xref: /linux/arch/um/drivers/ubd_kern.c (revision 9ce7677cfd7cd871adb457c80bea3b581b839641)
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5 
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19 
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22 
23 #include "linux/config.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/devfs_fs_kernel.h"
29 #include "linux/cdrom.h"
30 #include "linux/proc_fs.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
33 #include "linux/mm.h"
34 #include "linux/vmalloc.h"
35 #include "linux/blkpg.h"
36 #include "linux/genhd.h"
37 #include "linux/spinlock.h"
38 #include "linux/platform_device.h"
39 #include "asm/segment.h"
40 #include "asm/uaccess.h"
41 #include "asm/irq.h"
42 #include "asm/types.h"
43 #include "asm/tlbflush.h"
44 #include "user_util.h"
45 #include "mem_user.h"
46 #include "kern_util.h"
47 #include "kern.h"
48 #include "mconsole_kern.h"
49 #include "init.h"
50 #include "irq_user.h"
51 #include "irq_kern.h"
52 #include "ubd_user.h"
53 #include "os.h"
54 #include "mem.h"
55 #include "mem_kern.h"
56 #include "cow.h"
57 
58 enum ubd_req { UBD_READ, UBD_WRITE };
59 
60 struct io_thread_req {
61 	enum ubd_req op;
62 	int fds[2];
63 	unsigned long offsets[2];
64 	unsigned long long offset;
65 	unsigned long length;
66 	char *buffer;
67 	int sectorsize;
68 	unsigned long sector_mask;
69 	unsigned long long cow_offset;
70 	unsigned long bitmap_words[2];
71 	int error;
72 };
73 
74 extern int open_ubd_file(char *file, struct openflags *openflags,
75 			 char **backing_file_out, int *bitmap_offset_out,
76 			 unsigned long *bitmap_len_out, int *data_offset_out,
77 			 int *create_cow_out);
78 extern int create_cow_file(char *cow_file, char *backing_file,
79 			   struct openflags flags, int sectorsize,
80 			   int alignment, int *bitmap_offset_out,
81 			   unsigned long *bitmap_len_out,
82 			   int *data_offset_out);
83 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
84 extern void do_io(struct io_thread_req *req);
85 
86 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
87 {
88 	__u64 n;
89 	int bits, off;
90 
91 	bits = sizeof(data[0]) * 8;
92 	n = bit / bits;
93 	off = bit % bits;
94 	return((data[n] & (1 << off)) != 0);
95 }
96 
97 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
98 {
99 	__u64 n;
100 	int bits, off;
101 
102 	bits = sizeof(data[0]) * 8;
103 	n = bit / bits;
104 	off = bit % bits;
105 	data[n] |= (1 << off);
106 }
107 /*End stuff from ubd_user.h*/
108 
109 #define DRIVER_NAME "uml-blkdev"
110 
111 static DEFINE_SPINLOCK(ubd_io_lock);
112 static DEFINE_SPINLOCK(ubd_lock);
113 
114 static void (*do_ubd)(void);
115 
116 static int ubd_open(struct inode * inode, struct file * filp);
117 static int ubd_release(struct inode * inode, struct file * file);
118 static int ubd_ioctl(struct inode * inode, struct file * file,
119 		     unsigned int cmd, unsigned long arg);
120 
121 #define MAX_DEV (8)
122 
123 static struct block_device_operations ubd_blops = {
124         .owner		= THIS_MODULE,
125         .open		= ubd_open,
126         .release	= ubd_release,
127         .ioctl		= ubd_ioctl,
128 };
129 
130 /* Protected by the queue_lock */
131 static request_queue_t *ubd_queue;
132 
133 /* Protected by ubd_lock */
134 static int fake_major = MAJOR_NR;
135 
136 static struct gendisk *ubd_gendisk[MAX_DEV];
137 static struct gendisk *fake_gendisk[MAX_DEV];
138 
139 #ifdef CONFIG_BLK_DEV_UBD_SYNC
140 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
141 					 .cl = 1 })
142 #else
143 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
144 					 .cl = 1 })
145 #endif
146 
147 /* Not protected - changed only in ubd_setup_common and then only to
148  * to enable O_SYNC.
