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1)
1215 p = get_pages(pages);
1217 /* Initialize the virtqueue */
1219 vq->last_avail_idx = 0;
1222 vq->blocked = false;
1224 /* Initialize the configuration. */
1225 vq->config.num = num_descs;
1226 vq->config.irq = devices.next_irq++;
1227 vq->config.pfn = to_guest_phys(p) / getpagesize();
1229 /* Initialize the vring. */
1230 vring_init(&vq->vring, num_descs, p, getpagesize());
1232 /* Append virtqueue to this device's descriptor. We use
1233 * device_config() to get the end of the device's current virtqueues;
1234 * we check that we haven't added any config or feature information
1235 * yet, otherwise we'd be overwriting them. */
1236 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1237 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1238 dev->desc->num_vq++;
1240 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1242 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1244 for (i = &dev->vq; *i; i = &(*i)->next);
1247 /* Set the routine to call when the Guest does something to this
1249 vq->handle_output = handle_output;
1251 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1252 * don't have a handler */
1254 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1257 /* The first half of the feature bitmask is for us to advertise features. The
1258 * second half is for the Guest to accept features. */
1259 static void add_feature(struct device *dev, unsigned bit)
1261 u8 *features = get_feature_bits(dev);
1263 /* We can't extend the feature bits once we've added config bytes */
1264 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1265 assert(dev->desc->config_len == 0);
1266 dev->desc->feature_len = (bit / CHAR_BIT) + 1;
1269 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1272 /* This routine sets the configuration fields for an existing device's
1273 * descriptor. It only works for the last device, but that's OK because that's
1275 static void set_config(struct device *dev, unsigned len, const void *conf)
1277 /* Check we haven't overflowed our single page. */
1278 if (device_config(dev) + len > devices.descpage + getpagesize())
1279 errx(1, "Too many devices");
1281 /* Copy in the config information, and store the length. */
1282 memcpy(device_config(dev), conf, len);
1283 dev->desc->config_len = len;
1286 /* This routine does all the creation and setup of a new device, including
1287 * calling new_dev_desc() to allocate the descriptor and device memory.
1289 * See what I mean about userspace being boring? */
1290 static struct device *new_device(const char *name, u16 type, int fd,
1291 bool (*handle_input)(int, struct device *))
1293 struct device *dev = malloc(sizeof(*dev));
1295 /* Now we populate the fields one at a time. */
1297 /* If we have an input handler for this file descriptor, then we add it
1298 * to the device_list's fdset and maxfd. */
1300 add_device_fd(dev->fd);
1301 dev->desc = new_dev_desc(type);
1302 dev->handle_input = handle_input;
1307 /* Append to device list. Prepending to a single-linked list is
1308 * easier, but the user expects the devices to be arranged on the bus
1309 * in command-line order. The first network device on the command line
1310 * is eth0, the first block device /dev/vda, etc. */
1311 if (devices.lastdev)
1312 devices.lastdev->next = dev;
1315 devices.lastdev = dev;
1320 /* Our first setup routine is the console. It's a fairly simple device, but
1321 * UNIX tty handling makes it uglier than it could be. */
1322 static void setup_console(void)
1326 /* If we can save the initial standard input settings... */
1327 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1328 struct termios term = orig_term;
1329 /* Then we turn off echo, line buffering and ^C etc. We want a
1330 * raw input stream to the Guest. */
1331 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1332 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1333 /* If we exit gracefully, the original settings will be
1334 * restored so the user can see what they're typing. */
1335 atexit(restore_term);
1338 dev = new_device("console", VIRTIO_ID_CONSOLE,
1339 STDIN_FILENO, handle_console_input);
1340 /* We store the console state in dev->priv, and initialize it. */
1341 dev->priv = malloc(sizeof(struct console_abort));
1342 ((struct console_abort *)dev->priv)->count = 0;
1344 /* The console needs two virtqueues: the input then the output. When
1345 * they put something the input queue, we make sure we're listening to
1346 * stdin. When they put something in the output queue, we write it to
1348 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1349 add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output);
1351 verbose("device %u: console\n", devices.device_num++);
1355 static void timeout_alarm(int sig)
1357 write(timeoutpipe[1], "", 1);
1360 static void setup_timeout(void)
1362 if (pipe(timeoutpipe) != 0)
1363 err(1, "Creating timeout pipe");
1365 if (fcntl(timeoutpipe[1], F_SETFL,
1366 fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0)
1367 err(1, "Making timeout pipe nonblocking");
1369 add_device_fd(timeoutpipe[0]);
1370 signal(SIGALRM, timeout_alarm);
1373 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1374 * --sharenet=<name> option which opens or creates a named pipe. This can be
1375 * used to send packets to another guest in a 1:1 manner.
