xref: /openbmc/linux/arch/um/drivers/ubd_kern.c (revision 62e59c4e)
1 /*
2  * Copyright (C) 2018 Cambridge Greys Ltd
3  * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
4  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
5  * Licensed under the GPL
6  */
7 
8 /* 2001-09-28...2002-04-17
9  * Partition stuff by James_McMechan@hotmail.com
10  * old style ubd by setting UBD_SHIFT to 0
11  * 2002-09-27...2002-10-18 massive tinkering for 2.5
12  * partitions have changed in 2.5
13  * 2003-01-29 more tinkering for 2.5.59-1
14  * This should now address the sysfs problems and has
15  * the symlink for devfs to allow for booting with
16  * the common /dev/ubd/discX/... names rather than
17  * only /dev/ubdN/discN this version also has lots of
18  * clean ups preparing for ubd-many.
19  * James McMechan
20  */
21 
22 #define UBD_SHIFT 4
23 
24 #include <linux/module.h>
25 #include <linux/init.h>
26 #include <linux/blkdev.h>
27 #include <linux/blk-mq.h>
28 #include <linux/ata.h>
29 #include <linux/hdreg.h>
30 #include <linux/cdrom.h>
31 #include <linux/proc_fs.h>
32 #include <linux/seq_file.h>
33 #include <linux/ctype.h>
34 #include <linux/slab.h>
35 #include <linux/vmalloc.h>
36 #include <linux/platform_device.h>
37 #include <linux/scatterlist.h>
38 #include <asm/tlbflush.h>
39 #include <kern_util.h>
40 #include "mconsole_kern.h"
41 #include <init.h>
42 #include <irq_kern.h>
43 #include "ubd.h"
44 #include <os.h>
45 #include "cow.h"
46 
47 /* Max request size is determined by sector mask - 32K */
48 #define UBD_MAX_REQUEST (8 * sizeof(long))
49 
50 struct io_thread_req {
51 	struct request *req;
52 	int fds[2];
53 	unsigned long offsets[2];
54 	unsigned long long offset;
55 	unsigned long length;
56 	char *buffer;
57 	int sectorsize;
58 	unsigned long sector_mask;
59 	unsigned long long cow_offset;
60 	unsigned long bitmap_words[2];
61 	int error;
62 };
63 
64 
65 static struct io_thread_req * (*irq_req_buffer)[];
66 static struct io_thread_req *irq_remainder;
67 static int irq_remainder_size;
68 
69 static struct io_thread_req * (*io_req_buffer)[];
70 static struct io_thread_req *io_remainder;
71 static int io_remainder_size;
72 
73 
74 
75 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
76 {
77 	__u64 n;
78 	int bits, off;
79 
80 	bits = sizeof(data[0]) * 8;
81 	n = bit / bits;
82 	off = bit % bits;
83 	return (data[n] & (1 << off)) != 0;
84 }
85 
86 static inline void ubd_set_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 	data[n] |= (1 << off);
95 }
96 /*End stuff from ubd_user.h*/
97 
98 #define DRIVER_NAME "uml-blkdev"
99 
100 static DEFINE_MUTEX(ubd_lock);
101 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
102 
103 static int ubd_open(struct block_device *bdev, fmode_t mode);
104 static void ubd_release(struct gendisk *disk, fmode_t mode);
105 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
106 		     unsigned int cmd, unsigned long arg);
107 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
108 
109 #define MAX_DEV (16)
110 
111 static const struct block_device_operations ubd_blops = {
112         .owner		= THIS_MODULE,
113         .open		= ubd_open,
114         .release	= ubd_release,
115         .ioctl		= ubd_ioctl,
116 	.getgeo		= ubd_getgeo,
117 };
118 
119 /* Protected by ubd_lock */
120 static int fake_major = UBD_MAJOR;
121 static struct gendisk *ubd_gendisk[MAX_DEV];
122 static struct gendisk *fake_gendisk[MAX_DEV];
123 
124 #ifdef CONFIG_BLK_DEV_UBD_SYNC
125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
126 					 .cl = 1 })
127 #else
128 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
129 					 .cl = 1 })
130 #endif
131 static struct openflags global_openflags = OPEN_FLAGS;
132 
133 struct cow {
134 	/* backing file name */
135 	char *file;
136 	/* backing file fd */
137 	int fd;
138 	unsigned long *bitmap;
139 	unsigned long bitmap_len;
140 	int bitmap_offset;
141 	int data_offset;
142 };
143 
144 #define MAX_SG 64
145 
146 struct ubd {
147 	/* name (and fd, below) of the file opened for writing, either the
148 	 * backing or the cow file. */
149 	char *file;
150 	int count;
151 	int fd;
152 	__u64 size;
153 	struct openflags boot_openflags;
154 	struct openflags openflags;
155 	unsigned shared:1;
156 	unsigned no_cow:1;
157 	unsigned no_trim:1;
158 	struct cow cow;
159 	struct platform_device pdev;
160 	struct request_queue *queue;
161 	struct blk_mq_tag_set tag_set;
162 	spinlock_t lock;
163 };
164 
165 #define DEFAULT_COW { \
166 	.file =			NULL, \
167 	.fd =			-1,	\
168 	.bitmap =		NULL, \
169 	.bitmap_offset =	0, \
170 	.data_offset =		0, \
171 }
172 
173 #define DEFAULT_UBD { \
174 	.file = 		NULL, \
175 	.count =		0, \
176 	.fd =			-1, \
177 	.size =			-1, \
178 	.boot_openflags =	OPEN_FLAGS, \
179 	.openflags =		OPEN_FLAGS, \
180 	.no_cow =               0, \
181 	.no_trim =		0, \
182 	.shared =		0, \
183 	.cow =			DEFAULT_COW, \
184 	.lock =			__SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
185 }
186 
187 /* Protected by ubd_lock */
188 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
189 
190 /* Only changed by fake_ide_setup which is a setup */
191 static int fake_ide = 0;
192 static struct proc_dir_entry *proc_ide_root = NULL;
193 static struct proc_dir_entry *proc_ide = NULL;
194 
195 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
196 				 const struct blk_mq_queue_data *bd);
197 
198 static void make_proc_ide(void)
199 {
200 	proc_ide_root = proc_mkdir("ide", NULL);
201 	proc_ide = proc_mkdir("ide0", proc_ide_root);
202 }
203 
204 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
205 {
206 	seq_puts(m, "disk\n");
207 	return 0;
208 }
209 
210 static void make_ide_entries(const char *dev_name)
211 {
212 	struct proc_dir_entry *dir, *ent;
213 	char name[64];
214 
215 	if(proc_ide_root == NULL) make_proc_ide();
216 
217 	dir = proc_mkdir(dev_name, proc_ide);
218 	if(!dir) return;
219 
220 	ent = proc_create_single("media", S_IRUGO, dir,
221 			fake_ide_media_proc_show);
222 	if(!ent) return;
223 	snprintf(name, sizeof(name), "ide0/%s", dev_name);
224 	proc_symlink(dev_name, proc_ide_root, name);
225 }
226 
227 static int fake_ide_setup(char *str)
228 {
229 	fake_ide = 1;
230 	return 1;
231 }
232 
233 __setup("fake_ide", fake_ide_setup);
234 
235 __uml_help(fake_ide_setup,
236 "fake_ide\n"
237 "    Create ide0 entries that map onto ubd devices.\n\n"
238 );
239 
240 static int parse_unit(char **ptr)
241 {
242 	char *str = *ptr, *end;
243 	int n = -1;
244 
245 	if(isdigit(*str)) {
246 		n = simple_strtoul(str, &end, 0);
247 		if(end == str)
248 			return -1;
249 		*ptr = end;
250 	}
251 	else if (('a' <= *str) && (*str <= 'z')) {
252 		n = *str - 'a';
253 		str++;
254 		*ptr = str;
255 	}
256 	return n;
257 }
258 
259 /* If *index_out == -1 at exit, the passed option was a general one;
260  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
261  * should not be freed on exit.
