xref: /linux/arch/um/drivers/ubd_kern.c (revision 1cf855ded6962825da8623b467c4650a80776cad)
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 
110 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
111 		     unsigned int cmd, unsigned long arg);
112 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
113 
114 #define MAX_DEV (16)
115 
116 static const struct block_device_operations ubd_blops = {
117         .owner		= THIS_MODULE,
118         .ioctl		= ubd_ioctl,
119         .compat_ioctl	= blkdev_compat_ptr_ioctl,
120 	.getgeo		= ubd_getgeo,
121 };
122 
123 #ifdef CONFIG_BLK_DEV_UBD_SYNC
124 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
125 					 .cl = 1 })
126 #else
127 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
128 					 .cl = 1 })
129 #endif
130 static struct openflags global_openflags = OPEN_FLAGS;
131 
132 struct cow {
133 	/* backing file name */
134 	char *file;
135 	/* backing file fd */
136 	int fd;
137 	unsigned long *bitmap;
138 	unsigned long bitmap_len;
139 	int bitmap_offset;
140 	int data_offset;
141 };
142 
143 #define MAX_SG 64
144 
145 struct ubd {
146 	/* name (and fd, below) of the file opened for writing, either the
147 	 * backing or the cow file. */
148 	char *file;
149 	char *serial;
150 	int fd;
151 	__u64 size;
152 	struct openflags boot_openflags;
153 	struct openflags openflags;
154 	unsigned shared:1;
155 	unsigned no_cow:1;
156 	unsigned no_trim:1;
157 	struct cow cow;
158 	struct platform_device pdev;
159 	struct gendisk *disk;
160 	struct blk_mq_tag_set tag_set;
161 	spinlock_t lock;
162 };
163 
164 #define DEFAULT_COW { \
165 	.file =			NULL, \
166 	.fd =			-1,	\
167 	.bitmap =		NULL, \
168 	.bitmap_offset =	0, \
169 	.data_offset =		0, \
170 }
171 
172 #define DEFAULT_UBD { \
173 	.file = 		NULL, \
174 	.serial =		NULL, \
175 	.fd =			-1, \
176 	.size =			-1, \
177 	.boot_openflags =	OPEN_FLAGS, \
178 	.openflags =		OPEN_FLAGS, \
179 	.no_cow =               0, \
180 	.no_trim =		0, \
181 	.shared =		0, \
182 	.cow =			DEFAULT_COW, \
183 	.lock =			__SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
184 }
185 
186 /* Protected by ubd_lock */
187 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
188 
189 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
190 				 const struct blk_mq_queue_data *bd);
191 
192 static int fake_ide_setup(char *str)
193 {
194 	pr_warn("The fake_ide option has been removed\n");
195 	return 1;
196 }
197 __setup("fake_ide", fake_ide_setup);
198 
199 __uml_help(fake_ide_setup,
200 "fake_ide\n"
201 "    Obsolete stub.\n\n"
202 );
203 
204 static int parse_unit(char **ptr)
205 {
206 	char *str = *ptr, *end;
207 	int n = -1;
208 
209 	if(isdigit(*str)) {
210 		n = simple_strtoul(str, &end, 0);
211 		if(end == str)
212 			return -1;
213 		*ptr = end;
214 	}
215 	else if (('a' <= *str) && (*str <= 'z')) {
216 		n = *str - 'a';
217 		str++;
218 		*ptr = str;
219 	}
220 	return n;
221 }
222 
223 /* If *index_out == -1 at exit, the passed option was a general one;
224  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
225  * should not be freed on exit.
226  */
227 static int ubd_setup_common(char *str, int *index_out, char **error_out)
228 {
229 	struct ubd *ubd_dev;
230 	struct openflags flags = global_openflags;
231 	char *file, *backing_file, *serial;
232 	int n, err = 0, i;
233 
234 	if(index_out) *index_out = -1;
235 	n = *str;
236 	if(n == '='){
237 		str++;
238 		if(!strcmp(str, "sync")){
239 			global_openflags = of_sync(global_openflags);
240 			return err;
241 		}
242 
243 		pr_warn("fake major not supported any more\n");
244 		return 0;
245 	}
246 
247 	n = parse_unit(&str);
248 	if(n < 0){
249 		*error_out = "Couldn't parse device number";
250 		return -EINVAL;
251 	}
252 	if(n >= MAX_DEV){
253 		*error_out = "Device number out of range";
254 		return 1;
255 	}
256 
257 	err = -EBUSY;
258 	mutex_lock(&ubd_lock);
259 
260 	ubd_dev = &ubd_devs[n];
261 	if(ubd_dev->file != NULL){
262 		*error_out = "Device is already configured";
263 		goto out;
264 	}
265 
266 	if (index_out)
267 		*index_out = n;
268 
269 	err = -EINVAL;
270 	for (i = 0; i < sizeof("rscdt="); i++) {
271 		switch (*str) {
272 		case 'r':
273 			flags.w = 0;
274 			break;
275 		case 's':
276 			flags.s = 1;
277 			break;
278 		case 'd':
279 			ubd_dev->no_cow = 1;
280 			break;
281 		case 'c':
282 			ubd_dev->shared = 1;
283 			break;
284 		case 't':
285 			ubd_dev->no_trim = 1;
286 			break;
287 		case '=':
288 			str++;
289 			goto break_loop;
290 		default:
291 			*error_out = "Expected '=' or flag letter "
292 				"(r, s, c, t or d)";
293 			goto out;
294 		}
295 		str++;
296 	}
297 
298 	if (*str == '=')
299 		*error_out = "Too many flags specified";
300 	else
301 		*error_out = "Missing '='";
302 	goto out;
303 
304 break_loop:
305 	file = strsep(&str, ",:");
306 	if (*file == '\0')
307 		file = NULL;
308 
309 	backing_file = strsep(&str, ",:");
310 	if (backing_file && *backing_file == '\0')
311 		backing_file = NULL;
312 
313 	serial = strsep(&str, ",:");
314 	if (serial && *serial == '\0')
315 		serial = NULL;
