blob: 85d42d3d01a9a9bbeb739ce422271848cd4be3ff [file] [log] [blame]
Rusty Russellf938d2c2007-07-26 10:41:02 -07001/*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
Rusty Russell3c6b5bf2007-10-22 11:03:26 +10003 * 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
Rusty Russell15045272007-10-22 11:24:10 +10005 * or the Guest doing a NOTIFY out to the Launcher. :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -07006#include <linux/uaccess.h>
7#include <linux/miscdevice.h>
8#include <linux/fs.h>
Glauber de Oliveira Costaca94f2b2008-01-18 23:59:07 -02009#include <linux/sched.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070010#include "lg.h"
11
Rusty Russelle1e72962007-10-25 15:02:50 +100012/*L:055 When something happens, the Waker process needs a way to stop the
13 * kernel running the Guest and return to the Launcher. So the Waker writes
14 * LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher
15 * has done whatever needs attention, it writes LHREQ_BREAK and "0" to release
16 * the Waker. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020017static int break_guest_out(struct lg_cpu *cpu, const unsigned long __user*input)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070018{
19 unsigned long on;
20
Rusty Russelle1e72962007-10-25 15:02:50 +100021 /* Fetch whether they're turning break on or off. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070022 if (get_user(on, input) != 0)
23 return -EFAULT;
24
25 if (on) {
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020026 cpu->break_out = 1;
Rusty Russelle1e72962007-10-25 15:02:50 +100027 /* Pop it out of the Guest (may be running on different CPU) */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020028 wake_up_process(cpu->tsk);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070029 /* Wait for them to reset it */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020030 return wait_event_interruptible(cpu->break_wq, !cpu->break_out);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070031 } else {
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020032 cpu->break_out = 0;
33 wake_up(&cpu->break_wq);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070034 return 0;
35 }
36}
37
Rusty Russelldde79782007-07-26 10:41:03 -070038/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
39 * number to /dev/lguest. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020040static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070041{
Jes Sorensen511801d2007-10-22 11:03:31 +100042 unsigned long irq;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070043
44 if (get_user(irq, input) != 0)
45 return -EFAULT;
46 if (irq >= LGUEST_IRQS)
47 return -EINVAL;
Rusty Russelldde79782007-07-26 10:41:03 -070048 /* Next time the Guest runs, the core code will see if it can deliver
49 * this interrupt. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020050 set_bit(irq, cpu->irqs_pending);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070051 return 0;
52}
53
Rusty Russelldde79782007-07-26 10:41:03 -070054/*L:040 Once our Guest is initialized, the Launcher makes it run by reading
55 * from /dev/lguest. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070056static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
57{
58 struct lguest *lg = file->private_data;
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -020059 struct lg_cpu *cpu;
60 unsigned int cpu_id = *o;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070061
Rusty Russelldde79782007-07-26 10:41:03 -070062 /* You must write LHREQ_INITIALIZE first! */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070063 if (!lg)
64 return -EINVAL;
65
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -020066 /* Watch out for arbitrary vcpu indexes! */
67 if (cpu_id >= lg->nr_cpus)
68 return -EINVAL;
69
70 cpu = &lg->cpus[cpu_id];
71
Rusty Russelle1e72962007-10-25 15:02:50 +100072 /* If you're not the task which owns the Guest, go away. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020073 if (current != cpu->tsk)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070074 return -EPERM;
75
