xref: /linux/arch/um/drivers/ubd_kern.c (revision 776cfebb430c7b22c208b1b17add97f354d97cab)
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 "asm/segment.h"
39 #include "asm/uaccess.h"
40 #include "asm/irq.h"
41 #include "asm/types.h"
42 #include "asm/tlbflush.h"
43 #include "user_util.h"
44 #include "mem_user.h"
45 #include "kern_util.h"
46 #include "kern.h"
47 #include "mconsole_kern.h"
48 #include "init.h"
49 #include "irq_user.h"
50 #include "irq_kern.h"
51 #include "ubd_user.h"
52 #include "2_5compat.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, UBD_MMAP };
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 map_fd;
72 	unsigned long long map_offset;
73 	int error;
74 };
75 
76 extern int open_ubd_file(char *file, struct openflags *openflags,
77 			 char **backing_file_out, int *bitmap_offset_out,
78 			 unsigned long *bitmap_len_out, int *data_offset_out,
79 			 int *create_cow_out);
80 extern int create_cow_file(char *cow_file, char *backing_file,
81 			   struct openflags flags, int sectorsize,
82 			   int alignment, int *bitmap_offset_out,
83 			   unsigned long *bitmap_len_out,
84 			   int *data_offset_out);
85 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
86 extern void do_io(struct io_thread_req *req);
87 
88 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
89 {
90 	__u64 n;
91 	int bits, off;
92 
93 	bits = sizeof(data[0]) * 8;
94 	n = bit / bits;
95 	off = bit % bits;
96 	return((data[n] & (1 << off)) != 0);
97 }
98 
99 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
100 {
101 	__u64 n;
102 	int bits, off;
103 
104 	bits = sizeof(data[0]) * 8;
105 	n = bit / bits;
106 	off = bit % bits;
107 	data[n] |= (1 << off);
108 }
109 /*End stuff from ubd_user.h*/
110 
111 #define DRIVER_NAME "uml-blkdev"
112 
113 static DEFINE_SPINLOCK(ubd_io_lock);
114 static DEFINE_SPINLOCK(ubd_lock);
115 
116 static void (*do_ubd)(void);
117 
118 static int ubd_open(struct inode * inode, struct file * filp);
119 static int ubd_release(struct inode * inode, struct file * file);
120 static int ubd_ioctl(struct inode * inode, struct file * file,
121 		     unsigned int cmd, unsigned long arg);
122 
123 #define MAX_DEV (8)
124 
125 /* Changed in early boot */
126 static int ubd_do_mmap = 0;
127 #define UBD_MMAP_BLOCK_SIZE PAGE_SIZE
128 
129 static struct block_device_operations ubd_blops = {
130         .owner		= THIS_MODULE,
131         .open		= ubd_open,
132         .release	= ubd_release,
133         .ioctl		= ubd_ioctl,
134 };
135 
136 /* Protected by the queue_lock */
137 static request_queue_t *ubd_queue;
138 
139 /* Protected by ubd_lock */
140 static int fake_major = MAJOR_NR;
141 
142 static struct gendisk *ubd_gendisk[MAX_DEV];
143 static struct gendisk *fake_gendisk[MAX_DEV];
144 
145 #ifdef CONFIG_BLK_DEV_UBD_SYNC
146 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
147 					 .cl = 1 })
148 #else
149 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
150 					 .cl = 1 })
151 #endif
152 
153 /* Not protected - changed only in ubd_setup_common and then only to
154  * to enable O_SYNC.
155  */
156 static struct openflags global_openflags = OPEN_FLAGS;
157 
158 struct cow {
159 	/* This is the backing file, actually */
160 	char *file;
161 	int fd;
162 	unsigned long *bitmap;
163 	unsigned long bitmap_len;
164 	int bitmap_offset;
165         int data_offset;
166 };
167 
168 struct ubd {
169 	char *file;
170 	int count;
171 	int fd;
172 	__u64 size;
173 	struct openflags boot_openflags;
174 	struct openflags openflags;
175 	int no_cow;
176 	struct cow cow;
177 	struct platform_device pdev;
178 
179 	int map_writes;
180 	int map_reads;
181 	int nomap_writes;
182 	int nomap_reads;
183 	int write_maps;
184 };
185 
186 #define DEFAULT_COW { \
187 	.file =			NULL, \
188         .fd =			-1, \
189         .bitmap =		NULL, \
190 	.bitmap_offset =	0, \
191         .data_offset =		0, \
192 }
193 
194 #define DEFAULT_UBD { \
195 	.file = 		NULL, \
196 	.count =		0, \
197 	.fd =			-1, \
198 	.size =			-1, \
199 	.boot_openflags =	OPEN_FLAGS, \
200 	.openflags =		OPEN_FLAGS, \
201         .no_cow =               0, \
202         .cow =			DEFAULT_COW, \
203 	.map_writes		= 0, \
204 	.map_reads		= 0, \
205 	.nomap_writes		= 0, \
206 	.nomap_reads		= 0, \
207 	.write_maps		= 0, \
208 }
209 
210 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
211 
212 static int ubd0_init(void)
213 {
214 	struct ubd *dev = &ubd_dev[0];
215 
216 	if(dev->file == NULL)
217 		dev->file = "root_fs";
218 	return(0);
219 }
220 
221 __initcall(ubd0_init);
222 
223 /* Only changed by fake_ide_setup which is a setup */
224 static int fake_ide = 0;
225 static struct proc_dir_entry *proc_ide_root = NULL;
226 static struct proc_dir_entry *proc_ide = NULL;
227 
228 static void make_proc_ide(void)
229 {
230 	proc_ide_root = proc_mkdir("ide", NULL);
231 	proc_ide = proc_mkdir("ide0", proc_ide_root);
232 }
233 
234 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
235 			       int *eof, void *data)
236 {
237 	int len;
238 
239 	strcpy(page, "disk\n");
240 	len = strlen("disk\n");
241 	len -= off;
242 	if (len < count){
243 		*eof = 1;
244 		if (len <= 0) return 0;
245 	}
246 	else len = count;
247 	*start = page + off;
248 	return len;
249 }
250 
251 static void make_ide_entries(char *dev_name)
252 {
253 	struct proc_dir_entry *dir, *ent;
254 	char name[64];
255 
256 	if(proc_ide_root == NULL) make_proc_ide();
257 
258 	dir = proc_mkdir(dev_name, proc_ide);
259 	if(!dir) return;
260 
261 	ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
262 	if(!ent) return;
263 	ent->nlink = 1;
264 	ent->data = NULL;
265 	ent->read_proc = proc_ide_read_media;
266 	ent->write_proc = NULL;
267 	sprintf(name,"ide0/%s", dev_name);
268 	proc_symlink(dev_name, proc_ide_root, name);
269 }
270 
271 static int fake_ide_setup(char *str)
272 {
273 	fake_ide = 1;
274 	return(1);
275 }
276 
