xref: /linux/arch/um/drivers/ubd_kern.c (revision 375c4d1583948cf2439833e4a85d5a0aee853895)
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 /* Called without dev->lock held, and only in interrupt context. */
451 static void ubd_handler(void)
452 {
453 	int n;
454 	int count;
455 
456 	while(1){
457 		n = bulk_req_safe_read(
458 			thread_fd,
459 			irq_req_buffer,
460 			&irq_remainder,
461 			&irq_remainder_size,
462 			UBD_REQ_BUFFER_SIZE
463 		);
464 		if (n < 0) {
465 			if(n == -EAGAIN)
466 				break;
467 			printk(KERN_ERR "spurious interrupt in ubd_handler, "
468 			       "err = %d\n", -n);
469 			return;
470 		}
471 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
472 			struct io_thread_req *io_req = (*irq_req_buffer)[count];
473 
474 			if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
475 				blk_queue_max_discard_sectors(io_req->req->q, 0);
476 				blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
477 			}
478 			blk_mq_end_request(io_req->req, io_req->error);
479 			kfree(io_req);
480 		}
481 	}
482 }
483 
484 static irqreturn_t ubd_intr(int irq, void *dev)
485 {
486 	ubd_handler();
487 	return IRQ_HANDLED;
488 }
489 
490 /* Only changed by ubd_init, which is an initcall. */
491 static int io_pid = -1;
492 
493 static void kill_io_thread(void)
494 {
495 	if(io_pid != -1)
496 		os_kill_process(io_pid, 1);
497 }
498 
499 __uml_exitcall(kill_io_thread);
500 
501 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
502 {
503 	char *file;
504 	int fd;
505 	int err;
506 
507 	__u32 version;
508 	__u32 align;
509 	char *backing_file;
510 	time64_t mtime;
511 	unsigned long long size;
512 	int sector_size;
513 	int bitmap_offset;
514 
515 	if (ubd_dev->file && ubd_dev->cow.file) {
516 		file = ubd_dev->cow.file;
517 
518 		goto out;
519 	}
520 
521 	fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
522 	if (fd < 0)
523 		return fd;
524 
525 	err = read_cow_header(file_reader, &fd, &version, &backing_file, \
526 		&mtime, &size, &sector_size, &align, &bitmap_offset);
527 	os_close_file(fd);
528 
529 	if(err == -EINVAL)
530 		file = ubd_dev->file;
531 	else
532 		file = backing_file;
533 
534 out:
535 	return os_file_size(file, size_out);
536 }
537 
538 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
539 {
540 	int err;
541 
542 	err = os_pread_file(fd, buf, len, offset);
543 	if (err < 0)
544 		return err;
545 
546 	return 0;
547 }
548 
549 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
550 {
551 	time64_t modtime;
552 	unsigned long long actual;
553 	int err;
554 
555 	err = os_file_modtime(file, &modtime);
556 	if (err < 0) {
557 		printk(KERN_ERR "Failed to get modification time of backing "
558 		       "file \"%s\", err = %d\n", file, -err);
559 		return err;
560 	}
561 
562 	err = os_file_size(file, &actual);
563 	if (err < 0) {
564 		printk(KERN_ERR "Failed to get size of backing file \"%s\", "
565 		       "err = %d\n", file, -err);
566 		return err;
567 	}
568 
569 	if (actual != size) {
570 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
571 		 * the typecast.*/
572 		printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
573 		       "vs backing file\n", (unsigned long long) size, actual);
574 		return -EINVAL;
575 	}
576 	if (modtime != mtime) {
577 		printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
578 		       "backing file\n", mtime, modtime);
579 		return -EINVAL;
580 	}
581 	return 0;
582 }
583 
584 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
585 {
586 	struct uml_stat buf1, buf2;
587 	int err;
588 
589 	if (from_cmdline == NULL)
590 		return 0;
591 	if (!strcmp(from_cmdline, from_cow))
592 		return 0;
593 
594 	err = os_stat_file(from_cmdline, &buf1);
595 	if (err < 0) {
596 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
597 		       -err);
598 		return 0;
599 	}
600 	err = os_stat_file(from_cow, &buf2);
601 	if (err < 0) {
602 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
