About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / lguest / lguest.c




Custom Search

Based on kernel version 2.6.39.1. Page generated on 2011-06-03 13:47 EST.

1	/*P:100
2	 * This is the Launcher code, a simple program which lays out the "physical"
3	 * memory for the new Guest by mapping the kernel image and the virtual
4	 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5	 * control it.
6	:*/
7	#define _LARGEFILE64_SOURCE
8	#define _GNU_SOURCE
9	#include <stdio.h>
10	#include <string.h>
11	#include <unistd.h>
12	#include <err.h>
13	#include <stdint.h>
14	#include <stdlib.h>
15	#include <elf.h>
16	#include <sys/mman.h>
17	#include <sys/param.h>
18	#include <sys/types.h>
19	#include <sys/stat.h>
20	#include <sys/wait.h>
21	#include <sys/eventfd.h>
22	#include <fcntl.h>
23	#include <stdbool.h>
24	#include <errno.h>
25	#include <ctype.h>
26	#include <sys/socket.h>
27	#include <sys/ioctl.h>
28	#include <sys/time.h>
29	#include <time.h>
30	#include <netinet/in.h>
31	#include <net/if.h>
32	#include <linux/sockios.h>
33	#include <linux/if_tun.h>
34	#include <sys/uio.h>
35	#include <termios.h>
36	#include <getopt.h>
37	#include <assert.h>
38	#include <sched.h>
39	#include <limits.h>
40	#include <stddef.h>
41	#include <signal.h>
42	#include <pwd.h>
43	#include <grp.h>
44	
45	#include <linux/virtio_config.h>
46	#include <linux/virtio_net.h>
47	#include <linux/virtio_blk.h>
48	#include <linux/virtio_console.h>
49	#include <linux/virtio_rng.h>
50	#include <linux/virtio_ring.h>
51	#include <asm/bootparam.h>
52	#include "../../include/linux/lguest_launcher.h"
53	/*L:110
54	 * We can ignore the 42 include files we need for this program, but I do want
55	 * to draw attention to the use of kernel-style types.
56	 *
57	 * As Linus said, "C is a Spartan language, and so should your naming be."  I
58	 * like these abbreviations, so we define them here.  Note that u64 is always
59	 * unsigned long long, which works on all Linux systems: this means that we can
60	 * use %llu in printf for any u64.
61	 */
62	typedef unsigned long long u64;
63	typedef uint32_t u32;
64	typedef uint16_t u16;
65	typedef uint8_t u8;
66	/*:*/
67	
68	#define PAGE_PRESENT 0x7 	/* Present, RW, Execute */
69	#define BRIDGE_PFX "bridge:"
70	#ifndef SIOCBRADDIF
71	#define SIOCBRADDIF	0x89a2		/* add interface to bridge      */
72	#endif
73	/* We can have up to 256 pages for devices. */
74	#define DEVICE_PAGES 256
75	/* This will occupy 3 pages: it must be a power of 2. */
76	#define VIRTQUEUE_NUM 256
77	
78	/*L:120
79	 * verbose is both a global flag and a macro.  The C preprocessor allows
80	 * this, and although I wouldn't recommend it, it works quite nicely here.
81	 */
82	static bool verbose;
83	#define verbose(args...) \
84		do { if (verbose) printf(args); } while(0)
85	/*:*/
86	
87	/* The pointer to the start of guest memory. */
88	static void *guest_base;
89	/* The maximum guest physical address allowed, and maximum possible. */
90	static unsigned long guest_limit, guest_max;
91	/* The /dev/lguest file descriptor. */
92	static int lguest_fd;
93	
94	/* a per-cpu variable indicating whose vcpu is currently running */
95	static unsigned int __thread cpu_id;
96	
97	/* This is our list of devices. */
98	struct device_list {
99		/* Counter to assign interrupt numbers. */
100		unsigned int next_irq;
101	
102		/* Counter to print out convenient device numbers. */
103		unsigned int device_num;
104	
105		/* The descriptor page for the devices. */
106		u8 *descpage;
107	
108		/* A single linked list of devices. */
109		struct device *dev;
110		/* And a pointer to the last device for easy append. */
111		struct device *lastdev;
112	};
113	
114	/* The list of Guest devices, based on command line arguments. */
115	static struct device_list devices;
116	
117	/* The device structure describes a single device. */
118	struct device {
119		/* The linked-list pointer. */
120		struct device *next;
121	
122		/* The device's descriptor, as mapped into the Guest. */
123		struct lguest_device_desc *desc;
124	
125		/* We can't trust desc values once Guest has booted: we use these. */
126		unsigned int feature_len;
127		unsigned int num_vq;
128	
129		/* The name of this device, for --verbose. */
130		const char *name;
131	
132		/* Any queues attached to this device */
133		struct virtqueue *vq;
134	
135		/* Is it operational */
136		bool running;
137	
138		/* Does Guest want an intrrupt on empty? */
139		bool irq_on_empty;
140	
141		/* Device-specific data. */
142		void *priv;
143	};
144	
145	/* The virtqueue structure describes a queue attached to a device. */
146	struct virtqueue {
147		struct virtqueue *next;
148	
149		/* Which device owns me. */
150		struct device *dev;
151	
152		/* The configuration for this queue. */
153		struct lguest_vqconfig config;
154	
155		/* The actual ring of buffers. */
156		struct vring vring;
157	
158		/* Last available index we saw. */
159		u16 last_avail_idx;
160	
161		/* How many are used since we sent last irq? */
162		unsigned int pending_used;
163	
164		/* Eventfd where Guest notifications arrive. */
165		int eventfd;
166	
167		/* Function for the thread which is servicing this virtqueue. */
168		void (*service)(struct virtqueue *vq);
169		pid_t thread;
170	};
171	
172	/* Remember the arguments to the program so we can "reboot" */
173	static char **main_args;
174	
175	/* The original tty settings to restore on exit. */
176	static struct termios orig_term;
177	
178	/*
179	 * We have to be careful with barriers: our devices are all run in separate
180	 * threads and so we need to make sure that changes visible to the Guest happen
181	 * in precise order.
182	 */
183	#define wmb() __asm__ __volatile__("" : : : "memory")
184	#define mb() __asm__ __volatile__("" : : : "memory")
185	
186	/*
187	 * Convert an iovec element to the given type.
188	 *
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.
192	 *
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	 */
196	#define convert(iov, type) \
197		((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
198	
199	static void *_convert(struct iovec *iov, size_t size, size_t align,
200			      const char *name)
201	{
202		if (iov->iov_len != size)
203			errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
204		if ((unsigned long)iov->iov_base % align != 0)
205			errx(1, "Bad alignment %p for %s", iov->iov_base, name);
206		return iov->iov_base;
207	}
208	
209	/* Wrapper for the last available index.  Makes it easier to change. */
210	#define lg_last_avail(vq)	((vq)->last_avail_idx)
211	
212	/*
213	 * The virtio configuration space is defined to be little-endian.  x86 is
214	 * little-endian too, but it's nice to be explicit so we have these helpers.
215	 */
216	#define cpu_to_le16(v16) (v16)
217	#define cpu_to_le32(v32) (v32)
218	#define cpu_to_le64(v64) (v64)
219	#define le16_to_cpu(v16) (v16)
220	#define le32_to_cpu(v32) (v32)
221	#define le64_to_cpu(v64) (v64)
222	
223	/* Is this iovec empty? */
224	static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
225	{
226		unsigned int i;
227	
228		for (i = 0; i < num_iov; i++)
229			if (iov[i].iov_len)
230				return false;
231		return true;
232	}
233	
234	/* Take len bytes from the front of this iovec. */
235	static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
236	{
237		unsigned int i;
238	
239		for (i = 0; i < num_iov; i++) {
240			unsigned int used;
241	
242			used = iov[i].iov_len < len ? iov[i].iov_len : len;
243			iov[i].iov_base += used;
244			iov[i].iov_len -= used;
245			len -= used;
246		}
247		assert(len == 0);
248	}
249	
250	/* The device virtqueue descriptors are followed by feature bitmasks. */
251	static u8 *get_feature_bits(struct device *dev)
252	{
253		return (u8 *)(dev->desc + 1)
254			+ dev->num_vq * sizeof(struct lguest_vqconfig);
255	}
256	
257	/*L:100
258	 * The Launcher code itself takes us out into userspace, that scary place where
259	 * pointers run wild and free!  Unfortunately, like most userspace programs,
260	 * it's quite boring (which is why everyone likes to hack on the kernel!).
261	 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
262	 * you through this section.  Or, maybe not.
263	 *
264	 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
265	 * memory and stores it in "guest_base".  In other words, Guest physical ==
266	 * Launcher virtual with an offset.