262  */
263 static int ubd_setup_common(char *str, int *index_out, char **error_out)
264 {
265 	struct ubd *ubd_dev;
266 	struct openflags flags = global_openflags;
267 	char *backing_file;
268 	int n, err = 0, i;
269 
270 	if(index_out) *index_out = -1;
271 	n = *str;
272 	if(n == '='){
273 		char *end;
274 		int major;
275 
276 		str++;
277 		if(!strcmp(str, "sync")){
278 			global_openflags = of_sync(global_openflags);
279 			goto out1;
280 		}
281 
282 		err = -EINVAL;
283 		major = simple_strtoul(str, &end, 0);
284 		if((*end != '\0') || (end == str)){
285 			*error_out = "Didn't parse major number";
286 			goto out1;
287 		}
288 
289 		mutex_lock(&ubd_lock);
290 		if (fake_major != UBD_MAJOR) {
291 			*error_out = "Can't assign a fake major twice";
292 			goto out1;
293 		}
294 
295 		fake_major = major;
296 
297 		printk(KERN_INFO "Setting extra ubd major number to %d\n",
298 		       major);
299 		err = 0;
300 	out1:
301 		mutex_unlock(&ubd_lock);
302 		return err;
303 	}
304 
305 	n = parse_unit(&str);
306 	if(n < 0){
307 		*error_out = "Couldn't parse device number";
308 		return -EINVAL;
309 	}
310 	if(n >= MAX_DEV){
311 		*error_out = "Device number out of range";
312 		return 1;
313 	}
314 
315 	err = -EBUSY;
316 	mutex_lock(&ubd_lock);
317 
318 	ubd_dev = &ubd_devs[n];
319 	if(ubd_dev->file != NULL){
320 		*error_out = "Device is already configured";
321 		goto out;
322 	}
323 
324 	if (index_out)
325 		*index_out = n;
326 
327 	err = -EINVAL;
328 	for (i = 0; i < sizeof("rscdt="); i++) {
329 		switch (*str) {
330 		case 'r':
331 			flags.w = 0;
332 			break;
333 		case 's':
334 			flags.s = 1;
335 			break;
336 		case 'd':
337 			ubd_dev->no_cow = 1;
338 			break;
339 		case 'c':
340 			ubd_dev->shared = 1;
341 			break;
342 		case 't':
343 			ubd_dev->no_trim = 1;
344 			break;
345 		case '=':
346 			str++;
347 			goto break_loop;
348 		default:
349 			*error_out = "Expected '=' or flag letter "
350 				"(r, s, c, t or d)";
351 			goto out;
352 		}
353 		str++;
354 	}
355 
356 	if (*str == '=')
357 		*error_out = "Too many flags specified";
358 	else
359 		*error_out = "Missing '='";
360 	goto out;
361 
362 break_loop:
363 	backing_file = strchr(str, ',');
364 
365 	if (backing_file == NULL)
366 		backing_file = strchr(str, ':');
367 
368 	if(backing_file != NULL){
369 		if(ubd_dev->no_cow){
370 			*error_out = "Can't specify both 'd' and a cow file";
371 			goto out;
372 		}
373 		else {
374 			*backing_file = '\0';
375 			backing_file++;
376 		}
377 	}
378 	err = 0;
379 	ubd_dev->file = str;
380 	ubd_dev->cow.file = backing_file;
381 	ubd_dev->boot_openflags = flags;
382 out:
383 	mutex_unlock(&ubd_lock);
384 	return err;
385 }
386 
387 static int ubd_setup(char *str)
388 {
389 	char *error;
390 	int err;
391 
392 	err = ubd_setup_common(str, NULL, &error);
393 	if(err)
394 		printk(KERN_ERR "Failed to initialize device with \"%s\" : "
395 		       "%s\n", str, error);
396 	return 1;
397 }
398 
399 __setup("ubd", ubd_setup);
400 __uml_help(ubd_setup,
401 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
402 "    This is used to associate a device with a file in the underlying\n"
403 "    filesystem. When specifying two filenames, the first one is the\n"
404 "    COW name and the second is the backing file name. As separator you can\n"
405 "    use either a ':' or a ',': the first one allows writing things like;\n"
406 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
407 "    while with a ',' the shell would not expand the 2nd '~'.\n"
408 "    When using only one filename, UML will detect whether to treat it like\n"
409 "    a COW file or a backing file. To override this detection, add the 'd'\n"
410 "    flag:\n"
411 "	ubd0d=BackingFile\n"
412 "    Usually, there is a filesystem in the file, but \n"
413 "    that's not required. Swap devices containing swap files can be\n"
414 "    specified like this. Also, a file which doesn't contain a\n"
415 "    filesystem can have its contents read in the virtual \n"
416 "    machine by running 'dd' on the device. <n> must be in the range\n"
417 "    0 to 7. Appending an 'r' to the number will cause that device\n"
418 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
419 "    an 's' will cause data to be written to disk on the host immediately.\n"
420 "    'c' will cause the device to be treated as being shared between multiple\n"
421 "    UMLs and file locking will be turned off - this is appropriate for a\n"
422 "    cluster filesystem and inappropriate at almost all other times.\n\n"
423 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
424 );
425 
426 static int udb_setup(char *str)
427 {
428 	printk("udb%s specified on command line is almost certainly a ubd -> "
429 	       "udb TYPO\n", str);
430 	return 1;
431 }
432 
433 __setup("udb", udb_setup);
434 __uml_help(udb_setup,
435 "udb\n"
436 "    This option is here solely to catch ubd -> udb typos, which can be\n"
437 "    to impossible to catch visually unless you specifically look for\n"
438 "    them.  The only result of any option starting with 'udb' is an error\n"
439 "    in the boot output.\n\n"
440 );
441 
442 /* Only changed by ubd_init, which is an initcall. */
443 static int thread_fd = -1;
444 
445 /* Function to read several request pointers at a time
446 * handling fractional reads if (and as) needed
447 */
448 
449 static int bulk_req_safe_read(
450 	int fd,
451 	struct io_thread_req * (*request_buffer)[],
452 	struct io_thread_req **remainder,
453 	int *remainder_size,
454 	int max_recs
455 	)
456 {
457 	int n = 0;
458 	int res = 0;
459 
460 	if (*remainder_size > 0) {
461 		memmove(
462 			(char *) request_buffer,
463 			(char *) remainder, *remainder_size
