xref: /linux/arch/um/drivers/ubd_kern.c (revision 5e3b7009f116f684ac6b93d8924506154f3b1f6d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2018 Cambridge Greys Ltd
4  * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
5  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
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/major.h>
31 #include <linux/cdrom.h>
32 #include <linux/proc_fs.h>
33 #include <linux/seq_file.h>
34 #include <linux/ctype.h>
35 #include <linux/slab.h>
36 #include <linux/vmalloc.h>
37 #include <linux/platform_device.h>
38 #include <linux/scatterlist.h>
39 #include <asm/tlbflush.h>
40 #include <kern_util.h>
41 #include "mconsole_kern.h"
42 #include <init.h>
43 #include <irq_kern.h>
44 #include "ubd.h"
45 #include <os.h>
46 #include "cow.h"
47 
48 /* Max request size is determined by sector mask - 32K */
49 #define UBD_MAX_REQUEST (8 * sizeof(long))
50 
51 struct io_desc {
52 	char *buffer;
53 	unsigned long length;
54 	unsigned long sector_mask;
55 	unsigned long long cow_offset;
56 	unsigned long bitmap_words[2];
57 };
58 
59 struct io_thread_req {
60 	struct request *req;
61 	int fds[2];
62 	unsigned long offsets[2];
63 	unsigned long long offset;
64 	int sectorsize;
65 	int error;
66 
67 	int desc_cnt;
68 	/* io_desc has to be the last element of the struct */
69 	struct io_desc io_desc[];
70 };
71 
72 
73 static struct io_thread_req * (*irq_req_buffer)[];
74 static struct io_thread_req *irq_remainder;
75 static int irq_remainder_size;
76 
77 static struct io_thread_req * (*io_req_buffer)[];
78 static struct io_thread_req *io_remainder;
79 static int io_remainder_size;
80 
81 
82 
83 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
84 {
85 	__u64 n;
86 	int bits, off;
87 
88 	bits = sizeof(data[0]) * 8;
89 	n = bit / bits;
90 	off = bit % bits;
91 	return (data[n] & (1 << off)) != 0;
92 }
93 
94 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
95 {
96 	__u64 n;
97 	int bits, off;
98 
99 	bits = sizeof(data[0]) * 8;
100 	n = bit / bits;
101 	off = bit % bits;
102 	data[n] |= (1 << off);
103 }
104 /*End stuff from ubd_user.h*/
105 
106 #define DRIVER_NAME "uml-blkdev"
107 
108 static DEFINE_MUTEX(ubd_lock);
109 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
110 
111 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
112 		     unsigned int cmd, unsigned long arg);
113 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
114 
115 #define MAX_DEV (16)
116 
117 static const struct block_device_operations ubd_blops = {
118         .owner		= THIS_MODULE,
119         .ioctl		= ubd_ioctl,
120         .compat_ioctl	= blkdev_compat_ptr_ioctl,
121 	.getgeo		= ubd_getgeo,
122 };
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 	char *serial;
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 gendisk *disk;
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 	.serial =		NULL, \
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 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
191 				 const struct blk_mq_queue_data *bd);
192 
193 static int fake_ide_setup(char *str)
194 {
195 	pr_warn("The fake_ide option has been removed\n");
196 	return 1;
197 }
198 __setup("fake_ide", fake_ide_setup);
199 
200 __uml_help(fake_ide_setup,
201 "fake_ide\n"
202 "    Obsolete stub.\n\n"
203 );
204 
205 static int parse_unit(char **ptr)
206 {
207 	char *str = *ptr, *end;
208 	int n = -1;
209 
210 	if(isdigit(*str)) {
211 		n = simple_strtoul(str, &end, 0);
212 		if(end == str)
213 			return -1;
214 		*ptr = end;
215 	}
216 	else if (('a' <= *str) && (*str <= 'z')) {
217 		n = *str - 'a';
218 		str++;
219 		*ptr = str;
220 	}
221 	return n;
222 }
223 
224 /* If *index_out == -1 at exit, the passed option was a general one;
225  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
226  * should not be freed on exit.
