xref: /openbmc/linux/arch/um/drivers/ubd_kern.c (revision 6f4eaea2)
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/cdrom.h>
31 #include <linux/proc_fs.h>
32 #include <linux/seq_file.h>
33 #include <linux/ctype.h>
34 #include <linux/slab.h>
35 #include <linux/vmalloc.h>
36 #include <linux/platform_device.h>
37 #include <linux/scatterlist.h>
38 #include <asm/tlbflush.h>
39 #include <kern_util.h>
40 #include "mconsole_kern.h"
41 #include <init.h>
42 #include <irq_kern.h>
43 #include "ubd.h"
44 #include <os.h>
45 #include "cow.h"
46 
47 /* Max request size is determined by sector mask - 32K */
48 #define UBD_MAX_REQUEST (8 * sizeof(long))
49 
50 struct io_desc {
51 	char *buffer;
52 	unsigned long length;
53 	unsigned long sector_mask;
54 	unsigned long long cow_offset;
55 	unsigned long bitmap_words[2];
56 };
57 
58 struct io_thread_req {
59 	struct request *req;
60 	int fds[2];
61 	unsigned long offsets[2];
62 	unsigned long long offset;
63 	int sectorsize;
64 	int error;
65 
66 	int desc_cnt;
67 	/* io_desc has to be the last element of the struct */
68 	struct io_desc io_desc[];
69 };
70 
71 
72 static struct io_thread_req * (*irq_req_buffer)[];
73 static struct io_thread_req *irq_remainder;
74 static int irq_remainder_size;
75 
76 static struct io_thread_req * (*io_req_buffer)[];
77 static struct io_thread_req *io_remainder;
78 static int io_remainder_size;
79 
80 
81 
82 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
83 {
84 	__u64 n;
85 	int bits, off;
86 
87 	bits = sizeof(data[0]) * 8;
88 	n = bit / bits;
89 	off = bit % bits;
90 	return (data[n] & (1 << off)) != 0;
91 }
92 
93 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
94 {
95 	__u64 n;
96 	int bits, off;
97 
98 	bits = sizeof(data[0]) * 8;
99 	n = bit / bits;
100 	off = bit % bits;
101 	data[n] |= (1 << off);
102 }
103 /*End stuff from ubd_user.h*/
104 
105 #define DRIVER_NAME "uml-blkdev"
106 
107 static DEFINE_MUTEX(ubd_lock);
108 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
109 
110 static int ubd_open(struct block_device *bdev, fmode_t mode);
111 static void ubd_release(struct gendisk *disk, fmode_t mode);
112 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
113 		     unsigned int cmd, unsigned long arg);
114 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
115 
116 #define MAX_DEV (16)
117 
118 static const struct block_device_operations ubd_blops = {
119         .owner		= THIS_MODULE,
120         .open		= ubd_open,
121         .release	= ubd_release,
122         .ioctl		= ubd_ioctl,
123         .compat_ioctl	= blkdev_compat_ptr_ioctl,
124 	.getgeo		= ubd_getgeo,
125 };
126 
127 /* Protected by ubd_lock */
128 static int fake_major = UBD_MAJOR;
129 static struct gendisk *ubd_gendisk[MAX_DEV];
130 static struct gendisk *fake_gendisk[MAX_DEV];
131 
132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
134 					 .cl = 1 })
135 #else
136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
137 					 .cl = 1 })
138 #endif
139 static struct openflags global_openflags = OPEN_FLAGS;
140 
141 struct cow {
142 	/* backing file name */
143 	char *file;
144 	/* backing file fd */
145 	int fd;
146 	unsigned long *bitmap;
147 	unsigned long bitmap_len;
148 	int bitmap_offset;
149 	int data_offset;
150 };
151 
152 #define MAX_SG 64
153 
154 struct ubd {
155 	/* name (and fd, below) of the file opened for writing, either the
156 	 * backing or the cow file. */
157 	char *file;
158 	char *serial;
159 	int count;
160 	int fd;
161 	__u64 size;
162 	struct openflags boot_openflags;
163 	struct openflags openflags;
164 	unsigned shared:1;
165 	unsigned no_cow:1;
166 	unsigned no_trim:1;
167 	struct cow cow;
168 	struct platform_device pdev;
169 	struct request_queue *queue;
170 	struct blk_mq_tag_set tag_set;
171 	spinlock_t lock;
172 };
173 
174 #define DEFAULT_COW { \
175 	.file =			NULL, \
176 	.fd =			-1,	\
177 	.bitmap =		NULL, \
178 	.bitmap_offset =	0, \
179 	.data_offset =		0, \
180 }
181 
182 #define DEFAULT_UBD { \
183 	.file = 		NULL, \
184 	.serial =		NULL, \
185 	.count =		0, \
186 	.fd =			-1, \
187 	.size =			-1, \
188 	.boot_openflags =	OPEN_FLAGS, \
189 	.openflags =		OPEN_FLAGS, \
190 	.no_cow =               0, \
191 	.no_trim =		0, \
192 	.shared =		0, \
193 	.cow =			DEFAULT_COW, \
194 	.lock =			__SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
195 }
196 
197 /* Protected by ubd_lock */
198 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
199 
200 /* Only changed by fake_ide_setup which is a setup */
201 static int fake_ide = 0;
202 static struct proc_dir_entry *proc_ide_root = NULL;
203 static struct proc_dir_entry *proc_ide = NULL;
204 
205 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
206 				 const struct blk_mq_queue_data *bd);
207 
208 static void make_proc_ide(void)
209 {
210 	proc_ide_root = proc_mkdir("ide", NULL);
211 	proc_ide = proc_mkdir("ide0", proc_ide_root);
212 }
213 
214 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
215 {
216 	seq_puts(m, "disk\n");
217 	return 0;
218 }
219 
220 static void make_ide_entries(const char *dev_name)
221 {
222 	struct proc_dir_entry *dir, *ent;
223 	char name[64];
224 
225 	if(proc_ide_root == NULL) make_proc_ide();
226 
227 	dir = proc_mkdir(dev_name, proc_ide);
228 	if(!dir) return;
229 
230 	ent = proc_create_single("media", S_IRUGO, dir,
231 			fake_ide_media_proc_show);
232 	if(!ent) return;
233 	snprintf(name, sizeof(name), "ide0/%s", dev_name);
234 	proc_symlink(dev_name, proc_ide_root, name);
235 }
236 
237 static int fake_ide_setup(char *str)
238 {
239 	fake_ide = 1;
240 	return 1;
241 }
242 
243 __setup("fake_ide", fake_ide_setup);
244 
245 __uml_help(fake_ide_setup,
246 "fake_ide\n"
247 "    Create ide0 entries that map onto ubd devices.\n\n"
248 );
249 
250 static int parse_unit(char **ptr)
251 {
252 	char *str = *ptr, *end;
253 	int n = -1;
254 
255 	if(isdigit(*str)) {
256 		n = simple_strtoul(str, &end, 0);
257 		if(end == str)
258 			return -1;
259 		*ptr = end;
260 	}
261 	else if (('a' <= *str) && (*str <= 'z')) {
262 		n = *str - 'a';
263 		str++;
264 		*ptr = str;
265 	}
266 	return n;
267 }
268 
269 /* If *index_out == -1 at exit, the passed option was a general one;
270  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
271  * should not be freed on exit.