Rusty Russelldde79782007-07-26 10:41:03 -070076 /* If the guest is already dead, we indicate why */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070077 if (lg->dead) {
78 size_t len;
79
Rusty Russelldde79782007-07-26 10:41:03 -070080 /* lg->dead either contains an error code, or a string. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070081 if (IS_ERR(lg->dead))
82 return PTR_ERR(lg->dead);
83
Rusty Russelldde79782007-07-26 10:41:03 -070084 /* We can only return as much as the buffer they read with. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070085 len = min(size, strlen(lg->dead)+1);
86 if (copy_to_user(user, lg->dead, len) != 0)
87 return -EFAULT;
88 return len;
89 }
90
Rusty Russell15045272007-10-22 11:24:10 +100091 /* If we returned from read() last time because the Guest notified,
Rusty Russelldde79782007-07-26 10:41:03 -070092 * clear the flag. */
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -020093 if (cpu->pending_notify)
94 cpu->pending_notify = 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070095
Rusty Russelldde79782007-07-26 10:41:03 -070096 /* Run the Guest until something interesting happens. */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -020097 return run_guest(cpu, (unsigned long __user *)user);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070098}
99
Glauber de Oliveira Costa4dcc53d2008-01-07 11:05:24 -0200100static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
101{
102 if (id >= NR_CPUS)
103 return -EINVAL;
104
105 cpu->id = id;
106 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
107 cpu->lg->nr_cpus++;
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200108 init_clockdev(cpu);
Glauber de Oliveira Costa4dcc53d2008-01-07 11:05:24 -0200109
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200110 /* We need a complete page for the Guest registers: they are accessible
111 * to the Guest and we can only grant it access to whole pages. */
112 cpu->regs_page = get_zeroed_page(GFP_KERNEL);
113 if (!cpu->regs_page)
114 return -ENOMEM;
115
116 /* We actually put the registers at the bottom of the page. */
117 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
118
119 /* Now we initialize the Guest's registers, handing it the start
120 * address. */
121 lguest_arch_setup_regs(cpu, start_ip);
122
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200123 /* Initialize the queue for the waker to wait on */
124 init_waitqueue_head(&cpu->break_wq);
125
126 /* We keep a pointer to the Launcher task (ie. current task) for when
127 * other Guests want to wake this one (inter-Guest I/O). */
128 cpu->tsk = current;
129
130 /* We need to keep a pointer to the Launcher's memory map, because if
131 * the Launcher dies we need to clean it up. If we don't keep a
132 * reference, it is destroyed before close() is called. */
133 cpu->mm = get_task_mm(cpu->tsk);
134
Glauber de Oliveira Costaf34f8c52008-01-17 19:13:26 -0200135 /* We remember which CPU's pages this Guest used last, for optimization
136 * when the same Guest runs on the same CPU twice. */
137 cpu->last_pages = NULL;
138
Glauber de Oliveira Costa4dcc53d2008-01-07 11:05:24 -0200139 return 0;
140}
141
Rusty Russell47436aa2007-10-22 11:03:36 +1000142/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit)
Jes Sorensen511801d2007-10-22 11:03:31 +1000143 * values (in addition to the LHREQ_INITIALIZE value). These are:
Rusty Russelldde79782007-07-26 10:41:03 -0700144 *
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000145 * base: The start of the Guest-physical memory inside the Launcher memory.
146 *
Rusty Russelldde79782007-07-26 10:41:03 -0700147 * pfnlimit: The highest (Guest-physical) page number the Guest should be
Rusty Russelle1e72962007-10-25 15:02:50 +1000148 * allowed to access. The Guest memory lives inside the Launcher, so it sets
149 * this to ensure the Guest can only reach its own memory.
Rusty Russelldde79782007-07-26 10:41:03 -0700150 *
151 * pgdir: The (Guest-physical) address of the top of the initial Guest
152 * pagetables (which are set up by the Launcher).
153 *
154 * start: The first instruction to execute ("eip" in x86-speak).