149  */
150 static struct openflags global_openflags = OPEN_FLAGS;
151 
152 struct cow {
153 	/* This is the backing file, actually */
154 	char *file;
155 	int fd;
156 	unsigned long *bitmap;
157 	unsigned long bitmap_len;
158 	int bitmap_offset;
159         int data_offset;
160 };
161 
162 struct ubd {
163 	char *file;
164 	int count;
165 	int fd;
166 	__u64 size;
167 	struct openflags boot_openflags;
168 	struct openflags openflags;
169 	int no_cow;
170 	struct cow cow;
171 	struct platform_device pdev;
172 };
173 
174 #define DEFAULT_COW { \
175 	.file =			NULL, \
176         .fd =			-1, \
177         .bitmap =		NULL, \
178 	.bitmap_offset =	0, \
179         .data_offset =		0, \
180 }
181 
182 #define DEFAULT_UBD { \
183 	.file = 		NULL, \
184 	.count =		0, \
185 	.fd =			-1, \
186 	.size =			-1, \
187 	.boot_openflags =	OPEN_FLAGS, \
188 	.openflags =		OPEN_FLAGS, \
189         .no_cow =               0, \
190         .cow =			DEFAULT_COW, \
191 }
192 
193 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
194 
195 static int ubd0_init(void)
196 {
197 	struct ubd *dev = &ubd_dev[0];
198 
199 	if(dev->file == NULL)
200 		dev->file = "root_fs";
201 	return(0);
202 }
203 
204 __initcall(ubd0_init);
205 
206 /* Only changed by fake_ide_setup which is a setup */
207 static int fake_ide = 0;
208 static struct proc_dir_entry *proc_ide_root = NULL;
209 static struct proc_dir_entry *proc_ide = NULL;
210 
211 static void make_proc_ide(void)
212 {
213 	proc_ide_root = proc_mkdir("ide", NULL);
214 	proc_ide = proc_mkdir("ide0", proc_ide_root);
215 }
216 
217 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
218 			       int *eof, void *data)
219 {
220 	int len;
221 
222 	strcpy(page, "disk\n");
223 	len = strlen("disk\n");
224 	len -= off;
225 	if (len < count){
226 		*eof = 1;
227 		if (len <= 0) return 0;
228 	}
229 	else len = count;
230 	*start = page + off;
231 	return len;
232 }
233 
234 static void make_ide_entries(char *dev_name)
235 {
236 	struct proc_dir_entry *dir, *ent;
237 	char name[64];
238 
239 	if(proc_ide_root == NULL) make_proc_ide();
240 
241 	dir = proc_mkdir(dev_name, proc_ide);
242 	if(!dir) return;
243 
244 	ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
245 	if(!ent) return;
246 	ent->nlink = 1;
247 	ent->data = NULL;
248 	ent->read_proc = proc_ide_read_media;
249 	ent->write_proc = NULL;
250 	sprintf(name,"ide0/%s", dev_name);
251 	proc_symlink(dev_name, proc_ide_root, name);
252 }
253 
254 static int fake_ide_setup(char *str)
255 {
256 	fake_ide = 1;
257 	return(1);
258 }
259 
260 __setup("fake_ide", fake_ide_setup);
261 
262 __uml_help(fake_ide_setup,
263 "fake_ide\n"
264 "    Create ide0 entries that map onto ubd devices.\n\n"
265 );
266 
267 static int parse_unit(char **ptr)
268 {
269 	char *str = *ptr, *end;
270 	int n = -1;
271 
272 	if(isdigit(*str)) {
273 		n = simple_strtoul(str, &end, 0);
274 		if(end == str)
275 			return(-1);
276 		*ptr = end;
277 	}
278 	else if (('a' <= *str) && (*str <= 'h')) {
279 		n = *str - 'a';
280 		str++;
281 		*ptr = str;
282 	}
283 	return(n);
284 }
285 
286 static int ubd_setup_common(char *str, int *index_out)
287 {
288 	struct ubd *dev;
289 	struct openflags flags = global_openflags;
290 	char *backing_file;
291 	int n, err, i;
292 
293 	if(index_out) *index_out = -1;
294 	n = *str;
295 	if(n == '='){
296 		char *end;
297 		int major;
298 
299 		str++;
300 		if(!strcmp(str, "sync")){
301 			global_openflags = of_sync(global_openflags);
302 			return(0);
303 		}
304 		major = simple_strtoul(str, &end, 0);
305 		if((*end != '\0') || (end == str)){
306 			printk(KERN_ERR
307 			       "ubd_setup : didn't parse major number\n");
308 			return(1);
309 		}
310 
311 		err = 1;
312  		spin_lock(&ubd_lock);
313  		if(fake_major != MAJOR_NR){
314  			printk(KERN_ERR "Can't assign a fake major twice\n");
315  			goto out1;
316  		}
317 
318  		fake_major = major;
319 
320 		printk(KERN_INFO "Setting extra ubd major number to %d\n",
321 		       major);
322  		err = 0;
323  	out1:
324  		spin_unlock(&ubd_lock);
325 		return(err);
326 	}
327 
328 	n = parse_unit(&str);
329 	if(n < 0){
330 		printk(KERN_ERR "ubd_setup : couldn't parse unit number "
331 		       "'%s'\n", str);
332 		return(1);
333 	}
334 	if(n >= MAX_DEV){
335 		printk(KERN_ERR "ubd_setup : index %d out of range "
336 		       "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
337 		return(1);
338 	}
339 
340 	err = 1;
341 	spin_lock(&ubd_lock);
342 
343 	dev = &ubd_dev[n];
344 	if(dev->file != NULL){
345 		printk(KERN_ERR "ubd_setup : device already configured\n");
346 		goto out;
347 	}
348 
349 	if (index_out)
350 		*index_out = n;
351 
352 	for (i = 0; i < 4; i++) {
353 		switch (*str) {
354 		case 'r':
355 			flags.w = 0;
356 			break;
357 		case 's':
358 			flags.s = 1;
359 			break;
360 		case 'd':
361 			dev->no_cow = 1;
362 			break;
363 		case '=':
364 			str++;
365 			goto break_loop;
366 		default:
367 			printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r,s or d)\n");
368 			goto out;
369 		}
370 		str++;
371 	}
372 
373         if (*str == '=')
374 		printk(KERN_ERR "ubd_setup : Too many flags specified\n");