603 		       -err);
604 		return 1;
605 	}
606 	if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
607 		return 0;
608 
609 	printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
610 	       "\"%s\" specified in COW header of \"%s\"\n",
611 	       from_cmdline, from_cow, cow);
612 	return 1;
613 }
614 
615 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
616 		  char **backing_file_out, int *bitmap_offset_out,
617 		  unsigned long *bitmap_len_out, int *data_offset_out,
618 		  int *create_cow_out)
619 {
620 	time64_t mtime;
621 	unsigned long long size;
622 	__u32 version, align;
623 	char *backing_file;
624 	int fd, err, sectorsize, asked_switch, mode = 0644;
625 
626 	fd = os_open_file(file, *openflags, mode);
627 	if (fd < 0) {
628 		if ((fd == -ENOENT) && (create_cow_out != NULL))
629 			*create_cow_out = 1;
630 		if (!openflags->w ||
631 		    ((fd != -EROFS) && (fd != -EACCES)))
632 			return fd;
633 		openflags->w = 0;
634 		fd = os_open_file(file, *openflags, mode);
635 		if (fd < 0)
636 			return fd;
637 	}
638 
639 	if (shared)
640 		printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
641 	else {
642 		err = os_lock_file(fd, openflags->w);
643 		if (err < 0) {
644 			printk(KERN_ERR "Failed to lock '%s', err = %d\n",
645 			       file, -err);
646 			goto out_close;
647 		}
648 	}
649 
650 	/* Successful return case! */
651 	if (backing_file_out == NULL)
652 		return fd;
653 
654 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
655 			      &size, &sectorsize, &align, bitmap_offset_out);
656 	if (err && (*backing_file_out != NULL)) {
657 		printk(KERN_ERR "Failed to read COW header from COW file "
658 		       "\"%s\", errno = %d\n", file, -err);
659 		goto out_close;
660 	}
661 	if (err)
662 		return fd;
663 
664 	asked_switch = path_requires_switch(*backing_file_out, backing_file,
665 					    file);
666 
667 	/* Allow switching only if no mismatch. */
668 	if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
669 						   mtime)) {
670 		printk(KERN_ERR "Switching backing file to '%s'\n",
671 		       *backing_file_out);
672 		err = write_cow_header(file, fd, *backing_file_out,
673 				       sectorsize, align, &size);
674 		if (err) {
675 			printk(KERN_ERR "Switch failed, errno = %d\n", -err);
676 			goto out_close;
677 		}
678 	} else {
679 		*backing_file_out = backing_file;
680 		err = backing_file_mismatch(*backing_file_out, size, mtime);
681 		if (err)
682 			goto out_close;
683 	}
684 
685 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
686 		  bitmap_len_out, data_offset_out);
687 
688 	return fd;
689  out_close:
690 	os_close_file(fd);
691 	return err;
692 }
693 
694 static int create_cow_file(char *cow_file, char *backing_file,
695 		    struct openflags flags,
696 		    int sectorsize, int alignment, int *bitmap_offset_out,
697 		    unsigned long *bitmap_len_out, int *data_offset_out)
698 {
699 	int err, fd;
700 
701 	flags.c = 1;
702 	fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
703 	if (fd < 0) {
704 		err = fd;
705 		printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
706 		       cow_file, -err);
707 		goto out;
708 	}
709 
710 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
711 			    bitmap_offset_out, bitmap_len_out,
712 			    data_offset_out);
713 	if (!err)
714 		return fd;
715 	os_close_file(fd);
716  out:
717 	return err;
718 }
719 
720 static void ubd_close_dev(struct ubd *ubd_dev)
721 {
722 	os_close_file(ubd_dev->fd);
723 	if(ubd_dev->cow.file == NULL)
724 		return;
725 
726 	os_close_file(ubd_dev->cow.fd);
727 	vfree(ubd_dev->cow.bitmap);
728 	ubd_dev->cow.bitmap = NULL;
729 }
730 
731 static int ubd_open_dev(struct ubd *ubd_dev)
732 {
733 	struct openflags flags;
734 	char **back_ptr;
735 	int err, create_cow, *create_ptr;
736 	int fd;
737 
738 	ubd_dev->openflags = ubd_dev->boot_openflags;
739 	create_cow = 0;
740 	create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