267	 *
268	 * This can be tough to get your head around, but usually it just means that we
269	 * use these trivial conversion functions when the Guest gives us its
270	 * "physical" addresses:
271	 */
272	static void *from_guest_phys(unsigned long addr)
273	{
274		return guest_base + addr;
275	}
276	
277	static unsigned long to_guest_phys(const void *addr)
278	{
279		return (addr - guest_base);
280	}
281	
282	/*L:130
283	 * Loading the Kernel.
284	 *
285	 * We start with couple of simple helper routines.  open_or_die() avoids
286	 * error-checking code cluttering the callers:
287	 */
288	static int open_or_die(const char *name, int flags)
289	{
290		int fd = open(name, flags);
291		if (fd < 0)
292			err(1, "Failed to open %s", name);
293		return fd;
294	}
295	
296	/* map_zeroed_pages() takes a number of pages. */
297	static void *map_zeroed_pages(unsigned int num)
298	{
299		int fd = open_or_die("/dev/zero", O_RDONLY);
300		void *addr;
301	
302		/*
303		 * We use a private mapping (ie. if we write to the page, it will be
304		 * copied). We allocate an extra two pages PROT_NONE to act as guard
305		 * pages against read/write attempts that exceed allocated space.
306		 */
307		addr = mmap(NULL, getpagesize() * (num+2),
308			    PROT_NONE, MAP_PRIVATE, fd, 0);
309	
310		if (addr == MAP_FAILED)
311			err(1, "Mmapping %u pages of /dev/zero", num);
312	
313		if (mprotect(addr + getpagesize(), getpagesize() * num,
314			     PROT_READ|PROT_WRITE) == -1)
315			err(1, "mprotect rw %u pages failed", num);
316	
317		/*
318		 * One neat mmap feature is that you can close the fd, and it
319		 * stays mapped.
320		 */
321		close(fd);
322	
323		/* Return address after PROT_NONE page */
324		return addr + getpagesize();
325	}
326	
327	/* Get some more pages for a device. */
328	static void *get_pages(unsigned int num)
329	{
330		void *addr = from_guest_phys(guest_limit);
331	
332		guest_limit += num * getpagesize();
333		if (guest_limit > guest_max)
334			errx(1, "Not enough memory for devices");
335		return addr;
336	}
337	
338	/*
339	 * This routine is used to load the kernel or initrd.  It tries mmap, but if
340	 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
341	 * it falls back to reading the memory in.
342	 */
343	static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
344	{
345		ssize_t r;
346	
347		/*
348		 * We map writable even though for some segments are marked read-only.
349		 * The kernel really wants to be writable: it patches its own
350		 * instructions.
351		 *
352		 * MAP_PRIVATE means that the page won't be copied until a write is
353		 * done to it.  This allows us to share untouched memory between
354		 * Guests.
355		 */
356		if (mmap(addr, len, PROT_READ|PROT_WRITE,
357			 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
358			return;
359	
360		/* pread does a seek and a read in one shot: saves a few lines. */
361		r = pread(fd, addr, len, offset);
362		if (r != len)
363			err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
364	}
365	
366	/*
367	 * This routine takes an open vmlinux image, which is in ELF, and maps it into
368	 * the Guest memory.  ELF = Embedded Linking Format, which is the format used
369	 * by all modern binaries on Linux including the kernel.
370	 *
371	 * The ELF headers give *two* addresses: a physical address, and a virtual
372	 * address.  We use the physical address; the Guest will map itself to the
373	 * virtual address.
374	 *
375	 * We return the starting address.
376	 */
377	static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
378	{
379		Elf32_Phdr phdr[ehdr->e_phnum];
380		unsigned int i;
381	
382		/*
383		 * Sanity checks on the main ELF header: an x86 executable with a
384		 * reasonable number of correctly-sized program headers.
385		 */
386		if (ehdr->e_type != ET_EXEC
387		    || ehdr->e_machine != EM_386
388		    || ehdr->e_phentsize != sizeof(Elf32_Phdr)
389		    || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
390			errx(1, "Malformed elf header");
391	
392		/*
393		 * An ELF executable contains an ELF header and a number of "program"
394		 * headers which indicate which parts ("segments") of the program to
395		 * load where.
396		 */
397	
398		/* We read in all the program headers at once: */
399		if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
400			err(1, "Seeking to program headers");
401		if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
402			err(1, "Reading program headers");
403	
404		/*
405		 * Try all the headers: there are usually only three.  A read-only one,
406		 * a read-write one, and a "note" section which we don't load.
407		 */
408		for (i = 0; i < ehdr->e_phnum; i++) {
409			/* If this isn't a loadable segment, we ignore it */
410			if (phdr[i].p_type != PT_LOAD)
411				continue;
412	
413			verbose("Section %i: size %i addr %p\n",
414				i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
415	
416			/* We map this section of the file at its physical address. */
417			map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
418			       phdr[i].p_offset, phdr[i].p_filesz);
419		}
420	
421		/* The entry point is given in the ELF header. */
422		return ehdr->e_entry;
423	}
424	
425	/*L:150
426	 * A bzImage, unlike an ELF file, is not meant to be loaded.  You're supposed
427	 * to jump into it and it will unpack itself.  We used to have to perform some
428	 * hairy magic because the unpacking code scared me.
429	 *
430	 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
431	 * a small patch to jump over the tricky bits in the Guest, so now we just read
432	 * the funky header so we know where in the file to load, and away we go!
433	 */
434	static unsigned long load_bzimage(int fd)
435	{
436		struct boot_params boot;
437		int r;
438		/* Modern bzImages get loaded at 1M. */
439		void *p = from_guest_phys(0x100000);
440	
441		/*
442		 * Go back to the start of the file and read the header.  It should be
443		 * a Linux boot header (see Documentation/x86/i386/boot.txt)
444		 */
445		lseek(fd, 0, SEEK_SET);
446		read(fd, &boot, sizeof(boot));
447	
448		/* Inside the setup_hdr, we expect the magic "HdrS" */
449		if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
450			errx(1, "This doesn't look like a bzImage to me");
451	
452		/* Skip over the extra sectors of the header. */
453		lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
454	
455		/* Now read everything into memory. in nice big chunks. */
456		while ((r = read(fd, p, 65536)) > 0)
457			p += r;
458	
459		/* Finally, code32_start tells us where to enter the kernel. */
460		return boot.hdr.code32_start;
461	}
462	
463	/*L:140
464	 * Loading the kernel is easy when it's a "vmlinux", but most kernels
465	 * come wrapped up in the self-decompressing "bzImage" format.  With a little
466	 * work, we can load those, too.
467	 */
468	static unsigned long load_kernel(int fd)
469	{
470		Elf32_Ehdr hdr;
471	
472		/* Read in the first few bytes. */
473		if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
474			err(1, "Reading kernel");
475	
476		/* If it's an ELF file, it starts with "\177ELF" */
477		if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
478			return map_elf(fd, &hdr);
479	
480		/* Otherwise we assume it's a bzImage, and try to load it. */
481		return load_bzimage(fd);
482	}
483	
484	/*
485	 * This is a trivial little helper to align pages.  Andi Kleen hated it because
486	 * it calls getpagesize() twice: "it's dumb code."
487	 *
488	 * Kernel guys get really het up about optimization, even when it's not
489	 * necessary.  I leave this code as a reaction against that.
490	 */
491	static inline unsigned long page_align(unsigned long addr)
492	{
493		/* Add upwards and truncate downwards. */
494		return ((addr + getpagesize()-1) & ~(getpagesize()-1));
495	}
496	
497	/*L:180
498	 * An "initial ram disk" is a disk image loaded into memory along with the
499	 * kernel which the kernel can use to boot from without needing any drivers.
500	 * Most distributions now use this as standard: the initrd contains the code to
501	 * load the appropriate driver modules for the current machine.
502	 *
503	 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
504	 * kernels.  He sent me this (and tells me when I break it).
505	 */
506	static unsigned long load_initrd(const char *name, unsigned long mem)
507	{
508		int ifd;
509		struct stat st;
510		unsigned long len;
511	
512		ifd = open_or_die(name, O_RDONLY);
513		/* fstat() is needed to get the file size. */
514		if (fstat(ifd, &st) < 0)
515			err(1, "fstat() on initrd '%s'", name);
516	
517		/*
518		 * We map the initrd at the top of memory, but mmap wants it to be
519		 * page-aligned, so we round the size up for that.
520		 */
521		len = page_align(st.st_size);
522		map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
523		/*
524		 * Once a file is mapped, you can close the file descriptor.  It's a
525		 * little odd, but quite useful.