1377 * More sopisticated is to use one of the tools developed for project like UML
1380 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1381 * completely generic ("here's my vring, attach to your vring") and would work
1382 * for any traffic. Of course, namespace and permissions issues need to be
1383 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1384 * multiple inter-guest channels behind one interface, although it would
1385 * require some manner of hotplugging new virtio channels.
1387 * Finally, we could implement a virtio network switch in the kernel. :*/
1389 static u32 str2ip(const char *ipaddr)
1393 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1394 errx(1, "Failed to parse IP address '%s'", ipaddr);
1395 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1398 static void str2mac(const char *macaddr, unsigned char mac[6])
1401 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1402 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1403 errx(1, "Failed to parse mac address '%s'", macaddr);
1412 /* This code is "adapted" from libbridge: it attaches the Host end of the
1413 * network device to the bridge device specified by the command line.
1415 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1416 * dislike bridging), and I just try not to break it. */
1417 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1423 errx(1, "must specify bridge name");
1425 ifidx = if_nametoindex(if_name);
1427 errx(1, "interface %s does not exist!", if_name);
1429 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1430 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1431 ifr.ifr_ifindex = ifidx;
1432 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1433 err(1, "can't add %s to bridge %s", if_name, br_name);
1436 /* This sets up the Host end of the network device with an IP address, brings
1437 * it up so packets will flow, the copies the MAC address into the hwaddr
1439 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1442 struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr;
1444 memset(&ifr, 0, sizeof(ifr));
1445 strcpy(ifr.ifr_name, tapif);
1447 /* Don't read these incantations. Just cut & paste them like I did! */
1448 sin->sin_family = AF_INET;
1449 sin->sin_addr.s_addr = htonl(ipaddr);
1450 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1451 err(1, "Setting %s interface address", tapif);
1452 ifr.ifr_flags = IFF_UP;
1453 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1454 err(1, "Bringing interface %s up", tapif);
1457 static void get_mac(int fd, const char *tapif, unsigned char hwaddr[6])
1461 memset(&ifr, 0, sizeof(ifr));
1462 strcpy(ifr.ifr_name, tapif);
1464 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1465 * above). IF means Interface, and HWADDR is hardware address.
1467 if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
1468 err(1, "getting hw address for %s", tapif);
1469 memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6);
1472 static int get_tun_device(char tapif[IFNAMSIZ])
1477 /* Start with this zeroed. Messy but sure. */
1478 memset(&ifr, 0, sizeof(ifr));
1480 /* We open the /dev/net/tun device and tell it we want a tap device. A
1481 * tap device is like a tun device, only somehow different. To tell
1482 * the truth, I completely blundered my way through this code, but it
1484 netfd = open_or_die("/dev/net/tun", O_RDWR);
1485 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1486 strcpy(ifr.ifr_name, "tap%d");
1487 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1488 err(1, "configuring /dev/net/tun");
1490 if (ioctl(netfd, TUNSETOFFLOAD,
1491 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1492 err(1, "Could not set features for tun device");
1494 /* We don't need checksums calculated for packets coming in this
1495 * device: trust us! */
1496 ioctl(netfd, TUNSETNOCSUM, 1);
1498 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1502 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1503 * routing, but the principle is the same: it uses the "tun" device to inject
1504 * packets into the Host as if they came in from a normal network card. We
1505 * just shunt packets between the Guest and the tun device. */
1506 static void setup_tun_net(char *arg)
1510 u32 ip = INADDR_ANY;
1511 bool bridging = false;
1512 char tapif[IFNAMSIZ], *p;
1513 struct virtio_net_config conf;
