lguest_user.c revision 3c6b5bfa3cf3b4057788e08482a468cc3bc00780
1/*P:200 This contains all the /dev/lguest code, whereby the userspace launcher
2 * controls and communicates with the Guest.  For example, the first write will
3 * tell us the Guest's memory layout, pagetable, entry point and kernel address
4 * offset.  A read will run the Guest until something happens, such as a signal
5 * or the Guest doing a DMA out to the Launcher.  Writes are also used to get a
6 * DMA buffer registered by the Guest and to send the Guest an interrupt. :*/
7#include <linux/uaccess.h>
8#include <linux/miscdevice.h>
9#include <linux/fs.h>
10#include "lg.h"
11
12/*L:030 setup_regs() doesn't really belong in this file, but it gives us an
13 * early glimpse deeper into the Host so it's worth having here.
14 *
15 * Most of the Guest's registers are left alone: we used get_zeroed_page() to
16 * allocate the structure, so they will be 0. */
17static void setup_regs(struct lguest_regs *regs, unsigned long start)
18{
19	/* There are four "segment" registers which the Guest needs to boot:
20	 * The "code segment" register (cs) refers to the kernel code segment
21	 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
22	 * refer to the kernel data segment __KERNEL_DS.
23	 *
24	 * The privilege level is packed into the lower bits.  The Guest runs
25	 * at privilege level 1 (GUEST_PL).*/
26	regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
27	regs->cs = __KERNEL_CS|GUEST_PL;
28
29	/* The "eflags" register contains miscellaneous flags.  Bit 1 (0x002)
30	 * is supposed to always be "1".  Bit 9 (0x200) controls whether
31	 * interrupts are enabled.  We always leave interrupts enabled while
32	 * running the Guest. */
33	regs->eflags = 0x202;
34
35	/* The "Extended Instruction Pointer" register says where the Guest is
36	 * running. */
37	regs->eip = start;
38
39	/* %esi points to our boot information, at physical address 0, so don't
40	 * touch it. */
41}
42
43/*L:310 To send DMA into the Guest, the Launcher needs to be able to ask for a
44 * DMA buffer.  This is done by writing LHREQ_GETDMA and the key to
45 * /dev/lguest. */
46static long user_get_dma(struct lguest *lg, const u32 __user *input)
47{
48	unsigned long key, udma, irq;
49
50	/* Fetch the key they wrote to us. */
51	if (get_user(key, input) != 0)
52		return -EFAULT;
53	/* Look for a free Guest DMA buffer bound to that key. */
54	udma = get_dma_buffer(lg, key, &irq);
55	if (!udma)
56		return -ENOENT;
57
58	/* We need to tell the Launcher what interrupt the Guest expects after
59	 * the buffer is filled.  We stash it in udma->used_len. */
60	lgwrite_u32(lg, udma + offsetof(struct lguest_dma, used_len), irq);
61
62	/* The (guest-physical) address of the DMA buffer is returned from
63	 * the write(). */
64	return udma;
65}
66
67/*L:315 To force the Guest to stop running and return to the Launcher, the
68 * Waker sets writes LHREQ_BREAK and the value "1" to /dev/lguest.  The
69 * Launcher then writes LHREQ_BREAK and "0" to release the Waker. */
70static int break_guest_out(struct lguest *lg, const u32 __user *input)
71{
72	unsigned long on;
73
74	/* Fetch whether they're turning break on or off.. */
75	if (get_user(on, input) != 0)
76		return -EFAULT;
77
78	if (on) {
79		lg->break_out = 1;
80		/* Pop it out (may be running on different CPU) */
81		wake_up_process(lg->tsk);
82		/* Wait for them to reset it */
83		return wait_event_interruptible(lg->break_wq, !lg->break_out);
84	} else {
85		lg->break_out = 0;
86		wake_up(&lg->break_wq);
87		return 0;
88	}
89}
90
91/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
92 * number to /dev/lguest. */
93static int user_send_irq(struct lguest *lg, const u32 __user *input)
94{
95	u32 irq;
96
97	if (get_user(irq, input) != 0)
98		return -EFAULT;
99	if (irq >= LGUEST_IRQS)
100		return -EINVAL;
101	/* Next time the Guest runs, the core code will see if it can deliver
102	 * this interrupt. */
103	set_bit(irq, lg->irqs_pending);
104	return 0;
105}
106
107/*L:040 Once our Guest is initialized, the Launcher makes it run by reading
108 * from /dev/lguest. */
109static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
110{
111	struct lguest *lg = file->private_data;
112
113	/* You must write LHREQ_INITIALIZE first! */
114	if (!lg)
115		return -EINVAL;
116
117	/* If you're not the task which owns the guest, go away. */
118	if (current != lg->tsk)
119		return -EPERM;
120
121	/* If the guest is already dead, we indicate why */
122	if (lg->dead) {
123		size_t len;
124
125		/* lg->dead either contains an error code, or a string. */
126		if (IS_ERR(lg->dead))
127			return PTR_ERR(lg->dead);
128
129		/* We can only return as much as the buffer they read with. */
130		len = min(size, strlen(lg->dead)+1);
131		if (copy_to_user(user, lg->dead, len) != 0)
132			return -EFAULT;
133		return len;
134	}
135
136	/* If we returned from read() last time because the Guest sent DMA,
137	 * clear the flag. */
138	if (lg->dma_is_pending)
139		lg->dma_is_pending = 0;
140
141	/* Run the Guest until something interesting happens. */
142	return run_guest(lg, (unsigned long __user *)user);
143}
144
145/*L:020 The initialization write supplies 5 32-bit values (in addition to the
146 * 32-bit LHREQ_INITIALIZE value).  These are:
147 *
148 * base: The start of the Guest-physical memory inside the Launcher memory.