741 	back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
742 
743 	fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
744 				back_ptr, &ubd_dev->cow.bitmap_offset,
745 				&ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
746 				create_ptr);
747 
748 	if((fd == -ENOENT) && create_cow){
749 		fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
750 					  ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
751 					  &ubd_dev->cow.bitmap_offset,
752 					  &ubd_dev->cow.bitmap_len,
753 					  &ubd_dev->cow.data_offset);
754 		if(fd >= 0){
755 			printk(KERN_INFO "Creating \"%s\" as COW file for "
756 			       "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
757 		}
758 	}
759 
760 	if(fd < 0){
761 		printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
762 		       -fd);
763 		return fd;
764 	}
765 	ubd_dev->fd = fd;
766 
767 	if(ubd_dev->cow.file != NULL){
768 		err = -ENOMEM;
769 		ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
770 		if(ubd_dev->cow.bitmap == NULL){
771 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
772 			goto error;
773 		}
774 		flush_tlb_kernel_vm();
775 
776 		err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
777 				      ubd_dev->cow.bitmap_offset,
778 				      ubd_dev->cow.bitmap_len);
779 		if(err < 0)
780 			goto error;
781 
782 		flags = ubd_dev->openflags;
783 		flags.w = 0;
784 		err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
785 				    NULL, NULL, NULL, NULL);
786 		if(err < 0) goto error;
787 		ubd_dev->cow.fd = err;
788 	}
789 	return 0;
790  error:
791 	os_close_file(ubd_dev->fd);
792 	return err;
793 }
794 
795 static void ubd_device_release(struct device *dev)
796 {
797 	struct ubd *ubd_dev = dev_get_drvdata(dev);
798 
799 	blk_mq_free_tag_set(&ubd_dev->tag_set);
800 	*ubd_dev = ((struct ubd) DEFAULT_UBD);
801 }
802 
803 static ssize_t serial_show(struct device *dev,
804 			   struct device_attribute *attr, char *buf)
805 {
806 	struct gendisk *disk = dev_to_disk(dev);
807 	struct ubd *ubd_dev = disk->private_data;
808 
809 	if (!ubd_dev)
810 		return 0;
811 
812 	return sprintf(buf, "%s", ubd_dev->serial);
813 }
814 
815 static DEVICE_ATTR_RO(serial);
816 
817 static struct attribute *ubd_attrs[] = {
818 	&dev_attr_serial.attr,
819 	NULL,
820 };
821 
822 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
823 				     struct attribute *a, int n)
824 {
825 	return a->mode;
826 }
827 
828 static const struct attribute_group ubd_attr_group = {
829 	.attrs = ubd_attrs,
830 	.is_visible = ubd_attrs_are_visible,
831 };
832 
833 static const struct attribute_group *ubd_attr_groups[] = {
834 	&ubd_attr_group,
835 	NULL,
836 };
837 
838 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
839 
840 static const struct blk_mq_ops ubd_mq_ops = {
841 	.queue_rq = ubd_queue_rq,
842 };
843 
844 static int ubd_add(int n, char **error_out)
845 {
846 	struct ubd *ubd_dev = &ubd_devs[n];
847 	struct queue_limits lim = {
848 		.max_segments		= MAX_SG,
849 		.seg_boundary_mask	= PAGE_SIZE - 1,
850 	};
851 	struct gendisk *disk;
852 	int err = 0;
853 
854 	if(ubd_dev->file == NULL)
855 		goto out;
856 
857 	if (ubd_dev->cow.file)
858 		lim.max_hw_sectors = 8 * sizeof(long);
859 	if (!ubd_dev->no_trim) {
860 		lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
861 		lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
862 	}
863 
864 	err = ubd_file_size(ubd_dev, &ubd_dev->size);
865 	if(err < 0){
866 		*error_out = "Couldn't determine size of device's file";
867 		goto out;
868 	}
869 
870 	err = ubd_open_dev(ubd_dev);
871 	if (err) {
872 		pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
873 			'a' + n, ubd_dev->file, -err);
874 		goto out;
875 	}
876 
877 	ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
878 
879 	ubd_dev->tag_set.ops = &ubd_mq_ops;
880 	ubd_dev->tag_set.queue_depth = 64;
881 	ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