Rusty Russelldde79782007-07-26 10:41:03 -0700155 */
Jes Sorensen511801d2007-10-22 11:03:31 +1000156static int initialize(struct file *file, const unsigned long __user *input)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700157{
Rusty Russelldde79782007-07-26 10:41:03 -0700158 /* "struct lguest" contains everything we (the Host) know about a
159 * Guest. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700160 struct lguest *lg;
Rusty Russell48245cc2007-10-22 11:03:27 +1000161 int err;
Rusty Russell47436aa2007-10-22 11:03:36 +1000162 unsigned long args[4];
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700163
Rusty Russell48245cc2007-10-22 11:03:27 +1000164 /* We grab the Big Lguest lock, which protects against multiple
165 * simultaneous initializations. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700166 mutex_lock(&lguest_lock);
Rusty Russelldde79782007-07-26 10:41:03 -0700167 /* You can't initialize twice! Close the device and start again... */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700168 if (file->private_data) {
169 err = -EBUSY;
170 goto unlock;
171 }
172
173 if (copy_from_user(args, input, sizeof(args)) != 0) {
174 err = -EFAULT;
175 goto unlock;
176 }
177
Rusty Russell48245cc2007-10-22 11:03:27 +1000178 lg = kzalloc(sizeof(*lg), GFP_KERNEL);
179 if (!lg) {
180 err = -ENOMEM;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700181 goto unlock;
182 }
Rusty Russelldde79782007-07-26 10:41:03 -0700183
184 /* Populate the easy fields of our "struct lguest" */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000185 lg->mem_base = (void __user *)(long)args[0];
186 lg->pfn_limit = args[1];
Rusty Russelldde79782007-07-26 10:41:03 -0700187
Glauber de Oliveira Costa4dcc53d2008-01-07 11:05:24 -0200188 /* This is the first cpu */
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200189 err = lg_cpu_start(&lg->cpus[0], 0, args[3]);
Glauber de Oliveira Costa4dcc53d2008-01-07 11:05:24 -0200190 if (err)
191 goto release_guest;
192
Rusty Russelldde79782007-07-26 10:41:03 -0700193 /* Initialize the Guest's shadow page tables, using the toplevel
194 * address the Launcher gave us. This allocates memory, so can
195 * fail. */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000196 err = init_guest_pagetable(lg, args[2]);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700197 if (err)
198 goto free_regs;
199
Rusty Russelldde79782007-07-26 10:41:03 -0700200 /* We keep our "struct lguest" in the file's private_data. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700201 file->private_data = lg;
202
203 mutex_unlock(&lguest_lock);
204
Rusty Russelldde79782007-07-26 10:41:03 -0700205 /* And because this is a write() call, we return the length used. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700206 return sizeof(args);
207
208free_regs:
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200209 /* FIXME: This should be in free_vcpu */
210 free_page(lg->cpus[0].regs_page);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700211release_guest:
Adrian Bunk43054412007-11-14 16:59:00 -0800212 kfree(lg);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700213unlock:
214 mutex_unlock(&lguest_lock);
215 return err;
216}
217
Rusty Russelldde79782007-07-26 10:41:03 -0700218/*L:010 The first operation the Launcher does must be a write. All writes
Rusty Russelle1e72962007-10-25 15:02:50 +1000219 * start with an unsigned long number: for the first write this must be
Rusty Russelldde79782007-07-26 10:41:03 -0700220 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
Rusty Russell15045272007-10-22 11:24:10 +1000221 * writes of other values to send interrupts. */
Jes Sorensen511801d2007-10-22 11:03:31 +1000222static ssize_t write(struct file *file, const char __user *in,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700223 size_t size, loff_t *off)
224{
Rusty Russelldde79782007-07-26 10:41:03 -0700225 /* Once the guest is initialized, we hold the "struct lguest" in the
226 * file private data. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700227 struct lguest *lg = file->private_data;
Jes Sorensen511801d2007-10-22 11:03:31 +1000228 const unsigned long __user *input = (const unsigned long __user *)in;
229 unsigned long req;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200230 struct lg_cpu *uninitialized_var(cpu);
Glauber de Oliveira Costa7ea07a12008-01-07 11:05:26 -0200231 unsigned int cpu_id = *off;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700232
233 if (get_user(req, input) != 0)
234 return -EFAULT;
Jes Sorensen511801d2007-10-22 11:03:31 +1000235 input++;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700236
Rusty Russelldde79782007-07-26 10:41:03 -0700237 /* If you haven't initialized, you must do that first. */
Glauber de Oliveira Costa7ea07a12008-01-07 11:05:26 -0200238 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 }
Rusty Russelldde79782007-07-26 10:41:03 -0700245