464 		);
465 		n = *remainder_size;
466 	}
467 
468 	res = os_read_file(
469 			fd,
470 			((char *) request_buffer) + *remainder_size,
471 			sizeof(struct io_thread_req *)*max_recs
472 				- *remainder_size
473 		);
474 	if (res > 0) {
475 		n += res;
476 		if ((n % sizeof(struct io_thread_req *)) > 0) {
477 			/*
478 			* Read somehow returned not a multiple of dword
479 			* theoretically possible, but never observed in the
480 			* wild, so read routine must be able to handle it
481 			*/
482 			*remainder_size = n % sizeof(struct io_thread_req *);
483 			WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
484 			memmove(
485 				remainder,
486 				((char *) request_buffer) +
487 					(n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
488 				*remainder_size
489 			);
490 			n = n - *remainder_size;
491 		}
492 	} else {
493 		n = res;
494 	}
495 	return n;
496 }
497 
498 /* Called without dev->lock held, and only in interrupt context. */
499 static void ubd_handler(void)
500 {
501 	int n;
502 	int count;
503 
504 	while(1){
505 		n = bulk_req_safe_read(
506 			thread_fd,
507 			irq_req_buffer,
508 			&irq_remainder,
509 			&irq_remainder_size,
510 			UBD_REQ_BUFFER_SIZE
511 		);
512 		if (n < 0) {
513 			if(n == -EAGAIN)
514 				break;
515 			printk(KERN_ERR "spurious interrupt in ubd_handler, "
516 			       "err = %d\n", -n);
517 			return;
518 		}
519 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
520 			struct io_thread_req *io_req = (*irq_req_buffer)[count];
521 
522 			if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
523 				blk_queue_max_discard_sectors(io_req->req->q, 0);
524 				blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
525 				blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
526 			}
527 			if ((io_req->error) || (io_req->buffer == NULL))
528 				blk_mq_end_request(io_req->req, io_req->error);
529 			else {
530 				if (!blk_update_request(io_req->req, io_req->error, io_req->length))
531 					__blk_mq_end_request(io_req->req, io_req->error);
532 			}
533 			kfree(io_req);
534 		}
535 	}
536 }
537 
538 static irqreturn_t ubd_intr(int irq, void *dev)
539 {
540 	ubd_handler();
541 	return IRQ_HANDLED;
542 }
543 
544 /* Only changed by ubd_init, which is an initcall. */
545 static int io_pid = -1;
546 
547 static void kill_io_thread(void)
548 {
549 	if(io_pid != -1)
550 		os_kill_process(io_pid, 1);
551 }
552 
553 __uml_exitcall(kill_io_thread);
554 
555 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
556 {
557 	char *file;
558 	int fd;
559 	int err;
560 
561 	__u32 version;
562 	__u32 align;
563 	char *backing_file;
564 	time_t mtime;
565 	unsigned long long size;
566 	int sector_size;
567 	int bitmap_offset;
568 
569 	if (ubd_dev->file && ubd_dev->cow.file) {
570 		file = ubd_dev->cow.file;
571 
572 		goto out;
573 	}
574 
575 	fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
576 	if (fd < 0)
577 		return fd;
578 
579 	err = read_cow_header(file_reader, &fd, &version, &backing_file, \
580 		&mtime, &size, &sector_size, &align, &bitmap_offset);
581 	os_close_file(fd);
582 
583 	if(err == -EINVAL)
584 		file = ubd_dev->file;
585 	else
586 		file = backing_file;
587 
588 out:
589 	return os_file_size(file, size_out);
590 }
591 
592 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
593 {
594 	int err;
595 
596 	err = os_pread_file(fd, buf, len, offset);
597 	if (err < 0)
598 		return err;
599 
600 	return 0;
601 }
602 
603 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
604 {
605 	unsigned long modtime;
606 	unsigned long long actual;
607 	int err;
608 
609 	err = os_file_modtime(file, &modtime);
610 	if (err < 0) {
611 		printk(KERN_ERR "Failed to get modification time of backing "
612 		       "file \"%s\", err = %d\n", file, -err);
613 		return err;
614 	}
615 
616 	err = os_file_size(file, &actual);
617 	if (err < 0) {
618 		printk(KERN_ERR "Failed to get size of backing file \"%s\", "
619 		       "err = %d\n", file, -err);
620 		return err;
621 	}
622 
623 	if (actual != size) {
624 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
625 		 * the typecast.*/
626 		printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
627 		       "vs backing file\n", (unsigned long long) size, actual);
628 		return -EINVAL;
629 	}
630 	if (modtime != mtime) {
631 		printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
632 		       "backing file\n", mtime, modtime);
633 		return -EINVAL;
634 	}
635 	return 0;
636 }
637 
638 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
639 {
640 	struct uml_stat buf1, buf2;
641 	int err;
642 
643 	if (from_cmdline == NULL)
644 		return 0;
645 	if (!strcmp(from_cmdline, from_cow))
646 		return 0;
647 
648 	err = os_stat_file(from_cmdline, &buf1);
649 	if (err < 0) {
650 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
651 		       -err);
652 		return 0;
653 	}
654 	err = os_stat_file(from_cow, &buf2);
655 	if (err < 0) {
656 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
657 		       -err);
658 		return 1;
659 	}
660 	if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
661 		return 0;
662 
663 	printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
664 	       "\"%s\" specified in COW header of \"%s\"\n",
665 	       from_cmdline, from_cow, cow);
666 	return 1;
667 }
668 
669 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
670 		  char **backing_file_out, int *bitmap_offset_out,
671 		  unsigned long *bitmap_len_out, int *data_offset_out,
672 		  int *create_cow_out)
673 {
674 	time_t mtime;
675 	unsigned long long size;
676 	__u32 version, align;
677 	char *backing_file;
678 	int fd, err, sectorsize, asked_switch, mode = 0644;
679 
680 	fd = os_open_file(file, *openflags, mode);