316 
317 	if (backing_file && ubd_dev->no_cow) {
318 		*error_out = "Can't specify both 'd' and a cow file";
319 		goto out;
320 	}
321 
322 	err = 0;
323 	ubd_dev->file = file;
324 	ubd_dev->cow.file = backing_file;
325 	ubd_dev->serial = serial;
326 	ubd_dev->boot_openflags = flags;
327 out:
328 	mutex_unlock(&ubd_lock);
329 	return err;
330 }
331 
332 static int ubd_setup(char *str)
333 {
334 	char *error;
335 	int err;
336 
337 	err = ubd_setup_common(str, NULL, &error);
338 	if(err)
339 		printk(KERN_ERR "Failed to initialize device with \"%s\" : "
340 		       "%s\n", str, error);
341 	return 1;
342 }
343 
344 __setup("ubd", ubd_setup);
345 __uml_help(ubd_setup,
346 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
347 "    This is used to associate a device with a file in the underlying\n"
348 "    filesystem. When specifying two filenames, the first one is the\n"
349 "    COW name and the second is the backing file name. As separator you can\n"
350 "    use either a ':' or a ',': the first one allows writing things like;\n"
351 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
352 "    while with a ',' the shell would not expand the 2nd '~'.\n"
353 "    When using only one filename, UML will detect whether to treat it like\n"
354 "    a COW file or a backing file. To override this detection, add the 'd'\n"
355 "    flag:\n"
356 "	ubd0d=BackingFile\n"
357 "    Usually, there is a filesystem in the file, but \n"
358 "    that's not required. Swap devices containing swap files can be\n"
359 "    specified like this. Also, a file which doesn't contain a\n"
360 "    filesystem can have its contents read in the virtual \n"
361 "    machine by running 'dd' on the device. <n> must be in the range\n"
362 "    0 to 7. Appending an 'r' to the number will cause that device\n"
363 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
364 "    an 's' will cause data to be written to disk on the host immediately.\n"
365 "    'c' will cause the device to be treated as being shared between multiple\n"
366 "    UMLs and file locking will be turned off - this is appropriate for a\n"
367 "    cluster filesystem and inappropriate at almost all other times.\n\n"
368 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
369 "    An optional device serial number can be exposed using the serial parameter\n"
370 "    on the cmdline which is exposed as a sysfs entry. This is particularly\n"
371 "    useful when a unique number should be given to the device. Note when\n"
372 "    specifying a label, the filename2 must be also presented. It can be\n"
373 "    an empty string, in which case the backing file is not used:\n"
374 "       ubd0=File,,Serial\n"
375 );
376 
377 static int udb_setup(char *str)
378 {
379 	printk("udb%s specified on command line is almost certainly a ubd -> "
380 	       "udb TYPO\n", str);
381 	return 1;
382 }
383 
384 __setup("udb", udb_setup);
385 __uml_help(udb_setup,
386 "udb\n"
387 "    This option is here solely to catch ubd -> udb typos, which can be\n"
388 "    to impossible to catch visually unless you specifically look for\n"
389 "    them.  The only result of any option starting with 'udb' is an error\n"
390 "    in the boot output.\n\n"
391 );
392 
393 /* Only changed by ubd_init, which is an initcall. */
394 static int thread_fd = -1;
395 
396 /* Function to read several request pointers at a time
397 * handling fractional reads if (and as) needed
398 */
399 
400 static int bulk_req_safe_read(
401 	int fd,
402 	struct io_thread_req * (*request_buffer)[],
403 	struct io_thread_req **remainder,
404 	int *remainder_size,
405 	int max_recs
406 	)
407 {
408 	int n = 0;
409 	int res = 0;
410 
411 	if (*remainder_size > 0) {
412 		memmove(
413 			(char *) request_buffer,
414 			(char *) remainder, *remainder_size
415 		);
416 		n = *remainder_size;
417 	}
418 
419 	res = os_read_file(
420 			fd,
421 			((char *) request_buffer) + *remainder_size,
422 			sizeof(struct io_thread_req *)*max_recs
423 				- *remainder_size
424 		);
425 	if (res > 0) {
426 		n += res;
427 		if ((n % sizeof(struct io_thread_req *)) > 0) {
428 			/*
429 			* Read somehow returned not a multiple of dword
430 			* theoretically possible, but never observed in the
431 			* wild, so read routine must be able to handle it
432 			*/
433 			*remainder_size = n % sizeof(struct io_thread_req *);
434 			WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
435 			memmove(
436 				remainder,
437 				((char *) request_buffer) +
438 					(n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
439 				*remainder_size
440 			);
441 			n = n - *remainder_size;
442 		}
443 	} else {
444 		n = res;
445 	}
446 	return n;
447 }
448 
449 /* Called without dev->lock held, and only in interrupt context. */
450 static void ubd_handler(void)
451 {
452 	int n;
453 	int count;
454 
455 	while(1){
456 		n = bulk_req_safe_read(
457 			thread_fd,
458 			irq_req_buffer,
459 			&irq_remainder,
460 			&irq_remainder_size,
461 			UBD_REQ_BUFFER_SIZE
462 		);
463 		if (n < 0) {
464 			if(n == -EAGAIN)
465 				break;