227  */
228 static int ubd_setup_common(char *str, int *index_out, char **error_out)
229 {
230 	struct ubd *ubd_dev;
231 	struct openflags flags = global_openflags;
232 	char *file, *backing_file, *serial;
233 	int n, err = 0, i;
234 
235 	if(index_out) *index_out = -1;
236 	n = *str;
237 	if(n == '='){
238 		str++;
239 		if(!strcmp(str, "sync")){
240 			global_openflags = of_sync(global_openflags);
241 			return err;
242 		}
243 
244 		pr_warn("fake major not supported any more\n");
245 		return 0;
246 	}
247 
248 	n = parse_unit(&str);
249 	if(n < 0){
250 		*error_out = "Couldn't parse device number";
251 		return -EINVAL;
252 	}
253 	if(n >= MAX_DEV){
254 		*error_out = "Device number out of range";
255 		return 1;
256 	}
257 
258 	err = -EBUSY;
259 	mutex_lock(&ubd_lock);
260 
261 	ubd_dev = &ubd_devs[n];
262 	if(ubd_dev->file != NULL){
263 		*error_out = "Device is already configured";
264 		goto out;
265 	}
266 
267 	if (index_out)
268 		*index_out = n;
269 
270 	err = -EINVAL;
271 	for (i = 0; i < sizeof("rscdt="); i++) {
272 		switch (*str) {
273 		case 'r':
274 			flags.w = 0;
275 			break;
276 		case 's':
277 			flags.s = 1;
278 			break;
279 		case 'd':
280 			ubd_dev->no_cow = 1;
281 			break;
282 		case 'c':
283 			ubd_dev->shared = 1;
284 			break;
285 		case 't':
286 			ubd_dev->no_trim = 1;
287 			break;
288 		case '=':
289 			str++;
290 			goto break_loop;
291 		default:
292 			*error_out = "Expected '=' or flag letter "
293 				"(r, s, c, t or d)";
294 			goto out;
295 		}
296 		str++;
297 	}
298 
299 	if (*str == '=')
300 		*error_out = "Too many flags specified";
301 	else
302 		*error_out = "Missing '='";
303 	goto out;
304 
305 break_loop:
306 	file = strsep(&str, ",:");
307 	if (*file == '\0')
308 		file = NULL;
309 
310 	backing_file = strsep(&str, ",:");
311 	if (backing_file && *backing_file == '\0')
312 		backing_file = NULL;
313 
314 	serial = strsep(&str, ",:");
315 	if (serial && *serial == '\0')
316 		serial = NULL;
317 
318 	if (backing_file && ubd_dev->no_cow) {
319 		*error_out = "Can't specify both 'd' and a cow file";
320 		goto out;
321 	}
322 
323 	err = 0;
324 	ubd_dev->file = file;
325 	ubd_dev->cow.file = backing_file;
326 	ubd_dev->serial = serial;
327 	ubd_dev->boot_openflags = flags;
328 out:
329 	mutex_unlock(&ubd_lock);
330 	return err;
331 }
332 
333 static int ubd_setup(char *str)
334 {
335 	char *error;
336 	int err;
337 
338 	err = ubd_setup_common(str, NULL, &error);
339 	if(err)
340 		printk(KERN_ERR "Failed to initialize device with \"%s\" : "
341 		       "%s\n", str, error);
342 	return 1;
343 }
344 
345 __setup("ubd", ubd_setup);
346 __uml_help(ubd_setup,
347 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
348 "    This is used to associate a device with a file in the underlying\n"
349 "    filesystem. When specifying two filenames, the first one is the\n"
350 "    COW name and the second is the backing file name. As separator you can\n"
351 "    use either a ':' or a ',': the first one allows writing things like;\n"
352 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
353 "    while with a ',' the shell would not expand the 2nd '~'.\n"
354 "    When using only one filename, UML will detect whether to treat it like\n"
355 "    a COW file or a backing file. To override this detection, add the 'd'\n"
356 "    flag:\n"
357 "	ubd0d=BackingFile\n"
358 "    Usually, there is a filesystem in the file, but \n"
359 "    that's not required. Swap devices containing swap files can be\n"
360 "    specified like this. Also, a file which doesn't contain a\n"
361 "    filesystem can have its contents read in the virtual \n"
362 "    machine by running 'dd' on the device. <n> must be in the range\n"
363 "    0 to 7. Appending an 'r' to the number will cause that device\n"
364 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
365 "    an 's' will cause data to be written to disk on the host immediately.\n"
366 "    'c' will cause the device to be treated as being shared between multiple\n"
367 "    UMLs and file locking will be turned off - this is appropriate for a\n"
368 "    cluster filesystem and inappropriate at almost all other times.\n\n"
369 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
370 "    An optional device serial number can be exposed using the serial parameter\n"
371 "    on the cmdline which is exposed as a sysfs entry. This is particularly\n"
372 "    useful when a unique number should be given to the device. Note when\n"
373 "    specifying a label, the filename2 must be also presented. It can be\n"
374 "    an empty string, in which case the backing file is not used:\n"
375 "       ubd0=File,,Serial\n"
376 );
377 
378 static int udb_setup(char *str)
379 {
380 	printk("udb%s specified on command line is almost certainly a ubd -> "
381 	       "udb TYPO\n", str);
382 	return 1;
383 }
384 
385 __setup("udb", udb_setup);
386 __uml_help(udb_setup,
387 "udb\n"
388 "    This option is here solely to catch ubd -> udb typos, which can be\n"
389 "    to impossible to catch visually unless you specifically look for\n"
390 "    them.  The only result of any option starting with 'udb' is an error\n"
391 "    in the boot output.\n\n"
392 );
393 
394 /* Only changed by ubd_init, which is an initcall. */
395 static int thread_fd = -1;
396 
397 /* Function to read several request pointers at a time
398 * handling fractional reads if (and as) needed
399 */
400 
401 static int bulk_req_safe_read(
402 	int fd,
403 	struct io_thread_req * (*request_buffer)[],
404 	struct io_thread_req **remainder,
