1 |
62 |
marcus.erl |
/*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 NOTIFY out to the Launcher. :*/
|
6 |
|
|
#include <linux/uaccess.h>
|
7 |
|
|
#include <linux/miscdevice.h>
|
8 |
|
|
#include <linux/fs.h>
|
9 |
|
|
#include "lg.h"
|
10 |
|
|
|
11 |
|
|
/*L:055 When something happens, the Waker process needs a way to stop the
|
12 |
|
|
* kernel running the Guest and return to the Launcher. So the Waker writes
|
13 |
|
|
* LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher
|
14 |
|
|
* has done whatever needs attention, it writes LHREQ_BREAK and "0" to release
|
15 |
|
|
* the Waker. */
|
16 |
|
|
static int break_guest_out(struct lguest *lg, const unsigned long __user *input)
|
17 |
|
|
{
|
18 |
|
|
unsigned long on;
|
19 |
|
|
|
20 |
|
|
/* Fetch whether they're turning break on or off. */
|
21 |
|
|
if (get_user(on, input) != 0)
|
22 |
|
|
return -EFAULT;
|
23 |
|
|
|
24 |
|
|
if (on) {
|
25 |
|
|
lg->break_out = 1;
|
26 |
|
|
/* Pop it out of the Guest (may be running on different CPU) */
|
27 |
|
|
wake_up_process(lg->tsk);
|
28 |
|
|
/* Wait for them to reset it */
|
29 |
|
|
return wait_event_interruptible(lg->break_wq, !lg->break_out);
|
30 |
|
|
} else {
|
31 |
|
|
lg->break_out = 0;
|
32 |
|
|
wake_up(&lg->break_wq);
|
33 |
|
|
return 0;
|
34 |
|
|
}
|
35 |
|
|
}
|
36 |
|
|
|
37 |
|
|
/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
|
38 |
|
|
* number to /dev/lguest. */
|
39 |
|
|
static int user_send_irq(struct lguest *lg, const unsigned long __user *input)
|
40 |
|
|
{
|
41 |
|
|
unsigned long irq;
|
42 |
|
|
|
43 |
|
|
if (get_user(irq, input) != 0)
|
44 |
|
|
return -EFAULT;
|
45 |
|
|
if (irq >= LGUEST_IRQS)
|
46 |
|
|
return -EINVAL;
|
47 |
|
|
/* Next time the Guest runs, the core code will see if it can deliver
|
48 |
|
|
* this interrupt. */
|
49 |
|
|
set_bit(irq, lg->irqs_pending);
|
50 |
|
|
return 0;
|
51 |
|
|
}
|
52 |
|
|
|
53 |
|
|
/*L:040 Once our Guest is initialized, the Launcher makes it run by reading
|
54 |
|
|
* from /dev/lguest. */
|
55 |
|
|
static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
|
56 |
|
|
{
|
57 |
|
|
struct lguest *lg = file->private_data;
|
58 |
|
|
|
59 |
|
|
/* You must write LHREQ_INITIALIZE first! */
|
60 |
|
|
if (!lg)
|
61 |
|
|
return -EINVAL;
|
62 |
|
|
|
63 |
|
|
/* If you're not the task which owns the Guest, go away. */
|
64 |
|
|
if (current != lg->tsk)
|
65 |
|
|
return -EPERM;
|
66 |
|
|
|
67 |
|
|
/* If the guest is already dead, we indicate why */
|
68 |
|
|
if (lg->dead) {
|
69 |
|
|
size_t len;
|
70 |
|
|
|
71 |
|
|
/* lg->dead either contains an error code, or a string. */
|
72 |
|
|
if (IS_ERR(lg->dead))
|
73 |
|
|
return PTR_ERR(lg->dead);
|
74 |
|
|
|
75 |
|
|
/* We can only return as much as the buffer they read with. */
|
76 |
|
|
len = min(size, strlen(lg->dead)+1);
|
77 |
|
|
if (copy_to_user(user, lg->dead, len) != 0)
|
78 |
|
|
return -EFAULT;
|
79 |
|
|
return len;
|
80 |
|
|
}
|
81 |
|
|
|
82 |
|
|
/* If we returned from read() last time because the Guest notified,
|
83 |
|
|
* clear the flag. */
|
84 |
|
|
if (lg->pending_notify)
|
85 |
|
|
lg->pending_notify = 0;
|
86 |
|
|
|
87 |
|
|
/* Run the Guest until something interesting happens. */
|
88 |
|
|
return run_guest(lg, (unsigned long __user *)user);
|
89 |
|
|
}
|
90 |
|
|
|
91 |
|
|
/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit)
|
92 |
|
|
* values (in addition to the LHREQ_INITIALIZE value). These are:
|
93 |
|
|
*
|
94 |
|
|
* base: The start of the Guest-physical memory inside the Launcher memory.