246 /* Once the Guest is dead, all you can do is read() why it died. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700247 if (lg && lg->dead)
248 return -ENOENT;
249
250 /* If you're not the task which owns the Guest, you can only break */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200251 if (lg && current != cpu->tsk && req != LHREQ_BREAK)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700252 return -EPERM;
253
254 switch (req) {
255 case LHREQ_INITIALIZE:
Jes Sorensen511801d2007-10-22 11:03:31 +1000256 return initialize(file, input);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700257 case LHREQ_IRQ:
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200258 return user_send_irq(cpu, input);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700259 case LHREQ_BREAK:
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200260 return break_guest_out(cpu, input);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700261 default:
262 return -EINVAL;
263 }
264}
265
Rusty Russelldde79782007-07-26 10:41:03 -0700266/*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. :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700273static int close(struct inode *inode, struct file *file)
274{
275 struct lguest *lg = file->private_data;
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200276 unsigned int i;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700277
Rusty Russelldde79782007-07-26 10:41:03 -0700278 /* If we never successfully initialized, there's nothing to clean up */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700279 if (!lg)
280 return 0;
281
Rusty Russelldde79782007-07-26 10:41:03 -0700282 /* We need the big lock, to protect from inter-guest I/O and other
283 * Launchers initializing guests. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700284 mutex_lock(&lguest_lock);
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200285
286 /* Free up the shadow page tables for the Guest. */
287 free_guest_pagetable(lg);
288
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200289 for (i = 0; i < lg->nr_cpus; i++) {
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200290 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
291 hrtimer_cancel(&lg->cpus[i].hrt);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200292 /* We can free up the register page we allocated. */
293 free_page(lg->cpus[i].regs_page);
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200294 /* Now all the memory cleanups are done, it's safe to release
295 * the Launcher's memory management structure. */
296 mmput(lg->cpus[i].mm);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200297 }
Rusty Russelldde79782007-07-26 10:41:03 -0700298 /* If lg->dead doesn't contain an error code it will be NULL or a
299 * kmalloc()ed string, either of which is ok to hand to kfree(). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700300 if (!IS_ERR(lg->dead))
301 kfree(lg->dead);
Rusty Russelldde79782007-07-26 10:41:03 -0700302 /* We clear the entire structure, which also marks it as free for the
303 * next user. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700304 memset(lg, 0, sizeof(*lg));
Rusty Russelldde79782007-07-26 10:41:03 -0700305 /* Release lock and exit. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700306 mutex_unlock(&lguest_lock);
Rusty Russelldde79782007-07-26 10:41:03 -0700307
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700308 return 0;
309}
310
Rusty Russelldde79782007-07-26 10:41:03 -0700311/*L:000
312 * Welcome to our journey through the Launcher!
313 *
314 * The Launcher is the Host userspace program which sets up, runs and services
315 * the Guest. In fact, many comments in the Drivers which refer to "the Host"
316 * doing things are inaccurate: the Launcher does all the device handling for
Rusty Russelle1e72962007-10-25 15:02:50 +1000317 * the Guest, but the Guest can't know that.
Rusty Russelldde79782007-07-26 10:41:03 -0700318 *
319 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
320 * shall see more of that later.
321 *
322 * We begin our understanding with the Host kernel interface which the Launcher
323 * uses: reading and writing a character device called /dev/lguest. All the
324 * work happens in the read(), write() and close() routines: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700325static struct file_operations lguest_fops = {
326 .owner = THIS_MODULE,
327 .release = close,
328 .write = write,
329 .read = read,
330};
Rusty Russelldde79782007-07-26 10:41:03 -0700331
332/* This is a textbook example of a "misc" character device. Populate a "struct
333 * miscdevice" and register it with misc_register(). */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700334static struct miscdevice lguest_dev = {
335 .minor = MISC_DYNAMIC_MINOR,
336 .name = "lguest",
337 .fops = &lguest_fops,
338};
339
340int __init lguest_device_init(void)
341{
342 return misc_register(&lguest_dev);
343}
344
345void __exit lguest_device_remove(void)
346{
347 misc_deregister(&lguest_dev);
348}