405 	int *remainder_size,
406 	int max_recs
407 	)
408 {
409 	int n = 0;
410 	int res = 0;
411 
412 	if (*remainder_size > 0) {
413 		memmove(
414 			(char *) request_buffer,
415 			(char *) remainder, *remainder_size
416 		);
417 		n = *remainder_size;
418 	}
419 
420 	res = os_read_file(
421 			fd,
422 			((char *) request_buffer) + *remainder_size,
423 			sizeof(struct io_thread_req *)*max_recs
424 				- *remainder_size
425 		);
426 	if (res > 0) {
427 		n += res;
428 		if ((n % sizeof(struct io_thread_req *)) > 0) {
429 			/*
430 			* Read somehow returned not a multiple of dword
431 			* theoretically possible, but never observed in the
432 			* wild, so read routine must be able to handle it
433 			*/
434 			*remainder_size = n % sizeof(struct io_thread_req *);
435 			WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
436 			memmove(
437 				remainder,
438 				((char *) request_buffer) +
439 					(n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
440 				*remainder_size
441 			);
442 			n = n - *remainder_size;
443 		}
444 	} else {
445 		n = res;
446 	}
447 	return n;
448 }
449 
450 static void ubd_end_request(struct io_thread_req *io_req)
451 {
452 	if (io_req->error == BLK_STS_NOTSUPP) {
453 		if (req_op(io_req->req) == REQ_OP_DISCARD)
454 			blk_queue_disable_discard(io_req->req->q);
455 		else if (req_op(io_req->req) == REQ_OP_WRITE_ZEROES)
456 			blk_queue_disable_write_zeroes(io_req->req->q);
457 	}
458 	blk_mq_end_request(io_req->req, io_req->error);
459 	kfree(io_req);
460 }
461 
462 static irqreturn_t ubd_intr(int irq, void *dev)
463 {
464 	int len, i;
465 
466 	while ((len = bulk_req_safe_read(thread_fd, irq_req_buffer,
467 			&irq_remainder, &irq_remainder_size,
468 			UBD_REQ_BUFFER_SIZE)) >= 0) {
469 		for (i = 0; i < len / sizeof(struct io_thread_req *); i++)
470 			ubd_end_request((*irq_req_buffer)[i]);
471 	}
472 
473 	if (len < 0 && len != -EAGAIN)
474 		pr_err("spurious interrupt in %s, err = %d\n", __func__, len);
475 	return IRQ_HANDLED;
476 }
477 
478 /* Only changed by ubd_init, which is an initcall. */
479 static int io_pid = -1;
480 
481 static void kill_io_thread(void)
482 {
483 	if(io_pid != -1)
484 		os_kill_process(io_pid, 1);
485 }
486 
487 __uml_exitcall(kill_io_thread);
488 
489 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
490 {
491 	char *file;
492 	int fd;
493 	int err;
494 
495 	__u32 version;
496 	__u32 align;
497 	char *backing_file;
498 	time64_t mtime;
499 	unsigned long long size;
500 	int sector_size;
501 	int bitmap_offset;
502 
503 	if (ubd_dev->file && ubd_dev->cow.file) {
504 		file = ubd_dev->cow.file;
505 
506 		goto out;
507 	}
508 
509 	fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
510 	if (fd < 0)
511 		return fd;
512 
513 	err = read_cow_header(file_reader, &fd, &version, &backing_file, \
514 		&mtime, &size, &sector_size, &align, &bitmap_offset);
515 	os_close_file(fd);
516 
517 	if(err == -EINVAL)
518 		file = ubd_dev->file;
519 	else
520 		file = backing_file;
521 
522 out:
523 	return os_file_size(file, size_out);
524 }
525 
526 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
527 {
528 	int err;
529 
530 	err = os_pread_file(fd, buf, len, offset);
531 	if (err < 0)
532 		return err;
533 
534 	return 0;
535 }
536 
537 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
538 {
539 	time64_t modtime;
540 	unsigned long long actual;
541 	int err;
542 
543 	err = os_file_modtime(file, &modtime);
544 	if (err < 0) {
545 		printk(KERN_ERR "Failed to get modification time of backing "
546 		       "file \"%s\", err = %d\n", file, -err);
547 		return err;
548 	}
549 
550 	err = os_file_size(file, &actual);
551 	if (err < 0) {
552 		printk(KERN_ERR "Failed to get size of backing file \"%s\", "
553 		       "err = %d\n", file, -err);
554 		return err;
555 	}
556 
557 	if (actual != size) {
558 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
559 		 * the typecast.*/
560 		printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
561 		       "vs backing file\n", (unsigned long long) size, actual);
562 		return -EINVAL;
563 	}
564 	if (modtime != mtime) {
565 		printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
566 		       "backing file\n", mtime, modtime);
567 		return -EINVAL;
568 	}
569 	return 0;
570 }
571 
572 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
573 {
574 	struct uml_stat buf1, buf2;
575 	int err;
576 
577 	if (from_cmdline == NULL)
578 		return 0;
579 	if (!strcmp(from_cmdline, from_cow))
580 		return 0;
581 
582 	err = os_stat_file(from_cmdline, &buf1);
583 	if (err < 0) {
584 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
585 		       -err);
586 		return 0;
587 	}
588 	err = os_stat_file(from_cow, &buf2);
589 	if (err < 0) {
590 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
591 		       -err);
592 		return 1;
593 	}
594 	if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
595 		return 0;
596 
597 	printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
598 	       "\"%s\" specified in COW header of \"%s\"\n",
599 	       from_cmdline, from_cow, cow);
600 	return 1;
601 }
602 
603 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
604 		  char **backing_file_out, int *bitmap_offset_out,
605 		  unsigned long *bitmap_len_out, int *data_offset_out,
606 		  int *create_cow_out)
607 {
608 	time64_t mtime;
609 	unsigned long long size;
610 	__u32 version, align;
611 	char *backing_file;
612 	int fd, err, sectorsize, asked_switch, mode = 0644;
613 
614 	fd = os_open_file(file, *openflags, mode);