526		 */
527		close(ifd);
528		verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
529	
530		/* We return the initrd size. */
531		return len;
532	}
533	/*:*/
534	
535	/*
536	 * Simple routine to roll all the commandline arguments together with spaces
537	 * between them.
538	 */
539	static void concat(char *dst, char *args[])
540	{
541		unsigned int i, len = 0;
542	
543		for (i = 0; args[i]; i++) {
544			if (i) {
545				strcat(dst+len, " ");
546				len++;
547			}
548			strcpy(dst+len, args[i]);
549			len += strlen(args[i]);
550		}
551		/* In case it's empty. */
552		dst[len] = '\0';
553	}
554	
555	/*L:185
556	 * This is where we actually tell the kernel to initialize the Guest.  We
557	 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
558	 * the base of Guest "physical" memory, the top physical page to allow and the
559	 * entry point for the Guest.
560	 */
561	static void tell_kernel(unsigned long start)
562	{
563		unsigned long args[] = { LHREQ_INITIALIZE,
564					 (unsigned long)guest_base,
565					 guest_limit / getpagesize(), start };
566		verbose("Guest: %p - %p (%#lx)\n",
567			guest_base, guest_base + guest_limit, guest_limit);
568		lguest_fd = open_or_die("/dev/lguest", O_RDWR);
569		if (write(lguest_fd, args, sizeof(args)) < 0)
570			err(1, "Writing to /dev/lguest");
571	}
572	/*:*/
573	
574	/*L:200
575	 * Device Handling.
576	 *
577	 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
578	 * We need to make sure it's not trying to reach into the Launcher itself, so
579	 * we have a convenient routine which checks it and exits with an error message
580	 * if something funny is going on:
581	 */
582	static void *_check_pointer(unsigned long addr, unsigned int size,
583				    unsigned int line)
584	{
585		/*
586		 * Check if the requested address and size exceeds the allocated memory,
587		 * or addr + size wraps around.
588		 */
589		if ((addr + size) > guest_limit || (addr + size) < addr)
590			errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
591		/*
592		 * We return a pointer for the caller's convenience, now we know it's
593		 * safe to use.
594		 */
595		return from_guest_phys(addr);
596	}
597	/* A macro which transparently hands the line number to the real function. */
598	#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
599	
600	/*
601	 * Each buffer in the virtqueues is actually a chain of descriptors.  This
602	 * function returns the next descriptor in the chain, or vq->vring.num if we're
603	 * at the end.
604	 */
605	static unsigned next_desc(struct vring_desc *desc,
606				  unsigned int i, unsigned int max)
607	{
608		unsigned int next;
609	
610		/* If this descriptor says it doesn't chain, we're done. */
611		if (!(desc[i].flags & VRING_DESC_F_NEXT))
612			return max;
613	
614		/* Check they're not leading us off end of descriptors. */
615		next = desc[i].next;
616		/* Make sure compiler knows to grab that: we don't want it changing! */
617		wmb();
618	
619		if (next >= max)
620			errx(1, "Desc next is %u", next);
621	
622		return next;
623	}
624	
625	/*
626	 * This actually sends the interrupt for this virtqueue, if we've used a
627	 * buffer.
628	 */
629	static void trigger_irq(struct virtqueue *vq)
630	{
631		unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
632	
633		/* Don't inform them if nothing used. */
634		if (!vq->pending_used)
635			return;
636		vq->pending_used = 0;
637	
638		/* If they don't want an interrupt, don't send one... */
639		if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
640			/* ... unless they've asked us to force one on empty. */
641			if (!vq->dev->irq_on_empty
642			    || lg_last_avail(vq) != vq->vring.avail->idx)
643				return;
644		}
645	
646		/* Send the Guest an interrupt tell them we used something up. */
647		if (write(lguest_fd, buf, sizeof(buf)) != 0)
648			err(1, "Triggering irq %i", vq->config.irq);
649	}
650	
651	/*
652	 * This looks in the virtqueue for the first available buffer, and converts
653	 * it to an iovec for convenient access.  Since descriptors consist of some
654	 * number of output then some number of input descriptors, it's actually two
655	 * iovecs, but we pack them into one and note how many of each there were.
656	 *
657	 * This function waits if necessary, and returns the descriptor number found.
658	 */
659	static unsigned wait_for_vq_desc(struct virtqueue *vq,
660					 struct iovec iov[],
661					 unsigned int *out_num, unsigned int *in_num)
662	{
663		unsigned int i, head, max;
664		struct vring_desc *desc;
665		u16 last_avail = lg_last_avail(vq);
666	
667		/* There's nothing available? */
668		while (last_avail == vq->vring.avail->idx) {
669			u64 event;
670	
671			/*
672			 * Since we're about to sleep, now is a good time to tell the
673			 * Guest about what we've used up to now.
674			 */
675			trigger_irq(vq);
676	
677			/* OK, now we need to know about added descriptors. */
678			vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
679	
680			/*
681			 * They could have slipped one in as we were doing that: make
682			 * sure it's written, then check again.
683			 */
684			mb();
685			if (last_avail != vq->vring.avail->idx) {
686				vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
687				break;
688			}
689	
690			/* Nothing new?  Wait for eventfd to tell us they refilled. */
691			if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
692				errx(1, "Event read failed?");
693	
694			/* We don't need to be notified again. */
695			vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
696		}
697	
698		/* Check it isn't doing very strange things with descriptor numbers. */
699		if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
700			errx(1, "Guest moved used index from %u to %u",
701			     last_avail, vq->vring.avail->idx);
702	
703		/*
704		 * Grab the next descriptor number they're advertising, and increment
705		 * the index we've seen.
706		 */
707		head = vq->vring.avail->ring[last_avail % vq->vring.num];
708		lg_last_avail(vq)++;
709	
710		/* If their number is silly, that's a fatal mistake. */
711		if (head >= vq->vring.num)
712			errx(1, "Guest says index %u is available", head);
713	
714		/* When we start there are none of either input nor output. */
715		*out_num = *in_num = 0;
716	
717		max = vq->vring.num;
718		desc = vq->vring.desc;
719		i = head;
720	
721		/*
722		 * If this is an indirect entry, then this buffer contains a descriptor
723		 * table which we handle as if it's any normal descriptor chain.
724		 */
725		if (desc[i].flags & VRING_DESC_F_INDIRECT) {
726			if (desc[i].len % sizeof(struct vring_desc))
727				errx(1, "Invalid size for indirect buffer table");
728	
729			max = desc[i].len / sizeof(struct vring_desc);
730			desc = check_pointer(desc[i].addr, desc[i].len);
731			i = 0;
732		}
733	
734		do {
735			/* Grab the first descriptor, and check it's OK. */
736			iov[*out_num + *in_num].iov_len = desc[i].len;
737			iov[*out_num + *in_num].iov_base
738				= check_pointer(desc[i].addr, desc[i].len);
739			/* If this is an input descriptor, increment that count. */
740			if (desc[i].flags & VRING_DESC_F_WRITE)
741				(*in_num)++;
742			else {
743				/*
744				 * If it's an output descriptor, they're all supposed
745				 * to come before any input descriptors.
746				 */
747				if (*in_num)
748					errx(1, "Descriptor has out after in");
749				(*out_num)++;
750			}
751	
752			/* If we've got too many, that implies a descriptor loop. */
753			if (*out_num + *in_num > max)
754				errx(1, "Looped descriptor");
755		} while ((i = next_desc(desc, i, max)) != max);
756	
757		return head;
758	}
759	
760	/*
761	 * After we've used one of their buffers, we tell the Guest about it.  Sometime
762	 * later we'll want to send them an interrupt using trigger_irq(); note that
763	 * wait_for_vq_desc() does that for us if it has to wait.
764	 */
765	static void add_used(struct virtqueue *vq, unsigned int head, int len)
766	{
767		struct vring_used_elem *used;
768	
769		/*
770		 * The virtqueue contains a ring of used buffers.  Get a pointer to the
771		 * next entry in that used ring.
772		 */
773		used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
774		used->id = head;
775		used->len = len;
776		/* Make sure buffer is written before we update index. */
777		wmb();
778		vq->vring.used->idx++;
779		vq->pending_used++;
780	}
781	
782	/* And here's the combo meal deal.  Supersize me! */
783	static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
784	{
785		add_used(vq, head, len);
786		trigger_irq(vq);
787	}
788	
789	/*
790	 * The Console
791	 *
792	 * We associate some data with the console for our exit hack.