272  */
273 static int ubd_setup_common(char *str, int *index_out, char **error_out)
274 {
275 	struct ubd *ubd_dev;
276 	struct openflags flags = global_openflags;
277 	char *file, *backing_file, *serial;
278 	int n, err = 0, i;
279 
280 	if(index_out) *index_out = -1;
281 	n = *str;
282 	if(n == '='){
283 		char *end;
284 		int major;
285 
286 		str++;
287 		if(!strcmp(str, "sync")){
288 			global_openflags = of_sync(global_openflags);
289 			return err;
290 		}
291 
292 		err = -EINVAL;
293 		major = simple_strtoul(str, &end, 0);
294 		if((*end != '\0') || (end == str)){
295 			*error_out = "Didn't parse major number";
296 			return err;
297 		}
298 
299 		mutex_lock(&ubd_lock);
300 		if (fake_major != UBD_MAJOR) {
301 			*error_out = "Can't assign a fake major twice";
302 			goto out1;
303 		}
304 
305 		fake_major = major;
306 
307 		printk(KERN_INFO "Setting extra ubd major number to %d\n",
308 		       major);
309 		err = 0;
310 	out1:
311 		mutex_unlock(&ubd_lock);
312 		return err;
313 	}
314 
315 	n = parse_unit(&str);
316 	if(n < 0){
317 		*error_out = "Couldn't parse device number";
318 		return -EINVAL;
319 	}
320 	if(n >= MAX_DEV){
321 		*error_out = "Device number out of range";
322 		return 1;
323 	}
324 
325 	err = -EBUSY;
326 	mutex_lock(&ubd_lock);
327 
328 	ubd_dev = &ubd_devs[n];
329 	if(ubd_dev->file != NULL){
330 		*error_out = "Device is already configured";
331 		goto out;
332 	}
333 
334 	if (index_out)
335 		*index_out = n;
336 
337 	err = -EINVAL;
338 	for (i = 0; i < sizeof("rscdt="); i++) {
339 		switch (*str) {
340 		case 'r':
341 			flags.w = 0;
342 			break;
343 		case 's':
344 			flags.s = 1;
345 			break;
346 		case 'd':
347 			ubd_dev->no_cow = 1;
348 			break;
349 		case 'c':
350 			ubd_dev->shared = 1;
351 			break;
352 		case 't':
353 			ubd_dev->no_trim = 1;
354 			break;
355 		case '=':
356 			str++;
357 			goto break_loop;
358 		default:
359 			*error_out = "Expected '=' or flag letter "
360 				"(r, s, c, t or d)";
361 			goto out;
362 		}
363 		str++;
364 	}
365 
366 	if (*str == '=')
367 		*error_out = "Too many flags specified";
368 	else
369 		*error_out = "Missing '='";
370 	goto out;
371 
372 break_loop:
373 	file = strsep(&str, ",:");
374 	if (*file == '\0')
375 		file = NULL;
376 
377 	backing_file = strsep(&str, ",:");
378 	if (backing_file && *backing_file == '\0')
379 		backing_file = NULL;
380 
381 	serial = strsep(&str, ",:");
382 	if (serial && *serial == '\0')
383 		serial = NULL;
384 
385 	if (backing_file && ubd_dev->no_cow) {
386 		*error_out = "Can't specify both 'd' and a cow file";
387 		goto out;
388 	}
389 
390 	err = 0;
391 	ubd_dev->file = file;
392 	ubd_dev->cow.file = backing_file;
393 	ubd_dev->serial = serial;
394 	ubd_dev->boot_openflags = flags;
395 out:
396 	mutex_unlock(&ubd_lock);
397 	return err;
398 }
399 
400 static int ubd_setup(char *str)
401 {
402 	char *error;
403 	int err;
404 
405 	err = ubd_setup_common(str, NULL, &error);
406 	if(err)
407 		printk(KERN_ERR "Failed to initialize device with \"%s\" : "
408 		       "%s\n", str, error);
409 	return 1;
410 }
411 
412 __setup("ubd", ubd_setup);
413 __uml_help(ubd_setup,
414 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
415 "    This is used to associate a device with a file in the underlying\n"
416 "    filesystem. When specifying two filenames, the first one is the\n"
417 "    COW name and the second is the backing file name. As separator you can\n"
418 "    use either a ':' or a ',': the first one allows writing things like;\n"
419 "	ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
420 "    while with a ',' the shell would not expand the 2nd '~'.\n"
421 "    When using only one filename, UML will detect whether to treat it like\n"
422 "    a COW file or a backing file. To override this detection, add the 'd'\n"
423 "    flag:\n"
424 "	ubd0d=BackingFile\n"
425 "    Usually, there is a filesystem in the file, but \n"
426 "    that's not required. Swap devices containing swap files can be\n"
427 "    specified like this. Also, a file which doesn't contain a\n"
428 "    filesystem can have its contents read in the virtual \n"
429 "    machine by running 'dd' on the device. <n> must be in the range\n"
430 "    0 to 7. Appending an 'r' to the number will cause that device\n"
431 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
432 "    an 's' will cause data to be written to disk on the host immediately.\n"
433 "    'c' will cause the device to be treated as being shared between multiple\n"
434 "    UMLs and file locking will be turned off - this is appropriate for a\n"
435 "    cluster filesystem and inappropriate at almost all other times.\n\n"
436 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
437 "    An optional device serial number can be exposed using the serial parameter\n"
438 "    on the cmdline which is exposed as a sysfs entry. This is particularly\n"
439 "    useful when a unique number should be given to the device. Note when\n"
440 "    specifying a label, the filename2 must be also presented. It can be\n"
441 "    an empty string, in which case the backing file is not used:\n"
442 "       ubd0=File,,Serial\n"
443 );
444 
445 static int udb_setup(char *str)
446 {
447 	printk("udb%s specified on command line is almost certainly a ubd -> "
448 	       "udb TYPO\n", str);
449 	return 1;
450 }
451 
452 __setup("udb", udb_setup);
453 __uml_help(udb_setup,
454 "udb\n"
455 "    This option is here solely to catch ubd -> udb typos, which can be\n"
456 "    to impossible to catch visually unless you specifically look for\n"
457 "    them.  The only result of any option starting with 'udb' is an error\n"
458 "    in the boot output.\n\n"
459 );
460 
461 /* Only changed by ubd_init, which is an initcall. */
462 static int thread_fd = -1;
463 
464 /* Function to read several request pointers at a time
465 * handling fractional reads if (and as) needed
466 */
467 
468 static int bulk_req_safe_read(
469 	int fd,
470 	struct io_thread_req * (*request_buffer)[],
471 	struct io_thread_req **remainder,