615 	if (fd < 0) {
616 		if ((fd == -ENOENT) && (create_cow_out != NULL))
617 			*create_cow_out = 1;
618 		if (!openflags->w ||
619 		    ((fd != -EROFS) && (fd != -EACCES)))
620 			return fd;
621 		openflags->w = 0;
622 		fd = os_open_file(file, *openflags, mode);
623 		if (fd < 0)
624 			return fd;
625 	}
626 
627 	if (shared)
628 		printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
629 	else {
630 		err = os_lock_file(fd, openflags->w);
631 		if (err < 0) {
632 			printk(KERN_ERR "Failed to lock '%s', err = %d\n",
633 			       file, -err);
634 			goto out_close;
635 		}
636 	}
637 
638 	/* Successful return case! */
639 	if (backing_file_out == NULL)
640 		return fd;
641 
642 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
643 			      &size, &sectorsize, &align, bitmap_offset_out);
644 	if (err && (*backing_file_out != NULL)) {
645 		printk(KERN_ERR "Failed to read COW header from COW file "
646 		       "\"%s\", errno = %d\n", file, -err);
647 		goto out_close;
648 	}
649 	if (err)
650 		return fd;
651 
652 	asked_switch = path_requires_switch(*backing_file_out, backing_file,
653 					    file);
654 
655 	/* Allow switching only if no mismatch. */
656 	if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
657 						   mtime)) {
658 		printk(KERN_ERR "Switching backing file to '%s'\n",
659 		       *backing_file_out);
660 		err = write_cow_header(file, fd, *backing_file_out,
661 				       sectorsize, align, &size);
662 		if (err) {
663 			printk(KERN_ERR "Switch failed, errno = %d\n", -err);
664 			goto out_close;
665 		}
666 	} else {
667 		*backing_file_out = backing_file;
668 		err = backing_file_mismatch(*backing_file_out, size, mtime);
669 		if (err)
670 			goto out_close;
671 	}
672 
673 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
674 		  bitmap_len_out, data_offset_out);
675 
676 	return fd;
677  out_close:
678 	os_close_file(fd);
679 	return err;
680 }
681 
682 static int create_cow_file(char *cow_file, char *backing_file,
683 		    struct openflags flags,
684 		    int sectorsize, int alignment, int *bitmap_offset_out,
685 		    unsigned long *bitmap_len_out, int *data_offset_out)
686 {
687 	int err, fd;
688 
689 	flags.c = 1;
690 	fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
691 	if (fd < 0) {
692 		err = fd;
693 		printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
694 		       cow_file, -err);
695 		goto out;
696 	}
697 
698 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
699 			    bitmap_offset_out, bitmap_len_out,
700 			    data_offset_out);
701 	if (!err)
702 		return fd;
703 	os_close_file(fd);
704  out:
705 	return err;
706 }
707 
708 static void ubd_close_dev(struct ubd *ubd_dev)
709 {
710 	os_close_file(ubd_dev->fd);
711 	if(ubd_dev->cow.file == NULL)
712 		return;
713 
714 	os_close_file(ubd_dev->cow.fd);
715 	vfree(ubd_dev->cow.bitmap);
716 	ubd_dev->cow.bitmap = NULL;
717 }
718 
719 static int ubd_open_dev(struct ubd *ubd_dev)
720 {
721 	struct openflags flags;
722 	char **back_ptr;
723 	int err, create_cow, *create_ptr;
724 	int fd;
725 
726 	ubd_dev->openflags = ubd_dev->boot_openflags;
727 	create_cow = 0;
728 	create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
729 	back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
730 
731 	fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
732 				back_ptr, &ubd_dev->cow.bitmap_offset,
733 				&ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
734 				create_ptr);
735 
736 	if((fd == -ENOENT) && create_cow){
737 		fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
738 					  ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
739 					  &ubd_dev->cow.bitmap_offset,
740 					  &ubd_dev->cow.bitmap_len,
741 					  &ubd_dev->cow.data_offset);
742 		if(fd >= 0){
743 			printk(KERN_INFO "Creating \"%s\" as COW file for "
744 			       "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
745 		}
746 	}
747 
748 	if(fd < 0){
749 		printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
750 		       -fd);
751 		return fd;
752 	}
753 	ubd_dev->fd = fd;
754 
755 	if(ubd_dev->cow.file != NULL){
756 		err = -ENOMEM;
757 		ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
758 		if(ubd_dev->cow.bitmap == NULL){
759 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
760 			goto error;
761 		}
762 		flush_tlb_kernel_vm();
763 
764 		err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
765 				      ubd_dev->cow.bitmap_offset,
766 				      ubd_dev->cow.bitmap_len);
767 		if(err < 0)
768 			goto error;
769 
770 		flags = ubd_dev->openflags;
771 		flags.w = 0;
772 		err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
773 				    NULL, NULL, NULL, NULL);
774 		if(err < 0) goto error;
775 		ubd_dev->cow.fd = err;
776 	}
777 	return 0;
778  error:
779 	os_close_file(ubd_dev->fd);
780 	return err;
781 }
782 
783 static void ubd_device_release(struct device *dev)
784 {
785 	struct ubd *ubd_dev = dev_get_drvdata(dev);
786 
787 	blk_mq_free_tag_set(&ubd_dev->tag_set);
788 	*ubd_dev = ((struct ubd) DEFAULT_UBD);
789 }
790 
791 static ssize_t serial_show(struct device *dev,
792 			   struct device_attribute *attr, char *buf)
793 {
794 	struct gendisk *disk = dev_to_disk(dev);
795 	struct ubd *ubd_dev = disk->private_data;
796 
797 	if (!ubd_dev)
798 		return 0;
799 
800 	return sprintf(buf, "%s", ubd_dev->serial);
801 }
802 
803 static DEVICE_ATTR_RO(serial);
804 
805 static struct attribute *ubd_attrs[] = {
806 	&dev_attr_serial.attr,
807 	NULL,
808 };
809 