277 __setup("fake_ide", fake_ide_setup);
278 
279 __uml_help(fake_ide_setup,
280 "fake_ide\n"
281 "    Create ide0 entries that map onto ubd devices.\n\n"
282 );
283 
284 static int parse_unit(char **ptr)
285 {
286 	char *str = *ptr, *end;
287 	int n = -1;
288 
289 	if(isdigit(*str)) {
290 		n = simple_strtoul(str, &end, 0);
291 		if(end == str)
292 			return(-1);
293 		*ptr = end;
294 	}
295 	else if (('a' <= *str) && (*str <= 'h')) {
296 		n = *str - 'a';
297 		str++;
298 		*ptr = str;
299 	}
300 	return(n);
301 }
302 
303 static int ubd_setup_common(char *str, int *index_out)
304 {
305 	struct ubd *dev;
306 	struct openflags flags = global_openflags;
307 	char *backing_file;
308 	int n, err, i;
309 
310 	if(index_out) *index_out = -1;
311 	n = *str;
312 	if(n == '='){
313 		char *end;
314 		int major;
315 
316 		str++;
317 		if(!strcmp(str, "mmap")){
318 			CHOOSE_MODE(printk("mmap not supported by the ubd "
319 					   "driver in tt mode\n"),
320 				    ubd_do_mmap = 1);
321 			return(0);
322 		}
323 
324 		if(!strcmp(str, "sync")){
325 			global_openflags = of_sync(global_openflags);
326 			return(0);
327 		}
328 		major = simple_strtoul(str, &end, 0);
329 		if((*end != '\0') || (end == str)){
330 			printk(KERN_ERR
331 			       "ubd_setup : didn't parse major number\n");
332 			return(1);
333 		}
334 
335 		err = 1;
336  		spin_lock(&ubd_lock);
337  		if(fake_major != MAJOR_NR){
338  			printk(KERN_ERR "Can't assign a fake major twice\n");
339  			goto out1;
340  		}
341 
342  		fake_major = major;
343 
344 		printk(KERN_INFO "Setting extra ubd major number to %d\n",
345 		       major);
346  		err = 0;
347  	out1:
348  		spin_unlock(&ubd_lock);
349 		return(err);
350 	}
351 
352 	n = parse_unit(&str);
353 	if(n < 0){
354 		printk(KERN_ERR "ubd_setup : couldn't parse unit number "
355 		       "'%s'\n", str);
356 		return(1);
357 	}
358 	if(n >= MAX_DEV){
359 		printk(KERN_ERR "ubd_setup : index %d out of range "
360 		       "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
361 		return(1);
362 	}
363 
364 	err = 1;
365 	spin_lock(&ubd_lock);
366 
367 	dev = &ubd_dev[n];
368 	if(dev->file != NULL){
369 		printk(KERN_ERR "ubd_setup : device already configured\n");
370 		goto out;
371 	}
372 
373 	if (index_out)
374 		*index_out = n;
375 
376 	for (i = 0; i < 4; i++) {
377 		switch (*str) {
378 		case 'r':
379 			flags.w = 0;
380 			break;
381 		case 's':
382 			flags.s = 1;
383 			break;
384 		case 'd':
385 			dev->no_cow = 1;
386 			break;
387 		case '=':
388 			str++;
389 			goto break_loop;
390 		default:
391 			printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r,s or d)\n");
392 			goto out;
393 		}
394 		str++;
395 	}
396 
397         if (*str == '=')
398 		printk(KERN_ERR "ubd_setup : Too many flags specified\n");
399         else
400 		printk(KERN_ERR "ubd_setup : Expected '='\n");
401 	goto out;
402 
403 break_loop:
404 	err = 0;
405 	backing_file = strchr(str, ',');
406 
407 	if (!backing_file) {
408 		backing_file = strchr(str, ':');
409 	}
410 
411 	if(backing_file){
412 		if(dev->no_cow)
413 			printk(KERN_ERR "Can't specify both 'd' and a "
414 			       "cow file\n");
415 		else {
416 			*backing_file = '\0';
417 			backing_file++;
418 		}
419 	}
420 	dev->file = str;
421 	dev->cow.file = backing_file;
422 	dev->boot_openflags = flags;
423 out:
424 	spin_unlock(&ubd_lock);
425 	return(err);
426 }
427 
428 static int ubd_setup(char *str)
429 {
430 	ubd_setup_common(str, NULL);
431 	return(1);
432 }
433 
434 __setup("ubd", ubd_setup);
435 __uml_help(ubd_setup,
436 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
437 "    This is used to associate a device with a file in the underlying\n"
438 "    filesystem. When specifying two filenames, the first one is the\n"
439 "    COW name and the second is the backing file name. As separator you can\n"
440 "    use either a ':' or a ',': the first one allows writing things like;\n"
441 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
442 "    while with a ',' the shell would not expand the 2nd '~'.\n"
443 "    When using only one filename, UML will detect whether to thread it like\n"
444 "    a COW file or a backing file. To override this detection, add the 'd'\n"
445 "    flag:\n"
446 "	ubd0d=BackingFile\n"
447 "    Usually, there is a filesystem in the file, but \n"
448 "    that's not required. Swap devices containing swap files can be\n"
449 "    specified like this. Also, a file which doesn't contain a\n"
450 "    filesystem can have its contents read in the virtual \n"
451 "    machine by running 'dd' on the device. <n> must be in the range\n"
452 "    0 to 7. Appending an 'r' to the number will cause that device\n"
453 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
454 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
455 );
456 
457 static int udb_setup(char *str)
458 {
459 	printk("udb%s specified on command line is almost certainly a ubd -> "
460 	       "udb TYPO\n", str);
461 	return(1);
462 }
463 
464 __setup("udb", udb_setup);
465 __uml_help(udb_setup,
466 "udb\n"
467 "    This option is here solely to catch ubd -> udb typos, which can be\n\n"
468 "    to impossible to catch visually unless you specifically look for\n\n"
469 "    them.  The only result of any option starting with 'udb' is an error\n\n"
470 "    in the boot output.\n\n"
471 );
472 
473 static int fakehd_set = 0;
474 static int fakehd(char *str)
475 {
476 	printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
477 	fakehd_set = 1;
478 	return 1;
479 }
480 
481 __setup("fakehd", fakehd);
482 __uml_help(fakehd,
483 "fakehd\n"
484 "    Change the ubd device name to \"hd\".\n\n"
485 );
486 
487 static void do_ubd_request(request_queue_t * q);
488 
489 /* Only changed by ubd_init, which is an initcall. */
490 int thread_fd = -1;
491 
492 /* Changed by ubd_handler, which is serialized because interrupts only
493  * happen on CPU 0.