1515 netfd = get_tun_device(tapif);
1517 /* First we create a new network device. */
1518 dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input);
1520 /* Network devices need a receive and a send queue, just like
1522 add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd);
1523 add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output);
1525 /* We need a socket to perform the magic network ioctls to bring up the
1526 * tap interface, connect to the bridge etc. Any socket will do! */
1527 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1529 err(1, "opening IP socket");
1531 /* If the command line was --tunnet=bridge:<name> do bridging. */
1532 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1533 arg += strlen(BRIDGE_PFX);
1537 /* A mac address may follow the bridge name or IP address */
1538 p = strchr(arg, ':');
1540 str2mac(p+1, conf.mac);
1543 p = arg + strlen(arg);
1544 /* None supplied; query the randomly assigned mac. */
1545 get_mac(ipfd, tapif, conf.mac);
1548 /* arg is now either an IP address or a bridge name */
1550 add_to_bridge(ipfd, tapif, arg);
1554 /* Set up the tun device. */
1555 configure_device(ipfd, tapif, ip);
1557 /* Tell Guest what MAC address to use. */
1558 add_feature(dev, VIRTIO_NET_F_MAC);
1559 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1560 /* Expect Guest to handle everything except UFO */
1561 add_feature(dev, VIRTIO_NET_F_CSUM);
1562 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1563 add_feature(dev, VIRTIO_NET_F_MAC);
1564 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1565 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1566 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1567 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1568 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1569 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1570 set_config(dev, sizeof(conf), &conf);
1572 /* We don't need the socket any more; setup is done. */
1575 devices.device_num++;
1578 verbose("device %u: tun %s attached to bridge: %s\n",
1579 devices.device_num, tapif, arg);
1581 verbose("device %u: tun %s: %s\n",
1582 devices.device_num, tapif, arg);
1585 /* Our block (disk) device should be really simple: the Guest asks for a block
1586 * number and we read or write that position in the file. Unfortunately, that
1587 * was amazingly slow: the Guest waits until the read is finished before
1588 * running anything else, even if it could have been doing useful work.
1590 * We could use async I/O, except it's reputed to suck so hard that characters
1591 * actually go missing from your code when you try to use it.
1593 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1595 /* This hangs off device->priv. */
1598 /* The size of the file. */
1601 /* The file descriptor for the file. */
1604 /* IO thread listens on this file descriptor [0]. */
1607 /* IO thread writes to this file descriptor to mark it done, then
1608 * Launcher triggers interrupt to Guest. */
1615 * Remember that the block device is handled by a separate I/O thread. We head
1616 * straight into the core of that thread here:
1618 static bool service_io(struct device *dev)
1620 struct vblk_info *vblk = dev->priv;
1621 unsigned int head, out_num, in_num, wlen;
1624 struct virtio_blk_outhdr *out;
1625 struct iovec iov[dev->vq->vring.num];
1628 /* See if there's a request waiting. If not, nothing to do. */
1629 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1630 if (head == dev->vq->vring.num)
1633 /* Every block request should contain at least one output buffer
1634 * (detailing the location on disk and the type of request) and one
1635 * input buffer (to hold the result). */
1636 if (out_num == 0 || in_num == 0)
1637 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1638 head, out_num, in_num);
1640 out = convert(&iov[0], struct virtio_blk_outhdr);
1641 in = convert(&iov[out_num+in_num-1], u8);
1642 off = out->sector * 512;
1644 /* The block device implements "barriers", where the Guest indicates
1645 * that it wants all previous writes to occur before this write. We
1646 * don't have a way of asking our kernel to do a barrier, so we just
1647 * synchronize all the data in the file. Pretty poor, no? */
1648 if (out->type & VIRTIO_BLK_T_BARRIER)
1649 fdatasync(vblk->fd);
1651 /* In general the virtio block driver is allowed to try SCSI commands.