375         else
376 		printk(KERN_ERR "ubd_setup : Expected '='\n");
377 	goto out;
378 
379 break_loop:
380 	err = 0;
381 	backing_file = strchr(str, ',');
382 
383 	if (!backing_file) {
384 		backing_file = strchr(str, ':');
385 	}
386 
387 	if(backing_file){
388 		if(dev->no_cow)
389 			printk(KERN_ERR "Can't specify both 'd' and a "
390 			       "cow file\n");
391 		else {
392 			*backing_file = '\0';
393 			backing_file++;
394 		}
395 	}
396 	dev->file = str;
397 	dev->cow.file = backing_file;
398 	dev->boot_openflags = flags;
399 out:
400 	spin_unlock(&ubd_lock);
401 	return(err);
402 }
403 
404 static int ubd_setup(char *str)
405 {
406 	ubd_setup_common(str, NULL);
407 	return(1);
408 }
409 
410 __setup("ubd", ubd_setup);
411 __uml_help(ubd_setup,
412 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
413 "    This is used to associate a device with a file in the underlying\n"
414 "    filesystem. When specifying two filenames, the first one is the\n"
415 "    COW name and the second is the backing file name. As separator you can\n"
416 "    use either a ':' or a ',': the first one allows writing things like;\n"
417 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
418 "    while with a ',' the shell would not expand the 2nd '~'.\n"
419 "    When using only one filename, UML will detect whether to thread it like\n"
420 "    a COW file or a backing file. To override this detection, add the 'd'\n"
421 "    flag:\n"
422 "	ubd0d=BackingFile\n"
423 "    Usually, there is a filesystem in the file, but \n"
424 "    that's not required. Swap devices containing swap files can be\n"
425 "    specified like this. Also, a file which doesn't contain a\n"
426 "    filesystem can have its contents read in the virtual \n"
427 "    machine by running 'dd' on the device. <n> must be in the range\n"
428 "    0 to 7. Appending an 'r' to the number will cause that device\n"
429 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
430 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
431 );
432 
433 static int udb_setup(char *str)
434 {
435 	printk("udb%s specified on command line is almost certainly a ubd -> "
436 	       "udb TYPO\n", str);
437 	return(1);
438 }
439 
440 __setup("udb", udb_setup);
441 __uml_help(udb_setup,
442 "udb\n"
443 "    This option is here solely to catch ubd -> udb typos, which can be\n"
444 "    to impossible to catch visually unless you specifically look for\n"
445 "    them.  The only result of any option starting with 'udb' is an error\n"
446 "    in the boot output.\n\n"
447 );
448 
449 static int fakehd_set = 0;
450 static int fakehd(char *str)
451 {
452 	printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
453 	fakehd_set = 1;
454 	return 1;
455 }
456 
457 __setup("fakehd", fakehd);
458 __uml_help(fakehd,
459 "fakehd\n"
460 "    Change the ubd device name to \"hd\".\n\n"
461 );
462 
463 static void do_ubd_request(request_queue_t * q);
464 
465 /* Only changed by ubd_init, which is an initcall. */
466 int thread_fd = -1;
467 
468 /* Changed by ubd_handler, which is serialized because interrupts only
469  * happen on CPU 0.
470  */
471 int intr_count = 0;
472 
473 /* call ubd_finish if you need to serialize */
474 static void __ubd_finish(struct request *req, int error)
475 {
476 	int nsect;
477 
478 	if(error){
479 		end_request(req, 0);
480 		return;
481 	}
482 	nsect = req->current_nr_sectors;
483 	req->sector += nsect;
484 	req->buffer += nsect << 9;
485 	req->errors = 0;
486 	req->nr_sectors -= nsect;
487 	req->current_nr_sectors = 0;
488 	end_request(req, 1);
489 }
490 
491 static inline void ubd_finish(struct request *req, int error)
492 {
493  	spin_lock(&ubd_io_lock);
494 	__ubd_finish(req, error);
495 	spin_unlock(&ubd_io_lock);
496 }
497 
498 /* Called without ubd_io_lock held */
499 static void ubd_handler(void)
500 {
501 	struct io_thread_req req;
502 	struct request *rq = elv_next_request(ubd_queue);
503 	int n;
504 
505 	do_ubd = NULL;
506 	intr_count++;
507 	n = os_read_file(thread_fd, &req, sizeof(req));
508 	if(n != sizeof(req)){
509 		printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
510 		       "err = %d\n", os_getpid(), -n);
511 		spin_lock(&ubd_io_lock);
512 		end_request(rq, 0);
513 		spin_unlock(&ubd_io_lock);
514 		return;
515 	}
516 
517 	ubd_finish(rq, req.error);
518 	reactivate_fd(thread_fd, UBD_IRQ);
519 	do_ubd_request(ubd_queue);
520 }
521 
522 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
523 {
524 	ubd_handler();
525 	return(IRQ_HANDLED);
526 }
527 
528 /* Only changed by ubd_init, which is an initcall. */
529 static int io_pid = -1;
530 
531 void kill_io_thread(void)
532 {
533 	if(io_pid != -1)
534 		os_kill_process(io_pid, 1);
535 }
536 
537 __uml_exitcall(kill_io_thread);
538 
539 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
540 {
541 	char *file;
542 
543 	file = dev->cow.file ? dev->cow.file : dev->file;
544 	return(os_file_size(file, size_out));
545 }
546 
547 static void ubd_close(struct ubd *dev)
548 {
549 	os_close_file(dev->fd);
550 	if(dev->cow.file == NULL)
551 		return;
552 
553 	os_close_file(dev->cow.fd);
554 	vfree(dev->cow.bitmap);
555 	dev->cow.bitmap = NULL;