494  */
495 int intr_count = 0;
496 
497 /* call ubd_finish if you need to serialize */
498 static void __ubd_finish(struct request *req, int error)
499 {
500 	int nsect;
501 
502 	if(error){
503 		end_request(req, 0);
504 		return;
505 	}
506 	nsect = req->current_nr_sectors;
507 	req->sector += nsect;
508 	req->buffer += nsect << 9;
509 	req->errors = 0;
510 	req->nr_sectors -= nsect;
511 	req->current_nr_sectors = 0;
512 	end_request(req, 1);
513 }
514 
515 static inline void ubd_finish(struct request *req, int error)
516 {
517  	spin_lock(&ubd_io_lock);
518 	__ubd_finish(req, error);
519 	spin_unlock(&ubd_io_lock);
520 }
521 
522 /* Called without ubd_io_lock held */
523 static void ubd_handler(void)
524 {
525 	struct io_thread_req req;
526 	struct request *rq = elv_next_request(ubd_queue);
527 	int n, err;
528 
529 	do_ubd = NULL;
530 	intr_count++;
531 	n = os_read_file(thread_fd, &req, sizeof(req));
532 	if(n != sizeof(req)){
533 		printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
534 		       "err = %d\n", os_getpid(), -n);
535 		spin_lock(&ubd_io_lock);
536 		end_request(rq, 0);
537 		spin_unlock(&ubd_io_lock);
538 		return;
539 	}
540 
541 	if((req.op != UBD_MMAP) &&
542 	   ((req.offset != ((__u64) (rq->sector)) << 9) ||
543 	    (req.length != (rq->current_nr_sectors) << 9)))
544 		panic("I/O op mismatch");
545 
546 	if(req.map_fd != -1){
547 		err = physmem_subst_mapping(req.buffer, req.map_fd,
548 					    req.map_offset, 1);
549 		if(err)
550 			printk("ubd_handler - physmem_subst_mapping failed, "
551 			       "err = %d\n", -err);
552 	}
553 
554 	ubd_finish(rq, req.error);
555 	reactivate_fd(thread_fd, UBD_IRQ);
556 	do_ubd_request(ubd_queue);
557 }
558 
559 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
560 {
561 	ubd_handler();
562 	return(IRQ_HANDLED);
563 }
564 
565 /* Only changed by ubd_init, which is an initcall. */
566 static int io_pid = -1;
567 
568 void kill_io_thread(void)
569 {
570 	if(io_pid != -1)
571 		os_kill_process(io_pid, 1);
572 }
573 
574 __uml_exitcall(kill_io_thread);
575 
576 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
577 {
578 	char *file;
579 
580 	file = dev->cow.file ? dev->cow.file : dev->file;
581 	return(os_file_size(file, size_out));
582 }
583 
584 static void ubd_close(struct ubd *dev)
585 {
586 	if(ubd_do_mmap)
587 		physmem_forget_descriptor(dev->fd);
588 	os_close_file(dev->fd);
589 	if(dev->cow.file == NULL)
590 		return;
591 
592 	if(ubd_do_mmap)
593 		physmem_forget_descriptor(dev->cow.fd);
594 	os_close_file(dev->cow.fd);
595 	vfree(dev->cow.bitmap);
596 	dev->cow.bitmap = NULL;
597 }
598 
599 static int ubd_open_dev(struct ubd *dev)
600 {
601 	struct openflags flags;
602 	char **back_ptr;
603 	int err, create_cow, *create_ptr;
604 
605 	dev->openflags = dev->boot_openflags;
606 	create_cow = 0;
607 	create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
608 	back_ptr = dev->no_cow ? NULL : &dev->cow.file;
609 	dev->fd = open_ubd_file(dev->file, &dev->openflags, back_ptr,
610 				&dev->cow.bitmap_offset, &dev->cow.bitmap_len,
611 				&dev->cow.data_offset, create_ptr);
612 
613 	if((dev->fd == -ENOENT) && create_cow){
614 		dev->fd = create_cow_file(dev->file, dev->cow.file,
615 					  dev->openflags, 1 << 9, PAGE_SIZE,
616 					  &dev->cow.bitmap_offset,
617 					  &dev->cow.bitmap_len,
618 					  &dev->cow.data_offset);
619 		if(dev->fd >= 0){
620 			printk(KERN_INFO "Creating \"%s\" as COW file for "
621 			       "\"%s\"\n", dev->file, dev->cow.file);
622 		}
623 	}
624 
625 	if(dev->fd < 0){
626 		printk("Failed to open '%s', errno = %d\n", dev->file,
627 		       -dev->fd);
628 		return(dev->fd);
629 	}
630 
631 	if(dev->cow.file != NULL){
632 		err = -ENOMEM;
633 		dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
634 		if(dev->cow.bitmap == NULL){
635 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
636 			goto error;
637 		}
638 		flush_tlb_kernel_vm();
639 
640 		err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
641 				      dev->cow.bitmap_offset,
642 				      dev->cow.bitmap_len);
643 		if(err < 0)
644 			goto error;
645 
646 		flags = dev->openflags;
647 		flags.w = 0;
648 		err = open_ubd_file(dev->cow.file, &flags, NULL, NULL, NULL,
649 				    NULL, NULL);
650 		if(err < 0) goto error;
651 		dev->cow.fd = err;
652 	}
653 	return(0);
654  error:
655 	os_close_file(dev->fd);
656 	return(err);
657 }
658 
659 static int ubd_new_disk(int major, u64 size, int unit,
660 			struct gendisk **disk_out)
661 
662 {
663 	struct gendisk *disk;
664 	char from[sizeof("ubd/nnnnn\0")], to[sizeof("discnnnnn/disc\0")];
665 	int err;
666 
667 	disk = alloc_disk(1 << UBD_SHIFT);
668 	if(disk == NULL)
669 		return(-ENOMEM);
670 
671 	disk->major = major;
672 	disk->first_minor = unit << UBD_SHIFT;
673 	disk->fops = &ubd_blops;
674 	set_capacity(disk, size / 512);
675 	if(major == MAJOR_NR){
676 		sprintf(disk->disk_name, "ubd%c", 'a' + unit);
677 		sprintf(disk->devfs_name, "ubd/disc%d", unit);
678 		sprintf(from, "ubd/%d", unit);
679 		sprintf(to, "disc%d/disc", unit);
680 		err = devfs_mk_symlink(from, to);
681 		if(err)
682 			printk("ubd_new_disk failed to make link from %s to "
683 			       "%s, error = %d\n", from, to, err);
684 	}
685 	else {
686 		sprintf(disk->disk_name, "ubd_fake%d", unit);
687 		sprintf(disk->devfs_name, "ubd_fake/disc%d", unit);
688 	}
689 
690 	/* sysfs register (not for ide fake devices) */
691 	if (major == MAJOR_NR) {
692 		ubd_dev[unit].pdev.id   = unit;
693 		ubd_dev[unit].pdev.name = DRIVER_NAME;
694 		platform_device_register(&ubd_dev[unit].pdev);
695 		disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
696 	}
697 
698 	disk->private_data = &ubd_dev[unit];
699 	disk->queue = ubd_queue;