|
95 |
|
|
*
|
96 |
|
|
* pfnlimit: The highest (Guest-physical) page number the Guest should be
|
97 |
|
|
* allowed to access. The Guest memory lives inside the Launcher, so it sets
|
98 |
|
|
* this to ensure the Guest can only reach its own memory.
|
99 |
|
|
*
|
100 |
|
|
* pgdir: The (Guest-physical) address of the top of the initial Guest
|
101 |
|
|
* pagetables (which are set up by the Launcher).
|
102 |
|
|
*
|
103 |
|
|
* start: The first instruction to execute ("eip" in x86-speak).
|
104 |
|
|
*/
|
105 |
|
|
static int initialize(struct file *file, const unsigned long __user *input)
|
106 |
|
|
{
|
107 |
|
|
/* "struct lguest" contains everything we (the Host) know about a
|
108 |
|
|
* Guest. */
|
109 |
|
|
struct lguest *lg;
|
110 |
|
|
int err;
|
111 |
|
|
unsigned long args[4];
|
112 |
|
|
|
113 |
|
|
/* We grab the Big Lguest lock, which protects against multiple
|
114 |
|
|
* simultaneous initializations. */
|
115 |
|
|
mutex_lock(&lguest_lock);
|
116 |
|
|
/* You can't initialize twice! Close the device and start again... */
|
117 |
|
|
if (file->private_data) {
|
118 |
|
|
err = -EBUSY;
|
119 |
|
|
goto unlock;
|
120 |
|
|
}
|
121 |
|
|
|
122 |
|
|
if (copy_from_user(args, input, sizeof(args)) != 0) {
|
123 |
|
|
err = -EFAULT;
|
124 |
|
|
goto unlock;
|
125 |
|
|
}
|
126 |
|
|
|
127 |
|
|
lg = kzalloc(sizeof(*lg), GFP_KERNEL);
|
128 |
|
|
if (!lg) {
|
129 |
|
|
err = -ENOMEM;
|
130 |
|
|
goto unlock;
|
131 |
|
|
}
|
132 |
|
|
|
133 |
|
|
/* Populate the easy fields of our "struct lguest" */
|
134 |
|
|
lg->mem_base = (void __user *)(long)args[0];
|
135 |
|
|
lg->pfn_limit = args[1];
|
136 |
|
|
|
137 |
|
|
/* We need a complete page for the Guest registers: they are accessible
|
138 |
|
|
* to the Guest and we can only grant it access to whole pages. */
|
139 |
|
|
lg->regs_page = get_zeroed_page(GFP_KERNEL);
|
140 |
|
|
if (!lg->regs_page) {
|
141 |
|
|
err = -ENOMEM;
|
142 |
|
|
goto release_guest;
|
143 |
|
|
}
|
144 |
|
|
/* We actually put the registers at the bottom of the page. */
|
145 |
|
|
lg->regs = (void *)lg->regs_page + PAGE_SIZE - sizeof(*lg->regs);
|
146 |
|
|
|
147 |
|
|
/* Initialize the Guest's shadow page tables, using the toplevel
|
148 |
|
|
* address the Launcher gave us. This allocates memory, so can
|
149 |
|
|
* fail. */
|
150 |
|
|
err = init_guest_pagetable(lg, args[2]);
|
151 |
|
|
if (err)
|
152 |
|
|
goto free_regs;
|
153 |
|
|
|
154 |
|
|
/* Now we initialize the Guest's registers, handing it the start
|
155 |
|
|
* address. */
|
156 |
|
|
lguest_arch_setup_regs(lg, args[3]);
|
157 |
|
|
|
158 |
|
|
/* The timer for lguest's clock needs initialization. */
|
159 |
|
|
init_clockdev(lg);
|
160 |
|
|
|
161 |
|
|
/* We keep a pointer to the Launcher task (ie. current task) for when
|
162 |
|
|
* other Guests want to wake this one (inter-Guest I/O). */
|
163 |
|
|
lg->tsk = current;
|
164 |
|
|
/* We need to keep a pointer to the Launcher's memory map, because if