882 	ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
883 	ubd_dev->tag_set.driver_data = ubd_dev;
884 	ubd_dev->tag_set.nr_hw_queues = 1;
885 
886 	err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
887 	if (err)
888 		goto out_close;
889 
890 	disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
891 	if (IS_ERR(disk)) {
892 		err = PTR_ERR(disk);
893 		goto out_cleanup_tags;
894 	}
895 
896 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
897 	blk_queue_write_cache(disk->queue, true, false);
898 	disk->major = UBD_MAJOR;
899 	disk->first_minor = n << UBD_SHIFT;
900 	disk->minors = 1 << UBD_SHIFT;
901 	disk->fops = &ubd_blops;
902 	set_capacity(disk, ubd_dev->size / 512);
903 	sprintf(disk->disk_name, "ubd%c", 'a' + n);
904 	disk->private_data = ubd_dev;
905 	set_disk_ro(disk, !ubd_dev->openflags.w);
906 
907 	ubd_dev->pdev.id = n;
908 	ubd_dev->pdev.name = DRIVER_NAME;
909 	ubd_dev->pdev.dev.release = ubd_device_release;
910 	dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
911 	platform_device_register(&ubd_dev->pdev);
912 
913 	err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
914 	if (err)
915 		goto out_cleanup_disk;
916 
917 	return 0;
918 
919 out_cleanup_disk:
920 	put_disk(disk);
921 out_cleanup_tags:
922 	blk_mq_free_tag_set(&ubd_dev->tag_set);
923 out_close:
924 	ubd_close_dev(ubd_dev);
925 out:
926 	return err;
927 }
928 
929 static int ubd_config(char *str, char **error_out)
930 {
931 	int n, ret;
932 
933 	/* This string is possibly broken up and stored, so it's only
934 	 * freed if ubd_setup_common fails, or if only general options
935 	 * were set.
936 	 */
937 	str = kstrdup(str, GFP_KERNEL);
938 	if (str == NULL) {
939 		*error_out = "Failed to allocate memory";
940 		return -ENOMEM;
941 	}
942 
943 	ret = ubd_setup_common(str, &n, error_out);
944 	if (ret)
945 		goto err_free;
946 
947 	if (n == -1) {
948 		ret = 0;
949 		goto err_free;
950 	}
951 
952 	mutex_lock(&ubd_lock);
953 	ret = ubd_add(n, error_out);
954 	if (ret)
955 		ubd_devs[n].file = NULL;
956 	mutex_unlock(&ubd_lock);
957 
958 out:
959 	return ret;
960 
961 err_free:
962 	kfree(str);
963 	goto out;
964 }
965 
966 static int ubd_get_config(char *name, char *str, int size, char **error_out)
967 {
968 	struct ubd *ubd_dev;
969 	int n, len = 0;
970 
971 	n = parse_unit(&name);
972 	if((n >= MAX_DEV) || (n < 0)){
973 		*error_out = "ubd_get_config : device number out of range";
974 		return -1;
975 	}
976 
977 	ubd_dev = &ubd_devs[n];
978 	mutex_lock(&ubd_lock);
979 
980 	if(ubd_dev->file == NULL){
981 		CONFIG_CHUNK(str, size, len, "", 1);
982 		goto out;
983 	}
984 
985 	CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
986 
987 	if(ubd_dev->cow.file != NULL){
988 		CONFIG_CHUNK(str, size, len, ",", 0);
989 		CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
990 	}
991 	else CONFIG_CHUNK(str, size, len, "", 1);
992 
993  out:
994 	mutex_unlock(&ubd_lock);
995 	return len;
996 }
997 
998 static int ubd_id(char **str, int *start_out, int *end_out)
999 {
1000 	int n;
1001 
1002 	n = parse_unit(str);
1003 	*start_out = 0;
1004 	*end_out = MAX_DEV - 1;
1005 	return n;
1006 }
1007 
1008 static int ubd_remove(int n, char **error_out)
1009 {
1010 	struct ubd *ubd_dev;
1011 	int err = -ENODEV;
1012 
1013 	mutex_lock(&ubd_lock);
1014 
1015 	ubd_dev = &ubd_devs[n];
1016 
1017 	if(ubd_dev->file == NULL)
1018 		goto out;
1019 
1020 	if (ubd_dev->disk) {
1021 		/* you cannot remove a open disk */
1022 		err = -EBUSY;
1023 		if (disk_openers(ubd_dev->disk))
1024 			goto out;
1025 
1026 		del_gendisk(ubd_dev->disk);
1027 		ubd_close_dev(ubd_dev);
1028 		put_disk(ubd_dev->disk);
1029 	}
1030 
1031 	err = 0;
1032 	platform_device_unregister(&ubd_dev->pdev);
1033 out:
1034 	mutex_unlock(&ubd_lock);
1035 	return err;
1036 }
1037 
1038 /* All these are called by mconsole in process context and without
1039  * ubd-specific locks.  The structure itself is const except for .list.