556 }
557 
558 static int ubd_open_dev(struct ubd *dev)
559 {
560 	struct openflags flags;
561 	char **back_ptr;
562 	int err, create_cow, *create_ptr;
563 
564 	dev->openflags = dev->boot_openflags;
565 	create_cow = 0;
566 	create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
567 	back_ptr = dev->no_cow ? NULL : &dev->cow.file;
568 	dev->fd = open_ubd_file(dev->file, &dev->openflags, back_ptr,
569 				&dev->cow.bitmap_offset, &dev->cow.bitmap_len,
570 				&dev->cow.data_offset, create_ptr);
571 
572 	if((dev->fd == -ENOENT) && create_cow){
573 		dev->fd = create_cow_file(dev->file, dev->cow.file,
574 					  dev->openflags, 1 << 9, PAGE_SIZE,
575 					  &dev->cow.bitmap_offset,
576 					  &dev->cow.bitmap_len,
577 					  &dev->cow.data_offset);
578 		if(dev->fd >= 0){
579 			printk(KERN_INFO "Creating \"%s\" as COW file for "
580 			       "\"%s\"\n", dev->file, dev->cow.file);
581 		}
582 	}
583 
584 	if(dev->fd < 0){
585 		printk("Failed to open '%s', errno = %d\n", dev->file,
586 		       -dev->fd);
587 		return(dev->fd);
588 	}
589 
590 	if(dev->cow.file != NULL){
591 		err = -ENOMEM;
592 		dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
593 		if(dev->cow.bitmap == NULL){
594 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
595 			goto error;
596 		}
597 		flush_tlb_kernel_vm();
598 
599 		err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
600 				      dev->cow.bitmap_offset,
601 				      dev->cow.bitmap_len);
602 		if(err < 0)
603 			goto error;
604 
605 		flags = dev->openflags;
606 		flags.w = 0;
607 		err = open_ubd_file(dev->cow.file, &flags, NULL, NULL, NULL,
608 				    NULL, NULL);
609 		if(err < 0) goto error;
610 		dev->cow.fd = err;
611 	}
612 	return(0);
613  error:
614 	os_close_file(dev->fd);
615 	return(err);
616 }
617 
618 static int ubd_new_disk(int major, u64 size, int unit,
619 			struct gendisk **disk_out)
620 
621 {
622 	struct gendisk *disk;
623 	char from[sizeof("ubd/nnnnn\0")], to[sizeof("discnnnnn/disc\0")];
624 	int err;
625 
626 	disk = alloc_disk(1 << UBD_SHIFT);
627 	if(disk == NULL)
628 		return(-ENOMEM);
629 
630 	disk->major = major;
631 	disk->first_minor = unit << UBD_SHIFT;
632 	disk->fops = &ubd_blops;
633 	set_capacity(disk, size / 512);
634 	if(major == MAJOR_NR){
635 		sprintf(disk->disk_name, "ubd%c", 'a' + unit);
636 		sprintf(disk->devfs_name, "ubd/disc%d", unit);
637 		sprintf(from, "ubd/%d", unit);
638 		sprintf(to, "disc%d/disc", unit);
639 		err = devfs_mk_symlink(from, to);
640 		if(err)
641 			printk("ubd_new_disk failed to make link from %s to "
642 			       "%s, error = %d\n", from, to, err);
643 	}
644 	else {
645 		sprintf(disk->disk_name, "ubd_fake%d", unit);
646 		sprintf(disk->devfs_name, "ubd_fake/disc%d", unit);
647 	}
648 
649 	/* sysfs register (not for ide fake devices) */
650 	if (major == MAJOR_NR) {
651 		ubd_dev[unit].pdev.id   = unit;
652 		ubd_dev[unit].pdev.name = DRIVER_NAME;
653 		platform_device_register(&ubd_dev[unit].pdev);
654 		disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
655 	}
656 
657 	disk->private_data = &ubd_dev[unit];
658 	disk->queue = ubd_queue;
659 	add_disk(disk);
660 
661 	*disk_out = disk;
662 	return 0;
663 }
664 
665 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
666 
667 static int ubd_add(int n)
668 {
669 	struct ubd *dev = &ubd_dev[n];
670 	int err;
671 
672 	err = -ENODEV;
673 	if(dev->file == NULL)
674 		goto out;
675 
676 	if (ubd_open_dev(dev))
677 		goto out;
678 
679 	err = ubd_file_size(dev, &dev->size);
680 	if(err < 0)
681 		goto out_close;
682 
683 	dev->size = ROUND_BLOCK(dev->size);
684 
685 	err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
686 	if(err)
687 		goto out_close;
688 
689 	if(fake_major != MAJOR_NR)
690 		ubd_new_disk(fake_major, dev->size, n,
691 			     &fake_gendisk[n]);
692 
693 	/* perhaps this should also be under the "if (fake_major)" above */
694 	/* using the fake_disk->disk_name and also the fakehd_set name */
695 	if (fake_ide)
696 		make_ide_entries(ubd_gendisk[n]->disk_name);
697 
698 	err = 0;
699 out_close:
700 	ubd_close(dev);
701 out:
702 	return err;
703 }
704 
705 static int ubd_config(char *str)
706 {
707 	int n, err;
708 
709 	str = uml_strdup(str);
710 	if(str == NULL){
711 		printk(KERN_ERR "ubd_config failed to strdup string\n");
712 		return(1);
713 	}
714 	err = ubd_setup_common(str, &n);
715 	if(err){
716 		kfree(str);
717 		return(-1);
718 	}
719 	if(n == -1) return(0);
720 
721  	spin_lock(&ubd_lock);
722 	err = ubd_add(n);
723 	if(err)
724 		ubd_dev[n].file = NULL;
725  	spin_unlock(&ubd_lock);
726 
727 	return(err);
728 }
729 
730 static int ubd_get_config(char *name, char *str, int size, char **error_out)
731 {
732 	struct ubd *dev;
733 	int n, len = 0;
734 
735 	n = parse_unit(&name);
736 	if((n >= MAX_DEV) || (n < 0)){
737 		*error_out = "ubd_get_config : device number out of range";
738 		return(-1);
739 	}
740 
741 	dev = &ubd_dev[n];
742 	spin_lock(&ubd_lock);
743 
744 	if(dev->file == NULL){
745 		CONFIG_CHUNK(str, size, len, "", 1);
746 		goto out;
747 	}
748 
749 	CONFIG_CHUNK(str, size, len, dev->file, 0);
750 
751 	if(dev->cow.file != NULL){