466 			printk(KERN_ERR "spurious interrupt in ubd_handler, "
467 			       "err = %d\n", -n);
468 			return;
469 		}
470 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
471 			struct io_thread_req *io_req = (*irq_req_buffer)[count];
472 
473 			if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
474 				blk_queue_max_discard_sectors(io_req->req->q, 0);
475 				blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
476 			}
477 			blk_mq_end_request(io_req->req, io_req->error);
478 			kfree(io_req);
479 		}
480 	}
481 }
482 
483 static irqreturn_t ubd_intr(int irq, void *dev)
484 {
485 	ubd_handler();
486 	return IRQ_HANDLED;
487 }
488 
489 /* Only changed by ubd_init, which is an initcall. */
490 static int io_pid = -1;
491 
492 static void kill_io_thread(void)
493 {
494 	if(io_pid != -1)
495 		os_kill_process(io_pid, 1);
496 }
497 
498 __uml_exitcall(kill_io_thread);
499 
500 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
501 {
502 	char *file;
503 	int fd;
504 	int err;
505 
506 	__u32 version;
507 	__u32 align;
508 	char *backing_file;
509 	time64_t mtime;
510 	unsigned long long size;
511 	int sector_size;
512 	int bitmap_offset;
513 
514 	if (ubd_dev->file && ubd_dev->cow.file) {
515 		file = ubd_dev->cow.file;
516 
517 		goto out;
518 	}
519 
520 	fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
521 	if (fd < 0)
522 		return fd;
523 
524 	err = read_cow_header(file_reader, &fd, &version, &backing_file, \
525 		&mtime, &size, &sector_size, &align, &bitmap_offset);
526 	os_close_file(fd);
527 
528 	if(err == -EINVAL)
529 		file = ubd_dev->file;
530 	else
531 		file = backing_file;
532 
533 out:
534 	return os_file_size(file, size_out);
535 }
536 
537 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
538 {
539 	int err;
540 
541 	err = os_pread_file(fd, buf, len, offset);
542 	if (err < 0)
543 		return err;
544 
545 	return 0;
546 }
547 
548 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
549 {
550 	time64_t modtime;
551 	unsigned long long actual;
552 	int err;
553 
554 	err = os_file_modtime(file, &modtime);
555 	if (err < 0) {
556 		printk(KERN_ERR "Failed to get modification time of backing "
557 		       "file \"%s\", err = %d\n", file, -err);
558 		return err;
559 	}
560 
561 	err = os_file_size(file, &actual);
562 	if (err < 0) {
563 		printk(KERN_ERR "Failed to get size of backing file \"%s\", "
564 		       "err = %d\n", file, -err);
565 		return err;
566 	}
567 
568 	if (actual != size) {
569 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
570 		 * the typecast.*/
571 		printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
572 		       "vs backing file\n", (unsigned long long) size, actual);
573 		return -EINVAL;
574 	}
575 	if (modtime != mtime) {
576 		printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
577 		       "backing file\n", mtime, modtime);
578 		return -EINVAL;
579 	}
580 	return 0;
581 }
582 
583 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
584 {
585 	struct uml_stat buf1, buf2;
586 	int err;
587 
588 	if (from_cmdline == NULL)
589 		return 0;
590 	if (!strcmp(from_cmdline, from_cow))
591 		return 0;
592 
593 	err = os_stat_file(from_cmdline, &buf1);
594 	if (err < 0) {
595 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
596 		       -err);
597 		return 0;
598 	}
599 	err = os_stat_file(from_cow, &buf2);
600 	if (err < 0) {
601 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
602 		       -err);
603 		return 1;
604 	}
605 	if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
606 		return 0;
607 
608 	printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
609 	       "\"%s\" specified in COW header of \"%s\"\n",
610 	       from_cmdline, from_cow, cow);
611 	return 1;
612 }
613 
614 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
615 		  char **backing_file_out, int *bitmap_offset_out,
616 		  unsigned long *bitmap_len_out, int *data_offset_out,
617 		  int *create_cow_out)
618 {
619 	time64_t mtime;
620 	unsigned long long size;
621 	__u32 version, align;
622 	char *backing_file;
623 	int fd, err, sectorsize, asked_switch, mode = 0644;
624 
625 	fd = os_open_file(file, *openflags, mode);
626 	if (fd < 0) {
627 		if ((fd == -ENOENT) && (create_cow_out != NULL))
628 			*create_cow_out = 1;
629 		if (!openflags->w ||
630 		    ((fd != -EROFS) && (fd != -EACCES)))
631 			return fd;
632 		openflags->w = 0;
633 		fd = os_open_file(file, *openflags, mode);
634 		if (fd < 0)
635 			return fd;
636 	}
637 
638 	if (shared)
639 		printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
640 	else {
641 		err = os_lock_file(fd, openflags->w);
642 		if (err < 0) {
643 			printk(KERN_ERR "Failed to lock '%s', err = %d\n",
644 			       file, -err);
645 			goto out_close;
646 		}
647 	}
648 
649 	/* Successful return case! */
650 	if (backing_file_out == NULL)
651 		return fd;
652 
653 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
654 			      &size, &sectorsize, &align, bitmap_offset_out);
655 	if (err && (*backing_file_out != NULL)) {
656 		printk(KERN_ERR "Failed to read COW header from COW file "