700 	add_disk(disk);
701 
702 	*disk_out = disk;
703 	return 0;
704 }
705 
706 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
707 
708 static int ubd_add(int n)
709 {
710 	struct ubd *dev = &ubd_dev[n];
711 	int err;
712 
713 	if(dev->file == NULL)
714 		return(-ENODEV);
715 
716 	if (ubd_open_dev(dev))
717 		return(-ENODEV);
718 
719 	err = ubd_file_size(dev, &dev->size);
720 	if(err < 0)
721 		return(err);
722 
723 	dev->size = ROUND_BLOCK(dev->size);
724 
725 	err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
726 	if(err)
727 		return(err);
728 
729 	if(fake_major != MAJOR_NR)
730 		ubd_new_disk(fake_major, dev->size, n,
731 			     &fake_gendisk[n]);
732 
733 	/* perhaps this should also be under the "if (fake_major)" above */
734 	/* using the fake_disk->disk_name and also the fakehd_set name */
735 	if (fake_ide)
736 		make_ide_entries(ubd_gendisk[n]->disk_name);
737 
738 	ubd_close(dev);
739 	return 0;
740 }
741 
742 static int ubd_config(char *str)
743 {
744 	int n, err;
745 
746 	str = uml_strdup(str);
747 	if(str == NULL){
748 		printk(KERN_ERR "ubd_config failed to strdup string\n");
749 		return(1);
750 	}
751 	err = ubd_setup_common(str, &n);
752 	if(err){
753 		kfree(str);
754 		return(-1);
755 	}
756 	if(n == -1) return(0);
757 
758  	spin_lock(&ubd_lock);
759 	err = ubd_add(n);
760 	if(err)
761 		ubd_dev[n].file = NULL;
762  	spin_unlock(&ubd_lock);
763 
764 	return(err);
765 }
766 
767 static int ubd_get_config(char *name, char *str, int size, char **error_out)
768 {
769 	struct ubd *dev;
770 	int n, len = 0;
771 
772 	n = parse_unit(&name);
773 	if((n >= MAX_DEV) || (n < 0)){
774 		*error_out = "ubd_get_config : device number out of range";
775 		return(-1);
776 	}
777 
778 	dev = &ubd_dev[n];
779 	spin_lock(&ubd_lock);
780 
781 	if(dev->file == NULL){
782 		CONFIG_CHUNK(str, size, len, "", 1);
783 		goto out;
784 	}
785 
786 	CONFIG_CHUNK(str, size, len, dev->file, 0);
787 
788 	if(dev->cow.file != NULL){
789 		CONFIG_CHUNK(str, size, len, ",", 0);
790 		CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
791 	}
792 	else CONFIG_CHUNK(str, size, len, "", 1);
793 
794  out:
795 	spin_unlock(&ubd_lock);
796 	return(len);
797 }
798 
799 static int ubd_remove(char *str)
800 {
801 	struct ubd *dev;
802 	int n, err = -ENODEV;
803 
804 	n = parse_unit(&str);
805 
806 	if((n < 0) || (n >= MAX_DEV))
807 		return(err);
808 
809 	dev = &ubd_dev[n];
810 	if(dev->count > 0)
811 		return(-EBUSY);	/* you cannot remove a open disk */
812 
813 	err = 0;
814  	spin_lock(&ubd_lock);
815 
816 	if(ubd_gendisk[n] == NULL)
817 		goto out;
818 
819 	del_gendisk(ubd_gendisk[n]);
820 	put_disk(ubd_gendisk[n]);
821 	ubd_gendisk[n] = NULL;
822 
823 	if(fake_gendisk[n] != NULL){
824 		del_gendisk(fake_gendisk[n]);
825 		put_disk(fake_gendisk[n]);
826 		fake_gendisk[n] = NULL;
827 	}
828 
829 	platform_device_unregister(&dev->pdev);
830 	*dev = ((struct ubd) DEFAULT_UBD);
831 	err = 0;
832  out:
833  	spin_unlock(&ubd_lock);
834 	return(err);
835 }
836 
837 static struct mc_device ubd_mc = {
838 	.name		= "ubd",
839 	.config		= ubd_config,
840  	.get_config	= ubd_get_config,
841 	.remove		= ubd_remove,
842 };
843 
844 static int ubd_mc_init(void)
845 {
846 	mconsole_register_dev(&ubd_mc);
847 	return 0;
848 }
849 
850 __initcall(ubd_mc_init);
851 
852 static struct device_driver ubd_driver = {
853 	.name  = DRIVER_NAME,
854 	.bus   = &platform_bus_type,
855 };
856 
857 int ubd_init(void)
858 {
859         int i;
860 
861 	devfs_mk_dir("ubd");
862 	if (register_blkdev(MAJOR_NR, "ubd"))
863 		return -1;
864 
865 	ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
866 	if (!ubd_queue) {
867 		unregister_blkdev(MAJOR_NR, "ubd");
868 		return -1;
869 	}
870 
871 	if (fake_major != MAJOR_NR) {
872 		char name[sizeof("ubd_nnn\0")];
873 
874 		snprintf(name, sizeof(name), "ubd_%d", fake_major);
875 		devfs_mk_dir(name);
876 		if (register_blkdev(fake_major, "ubd"))
877 			return -1;
878 	}
879 	driver_register(&ubd_driver);
880 	for (i = 0; i < MAX_DEV; i++)
881 		ubd_add(i);
882 	return 0;
883 }
884 
885 late_initcall(ubd_init);
886 
887 int ubd_driver_init(void){
888 	unsigned long stack;
889 	int err;
890 
891 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
892 	if(global_openflags.s){
893 		printk(KERN_INFO "ubd: Synchronous mode\n");
894 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
895 		 * enough. So use anyway the io thread. */
896 	}
897 	stack = alloc_stack(0, 0);
898 	io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
899 				 &thread_fd);
900 	if(io_pid < 0){
901 		printk(KERN_ERR
902 		       "ubd : Failed to start I/O thread (errno = %d) - "
903 		       "falling back to synchronous I/O\n", -io_pid);
904 		io_pid = -1;
905 		return(0);
906 	}
907 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
908 			     SA_INTERRUPT, "ubd", ubd_dev);
909 	if(err != 0)
910 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
911 	return(err);
912 }
913 
914 device_initcall(ubd_driver_init);
915 
916 static int ubd_open(struct inode *inode, struct file *filp)
917 {
918 	struct gendisk *disk = inode->i_bdev->bd_disk;
919 	struct ubd *dev = disk->private_data;
920 	int err = 0;
921 
922 	if(dev->count == 0){
923 		err = ubd_open_dev(dev);
924 		if(err){
925 			printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
926 			       disk->disk_name, dev->file, -err);
927 			goto out;
928 		}
929 	}
930 	dev->count++;
931 	set_disk_ro(disk, !dev->openflags.w);
932 
933 	/* This should no more be needed. And it didn't work anyway to exclude
934 	 * read-write remounting of filesystems.*/
935 	/*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
936 	        if(--dev->count == 0) ubd_close(dev);
937 	        err = -EROFS;