752 		CONFIG_CHUNK(str, size, len, ",", 0);
753 		CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
754 	}
755 	else CONFIG_CHUNK(str, size, len, "", 1);
756 
757  out:
758 	spin_unlock(&ubd_lock);
759 	return(len);
760 }
761 
762 static int ubd_id(char **str, int *start_out, int *end_out)
763 {
764         int n;
765 
766 	n = parse_unit(str);
767         *start_out = 0;
768         *end_out = MAX_DEV - 1;
769         return n;
770 }
771 
772 static int ubd_remove(int n)
773 {
774 	struct ubd *dev;
775 	int err = -ENODEV;
776 
777 	spin_lock(&ubd_lock);
778 
779 	if(ubd_gendisk[n] == NULL)
780 		goto out;
781 
782 	dev = &ubd_dev[n];
783 
784 	if(dev->file == NULL)
785 		goto out;
786 
787 	/* you cannot remove a open disk */
788 	err = -EBUSY;
789 	if(dev->count > 0)
790 		goto out;
791 
792 	del_gendisk(ubd_gendisk[n]);
793 	put_disk(ubd_gendisk[n]);
794 	ubd_gendisk[n] = NULL;
795 
796 	if(fake_gendisk[n] != NULL){
797 		del_gendisk(fake_gendisk[n]);
798 		put_disk(fake_gendisk[n]);
799 		fake_gendisk[n] = NULL;
800 	}
801 
802 	platform_device_unregister(&dev->pdev);
803 	*dev = ((struct ubd) DEFAULT_UBD);
804 	err = 0;
805 out:
806 	spin_unlock(&ubd_lock);
807 	return err;
808 }
809 
810 static struct mc_device ubd_mc = {
811 	.name		= "ubd",
812 	.config		= ubd_config,
813  	.get_config	= ubd_get_config,
814 	.id		= ubd_id,
815 	.remove		= ubd_remove,
816 };
817 
818 static int ubd_mc_init(void)
819 {
820 	mconsole_register_dev(&ubd_mc);
821 	return 0;
822 }
823 
824 __initcall(ubd_mc_init);
825 
826 static struct platform_driver ubd_driver = {
827 	.driver = {
828 		.name  = DRIVER_NAME,
829 	},
830 };
831 
832 int ubd_init(void)
833 {
834         int i;
835 
836 	devfs_mk_dir("ubd");
837 	if (register_blkdev(MAJOR_NR, "ubd"))
838 		return -1;
839 
840 	ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
841 	if (!ubd_queue) {
842 		unregister_blkdev(MAJOR_NR, "ubd");
843 		return -1;
844 	}
845 
846 	if (fake_major != MAJOR_NR) {
847 		char name[sizeof("ubd_nnn\0")];
848 
849 		snprintf(name, sizeof(name), "ubd_%d", fake_major);
850 		devfs_mk_dir(name);
851 		if (register_blkdev(fake_major, "ubd"))
852 			return -1;
853 	}
854 	platform_driver_register(&ubd_driver);
855 	for (i = 0; i < MAX_DEV; i++)
856 		ubd_add(i);
857 	return 0;
858 }
859 
860 late_initcall(ubd_init);
861 
862 int ubd_driver_init(void){
863 	unsigned long stack;
864 	int err;
865 
866 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
867 	if(global_openflags.s){
868 		printk(KERN_INFO "ubd: Synchronous mode\n");
869 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
870 		 * enough. So use anyway the io thread. */
871 	}
872 	stack = alloc_stack(0, 0);
873 	io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
874 				 &thread_fd);
875 	if(io_pid < 0){
876 		printk(KERN_ERR
877 		       "ubd : Failed to start I/O thread (errno = %d) - "
878 		       "falling back to synchronous I/O\n", -io_pid);
879 		io_pid = -1;
880 		return(0);
881 	}
882 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
883 			     SA_INTERRUPT, "ubd", ubd_dev);
884 	if(err != 0)
885 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
886 	return(err);
887 }
888 
889 device_initcall(ubd_driver_init);
890 
891 static int ubd_open(struct inode *inode, struct file *filp)
892 {
893 	struct gendisk *disk = inode->i_bdev->bd_disk;
894 	struct ubd *dev = disk->private_data;
895 	int err = 0;
896 
897 	if(dev->count == 0){
898 		err = ubd_open_dev(dev);
899 		if(err){
900 			printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
901 			       disk->disk_name, dev->file, -err);
902 			goto out;
903 		}
904 	}
905 	dev->count++;
906 	set_disk_ro(disk, !dev->openflags.w);
907 
908 	/* This should no more be needed. And it didn't work anyway to exclude
909 	 * read-write remounting of filesystems.*/
910 	/*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
911 	        if(--dev->count == 0) ubd_close(dev);
912 	        err = -EROFS;
913 	}*/
914  out:
915 	return(err);
916 }
917 
918 static int ubd_release(struct inode * inode, struct file * file)
919 {
920 	struct gendisk *disk = inode->i_bdev->bd_disk;
921 	struct ubd *dev = disk->private_data;
922 
923 	if(--dev->count == 0)
924 		ubd_close(dev);
925 	return(0);
926 }
927 
928 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
929 			  __u64 *cow_offset, unsigned long *bitmap,
930 			  __u64 bitmap_offset, unsigned long *bitmap_words,
931 			  __u64 bitmap_len)
932 {
933 	__u64 sector = io_offset >> 9;
934 	int i, update_bitmap = 0;
935 
936 	for(i = 0; i < length >> 9; i++){
937 		if(cow_mask != NULL)
938 			ubd_set_bit(i, (unsigned char *) cow_mask);
939 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
940 			continue;
941 
942 		update_bitmap = 1;
943 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
944 	}
945 
946 	if(!update_bitmap)
947 		return;
948 
949 	*cow_offset = sector / (sizeof(unsigned long) * 8);
950 
951 	/* This takes care of the case where we're exactly at the end of the
952 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
953 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
954 	 * for the original diagnosis.