1040  */
1041 static struct mc_device ubd_mc = {
1042 	.list		= LIST_HEAD_INIT(ubd_mc.list),
1043 	.name		= "ubd",
1044 	.config		= ubd_config,
1045 	.get_config	= ubd_get_config,
1046 	.id		= ubd_id,
1047 	.remove		= ubd_remove,
1048 };
1049 
1050 static int __init ubd_mc_init(void)
1051 {
1052 	mconsole_register_dev(&ubd_mc);
1053 	return 0;
1054 }
1055 
1056 __initcall(ubd_mc_init);
1057 
1058 static int __init ubd0_init(void)
1059 {
1060 	struct ubd *ubd_dev = &ubd_devs[0];
1061 
1062 	mutex_lock(&ubd_lock);
1063 	if(ubd_dev->file == NULL)
1064 		ubd_dev->file = "root_fs";
1065 	mutex_unlock(&ubd_lock);
1066 
1067 	return 0;
1068 }
1069 
1070 __initcall(ubd0_init);
1071 
1072 /* Used in ubd_init, which is an initcall */
1073 static struct platform_driver ubd_driver = {
1074 	.driver = {
1075 		.name  = DRIVER_NAME,
1076 	},
1077 };
1078 
1079 static int __init ubd_init(void)
1080 {
1081 	char *error;
1082 	int i, err;
1083 
1084 	if (register_blkdev(UBD_MAJOR, "ubd"))
1085 		return -1;
1086 
1087 	irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1088 				       sizeof(struct io_thread_req *),
1089 				       GFP_KERNEL
1090 		);
1091 	irq_remainder = 0;
1092 
1093 	if (irq_req_buffer == NULL) {
1094 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1095 		return -ENOMEM;
1096 	}
1097 	io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1098 				      sizeof(struct io_thread_req *),
1099 				      GFP_KERNEL
1100 		);
1101 
1102 	io_remainder = 0;
1103 
1104 	if (io_req_buffer == NULL) {
1105 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1106 		return -ENOMEM;
1107 	}
1108 	platform_driver_register(&ubd_driver);
1109 	mutex_lock(&ubd_lock);
1110 	for (i = 0; i < MAX_DEV; i++){
1111 		err = ubd_add(i, &error);
1112 		if(err)
1113 			printk(KERN_ERR "Failed to initialize ubd device %d :"
1114 			       "%s\n", i, error);
1115 	}
1116 	mutex_unlock(&ubd_lock);
1117 	return 0;
1118 }
1119 
1120 late_initcall(ubd_init);
1121 
1122 static int __init ubd_driver_init(void){
1123 	unsigned long stack;
1124 	int err;
1125 
1126 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1127 	if(global_openflags.s){
1128 		printk(KERN_INFO "ubd: Synchronous mode\n");
1129 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1130 		 * enough. So use anyway the io thread. */
1131 	}
1132 	stack = alloc_stack(0, 0);
1133 	io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1134 	if(io_pid < 0){
1135 		printk(KERN_ERR
1136 		       "ubd : Failed to start I/O thread (errno = %d) - "
1137 		       "falling back to synchronous I/O\n", -io_pid);
1138 		io_pid = -1;
1139 		return 0;
1140 	}
1141 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1142 			     0, "ubd", ubd_devs);
1143 	if(err < 0)
1144 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1145 	return 0;
1146 }
1147 
1148 device_initcall(ubd_driver_init);
1149 
1150 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1151 			  __u64 *cow_offset, unsigned long *bitmap,
1152 			  __u64 bitmap_offset, unsigned long *bitmap_words,
1153 			  __u64 bitmap_len)
1154 {
1155 	__u64 sector = io_offset >> SECTOR_SHIFT;
1156 	int i, update_bitmap = 0;
1157 
1158 	for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1159 		if(cow_mask != NULL)
1160 			ubd_set_bit(i, (unsigned char *) cow_mask);
1161 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1162 			continue;
1163 
1164 		update_bitmap = 1;
1165 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
1166 	}
1167 
1168 	if(!update_bitmap)
1169 		return;
1170 
1171 	*cow_offset = sector / (sizeof(unsigned long) * 8);
1172 
1173 	/* This takes care of the case where we're exactly at the end of the
1174 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
1175 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1176 	 * for the original diagnosis.