472 	int *remainder_size,
473 	int max_recs
474 	)
475 {
476 	int n = 0;
477 	int res = 0;
478 
479 	if (*remainder_size > 0) {
480 		memmove(
481 			(char *) request_buffer,
482 			(char *) remainder, *remainder_size
483 		);
484 		n = *remainder_size;
485 	}
486 
487 	res = os_read_file(
488 			fd,
489 			((char *) request_buffer) + *remainder_size,
490 			sizeof(struct io_thread_req *)*max_recs
491 				- *remainder_size
492 		);
493 	if (res > 0) {
494 		n += res;
495 		if ((n % sizeof(struct io_thread_req *)) > 0) {
496 			/*
497 			* Read somehow returned not a multiple of dword
498 			* theoretically possible, but never observed in the
499 			* wild, so read routine must be able to handle it
500 			*/
501 			*remainder_size = n % sizeof(struct io_thread_req *);
502 			WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
503 			memmove(
504 				remainder,
505 				((char *) request_buffer) +
506 					(n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
507 				*remainder_size
508 			);
509 			n = n - *remainder_size;
510 		}
511 	} else {
512 		n = res;
513 	}
514 	return n;
515 }
516 
517 /* Called without dev->lock held, and only in interrupt context. */
518 static void ubd_handler(void)
519 {
520 	int n;
521 	int count;
522 
523 	while(1){
524 		n = bulk_req_safe_read(
525 			thread_fd,
526 			irq_req_buffer,
527 			&irq_remainder,
528 			&irq_remainder_size,
529 			UBD_REQ_BUFFER_SIZE
530 		);
531 		if (n < 0) {
532 			if(n == -EAGAIN)
533 				break;
534 			printk(KERN_ERR "spurious interrupt in ubd_handler, "
535 			       "err = %d\n", -n);
536 			return;
537 		}
538 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
539 			struct io_thread_req *io_req = (*irq_req_buffer)[count];
540 
541 			if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
542 				blk_queue_max_discard_sectors(io_req->req->q, 0);
543 				blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
544 				blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
545 			}
546 			blk_mq_end_request(io_req->req, io_req->error);
547 			kfree(io_req);
548 		}
549 	}
550 }
551 
552 static irqreturn_t ubd_intr(int irq, void *dev)
553 {
554 	ubd_handler();
555 	return IRQ_HANDLED;
556 }
557 
558 /* Only changed by ubd_init, which is an initcall. */
559 static int io_pid = -1;
560 
561 static void kill_io_thread(void)
562 {
563 	if(io_pid != -1)
564 		os_kill_process(io_pid, 1);
565 }
566 
567 __uml_exitcall(kill_io_thread);
568 
569 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
570 {
571 	char *file;
572 	int fd;
573 	int err;
574 
575 	__u32 version;
576 	__u32 align;
577 	char *backing_file;
578 	time64_t mtime;
579 	unsigned long long size;
580 	int sector_size;
581 	int bitmap_offset;
582 
583 	if (ubd_dev->file && ubd_dev->cow.file) {
584 		file = ubd_dev->cow.file;
585 
586 		goto out;
587 	}
588 
589 	fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
590 	if (fd < 0)
591 		return fd;
592 
593 	err = read_cow_header(file_reader, &fd, &version, &backing_file, \
594 		&mtime, &size, &sector_size, &align, &bitmap_offset);
595 	os_close_file(fd);
596 
597 	if(err == -EINVAL)
598 		file = ubd_dev->file;
599 	else
600 		file = backing_file;
601 
602 out:
603 	return os_file_size(file, size_out);
604 }
605 
606 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
607 {
608 	int err;
609 
610 	err = os_pread_file(fd, buf, len, offset);
611 	if (err < 0)
612 		return err;
613 
614 	return 0;
615 }
616 
617 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
618 {
619 	time64_t modtime;
620 	unsigned long long actual;
621 	int err;
622 
623 	err = os_file_modtime(file, &modtime);
624 	if (err < 0) {
625 		printk(KERN_ERR "Failed to get modification time of backing "
626 		       "file \"%s\", err = %d\n", file, -err);
627 		return err;
628 	}
629 
630 	err = os_file_size(file, &actual);
631 	if (err < 0) {
632 		printk(KERN_ERR "Failed to get size of backing file \"%s\", "
633 		       "err = %d\n", file, -err);
634 		return err;
635 	}
636 
637 	if (actual != size) {
638 		/*__u64 can be a long on AMD64 and with %lu GCC complains; so
639 		 * the typecast.*/
640 		printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
641 		       "vs backing file\n", (unsigned long long) size, actual);
642 		return -EINVAL;
643 	}
644 	if (modtime != mtime) {
645 		printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
646 		       "backing file\n", mtime, modtime);
647 		return -EINVAL;
648 	}
649 	return 0;
650 }
651 
652 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
653 {
654 	struct uml_stat buf1, buf2;
655 	int err;
656 
657 	if (from_cmdline == NULL)
658 		return 0;
659 	if (!strcmp(from_cmdline, from_cow))
660 		return 0;
661 
662 	err = os_stat_file(from_cmdline, &buf1);
663 	if (err < 0) {
664 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
665 		       -err);
666 		return 0;
667 	}
668 	err = os_stat_file(from_cow, &buf2);
669 	if (err < 0) {
670 		printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
671 		       -err);
672 		return 1;
673 	}
674 	if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
675 		return 0;
676 
677 	printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
678 	       "\"%s\" specified in COW header of \"%s\"\n",
679 	       from_cmdline, from_cow, cow);
680 	return 1;
681 }
682 
683 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
684 		  char **backing_file_out, int *bitmap_offset_out,
685 		  unsigned long *bitmap_len_out, int *data_offset_out,
686 		  int *create_cow_out)
687 {
688 	time64_t mtime;
689 	unsigned long long size;
690 	__u32 version, align;
691 	char *backing_file;
692 	int fd, err, sectorsize, asked_switch, mode = 0644;
693 
694 	fd = os_open_file(file, *openflags, mode);
695 	if (fd < 0) {
696 		if ((fd == -ENOENT) && (create_cow_out != NULL))
697 			*create_cow_out = 1;
698 		if (!openflags->w ||
699 		    ((fd != -EROFS) && (fd != -EACCES)))
700 			return fd;
701 		openflags->w = 0;
702 		fd = os_open_file(file, *openflags, mode);
703 		if (fd < 0)