938 	}*/
939  out:
940 	return(err);
941 }
942 
943 static int ubd_release(struct inode * inode, struct file * file)
944 {
945 	struct gendisk *disk = inode->i_bdev->bd_disk;
946 	struct ubd *dev = disk->private_data;
947 
948 	if(--dev->count == 0)
949 		ubd_close(dev);
950 	return(0);
951 }
952 
953 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
954 			  __u64 *cow_offset, unsigned long *bitmap,
955 			  __u64 bitmap_offset, unsigned long *bitmap_words,
956 			  __u64 bitmap_len)
957 {
958 	__u64 sector = io_offset >> 9;
959 	int i, update_bitmap = 0;
960 
961 	for(i = 0; i < length >> 9; i++){
962 		if(cow_mask != NULL)
963 			ubd_set_bit(i, (unsigned char *) cow_mask);
964 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
965 			continue;
966 
967 		update_bitmap = 1;
968 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
969 	}
970 
971 	if(!update_bitmap)
972 		return;
973 
974 	*cow_offset = sector / (sizeof(unsigned long) * 8);
975 
976 	/* This takes care of the case where we're exactly at the end of the
977 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
978 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
979 	 * for the original diagnosis.
980 	 */
981 	if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
982 			   sizeof(unsigned long) - 1))
983 		(*cow_offset)--;
984 
985 	bitmap_words[0] = bitmap[*cow_offset];
986 	bitmap_words[1] = bitmap[*cow_offset + 1];
987 
988 	*cow_offset *= sizeof(unsigned long);
989 	*cow_offset += bitmap_offset;
990 }
991 
992 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
993 		       __u64 bitmap_offset, __u64 bitmap_len)
994 {
995 	__u64 sector = req->offset >> 9;
996 	int i;
997 
998 	if(req->length > (sizeof(req->sector_mask) * 8) << 9)
999 		panic("Operation too long");
1000 
1001 	if(req->op == UBD_READ) {
1002 		for(i = 0; i < req->length >> 9; i++){
1003 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1004 				ubd_set_bit(i, (unsigned char *)
1005 					    &req->sector_mask);
1006                 }
1007 	}
1008 	else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1009 			   &req->cow_offset, bitmap, bitmap_offset,
1010 			   req->bitmap_words, bitmap_len);
1011 }
1012 
1013 static int mmap_fd(struct request *req, struct ubd *dev, __u64 offset)
1014 {
1015 	__u64 sector;
1016 	unsigned char *bitmap;
1017 	int bit, i;
1018 
1019 	/* mmap must have been requested on the command line */
1020 	if(!ubd_do_mmap)
1021 		return(-1);
1022 
1023 	/* The buffer must be page aligned */
1024 	if(((unsigned long) req->buffer % UBD_MMAP_BLOCK_SIZE) != 0)
1025 		return(-1);
1026 
1027 	/* The request must be a page long */
1028 	if((req->current_nr_sectors << 9) != PAGE_SIZE)
1029 		return(-1);
1030 
1031 	if(dev->cow.file == NULL)
1032 		return(dev->fd);
1033 
1034 	sector = offset >> 9;
1035 	bitmap = (unsigned char *) dev->cow.bitmap;
1036 	bit = ubd_test_bit(sector, bitmap);
1037 
1038 	for(i = 1; i < req->current_nr_sectors; i++){
1039 		if(ubd_test_bit(sector + i, bitmap) != bit)
1040 			return(-1);
1041 	}
1042 
1043 	if(bit || (rq_data_dir(req) == WRITE))
1044 		offset += dev->cow.data_offset;
1045 
1046 	/* The data on disk must be page aligned */
1047 	if((offset % UBD_MMAP_BLOCK_SIZE) != 0)
1048 		return(-1);
1049 
1050 	return(bit ? dev->fd : dev->cow.fd);
1051 }
1052 
1053 static int prepare_mmap_request(struct ubd *dev, int fd, __u64 offset,
1054 				struct request *req,
1055 				struct io_thread_req *io_req)
1056 {
1057 	int err;
1058 
1059 	if(rq_data_dir(req) == WRITE){
1060 		/* Writes are almost no-ops since the new data is already in the
1061 		 * host page cache
1062 		 */
1063 		dev->map_writes++;
1064 		if(dev->cow.file != NULL)
1065 			cowify_bitmap(io_req->offset, io_req->length,
1066 				      &io_req->sector_mask, &io_req->cow_offset,
1067 				      dev->cow.bitmap, dev->cow.bitmap_offset,
1068 				      io_req->bitmap_words,
1069 				      dev->cow.bitmap_len);
1070 	}
1071 	else {
1072 		int w;
1073 
1074 		if((dev->cow.file != NULL) && (fd == dev->cow.fd))
1075 			w = 0;
1076 		else w = dev->openflags.w;
1077 
1078 		if((dev->cow.file != NULL) && (fd == dev->fd))
1079 			offset += dev->cow.data_offset;
1080 
1081 		err = physmem_subst_mapping(req->buffer, fd, offset, w);
1082 		if(err){
1083 			printk("physmem_subst_mapping failed, err = %d\n",
1084 			       -err);
1085 			return(1);
1086 		}
1087 		dev->map_reads++;
1088 	}
1089 	io_req->op = UBD_MMAP;
1090 	io_req->buffer = req->buffer;
1091 	return(0);
1092 }
1093 
1094 /* Called with ubd_io_lock held */
1095 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1096 {
1097 	struct gendisk *disk = req->rq_disk;
1098 	struct ubd *dev = disk->private_data;
1099 	__u64 offset;
1100 	int len, fd;
1101 
1102 	if(req->rq_status == RQ_INACTIVE) return(1);
1103 
1104 	/* This should be impossible now */
1105 	if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
1106 		printk("Write attempted on readonly ubd device %s\n",
1107 		       disk->disk_name);
1108 		end_request(req, 0);
1109 		return(1);
1110 	}
1111 
1112 	offset = ((__u64) req->sector) << 9;
1113 	len = req->current_nr_sectors << 9;
1114 
1115 	io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1116 	io_req->fds[1] = dev->fd;
1117 	io_req->map_fd = -1;
1118 	io_req->cow_offset = -1;
1119 	io_req->offset = offset;
1120 	io_req->length = len;
1121 	io_req->error = 0;
1122 	io_req->sector_mask = 0;
1123 
1124 	fd = mmap_fd(req, dev, io_req->offset);
1125 	if(fd > 0){
1126 		/* If mmapping is otherwise OK, but the first access to the
1127 		 * page is a write, then it's not mapped in yet.  So we have
1128 		 * to write the data to disk first, then we can map the disk
1129 		 * page in and continue normally from there.