955 	 */
956 	if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
957 			   sizeof(unsigned long) - 1))
958 		(*cow_offset)--;
959 
960 	bitmap_words[0] = bitmap[*cow_offset];
961 	bitmap_words[1] = bitmap[*cow_offset + 1];
962 
963 	*cow_offset *= sizeof(unsigned long);
964 	*cow_offset += bitmap_offset;
965 }
966 
967 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
968 		       __u64 bitmap_offset, __u64 bitmap_len)
969 {
970 	__u64 sector = req->offset >> 9;
971 	int i;
972 
973 	if(req->length > (sizeof(req->sector_mask) * 8) << 9)
974 		panic("Operation too long");
975 
976 	if(req->op == UBD_READ) {
977 		for(i = 0; i < req->length >> 9; i++){
978 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
979 				ubd_set_bit(i, (unsigned char *)
980 					    &req->sector_mask);
981                 }
982 	}
983 	else cowify_bitmap(req->offset, req->length, &req->sector_mask,
984 			   &req->cow_offset, bitmap, bitmap_offset,
985 			   req->bitmap_words, bitmap_len);
986 }
987 
988 /* Called with ubd_io_lock held */
989 static int prepare_request(struct request *req, struct io_thread_req *io_req)
990 {
991 	struct gendisk *disk = req->rq_disk;
992 	struct ubd *dev = disk->private_data;
993 	__u64 offset;
994 	int len;
995 
996 	if(req->rq_status == RQ_INACTIVE) return(1);
997 
998 	/* This should be impossible now */
999 	if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
1000 		printk("Write attempted on readonly ubd device %s\n",
1001 		       disk->disk_name);
1002 		end_request(req, 0);
1003 		return(1);
1004 	}
1005 
1006 	offset = ((__u64) req->sector) << 9;
1007 	len = req->current_nr_sectors << 9;
1008 
1009 	io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1010 	io_req->fds[1] = dev->fd;
1011 	io_req->cow_offset = -1;
1012 	io_req->offset = offset;
1013 	io_req->length = len;
1014 	io_req->error = 0;
1015 	io_req->sector_mask = 0;
1016 
1017 	io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1018 	io_req->offsets[0] = 0;
1019 	io_req->offsets[1] = dev->cow.data_offset;
1020 	io_req->buffer = req->buffer;
1021 	io_req->sectorsize = 1 << 9;
1022 
1023 	if(dev->cow.file != NULL)
1024 		cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1025 			   dev->cow.bitmap_len);
1026 
1027 	return(0);
1028 }
1029 
1030 /* Called with ubd_io_lock held */
1031 static void do_ubd_request(request_queue_t *q)
1032 {
1033 	struct io_thread_req io_req;
1034 	struct request *req;
1035 	int err, n;
1036 
1037 	if(thread_fd == -1){
1038 		while((req = elv_next_request(q)) != NULL){
1039 			err = prepare_request(req, &io_req);
1040 			if(!err){
1041 				do_io(&io_req);
1042 				__ubd_finish(req, io_req.error);
1043 			}
1044 		}
1045 	}
1046 	else {
1047 		if(do_ubd || (req = elv_next_request(q)) == NULL)
1048 			return;
1049 		err = prepare_request(req, &io_req);
1050 		if(!err){
1051 			do_ubd = ubd_handler;
1052 			n = os_write_file(thread_fd, (char *) &io_req,
1053 					 sizeof(io_req));
1054 			if(n != sizeof(io_req))
1055 				printk("write to io thread failed, "
1056 				       "errno = %d\n", -n);
1057 		}
1058 	}
1059 }
1060 
1061 static int ubd_ioctl(struct inode * inode, struct file * file,
1062 		     unsigned int cmd, unsigned long arg)
1063 {
1064 	struct hd_geometry __user *loc = (struct hd_geometry __user *) arg;
1065 	struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1066 	struct hd_driveid ubd_id = {
1067 		.cyls		= 0,
1068 		.heads		= 128,
1069 		.sectors	= 32,
1070 	};
1071 
1072 	switch (cmd) {
1073 	        struct hd_geometry g;
1074 		struct cdrom_volctrl volume;
1075 	case HDIO_GETGEO:
1076 		if(!loc) return(-EINVAL);
1077 		g.heads = 128;
1078 		g.sectors = 32;
1079 		g.cylinders = dev->size / (128 * 32 * 512);
1080 		g.start = get_start_sect(inode->i_bdev);
1081 		return(copy_to_user(loc, &g, sizeof(g)) ? -EFAULT : 0);
1082 
1083 	case HDIO_GET_IDENTITY:
1084 		ubd_id.cyls = dev->size / (128 * 32 * 512);
1085 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1086 				 sizeof(ubd_id)))
1087 			return(-EFAULT);
1088 		return(0);
1089 
1090 	case CDROMVOLREAD:
1091 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1092 			return(-EFAULT);
1093 		volume.channel0 = 255;
1094 		volume.channel1 = 255;
1095 		volume.channel2 = 255;
1096 		volume.channel3 = 255;
1097 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1098 			return(-EFAULT);