793	 */
794	struct console_abort {
795		/* How many times have they hit ^C? */
796		int count;
797		/* When did they start? */
798		struct timeval start;
799	};
800	
801	/* This is the routine which handles console input (ie. stdin). */
802	static void console_input(struct virtqueue *vq)
803	{
804		int len;
805		unsigned int head, in_num, out_num;
806		struct console_abort *abort = vq->dev->priv;
807		struct iovec iov[vq->vring.num];
808	
809		/* Make sure there's a descriptor available. */
810		head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
811		if (out_num)
812			errx(1, "Output buffers in console in queue?");
813	
814		/* Read into it.  This is where we usually wait. */
815		len = readv(STDIN_FILENO, iov, in_num);
816		if (len <= 0) {
817			/* Ran out of input? */
818			warnx("Failed to get console input, ignoring console.");
819			/*
820			 * For simplicity, dying threads kill the whole Launcher.  So
821			 * just nap here.
822			 */
823			for (;;)
824				pause();
825		}
826	
827		/* Tell the Guest we used a buffer. */
828		add_used_and_trigger(vq, head, len);
829	
830		/*
831		 * Three ^C within one second?  Exit.
832		 *
833		 * This is such a hack, but works surprisingly well.  Each ^C has to
834		 * be in a buffer by itself, so they can't be too fast.  But we check
835		 * that we get three within about a second, so they can't be too
836		 * slow.
837		 */
838		if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
839			abort->count = 0;
840			return;
841		}
842	
843		abort->count++;
844		if (abort->count == 1)
845			gettimeofday(&abort->start, NULL);
846		else if (abort->count == 3) {
847			struct timeval now;
848			gettimeofday(&now, NULL);
849			/* Kill all Launcher processes with SIGINT, like normal ^C */
850			if (now.tv_sec <= abort->start.tv_sec+1)
851				kill(0, SIGINT);
852			abort->count = 0;
853		}
854	}
855	
856	/* This is the routine which handles console output (ie. stdout). */
857	static void console_output(struct virtqueue *vq)
858	{
859		unsigned int head, out, in;
860		struct iovec iov[vq->vring.num];
861	
862		/* We usually wait in here, for the Guest to give us something. */
863		head = wait_for_vq_desc(vq, iov, &out, &in);
864		if (in)
865			errx(1, "Input buffers in console output queue?");
866	
867		/* writev can return a partial write, so we loop here. */
868		while (!iov_empty(iov, out)) {
869			int len = writev(STDOUT_FILENO, iov, out);
870			if (len <= 0)
871				err(1, "Write to stdout gave %i", len);
872			iov_consume(iov, out, len);
873		}
874	
875		/*
876		 * We're finished with that buffer: if we're going to sleep,
877		 * wait_for_vq_desc() will prod the Guest with an interrupt.
878		 */
879		add_used(vq, head, 0);
880	}
881	
882	/*
883	 * The Network
884	 *
885	 * Handling output for network is also simple: we get all the output buffers
886	 * and write them to /dev/net/tun.
887	 */
888	struct net_info {
889		int tunfd;
890	};
891	
892	static void net_output(struct virtqueue *vq)
893	{
894		struct net_info *net_info = vq->dev->priv;
895		unsigned int head, out, in;
896		struct iovec iov[vq->vring.num];
897	
898		/* We usually wait in here for the Guest to give us a packet. */
899		head = wait_for_vq_desc(vq, iov, &out, &in);
900		if (in)
901			errx(1, "Input buffers in net output queue?");
902		/*
903		 * Send the whole thing through to /dev/net/tun.  It expects the exact
904		 * same format: what a coincidence!
905		 */
906		if (writev(net_info->tunfd, iov, out) < 0)
907			errx(1, "Write to tun failed?");
908	
909		/*
910		 * Done with that one; wait_for_vq_desc() will send the interrupt if
911		 * all packets are processed.
912		 */
913		add_used(vq, head, 0);
914	}
915	
916	/*
917	 * Handling network input is a bit trickier, because I've tried to optimize it.
918	 *
919	 * First we have a helper routine which tells is if from this file descriptor
920	 * (ie. the /dev/net/tun device) will block:
921	 */
922	static bool will_block(int fd)
923	{
924		fd_set fdset;
925		struct timeval zero = { 0, 0 };
926		FD_ZERO(&fdset);
927		FD_SET(fd, &fdset);
928		return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
929	}
930	
931	/*
932	 * This handles packets coming in from the tun device to our Guest.  Like all
933	 * service routines, it gets called again as soon as it returns, so you don't
934	 * see a while(1) loop here.
935	 */
936	static void net_input(struct virtqueue *vq)
937	{
938		int len;
939		unsigned int head, out, in;
940		struct iovec iov[vq->vring.num];
941		struct net_info *net_info = vq->dev->priv;
942	
943		/*
944		 * Get a descriptor to write an incoming packet into.  This will also
945		 * send an interrupt if they're out of descriptors.
946		 */
947		head = wait_for_vq_desc(vq, iov, &out, &in);
948		if (out)
949			errx(1, "Output buffers in net input queue?");
950	
951		/*
952		 * If it looks like we'll block reading from the tun device, send them
953		 * an interrupt.
954		 */
955		if (vq->pending_used && will_block(net_info->tunfd))
956			trigger_irq(vq);
957	
958		/*
959		 * Read in the packet.  This is where we normally wait (when there's no
960		 * incoming network traffic).
961		 */
962		len = readv(net_info->tunfd, iov, in);
963		if (len <= 0)
964			err(1, "Failed to read from tun.");
965	
966		/*
967		 * Mark that packet buffer as used, but don't interrupt here.  We want
968		 * to wait until we've done as much work as we can.
969		 */
970		add_used(vq, head, len);
971	}
972	/*:*/
973	
974	/* This is the helper to create threads: run the service routine in a loop. */
975	static int do_thread(void *_vq)
976	{
977		struct virtqueue *vq = _vq;
978	
979		for (;;)
980			vq->service(vq);
981		return 0;
982	}
983	
984	/*
985	 * When a child dies, we kill our entire process group with SIGTERM.  This
986	 * also has the side effect that the shell restores the console for us!
987	 */
988	static void kill_launcher(int signal)
989	{
990		kill(0, SIGTERM);
991	}
992	
993	static void reset_device(struct device *dev)
994	{
995		struct virtqueue *vq;
996	
997		verbose("Resetting device %s\n", dev->name);
998	
999		/* Clear any features they've acked. */
1000		memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
1001	
1002		/* We're going to be explicitly killing threads, so ignore them. */
1003		signal(SIGCHLD, SIG_IGN);
1004	
1005		/* Zero out the virtqueues, get rid of their threads */
1006		for (vq = dev->vq; vq; vq = vq->next) {
1007			if (vq->thread != (pid_t)-1) {
1008				kill(vq->thread, SIGTERM);
1009				waitpid(vq->thread, NULL, 0);
1010				vq->thread = (pid_t)-1;
1011			}
1012			memset(vq->vring.desc, 0,
1013			       vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1014			lg_last_avail(vq) = 0;
1015		}
1016		dev->running = false;
1017	
1018		/* Now we care if threads die. */
1019		signal(SIGCHLD, (void *)kill_launcher);
1020	}
1021	
1022	/*L:216
1023	 * This actually creates the thread which services the virtqueue for a device.
1024	 */
1025	static void create_thread(struct virtqueue *vq)
1026	{
1027		/*
1028		 * Create stack for thread.  Since the stack grows upwards, we point
1029		 * the stack pointer to the end of this region.
1030		 */
1031		char *stack = malloc(32768);
1032		unsigned long args[] = { LHREQ_EVENTFD,
1033					 vq->config.pfn*getpagesize(), 0 };
1034	
1035		/* Create a zero-initialized eventfd. */
1036		vq->eventfd = eventfd(0, 0);
1037		if (vq->eventfd < 0)
1038			err(1, "Creating eventfd");
1039		args[2] = vq->eventfd;
1040	
1041		/*
1042		 * Attach an eventfd to this virtqueue: it will go off when the Guest
1043		 * does an LHCALL_NOTIFY for this vq.
1044		 */
1045		if (write(lguest_fd, &args, sizeof(args)) != 0)
1046			err(1, "Attaching eventfd");
1047	
1048		/*
1049		 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1050		 * we get a signal if it dies.