|
165 |
|
|
* the Launcher dies we need to clean it up. If we don't keep a
|
166 |
|
|
* reference, it is destroyed before close() is called. */
|
167 |
|
|
lg->mm = get_task_mm(lg->tsk);
|
168 |
|
|
|
169 |
|
|
/* Initialize the queue for the waker to wait on */
|
170 |
|
|
init_waitqueue_head(&lg->break_wq);
|
171 |
|
|
|
172 |
|
|
/* We remember which CPU's pages this Guest used last, for optimization
|
173 |
|
|
* when the same Guest runs on the same CPU twice. */
|
174 |
|
|
lg->last_pages = NULL;
|
175 |
|
|
|
176 |
|
|
/* We keep our "struct lguest" in the file's private_data. */
|
177 |
|
|
file->private_data = lg;
|
178 |
|
|
|
179 |
|
|
mutex_unlock(&lguest_lock);
|
180 |
|
|
|
181 |
|
|
/* And because this is a write() call, we return the length used. */
|
182 |
|
|
return sizeof(args);
|
183 |
|
|
|
184 |
|
|
free_regs:
|
185 |
|
|
free_page(lg->regs_page);
|
186 |
|
|
release_guest:
|
187 |
|
|
kfree(lg);
|
188 |
|
|
unlock:
|
189 |
|
|
mutex_unlock(&lguest_lock);
|
190 |
|
|
return err;
|
191 |
|
|
}
|
192 |
|
|
|
193 |
|
|
/*L:010 The first operation the Launcher does must be a write. All writes
|
194 |
|
|
* start with an unsigned long number: for the first write this must be
|
195 |
|
|
* LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
|
196 |
|
|
* writes of other values to send interrupts. */
|
197 |
|
|
static ssize_t write(struct file *file, const char __user *in,
|
198 |
|
|
size_t size, loff_t *off)
|
199 |
|
|
{
|
200 |
|
|
/* Once the guest is initialized, we hold the "struct lguest" in the
|
201 |
|
|
* file private data. */
|
202 |
|
|
struct lguest *lg = file->private_data;
|
203 |
|
|
const unsigned long __user *input = (const unsigned long __user *)in;
|
204 |
|
|
unsigned long req;
|
205 |
|
|
|
206 |
|
|
if (get_user(req, input) != 0)
|
207 |
|
|
return -EFAULT;
|
208 |
|
|
input++;
|
209 |
|
|
|
210 |
|
|
/* If you haven't initialized, you must do that first. */
|
211 |
|
|
if (req != LHREQ_INITIALIZE && !lg)
|
212 |
|
|
return -EINVAL;
|
213 |
|
|
|
214 |
|
|
/* Once the Guest is dead, all you can do is read() why it died. */
|
215 |
|
|
if (lg && lg->dead)
|
216 |
|
|
return -ENOENT;
|
217 |
|
|
|
218 |
|
|
/* If you're not the task which owns the Guest, you can only break */
|
219 |
|
|
if (lg && current != lg->tsk && req != LHREQ_BREAK)
|
220 |
|
|
return -EPERM;
|
221 |
|
|
|
222 |
|
|
switch (req) {
|
223 |
|
|
case LHREQ_INITIALIZE:
|
224 |
|
|
return initialize(file, input);
|
225 |
|
|
case LHREQ_IRQ:
|
226 |
|
|
return user_send_irq(lg, input);
|
227 |
|
|
case LHREQ_BREAK:
|
228 |
|
|
return break_guest_out(lg, input);
|
229 |
|
|
default:
|
230 |
|
|
return -EINVAL;
|
231 |
|
|
}
|
232 |
|
|
}
|
233 |
|
|
|
234 |
|
|
/*L:060 The final piece of interface code is the close() routine. It reverses
|
235 |
|
|
* everything done in initialize(). This is usually called because the
|
236 |
|
|
* Launcher exited.