1177 	 */
1178 	if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1179 					 sizeof(unsigned long)) - 1))
1180 		(*cow_offset)--;
1181 
1182 	bitmap_words[0] = bitmap[*cow_offset];
1183 	bitmap_words[1] = bitmap[*cow_offset + 1];
1184 
1185 	*cow_offset *= sizeof(unsigned long);
1186 	*cow_offset += bitmap_offset;
1187 }
1188 
1189 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1190 		       unsigned long offset, unsigned long *bitmap,
1191 		       __u64 bitmap_offset, __u64 bitmap_len)
1192 {
1193 	__u64 sector = offset >> SECTOR_SHIFT;
1194 	int i;
1195 
1196 	if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1197 		panic("Operation too long");
1198 
1199 	if (req_op(req->req) == REQ_OP_READ) {
1200 		for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1201 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1202 				ubd_set_bit(i, (unsigned char *)
1203 					    &segment->sector_mask);
1204 		}
1205 	} else {
1206 		cowify_bitmap(offset, segment->length, &segment->sector_mask,
1207 			      &segment->cow_offset, bitmap, bitmap_offset,
1208 			      segment->bitmap_words, bitmap_len);
1209 	}
1210 }
1211 
1212 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1213 			struct request *req)
1214 {
1215 	struct bio_vec bvec;
1216 	struct req_iterator iter;
1217 	int i = 0;
1218 	unsigned long byte_offset = io_req->offset;
1219 	enum req_op op = req_op(req);
1220 
1221 	if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1222 		io_req->io_desc[0].buffer = NULL;
1223 		io_req->io_desc[0].length = blk_rq_bytes(req);
1224 	} else {
1225 		rq_for_each_segment(bvec, req, iter) {
1226 			BUG_ON(i >= io_req->desc_cnt);
1227 
1228 			io_req->io_desc[i].buffer = bvec_virt(&bvec);
1229 			io_req->io_desc[i].length = bvec.bv_len;
1230 			i++;
1231 		}
1232 	}
1233 
1234 	if (dev->cow.file) {
1235 		for (i = 0; i < io_req->desc_cnt; i++) {
1236 			cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1237 				   dev->cow.bitmap, dev->cow.bitmap_offset,
1238 				   dev->cow.bitmap_len);
1239 			byte_offset += io_req->io_desc[i].length;
1240 		}
1241 
1242 	}
1243 }
1244 
1245 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1246 					   int desc_cnt)
1247 {
1248 	struct io_thread_req *io_req;
1249 	int i;
1250 
1251 	io_req = kmalloc(sizeof(*io_req) +
1252 			 (desc_cnt * sizeof(struct io_desc)),
1253 			 GFP_ATOMIC);
1254 	if (!io_req)
1255 		return NULL;
1256 
1257 	io_req->req = req;
1258 	if (dev->cow.file)
1259 		io_req->fds[0] = dev->cow.fd;
1260 	else
1261 		io_req->fds[0] = dev->fd;
1262 	io_req->error = 0;
1263 	io_req->sectorsize = SECTOR_SIZE;
1264 	io_req->fds[1] = dev->fd;
1265 	io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1266 	io_req->offsets[0] = 0;
1267 	io_req->offsets[1] = dev->cow.data_offset;
1268 
1269 	for (i = 0 ; i < desc_cnt; i++) {
1270 		io_req->io_desc[i].sector_mask = 0;
1271 		io_req->io_desc[i].cow_offset = -1;
1272 	}
1273 
1274 	return io_req;
1275 }
1276 
1277 static int ubd_submit_request(struct ubd *dev, struct request *req)
1278 {
1279 	int segs = 0;
1280 	struct io_thread_req *io_req;
1281 	int ret;
1282 	enum req_op op = req_op(req);
1283 
1284 	if (op == REQ_OP_FLUSH)
1285 		segs = 0;
1286 	else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1287 		segs = 1;
1288 	else
1289 		segs = blk_rq_nr_phys_segments(req);