704 			return fd;
705 	}
706 
707 	if (shared)
708 		printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
709 	else {
710 		err = os_lock_file(fd, openflags->w);
711 		if (err < 0) {
712 			printk(KERN_ERR "Failed to lock '%s', err = %d\n",
713 			       file, -err);
714 			goto out_close;
715 		}
716 	}
717 
718 	/* Successful return case! */
719 	if (backing_file_out == NULL)
720 		return fd;
721 
722 	err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
723 			      &size, &sectorsize, &align, bitmap_offset_out);
724 	if (err && (*backing_file_out != NULL)) {
725 		printk(KERN_ERR "Failed to read COW header from COW file "
726 		       "\"%s\", errno = %d\n", file, -err);
727 		goto out_close;
728 	}
729 	if (err)
730 		return fd;
731 
732 	asked_switch = path_requires_switch(*backing_file_out, backing_file,
733 					    file);
734 
735 	/* Allow switching only if no mismatch. */
736 	if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
737 						   mtime)) {
738 		printk(KERN_ERR "Switching backing file to '%s'\n",
739 		       *backing_file_out);
740 		err = write_cow_header(file, fd, *backing_file_out,
741 				       sectorsize, align, &size);
742 		if (err) {
743 			printk(KERN_ERR "Switch failed, errno = %d\n", -err);
744 			goto out_close;
745 		}
746 	} else {
747 		*backing_file_out = backing_file;
748 		err = backing_file_mismatch(*backing_file_out, size, mtime);
749 		if (err)
750 			goto out_close;
751 	}
752 
753 	cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
754 		  bitmap_len_out, data_offset_out);
755 
756 	return fd;
757  out_close:
758 	os_close_file(fd);
759 	return err;
760 }
761 
762 static int create_cow_file(char *cow_file, char *backing_file,
763 		    struct openflags flags,
764 		    int sectorsize, int alignment, int *bitmap_offset_out,
765 		    unsigned long *bitmap_len_out, int *data_offset_out)
766 {
767 	int err, fd;
768 
769 	flags.c = 1;
770 	fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
771 	if (fd < 0) {
772 		err = fd;
773 		printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
774 		       cow_file, -err);
775 		goto out;
776 	}
777 
778 	err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
779 			    bitmap_offset_out, bitmap_len_out,
780 			    data_offset_out);
781 	if (!err)
782 		return fd;
783 	os_close_file(fd);
784  out:
785 	return err;
786 }
787 
788 static void ubd_close_dev(struct ubd *ubd_dev)
789 {
790 	os_close_file(ubd_dev->fd);
791 	if(ubd_dev->cow.file == NULL)
792 		return;
793 
794 	os_close_file(ubd_dev->cow.fd);
795 	vfree(ubd_dev->cow.bitmap);
796 	ubd_dev->cow.bitmap = NULL;
797 }
798 
799 static int ubd_open_dev(struct ubd *ubd_dev)
800 {
801 	struct openflags flags;
802 	char **back_ptr;
803 	int err, create_cow, *create_ptr;
804 	int fd;
805 
806 	ubd_dev->openflags = ubd_dev->boot_openflags;
807 	create_cow = 0;
808 	create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
809 	back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
810 
811 	fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
812 				back_ptr, &ubd_dev->cow.bitmap_offset,
813 				&ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
814 				create_ptr);
815 
816 	if((fd == -ENOENT) && create_cow){
817 		fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
818 					  ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
819 					  &ubd_dev->cow.bitmap_offset,
820 					  &ubd_dev->cow.bitmap_len,
821 					  &ubd_dev->cow.data_offset);
822 		if(fd >= 0){
823 			printk(KERN_INFO "Creating \"%s\" as COW file for "
824 			       "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
825 		}
826 	}
827 
828 	if(fd < 0){
829 		printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
830 		       -fd);
831 		return fd;
832 	}
833 	ubd_dev->fd = fd;
834 
835 	if(ubd_dev->cow.file != NULL){
836 		blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
837 
838 		err = -ENOMEM;
839 		ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
840 		if(ubd_dev->cow.bitmap == NULL){
841 			printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
842 			goto error;
843 		}
844 		flush_tlb_kernel_vm();
845 
846 		err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
847 				      ubd_dev->cow.bitmap_offset,
848 				      ubd_dev->cow.bitmap_len);
849 		if(err < 0)
850 			goto error;
851 
852 		flags = ubd_dev->openflags;
853 		flags.w = 0;
854 		err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
855 				    NULL, NULL, NULL, NULL);
856 		if(err < 0) goto error;
857 		ubd_dev->cow.fd = err;
858 	}
859 	if (ubd_dev->no_trim == 0) {
860 		ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
861 		ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
862 		blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
863 		blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
864 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
865 	}
866 	blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
867 	return 0;
868  error:
869 	os_close_file(ubd_dev->fd);
870 	return err;
871 }
872 
873 static void ubd_device_release(struct device *dev)
874 {
875 	struct ubd *ubd_dev = dev_get_drvdata(dev);
876 
877 	blk_cleanup_queue(ubd_dev->queue);
878 	blk_mq_free_tag_set(&ubd_dev->tag_set);
879 	*ubd_dev = ((struct ubd) DEFAULT_UBD);
880 }
881 
882 static ssize_t serial_show(struct device *dev,
883 			   struct device_attribute *attr, char *buf)
884 {
885 	struct gendisk *disk = dev_to_disk(dev);
886 	struct ubd *ubd_dev = disk->private_data;
887 
888 	if (!ubd_dev)
889 		return 0;
890 
891 	return sprintf(buf, "%s", ubd_dev->serial);
892 }
893 
894 static DEVICE_ATTR_RO(serial);
895 
896 static struct attribute *ubd_attrs[] = {
897 	&dev_attr_serial.attr,
898 	NULL,
899 };
900 
901 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
902 				     struct attribute *a, int n)
903 {
904 	return a->mode;
905 }
906 
907 static const struct attribute_group ubd_attr_group = {
908 	.attrs = ubd_attrs,
909 	.is_visible = ubd_attrs_are_visible,
910 };
911 