|
237 |
|
|
*
|
238 |
|
|
* Note that the close routine returns 0 or a negative error number: it can't
|
239 |
|
|
* really fail, but it can whine. I blame Sun for this wart, and K&R C for
|
240 |
|
|
* letting them do it. :*/
|
241 |
|
|
static int close(struct inode *inode, struct file *file)
|
242 |
|
|
{
|
243 |
|
|
struct lguest *lg = file->private_data;
|
244 |
|
|
|
245 |
|
|
/* If we never successfully initialized, there's nothing to clean up */
|
246 |
|
|
if (!lg)
|
247 |
|
|
return 0;
|
248 |
|
|
|
249 |
|
|
/* We need the big lock, to protect from inter-guest I/O and other
|
250 |
|
|
* Launchers initializing guests. */
|
251 |
|
|
mutex_lock(&lguest_lock);
|
252 |
|
|
/* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
|
253 |
|
|
hrtimer_cancel(&lg->hrt);
|
254 |
|
|
/* Free up the shadow page tables for the Guest. */
|
255 |
|
|
free_guest_pagetable(lg);
|
256 |
|
|
/* Now all the memory cleanups are done, it's safe to release the
|
257 |
|
|
* Launcher's memory management structure. */
|
258 |
|
|
mmput(lg->mm);
|
259 |
|
|
/* If lg->dead doesn't contain an error code it will be NULL or a
|
260 |
|
|
* kmalloc()ed string, either of which is ok to hand to kfree(). */
|
261 |
|
|
if (!IS_ERR(lg->dead))
|
262 |
|
|
kfree(lg->dead);
|
263 |
|
|
/* We can free up the register page we allocated. */
|
264 |
|
|
free_page(lg->regs_page);
|
265 |
|
|
/* We clear the entire structure, which also marks it as free for the
|
266 |
|
|
* next user. */
|
267 |
|
|
memset(lg, 0, sizeof(*lg));
|
268 |
|
|
/* Release lock and exit. */
|
269 |
|
|
mutex_unlock(&lguest_lock);
|
270 |
|
|
|
271 |
|
|
return 0;
|
272 |
|
|
}
|
273 |
|
|
|
274 |
|
|
/*L:000
|
275 |
|
|
* Welcome to our journey through the Launcher!
|
276 |
|
|
*
|
277 |
|
|
* The Launcher is the Host userspace program which sets up, runs and services
|
278 |
|
|
* the Guest. In fact, many comments in the Drivers which refer to "the Host"
|
279 |
|
|
* doing things are inaccurate: the Launcher does all the device handling for
|
280 |
|
|
* the Guest, but the Guest can't know that.
|
281 |
|
|
*
|
282 |
|
|
* Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
|
283 |
|
|
* shall see more of that later.
|
284 |
|
|
*
|
285 |
|
|
* We begin our understanding with the Host kernel interface which the Launcher
|
286 |
|
|
* uses: reading and writing a character device called /dev/lguest. All the
|
287 |
|
|
* work happens in the read(), write() and close() routines: */
|
288 |
|
|
static struct file_operations lguest_fops = {
|
289 |
|
|
.owner = THIS_MODULE,
|
290 |
|
|
.release = close,
|
291 |
|
|
.write = write,
|
292 |
|
|
.read = read,
|
293 |
|
|
};
|
294 |
|
|
|
295 |
|
|
/* This is a textbook example of a "misc" character device. Populate a "struct
|
296 |
|
|
* miscdevice" and register it with misc_register(). */
|
297 |
|
|
static struct miscdevice lguest_dev = {
|
298 |
|
|
.minor = MISC_DYNAMIC_MINOR,
|
299 |
|
|
.name = "lguest",
|
300 |
|
|
.fops = &lguest_fops,
|
301 |
|
|
};
|
302 |
|
|
|
303 |
|
|
int __init lguest_device_init(void)
|
304 |
|
|
{
|
305 |
|
|
return misc_register(&lguest_dev);
|
306 |
|
|
}
|
307 |
|
|
|
308 |
|
|
void __exit lguest_device_remove(void)
|
309 |
|
|
{
|
310 |
|
|
misc_deregister(&lguest_dev);
|
311 |
|
|
}
|