657 		       "\"%s\", errno = %d\n", file, -err);
658 		goto out_close;
659 	}
660 	if (err)
661 		return fd;
662 
663 	asked_switch = path_requires_switch(*backing_file_out, backing_file,
664 					    file);
665 
666 	/* Allow switching only if no mismatch. */
667 	if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
668 						   mtime)) {
669 		printk(KERN_ERR "Switching backing file to '%s'\n",
670 		       *backing_file_out);
671 		err = write_cow_header(file, fd, *backing_file_out,
672 				       sectorsize, align, &size);
673 		if (err) {
674 			printk(KERN_ERR "Switch failed, errno = %d\n", -err);
675 			goto out_close;
676 		}
677 	} else {
678 		*backing_file_out = backing_file;
679 		err = backing_file_mismatch(*backing_file_out, size, mtime);
680 		if (err)
681 			goto out_close;
682 	}
683 
684 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
685 		  bitmap_len_out, data_offset_out);
686 
687 	return fd;
688  out_close:
689 	os_close_file(fd);
690 	return err;
691 }
692 
693 static int create_cow_file(char *cow_file, char *backing_file,
694 		    struct openflags flags,
695 		    int sectorsize, int alignment, int *bitmap_offset_out,
696 		    unsigned long *bitmap_len_out, int *data_offset_out)
697 {
698 	int err, fd;
699 
700 	flags.c = 1;
701 	fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
702 	if (fd < 0) {
703 		err = fd;
704 		printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
705 		       cow_file, -err);
706 		goto out;
707 	}
708 
709 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
710 			    bitmap_offset_out, bitmap_len_out,
711 			    data_offset_out);
712 	if (!err)
713 		return fd;
714 	os_close_file(fd);
715  out:
716 	return err;
717 }
718 
719 static void ubd_close_dev(struct ubd *ubd_dev)
720 {
721 	os_close_file(ubd_dev->fd);
722 	if(ubd_dev->cow.file == NULL)
723 		return;
724 
725 	os_close_file(ubd_dev->cow.fd);
726 	vfree(ubd_dev->cow.bitmap);
727 	ubd_dev->cow.bitmap = NULL;
728 }
729 
730 static int ubd_open_dev(struct ubd *ubd_dev)
731 {
732 	struct openflags flags;
733 	char **back_ptr;
734 	int err, create_cow, *create_ptr;
735 	int fd;
736 
737 	ubd_dev->openflags = ubd_dev->boot_openflags;
738 	create_cow = 0;
739 	create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
740 	back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
741 
742 	fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
743 				back_ptr, &ubd_dev->cow.bitmap_offset,
744 				&ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
745 				create_ptr);
746 
747 	if((fd == -ENOENT) && create_cow){
748 		fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
749 					  ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
750 					  &ubd_dev->cow.bitmap_offset,
751 					  &ubd_dev->cow.bitmap_len,
752 					  &ubd_dev->cow.data_offset);
753 		if(fd >= 0){
754 			printk(KERN_INFO "Creating \"%s\" as COW file for "
755 			       "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
756 		}
757 	}
758 
759 	if(fd < 0){
760 		printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
761 		       -fd);
762 		return fd;
763 	}
764 	ubd_dev->fd = fd;
765 
766 	if(ubd_dev->cow.file != NULL){
767 		err = -ENOMEM;
768 		ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
769 		if(ubd_dev->cow.bitmap == NULL){
770 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
771 			goto error;
772 		}
773 		flush_tlb_kernel_vm();
774 
775 		err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
776 				      ubd_dev->cow.bitmap_offset,
777 				      ubd_dev->cow.bitmap_len);
778 		if(err < 0)
779 			goto error;
780 
781 		flags = ubd_dev->openflags;
782 		flags.w = 0;
783 		err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
784 				    NULL, NULL, NULL, NULL);
785 		if(err < 0) goto error;
786 		ubd_dev->cow.fd = err;
787 	}
788 	return 0;
789  error:
790 	os_close_file(ubd_dev->fd);
791 	return err;
792 }
793 
794 static void ubd_device_release(struct device *dev)
795 {
796 	struct ubd *ubd_dev = dev_get_drvdata(dev);
797 
798 	blk_mq_free_tag_set(&ubd_dev->tag_set);
799 	*ubd_dev = ((struct ubd) DEFAULT_UBD);
800 }
801 
802 static ssize_t serial_show(struct device *dev,
803 			   struct device_attribute *attr, char *buf)
804 {
805 	struct gendisk *disk = dev_to_disk(dev);
806 	struct ubd *ubd_dev = disk->private_data;
807 
808 	if (!ubd_dev)
809 		return 0;
810 
811 	return sprintf(buf, "%s", ubd_dev->serial);
812 }
813 
814 static DEVICE_ATTR_RO(serial);
815 
816 static struct attribute *ubd_attrs[] = {
817 	&dev_attr_serial.attr,
818 	NULL,
819 };
820 
821 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
822 				     struct attribute *a, int n)
823 {
824 	return a->mode;
825 }
826 
827 static const struct attribute_group ubd_attr_group = {
828 	.attrs = ubd_attrs,
829 	.is_visible = ubd_attrs_are_visible,
830 };
831 
832 static const struct attribute_group *ubd_attr_groups[] = {
833 	&ubd_attr_group,
834 	NULL,
835 };
836 
837 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
838 
839 static const struct blk_mq_ops ubd_mq_ops = {