1130 		 */
1131 		if((rq_data_dir(req) == WRITE) && !is_remapped(req->buffer)){
1132 			io_req->map_fd = dev->fd;
1133 			io_req->map_offset = io_req->offset +
1134 				dev->cow.data_offset;
1135 			dev->write_maps++;
1136 		}
1137 		else return(prepare_mmap_request(dev, fd, io_req->offset, req,
1138 						 io_req));
1139 	}
1140 
1141 	if(rq_data_dir(req) == READ)
1142 		dev->nomap_reads++;
1143 	else dev->nomap_writes++;
1144 
1145 	io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1146 	io_req->offsets[0] = 0;
1147 	io_req->offsets[1] = dev->cow.data_offset;
1148 	io_req->buffer = req->buffer;
1149 	io_req->sectorsize = 1 << 9;
1150 
1151 	if(dev->cow.file != NULL)
1152 		cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1153 			   dev->cow.bitmap_len);
1154 
1155 	return(0);
1156 }
1157 
1158 /* Called with ubd_io_lock held */
1159 static void do_ubd_request(request_queue_t *q)
1160 {
1161 	struct io_thread_req io_req;
1162 	struct request *req;
1163 	int err, n;
1164 
1165 	if(thread_fd == -1){
1166 		while((req = elv_next_request(q)) != NULL){
1167 			err = prepare_request(req, &io_req);
1168 			if(!err){
1169 				do_io(&io_req);
1170 				__ubd_finish(req, io_req.error);
1171 			}
1172 		}
1173 	}
1174 	else {
1175 		if(do_ubd || (req = elv_next_request(q)) == NULL)
1176 			return;
1177 		err = prepare_request(req, &io_req);
1178 		if(!err){
1179 			do_ubd = ubd_handler;
1180 			n = os_write_file(thread_fd, (char *) &io_req,
1181 					 sizeof(io_req));
1182 			if(n != sizeof(io_req))
1183 				printk("write to io thread failed, "
1184 				       "errno = %d\n", -n);
1185 		}
1186 	}
1187 }
1188 
1189 static int ubd_ioctl(struct inode * inode, struct file * file,
1190 		     unsigned int cmd, unsigned long arg)
1191 {
1192 	struct hd_geometry __user *loc = (struct hd_geometry __user *) arg;
1193 	struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1194 	struct hd_driveid ubd_id = {
1195 		.cyls		= 0,
1196 		.heads		= 128,
1197 		.sectors	= 32,
1198 	};
1199 
1200 	switch (cmd) {
1201 	        struct hd_geometry g;
1202 		struct cdrom_volctrl volume;
1203 	case HDIO_GETGEO:
1204 		if(!loc) return(-EINVAL);
1205 		g.heads = 128;
1206 		g.sectors = 32;
1207 		g.cylinders = dev->size / (128 * 32 * 512);
1208 		g.start = get_start_sect(inode->i_bdev);
1209 		return(copy_to_user(loc, &g, sizeof(g)) ? -EFAULT : 0);
1210 
1211 	case HDIO_GET_IDENTITY:
1212 		ubd_id.cyls = dev->size / (128 * 32 * 512);
1213 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1214 				 sizeof(ubd_id)))
1215 			return(-EFAULT);
1216 		return(0);
1217 
1218 	case CDROMVOLREAD:
1219 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1220 			return(-EFAULT);
1221 		volume.channel0 = 255;
1222 		volume.channel1 = 255;
1223 		volume.channel2 = 255;
1224 		volume.channel3 = 255;
1225 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1226 			return(-EFAULT);
1227 		return(0);
1228 	}
1229 	return(-EINVAL);
1230 }
1231 
1232 static int ubd_check_remapped(int fd, unsigned long address, int is_write,
1233 			      __u64 offset)
1234 {
1235 	__u64 bitmap_offset;
1236 	unsigned long new_bitmap[2];
1237 	int i, err, n;
1238 
1239 	/* If it's not a write access, we can't do anything about it */
1240 	if(!is_write)
1241 		return(0);
1242 
1243 	/* We have a write */
1244 	for(i = 0; i < sizeof(ubd_dev) / sizeof(ubd_dev[0]); i++){
1245 		struct ubd *dev = &ubd_dev[i];
1246 
1247 		if((dev->fd != fd) && (dev->cow.fd != fd))
1248 			continue;
1249 
1250 		/* It's a write to a ubd device */
1251 
1252 		/* This should be impossible now */
1253 		if(!dev->openflags.w){
1254 			/* It's a write access on a read-only device - probably
1255 			 * shouldn't happen.  If the kernel is trying to change
1256 			 * something with no intention of writing it back out,
1257 			 * then this message will clue us in that this needs
1258 			 * fixing
1259 			 */
1260 			printk("Write access to mapped page from readonly ubd "
1261 			       "device %d\n", i);
1262 			return(0);
1263 		}
1264 
1265 		/* It's a write to a writeable ubd device - it must be COWed
1266 		 * because, otherwise, the page would have been mapped in
1267 		 * writeable
1268 		 */
1269 
1270 		if(!dev->cow.file)
1271 			panic("Write fault on writeable non-COW ubd device %d",
1272 			      i);
1273 
1274 		/* It should also be an access to the backing file since the
1275 		 * COW pages should be mapped in read-write
1276 		 */
1277 
1278 		if(fd == dev->fd)
1279 			panic("Write fault on a backing page of ubd "
1280 			      "device %d\n", i);
1281 
1282 		/* So, we do the write, copying the backing data to the COW
1283 		 * file...