1652 * It'd be nice if we supported eject, for example, but we don't. */
1653 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1654 fprintf(stderr, "Scsi commands unsupported\n");
1655 *in = VIRTIO_BLK_S_UNSUPP;
1657 } else if (out->type & VIRTIO_BLK_T_OUT) {
1660 /* Move to the right location in the block file. This can fail
1661 * if they try to write past end. */
1662 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1663 err(1, "Bad seek to sector %llu", out->sector);
1665 ret = writev(vblk->fd, iov+1, out_num-1);
1666 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1668 /* Grr... Now we know how long the descriptor they sent was, we
1669 * make sure they didn't try to write over the end of the block
1670 * file (possibly extending it). */
1671 if (ret > 0 && off + ret > vblk->len) {
1672 /* Trim it back to the correct length */
1673 ftruncate64(vblk->fd, vblk->len);
1674 /* Die, bad Guest, die. */
1675 errx(1, "Write past end %llu+%u", off, ret);
1678 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1682 /* Move to the right location in the block file. This can fail
1683 * if they try to read past end. */
1684 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1685 err(1, "Bad seek to sector %llu", out->sector);
1687 ret = readv(vblk->fd, iov+1, in_num-1);
1688 verbose("READ from sector %llu: %i\n", out->sector, ret);
1690 wlen = sizeof(*in) + ret;
1691 *in = VIRTIO_BLK_S_OK;
1694 *in = VIRTIO_BLK_S_IOERR;
1698 /* We can't trigger an IRQ, because we're not the Launcher. It does
1699 * that when we tell it we're done. */
1700 add_used(dev->vq, head, wlen);
1704 /* This is the thread which actually services the I/O. */
1705 static int io_thread(void *_dev)
1707 struct device *dev = _dev;
1708 struct vblk_info *vblk = dev->priv;
1711 /* Close other side of workpipe so we get 0 read when main dies. */
1712 close(vblk->workpipe[1]);
1713 /* Close the other side of the done_fd pipe. */
1716 /* When this read fails, it means Launcher died, so we follow. */
1717 while (read(vblk->workpipe[0], &c, 1) == 1) {
1718 /* We acknowledge each request immediately to reduce latency,
1719 * rather than waiting until we've done them all. I haven't
1720 * measured to see if it makes any difference.
1722 * That would be an interesting test, wouldn't it? You could
1723 * also try having more than one I/O thread. */
1724 while (service_io(dev))
1725 write(vblk->done_fd, &c, 1);
1730 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1731 * when that thread tells us it's completed some I/O. */
1732 static bool handle_io_finish(int fd, struct device *dev)
1736 /* If the I/O thread died, presumably it printed the error, so we
1738 if (read(dev->fd, &c, 1) != 1)
1741 /* It did some work, so trigger the irq. */
1742 trigger_irq(fd, dev->vq);
1746 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1747 static void handle_virtblk_output(int fd, struct virtqueue *vq, bool timeout)
1749 struct vblk_info *vblk = vq->dev->priv;
1752 /* Wake up I/O thread and tell it to go to work! */
1753 if (write(vblk->workpipe[1], &c, 1) != 1)
1754 /* Presumably it indicated why it died. */
1758 /*L:198 This actually sets up a virtual block device. */
1759 static void setup_block_file(const char *filename)
1763 struct vblk_info *vblk;
1765 struct virtio_blk_config conf;
1767 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1770 /* The device responds to return from I/O thread. */
1771 dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish);
1773 /* The device has one virtqueue, where the Guest places requests. */
1774 add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output);
1776 /* Allocate the room for our own bookkeeping */
1777 vblk = dev->priv = malloc(sizeof(*vblk));
1779 /* First we open the file and store the length. */
1780 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1781 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1783 /* We support barriers. */
1784 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1786 /* Tell Guest how many sectors this device has. */
1787 conf.capacity = cpu_to_le64(vblk->len / 512);
1789 /* Tell Guest not to put in too many descriptors at once: two are used
1790 * for the in and out elements. */
1791 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1792 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1794 set_config(dev, sizeof(conf), &conf);
1796 /* The I/O thread writes to this end of the pipe when done. */
1797 vblk->done_fd = p[1];
1799 /* This is the second pipe, which is how we tell the I/O thread about
1801 pipe(vblk->workpipe);
1803 /* Create stack for thread and run it. Since stack grows upwards, we
1804 * point the stack pointer to the end of this region. */
1805 stack = malloc(32768);
1806 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1807 * becoming a zombie. */
1808 if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1)
1809 err(1, "Creating clone");
1811 /* We don't need to keep the I/O thread's end of the pipes open. */
1812 close(vblk->done_fd);
1813 close(vblk->workpipe[0]);
1815 verbose("device %u: virtblock %llu sectors\n",
1816 devices.device_num, le64_to_cpu(conf.capacity));
1819 /* Our random number generator device reads from /dev/random into the Guest's
1820 * input buffers. The usual case is that the Guest doesn't want random numbers
1821 * and so has no buffers although /dev/random is still readable, whereas
1822 * console is the reverse.