840 	.queue_rq = ubd_queue_rq,
841 };
842 
843 static int ubd_add(int n, char **error_out)
844 {
845 	struct ubd *ubd_dev = &ubd_devs[n];
846 	struct queue_limits lim = {
847 		.max_segments		= MAX_SG,
848 		.seg_boundary_mask	= PAGE_SIZE - 1,
849 	};
850 	struct gendisk *disk;
851 	int err = 0;
852 
853 	if(ubd_dev->file == NULL)
854 		goto out;
855 
856 	if (ubd_dev->cow.file)
857 		lim.max_hw_sectors = 8 * sizeof(long);
858 	if (!ubd_dev->no_trim) {
859 		lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
860 		lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
861 	}
862 
863 	err = ubd_file_size(ubd_dev, &ubd_dev->size);
864 	if(err < 0){
865 		*error_out = "Couldn't determine size of device's file";
866 		goto out;
867 	}
868 
869 	err = ubd_open_dev(ubd_dev);
870 	if (err) {
871 		pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
872 			'a' + n, ubd_dev->file, -err);
873 		goto out;
874 	}
875 
876 	ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
877 
878 	ubd_dev->tag_set.ops = &ubd_mq_ops;
879 	ubd_dev->tag_set.queue_depth = 64;
880 	ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
881 	ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
882 	ubd_dev->tag_set.driver_data = ubd_dev;
883 	ubd_dev->tag_set.nr_hw_queues = 1;
884 
885 	err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
886 	if (err)
887 		goto out_close;
888 
889 	disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
890 	if (IS_ERR(disk)) {
891 		err = PTR_ERR(disk);
892 		goto out_cleanup_tags;
893 	}
894 
895 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
896 	blk_queue_write_cache(disk->queue, true, false);
897 	disk->major = UBD_MAJOR;
898 	disk->first_minor = n << UBD_SHIFT;
899 	disk->minors = 1 << UBD_SHIFT;
900 	disk->fops = &ubd_blops;
901 	set_capacity(disk, ubd_dev->size / 512);
902 	sprintf(disk->disk_name, "ubd%c", 'a' + n);
903 	disk->private_data = ubd_dev;
904 	set_disk_ro(disk, !ubd_dev->openflags.w);
905 
906 	ubd_dev->pdev.id = n;
907 	ubd_dev->pdev.name = DRIVER_NAME;
908 	ubd_dev->pdev.dev.release = ubd_device_release;
909 	dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
910 	platform_device_register(&ubd_dev->pdev);
911 
912 	err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
913 	if (err)
914 		goto out_cleanup_disk;
915 
916 	return 0;
917 
918 out_cleanup_disk:
919 	put_disk(disk);
920 out_cleanup_tags:
921 	blk_mq_free_tag_set(&ubd_dev->tag_set);
922 out_close:
923 	ubd_close_dev(ubd_dev);
924 out:
925 	return err;
926 }
927 
928 static int ubd_config(char *str, char **error_out)
929 {
930 	int n, ret;
931 
932 	/* This string is possibly broken up and stored, so it's only
933 	 * freed if ubd_setup_common fails, or if only general options
934 	 * were set.
935 	 */
936 	str = kstrdup(str, GFP_KERNEL);
937 	if (str == NULL) {
938 		*error_out = "Failed to allocate memory";
939 		return -ENOMEM;
940 	}
941 
942 	ret = ubd_setup_common(str, &n, error_out);
943 	if (ret)
944 		goto err_free;
945 
946 	if (n == -1) {
947 		ret = 0;
948 		goto err_free;
949 	}
950 
951 	mutex_lock(&ubd_lock);
952 	ret = ubd_add(n, error_out);
953 	if (ret)
954 		ubd_devs[n].file = NULL;
955 	mutex_unlock(&ubd_lock);
956 
957 out:
958 	return ret;
959 
960 err_free:
961 	kfree(str);
962 	goto out;
963 }
964 
965 static int ubd_get_config(char *name, char *str, int size, char **error_out)
966 {
967 	struct ubd *ubd_dev;
968 	int n, len = 0;
969 
970 	n = parse_unit(&name);
971 	if((n >= MAX_DEV) || (n < 0)){
972 		*error_out = "ubd_get_config : device number out of range";
973 		return -1;
974 	}
975 
976 	ubd_dev = &ubd_devs[n];
977 	mutex_lock(&ubd_lock);
978 
979 	if(ubd_dev->file == NULL){
980 		CONFIG_CHUNK(str, size, len, "", 1);
981 		goto out;
982 	}
983 
984 	CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
985 
986 	if(ubd_dev->cow.file != NULL){
987 		CONFIG_CHUNK(str, size, len, ",", 0);
988 		CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
989 	}
990 	else CONFIG_CHUNK(str, size, len, "", 1);
991 
992  out:
993 	mutex_unlock(&ubd_lock);
994 	return len;
995 }
996 
997 static int ubd_id(char **str, int *start_out, int *end_out)
998 {
999 	int n;
1000 
1001 	n = parse_unit(str);
1002 	*start_out = 0;
1003 	*end_out = MAX_DEV - 1;
1004 	return n;
1005 }
1006 
1007 static int ubd_remove(int n, char **error_out)
1008 {
1009 	struct ubd *ubd_dev;
1010 	int err = -ENODEV;
1011 
1012 	mutex_lock(&ubd_lock);
1013 
1014 	ubd_dev = &ubd_devs[n];
1015 
1016 	if(ubd_dev->file == NULL)
1017 		goto out;
1018 
1019 	if (ubd_dev->disk) {
1020 		/* you cannot remove a open disk */
1021 		err = -EBUSY;
1022 		if (disk_openers(ubd_dev->disk))
1023 			goto out;
1024 
1025 		del_gendisk(ubd_dev->disk);
1026 		ubd_close_dev(ubd_dev);
1027 		put_disk(ubd_dev->disk);
1028 	}
1029 
1030 	err = 0;
1031 	platform_device_unregister(&ubd_dev->pdev);
1032 out:
1033 	mutex_unlock(&ubd_lock);
1034 	return err;
1035 }
1036 
1037 /* All these are called by mconsole in process context and without
1038  * ubd-specific locks.  The structure itself is const except for .list.