681 	if (fd < 0) {
682 		if ((fd == -ENOENT) && (create_cow_out != NULL))
683 			*create_cow_out = 1;
684 		if (!openflags->w ||
685 		    ((fd != -EROFS) && (fd != -EACCES)))
686 			return fd;
687 		openflags->w = 0;
688 		fd = os_open_file(file, *openflags, mode);
689 		if (fd < 0)
690 			return fd;
691 	}
692 
693 	if (shared)
694 		printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
695 	else {
696 		err = os_lock_file(fd, openflags->w);
697 		if (err < 0) {
698 			printk(KERN_ERR "Failed to lock '%s', err = %d\n",
699 			       file, -err);
700 			goto out_close;
701 		}
702 	}
703 
704 	/* Successful return case! */
705 	if (backing_file_out == NULL)
706 		return fd;
707 
708 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
709 			      &size, &sectorsize, &align, bitmap_offset_out);
710 	if (err && (*backing_file_out != NULL)) {
711 		printk(KERN_ERR "Failed to read COW header from COW file "
712 		       "\"%s\", errno = %d\n", file, -err);
713 		goto out_close;
714 	}
715 	if (err)
716 		return fd;
717 
718 	asked_switch = path_requires_switch(*backing_file_out, backing_file,
719 					    file);
720 
721 	/* Allow switching only if no mismatch. */
722 	if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
723 						   mtime)) {
724 		printk(KERN_ERR "Switching backing file to '%s'\n",
725 		       *backing_file_out);
726 		err = write_cow_header(file, fd, *backing_file_out,
727 				       sectorsize, align, &size);
728 		if (err) {
729 			printk(KERN_ERR "Switch failed, errno = %d\n", -err);
730 			goto out_close;
731 		}
732 	} else {
733 		*backing_file_out = backing_file;
734 		err = backing_file_mismatch(*backing_file_out, size, mtime);
735 		if (err)
736 			goto out_close;
737 	}
738 
739 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
740 		  bitmap_len_out, data_offset_out);
741 
742 	return fd;
743  out_close:
744 	os_close_file(fd);
745 	return err;
746 }
747 
748 static int create_cow_file(char *cow_file, char *backing_file,
749 		    struct openflags flags,
750 		    int sectorsize, int alignment, int *bitmap_offset_out,
751 		    unsigned long *bitmap_len_out, int *data_offset_out)
752 {
753 	int err, fd;
754 
755 	flags.c = 1;
756 	fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
757 	if (fd < 0) {
758 		err = fd;
759 		printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
760 		       cow_file, -err);
761 		goto out;
762 	}
763 
764 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
765 			    bitmap_offset_out, bitmap_len_out,
766 			    data_offset_out);
767 	if (!err)
768 		return fd;
769 	os_close_file(fd);
770  out:
771 	return err;
772 }
773 
774 static void ubd_close_dev(struct ubd *ubd_dev)
775 {
776 	os_close_file(ubd_dev->fd);
777 	if(ubd_dev->cow.file == NULL)
778 		return;
779 
780 	os_close_file(ubd_dev->cow.fd);
781 	vfree(ubd_dev->cow.bitmap);
782 	ubd_dev->cow.bitmap = NULL;
783 }
784 
785 static int ubd_open_dev(struct ubd *ubd_dev)
786 {
787 	struct openflags flags;
788 	char **back_ptr;
789 	int err, create_cow, *create_ptr;
790 	int fd;
791 
792 	ubd_dev->openflags = ubd_dev->boot_openflags;
793 	create_cow = 0;
794 	create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
795 	back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
796 
797 	fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
798 				back_ptr, &ubd_dev->cow.bitmap_offset,
799 				&ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
800 				create_ptr);
801 
802 	if((fd == -ENOENT) && create_cow){
803 		fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
804 					  ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
805 					  &ubd_dev->cow.bitmap_offset,
806 					  &ubd_dev->cow.bitmap_len,
807 					  &ubd_dev->cow.data_offset);
808 		if(fd >= 0){
809 			printk(KERN_INFO "Creating \"%s\" as COW file for "
810 			       "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
811 		}
812 	}
813 
814 	if(fd < 0){
815 		printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
816 		       -fd);
817 		return fd;
818 	}
819 	ubd_dev->fd = fd;
820 
821 	if(ubd_dev->cow.file != NULL){
822 		blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
823 
824 		err = -ENOMEM;
825 		ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
826 		if(ubd_dev->cow.bitmap == NULL){
827 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
828 			goto error;
829 		}
830 		flush_tlb_kernel_vm();
831 
832 		err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
833 				      ubd_dev->cow.bitmap_offset,
834 				      ubd_dev->cow.bitmap_len);
835 		if(err < 0)
836 			goto error;
837 
838 		flags = ubd_dev->openflags;
839 		flags.w = 0;
840 		err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
841 				    NULL, NULL, NULL, NULL);
842 		if(err < 0) goto error;
843 		ubd_dev->cow.fd = err;
844 	}
845 	if (ubd_dev->no_trim == 0) {
846 		ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
847 		ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
848 		blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
849 		blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
850 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
851 	}
852 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
853 	return 0;
854  error:
855 	os_close_file(ubd_dev->fd);
856 	return err;
857 }
858 
859 static void ubd_device_release(struct device *dev)
860 {
861 	struct ubd *ubd_dev = dev_get_drvdata(dev);
862 
863 	blk_cleanup_queue(ubd_dev->queue);
864 	blk_mq_free_tag_set(&ubd_dev->tag_set);
865 	*ubd_dev = ((struct ubd) DEFAULT_UBD);
866 }
867 
868 static int ubd_disk_register(int major, u64 size, int unit,
869 			     struct gendisk **disk_out)
870 {
871 	struct device *parent = NULL;