810 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
811 				     struct attribute *a, int n)
812 {
813 	return a->mode;
814 }
815 
816 static const struct attribute_group ubd_attr_group = {
817 	.attrs = ubd_attrs,
818 	.is_visible = ubd_attrs_are_visible,
819 };
820 
821 static const struct attribute_group *ubd_attr_groups[] = {
822 	&ubd_attr_group,
823 	NULL,
824 };
825 
826 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
827 
828 static const struct blk_mq_ops ubd_mq_ops = {
829 	.queue_rq = ubd_queue_rq,
830 };
831 
832 static int ubd_add(int n, char **error_out)
833 {
834 	struct ubd *ubd_dev = &ubd_devs[n];
835 	struct queue_limits lim = {
836 		.max_segments		= MAX_SG,
837 		.seg_boundary_mask	= PAGE_SIZE - 1,
838 	};
839 	struct gendisk *disk;
840 	int err = 0;
841 
842 	if(ubd_dev->file == NULL)
843 		goto out;
844 
845 	if (ubd_dev->cow.file)
846 		lim.max_hw_sectors = 8 * sizeof(long);
847 	if (!ubd_dev->no_trim) {
848 		lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
849 		lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
850 	}
851 
852 	err = ubd_file_size(ubd_dev, &ubd_dev->size);
853 	if(err < 0){
854 		*error_out = "Couldn't determine size of device's file";
855 		goto out;
856 	}
857 
858 	err = ubd_open_dev(ubd_dev);
859 	if (err) {
860 		pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
861 			'a' + n, ubd_dev->file, -err);
862 		goto out;
863 	}
864 
865 	ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
866 
867 	ubd_dev->tag_set.ops = &ubd_mq_ops;
868 	ubd_dev->tag_set.queue_depth = 64;
869 	ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
870 	ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
871 	ubd_dev->tag_set.driver_data = ubd_dev;
872 	ubd_dev->tag_set.nr_hw_queues = 1;
873 
874 	err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
875 	if (err)
876 		goto out_close;
877 
878 	disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
879 	if (IS_ERR(disk)) {
880 		err = PTR_ERR(disk);
881 		goto out_cleanup_tags;
882 	}
883 
884 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
885 	blk_queue_write_cache(disk->queue, true, false);
886 	disk->major = UBD_MAJOR;
887 	disk->first_minor = n << UBD_SHIFT;
888 	disk->minors = 1 << UBD_SHIFT;
889 	disk->fops = &ubd_blops;
890 	set_capacity(disk, ubd_dev->size / 512);
891 	sprintf(disk->disk_name, "ubd%c", 'a' + n);
892 	disk->private_data = ubd_dev;
893 	set_disk_ro(disk, !ubd_dev->openflags.w);
894 
895 	ubd_dev->pdev.id = n;
896 	ubd_dev->pdev.name = DRIVER_NAME;
897 	ubd_dev->pdev.dev.release = ubd_device_release;
898 	dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
899 	platform_device_register(&ubd_dev->pdev);
900 
901 	err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
902 	if (err)
903 		goto out_cleanup_disk;
904 
905 	return 0;
906 
907 out_cleanup_disk:
908 	put_disk(disk);
909 out_cleanup_tags:
910 	blk_mq_free_tag_set(&ubd_dev->tag_set);
911 out_close:
912 	ubd_close_dev(ubd_dev);
913 out:
914 	return err;
915 }
916 
917 static int ubd_config(char *str, char **error_out)
918 {
919 	int n, ret;
920 
921 	/* This string is possibly broken up and stored, so it's only
922 	 * freed if ubd_setup_common fails, or if only general options
923 	 * were set.
924 	 */
925 	str = kstrdup(str, GFP_KERNEL);
926 	if (str == NULL) {
927 		*error_out = "Failed to allocate memory";
928 		return -ENOMEM;
929 	}
930 
931 	ret = ubd_setup_common(str, &n, error_out);
932 	if (ret)
933 		goto err_free;
934 
935 	if (n == -1) {
936 		ret = 0;
937 		goto err_free;
938 	}
939 
940 	mutex_lock(&ubd_lock);
941 	ret = ubd_add(n, error_out);
942 	if (ret)
943 		ubd_devs[n].file = NULL;
944 	mutex_unlock(&ubd_lock);
945 
946 out:
947 	return ret;
948 
949 err_free:
950 	kfree(str);
951 	goto out;
952 }
953 
954 static int ubd_get_config(char *name, char *str, int size, char **error_out)
955 {
956 	struct ubd *ubd_dev;
957 	int n, len = 0;
958 
959 	n = parse_unit(&name);
960 	if((n >= MAX_DEV) || (n < 0)){
961 		*error_out = "ubd_get_config : device number out of range";
962 		return -1;
963 	}
964 
965 	ubd_dev = &ubd_devs[n];
966 	mutex_lock(&ubd_lock);
967 
968 	if(ubd_dev->file == NULL){
969 		CONFIG_CHUNK(str, size, len, "", 1);
970 		goto out;
971 	}
972 
973 	CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
974 
975 	if(ubd_dev->cow.file != NULL){
976 		CONFIG_CHUNK(str, size, len, ",", 0);
977 		CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
978 	}
979 	else CONFIG_CHUNK(str, size, len, "", 1);
980 
981  out:
982 	mutex_unlock(&ubd_lock);
983 	return len;
984 }
985 
986 static int ubd_id(char **str, int *start_out, int *end_out)
987 {
988 	int n;
989 
990 	n = parse_unit(str);
991 	*start_out = 0;
992 	*end_out = MAX_DEV - 1;
993 	return n;
994 }
995 
996 static int ubd_remove(int n, char **error_out)
997 {
998 	struct ubd *ubd_dev;
999 	int err = -ENODEV;
1000 
1001 	mutex_lock(&ubd_lock);
1002 
1003 	ubd_dev = &ubd_devs[n];
1004 
1005 	if(ubd_dev->file == NULL)
1006 		goto out;
1007 
1008 	if (ubd_dev->disk) {
1009 		/* you cannot remove a open disk */
1010 		err = -EBUSY;
1011 		if (disk_openers(ubd_dev->disk))
1012 			goto out;
1013 
1014 		del_gendisk(ubd_dev->disk);
1015 		ubd_close_dev(ubd_dev);
1016 		put_disk(ubd_dev->disk);
1017 	}
1018 
1019 	err = 0;
1020 	platform_device_unregister(&ubd_dev->pdev);
1021 out:
1022 	mutex_unlock(&ubd_lock);
1023 	return err;
1024 }
1025 
1026 /* All these are called by mconsole in process context and without
1027  * ubd-specific locks.  The structure itself is const except for .list.