912 static const struct attribute_group *ubd_attr_groups[] = {
913 	&ubd_attr_group,
914 	NULL,
915 };
916 
917 static int ubd_disk_register(int major, u64 size, int unit,
918 			     struct gendisk **disk_out)
919 {
920 	struct device *parent = NULL;
921 	struct gendisk *disk;
922 
923 	disk = alloc_disk(1 << UBD_SHIFT);
924 	if(disk == NULL)
925 		return -ENOMEM;
926 
927 	disk->major = major;
928 	disk->first_minor = unit << UBD_SHIFT;
929 	disk->fops = &ubd_blops;
930 	set_capacity(disk, size / 512);
931 	if (major == UBD_MAJOR)
932 		sprintf(disk->disk_name, "ubd%c", 'a' + unit);
933 	else
934 		sprintf(disk->disk_name, "ubd_fake%d", unit);
935 
936 	/* sysfs register (not for ide fake devices) */
937 	if (major == UBD_MAJOR) {
938 		ubd_devs[unit].pdev.id   = unit;
939 		ubd_devs[unit].pdev.name = DRIVER_NAME;
940 		ubd_devs[unit].pdev.dev.release = ubd_device_release;
941 		dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
942 		platform_device_register(&ubd_devs[unit].pdev);
943 		parent = &ubd_devs[unit].pdev.dev;
944 	}
945 
946 	disk->private_data = &ubd_devs[unit];
947 	disk->queue = ubd_devs[unit].queue;
948 	device_add_disk(parent, disk, ubd_attr_groups);
949 
950 	*disk_out = disk;
951 	return 0;
952 }
953 
954 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
955 
956 static const struct blk_mq_ops ubd_mq_ops = {
957 	.queue_rq = ubd_queue_rq,
958 };
959 
960 static int ubd_add(int n, char **error_out)
961 {
962 	struct ubd *ubd_dev = &ubd_devs[n];
963 	int err = 0;
964 
965 	if(ubd_dev->file == NULL)
966 		goto out;
967 
968 	err = ubd_file_size(ubd_dev, &ubd_dev->size);
969 	if(err < 0){
970 		*error_out = "Couldn't determine size of device's file";
971 		goto out;
972 	}
973 
974 	ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
975 
976 	ubd_dev->tag_set.ops = &ubd_mq_ops;
977 	ubd_dev->tag_set.queue_depth = 64;
978 	ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
979 	ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
980 	ubd_dev->tag_set.driver_data = ubd_dev;
981 	ubd_dev->tag_set.nr_hw_queues = 1;
982 
983 	err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
984 	if (err)
985 		goto out;
986 
987 	ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
988 	if (IS_ERR(ubd_dev->queue)) {
989 		err = PTR_ERR(ubd_dev->queue);
990 		goto out_cleanup_tags;
991 	}
992 
993 	ubd_dev->queue->queuedata = ubd_dev;
994 	blk_queue_write_cache(ubd_dev->queue, true, false);
995 
996 	blk_queue_max_segments(ubd_dev->queue, MAX_SG);
997 	blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
998 	err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
999 	if(err){
1000 		*error_out = "Failed to register device";
1001 		goto out_cleanup_tags;
1002 	}
1003 
1004 	if (fake_major != UBD_MAJOR)
1005 		ubd_disk_register(fake_major, ubd_dev->size, n,
1006 				  &fake_gendisk[n]);
1007 
1008 	/*
1009 	 * Perhaps this should also be under the "if (fake_major)" above
1010 	 * using the fake_disk->disk_name
1011 	 */
1012 	if (fake_ide)
1013 		make_ide_entries(ubd_gendisk[n]->disk_name);
1014 
1015 	err = 0;
1016 out:
1017 	return err;
1018 
1019 out_cleanup_tags:
1020 	blk_mq_free_tag_set(&ubd_dev->tag_set);
1021 	if (!(IS_ERR(ubd_dev->queue)))
1022 		blk_cleanup_queue(ubd_dev->queue);
1023 	goto out;
1024 }
1025 
1026 static int ubd_config(char *str, char **error_out)
1027 {
1028 	int n, ret;
1029 
1030 	/* This string is possibly broken up and stored, so it's only
1031 	 * freed if ubd_setup_common fails, or if only general options
1032 	 * were set.
1033 	 */
1034 	str = kstrdup(str, GFP_KERNEL);
1035 	if (str == NULL) {
1036 		*error_out = "Failed to allocate memory";
1037 		return -ENOMEM;
1038 	}
1039 
1040 	ret = ubd_setup_common(str, &n, error_out);
1041 	if (ret)
1042 		goto err_free;
1043 
1044 	if (n == -1) {
1045 		ret = 0;
1046 		goto err_free;
1047 	}
1048 
1049 	mutex_lock(&ubd_lock);
1050 	ret = ubd_add(n, error_out);
1051 	if (ret)
1052 		ubd_devs[n].file = NULL;
1053 	mutex_unlock(&ubd_lock);
1054 
1055 out:
1056 	return ret;
1057 
1058 err_free:
1059 	kfree(str);
1060 	goto out;
1061 }
1062 
1063 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1064 {
1065 	struct ubd *ubd_dev;
1066 	int n, len = 0;
1067 
1068 	n = parse_unit(&name);
1069 	if((n >= MAX_DEV) || (n < 0)){
1070 		*error_out = "ubd_get_config : device number out of range";
1071 		return -1;
1072 	}
1073 
1074 	ubd_dev = &ubd_devs[n];
1075 	mutex_lock(&ubd_lock);
1076 
1077 	if(ubd_dev->file == NULL){
1078 		CONFIG_CHUNK(str, size, len, "", 1);
1079 		goto out;
1080 	}
1081 
1082 	CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1083 
1084 	if(ubd_dev->cow.file != NULL){
1085 		CONFIG_CHUNK(str, size, len, ",", 0);
1086 		CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1087 	}
1088 	else CONFIG_CHUNK(str, size, len, "", 1);
1089 
1090  out:
1091 	mutex_unlock(&ubd_lock);
1092 	return len;
1093 }
1094 
1095 static int ubd_id(char **str, int *start_out, int *end_out)
1096 {
1097 	int n;
1098 
1099 	n = parse_unit(str);
1100 	*start_out = 0;
1101 	*end_out = MAX_DEV - 1;
1102 	return n;
1103 }
1104 
1105 static int ubd_remove(int n, char **error_out)
1106 {
1107 	struct gendisk *disk = ubd_gendisk[n];
1108 	struct ubd *ubd_dev;
1109 	int err = -ENODEV;
1110 
1111 	mutex_lock(&ubd_lock);
1112 
1113 	ubd_dev = &ubd_devs[n];
1114 
1115 	if(ubd_dev->file == NULL)
1116 		goto out;
1117 
1118 	/* you cannot remove a open disk */
1119 	err = -EBUSY;
1120 	if(ubd_dev->count > 0)
1121 		goto out;
1122 
1123 	ubd_gendisk[n] = NULL;
1124 	if(disk != NULL){
1125 		del_gendisk(disk);
1126 		put_disk(disk);
1127 	}
1128 
1129 	if(fake_gendisk[n] != NULL){
1130 		del_gendisk(fake_gendisk[n]);
1131 		put_disk(fake_gendisk[n]);
1132 		fake_gendisk[n] = NULL;
1133 	}
1134 
1135 	err = 0;
1136 	platform_device_unregister(&ubd_dev->pdev);
1137 out:
1138 	mutex_unlock(&ubd_lock);
1139 	return err;
1140 }
1141 
1142 /* All these are called by mconsole in process context and without
1143  * ubd-specific locks.  The structure itself is const except for .list.
