Commit | Line | Data |
---|---|---|
f938d2c8 RR |
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 | |
3c6b5bfa RR |
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 | |
15045275 | 5 | * or the Guest doing a NOTIFY out to the Launcher. :*/ |
d7e28ffe RR |
6 | #include <linux/uaccess.h> |
7 | #include <linux/miscdevice.h> | |
8 | #include <linux/fs.h> | |
9 | #include "lg.h" | |
10 | ||
e1e72965 RR |
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. */ | |
511801dc | 16 | static int break_guest_out(struct lguest *lg, const unsigned long __user *input) |
d7e28ffe RR |
17 | { |
18 | unsigned long on; | |
19 | ||
e1e72965 | 20 | /* Fetch whether they're turning break on or off. */ |
d7e28ffe RR |
21 | if (get_user(on, input) != 0) |
22 | return -EFAULT; | |
23 | ||
24 | if (on) { | |
25 | lg->break_out = 1; | |
e1e72965 | 26 | /* Pop it out of the Guest (may be running on different CPU) */ |
d7e28ffe RR |
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 | ||
dde79789 RR |
37 | /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt |
38 | * number to /dev/lguest. */ | |
511801dc | 39 | static int user_send_irq(struct lguest *lg, const unsigned long __user *input) |
d7e28ffe | 40 | { |
511801dc | 41 | unsigned long irq; |
d7e28ffe RR |
42 | |
43 | if (get_user(irq, input) != 0) | |
44 | return -EFAULT; | |
45 | if (irq >= LGUEST_IRQS) | |
46 | return -EINVAL; | |
dde79789 RR |
47 | /* Next time the Guest runs, the core code will see if it can deliver |
48 | * this interrupt. */ | |
d7e28ffe RR |
49 | set_bit(irq, lg->irqs_pending); |
50 | return 0; | |
51 | } | |
52 | ||
dde79789 RR |
53 | /*L:040 Once our Guest is initialized, the Launcher makes it run by reading |
54 | * from /dev/lguest. */ | |
d7e28ffe RR |
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; | |
d0953d42 GOC |
58 | struct lg_cpu *cpu; |
59 | unsigned int cpu_id = *o; | |
d7e28ffe | 60 | |
dde79789 | 61 | /* You must write LHREQ_INITIALIZE first! */ |
d7e28ffe RR |
62 | if (!lg) |
63 | return -EINVAL; | |
64 | ||
d0953d42 GOC |
65 | /* Watch out for arbitrary vcpu indexes! */ |
66 | if (cpu_id >= lg->nr_cpus) | |
67 | return -EINVAL; | |
68 | ||
69 | cpu = &lg->cpus[cpu_id]; | |
70 | ||
e1e72965 | 71 | /* If you're not the task which owns the Guest, go away. */ |
d7e28ffe RR |
72 | if (current != lg->tsk) |
73 | return -EPERM; | |
74 | ||
dde79789 | 75 | /* If the guest is already dead, we indicate why */ |
d7e28ffe RR |
76 | if (lg->dead) { |
77 | size_t len; | |
78 | ||
dde79789 | 79 | /* lg->dead either contains an error code, or a string. */ |
d7e28ffe RR |
80 | if (IS_ERR(lg->dead)) |
81 | return PTR_ERR(lg->dead); | |
82 | ||
dde79789 | 83 | /* We can only return as much as the buffer they read with. */ |
d7e28ffe RR |
84 | len = min(size, strlen(lg->dead)+1); |
85 | if (copy_to_user(user, lg->dead, len) != 0) | |
86 | return -EFAULT; | |
87 | return len; | |
88 | } | |
89 | ||
15045275 | 90 | /* If we returned from read() last time because the Guest notified, |
dde79789 | 91 | * clear the flag. */ |
15045275 RR |
92 | if (lg->pending_notify) |
93 | lg->pending_notify = 0; | |
d7e28ffe | 94 | |
dde79789 | 95 | /* Run the Guest until something interesting happens. */ |
d0953d42 | 96 | return run_guest(cpu, (unsigned long __user *)user); |
d7e28ffe RR |
97 | } |
98 | ||
4dcc53da GOC |
99 | static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) |
100 | { | |
101 | if (id >= NR_CPUS) | |
102 | return -EINVAL; | |
103 | ||
104 | cpu->id = id; | |