1824 * The same logic applies, however. */
1825 static bool handle_rng_input(int fd, struct device *dev)
1828 unsigned int head, in_num, out_num, totlen = 0;
1829 struct iovec iov[dev->vq->vring.num];
1831 /* First we need a buffer from the Guests's virtqueue. */
1832 head = get_vq_desc(dev->vq, iov, &out_num, &in_num);
1834 /* If they're not ready for input, stop listening to this file
1835 * descriptor. We'll start again once they add an input buffer. */
1836 if (head == dev->vq->vring.num)
1840 errx(1, "Output buffers in rng?");
1842 /* This is why we convert to iovecs: the readv() call uses them, and so
1843 * it reads straight into the Guest's buffer. We loop to make sure we
1845 while (!iov_empty(iov, in_num)) {
1846 len = readv(dev->fd, iov, in_num);
1848 err(1, "Read from /dev/random gave %i", len);
1849 iov_consume(iov, in_num, len);
1853 /* Tell the Guest about the new input. */
1854 add_used_and_trigger(fd, dev->vq, head, totlen);
1856 /* Everything went OK! */
1860 /* And this creates a "hardware" random number device for the Guest. */
1861 static void setup_rng(void)
1866 fd = open_or_die("/dev/random", O_RDONLY);
1868 /* The device responds to return from I/O thread. */
1869 dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input);
1871 /* The device has one virtqueue, where the Guest places inbufs. */
1872 add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd);
1874 verbose("device %u: rng\n", devices.device_num++);
1876 /* That's the end of device setup. */
1878 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1879 static void __attribute__((noreturn)) restart_guest(void)
1883 /* Closing pipes causes the Waker thread and io_threads to die, and
1884 * closing /dev/lguest cleans up the Guest. Since we don't track all
1885 * open fds, we simply close everything beyond stderr. */
1886 for (i = 3; i < FD_SETSIZE; i++)
1888 execv(main_args[0], main_args);
1889 err(1, "Could not exec %s", main_args[0]);
1892 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1893 * its input and output, and finally, lays it to rest. */
1894 static void __attribute__((noreturn)) run_guest(int lguest_fd)
1897 unsigned long args[] = { LHREQ_BREAK, 0 };
1898 unsigned long notify_addr;
1901 /* We read from the /dev/lguest device to run the Guest. */
1902 readval = pread(lguest_fd, ¬ify_addr,
1903 sizeof(notify_addr), cpu_id);
1905 /* One unsigned long means the Guest did HCALL_NOTIFY */
1906 if (readval == sizeof(notify_addr)) {
1907 verbose("Notify on address %#lx\n", notify_addr);
1908 handle_output(lguest_fd, notify_addr);
1910 /* ENOENT means the Guest died. Reading tells us why. */
1911 } else if (errno == ENOENT) {
1912 char reason[1024] = { 0 };
1913 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1914 errx(1, "%s", reason);
1915 /* ERESTART means that we need to reboot the guest */
1916 } else if (errno == ERESTART) {
1918 /* EAGAIN means a signal (timeout).
1919 * Anything else means a bug or incompatible change. */
1920 } else if (errno != EAGAIN)
1921 err(1, "Running guest failed");
1923 /* Only service input on thread for CPU 0. */
1927 /* Service input, then unset the BREAK to release the Waker. */
1928 handle_input(lguest_fd);
1929 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1930 err(1, "Resetting break");
1934 * This is the end of the Launcher. The good news: we are over halfway
1935 * through! The bad news: the most fiendish part of the code still lies ahead
1938 * Are you ready? Take a deep breath and join me in the core of the Host, in
1942 static struct option opts[] = {
1943 { "verbose", 0, NULL, 'v' },
1944 { "tunnet", 1, NULL, 't' },
1945 { "block", 1, NULL, 'b' },
1946 { "rng", 0, NULL, 'r' },
1947 { "initrd", 1, NULL, 'i' },
1950 static void usage(void)
1952 errx(1, "Usage: lguest [--verbose] "
1953 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1954 "|--block=<filename>|--initrd=<filename>]...\n"
1955 "<mem-in-mb> vmlinux [args...]");
1958 /*L:105 The main routine is where the real work begins: */
1959 int main(int argc, char *argv[])
1961 /* Memory, top-level pagetable, code startpoint and size of the
1962 * (optional) initrd. */
1963 unsigned long mem = 0, pgdir, start, initrd_size = 0;
1964 /* Two temporaries and the /dev/lguest file descriptor. */
1965 int i, c, lguest_fd;
1966 /* The boot information for the Guest. */