1144  */
1145 static struct mc_device ubd_mc = {
1146 	.list		= LIST_HEAD_INIT(ubd_mc.list),
1147 	.name		= "ubd",
1148 	.config		= ubd_config,
1149 	.get_config	= ubd_get_config,
1150 	.id		= ubd_id,
1151 	.remove		= ubd_remove,
1152 };
1153 
1154 static int __init ubd_mc_init(void)
1155 {
1156 	mconsole_register_dev(&ubd_mc);
1157 	return 0;
1158 }
1159 
1160 __initcall(ubd_mc_init);
1161 
1162 static int __init ubd0_init(void)
1163 {
1164 	struct ubd *ubd_dev = &ubd_devs[0];
1165 
1166 	mutex_lock(&ubd_lock);
1167 	if(ubd_dev->file == NULL)
1168 		ubd_dev->file = "root_fs";
1169 	mutex_unlock(&ubd_lock);
1170 
1171 	return 0;
1172 }
1173 
1174 __initcall(ubd0_init);
1175 
1176 /* Used in ubd_init, which is an initcall */
1177 static struct platform_driver ubd_driver = {
1178 	.driver = {
1179 		.name  = DRIVER_NAME,
1180 	},
1181 };
1182 
1183 static int __init ubd_init(void)
1184 {
1185 	char *error;
1186 	int i, err;
1187 
1188 	if (register_blkdev(UBD_MAJOR, "ubd"))
1189 		return -1;
1190 
1191 	if (fake_major != UBD_MAJOR) {
1192 		char name[sizeof("ubd_nnn\0")];
1193 
1194 		snprintf(name, sizeof(name), "ubd_%d", fake_major);
1195 		if (register_blkdev(fake_major, "ubd"))
1196 			return -1;
1197 	}
1198 
1199 	irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1200 				       sizeof(struct io_thread_req *),
1201 				       GFP_KERNEL
1202 		);
1203 	irq_remainder = 0;
1204 
1205 	if (irq_req_buffer == NULL) {
1206 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1207 		return -1;
1208 	}
1209 	io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1210 				      sizeof(struct io_thread_req *),
1211 				      GFP_KERNEL
1212 		);
1213 
1214 	io_remainder = 0;
1215 
1216 	if (io_req_buffer == NULL) {
1217 		printk(KERN_ERR "Failed to initialize ubd buffering\n");
1218 		return -1;
1219 	}
1220 	platform_driver_register(&ubd_driver);
1221 	mutex_lock(&ubd_lock);
1222 	for (i = 0; i < MAX_DEV; i++){
1223 		err = ubd_add(i, &error);
1224 		if(err)
1225 			printk(KERN_ERR "Failed to initialize ubd device %d :"
1226 			       "%s\n", i, error);
1227 	}
1228 	mutex_unlock(&ubd_lock);
1229 	return 0;
1230 }
1231 
1232 late_initcall(ubd_init);
1233 
1234 static int __init ubd_driver_init(void){
1235 	unsigned long stack;
1236 	int err;
1237 
1238 	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1239 	if(global_openflags.s){
1240 		printk(KERN_INFO "ubd: Synchronous mode\n");
1241 		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1242 		 * enough. So use anyway the io thread. */
1243 	}
1244 	stack = alloc_stack(0, 0);
1245 	io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1246 				 &thread_fd);
1247 	if(io_pid < 0){
1248 		printk(KERN_ERR
1249 		       "ubd : Failed to start I/O thread (errno = %d) - "
1250 		       "falling back to synchronous I/O\n", -io_pid);
1251 		io_pid = -1;
1252 		return 0;
1253 	}
1254 	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1255 			     0, "ubd", ubd_devs);
1256 	if(err < 0)
1257 		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1258 	return 0;
1259 }
1260 
1261 device_initcall(ubd_driver_init);
1262 
1263 static int ubd_open(struct block_device *bdev, fmode_t mode)
1264 {
1265 	struct gendisk *disk = bdev->bd_disk;
1266 	struct ubd *ubd_dev = disk->private_data;
1267 	int err = 0;
1268 
1269 	mutex_lock(&ubd_mutex);
1270 	if(ubd_dev->count == 0){
1271 		err = ubd_open_dev(ubd_dev);
1272 		if(err){
1273 			printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1274 			       disk->disk_name, ubd_dev->file, -err);
1275 			goto out;
1276 		}
1277 	}
1278 	ubd_dev->count++;
1279 	set_disk_ro(disk, !ubd_dev->openflags.w);
1280 
1281 	/* This should no more be needed. And it didn't work anyway to exclude
1282 	 * read-write remounting of filesystems.*/
1283 	/*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1284 	        if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1285 	        err = -EROFS;
1286 	}*/
1287 out:
1288 	mutex_unlock(&ubd_mutex);
1289 	return err;
1290 }
1291 
1292 static void ubd_release(struct gendisk *disk, fmode_t mode)
1293 {
1294 	struct ubd *ubd_dev = disk->private_data;
1295 
1296 	mutex_lock(&ubd_mutex);
1297 	if(--ubd_dev->count == 0)
1298 		ubd_close_dev(ubd_dev);
1299 	mutex_unlock(&ubd_mutex);
1300 }
1301 
1302 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1303 			  __u64 *cow_offset, unsigned long *bitmap,
1304 			  __u64 bitmap_offset, unsigned long *bitmap_words,
1305 			  __u64 bitmap_len)
1306 {
1307 	__u64 sector = io_offset >> SECTOR_SHIFT;
1308 	int i, update_bitmap = 0;
1309 
1310 	for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1311 		if(cow_mask != NULL)
1312 			ubd_set_bit(i, (unsigned char *) cow_mask);
1313 		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1314 			continue;
1315 
1316 		update_bitmap = 1;
1317 		ubd_set_bit(sector + i, (unsigned char *) bitmap);
1318 	}
1319 
1320 	if(!update_bitmap)
1321 		return;
1322 
1323 	*cow_offset = sector / (sizeof(unsigned long) * 8);
1324 
1325 	/* This takes care of the case where we're exactly at the end of the
1326 	 * device, and *cow_offset + 1 is off the end.  So, just back it up
1327 	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1328 	 * for the original diagnosis.
