lguest_user.c revision dde797899ac17ebb812b7566044124d785e98dc7
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 memory size, pagetable, entry point and kernel address offset. 4 * A read will run the Guest until a signal is pending (-EINTR), or the Guest 5 * does a DMA out to the Launcher. Writes are also used to get a DMA buffer 6 * 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 4 32-bit values (in addition to the 146 * 32-bit LHREQ_INITIALIZE value). These are: 147 * 148 * pfnlimit: The highest (Guest-physical) page number the Guest should be 149 * allowed to access. The Launcher has to live in Guest memory, so it sets 150 * this to ensure the Guest can't reach it. 151 * 152 * pgdir: The (Guest-physical) address of the top of the initial Guest 153 * pagetables (which are set up by the Launcher). 154 * 155 * start: The first instruction to execute ("eip" in x86-speak). 156 * 157 * page_offset: The PAGE_OFFSET constant in the Guest kernel. We should 158 * probably wean the code off this, but it's a very useful constant! Any 159 * address above this is within the Guest kernel, and any kernel address can 160 * quickly converted from physical to virtual by adding PAGE_OFFSET. It's 161 * 0xC0000000 (3G) by default, but it's configurable at kernel build time. 162 */ 163static int initialize(struct file *file, const u32 __user *input) 164{ 165 /* "struct lguest" contains everything we (the Host) know about a 166 * Guest. */ 167 struct lguest *lg; 168 int err, i; 169 u32 args[4]; 170 171 /* We grab the Big Lguest lock, which protects the global array 172 * "lguests" and multiple simultaneous initializations. */ 173 mutex_lock(&lguest_lock); 174 /* You can't initialize twice! Close the device and start again... */ 175 if (file->private_data) { 176 err = -EBUSY; 177 goto unlock; 178 } 179 180 if (copy_from_user(args, input, sizeof(args)) != 0) { 181 err = -EFAULT; 182 goto unlock; 183 } 184 185 /* Find an unused guest. */ 186 i = find_free_guest(); 187 if (i < 0) { 188 err = -ENOSPC; 189 goto unlock; 190 } 191 /* OK, we have an index into the "lguest" array: "lg" is a convenient 192 * pointer. */ 193 lg = &lguests[i]; 194 195 /* Populate the easy fields of our "struct lguest" */ 196 lg->guestid = i; 197 lg->pfn_limit = args[0]; 198 lg->page_offset = args[3]; 199 200 /* We need a complete page for the Guest registers: they are accessible 201 * to the Guest and we can only grant it access to whole pages. */ 202 lg->regs_page = get_zeroed_page(GFP_KERNEL); 203 if (!lg->regs_page) { 204 err = -ENOMEM; 205 goto release_guest; 206 } 207 /* We actually put the registers at the bottom of the page. */ 208 lg->regs = (void *)lg->regs_page + PAGE_SIZE - sizeof(*lg->regs); 209 210 /* Initialize the Guest's shadow page tables, using the toplevel 211 * address the Launcher gave us. This allocates memory, so can 212 * fail. */ 213 err = init_guest_pagetable(lg, args[1]); 214 if (err) 215 goto free_regs; 216 217 /* Now we initialize the Guest's registers, handing it the start 218 * address. */ 219 setup_regs(lg->regs, args[2]); 220 221 /* There are a couple of GDT entries the Guest expects when first 222 * booting. */ 223 setup_guest_gdt(lg); 224 225 /* The timer for lguest's clock needs initialization. */ 226 init_clockdev(lg); 227 228 /* We keep a pointer to the Launcher task (ie. current task) for when 229 * other Guests want to wake this one (inter-Guest I/O). */ 230 lg->tsk = current; 231 /* We need to keep a pointer to the Launcher's memory map, because if 232 * the Launcher dies we need to clean it up. If we don't keep a 233 * reference, it is destroyed before close() is called. */ 234 lg->mm = get_task_mm(lg->tsk); 235 236 /* Initialize the queue for the waker to wait on */ 237 init_waitqueue_head(&lg->break_wq); 238 239 /* We remember which CPU's pages this Guest used last, for optimization 240 * when the same Guest runs on the same CPU twice. */ 241 lg->last_pages = NULL; 242 243 /* We keep our "struct lguest" in the file's private_data. */ 244 file->private_data = lg; 245 246 mutex_unlock(&lguest_lock); 247 248 /* And because this is a write() call, we return the length used. */ 249 return sizeof(args); 250 251free_regs: 252 free_page(lg->regs_page); 253release_guest: 254 memset(lg, 0, sizeof(*lg)); 255unlock: 256 mutex_unlock(&lguest_lock); 257 return err; 