105 | cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); | |
106 | cpu->lg->nr_cpus++; | |
107 | ||
108 | return 0; | |
109 | } | |
110 | ||
47436aa4 | 111 | /*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit) |
511801dc | 112 | * values (in addition to the LHREQ_INITIALIZE value). These are: |
dde79789 | 113 | * |
3c6b5bfa RR |
114 | * base: The start of the Guest-physical memory inside the Launcher memory. |
115 | * | |
dde79789 | 116 | * pfnlimit: The highest (Guest-physical) page number the Guest should be |
e1e72965 RR |
117 | * allowed to access. The Guest memory lives inside the Launcher, so it sets |
118 | * this to ensure the Guest can only reach its own memory. | |
dde79789 RR |
119 | * |
120 | * pgdir: The (Guest-physical) address of the top of the initial Guest | |
121 | * pagetables (which are set up by the Launcher). | |
122 | * | |
123 | * start: The first instruction to execute ("eip" in x86-speak). | |
dde79789 | 124 | */ |
511801dc | 125 | static int initialize(struct file *file, const unsigned long __user *input) |
d7e28ffe | 126 | { |
dde79789 RR |
127 | /* "struct lguest" contains everything we (the Host) know about a |
128 | * Guest. */ | |
d7e28ffe | 129 | struct lguest *lg; |
48245cc0 | 130 | int err; |
47436aa4 | 131 | unsigned long args[4]; |
d7e28ffe | 132 | |
48245cc0 RR |
133 | /* We grab the Big Lguest lock, which protects against multiple |
134 | * simultaneous initializations. */ | |
d7e28ffe | 135 | mutex_lock(&lguest_lock); |
dde79789 | 136 | /* You can't initialize twice! Close the device and start again... */ |
d7e28ffe RR |
137 | if (file->private_data) { |
138 | err = -EBUSY; | |
139 | goto unlock; | |
140 | } | |
141 | ||
142 | if (copy_from_user(args, input, sizeof(args)) != 0) { | |
143 | err = -EFAULT; | |
144 | goto unlock; | |
145 | } | |
146 | ||
48245cc0 RR |
147 | lg = kzalloc(sizeof(*lg), GFP_KERNEL); |
148 | if (!lg) { | |
149 | err = -ENOMEM; | |
d7e28ffe RR |
150 | goto unlock; |
151 | } | |
dde79789 RR |
152 | |
153 | /* Populate the easy fields of our "struct lguest" */ | |
3c6b5bfa RR |
154 | lg->mem_base = (void __user *)(long)args[0]; |
155 | lg->pfn_limit = args[1]; | |
dde79789 | 156 | |
4dcc53da | 157 | /* This is the first cpu */ |
d0953d42 | 158 | err = lg_cpu_start(&lg->cpus[0], 0, args[3]); |
4dcc53da GOC |
159 | if (err) |
160 | goto release_guest; | |
161 | ||
dde79789 RR |
162 | /* We need a complete page for the Guest registers: they are accessible |
163 | * to the Guest and we can only grant it access to whole pages. */ | |
d7e28ffe RR |
164 | lg->regs_page = get_zeroed_page(GFP_KERNEL); |
165 | if (!lg->regs_page) { | |
166 | err = -ENOMEM; | |
167 | goto release_guest; | |
168 | } | |
dde79789 | 169 | /* We actually put the registers at the bottom of the page. */ |
d7e28ffe RR |
170 | lg->regs = (void *)lg->regs_page + PAGE_SIZE - sizeof(*lg->regs); |
171 | ||
dde79789 RR |
172 | /* Initialize the Guest's shadow page tables, using the toplevel |
173 | * address the Launcher gave us. This allocates memory, so can | |
174 | * fail. */ | |
3c6b5bfa | 175 | err = init_guest_pagetable(lg, args[2]); |
d7e28ffe RR |
176 | if (err) |
177 | goto free_regs; | |
178 | ||
dde79789 RR |
179 | /* Now we initialize the Guest's registers, handing it the start |
180 | * address. */ | |
d612cde0 | 181 | lguest_arch_setup_regs(lg, args[3]); |
dde79789 RR |
182 | |
183 | /* The timer for lguest's clock needs initialization. */ | |
d7e28ffe | 184 | init_clockdev(lg); |