1329 	 */
1330 	if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1331 					 sizeof(unsigned long)) - 1))
1332 		(*cow_offset)--;
1333 
1334 	bitmap_words[0] = bitmap[*cow_offset];
1335 	bitmap_words[1] = bitmap[*cow_offset + 1];
1336 
1337 	*cow_offset *= sizeof(unsigned long);
1338 	*cow_offset += bitmap_offset;
1339 }
1340 
1341 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1342 		       unsigned long offset, unsigned long *bitmap,
1343 		       __u64 bitmap_offset, __u64 bitmap_len)
1344 {
1345 	__u64 sector = offset >> SECTOR_SHIFT;
1346 	int i;
1347 
1348 	if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1349 		panic("Operation too long");
1350 
1351 	if (req_op(req->req) == REQ_OP_READ) {
1352 		for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1353 			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1354 				ubd_set_bit(i, (unsigned char *)
1355 					    &segment->sector_mask);
1356 		}
1357 	} else {
1358 		cowify_bitmap(offset, segment->length, &segment->sector_mask,
1359 			      &segment->cow_offset, bitmap, bitmap_offset,
1360 			      segment->bitmap_words, bitmap_len);
1361 	}
1362 }
1363 
1364 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1365 			struct request *req)
1366 {
1367 	struct bio_vec bvec;
1368 	struct req_iterator iter;
1369 	int i = 0;
1370 	unsigned long byte_offset = io_req->offset;
1371 	int op = req_op(req);
1372 
1373 	if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1374 		io_req->io_desc[0].buffer = NULL;
1375 		io_req->io_desc[0].length = blk_rq_bytes(req);
1376 	} else {
1377 		rq_for_each_segment(bvec, req, iter) {
1378 			BUG_ON(i >= io_req->desc_cnt);
1379 
1380 			io_req->io_desc[i].buffer =
1381 				page_address(bvec.bv_page) + bvec.bv_offset;
1382 			io_req->io_desc[i].length = bvec.bv_len;
1383 			i++;
1384 		}
1385 	}
1386 
1387 	if (dev->cow.file) {
1388 		for (i = 0; i < io_req->desc_cnt; i++) {
1389 			cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1390 				   dev->cow.bitmap, dev->cow.bitmap_offset,
1391 				   dev->cow.bitmap_len);
1392 			byte_offset += io_req->io_desc[i].length;
1393 		}
1394 
1395 	}
1396 }
1397 
1398 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1399 					   int desc_cnt)
1400 {
1401 	struct io_thread_req *io_req;
1402 	int i;
1403 
1404 	io_req = kmalloc(sizeof(*io_req) +
1405 			 (desc_cnt * sizeof(struct io_desc)),
1406 			 GFP_ATOMIC);
1407 	if (!io_req)
1408 		return NULL;
1409 
1410 	io_req->req = req;
1411 	if (dev->cow.file)
1412 		io_req->fds[0] = dev->cow.fd;
1413 	else
1414 		io_req->fds[0] = dev->fd;
1415 	io_req->error = 0;
1416 	io_req->sectorsize = SECTOR_SIZE;
1417 	io_req->fds[1] = dev->fd;
1418 	io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1419 	io_req->offsets[0] = 0;
1420 	io_req->offsets[1] = dev->cow.data_offset;
1421 
1422 	for (i = 0 ; i < desc_cnt; i++) {
1423 		io_req->io_desc[i].sector_mask = 0;
1424 		io_req->io_desc[i].cow_offset = -1;
1425 	}
1426 
1427 	return io_req;
1428 }
1429 
1430 static int ubd_submit_request(struct ubd *dev, struct request *req)
1431 {
1432 	int segs = 0;
1433 	struct io_thread_req *io_req;
1434 	int ret;
1435 	int op = req_op(req);
1436 
1437 	if (op == REQ_OP_FLUSH)
1438 		segs = 0;
1439 	else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1440 		segs = 1;
1441 	else
1442 		segs = blk_rq_nr_phys_segments(req);
1443 
1444 	io_req = ubd_alloc_req(dev, req, segs);
1445 	if (!io_req)
1446 		return -ENOMEM;
1447 
1448 	io_req->desc_cnt = segs;
1449 	if (segs)
1450 		ubd_map_req(dev, io_req, req);
1451 
1452 	ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1453 	if (ret != sizeof(io_req)) {
1454 		if (ret != -EAGAIN)
1455 			pr_err("write to io thread failed: %d\n", -ret);
1456 		kfree(io_req);
1457 	}
1458 	return ret;
1459 }
1460 
1461 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1462 				 const struct blk_mq_queue_data *bd)
1463 {
1464 	struct ubd *ubd_dev = hctx->queue->queuedata;
1465 	struct request *req = bd->rq;
1466 	int ret = 0, res = BLK_STS_OK;
1467 
1468 	blk_mq_start_request(req);
1469 
1470 	spin_lock_irq(&ubd_dev->lock);
1471 
1472 	switch (req_op(req)) {
1473 	case REQ_OP_FLUSH:
1474 	case REQ_OP_READ:
1475 	case REQ_OP_WRITE:
1476 	case REQ_OP_DISCARD:
1477 	case REQ_OP_WRITE_ZEROES:
1478 		ret = ubd_submit_request(ubd_dev, req);
1479 		break;
1480 	default:
1481 		WARN_ON_ONCE(1);
1482 		res = BLK_STS_NOTSUPP;
1483 	}
1484 
1485 	spin_unlock_irq(&ubd_dev->lock);
1486 
1487 	if (ret < 0) {
1488 		if (ret == -ENOMEM)
1489 			res = BLK_STS_RESOURCE;
1490 		else
1491 			res = BLK_STS_DEV_RESOURCE;
1492 	}
1493 
1494 	return res;
1495 }
1496 
1497 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1498 {
1499 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1500 
1501 	geo->heads = 128;
1502 	geo->sectors = 32;
1503 	geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1504 	return 0;
1505 }
1506 
1507 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1508 		     unsigned int cmd, unsigned long arg)
1509 {
1510 	struct ubd *ubd_dev = bdev->bd_disk->private_data;
1511 	u16 ubd_id[ATA_ID_WORDS];
1512 
1513 	switch (cmd) {
1514 		struct cdrom_volctrl volume;
1515 	case HDIO_GET_IDENTITY:
1516 		memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1517 		ubd_id[ATA_ID_CYLS]	= ubd_dev->size / (128 * 32 * 512);
1518 		ubd_id[ATA_ID_HEADS]	= 128;
1519 		ubd_id[ATA_ID_SECTORS]	= 32;
1520 		if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1521 				 sizeof(ubd_id)))
1522 			return -EFAULT;
1523 		return 0;
1524 
1525 	case CDROMVOLREAD:
1526 		if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1527 			return -EFAULT;
1528 		volume.channel0 = 255;
1529 		volume.channel1 = 255;
1530 		volume.channel2 = 255;
1531 		volume.channel3 = 255;
1532 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1533 			return -EFAULT;
1534 		return 0;
1535 	}
1536 	return -EINVAL;
1537 }
1538 
1539 static int map_error(int error_code)
1540 {
1541 	switch (error_code) {
1542 	case 0:
1543 		return BLK_STS_OK;
1544 	case ENOSYS:
1545 	case EOPNOTSUPP:
1546 		return BLK_STS_NOTSUPP;
1547 	case ENOSPC:
1548 		return BLK_STS_NOSPC;
1549 	}
1550 	return BLK_STS_IOERR;
1551 }
1552 
1553 /*
1554  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1555  *
1556  * The following functions are part of UML hypervisor code.