1284 		 */
1285 
1286 		err = os_seek_file(dev->fd, offset + dev->cow.data_offset);
1287 		if(err < 0)
1288 			panic("Couldn't seek to %lld in COW file of ubd "
1289 			      "device %d, err = %d",
1290 			      offset + dev->cow.data_offset, i, -err);
1291 
1292 		n = os_write_file(dev->fd, (void *) address, PAGE_SIZE);
1293 		if(n != PAGE_SIZE)
1294 			panic("Couldn't copy data to COW file of ubd "
1295 			      "device %d, err = %d", i, -n);
1296 
1297 		/* ... updating the COW bitmap... */
1298 
1299 		cowify_bitmap(offset, PAGE_SIZE, NULL, &bitmap_offset,
1300 			      dev->cow.bitmap, dev->cow.bitmap_offset,
1301 			      new_bitmap, dev->cow.bitmap_len);
1302 
1303 		err = os_seek_file(dev->fd, bitmap_offset);
1304 		if(err < 0)
1305 			panic("Couldn't seek to %lld in COW file of ubd "
1306 			      "device %d, err = %d", bitmap_offset, i, -err);
1307 
1308 		n = os_write_file(dev->fd, new_bitmap, sizeof(new_bitmap));
1309 		if(n != sizeof(new_bitmap))
1310 			panic("Couldn't update bitmap  of ubd device %d, "
1311 			      "err = %d", i, -n);
1312 
1313 		/* Maybe we can map the COW page in, and maybe we can't.  If
1314 		 * it is a pre-V3 COW file, we can't, since the alignment will
1315 		 * be wrong.  If it is a V3 or later COW file which has been
1316 		 * moved to a system with a larger page size, then maybe we
1317 		 * can't, depending on the exact location of the page.
1318 		 */
1319 
1320 		offset += dev->cow.data_offset;
1321 
1322 		/* Remove the remapping, putting the original anonymous page
1323 		 * back.  If the COW file can be mapped in, that is done.
1324 		 * Otherwise, the COW page is read in.
1325 		 */
1326 
1327 		if(!physmem_remove_mapping((void *) address))
1328 			panic("Address 0x%lx not remapped by ubd device %d",
1329 			      address, i);
1330 		if((offset % UBD_MMAP_BLOCK_SIZE) == 0)
1331 			physmem_subst_mapping((void *) address, dev->fd,
1332 					      offset, 1);
1333 		else {
1334 			err = os_seek_file(dev->fd, offset);
1335 			if(err < 0)
1336 				panic("Couldn't seek to %lld in COW file of "
1337 				      "ubd device %d, err = %d", offset, i,
1338 				      -err);
1339 
1340 			n = os_read_file(dev->fd, (void *) address, PAGE_SIZE);
1341 			if(n != PAGE_SIZE)
1342 				panic("Failed to read page from offset %llx of "
1343 				      "COW file of ubd device %d, err = %d",
1344 				      offset, i, -n);
1345 		}
1346 
1347 		return(1);
1348 	}
1349 
1350 	/* It's not a write on a ubd device */
1351 	return(0);
1352 }
1353 
1354 static struct remapper ubd_remapper = {
1355 	.list	= LIST_HEAD_INIT(ubd_remapper.list),
1356 	.proc	= ubd_check_remapped,
1357 };
1358 
1359 static int ubd_remapper_setup(void)
1360 {
1361 	if(ubd_do_mmap)
1362 		register_remapper(&ubd_remapper);
1363 
1364 	return(0);
1365 }
1366 
1367 __initcall(ubd_remapper_setup);
1368 
1369 static int same_backing_files(char *from_cmdline, char *from_cow, char *cow)
1370 {
1371 	struct uml_stat buf1, buf2;
1372 	int err;
1373 
1374 	if(from_cmdline == NULL) return(1);
1375 	if(!strcmp(from_cmdline, from_cow)) return(1);
1376 
1377 	err = os_stat_file(from_cmdline, &buf1);
1378 	if(err < 0){
1379 		printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1380 		return(1);
1381 	}
1382 	err = os_stat_file(from_cow, &buf2);
1383 	if(err < 0){
1384 		printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1385 		return(1);
1386 	}
1387 	if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1388 		return(1);
1389 
1390 	printk("Backing file mismatch - \"%s\" requested,\n"
1391 	       "\"%s\" specified in COW header of \"%s\"\n",
1392 	       from_cmdline, from_cow, cow);
1393 	return(0);
1394 }
1395 
1396 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1397 {
1398 	unsigned long modtime;
1399 	long long actual;
1400 	int err;
1401 
1402 	err = os_file_modtime(file, &modtime);
1403 	if(err < 0){
1404 		printk("Failed to get modification time of backing file "
1405 		       "\"%s\", err = %d\n", file, -err);
1406 		return(err);
1407 	}
1408 
1409 	err = os_file_size(file, &actual);
1410 	if(err < 0){
1411 		printk("Failed to get size of backing file \"%s\", "
1412 		       "err = %d\n", file, -err);
1413 		return(err);
1414 	}
1415 
1416   	if(actual != size){
1417 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
1418 		 * the typecast.*/
1419 		printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1420 		       "file\n", (unsigned long long) size, actual);
1421 		return(-EINVAL);
1422 	}
1423 	if(modtime != mtime){
1424 		printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1425 		       "file\n", mtime, modtime);
1426 		return(-EINVAL);
1427 	}
1428 	return(0);
1429 }
1430 
1431 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1432 {
1433 	int err;
1434 
1435 	err = os_seek_file(fd, offset);
1436 	if(err < 0)
1437 		return(err);
1438 
1439 	err = os_read_file(fd, buf, len);
1440 	if(err < 0)
1441 		return(err);
1442 
1443 	return(0);
1444 }
1445 
1446 int open_ubd_file(char *file, struct openflags *openflags,
1447 		  char **backing_file_out, int *bitmap_offset_out,
1448 		  unsigned long *bitmap_len_out, int *data_offset_out,
1449 		  int *create_cow_out)
1450 {
1451 	time_t mtime;
1452 	unsigned long long size;
1453 	__u32 version, align;
1454 	char *backing_file;
1455 	int fd, err, sectorsize, same, mode = 0644;
1456 
1457 	fd = os_open_file(file, *openflags, mode);
1458 	if(fd < 0){
1459 		if((fd == -ENOENT) && (create_cow_out != NULL))
1460 			*create_cow_out = 1;
1461                 if(!openflags->w ||
1462                    ((fd != -EROFS) && (fd != -EACCES))) return(fd);
1463 		openflags->w = 0;
1464 		fd = os_open_file(file, *openflags, mode);
1465 		if(fd < 0)
1466 			return(fd);
1467         }
1468 
1469 	err = os_lock_file(fd, openflags->w);
1470 	if(err < 0){
1471 		printk("Failed to lock '%s', err = %d\n", file, -err);