872 	struct gendisk *disk;
873 
874 	disk = alloc_disk(1 << UBD_SHIFT);
875 	if(disk == NULL)
876 		return -ENOMEM;
877 
878 	disk->major = major;
879 	disk->first_minor = unit << UBD_SHIFT;
880 	disk->fops = &ubd_blops;
881 	set_capacity(disk, size / 512);
882 	if (major == UBD_MAJOR)
883 		sprintf(disk->disk_name, "ubd%c", 'a' + unit);
884 	else
885 		sprintf(disk->disk_name, "ubd_fake%d", unit);
886 
887 	/* sysfs register (not for ide fake devices) */
888 	if (major == UBD_MAJOR) {
889 		ubd_devs[unit].pdev.id   = unit;
890 		ubd_devs[unit].pdev.name = DRIVER_NAME;
891 		ubd_devs[unit].pdev.dev.release = ubd_device_release;
892 		dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
893 		platform_device_register(&ubd_devs[unit].pdev);
894 		parent = &ubd_devs[unit].pdev.dev;
895 	}
896 
897 	disk->private_data = &ubd_devs[unit];
898 	disk->queue = ubd_devs[unit].queue;
899 	device_add_disk(parent, disk, NULL);
900 
901 	*disk_out = disk;
902 	return 0;
903 }
904 
905 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
906 
907 static const struct blk_mq_ops ubd_mq_ops = {
908 	.queue_rq = ubd_queue_rq,
909 };
910 
911 static int ubd_add(int n, char **error_out)
912 {
913 	struct ubd *ubd_dev = &ubd_devs[n];
914 	int err = 0;
915 
916 	if(ubd_dev->file == NULL)
917 		goto out;
918 
919 	err = ubd_file_size(ubd_dev, &ubd_dev->size);
920 	if(err < 0){
921 		*error_out = "Couldn't determine size of device's file";
922 		goto out;
923 	}
924 
925 	ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
926 
927 	ubd_dev->tag_set.ops = &ubd_mq_ops;
928 	ubd_dev->tag_set.queue_depth = 64;
929 	ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
930 	ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
931 	ubd_dev->tag_set.driver_data = ubd_dev;
932 	ubd_dev->tag_set.nr_hw_queues = 1;
933 
934 	err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
935 	if (err)
936 		goto out;
937 
938 	ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
939 	if (IS_ERR(ubd_dev->queue)) {
940 		err = PTR_ERR(ubd_dev->queue);
941 		goto out_cleanup_tags;
942 	}
943 
944 	ubd_dev->queue->queuedata = ubd_dev;
945 	blk_queue_write_cache(ubd_dev->queue, true, false);
946 
947 	blk_queue_max_segments(ubd_dev->queue, MAX_SG);
948 	err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
949 	if(err){
950 		*error_out = "Failed to register device";
951 		goto out_cleanup_tags;
952 	}
953 
954 	if (fake_major != UBD_MAJOR)
955 		ubd_disk_register(fake_major, ubd_dev->size, n,
956 				  &fake_gendisk[n]);
957 
958 	/*
959 	 * Perhaps this should also be under the "if (fake_major)" above
960 	 * using the fake_disk->disk_name
961 	 */
962 	if (fake_ide)
963 		make_ide_entries(ubd_gendisk[n]->disk_name);
964 
965 	err = 0;
966 out:
967 	return err;
968 
969 out_cleanup_tags:
970 	blk_mq_free_tag_set(&ubd_dev->tag_set);
971 	if (!(IS_ERR(ubd_dev->queue)))
972 		blk_cleanup_queue(ubd_dev->queue);
973 	goto out;
974 }
975 
976 static int ubd_config(char *str, char **error_out)
977 {
978 	int n, ret;
979 
980 	/* This string is possibly broken up and stored, so it's only
981 	 * freed if ubd_setup_common fails, or if only general options
982 	 * were set.
983 	 */
984 	str = kstrdup(str, GFP_KERNEL);
985 	if (str == NULL) {
986 		*error_out = "Failed to allocate memory";
987 		return -ENOMEM;
988 	}
989 
990 	ret = ubd_setup_common(str, &n, error_out);
991 	if (ret)
992 		goto err_free;
993 
994 	if (n == -1) {
995 		ret = 0;
996 		goto err_free;
997 	}
998 
999 	mutex_lock(&ubd_lock);
1000 	ret = ubd_add(n, error_out);
1001 	if (ret)
1002 		ubd_devs[n].file = NULL;
1003 	mutex_unlock(&ubd_lock);
1004 
1005 out:
1006 	return ret;
1007 
1008 err_free:
1009 	kfree(str);
1010 	goto out;
1011 }
1012 
1013 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1014 {
1015 	struct ubd *ubd_dev;
1016 	int n, len = 0;
1017 
1018 	n = parse_unit(&name);
1019 	if((n >= MAX_DEV) || (n < 0)){
1020 		*error_out = "ubd_get_config : device number out of range";
1021 		return -1;
1022 	}
1023 
1024 	ubd_dev = &ubd_devs[n];
1025 	mutex_lock(&ubd_lock);
1026 
1027 	if(ubd_dev->file == NULL){
1028 		CONFIG_CHUNK(str, size, len, "", 1);
1029 		goto out;
1030 	}
1031 
1032 	CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1033 
1034 	if(ubd_dev->cow.file != NULL){
1035 		CONFIG_CHUNK(str, size, len, ",", 0);
1036 		CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1037 	}
1038 	else CONFIG_CHUNK(str, size, len, "", 1);
1039 
1040  out:
1041 	mutex_unlock(&ubd_lock);
1042 	return len;
1043 }
1044 
1045 static int ubd_id(char **str, int *start_out, int *end_out)
1046 {
1047 	int n;
1048 
1049 	n = parse_unit(str);
1050 	*start_out = 0;
1051 	*end_out = MAX_DEV - 1;
1052 	return n;
1053 }
1054 
1055 static int ubd_remove(int n, char **error_out)
1056 {
1057 	struct gendisk *disk = ubd_gendisk[n];
1058 	struct ubd *ubd_dev;
1059 	int err = -ENODEV;
1060 
1061 	mutex_lock(&ubd_lock);
1062 
1063 	ubd_dev = &ubd_devs[n];
1064 
1065 	if(ubd_dev->file == NULL)
1066 		goto out;
1067 
1068 	/* you cannot remove a open disk */
1069 	err = -EBUSY;
1070 	if(ubd_dev->count > 0)
1071 		goto out;
1072 
1073 	ubd_gendisk[n] = NULL;
1074 	if(disk != NULL){
1075 		del_gendisk(disk);
1076 		put_disk(disk);
1077 	}
1078 
1079 	if(fake_gendisk[n] != NULL){
1080 		del_gendisk(fake_gendisk[n]);
1081 		put_disk(fake_gendisk[n]);
1082 		fake_gendisk[n] = NULL;
1083 	}
1084 
1085 	err = 0;
1086 	platform_device_unregister(&ubd_dev->pdev);
1087 out:
1088 	mutex_unlock(&ubd_lock);
1089 	return err;
1090 }
1091 
1092 /* All these are called by mconsole in process context and without
1093  * ubd-specific locks.  The structure itself is const except for .list.