1028  */
1029 static struct mc_device ubd_mc = {
1030 	.list		= LIST_HEAD_INIT(ubd_mc.list),
1031 	.name		= "ubd",
1032 	.config		= ubd_config,
1033 	.get_config	= ubd_get_config,
1034 	.id		= ubd_id,
1035 	.remove		= ubd_remove,
1036 };
1037 
1038 static int __init ubd_mc_init(void)
1039 {
1040 	mconsole_register_dev(&ubd_mc);
1041 	return 0;
1042 }
1043 
1044 __initcall(ubd_mc_init);
1045 
1046 static int __init ubd0_init(void)
1047 {
1048 	struct ubd *ubd_dev = &ubd_devs[0];
1049 
1050 	mutex_lock(&ubd_lock);
1051 	if(ubd_dev->file == NULL)
1052 		ubd_dev->file = "root_fs";
1053 	mutex_unlock(&ubd_lock);
1054 
1055 	return 0;
1056 }
1057 
1058 __initcall(ubd0_init);
1059 
1060 /* Used in ubd_init, which is an initcall */
1061 static struct platform_driver ubd_driver = {
1062 	.driver = {
1063 		.name  = DRIVER_NAME,
1064 	},
1065 };
1066 
1067 static int __init ubd_init(void)
1068 {
1069 	char *error;
1070 	int i, err;
1071 
1072 	if (register_blkdev(UBD_MAJOR, "ubd"))
1073 		return -1;
1074 
1075 	irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1076 				       sizeof(struct io_thread_req *),
1077 				       GFP_KERNEL
1078 		);
1079 	irq_remainder = 0;
1080 
1081 	if (irq_req_buffer == NULL) {
1082 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1083 		return -ENOMEM;
1084 	}
1085 	io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1086 				      sizeof(struct io_thread_req *),
1087 				      GFP_KERNEL
1088 		);
1089 
1090 	io_remainder = 0;
1091 
1092 	if (io_req_buffer == NULL) {
1093 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1094 		return -ENOMEM;
1095 	}
1096 	platform_driver_register(&ubd_driver);
1097 	mutex_lock(&ubd_lock);
1098 	for (i = 0; i < MAX_DEV; i++){
1099 		err = ubd_add(i, &error);
1100 		if(err)
1101 			printk(KERN_ERR "Failed to initialize ubd device %d :"
1102 			       "%s\n", i, error);
1103 	}
1104 	mutex_unlock(&ubd_lock);
1105 	return 0;
1106 }
1107 
1108 late_initcall(ubd_init);
1109 
1110 static int __init ubd_driver_init(void){
1111 	unsigned long stack;
1112 	int err;
1113 
1114 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1115 	if(global_openflags.s){
1116 		printk(KERN_INFO "ubd: Synchronous mode\n");
1117 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1118 		 * enough. So use anyway the io thread. */
1119 	}
1120 	stack = alloc_stack(0, 0);
1121 	io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1122 	if(io_pid < 0){
1123 		printk(KERN_ERR
1124 		       "ubd : Failed to start I/O thread (errno = %d) - "
1125 		       "falling back to synchronous I/O\n", -io_pid);
1126 		io_pid = -1;
1127 		return 0;
1128 	}
1129 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1130 			     0, "ubd", ubd_devs);
1131 	if(err < 0)
1132 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1133 	return 0;
1134 }
1135 
1136 device_initcall(ubd_driver_init);
1137 
1138 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1139 			  __u64 *cow_offset, unsigned long *bitmap,
1140 			  __u64 bitmap_offset, unsigned long *bitmap_words,
1141 			  __u64 bitmap_len)
1142 {
1143 	__u64 sector = io_offset >> SECTOR_SHIFT;
1144 	int i, update_bitmap = 0;
1145 
1146 	for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1147 		if(cow_mask != NULL)
1148 			ubd_set_bit(i, (unsigned char *) cow_mask);
1149 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1150 			continue;
1151 
1152 		update_bitmap = 1;
1153 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
1154 	}
1155 
1156 	if(!update_bitmap)
1157 		return;
1158 
1159 	*cow_offset = sector / (sizeof(unsigned long) * 8);
1160 
1161 	/* This takes care of the case where we're exactly at the end of the
1162 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
1163 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1164 	 * for the original diagnosis.