1472 		goto out_close;
1473 	}
1474 
1475 	if(backing_file_out == NULL) return(fd);
1476 
1477 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1478 			      &size, &sectorsize, &align, bitmap_offset_out);
1479 	if(err && (*backing_file_out != NULL)){
1480 		printk("Failed to read COW header from COW file \"%s\", "
1481 		       "errno = %d\n", file, -err);
1482 		goto out_close;
1483 	}
1484 	if(err) return(fd);
1485 
1486 	if(backing_file_out == NULL) return(fd);
1487 
1488 	same = same_backing_files(*backing_file_out, backing_file, file);
1489 
1490 	if(!same && !backing_file_mismatch(*backing_file_out, size, mtime)){
1491 		printk("Switching backing file to '%s'\n", *backing_file_out);
1492 		err = write_cow_header(file, fd, *backing_file_out,
1493 				       sectorsize, align, &size);
1494 		if(err){
1495 			printk("Switch failed, errno = %d\n", -err);
1496 			return(err);
1497 		}
1498 	}
1499 	else {
1500 		*backing_file_out = backing_file;
1501 		err = backing_file_mismatch(*backing_file_out, size, mtime);
1502 		if(err) goto out_close;
1503 	}
1504 
1505 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1506 		  bitmap_len_out, data_offset_out);
1507 
1508         return(fd);
1509  out_close:
1510 	os_close_file(fd);
1511 	return(err);
1512 }
1513 
1514 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1515 		    int sectorsize, int alignment, int *bitmap_offset_out,
1516 		    unsigned long *bitmap_len_out, int *data_offset_out)
1517 {
1518 	int err, fd;
1519 
1520 	flags.c = 1;
1521 	fd = open_ubd_file(cow_file, &flags, NULL, NULL, NULL, NULL, NULL);
1522 	if(fd < 0){
1523 		err = fd;
1524 		printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1525 		       -err);
1526 		goto out;
1527 	}
1528 
1529 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1530 			    bitmap_offset_out, bitmap_len_out,
1531 			    data_offset_out);
1532 	if(!err)
1533 		return(fd);
1534 	os_close_file(fd);
1535  out:
1536 	return(err);
1537 }
1538 
1539 static int update_bitmap(struct io_thread_req *req)
1540 {
1541 	int n;
1542 
1543 	if(req->cow_offset == -1)
1544 		return(0);
1545 
1546 	n = os_seek_file(req->fds[1], req->cow_offset);
1547 	if(n < 0){
1548 		printk("do_io - bitmap lseek failed : err = %d\n", -n);
1549 		return(1);
1550 	}
1551 
1552 	n = os_write_file(req->fds[1], &req->bitmap_words,
1553 		          sizeof(req->bitmap_words));
1554 	if(n != sizeof(req->bitmap_words)){
1555 		printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1556 		       req->fds[1]);
1557 		return(1);
1558 	}
1559 
1560 	return(0);
1561 }
1562 
1563 void do_io(struct io_thread_req *req)
1564 {
1565 	char *buf;
1566 	unsigned long len;
1567 	int n, nsectors, start, end, bit;
1568 	int err;
1569 	__u64 off;
1570 
1571 	if(req->op == UBD_MMAP){
1572 		/* Touch the page to force the host to do any necessary IO to
1573 		 * get it into memory
1574 		 */
1575 		n = *((volatile int *) req->buffer);
1576 		req->error = update_bitmap(req);
1577 		return;
1578 	}
1579 
1580 	nsectors = req->length / req->sectorsize;
1581 	start = 0;
1582 	do {
1583 		bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1584 		end = start;
1585 		while((end < nsectors) &&
1586 		      (ubd_test_bit(end, (unsigned char *)
1587 				    &req->sector_mask) == bit))
1588 			end++;
1589 
1590 		off = req->offset + req->offsets[bit] +
1591 			start * req->sectorsize;
1592 		len = (end - start) * req->sectorsize;
1593 		buf = &req->buffer[start * req->sectorsize];
1594 
1595 		err = os_seek_file(req->fds[bit], off);
1596 		if(err < 0){
1597 			printk("do_io - lseek failed : err = %d\n", -err);
1598 			req->error = 1;
1599 			return;
1600 		}
1601 		if(req->op == UBD_READ){
1602 			n = 0;
1603 			do {
1604 				buf = &buf[n];
1605 				len -= n;
1606 				n = os_read_file(req->fds[bit], buf, len);
1607 				if (n < 0) {
1608 					printk("do_io - read failed, err = %d "
1609 					       "fd = %d\n", -n, req->fds[bit]);
1610 					req->error = 1;
1611 					return;
1612 				}
1613 			} while((n < len) && (n != 0));
1614 			if (n < len) memset(&buf[n], 0, len - n);
1615 		} else {
1616 			n = os_write_file(req->fds[bit], buf, len);
1617 			if(n != len){
1618 				printk("do_io - write failed err = %d "
1619 				       "fd = %d\n", -n, req->fds[bit]);
1620 				req->error = 1;
1621 				return;
1622 			}
1623 		}
1624 
1625 		start = end;
1626 	} while(start < nsectors);
1627 
1628 	req->error = update_bitmap(req);
1629 }
1630 
1631 /* Changed in start_io_thread, which is serialized by being called only
1632  * from ubd_init, which is an initcall.
1633  */
1634 int kernel_fd = -1;
1635 
1636 /* Only changed by the io thread */
1637 int io_count = 0;
1638 
1639 int io_thread(void *arg)
1640 {
1641 	struct io_thread_req req;
1642 	int n;
1643 
1644 	ignore_sigwinch_sig();
1645 	while(1){
1646 		n = os_read_file(kernel_fd, &req, sizeof(req));
1647 		if(n != sizeof(req)){
1648 			if(n < 0)
1649 				printk("io_thread - read failed, fd = %d, "
1650 				       "err = %d\n", kernel_fd, -n);
1651 			else {
1652 				printk("io_thread - short read, fd = %d, "
1653 				       "length = %d\n", kernel_fd, n);
1654 			}
1655 			continue;
1656 		}
1657 		io_count++;
1658 		do_io(&req);
1659 		n = os_write_file(kernel_fd, &req, sizeof(req));
1660 		if(n != sizeof(req))
1661 			printk("io_thread - write failed, fd = %d, err = %d\n",
1662 			       kernel_fd, -n);
1663 	}
1664 }
1665 
1666 /*
1667  * Overrides for Emacs so that we follow Linus's tabbing style.
1668  * Emacs will notice this stuff at the end of the file and automatically
1669  * adjust the settings for this buffer only.  This must remain at the end
1670  * of the file.
1671  * ---------------------------------------------------------------------------
1672  * Local variables:
1673  * c-file-style: "linux"
1674  * End:
1675  */
1676