1557  * All functions from here onwards are executed as a helper
1558  * thread and are not allowed to execute any kernel functions.
1559  *
1560  * Any communication must occur strictly via shared memory and IPC.
1561  *
1562  * Do not add printks, locks, kernel memory operations, etc - it
1563  * will result in unpredictable behaviour and/or crashes.
1564  */
1565 
1566 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1567 {
1568 	int n;
1569 
1570 	if (segment->cow_offset == -1)
1571 		return map_error(0);
1572 
1573 	n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1574 			  sizeof(segment->bitmap_words), segment->cow_offset);
1575 	if (n != sizeof(segment->bitmap_words))
1576 		return map_error(-n);
1577 
1578 	return map_error(0);
1579 }
1580 
1581 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1582 {
1583 	char *buf = NULL;
1584 	unsigned long len;
1585 	int n, nsectors, start, end, bit;
1586 	__u64 off;
1587 
1588 	/* FLUSH is really a special case, we cannot "case" it with others */
1589 
1590 	if (req_op(req->req) == REQ_OP_FLUSH) {
1591 		/* fds[0] is always either the rw image or our cow file */
1592 		req->error = map_error(-os_sync_file(req->fds[0]));
1593 		return;
1594 	}
1595 
1596 	nsectors = desc->length / req->sectorsize;
1597 	start = 0;
1598 	do {
1599 		bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1600 		end = start;
1601 		while((end < nsectors) &&
1602 		      (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1603 			end++;
1604 
1605 		off = req->offset + req->offsets[bit] +
1606 			start * req->sectorsize;
1607 		len = (end - start) * req->sectorsize;
1608 		if (desc->buffer != NULL)
1609 			buf = &desc->buffer[start * req->sectorsize];
1610 
1611 		switch (req_op(req->req)) {
1612 		case REQ_OP_READ:
1613 			n = 0;
1614 			do {
1615 				buf = &buf[n];
1616 				len -= n;
1617 				n = os_pread_file(req->fds[bit], buf, len, off);
1618 				if (n < 0) {
1619 					req->error = map_error(-n);
1620 					return;
1621 				}
1622 			} while((n < len) && (n != 0));
1623 			if (n < len) memset(&buf[n], 0, len - n);
1624 			break;
1625 		case REQ_OP_WRITE:
1626 			n = os_pwrite_file(req->fds[bit], buf, len, off);
1627 			if(n != len){
1628 				req->error = map_error(-n);
1629 				return;
1630 			}
1631 			break;
1632 		case REQ_OP_DISCARD:
1633 		case REQ_OP_WRITE_ZEROES:
1634 			n = os_falloc_punch(req->fds[bit], off, len);
1635 			if (n) {
1636 				req->error = map_error(-n);
1637 				return;
1638 			}
1639 			break;
1640 		default:
1641 			WARN_ON_ONCE(1);
1642 			req->error = BLK_STS_NOTSUPP;
1643 			return;
1644 		}
1645 
1646 		start = end;
1647 	} while(start < nsectors);
1648 
1649 	req->offset += len;
1650 	req->error = update_bitmap(req, desc);
1651 }
1652 
1653 /* Changed in start_io_thread, which is serialized by being called only
1654  * from ubd_init, which is an initcall.
1655  */
1656 int kernel_fd = -1;
1657 
1658 /* Only changed by the io thread. XXX: currently unused. */
1659 static int io_count = 0;
1660 
1661 int io_thread(void *arg)
1662 {
1663 	int n, count, written, res;
1664 
1665 	os_fix_helper_signals();
1666 
1667 	while(1){
1668 		n = bulk_req_safe_read(
1669 			kernel_fd,
1670 			io_req_buffer,
1671 			&io_remainder,
1672 			&io_remainder_size,
1673 			UBD_REQ_BUFFER_SIZE
1674 		);
1675 		if (n <= 0) {
1676 			if (n == -EAGAIN)
1677 				ubd_read_poll(-1);
1678 
1679 			continue;
1680 		}
1681 
1682 		for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1683 			struct io_thread_req *req = (*io_req_buffer)[count];
1684 			int i;
1685 
1686 			io_count++;
1687 			for (i = 0; !req->error && i < req->desc_cnt; i++)
1688 				do_io(req, &(req->io_desc[i]));
1689 
1690 		}
1691 
1692 		written = 0;
1693 
1694 		do {
1695 			res = os_write_file(kernel_fd,
1696 					    ((char *) io_req_buffer) + written,
1697 					    n - written);
1698 			if (res >= 0) {
1699 				written += res;
1700 			}
1701 			if (written < n) {
1702 				ubd_write_poll(-1);
1703 			}
1704 		} while (written < n);
1705 	}
1706 
1707 	return 0;
1708 }
1709