1099 		return(0);
1100 	}
1101 	return(-EINVAL);
1102 }
1103 
1104 static int same_backing_files(char *from_cmdline, char *from_cow, char *cow)
1105 {
1106 	struct uml_stat buf1, buf2;
1107 	int err;
1108 
1109 	if(from_cmdline == NULL) return(1);
1110 	if(!strcmp(from_cmdline, from_cow)) return(1);
1111 
1112 	err = os_stat_file(from_cmdline, &buf1);
1113 	if(err < 0){
1114 		printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1115 		return(1);
1116 	}
1117 	err = os_stat_file(from_cow, &buf2);
1118 	if(err < 0){
1119 		printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1120 		return(1);
1121 	}
1122 	if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1123 		return(1);
1124 
1125 	printk("Backing file mismatch - \"%s\" requested,\n"
1126 	       "\"%s\" specified in COW header of \"%s\"\n",
1127 	       from_cmdline, from_cow, cow);
1128 	return(0);
1129 }
1130 
1131 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1132 {
1133 	unsigned long modtime;
1134 	long long actual;
1135 	int err;
1136 
1137 	err = os_file_modtime(file, &modtime);
1138 	if(err < 0){
1139 		printk("Failed to get modification time of backing file "
1140 		       "\"%s\", err = %d\n", file, -err);
1141 		return(err);
1142 	}
1143 
1144 	err = os_file_size(file, &actual);
1145 	if(err < 0){
1146 		printk("Failed to get size of backing file \"%s\", "
1147 		       "err = %d\n", file, -err);
1148 		return(err);
1149 	}
1150 
1151   	if(actual != size){
1152 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
1153 		 * the typecast.*/
1154 		printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1155 		       "file\n", (unsigned long long) size, actual);
1156 		return(-EINVAL);
1157 	}
1158 	if(modtime != mtime){
1159 		printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1160 		       "file\n", mtime, modtime);
1161 		return(-EINVAL);
1162 	}
1163 	return(0);
1164 }
1165 
1166 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1167 {
1168 	int err;
1169 
1170 	err = os_seek_file(fd, offset);
1171 	if(err < 0)
1172 		return(err);
1173 
1174 	err = os_read_file(fd, buf, len);
1175 	if(err < 0)
1176 		return(err);
1177 
1178 	return(0);
1179 }
1180 
1181 int open_ubd_file(char *file, struct openflags *openflags,
1182 		  char **backing_file_out, int *bitmap_offset_out,
1183 		  unsigned long *bitmap_len_out, int *data_offset_out,
1184 		  int *create_cow_out)
1185 {
1186 	time_t mtime;
1187 	unsigned long long size;
1188 	__u32 version, align;
1189 	char *backing_file;
1190 	int fd, err, sectorsize, same, mode = 0644;
1191 
1192 	fd = os_open_file(file, *openflags, mode);
1193 	if(fd < 0){
1194 		if((fd == -ENOENT) && (create_cow_out != NULL))
1195 			*create_cow_out = 1;
1196                 if(!openflags->w ||
1197                    ((fd != -EROFS) && (fd != -EACCES))) return(fd);
1198 		openflags->w = 0;
1199 		fd = os_open_file(file, *openflags, mode);
1200 		if(fd < 0)
1201 			return(fd);
1202         }
1203 
1204 	err = os_lock_file(fd, openflags->w);
1205 	if(err < 0){
1206 		printk("Failed to lock '%s', err = %d\n", file, -err);
1207 		goto out_close;
1208 	}
1209 
1210 	if(backing_file_out == NULL) return(fd);
1211 
1212 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1213 			      &size, &sectorsize, &align, bitmap_offset_out);
1214 	if(err && (*backing_file_out != NULL)){
1215 		printk("Failed to read COW header from COW file \"%s\", "
1216 		       "errno = %d\n", file, -err);
1217 		goto out_close;
1218 	}
1219 	if(err) return(fd);
1220 
1221 	if(backing_file_out == NULL) return(fd);
1222 
1223 	same = same_backing_files(*backing_file_out, backing_file, file);
1224 
1225 	if(!same && !backing_file_mismatch(*backing_file_out, size, mtime)){
1226 		printk("Switching backing file to '%s'\n", *backing_file_out);
1227 		err = write_cow_header(file, fd, *backing_file_out,
1228 				       sectorsize, align, &size);
1229 		if(err){
1230 			printk("Switch failed, errno = %d\n", -err);
1231 			return(err);
1232 		}
1233 	}
1234 	else {
1235 		*backing_file_out = backing_file;
1236 		err = backing_file_mismatch(*backing_file_out, size, mtime);
1237 		if(err) goto out_close;
1238 	}
1239 
1240 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1241 		  bitmap_len_out, data_offset_out);
1242 
1243         return(fd);
1244  out_close:
1245 	os_close_file(fd);
1246 	return(err);
1247 }
1248 
1249 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1250 		    int sectorsize, int alignment, int *bitmap_offset_out,
1251 		    unsigned long *bitmap_len_out, int *data_offset_out)
1252 {
1253 	int err, fd;
1254 
1255 	flags.c = 1;
1256 	fd = open_ubd_file(cow_file, &flags, NULL, NULL, NULL, NULL, NULL);
1257 	if(fd < 0){
1258 		err = fd;
1259 		printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1260 		       -err);
1261 		goto out;
1262 	}
1263 
1264 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1265 			    bitmap_offset_out, bitmap_len_out,
1266 			    data_offset_out);
1267 	if(!err)
1268 		return(fd);
1269 	os_close_file(fd);
1270  out:
1271 	return(err);
1272 }
1273 
1274 static int update_bitmap(struct io_thread_req *req)
1275 {
1276 	int n;
1277 
1278 	if(req->cow_offset == -1)
1279 		return(0);
1280 
1281 	n = os_seek_file(req->fds[1], req->cow_offset);
1282 	if(n < 0){
1283 		printk("do_io - bitmap lseek failed : err = %d\n", -n);
1284 		return(1);
1285 	}
1286 
1287 	n = os_write_file(req->fds[1], &req->bitmap_words,
1288 		          sizeof(req->bitmap_words));
1289 	if(n != sizeof(req->bitmap_words)){
1290 		printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1291 		       req->fds[1]);
1292 		return(1);
1293 	}
1294 
1295 	return(0);
1296 }
1297 
1298 void do_io(struct io_thread_req *req)
1299 {
1300 	char *buf;
1301 	unsigned long len;
1302 	int n, nsectors, start, end, bit;
1303 	int err;
1304 	__u64 off;
1305 
1306 	nsectors = req->length / req->sectorsize;
1307 	start = 0;
1308 	do {
1309 		bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1310 		end = start;
1311 		while((end < nsectors) &&
1312 		      (ubd_test_bit(end, (unsigned char *)
1313 				    &req->sector_mask) == bit))
1314 			end++;
1315 
1316 		off = req->offset + req->offsets[bit] +
1317 			start * req->sectorsize;
1318 		len = (end - start) * req->sectorsize;
1319 		buf = &req->buffer[start * req->sectorsize];
1320 
1321 		err = os_seek_file(req->fds[bit], off);
1322 		if(err < 0){
1323 			printk("do_io - lseek failed : err = %d\n", -err);
1324 			req->error = 1;
1325 			return;
1326 		}
1327 		if(req->op == UBD_READ){
1328 			n = 0;
1329 			do {
1330 				buf = &buf[n];
1331 				len -= n;
1332 				n = os_read_file(req->fds[bit], buf, len);
1333 				if (n < 0) {
1334 					printk("do_io - read failed, err = %d "
1335 					       "fd = %d\n", -n, req->fds[bit]);
1336 					req->error = 1;
1337 					return;
1338 				}
1339 			} while((n < len) && (n != 0));
1340 			if (n < len) memset(&buf[n], 0, len - n);
1341 		} else {
1342 			n = os_write_file(req->fds[bit], buf, len);
1343 			if(n != len){
1344 				printk("do_io - write failed err = %d "
1345 				       "fd = %d\n", -n, req->fds[bit]);
1346 				req->error = 1;
1347 				return;
1348 			}
1349 		}
1350 
1351 		start = end;
1352 	} while(start < nsectors);
1353 
1354 	req->error = update_bitmap(req);
1355 }
1356 
1357 /* Changed in start_io_thread, which is serialized by being called only
1358  * from ubd_init, which is an initcall.
1359  */
1360 int kernel_fd = -1;
1361 
1362 /* Only changed by the io thread */
1363 int io_count = 0;
1364 
1365 int io_thread(void *arg)
1366 {
1367 	struct io_thread_req req;
1368 	int n;
1369 
1370 	ignore_sigwinch_sig();
1371 	while(1){
1372 		n = os_read_file(kernel_fd, &req, sizeof(req));
1373 		if(n != sizeof(req)){
1374 			if(n < 0)
1375 				printk("io_thread - read failed, fd = %d, "
1376 				       "err = %d\n", kernel_fd, -n);
1377 			else {
1378 				printk("io_thread - short read, fd = %d, "
1379 				       "length = %d\n", kernel_fd, n);
1380 			}
1381 			continue;
1382 		}
1383 		io_count++;
1384 		do_io(&req);
1385 		n = os_write_file(kernel_fd, &req, sizeof(req));
1386 		if(n != sizeof(req))
1387 			printk("io_thread - write failed, fd = %d, err = %d\n",
1388 			       kernel_fd, -n);
1389 	}
1390 }
1391 
1392 /*
1393  * Overrides for Emacs so that we follow Linus's tabbing style.
1394  * Emacs will notice this stuff at the end of the file and automatically
1395  * adjust the settings for this buffer only.  This must remain at the end
1396  * of the file.
1397  * ---------------------------------------------------------------------------
1398  * Local variables:
1399  * c-file-style: "linux"
1400  * End:
1401  */
1402