dde79789 RR |
185 | |
186 | /* We keep a pointer to the Launcher task (ie. current task) for when | |
187 | * other Guests want to wake this one (inter-Guest I/O). */ | |
d7e28ffe | 188 | lg->tsk = current; |
dde79789 RR |
189 | /* We need to keep a pointer to the Launcher's memory map, because if |
190 | * the Launcher dies we need to clean it up. If we don't keep a | |
191 | * reference, it is destroyed before close() is called. */ | |
d7e28ffe | 192 | lg->mm = get_task_mm(lg->tsk); |
dde79789 RR |
193 | |
194 | /* Initialize the queue for the waker to wait on */ | |
d7e28ffe | 195 | init_waitqueue_head(&lg->break_wq); |
dde79789 RR |
196 | |
197 | /* We remember which CPU's pages this Guest used last, for optimization | |
198 | * when the same Guest runs on the same CPU twice. */ | |
d7e28ffe | 199 | lg->last_pages = NULL; |
dde79789 RR |
200 | |
201 | /* We keep our "struct lguest" in the file's private_data. */ | |
d7e28ffe RR |
202 | file->private_data = lg; |
203 | ||
204 | mutex_unlock(&lguest_lock); | |
205 | ||
dde79789 | 206 | /* And because this is a write() call, we return the length used. */ |
d7e28ffe RR |
207 | return sizeof(args); |
208 | ||
209 | free_regs: | |
210 | free_page(lg->regs_page); | |
211 | release_guest: | |
43054412 | 212 | kfree(lg); |
d7e28ffe RR |
213 | unlock: |
214 | mutex_unlock(&lguest_lock); | |
215 | return err; | |
216 | } | |
217 | ||
dde79789 | 218 | /*L:010 The first operation the Launcher does must be a write. All writes |
e1e72965 | 219 | * start with an unsigned long number: for the first write this must be |
dde79789 | 220 | * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use |
15045275 | 221 | * writes of other values to send interrupts. */ |
511801dc | 222 | static ssize_t write(struct file *file, const char __user *in, |
d7e28ffe RR |
223 | size_t size, loff_t *off) |
224 | { | |
dde79789 RR |
225 | /* Once the guest is initialized, we hold the "struct lguest" in the |
226 | * file private data. */ | |
d7e28ffe | 227 | struct lguest *lg = file->private_data; |
511801dc JS |
228 | const unsigned long __user *input = (const unsigned long __user *)in; |
229 | unsigned long req; | |
7ea07a15 GOC |
230 | struct lg_cpu *cpu; |
231 | unsigned int cpu_id = *off; | |
d7e28ffe RR |
232 | |
233 | if (get_user(req, input) != 0) | |
234 | return -EFAULT; | |
511801dc | 235 | input++; |
d7e28ffe | 236 | |
dde79789 | 237 | /* If you haven't initialized, you must do that first. */ |
7ea07a15 GOC |
238 | if (req != LHREQ_INITIALIZE) { |
239 | if (!lg || (cpu_id >= lg->nr_cpus)) | |
240 | return -EINVAL; | |
241 | cpu = &lg->cpus[cpu_id]; | |
242 | if (!cpu) | |
243 | return -EINVAL; | |
244 | } | |
dde79789 RR |
245 | |
246 | /* Once the Guest is dead, all you can do is read() why it died. */ | |
d7e28ffe RR |
247 | if (lg && lg->dead) |
248 | return -ENOENT; | |
249 | ||
250 | /* If you're not the task which owns the Guest, you can only break */ | |
251 | if (lg && current != lg->tsk && req != LHREQ_BREAK) | |
252 | return -EPERM; | |
253 | ||
254 | switch (req) { | |
255 | case LHREQ_INITIALIZE: | |
511801dc | 256 | return initialize(file, input); |
d7e28ffe | 257 | case LHREQ_IRQ: |
511801dc | 258 | return user_send_irq(lg, input); |
d7e28ffe | 259 | case LHREQ_BREAK: |
511801dc | 260 | return break_guest_out(lg, input); |
d7e28ffe RR |
261 | default: |
262 | return -EINVAL; | |
263 | } | |
264 | } | |
265 | ||
dde79789 RR |
266 | /*L:060 The final piece of interface code is the close() routine. It reverses |