1039  */
1040 static struct mc_device ubd_mc = {
1041 	.list		= LIST_HEAD_INIT(ubd_mc.list),
1042 	.name		= "ubd",
1043 	.config		= ubd_config,
1044 	.get_config	= ubd_get_config,
1045 	.id		= ubd_id,
1046 	.remove		= ubd_remove,
1047 };
1048 
1049 static int __init ubd_mc_init(void)
1050 {
1051 	mconsole_register_dev(&ubd_mc);
1052 	return 0;
1053 }
1054 
1055 __initcall(ubd_mc_init);
1056 
1057 static int __init ubd0_init(void)
1058 {
1059 	struct ubd *ubd_dev = &ubd_devs[0];
1060 
1061 	mutex_lock(&ubd_lock);
1062 	if(ubd_dev->file == NULL)
1063 		ubd_dev->file = "root_fs";
1064 	mutex_unlock(&ubd_lock);
1065 
1066 	return 0;
1067 }
1068 
1069 __initcall(ubd0_init);
1070 
1071 /* Used in ubd_init, which is an initcall */
1072 static struct platform_driver ubd_driver = {
1073 	.driver = {
1074 		.name  = DRIVER_NAME,
1075 	},
1076 };
1077 
1078 static int __init ubd_init(void)
1079 {
1080 	char *error;
1081 	int i, err;
1082 
1083 	if (register_blkdev(UBD_MAJOR, "ubd"))
1084 		return -1;
1085 
1086 	irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1087 				       sizeof(struct io_thread_req *),
1088 				       GFP_KERNEL
1089 		);
1090 	irq_remainder = 0;
1091 
1092 	if (irq_req_buffer == NULL) {
1093 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1094 		return -ENOMEM;
1095 	}
1096 	io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1097 				      sizeof(struct io_thread_req *),
1098 				      GFP_KERNEL
1099 		);
1100 
1101 	io_remainder = 0;
1102 
1103 	if (io_req_buffer == NULL) {
1104 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1105 		return -ENOMEM;
1106 	}
1107 	platform_driver_register(&ubd_driver);
1108 	mutex_lock(&ubd_lock);
1109 	for (i = 0; i < MAX_DEV; i++){
1110 		err = ubd_add(i, &error);
1111 		if(err)
1112 			printk(KERN_ERR "Failed to initialize ubd device %d :"
1113 			       "%s\n", i, error);
1114 	}
1115 	mutex_unlock(&ubd_lock);
1116 	return 0;
1117 }
1118 
1119 late_initcall(ubd_init);
1120 
1121 static int __init ubd_driver_init(void){
1122 	unsigned long stack;
1123 	int err;
1124 
1125 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1126 	if(global_openflags.s){
1127 		printk(KERN_INFO "ubd: Synchronous mode\n");
1128 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1129 		 * enough. So use anyway the io thread. */
1130 	}
1131 	stack = alloc_stack(0, 0);
1132 	io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1133 	if(io_pid < 0){
1134 		printk(KERN_ERR
1135 		       "ubd : Failed to start I/O thread (errno = %d) - "
1136 		       "falling back to synchronous I/O\n", -io_pid);
1137 		io_pid = -1;
1138 		return 0;
1139 	}
1140 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1141 			     0, "ubd", ubd_devs);
1142 	if(err < 0)
1143 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1144 	return 0;
1145 }
1146 
1147 device_initcall(ubd_driver_init);
1148 
1149 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1150 			  __u64 *cow_offset, unsigned long *bitmap,
1151 			  __u64 bitmap_offset, unsigned long *bitmap_words,
1152 			  __u64 bitmap_len)
1153 {
1154 	__u64 sector = io_offset >> SECTOR_SHIFT;
1155 	int i, update_bitmap = 0;
1156 
1157 	for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1158 		if(cow_mask != NULL)
1159 			ubd_set_bit(i, (unsigned char *) cow_mask);
1160 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1161 			continue;
1162 
1163 		update_bitmap = 1;
1164 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
1165 	}
1166 
1167 	if(!update_bitmap)
1168 		return;
1169 
1170 	*cow_offset = sector / (sizeof(unsigned long) * 8);
1171 
1172 	/* This takes care of the case where we're exactly at the end of the
1173 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
1174 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1175 	 * for the original diagnosis.