1051		 */
1052		vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1053		if (vq->thread == (pid_t)-1)
1054			err(1, "Creating clone");
1055	
1056		/* We close our local copy now the child has it. */
1057		close(vq->eventfd);
1058	}
1059	
1060	static bool accepted_feature(struct device *dev, unsigned int bit)
1061	{
1062		const u8 *features = get_feature_bits(dev) + dev->feature_len;
1063	
1064		if (dev->feature_len < bit / CHAR_BIT)
1065			return false;
1066		return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT));
1067	}
1068	
1069	static void start_device(struct device *dev)
1070	{
1071		unsigned int i;
1072		struct virtqueue *vq;
1073	
1074		verbose("Device %s OK: offered", dev->name);
1075		for (i = 0; i < dev->feature_len; i++)
1076			verbose(" %02x", get_feature_bits(dev)[i]);
1077		verbose(", accepted");
1078		for (i = 0; i < dev->feature_len; i++)
1079			verbose(" %02x", get_feature_bits(dev)
1080				[dev->feature_len+i]);
1081	
1082		dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1083	
1084		for (vq = dev->vq; vq; vq = vq->next) {
1085			if (vq->service)
1086				create_thread(vq);
1087		}
1088		dev->running = true;
1089	}
1090	
1091	static void cleanup_devices(void)
1092	{
1093		struct device *dev;
1094	
1095		for (dev = devices.dev; dev; dev = dev->next)
1096			reset_device(dev);
1097	
1098		/* If we saved off the original terminal settings, restore them now. */
1099		if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1100			tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1101	}
1102	
1103	/* When the Guest tells us they updated the status field, we handle it. */
1104	static void update_device_status(struct device *dev)
1105	{
1106		/* A zero status is a reset, otherwise it's a set of flags. */
1107		if (dev->desc->status == 0)
1108			reset_device(dev);
1109		else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1110			warnx("Device %s configuration FAILED", dev->name);
1111			if (dev->running)
1112				reset_device(dev);
1113		} else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
1114			if (!dev->running)
1115				start_device(dev);
1116		}
1117	}
1118	
1119	/*L:215
1120	 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY.  In
1121	 * particular, it's used to notify us of device status changes during boot.
1122	 */
1123	static void handle_output(unsigned long addr)
1124	{
1125		struct device *i;
1126	
1127		/* Check each device. */
1128		for (i = devices.dev; i; i = i->next) {
1129			struct virtqueue *vq;
1130	
1131			/*
1132			 * Notifications to device descriptors mean they updated the
1133			 * device status.
1134			 */
1135			if (from_guest_phys(addr) == i->desc) {
1136				update_device_status(i);
1137				return;
1138			}
1139	
1140			/*
1141			 * Devices *can* be used before status is set to DRIVER_OK.
1142			 * The original plan was that they would never do this: they
1143			 * would always finish setting up their status bits before
1144			 * actually touching the virtqueues.  In practice, we allowed
1145			 * them to, and they do (eg. the disk probes for partition
1146			 * tables as part of initialization).
1147			 *
1148			 * If we see this, we start the device: once it's running, we
1149			 * expect the device to catch all the notifications.
1150			 */
1151			for (vq = i->vq; vq; vq = vq->next) {
1152				if (addr != vq->config.pfn*getpagesize())
1153					continue;
1154				if (i->running)
1155					errx(1, "Notification on running %s", i->name);
1156				/* This just calls create_thread() for each virtqueue */
1157				start_device(i);
1158				return;
1159			}
1160		}
1161	
1162		/*
1163		 * Early console write is done using notify on a nul-terminated string
1164		 * in Guest memory.  It's also great for hacking debugging messages
1165		 * into a Guest.
1166		 */
1167		if (addr >= guest_limit)
1168			errx(1, "Bad NOTIFY %#lx", addr);
1169	
1170		write(STDOUT_FILENO, from_guest_phys(addr),
1171		      strnlen(from_guest_phys(addr), guest_limit - addr));
1172	}
1173	
1174	/*L:190
1175	 * Device Setup
1176	 *
1177	 * All devices need a descriptor so the Guest knows it exists, and a "struct
1178	 * device" so the Launcher can keep track of it.  We have common helper
1179	 * routines to allocate and manage them.
1180	 */
1181	
1182	/*
1183	 * The layout of the device page is a "struct lguest_device_desc" followed by a
1184	 * number of virtqueue descriptors, then two sets of feature bits, then an
1185	 * array of configuration bytes.  This routine returns the configuration
1186	 * pointer.
1187	 */
1188	static u8 *device_config(const struct device *dev)
1189	{
1190		return (void *)(dev->desc + 1)
1191			+ dev->num_vq * sizeof(struct lguest_vqconfig)
1192			+ dev->feature_len * 2;
1193	}
1194	
1195	/*
1196	 * This routine allocates a new "struct lguest_device_desc" from descriptor
1197	 * table page just above the Guest's normal memory.  It returns a pointer to
1198	 * that descriptor.
1199	 */
1200	static struct lguest_device_desc *new_dev_desc(u16 type)
1201	{
1202		struct lguest_device_desc d = { .type = type };
1203		void *p;
1204	
1205		/* Figure out where the next device config is, based on the last one. */
1206		if (devices.lastdev)
1207			p = device_config(devices.lastdev)
1208				+ devices.lastdev->desc->config_len;
1209		else
1210			p = devices.descpage;
1211	
1212		/* We only have one page for all the descriptors. */
1213		if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1214			errx(1, "Too many devices");
1215	
1216		/* p might not be aligned, so we memcpy in. */
1217		return memcpy(p, &d, sizeof(d));
1218	}
1219	
1220	/*
1221	 * Each device descriptor is followed by the description of its virtqueues.  We
1222	 * specify how many descriptors the virtqueue is to have.
1223	 */
1224	static void add_virtqueue(struct device *dev, unsigned int num_descs,
1225				  void (*service)(struct virtqueue *))
1226	{
1227		unsigned int pages;
1228		struct virtqueue **i, *vq = malloc(sizeof(*vq));
1229		void *p;
1230	
1231		/* First we need some memory for this virtqueue. */
1232		pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1233			/ getpagesize();
1234		p = get_pages(pages);
1235	
1236		/* Initialize the virtqueue */
1237		vq->next = NULL;
1238		vq->last_avail_idx = 0;
1239		vq->dev = dev;
1240	
1241		/*
1242		 * This is the routine the service thread will run, and its Process ID
1243		 * once it's running.
1244		 */
1245		vq->service = service;
1246		vq->thread = (pid_t)-1;
1247	
1248		/* Initialize the configuration. */
1249		vq->config.num = num_descs;
1250		vq->config.irq = devices.next_irq++;
1251		vq->config.pfn = to_guest_phys(p) / getpagesize();
1252	
1253		/* Initialize the vring. */
1254		vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1255	
1256		/*
1257		 * Append virtqueue to this device's descriptor.  We use
1258		 * device_config() to get the end of the device's current virtqueues;
1259		 * we check that we haven't added any config or feature information
1260		 * yet, otherwise we'd be overwriting them.
1261		 */
1262		assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1263		memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1264		dev->num_vq++;
1265		dev->desc->num_vq++;
1266	
1267		verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1268	
1269		/*
1270		 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1271		 * second.
1272		 */
1273		for (i = &dev->vq; *i; i = &(*i)->next);
1274		*i = vq;
1275	}
1276	
1277	/*
1278	 * The first half of the feature bitmask is for us to advertise features.  The
1279	 * second half is for the Guest to accept features.
1280	 */
1281	static void add_feature(struct device *dev, unsigned bit)
1282	{
1283		u8 *features = get_feature_bits(dev);
1284	
1285		/* We can't extend the feature bits once we've added config bytes */
1286		if (dev->desc->feature_len <= bit / CHAR_BIT) {
1287			assert(dev->desc->config_len == 0);
1288			dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1289		}
1290	
1291		features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1292	}
1293	
1294	/*
1295	 * This routine sets the configuration fields for an existing device's
1296	 * descriptor.  It only works for the last device, but that's OK because that's
1297	 * how we use it.
1298	 */
1299	static void set_config(struct device *dev, unsigned len, const void *conf)
1300	{
1301		/* Check we haven't overflowed our single page. */
1302		if (device_config(dev) + len > devices.descpage + getpagesize())
1303			errx(1, "Too many devices");
1304	
1305		/* Copy in the config information, and store the length. */
1306		memcpy(device_config(dev), conf, len);
1307		dev->desc->config_len = len;
1308	
1309		/* Size must fit in config_len field (8 bits)! */
1310		assert(dev->desc->config_len == len);
1311	}
1312	
1313	/*
1314	 * This routine does all the creation and setup of a new device, including
1315	 * calling new_dev_desc() to allocate the descriptor and device memory.  We
1316	 * don't actually start the service threads until later.
1317	 *
1318	 * See what I mean about userspace being boring?
1319	 */
1320	static struct device *new_device(const char *name, u16 type)
1321	{
1322		struct device *dev = malloc(sizeof(*dev));
1323	
1324		/* Now we populate the fields one at a time. */
1325		dev->desc = new_dev_desc(type);
1326		dev->name = name;
1327		dev->vq = NULL;
1328		dev->feature_len = 0;
1329		dev->num_vq = 0;
1330		dev->running = false;
1331	
1332		/*
1333		 * Append to device list.  Prepending to a single-linked list is
1334		 * easier, but the user expects the devices to be arranged on the bus
1335		 * in command-line order.  The first network device on the command line
1336		 * is eth0, the first block device /dev/vda, etc.