149 *
150 * pfnlimit: The highest (Guest-physical) page number the Guest should be
151 * allowed to access.  The Launcher has to live in Guest memory, so it sets
152 * this to ensure the Guest can't reach it.
153 *
154 * pgdir: The (Guest-physical) address of the top of the initial Guest
155 * pagetables (which are set up by the Launcher).
156 *
157 * start: The first instruction to execute ("eip" in x86-speak).
158 *
159 * page_offset: The PAGE_OFFSET constant in the Guest kernel.  We should
160 * probably wean the code off this, but it's a very useful constant!  Any
161 * address above this is within the Guest kernel, and any kernel address can
162 * quickly converted from physical to virtual by adding PAGE_OFFSET.  It's
163 * 0xC0000000 (3G) by default, but it's configurable at kernel build time.
164 */
165static int initialize(struct file *file, const u32 __user *input)
166{
167	/* "struct lguest" contains everything we (the Host) know about a
168	 * Guest. */
169	struct lguest *lg;
170	int err, i;
171	u32 args[5];
172
173	/* We grab the Big Lguest lock, which protects the global array
174	 * "lguests" and multiple simultaneous initializations. */
175	mutex_lock(&lguest_lock);
176	/* You can't initialize twice!  Close the device and start again... */
177	if (file->private_data) {
178		err = -EBUSY;
179		goto unlock;
180	}
181
182	if (copy_from_user(args, input, sizeof(args)) != 0) {
183		err = -EFAULT;
184		goto unlock;
185	}
186
187	/* Find an unused guest. */
188	i = find_free_guest();
189	if (i < 0) {
190		err = -ENOSPC;
191		goto unlock;
192	}
193	/* OK, we have an index into the "lguest" array: "lg" is a convenient
194	 * pointer. */
195	lg = &lguests[i];
196
197	/* Populate the easy fields of our "struct lguest" */
198	lg->guestid = i;
199	lg->mem_base = (void __user *)(long)args[0];
200	lg->pfn_limit = args[1];
201	lg->page_offset = args[4];
202
203	/* We need a complete page for the Guest registers: they are accessible
204	 * to the Guest and we can only grant it access to whole pages. */
205	lg->regs_page = get_zeroed_page(GFP_KERNEL);
206	if (!lg->regs_page) {
207		err = -ENOMEM;
208		goto release_guest;
209	}
210	/* We actually put the registers at the bottom of the page. */
211	lg->regs = (void *)lg->regs_page + PAGE_SIZE - sizeof(*lg->regs);
212
213	/* Initialize the Guest's shadow page tables, using the toplevel
214	 * address the Launcher gave us.  This allocates memory, so can
215	 * fail. */
216	err = init_guest_pagetable(lg, args[2]);
217	if (err)
218		goto free_regs;
219
220	/* Now we initialize the Guest's registers, handing it the start
221	 * address. */
222	setup_regs(lg->regs, args[3]);
223
224	/* There are a couple of GDT entries the Guest expects when first
225	 * booting. */
226	setup_guest_gdt(lg);
227
228	/* The timer for lguest's clock needs initialization. */
229	init_clockdev(lg);
230
231	/* We keep a pointer to the Launcher task (ie. current task) for when
232	 * other Guests want to wake this one (inter-Guest I/O). */
233	lg->tsk = current;
234	/* We need to keep a pointer to the Launcher's memory map, because if
235	 * the Launcher dies we need to clean it up.  If we don't keep a
236	 * reference, it is destroyed before close() is called. */
237	lg->mm = get_task_mm(lg->tsk);
238
239	/* Initialize the queue for the waker to wait on */
240	init_waitqueue_head(&lg->break_wq);
241
242	/* We remember which CPU's pages this Guest used last, for optimization
243	 * when the same Guest runs on the same CPU twice. */
244	lg->last_pages = NULL;
245
246	/* We keep our "struct lguest" in the file's private_data. */
247	file->private_data = lg;
248
249	mutex_unlock(&lguest_lock);
250
251	/* And because this is a write() call, we return the length used. */
252	return sizeof(args);
253
254free_regs:
255	free_page(lg->regs_page);
256release_guest:
257	memset(lg, 0, sizeof(*lg));
258unlock:
259	mutex_unlock(&lguest_lock);
260	return err;
261}
262
263/*L:010 The first operation the Launcher does must be a write.  All writes
264 * start with a 32 bit number: for the first write this must be
265 * LHREQ_INITIALIZE to set up the Guest.  After that the Launcher can use