1094  */
1095 static struct mc_device ubd_mc = {
1096 	.list		= LIST_HEAD_INIT(ubd_mc.list),
1097 	.name		= "ubd",
1098 	.config		= ubd_config,
1099 	.get_config	= ubd_get_config,
1100 	.id		= ubd_id,
1101 	.remove		= ubd_remove,
1102 };
1103 
1104 static int __init ubd_mc_init(void)
1105 {
1106 	mconsole_register_dev(&ubd_mc);
1107 	return 0;
1108 }
1109 
1110 __initcall(ubd_mc_init);
1111 
1112 static int __init ubd0_init(void)
1113 {
1114 	struct ubd *ubd_dev = &ubd_devs[0];
1115 
1116 	mutex_lock(&ubd_lock);
1117 	if(ubd_dev->file == NULL)
1118 		ubd_dev->file = "root_fs";
1119 	mutex_unlock(&ubd_lock);
1120 
1121 	return 0;
1122 }
1123 
1124 __initcall(ubd0_init);
1125 
1126 /* Used in ubd_init, which is an initcall */
1127 static struct platform_driver ubd_driver = {
1128 	.driver = {
1129 		.name  = DRIVER_NAME,
1130 	},
1131 };
1132 
1133 static int __init ubd_init(void)
1134 {
1135 	char *error;
1136 	int i, err;
1137 
1138 	if (register_blkdev(UBD_MAJOR, "ubd"))
1139 		return -1;
1140 
1141 	if (fake_major != UBD_MAJOR) {
1142 		char name[sizeof("ubd_nnn\0")];
1143 
1144 		snprintf(name, sizeof(name), "ubd_%d", fake_major);
1145 		if (register_blkdev(fake_major, "ubd"))
1146 			return -1;
1147 	}
1148 
1149 	irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1150 				       sizeof(struct io_thread_req *),
1151 				       GFP_KERNEL
1152 		);
1153 	irq_remainder = 0;
1154 
1155 	if (irq_req_buffer == NULL) {
1156 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1157 		return -1;
1158 	}
1159 	io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1160 				      sizeof(struct io_thread_req *),
1161 				      GFP_KERNEL
1162 		);
1163 
1164 	io_remainder = 0;
1165 
1166 	if (io_req_buffer == NULL) {
1167 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1168 		return -1;
1169 	}
1170 	platform_driver_register(&ubd_driver);
1171 	mutex_lock(&ubd_lock);
1172 	for (i = 0; i < MAX_DEV; i++){
1173 		err = ubd_add(i, &error);
1174 		if(err)
1175 			printk(KERN_ERR "Failed to initialize ubd device %d :"
1176 			       "%s\n", i, error);
1177 	}
1178 	mutex_unlock(&ubd_lock);
1179 	return 0;
1180 }
1181 
1182 late_initcall(ubd_init);
1183 
1184 static int __init ubd_driver_init(void){
1185 	unsigned long stack;
1186 	int err;
1187 
1188 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1189 	if(global_openflags.s){
1190 		printk(KERN_INFO "ubd: Synchronous mode\n");
1191 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1192 		 * enough. So use anyway the io thread. */
1193 	}
1194 	stack = alloc_stack(0, 0);
1195 	io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1196 				 &thread_fd);
1197 	if(io_pid < 0){
1198 		printk(KERN_ERR
1199 		       "ubd : Failed to start I/O thread (errno = %d) - "
1200 		       "falling back to synchronous I/O\n", -io_pid);
1201 		io_pid = -1;
1202 		return 0;
1203 	}
1204 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1205 			     0, "ubd", ubd_devs);
1206 	if(err != 0)
1207 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1208 	return 0;
1209 }
1210 
1211 device_initcall(ubd_driver_init);
1212 
1213 static int ubd_open(struct block_device *bdev, fmode_t mode)
1214 {
1215 	struct gendisk *disk = bdev->bd_disk;
1216 	struct ubd *ubd_dev = disk->private_data;
1217 	int err = 0;
1218 
1219 	mutex_lock(&ubd_mutex);
1220 	if(ubd_dev->count == 0){
1221 		err = ubd_open_dev(ubd_dev);
1222 		if(err){
1223 			printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1224 			       disk->disk_name, ubd_dev->file, -err);
1225 			goto out;
1226 		}
1227 	}
1228 	ubd_dev->count++;
1229 	set_disk_ro(disk, !ubd_dev->openflags.w);
1230 
1231 	/* This should no more be needed. And it didn't work anyway to exclude
1232 	 * read-write remounting of filesystems.*/
1233 	/*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1234 	        if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1235 	        err = -EROFS;
1236 	}*/
1237 out:
1238 	mutex_unlock(&ubd_mutex);
1239 	return err;
1240 }
1241 
1242 static void ubd_release(struct gendisk *disk, fmode_t mode)
1243 {
1244 	struct ubd *ubd_dev = disk->private_data;
1245 
1246 	mutex_lock(&ubd_mutex);
1247 	if(--ubd_dev->count == 0)
1248 		ubd_close_dev(ubd_dev);
1249 	mutex_unlock(&ubd_mutex);
1250 }
1251 
1252 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1253 			  __u64 *cow_offset, unsigned long *bitmap,
1254 			  __u64 bitmap_offset, unsigned long *bitmap_words,
1255 			  __u64 bitmap_len)
1256 {
1257 	__u64 sector = io_offset >> SECTOR_SHIFT;
1258 	int i, update_bitmap = 0;
1259 
1260 	for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1261 		if(cow_mask != NULL)
1262 			ubd_set_bit(i, (unsigned char *) cow_mask);
1263 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1264 			continue;
1265 
1266 		update_bitmap = 1;
1267 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
1268 	}
1269 
1270 	if(!update_bitmap)
1271 		return;
1272 
1273 	*cow_offset = sector / (sizeof(unsigned long) * 8);
1274 
1275 	/* This takes care of the case where we're exactly at the end of the
1276 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
1277 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1278 	 * for the original diagnosis.