1176 	 */
1177 	if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1178 					 sizeof(unsigned long)) - 1))
1179 		(*cow_offset)--;
1180 
1181 	bitmap_words[0] = bitmap[*cow_offset];
1182 	bitmap_words[1] = bitmap[*cow_offset + 1];
1183 
1184 	*cow_offset *= sizeof(unsigned long);
1185 	*cow_offset += bitmap_offset;
1186 }
1187 
1188 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1189 		       unsigned long offset, unsigned long *bitmap,
1190 		       __u64 bitmap_offset, __u64 bitmap_len)
1191 {
1192 	__u64 sector = offset >> SECTOR_SHIFT;
1193 	int i;
1194 
1195 	if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1196 		panic("Operation too long");
1197 
1198 	if (req_op(req->req) == REQ_OP_READ) {
1199 		for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1200 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1201 				ubd_set_bit(i, (unsigned char *)
1202 					    &segment->sector_mask);
1203 		}
1204 	} else {
1205 		cowify_bitmap(offset, segment->length, &segment->sector_mask,
1206 			      &segment->cow_offset, bitmap, bitmap_offset,
1207 			      segment->bitmap_words, bitmap_len);
1208 	}
1209 }
1210 
1211 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1212 			struct request *req)
1213 {
1214 	struct bio_vec bvec;
1215 	struct req_iterator iter;
1216 	int i = 0;
1217 	unsigned long byte_offset = io_req->offset;
1218 	enum req_op op = req_op(req);
1219 
1220 	if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1221 		io_req->io_desc[0].buffer = NULL;
1222 		io_req->io_desc[0].length = blk_rq_bytes(req);
1223 	} else {
1224 		rq_for_each_segment(bvec, req, iter) {
1225 			BUG_ON(i >= io_req->desc_cnt);
1226 
1227 			io_req->io_desc[i].buffer = bvec_virt(&bvec);
1228 			io_req->io_desc[i].length = bvec.bv_len;
1229 			i++;
1230 		}
1231 	}
1232 
1233 	if (dev->cow.file) {
1234 		for (i = 0; i < io_req->desc_cnt; i++) {
1235 			cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1236 				   dev->cow.bitmap, dev->cow.bitmap_offset,
1237 				   dev->cow.bitmap_len);
1238 			byte_offset += io_req->io_desc[i].length;
1239 		}
1240 
1241 	}
1242 }
1243 
1244 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1245 					   int desc_cnt)
1246 {
1247 	struct io_thread_req *io_req;
1248 	int i;
1249 
1250 	io_req = kmalloc(sizeof(*io_req) +
1251 			 (desc_cnt * sizeof(struct io_desc)),
1252 			 GFP_ATOMIC);
1253 	if (!io_req)
1254 		return NULL;
1255 
1256 	io_req->req = req;
1257 	if (dev->cow.file)
1258 		io_req->fds[0] = dev->cow.fd;
1259 	else
1260 		io_req->fds[0] = dev->fd;
1261 	io_req->error = 0;
1262 	io_req->sectorsize = SECTOR_SIZE;
1263 	io_req->fds[1] = dev->fd;
1264 	io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1265 	io_req->offsets[0] = 0;
1266 	io_req->offsets[1] = dev->cow.data_offset;
1267 
1268 	for (i = 0 ; i < desc_cnt; i++) {
1269 		io_req->io_desc[i].sector_mask = 0;
1270 		io_req->io_desc[i].cow_offset = -1;
1271 	}
1272 
1273 	return io_req;
1274 }
1275 
1276 static int ubd_submit_request(struct ubd *dev, struct request *req)
1277 {
1278 	int segs = 0;
1279 	struct io_thread_req *io_req;
1280 	int ret;
1281 	enum req_op op = req_op(req);
1282 
1283 	if (op == REQ_OP_FLUSH)
1284 		segs = 0;
1285 	else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1286 		segs = 1;
1287 	else
1288 		segs = blk_rq_nr_phys_segments(req);
1289 
1290 	io_req = ubd_alloc_req(dev, req, segs);
1291 	if (!io_req)
1292 		return -ENOMEM;
1293 
1294 	io_req->desc_cnt = segs;
1295 	if (segs)
1296 		ubd_map_req(dev, io_req, req);
1297 
1298 	ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1299 	if (ret != sizeof(io_req)) {
1300 		if (ret != -EAGAIN)
1301 			pr_err("write to io thread failed: %d\n", -ret);
1302 		kfree(io_req);
1303 	}
1304 	return ret;
1305 }
1306 
1307 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1308 				 const struct blk_mq_queue_data *bd)
1309 {
1310 	struct ubd *ubd_dev = hctx->queue->queuedata;
1311 	struct request *req = bd->rq;
1312 	int ret = 0, res = BLK_STS_OK;
1313 
1314 	blk_mq_start_request(req);
1315 
1316 	spin_lock_irq(&ubd_dev->lock);
1317 
1318 	switch (req_op(req)) {
1319 	case REQ_OP_FLUSH:
1320 	case REQ_OP_READ:
1321 	case REQ_OP_WRITE:
1322 	case REQ_OP_DISCARD:
1323 	case REQ_OP_WRITE_ZEROES:
1324 		ret = ubd_submit_request(ubd_dev, req);
1325 		break;
1326 	default:
1327 		WARN_ON_ONCE(1);
1328 		res = BLK_STS_NOTSUPP;
1329 	}
1330 
1331 	spin_unlock_irq(&ubd_dev->lock);
1332 
1333 	if (ret < 0) {
1334 		if (ret == -ENOMEM)
1335 			res = BLK_STS_RESOURCE;
1336 		else
1337 			res = BLK_STS_DEV_RESOURCE;
1338 	}
1339 
1340 	return res;
1341 }
1342 
1343 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1344 {
1345 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1346 
1347 	geo->heads = 128;
1348 	geo->sectors = 32;
1349 	geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1350 	return 0;
1351 }
1352 
1353 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1354 		     unsigned int cmd, unsigned long arg)
1355 {
1356 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1357 	u16 ubd_id[ATA_ID_WORDS];
1358 
1359 	switch (cmd) {
1360 		struct cdrom_volctrl volume;
1361 	case HDIO_GET_IDENTITY:
1362 		memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1363 		ubd_id[ATA_ID_CYLS]	= ubd_dev->size / (128 * 32 * 512);
1364 		ubd_id[ATA_ID_HEADS]	= 128;
1365 		ubd_id[ATA_ID_SECTORS]	= 32;
1366 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1367 				 sizeof(ubd_id)))
1368 			return -EFAULT;
1369 		return 0;
1370 
1371 	case CDROMVOLREAD:
1372 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1373 			return -EFAULT;
1374 		volume.channel0 = 255;
1375 		volume.channel1 = 255;
1376 		volume.channel2 = 255;
1377 		volume.channel3 = 255;
1378 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1379 			return -EFAULT;
1380 		return 0;
1381 	}
1382 	return -EINVAL;
1383 }
1384 
1385 static int map_error(int error_code)
1386 {
1387 	switch (error_code) {
1388 	case 0:
1389 		return BLK_STS_OK;
1390 	case ENOSYS:
1391 	case EOPNOTSUPP:
1392 		return BLK_STS_NOTSUPP;
1393 	case ENOSPC:
1394 		return BLK_STS_NOSPC;
1395 	}
1396 	return BLK_STS_IOERR;
1397 }
1398 
1399 /*
1400  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1401  *
1402  * The following functions are part of UML hypervisor code.