1290 
1291 	io_req = ubd_alloc_req(dev, req, segs);
1292 	if (!io_req)
1293 		return -ENOMEM;
1294 
1295 	io_req->desc_cnt = segs;
1296 	if (segs)
1297 		ubd_map_req(dev, io_req, req);
1298 
1299 	ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1300 	if (ret != sizeof(io_req)) {
1301 		if (ret != -EAGAIN)
1302 			pr_err("write to io thread failed: %d\n", -ret);
1303 		kfree(io_req);
1304 	}
1305 	return ret;
1306 }
1307 
1308 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1309 				 const struct blk_mq_queue_data *bd)
1310 {
1311 	struct ubd *ubd_dev = hctx->queue->queuedata;
1312 	struct request *req = bd->rq;
1313 	int ret = 0, res = BLK_STS_OK;
1314 
1315 	blk_mq_start_request(req);
1316 
1317 	spin_lock_irq(&ubd_dev->lock);
1318 
1319 	switch (req_op(req)) {
1320 	case REQ_OP_FLUSH:
1321 	case REQ_OP_READ:
1322 	case REQ_OP_WRITE:
1323 	case REQ_OP_DISCARD:
1324 	case REQ_OP_WRITE_ZEROES:
1325 		ret = ubd_submit_request(ubd_dev, req);
1326 		break;
1327 	default:
1328 		WARN_ON_ONCE(1);
1329 		res = BLK_STS_NOTSUPP;
1330 	}
1331 
1332 	spin_unlock_irq(&ubd_dev->lock);
1333 
1334 	if (ret < 0) {
1335 		if (ret == -ENOMEM)
1336 			res = BLK_STS_RESOURCE;
1337 		else
1338 			res = BLK_STS_DEV_RESOURCE;
1339 	}
1340 
1341 	return res;
1342 }
1343 
1344 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1345 {
1346 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1347 
1348 	geo->heads = 128;
1349 	geo->sectors = 32;
1350 	geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1351 	return 0;
1352 }
1353 
1354 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1355 		     unsigned int cmd, unsigned long arg)
1356 {
1357 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1358 	u16 ubd_id[ATA_ID_WORDS];
1359 
1360 	switch (cmd) {
1361 		struct cdrom_volctrl volume;
1362 	case HDIO_GET_IDENTITY:
1363 		memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1364 		ubd_id[ATA_ID_CYLS]	= ubd_dev->size / (128 * 32 * 512);
1365 		ubd_id[ATA_ID_HEADS]	= 128;
1366 		ubd_id[ATA_ID_SECTORS]	= 32;
1367 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1368 				 sizeof(ubd_id)))
1369 			return -EFAULT;
1370 		return 0;
1371 
1372 	case CDROMVOLREAD:
1373 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1374 			return -EFAULT;
1375 		volume.channel0 = 255;
1376 		volume.channel1 = 255;
1377 		volume.channel2 = 255;
1378 		volume.channel3 = 255;
1379 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1380 			return -EFAULT;
1381 		return 0;
1382 	}
1383 	return -EINVAL;
1384 }
1385 
1386 static int map_error(int error_code)
1387 {
1388 	switch (error_code) {
1389 	case 0:
1390 		return BLK_STS_OK;
1391 	case ENOSYS:
1392 	case EOPNOTSUPP:
1393 		return BLK_STS_NOTSUPP;
1394 	case ENOSPC:
1395 		return BLK_STS_NOSPC;
1396 	}
1397 	return BLK_STS_IOERR;
1398 }
1399 
1400 /*
1401  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1402  *
1403  * The following functions are part of UML hypervisor code.
1404  * All functions from here onwards are executed as a helper
1405  * thread and are not allowed to execute any kernel functions.
1406  *
1407  * Any communication must occur strictly via shared memory and IPC.
1408  *
1409  * Do not add printks, locks, kernel memory operations, etc - it
1410  * will result in unpredictable behaviour and/or crashes.