1165 	 */
1166 	if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1167 					 sizeof(unsigned long)) - 1))
1168 		(*cow_offset)--;
1169 
1170 	bitmap_words[0] = bitmap[*cow_offset];
1171 	bitmap_words[1] = bitmap[*cow_offset + 1];
1172 
1173 	*cow_offset *= sizeof(unsigned long);
1174 	*cow_offset += bitmap_offset;
1175 }
1176 
1177 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1178 		       unsigned long offset, unsigned long *bitmap,
1179 		       __u64 bitmap_offset, __u64 bitmap_len)
1180 {
1181 	__u64 sector = offset >> SECTOR_SHIFT;
1182 	int i;
1183 
1184 	if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1185 		panic("Operation too long");
1186 
1187 	if (req_op(req->req) == REQ_OP_READ) {
1188 		for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1189 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1190 				ubd_set_bit(i, (unsigned char *)
1191 					    &segment->sector_mask);
1192 		}
1193 	} else {
1194 		cowify_bitmap(offset, segment->length, &segment->sector_mask,
1195 			      &segment->cow_offset, bitmap, bitmap_offset,
1196 			      segment->bitmap_words, bitmap_len);
1197 	}
1198 }
1199 
1200 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1201 			struct request *req)
1202 {
1203 	struct bio_vec bvec;
1204 	struct req_iterator iter;
1205 	int i = 0;
1206 	unsigned long byte_offset = io_req->offset;
1207 	enum req_op op = req_op(req);
1208 
1209 	if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1210 		io_req->io_desc[0].buffer = NULL;
1211 		io_req->io_desc[0].length = blk_rq_bytes(req);
1212 	} else {
1213 		rq_for_each_segment(bvec, req, iter) {
1214 			BUG_ON(i >= io_req->desc_cnt);
1215 
1216 			io_req->io_desc[i].buffer = bvec_virt(&bvec);
1217 			io_req->io_desc[i].length = bvec.bv_len;
1218 			i++;
1219 		}
1220 	}
1221 
1222 	if (dev->cow.file) {
1223 		for (i = 0; i < io_req->desc_cnt; i++) {
1224 			cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1225 				   dev->cow.bitmap, dev->cow.bitmap_offset,
1226 				   dev->cow.bitmap_len);
1227 			byte_offset += io_req->io_desc[i].length;
1228 		}
1229 
1230 	}
1231 }
1232 
1233 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1234 					   int desc_cnt)
1235 {
1236 	struct io_thread_req *io_req;
1237 	int i;
1238 
1239 	io_req = kmalloc(sizeof(*io_req) +
1240 			 (desc_cnt * sizeof(struct io_desc)),
1241 			 GFP_ATOMIC);
1242 	if (!io_req)
1243 		return NULL;
1244 
1245 	io_req->req = req;
1246 	if (dev->cow.file)
1247 		io_req->fds[0] = dev->cow.fd;
1248 	else
1249 		io_req->fds[0] = dev->fd;
1250 	io_req->error = 0;
1251 	io_req->sectorsize = SECTOR_SIZE;
1252 	io_req->fds[1] = dev->fd;
1253 	io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1254 	io_req->offsets[0] = 0;
1255 	io_req->offsets[1] = dev->cow.data_offset;
1256 
1257 	for (i = 0 ; i < desc_cnt; i++) {
1258 		io_req->io_desc[i].sector_mask = 0;
1259 		io_req->io_desc[i].cow_offset = -1;
1260 	}
1261 
1262 	return io_req;
1263 }
1264 
1265 static int ubd_submit_request(struct ubd *dev, struct request *req)
1266 {
1267 	int segs = 0;
1268 	struct io_thread_req *io_req;
1269 	int ret;
1270 	enum req_op op = req_op(req);
1271 
1272 	if (op == REQ_OP_FLUSH)
1273 		segs = 0;
1274 	else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1275 		segs = 1;
1276 	else
1277 		segs = blk_rq_nr_phys_segments(req);
1278 
1279 	io_req = ubd_alloc_req(dev, req, segs);
1280 	if (!io_req)
1281 		return -ENOMEM;
1282 
1283 	io_req->desc_cnt = segs;
1284 	if (segs)
1285 		ubd_map_req(dev, io_req, req);
1286 
1287 	ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1288 	if (ret != sizeof(io_req)) {
1289 		if (ret != -EAGAIN)
1290 			pr_err("write to io thread failed: %d\n", -ret);
1291 		kfree(io_req);
1292 	}
1293 	return ret;
1294 }
1295 
1296 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1297 				 const struct blk_mq_queue_data *bd)
1298 {
1299 	struct ubd *ubd_dev = hctx->queue->queuedata;
1300 	struct request *req = bd->rq;
1301 	int ret = 0, res = BLK_STS_OK;
1302 
1303 	blk_mq_start_request(req);
1304 
1305 	spin_lock_irq(&ubd_dev->lock);
1306 
1307 	switch (req_op(req)) {
1308 	case REQ_OP_FLUSH:
1309 	case REQ_OP_READ:
1310 	case REQ_OP_WRITE:
1311 	case REQ_OP_DISCARD:
1312 	case REQ_OP_WRITE_ZEROES:
1313 		ret = ubd_submit_request(ubd_dev, req);
1314 		break;
1315 	default:
1316 		WARN_ON_ONCE(1);
1317 		res = BLK_STS_NOTSUPP;
1318 	}
1319 
1320 	spin_unlock_irq(&ubd_dev->lock);
1321 
1322 	if (ret < 0) {
1323 		if (ret == -ENOMEM)
1324 			res = BLK_STS_RESOURCE;
1325 		else
1326 			res = BLK_STS_DEV_RESOURCE;
1327 	}
1328 
1329 	return res;
1330 }
1331 
1332 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1333 {
1334 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1335 
1336 	geo->heads = 128;
1337 	geo->sectors = 32;
1338 	geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1339 	return 0;
1340 }
1341 
1342 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1343 		     unsigned int cmd, unsigned long arg)
1344 {
1345 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1346 	u16 ubd_id[ATA_ID_WORDS];
1347 
1348 	switch (cmd) {
1349 		struct cdrom_volctrl volume;
1350 	case HDIO_GET_IDENTITY:
1351 		memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1352 		ubd_id[ATA_ID_CYLS]	= ubd_dev->size / (128 * 32 * 512);
1353 		ubd_id[ATA_ID_HEADS]	= 128;
1354 		ubd_id[ATA_ID_SECTORS]	= 32;
1355 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1356 				 sizeof(ubd_id)))
1357 			return -EFAULT;
1358 		return 0;
1359 
1360 	case CDROMVOLREAD:
1361 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1362 			return -EFAULT;
1363 		volume.channel0 = 255;
1364 		volume.channel1 = 255;
1365 		volume.channel2 = 255;
1366 		volume.channel3 = 255;
1367 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1368 			return -EFAULT;
1369 		return 0;
1370 	}
1371 	return -EINVAL;
1372 }
1373 
1374 static int map_error(int error_code)
1375 {
1376 	switch (error_code) {
1377 	case 0:
1378 		return BLK_STS_OK;
1379 	case ENOSYS:
1380 	case EOPNOTSUPP:
1381 		return BLK_STS_NOTSUPP;
1382 	case ENOSPC:
1383 		return BLK_STS_NOSPC;
1384 	}
1385 	return BLK_STS_IOERR;
1386 }
1387 
1388 /*
1389  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1390  *
1391  * The following functions are part of UML hypervisor code.