267 | * everything done in initialize(). This is usually called because the | |
268 | * Launcher exited. | |
269 | * | |
270 | * Note that the close routine returns 0 or a negative error number: it can't | |
271 | * really fail, but it can whine. I blame Sun for this wart, and K&R C for | |
272 | * letting them do it. :*/ | |
d7e28ffe RR |
273 | static int close(struct inode *inode, struct file *file) |
274 | { | |
275 | struct lguest *lg = file->private_data; | |
276 | ||
dde79789 | 277 | /* If we never successfully initialized, there's nothing to clean up */ |
d7e28ffe RR |
278 | if (!lg) |
279 | return 0; | |
280 | ||
dde79789 RR |
281 | /* We need the big lock, to protect from inter-guest I/O and other |
282 | * Launchers initializing guests. */ | |
d7e28ffe RR |
283 | mutex_lock(&lguest_lock); |
284 | /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ | |
285 | hrtimer_cancel(&lg->hrt); | |
dde79789 | 286 | /* Free up the shadow page tables for the Guest. */ |
d7e28ffe | 287 | free_guest_pagetable(lg); |
dde79789 RR |
288 | /* Now all the memory cleanups are done, it's safe to release the |
289 | * Launcher's memory management structure. */ | |
d7e28ffe | 290 | mmput(lg->mm); |
dde79789 RR |
291 | /* If lg->dead doesn't contain an error code it will be NULL or a |
292 | * kmalloc()ed string, either of which is ok to hand to kfree(). */ | |
d7e28ffe RR |
293 | if (!IS_ERR(lg->dead)) |
294 | kfree(lg->dead); | |
dde79789 | 295 | /* We can free up the register page we allocated. */ |
d7e28ffe | 296 | free_page(lg->regs_page); |
dde79789 RR |
297 | /* We clear the entire structure, which also marks it as free for the |
298 | * next user. */ | |
d7e28ffe | 299 | memset(lg, 0, sizeof(*lg)); |
dde79789 | 300 | /* Release lock and exit. */ |
d7e28ffe | 301 | mutex_unlock(&lguest_lock); |
dde79789 | 302 | |
d7e28ffe RR |
303 | return 0; |
304 | } | |
305 | ||
dde79789 RR |
306 | /*L:000 |
307 | * Welcome to our journey through the Launcher! | |
308 | * | |
309 | * The Launcher is the Host userspace program which sets up, runs and services | |
310 | * the Guest. In fact, many comments in the Drivers which refer to "the Host" | |
311 | * doing things are inaccurate: the Launcher does all the device handling for | |
e1e72965 | 312 | * the Guest, but the Guest can't know that. |
dde79789 RR |
313 | * |
314 | * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we | |
315 | * shall see more of that later. | |
316 | * | |
317 | * We begin our understanding with the Host kernel interface which the Launcher | |
318 | * uses: reading and writing a character device called /dev/lguest. All the | |
319 | * work happens in the read(), write() and close() routines: */ | |
d7e28ffe RR |
320 | static struct file_operations lguest_fops = { |
321 | .owner = THIS_MODULE, | |
322 | .release = close, | |
323 | .write = write, | |
324 | .read = read, | |
325 | }; | |
dde79789 RR |
326 | |
327 | /* This is a textbook example of a "misc" character device. Populate a "struct | |
328 | * miscdevice" and register it with misc_register(). */ | |
d7e28ffe RR |
329 | static struct miscdevice lguest_dev = { |
330 | .minor = MISC_DYNAMIC_MINOR, | |
331 | .name = "lguest", | |
332 | .fops = &lguest_fops, | |
333 | }; | |
334 | ||
335 | int __init lguest_device_init(void) | |
336 | { | |
337 | return misc_register(&lguest_dev); | |
338 | } | |
339 | ||
340 | void __exit lguest_device_remove(void) | |
341 | { | |
342 | misc_deregister(&lguest_dev); | |
343 | } |