1411  */
1412 
1413 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1414 {
1415 	int n;
1416 
1417 	if (segment->cow_offset == -1)
1418 		return map_error(0);
1419 
1420 	n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1421 			  sizeof(segment->bitmap_words), segment->cow_offset);
1422 	if (n != sizeof(segment->bitmap_words))
1423 		return map_error(-n);
1424 
1425 	return map_error(0);
1426 }
1427 
1428 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1429 {
1430 	char *buf = NULL;
1431 	unsigned long len;
1432 	int n, nsectors, start, end, bit;
1433 	__u64 off;
1434 
1435 	/* FLUSH is really a special case, we cannot "case" it with others */
1436 
1437 	if (req_op(req->req) == REQ_OP_FLUSH) {
1438 		/* fds[0] is always either the rw image or our cow file */
1439 		req->error = map_error(-os_sync_file(req->fds[0]));
1440 		return;
1441 	}
1442 
1443 	nsectors = desc->length / req->sectorsize;
1444 	start = 0;
1445 	do {
1446 		bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1447 		end = start;
1448 		while((end < nsectors) &&
1449 		      (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1450 			end++;
1451 
1452 		off = req->offset + req->offsets[bit] +
1453 			start * req->sectorsize;
1454 		len = (end - start) * req->sectorsize;
1455 		if (desc->buffer != NULL)
1456 			buf = &desc->buffer[start * req->sectorsize];
1457 
1458 		switch (req_op(req->req)) {
1459 		case REQ_OP_READ:
1460 			n = 0;
1461 			do {
1462 				buf = &buf[n];
1463 				len -= n;
1464 				n = os_pread_file(req->fds[bit], buf, len, off);
1465 				if (n < 0) {
1466 					req->error = map_error(-n);
1467 					return;
1468 				}
1469 			} while((n < len) && (n != 0));
1470 			if (n < len) memset(&buf[n], 0, len - n);
1471 			break;
1472 		case REQ_OP_WRITE:
1473 			n = os_pwrite_file(req->fds[bit], buf, len, off);
1474 			if(n != len){
1475 				req->error = map_error(-n);
1476 				return;
1477 			}
1478 			break;
1479 		case REQ_OP_DISCARD:
1480 			n = os_falloc_punch(req->fds[bit], off, len);
1481 			if (n) {
1482 				req->error = map_error(-n);
1483 				return;
1484 			}
1485 			break;
1486 		case REQ_OP_WRITE_ZEROES:
1487 			n = os_falloc_zeroes(req->fds[bit], off, len);
1488 			if (n) {
1489 				req->error = map_error(-n);
1490 				return;
1491 			}
1492 			break;
1493 		default:
1494 			WARN_ON_ONCE(1);
1495 			req->error = BLK_STS_NOTSUPP;
1496 			return;
1497 		}
1498 
1499 		start = end;
1500 	} while(start < nsectors);
1501 
1502 	req->offset += len;
1503 	req->error = update_bitmap(req, desc);
1504 }
1505 
1506 /* Changed in start_io_thread, which is serialized by being called only
1507  * from ubd_init, which is an initcall.
1508  */
1509 int kernel_fd = -1;
1510 
1511 /* Only changed by the io thread. XXX: currently unused. */
1512 static int io_count;
1513 
1514 int io_thread(void *arg)
1515 {
1516 	int n, count, written, res;
1517 
1518 	os_fix_helper_signals();
1519 
1520 	while(1){
1521 		n = bulk_req_safe_read(
1522 			kernel_fd,
1523 			io_req_buffer,
1524 			&io_remainder,
1525 			&io_remainder_size,
1526 			UBD_REQ_BUFFER_SIZE
1527 		);
1528 		if (n <= 0) {
1529 			if (n == -EAGAIN)
1530 				ubd_read_poll(-1);
1531 
1532 			continue;
1533 		}
1534 
1535 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1536 			struct io_thread_req *req = (*io_req_buffer)[count];
1537 			int i;
1538 
1539 			io_count++;
1540 			for (i = 0; !req->error && i < req->desc_cnt; i++)
1541 				do_io(req, &(req->io_desc[i]));
1542 
1543 		}
1544 
1545 		written = 0;
1546 
1547 		do {
1548 			res = os_write_file(kernel_fd,
1549 					    ((char *) io_req_buffer) + written,
1550 					    n - written);
1551 			if (res >= 0) {
1552 				written += res;
1553 			}
1554 			if (written < n) {
1555 				ubd_write_poll(-1);
1556 			}
1557 		} while (written < n);
1558 	}
1559 
1560 	return 0;
1561 }
1562