1337		 */
1338		if (devices.lastdev)
1339			devices.lastdev->next = dev;
1340		else
1341			devices.dev = dev;
1342		devices.lastdev = dev;
1343	
1344		return dev;
1345	}
1346	
1347	/*
1348	 * Our first setup routine is the console.  It's a fairly simple device, but
1349	 * UNIX tty handling makes it uglier than it could be.
1350	 */
1351	static void setup_console(void)
1352	{
1353		struct device *dev;
1354	
1355		/* If we can save the initial standard input settings... */
1356		if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1357			struct termios term = orig_term;
1358			/*
1359			 * Then we turn off echo, line buffering and ^C etc: We want a
1360			 * raw input stream to the Guest.
1361			 */
1362			term.c_lflag &= ~(ISIG|ICANON|ECHO);
1363			tcsetattr(STDIN_FILENO, TCSANOW, &term);
1364		}
1365	
1366		dev = new_device("console", VIRTIO_ID_CONSOLE);
1367	
1368		/* We store the console state in dev->priv, and initialize it. */
1369		dev->priv = malloc(sizeof(struct console_abort));
1370		((struct console_abort *)dev->priv)->count = 0;
1371	
1372		/*
1373		 * The console needs two virtqueues: the input then the output.  When
1374		 * they put something the input queue, we make sure we're listening to
1375		 * stdin.  When they put something in the output queue, we write it to
1376		 * stdout.
1377		 */
1378		add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1379		add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1380	
1381		verbose("device %u: console\n", ++devices.device_num);
1382	}
1383	/*:*/
1384	
1385	/*M:010
1386	 * Inter-guest networking is an interesting area.  Simplest is to have a
1387	 * --sharenet=<name> option which opens or creates a named pipe.  This can be
1388	 * used to send packets to another guest in a 1:1 manner.
1389	 *
1390	 * More sopisticated is to use one of the tools developed for project like UML
1391	 * to do networking.
1392	 *
1393	 * Faster is to do virtio bonding in kernel.  Doing this 1:1 would be
1394	 * completely generic ("here's my vring, attach to your vring") and would work
1395	 * for any traffic.  Of course, namespace and permissions issues need to be
1396	 * dealt with.  A more sophisticated "multi-channel" virtio_net.c could hide
1397	 * multiple inter-guest channels behind one interface, although it would
1398	 * require some manner of hotplugging new virtio channels.
1399	 *
1400	 * Finally, we could implement a virtio network switch in the kernel.
1401	:*/
1402	
1403	static u32 str2ip(const char *ipaddr)
1404	{
1405		unsigned int b[4];
1406	
1407		if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1408			errx(1, "Failed to parse IP address '%s'", ipaddr);
1409		return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1410	}
1411	
1412	static void str2mac(const char *macaddr, unsigned char mac[6])
1413	{
1414		unsigned int m[6];
1415		if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1416			   &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1417			errx(1, "Failed to parse mac address '%s'", macaddr);
1418		mac[0] = m[0];
1419		mac[1] = m[1];
1420		mac[2] = m[2];
1421		mac[3] = m[3];
1422		mac[4] = m[4];
1423		mac[5] = m[5];
1424	}
1425	
1426	/*
1427	 * This code is "adapted" from libbridge: it attaches the Host end of the
1428	 * network device to the bridge device specified by the command line.
1429	 *
1430	 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1431	 * dislike bridging), and I just try not to break it.
1432	 */
1433	static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1434	{
1435		int ifidx;
1436		struct ifreq ifr;
1437	
1438		if (!*br_name)
1439			errx(1, "must specify bridge name");
1440	
1441		ifidx = if_nametoindex(if_name);
1442		if (!ifidx)
1443			errx(1, "interface %s does not exist!", if_name);
1444	
1445		strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1446		ifr.ifr_name[IFNAMSIZ-1] = '\0';
1447		ifr.ifr_ifindex = ifidx;
1448		if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1449			err(1, "can't add %s to bridge %s", if_name, br_name);
1450	}
1451	
1452	/*
1453	 * This sets up the Host end of the network device with an IP address, brings
1454	 * it up so packets will flow, the copies the MAC address into the hwaddr
1455	 * pointer.
1456	 */
1457	static void configure_device(int fd, const char *tapif, u32 ipaddr)
1458	{
1459		struct ifreq ifr;
1460		struct sockaddr_in sin;
1461	
1462		memset(&ifr, 0, sizeof(ifr));
1463		strcpy(ifr.ifr_name, tapif);
1464	
1465		/* Don't read these incantations.  Just cut & paste them like I did! */
1466		sin.sin_family = AF_INET;
1467		sin.sin_addr.s_addr = htonl(ipaddr);
1468		memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1469		if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1470			err(1, "Setting %s interface address", tapif);
1471		ifr.ifr_flags = IFF_UP;
1472		if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1473			err(1, "Bringing interface %s up", tapif);
1474	}
1475	
1476	static int get_tun_device(char tapif[IFNAMSIZ])
1477	{
1478		struct ifreq ifr;
1479		int netfd;
1480	
1481		/* Start with this zeroed.  Messy but sure. */
1482		memset(&ifr, 0, sizeof(ifr));
1483	
1484		/*
1485		 * We open the /dev/net/tun device and tell it we want a tap device.  A
1486		 * tap device is like a tun device, only somehow different.  To tell
1487		 * the truth, I completely blundered my way through this code, but it
1488		 * works now!
1489		 */
1490		netfd = open_or_die("/dev/net/tun", O_RDWR);
1491		ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1492		strcpy(ifr.ifr_name, "tap%d");
1493		if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1494			err(1, "configuring /dev/net/tun");
1495	
1496		if (ioctl(netfd, TUNSETOFFLOAD,
1497			  TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1498			err(1, "Could not set features for tun device");
1499	
1500		/*
1501		 * We don't need checksums calculated for packets coming in this
1502		 * device: trust us!
1503		 */
1504		ioctl(netfd, TUNSETNOCSUM, 1);
1505	
1506		memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1507		return netfd;
1508	}
1509	
1510	/*L:195
1511	 * Our network is a Host<->Guest network.  This can either use bridging or
1512	 * routing, but the principle is the same: it uses the "tun" device to inject
1513	 * packets into the Host as if they came in from a normal network card.  We
1514	 * just shunt packets between the Guest and the tun device.
1515	 */
1516	static void setup_tun_net(char *arg)
1517	{
1518		struct device *dev;
1519		struct net_info *net_info = malloc(sizeof(*net_info));
1520		int ipfd;
1521		u32 ip = INADDR_ANY;
1522		bool bridging = false;
1523		char tapif[IFNAMSIZ], *p;
1524		struct virtio_net_config conf;
1525	
1526		net_info->tunfd = get_tun_device(tapif);
1527	
1528		/* First we create a new network device. */
1529		dev = new_device("net", VIRTIO_ID_NET);
1530		dev->priv = net_info;
1531	
1532		/* Network devices need a recv and a send queue, just like console. */
1533		add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1534		add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1535	
1536		/*
1537		 * We need a socket to perform the magic network ioctls to bring up the
1538		 * tap interface, connect to the bridge etc.  Any socket will do!
1539		 */
1540		ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1541		if (ipfd < 0)
1542			err(1, "opening IP socket");
1543	
1544		/* If the command line was --tunnet=bridge:<name> do bridging. */
1545		if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1546			arg += strlen(BRIDGE_PFX);
1547			bridging = true;
1548		}
1549	
1550		/* A mac address may follow the bridge name or IP address */
1551		p = strchr(arg, ':');
1552		if (p) {
1553			str2mac(p+1, conf.mac);
1554			add_feature(dev, VIRTIO_NET_F_MAC);
1555			*p = '\0';
1556		}
1557	
1558		/* arg is now either an IP address or a bridge name */
1559		if (bridging)
1560			add_to_bridge(ipfd, tapif, arg);
1561		else
1562			ip = str2ip(arg);
1563	
1564		/* Set up the tun device. */
1565		configure_device(ipfd, tapif, ip);
1566	
1567		add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1568		/* Expect Guest to handle everything except UFO */
1569		add_feature(dev, VIRTIO_NET_F_CSUM);
1570		add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1571		add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1572		add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1573		add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1574		add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1575		add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1576		add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1577		/* We handle indirect ring entries */
1578		add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1579		set_config(dev, sizeof(conf), &conf);
1580	
1581		/* We don't need the socket any more; setup is done. */
1582		close(ipfd);
1583	
1584		devices.device_num++;
1585	
1586		if (bridging)
1587			verbose("device %u: tun %s attached to bridge: %s\n",
1588				devices.device_num, tapif, arg);
1589		else
1590			verbose("device %u: tun %s: %s\n",
1591				devices.device_num, tapif, arg);
1592	}
1593	/*:*/
1594	
1595	/* This hangs off device->priv. */
1596	struct vblk_info {
1597		/* The size of the file. */
1598		off64_t len;
1599	
1600		/* The file descriptor for the file. */
1601		int fd;
1602	
1603	};
1604	
1605	/*L:210
1606	 * The Disk
1607	 *
1608	 * The disk only has one virtqueue, so it only has one thread.  It is really
1609	 * simple: the Guest asks for a block number and we read or write that position
1610	 * in the file.