1967 struct boot_params *boot;
1968 /* If they specify an initrd file to load. */
1969 const char *initrd_name = NULL;
1971 /* Save the args: we "reboot" by execing ourselves again. */
1973 /* We don't "wait" for the children, so prevent them from becoming
1975 signal(SIGCHLD, SIG_IGN);
1977 /* First we initialize the device list. Since console and network
1978 * device receive input from a file descriptor, we keep an fdset
1979 * (infds) and the maximum fd number (max_infd) with the head of the
1980 * list. We also keep a pointer to the last device. Finally, we keep
1981 * the next interrupt number to use for devices (1: remember that 0 is
1982 * used by the timer). */
1983 FD_ZERO(&devices.infds);
1984 devices.max_infd = -1;
1985 devices.lastdev = NULL;
1986 devices.next_irq = 1;
1989 /* We need to know how much memory so we can set up the device
1990 * descriptor and memory pages for the devices as we parse the command
1991 * line. So we quickly look through the arguments to find the amount
1993 for (i = 1; i < argc; i++) {
1994 if (argv[i][0] != '-') {
1995 mem = atoi(argv[i]) * 1024 * 1024;
1996 /* We start by mapping anonymous pages over all of
1997 * guest-physical memory range. This fills it with 0,
1998 * and ensures that the Guest won't be killed when it
1999 * tries to access it. */
2000 guest_base = map_zeroed_pages(mem / getpagesize()
2003 guest_max = mem + DEVICE_PAGES*getpagesize();
2004 devices.descpage = get_pages(1);
2009 /* The options are fairly straight-forward */
2010 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
2016 setup_tun_net(optarg);
2019 setup_block_file(optarg);
2025 initrd_name = optarg;
2028 warnx("Unknown argument %s", argv[optind]);
2032 /* After the other arguments we expect memory and kernel image name,
2033 * followed by command line arguments for the kernel. */
2034 if (optind + 2 > argc)
2037 verbose("Guest base is at %p\n", guest_base);
2039 /* We always have a console device */
2042 /* We can timeout waiting for Guest network transmit. */
2045 /* Now we load the kernel */
2046 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
2048 /* Boot information is stashed at physical address 0 */
2049 boot = from_guest_phys(0);
2051 /* Map the initrd image if requested (at top of physical memory) */
2053 initrd_size = load_initrd(initrd_name, mem);
2054 /* These are the location in the Linux boot header where the
2055 * start and size of the initrd are expected to be found. */
2056 boot->hdr.ramdisk_image = mem - initrd_size;
2057 boot->hdr.ramdisk_size = initrd_size;
2058 /* The bootloader type 0xFF means "unknown"; that's OK. */
2059 boot->hdr.type_of_loader = 0xFF;
2062 /* Set up the initial linear pagetables, starting below the initrd. */
2063 pgdir = setup_pagetables(mem, initrd_size);
2065 /* The Linux boot header contains an "E820" memory map: ours is a
2066 * simple, single region. */
2067 boot->e820_entries = 1;
2068 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
2069 /* The boot header contains a command line pointer: we put the command
2070 * line after the boot header. */
2071 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2072 /* We use a simple helper to copy the arguments separated by spaces. */
2073 concat((char *)(boot + 1), argv+optind+2);
2075 /* Boot protocol version: 2.07 supports the fields for lguest. */
2076 boot->hdr.version = 0x207;
2078 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2079 boot->hdr.hardware_subarch = 1;
2081 /* Tell the entry path not to try to reload segment registers. */
2082 boot->hdr.loadflags |= KEEP_SEGMENTS;
2084 /* We tell the kernel to initialize the Guest: this returns the open
2085 * /dev/lguest file descriptor. */
2086 lguest_fd = tell_kernel(pgdir, start);
2088 /* We fork off a child process, which wakes the Launcher whenever one
2089 * of the input file descriptors needs attention. We call this the
2090 * Waker, and we'll cover it in a moment. */
2091 waker_fd = setup_waker(lguest_fd);
2093 /* Finally, run the Guest. This doesn't return. */
2094 run_guest(lguest_fd);
2099 * Mastery is done: you now know everything I do.
2101 * But surely you have seen code, features and bugs in your wanderings which
2102 * you now yearn to attack? That is the real game, and I look forward to you
2103 * patching and forking lguest into the Your-Name-Here-visor.
2105 * Farewell, and good coding!