1403  * All functions from here onwards are executed as a helper
1404  * thread and are not allowed to execute any kernel functions.
1405  *
1406  * Any communication must occur strictly via shared memory and IPC.
1407  *
1408  * Do not add printks, locks, kernel memory operations, etc - it
1409  * will result in unpredictable behaviour and/or crashes.
1410  */
1411 
1412 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1413 {
1414 	int n;
1415 
1416 	if (segment->cow_offset == -1)
1417 		return map_error(0);
1418 
1419 	n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1420 			  sizeof(segment->bitmap_words), segment->cow_offset);
1421 	if (n != sizeof(segment->bitmap_words))
1422 		return map_error(-n);
1423 
1424 	return map_error(0);
1425 }
1426 
1427 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1428 {
1429 	char *buf = NULL;
1430 	unsigned long len;
1431 	int n, nsectors, start, end, bit;
1432 	__u64 off;
1433 
1434 	/* FLUSH is really a special case, we cannot "case" it with others */
1435 
1436 	if (req_op(req->req) == REQ_OP_FLUSH) {
1437 		/* fds[0] is always either the rw image or our cow file */
1438 		req->error = map_error(-os_sync_file(req->fds[0]));
1439 		return;
1440 	}
1441 
1442 	nsectors = desc->length / req->sectorsize;
1443 	start = 0;
1444 	do {
1445 		bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1446 		end = start;
1447 		while((end < nsectors) &&
1448 		      (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1449 			end++;
1450 
1451 		off = req->offset + req->offsets[bit] +
1452 			start * req->sectorsize;
1453 		len = (end - start) * req->sectorsize;
1454 		if (desc->buffer != NULL)
1455 			buf = &desc->buffer[start * req->sectorsize];
1456 
1457 		switch (req_op(req->req)) {
1458 		case REQ_OP_READ:
1459 			n = 0;
1460 			do {
1461 				buf = &buf[n];
1462 				len -= n;
1463 				n = os_pread_file(req->fds[bit], buf, len, off);
1464 				if (n < 0) {
1465 					req->error = map_error(-n);
1466 					return;
1467 				}
1468 			} while((n < len) && (n != 0));
1469 			if (n < len) memset(&buf[n], 0, len - n);
1470 			break;
1471 		case REQ_OP_WRITE:
1472 			n = os_pwrite_file(req->fds[bit], buf, len, off);
1473 			if(n != len){
1474 				req->error = map_error(-n);
1475 				return;
1476 			}
1477 			break;
1478 		case REQ_OP_DISCARD:
1479 			n = os_falloc_punch(req->fds[bit], off, len);
1480 			if (n) {
1481 				req->error = map_error(-n);
1482 				return;
1483 			}
1484 			break;
1485 		case REQ_OP_WRITE_ZEROES:
1486 			n = os_falloc_zeroes(req->fds[bit], off, len);
1487 			if (n) {
1488 				req->error = map_error(-n);
1489 				return;
1490 			}
1491 			break;
1492 		default:
1493 			WARN_ON_ONCE(1);
1494 			req->error = BLK_STS_NOTSUPP;
1495 			return;
1496 		}
1497 
1498 		start = end;
1499 	} while(start < nsectors);
1500 
1501 	req->offset += len;
1502 	req->error = update_bitmap(req, desc);
1503 }
1504 
1505 /* Changed in start_io_thread, which is serialized by being called only
1506  * from ubd_init, which is an initcall.
1507  */
1508 int kernel_fd = -1;
1509 
1510 /* Only changed by the io thread. XXX: currently unused. */
1511 static int io_count;
1512 
1513 int io_thread(void *arg)
1514 {
1515 	int n, count, written, res;
1516 
1517 	os_fix_helper_signals();
1518 
1519 	while(1){
1520 		n = bulk_req_safe_read(
1521 			kernel_fd,
1522 			io_req_buffer,
1523 			&io_remainder,
1524 			&io_remainder_size,
1525 			UBD_REQ_BUFFER_SIZE
1526 		);
1527 		if (n <= 0) {
1528 			if (n == -EAGAIN)
1529 				ubd_read_poll(-1);
1530 
1531 			continue;
1532 		}
1533 
1534 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1535 			struct io_thread_req *req = (*io_req_buffer)[count];
1536 			int i;
1537 
1538 			io_count++;
1539 			for (i = 0; !req->error && i < req->desc_cnt; i++)
1540 				do_io(req, &(req->io_desc[i]));
1541 
1542 		}
1543 
1544 		written = 0;
1545 
1546 		do {
1547 			res = os_write_file(kernel_fd,
1548 					    ((char *) io_req_buffer) + written,
1549 					    n - written);
1550 			if (res >= 0) {
1551 				written += res;
1552 			}
1553 			if (written < n) {
1554 				ubd_write_poll(-1);
1555 			}
1556 		} while (written < n);
1557 	}
1558 
1559 	return 0;
1560 }
1561