1611	 *
1612	 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1613	 * slow: the Guest waits until the read is finished before running anything
1614	 * else, even if it could have been doing useful work.
1615	 *
1616	 * We could have used async I/O, except it's reputed to suck so hard that
1617	 * characters actually go missing from your code when you try to use it.
1618	 */
1619	static void blk_request(struct virtqueue *vq)
1620	{
1621		struct vblk_info *vblk = vq->dev->priv;
1622		unsigned int head, out_num, in_num, wlen;
1623		int ret;
1624		u8 *in;
1625		struct virtio_blk_outhdr *out;
1626		struct iovec iov[vq->vring.num];
1627		off64_t off;
1628	
1629		/*
1630		 * Get the next request, where we normally wait.  It triggers the
1631		 * interrupt to acknowledge previously serviced requests (if any).
1632		 */
1633		head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1634	
1635		/*
1636		 * Every block request should contain at least one output buffer
1637		 * (detailing the location on disk and the type of request) and one
1638		 * input buffer (to hold the result).
1639		 */
1640		if (out_num == 0 || in_num == 0)
1641			errx(1, "Bad virtblk cmd %u out=%u in=%u",
1642			     head, out_num, in_num);
1643	
1644		out = convert(&iov[0], struct virtio_blk_outhdr);
1645		in = convert(&iov[out_num+in_num-1], u8);
1646		/*
1647		 * For historical reasons, block operations are expressed in 512 byte
1648		 * "sectors".
1649		 */
1650		off = out->sector * 512;
1651	
1652		/*
1653		 * In general the virtio block driver is allowed to try SCSI commands.
1654		 * It'd be nice if we supported eject, for example, but we don't.
1655		 */
1656		if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1657			fprintf(stderr, "Scsi commands unsupported\n");
1658			*in = VIRTIO_BLK_S_UNSUPP;
1659			wlen = sizeof(*in);
1660		} else if (out->type & VIRTIO_BLK_T_OUT) {
1661			/*
1662			 * Write
1663			 *
1664			 * Move to the right location in the block file.  This can fail
1665			 * if they try to write past end.
1666			 */
1667			if (lseek64(vblk->fd, off, SEEK_SET) != off)
1668				err(1, "Bad seek to sector %llu", out->sector);
1669	
1670			ret = writev(vblk->fd, iov+1, out_num-1);
1671			verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1672	
1673			/*
1674			 * Grr... Now we know how long the descriptor they sent was, we
1675			 * make sure they didn't try to write over the end of the block
1676			 * file (possibly extending it).
1677			 */
1678			if (ret > 0 && off + ret > vblk->len) {
1679				/* Trim it back to the correct length */
1680				ftruncate64(vblk->fd, vblk->len);
1681				/* Die, bad Guest, die. */
1682				errx(1, "Write past end %llu+%u", off, ret);
1683			}
1684	
1685			wlen = sizeof(*in);
1686			*in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1687		} else if (out->type & VIRTIO_BLK_T_FLUSH) {
1688			/* Flush */
1689			ret = fdatasync(vblk->fd);
1690			verbose("FLUSH fdatasync: %i\n", ret);
1691			wlen = sizeof(*in);
1692			*in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1693		} else {
1694			/*
1695			 * Read
1696			 *
1697			 * Move to the right location in the block file.  This can fail
1698			 * if they try to read past end.
1699			 */
1700			if (lseek64(vblk->fd, off, SEEK_SET) != off)
1701				err(1, "Bad seek to sector %llu", out->sector);
1702	
1703			ret = readv(vblk->fd, iov+1, in_num-1);
1704			verbose("READ from sector %llu: %i\n", out->sector, ret);
1705			if (ret >= 0) {
1706				wlen = sizeof(*in) + ret;
1707				*in = VIRTIO_BLK_S_OK;
1708			} else {
1709				wlen = sizeof(*in);
1710				*in = VIRTIO_BLK_S_IOERR;
1711			}
1712		}
1713	
1714		/* Finished that request. */
1715		add_used(vq, head, wlen);
1716	}
1717	
1718	/*L:198 This actually sets up a virtual block device. */
1719	static void setup_block_file(const char *filename)
1720	{
1721		struct device *dev;
1722		struct vblk_info *vblk;
1723		struct virtio_blk_config conf;
1724	
1725		/* Creat the device. */
1726		dev = new_device("block", VIRTIO_ID_BLOCK);
1727	
1728		/* The device has one virtqueue, where the Guest places requests. */
1729		add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1730	
1731		/* Allocate the room for our own bookkeeping */
1732		vblk = dev->priv = malloc(sizeof(*vblk));
1733	
1734		/* First we open the file and store the length. */
1735		vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1736		vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1737	
1738		/* We support FLUSH. */
1739		add_feature(dev, VIRTIO_BLK_F_FLUSH);
1740	
1741		/* Tell Guest how many sectors this device has. */
1742		conf.capacity = cpu_to_le64(vblk->len / 512);
1743	
1744		/*
1745		 * Tell Guest not to put in too many descriptors at once: two are used
1746		 * for the in and out elements.
1747		 */
1748		add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1749		conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1750	
1751		/* Don't try to put whole struct: we have 8 bit limit. */
1752		set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1753	
1754		verbose("device %u: virtblock %llu sectors\n",
1755			++devices.device_num, le64_to_cpu(conf.capacity));
1756	}
1757	
1758	/*L:211
1759	 * Our random number generator device reads from /dev/random into the Guest's
1760	 * input buffers.  The usual case is that the Guest doesn't want random numbers
1761	 * and so has no buffers although /dev/random is still readable, whereas
1762	 * console is the reverse.
1763	 *
1764	 * The same logic applies, however.
1765	 */
1766	struct rng_info {
1767		int rfd;
1768	};
1769	
1770	static void rng_input(struct virtqueue *vq)
1771	{
1772		int len;
1773		unsigned int head, in_num, out_num, totlen = 0;
1774		struct rng_info *rng_info = vq->dev->priv;
1775		struct iovec iov[vq->vring.num];
1776	
1777		/* First we need a buffer from the Guests's virtqueue. */
1778		head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1779		if (out_num)
1780			errx(1, "Output buffers in rng?");
1781	
1782		/*
1783		 * Just like the console write, we loop to cover the whole iovec.
1784		 * In this case, short reads actually happen quite a bit.
1785		 */
1786		while (!iov_empty(iov, in_num)) {
1787			len = readv(rng_info->rfd, iov, in_num);
1788			if (len <= 0)
1789				err(1, "Read from /dev/random gave %i", len);
1790			iov_consume(iov, in_num, len);
1791			totlen += len;
1792		}
1793	
1794		/* Tell the Guest about the new input. */
1795		add_used(vq, head, totlen);
1796	}
1797	
1798	/*L:199
1799	 * This creates a "hardware" random number device for the Guest.
1800	 */
1801	static void setup_rng(void)
1802	{
1803		struct device *dev;
1804		struct rng_info *rng_info = malloc(sizeof(*rng_info));
1805	
1806		/* Our device's privat info simply contains the /dev/random fd. */
1807		rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1808	
1809		/* Create the new device. */
1810		dev = new_device("rng", VIRTIO_ID_RNG);
1811		dev->priv = rng_info;
1812	
1813		/* The device has one virtqueue, where the Guest places inbufs. */
1814		add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1815	
1816		verbose("device %u: rng\n", devices.device_num++);
1817	}
1818	/* That's the end of device setup. */
1819	
1820	/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1821	static void __attribute__((noreturn)) restart_guest(void)
1822	{
1823		unsigned int i;
1824	
1825		/*
1826		 * Since we don't track all open fds, we simply close everything beyond
1827		 * stderr.
1828		 */
1829		for (i = 3; i < FD_SETSIZE; i++)
1830			close(i);
1831	
1832		/* Reset all the devices (kills all threads). */
1833		cleanup_devices();
1834	
1835		execv(main_args[0], main_args);
1836		err(1, "Could not exec %s", main_args[0]);
1837	}
1838	
1839	/*L:220
1840	 * Finally we reach the core of the Launcher which runs the Guest, serves
1841	 * its input and output, and finally, lays it to rest.