1279 	 */
1280 	if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1281 					 sizeof(unsigned long)) - 1))
1282 		(*cow_offset)--;
1283 
1284 	bitmap_words[0] = bitmap[*cow_offset];
1285 	bitmap_words[1] = bitmap[*cow_offset + 1];
1286 
1287 	*cow_offset *= sizeof(unsigned long);
1288 	*cow_offset += bitmap_offset;
1289 }
1290 
1291 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1292 		       __u64 bitmap_offset, __u64 bitmap_len)
1293 {
1294 	__u64 sector = req->offset >> SECTOR_SHIFT;
1295 	int i;
1296 
1297 	if (req->length > (sizeof(req->sector_mask) * 8) << SECTOR_SHIFT)
1298 		panic("Operation too long");
1299 
1300 	if (req_op(req->req) == REQ_OP_READ) {
1301 		for (i = 0; i < req->length >> SECTOR_SHIFT; i++) {
1302 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1303 				ubd_set_bit(i, (unsigned char *)
1304 					    &req->sector_mask);
1305 		}
1306 	}
1307 	else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1308 			   &req->cow_offset, bitmap, bitmap_offset,
1309 			   req->bitmap_words, bitmap_len);
1310 }
1311 
1312 static int ubd_queue_one_vec(struct blk_mq_hw_ctx *hctx, struct request *req,
1313 		u64 off, struct bio_vec *bvec)
1314 {
1315 	struct ubd *dev = hctx->queue->queuedata;
1316 	struct io_thread_req *io_req;
1317 	int ret;
1318 
1319 	io_req = kmalloc(sizeof(struct io_thread_req), GFP_ATOMIC);
1320 	if (!io_req)
1321 		return -ENOMEM;
1322 
1323 	io_req->req = req;
1324 	if (dev->cow.file)
1325 		io_req->fds[0] = dev->cow.fd;
1326 	else
1327 		io_req->fds[0] = dev->fd;
1328 	io_req->error = 0;
1329 
1330 	if (bvec != NULL) {
1331 		io_req->buffer = page_address(bvec->bv_page) + bvec->bv_offset;
1332 		io_req->length = bvec->bv_len;
1333 	} else {
1334 		io_req->buffer = NULL;
1335 		io_req->length = blk_rq_bytes(req);
1336 	}
1337 
1338 	io_req->sectorsize = SECTOR_SIZE;
1339 	io_req->fds[1] = dev->fd;
1340 	io_req->cow_offset = -1;
1341 	io_req->offset = off;
1342 	io_req->sector_mask = 0;
1343 	io_req->offsets[0] = 0;
1344 	io_req->offsets[1] = dev->cow.data_offset;
1345 
1346 	if (dev->cow.file)
1347 		cowify_req(io_req, dev->cow.bitmap,
1348 			   dev->cow.bitmap_offset, dev->cow.bitmap_len);
1349 
1350 	ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1351 	if (ret != sizeof(io_req)) {
1352 		if (ret != -EAGAIN)
1353 			pr_err("write to io thread failed: %d\n", -ret);
1354 		kfree(io_req);
1355 	}
1356 	return ret;
1357 }
1358 
1359 static int queue_rw_req(struct blk_mq_hw_ctx *hctx, struct request *req)
1360 {
1361 	struct req_iterator iter;
1362 	struct bio_vec bvec;
1363 	int ret;
1364 	u64 off = (u64)blk_rq_pos(req) << SECTOR_SHIFT;
1365 
1366 	rq_for_each_segment(bvec, req, iter) {
1367 		ret = ubd_queue_one_vec(hctx, req, off, &bvec);
1368 		if (ret < 0)
1369 			return ret;
1370 		off += bvec.bv_len;
1371 	}
1372 	return 0;
1373 }
1374 
1375 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1376 				 const struct blk_mq_queue_data *bd)
1377 {
1378 	struct ubd *ubd_dev = hctx->queue->queuedata;
1379 	struct request *req = bd->rq;
1380 	int ret = 0, res = BLK_STS_OK;
1381 
1382 	blk_mq_start_request(req);
1383 
1384 	spin_lock_irq(&ubd_dev->lock);
1385 
1386 	switch (req_op(req)) {
1387 	/* operations with no lentgth/offset arguments */
1388 	case REQ_OP_FLUSH:
1389 		ret = ubd_queue_one_vec(hctx, req, 0, NULL);
1390 		break;
1391 	case REQ_OP_READ:
1392 	case REQ_OP_WRITE:
1393 		ret = queue_rw_req(hctx, req);
1394 		break;
1395 	case REQ_OP_DISCARD:
1396 	case REQ_OP_WRITE_ZEROES:
1397 		ret = ubd_queue_one_vec(hctx, req, (u64)blk_rq_pos(req) << 9, NULL);
1398 		break;
1399 	default:
1400 		WARN_ON_ONCE(1);
1401 		res = BLK_STS_NOTSUPP;
1402 	}
1403 
1404 	spin_unlock_irq(&ubd_dev->lock);
1405 
1406 	if (ret < 0)
1407 		blk_mq_requeue_request(req, true);
1408 
1409 	return res;
1410 }
1411 
1412 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1413 {
1414 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1415 
1416 	geo->heads = 128;
1417 	geo->sectors = 32;
1418 	geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1419 	return 0;
1420 }
1421 
1422 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1423 		     unsigned int cmd, unsigned long arg)
1424 {
1425 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1426 	u16 ubd_id[ATA_ID_WORDS];
1427 
1428 	switch (cmd) {
1429 		struct cdrom_volctrl volume;
1430 	case HDIO_GET_IDENTITY:
1431 		memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1432 		ubd_id[ATA_ID_CYLS]	= ubd_dev->size / (128 * 32 * 512);
1433 		ubd_id[ATA_ID_HEADS]	= 128;
1434 		ubd_id[ATA_ID_SECTORS]	= 32;
1435 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1436 				 sizeof(ubd_id)))
1437 			return -EFAULT;
1438 		return 0;
1439 
1440 	case CDROMVOLREAD:
1441 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1442 			return -EFAULT;
1443 		volume.channel0 = 255;
1444 		volume.channel1 = 255;
1445 		volume.channel2 = 255;
1446 		volume.channel3 = 255;
1447 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1448 			return -EFAULT;
1449 		return 0;
1450 	}
1451 	return -EINVAL;
1452 }
1453 
1454 static int map_error(int error_code)
1455 {
1456 	switch (error_code) {
1457 	case 0:
1458 		return BLK_STS_OK;
1459 	case ENOSYS:
1460 	case EOPNOTSUPP:
1461 		return BLK_STS_NOTSUPP;
1462 	case ENOSPC:
1463 		return BLK_STS_NOSPC;
1464 	}
1465 	return BLK_STS_IOERR;
1466 }
1467 
1468 /*
1469  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1470  *
1471  * The following functions are part of UML hypervisor code.