1392  * All functions from here onwards are executed as a helper
1393  * thread and are not allowed to execute any kernel functions.
1394  *
1395  * Any communication must occur strictly via shared memory and IPC.
1396  *
1397  * Do not add printks, locks, kernel memory operations, etc - it
1398  * will result in unpredictable behaviour and/or crashes.
1399  */
1400 
1401 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1402 {
1403 	int n;
1404 
1405 	if (segment->cow_offset == -1)
1406 		return map_error(0);
1407 
1408 	n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1409 			  sizeof(segment->bitmap_words), segment->cow_offset);
1410 	if (n != sizeof(segment->bitmap_words))
1411 		return map_error(-n);
1412 
1413 	return map_error(0);
1414 }
1415 
1416 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1417 {
1418 	char *buf = NULL;
1419 	unsigned long len;
1420 	int n, nsectors, start, end, bit;
1421 	__u64 off;
1422 
1423 	/* FLUSH is really a special case, we cannot "case" it with others */
1424 
1425 	if (req_op(req->req) == REQ_OP_FLUSH) {
1426 		/* fds[0] is always either the rw image or our cow file */
1427 		req->error = map_error(-os_sync_file(req->fds[0]));
1428 		return;
1429 	}
1430 
1431 	nsectors = desc->length / req->sectorsize;
1432 	start = 0;
1433 	do {
1434 		bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1435 		end = start;
1436 		while((end < nsectors) &&
1437 		      (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1438 			end++;
1439 
1440 		off = req->offset + req->offsets[bit] +
1441 			start * req->sectorsize;
1442 		len = (end - start) * req->sectorsize;
1443 		if (desc->buffer != NULL)
1444 			buf = &desc->buffer[start * req->sectorsize];
1445 
1446 		switch (req_op(req->req)) {
1447 		case REQ_OP_READ:
1448 			n = 0;
1449 			do {
1450 				buf = &buf[n];
1451 				len -= n;
1452 				n = os_pread_file(req->fds[bit], buf, len, off);
1453 				if (n < 0) {
1454 					req->error = map_error(-n);
1455 					return;
1456 				}
1457 			} while((n < len) && (n != 0));
1458 			if (n < len) memset(&buf[n], 0, len - n);
1459 			break;
1460 		case REQ_OP_WRITE:
1461 			n = os_pwrite_file(req->fds[bit], buf, len, off);
1462 			if(n != len){
1463 				req->error = map_error(-n);
1464 				return;
1465 			}
1466 			break;
1467 		case REQ_OP_DISCARD:
1468 			n = os_falloc_punch(req->fds[bit], off, len);
1469 			if (n) {
1470 				req->error = map_error(-n);
1471 				return;
1472 			}
1473 			break;
1474 		case REQ_OP_WRITE_ZEROES:
1475 			n = os_falloc_zeroes(req->fds[bit], off, len);
1476 			if (n) {
1477 				req->error = map_error(-n);
1478 				return;
1479 			}
1480 			break;
1481 		default:
1482 			WARN_ON_ONCE(1);
1483 			req->error = BLK_STS_NOTSUPP;
1484 			return;
1485 		}
1486 
1487 		start = end;
1488 	} while(start < nsectors);
1489 
1490 	req->offset += len;
1491 	req->error = update_bitmap(req, desc);
1492 }
1493 
1494 /* Changed in start_io_thread, which is serialized by being called only
1495  * from ubd_init, which is an initcall.
1496  */
1497 int kernel_fd = -1;
1498 
1499 /* Only changed by the io thread. XXX: currently unused. */
1500 static int io_count;
1501 
1502 int io_thread(void *arg)
1503 {
1504 	int n, count, written, res;
1505 
1506 	os_fix_helper_signals();
1507 
1508 	while(1){
1509 		n = bulk_req_safe_read(
1510 			kernel_fd,
1511 			io_req_buffer,
1512 			&io_remainder,
1513 			&io_remainder_size,
1514 			UBD_REQ_BUFFER_SIZE
1515 		);
1516 		if (n <= 0) {
1517 			if (n == -EAGAIN)
1518 				ubd_read_poll(-1);
1519 
1520 			continue;
1521 		}
1522 
1523 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1524 			struct io_thread_req *req = (*io_req_buffer)[count];
1525 			int i;
1526 
1527 			io_count++;
1528 			for (i = 0; !req->error && i < req->desc_cnt; i++)
1529 				do_io(req, &(req->io_desc[i]));
1530 
1531 		}
1532 
1533 		written = 0;
1534 
1535 		do {
1536 			res = os_write_file(kernel_fd,
1537 					    ((char *) io_req_buffer) + written,
1538 					    n - written);
1539 			if (res >= 0) {
1540 				written += res;
1541 			}
1542 			if (written < n) {
1543 				ubd_write_poll(-1);
1544 			}
1545 		} while (written < n);
1546 	}
1547 
1548 	return 0;
1549 }
1550