266 * writes of other values to get DMA buffers and send interrupts. */
267static ssize_t write(struct file *file, const char __user *input,
268		     size_t size, loff_t *off)
269{
270	/* Once the guest is initialized, we hold the "struct lguest" in the
271	 * file private data. */
272	struct lguest *lg = file->private_data;
273	u32 req;
274
275	if (get_user(req, input) != 0)
276		return -EFAULT;
277	input += sizeof(req);
278
279	/* If you haven't initialized, you must do that first. */
280	if (req != LHREQ_INITIALIZE && !lg)
281		return -EINVAL;
282
283	/* Once the Guest is dead, all you can do is read() why it died. */
284	if (lg && lg->dead)
285		return -ENOENT;
286
287	/* If you're not the task which owns the Guest, you can only break */
288	if (lg && current != lg->tsk && req != LHREQ_BREAK)
289		return -EPERM;
290
291	switch (req) {
292	case LHREQ_INITIALIZE:
293		return initialize(file, (const u32 __user *)input);
294	case LHREQ_GETDMA:
295		return user_get_dma(lg, (const u32 __user *)input);
296	case LHREQ_IRQ:
297		return user_send_irq(lg, (const u32 __user *)input);
298	case LHREQ_BREAK:
299		return break_guest_out(lg, (const u32 __user *)input);
300	default:
301		return -EINVAL;
302	}
303}
304
305/*L:060 The final piece of interface code is the close() routine.  It reverses
306 * everything done in initialize().  This is usually called because the
307 * Launcher exited.
308 *
309 * Note that the close routine returns 0 or a negative error number: it can't
310 * really fail, but it can whine.  I blame Sun for this wart, and K&R C for
311 * letting them do it. :*/
312static int close(struct inode *inode, struct file *file)
313{
314	struct lguest *lg = file->private_data;
315
316	/* If we never successfully initialized, there's nothing to clean up */
317	if (!lg)
318		return 0;
319
320	/* We need the big lock, to protect from inter-guest I/O and other
321	 * Launchers initializing guests. */
322	mutex_lock(&lguest_lock);
323	/* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
324	hrtimer_cancel(&lg->hrt);
325	/* Free any DMA buffers the Guest had bound. */
326	release_all_dma(lg);
327	/* Free up the shadow page tables for the Guest. */
328	free_guest_pagetable(lg);
329	/* Now all the memory cleanups are done, it's safe to release the
330	 * Launcher's memory management structure. */
331	mmput(lg->mm);
332	/* If lg->dead doesn't contain an error code it will be NULL or a
333	 * kmalloc()ed string, either of which is ok to hand to kfree(). */
334	if (!IS_ERR(lg->dead))
335		kfree(lg->dead);
336	/* We can free up the register page we allocated. */
337	free_page(lg->regs_page);
338	/* We clear the entire structure, which also marks it as free for the
339	 * next user. */
340	memset(lg, 0, sizeof(*lg));
341	/* Release lock and exit. */
342	mutex_unlock(&lguest_lock);
343
344	return 0;
345}
346
347/*L:000
348 * Welcome to our journey through the Launcher!
349 *
350 * The Launcher is the Host userspace program which sets up, runs and services
351 * the Guest.  In fact, many comments in the Drivers which refer to "the Host"
352 * doing things are inaccurate: the Launcher does all the device handling for
353 * the Guest.  The Guest can't tell what's done by the the Launcher and what by
354 * the Host.
355 *
356 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
357 * shall see more of that later.
358 *
359 * We begin our understanding with the Host kernel interface which the Launcher
360 * uses: reading and writing a character device called /dev/lguest.  All the
361 * work happens in the read(), write() and close() routines: */
362static struct file_operations lguest_fops = {
363	.owner	 = THIS_MODULE,
364	.release = close,
365	.write	 = write,
366	.read	 = read,
367};
368
369/* This is a textbook example of a "misc" character device.  Populate a "struct
370 * miscdevice" and register it with misc_register(). */
371static struct miscdevice lguest_dev = {
372	.minor	= MISC_DYNAMIC_MINOR,
373	.name	= "lguest",
374	.fops	= &lguest_fops,
375};
376
377int __init lguest_device_init(void)
378{
379	return misc_register(&lguest_dev);
380}
381
382void __exit lguest_device_remove(void)
383{
384	misc_deregister(&lguest_dev);
385}
386