1842	 */
1843	static void __attribute__((noreturn)) run_guest(void)
1844	{
1845		for (;;) {
1846			unsigned long notify_addr;
1847			int readval;
1848	
1849			/* We read from the /dev/lguest device to run the Guest. */
1850			readval = pread(lguest_fd, &notify_addr,
1851					sizeof(notify_addr), cpu_id);
1852	
1853			/* One unsigned long means the Guest did HCALL_NOTIFY */
1854			if (readval == sizeof(notify_addr)) {
1855				verbose("Notify on address %#lx\n", notify_addr);
1856				handle_output(notify_addr);
1857			/* ENOENT means the Guest died.  Reading tells us why. */
1858			} else if (errno == ENOENT) {
1859				char reason[1024] = { 0 };
1860				pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1861				errx(1, "%s", reason);
1862			/* ERESTART means that we need to reboot the guest */
1863			} else if (errno == ERESTART) {
1864				restart_guest();
1865			/* Anything else means a bug or incompatible change. */
1866			} else
1867				err(1, "Running guest failed");
1868		}
1869	}
1870	/*L:240
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
1873	 * of us.
1874	 *
1875	 * Are you ready?  Take a deep breath and join me in the core of the Host, in
1876	 * "make Host".
1877	:*/
1878	
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' },
1885		{ "username", 1, NULL, 'u' },
1886		{ "chroot", 1, NULL, 'c' },
1887		{ NULL },
1888	};
1889	static void usage(void)
1890	{
1891		errx(1, "Usage: lguest [--verbose] "
1892		     "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1893		     "|--block=<filename>|--initrd=<filename>]...\n"
1894		     "<mem-in-mb> vmlinux [args...]");
1895	}
1896	
1897	/*L:105 The main routine is where the real work begins: */
1898	int main(int argc, char *argv[])
1899	{
1900		/* Memory, code startpoint and size of the (optional) initrd. */
1901		unsigned long mem = 0, start, initrd_size = 0;
1902		/* Two temporaries. */
1903		int i, c;
1904		/* The boot information for the Guest. */
1905		struct boot_params *boot;
1906		/* If they specify an initrd file to load. */
1907		const char *initrd_name = NULL;
1908	
1909		/* Password structure for initgroups/setres[gu]id */
1910		struct passwd *user_details = NULL;
1911	
1912		/* Directory to chroot to */
1913		char *chroot_path = NULL;
1914	
1915		/* Save the args: we "reboot" by execing ourselves again. */
1916		main_args = argv;
1917	
1918		/*
1919		 * First we initialize the device list.  We keep a pointer to the last
1920		 * device, and the next interrupt number to use for devices (1:
1921		 * remember that 0 is used by the timer).
1922		 */
1923		devices.lastdev = NULL;
1924		devices.next_irq = 1;
1925	
1926		/* We're CPU 0.  In fact, that's the only CPU possible right now. */
1927		cpu_id = 0;
1928	
1929		/*
1930		 * We need to know how much memory so we can set up the device
1931		 * descriptor and memory pages for the devices as we parse the command
1932		 * line.  So we quickly look through the arguments to find the amount
1933		 * of memory now.
1934		 */
1935		for (i = 1; i < argc; i++) {
1936			if (argv[i][0] != '-') {
1937				mem = atoi(argv[i]) * 1024 * 1024;
1938				/*
1939				 * We start by mapping anonymous pages over all of
1940				 * guest-physical memory range.  This fills it with 0,
1941				 * and ensures that the Guest won't be killed when it
1942				 * tries to access it.
1943				 */
1944				guest_base = map_zeroed_pages(mem / getpagesize()
1945							      + DEVICE_PAGES);
1946				guest_limit = mem;
1947				guest_max = mem + DEVICE_PAGES*getpagesize();
1948				devices.descpage = get_pages(1);
1949				break;
1950			}
1951		}
1952	
1953		/* The options are fairly straight-forward */
1954		while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1955			switch (c) {
1956			case 'v':
1957				verbose = true;
1958				break;
1959			case 't':
1960				setup_tun_net(optarg);
1961				break;
1962			case 'b':
1963				setup_block_file(optarg);
1964				break;
1965			case 'r':
1966				setup_rng();
1967				break;
1968			case 'i':
1969				initrd_name = optarg;
1970				break;
1971			case 'u':
1972				user_details = getpwnam(optarg);
1973				if (!user_details)
1974					err(1, "getpwnam failed, incorrect username?");
1975				break;
1976			case 'c':
1977				chroot_path = optarg;
1978				break;
1979			default:
1980				warnx("Unknown argument %s", argv[optind]);
1981				usage();
1982			}
1983		}
1984		/*
1985		 * After the other arguments we expect memory and kernel image name,
1986		 * followed by command line arguments for the kernel.
1987		 */
1988		if (optind + 2 > argc)
1989			usage();
1990	
1991		verbose("Guest base is at %p\n", guest_base);
1992	
1993		/* We always have a console device */
1994		setup_console();
1995	
1996		/* Now we load the kernel */
1997		start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1998	
1999		/* Boot information is stashed at physical address 0 */
2000		boot = from_guest_phys(0);
2001	
2002		/* Map the initrd image if requested (at top of physical memory) */
2003		if (initrd_name) {
2004			initrd_size = load_initrd(initrd_name, mem);
2005			/*
2006			 * These are the location in the Linux boot header where the
2007			 * start and size of the initrd are expected to be found.
2008			 */
2009			boot->hdr.ramdisk_image = mem - initrd_size;
2010			boot->hdr.ramdisk_size = initrd_size;
2011			/* The bootloader type 0xFF means "unknown"; that's OK. */
2012			boot->hdr.type_of_loader = 0xFF;
2013		}
2014	
2015		/*
2016		 * The Linux boot header contains an "E820" memory map: ours is a
2017		 * simple, single region.
2018		 */
2019		boot->e820_entries = 1;
2020		boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
2021		/*
2022		 * The boot header contains a command line pointer: we put the command
2023		 * line after the boot header.
2024		 */
2025		boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2026		/* We use a simple helper to copy the arguments separated by spaces. */
2027		concat((char *)(boot + 1), argv+optind+2);
2028	
2029		/* Boot protocol version: 2.07 supports the fields for lguest. */
2030		boot->hdr.version = 0x207;
2031	
2032		/* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2033		boot->hdr.hardware_subarch = 1;
2034	
2035		/* Tell the entry path not to try to reload segment registers. */
2036		boot->hdr.loadflags |= KEEP_SEGMENTS;
2037	
2038		/*
2039		 * We tell the kernel to initialize the Guest: this returns the open
2040		 * /dev/lguest file descriptor.
2041		 */
2042		tell_kernel(start);
2043	
2044		/* Ensure that we terminate if a device-servicing child dies. */
2045		signal(SIGCHLD, kill_launcher);
2046	
2047		/* If we exit via err(), this kills all the threads, restores tty. */
2048		atexit(cleanup_devices);
2049	
2050		/* If requested, chroot to a directory */
2051		if (chroot_path) {
2052			if (chroot(chroot_path) != 0)
2053				err(1, "chroot(\"%s\") failed", chroot_path);
2054	
2055			if (chdir("/") != 0)
2056				err(1, "chdir(\"/\") failed");
2057	
2058			verbose("chroot done\n");
2059		}
2060	
2061		/* If requested, drop privileges */
2062		if (user_details) {
2063			uid_t u;
2064			gid_t g;
2065	
2066			u = user_details->pw_uid;
2067			g = user_details->pw_gid;
2068	
2069			if (initgroups(user_details->pw_name, g) != 0)
2070				err(1, "initgroups failed");
2071	
2072			if (setresgid(g, g, g) != 0)
2073				err(1, "setresgid failed");
2074	
2075			if (setresuid(u, u, u) != 0)
2076				err(1, "setresuid failed");
2077	
2078			verbose("Dropping privileges completed\n");
2079		}
2080	
2081		/* Finally, run the Guest.  This doesn't return. */
2082		run_guest();
2083	}
2084	/*:*/
2085	
2086	/*M:999
2087	 * Mastery is done: you now know everything I do.
2088	 *
2089	 * But surely you have seen code, features and bugs in your wanderings which
2090	 * you now yearn to attack?  That is the real game, and I look forward to you
2091	 * patching and forking lguest into the Your-Name-Here-visor.
2092	 *
2093	 * Farewell, and good coding!
2094	 * Rusty Russell.
2095	 */
Hide Line Numbers
About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Information is copyright its respective author. All material is available from the Linux Kernel Source distributed under a GPL License. This page is provided as a free service by mjmwired.net.