258} 259 260/*L:010 The first operation the Launcher does must be a write. All writes 261 * start with a 32 bit number: for the first write this must be 262 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 263 * writes of other values to get DMA buffers and send interrupts. */ 264static ssize_t write(struct file *file, const char __user *input, 265 size_t size, loff_t *off) 266{ 267 /* Once the guest is initialized, we hold the "struct lguest" in the 268 * file private data. */ 269 struct lguest *lg = file->private_data; 270 u32 req; 271 272 if (get_user(req, input) != 0) 273 return -EFAULT; 274 input += sizeof(req); 275 276 /* If you haven't initialized, you must do that first. */ 277 if (req != LHREQ_INITIALIZE && !lg) 278 return -EINVAL; 279 280 /* Once the Guest is dead, all you can do is read() why it died. */ 281 if (lg && lg->dead) 282 return -ENOENT; 283 284 /* If you're not the task which owns the Guest, you can only break */ 285 if (lg && current != lg->tsk && req != LHREQ_BREAK) 286 return -EPERM; 287 288 switch (req) { 289 case LHREQ_INITIALIZE: 290 return initialize(file, (const u32 __user *)input); 291 case LHREQ_GETDMA: 292 return user_get_dma(lg, (const u32 __user *)input); 293 case LHREQ_IRQ: 294 return user_send_irq(lg, (const u32 __user *)input); 295 case LHREQ_BREAK: 296 return break_guest_out(lg, (const u32 __user *)input); 297 default: 298 return -EINVAL; 299 } 300} 301 302/*L:060 The final piece of interface code is the close() routine. It reverses 303 * everything done in initialize(). This is usually called because the 304 * Launcher exited. 305 * 306 * Note that the close routine returns 0 or a negative error number: it can't 307 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 308 * letting them do it. :*/ 309static int close(struct inode *inode, struct file *file) 310{ 311 struct lguest *lg = file->private_data; 312 313 /* If we never successfully initialized, there's nothing to clean up */ 314 if (!lg) 315 return 0; 316 317 /* We need the big lock, to protect from inter-guest I/O and other 318 * Launchers initializing guests. */ 319 mutex_lock(&lguest_lock); 320 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 321 hrtimer_cancel(&lg->hrt); 322 /* Free any DMA buffers the Guest had bound. */ 323 release_all_dma(lg); 324 /* Free up the shadow page tables for the Guest. */ 325 free_guest_pagetable(lg); 326 /* Now all the memory cleanups are done, it's safe to release the 327 * Launcher's memory management structure. */ 328 mmput(lg->mm); 329 /* If lg->dead doesn't contain an error code it will be NULL or a 330 * kmalloc()ed string, either of which is ok to hand to kfree(). */ 331 if (!IS_ERR(lg->dead)) 332 kfree(lg->dead); 333 /* We can free up the register page we allocated. */ 334 free_page(lg->regs_page); 335 /* We clear the entire structure, which also marks it as free for the 336 * next user. */ 337 memset(lg, 0, sizeof(*lg)); 338 /* Release lock and exit. */ 339 mutex_unlock(&lguest_lock); 340 341 return 0; 342} 343 344/*L:000 345 * Welcome to our journey through the Launcher! 346 * 347 * The Launcher is the Host userspace program which sets up, runs and services 348 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 349 * doing things are inaccurate: the Launcher does all the device handling for 350 * the Guest. The Guest can't tell what's done by the the Launcher and what by 351 * the Host. 352 * 353 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 354 * shall see more of that later. 355 * 356 * We begin our understanding with the Host kernel interface which the Launcher 357 * uses: reading and writing a character device called /dev/lguest. All the 358 * work happens in the read(), write() and close() routines: */ 359static struct file_operations lguest_fops = { 360 .owner = THIS_MODULE, 361 .release = close, 362 .write = write, 363 .read = read, 364}; 365 366/* This is a textbook example of a "misc" character device. Populate a "struct 367 * miscdevice" and register it with misc_register(). */ 368static struct miscdevice lguest_dev = { 369 .minor = MISC_DYNAMIC_MINOR, 370 .name = "lguest", 371 .fops = &lguest_fops, 372}; 373 374int __init lguest_device_init(void) 375{ 376 return misc_register(&lguest_dev); 377} 378 379void __exit lguest_device_remove(void) 380{ 381 misc_deregister(&lguest_dev); 382} 383