1472  * All functions from here onwards are executed as a helper
1473  * thread and are not allowed to execute any kernel functions.
1474  *
1475  * Any communication must occur strictly via shared memory and IPC.
1476  *
1477  * Do not add printks, locks, kernel memory operations, etc - it
1478  * will result in unpredictable behaviour and/or crashes.
1479  */
1480 
1481 static int update_bitmap(struct io_thread_req *req)
1482 {
1483 	int n;
1484 
1485 	if(req->cow_offset == -1)
1486 		return map_error(0);
1487 
1488 	n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1489 			  sizeof(req->bitmap_words), req->cow_offset);
1490 	if (n != sizeof(req->bitmap_words))
1491 		return map_error(-n);
1492 
1493 	return map_error(0);
1494 }
1495 
1496 static void do_io(struct io_thread_req *req)
1497 {
1498 	char *buf = NULL;
1499 	unsigned long len;
1500 	int n, nsectors, start, end, bit;
1501 	__u64 off;
1502 
1503 	/* FLUSH is really a special case, we cannot "case" it with others */
1504 
1505 	if (req_op(req->req) == REQ_OP_FLUSH) {
1506 		/* fds[0] is always either the rw image or our cow file */
1507 		req->error = map_error(-os_sync_file(req->fds[0]));
1508 		return;
1509 	}
1510 
1511 	nsectors = req->length / req->sectorsize;
1512 	start = 0;
1513 	do {
1514 		bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1515 		end = start;
1516 		while((end < nsectors) &&
1517 		      (ubd_test_bit(end, (unsigned char *)
1518 				    &req->sector_mask) == bit))
1519 			end++;
1520 
1521 		off = req->offset + req->offsets[bit] +
1522 			start * req->sectorsize;
1523 		len = (end - start) * req->sectorsize;
1524 		if (req->buffer != NULL)
1525 			buf = &req->buffer[start * req->sectorsize];
1526 
1527 		switch (req_op(req->req)) {
1528 		case REQ_OP_READ:
1529 			n = 0;
1530 			do {
1531 				buf = &buf[n];
1532 				len -= n;
1533 				n = os_pread_file(req->fds[bit], buf, len, off);
1534 				if (n < 0) {
1535 					req->error = map_error(-n);
1536 					return;
1537 				}
1538 			} while((n < len) && (n != 0));
1539 			if (n < len) memset(&buf[n], 0, len - n);
1540 			break;
1541 		case REQ_OP_WRITE:
1542 			n = os_pwrite_file(req->fds[bit], buf, len, off);
1543 			if(n != len){
1544 				req->error = map_error(-n);
1545 				return;
1546 			}
1547 			break;
1548 		case REQ_OP_DISCARD:
1549 		case REQ_OP_WRITE_ZEROES:
1550 			n = os_falloc_punch(req->fds[bit], off, len);
1551 			if (n) {
1552 				req->error = map_error(-n);
1553 				return;
1554 			}
1555 			break;
1556 		default:
1557 			WARN_ON_ONCE(1);
1558 			req->error = BLK_STS_NOTSUPP;
1559 			return;
1560 		}
1561 
1562 		start = end;
1563 	} while(start < nsectors);
1564 
1565 	req->error = update_bitmap(req);
1566 }
1567 
1568 /* Changed in start_io_thread, which is serialized by being called only
1569  * from ubd_init, which is an initcall.
1570  */
1571 int kernel_fd = -1;
1572 
1573 /* Only changed by the io thread. XXX: currently unused. */
1574 static int io_count = 0;
1575 
1576 int io_thread(void *arg)
1577 {
1578 	int n, count, written, res;
1579 
1580 	os_fix_helper_signals();
1581 
1582 	while(1){
1583 		n = bulk_req_safe_read(
1584 			kernel_fd,
1585 			io_req_buffer,
1586 			&io_remainder,
1587 			&io_remainder_size,
1588 			UBD_REQ_BUFFER_SIZE
1589 		);
1590 		if (n < 0) {
1591 			if (n == -EAGAIN) {
1592 				ubd_read_poll(-1);
1593 				continue;
1594 			}
1595 		}
1596 
1597 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1598 			io_count++;
1599 			do_io((*io_req_buffer)[count]);
1600 		}
1601 
1602 		written = 0;
1603 
1604 		do {
1605 			res = os_write_file(kernel_fd, ((char *) io_req_buffer) + written, n);
1606 			if (res >= 0) {
1607 				written += res;
1608 			}
1609 			if (written < n) {
1610 				ubd_write_poll(-1);
1611 			}
1612 		} while (written < n);
1613 	}
1614 
1615 	return 0;
1616 }
1617